ref: fe7cabe8b6ef9d2d16c25eb3de15d856d7fd8ace
parent: a73f0f4170b54ef5f8ab82f4801b39afd43be903
parent: 7971846a5ef42d22385a4dffd1a2aa0b6ee93728
author: Hui Su <huisu@google.com>
date: Tue Jul 28 16:41:01 EDT 2015
Merge "Move intra prediction functions from vp9/common/ to vpx_dsp/"
--- a/build/make/rtcd.pl
+++ b/build/make/rtcd.pl
@@ -319,6 +319,8 @@
print <<EOF;
#if HAVE_DSPR2
+void vpx_dsputil_static_init();
+vpx_dsputil_static_init();
#if CONFIG_VP8
void dsputil_static_init();
dsputil_static_init();
--- a/test/test_intra_pred_speed.cc
+++ b/test/test_intra_pred_speed.cc
@@ -14,7 +14,7 @@
#include "third_party/googletest/src/include/gtest/gtest.h"
-#include "./vp9_rtcd.h"
+#include "./vpx_dsp_rtcd.h"
#include "test/acm_random.h"
#include "test/clear_system_state.h"
#include "test/md5_helper.h"
--- a/test/vp9_intrapred_test.cc
+++ b/test/vp9_intrapred_test.cc
@@ -16,7 +16,7 @@
#include "third_party/googletest/src/include/gtest/gtest.h"
#include "./vpx_config.h"
-#include "./vp9_rtcd.h"
+#include "./vpx_dsp_rtcd.h"
#include "vp9/common/vp9_blockd.h"
#include "vp9/common/vp9_pred_common.h"
#include "vpx_mem/vpx_mem.h"
--- a/vp9/common/arm/neon/vp9_reconintra_neon.c
+++ /dev/null
@@ -1,822 +1,0 @@
-/*
- * Copyright (c) 2014 The WebM project authors. All Rights Reserved.
- *
- * Use of this source code is governed by a BSD-style license
- * that can be found in the LICENSE file in the root of the source
- * tree. An additional intellectual property rights grant can be found
- * in the file PATENTS. All contributing project authors may
- * be found in the AUTHORS file in the root of the source tree.
- */
-
-#include <arm_neon.h>
-
-#include "./vp9_rtcd.h"
-#include "./vpx_config.h"
-#include "vpx/vpx_integer.h"
-
-//------------------------------------------------------------------------------
-// DC 4x4
-
-// 'do_above' and 'do_left' facilitate branch removal when inlined.
-static INLINE void dc_4x4(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above, const uint8_t *left,
- int do_above, int do_left) {- uint16x8_t sum_top;
- uint16x8_t sum_left;
- uint8x8_t dc0;
-
- if (do_above) {- const uint8x8_t A = vld1_u8(above); // top row
- const uint16x4_t p0 = vpaddl_u8(A); // cascading summation of the top
- const uint16x4_t p1 = vpadd_u16(p0, p0);
- sum_top = vcombine_u16(p1, p1);
- }
-
- if (do_left) {- const uint8x8_t L = vld1_u8(left); // left border
- const uint16x4_t p0 = vpaddl_u8(L); // cascading summation of the left
- const uint16x4_t p1 = vpadd_u16(p0, p0);
- sum_left = vcombine_u16(p1, p1);
- }
-
- if (do_above && do_left) {- const uint16x8_t sum = vaddq_u16(sum_left, sum_top);
- dc0 = vrshrn_n_u16(sum, 3);
- } else if (do_above) {- dc0 = vrshrn_n_u16(sum_top, 2);
- } else if (do_left) {- dc0 = vrshrn_n_u16(sum_left, 2);
- } else {- dc0 = vdup_n_u8(0x80);
- }
-
- {- const uint8x8_t dc = vdup_lane_u8(dc0, 0);
- int i;
- for (i = 0; i < 4; ++i) {- vst1_lane_u32((uint32_t*)(dst + i * stride), vreinterpret_u32_u8(dc), 0);
- }
- }
-}
-
-void vp9_dc_predictor_4x4_neon(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above, const uint8_t *left) {- dc_4x4(dst, stride, above, left, 1, 1);
-}
-
-void vp9_dc_left_predictor_4x4_neon(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above, const uint8_t *left) {- (void)above;
- dc_4x4(dst, stride, NULL, left, 0, 1);
-}
-
-void vp9_dc_top_predictor_4x4_neon(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above, const uint8_t *left) {- (void)left;
- dc_4x4(dst, stride, above, NULL, 1, 0);
-}
-
-void vp9_dc_128_predictor_4x4_neon(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above, const uint8_t *left) {- (void)above;
- (void)left;
- dc_4x4(dst, stride, NULL, NULL, 0, 0);
-}
-
-//------------------------------------------------------------------------------
-// DC 8x8
-
-// 'do_above' and 'do_left' facilitate branch removal when inlined.
-static INLINE void dc_8x8(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above, const uint8_t *left,
- int do_above, int do_left) {- uint16x8_t sum_top;
- uint16x8_t sum_left;
- uint8x8_t dc0;
-
- if (do_above) {- const uint8x8_t A = vld1_u8(above); // top row
- const uint16x4_t p0 = vpaddl_u8(A); // cascading summation of the top
- const uint16x4_t p1 = vpadd_u16(p0, p0);
- const uint16x4_t p2 = vpadd_u16(p1, p1);
- sum_top = vcombine_u16(p2, p2);
- }
-
- if (do_left) {- const uint8x8_t L = vld1_u8(left); // left border
- const uint16x4_t p0 = vpaddl_u8(L); // cascading summation of the left
- const uint16x4_t p1 = vpadd_u16(p0, p0);
- const uint16x4_t p2 = vpadd_u16(p1, p1);
- sum_left = vcombine_u16(p2, p2);
- }
-
- if (do_above && do_left) {- const uint16x8_t sum = vaddq_u16(sum_left, sum_top);
- dc0 = vrshrn_n_u16(sum, 4);
- } else if (do_above) {- dc0 = vrshrn_n_u16(sum_top, 3);
- } else if (do_left) {- dc0 = vrshrn_n_u16(sum_left, 3);
- } else {- dc0 = vdup_n_u8(0x80);
- }
-
- {- const uint8x8_t dc = vdup_lane_u8(dc0, 0);
- int i;
- for (i = 0; i < 8; ++i) {- vst1_u32((uint32_t*)(dst + i * stride), vreinterpret_u32_u8(dc));
- }
- }
-}
-
-void vp9_dc_predictor_8x8_neon(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above, const uint8_t *left) {- dc_8x8(dst, stride, above, left, 1, 1);
-}
-
-void vp9_dc_left_predictor_8x8_neon(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above, const uint8_t *left) {- (void)above;
- dc_8x8(dst, stride, NULL, left, 0, 1);
-}
-
-void vp9_dc_top_predictor_8x8_neon(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above, const uint8_t *left) {- (void)left;
- dc_8x8(dst, stride, above, NULL, 1, 0);
-}
-
-void vp9_dc_128_predictor_8x8_neon(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above, const uint8_t *left) {- (void)above;
- (void)left;
- dc_8x8(dst, stride, NULL, NULL, 0, 0);
-}
-
-//------------------------------------------------------------------------------
-// DC 16x16
-
-// 'do_above' and 'do_left' facilitate branch removal when inlined.
-static INLINE void dc_16x16(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above, const uint8_t *left,
- int do_above, int do_left) {- uint16x8_t sum_top;
- uint16x8_t sum_left;
- uint8x8_t dc0;
-
- if (do_above) {- const uint8x16_t A = vld1q_u8(above); // top row
- const uint16x8_t p0 = vpaddlq_u8(A); // cascading summation of the top
- const uint16x4_t p1 = vadd_u16(vget_low_u16(p0), vget_high_u16(p0));
- const uint16x4_t p2 = vpadd_u16(p1, p1);
- const uint16x4_t p3 = vpadd_u16(p2, p2);
- sum_top = vcombine_u16(p3, p3);
- }
-
- if (do_left) {- const uint8x16_t L = vld1q_u8(left); // left row
- const uint16x8_t p0 = vpaddlq_u8(L); // cascading summation of the left
- const uint16x4_t p1 = vadd_u16(vget_low_u16(p0), vget_high_u16(p0));
- const uint16x4_t p2 = vpadd_u16(p1, p1);
- const uint16x4_t p3 = vpadd_u16(p2, p2);
- sum_left = vcombine_u16(p3, p3);
- }
-
- if (do_above && do_left) {- const uint16x8_t sum = vaddq_u16(sum_left, sum_top);
- dc0 = vrshrn_n_u16(sum, 5);
- } else if (do_above) {- dc0 = vrshrn_n_u16(sum_top, 4);
- } else if (do_left) {- dc0 = vrshrn_n_u16(sum_left, 4);
- } else {- dc0 = vdup_n_u8(0x80);
- }
-
- {- const uint8x16_t dc = vdupq_lane_u8(dc0, 0);
- int i;
- for (i = 0; i < 16; ++i) {- vst1q_u8(dst + i * stride, dc);
- }
- }
-}
-
-void vp9_dc_predictor_16x16_neon(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above, const uint8_t *left) {- dc_16x16(dst, stride, above, left, 1, 1);
-}
-
-void vp9_dc_left_predictor_16x16_neon(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above,
- const uint8_t *left) {- (void)above;
- dc_16x16(dst, stride, NULL, left, 0, 1);
-}
-
-void vp9_dc_top_predictor_16x16_neon(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above,
- const uint8_t *left) {- (void)left;
- dc_16x16(dst, stride, above, NULL, 1, 0);
-}
-
-void vp9_dc_128_predictor_16x16_neon(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above,
- const uint8_t *left) {- (void)above;
- (void)left;
- dc_16x16(dst, stride, NULL, NULL, 0, 0);
-}
-
-//------------------------------------------------------------------------------
-// DC 32x32
-
-// 'do_above' and 'do_left' facilitate branch removal when inlined.
-static INLINE void dc_32x32(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above, const uint8_t *left,
- int do_above, int do_left) {- uint16x8_t sum_top;
- uint16x8_t sum_left;
- uint8x8_t dc0;
-
- if (do_above) {- const uint8x16_t A0 = vld1q_u8(above); // top row
- const uint8x16_t A1 = vld1q_u8(above + 16);
- const uint16x8_t p0 = vpaddlq_u8(A0); // cascading summation of the top
- const uint16x8_t p1 = vpaddlq_u8(A1);
- const uint16x8_t p2 = vaddq_u16(p0, p1);
- const uint16x4_t p3 = vadd_u16(vget_low_u16(p2), vget_high_u16(p2));
- const uint16x4_t p4 = vpadd_u16(p3, p3);
- const uint16x4_t p5 = vpadd_u16(p4, p4);
- sum_top = vcombine_u16(p5, p5);
- }
-
- if (do_left) {- const uint8x16_t L0 = vld1q_u8(left); // left row
- const uint8x16_t L1 = vld1q_u8(left + 16);
- const uint16x8_t p0 = vpaddlq_u8(L0); // cascading summation of the left
- const uint16x8_t p1 = vpaddlq_u8(L1);
- const uint16x8_t p2 = vaddq_u16(p0, p1);
- const uint16x4_t p3 = vadd_u16(vget_low_u16(p2), vget_high_u16(p2));
- const uint16x4_t p4 = vpadd_u16(p3, p3);
- const uint16x4_t p5 = vpadd_u16(p4, p4);
- sum_left = vcombine_u16(p5, p5);
- }
-
- if (do_above && do_left) {- const uint16x8_t sum = vaddq_u16(sum_left, sum_top);
- dc0 = vrshrn_n_u16(sum, 6);
- } else if (do_above) {- dc0 = vrshrn_n_u16(sum_top, 5);
- } else if (do_left) {- dc0 = vrshrn_n_u16(sum_left, 5);
- } else {- dc0 = vdup_n_u8(0x80);
- }
-
- {- const uint8x16_t dc = vdupq_lane_u8(dc0, 0);
- int i;
- for (i = 0; i < 32; ++i) {- vst1q_u8(dst + i * stride, dc);
- vst1q_u8(dst + i * stride + 16, dc);
- }
- }
-}
-
-void vp9_dc_predictor_32x32_neon(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above, const uint8_t *left) {- dc_32x32(dst, stride, above, left, 1, 1);
-}
-
-void vp9_dc_left_predictor_32x32_neon(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above,
- const uint8_t *left) {- (void)above;
- dc_32x32(dst, stride, NULL, left, 0, 1);
-}
-
-void vp9_dc_top_predictor_32x32_neon(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above,
- const uint8_t *left) {- (void)left;
- dc_32x32(dst, stride, above, NULL, 1, 0);
-}
-
-void vp9_dc_128_predictor_32x32_neon(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above,
- const uint8_t *left) {- (void)above;
- (void)left;
- dc_32x32(dst, stride, NULL, NULL, 0, 0);
-}
-
-// -----------------------------------------------------------------------------
-
-void vp9_d45_predictor_4x4_neon(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above, const uint8_t *left) {- const uint64x1_t A0 = vreinterpret_u64_u8(vld1_u8(above)); // top row
- const uint64x1_t A1 = vshr_n_u64(A0, 8);
- const uint64x1_t A2 = vshr_n_u64(A0, 16);
- const uint8x8_t ABCDEFGH = vreinterpret_u8_u64(A0);
- const uint8x8_t BCDEFGH0 = vreinterpret_u8_u64(A1);
- const uint8x8_t CDEFGH00 = vreinterpret_u8_u64(A2);
- const uint8x8_t avg1 = vhadd_u8(ABCDEFGH, CDEFGH00);
- const uint8x8_t avg2 = vrhadd_u8(avg1, BCDEFGH0);
- const uint64x1_t avg2_u64 = vreinterpret_u64_u8(avg2);
- const uint32x2_t r0 = vreinterpret_u32_u8(avg2);
- const uint32x2_t r1 = vreinterpret_u32_u64(vshr_n_u64(avg2_u64, 8));
- const uint32x2_t r2 = vreinterpret_u32_u64(vshr_n_u64(avg2_u64, 16));
- const uint32x2_t r3 = vreinterpret_u32_u64(vshr_n_u64(avg2_u64, 24));
- (void)left;
- vst1_lane_u32((uint32_t *)(dst + 0 * stride), r0, 0);
- vst1_lane_u32((uint32_t *)(dst + 1 * stride), r1, 0);
- vst1_lane_u32((uint32_t *)(dst + 2 * stride), r2, 0);
- vst1_lane_u32((uint32_t *)(dst + 3 * stride), r3, 0);
- dst[3 * stride + 3] = above[7];
-}
-
-void vp9_d45_predictor_8x8_neon(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above, const uint8_t *left) {- static const uint8_t shuffle1[8] = { 1, 2, 3, 4, 5, 6, 7, 7 };- static const uint8_t shuffle2[8] = { 2, 3, 4, 5, 6, 7, 7, 7 };- const uint8x8_t sh_12345677 = vld1_u8(shuffle1);
- const uint8x8_t sh_23456777 = vld1_u8(shuffle2);
- const uint8x8_t A0 = vld1_u8(above); // top row
- const uint8x8_t A1 = vtbl1_u8(A0, sh_12345677);
- const uint8x8_t A2 = vtbl1_u8(A0, sh_23456777);
- const uint8x8_t avg1 = vhadd_u8(A0, A2);
- uint8x8_t row = vrhadd_u8(avg1, A1);
- int i;
- (void)left;
- for (i = 0; i < 7; ++i) {- vst1_u8(dst + i * stride, row);
- row = vtbl1_u8(row, sh_12345677);
- }
- vst1_u8(dst + i * stride, row);
-}
-
-void vp9_d45_predictor_16x16_neon(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above, const uint8_t *left) {- const uint8x16_t A0 = vld1q_u8(above); // top row
- const uint8x16_t above_right = vld1q_dup_u8(above + 15);
- const uint8x16_t A1 = vextq_u8(A0, above_right, 1);
- const uint8x16_t A2 = vextq_u8(A0, above_right, 2);
- const uint8x16_t avg1 = vhaddq_u8(A0, A2);
- uint8x16_t row = vrhaddq_u8(avg1, A1);
- int i;
- (void)left;
- for (i = 0; i < 15; ++i) {- vst1q_u8(dst + i * stride, row);
- row = vextq_u8(row, above_right, 1);
- }
- vst1q_u8(dst + i * stride, row);
-}
-
-// -----------------------------------------------------------------------------
-
-void vp9_d135_predictor_4x4_neon(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above, const uint8_t *left) {- const uint8x8_t XABCD_u8 = vld1_u8(above - 1);
- const uint64x1_t XABCD = vreinterpret_u64_u8(XABCD_u8);
- const uint64x1_t ____XABC = vshl_n_u64(XABCD, 32);
- const uint32x2_t zero = vdup_n_u32(0);
- const uint32x2_t IJKL = vld1_lane_u32((const uint32_t *)left, zero, 0);
- const uint8x8_t IJKL_u8 = vreinterpret_u8_u32(IJKL);
- const uint64x1_t LKJI____ = vreinterpret_u64_u8(vrev32_u8(IJKL_u8));
- const uint64x1_t LKJIXABC = vorr_u64(LKJI____, ____XABC);
- const uint8x8_t KJIXABC_ = vreinterpret_u8_u64(vshr_n_u64(LKJIXABC, 8));
- const uint8x8_t JIXABC__ = vreinterpret_u8_u64(vshr_n_u64(LKJIXABC, 16));
- const uint8_t D = vget_lane_u8(XABCD_u8, 4);
- const uint8x8_t JIXABCD_ = vset_lane_u8(D, JIXABC__, 6);
- const uint8x8_t LKJIXABC_u8 = vreinterpret_u8_u64(LKJIXABC);
- const uint8x8_t avg1 = vhadd_u8(JIXABCD_, LKJIXABC_u8);
- const uint8x8_t avg2 = vrhadd_u8(avg1, KJIXABC_);
- const uint64x1_t avg2_u64 = vreinterpret_u64_u8(avg2);
- const uint32x2_t r3 = vreinterpret_u32_u8(avg2);
- const uint32x2_t r2 = vreinterpret_u32_u64(vshr_n_u64(avg2_u64, 8));
- const uint32x2_t r1 = vreinterpret_u32_u64(vshr_n_u64(avg2_u64, 16));
- const uint32x2_t r0 = vreinterpret_u32_u64(vshr_n_u64(avg2_u64, 24));
- vst1_lane_u32((uint32_t *)(dst + 0 * stride), r0, 0);
- vst1_lane_u32((uint32_t *)(dst + 1 * stride), r1, 0);
- vst1_lane_u32((uint32_t *)(dst + 2 * stride), r2, 0);
- vst1_lane_u32((uint32_t *)(dst + 3 * stride), r3, 0);
-}
-
-#if !HAVE_NEON_ASM
-
-void vp9_v_predictor_4x4_neon(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above, const uint8_t *left) {- int i;
- uint32x2_t d0u32 = vdup_n_u32(0);
- (void)left;
-
- d0u32 = vld1_lane_u32((const uint32_t *)above, d0u32, 0);
- for (i = 0; i < 4; i++, dst += stride)
- vst1_lane_u32((uint32_t *)dst, d0u32, 0);
-}
-
-void vp9_v_predictor_8x8_neon(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above, const uint8_t *left) {- int i;
- uint8x8_t d0u8 = vdup_n_u8(0);
- (void)left;
-
- d0u8 = vld1_u8(above);
- for (i = 0; i < 8; i++, dst += stride)
- vst1_u8(dst, d0u8);
-}
-
-void vp9_v_predictor_16x16_neon(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above, const uint8_t *left) {- int i;
- uint8x16_t q0u8 = vdupq_n_u8(0);
- (void)left;
-
- q0u8 = vld1q_u8(above);
- for (i = 0; i < 16; i++, dst += stride)
- vst1q_u8(dst, q0u8);
-}
-
-void vp9_v_predictor_32x32_neon(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above, const uint8_t *left) {- int i;
- uint8x16_t q0u8 = vdupq_n_u8(0);
- uint8x16_t q1u8 = vdupq_n_u8(0);
- (void)left;
-
- q0u8 = vld1q_u8(above);
- q1u8 = vld1q_u8(above + 16);
- for (i = 0; i < 32; i++, dst += stride) {- vst1q_u8(dst, q0u8);
- vst1q_u8(dst + 16, q1u8);
- }
-}
-
-void vp9_h_predictor_4x4_neon(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above, const uint8_t *left) {- uint8x8_t d0u8 = vdup_n_u8(0);
- uint32x2_t d1u32 = vdup_n_u32(0);
- (void)above;
-
- d1u32 = vld1_lane_u32((const uint32_t *)left, d1u32, 0);
-
- d0u8 = vdup_lane_u8(vreinterpret_u8_u32(d1u32), 0);
- vst1_lane_u32((uint32_t *)dst, vreinterpret_u32_u8(d0u8), 0);
- dst += stride;
- d0u8 = vdup_lane_u8(vreinterpret_u8_u32(d1u32), 1);
- vst1_lane_u32((uint32_t *)dst, vreinterpret_u32_u8(d0u8), 0);
- dst += stride;
- d0u8 = vdup_lane_u8(vreinterpret_u8_u32(d1u32), 2);
- vst1_lane_u32((uint32_t *)dst, vreinterpret_u32_u8(d0u8), 0);
- dst += stride;
- d0u8 = vdup_lane_u8(vreinterpret_u8_u32(d1u32), 3);
- vst1_lane_u32((uint32_t *)dst, vreinterpret_u32_u8(d0u8), 0);
-}
-
-void vp9_h_predictor_8x8_neon(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above, const uint8_t *left) {- uint8x8_t d0u8 = vdup_n_u8(0);
- uint64x1_t d1u64 = vdup_n_u64(0);
- (void)above;
-
- d1u64 = vld1_u64((const uint64_t *)left);
-
- d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 0);
- vst1_u8(dst, d0u8);
- dst += stride;
- d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 1);
- vst1_u8(dst, d0u8);
- dst += stride;
- d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 2);
- vst1_u8(dst, d0u8);
- dst += stride;
- d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 3);
- vst1_u8(dst, d0u8);
- dst += stride;
- d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 4);
- vst1_u8(dst, d0u8);
- dst += stride;
- d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 5);
- vst1_u8(dst, d0u8);
- dst += stride;
- d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 6);
- vst1_u8(dst, d0u8);
- dst += stride;
- d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 7);
- vst1_u8(dst, d0u8);
-}
-
-void vp9_h_predictor_16x16_neon(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above, const uint8_t *left) {- int j;
- uint8x8_t d2u8 = vdup_n_u8(0);
- uint8x16_t q0u8 = vdupq_n_u8(0);
- uint8x16_t q1u8 = vdupq_n_u8(0);
- (void)above;
-
- q1u8 = vld1q_u8(left);
- d2u8 = vget_low_u8(q1u8);
- for (j = 0; j < 2; j++, d2u8 = vget_high_u8(q1u8)) {- q0u8 = vdupq_lane_u8(d2u8, 0);
- vst1q_u8(dst, q0u8);
- dst += stride;
- q0u8 = vdupq_lane_u8(d2u8, 1);
- vst1q_u8(dst, q0u8);
- dst += stride;
- q0u8 = vdupq_lane_u8(d2u8, 2);
- vst1q_u8(dst, q0u8);
- dst += stride;
- q0u8 = vdupq_lane_u8(d2u8, 3);
- vst1q_u8(dst, q0u8);
- dst += stride;
- q0u8 = vdupq_lane_u8(d2u8, 4);
- vst1q_u8(dst, q0u8);
- dst += stride;
- q0u8 = vdupq_lane_u8(d2u8, 5);
- vst1q_u8(dst, q0u8);
- dst += stride;
- q0u8 = vdupq_lane_u8(d2u8, 6);
- vst1q_u8(dst, q0u8);
- dst += stride;
- q0u8 = vdupq_lane_u8(d2u8, 7);
- vst1q_u8(dst, q0u8);
- dst += stride;
- }
-}
-
-void vp9_h_predictor_32x32_neon(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above, const uint8_t *left) {- int j, k;
- uint8x8_t d2u8 = vdup_n_u8(0);
- uint8x16_t q0u8 = vdupq_n_u8(0);
- uint8x16_t q1u8 = vdupq_n_u8(0);
- (void)above;
-
- for (k = 0; k < 2; k++, left += 16) {- q1u8 = vld1q_u8(left);
- d2u8 = vget_low_u8(q1u8);
- for (j = 0; j < 2; j++, d2u8 = vget_high_u8(q1u8)) {- q0u8 = vdupq_lane_u8(d2u8, 0);
- vst1q_u8(dst, q0u8);
- vst1q_u8(dst + 16, q0u8);
- dst += stride;
- q0u8 = vdupq_lane_u8(d2u8, 1);
- vst1q_u8(dst, q0u8);
- vst1q_u8(dst + 16, q0u8);
- dst += stride;
- q0u8 = vdupq_lane_u8(d2u8, 2);
- vst1q_u8(dst, q0u8);
- vst1q_u8(dst + 16, q0u8);
- dst += stride;
- q0u8 = vdupq_lane_u8(d2u8, 3);
- vst1q_u8(dst, q0u8);
- vst1q_u8(dst + 16, q0u8);
- dst += stride;
- q0u8 = vdupq_lane_u8(d2u8, 4);
- vst1q_u8(dst, q0u8);
- vst1q_u8(dst + 16, q0u8);
- dst += stride;
- q0u8 = vdupq_lane_u8(d2u8, 5);
- vst1q_u8(dst, q0u8);
- vst1q_u8(dst + 16, q0u8);
- dst += stride;
- q0u8 = vdupq_lane_u8(d2u8, 6);
- vst1q_u8(dst, q0u8);
- vst1q_u8(dst + 16, q0u8);
- dst += stride;
- q0u8 = vdupq_lane_u8(d2u8, 7);
- vst1q_u8(dst, q0u8);
- vst1q_u8(dst + 16, q0u8);
- dst += stride;
- }
- }
-}
-
-void vp9_tm_predictor_4x4_neon(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above, const uint8_t *left) {- int i;
- uint16x8_t q1u16, q3u16;
- int16x8_t q1s16;
- uint8x8_t d0u8 = vdup_n_u8(0);
- uint32x2_t d2u32 = vdup_n_u32(0);
-
- d0u8 = vld1_dup_u8(above - 1);
- d2u32 = vld1_lane_u32((const uint32_t *)above, d2u32, 0);
- q3u16 = vsubl_u8(vreinterpret_u8_u32(d2u32), d0u8);
- for (i = 0; i < 4; i++, dst += stride) {- q1u16 = vdupq_n_u16((uint16_t)left[i]);
- q1s16 = vaddq_s16(vreinterpretq_s16_u16(q1u16),
- vreinterpretq_s16_u16(q3u16));
- d0u8 = vqmovun_s16(q1s16);
- vst1_lane_u32((uint32_t *)dst, vreinterpret_u32_u8(d0u8), 0);
- }
-}
-
-void vp9_tm_predictor_8x8_neon(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above, const uint8_t *left) {- int j;
- uint16x8_t q0u16, q3u16, q10u16;
- int16x8_t q0s16;
- uint16x4_t d20u16;
- uint8x8_t d0u8, d2u8, d30u8;
-
- d0u8 = vld1_dup_u8(above - 1);
- d30u8 = vld1_u8(left);
- d2u8 = vld1_u8(above);
- q10u16 = vmovl_u8(d30u8);
- q3u16 = vsubl_u8(d2u8, d0u8);
- d20u16 = vget_low_u16(q10u16);
- for (j = 0; j < 2; j++, d20u16 = vget_high_u16(q10u16)) {- q0u16 = vdupq_lane_u16(d20u16, 0);
- q0s16 = vaddq_s16(vreinterpretq_s16_u16(q3u16),
- vreinterpretq_s16_u16(q0u16));
- d0u8 = vqmovun_s16(q0s16);
- vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d0u8));
- dst += stride;
- q0u16 = vdupq_lane_u16(d20u16, 1);
- q0s16 = vaddq_s16(vreinterpretq_s16_u16(q3u16),
- vreinterpretq_s16_u16(q0u16));
- d0u8 = vqmovun_s16(q0s16);
- vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d0u8));
- dst += stride;
- q0u16 = vdupq_lane_u16(d20u16, 2);
- q0s16 = vaddq_s16(vreinterpretq_s16_u16(q3u16),
- vreinterpretq_s16_u16(q0u16));
- d0u8 = vqmovun_s16(q0s16);
- vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d0u8));
- dst += stride;
- q0u16 = vdupq_lane_u16(d20u16, 3);
- q0s16 = vaddq_s16(vreinterpretq_s16_u16(q3u16),
- vreinterpretq_s16_u16(q0u16));
- d0u8 = vqmovun_s16(q0s16);
- vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d0u8));
- dst += stride;
- }
-}
-
-void vp9_tm_predictor_16x16_neon(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above, const uint8_t *left) {- int j, k;
- uint16x8_t q0u16, q2u16, q3u16, q8u16, q10u16;
- uint8x16_t q0u8, q1u8;
- int16x8_t q0s16, q1s16, q8s16, q11s16;
- uint16x4_t d20u16;
- uint8x8_t d2u8, d3u8, d18u8, d22u8, d23u8;
-
- q0u8 = vld1q_dup_u8(above - 1);
- q1u8 = vld1q_u8(above);
- q2u16 = vsubl_u8(vget_low_u8(q1u8), vget_low_u8(q0u8));
- q3u16 = vsubl_u8(vget_high_u8(q1u8), vget_high_u8(q0u8));
- for (k = 0; k < 2; k++, left += 8) {- d18u8 = vld1_u8(left);
- q10u16 = vmovl_u8(d18u8);
- d20u16 = vget_low_u16(q10u16);
- for (j = 0; j < 2; j++, d20u16 = vget_high_u16(q10u16)) {- q0u16 = vdupq_lane_u16(d20u16, 0);
- q8u16 = vdupq_lane_u16(d20u16, 1);
- q1s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
- vreinterpretq_s16_u16(q2u16));
- q0s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
- vreinterpretq_s16_u16(q3u16));
- q11s16 = vaddq_s16(vreinterpretq_s16_u16(q8u16),
- vreinterpretq_s16_u16(q2u16));
- q8s16 = vaddq_s16(vreinterpretq_s16_u16(q8u16),
- vreinterpretq_s16_u16(q3u16));
- d2u8 = vqmovun_s16(q1s16);
- d3u8 = vqmovun_s16(q0s16);
- d22u8 = vqmovun_s16(q11s16);
- d23u8 = vqmovun_s16(q8s16);
- vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d2u8));
- vst1_u64((uint64_t *)(dst + 8), vreinterpret_u64_u8(d3u8));
- dst += stride;
- vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d22u8));
- vst1_u64((uint64_t *)(dst + 8), vreinterpret_u64_u8(d23u8));
- dst += stride;
-
- q0u16 = vdupq_lane_u16(d20u16, 2);
- q8u16 = vdupq_lane_u16(d20u16, 3);
- q1s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
- vreinterpretq_s16_u16(q2u16));
- q0s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
- vreinterpretq_s16_u16(q3u16));
- q11s16 = vaddq_s16(vreinterpretq_s16_u16(q8u16),
- vreinterpretq_s16_u16(q2u16));
- q8s16 = vaddq_s16(vreinterpretq_s16_u16(q8u16),
- vreinterpretq_s16_u16(q3u16));
- d2u8 = vqmovun_s16(q1s16);
- d3u8 = vqmovun_s16(q0s16);
- d22u8 = vqmovun_s16(q11s16);
- d23u8 = vqmovun_s16(q8s16);
- vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d2u8));
- vst1_u64((uint64_t *)(dst + 8), vreinterpret_u64_u8(d3u8));
- dst += stride;
- vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d22u8));
- vst1_u64((uint64_t *)(dst + 8), vreinterpret_u64_u8(d23u8));
- dst += stride;
- }
- }
-}
-
-void vp9_tm_predictor_32x32_neon(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above, const uint8_t *left) {- int j, k;
- uint16x8_t q0u16, q3u16, q8u16, q9u16, q10u16, q11u16;
- uint8x16_t q0u8, q1u8, q2u8;
- int16x8_t q12s16, q13s16, q14s16, q15s16;
- uint16x4_t d6u16;
- uint8x8_t d0u8, d1u8, d2u8, d3u8, d26u8;
-
- q0u8 = vld1q_dup_u8(above - 1);
- q1u8 = vld1q_u8(above);
- q2u8 = vld1q_u8(above + 16);
- q8u16 = vsubl_u8(vget_low_u8(q1u8), vget_low_u8(q0u8));
- q9u16 = vsubl_u8(vget_high_u8(q1u8), vget_high_u8(q0u8));
- q10u16 = vsubl_u8(vget_low_u8(q2u8), vget_low_u8(q0u8));
- q11u16 = vsubl_u8(vget_high_u8(q2u8), vget_high_u8(q0u8));
- for (k = 0; k < 4; k++, left += 8) {- d26u8 = vld1_u8(left);
- q3u16 = vmovl_u8(d26u8);
- d6u16 = vget_low_u16(q3u16);
- for (j = 0; j < 2; j++, d6u16 = vget_high_u16(q3u16)) {- q0u16 = vdupq_lane_u16(d6u16, 0);
- q12s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
- vreinterpretq_s16_u16(q8u16));
- q13s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
- vreinterpretq_s16_u16(q9u16));
- q14s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
- vreinterpretq_s16_u16(q10u16));
- q15s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
- vreinterpretq_s16_u16(q11u16));
- d0u8 = vqmovun_s16(q12s16);
- d1u8 = vqmovun_s16(q13s16);
- d2u8 = vqmovun_s16(q14s16);
- d3u8 = vqmovun_s16(q15s16);
- q0u8 = vcombine_u8(d0u8, d1u8);
- q1u8 = vcombine_u8(d2u8, d3u8);
- vst1q_u64((uint64_t *)dst, vreinterpretq_u64_u8(q0u8));
- vst1q_u64((uint64_t *)(dst + 16), vreinterpretq_u64_u8(q1u8));
- dst += stride;
-
- q0u16 = vdupq_lane_u16(d6u16, 1);
- q12s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
- vreinterpretq_s16_u16(q8u16));
- q13s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
- vreinterpretq_s16_u16(q9u16));
- q14s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
- vreinterpretq_s16_u16(q10u16));
- q15s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
- vreinterpretq_s16_u16(q11u16));
- d0u8 = vqmovun_s16(q12s16);
- d1u8 = vqmovun_s16(q13s16);
- d2u8 = vqmovun_s16(q14s16);
- d3u8 = vqmovun_s16(q15s16);
- q0u8 = vcombine_u8(d0u8, d1u8);
- q1u8 = vcombine_u8(d2u8, d3u8);
- vst1q_u64((uint64_t *)dst, vreinterpretq_u64_u8(q0u8));
- vst1q_u64((uint64_t *)(dst + 16), vreinterpretq_u64_u8(q1u8));
- dst += stride;
-
- q0u16 = vdupq_lane_u16(d6u16, 2);
- q12s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
- vreinterpretq_s16_u16(q8u16));
- q13s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
- vreinterpretq_s16_u16(q9u16));
- q14s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
- vreinterpretq_s16_u16(q10u16));
- q15s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
- vreinterpretq_s16_u16(q11u16));
- d0u8 = vqmovun_s16(q12s16);
- d1u8 = vqmovun_s16(q13s16);
- d2u8 = vqmovun_s16(q14s16);
- d3u8 = vqmovun_s16(q15s16);
- q0u8 = vcombine_u8(d0u8, d1u8);
- q1u8 = vcombine_u8(d2u8, d3u8);
- vst1q_u64((uint64_t *)dst, vreinterpretq_u64_u8(q0u8));
- vst1q_u64((uint64_t *)(dst + 16), vreinterpretq_u64_u8(q1u8));
- dst += stride;
-
- q0u16 = vdupq_lane_u16(d6u16, 3);
- q12s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
- vreinterpretq_s16_u16(q8u16));
- q13s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
- vreinterpretq_s16_u16(q9u16));
- q14s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
- vreinterpretq_s16_u16(q10u16));
- q15s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
- vreinterpretq_s16_u16(q11u16));
- d0u8 = vqmovun_s16(q12s16);
- d1u8 = vqmovun_s16(q13s16);
- d2u8 = vqmovun_s16(q14s16);
- d3u8 = vqmovun_s16(q15s16);
- q0u8 = vcombine_u8(d0u8, d1u8);
- q1u8 = vcombine_u8(d2u8, d3u8);
- vst1q_u64((uint64_t *)dst, vreinterpretq_u64_u8(q0u8));
- vst1q_u64((uint64_t *)(dst + 16), vreinterpretq_u64_u8(q1u8));
- dst += stride;
- }
- }
-}
-#endif // !HAVE_NEON_ASM
--- a/vp9/common/arm/neon/vp9_reconintra_neon_asm.asm
+++ /dev/null
@@ -1,630 +1,0 @@
-;
-; Copyright (c) 2014 The WebM project authors. All Rights Reserved.
-;
-; Use of this source code is governed by a BSD-style license
-; that can be found in the LICENSE file in the root of the source
-; tree. An additional intellectual property rights grant can be found
-; in the file PATENTS. All contributing project authors may
-; be found in the AUTHORS file in the root of the source tree.
-;
-
- EXPORT |vp9_v_predictor_4x4_neon|
- EXPORT |vp9_v_predictor_8x8_neon|
- EXPORT |vp9_v_predictor_16x16_neon|
- EXPORT |vp9_v_predictor_32x32_neon|
- EXPORT |vp9_h_predictor_4x4_neon|
- EXPORT |vp9_h_predictor_8x8_neon|
- EXPORT |vp9_h_predictor_16x16_neon|
- EXPORT |vp9_h_predictor_32x32_neon|
- EXPORT |vp9_tm_predictor_4x4_neon|
- EXPORT |vp9_tm_predictor_8x8_neon|
- EXPORT |vp9_tm_predictor_16x16_neon|
- EXPORT |vp9_tm_predictor_32x32_neon|
- ARM
- REQUIRE8
- PRESERVE8
-
- AREA ||.text||, CODE, READONLY, ALIGN=2
-
-;void vp9_v_predictor_4x4_neon(uint8_t *dst, ptrdiff_t y_stride,
-; const uint8_t *above,
-; const uint8_t *left)
-; r0 uint8_t *dst
-; r1 ptrdiff_t y_stride
-; r2 const uint8_t *above
-; r3 const uint8_t *left
-
-|vp9_v_predictor_4x4_neon| PROC
- vld1.32 {d0[0]}, [r2]- vst1.32 {d0[0]}, [r0], r1- vst1.32 {d0[0]}, [r0], r1- vst1.32 {d0[0]}, [r0], r1- vst1.32 {d0[0]}, [r0], r1- bx lr
- ENDP ; |vp9_v_predictor_4x4_neon|
-
-;void vp9_v_predictor_8x8_neon(uint8_t *dst, ptrdiff_t y_stride,
-; const uint8_t *above,
-; const uint8_t *left)
-; r0 uint8_t *dst
-; r1 ptrdiff_t y_stride
-; r2 const uint8_t *above
-; r3 const uint8_t *left
-
-|vp9_v_predictor_8x8_neon| PROC
- vld1.8 {d0}, [r2]- vst1.8 {d0}, [r0], r1- vst1.8 {d0}, [r0], r1- vst1.8 {d0}, [r0], r1- vst1.8 {d0}, [r0], r1- vst1.8 {d0}, [r0], r1- vst1.8 {d0}, [r0], r1- vst1.8 {d0}, [r0], r1- vst1.8 {d0}, [r0], r1- bx lr
- ENDP ; |vp9_v_predictor_8x8_neon|
-
-;void vp9_v_predictor_16x16_neon(uint8_t *dst, ptrdiff_t y_stride,
-; const uint8_t *above,
-; const uint8_t *left)
-; r0 uint8_t *dst
-; r1 ptrdiff_t y_stride
-; r2 const uint8_t *above
-; r3 const uint8_t *left
-
-|vp9_v_predictor_16x16_neon| PROC
- vld1.8 {q0}, [r2]- vst1.8 {q0}, [r0], r1- vst1.8 {q0}, [r0], r1- vst1.8 {q0}, [r0], r1- vst1.8 {q0}, [r0], r1- vst1.8 {q0}, [r0], r1- vst1.8 {q0}, [r0], r1- vst1.8 {q0}, [r0], r1- vst1.8 {q0}, [r0], r1- vst1.8 {q0}, [r0], r1- vst1.8 {q0}, [r0], r1- vst1.8 {q0}, [r0], r1- vst1.8 {q0}, [r0], r1- vst1.8 {q0}, [r0], r1- vst1.8 {q0}, [r0], r1- vst1.8 {q0}, [r0], r1- vst1.8 {q0}, [r0], r1- bx lr
- ENDP ; |vp9_v_predictor_16x16_neon|
-
-;void vp9_v_predictor_32x32_neon(uint8_t *dst, ptrdiff_t y_stride,
-; const uint8_t *above,
-; const uint8_t *left)
-; r0 uint8_t *dst
-; r1 ptrdiff_t y_stride
-; r2 const uint8_t *above
-; r3 const uint8_t *left
-
-|vp9_v_predictor_32x32_neon| PROC
- vld1.8 {q0, q1}, [r2]- mov r2, #2
-loop_v
- vst1.8 {q0, q1}, [r0], r1- vst1.8 {q0, q1}, [r0], r1- vst1.8 {q0, q1}, [r0], r1- vst1.8 {q0, q1}, [r0], r1- vst1.8 {q0, q1}, [r0], r1- vst1.8 {q0, q1}, [r0], r1- vst1.8 {q0, q1}, [r0], r1- vst1.8 {q0, q1}, [r0], r1- vst1.8 {q0, q1}, [r0], r1- vst1.8 {q0, q1}, [r0], r1- vst1.8 {q0, q1}, [r0], r1- vst1.8 {q0, q1}, [r0], r1- vst1.8 {q0, q1}, [r0], r1- vst1.8 {q0, q1}, [r0], r1- vst1.8 {q0, q1}, [r0], r1- vst1.8 {q0, q1}, [r0], r1- subs r2, r2, #1
- bgt loop_v
- bx lr
- ENDP ; |vp9_v_predictor_32x32_neon|
-
-;void vp9_h_predictor_4x4_neon(uint8_t *dst, ptrdiff_t y_stride,
-; const uint8_t *above,
-; const uint8_t *left)
-; r0 uint8_t *dst
-; r1 ptrdiff_t y_stride
-; r2 const uint8_t *above
-; r3 const uint8_t *left
-
-|vp9_h_predictor_4x4_neon| PROC
- vld1.32 {d1[0]}, [r3]- vdup.8 d0, d1[0]
- vst1.32 {d0[0]}, [r0], r1- vdup.8 d0, d1[1]
- vst1.32 {d0[0]}, [r0], r1- vdup.8 d0, d1[2]
- vst1.32 {d0[0]}, [r0], r1- vdup.8 d0, d1[3]
- vst1.32 {d0[0]}, [r0], r1- bx lr
- ENDP ; |vp9_h_predictor_4x4_neon|
-
-;void vp9_h_predictor_8x8_neon(uint8_t *dst, ptrdiff_t y_stride,
-; const uint8_t *above,
-; const uint8_t *left)
-; r0 uint8_t *dst
-; r1 ptrdiff_t y_stride
-; r2 const uint8_t *above
-; r3 const uint8_t *left
-
-|vp9_h_predictor_8x8_neon| PROC
- vld1.64 {d1}, [r3]- vdup.8 d0, d1[0]
- vst1.64 {d0}, [r0], r1- vdup.8 d0, d1[1]
- vst1.64 {d0}, [r0], r1- vdup.8 d0, d1[2]
- vst1.64 {d0}, [r0], r1- vdup.8 d0, d1[3]
- vst1.64 {d0}, [r0], r1- vdup.8 d0, d1[4]
- vst1.64 {d0}, [r0], r1- vdup.8 d0, d1[5]
- vst1.64 {d0}, [r0], r1- vdup.8 d0, d1[6]
- vst1.64 {d0}, [r0], r1- vdup.8 d0, d1[7]
- vst1.64 {d0}, [r0], r1- bx lr
- ENDP ; |vp9_h_predictor_8x8_neon|
-
-;void vp9_h_predictor_16x16_neon(uint8_t *dst, ptrdiff_t y_stride,
-; const uint8_t *above,
-; const uint8_t *left)
-; r0 uint8_t *dst
-; r1 ptrdiff_t y_stride
-; r2 const uint8_t *above
-; r3 const uint8_t *left
-
-|vp9_h_predictor_16x16_neon| PROC
- vld1.8 {q1}, [r3]- vdup.8 q0, d2[0]
- vst1.8 {q0}, [r0], r1- vdup.8 q0, d2[1]
- vst1.8 {q0}, [r0], r1- vdup.8 q0, d2[2]
- vst1.8 {q0}, [r0], r1- vdup.8 q0, d2[3]
- vst1.8 {q0}, [r0], r1- vdup.8 q0, d2[4]
- vst1.8 {q0}, [r0], r1- vdup.8 q0, d2[5]
- vst1.8 {q0}, [r0], r1- vdup.8 q0, d2[6]
- vst1.8 {q0}, [r0], r1- vdup.8 q0, d2[7]
- vst1.8 {q0}, [r0], r1- vdup.8 q0, d3[0]
- vst1.8 {q0}, [r0], r1- vdup.8 q0, d3[1]
- vst1.8 {q0}, [r0], r1- vdup.8 q0, d3[2]
- vst1.8 {q0}, [r0], r1- vdup.8 q0, d3[3]
- vst1.8 {q0}, [r0], r1- vdup.8 q0, d3[4]
- vst1.8 {q0}, [r0], r1- vdup.8 q0, d3[5]
- vst1.8 {q0}, [r0], r1- vdup.8 q0, d3[6]
- vst1.8 {q0}, [r0], r1- vdup.8 q0, d3[7]
- vst1.8 {q0}, [r0], r1- bx lr
- ENDP ; |vp9_h_predictor_16x16_neon|
-
-;void vp9_h_predictor_32x32_neon(uint8_t *dst, ptrdiff_t y_stride,
-; const uint8_t *above,
-; const uint8_t *left)
-; r0 uint8_t *dst
-; r1 ptrdiff_t y_stride
-; r2 const uint8_t *above
-; r3 const uint8_t *left
-
-|vp9_h_predictor_32x32_neon| PROC
- sub r1, r1, #16
- mov r2, #2
-loop_h
- vld1.8 {q1}, [r3]!- vdup.8 q0, d2[0]
- vst1.8 {q0}, [r0]!- vst1.8 {q0}, [r0], r1- vdup.8 q0, d2[1]
- vst1.8 {q0}, [r0]!- vst1.8 {q0}, [r0], r1- vdup.8 q0, d2[2]
- vst1.8 {q0}, [r0]!- vst1.8 {q0}, [r0], r1- vdup.8 q0, d2[3]
- vst1.8 {q0}, [r0]!- vst1.8 {q0}, [r0], r1- vdup.8 q0, d2[4]
- vst1.8 {q0}, [r0]!- vst1.8 {q0}, [r0], r1- vdup.8 q0, d2[5]
- vst1.8 {q0}, [r0]!- vst1.8 {q0}, [r0], r1- vdup.8 q0, d2[6]
- vst1.8 {q0}, [r0]!- vst1.8 {q0}, [r0], r1- vdup.8 q0, d2[7]
- vst1.8 {q0}, [r0]!- vst1.8 {q0}, [r0], r1- vdup.8 q0, d3[0]
- vst1.8 {q0}, [r0]!- vst1.8 {q0}, [r0], r1- vdup.8 q0, d3[1]
- vst1.8 {q0}, [r0]!- vst1.8 {q0}, [r0], r1- vdup.8 q0, d3[2]
- vst1.8 {q0}, [r0]!- vst1.8 {q0}, [r0], r1- vdup.8 q0, d3[3]
- vst1.8 {q0}, [r0]!- vst1.8 {q0}, [r0], r1- vdup.8 q0, d3[4]
- vst1.8 {q0}, [r0]!- vst1.8 {q0}, [r0], r1- vdup.8 q0, d3[5]
- vst1.8 {q0}, [r0]!- vst1.8 {q0}, [r0], r1- vdup.8 q0, d3[6]
- vst1.8 {q0}, [r0]!- vst1.8 {q0}, [r0], r1- vdup.8 q0, d3[7]
- vst1.8 {q0}, [r0]!- vst1.8 {q0}, [r0], r1- subs r2, r2, #1
- bgt loop_h
- bx lr
- ENDP ; |vp9_h_predictor_32x32_neon|
-
-;void vp9_tm_predictor_4x4_neon (uint8_t *dst, ptrdiff_t y_stride,
-; const uint8_t *above,
-; const uint8_t *left)
-; r0 uint8_t *dst
-; r1 ptrdiff_t y_stride
-; r2 const uint8_t *above
-; r3 const uint8_t *left
-
-|vp9_tm_predictor_4x4_neon| PROC
- ; Load ytop_left = above[-1];
- sub r12, r2, #1
- vld1.u8 {d0[]}, [r12]-
- ; Load above 4 pixels
- vld1.32 {d2[0]}, [r2]-
- ; Compute above - ytop_left
- vsubl.u8 q3, d2, d0
-
- ; Load left row by row and compute left + (above - ytop_left)
- ; 1st row and 2nd row
- vld1.u8 {d2[]}, [r3]!- vld1.u8 {d4[]}, [r3]!- vmovl.u8 q1, d2
- vmovl.u8 q2, d4
- vadd.s16 q1, q1, q3
- vadd.s16 q2, q2, q3
- vqmovun.s16 d0, q1
- vqmovun.s16 d1, q2
- vst1.32 {d0[0]}, [r0], r1- vst1.32 {d1[0]}, [r0], r1-
- ; 3rd row and 4th row
- vld1.u8 {d2[]}, [r3]!- vld1.u8 {d4[]}, [r3]- vmovl.u8 q1, d2
- vmovl.u8 q2, d4
- vadd.s16 q1, q1, q3
- vadd.s16 q2, q2, q3
- vqmovun.s16 d0, q1
- vqmovun.s16 d1, q2
- vst1.32 {d0[0]}, [r0], r1- vst1.32 {d1[0]}, [r0], r1- bx lr
- ENDP ; |vp9_tm_predictor_4x4_neon|
-
-;void vp9_tm_predictor_8x8_neon (uint8_t *dst, ptrdiff_t y_stride,
-; const uint8_t *above,
-; const uint8_t *left)
-; r0 uint8_t *dst
-; r1 ptrdiff_t y_stride
-; r2 const uint8_t *above
-; r3 const uint8_t *left
-
-|vp9_tm_predictor_8x8_neon| PROC
- ; Load ytop_left = above[-1];
- sub r12, r2, #1
- vld1.8 {d0[]}, [r12]-
- ; preload 8 left
- vld1.8 {d30}, [r3]-
- ; Load above 8 pixels
- vld1.64 {d2}, [r2]-
- vmovl.u8 q10, d30
-
- ; Compute above - ytop_left
- vsubl.u8 q3, d2, d0
-
- ; Load left row by row and compute left + (above - ytop_left)
- ; 1st row and 2nd row
- vdup.16 q0, d20[0]
- vdup.16 q1, d20[1]
- vadd.s16 q0, q3, q0
- vadd.s16 q1, q3, q1
-
- ; 3rd row and 4th row
- vdup.16 q8, d20[2]
- vdup.16 q9, d20[3]
- vadd.s16 q8, q3, q8
- vadd.s16 q9, q3, q9
-
- vqmovun.s16 d0, q0
- vqmovun.s16 d1, q1
- vqmovun.s16 d2, q8
- vqmovun.s16 d3, q9
-
- vst1.64 {d0}, [r0], r1- vst1.64 {d1}, [r0], r1- vst1.64 {d2}, [r0], r1- vst1.64 {d3}, [r0], r1-
- ; 5th row and 6th row
- vdup.16 q0, d21[0]
- vdup.16 q1, d21[1]
- vadd.s16 q0, q3, q0
- vadd.s16 q1, q3, q1
-
- ; 7th row and 8th row
- vdup.16 q8, d21[2]
- vdup.16 q9, d21[3]
- vadd.s16 q8, q3, q8
- vadd.s16 q9, q3, q9
-
- vqmovun.s16 d0, q0
- vqmovun.s16 d1, q1
- vqmovun.s16 d2, q8
- vqmovun.s16 d3, q9
-
- vst1.64 {d0}, [r0], r1- vst1.64 {d1}, [r0], r1- vst1.64 {d2}, [r0], r1- vst1.64 {d3}, [r0], r1-
- bx lr
- ENDP ; |vp9_tm_predictor_8x8_neon|
-
-;void vp9_tm_predictor_16x16_neon (uint8_t *dst, ptrdiff_t y_stride,
-; const uint8_t *above,
-; const uint8_t *left)
-; r0 uint8_t *dst
-; r1 ptrdiff_t y_stride
-; r2 const uint8_t *above
-; r3 const uint8_t *left
-
-|vp9_tm_predictor_16x16_neon| PROC
- ; Load ytop_left = above[-1];
- sub r12, r2, #1
- vld1.8 {d0[]}, [r12]-
- ; Load above 8 pixels
- vld1.8 {q1}, [r2]-
- ; preload 8 left into r12
- vld1.8 {d18}, [r3]!-
- ; Compute above - ytop_left
- vsubl.u8 q2, d2, d0
- vsubl.u8 q3, d3, d0
-
- vmovl.u8 q10, d18
-
- ; Load left row by row and compute left + (above - ytop_left)
- ; Process 8 rows in each single loop and loop 2 times to process 16 rows.
- mov r2, #2
-
-loop_16x16_neon
- ; Process two rows.
- vdup.16 q0, d20[0]
- vdup.16 q8, d20[1]
- vadd.s16 q1, q0, q2
- vadd.s16 q0, q0, q3
- vadd.s16 q11, q8, q2
- vadd.s16 q8, q8, q3
- vqmovun.s16 d2, q1
- vqmovun.s16 d3, q0
- vqmovun.s16 d22, q11
- vqmovun.s16 d23, q8
- vdup.16 q0, d20[2] ; proload next 2 rows data
- vdup.16 q8, d20[3]
- vst1.64 {d2,d3}, [r0], r1- vst1.64 {d22,d23}, [r0], r1-
- ; Process two rows.
- vadd.s16 q1, q0, q2
- vadd.s16 q0, q0, q3
- vadd.s16 q11, q8, q2
- vadd.s16 q8, q8, q3
- vqmovun.s16 d2, q1
- vqmovun.s16 d3, q0
- vqmovun.s16 d22, q11
- vqmovun.s16 d23, q8
- vdup.16 q0, d21[0] ; proload next 2 rows data
- vdup.16 q8, d21[1]
- vst1.64 {d2,d3}, [r0], r1- vst1.64 {d22,d23}, [r0], r1-
- vadd.s16 q1, q0, q2
- vadd.s16 q0, q0, q3
- vadd.s16 q11, q8, q2
- vadd.s16 q8, q8, q3
- vqmovun.s16 d2, q1
- vqmovun.s16 d3, q0
- vqmovun.s16 d22, q11
- vqmovun.s16 d23, q8
- vdup.16 q0, d21[2] ; proload next 2 rows data
- vdup.16 q8, d21[3]
- vst1.64 {d2,d3}, [r0], r1- vst1.64 {d22,d23}, [r0], r1-
-
- vadd.s16 q1, q0, q2
- vadd.s16 q0, q0, q3
- vadd.s16 q11, q8, q2
- vadd.s16 q8, q8, q3
- vqmovun.s16 d2, q1
- vqmovun.s16 d3, q0
- vqmovun.s16 d22, q11
- vqmovun.s16 d23, q8
- vld1.8 {d18}, [r3]! ; preload 8 left into r12- vmovl.u8 q10, d18
- vst1.64 {d2,d3}, [r0], r1- vst1.64 {d22,d23}, [r0], r1-
- subs r2, r2, #1
- bgt loop_16x16_neon
-
- bx lr
- ENDP ; |vp9_tm_predictor_16x16_neon|
-
-;void vp9_tm_predictor_32x32_neon (uint8_t *dst, ptrdiff_t y_stride,
-; const uint8_t *above,
-; const uint8_t *left)
-; r0 uint8_t *dst
-; r1 ptrdiff_t y_stride
-; r2 const uint8_t *above
-; r3 const uint8_t *left
-
-|vp9_tm_predictor_32x32_neon| PROC
- ; Load ytop_left = above[-1];
- sub r12, r2, #1
- vld1.8 {d0[]}, [r12]-
- ; Load above 32 pixels
- vld1.8 {q1}, [r2]!- vld1.8 {q2}, [r2]-
- ; preload 8 left pixels
- vld1.8 {d26}, [r3]!-
- ; Compute above - ytop_left
- vsubl.u8 q8, d2, d0
- vsubl.u8 q9, d3, d0
- vsubl.u8 q10, d4, d0
- vsubl.u8 q11, d5, d0
-
- vmovl.u8 q3, d26
-
- ; Load left row by row and compute left + (above - ytop_left)
- ; Process 8 rows in each single loop and loop 4 times to process 32 rows.
- mov r2, #4
-
-loop_32x32_neon
- ; Process two rows.
- vdup.16 q0, d6[0]
- vdup.16 q2, d6[1]
- vadd.s16 q12, q0, q8
- vadd.s16 q13, q0, q9
- vadd.s16 q14, q0, q10
- vadd.s16 q15, q0, q11
- vqmovun.s16 d0, q12
- vqmovun.s16 d1, q13
- vadd.s16 q12, q2, q8
- vadd.s16 q13, q2, q9
- vqmovun.s16 d2, q14
- vqmovun.s16 d3, q15
- vadd.s16 q14, q2, q10
- vadd.s16 q15, q2, q11
- vst1.64 {d0-d3}, [r0], r1- vqmovun.s16 d24, q12
- vqmovun.s16 d25, q13
- vqmovun.s16 d26, q14
- vqmovun.s16 d27, q15
- vdup.16 q1, d6[2]
- vdup.16 q2, d6[3]
- vst1.64 {d24-d27}, [r0], r1-
- ; Process two rows.
- vadd.s16 q12, q1, q8
- vadd.s16 q13, q1, q9
- vadd.s16 q14, q1, q10
- vadd.s16 q15, q1, q11
- vqmovun.s16 d0, q12
- vqmovun.s16 d1, q13
- vadd.s16 q12, q2, q8
- vadd.s16 q13, q2, q9
- vqmovun.s16 d2, q14
- vqmovun.s16 d3, q15
- vadd.s16 q14, q2, q10
- vadd.s16 q15, q2, q11
- vst1.64 {d0-d3}, [r0], r1- vqmovun.s16 d24, q12
- vqmovun.s16 d25, q13
- vqmovun.s16 d26, q14
- vqmovun.s16 d27, q15
- vdup.16 q0, d7[0]
- vdup.16 q2, d7[1]
- vst1.64 {d24-d27}, [r0], r1-
- ; Process two rows.
- vadd.s16 q12, q0, q8
- vadd.s16 q13, q0, q9
- vadd.s16 q14, q0, q10
- vadd.s16 q15, q0, q11
- vqmovun.s16 d0, q12
- vqmovun.s16 d1, q13
- vadd.s16 q12, q2, q8
- vadd.s16 q13, q2, q9
- vqmovun.s16 d2, q14
- vqmovun.s16 d3, q15
- vadd.s16 q14, q2, q10
- vadd.s16 q15, q2, q11
- vst1.64 {d0-d3}, [r0], r1- vqmovun.s16 d24, q12
- vqmovun.s16 d25, q13
- vqmovun.s16 d26, q14
- vqmovun.s16 d27, q15
- vdup.16 q0, d7[2]
- vdup.16 q2, d7[3]
- vst1.64 {d24-d27}, [r0], r1-
- ; Process two rows.
- vadd.s16 q12, q0, q8
- vadd.s16 q13, q0, q9
- vadd.s16 q14, q0, q10
- vadd.s16 q15, q0, q11
- vqmovun.s16 d0, q12
- vqmovun.s16 d1, q13
- vadd.s16 q12, q2, q8
- vadd.s16 q13, q2, q9
- vqmovun.s16 d2, q14
- vqmovun.s16 d3, q15
- vadd.s16 q14, q2, q10
- vadd.s16 q15, q2, q11
- vst1.64 {d0-d3}, [r0], r1- vqmovun.s16 d24, q12
- vqmovun.s16 d25, q13
- vld1.8 {d0}, [r3]! ; preload 8 left pixels- vqmovun.s16 d26, q14
- vqmovun.s16 d27, q15
- vmovl.u8 q3, d0
- vst1.64 {d24-d27}, [r0], r1-
- subs r2, r2, #1
- bgt loop_32x32_neon
-
- bx lr
- ENDP ; |vp9_tm_predictor_32x32_neon|
-
- END
--- a/vp9/common/mips/dspr2/vp9_intrapred16_dspr2.c
+++ /dev/null
@@ -1,332 +1,0 @@
-/*
- * Copyright (c) 2013 The WebM project authors. All Rights Reserved.
- *
- * Use of this source code is governed by a BSD-style license
- * that can be found in the LICENSE file in the root of the source
- * tree. An additional intellectual property rights grant can be found
- * in the file PATENTS. All contributing project authors may
- * be found in the AUTHORS file in the root of the source tree.
- */
-#include <stdlib.h>
-
-#include "./vp9_rtcd.h"
-#include "vp9/common/vp9_common.h"
-#include "vp9/common/mips/dspr2/vp9_common_dspr2.h"
-
-#if HAVE_DSPR2
-void vp9_h_predictor_16x16_dspr2(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above, const uint8_t *left) {- int32_t tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp8;
- int32_t tmp9, tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, tmp16;
-
- __asm__ __volatile__ (
- "lb %[tmp1], (%[left]) \n\t"
- "lb %[tmp2], 1(%[left]) \n\t"
- "lb %[tmp3], 2(%[left]) \n\t"
- "lb %[tmp4], 3(%[left]) \n\t"
- "lb %[tmp5], 4(%[left]) \n\t"
- "lb %[tmp6], 5(%[left]) \n\t"
- "lb %[tmp7], 6(%[left]) \n\t"
- "lb %[tmp8], 7(%[left]) \n\t"
- "lb %[tmp9], 8(%[left]) \n\t"
- "lb %[tmp10], 9(%[left]) \n\t"
- "lb %[tmp11], 10(%[left]) \n\t"
- "lb %[tmp12], 11(%[left]) \n\t"
- "lb %[tmp13], 12(%[left]) \n\t"
- "lb %[tmp14], 13(%[left]) \n\t"
- "lb %[tmp15], 14(%[left]) \n\t"
- "lb %[tmp16], 15(%[left]) \n\t"
-
- "replv.qb %[tmp1], %[tmp1] \n\t"
- "replv.qb %[tmp2], %[tmp2] \n\t"
- "replv.qb %[tmp3], %[tmp3] \n\t"
- "replv.qb %[tmp4], %[tmp4] \n\t"
- "replv.qb %[tmp5], %[tmp5] \n\t"
- "replv.qb %[tmp6], %[tmp6] \n\t"
- "replv.qb %[tmp7], %[tmp7] \n\t"
- "replv.qb %[tmp8], %[tmp8] \n\t"
- "replv.qb %[tmp9], %[tmp9] \n\t"
- "replv.qb %[tmp10], %[tmp10] \n\t"
- "replv.qb %[tmp11], %[tmp11] \n\t"
- "replv.qb %[tmp12], %[tmp12] \n\t"
- "replv.qb %[tmp13], %[tmp13] \n\t"
- "replv.qb %[tmp14], %[tmp14] \n\t"
- "replv.qb %[tmp15], %[tmp15] \n\t"
- "replv.qb %[tmp16], %[tmp16] \n\t"
-
- "sw %[tmp1], (%[dst]) \n\t"
- "sw %[tmp1], 4(%[dst]) \n\t"
- "sw %[tmp1], 8(%[dst]) \n\t"
- "sw %[tmp1], 12(%[dst]) \n\t"
-
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[tmp2], (%[dst]) \n\t"
- "sw %[tmp2], 4(%[dst]) \n\t"
- "sw %[tmp2], 8(%[dst]) \n\t"
- "sw %[tmp2], 12(%[dst]) \n\t"
-
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[tmp3], (%[dst]) \n\t"
- "sw %[tmp3], 4(%[dst]) \n\t"
- "sw %[tmp3], 8(%[dst]) \n\t"
- "sw %[tmp3], 12(%[dst]) \n\t"
-
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[tmp4], (%[dst]) \n\t"
- "sw %[tmp4], 4(%[dst]) \n\t"
- "sw %[tmp4], 8(%[dst]) \n\t"
- "sw %[tmp4], 12(%[dst]) \n\t"
-
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[tmp5], (%[dst]) \n\t"
- "sw %[tmp5], 4(%[dst]) \n\t"
- "sw %[tmp5], 8(%[dst]) \n\t"
- "sw %[tmp5], 12(%[dst]) \n\t"
-
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[tmp6], (%[dst]) \n\t"
- "sw %[tmp6], 4(%[dst]) \n\t"
- "sw %[tmp6], 8(%[dst]) \n\t"
- "sw %[tmp6], 12(%[dst]) \n\t"
-
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[tmp7], (%[dst]) \n\t"
- "sw %[tmp7], 4(%[dst]) \n\t"
- "sw %[tmp7], 8(%[dst]) \n\t"
- "sw %[tmp7], 12(%[dst]) \n\t"
-
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[tmp8], (%[dst]) \n\t"
- "sw %[tmp8], 4(%[dst]) \n\t"
- "sw %[tmp8], 8(%[dst]) \n\t"
- "sw %[tmp8], 12(%[dst]) \n\t"
-
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[tmp9], (%[dst]) \n\t"
- "sw %[tmp9], 4(%[dst]) \n\t"
- "sw %[tmp9], 8(%[dst]) \n\t"
- "sw %[tmp9], 12(%[dst]) \n\t"
-
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[tmp10], (%[dst]) \n\t"
- "sw %[tmp10], 4(%[dst]) \n\t"
- "sw %[tmp10], 8(%[dst]) \n\t"
- "sw %[tmp10], 12(%[dst]) \n\t"
-
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[tmp11], (%[dst]) \n\t"
- "sw %[tmp11], 4(%[dst]) \n\t"
- "sw %[tmp11], 8(%[dst]) \n\t"
- "sw %[tmp11], 12(%[dst]) \n\t"
-
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[tmp12], (%[dst]) \n\t"
- "sw %[tmp12], 4(%[dst]) \n\t"
- "sw %[tmp12], 8(%[dst]) \n\t"
- "sw %[tmp12], 12(%[dst]) \n\t"
-
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[tmp13], (%[dst]) \n\t"
- "sw %[tmp13], 4(%[dst]) \n\t"
- "sw %[tmp13], 8(%[dst]) \n\t"
- "sw %[tmp13], 12(%[dst]) \n\t"
-
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[tmp14], (%[dst]) \n\t"
- "sw %[tmp14], 4(%[dst]) \n\t"
- "sw %[tmp14], 8(%[dst]) \n\t"
- "sw %[tmp14], 12(%[dst]) \n\t"
-
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[tmp15], (%[dst]) \n\t"
- "sw %[tmp15], 4(%[dst]) \n\t"
- "sw %[tmp15], 8(%[dst]) \n\t"
- "sw %[tmp15], 12(%[dst]) \n\t"
-
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[tmp16], (%[dst]) \n\t"
- "sw %[tmp16], 4(%[dst]) \n\t"
- "sw %[tmp16], 8(%[dst]) \n\t"
- "sw %[tmp16], 12(%[dst]) \n\t"
-
- : [tmp1] "=&r" (tmp1), [tmp2] "=&r" (tmp2),
- [tmp3] "=&r" (tmp3), [tmp4] "=&r" (tmp4),
- [tmp5] "=&r" (tmp5), [tmp7] "=&r" (tmp7),
- [tmp6] "=&r" (tmp6), [tmp8] "=&r" (tmp8),
- [tmp9] "=&r" (tmp9), [tmp10] "=&r" (tmp10),
- [tmp11] "=&r" (tmp11), [tmp12] "=&r" (tmp12),
- [tmp13] "=&r" (tmp13), [tmp14] "=&r" (tmp14),
- [tmp15] "=&r" (tmp15), [tmp16] "=&r" (tmp16)
- : [left] "r" (left), [dst] "r" (dst), [stride] "r" (stride)
- );
-}
-
-void vp9_dc_predictor_16x16_dspr2(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above, const uint8_t *left) {- int32_t expected_dc;
- int32_t average;
- int32_t tmp, above1, above_l1, above_r1, left1, left_r1, left_l1;
- int32_t above2, left2;
-
- __asm__ __volatile__ (
- "lw %[above1], (%[above]) \n\t"
- "lw %[above2], 4(%[above]) \n\t"
- "lw %[left1], (%[left]) \n\t"
- "lw %[left2], 4(%[left]) \n\t"
-
- "preceu.ph.qbl %[above_l1], %[above1] \n\t"
- "preceu.ph.qbr %[above_r1], %[above1] \n\t"
- "preceu.ph.qbl %[left_l1], %[left1] \n\t"
- "preceu.ph.qbr %[left_r1], %[left1] \n\t"
-
- "addu.ph %[average], %[above_r1], %[above_l1] \n\t"
- "addu.ph %[average], %[average], %[left_l1] \n\t"
- "addu.ph %[average], %[average], %[left_r1] \n\t"
-
- "preceu.ph.qbl %[above_l1], %[above2] \n\t"
- "preceu.ph.qbr %[above_r1], %[above2] \n\t"
- "preceu.ph.qbl %[left_l1], %[left2] \n\t"
- "preceu.ph.qbr %[left_r1], %[left2] \n\t"
-
- "addu.ph %[average], %[average], %[above_l1] \n\t"
- "addu.ph %[average], %[average], %[above_r1] \n\t"
- "addu.ph %[average], %[average], %[left_l1] \n\t"
- "addu.ph %[average], %[average], %[left_r1] \n\t"
-
- "lw %[above1], 8(%[above]) \n\t"
- "lw %[above2], 12(%[above]) \n\t"
- "lw %[left1], 8(%[left]) \n\t"
- "lw %[left2], 12(%[left]) \n\t"
-
- "preceu.ph.qbl %[above_l1], %[above1] \n\t"
- "preceu.ph.qbr %[above_r1], %[above1] \n\t"
- "preceu.ph.qbl %[left_l1], %[left1] \n\t"
- "preceu.ph.qbr %[left_r1], %[left1] \n\t"
-
- "addu.ph %[average], %[average], %[above_l1] \n\t"
- "addu.ph %[average], %[average], %[above_r1] \n\t"
- "addu.ph %[average], %[average], %[left_l1] \n\t"
- "addu.ph %[average], %[average], %[left_r1] \n\t"
-
- "preceu.ph.qbl %[above_l1], %[above2] \n\t"
- "preceu.ph.qbr %[above_r1], %[above2] \n\t"
- "preceu.ph.qbl %[left_l1], %[left2] \n\t"
- "preceu.ph.qbr %[left_r1], %[left2] \n\t"
-
- "addu.ph %[average], %[average], %[above_l1] \n\t"
- "addu.ph %[average], %[average], %[above_r1] \n\t"
- "addu.ph %[average], %[average], %[left_l1] \n\t"
- "addu.ph %[average], %[average], %[left_r1] \n\t"
-
- "addiu %[average], %[average], 16 \n\t"
- "srl %[tmp], %[average], 16 \n\t"
- "addu.ph %[average], %[tmp], %[average] \n\t"
- "srl %[expected_dc], %[average], 5 \n\t"
- "replv.qb %[expected_dc], %[expected_dc] \n\t"
-
- "sw %[expected_dc], (%[dst]) \n\t"
- "sw %[expected_dc], 4(%[dst]) \n\t"
- "sw %[expected_dc], 8(%[dst]) \n\t"
- "sw %[expected_dc], 12(%[dst]) \n\t"
-
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[expected_dc], (%[dst]) \n\t"
- "sw %[expected_dc], 4(%[dst]) \n\t"
- "sw %[expected_dc], 8(%[dst]) \n\t"
- "sw %[expected_dc], 12(%[dst]) \n\t"
-
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[expected_dc], (%[dst]) \n\t"
- "sw %[expected_dc], 4(%[dst]) \n\t"
- "sw %[expected_dc], 8(%[dst]) \n\t"
- "sw %[expected_dc], 12(%[dst]) \n\t"
-
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[expected_dc], (%[dst]) \n\t"
- "sw %[expected_dc], 4(%[dst]) \n\t"
- "sw %[expected_dc], 8(%[dst]) \n\t"
- "sw %[expected_dc], 12(%[dst]) \n\t"
-
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[expected_dc], (%[dst]) \n\t"
- "sw %[expected_dc], 4(%[dst]) \n\t"
- "sw %[expected_dc], 8(%[dst]) \n\t"
- "sw %[expected_dc], 12(%[dst]) \n\t"
-
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[expected_dc], (%[dst]) \n\t"
- "sw %[expected_dc], 4(%[dst]) \n\t"
- "sw %[expected_dc], 8(%[dst]) \n\t"
- "sw %[expected_dc], 12(%[dst]) \n\t"
-
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[expected_dc], (%[dst]) \n\t"
- "sw %[expected_dc], 4(%[dst]) \n\t"
- "sw %[expected_dc], 8(%[dst]) \n\t"
- "sw %[expected_dc], 12(%[dst]) \n\t"
-
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[expected_dc], (%[dst]) \n\t"
- "sw %[expected_dc], 4(%[dst]) \n\t"
- "sw %[expected_dc], 8(%[dst]) \n\t"
- "sw %[expected_dc], 12(%[dst]) \n\t"
-
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[expected_dc], (%[dst]) \n\t"
- "sw %[expected_dc], 4(%[dst]) \n\t"
- "sw %[expected_dc], 8(%[dst]) \n\t"
- "sw %[expected_dc], 12(%[dst]) \n\t"
-
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[expected_dc], (%[dst]) \n\t"
- "sw %[expected_dc], 4(%[dst]) \n\t"
- "sw %[expected_dc], 8(%[dst]) \n\t"
- "sw %[expected_dc], 12(%[dst]) \n\t"
-
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[expected_dc], (%[dst]) \n\t"
- "sw %[expected_dc], 4(%[dst]) \n\t"
- "sw %[expected_dc], 8(%[dst]) \n\t"
- "sw %[expected_dc], 12(%[dst]) \n\t"
-
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[expected_dc], (%[dst]) \n\t"
- "sw %[expected_dc], 4(%[dst]) \n\t"
- "sw %[expected_dc], 8(%[dst]) \n\t"
- "sw %[expected_dc], 12(%[dst]) \n\t"
-
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[expected_dc], (%[dst]) \n\t"
- "sw %[expected_dc], 4(%[dst]) \n\t"
- "sw %[expected_dc], 8(%[dst]) \n\t"
- "sw %[expected_dc], 12(%[dst]) \n\t"
-
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[expected_dc], (%[dst]) \n\t"
- "sw %[expected_dc], 4(%[dst]) \n\t"
- "sw %[expected_dc], 8(%[dst]) \n\t"
- "sw %[expected_dc], 12(%[dst]) \n\t"
-
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[expected_dc], (%[dst]) \n\t"
- "sw %[expected_dc], 4(%[dst]) \n\t"
- "sw %[expected_dc], 8(%[dst]) \n\t"
- "sw %[expected_dc], 12(%[dst]) \n\t"
-
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[expected_dc], (%[dst]) \n\t"
- "sw %[expected_dc], 4(%[dst]) \n\t"
- "sw %[expected_dc], 8(%[dst]) \n\t"
- "sw %[expected_dc], 12(%[dst]) \n\t"
-
- : [left1] "=&r" (left1), [above1] "=&r" (above1),
- [left_l1] "=&r" (left_l1), [above_l1] "=&r" (above_l1),
- [left_r1] "=&r" (left_r1), [above_r1] "=&r" (above_r1),
- [above2] "=&r" (above2), [left2] "=&r" (left2),
- [average] "=&r" (average), [tmp] "=&r" (tmp),
- [expected_dc] "=&r" (expected_dc)
- : [above] "r" (above), [left] "r" (left),
- [dst] "r" (dst), [stride] "r" (stride)
- );
-}
-#endif // #if HAVE_DSPR2
--- a/vp9/common/mips/dspr2/vp9_intrapred4_dspr2.c
+++ /dev/null
@@ -1,232 +1,0 @@
-/*
- * Copyright (c) 2013 The WebM project authors. All Rights Reserved.
- *
- * Use of this source code is governed by a BSD-style license
- * that can be found in the LICENSE file in the root of the source
- * tree. An additional intellectual property rights grant can be found
- * in the file PATENTS. All contributing project authors may
- * be found in the AUTHORS file in the root of the source tree.
- */
-#include <stdlib.h>
-
-#include "./vp9_rtcd.h"
-#include "vp9/common/vp9_common.h"
-#include "vp9/common/mips/dspr2/vp9_common_dspr2.h"
-
-#if HAVE_DSPR2
-void vp9_h_predictor_4x4_dspr2(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above, const uint8_t *left) {- int32_t tmp1, tmp2, tmp3, tmp4;
-
- __asm__ __volatile__ (
- "lb %[tmp1], (%[left]) \n\t"
- "lb %[tmp2], 1(%[left]) \n\t"
- "lb %[tmp3], 2(%[left]) \n\t"
- "lb %[tmp4], 3(%[left]) \n\t"
- "replv.qb %[tmp1], %[tmp1] \n\t"
- "replv.qb %[tmp2], %[tmp2] \n\t"
- "replv.qb %[tmp3], %[tmp3] \n\t"
- "replv.qb %[tmp4], %[tmp4] \n\t"
- "sw %[tmp1], (%[dst]) \n\t"
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[tmp2], (%[dst]) \n\t"
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[tmp3], (%[dst]) \n\t"
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[tmp4], (%[dst]) \n\t"
-
- : [tmp1] "=&r" (tmp1), [tmp2] "=&r" (tmp2),
- [tmp3] "=&r" (tmp3), [tmp4] "=&r" (tmp4)
- : [left] "r" (left), [dst] "r" (dst), [stride] "r" (stride)
- );
-}
-
-void vp9_dc_predictor_4x4_dspr2(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above, const uint8_t *left) {- int32_t expected_dc;
- int32_t average;
- int32_t tmp, above_c, above_l, above_r, left_c, left_r, left_l;
-
- __asm__ __volatile__ (
- "lw %[above_c], (%[above]) \n\t"
- "lw %[left_c], (%[left]) \n\t"
-
- "preceu.ph.qbl %[above_l], %[above_c] \n\t"
- "preceu.ph.qbr %[above_r], %[above_c] \n\t"
- "preceu.ph.qbl %[left_l], %[left_c] \n\t"
- "preceu.ph.qbr %[left_r], %[left_c] \n\t"
-
- "addu.ph %[average], %[above_r], %[above_l] \n\t"
- "addu.ph %[average], %[average], %[left_l] \n\t"
- "addu.ph %[average], %[average], %[left_r] \n\t"
- "addiu %[average], %[average], 4 \n\t"
- "srl %[tmp], %[average], 16 \n\t"
- "addu.ph %[average], %[tmp], %[average] \n\t"
- "srl %[expected_dc], %[average], 3 \n\t"
- "replv.qb %[expected_dc], %[expected_dc] \n\t"
-
- "sw %[expected_dc], (%[dst]) \n\t"
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[expected_dc], (%[dst]) \n\t"
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[expected_dc], (%[dst]) \n\t"
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[expected_dc], (%[dst]) \n\t"
-
- : [above_c] "=&r" (above_c), [above_l] "=&r" (above_l),
- [above_r] "=&r" (above_r), [left_c] "=&r" (left_c),
- [left_l] "=&r" (left_l), [left_r] "=&r" (left_r),
- [average] "=&r" (average), [tmp] "=&r" (tmp),
- [expected_dc] "=&r" (expected_dc)
- : [above] "r" (above), [left] "r" (left),
- [dst] "r" (dst), [stride] "r" (stride)
- );
-}
-
-void vp9_tm_predictor_4x4_dspr2(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above, const uint8_t *left) {- int32_t abovel, abover;
- int32_t left0, left1, left2, left3;
- int32_t res0, res1;
- int32_t resl;
- int32_t resr;
- int32_t top_left;
- uint8_t *cm = vp9_ff_cropTbl;
-
- __asm__ __volatile__ (
- "ulw %[resl], (%[above]) \n\t"
-
- "lbu %[left0], (%[left]) \n\t"
- "lbu %[left1], 1(%[left]) \n\t"
- "lbu %[left2], 2(%[left]) \n\t"
- "lbu %[left3], 3(%[left]) \n\t"
-
- "lbu %[top_left], -1(%[above]) \n\t"
-
- "preceu.ph.qbl %[abovel], %[resl] \n\t"
- "preceu.ph.qbr %[abover], %[resl] \n\t"
-
- "replv.ph %[left0], %[left0] \n\t"
- "replv.ph %[left1], %[left1] \n\t"
- "replv.ph %[left2], %[left2] \n\t"
- "replv.ph %[left3], %[left3] \n\t"
-
- "replv.ph %[top_left], %[top_left] \n\t"
-
- "addu.ph %[resl], %[abovel], %[left0] \n\t"
- "subu.ph %[resl], %[resl], %[top_left] \n\t"
-
- "addu.ph %[resr], %[abover], %[left0] \n\t"
- "subu.ph %[resr], %[resr], %[top_left] \n\t"
-
- "sll %[res0], %[resr], 16 \n\t"
- "sra %[res0], %[res0], 16 \n\t"
- "lbux %[res0], %[res0](%[cm]) \n\t"
-
- "sra %[res1], %[resr], 16 \n\t"
- "lbux %[res1], %[res1](%[cm]) \n\t"
- "sb %[res0], (%[dst]) \n\t"
-
- "sll %[res0], %[resl], 16 \n\t"
- "sra %[res0], %[res0], 16 \n\t"
- "lbux %[res0], %[res0](%[cm]) \n\t"
- "sb %[res1], 1(%[dst]) \n\t"
-
- "sra %[res1], %[resl], 16 \n\t"
- "lbux %[res1], %[res1](%[cm]) \n\t"
-
- "addu.ph %[resl], %[abovel], %[left1] \n\t"
- "subu.ph %[resl], %[resl], %[top_left] \n\t"
-
- "addu.ph %[resr], %[abover], %[left1] \n\t"
- "subu.ph %[resr], %[resr], %[top_left] \n\t"
-
- "sb %[res0], 2(%[dst]) \n\t"
- "sb %[res1], 3(%[dst]) \n\t"
-
- "add %[dst], %[dst], %[stride] \n\t"
-
- "sll %[res0], %[resr], 16 \n\t"
- "sra %[res0], %[res0], 16 \n\t"
- "lbux %[res0], %[res0](%[cm]) \n\t"
-
- "sra %[res1], %[resr], 16 \n\t"
- "lbux %[res1], %[res1](%[cm]) \n\t"
- "sb %[res0], (%[dst]) \n\t"
-
- "sll %[res0], %[resl], 16 \n\t"
- "sra %[res0], %[res0], 16 \n\t"
- "lbux %[res0], %[res0](%[cm]) \n\t"
-
- "sb %[res1], 1(%[dst]) \n\t"
- "sra %[res1], %[resl], 16 \n\t"
- "lbux %[res1], %[res1](%[cm]) \n\t"
-
- "addu.ph %[resl], %[abovel], %[left2] \n\t"
- "subu.ph %[resl], %[resl], %[top_left] \n\t"
-
- "addu.ph %[resr], %[abover], %[left2] \n\t"
- "subu.ph %[resr], %[resr], %[top_left] \n\t"
-
- "sb %[res0], 2(%[dst]) \n\t"
- "sb %[res1], 3(%[dst]) \n\t"
-
- "add %[dst], %[dst], %[stride] \n\t"
-
- "sll %[res0], %[resr], 16 \n\t"
- "sra %[res0], %[res0], 16 \n\t"
- "lbux %[res0], %[res0](%[cm]) \n\t"
-
-
- "sra %[res1], %[resr], 16 \n\t"
- "lbux %[res1], %[res1](%[cm]) \n\t"
- "sb %[res0], (%[dst]) \n\t"
-
- "sll %[res0], %[resl], 16 \n\t"
- "sra %[res0], %[res0], 16 \n\t"
- "lbux %[res0], %[res0](%[cm]) \n\t"
-
-
- "sb %[res1], 1(%[dst]) \n\t"
- "sra %[res1], %[resl], 16 \n\t"
- "lbux %[res1], %[res1](%[cm]) \n\t"
-
- "addu.ph %[resl], %[abovel], %[left3] \n\t"
- "subu.ph %[resl], %[resl], %[top_left] \n\t"
-
- "addu.ph %[resr], %[abover], %[left3] \n\t"
- "subu.ph %[resr], %[resr], %[top_left] \n\t"
-
- "sb %[res0], 2(%[dst]) \n\t"
- "sb %[res1], 3(%[dst]) \n\t"
-
- "add %[dst], %[dst], %[stride] \n\t"
-
- "sll %[res0], %[resr], 16 \n\t"
- "sra %[res0], %[res0], 16 \n\t"
- "lbux %[res0], %[res0](%[cm]) \n\t"
-
- "sra %[res1], %[resr], 16 \n\t"
- "lbux %[res1], %[res1](%[cm]) \n\t"
- "sb %[res0], (%[dst]) \n\t"
-
- "sll %[res0], %[resl], 16 \n\t"
- "sra %[res0], %[res0], 16 \n\t"
- "lbux %[res0], %[res0](%[cm]) \n\t"
- "sb %[res1], 1(%[dst]) \n\t"
-
- "sra %[res1], %[resl], 16 \n\t"
- "lbux %[res1], %[res1](%[cm]) \n\t"
-
- "sb %[res0], 2(%[dst]) \n\t"
- "sb %[res1], 3(%[dst]) \n\t"
-
- : [abovel] "=&r" (abovel), [abover] "=&r" (abover),
- [left0] "=&r" (left0), [left1] "=&r" (left1), [left2] "=&r" (left2),
- [res0] "=&r" (res0), [res1] "=&r" (res1), [left3] "=&r" (left3),
- [resl] "=&r" (resl), [resr] "=&r" (resr), [top_left] "=&r" (top_left)
- : [above] "r" (above), [left] "r" (left),
- [dst] "r" (dst), [stride] "r" (stride), [cm] "r" (cm)
- );
-}
-#endif // #if HAVE_DSPR2
--- a/vp9/common/mips/dspr2/vp9_intrapred8_dspr2.c
+++ /dev/null
@@ -1,610 +1,0 @@
-/*
- * Copyright (c) 2013 The WebM project authors. All Rights Reserved.
- *
- * Use of this source code is governed by a BSD-style license
- * that can be found in the LICENSE file in the root of the source
- * tree. An additional intellectual property rights grant can be found
- * in the file PATENTS. All contributing project authors may
- * be found in the AUTHORS file in the root of the source tree.
- */
-#include <stdlib.h>
-
-#include "./vp9_rtcd.h"
-#include "vp9/common/vp9_common.h"
-#include "vp9/common/mips/dspr2/vp9_common_dspr2.h"
-
-#if HAVE_DSPR2
-void vp9_h_predictor_8x8_dspr2(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above, const uint8_t *left) {- int32_t tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp8;
-
- __asm__ __volatile__ (
- "lb %[tmp1], (%[left]) \n\t"
- "lb %[tmp2], 1(%[left]) \n\t"
- "lb %[tmp3], 2(%[left]) \n\t"
- "lb %[tmp4], 3(%[left]) \n\t"
- "lb %[tmp5], 4(%[left]) \n\t"
- "lb %[tmp6], 5(%[left]) \n\t"
- "lb %[tmp7], 6(%[left]) \n\t"
- "lb %[tmp8], 7(%[left]) \n\t"
-
- "replv.qb %[tmp1], %[tmp1] \n\t"
- "replv.qb %[tmp2], %[tmp2] \n\t"
- "replv.qb %[tmp3], %[tmp3] \n\t"
- "replv.qb %[tmp4], %[tmp4] \n\t"
- "replv.qb %[tmp5], %[tmp5] \n\t"
- "replv.qb %[tmp6], %[tmp6] \n\t"
- "replv.qb %[tmp7], %[tmp7] \n\t"
- "replv.qb %[tmp8], %[tmp8] \n\t"
-
- "sw %[tmp1], (%[dst]) \n\t"
- "sw %[tmp1], 4(%[dst]) \n\t"
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[tmp2], (%[dst]) \n\t"
- "sw %[tmp2], 4(%[dst]) \n\t"
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[tmp3], (%[dst]) \n\t"
- "sw %[tmp3], 4(%[dst]) \n\t"
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[tmp4], (%[dst]) \n\t"
- "sw %[tmp4], 4(%[dst]) \n\t"
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[tmp5], (%[dst]) \n\t"
- "sw %[tmp5], 4(%[dst]) \n\t"
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[tmp6], (%[dst]) \n\t"
- "sw %[tmp6], 4(%[dst]) \n\t"
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[tmp7], (%[dst]) \n\t"
- "sw %[tmp7], 4(%[dst]) \n\t"
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[tmp8], (%[dst]) \n\t"
- "sw %[tmp8], 4(%[dst]) \n\t"
-
- : [tmp1] "=&r" (tmp1), [tmp2] "=&r" (tmp2),
- [tmp3] "=&r" (tmp3), [tmp4] "=&r" (tmp4),
- [tmp5] "=&r" (tmp5), [tmp7] "=&r" (tmp7),
- [tmp6] "=&r" (tmp6), [tmp8] "=&r" (tmp8)
- : [left] "r" (left), [dst] "r" (dst),
- [stride] "r" (stride)
- );
-}
-
-void vp9_dc_predictor_8x8_dspr2(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above, const uint8_t *left) {- int32_t expected_dc;
- int32_t average;
- int32_t tmp, above1, above_l1, above_r1, left1, left_r1, left_l1;
- int32_t above2, above_l2, above_r2, left2, left_r2, left_l2;
-
- __asm__ __volatile__ (
- "lw %[above1], (%[above]) \n\t"
- "lw %[above2], 4(%[above]) \n\t"
- "lw %[left1], (%[left]) \n\t"
- "lw %[left2], 4(%[left]) \n\t"
-
- "preceu.ph.qbl %[above_l1], %[above1] \n\t"
- "preceu.ph.qbr %[above_r1], %[above1] \n\t"
- "preceu.ph.qbl %[left_l1], %[left1] \n\t"
- "preceu.ph.qbr %[left_r1], %[left1] \n\t"
-
- "preceu.ph.qbl %[above_l2], %[above2] \n\t"
- "preceu.ph.qbr %[above_r2], %[above2] \n\t"
- "preceu.ph.qbl %[left_l2], %[left2] \n\t"
- "preceu.ph.qbr %[left_r2], %[left2] \n\t"
-
- "addu.ph %[average], %[above_r1], %[above_l1] \n\t"
- "addu.ph %[average], %[average], %[left_l1] \n\t"
- "addu.ph %[average], %[average], %[left_r1] \n\t"
-
- "addu.ph %[average], %[average], %[above_l2] \n\t"
- "addu.ph %[average], %[average], %[above_r2] \n\t"
- "addu.ph %[average], %[average], %[left_l2] \n\t"
- "addu.ph %[average], %[average], %[left_r2] \n\t"
-
- "addiu %[average], %[average], 8 \n\t"
-
- "srl %[tmp], %[average], 16 \n\t"
- "addu.ph %[average], %[tmp], %[average] \n\t"
- "srl %[expected_dc], %[average], 4 \n\t"
- "replv.qb %[expected_dc], %[expected_dc] \n\t"
-
- "sw %[expected_dc], (%[dst]) \n\t"
- "sw %[expected_dc], 4(%[dst]) \n\t"
-
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[expected_dc], (%[dst]) \n\t"
- "sw %[expected_dc], 4(%[dst]) \n\t"
-
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[expected_dc], (%[dst]) \n\t"
- "sw %[expected_dc], 4(%[dst]) \n\t"
-
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[expected_dc], (%[dst]) \n\t"
- "sw %[expected_dc], 4(%[dst]) \n\t"
-
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[expected_dc], (%[dst]) \n\t"
- "sw %[expected_dc], 4(%[dst]) \n\t"
-
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[expected_dc], (%[dst]) \n\t"
- "sw %[expected_dc], 4(%[dst]) \n\t"
-
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[expected_dc], (%[dst]) \n\t"
- "sw %[expected_dc], 4(%[dst]) \n\t"
-
- "add %[dst], %[dst], %[stride] \n\t"
- "sw %[expected_dc], (%[dst]) \n\t"
- "sw %[expected_dc], 4(%[dst]) \n\t"
-
- : [above1] "=&r" (above1), [above_l1] "=&r" (above_l1),
- [above_r1] "=&r" (above_r1), [left1] "=&r" (left1),
- [left_l1] "=&r" (left_l1), [left_r1] "=&r" (left_r1),
- [above2] "=&r" (above2), [above_l2] "=&r" (above_l2),
- [above_r2] "=&r" (above_r2), [left2] "=&r" (left2),
- [left_l2] "=&r" (left_l2), [left_r2] "=&r" (left_r2),
- [average] "=&r" (average), [tmp] "=&r" (tmp),
- [expected_dc] "=&r" (expected_dc)
- : [above] "r" (above), [left] "r" (left), [dst] "r" (dst),
- [stride] "r" (stride)
- );
-}
-
-void vp9_tm_predictor_8x8_dspr2(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above, const uint8_t *left) {- int32_t abovel, abover;
- int32_t abovel_1, abover_1;
- int32_t left0;
- int32_t res0, res1, res2, res3;
- int32_t reshw;
- int32_t top_left;
- uint8_t *cm = vp9_ff_cropTbl;
-
- __asm__ __volatile__ (
- "ulw %[reshw], (%[above]) \n\t"
- "ulw %[top_left], 4(%[above]) \n\t"
-
- "lbu %[left0], (%[left]) \n\t"
-
- "preceu.ph.qbl %[abovel], %[reshw] \n\t"
- "preceu.ph.qbr %[abover], %[reshw] \n\t"
- "preceu.ph.qbl %[abovel_1], %[top_left] \n\t"
- "preceu.ph.qbr %[abover_1], %[top_left] \n\t"
-
- "lbu %[top_left], -1(%[above]) \n\t"
- "replv.ph %[left0], %[left0] \n\t"
-
- "replv.ph %[top_left], %[top_left] \n\t"
-
- "addu.ph %[reshw], %[abovel], %[left0] \n\t"
- "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
-
- "sll %[res2], %[reshw], 16 \n\t"
- "sra %[res2], %[res2], 16 \n\t"
- "sra %[res3], %[reshw], 16 \n\t"
-
- "addu.ph %[reshw], %[abover], %[left0] \n\t"
- "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
-
- "sll %[res0], %[reshw], 16 \n\t"
- "sra %[res0], %[res0], 16 \n\t"
- "sra %[res1], %[reshw], 16 \n\t"
-
- "lbux %[res0], %[res0](%[cm]) \n\t"
- "lbux %[res1], %[res1](%[cm]) \n\t"
- "lbux %[res2], %[res2](%[cm]) \n\t"
- "lbux %[res3], %[res3](%[cm]) \n\t"
-
- "sb %[res0], (%[dst]) \n\t"
- "sb %[res1], 1(%[dst]) \n\t"
- "sb %[res2], 2(%[dst]) \n\t"
- "sb %[res3], 3(%[dst]) \n\t"
-
- "addu.ph %[reshw], %[abovel_1], %[left0] \n\t"
- "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
-
- "sll %[res2], %[reshw], 16 \n\t"
- "sra %[res2], %[res2], 16 \n\t"
- "sra %[res3], %[reshw], 16 \n\t"
-
- "addu.ph %[reshw], %[abover_1], %[left0] \n\t"
- "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
-
- "sll %[res0], %[reshw], 16 \n\t"
- "sra %[res0], %[res0], 16 \n\t"
- "sra %[res1], %[reshw], 16 \n\t"
-
- "lbu %[left0], 1(%[left]) \n\t"
-
- "lbux %[res0], %[res0](%[cm]) \n\t"
- "lbux %[res1], %[res1](%[cm]) \n\t"
- "lbux %[res2], %[res2](%[cm]) \n\t"
- "lbux %[res3], %[res3](%[cm]) \n\t"
-
- "sb %[res0], 4(%[dst]) \n\t"
- "sb %[res1], 5(%[dst]) \n\t"
- "sb %[res2], 6(%[dst]) \n\t"
- "sb %[res3], 7(%[dst]) \n\t"
-
- "replv.ph %[left0], %[left0] \n\t"
- "add %[dst], %[dst], %[stride] \n\t"
-
- "addu.ph %[reshw], %[abovel], %[left0] \n\t"
- "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
-
- "sll %[res2], %[reshw], 16 \n\t"
- "sra %[res2], %[res2], 16 \n\t"
- "sra %[res3], %[reshw], 16 \n\t"
-
- "addu.ph %[reshw], %[abover], %[left0] \n\t"
- "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
-
- "sll %[res0], %[reshw], 16 \n\t"
- "sra %[res0], %[res0], 16 \n\t"
- "sra %[res1], %[reshw], 16 \n\t"
-
- "lbux %[res0], %[res0](%[cm]) \n\t"
- "lbux %[res1], %[res1](%[cm]) \n\t"
- "lbux %[res2], %[res2](%[cm]) \n\t"
- "lbux %[res3], %[res3](%[cm]) \n\t"
-
- "sb %[res0], (%[dst]) \n\t"
- "sb %[res1], 1(%[dst]) \n\t"
- "sb %[res2], 2(%[dst]) \n\t"
- "sb %[res3], 3(%[dst]) \n\t"
-
- "addu.ph %[reshw], %[abovel_1], %[left0] \n\t"
- "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
-
- "sll %[res2], %[reshw], 16 \n\t"
- "sra %[res2], %[res2], 16 \n\t"
- "sra %[res3], %[reshw], 16 \n\t"
-
- "addu.ph %[reshw], %[abover_1], %[left0] \n\t"
- "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
-
- "sll %[res0], %[reshw], 16 \n\t"
- "sra %[res0], %[res0], 16 \n\t"
- "sra %[res1], %[reshw], 16 \n\t"
-
- "lbu %[left0], 2(%[left]) \n\t"
-
- "lbux %[res0], %[res0](%[cm]) \n\t"
- "lbux %[res1], %[res1](%[cm]) \n\t"
- "lbux %[res2], %[res2](%[cm]) \n\t"
- "lbux %[res3], %[res3](%[cm]) \n\t"
-
- "sb %[res0], 4(%[dst]) \n\t"
- "sb %[res1], 5(%[dst]) \n\t"
- "sb %[res2], 6(%[dst]) \n\t"
- "sb %[res3], 7(%[dst]) \n\t"
-
- "replv.ph %[left0], %[left0] \n\t"
- "add %[dst], %[dst], %[stride] \n\t"
-
- "addu.ph %[reshw], %[abovel], %[left0] \n\t"
- "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
-
- "sll %[res2], %[reshw], 16 \n\t"
- "sra %[res2], %[res2], 16 \n\t"
- "sra %[res3], %[reshw], 16 \n\t"
-
- "addu.ph %[reshw], %[abover], %[left0] \n\t"
- "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
-
- "sll %[res0], %[reshw], 16 \n\t"
- "sra %[res0], %[res0], 16 \n\t"
- "sra %[res1], %[reshw], 16 \n\t"
-
- "lbux %[res0], %[res0](%[cm]) \n\t"
- "lbux %[res1], %[res1](%[cm]) \n\t"
- "lbux %[res2], %[res2](%[cm]) \n\t"
- "lbux %[res3], %[res3](%[cm]) \n\t"
-
- "sb %[res0], (%[dst]) \n\t"
- "sb %[res1], 1(%[dst]) \n\t"
- "sb %[res2], 2(%[dst]) \n\t"
- "sb %[res3], 3(%[dst]) \n\t"
-
- "addu.ph %[reshw], %[abovel_1], %[left0] \n\t"
- "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
-
- "sll %[res2], %[reshw], 16 \n\t"
- "sra %[res2], %[res2], 16 \n\t"
- "sra %[res3], %[reshw], 16 \n\t"
-
- "addu.ph %[reshw], %[abover_1], %[left0] \n\t"
- "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
-
- "sll %[res0], %[reshw], 16 \n\t"
- "sra %[res0], %[res0], 16 \n\t"
- "sra %[res1], %[reshw], 16 \n\t"
-
- "lbu %[left0], 3(%[left]) \n\t"
-
- "lbux %[res0], %[res0](%[cm]) \n\t"
- "lbux %[res1], %[res1](%[cm]) \n\t"
- "lbux %[res2], %[res2](%[cm]) \n\t"
- "lbux %[res3], %[res3](%[cm]) \n\t"
-
- "sb %[res0], 4(%[dst]) \n\t"
- "sb %[res1], 5(%[dst]) \n\t"
- "sb %[res2], 6(%[dst]) \n\t"
- "sb %[res3], 7(%[dst]) \n\t"
-
- "replv.ph %[left0], %[left0] \n\t"
- "add %[dst], %[dst], %[stride] \n\t"
-
- "addu.ph %[reshw], %[abovel], %[left0] \n\t"
- "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
-
- "sll %[res2], %[reshw], 16 \n\t"
- "sra %[res2], %[res2], 16 \n\t"
- "sra %[res3], %[reshw], 16 \n\t"
-
- "addu.ph %[reshw], %[abover], %[left0] \n\t"
- "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
-
- "sll %[res0], %[reshw], 16 \n\t"
- "sra %[res0], %[res0], 16 \n\t"
- "sra %[res1], %[reshw], 16 \n\t"
-
- "lbux %[res0], %[res0](%[cm]) \n\t"
- "lbux %[res1], %[res1](%[cm]) \n\t"
- "lbux %[res2], %[res2](%[cm]) \n\t"
- "lbux %[res3], %[res3](%[cm]) \n\t"
-
- "sb %[res0], (%[dst]) \n\t"
- "sb %[res1], 1(%[dst]) \n\t"
- "sb %[res2], 2(%[dst]) \n\t"
- "sb %[res3], 3(%[dst]) \n\t"
-
- "addu.ph %[reshw], %[abovel_1], %[left0] \n\t"
- "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
-
- "sll %[res2], %[reshw], 16 \n\t"
- "sra %[res2], %[res2], 16 \n\t"
- "sra %[res3], %[reshw], 16 \n\t"
-
- "addu.ph %[reshw], %[abover_1], %[left0] \n\t"
- "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
-
- "sll %[res0], %[reshw], 16 \n\t"
- "sra %[res0], %[res0], 16 \n\t"
- "sra %[res1], %[reshw], 16 \n\t"
-
- "lbu %[left0], 4(%[left]) \n\t"
-
- "lbux %[res0], %[res0](%[cm]) \n\t"
- "lbux %[res1], %[res1](%[cm]) \n\t"
- "lbux %[res2], %[res2](%[cm]) \n\t"
- "lbux %[res3], %[res3](%[cm]) \n\t"
-
- "sb %[res0], 4(%[dst]) \n\t"
- "sb %[res1], 5(%[dst]) \n\t"
- "sb %[res2], 6(%[dst]) \n\t"
- "sb %[res3], 7(%[dst]) \n\t"
-
- "replv.ph %[left0], %[left0] \n\t"
- "add %[dst], %[dst], %[stride] \n\t"
-
- "addu.ph %[reshw], %[abovel], %[left0] \n\t"
- "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
-
- "sll %[res2], %[reshw], 16 \n\t"
- "sra %[res2], %[res2], 16 \n\t"
- "sra %[res3], %[reshw], 16 \n\t"
-
- "addu.ph %[reshw], %[abover], %[left0] \n\t"
- "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
-
- "sll %[res0], %[reshw], 16 \n\t"
- "sra %[res0], %[res0], 16 \n\t"
- "sra %[res1], %[reshw], 16 \n\t"
-
- "lbux %[res0], %[res0](%[cm]) \n\t"
- "lbux %[res1], %[res1](%[cm]) \n\t"
- "lbux %[res2], %[res2](%[cm]) \n\t"
- "lbux %[res3], %[res3](%[cm]) \n\t"
-
- "sb %[res0], (%[dst]) \n\t"
- "sb %[res1], 1(%[dst]) \n\t"
- "sb %[res2], 2(%[dst]) \n\t"
- "sb %[res3], 3(%[dst]) \n\t"
-
- "addu.ph %[reshw], %[abovel_1], %[left0] \n\t"
- "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
-
- "sll %[res2], %[reshw], 16 \n\t"
- "sra %[res2], %[res2], 16 \n\t"
- "sra %[res3], %[reshw], 16 \n\t"
-
- "addu.ph %[reshw], %[abover_1], %[left0] \n\t"
- "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
-
- "sll %[res0], %[reshw], 16 \n\t"
- "sra %[res0], %[res0], 16 \n\t"
- "sra %[res1], %[reshw], 16 \n\t"
-
- "lbu %[left0], 5(%[left]) \n\t"
-
- "lbux %[res0], %[res0](%[cm]) \n\t"
- "lbux %[res1], %[res1](%[cm]) \n\t"
- "lbux %[res2], %[res2](%[cm]) \n\t"
- "lbux %[res3], %[res3](%[cm]) \n\t"
-
- "sb %[res0], 4(%[dst]) \n\t"
- "sb %[res1], 5(%[dst]) \n\t"
- "sb %[res2], 6(%[dst]) \n\t"
- "sb %[res3], 7(%[dst]) \n\t"
-
- "replv.ph %[left0], %[left0] \n\t"
- "add %[dst], %[dst], %[stride] \n\t"
-
- "addu.ph %[reshw], %[abovel], %[left0] \n\t"
- "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
-
- "sll %[res2], %[reshw], 16 \n\t"
- "sra %[res2], %[res2], 16 \n\t"
- "sra %[res3], %[reshw], 16 \n\t"
-
- "addu.ph %[reshw], %[abover], %[left0] \n\t"
- "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
-
- "sll %[res0], %[reshw], 16 \n\t"
- "sra %[res0], %[res0], 16 \n\t"
- "sra %[res1], %[reshw], 16 \n\t"
-
- "lbux %[res0], %[res0](%[cm]) \n\t"
- "lbux %[res1], %[res1](%[cm]) \n\t"
- "lbux %[res2], %[res2](%[cm]) \n\t"
- "lbux %[res3], %[res3](%[cm]) \n\t"
-
- "sb %[res0], (%[dst]) \n\t"
- "sb %[res1], 1(%[dst]) \n\t"
- "sb %[res2], 2(%[dst]) \n\t"
- "sb %[res3], 3(%[dst]) \n\t"
-
- "addu.ph %[reshw], %[abovel_1], %[left0] \n\t"
- "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
-
- "sll %[res2], %[reshw], 16 \n\t"
- "sra %[res2], %[res2], 16 \n\t"
- "sra %[res3], %[reshw], 16 \n\t"
-
- "addu.ph %[reshw], %[abover_1], %[left0] \n\t"
- "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
-
- "sll %[res0], %[reshw], 16 \n\t"
- "sra %[res0], %[res0], 16 \n\t"
- "sra %[res1], %[reshw], 16 \n\t"
-
- "lbu %[left0], 6(%[left]) \n\t"
-
- "lbux %[res0], %[res0](%[cm]) \n\t"
- "lbux %[res1], %[res1](%[cm]) \n\t"
- "lbux %[res2], %[res2](%[cm]) \n\t"
- "lbux %[res3], %[res3](%[cm]) \n\t"
-
- "sb %[res0], 4(%[dst]) \n\t"
- "sb %[res1], 5(%[dst]) \n\t"
- "sb %[res2], 6(%[dst]) \n\t"
- "sb %[res3], 7(%[dst]) \n\t"
-
- "replv.ph %[left0], %[left0] \n\t"
- "add %[dst], %[dst], %[stride] \n\t"
-
- "addu.ph %[reshw], %[abovel], %[left0] \n\t"
- "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
-
- "sll %[res2], %[reshw], 16 \n\t"
- "sra %[res2], %[res2], 16 \n\t"
- "sra %[res3], %[reshw], 16 \n\t"
-
- "addu.ph %[reshw], %[abover], %[left0] \n\t"
- "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
-
- "sll %[res0], %[reshw], 16 \n\t"
- "sra %[res0], %[res0], 16 \n\t"
- "sra %[res1], %[reshw], 16 \n\t"
-
- "lbux %[res0], %[res0](%[cm]) \n\t"
- "lbux %[res1], %[res1](%[cm]) \n\t"
- "lbux %[res2], %[res2](%[cm]) \n\t"
- "lbux %[res3], %[res3](%[cm]) \n\t"
-
- "sb %[res0], (%[dst]) \n\t"
- "sb %[res1], 1(%[dst]) \n\t"
- "sb %[res2], 2(%[dst]) \n\t"
- "sb %[res3], 3(%[dst]) \n\t"
-
- "addu.ph %[reshw], %[abovel_1], %[left0] \n\t"
- "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
-
- "sll %[res2], %[reshw], 16 \n\t"
- "sra %[res2], %[res2], 16 \n\t"
- "sra %[res3], %[reshw], 16 \n\t"
-
- "addu.ph %[reshw], %[abover_1], %[left0] \n\t"
- "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
-
- "sll %[res0], %[reshw], 16 \n\t"
- "sra %[res0], %[res0], 16 \n\t"
- "sra %[res1], %[reshw], 16 \n\t"
-
- "lbu %[left0], 7(%[left]) \n\t"
-
- "lbux %[res0], %[res0](%[cm]) \n\t"
- "lbux %[res1], %[res1](%[cm]) \n\t"
- "lbux %[res2], %[res2](%[cm]) \n\t"
- "lbux %[res3], %[res3](%[cm]) \n\t"
-
- "sb %[res0], 4(%[dst]) \n\t"
- "sb %[res1], 5(%[dst]) \n\t"
- "sb %[res2], 6(%[dst]) \n\t"
- "sb %[res3], 7(%[dst]) \n\t"
-
- "replv.ph %[left0], %[left0] \n\t"
- "add %[dst], %[dst], %[stride] \n\t"
-
- "addu.ph %[reshw], %[abovel], %[left0] \n\t"
- "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
-
- "sll %[res2], %[reshw], 16 \n\t"
- "sra %[res2], %[res2], 16 \n\t"
- "sra %[res3], %[reshw], 16 \n\t"
-
- "addu.ph %[reshw], %[abover], %[left0] \n\t"
- "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
-
- "sll %[res0], %[reshw], 16 \n\t"
- "sra %[res0], %[res0], 16 \n\t"
- "sra %[res1], %[reshw], 16 \n\t"
-
- "lbux %[res0], %[res0](%[cm]) \n\t"
- "lbux %[res1], %[res1](%[cm]) \n\t"
- "lbux %[res2], %[res2](%[cm]) \n\t"
- "lbux %[res3], %[res3](%[cm]) \n\t"
-
- "sb %[res0], (%[dst]) \n\t"
- "sb %[res1], 1(%[dst]) \n\t"
- "sb %[res2], 2(%[dst]) \n\t"
- "sb %[res3], 3(%[dst]) \n\t"
-
- "addu.ph %[reshw], %[abovel_1], %[left0] \n\t"
- "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
-
- "sll %[res2], %[reshw], 16 \n\t"
- "sra %[res2], %[res2], 16 \n\t"
- "sra %[res3], %[reshw], 16 \n\t"
-
- "addu.ph %[reshw], %[abover_1], %[left0] \n\t"
- "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
-
- "sll %[res0], %[reshw], 16 \n\t"
- "sra %[res0], %[res0], 16 \n\t"
- "sra %[res1], %[reshw], 16 \n\t"
-
- "lbux %[res0], %[res0](%[cm]) \n\t"
- "lbux %[res1], %[res1](%[cm]) \n\t"
- "lbux %[res2], %[res2](%[cm]) \n\t"
- "lbux %[res3], %[res3](%[cm]) \n\t"
-
- "sb %[res0], 4(%[dst]) \n\t"
- "sb %[res1], 5(%[dst]) \n\t"
- "sb %[res2], 6(%[dst]) \n\t"
- "sb %[res3], 7(%[dst]) \n\t"
-
- : [abovel] "=&r" (abovel), [abover] "=&r" (abover),
- [abovel_1] "=&r" (abovel_1), [abover_1] "=&r" (abover_1),
- [left0] "=&r" (left0), [res2] "=&r" (res2), [res3] "=&r" (res3),
- [res0] "=&r" (res0), [res1] "=&r" (res1),
- [reshw] "=&r" (reshw), [top_left] "=&r" (top_left)
- : [above] "r" (above), [left] "r" (left),
- [dst] "r" (dst), [stride] "r" (stride), [cm] "r" (cm)
- );
-}
-#endif // #if HAVE_DSPR2
--- a/vp9/common/mips/msa/vp9_intra_predict_msa.c
+++ /dev/null
@@ -1,737 +1,0 @@
-/*
- * Copyright (c) 2015 The WebM project authors. All Rights Reserved.
- *
- * Use of this source code is governed by a BSD-style license
- * that can be found in the LICENSE file in the root of the source
- * tree. An additional intellectual property rights grant can be found
- * in the file PATENTS. All contributing project authors may
- * be found in the AUTHORS file in the root of the source tree.
- */
-
-#include "./vp9_rtcd.h"
-#include "vpx_dsp/mips/macros_msa.h"
-
-#define IPRED_SUBS_UH2_UH(in0, in1, out0, out1) { \- out0 = __msa_subs_u_h(out0, in0); \
- out1 = __msa_subs_u_h(out1, in1); \
-}
-
-static void intra_predict_vert_4x4_msa(const uint8_t *src, uint8_t *dst,
- int32_t dst_stride) {- uint32_t src_data;
-
- src_data = LW(src);
-
- SW4(src_data, src_data, src_data, src_data, dst, dst_stride);
-}
-
-static void intra_predict_vert_8x8_msa(const uint8_t *src, uint8_t *dst,
- int32_t dst_stride) {- uint32_t row;
- uint32_t src_data1, src_data2;
-
- src_data1 = LW(src);
- src_data2 = LW(src + 4);
-
- for (row = 8; row--;) {- SW(src_data1, dst);
- SW(src_data2, (dst + 4));
- dst += dst_stride;
- }
-}
-
-static void intra_predict_vert_16x16_msa(const uint8_t *src, uint8_t *dst,
- int32_t dst_stride) {- uint32_t row;
- v16u8 src0;
-
- src0 = LD_UB(src);
-
- for (row = 16; row--;) {- ST_UB(src0, dst);
- dst += dst_stride;
- }
-}
-
-static void intra_predict_vert_32x32_msa(const uint8_t *src, uint8_t *dst,
- int32_t dst_stride) {- uint32_t row;
- v16u8 src1, src2;
-
- src1 = LD_UB(src);
- src2 = LD_UB(src + 16);
-
- for (row = 32; row--;) {- ST_UB2(src1, src2, dst, 16);
- dst += dst_stride;
- }
-}
-
-static void intra_predict_horiz_4x4_msa(const uint8_t *src, uint8_t *dst,
- int32_t dst_stride) {- uint32_t out0, out1, out2, out3;
-
- out0 = src[0] * 0x01010101;
- out1 = src[1] * 0x01010101;
- out2 = src[2] * 0x01010101;
- out3 = src[3] * 0x01010101;
-
- SW4(out0, out1, out2, out3, dst, dst_stride);
-}
-
-static void intra_predict_horiz_8x8_msa(const uint8_t *src, uint8_t *dst,
- int32_t dst_stride) {- uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
-
- out0 = src[0] * 0x0101010101010101ull;
- out1 = src[1] * 0x0101010101010101ull;
- out2 = src[2] * 0x0101010101010101ull;
- out3 = src[3] * 0x0101010101010101ull;
- out4 = src[4] * 0x0101010101010101ull;
- out5 = src[5] * 0x0101010101010101ull;
- out6 = src[6] * 0x0101010101010101ull;
- out7 = src[7] * 0x0101010101010101ull;
-
- SD4(out0, out1, out2, out3, dst, dst_stride);
- dst += (4 * dst_stride);
- SD4(out4, out5, out6, out7, dst, dst_stride);
-}
-
-static void intra_predict_horiz_16x16_msa(const uint8_t *src, uint8_t *dst,
- int32_t dst_stride) {- uint32_t row;
- uint8_t inp0, inp1, inp2, inp3;
- v16u8 src0, src1, src2, src3;
-
- for (row = 4; row--;) {- inp0 = src[0];
- inp1 = src[1];
- inp2 = src[2];
- inp3 = src[3];
- src += 4;
-
- src0 = (v16u8)__msa_fill_b(inp0);
- src1 = (v16u8)__msa_fill_b(inp1);
- src2 = (v16u8)__msa_fill_b(inp2);
- src3 = (v16u8)__msa_fill_b(inp3);
-
- ST_UB4(src0, src1, src2, src3, dst, dst_stride);
- dst += (4 * dst_stride);
- }
-}
-
-static void intra_predict_horiz_32x32_msa(const uint8_t *src, uint8_t *dst,
- int32_t dst_stride) {- uint32_t row;
- uint8_t inp0, inp1, inp2, inp3;
- v16u8 src0, src1, src2, src3;
-
- for (row = 8; row--;) {- inp0 = src[0];
- inp1 = src[1];
- inp2 = src[2];
- inp3 = src[3];
- src += 4;
-
- src0 = (v16u8)__msa_fill_b(inp0);
- src1 = (v16u8)__msa_fill_b(inp1);
- src2 = (v16u8)__msa_fill_b(inp2);
- src3 = (v16u8)__msa_fill_b(inp3);
-
- ST_UB2(src0, src0, dst, 16);
- dst += dst_stride;
- ST_UB2(src1, src1, dst, 16);
- dst += dst_stride;
- ST_UB2(src2, src2, dst, 16);
- dst += dst_stride;
- ST_UB2(src3, src3, dst, 16);
- dst += dst_stride;
- }
-}
-
-static void intra_predict_dc_4x4_msa(const uint8_t *src_top,
- const uint8_t *src_left,
- uint8_t *dst, int32_t dst_stride) {- uint32_t val0, val1;
- v16i8 store, src = { 0 };- v8u16 sum_h;
- v4u32 sum_w;
- v2u64 sum_d;
-
- val0 = LW(src_top);
- val1 = LW(src_left);
- INSERT_W2_SB(val0, val1, src);
- sum_h = __msa_hadd_u_h((v16u8)src, (v16u8)src);
- sum_w = __msa_hadd_u_w(sum_h, sum_h);
- sum_d = __msa_hadd_u_d(sum_w, sum_w);
- sum_w = (v4u32)__msa_srari_w((v4i32)sum_d, 3);
- store = __msa_splati_b((v16i8)sum_w, 0);
- val0 = __msa_copy_u_w((v4i32)store, 0);
-
- SW4(val0, val0, val0, val0, dst, dst_stride);
-}
-
-static void intra_predict_dc_tl_4x4_msa(const uint8_t *src, uint8_t *dst,
- int32_t dst_stride) {- uint32_t val0;
- v16i8 store, data = { 0 };- v8u16 sum_h;
- v4u32 sum_w;
-
- val0 = LW(src);
- data = (v16i8)__msa_insert_w((v4i32)data, 0, val0);
- sum_h = __msa_hadd_u_h((v16u8)data, (v16u8)data);
- sum_w = __msa_hadd_u_w(sum_h, sum_h);
- sum_w = (v4u32)__msa_srari_w((v4i32)sum_w, 2);
- store = __msa_splati_b((v16i8)sum_w, 0);
- val0 = __msa_copy_u_w((v4i32)store, 0);
-
- SW4(val0, val0, val0, val0, dst, dst_stride);
-}
-
-static void intra_predict_128dc_4x4_msa(uint8_t *dst, int32_t dst_stride) {- uint32_t out;
- const v16i8 store = __msa_ldi_b(128);
-
- out = __msa_copy_u_w((v4i32)store, 0);
-
- SW4(out, out, out, out, dst, dst_stride);
-}
-
-static void intra_predict_dc_8x8_msa(const uint8_t *src_top,
- const uint8_t *src_left,
- uint8_t *dst, int32_t dst_stride) {- uint64_t val0, val1;
- v16i8 store;
- v16u8 src = { 0 };- v8u16 sum_h;
- v4u32 sum_w;
- v2u64 sum_d;
-
- val0 = LD(src_top);
- val1 = LD(src_left);
- INSERT_D2_UB(val0, val1, src);
- sum_h = __msa_hadd_u_h(src, src);
- sum_w = __msa_hadd_u_w(sum_h, sum_h);
- sum_d = __msa_hadd_u_d(sum_w, sum_w);
- sum_w = (v4u32)__msa_pckev_w((v4i32)sum_d, (v4i32)sum_d);
- sum_d = __msa_hadd_u_d(sum_w, sum_w);
- sum_w = (v4u32)__msa_srari_w((v4i32)sum_d, 4);
- store = __msa_splati_b((v16i8)sum_w, 0);
- val0 = __msa_copy_u_d((v2i64)store, 0);
-
- SD4(val0, val0, val0, val0, dst, dst_stride);
- dst += (4 * dst_stride);
- SD4(val0, val0, val0, val0, dst, dst_stride);
-}
-
-static void intra_predict_dc_tl_8x8_msa(const uint8_t *src, uint8_t *dst,
- int32_t dst_stride) {- uint64_t val0;
- v16i8 store;
- v16u8 data = { 0 };- v8u16 sum_h;
- v4u32 sum_w;
- v2u64 sum_d;
-
- val0 = LD(src);
- data = (v16u8)__msa_insert_d((v2i64)data, 0, val0);
- sum_h = __msa_hadd_u_h(data, data);
- sum_w = __msa_hadd_u_w(sum_h, sum_h);
- sum_d = __msa_hadd_u_d(sum_w, sum_w);
- sum_w = (v4u32)__msa_srari_w((v4i32)sum_d, 3);
- store = __msa_splati_b((v16i8)sum_w, 0);
- val0 = __msa_copy_u_d((v2i64)store, 0);
-
- SD4(val0, val0, val0, val0, dst, dst_stride);
- dst += (4 * dst_stride);
- SD4(val0, val0, val0, val0, dst, dst_stride);
-}
-
-static void intra_predict_128dc_8x8_msa(uint8_t *dst, int32_t dst_stride) {- uint64_t out;
- const v16i8 store = __msa_ldi_b(128);
-
- out = __msa_copy_u_d((v2i64)store, 0);
-
- SD4(out, out, out, out, dst, dst_stride);
- dst += (4 * dst_stride);
- SD4(out, out, out, out, dst, dst_stride);
-}
-
-static void intra_predict_dc_16x16_msa(const uint8_t *src_top,
- const uint8_t *src_left,
- uint8_t *dst, int32_t dst_stride) {- v16u8 top, left, out;
- v8u16 sum_h, sum_top, sum_left;
- v4u32 sum_w;
- v2u64 sum_d;
-
- top = LD_UB(src_top);
- left = LD_UB(src_left);
- HADD_UB2_UH(top, left, sum_top, sum_left);
- sum_h = sum_top + sum_left;
- sum_w = __msa_hadd_u_w(sum_h, sum_h);
- sum_d = __msa_hadd_u_d(sum_w, sum_w);
- sum_w = (v4u32)__msa_pckev_w((v4i32)sum_d, (v4i32)sum_d);
- sum_d = __msa_hadd_u_d(sum_w, sum_w);
- sum_w = (v4u32)__msa_srari_w((v4i32)sum_d, 5);
- out = (v16u8)__msa_splati_b((v16i8)sum_w, 0);
-
- ST_UB8(out, out, out, out, out, out, out, out, dst, dst_stride);
- dst += (8 * dst_stride);
- ST_UB8(out, out, out, out, out, out, out, out, dst, dst_stride);
-}
-
-static void intra_predict_dc_tl_16x16_msa(const uint8_t *src, uint8_t *dst,
- int32_t dst_stride) {- v16u8 data, out;
- v8u16 sum_h;
- v4u32 sum_w;
- v2u64 sum_d;
-
- data = LD_UB(src);
- sum_h = __msa_hadd_u_h(data, data);
- sum_w = __msa_hadd_u_w(sum_h, sum_h);
- sum_d = __msa_hadd_u_d(sum_w, sum_w);
- sum_w = (v4u32)__msa_pckev_w((v4i32)sum_d, (v4i32)sum_d);
- sum_d = __msa_hadd_u_d(sum_w, sum_w);
- sum_w = (v4u32)__msa_srari_w((v4i32)sum_d, 4);
- out = (v16u8)__msa_splati_b((v16i8)sum_w, 0);
-
- ST_UB8(out, out, out, out, out, out, out, out, dst, dst_stride);
- dst += (8 * dst_stride);
- ST_UB8(out, out, out, out, out, out, out, out, dst, dst_stride);
-}
-
-static void intra_predict_128dc_16x16_msa(uint8_t *dst, int32_t dst_stride) {- const v16u8 out = (v16u8)__msa_ldi_b(128);
-
- ST_UB8(out, out, out, out, out, out, out, out, dst, dst_stride);
- dst += (8 * dst_stride);
- ST_UB8(out, out, out, out, out, out, out, out, dst, dst_stride);
-}
-
-static void intra_predict_dc_32x32_msa(const uint8_t *src_top,
- const uint8_t *src_left,
- uint8_t *dst, int32_t dst_stride) {- uint32_t row;
- v16u8 top0, top1, left0, left1, out;
- v8u16 sum_h, sum_top0, sum_top1, sum_left0, sum_left1;
- v4u32 sum_w;
- v2u64 sum_d;
-
- LD_UB2(src_top, 16, top0, top1);
- LD_UB2(src_left, 16, left0, left1);
- HADD_UB2_UH(top0, top1, sum_top0, sum_top1);
- HADD_UB2_UH(left0, left1, sum_left0, sum_left1);
- sum_h = sum_top0 + sum_top1;
- sum_h += sum_left0 + sum_left1;
- sum_w = __msa_hadd_u_w(sum_h, sum_h);
- sum_d = __msa_hadd_u_d(sum_w, sum_w);
- sum_w = (v4u32)__msa_pckev_w((v4i32)sum_d, (v4i32)sum_d);
- sum_d = __msa_hadd_u_d(sum_w, sum_w);
- sum_w = (v4u32)__msa_srari_w((v4i32)sum_d, 6);
- out = (v16u8)__msa_splati_b((v16i8)sum_w, 0);
-
- for (row = 16; row--;) {- ST_UB2(out, out, dst, 16);
- dst += dst_stride;
- ST_UB2(out, out, dst, 16);
- dst += dst_stride;
- }
-}
-
-static void intra_predict_dc_tl_32x32_msa(const uint8_t *src, uint8_t *dst,
- int32_t dst_stride) {- uint32_t row;
- v16u8 data0, data1, out;
- v8u16 sum_h, sum_data0, sum_data1;
- v4u32 sum_w;
- v2u64 sum_d;
-
- LD_UB2(src, 16, data0, data1);
- HADD_UB2_UH(data0, data1, sum_data0, sum_data1);
- sum_h = sum_data0 + sum_data1;
- sum_w = __msa_hadd_u_w(sum_h, sum_h);
- sum_d = __msa_hadd_u_d(sum_w, sum_w);
- sum_w = (v4u32)__msa_pckev_w((v4i32)sum_d, (v4i32)sum_d);
- sum_d = __msa_hadd_u_d(sum_w, sum_w);
- sum_w = (v4u32)__msa_srari_w((v4i32)sum_d, 5);
- out = (v16u8)__msa_splati_b((v16i8)sum_w, 0);
-
- for (row = 16; row--;) {- ST_UB2(out, out, dst, 16);
- dst += dst_stride;
- ST_UB2(out, out, dst, 16);
- dst += dst_stride;
- }
-}
-
-static void intra_predict_128dc_32x32_msa(uint8_t *dst, int32_t dst_stride) {- uint32_t row;
- const v16u8 out = (v16u8)__msa_ldi_b(128);
-
- for (row = 16; row--;) {- ST_UB2(out, out, dst, 16);
- dst += dst_stride;
- ST_UB2(out, out, dst, 16);
- dst += dst_stride;
- }
-}
-
-static void intra_predict_tm_4x4_msa(const uint8_t *src_top_ptr,
- const uint8_t *src_left,
- uint8_t *dst, int32_t dst_stride) {- uint32_t val;
- uint8_t top_left = src_top_ptr[-1];
- v16i8 src_left0, src_left1, src_left2, src_left3, tmp0, tmp1, src_top = { 0 };- v16u8 src0, src1, src2, src3;
- v8u16 src_top_left, vec0, vec1, vec2, vec3;
-
- src_top_left = (v8u16)__msa_fill_h(top_left);
- val = LW(src_top_ptr);
- src_top = (v16i8)__msa_insert_w((v4i32)src_top, 0, val);
-
- src_left0 = __msa_fill_b(src_left[0]);
- src_left1 = __msa_fill_b(src_left[1]);
- src_left2 = __msa_fill_b(src_left[2]);
- src_left3 = __msa_fill_b(src_left[3]);
-
- ILVR_B4_UB(src_left0, src_top, src_left1, src_top, src_left2, src_top,
- src_left3, src_top, src0, src1, src2, src3);
- HADD_UB4_UH(src0, src1, src2, src3, vec0, vec1, vec2, vec3);
- IPRED_SUBS_UH2_UH(src_top_left, src_top_left, vec0, vec1);
- IPRED_SUBS_UH2_UH(src_top_left, src_top_left, vec2, vec3);
- SAT_UH4_UH(vec0, vec1, vec2, vec3, 7);
- PCKEV_B2_SB(vec1, vec0, vec3, vec2, tmp0, tmp1);
- ST4x4_UB(tmp0, tmp1, 0, 2, 0, 2, dst, dst_stride);
-}
-
-static void intra_predict_tm_8x8_msa(const uint8_t *src_top_ptr,
- const uint8_t *src_left,
- uint8_t *dst, int32_t dst_stride) {- uint64_t val;
- uint8_t top_left = src_top_ptr[-1];
- uint32_t loop_cnt;
- v16i8 src_left0, src_left1, src_left2, src_left3, tmp0, tmp1, src_top = { 0 };- v8u16 src_top_left, vec0, vec1, vec2, vec3;
- v16u8 src0, src1, src2, src3;
-
- val = LD(src_top_ptr);
- src_top = (v16i8)__msa_insert_d((v2i64)src_top, 0, val);
- src_top_left = (v8u16)__msa_fill_h(top_left);
-
- for (loop_cnt = 2; loop_cnt--;) {- src_left0 = __msa_fill_b(src_left[0]);
- src_left1 = __msa_fill_b(src_left[1]);
- src_left2 = __msa_fill_b(src_left[2]);
- src_left3 = __msa_fill_b(src_left[3]);
- src_left += 4;
-
- ILVR_B4_UB(src_left0, src_top, src_left1, src_top, src_left2, src_top,
- src_left3, src_top, src0, src1, src2, src3);
- HADD_UB4_UH(src0, src1, src2, src3, vec0, vec1, vec2, vec3);
- IPRED_SUBS_UH2_UH(src_top_left, src_top_left, vec0, vec1);
- IPRED_SUBS_UH2_UH(src_top_left, src_top_left, vec2, vec3);
- SAT_UH4_UH(vec0, vec1, vec2, vec3, 7);
- PCKEV_B2_SB(vec1, vec0, vec3, vec2, tmp0, tmp1);
- ST8x4_UB(tmp0, tmp1, dst, dst_stride);
- dst += (4 * dst_stride);
- }
-}
-
-static void intra_predict_tm_16x16_msa(const uint8_t *src_top_ptr,
- const uint8_t *src_left,
- uint8_t *dst, int32_t dst_stride) {- uint8_t top_left = src_top_ptr[-1];
- uint32_t loop_cnt;
- v16i8 src_top, src_left0, src_left1, src_left2, src_left3;
- v8u16 src_top_left, res_r, res_l;
-
- src_top = LD_SB(src_top_ptr);
- src_top_left = (v8u16)__msa_fill_h(top_left);
-
- for (loop_cnt = 4; loop_cnt--;) {- src_left0 = __msa_fill_b(src_left[0]);
- src_left1 = __msa_fill_b(src_left[1]);
- src_left2 = __msa_fill_b(src_left[2]);
- src_left3 = __msa_fill_b(src_left[3]);
- src_left += 4;
-
- ILVRL_B2_UH(src_left0, src_top, res_r, res_l);
- HADD_UB2_UH(res_r, res_l, res_r, res_l);
- IPRED_SUBS_UH2_UH(src_top_left, src_top_left, res_r, res_l);
-
- SAT_UH2_UH(res_r, res_l, 7);
- PCKEV_ST_SB(res_r, res_l, dst);
- dst += dst_stride;
-
- ILVRL_B2_UH(src_left1, src_top, res_r, res_l);
- HADD_UB2_UH(res_r, res_l, res_r, res_l);
- IPRED_SUBS_UH2_UH(src_top_left, src_top_left, res_r, res_l);
- SAT_UH2_UH(res_r, res_l, 7);
- PCKEV_ST_SB(res_r, res_l, dst);
- dst += dst_stride;
-
- ILVRL_B2_UH(src_left2, src_top, res_r, res_l);
- HADD_UB2_UH(res_r, res_l, res_r, res_l);
- IPRED_SUBS_UH2_UH(src_top_left, src_top_left, res_r, res_l);
- SAT_UH2_UH(res_r, res_l, 7);
- PCKEV_ST_SB(res_r, res_l, dst);
- dst += dst_stride;
-
- ILVRL_B2_UH(src_left3, src_top, res_r, res_l);
- HADD_UB2_UH(res_r, res_l, res_r, res_l);
- IPRED_SUBS_UH2_UH(src_top_left, src_top_left, res_r, res_l);
- SAT_UH2_UH(res_r, res_l, 7);
- PCKEV_ST_SB(res_r, res_l, dst);
- dst += dst_stride;
- }
-}
-
-static void intra_predict_tm_32x32_msa(const uint8_t *src_top,
- const uint8_t *src_left,
- uint8_t *dst, int32_t dst_stride) {- uint8_t top_left = src_top[-1];
- uint32_t loop_cnt;
- v16i8 src_top0, src_top1, src_left0, src_left1, src_left2, src_left3;
- v8u16 src_top_left, res_r0, res_r1, res_l0, res_l1;
-
- LD_SB2(src_top, 16, src_top0, src_top1);
- src_top_left = (v8u16)__msa_fill_h(top_left);
-
- for (loop_cnt = 8; loop_cnt--;) {- src_left0 = __msa_fill_b(src_left[0]);
- src_left1 = __msa_fill_b(src_left[1]);
- src_left2 = __msa_fill_b(src_left[2]);
- src_left3 = __msa_fill_b(src_left[3]);
- src_left += 4;
-
- ILVR_B2_UH(src_left0, src_top0, src_left0, src_top1, res_r0, res_r1);
- ILVL_B2_UH(src_left0, src_top0, src_left0, src_top1, res_l0, res_l1);
- HADD_UB4_UH(res_r0, res_l0, res_r1, res_l1, res_r0, res_l0, res_r1, res_l1);
- IPRED_SUBS_UH2_UH(src_top_left, src_top_left, res_r0, res_l0);
- IPRED_SUBS_UH2_UH(src_top_left, src_top_left, res_r1, res_l1);
- SAT_UH4_UH(res_r0, res_l0, res_r1, res_l1, 7);
- PCKEV_ST_SB(res_r0, res_l0, dst);
- PCKEV_ST_SB(res_r1, res_l1, dst + 16);
- dst += dst_stride;
-
- ILVR_B2_UH(src_left1, src_top0, src_left1, src_top1, res_r0, res_r1);
- ILVL_B2_UH(src_left1, src_top0, src_left1, src_top1, res_l0, res_l1);
- HADD_UB4_UH(res_r0, res_l0, res_r1, res_l1, res_r0, res_l0, res_r1, res_l1);
- IPRED_SUBS_UH2_UH(src_top_left, src_top_left, res_r0, res_l0);
- IPRED_SUBS_UH2_UH(src_top_left, src_top_left, res_r1, res_l1);
- SAT_UH4_UH(res_r0, res_l0, res_r1, res_l1, 7);
- PCKEV_ST_SB(res_r0, res_l0, dst);
- PCKEV_ST_SB(res_r1, res_l1, dst + 16);
- dst += dst_stride;
-
- ILVR_B2_UH(src_left2, src_top0, src_left2, src_top1, res_r0, res_r1);
- ILVL_B2_UH(src_left2, src_top0, src_left2, src_top1, res_l0, res_l1);
- HADD_UB4_UH(res_r0, res_l0, res_r1, res_l1, res_r0, res_l0, res_r1, res_l1);
- IPRED_SUBS_UH2_UH(src_top_left, src_top_left, res_r0, res_l0);
- IPRED_SUBS_UH2_UH(src_top_left, src_top_left, res_r1, res_l1);
- SAT_UH4_UH(res_r0, res_l0, res_r1, res_l1, 7);
- PCKEV_ST_SB(res_r0, res_l0, dst);
- PCKEV_ST_SB(res_r1, res_l1, dst + 16);
- dst += dst_stride;
-
- ILVR_B2_UH(src_left3, src_top0, src_left3, src_top1, res_r0, res_r1);
- ILVL_B2_UH(src_left3, src_top0, src_left3, src_top1, res_l0, res_l1);
- HADD_UB4_UH(res_r0, res_l0, res_r1, res_l1, res_r0, res_l0, res_r1, res_l1);
- IPRED_SUBS_UH2_UH(src_top_left, src_top_left, res_r0, res_l0);
- IPRED_SUBS_UH2_UH(src_top_left, src_top_left, res_r1, res_l1);
- SAT_UH4_UH(res_r0, res_l0, res_r1, res_l1, 7);
- PCKEV_ST_SB(res_r0, res_l0, dst);
- PCKEV_ST_SB(res_r1, res_l1, dst + 16);
- dst += dst_stride;
- }
-}
-
-void vp9_v_predictor_4x4_msa(uint8_t *dst, ptrdiff_t y_stride,
- const uint8_t *above, const uint8_t *left) {- (void)left;
-
- intra_predict_vert_4x4_msa(above, dst, y_stride);
-}
-
-void vp9_v_predictor_8x8_msa(uint8_t *dst, ptrdiff_t y_stride,
- const uint8_t *above, const uint8_t *left) {- (void)left;
-
- intra_predict_vert_8x8_msa(above, dst, y_stride);
-}
-
-void vp9_v_predictor_16x16_msa(uint8_t *dst, ptrdiff_t y_stride,
- const uint8_t *above, const uint8_t *left) {- (void)left;
-
- intra_predict_vert_16x16_msa(above, dst, y_stride);
-}
-
-void vp9_v_predictor_32x32_msa(uint8_t *dst, ptrdiff_t y_stride,
- const uint8_t *above, const uint8_t *left) {- (void)left;
-
- intra_predict_vert_32x32_msa(above, dst, y_stride);
-}
-
-void vp9_h_predictor_4x4_msa(uint8_t *dst, ptrdiff_t y_stride,
- const uint8_t *above, const uint8_t *left) {- (void)above;
-
- intra_predict_horiz_4x4_msa(left, dst, y_stride);
-}
-
-void vp9_h_predictor_8x8_msa(uint8_t *dst, ptrdiff_t y_stride,
- const uint8_t *above, const uint8_t *left) {- (void)above;
-
- intra_predict_horiz_8x8_msa(left, dst, y_stride);
-}
-
-void vp9_h_predictor_16x16_msa(uint8_t *dst, ptrdiff_t y_stride,
- const uint8_t *above, const uint8_t *left) {- (void)above;
-
- intra_predict_horiz_16x16_msa(left, dst, y_stride);
-}
-
-void vp9_h_predictor_32x32_msa(uint8_t *dst, ptrdiff_t y_stride,
- const uint8_t *above, const uint8_t *left) {- (void)above;
-
- intra_predict_horiz_32x32_msa(left, dst, y_stride);
-}
-
-void vp9_dc_predictor_4x4_msa(uint8_t *dst, ptrdiff_t y_stride,
- const uint8_t *above, const uint8_t *left) {- intra_predict_dc_4x4_msa(above, left, dst, y_stride);
-}
-
-void vp9_dc_predictor_8x8_msa(uint8_t *dst, ptrdiff_t y_stride,
- const uint8_t *above, const uint8_t *left) {- intra_predict_dc_8x8_msa(above, left, dst, y_stride);
-}
-
-void vp9_dc_predictor_16x16_msa(uint8_t *dst, ptrdiff_t y_stride,
- const uint8_t *above, const uint8_t *left) {- intra_predict_dc_16x16_msa(above, left, dst, y_stride);
-}
-
-void vp9_dc_predictor_32x32_msa(uint8_t *dst, ptrdiff_t y_stride,
- const uint8_t *above, const uint8_t *left) {- intra_predict_dc_32x32_msa(above, left, dst, y_stride);
-}
-
-void vp9_dc_top_predictor_4x4_msa(uint8_t *dst, ptrdiff_t y_stride,
- const uint8_t *above, const uint8_t *left) {- (void)left;
-
- intra_predict_dc_tl_4x4_msa(above, dst, y_stride);
-}
-
-void vp9_dc_top_predictor_8x8_msa(uint8_t *dst, ptrdiff_t y_stride,
- const uint8_t *above, const uint8_t *left) {- (void)left;
-
- intra_predict_dc_tl_8x8_msa(above, dst, y_stride);
-}
-
-void vp9_dc_top_predictor_16x16_msa(uint8_t *dst, ptrdiff_t y_stride,
- const uint8_t *above, const uint8_t *left) {- (void)left;
-
- intra_predict_dc_tl_16x16_msa(above, dst, y_stride);
-}
-
-void vp9_dc_top_predictor_32x32_msa(uint8_t *dst, ptrdiff_t y_stride,
- const uint8_t *above, const uint8_t *left) {- (void)left;
-
- intra_predict_dc_tl_32x32_msa(above, dst, y_stride);
-}
-
-void vp9_dc_left_predictor_4x4_msa(uint8_t *dst, ptrdiff_t y_stride,
- const uint8_t *above, const uint8_t *left) {- (void)above;
-
- intra_predict_dc_tl_4x4_msa(left, dst, y_stride);
-}
-
-void vp9_dc_left_predictor_8x8_msa(uint8_t *dst, ptrdiff_t y_stride,
- const uint8_t *above, const uint8_t *left) {- (void)above;
-
- intra_predict_dc_tl_8x8_msa(left, dst, y_stride);
-}
-
-void vp9_dc_left_predictor_16x16_msa(uint8_t *dst, ptrdiff_t y_stride,
- const uint8_t *above,
- const uint8_t *left) {- (void)above;
-
- intra_predict_dc_tl_16x16_msa(left, dst, y_stride);
-}
-
-void vp9_dc_left_predictor_32x32_msa(uint8_t *dst, ptrdiff_t y_stride,
- const uint8_t *above,
- const uint8_t *left) {- (void)above;
-
- intra_predict_dc_tl_32x32_msa(left, dst, y_stride);
-}
-
-void vp9_dc_128_predictor_4x4_msa(uint8_t *dst, ptrdiff_t y_stride,
- const uint8_t *above, const uint8_t *left) {- (void)above;
- (void)left;
-
- intra_predict_128dc_4x4_msa(dst, y_stride);
-}
-
-void vp9_dc_128_predictor_8x8_msa(uint8_t *dst, ptrdiff_t y_stride,
- const uint8_t *above, const uint8_t *left) {- (void)above;
- (void)left;
-
- intra_predict_128dc_8x8_msa(dst, y_stride);
-}
-
-void vp9_dc_128_predictor_16x16_msa(uint8_t *dst, ptrdiff_t y_stride,
- const uint8_t *above, const uint8_t *left) {- (void)above;
- (void)left;
-
- intra_predict_128dc_16x16_msa(dst, y_stride);
-}
-
-void vp9_dc_128_predictor_32x32_msa(uint8_t *dst, ptrdiff_t y_stride,
- const uint8_t *above, const uint8_t *left) {- (void)above;
- (void)left;
-
- intra_predict_128dc_32x32_msa(dst, y_stride);
-}
-
-void vp9_tm_predictor_4x4_msa(uint8_t *dst, ptrdiff_t y_stride,
- const uint8_t *above, const uint8_t *left) {- intra_predict_tm_4x4_msa(above, left, dst, y_stride);
-}
-
-void vp9_tm_predictor_8x8_msa(uint8_t *dst, ptrdiff_t y_stride,
- const uint8_t *above, const uint8_t *left) {- intra_predict_tm_8x8_msa(above, left, dst, y_stride);
-}
-
-void vp9_tm_predictor_16x16_msa(uint8_t *dst, ptrdiff_t y_stride,
- const uint8_t *above, const uint8_t *left) {- intra_predict_tm_16x16_msa(above, left, dst, y_stride);
-}
-
-void vp9_tm_predictor_32x32_msa(uint8_t *dst, ptrdiff_t y_stride,
- const uint8_t *above, const uint8_t *left) {- intra_predict_tm_32x32_msa(above, left, dst, y_stride);
-}
--- a/vp9/common/vp9_common.h
+++ b/vp9/common/vp9_common.h
@@ -44,20 +44,6 @@
return num_values > 0 ? get_msb(num_values) + 1 : 0;
}
-#if CONFIG_VP9_HIGHBITDEPTH
-static INLINE uint16_t clip_pixel_highbd(int val, int bd) {- switch (bd) {- case 8:
- default:
- return (uint16_t)clamp(val, 0, 255);
- case 10:
- return (uint16_t)clamp(val, 0, 1023);
- case 12:
- return (uint16_t)clamp(val, 0, 4095);
- }
-}
-#endif // CONFIG_VP9_HIGHBITDEPTH
-
#if CONFIG_DEBUG
#define CHECK_MEM_ERROR(cm, lval, expr) do { \lval = (expr); \
--- a/vp9/common/vp9_reconintra.c
+++ b/vp9/common/vp9_reconintra.c
@@ -9,7 +9,7 @@
*/
#include "./vpx_config.h"
-#include "./vp9_rtcd.h"
+#include "./vpx_dsp_rtcd.h"
#include "vpx_mem/vpx_mem.h"
#include "vpx_ports/mem.h"
@@ -49,683 +49,6 @@
NEED_ABOVERIGHT, // D63
NEED_LEFT | NEED_ABOVE, // TM
};
-
-// This serves as a wrapper function, so that all the prediction functions
-// can be unified and accessed as a pointer array. Note that the boundary
-// above and left are not necessarily used all the time.
-#define intra_pred_sized(type, size) \
- void vp9_##type##_predictor_##size##x##size##_c(uint8_t *dst, \
- ptrdiff_t stride, \
- const uint8_t *above, \
- const uint8_t *left) { \- type##_predictor(dst, stride, size, above, left); \
- }
-
-#if CONFIG_VP9_HIGHBITDEPTH
-#define intra_pred_highbd_sized(type, size) \
- void vp9_highbd_##type##_predictor_##size##x##size##_c( \
- uint16_t *dst, ptrdiff_t stride, const uint16_t *above, \
- const uint16_t *left, int bd) { \- highbd_##type##_predictor(dst, stride, size, above, left, bd); \
- }
-
-#define intra_pred_allsizes(type) \
- intra_pred_sized(type, 4) \
- intra_pred_sized(type, 8) \
- intra_pred_sized(type, 16) \
- intra_pred_sized(type, 32) \
- intra_pred_highbd_sized(type, 4) \
- intra_pred_highbd_sized(type, 8) \
- intra_pred_highbd_sized(type, 16) \
- intra_pred_highbd_sized(type, 32)
-
-#define intra_pred_no_4x4(type) \
- intra_pred_sized(type, 8) \
- intra_pred_sized(type, 16) \
- intra_pred_sized(type, 32) \
- intra_pred_highbd_sized(type, 4) \
- intra_pred_highbd_sized(type, 8) \
- intra_pred_highbd_sized(type, 16) \
- intra_pred_highbd_sized(type, 32)
-
-#else
-
-#define intra_pred_allsizes(type) \
- intra_pred_sized(type, 4) \
- intra_pred_sized(type, 8) \
- intra_pred_sized(type, 16) \
- intra_pred_sized(type, 32)
-
-#define intra_pred_no_4x4(type) \
- intra_pred_sized(type, 8) \
- intra_pred_sized(type, 16) \
- intra_pred_sized(type, 32)
-#endif // CONFIG_VP9_HIGHBITDEPTH
-
-#define DST(x, y) dst[(x) + (y) * stride]
-#define AVG3(a, b, c) (((a) + 2 * (b) + (c) + 2) >> 2)
-#define AVG2(a, b) (((a) + (b) + 1) >> 1)
-
-#if CONFIG_VP9_HIGHBITDEPTH
-static INLINE void highbd_d207_predictor(uint16_t *dst, ptrdiff_t stride,
- int bs, const uint16_t *above,
- const uint16_t *left, int bd) {- int r, c;
- (void) above;
- (void) bd;
-
- // First column.
- for (r = 0; r < bs - 1; ++r) {- dst[r * stride] = AVG2(left[r], left[r + 1]);
- }
- dst[(bs - 1) * stride] = left[bs - 1];
- dst++;
-
- // Second column.
- for (r = 0; r < bs - 2; ++r) {- dst[r * stride] = AVG3(left[r], left[r + 1], left[r + 2]);
- }
- dst[(bs - 2) * stride] = AVG3(left[bs - 2], left[bs - 1], left[bs - 1]);
- dst[(bs - 1) * stride] = left[bs - 1];
- dst++;
-
- // Rest of last row.
- for (c = 0; c < bs - 2; ++c)
- dst[(bs - 1) * stride + c] = left[bs - 1];
-
- for (r = bs - 2; r >= 0; --r) {- for (c = 0; c < bs - 2; ++c)
- dst[r * stride + c] = dst[(r + 1) * stride + c - 2];
- }
-}
-
-static INLINE void highbd_d63_predictor(uint16_t *dst, ptrdiff_t stride,
- int bs, const uint16_t *above,
- const uint16_t *left, int bd) {- int r, c;
- (void) left;
- (void) bd;
- for (r = 0; r < bs; ++r) {- for (c = 0; c < bs; ++c) {- dst[c] = r & 1 ? AVG3(above[(r >> 1) + c], above[(r >> 1) + c + 1],
- above[(r >> 1) + c + 2])
- : AVG2(above[(r >> 1) + c], above[(r >> 1) + c + 1]);
- }
- dst += stride;
- }
-}
-
-static INLINE void highbd_d45_predictor(uint16_t *dst, ptrdiff_t stride, int bs,
- const uint16_t *above,
- const uint16_t *left, int bd) {- int r, c;
- (void) left;
- (void) bd;
- for (r = 0; r < bs; ++r) {- for (c = 0; c < bs; ++c) {- dst[c] = r + c + 2 < bs * 2 ? AVG3(above[r + c], above[r + c + 1],
- above[r + c + 2])
- : above[bs * 2 - 1];
- }
- dst += stride;
- }
-}
-
-static INLINE void highbd_d117_predictor(uint16_t *dst, ptrdiff_t stride,
- int bs, const uint16_t *above,
- const uint16_t *left, int bd) {- int r, c;
- (void) bd;
-
- // first row
- for (c = 0; c < bs; c++)
- dst[c] = AVG2(above[c - 1], above[c]);
- dst += stride;
-
- // second row
- dst[0] = AVG3(left[0], above[-1], above[0]);
- for (c = 1; c < bs; c++)
- dst[c] = AVG3(above[c - 2], above[c - 1], above[c]);
- dst += stride;
-
- // the rest of first col
- dst[0] = AVG3(above[-1], left[0], left[1]);
- for (r = 3; r < bs; ++r)
- dst[(r - 2) * stride] = AVG3(left[r - 3], left[r - 2], left[r - 1]);
-
- // the rest of the block
- for (r = 2; r < bs; ++r) {- for (c = 1; c < bs; c++)
- dst[c] = dst[-2 * stride + c - 1];
- dst += stride;
- }
-}
-
-static INLINE void highbd_d135_predictor(uint16_t *dst, ptrdiff_t stride,
- int bs, const uint16_t *above,
- const uint16_t *left, int bd) {- int r, c;
- (void) bd;
- dst[0] = AVG3(left[0], above[-1], above[0]);
- for (c = 1; c < bs; c++)
- dst[c] = AVG3(above[c - 2], above[c - 1], above[c]);
-
- dst[stride] = AVG3(above[-1], left[0], left[1]);
- for (r = 2; r < bs; ++r)
- dst[r * stride] = AVG3(left[r - 2], left[r - 1], left[r]);
-
- dst += stride;
- for (r = 1; r < bs; ++r) {- for (c = 1; c < bs; c++)
- dst[c] = dst[-stride + c - 1];
- dst += stride;
- }
-}
-
-static INLINE void highbd_d153_predictor(uint16_t *dst, ptrdiff_t stride,
- int bs, const uint16_t *above,
- const uint16_t *left, int bd) {- int r, c;
- (void) bd;
- dst[0] = AVG2(above[-1], left[0]);
- for (r = 1; r < bs; r++)
- dst[r * stride] = AVG2(left[r - 1], left[r]);
- dst++;
-
- dst[0] = AVG3(left[0], above[-1], above[0]);
- dst[stride] = AVG3(above[-1], left[0], left[1]);
- for (r = 2; r < bs; r++)
- dst[r * stride] = AVG3(left[r - 2], left[r - 1], left[r]);
- dst++;
-
- for (c = 0; c < bs - 2; c++)
- dst[c] = AVG3(above[c - 1], above[c], above[c + 1]);
- dst += stride;
-
- for (r = 1; r < bs; ++r) {- for (c = 0; c < bs - 2; c++)
- dst[c] = dst[-stride + c - 2];
- dst += stride;
- }
-}
-
-static INLINE void highbd_v_predictor(uint16_t *dst, ptrdiff_t stride,
- int bs, const uint16_t *above,
- const uint16_t *left, int bd) {- int r;
- (void) left;
- (void) bd;
- for (r = 0; r < bs; r++) {- memcpy(dst, above, bs * sizeof(uint16_t));
- dst += stride;
- }
-}
-
-static INLINE void highbd_h_predictor(uint16_t *dst, ptrdiff_t stride,
- int bs, const uint16_t *above,
- const uint16_t *left, int bd) {- int r;
- (void) above;
- (void) bd;
- for (r = 0; r < bs; r++) {- vpx_memset16(dst, left[r], bs);
- dst += stride;
- }
-}
-
-static INLINE void highbd_tm_predictor(uint16_t *dst, ptrdiff_t stride,
- int bs, const uint16_t *above,
- const uint16_t *left, int bd) {- int r, c;
- int ytop_left = above[-1];
- (void) bd;
-
- for (r = 0; r < bs; r++) {- for (c = 0; c < bs; c++)
- dst[c] = clip_pixel_highbd(left[r] + above[c] - ytop_left, bd);
- dst += stride;
- }
-}
-
-static INLINE void highbd_dc_128_predictor(uint16_t *dst, ptrdiff_t stride,
- int bs, const uint16_t *above,
- const uint16_t *left, int bd) {- int r;
- (void) above;
- (void) left;
-
- for (r = 0; r < bs; r++) {- vpx_memset16(dst, 128 << (bd - 8), bs);
- dst += stride;
- }
-}
-
-static INLINE void highbd_dc_left_predictor(uint16_t *dst, ptrdiff_t stride,
- int bs, const uint16_t *above,
- const uint16_t *left, int bd) {- int i, r, expected_dc, sum = 0;
- (void) above;
- (void) bd;
-
- for (i = 0; i < bs; i++)
- sum += left[i];
- expected_dc = (sum + (bs >> 1)) / bs;
-
- for (r = 0; r < bs; r++) {- vpx_memset16(dst, expected_dc, bs);
- dst += stride;
- }
-}
-
-static INLINE void highbd_dc_top_predictor(uint16_t *dst, ptrdiff_t stride,
- int bs, const uint16_t *above,
- const uint16_t *left, int bd) {- int i, r, expected_dc, sum = 0;
- (void) left;
- (void) bd;
-
- for (i = 0; i < bs; i++)
- sum += above[i];
- expected_dc = (sum + (bs >> 1)) / bs;
-
- for (r = 0; r < bs; r++) {- vpx_memset16(dst, expected_dc, bs);
- dst += stride;
- }
-}
-
-static INLINE void highbd_dc_predictor(uint16_t *dst, ptrdiff_t stride,
- int bs, const uint16_t *above,
- const uint16_t *left, int bd) {- int i, r, expected_dc, sum = 0;
- const int count = 2 * bs;
- (void) bd;
-
- for (i = 0; i < bs; i++) {- sum += above[i];
- sum += left[i];
- }
-
- expected_dc = (sum + (count >> 1)) / count;
-
- for (r = 0; r < bs; r++) {- vpx_memset16(dst, expected_dc, bs);
- dst += stride;
- }
-}
-#endif // CONFIG_VP9_HIGHBITDEPTH
-
-void vp9_d207_predictor_4x4_c(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above, const uint8_t *left) {- const int I = left[0];
- const int J = left[1];
- const int K = left[2];
- const int L = left[3];
- (void)above;
- DST(0, 0) = AVG2(I, J);
- DST(2, 0) = DST(0, 1) = AVG2(J, K);
- DST(2, 1) = DST(0, 2) = AVG2(K, L);
- DST(1, 0) = AVG3(I, J, K);
- DST(3, 0) = DST(1, 1) = AVG3(J, K, L);
- DST(3, 1) = DST(1, 2) = AVG3(K, L, L);
- DST(3, 2) = DST(2, 2) =
- DST(0, 3) = DST(1, 3) = DST(2, 3) = DST(3, 3) = L;
-}
-
-static INLINE void d207_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
- const uint8_t *above, const uint8_t *left) {- int r, c;
- (void) above;
- // first column
- for (r = 0; r < bs - 1; ++r)
- dst[r * stride] = AVG2(left[r], left[r + 1]);
- dst[(bs - 1) * stride] = left[bs - 1];
- dst++;
-
- // second column
- for (r = 0; r < bs - 2; ++r)
- dst[r * stride] = AVG3(left[r], left[r + 1], left[r + 2]);
- dst[(bs - 2) * stride] = AVG3(left[bs - 2], left[bs - 1], left[bs - 1]);
- dst[(bs - 1) * stride] = left[bs - 1];
- dst++;
-
- // rest of last row
- for (c = 0; c < bs - 2; ++c)
- dst[(bs - 1) * stride + c] = left[bs - 1];
-
- for (r = bs - 2; r >= 0; --r)
- for (c = 0; c < bs - 2; ++c)
- dst[r * stride + c] = dst[(r + 1) * stride + c - 2];
-}
-intra_pred_no_4x4(d207)
-
-void vp9_d63_predictor_4x4_c(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above, const uint8_t *left) {- const int A = above[0];
- const int B = above[1];
- const int C = above[2];
- const int D = above[3];
- const int E = above[4];
- const int F = above[5];
- const int G = above[6];
- (void)left;
- DST(0, 0) = AVG2(A, B);
- DST(1, 0) = DST(0, 2) = AVG2(B, C);
- DST(2, 0) = DST(1, 2) = AVG2(C, D);
- DST(3, 0) = DST(2, 2) = AVG2(D, E);
- DST(3, 2) = AVG2(E, F); // differs from vp8
-
- DST(0, 1) = AVG3(A, B, C);
- DST(1, 1) = DST(0, 3) = AVG3(B, C, D);
- DST(2, 1) = DST(1, 3) = AVG3(C, D, E);
- DST(3, 1) = DST(2, 3) = AVG3(D, E, F);
- DST(3, 3) = AVG3(E, F, G); // differs from vp8
-}
-
-static INLINE void d63_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
- const uint8_t *above, const uint8_t *left) {- int r, c;
- int size;
- (void)left;
- for (c = 0; c < bs; ++c) {- dst[c] = AVG2(above[c], above[c + 1]);
- dst[stride + c] = AVG3(above[c], above[c + 1], above[c + 2]);
- }
- for (r = 2, size = bs - 2; r < bs; r += 2, --size) {- memcpy(dst + (r + 0) * stride, dst + (r >> 1), size);
- memset(dst + (r + 0) * stride + size, above[bs - 1], bs - size);
- memcpy(dst + (r + 1) * stride, dst + stride + (r >> 1), size);
- memset(dst + (r + 1) * stride + size, above[bs - 1], bs - size);
- }
-}
-intra_pred_no_4x4(d63)
-
-void vp9_d45_predictor_4x4_c(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above, const uint8_t *left) {- const int A = above[0];
- const int B = above[1];
- const int C = above[2];
- const int D = above[3];
- const int E = above[4];
- const int F = above[5];
- const int G = above[6];
- const int H = above[7];
- (void)stride;
- (void)left;
- DST(0, 0) = AVG3(A, B, C);
- DST(1, 0) = DST(0, 1) = AVG3(B, C, D);
- DST(2, 0) = DST(1, 1) = DST(0, 2) = AVG3(C, D, E);
- DST(3, 0) = DST(2, 1) = DST(1, 2) = DST(0, 3) = AVG3(D, E, F);
- DST(3, 1) = DST(2, 2) = DST(1, 3) = AVG3(E, F, G);
- DST(3, 2) = DST(2, 3) = AVG3(F, G, H);
- DST(3, 3) = H; // differs from vp8
-}
-
-static INLINE void d45_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
- const uint8_t *above, const uint8_t *left) {- const uint8_t above_right = above[bs - 1];
- const uint8_t *const dst_row0 = dst;
- int x, size;
- (void)left;
-
- for (x = 0; x < bs - 1; ++x) {- dst[x] = AVG3(above[x], above[x + 1], above[x + 2]);
- }
- dst[bs - 1] = above_right;
- dst += stride;
- for (x = 1, size = bs - 2; x < bs; ++x, --size) {- memcpy(dst, dst_row0 + x, size);
- memset(dst + size, above_right, x + 1);
- dst += stride;
- }
-}
-intra_pred_no_4x4(d45)
-
-void vp9_d117_predictor_4x4_c(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above, const uint8_t *left) {- const int I = left[0];
- const int J = left[1];
- const int K = left[2];
- const int X = above[-1];
- const int A = above[0];
- const int B = above[1];
- const int C = above[2];
- const int D = above[3];
- DST(0, 0) = DST(1, 2) = AVG2(X, A);
- DST(1, 0) = DST(2, 2) = AVG2(A, B);
- DST(2, 0) = DST(3, 2) = AVG2(B, C);
- DST(3, 0) = AVG2(C, D);
-
- DST(0, 3) = AVG3(K, J, I);
- DST(0, 2) = AVG3(J, I, X);
- DST(0, 1) = DST(1, 3) = AVG3(I, X, A);
- DST(1, 1) = DST(2, 3) = AVG3(X, A, B);
- DST(2, 1) = DST(3, 3) = AVG3(A, B, C);
- DST(3, 1) = AVG3(B, C, D);
-}
-
-static INLINE void d117_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
- const uint8_t *above, const uint8_t *left) {- int r, c;
-
- // first row
- for (c = 0; c < bs; c++)
- dst[c] = AVG2(above[c - 1], above[c]);
- dst += stride;
-
- // second row
- dst[0] = AVG3(left[0], above[-1], above[0]);
- for (c = 1; c < bs; c++)
- dst[c] = AVG3(above[c - 2], above[c - 1], above[c]);
- dst += stride;
-
- // the rest of first col
- dst[0] = AVG3(above[-1], left[0], left[1]);
- for (r = 3; r < bs; ++r)
- dst[(r - 2) * stride] = AVG3(left[r - 3], left[r - 2], left[r - 1]);
-
- // the rest of the block
- for (r = 2; r < bs; ++r) {- for (c = 1; c < bs; c++)
- dst[c] = dst[-2 * stride + c - 1];
- dst += stride;
- }
-}
-intra_pred_no_4x4(d117)
-
-void vp9_d135_predictor_4x4_c(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above, const uint8_t *left) {- const int I = left[0];
- const int J = left[1];
- const int K = left[2];
- const int L = left[3];
- const int X = above[-1];
- const int A = above[0];
- const int B = above[1];
- const int C = above[2];
- const int D = above[3];
- (void)stride;
- DST(0, 3) = AVG3(J, K, L);
- DST(1, 3) = DST(0, 2) = AVG3(I, J, K);
- DST(2, 3) = DST(1, 2) = DST(0, 1) = AVG3(X, I, J);
- DST(3, 3) = DST(2, 2) = DST(1, 1) = DST(0, 0) = AVG3(A, X, I);
- DST(3, 2) = DST(2, 1) = DST(1, 0) = AVG3(B, A, X);
- DST(3, 1) = DST(2, 0) = AVG3(C, B, A);
- DST(3, 0) = AVG3(D, C, B);
-}
-
-static INLINE void d135_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
- const uint8_t *above, const uint8_t *left) {- int r, c;
- dst[0] = AVG3(left[0], above[-1], above[0]);
- for (c = 1; c < bs; c++)
- dst[c] = AVG3(above[c - 2], above[c - 1], above[c]);
-
- dst[stride] = AVG3(above[-1], left[0], left[1]);
- for (r = 2; r < bs; ++r)
- dst[r * stride] = AVG3(left[r - 2], left[r - 1], left[r]);
-
- dst += stride;
- for (r = 1; r < bs; ++r) {- for (c = 1; c < bs; c++)
- dst[c] = dst[-stride + c - 1];
- dst += stride;
- }
-}
-intra_pred_no_4x4(d135)
-
-void vp9_d153_predictor_4x4_c(uint8_t *dst, ptrdiff_t stride,
- const uint8_t *above, const uint8_t *left) {- const int I = left[0];
- const int J = left[1];
- const int K = left[2];
- const int L = left[3];
- const int X = above[-1];
- const int A = above[0];
- const int B = above[1];
- const int C = above[2];
-
- DST(0, 0) = DST(2, 1) = AVG2(I, X);
- DST(0, 1) = DST(2, 2) = AVG2(J, I);
- DST(0, 2) = DST(2, 3) = AVG2(K, J);
- DST(0, 3) = AVG2(L, K);
-
- DST(3, 0) = AVG3(A, B, C);
- DST(2, 0) = AVG3(X, A, B);
- DST(1, 0) = DST(3, 1) = AVG3(I, X, A);
- DST(1, 1) = DST(3, 2) = AVG3(J, I, X);
- DST(1, 2) = DST(3, 3) = AVG3(K, J, I);
- DST(1, 3) = AVG3(L, K, J);
-}
-
-static INLINE void d153_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
- const uint8_t *above, const uint8_t *left) {- int r, c;
- dst[0] = AVG2(above[-1], left[0]);
- for (r = 1; r < bs; r++)
- dst[r * stride] = AVG2(left[r - 1], left[r]);
- dst++;
-
- dst[0] = AVG3(left[0], above[-1], above[0]);
- dst[stride] = AVG3(above[-1], left[0], left[1]);
- for (r = 2; r < bs; r++)
- dst[r * stride] = AVG3(left[r - 2], left[r - 1], left[r]);
- dst++;
-
- for (c = 0; c < bs - 2; c++)
- dst[c] = AVG3(above[c - 1], above[c], above[c + 1]);
- dst += stride;
-
- for (r = 1; r < bs; ++r) {- for (c = 0; c < bs - 2; c++)
- dst[c] = dst[-stride + c - 2];
- dst += stride;
- }
-}
-intra_pred_no_4x4(d153)
-
-static INLINE void v_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
- const uint8_t *above, const uint8_t *left) {- int r;
- (void) left;
-
- for (r = 0; r < bs; r++) {- memcpy(dst, above, bs);
- dst += stride;
- }
-}
-intra_pred_allsizes(v)
-
-static INLINE void h_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
- const uint8_t *above, const uint8_t *left) {- int r;
- (void) above;
-
- for (r = 0; r < bs; r++) {- memset(dst, left[r], bs);
- dst += stride;
- }
-}
-intra_pred_allsizes(h)
-
-static INLINE void tm_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
- const uint8_t *above, const uint8_t *left) {- int r, c;
- int ytop_left = above[-1];
-
- for (r = 0; r < bs; r++) {- for (c = 0; c < bs; c++)
- dst[c] = clip_pixel(left[r] + above[c] - ytop_left);
- dst += stride;
- }
-}
-intra_pred_allsizes(tm)
-
-static INLINE void dc_128_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
- const uint8_t *above, const uint8_t *left) {- int r;
- (void) above;
- (void) left;
-
- for (r = 0; r < bs; r++) {- memset(dst, 128, bs);
- dst += stride;
- }
-}
-intra_pred_allsizes(dc_128)
-
-static INLINE void dc_left_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
- const uint8_t *above,
- const uint8_t *left) {- int i, r, expected_dc, sum = 0;
- (void) above;
-
- for (i = 0; i < bs; i++)
- sum += left[i];
- expected_dc = (sum + (bs >> 1)) / bs;
-
- for (r = 0; r < bs; r++) {- memset(dst, expected_dc, bs);
- dst += stride;
- }
-}
-intra_pred_allsizes(dc_left)
-
-static INLINE void dc_top_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
- const uint8_t *above, const uint8_t *left) {- int i, r, expected_dc, sum = 0;
- (void) left;
-
- for (i = 0; i < bs; i++)
- sum += above[i];
- expected_dc = (sum + (bs >> 1)) / bs;
-
- for (r = 0; r < bs; r++) {- memset(dst, expected_dc, bs);
- dst += stride;
- }
-}
-intra_pred_allsizes(dc_top)
-
-static INLINE void dc_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
- const uint8_t *above, const uint8_t *left) {- int i, r, expected_dc, sum = 0;
- const int count = 2 * bs;
-
- for (i = 0; i < bs; i++) {- sum += above[i];
- sum += left[i];
- }
-
- expected_dc = (sum + (count >> 1)) / count;
-
- for (r = 0; r < bs; r++) {- memset(dst, expected_dc, bs);
- dst += stride;
- }
-}
-intra_pred_allsizes(dc)
-#undef intra_pred_allsizes
typedef void (*intra_pred_fn)(uint8_t *dst, ptrdiff_t stride,
const uint8_t *above, const uint8_t *left);
--- a/vp9/common/vp9_rtcd_defs.pl
+++ b/vp9/common/vp9_rtcd_defs.pl
@@ -61,165 +61,6 @@
}
#
-# RECON
-#
-add_proto qw/void vp9_d207_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_d207_predictor_4x4/, "$ssse3_x86inc";
-
-add_proto qw/void vp9_d45_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_d45_predictor_4x4 neon/, "$ssse3_x86inc";
-
-add_proto qw/void vp9_d63_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_d63_predictor_4x4/, "$ssse3_x86inc";
-
-add_proto qw/void vp9_h_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_h_predictor_4x4 neon dspr2 msa/, "$ssse3_x86inc";
-
-add_proto qw/void vp9_d117_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_d117_predictor_4x4/;
-
-add_proto qw/void vp9_d135_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_d135_predictor_4x4 neon/;
-
-add_proto qw/void vp9_d153_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_d153_predictor_4x4/, "$ssse3_x86inc";
-
-add_proto qw/void vp9_v_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_v_predictor_4x4 neon msa/, "$sse_x86inc";
-
-add_proto qw/void vp9_tm_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_tm_predictor_4x4 neon dspr2 msa/, "$sse_x86inc";
-
-add_proto qw/void vp9_dc_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_dc_predictor_4x4 dspr2 msa neon/, "$sse_x86inc";
-
-add_proto qw/void vp9_dc_top_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_dc_top_predictor_4x4 msa neon/, "$sse_x86inc";
-
-add_proto qw/void vp9_dc_left_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_dc_left_predictor_4x4 msa neon/, "$sse_x86inc";
-
-add_proto qw/void vp9_dc_128_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_dc_128_predictor_4x4 msa neon/, "$sse_x86inc";
-
-add_proto qw/void vp9_d207_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_d207_predictor_8x8/, "$ssse3_x86inc";
-
-add_proto qw/void vp9_d45_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_d45_predictor_8x8 neon/, "$ssse3_x86inc";
-
-add_proto qw/void vp9_d63_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_d63_predictor_8x8/, "$ssse3_x86inc";
-
-add_proto qw/void vp9_h_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_h_predictor_8x8 neon dspr2 msa/, "$ssse3_x86inc";
-
-add_proto qw/void vp9_d117_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_d117_predictor_8x8/;
-
-add_proto qw/void vp9_d135_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_d135_predictor_8x8/;
-
-add_proto qw/void vp9_d153_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_d153_predictor_8x8/, "$ssse3_x86inc";
-
-add_proto qw/void vp9_v_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_v_predictor_8x8 neon msa/, "$sse_x86inc";
-
-add_proto qw/void vp9_tm_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_tm_predictor_8x8 neon dspr2 msa/, "$sse2_x86inc";
-
-add_proto qw/void vp9_dc_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_dc_predictor_8x8 dspr2 neon msa/, "$sse_x86inc";
-
-add_proto qw/void vp9_dc_top_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_dc_top_predictor_8x8 neon msa/, "$sse_x86inc";
-
-add_proto qw/void vp9_dc_left_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_dc_left_predictor_8x8 neon msa/, "$sse_x86inc";
-
-add_proto qw/void vp9_dc_128_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_dc_128_predictor_8x8 neon msa/, "$sse_x86inc";
-
-add_proto qw/void vp9_d207_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_d207_predictor_16x16/, "$ssse3_x86inc";
-
-add_proto qw/void vp9_d45_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_d45_predictor_16x16 neon/, "$ssse3_x86inc";
-
-add_proto qw/void vp9_d63_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_d63_predictor_16x16/, "$ssse3_x86inc";
-
-add_proto qw/void vp9_h_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_h_predictor_16x16 neon dspr2 msa/, "$ssse3_x86inc";
-
-add_proto qw/void vp9_d117_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_d117_predictor_16x16/;
-
-add_proto qw/void vp9_d135_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_d135_predictor_16x16/;
-
-add_proto qw/void vp9_d153_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_d153_predictor_16x16/, "$ssse3_x86inc";
-
-add_proto qw/void vp9_v_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_v_predictor_16x16 neon msa/, "$sse2_x86inc";
-
-add_proto qw/void vp9_tm_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_tm_predictor_16x16 neon msa/, "$sse2_x86inc";
-
-add_proto qw/void vp9_dc_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_dc_predictor_16x16 dspr2 neon msa/, "$sse2_x86inc";
-
-add_proto qw/void vp9_dc_top_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_dc_top_predictor_16x16 neon msa/, "$sse2_x86inc";
-
-add_proto qw/void vp9_dc_left_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_dc_left_predictor_16x16 neon msa/, "$sse2_x86inc";
-
-add_proto qw/void vp9_dc_128_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_dc_128_predictor_16x16 neon msa/, "$sse2_x86inc";
-
-add_proto qw/void vp9_d207_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_d207_predictor_32x32/, "$ssse3_x86inc";
-
-add_proto qw/void vp9_d45_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_d45_predictor_32x32/, "$ssse3_x86inc";
-
-add_proto qw/void vp9_d63_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_d63_predictor_32x32/, "$ssse3_x86inc";
-
-add_proto qw/void vp9_h_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_h_predictor_32x32 neon msa/, "$ssse3_x86inc";
-
-add_proto qw/void vp9_d117_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_d117_predictor_32x32/;
-
-add_proto qw/void vp9_d135_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_d135_predictor_32x32/;
-
-add_proto qw/void vp9_d153_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_d153_predictor_32x32/, "$ssse3_x86inc";
-
-add_proto qw/void vp9_v_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_v_predictor_32x32 neon msa/, "$sse2_x86inc";
-
-add_proto qw/void vp9_tm_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_tm_predictor_32x32 neon msa/, "$sse2_x86_64_x86inc";
-
-add_proto qw/void vp9_dc_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_dc_predictor_32x32 msa neon/, "$sse2_x86inc";
-
-add_proto qw/void vp9_dc_top_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_dc_top_predictor_32x32 msa neon/, "$sse2_x86inc";
-
-add_proto qw/void vp9_dc_left_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_dc_left_predictor_32x32 msa neon/, "$sse2_x86inc";
-
-add_proto qw/void vp9_dc_128_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vp9_dc_128_predictor_32x32 msa neon/, "$sse2_x86inc";
-
-#
# post proc
#
if (vpx_config("CONFIG_VP9_POSTPROC") eq "yes") {@@ -438,165 +279,6 @@
# High bitdepth functions
if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {- #
- # Intra prediction
- #
- add_proto qw/void vp9_highbd_d207_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_d207_predictor_4x4/;
-
- add_proto qw/void vp9_highbd_d45_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_d45_predictor_4x4/;
-
- add_proto qw/void vp9_highbd_d63_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_d63_predictor_4x4/;
-
- add_proto qw/void vp9_highbd_h_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_h_predictor_4x4/;
-
- add_proto qw/void vp9_highbd_d117_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_d117_predictor_4x4/;
-
- add_proto qw/void vp9_highbd_d135_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_d135_predictor_4x4/;
-
- add_proto qw/void vp9_highbd_d153_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_d153_predictor_4x4/;
-
- add_proto qw/void vp9_highbd_v_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_v_predictor_4x4/, "$sse_x86inc";
-
- add_proto qw/void vp9_highbd_tm_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_tm_predictor_4x4/, "$sse_x86inc";
-
- add_proto qw/void vp9_highbd_dc_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_dc_predictor_4x4/, "$sse_x86inc";
-
- add_proto qw/void vp9_highbd_dc_top_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_dc_top_predictor_4x4/;
-
- add_proto qw/void vp9_highbd_dc_left_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_dc_left_predictor_4x4/;
-
- add_proto qw/void vp9_highbd_dc_128_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_dc_128_predictor_4x4/;
-
- add_proto qw/void vp9_highbd_d207_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_d207_predictor_8x8/;
-
- add_proto qw/void vp9_highbd_d45_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_d45_predictor_8x8/;
-
- add_proto qw/void vp9_highbd_d63_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_d63_predictor_8x8/;
-
- add_proto qw/void vp9_highbd_h_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_h_predictor_8x8/;
-
- add_proto qw/void vp9_highbd_d117_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_d117_predictor_8x8/;
-
- add_proto qw/void vp9_highbd_d135_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_d135_predictor_8x8/;
-
- add_proto qw/void vp9_highbd_d153_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_d153_predictor_8x8/;
-
- add_proto qw/void vp9_highbd_v_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_v_predictor_8x8/, "$sse2_x86inc";
-
- add_proto qw/void vp9_highbd_tm_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_tm_predictor_8x8/, "$sse2_x86inc";
-
- add_proto qw/void vp9_highbd_dc_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_dc_predictor_8x8/, "$sse2_x86inc";;
-
- add_proto qw/void vp9_highbd_dc_top_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_dc_top_predictor_8x8/;
-
- add_proto qw/void vp9_highbd_dc_left_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_dc_left_predictor_8x8/;
-
- add_proto qw/void vp9_highbd_dc_128_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_dc_128_predictor_8x8/;
-
- add_proto qw/void vp9_highbd_d207_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_d207_predictor_16x16/;
-
- add_proto qw/void vp9_highbd_d45_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_d45_predictor_16x16/;
-
- add_proto qw/void vp9_highbd_d63_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_d63_predictor_16x16/;
-
- add_proto qw/void vp9_highbd_h_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_h_predictor_16x16/;
-
- add_proto qw/void vp9_highbd_d117_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_d117_predictor_16x16/;
-
- add_proto qw/void vp9_highbd_d135_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_d135_predictor_16x16/;
-
- add_proto qw/void vp9_highbd_d153_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_d153_predictor_16x16/;
-
- add_proto qw/void vp9_highbd_v_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_v_predictor_16x16/, "$sse2_x86inc";
-
- add_proto qw/void vp9_highbd_tm_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_tm_predictor_16x16/, "$sse2_x86_64_x86inc";
-
- add_proto qw/void vp9_highbd_dc_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_dc_predictor_16x16/, "$sse2_x86inc";
-
- add_proto qw/void vp9_highbd_dc_top_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_dc_top_predictor_16x16/;
-
- add_proto qw/void vp9_highbd_dc_left_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_dc_left_predictor_16x16/;
-
- add_proto qw/void vp9_highbd_dc_128_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_dc_128_predictor_16x16/;
-
- add_proto qw/void vp9_highbd_d207_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_d207_predictor_32x32/;
-
- add_proto qw/void vp9_highbd_d45_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_d45_predictor_32x32/;
-
- add_proto qw/void vp9_highbd_d63_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_d63_predictor_32x32/;
-
- add_proto qw/void vp9_highbd_h_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_h_predictor_32x32/;
-
- add_proto qw/void vp9_highbd_d117_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_d117_predictor_32x32/;
-
- add_proto qw/void vp9_highbd_d135_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_d135_predictor_32x32/;
-
- add_proto qw/void vp9_highbd_d153_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_d153_predictor_32x32/;
-
- add_proto qw/void vp9_highbd_v_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_v_predictor_32x32/, "$sse2_x86inc";
-
- add_proto qw/void vp9_highbd_tm_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_tm_predictor_32x32/, "$sse2_x86_64_x86inc";
-
- add_proto qw/void vp9_highbd_dc_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_dc_predictor_32x32/, "$sse2_x86_64_x86inc";
-
- add_proto qw/void vp9_highbd_dc_top_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_dc_top_predictor_32x32/;
-
- add_proto qw/void vp9_highbd_dc_left_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_dc_left_predictor_32x32/;
-
- add_proto qw/void vp9_highbd_dc_128_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vp9_highbd_dc_128_predictor_32x32/;
-
#
# Sub Pixel Filters
#
--- a/vp9/common/x86/vp9_high_intrapred_sse2.asm
+++ /dev/null
@@ -1,476 +1,0 @@
-;
-; Copyright (c) 2014 The WebM project authors. All Rights Reserved.
-;
-; Use of this source code is governed by a BSD-style license
-; that can be found in the LICENSE file in the root of the source
-; tree. An additional intellectual property rights grant can be found
-; in the file PATENTS. All contributing project authors may
-; be found in the AUTHORS file in the root of the source tree.
-;
-
-%include "third_party/x86inc/x86inc.asm"
-
-SECTION_RODATA
-pw_4: times 8 dw 4
-pw_8: times 8 dw 8
-pw_16: times 4 dd 16
-pw_32: times 4 dd 32
-
-SECTION .text
-INIT_MMX sse
-cglobal highbd_dc_predictor_4x4, 4, 5, 4, dst, stride, above, left, goffset
- GET_GOT goffsetq
-
- movq m0, [aboveq]
- movq m2, [leftq]
- DEFINE_ARGS dst, stride, one
- mov oned, 0x0001
- pxor m1, m1
- movd m3, oned
- pshufw m3, m3, 0x0
- paddw m0, m2
- pmaddwd m0, m3
- packssdw m0, m1
- pmaddwd m0, m3
- paddw m0, [GLOBAL(pw_4)]
- psraw m0, 3
- pshufw m0, m0, 0x0
- movq [dstq ], m0
- movq [dstq+strideq*2], m0
- lea dstq, [dstq+strideq*4]
- movq [dstq ], m0
- movq [dstq+strideq*2], m0
-
- RESTORE_GOT
- RET
-
-INIT_XMM sse2
-cglobal highbd_dc_predictor_8x8, 4, 5, 4, dst, stride, above, left, goffset
- GET_GOT goffsetq
-
- pxor m1, m1
- mova m0, [aboveq]
- mova m2, [leftq]
- DEFINE_ARGS dst, stride, stride3, one
- mov oned, 0x00010001
- lea stride3q, [strideq*3]
- movd m3, oned
- pshufd m3, m3, 0x0
- paddw m0, m2
- pmaddwd m0, m3
- packssdw m0, m1
- pmaddwd m0, m3
- packssdw m0, m1
- pmaddwd m0, m3
- paddw m0, [GLOBAL(pw_8)]
- psrlw m0, 4
- pshuflw m0, m0, 0x0
- punpcklqdq m0, m0
- mova [dstq ], m0
- mova [dstq+strideq*2 ], m0
- mova [dstq+strideq*4 ], m0
- mova [dstq+stride3q*2], m0
- lea dstq, [dstq+strideq*8]
- mova [dstq ], m0
- mova [dstq+strideq*2 ], m0
- mova [dstq+strideq*4 ], m0
- mova [dstq+stride3q*2], m0
-
- RESTORE_GOT
- RET
-
-INIT_XMM sse2
-cglobal highbd_dc_predictor_16x16, 4, 5, 5, dst, stride, above, left, goffset
- GET_GOT goffsetq
-
- pxor m1, m1
- mova m0, [aboveq]
- mova m3, [aboveq+16]
- mova m2, [leftq]
- mova m4, [leftq+16]
- DEFINE_ARGS dst, stride, stride3, lines4
- lea stride3q, [strideq*3]
- mov lines4d, 4
- paddw m0, m2
- paddw m0, m3
- paddw m0, m4
- movhlps m2, m0
- paddw m0, m2
- punpcklwd m0, m1
- movhlps m2, m0
- paddd m0, m2
- punpckldq m0, m1
- movhlps m2, m0
- paddd m0, m2
- paddd m0, [GLOBAL(pw_16)]
- psrad m0, 5
- pshuflw m0, m0, 0x0
- punpcklqdq m0, m0
-.loop:
- mova [dstq ], m0
- mova [dstq +16], m0
- mova [dstq+strideq*2 ], m0
- mova [dstq+strideq*2 +16], m0
- mova [dstq+strideq*4 ], m0
- mova [dstq+strideq*4 +16], m0
- mova [dstq+stride3q*2 ], m0
- mova [dstq+stride3q*2+16], m0
- lea dstq, [dstq+strideq*8]
- dec lines4d
- jnz .loop
-
- RESTORE_GOT
- REP_RET
-
-%if ARCH_X86_64
-INIT_XMM sse2
-cglobal highbd_dc_predictor_32x32, 4, 5, 9, dst, stride, above, left, goffset
- GET_GOT goffsetq
-
- pxor m1, m1
- mova m0, [aboveq]
- mova m2, [aboveq+16]
- mova m3, [aboveq+32]
- mova m4, [aboveq+48]
- mova m5, [leftq]
- mova m6, [leftq+16]
- mova m7, [leftq+32]
- mova m8, [leftq+48]
- DEFINE_ARGS dst, stride, stride3, lines4
- lea stride3q, [strideq*3]
- mov lines4d, 8
- paddw m0, m2
- paddw m0, m3
- paddw m0, m4
- paddw m0, m5
- paddw m0, m6
- paddw m0, m7
- paddw m0, m8
- movhlps m2, m0
- paddw m0, m2
- punpcklwd m0, m1
- movhlps m2, m0
- paddd m0, m2
- punpckldq m0, m1
- movhlps m2, m0
- paddd m0, m2
- paddd m0, [GLOBAL(pw_32)]
- psrad m0, 6
- pshuflw m0, m0, 0x0
- punpcklqdq m0, m0
-.loop:
- mova [dstq ], m0
- mova [dstq +16 ], m0
- mova [dstq +32 ], m0
- mova [dstq +48 ], m0
- mova [dstq+strideq*2 ], m0
- mova [dstq+strideq*2+16 ], m0
- mova [dstq+strideq*2+32 ], m0
- mova [dstq+strideq*2+48 ], m0
- mova [dstq+strideq*4 ], m0
- mova [dstq+strideq*4+16 ], m0
- mova [dstq+strideq*4+32 ], m0
- mova [dstq+strideq*4+48 ], m0
- mova [dstq+stride3q*2 ], m0
- mova [dstq+stride3q*2 +16], m0
- mova [dstq+stride3q*2 +32], m0
- mova [dstq+stride3q*2 +48], m0
- lea dstq, [dstq+strideq*8]
- dec lines4d
- jnz .loop
-
- RESTORE_GOT
- REP_RET
-%endif
-
-INIT_MMX sse
-cglobal highbd_v_predictor_4x4, 3, 3, 1, dst, stride, above
- movq m0, [aboveq]
- movq [dstq ], m0
- movq [dstq+strideq*2], m0
- lea dstq, [dstq+strideq*4]
- movq [dstq ], m0
- movq [dstq+strideq*2], m0
- RET
-
-INIT_XMM sse2
-cglobal highbd_v_predictor_8x8, 3, 3, 1, dst, stride, above
- mova m0, [aboveq]
- DEFINE_ARGS dst, stride, stride3
- lea stride3q, [strideq*3]
- mova [dstq ], m0
- mova [dstq+strideq*2 ], m0
- mova [dstq+strideq*4 ], m0
- mova [dstq+stride3q*2], m0
- lea dstq, [dstq+strideq*8]
- mova [dstq ], m0
- mova [dstq+strideq*2 ], m0
- mova [dstq+strideq*4 ], m0
- mova [dstq+stride3q*2], m0
- RET
-
-INIT_XMM sse2
-cglobal highbd_v_predictor_16x16, 3, 4, 2, dst, stride, above
- mova m0, [aboveq]
- mova m1, [aboveq+16]
- DEFINE_ARGS dst, stride, stride3, nlines4
- lea stride3q, [strideq*3]
- mov nlines4d, 4
-.loop:
- mova [dstq ], m0
- mova [dstq +16], m1
- mova [dstq+strideq*2 ], m0
- mova [dstq+strideq*2 +16], m1
- mova [dstq+strideq*4 ], m0
- mova [dstq+strideq*4 +16], m1
- mova [dstq+stride3q*2 ], m0
- mova [dstq+stride3q*2+16], m1
- lea dstq, [dstq+strideq*8]
- dec nlines4d
- jnz .loop
- REP_RET
-
-INIT_XMM sse2
-cglobal highbd_v_predictor_32x32, 3, 4, 4, dst, stride, above
- mova m0, [aboveq]
- mova m1, [aboveq+16]
- mova m2, [aboveq+32]
- mova m3, [aboveq+48]
- DEFINE_ARGS dst, stride, stride3, nlines4
- lea stride3q, [strideq*3]
- mov nlines4d, 8
-.loop:
- mova [dstq ], m0
- mova [dstq +16], m1
- mova [dstq +32], m2
- mova [dstq +48], m3
- mova [dstq+strideq*2 ], m0
- mova [dstq+strideq*2 +16], m1
- mova [dstq+strideq*2 +32], m2
- mova [dstq+strideq*2 +48], m3
- mova [dstq+strideq*4 ], m0
- mova [dstq+strideq*4 +16], m1
- mova [dstq+strideq*4 +32], m2
- mova [dstq+strideq*4 +48], m3
- mova [dstq+stride3q*2 ], m0
- mova [dstq+stride3q*2 +16], m1
- mova [dstq+stride3q*2 +32], m2
- mova [dstq+stride3q*2 +48], m3
- lea dstq, [dstq+strideq*8]
- dec nlines4d
- jnz .loop
- REP_RET
-
-INIT_MMX sse
-cglobal highbd_tm_predictor_4x4, 5, 6, 5, dst, stride, above, left, bps, one
- movd m1, [aboveq-2]
- movq m0, [aboveq]
- pshufw m1, m1, 0x0
- ; Get the values to compute the maximum value at this bit depth
- mov oned, 1
- movd m3, oned
- movd m4, bpsd
- pshufw m3, m3, 0x0
- DEFINE_ARGS dst, stride, line, left
- mov lineq, -2
- mova m2, m3
- psllw m3, m4
- add leftq, 8
- psubw m3, m2 ; max possible value
- pxor m4, m4 ; min possible value
- psubw m0, m1
-.loop:
- movq m1, [leftq+lineq*4]
- movq m2, [leftq+lineq*4+2]
- pshufw m1, m1, 0x0
- pshufw m2, m2, 0x0
- paddw m1, m0
- paddw m2, m0
- ;Clamp to the bit-depth
- pminsw m1, m3
- pminsw m2, m3
- pmaxsw m1, m4
- pmaxsw m2, m4
- ;Store the values
- movq [dstq ], m1
- movq [dstq+strideq*2], m2
- lea dstq, [dstq+strideq*4]
- inc lineq
- jnz .loop
- REP_RET
-
-INIT_XMM sse2
-cglobal highbd_tm_predictor_8x8, 5, 6, 5, dst, stride, above, left, bps, one
- movd m1, [aboveq-2]
- mova m0, [aboveq]
- pshuflw m1, m1, 0x0
- ; Get the values to compute the maximum value at this bit depth
- mov oned, 1
- pxor m3, m3
- pxor m4, m4
- pinsrw m3, oned, 0
- pinsrw m4, bpsd, 0
- pshuflw m3, m3, 0x0
- DEFINE_ARGS dst, stride, line, left
- punpcklqdq m3, m3
- mov lineq, -4
- mova m2, m3
- punpcklqdq m1, m1
- psllw m3, m4
- add leftq, 16
- psubw m3, m2 ; max possible value
- pxor m4, m4 ; min possible value
- psubw m0, m1
-.loop:
- movd m1, [leftq+lineq*4]
- movd m2, [leftq+lineq*4+2]
- pshuflw m1, m1, 0x0
- pshuflw m2, m2, 0x0
- punpcklqdq m1, m1
- punpcklqdq m2, m2
- paddw m1, m0
- paddw m2, m0
- ;Clamp to the bit-depth
- pminsw m1, m3
- pminsw m2, m3
- pmaxsw m1, m4
- pmaxsw m2, m4
- ;Store the values
- mova [dstq ], m1
- mova [dstq+strideq*2], m2
- lea dstq, [dstq+strideq*4]
- inc lineq
- jnz .loop
- REP_RET
-
-%if ARCH_X86_64
-INIT_XMM sse2
-cglobal highbd_tm_predictor_16x16, 5, 6, 9, dst, stride, above, left, bps, one
- movd m2, [aboveq-2]
- mova m0, [aboveq]
- mova m1, [aboveq+16]
- pshuflw m2, m2, 0x0
- ; Get the values to compute the maximum value at this bit depth
- mov oned, 1
- pxor m7, m7
- pxor m8, m8
- pinsrw m7, oned, 0
- pinsrw m8, bpsd, 0
- pshuflw m7, m7, 0x0
- DEFINE_ARGS dst, stride, line, left
- punpcklqdq m7, m7
- mov lineq, -8
- mova m5, m7
- punpcklqdq m2, m2
- psllw m7, m8
- add leftq, 32
- psubw m7, m5 ; max possible value
- pxor m8, m8 ; min possible value
- psubw m0, m2
- psubw m1, m2
-.loop:
- movd m2, [leftq+lineq*4]
- movd m3, [leftq+lineq*4+2]
- pshuflw m2, m2, 0x0
- pshuflw m3, m3, 0x0
- punpcklqdq m2, m2
- punpcklqdq m3, m3
- paddw m4, m2, m0
- paddw m5, m3, m0
- paddw m2, m1
- paddw m3, m1
- ;Clamp to the bit-depth
- pminsw m4, m7
- pminsw m5, m7
- pminsw m2, m7
- pminsw m3, m7
- pmaxsw m4, m8
- pmaxsw m5, m8
- pmaxsw m2, m8
- pmaxsw m3, m8
- ;Store the values
- mova [dstq ], m4
- mova [dstq+strideq*2 ], m5
- mova [dstq +16], m2
- mova [dstq+strideq*2+16], m3
- lea dstq, [dstq+strideq*4]
- inc lineq
- jnz .loop
- REP_RET
-
-INIT_XMM sse2
-cglobal highbd_tm_predictor_32x32, 5, 6, 12, dst, stride, above, left, bps, one
- movd m0, [aboveq-2]
- mova m1, [aboveq]
- mova m2, [aboveq+16]
- mova m3, [aboveq+32]
- mova m4, [aboveq+48]
- pshuflw m0, m0, 0x0
- ; Get the values to compute the maximum value at this bit depth
- mov oned, 1
- pxor m10, m10
- pxor m11, m11
- pinsrw m10, oned, 0
- pinsrw m11, bpsd, 0
- pshuflw m10, m10, 0x0
- DEFINE_ARGS dst, stride, line, left
- punpcklqdq m10, m10
- mov lineq, -16
- mova m5, m10
- punpcklqdq m0, m0
- psllw m10, m11
- add leftq, 64
- psubw m10, m5 ; max possible value
- pxor m11, m11 ; min possible value
- psubw m1, m0
- psubw m2, m0
- psubw m3, m0
- psubw m4, m0
-.loop:
- movd m5, [leftq+lineq*4]
- movd m6, [leftq+lineq*4+2]
- pshuflw m5, m5, 0x0
- pshuflw m6, m6, 0x0
- punpcklqdq m5, m5
- punpcklqdq m6, m6
- paddw m7, m5, m1
- paddw m8, m5, m2
- paddw m9, m5, m3
- paddw m5, m4
- ;Clamp these values to the bit-depth
- pminsw m7, m10
- pminsw m8, m10
- pminsw m9, m10
- pminsw m5, m10
- pmaxsw m7, m11
- pmaxsw m8, m11
- pmaxsw m9, m11
- pmaxsw m5, m11
- ;Store these values
- mova [dstq ], m7
- mova [dstq +16], m8
- mova [dstq +32], m9
- mova [dstq +48], m5
- paddw m7, m6, m1
- paddw m8, m6, m2
- paddw m9, m6, m3
- paddw m6, m4
- ;Clamp these values to the bit-depth
- pminsw m7, m10
- pminsw m8, m10
- pminsw m9, m10
- pminsw m6, m10
- pmaxsw m7, m11
- pmaxsw m8, m11
- pmaxsw m9, m11
- pmaxsw m6, m11
- ;Store these values
- mova [dstq+strideq*2 ], m7
- mova [dstq+strideq*2+16], m8
- mova [dstq+strideq*2+32], m9
- mova [dstq+strideq*2+48], m6
- lea dstq, [dstq+strideq*4]
- inc lineq
- jnz .loop
- REP_RET
-%endif
--- a/vp9/common/x86/vp9_intrapred_sse2.asm
+++ /dev/null
@@ -1,667 +1,0 @@
-;
-; Copyright (c) 2010 The WebM project authors. All Rights Reserved.
-;
-; Use of this source code is governed by a BSD-style license
-; that can be found in the LICENSE file in the root of the source
-; tree. An additional intellectual property rights grant can be found
-; in the file PATENTS. All contributing project authors may
-; be found in the AUTHORS file in the root of the source tree.
-;
-
-%include "third_party/x86inc/x86inc.asm"
-
-SECTION_RODATA
-pw_4: times 8 dw 4
-pw_8: times 8 dw 8
-pw_16: times 8 dw 16
-pw_32: times 8 dw 32
-dc_128: times 16 db 128
-pw2_4: times 8 dw 2
-pw2_8: times 8 dw 4
-pw2_16: times 8 dw 8
-pw2_32: times 8 dw 16
-
-SECTION .text
-
-INIT_MMX sse
-cglobal dc_predictor_4x4, 4, 5, 2, dst, stride, above, left, goffset
- GET_GOT goffsetq
-
- pxor m1, m1
- movd m0, [aboveq]
- punpckldq m0, [leftq]
- psadbw m0, m1
- paddw m0, [GLOBAL(pw_4)]
- psraw m0, 3
- pshufw m0, m0, 0x0
- packuswb m0, m0
- movd [dstq ], m0
- movd [dstq+strideq], m0
- lea dstq, [dstq+strideq*2]
- movd [dstq ], m0
- movd [dstq+strideq], m0
-
- RESTORE_GOT
- RET
-
-INIT_MMX sse
-cglobal dc_left_predictor_4x4, 4, 5, 2, dst, stride, above, left, goffset
- GET_GOT goffsetq
-
- pxor m1, m1
- movd m0, [leftq]
- psadbw m0, m1
- paddw m0, [GLOBAL(pw2_4)]
- psraw m0, 2
- pshufw m0, m0, 0x0
- packuswb m0, m0
- movd [dstq ], m0
- movd [dstq+strideq], m0
- lea dstq, [dstq+strideq*2]
- movd [dstq ], m0
- movd [dstq+strideq], m0
-
- RESTORE_GOT
- RET
-
-INIT_MMX sse
-cglobal dc_top_predictor_4x4, 4, 5, 2, dst, stride, above, left, goffset
- GET_GOT goffsetq
-
- pxor m1, m1
- movd m0, [aboveq]
- psadbw m0, m1
- paddw m0, [GLOBAL(pw2_4)]
- psraw m0, 2
- pshufw m0, m0, 0x0
- packuswb m0, m0
- movd [dstq ], m0
- movd [dstq+strideq], m0
- lea dstq, [dstq+strideq*2]
- movd [dstq ], m0
- movd [dstq+strideq], m0
-
- RESTORE_GOT
- RET
-
-INIT_MMX sse
-cglobal dc_predictor_8x8, 4, 5, 3, dst, stride, above, left, goffset
- GET_GOT goffsetq
-
- pxor m1, m1
- movq m0, [aboveq]
- movq m2, [leftq]
- DEFINE_ARGS dst, stride, stride3
- lea stride3q, [strideq*3]
- psadbw m0, m1
- psadbw m2, m1
- paddw m0, m2
- paddw m0, [GLOBAL(pw_8)]
- psraw m0, 4
- pshufw m0, m0, 0x0
- packuswb m0, m0
- movq [dstq ], m0
- movq [dstq+strideq ], m0
- movq [dstq+strideq*2], m0
- movq [dstq+stride3q ], m0
- lea dstq, [dstq+strideq*4]
- movq [dstq ], m0
- movq [dstq+strideq ], m0
- movq [dstq+strideq*2], m0
- movq [dstq+stride3q ], m0
-
- RESTORE_GOT
- RET
-
-INIT_MMX sse
-cglobal dc_top_predictor_8x8, 4, 5, 3, dst, stride, above, left, goffset
- GET_GOT goffsetq
-
- pxor m1, m1
- movq m0, [aboveq]
- DEFINE_ARGS dst, stride, stride3
- lea stride3q, [strideq*3]
- psadbw m0, m1
- paddw m0, [GLOBAL(pw2_8)]
- psraw m0, 3
- pshufw m0, m0, 0x0
- packuswb m0, m0
- movq [dstq ], m0
- movq [dstq+strideq ], m0
- movq [dstq+strideq*2], m0
- movq [dstq+stride3q ], m0
- lea dstq, [dstq+strideq*4]
- movq [dstq ], m0
- movq [dstq+strideq ], m0
- movq [dstq+strideq*2], m0
- movq [dstq+stride3q ], m0
-
- RESTORE_GOT
- RET
-
-INIT_MMX sse
-cglobal dc_left_predictor_8x8, 4, 5, 3, dst, stride, above, left, goffset
- GET_GOT goffsetq
-
- pxor m1, m1
- movq m0, [leftq]
- DEFINE_ARGS dst, stride, stride3
- lea stride3q, [strideq*3]
- psadbw m0, m1
- paddw m0, [GLOBAL(pw2_8)]
- psraw m0, 3
- pshufw m0, m0, 0x0
- packuswb m0, m0
- movq [dstq ], m0
- movq [dstq+strideq ], m0
- movq [dstq+strideq*2], m0
- movq [dstq+stride3q ], m0
- lea dstq, [dstq+strideq*4]
- movq [dstq ], m0
- movq [dstq+strideq ], m0
- movq [dstq+strideq*2], m0
- movq [dstq+stride3q ], m0
-
- RESTORE_GOT
- RET
-
-INIT_MMX sse
-cglobal dc_128_predictor_4x4, 4, 5, 3, dst, stride, above, left, goffset
- GET_GOT goffsetq
-
- DEFINE_ARGS dst, stride, stride3
- lea stride3q, [strideq*3]
- movd m0, [GLOBAL(dc_128)]
- movd [dstq ], m0
- movd [dstq+strideq ], m0
- movd [dstq+strideq*2], m0
- movd [dstq+stride3q ], m0
- RESTORE_GOT
- RET
-
-INIT_MMX sse
-cglobal dc_128_predictor_8x8, 4, 5, 3, dst, stride, above, left, goffset
- GET_GOT goffsetq
-
- DEFINE_ARGS dst, stride, stride3
- lea stride3q, [strideq*3]
- movq m0, [GLOBAL(dc_128)]
- movq [dstq ], m0
- movq [dstq+strideq ], m0
- movq [dstq+strideq*2], m0
- movq [dstq+stride3q ], m0
- lea dstq, [dstq+strideq*4]
- movq [dstq ], m0
- movq [dstq+strideq ], m0
- movq [dstq+strideq*2], m0
- movq [dstq+stride3q ], m0
- RESTORE_GOT
- RET
-
-INIT_XMM sse2
-cglobal dc_predictor_16x16, 4, 5, 3, dst, stride, above, left, goffset
- GET_GOT goffsetq
-
- pxor m1, m1
- mova m0, [aboveq]
- mova m2, [leftq]
- DEFINE_ARGS dst, stride, stride3, lines4
- lea stride3q, [strideq*3]
- mov lines4d, 4
- psadbw m0, m1
- psadbw m2, m1
- paddw m0, m2
- movhlps m2, m0
- paddw m0, m2
- paddw m0, [GLOBAL(pw_16)]
- psraw m0, 5
- pshuflw m0, m0, 0x0
- punpcklqdq m0, m0
- packuswb m0, m0
-.loop:
- mova [dstq ], m0
- mova [dstq+strideq ], m0
- mova [dstq+strideq*2], m0
- mova [dstq+stride3q ], m0
- lea dstq, [dstq+strideq*4]
- dec lines4d
- jnz .loop
-
- RESTORE_GOT
- REP_RET
-
-
-INIT_XMM sse2
-cglobal dc_top_predictor_16x16, 4, 5, 3, dst, stride, above, left, goffset
- GET_GOT goffsetq
-
- pxor m1, m1
- pxor m2, m2
- mova m0, [aboveq]
- DEFINE_ARGS dst, stride, stride3, lines4
- lea stride3q, [strideq*3]
- mov lines4d, 4
- psadbw m0, m1
- psadbw m2, m1
- paddw m0, m2
- movhlps m2, m0
- paddw m0, m2
- paddw m0, [GLOBAL(pw2_16)]
- psraw m0, 4
- pshuflw m0, m0, 0x0
- punpcklqdq m0, m0
- packuswb m0, m0
-.loop:
- mova [dstq ], m0
- mova [dstq+strideq ], m0
- mova [dstq+strideq*2], m0
- mova [dstq+stride3q ], m0
- lea dstq, [dstq+strideq*4]
- dec lines4d
- jnz .loop
-
- RESTORE_GOT
- REP_RET
-
-INIT_XMM sse2
-cglobal dc_left_predictor_16x16, 4, 5, 3, dst, stride, above, left, goffset
- GET_GOT goffsetq
-
- pxor m1, m1
- pxor m2, m2
- mova m0, [leftq]
- DEFINE_ARGS dst, stride, stride3, lines4
- lea stride3q, [strideq*3]
- mov lines4d, 4
- psadbw m0, m1
- psadbw m2, m1
- paddw m0, m2
- movhlps m2, m0
- paddw m0, m2
- paddw m0, [GLOBAL(pw2_16)]
- psraw m0, 4
- pshuflw m0, m0, 0x0
- punpcklqdq m0, m0
- packuswb m0, m0
-.loop:
- mova [dstq ], m0
- mova [dstq+strideq ], m0
- mova [dstq+strideq*2], m0
- mova [dstq+stride3q ], m0
- lea dstq, [dstq+strideq*4]
- dec lines4d
- jnz .loop
-
- RESTORE_GOT
- REP_RET
-
-INIT_XMM sse2
-cglobal dc_128_predictor_16x16, 4, 5, 3, dst, stride, above, left, goffset
- GET_GOT goffsetq
-
- DEFINE_ARGS dst, stride, stride3, lines4
- lea stride3q, [strideq*3]
- mov lines4d, 4
- mova m0, [GLOBAL(dc_128)]
-.loop:
- mova [dstq ], m0
- mova [dstq+strideq ], m0
- mova [dstq+strideq*2], m0
- mova [dstq+stride3q ], m0
- lea dstq, [dstq+strideq*4]
- dec lines4d
- jnz .loop
- RESTORE_GOT
- RET
-
-
-INIT_XMM sse2
-cglobal dc_predictor_32x32, 4, 5, 5, dst, stride, above, left, goffset
- GET_GOT goffsetq
-
- pxor m1, m1
- mova m0, [aboveq]
- mova m2, [aboveq+16]
- mova m3, [leftq]
- mova m4, [leftq+16]
- DEFINE_ARGS dst, stride, stride3, lines4
- lea stride3q, [strideq*3]
- mov lines4d, 8
- psadbw m0, m1
- psadbw m2, m1
- psadbw m3, m1
- psadbw m4, m1
- paddw m0, m2
- paddw m0, m3
- paddw m0, m4
- movhlps m2, m0
- paddw m0, m2
- paddw m0, [GLOBAL(pw_32)]
- psraw m0, 6
- pshuflw m0, m0, 0x0
- punpcklqdq m0, m0
- packuswb m0, m0
-.loop:
- mova [dstq ], m0
- mova [dstq +16], m0
- mova [dstq+strideq ], m0
- mova [dstq+strideq +16], m0
- mova [dstq+strideq*2 ], m0
- mova [dstq+strideq*2+16], m0
- mova [dstq+stride3q ], m0
- mova [dstq+stride3q +16], m0
- lea dstq, [dstq+strideq*4]
- dec lines4d
- jnz .loop
-
- RESTORE_GOT
- REP_RET
-
-INIT_XMM sse2
-cglobal dc_top_predictor_32x32, 4, 5, 5, dst, stride, above, left, goffset
- GET_GOT goffsetq
-
- pxor m1, m1
- mova m0, [aboveq]
- mova m2, [aboveq+16]
- DEFINE_ARGS dst, stride, stride3, lines4
- lea stride3q, [strideq*3]
- mov lines4d, 8
- psadbw m0, m1
- psadbw m2, m1
- paddw m0, m2
- movhlps m2, m0
- paddw m0, m2
- paddw m0, [GLOBAL(pw2_32)]
- psraw m0, 5
- pshuflw m0, m0, 0x0
- punpcklqdq m0, m0
- packuswb m0, m0
-.loop:
- mova [dstq ], m0
- mova [dstq +16], m0
- mova [dstq+strideq ], m0
- mova [dstq+strideq +16], m0
- mova [dstq+strideq*2 ], m0
- mova [dstq+strideq*2+16], m0
- mova [dstq+stride3q ], m0
- mova [dstq+stride3q +16], m0
- lea dstq, [dstq+strideq*4]
- dec lines4d
- jnz .loop
-
- RESTORE_GOT
- REP_RET
-
-INIT_XMM sse2
-cglobal dc_left_predictor_32x32, 4, 5, 5, dst, stride, above, left, goffset
- GET_GOT goffsetq
-
- pxor m1, m1
- mova m0, [leftq]
- mova m2, [leftq+16]
- DEFINE_ARGS dst, stride, stride3, lines4
- lea stride3q, [strideq*3]
- mov lines4d, 8
- psadbw m0, m1
- psadbw m2, m1
- paddw m0, m2
- movhlps m2, m0
- paddw m0, m2
- paddw m0, [GLOBAL(pw2_32)]
- psraw m0, 5
- pshuflw m0, m0, 0x0
- punpcklqdq m0, m0
- packuswb m0, m0
-.loop:
- mova [dstq ], m0
- mova [dstq +16], m0
- mova [dstq+strideq ], m0
- mova [dstq+strideq +16], m0
- mova [dstq+strideq*2 ], m0
- mova [dstq+strideq*2+16], m0
- mova [dstq+stride3q ], m0
- mova [dstq+stride3q +16], m0
- lea dstq, [dstq+strideq*4]
- dec lines4d
- jnz .loop
-
- RESTORE_GOT
- REP_RET
-
-INIT_XMM sse2
-cglobal dc_128_predictor_32x32, 4, 5, 3, dst, stride, above, left, goffset
- GET_GOT goffsetq
-
- DEFINE_ARGS dst, stride, stride3, lines4
- lea stride3q, [strideq*3]
- mov lines4d, 8
- mova m0, [GLOBAL(dc_128)]
-.loop:
- mova [dstq ], m0
- mova [dstq +16], m0
- mova [dstq+strideq ], m0
- mova [dstq+strideq +16], m0
- mova [dstq+strideq*2 ], m0
- mova [dstq+strideq*2+16], m0
- mova [dstq+stride3q ], m0
- mova [dstq+stride3q +16], m0
- lea dstq, [dstq+strideq*4]
- dec lines4d
- jnz .loop
- RESTORE_GOT
- RET
-
-INIT_MMX sse
-cglobal v_predictor_4x4, 3, 3, 1, dst, stride, above
- movd m0, [aboveq]
- movd [dstq ], m0
- movd [dstq+strideq], m0
- lea dstq, [dstq+strideq*2]
- movd [dstq ], m0
- movd [dstq+strideq], m0
- RET
-
-INIT_MMX sse
-cglobal v_predictor_8x8, 3, 3, 1, dst, stride, above
- movq m0, [aboveq]
- DEFINE_ARGS dst, stride, stride3
- lea stride3q, [strideq*3]
- movq [dstq ], m0
- movq [dstq+strideq ], m0
- movq [dstq+strideq*2], m0
- movq [dstq+stride3q ], m0
- lea dstq, [dstq+strideq*4]
- movq [dstq ], m0
- movq [dstq+strideq ], m0
- movq [dstq+strideq*2], m0
- movq [dstq+stride3q ], m0
- RET
-
-INIT_XMM sse2
-cglobal v_predictor_16x16, 3, 4, 1, dst, stride, above
- mova m0, [aboveq]
- DEFINE_ARGS dst, stride, stride3, nlines4
- lea stride3q, [strideq*3]
- mov nlines4d, 4
-.loop:
- mova [dstq ], m0
- mova [dstq+strideq ], m0
- mova [dstq+strideq*2], m0
- mova [dstq+stride3q ], m0
- lea dstq, [dstq+strideq*4]
- dec nlines4d
- jnz .loop
- REP_RET
-
-INIT_XMM sse2
-cglobal v_predictor_32x32, 3, 4, 2, dst, stride, above
- mova m0, [aboveq]
- mova m1, [aboveq+16]
- DEFINE_ARGS dst, stride, stride3, nlines4
- lea stride3q, [strideq*3]
- mov nlines4d, 8
-.loop:
- mova [dstq ], m0
- mova [dstq +16], m1
- mova [dstq+strideq ], m0
- mova [dstq+strideq +16], m1
- mova [dstq+strideq*2 ], m0
- mova [dstq+strideq*2+16], m1
- mova [dstq+stride3q ], m0
- mova [dstq+stride3q +16], m1
- lea dstq, [dstq+strideq*4]
- dec nlines4d
- jnz .loop
- REP_RET
-
-INIT_MMX sse
-cglobal tm_predictor_4x4, 4, 4, 4, dst, stride, above, left
- pxor m1, m1
- movd m2, [aboveq-1]
- movd m0, [aboveq]
- punpcklbw m2, m1
- punpcklbw m0, m1
- pshufw m2, m2, 0x0
- DEFINE_ARGS dst, stride, line, left
- mov lineq, -2
- add leftq, 4
- psubw m0, m2
-.loop:
- movd m2, [leftq+lineq*2]
- movd m3, [leftq+lineq*2+1]
- punpcklbw m2, m1
- punpcklbw m3, m1
- pshufw m2, m2, 0x0
- pshufw m3, m3, 0x0
- paddw m2, m0
- paddw m3, m0
- packuswb m2, m2
- packuswb m3, m3
- movd [dstq ], m2
- movd [dstq+strideq], m3
- lea dstq, [dstq+strideq*2]
- inc lineq
- jnz .loop
- REP_RET
-
-INIT_XMM sse2
-cglobal tm_predictor_8x8, 4, 4, 4, dst, stride, above, left
- pxor m1, m1
- movd m2, [aboveq-1]
- movq m0, [aboveq]
- punpcklbw m2, m1
- punpcklbw m0, m1
- pshuflw m2, m2, 0x0
- DEFINE_ARGS dst, stride, line, left
- mov lineq, -4
- punpcklqdq m2, m2
- add leftq, 8
- psubw m0, m2
-.loop:
- movd m2, [leftq+lineq*2]
- movd m3, [leftq+lineq*2+1]
- punpcklbw m2, m1
- punpcklbw m3, m1
- pshuflw m2, m2, 0x0
- pshuflw m3, m3, 0x0
- punpcklqdq m2, m2
- punpcklqdq m3, m3
- paddw m2, m0
- paddw m3, m0
- packuswb m2, m3
- movq [dstq ], m2
- movhps [dstq+strideq], m2
- lea dstq, [dstq+strideq*2]
- inc lineq
- jnz .loop
- REP_RET
-
-INIT_XMM sse2
-cglobal tm_predictor_16x16, 4, 4, 7, dst, stride, above, left
- pxor m1, m1
- movd m2, [aboveq-1]
- mova m0, [aboveq]
- punpcklbw m2, m1
- punpckhbw m4, m0, m1
- punpcklbw m0, m1
- pshuflw m2, m2, 0x0
- DEFINE_ARGS dst, stride, line, left
- mov lineq, -8
- punpcklqdq m2, m2
- add leftq, 16
- psubw m0, m2
- psubw m4, m2
-.loop:
- movd m2, [leftq+lineq*2]
- movd m3, [leftq+lineq*2+1]
- punpcklbw m2, m1
- punpcklbw m3, m1
- pshuflw m2, m2, 0x0
- pshuflw m3, m3, 0x0
- punpcklqdq m2, m2
- punpcklqdq m3, m3
- paddw m5, m2, m0
- paddw m6, m3, m0
- paddw m2, m4
- paddw m3, m4
- packuswb m5, m2
- packuswb m6, m3
- mova [dstq ], m5
- mova [dstq+strideq], m6
- lea dstq, [dstq+strideq*2]
- inc lineq
- jnz .loop
- REP_RET
-
-%if ARCH_X86_64
-INIT_XMM sse2
-cglobal tm_predictor_32x32, 4, 4, 10, dst, stride, above, left
- pxor m1, m1
- movd m2, [aboveq-1]
- mova m0, [aboveq]
- mova m4, [aboveq+16]
- punpcklbw m2, m1
- punpckhbw m3, m0, m1
- punpckhbw m5, m4, m1
- punpcklbw m0, m1
- punpcklbw m4, m1
- pshuflw m2, m2, 0x0
- DEFINE_ARGS dst, stride, line, left
- mov lineq, -16
- punpcklqdq m2, m2
- add leftq, 32
- psubw m0, m2
- psubw m3, m2
- psubw m4, m2
- psubw m5, m2
-.loop:
- movd m2, [leftq+lineq*2]
- movd m6, [leftq+lineq*2+1]
- punpcklbw m2, m1
- punpcklbw m6, m1
- pshuflw m2, m2, 0x0
- pshuflw m6, m6, 0x0
- punpcklqdq m2, m2
- punpcklqdq m6, m6
- paddw m7, m2, m0
- paddw m8, m2, m3
- paddw m9, m2, m4
- paddw m2, m5
- packuswb m7, m8
- packuswb m9, m2
- paddw m2, m6, m0
- paddw m8, m6, m3
- mova [dstq ], m7
- paddw m7, m6, m4
- paddw m6, m5
- mova [dstq +16], m9
- packuswb m2, m8
- packuswb m7, m6
- mova [dstq+strideq ], m2
- mova [dstq+strideq+16], m7
- lea dstq, [dstq+strideq*2]
- inc lineq
- jnz .loop
- REP_RET
-%endif
--- a/vp9/common/x86/vp9_intrapred_ssse3.asm
+++ /dev/null
@@ -1,1036 +1,0 @@
-;
-; Copyright (c) 2010 The WebM project authors. All Rights Reserved.
-;
-; Use of this source code is governed by a BSD-style license
-; that can be found in the LICENSE file in the root of the source
-; tree. An additional intellectual property rights grant can be found
-; in the file PATENTS. All contributing project authors may
-; be found in the AUTHORS file in the root of the source tree.
-;
-
-%include "third_party/x86inc/x86inc.asm"
-
-SECTION_RODATA
-
-pb_1: times 16 db 1
-sh_b01234577: db 0, 1, 2, 3, 4, 5, 7, 7, 0, 0, 0, 0, 0, 0, 0, 0
-sh_b12345677: db 1, 2, 3, 4, 5, 6, 7, 7, 0, 0, 0, 0, 0, 0, 0, 0
-sh_b23456777: db 2, 3, 4, 5, 6, 7, 7, 7, 0, 0, 0, 0, 0, 0, 0, 0
-sh_b0123456777777777: db 0, 1, 2, 3, 4, 5, 6, 7, 7, 7, 7, 7, 7, 7, 7, 7
-sh_b1234567777777777: db 1, 2, 3, 4, 5, 6, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7
-sh_b2345677777777777: db 2, 3, 4, 5, 6, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7
-sh_b123456789abcdeff: db 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 15
-sh_b23456789abcdefff: db 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 15, 15
-sh_b32104567: db 3, 2, 1, 0, 4, 5, 6, 7, 0, 0, 0, 0, 0, 0, 0, 0
-sh_b8091a2b345: db 8, 0, 9, 1, 10, 2, 11, 3, 4, 5, 0, 0, 0, 0, 0, 0
-sh_b76543210: db 7, 6, 5, 4, 3, 2, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0
-sh_b65432108: db 6, 5, 4, 3, 2, 1, 0, 8, 0, 0, 0, 0, 0, 0, 0, 0
-sh_b54321089: db 5, 4, 3, 2, 1, 0, 8, 9, 0, 0, 0, 0, 0, 0, 0, 0
-sh_b89abcdef: db 8, 9, 10, 11, 12, 13, 14, 15, 0, 0, 0, 0, 0, 0, 0, 0
-sh_bfedcba9876543210: db 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0
-sh_b1233: db 1, 2, 3, 3, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
-sh_b2333: db 2, 3, 3, 3, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
-
-SECTION .text
-
-INIT_MMX ssse3
-cglobal h_predictor_4x4, 2, 4, 3, dst, stride, line, left
- movifnidn leftq, leftmp
- add leftq, 4
- mov lineq, -2
- pxor m0, m0
-.loop:
- movd m1, [leftq+lineq*2 ]
- movd m2, [leftq+lineq*2+1]
- pshufb m1, m0
- pshufb m2, m0
- movd [dstq ], m1
- movd [dstq+strideq], m2
- lea dstq, [dstq+strideq*2]
- inc lineq
- jnz .loop
- REP_RET
-
-INIT_MMX ssse3
-cglobal h_predictor_8x8, 2, 4, 3, dst, stride, line, left
- movifnidn leftq, leftmp
- add leftq, 8
- mov lineq, -4
- pxor m0, m0
-.loop:
- movd m1, [leftq+lineq*2 ]
- movd m2, [leftq+lineq*2+1]
- pshufb m1, m0
- pshufb m2, m0
- movq [dstq ], m1
- movq [dstq+strideq], m2
- lea dstq, [dstq+strideq*2]
- inc lineq
- jnz .loop
- REP_RET
-
-INIT_XMM ssse3
-cglobal h_predictor_16x16, 2, 4, 3, dst, stride, line, left
- movifnidn leftq, leftmp
- add leftq, 16
- mov lineq, -8
- pxor m0, m0
-.loop:
- movd m1, [leftq+lineq*2 ]
- movd m2, [leftq+lineq*2+1]
- pshufb m1, m0
- pshufb m2, m0
- mova [dstq ], m1
- mova [dstq+strideq], m2
- lea dstq, [dstq+strideq*2]
- inc lineq
- jnz .loop
- REP_RET
-
-INIT_XMM ssse3
-cglobal h_predictor_32x32, 2, 4, 3, dst, stride, line, left
- movifnidn leftq, leftmp
- add leftq, 32
- mov lineq, -16
- pxor m0, m0
-.loop:
- movd m1, [leftq+lineq*2 ]
- movd m2, [leftq+lineq*2+1]
- pshufb m1, m0
- pshufb m2, m0
- mova [dstq ], m1
- mova [dstq +16], m1
- mova [dstq+strideq ], m2
- mova [dstq+strideq+16], m2
- lea dstq, [dstq+strideq*2]
- inc lineq
- jnz .loop
- REP_RET
-
-INIT_MMX ssse3
-cglobal d45_predictor_4x4, 3, 4, 4, dst, stride, above, goffset
- GET_GOT goffsetq
-
- movq m0, [aboveq]
- pshufb m2, m0, [GLOBAL(sh_b23456777)]
- pshufb m1, m0, [GLOBAL(sh_b01234577)]
- pshufb m0, [GLOBAL(sh_b12345677)]
- pavgb m3, m2, m1
- pxor m2, m1
- pand m2, [GLOBAL(pb_1)]
- psubb m3, m2
- pavgb m0, m3
-
- ; store 4 lines
- movd [dstq ], m0
- psrlq m0, 8
- movd [dstq+strideq], m0
- lea dstq, [dstq+strideq*2]
- psrlq m0, 8
- movd [dstq ], m0
- psrlq m0, 8
- movd [dstq+strideq], m0
-
- RESTORE_GOT
- RET
-
-INIT_MMX ssse3
-cglobal d45_predictor_8x8, 3, 4, 4, dst, stride, above, goffset
- GET_GOT goffsetq
-
- movq m0, [aboveq]
- mova m1, [GLOBAL(sh_b12345677)]
- DEFINE_ARGS dst, stride, stride3
- lea stride3q, [strideq*3]
- pshufb m2, m0, [GLOBAL(sh_b23456777)]
- pavgb m3, m2, m0
- pxor m2, m0
- pshufb m0, m1
- pand m2, [GLOBAL(pb_1)]
- psubb m3, m2
- pavgb m0, m3
-
- ; store 4 lines
- movq [dstq ], m0
- pshufb m0, m1
- movq [dstq+strideq ], m0
- pshufb m0, m1
- movq [dstq+strideq*2], m0
- pshufb m0, m1
- movq [dstq+stride3q ], m0
- pshufb m0, m1
- lea dstq, [dstq+strideq*4]
-
- ; store next 4 lines
- movq [dstq ], m0
- pshufb m0, m1
- movq [dstq+strideq ], m0
- pshufb m0, m1
- movq [dstq+strideq*2], m0
- pshufb m0, m1
- movq [dstq+stride3q ], m0
-
- RESTORE_GOT
- RET
-
-INIT_XMM ssse3
-cglobal d45_predictor_16x16, 3, 6, 4, dst, stride, above, dst8, line, goffset
- GET_GOT goffsetq
-
- mova m0, [aboveq]
- DEFINE_ARGS dst, stride, stride3, dst8, line
- lea stride3q, [strideq*3]
- lea dst8q, [dstq+strideq*8]
- mova m1, [GLOBAL(sh_b123456789abcdeff)]
- pshufb m2, m0, [GLOBAL(sh_b23456789abcdefff)]
- pavgb m3, m2, m0
- pxor m2, m0
- pshufb m0, m1
- pand m2, [GLOBAL(pb_1)]
- psubb m3, m2
- pavgb m0, m3
-
- ; first 4 lines and first half of 3rd 4 lines
- mov lined, 2
-.loop:
- mova [dstq ], m0
- movhps [dst8q ], m0
- pshufb m0, m1
- mova [dstq +strideq ], m0
- movhps [dst8q+strideq ], m0
- pshufb m0, m1
- mova [dstq +strideq*2 ], m0
- movhps [dst8q+strideq*2 ], m0
- pshufb m0, m1
- mova [dstq +stride3q ], m0
- movhps [dst8q+stride3q ], m0
- pshufb m0, m1
- lea dstq, [dstq +strideq*4]
- lea dst8q, [dst8q+strideq*4]
- dec lined
- jnz .loop
-
- ; bottom-right 8x8 block
- movhps [dstq +8], m0
- movhps [dstq+strideq +8], m0
- movhps [dstq+strideq*2+8], m0
- movhps [dstq+stride3q +8], m0
- lea dstq, [dstq+strideq*4]
- movhps [dstq +8], m0
- movhps [dstq+strideq +8], m0
- movhps [dstq+strideq*2+8], m0
- movhps [dstq+stride3q +8], m0
-
- RESTORE_GOT
- RET
-
-INIT_XMM ssse3
-cglobal d45_predictor_32x32, 3, 6, 7, dst, stride, above, dst16, line, goffset
- GET_GOT goffsetq
-
- mova m0, [aboveq]
- mova m4, [aboveq+16]
- DEFINE_ARGS dst, stride, stride3, dst16, line
- lea stride3q, [strideq*3]
- lea dst16q, [dstq +strideq*8]
- lea dst16q, [dst16q+strideq*8]
- mova m1, [GLOBAL(sh_b123456789abcdeff)]
- pshufb m2, m4, [GLOBAL(sh_b23456789abcdefff)]
- pavgb m3, m2, m4
- pxor m2, m4
- palignr m5, m4, m0, 1
- palignr m6, m4, m0, 2
- pshufb m4, m1
- pand m2, [GLOBAL(pb_1)]
- psubb m3, m2
- pavgb m4, m3
- pavgb m3, m0, m6
- pxor m0, m6
- pand m0, [GLOBAL(pb_1)]
- psubb m3, m0
- pavgb m5, m3
-
- ; write 4x4 lines (and the first half of the second 4x4 lines)
- mov lined, 4
-.loop:
- mova [dstq ], m5
- mova [dstq +16], m4
- mova [dst16q ], m4
- palignr m3, m4, m5, 1
- pshufb m4, m1
- mova [dstq +strideq ], m3
- mova [dstq +strideq +16], m4
- mova [dst16q+strideq ], m4
- palignr m5, m4, m3, 1
- pshufb m4, m1
- mova [dstq +strideq*2 ], m5
- mova [dstq +strideq*2+16], m4
- mova [dst16q+strideq*2 ], m4
- palignr m3, m4, m5, 1
- pshufb m4, m1
- mova [dstq +stride3q ], m3
- mova [dstq +stride3q +16], m4
- mova [dst16q+stride3q ], m4
- palignr m5, m4, m3, 1
- pshufb m4, m1
- lea dstq, [dstq +strideq*4]
- lea dst16q, [dst16q+strideq*4]
- dec lined
- jnz .loop
-
- ; write second half of second 4x4 lines
- mova [dstq +16], m4
- mova [dstq +strideq +16], m4
- mova [dstq +strideq*2+16], m4
- mova [dstq +stride3q +16], m4
- lea dstq, [dstq +strideq*4]
- mova [dstq +16], m4
- mova [dstq +strideq +16], m4
- mova [dstq +strideq*2+16], m4
- mova [dstq +stride3q +16], m4
- lea dstq, [dstq +strideq*4]
- mova [dstq +16], m4
- mova [dstq +strideq +16], m4
- mova [dstq +strideq*2+16], m4
- mova [dstq +stride3q +16], m4
- lea dstq, [dstq +strideq*4]
- mova [dstq +16], m4
- mova [dstq +strideq +16], m4
- mova [dstq +strideq*2+16], m4
- mova [dstq +stride3q +16], m4
-
- RESTORE_GOT
- RET
-
-; ------------------------------------------
-; input: x, y, z, result
-;
-; trick from pascal
-; (x+2y+z+2)>>2 can be calculated as:
-; result = avg(x,z)
-; result -= xor(x,z) & 1
-; result = avg(result,y)
-; ------------------------------------------
-%macro X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 4
- pavgb %4, %1, %3
- pxor %3, %1
- pand %3, [GLOBAL(pb_1)]
- psubb %4, %3
- pavgb %4, %2
-%endmacro
-
-INIT_XMM ssse3
-cglobal d63_predictor_4x4, 3, 4, 5, dst, stride, above, goffset
- GET_GOT goffsetq
-
- movq m3, [aboveq]
- pshufb m1, m3, [GLOBAL(sh_b23456777)]
- pshufb m2, m3, [GLOBAL(sh_b12345677)]
-
- X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m3, m2, m1, m4
- pavgb m3, m2
-
- ; store 4 lines
- movd [dstq ], m3
- movd [dstq+strideq], m4
- lea dstq, [dstq+strideq*2]
- psrldq m3, 1
- psrldq m4, 1
- movd [dstq ], m3
- movd [dstq+strideq], m4
- RESTORE_GOT
- RET
-
-INIT_XMM ssse3
-cglobal d63_predictor_8x8, 3, 4, 5, dst, stride, above, goffset
- GET_GOT goffsetq
-
- movq m3, [aboveq]
- DEFINE_ARGS dst, stride, stride3
- lea stride3q, [strideq*3]
- pshufb m1, m3, [GLOBAL(sh_b2345677777777777)]
- pshufb m0, m3, [GLOBAL(sh_b0123456777777777)]
- pshufb m2, m3, [GLOBAL(sh_b1234567777777777)]
- pshufb m3, [GLOBAL(sh_b0123456777777777)]
-
- X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m0, m2, m1, m4
- pavgb m3, m2
-
- ; store 4 lines
- movq [dstq ], m3
- movq [dstq+strideq], m4
- psrldq m3, 1
- psrldq m4, 1
- movq [dstq+strideq*2], m3
- movq [dstq+stride3q ], m4
- lea dstq, [dstq+strideq*4]
- psrldq m3, 1
- psrldq m4, 1
-
- ; store 4 lines
- movq [dstq ], m3
- movq [dstq+strideq], m4
- psrldq m3, 1
- psrldq m4, 1
- movq [dstq+strideq*2], m3
- movq [dstq+stride3q ], m4
- RESTORE_GOT
- RET
-
-INIT_XMM ssse3
-cglobal d63_predictor_16x16, 3, 5, 5, dst, stride, above, line, goffset
- GET_GOT goffsetq
-
- mova m0, [aboveq]
- DEFINE_ARGS dst, stride, stride3, line
- lea stride3q, [strideq*3]
- mova m1, [GLOBAL(sh_b123456789abcdeff)]
- pshufb m2, m0, [GLOBAL(sh_b23456789abcdefff)]
- pshufb m3, m0, m1
-
- X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m0, m3, m2, m4
- pavgb m0, m3
-
- mov lined, 4
-.loop:
- mova [dstq ], m0
- mova [dstq+strideq ], m4
- pshufb m0, m1
- pshufb m4, m1
- mova [dstq+strideq*2], m0
- mova [dstq+stride3q ], m4
- pshufb m0, m1
- pshufb m4, m1
- lea dstq, [dstq+strideq*4]
- dec lined
- jnz .loop
- RESTORE_GOT
- REP_RET
-
-INIT_XMM ssse3
-cglobal d63_predictor_32x32, 3, 5, 8, dst, stride, above, line, goffset
- GET_GOT goffsetq
-
- mova m0, [aboveq]
- mova m7, [aboveq+16]
- DEFINE_ARGS dst, stride, stride3, line
- mova m1, [GLOBAL(sh_b123456789abcdeff)]
- lea stride3q, [strideq*3]
- pshufb m2, m7, [GLOBAL(sh_b23456789abcdefff)]
- pshufb m3, m7, m1
-
- X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m7, m3, m2, m4
- palignr m6, m7, m0, 1
- palignr m5, m7, m0, 2
- pavgb m7, m3
-
- X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m0, m6, m5, m2
- pavgb m0, m6
-
- mov lined, 8
-.loop:
- mova [dstq ], m0
- mova [dstq +16], m7
- mova [dstq+strideq ], m2
- mova [dstq+strideq +16], m4
- palignr m3, m7, m0, 1
- palignr m5, m4, m2, 1
- pshufb m7, m1
- pshufb m4, m1
-
- mova [dstq+strideq*2 ], m3
- mova [dstq+strideq*2+16], m7
- mova [dstq+stride3q ], m5
- mova [dstq+stride3q +16], m4
- palignr m0, m7, m3, 1
- palignr m2, m4, m5, 1
- pshufb m7, m1
- pshufb m4, m1
- lea dstq, [dstq+strideq*4]
- dec lined
- jnz .loop
- RESTORE_GOT
- REP_RET
-
-INIT_XMM ssse3
-cglobal d153_predictor_4x4, 4, 5, 4, dst, stride, above, left, goffset
- GET_GOT goffsetq
- movd m0, [leftq] ; l1, l2, l3, l4
- movd m1, [aboveq-1] ; tl, t1, t2, t3
- punpckldq m0, m1 ; l1, l2, l3, l4, tl, t1, t2, t3
- pshufb m0, [GLOBAL(sh_b32104567)]; l4, l3, l2, l1, tl, t1, t2, t3
- psrldq m1, m0, 1 ; l3, l2, l1, tl, t1, t2, t3
- psrldq m2, m0, 2 ; l2, l1, tl, t1, t2, t3
- ; comments below are for a predictor like this
- ; A1 B1 C1 D1
- ; A2 B2 A1 B1
- ; A3 B3 A2 B2
- ; A4 B4 A3 B3
- X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m0, m1, m2, m3 ; 3-tap avg B4 B3 B2 B1 C1 D1
- pavgb m1, m0 ; 2-tap avg A4 A3 A2 A1
-
- punpcklqdq m3, m1 ; B4 B3 B2 B1 C1 D1 x x A4 A3 A2 A1 ..
-
- DEFINE_ARGS dst, stride, stride3
- lea stride3q, [strideq*3]
- pshufb m3, [GLOBAL(sh_b8091a2b345)] ; A4 B4 A3 B3 A2 B2 A1 B1 C1 D1 ..
- movd [dstq+stride3q ], m3
- psrldq m3, 2 ; A3 B3 A2 B2 A1 B1 C1 D1 ..
- movd [dstq+strideq*2], m3
- psrldq m3, 2 ; A2 B2 A1 B1 C1 D1 ..
- movd [dstq+strideq ], m3
- psrldq m3, 2 ; A1 B1 C1 D1 ..
- movd [dstq ], m3
- RESTORE_GOT
- RET
-
-INIT_XMM ssse3
-cglobal d153_predictor_8x8, 4, 5, 8, dst, stride, above, left, goffset
- GET_GOT goffsetq
- movq m0, [leftq] ; [0- 7] l1-8 [byte]
- movhps m0, [aboveq-1] ; [8-15] tl, t1-7 [byte]
- pshufb m1, m0, [GLOBAL(sh_b76543210)] ; l8-1 [word]
- pshufb m2, m0, [GLOBAL(sh_b65432108)] ; l7-1,tl [word]
- pshufb m3, m0, [GLOBAL(sh_b54321089)] ; l6-1,tl,t1 [word]
- pshufb m0, [GLOBAL(sh_b89abcdef)] ; tl,t1-7 [word]
- psrldq m4, m0, 1 ; t1-7 [word]
- psrldq m5, m0, 2 ; t2-7 [word]
- ; comments below are for a predictor like this
- ; A1 B1 C1 D1 E1 F1 G1 H1
- ; A2 B2 A1 B1 C1 D1 E1 F1
- ; A3 B3 A2 B2 A1 B1 C1 D1
- ; A4 B4 A3 B3 A2 B2 A1 B1
- ; A5 B5 A4 B4 A3 B3 A2 B2
- ; A6 B6 A5 B5 A4 B4 A3 B3
- ; A7 B7 A6 B6 A5 B5 A4 B4
- ; A8 B8 A7 B7 A6 B6 A5 B5
- pavgb m6, m1, m2 ; 2-tap avg A8-A1
-
- X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m0, m4, m5, m7 ; 3-tap avg C-H1
-
- X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m1, m2, m3, m0 ; 3-tap avg B8-1
-
- punpcklbw m6, m0 ; A-B8, A-B7 ... A-B2, A-B1
-
- DEFINE_ARGS dst, stride, stride3
- lea stride3q, [strideq*3]
-
- movhps [dstq+stride3q], m6 ; A-B4, A-B3, A-B2, A-B1
- palignr m0, m7, m6, 10 ; A-B3, A-B2, A-B1, C-H1
- movq [dstq+strideq*2], m0
- psrldq m0, 2 ; A-B2, A-B1, C-H1
- movq [dstq+strideq ], m0
- psrldq m0, 2 ; A-H1
- movq [dstq ], m0
- lea dstq, [dstq+strideq*4]
- movq [dstq+stride3q ], m6 ; A-B8, A-B7, A-B6, A-B5
- psrldq m6, 2 ; A-B7, A-B6, A-B5, A-B4
- movq [dstq+strideq*2], m6
- psrldq m6, 2 ; A-B6, A-B5, A-B4, A-B3
- movq [dstq+strideq ], m6
- psrldq m6, 2 ; A-B5, A-B4, A-B3, A-B2
- movq [dstq ], m6
- RESTORE_GOT
- RET
-
-INIT_XMM ssse3
-cglobal d153_predictor_16x16, 4, 5, 8, dst, stride, above, left, goffset
- GET_GOT goffsetq
- mova m0, [leftq]
- movu m7, [aboveq-1]
- ; comments below are for a predictor like this
- ; A1 B1 C1 D1 E1 F1 G1 H1 I1 J1 K1 L1 M1 N1 O1 P1
- ; A2 B2 A1 B1 C1 D1 E1 F1 G1 H1 I1 J1 K1 L1 M1 N1
- ; A3 B3 A2 B2 A1 B1 C1 D1 E1 F1 G1 H1 I1 J1 K1 L1
- ; A4 B4 A3 B3 A2 B2 A1 B1 C1 D1 E1 F1 G1 H1 I1 J1
- ; A5 B5 A4 B4 A3 B3 A2 B2 A1 B1 C1 D1 E1 F1 G1 H1
- ; A6 B6 A5 B5 A4 B4 A3 B3 A2 B2 A1 B1 C1 D1 E1 F1
- ; A7 B7 A6 B6 A5 B5 A4 B4 A3 B3 A2 B2 A1 B1 C1 D1
- ; A8 B8 A7 B7 A6 B6 A5 B5 A4 B4 A3 B3 A2 B2 A1 B1
- ; A9 B9 A8 B8 A7 B7 A6 B6 A5 B5 A4 B4 A3 B3 A2 B2
- ; Aa Ba A9 B9 A8 B8 A7 B7 A6 B6 A5 B5 A4 B4 A3 B3
- ; Ab Bb Aa Ba A9 B9 A8 B8 A7 B7 A6 B6 A5 B5 A4 B4
- ; Ac Bc Ab Bb Aa Ba A9 B9 A8 B8 A7 B7 A6 B6 A5 B5
- ; Ad Bd Ac Bc Ab Bb Aa Ba A9 B9 A8 B8 A7 B7 A6 B6
- ; Ae Be Ad Bd Ac Bc Ab Bb Aa Ba A9 B9 A8 B8 A7 B7
- ; Af Bf Ae Be Ad Bd Ac Bc Ab Bb Aa Ba A9 B9 A8 B8
- ; Ag Bg Af Bf Ae Be Ad Bd Ac Bc Ab Bb Aa Ba A9 B9
- pshufb m6, m7, [GLOBAL(sh_bfedcba9876543210)]
- palignr m5, m0, m6, 15
- palignr m3, m0, m6, 14
-
- X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m0, m5, m3, m4 ; 3-tap avg B3-Bg
- pshufb m1, m0, [GLOBAL(sh_b123456789abcdeff)]
- pavgb m5, m0 ; A1 - Ag
-
- punpcklbw m0, m4, m5 ; A-B8 ... A-B1
- punpckhbw m4, m5 ; A-B9 ... A-Bg
-
- pshufb m3, m7, [GLOBAL(sh_b123456789abcdeff)]
- pshufb m5, m7, [GLOBAL(sh_b23456789abcdefff)]
-
- X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m7, m3, m5, m1 ; 3-tap avg C1-P1
-
- pshufb m6, m0, [GLOBAL(sh_bfedcba9876543210)]
- DEFINE_ARGS dst, stride, stride3
- lea stride3q, [strideq*3]
- palignr m2, m1, m6, 14
- mova [dstq ], m2
- palignr m2, m1, m6, 12
- mova [dstq+strideq ], m2
- palignr m2, m1, m6, 10
- mova [dstq+strideq*2], m2
- palignr m2, m1, m6, 8
- mova [dstq+stride3q ], m2
- lea dstq, [dstq+strideq*4]
- palignr m2, m1, m6, 6
- mova [dstq ], m2
- palignr m2, m1, m6, 4
- mova [dstq+strideq ], m2
- palignr m2, m1, m6, 2
- mova [dstq+strideq*2], m2
- pshufb m4, [GLOBAL(sh_bfedcba9876543210)]
- mova [dstq+stride3q ], m6
- lea dstq, [dstq+strideq*4]
-
- palignr m2, m6, m4, 14
- mova [dstq ], m2
- palignr m2, m6, m4, 12
- mova [dstq+strideq ], m2
- palignr m2, m6, m4, 10
- mova [dstq+strideq*2], m2
- palignr m2, m6, m4, 8
- mova [dstq+stride3q ], m2
- lea dstq, [dstq+strideq*4]
- palignr m2, m6, m4, 6
- mova [dstq ], m2
- palignr m2, m6, m4, 4
- mova [dstq+strideq ], m2
- palignr m2, m6, m4, 2
- mova [dstq+strideq*2], m2
- mova [dstq+stride3q ], m4
- RESTORE_GOT
- RET
-
-INIT_XMM ssse3
-cglobal d153_predictor_32x32, 4, 5, 8, dst, stride, above, left, goffset
- GET_GOT goffsetq
- mova m0, [leftq]
- movu m7, [aboveq-1]
- movu m1, [aboveq+15]
-
- pshufb m4, m1, [GLOBAL(sh_b123456789abcdeff)]
- pshufb m6, m1, [GLOBAL(sh_b23456789abcdefff)]
-
- X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m1, m4, m6, m2 ; 3-tap avg above [high]
-
- palignr m3, m1, m7, 1
- palignr m5, m1, m7, 2
-
- X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m7, m3, m5, m1 ; 3-tap avg above [low]
-
- pshufb m7, [GLOBAL(sh_bfedcba9876543210)]
- palignr m5, m0, m7, 15
- palignr m3, m0, m7, 14
-
- X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m0, m5, m3, m4 ; 3-tap avg B3-Bg
- pavgb m5, m0 ; A1 - Ag
- punpcklbw m6, m4, m5 ; A-B8 ... A-B1
- punpckhbw m4, m5 ; A-B9 ... A-Bg
- pshufb m6, [GLOBAL(sh_bfedcba9876543210)]
- pshufb m4, [GLOBAL(sh_bfedcba9876543210)]
-
- DEFINE_ARGS dst, stride, stride3, left, line
- lea stride3q, [strideq*3]
-
- palignr m5, m2, m1, 14
- palignr m7, m1, m6, 14
- mova [dstq ], m7
- mova [dstq+16 ], m5
- palignr m5, m2, m1, 12
- palignr m7, m1, m6, 12
- mova [dstq+strideq ], m7
- mova [dstq+strideq+16 ], m5
- palignr m5, m2, m1, 10
- palignr m7, m1, m6, 10
- mova [dstq+strideq*2 ], m7
- mova [dstq+strideq*2+16], m5
- palignr m5, m2, m1, 8
- palignr m7, m1, m6, 8
- mova [dstq+stride3q ], m7
- mova [dstq+stride3q+16 ], m5
- lea dstq, [dstq+strideq*4]
- palignr m5, m2, m1, 6
- palignr m7, m1, m6, 6
- mova [dstq ], m7
- mova [dstq+16 ], m5
- palignr m5, m2, m1, 4
- palignr m7, m1, m6, 4
- mova [dstq+strideq ], m7
- mova [dstq+strideq+16 ], m5
- palignr m5, m2, m1, 2
- palignr m7, m1, m6, 2
- mova [dstq+strideq*2 ], m7
- mova [dstq+strideq*2+16], m5
- mova [dstq+stride3q ], m6
- mova [dstq+stride3q+16 ], m1
- lea dstq, [dstq+strideq*4]
-
- palignr m5, m1, m6, 14
- palignr m3, m6, m4, 14
- mova [dstq ], m3
- mova [dstq+16 ], m5
- palignr m5, m1, m6, 12
- palignr m3, m6, m4, 12
- mova [dstq+strideq ], m3
- mova [dstq+strideq+16 ], m5
- palignr m5, m1, m6, 10
- palignr m3, m6, m4, 10
- mova [dstq+strideq*2 ], m3
- mova [dstq+strideq*2+16], m5
- palignr m5, m1, m6, 8
- palignr m3, m6, m4, 8
- mova [dstq+stride3q ], m3
- mova [dstq+stride3q+16 ], m5
- lea dstq, [dstq+strideq*4]
- palignr m5, m1, m6, 6
- palignr m3, m6, m4, 6
- mova [dstq ], m3
- mova [dstq+16 ], m5
- palignr m5, m1, m6, 4
- palignr m3, m6, m4, 4
- mova [dstq+strideq ], m3
- mova [dstq+strideq+16 ], m5
- palignr m5, m1, m6, 2
- palignr m3, m6, m4, 2
- mova [dstq+strideq*2 ], m3
- mova [dstq+strideq*2+16], m5
- mova [dstq+stride3q ], m4
- mova [dstq+stride3q+16 ], m6
- lea dstq, [dstq+strideq*4]
-
- mova m7, [leftq]
- mova m3, [leftq+16]
- palignr m5, m3, m7, 15
- palignr m0, m3, m7, 14
-
- X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m3, m5, m0, m2 ; 3-tap avg Bh -
- pavgb m5, m3 ; Ah -
- punpcklbw m3, m2, m5 ; A-B8 ... A-B1
- punpckhbw m2, m5 ; A-B9 ... A-Bg
- pshufb m3, [GLOBAL(sh_bfedcba9876543210)]
- pshufb m2, [GLOBAL(sh_bfedcba9876543210)]
-
- palignr m7, m6, m4, 14
- palignr m0, m4, m3, 14
- mova [dstq ], m0
- mova [dstq+16 ], m7
- palignr m7, m6, m4, 12
- palignr m0, m4, m3, 12
- mova [dstq+strideq ], m0
- mova [dstq+strideq+16 ], m7
- palignr m7, m6, m4, 10
- palignr m0, m4, m3, 10
- mova [dstq+strideq*2 ], m0
- mova [dstq+strideq*2+16], m7
- palignr m7, m6, m4, 8
- palignr m0, m4, m3, 8
- mova [dstq+stride3q ], m0
- mova [dstq+stride3q+16 ], m7
- lea dstq, [dstq+strideq*4]
- palignr m7, m6, m4, 6
- palignr m0, m4, m3, 6
- mova [dstq ], m0
- mova [dstq+16 ], m7
- palignr m7, m6, m4, 4
- palignr m0, m4, m3, 4
- mova [dstq+strideq ], m0
- mova [dstq+strideq+16 ], m7
- palignr m7, m6, m4, 2
- palignr m0, m4, m3, 2
- mova [dstq+strideq*2 ], m0
- mova [dstq+strideq*2+16], m7
- mova [dstq+stride3q ], m3
- mova [dstq+stride3q+16 ], m4
- lea dstq, [dstq+strideq*4]
-
- palignr m7, m4, m3, 14
- palignr m0, m3, m2, 14
- mova [dstq ], m0
- mova [dstq+16 ], m7
- palignr m7, m4, m3, 12
- palignr m0, m3, m2, 12
- mova [dstq+strideq ], m0
- mova [dstq+strideq+16 ], m7
- palignr m7, m4, m3, 10
- palignr m0, m3, m2, 10
- mova [dstq+strideq*2 ], m0
- mova [dstq+strideq*2+16], m7
- palignr m7, m4, m3, 8
- palignr m0, m3, m2, 8
- mova [dstq+stride3q ], m0
- mova [dstq+stride3q+16 ], m7
- lea dstq, [dstq+strideq*4]
- palignr m7, m4, m3, 6
- palignr m0, m3, m2, 6
- mova [dstq ], m0
- mova [dstq+16 ], m7
- palignr m7, m4, m3, 4
- palignr m0, m3, m2, 4
- mova [dstq+strideq ], m0
- mova [dstq+strideq+16 ], m7
- palignr m7, m4, m3, 2
- palignr m0, m3, m2, 2
- mova [dstq+strideq*2 ], m0
- mova [dstq+strideq*2+16], m7
- mova [dstq+stride3q ], m2
- mova [dstq+stride3q+16 ], m3
-
- RESTORE_GOT
- RET
-
-INIT_MMX ssse3
-cglobal d207_predictor_4x4, 4, 5, 4, dst, stride, unused, left, goffset
- GET_GOT goffsetq
- movd m0, [leftq] ; abcd [byte]
- pshufb m1, m0, [GLOBAL(sh_b1233)] ; bcdd [byte]
- pshufb m3, m0, [GLOBAL(sh_b2333)] ; cddd
-
- X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m0, m1, m3, m2
- pavgb m1, m0 ; ab, bc, cd, d [byte]
-
- punpcklbw m1, m2 ; ab, a2bc, bc, b2cd, cd, c3d, d, d
- movd [dstq ], m1
- psrlq m1, 16 ; bc, b2cd, cd, c3d, d, d
- movd [dstq+strideq], m1
- lea dstq, [dstq+strideq*2]
- psrlq m1, 16 ; cd, c3d, d, d
- movd [dstq ], m1
- pshufw m1, m1, q1111 ; d, d, d, d
- movd [dstq+strideq], m1
- RESTORE_GOT
- RET
-
-INIT_XMM ssse3
-cglobal d207_predictor_8x8, 4, 5, 4, dst, stride, stride3, left, goffset
- GET_GOT goffsetq
- movq m3, [leftq] ; abcdefgh [byte]
- lea stride3q, [strideq*3]
-
- pshufb m1, m3, [GLOBAL(sh_b2345677777777777)]
- pshufb m0, m3, [GLOBAL(sh_b0123456777777777)]
- pshufb m2, m3, [GLOBAL(sh_b1234567777777777)]
-
- X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m0, m2, m1, m3
- pavgb m0, m2
- punpcklbw m0, m3 ; interleaved output
-
- movq [dstq ], m0
- psrldq m0, 2
- movq [dstq+strideq ], m0
- psrldq m0, 2
- movq [dstq+strideq*2], m0
- psrldq m0, 2
- movq [dstq+stride3q ], m0
- lea dstq, [dstq+strideq*4]
- pshufhw m0, m0, q0000 ; de, d2ef, ef, e2fg, fg, f2gh, gh, g3h, 8xh
- psrldq m0, 2
- movq [dstq ], m0
- psrldq m0, 2
- movq [dstq+strideq ], m0
- psrldq m0, 2
- movq [dstq+strideq*2], m0
- psrldq m0, 2
- movq [dstq+stride3q ], m0
- RESTORE_GOT
- RET
-
-INIT_XMM ssse3
-cglobal d207_predictor_16x16, 4, 5, 5, dst, stride, stride3, left, goffset
- GET_GOT goffsetq
- lea stride3q, [strideq*3]
- mova m0, [leftq] ; abcdefghijklmnop [byte]
- pshufb m1, m0, [GLOBAL(sh_b123456789abcdeff)] ; bcdefghijklmnopp
- pshufb m2, m0, [GLOBAL(sh_b23456789abcdefff)]
-
- X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m0, m1, m2, m3
- pavgb m1, m0 ; ab, bc, cd .. no, op, pp [byte]
-
- punpckhbw m4, m1, m3 ; interleaved input
- punpcklbw m1, m3 ; interleaved output
- mova [dstq ], m1
- palignr m3, m4, m1, 2
- mova [dstq+strideq ], m3
- palignr m3, m4, m1, 4
- mova [dstq+strideq*2], m3
- palignr m3, m4, m1, 6
- mova [dstq+stride3q ], m3
- lea dstq, [dstq+strideq*4]
- palignr m3, m4, m1, 8
- mova [dstq ], m3
- palignr m3, m4, m1, 10
- mova [dstq+strideq ], m3
- palignr m3, m4, m1, 12
- mova [dstq+strideq*2], m3
- palignr m3, m4, m1, 14
- mova [dstq+stride3q ], m3
- DEFINE_ARGS dst, stride, stride3, line
- mov lined, 2
- mova m0, [GLOBAL(sh_b23456789abcdefff)]
-.loop:
- lea dstq, [dstq+strideq*4]
- mova [dstq ], m4
- pshufb m4, m0
- mova [dstq+strideq ], m4
- pshufb m4, m0
- mova [dstq+strideq*2], m4
- pshufb m4, m0
- mova [dstq+stride3q ], m4
- pshufb m4, m0
- dec lined
- jnz .loop
- RESTORE_GOT
- REP_RET
-
-INIT_XMM ssse3
-cglobal d207_predictor_32x32, 4, 5, 8, dst, stride, stride3, left, goffset
- GET_GOT goffsetq
- lea stride3q, [strideq*3]
- mova m1, [leftq] ; 0-15 [byte]
- mova m2, [leftq+16] ; 16-31 [byte]
- pshufb m0, m2, [GLOBAL(sh_b23456789abcdefff)]
- pshufb m4, m2, [GLOBAL(sh_b123456789abcdeff)]
-
- X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m2, m4, m0, m3
- palignr m6, m2, m1, 1
- palignr m5, m2, m1, 2
- pavgb m2, m4 ; high 16px even lines
-
- X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m1, m6, m5, m0
- pavgb m1, m6 ; low 16px even lines
-
- punpckhbw m6, m1, m0 ; interleaved output 2
- punpcklbw m1, m0 ; interleaved output 1
-
- punpckhbw m7, m2, m3 ; interleaved output 4
- punpcklbw m2, m3 ; interleaved output 3
-
- ; output 1st 8 lines (and half of 2nd 8 lines)
- DEFINE_ARGS dst, stride, stride3, dst8
- lea dst8q, [dstq+strideq*8]
- mova [dstq ], m1
- mova [dstq +16], m6
- mova [dst8q ], m6
- palignr m0, m6, m1, 2
- palignr m4, m2, m6, 2
- mova [dstq +strideq ], m0
- mova [dstq +strideq +16], m4
- mova [dst8q+strideq ], m4
- palignr m0, m6, m1, 4
- palignr m4, m2, m6, 4
- mova [dstq +strideq*2 ], m0
- mova [dstq +strideq*2+16], m4
- mova [dst8q+strideq*2 ], m4
- palignr m0, m6, m1, 6
- palignr m4, m2, m6, 6
- mova [dstq +stride3q ], m0
- mova [dstq +stride3q +16], m4
- mova [dst8q+stride3q ], m4
- lea dstq, [dstq +strideq*4]
- lea dst8q, [dst8q+strideq*4]
- palignr m0, m6, m1, 8
- palignr m4, m2, m6, 8
- mova [dstq ], m0
- mova [dstq +16], m4
- mova [dst8q ], m4
- palignr m0, m6, m1, 10
- palignr m4, m2, m6, 10
- mova [dstq +strideq ], m0
- mova [dstq +strideq +16], m4
- mova [dst8q+strideq ], m4
- palignr m0, m6, m1, 12
- palignr m4, m2, m6, 12
- mova [dstq +strideq*2 ], m0
- mova [dstq +strideq*2+16], m4
- mova [dst8q+strideq*2 ], m4
- palignr m0, m6, m1, 14
- palignr m4, m2, m6, 14
- mova [dstq +stride3q ], m0
- mova [dstq +stride3q +16], m4
- mova [dst8q+stride3q ], m4
- lea dstq, [dstq+strideq*4]
- lea dst8q, [dst8q+strideq*4]
-
- ; output 2nd half of 2nd 8 lines and half of 3rd 8 lines
- mova [dstq +16], m2
- mova [dst8q ], m2
- palignr m4, m7, m2, 2
- mova [dstq +strideq +16], m4
- mova [dst8q+strideq ], m4
- palignr m4, m7, m2, 4
- mova [dstq +strideq*2+16], m4
- mova [dst8q+strideq*2 ], m4
- palignr m4, m7, m2, 6
- mova [dstq +stride3q +16], m4
- mova [dst8q+stride3q ], m4
- lea dstq, [dstq+strideq*4]
- lea dst8q, [dst8q+strideq*4]
- palignr m4, m7, m2, 8
- mova [dstq +16], m4
- mova [dst8q ], m4
- palignr m4, m7, m2, 10
- mova [dstq +strideq +16], m4
- mova [dst8q+strideq ], m4
- palignr m4, m7, m2, 12
- mova [dstq +strideq*2+16], m4
- mova [dst8q+strideq*2 ], m4
- palignr m4, m7, m2, 14
- mova [dstq +stride3q +16], m4
- mova [dst8q+stride3q ], m4
- lea dstq, [dstq+strideq*4]
- lea dst8q, [dst8q+strideq*4]
-
- ; output 2nd half of 3rd 8 lines and half of 4th 8 lines
- mova m0, [GLOBAL(sh_b23456789abcdefff)]
- mova [dstq +16], m7
- mova [dst8q ], m7
- pshufb m7, m0
- mova [dstq +strideq +16], m7
- mova [dst8q+strideq ], m7
- pshufb m7, m0
- mova [dstq +strideq*2+16], m7
- mova [dst8q+strideq*2 ], m7
- pshufb m7, m0
- mova [dstq +stride3q +16], m7
- mova [dst8q+stride3q ], m7
- pshufb m7, m0
- lea dstq, [dstq+strideq*4]
- lea dst8q, [dst8q+strideq*4]
- mova [dstq +16], m7
- mova [dst8q ], m7
- pshufb m7, m0
- mova [dstq +strideq +16], m7
- mova [dst8q+strideq ], m7
- pshufb m7, m0
- mova [dstq +strideq*2+16], m7
- mova [dst8q+strideq*2 ], m7
- pshufb m7, m0
- mova [dstq +stride3q +16], m7
- mova [dst8q+stride3q ], m7
- pshufb m7, m0
- lea dstq, [dstq+strideq*4]
-
- ; output last half of 4th 8 lines
- mova [dstq +16], m7
- mova [dstq +strideq +16], m7
- mova [dstq +strideq*2+16], m7
- mova [dstq +stride3q +16], m7
- lea dstq, [dstq+strideq*4]
- mova [dstq +16], m7
- mova [dstq +strideq +16], m7
- mova [dstq +strideq*2+16], m7
- mova [dstq +stride3q +16], m7
-
- ; done!
- RESTORE_GOT
- RET
--- a/vp9/vp9_common.mk
+++ b/vp9/vp9_common.mk
@@ -84,17 +84,12 @@
ifeq ($(CONFIG_USE_X86INC),yes)
VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_copy_sse2.asm
-VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_intrapred_sse2.asm
-VP9_COMMON_SRCS-$(HAVE_SSSE3) += common/x86/vp9_intrapred_ssse3.asm
endif
ifeq ($(CONFIG_VP9_HIGHBITDEPTH),yes)
VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_high_subpixel_8t_sse2.asm
VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_high_subpixel_bilinear_sse2.asm
-ifeq ($(CONFIG_USE_X86INC),yes)
-VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_high_intrapred_sse2.asm
endif
-endif
# common (c)
VP9_COMMON_SRCS-$(HAVE_DSPR2) += common/mips/dspr2/vp9_common_dspr2.h
@@ -108,9 +103,6 @@
VP9_COMMON_SRCS-$(HAVE_DSPR2) += common/mips/dspr2/vp9_convolve8_dspr2.c
VP9_COMMON_SRCS-$(HAVE_DSPR2) += common/mips/dspr2/vp9_convolve8_horiz_dspr2.c
VP9_COMMON_SRCS-$(HAVE_DSPR2) += common/mips/dspr2/vp9_convolve8_vert_dspr2.c
-VP9_COMMON_SRCS-$(HAVE_DSPR2) += common/mips/dspr2/vp9_intrapred4_dspr2.c
-VP9_COMMON_SRCS-$(HAVE_DSPR2) += common/mips/dspr2/vp9_intrapred8_dspr2.c
-VP9_COMMON_SRCS-$(HAVE_DSPR2) += common/mips/dspr2/vp9_intrapred16_dspr2.c
ifneq ($(CONFIG_VP9_HIGHBITDEPTH),yes)
VP9_COMMON_SRCS-$(HAVE_DSPR2) += common/mips/dspr2/vp9_itrans4_dspr2.c
@@ -135,7 +127,6 @@
VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_idct16x16_msa.c
VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_idct32x32_msa.c
VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_idct_msa.h
-VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_intra_predict_msa.c
ifeq ($(CONFIG_VP9_POSTPROC),yes)
VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_mfqe_msa.c
@@ -174,7 +165,6 @@
VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct4x4_add_neon_asm$(ASM)
VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct8x8_1_add_neon_asm$(ASM)
VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct8x8_add_neon_asm$(ASM)
-VP9_COMMON_SRCS-yes += common/arm/neon/vp9_reconintra_neon_asm$(ASM)
else
ifeq ($(HAVE_NEON), yes)
VP9_COMMON_SRCS-yes += common/arm/neon/vp9_convolve8_avg_neon.c
@@ -193,7 +183,5 @@
VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct8x8_add_neon.c
endif # HAVE_NEON
endif # HAVE_NEON_ASM
-
-VP9_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/vp9_reconintra_neon.c
$(eval $(call rtcd_h_template,vp9_rtcd,vp9/common/vp9_rtcd_defs.pl))
--- /dev/null
+++ b/vpx_dsp/arm/intrapred_neon.c
@@ -1,0 +1,822 @@
+/*
+ * Copyright (c) 2014 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <arm_neon.h>
+
+#include "./vpx_config.h"
+#include "./vpx_dsp_rtcd.h"
+#include "vpx/vpx_integer.h"
+
+//------------------------------------------------------------------------------
+// DC 4x4
+
+// 'do_above' and 'do_left' facilitate branch removal when inlined.
+static INLINE void dc_4x4(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left,
+ int do_above, int do_left) {+ uint16x8_t sum_top;
+ uint16x8_t sum_left;
+ uint8x8_t dc0;
+
+ if (do_above) {+ const uint8x8_t A = vld1_u8(above); // top row
+ const uint16x4_t p0 = vpaddl_u8(A); // cascading summation of the top
+ const uint16x4_t p1 = vpadd_u16(p0, p0);
+ sum_top = vcombine_u16(p1, p1);
+ }
+
+ if (do_left) {+ const uint8x8_t L = vld1_u8(left); // left border
+ const uint16x4_t p0 = vpaddl_u8(L); // cascading summation of the left
+ const uint16x4_t p1 = vpadd_u16(p0, p0);
+ sum_left = vcombine_u16(p1, p1);
+ }
+
+ if (do_above && do_left) {+ const uint16x8_t sum = vaddq_u16(sum_left, sum_top);
+ dc0 = vrshrn_n_u16(sum, 3);
+ } else if (do_above) {+ dc0 = vrshrn_n_u16(sum_top, 2);
+ } else if (do_left) {+ dc0 = vrshrn_n_u16(sum_left, 2);
+ } else {+ dc0 = vdup_n_u8(0x80);
+ }
+
+ {+ const uint8x8_t dc = vdup_lane_u8(dc0, 0);
+ int i;
+ for (i = 0; i < 4; ++i) {+ vst1_lane_u32((uint32_t*)(dst + i * stride), vreinterpret_u32_u8(dc), 0);
+ }
+ }
+}
+
+void vp9_dc_predictor_4x4_neon(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left) {+ dc_4x4(dst, stride, above, left, 1, 1);
+}
+
+void vp9_dc_left_predictor_4x4_neon(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left) {+ (void)above;
+ dc_4x4(dst, stride, NULL, left, 0, 1);
+}
+
+void vp9_dc_top_predictor_4x4_neon(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left) {+ (void)left;
+ dc_4x4(dst, stride, above, NULL, 1, 0);
+}
+
+void vp9_dc_128_predictor_4x4_neon(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left) {+ (void)above;
+ (void)left;
+ dc_4x4(dst, stride, NULL, NULL, 0, 0);
+}
+
+//------------------------------------------------------------------------------
+// DC 8x8
+
+// 'do_above' and 'do_left' facilitate branch removal when inlined.
+static INLINE void dc_8x8(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left,
+ int do_above, int do_left) {+ uint16x8_t sum_top;
+ uint16x8_t sum_left;
+ uint8x8_t dc0;
+
+ if (do_above) {+ const uint8x8_t A = vld1_u8(above); // top row
+ const uint16x4_t p0 = vpaddl_u8(A); // cascading summation of the top
+ const uint16x4_t p1 = vpadd_u16(p0, p0);
+ const uint16x4_t p2 = vpadd_u16(p1, p1);
+ sum_top = vcombine_u16(p2, p2);
+ }
+
+ if (do_left) {+ const uint8x8_t L = vld1_u8(left); // left border
+ const uint16x4_t p0 = vpaddl_u8(L); // cascading summation of the left
+ const uint16x4_t p1 = vpadd_u16(p0, p0);
+ const uint16x4_t p2 = vpadd_u16(p1, p1);
+ sum_left = vcombine_u16(p2, p2);
+ }
+
+ if (do_above && do_left) {+ const uint16x8_t sum = vaddq_u16(sum_left, sum_top);
+ dc0 = vrshrn_n_u16(sum, 4);
+ } else if (do_above) {+ dc0 = vrshrn_n_u16(sum_top, 3);
+ } else if (do_left) {+ dc0 = vrshrn_n_u16(sum_left, 3);
+ } else {+ dc0 = vdup_n_u8(0x80);
+ }
+
+ {+ const uint8x8_t dc = vdup_lane_u8(dc0, 0);
+ int i;
+ for (i = 0; i < 8; ++i) {+ vst1_u32((uint32_t*)(dst + i * stride), vreinterpret_u32_u8(dc));
+ }
+ }
+}
+
+void vp9_dc_predictor_8x8_neon(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left) {+ dc_8x8(dst, stride, above, left, 1, 1);
+}
+
+void vp9_dc_left_predictor_8x8_neon(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left) {+ (void)above;
+ dc_8x8(dst, stride, NULL, left, 0, 1);
+}
+
+void vp9_dc_top_predictor_8x8_neon(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left) {+ (void)left;
+ dc_8x8(dst, stride, above, NULL, 1, 0);
+}
+
+void vp9_dc_128_predictor_8x8_neon(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left) {+ (void)above;
+ (void)left;
+ dc_8x8(dst, stride, NULL, NULL, 0, 0);
+}
+
+//------------------------------------------------------------------------------
+// DC 16x16
+
+// 'do_above' and 'do_left' facilitate branch removal when inlined.
+static INLINE void dc_16x16(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left,
+ int do_above, int do_left) {+ uint16x8_t sum_top;
+ uint16x8_t sum_left;
+ uint8x8_t dc0;
+
+ if (do_above) {+ const uint8x16_t A = vld1q_u8(above); // top row
+ const uint16x8_t p0 = vpaddlq_u8(A); // cascading summation of the top
+ const uint16x4_t p1 = vadd_u16(vget_low_u16(p0), vget_high_u16(p0));
+ const uint16x4_t p2 = vpadd_u16(p1, p1);
+ const uint16x4_t p3 = vpadd_u16(p2, p2);
+ sum_top = vcombine_u16(p3, p3);
+ }
+
+ if (do_left) {+ const uint8x16_t L = vld1q_u8(left); // left row
+ const uint16x8_t p0 = vpaddlq_u8(L); // cascading summation of the left
+ const uint16x4_t p1 = vadd_u16(vget_low_u16(p0), vget_high_u16(p0));
+ const uint16x4_t p2 = vpadd_u16(p1, p1);
+ const uint16x4_t p3 = vpadd_u16(p2, p2);
+ sum_left = vcombine_u16(p3, p3);
+ }
+
+ if (do_above && do_left) {+ const uint16x8_t sum = vaddq_u16(sum_left, sum_top);
+ dc0 = vrshrn_n_u16(sum, 5);
+ } else if (do_above) {+ dc0 = vrshrn_n_u16(sum_top, 4);
+ } else if (do_left) {+ dc0 = vrshrn_n_u16(sum_left, 4);
+ } else {+ dc0 = vdup_n_u8(0x80);
+ }
+
+ {+ const uint8x16_t dc = vdupq_lane_u8(dc0, 0);
+ int i;
+ for (i = 0; i < 16; ++i) {+ vst1q_u8(dst + i * stride, dc);
+ }
+ }
+}
+
+void vp9_dc_predictor_16x16_neon(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left) {+ dc_16x16(dst, stride, above, left, 1, 1);
+}
+
+void vp9_dc_left_predictor_16x16_neon(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above,
+ const uint8_t *left) {+ (void)above;
+ dc_16x16(dst, stride, NULL, left, 0, 1);
+}
+
+void vp9_dc_top_predictor_16x16_neon(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above,
+ const uint8_t *left) {+ (void)left;
+ dc_16x16(dst, stride, above, NULL, 1, 0);
+}
+
+void vp9_dc_128_predictor_16x16_neon(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above,
+ const uint8_t *left) {+ (void)above;
+ (void)left;
+ dc_16x16(dst, stride, NULL, NULL, 0, 0);
+}
+
+//------------------------------------------------------------------------------
+// DC 32x32
+
+// 'do_above' and 'do_left' facilitate branch removal when inlined.
+static INLINE void dc_32x32(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left,
+ int do_above, int do_left) {+ uint16x8_t sum_top;
+ uint16x8_t sum_left;
+ uint8x8_t dc0;
+
+ if (do_above) {+ const uint8x16_t A0 = vld1q_u8(above); // top row
+ const uint8x16_t A1 = vld1q_u8(above + 16);
+ const uint16x8_t p0 = vpaddlq_u8(A0); // cascading summation of the top
+ const uint16x8_t p1 = vpaddlq_u8(A1);
+ const uint16x8_t p2 = vaddq_u16(p0, p1);
+ const uint16x4_t p3 = vadd_u16(vget_low_u16(p2), vget_high_u16(p2));
+ const uint16x4_t p4 = vpadd_u16(p3, p3);
+ const uint16x4_t p5 = vpadd_u16(p4, p4);
+ sum_top = vcombine_u16(p5, p5);
+ }
+
+ if (do_left) {+ const uint8x16_t L0 = vld1q_u8(left); // left row
+ const uint8x16_t L1 = vld1q_u8(left + 16);
+ const uint16x8_t p0 = vpaddlq_u8(L0); // cascading summation of the left
+ const uint16x8_t p1 = vpaddlq_u8(L1);
+ const uint16x8_t p2 = vaddq_u16(p0, p1);
+ const uint16x4_t p3 = vadd_u16(vget_low_u16(p2), vget_high_u16(p2));
+ const uint16x4_t p4 = vpadd_u16(p3, p3);
+ const uint16x4_t p5 = vpadd_u16(p4, p4);
+ sum_left = vcombine_u16(p5, p5);
+ }
+
+ if (do_above && do_left) {+ const uint16x8_t sum = vaddq_u16(sum_left, sum_top);
+ dc0 = vrshrn_n_u16(sum, 6);
+ } else if (do_above) {+ dc0 = vrshrn_n_u16(sum_top, 5);
+ } else if (do_left) {+ dc0 = vrshrn_n_u16(sum_left, 5);
+ } else {+ dc0 = vdup_n_u8(0x80);
+ }
+
+ {+ const uint8x16_t dc = vdupq_lane_u8(dc0, 0);
+ int i;
+ for (i = 0; i < 32; ++i) {+ vst1q_u8(dst + i * stride, dc);
+ vst1q_u8(dst + i * stride + 16, dc);
+ }
+ }
+}
+
+void vp9_dc_predictor_32x32_neon(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left) {+ dc_32x32(dst, stride, above, left, 1, 1);
+}
+
+void vp9_dc_left_predictor_32x32_neon(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above,
+ const uint8_t *left) {+ (void)above;
+ dc_32x32(dst, stride, NULL, left, 0, 1);
+}
+
+void vp9_dc_top_predictor_32x32_neon(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above,
+ const uint8_t *left) {+ (void)left;
+ dc_32x32(dst, stride, above, NULL, 1, 0);
+}
+
+void vp9_dc_128_predictor_32x32_neon(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above,
+ const uint8_t *left) {+ (void)above;
+ (void)left;
+ dc_32x32(dst, stride, NULL, NULL, 0, 0);
+}
+
+// -----------------------------------------------------------------------------
+
+void vp9_d45_predictor_4x4_neon(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left) {+ const uint64x1_t A0 = vreinterpret_u64_u8(vld1_u8(above)); // top row
+ const uint64x1_t A1 = vshr_n_u64(A0, 8);
+ const uint64x1_t A2 = vshr_n_u64(A0, 16);
+ const uint8x8_t ABCDEFGH = vreinterpret_u8_u64(A0);
+ const uint8x8_t BCDEFGH0 = vreinterpret_u8_u64(A1);
+ const uint8x8_t CDEFGH00 = vreinterpret_u8_u64(A2);
+ const uint8x8_t avg1 = vhadd_u8(ABCDEFGH, CDEFGH00);
+ const uint8x8_t avg2 = vrhadd_u8(avg1, BCDEFGH0);
+ const uint64x1_t avg2_u64 = vreinterpret_u64_u8(avg2);
+ const uint32x2_t r0 = vreinterpret_u32_u8(avg2);
+ const uint32x2_t r1 = vreinterpret_u32_u64(vshr_n_u64(avg2_u64, 8));
+ const uint32x2_t r2 = vreinterpret_u32_u64(vshr_n_u64(avg2_u64, 16));
+ const uint32x2_t r3 = vreinterpret_u32_u64(vshr_n_u64(avg2_u64, 24));
+ (void)left;
+ vst1_lane_u32((uint32_t *)(dst + 0 * stride), r0, 0);
+ vst1_lane_u32((uint32_t *)(dst + 1 * stride), r1, 0);
+ vst1_lane_u32((uint32_t *)(dst + 2 * stride), r2, 0);
+ vst1_lane_u32((uint32_t *)(dst + 3 * stride), r3, 0);
+ dst[3 * stride + 3] = above[7];
+}
+
+void vp9_d45_predictor_8x8_neon(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left) {+ static const uint8_t shuffle1[8] = { 1, 2, 3, 4, 5, 6, 7, 7 };+ static const uint8_t shuffle2[8] = { 2, 3, 4, 5, 6, 7, 7, 7 };+ const uint8x8_t sh_12345677 = vld1_u8(shuffle1);
+ const uint8x8_t sh_23456777 = vld1_u8(shuffle2);
+ const uint8x8_t A0 = vld1_u8(above); // top row
+ const uint8x8_t A1 = vtbl1_u8(A0, sh_12345677);
+ const uint8x8_t A2 = vtbl1_u8(A0, sh_23456777);
+ const uint8x8_t avg1 = vhadd_u8(A0, A2);
+ uint8x8_t row = vrhadd_u8(avg1, A1);
+ int i;
+ (void)left;
+ for (i = 0; i < 7; ++i) {+ vst1_u8(dst + i * stride, row);
+ row = vtbl1_u8(row, sh_12345677);
+ }
+ vst1_u8(dst + i * stride, row);
+}
+
+void vp9_d45_predictor_16x16_neon(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left) {+ const uint8x16_t A0 = vld1q_u8(above); // top row
+ const uint8x16_t above_right = vld1q_dup_u8(above + 15);
+ const uint8x16_t A1 = vextq_u8(A0, above_right, 1);
+ const uint8x16_t A2 = vextq_u8(A0, above_right, 2);
+ const uint8x16_t avg1 = vhaddq_u8(A0, A2);
+ uint8x16_t row = vrhaddq_u8(avg1, A1);
+ int i;
+ (void)left;
+ for (i = 0; i < 15; ++i) {+ vst1q_u8(dst + i * stride, row);
+ row = vextq_u8(row, above_right, 1);
+ }
+ vst1q_u8(dst + i * stride, row);
+}
+
+// -----------------------------------------------------------------------------
+
+void vp9_d135_predictor_4x4_neon(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left) {+ const uint8x8_t XABCD_u8 = vld1_u8(above - 1);
+ const uint64x1_t XABCD = vreinterpret_u64_u8(XABCD_u8);
+ const uint64x1_t ____XABC = vshl_n_u64(XABCD, 32);
+ const uint32x2_t zero = vdup_n_u32(0);
+ const uint32x2_t IJKL = vld1_lane_u32((const uint32_t *)left, zero, 0);
+ const uint8x8_t IJKL_u8 = vreinterpret_u8_u32(IJKL);
+ const uint64x1_t LKJI____ = vreinterpret_u64_u8(vrev32_u8(IJKL_u8));
+ const uint64x1_t LKJIXABC = vorr_u64(LKJI____, ____XABC);
+ const uint8x8_t KJIXABC_ = vreinterpret_u8_u64(vshr_n_u64(LKJIXABC, 8));
+ const uint8x8_t JIXABC__ = vreinterpret_u8_u64(vshr_n_u64(LKJIXABC, 16));
+ const uint8_t D = vget_lane_u8(XABCD_u8, 4);
+ const uint8x8_t JIXABCD_ = vset_lane_u8(D, JIXABC__, 6);
+ const uint8x8_t LKJIXABC_u8 = vreinterpret_u8_u64(LKJIXABC);
+ const uint8x8_t avg1 = vhadd_u8(JIXABCD_, LKJIXABC_u8);
+ const uint8x8_t avg2 = vrhadd_u8(avg1, KJIXABC_);
+ const uint64x1_t avg2_u64 = vreinterpret_u64_u8(avg2);
+ const uint32x2_t r3 = vreinterpret_u32_u8(avg2);
+ const uint32x2_t r2 = vreinterpret_u32_u64(vshr_n_u64(avg2_u64, 8));
+ const uint32x2_t r1 = vreinterpret_u32_u64(vshr_n_u64(avg2_u64, 16));
+ const uint32x2_t r0 = vreinterpret_u32_u64(vshr_n_u64(avg2_u64, 24));
+ vst1_lane_u32((uint32_t *)(dst + 0 * stride), r0, 0);
+ vst1_lane_u32((uint32_t *)(dst + 1 * stride), r1, 0);
+ vst1_lane_u32((uint32_t *)(dst + 2 * stride), r2, 0);
+ vst1_lane_u32((uint32_t *)(dst + 3 * stride), r3, 0);
+}
+
+#if !HAVE_NEON_ASM
+
+void vp9_v_predictor_4x4_neon(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left) {+ int i;
+ uint32x2_t d0u32 = vdup_n_u32(0);
+ (void)left;
+
+ d0u32 = vld1_lane_u32((const uint32_t *)above, d0u32, 0);
+ for (i = 0; i < 4; i++, dst += stride)
+ vst1_lane_u32((uint32_t *)dst, d0u32, 0);
+}
+
+void vp9_v_predictor_8x8_neon(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left) {+ int i;
+ uint8x8_t d0u8 = vdup_n_u8(0);
+ (void)left;
+
+ d0u8 = vld1_u8(above);
+ for (i = 0; i < 8; i++, dst += stride)
+ vst1_u8(dst, d0u8);
+}
+
+void vp9_v_predictor_16x16_neon(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left) {+ int i;
+ uint8x16_t q0u8 = vdupq_n_u8(0);
+ (void)left;
+
+ q0u8 = vld1q_u8(above);
+ for (i = 0; i < 16; i++, dst += stride)
+ vst1q_u8(dst, q0u8);
+}
+
+void vp9_v_predictor_32x32_neon(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left) {+ int i;
+ uint8x16_t q0u8 = vdupq_n_u8(0);
+ uint8x16_t q1u8 = vdupq_n_u8(0);
+ (void)left;
+
+ q0u8 = vld1q_u8(above);
+ q1u8 = vld1q_u8(above + 16);
+ for (i = 0; i < 32; i++, dst += stride) {+ vst1q_u8(dst, q0u8);
+ vst1q_u8(dst + 16, q1u8);
+ }
+}
+
+void vp9_h_predictor_4x4_neon(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left) {+ uint8x8_t d0u8 = vdup_n_u8(0);
+ uint32x2_t d1u32 = vdup_n_u32(0);
+ (void)above;
+
+ d1u32 = vld1_lane_u32((const uint32_t *)left, d1u32, 0);
+
+ d0u8 = vdup_lane_u8(vreinterpret_u8_u32(d1u32), 0);
+ vst1_lane_u32((uint32_t *)dst, vreinterpret_u32_u8(d0u8), 0);
+ dst += stride;
+ d0u8 = vdup_lane_u8(vreinterpret_u8_u32(d1u32), 1);
+ vst1_lane_u32((uint32_t *)dst, vreinterpret_u32_u8(d0u8), 0);
+ dst += stride;
+ d0u8 = vdup_lane_u8(vreinterpret_u8_u32(d1u32), 2);
+ vst1_lane_u32((uint32_t *)dst, vreinterpret_u32_u8(d0u8), 0);
+ dst += stride;
+ d0u8 = vdup_lane_u8(vreinterpret_u8_u32(d1u32), 3);
+ vst1_lane_u32((uint32_t *)dst, vreinterpret_u32_u8(d0u8), 0);
+}
+
+void vp9_h_predictor_8x8_neon(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left) {+ uint8x8_t d0u8 = vdup_n_u8(0);
+ uint64x1_t d1u64 = vdup_n_u64(0);
+ (void)above;
+
+ d1u64 = vld1_u64((const uint64_t *)left);
+
+ d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 0);
+ vst1_u8(dst, d0u8);
+ dst += stride;
+ d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 1);
+ vst1_u8(dst, d0u8);
+ dst += stride;
+ d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 2);
+ vst1_u8(dst, d0u8);
+ dst += stride;
+ d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 3);
+ vst1_u8(dst, d0u8);
+ dst += stride;
+ d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 4);
+ vst1_u8(dst, d0u8);
+ dst += stride;
+ d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 5);
+ vst1_u8(dst, d0u8);
+ dst += stride;
+ d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 6);
+ vst1_u8(dst, d0u8);
+ dst += stride;
+ d0u8 = vdup_lane_u8(vreinterpret_u8_u64(d1u64), 7);
+ vst1_u8(dst, d0u8);
+}
+
+void vp9_h_predictor_16x16_neon(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left) {+ int j;
+ uint8x8_t d2u8 = vdup_n_u8(0);
+ uint8x16_t q0u8 = vdupq_n_u8(0);
+ uint8x16_t q1u8 = vdupq_n_u8(0);
+ (void)above;
+
+ q1u8 = vld1q_u8(left);
+ d2u8 = vget_low_u8(q1u8);
+ for (j = 0; j < 2; j++, d2u8 = vget_high_u8(q1u8)) {+ q0u8 = vdupq_lane_u8(d2u8, 0);
+ vst1q_u8(dst, q0u8);
+ dst += stride;
+ q0u8 = vdupq_lane_u8(d2u8, 1);
+ vst1q_u8(dst, q0u8);
+ dst += stride;
+ q0u8 = vdupq_lane_u8(d2u8, 2);
+ vst1q_u8(dst, q0u8);
+ dst += stride;
+ q0u8 = vdupq_lane_u8(d2u8, 3);
+ vst1q_u8(dst, q0u8);
+ dst += stride;
+ q0u8 = vdupq_lane_u8(d2u8, 4);
+ vst1q_u8(dst, q0u8);
+ dst += stride;
+ q0u8 = vdupq_lane_u8(d2u8, 5);
+ vst1q_u8(dst, q0u8);
+ dst += stride;
+ q0u8 = vdupq_lane_u8(d2u8, 6);
+ vst1q_u8(dst, q0u8);
+ dst += stride;
+ q0u8 = vdupq_lane_u8(d2u8, 7);
+ vst1q_u8(dst, q0u8);
+ dst += stride;
+ }
+}
+
+void vp9_h_predictor_32x32_neon(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left) {+ int j, k;
+ uint8x8_t d2u8 = vdup_n_u8(0);
+ uint8x16_t q0u8 = vdupq_n_u8(0);
+ uint8x16_t q1u8 = vdupq_n_u8(0);
+ (void)above;
+
+ for (k = 0; k < 2; k++, left += 16) {+ q1u8 = vld1q_u8(left);
+ d2u8 = vget_low_u8(q1u8);
+ for (j = 0; j < 2; j++, d2u8 = vget_high_u8(q1u8)) {+ q0u8 = vdupq_lane_u8(d2u8, 0);
+ vst1q_u8(dst, q0u8);
+ vst1q_u8(dst + 16, q0u8);
+ dst += stride;
+ q0u8 = vdupq_lane_u8(d2u8, 1);
+ vst1q_u8(dst, q0u8);
+ vst1q_u8(dst + 16, q0u8);
+ dst += stride;
+ q0u8 = vdupq_lane_u8(d2u8, 2);
+ vst1q_u8(dst, q0u8);
+ vst1q_u8(dst + 16, q0u8);
+ dst += stride;
+ q0u8 = vdupq_lane_u8(d2u8, 3);
+ vst1q_u8(dst, q0u8);
+ vst1q_u8(dst + 16, q0u8);
+ dst += stride;
+ q0u8 = vdupq_lane_u8(d2u8, 4);
+ vst1q_u8(dst, q0u8);
+ vst1q_u8(dst + 16, q0u8);
+ dst += stride;
+ q0u8 = vdupq_lane_u8(d2u8, 5);
+ vst1q_u8(dst, q0u8);
+ vst1q_u8(dst + 16, q0u8);
+ dst += stride;
+ q0u8 = vdupq_lane_u8(d2u8, 6);
+ vst1q_u8(dst, q0u8);
+ vst1q_u8(dst + 16, q0u8);
+ dst += stride;
+ q0u8 = vdupq_lane_u8(d2u8, 7);
+ vst1q_u8(dst, q0u8);
+ vst1q_u8(dst + 16, q0u8);
+ dst += stride;
+ }
+ }
+}
+
+void vp9_tm_predictor_4x4_neon(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left) {+ int i;
+ uint16x8_t q1u16, q3u16;
+ int16x8_t q1s16;
+ uint8x8_t d0u8 = vdup_n_u8(0);
+ uint32x2_t d2u32 = vdup_n_u32(0);
+
+ d0u8 = vld1_dup_u8(above - 1);
+ d2u32 = vld1_lane_u32((const uint32_t *)above, d2u32, 0);
+ q3u16 = vsubl_u8(vreinterpret_u8_u32(d2u32), d0u8);
+ for (i = 0; i < 4; i++, dst += stride) {+ q1u16 = vdupq_n_u16((uint16_t)left[i]);
+ q1s16 = vaddq_s16(vreinterpretq_s16_u16(q1u16),
+ vreinterpretq_s16_u16(q3u16));
+ d0u8 = vqmovun_s16(q1s16);
+ vst1_lane_u32((uint32_t *)dst, vreinterpret_u32_u8(d0u8), 0);
+ }
+}
+
+void vp9_tm_predictor_8x8_neon(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left) {+ int j;
+ uint16x8_t q0u16, q3u16, q10u16;
+ int16x8_t q0s16;
+ uint16x4_t d20u16;
+ uint8x8_t d0u8, d2u8, d30u8;
+
+ d0u8 = vld1_dup_u8(above - 1);
+ d30u8 = vld1_u8(left);
+ d2u8 = vld1_u8(above);
+ q10u16 = vmovl_u8(d30u8);
+ q3u16 = vsubl_u8(d2u8, d0u8);
+ d20u16 = vget_low_u16(q10u16);
+ for (j = 0; j < 2; j++, d20u16 = vget_high_u16(q10u16)) {+ q0u16 = vdupq_lane_u16(d20u16, 0);
+ q0s16 = vaddq_s16(vreinterpretq_s16_u16(q3u16),
+ vreinterpretq_s16_u16(q0u16));
+ d0u8 = vqmovun_s16(q0s16);
+ vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d0u8));
+ dst += stride;
+ q0u16 = vdupq_lane_u16(d20u16, 1);
+ q0s16 = vaddq_s16(vreinterpretq_s16_u16(q3u16),
+ vreinterpretq_s16_u16(q0u16));
+ d0u8 = vqmovun_s16(q0s16);
+ vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d0u8));
+ dst += stride;
+ q0u16 = vdupq_lane_u16(d20u16, 2);
+ q0s16 = vaddq_s16(vreinterpretq_s16_u16(q3u16),
+ vreinterpretq_s16_u16(q0u16));
+ d0u8 = vqmovun_s16(q0s16);
+ vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d0u8));
+ dst += stride;
+ q0u16 = vdupq_lane_u16(d20u16, 3);
+ q0s16 = vaddq_s16(vreinterpretq_s16_u16(q3u16),
+ vreinterpretq_s16_u16(q0u16));
+ d0u8 = vqmovun_s16(q0s16);
+ vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d0u8));
+ dst += stride;
+ }
+}
+
+void vp9_tm_predictor_16x16_neon(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left) {+ int j, k;
+ uint16x8_t q0u16, q2u16, q3u16, q8u16, q10u16;
+ uint8x16_t q0u8, q1u8;
+ int16x8_t q0s16, q1s16, q8s16, q11s16;
+ uint16x4_t d20u16;
+ uint8x8_t d2u8, d3u8, d18u8, d22u8, d23u8;
+
+ q0u8 = vld1q_dup_u8(above - 1);
+ q1u8 = vld1q_u8(above);
+ q2u16 = vsubl_u8(vget_low_u8(q1u8), vget_low_u8(q0u8));
+ q3u16 = vsubl_u8(vget_high_u8(q1u8), vget_high_u8(q0u8));
+ for (k = 0; k < 2; k++, left += 8) {+ d18u8 = vld1_u8(left);
+ q10u16 = vmovl_u8(d18u8);
+ d20u16 = vget_low_u16(q10u16);
+ for (j = 0; j < 2; j++, d20u16 = vget_high_u16(q10u16)) {+ q0u16 = vdupq_lane_u16(d20u16, 0);
+ q8u16 = vdupq_lane_u16(d20u16, 1);
+ q1s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
+ vreinterpretq_s16_u16(q2u16));
+ q0s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
+ vreinterpretq_s16_u16(q3u16));
+ q11s16 = vaddq_s16(vreinterpretq_s16_u16(q8u16),
+ vreinterpretq_s16_u16(q2u16));
+ q8s16 = vaddq_s16(vreinterpretq_s16_u16(q8u16),
+ vreinterpretq_s16_u16(q3u16));
+ d2u8 = vqmovun_s16(q1s16);
+ d3u8 = vqmovun_s16(q0s16);
+ d22u8 = vqmovun_s16(q11s16);
+ d23u8 = vqmovun_s16(q8s16);
+ vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d2u8));
+ vst1_u64((uint64_t *)(dst + 8), vreinterpret_u64_u8(d3u8));
+ dst += stride;
+ vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d22u8));
+ vst1_u64((uint64_t *)(dst + 8), vreinterpret_u64_u8(d23u8));
+ dst += stride;
+
+ q0u16 = vdupq_lane_u16(d20u16, 2);
+ q8u16 = vdupq_lane_u16(d20u16, 3);
+ q1s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
+ vreinterpretq_s16_u16(q2u16));
+ q0s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
+ vreinterpretq_s16_u16(q3u16));
+ q11s16 = vaddq_s16(vreinterpretq_s16_u16(q8u16),
+ vreinterpretq_s16_u16(q2u16));
+ q8s16 = vaddq_s16(vreinterpretq_s16_u16(q8u16),
+ vreinterpretq_s16_u16(q3u16));
+ d2u8 = vqmovun_s16(q1s16);
+ d3u8 = vqmovun_s16(q0s16);
+ d22u8 = vqmovun_s16(q11s16);
+ d23u8 = vqmovun_s16(q8s16);
+ vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d2u8));
+ vst1_u64((uint64_t *)(dst + 8), vreinterpret_u64_u8(d3u8));
+ dst += stride;
+ vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d22u8));
+ vst1_u64((uint64_t *)(dst + 8), vreinterpret_u64_u8(d23u8));
+ dst += stride;
+ }
+ }
+}
+
+void vp9_tm_predictor_32x32_neon(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left) {+ int j, k;
+ uint16x8_t q0u16, q3u16, q8u16, q9u16, q10u16, q11u16;
+ uint8x16_t q0u8, q1u8, q2u8;
+ int16x8_t q12s16, q13s16, q14s16, q15s16;
+ uint16x4_t d6u16;
+ uint8x8_t d0u8, d1u8, d2u8, d3u8, d26u8;
+
+ q0u8 = vld1q_dup_u8(above - 1);
+ q1u8 = vld1q_u8(above);
+ q2u8 = vld1q_u8(above + 16);
+ q8u16 = vsubl_u8(vget_low_u8(q1u8), vget_low_u8(q0u8));
+ q9u16 = vsubl_u8(vget_high_u8(q1u8), vget_high_u8(q0u8));
+ q10u16 = vsubl_u8(vget_low_u8(q2u8), vget_low_u8(q0u8));
+ q11u16 = vsubl_u8(vget_high_u8(q2u8), vget_high_u8(q0u8));
+ for (k = 0; k < 4; k++, left += 8) {+ d26u8 = vld1_u8(left);
+ q3u16 = vmovl_u8(d26u8);
+ d6u16 = vget_low_u16(q3u16);
+ for (j = 0; j < 2; j++, d6u16 = vget_high_u16(q3u16)) {+ q0u16 = vdupq_lane_u16(d6u16, 0);
+ q12s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
+ vreinterpretq_s16_u16(q8u16));
+ q13s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
+ vreinterpretq_s16_u16(q9u16));
+ q14s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
+ vreinterpretq_s16_u16(q10u16));
+ q15s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
+ vreinterpretq_s16_u16(q11u16));
+ d0u8 = vqmovun_s16(q12s16);
+ d1u8 = vqmovun_s16(q13s16);
+ d2u8 = vqmovun_s16(q14s16);
+ d3u8 = vqmovun_s16(q15s16);
+ q0u8 = vcombine_u8(d0u8, d1u8);
+ q1u8 = vcombine_u8(d2u8, d3u8);
+ vst1q_u64((uint64_t *)dst, vreinterpretq_u64_u8(q0u8));
+ vst1q_u64((uint64_t *)(dst + 16), vreinterpretq_u64_u8(q1u8));
+ dst += stride;
+
+ q0u16 = vdupq_lane_u16(d6u16, 1);
+ q12s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
+ vreinterpretq_s16_u16(q8u16));
+ q13s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
+ vreinterpretq_s16_u16(q9u16));
+ q14s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
+ vreinterpretq_s16_u16(q10u16));
+ q15s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
+ vreinterpretq_s16_u16(q11u16));
+ d0u8 = vqmovun_s16(q12s16);
+ d1u8 = vqmovun_s16(q13s16);
+ d2u8 = vqmovun_s16(q14s16);
+ d3u8 = vqmovun_s16(q15s16);
+ q0u8 = vcombine_u8(d0u8, d1u8);
+ q1u8 = vcombine_u8(d2u8, d3u8);
+ vst1q_u64((uint64_t *)dst, vreinterpretq_u64_u8(q0u8));
+ vst1q_u64((uint64_t *)(dst + 16), vreinterpretq_u64_u8(q1u8));
+ dst += stride;
+
+ q0u16 = vdupq_lane_u16(d6u16, 2);
+ q12s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
+ vreinterpretq_s16_u16(q8u16));
+ q13s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
+ vreinterpretq_s16_u16(q9u16));
+ q14s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
+ vreinterpretq_s16_u16(q10u16));
+ q15s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
+ vreinterpretq_s16_u16(q11u16));
+ d0u8 = vqmovun_s16(q12s16);
+ d1u8 = vqmovun_s16(q13s16);
+ d2u8 = vqmovun_s16(q14s16);
+ d3u8 = vqmovun_s16(q15s16);
+ q0u8 = vcombine_u8(d0u8, d1u8);
+ q1u8 = vcombine_u8(d2u8, d3u8);
+ vst1q_u64((uint64_t *)dst, vreinterpretq_u64_u8(q0u8));
+ vst1q_u64((uint64_t *)(dst + 16), vreinterpretq_u64_u8(q1u8));
+ dst += stride;
+
+ q0u16 = vdupq_lane_u16(d6u16, 3);
+ q12s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
+ vreinterpretq_s16_u16(q8u16));
+ q13s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
+ vreinterpretq_s16_u16(q9u16));
+ q14s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
+ vreinterpretq_s16_u16(q10u16));
+ q15s16 = vaddq_s16(vreinterpretq_s16_u16(q0u16),
+ vreinterpretq_s16_u16(q11u16));
+ d0u8 = vqmovun_s16(q12s16);
+ d1u8 = vqmovun_s16(q13s16);
+ d2u8 = vqmovun_s16(q14s16);
+ d3u8 = vqmovun_s16(q15s16);
+ q0u8 = vcombine_u8(d0u8, d1u8);
+ q1u8 = vcombine_u8(d2u8, d3u8);
+ vst1q_u64((uint64_t *)dst, vreinterpretq_u64_u8(q0u8));
+ vst1q_u64((uint64_t *)(dst + 16), vreinterpretq_u64_u8(q1u8));
+ dst += stride;
+ }
+ }
+}
+#endif // !HAVE_NEON_ASM
--- /dev/null
+++ b/vpx_dsp/arm/intrapred_neon_asm.asm
@@ -1,0 +1,630 @@
+;
+; Copyright (c) 2014 The WebM project authors. All Rights Reserved.
+;
+; Use of this source code is governed by a BSD-style license
+; that can be found in the LICENSE file in the root of the source
+; tree. An additional intellectual property rights grant can be found
+; in the file PATENTS. All contributing project authors may
+; be found in the AUTHORS file in the root of the source tree.
+;
+
+ EXPORT |vp9_v_predictor_4x4_neon|
+ EXPORT |vp9_v_predictor_8x8_neon|
+ EXPORT |vp9_v_predictor_16x16_neon|
+ EXPORT |vp9_v_predictor_32x32_neon|
+ EXPORT |vp9_h_predictor_4x4_neon|
+ EXPORT |vp9_h_predictor_8x8_neon|
+ EXPORT |vp9_h_predictor_16x16_neon|
+ EXPORT |vp9_h_predictor_32x32_neon|
+ EXPORT |vp9_tm_predictor_4x4_neon|
+ EXPORT |vp9_tm_predictor_8x8_neon|
+ EXPORT |vp9_tm_predictor_16x16_neon|
+ EXPORT |vp9_tm_predictor_32x32_neon|
+ ARM
+ REQUIRE8
+ PRESERVE8
+
+ AREA ||.text||, CODE, READONLY, ALIGN=2
+
+;void vp9_v_predictor_4x4_neon(uint8_t *dst, ptrdiff_t y_stride,
+; const uint8_t *above,
+; const uint8_t *left)
+; r0 uint8_t *dst
+; r1 ptrdiff_t y_stride
+; r2 const uint8_t *above
+; r3 const uint8_t *left
+
+|vp9_v_predictor_4x4_neon| PROC
+ vld1.32 {d0[0]}, [r2]+ vst1.32 {d0[0]}, [r0], r1+ vst1.32 {d0[0]}, [r0], r1+ vst1.32 {d0[0]}, [r0], r1+ vst1.32 {d0[0]}, [r0], r1+ bx lr
+ ENDP ; |vp9_v_predictor_4x4_neon|
+
+;void vp9_v_predictor_8x8_neon(uint8_t *dst, ptrdiff_t y_stride,
+; const uint8_t *above,
+; const uint8_t *left)
+; r0 uint8_t *dst
+; r1 ptrdiff_t y_stride
+; r2 const uint8_t *above
+; r3 const uint8_t *left
+
+|vp9_v_predictor_8x8_neon| PROC
+ vld1.8 {d0}, [r2]+ vst1.8 {d0}, [r0], r1+ vst1.8 {d0}, [r0], r1+ vst1.8 {d0}, [r0], r1+ vst1.8 {d0}, [r0], r1+ vst1.8 {d0}, [r0], r1+ vst1.8 {d0}, [r0], r1+ vst1.8 {d0}, [r0], r1+ vst1.8 {d0}, [r0], r1+ bx lr
+ ENDP ; |vp9_v_predictor_8x8_neon|
+
+;void vp9_v_predictor_16x16_neon(uint8_t *dst, ptrdiff_t y_stride,
+; const uint8_t *above,
+; const uint8_t *left)
+; r0 uint8_t *dst
+; r1 ptrdiff_t y_stride
+; r2 const uint8_t *above
+; r3 const uint8_t *left
+
+|vp9_v_predictor_16x16_neon| PROC
+ vld1.8 {q0}, [r2]+ vst1.8 {q0}, [r0], r1+ vst1.8 {q0}, [r0], r1+ vst1.8 {q0}, [r0], r1+ vst1.8 {q0}, [r0], r1+ vst1.8 {q0}, [r0], r1+ vst1.8 {q0}, [r0], r1+ vst1.8 {q0}, [r0], r1+ vst1.8 {q0}, [r0], r1+ vst1.8 {q0}, [r0], r1+ vst1.8 {q0}, [r0], r1+ vst1.8 {q0}, [r0], r1+ vst1.8 {q0}, [r0], r1+ vst1.8 {q0}, [r0], r1+ vst1.8 {q0}, [r0], r1+ vst1.8 {q0}, [r0], r1+ vst1.8 {q0}, [r0], r1+ bx lr
+ ENDP ; |vp9_v_predictor_16x16_neon|
+
+;void vp9_v_predictor_32x32_neon(uint8_t *dst, ptrdiff_t y_stride,
+; const uint8_t *above,
+; const uint8_t *left)
+; r0 uint8_t *dst
+; r1 ptrdiff_t y_stride
+; r2 const uint8_t *above
+; r3 const uint8_t *left
+
+|vp9_v_predictor_32x32_neon| PROC
+ vld1.8 {q0, q1}, [r2]+ mov r2, #2
+loop_v
+ vst1.8 {q0, q1}, [r0], r1+ vst1.8 {q0, q1}, [r0], r1+ vst1.8 {q0, q1}, [r0], r1+ vst1.8 {q0, q1}, [r0], r1+ vst1.8 {q0, q1}, [r0], r1+ vst1.8 {q0, q1}, [r0], r1+ vst1.8 {q0, q1}, [r0], r1+ vst1.8 {q0, q1}, [r0], r1+ vst1.8 {q0, q1}, [r0], r1+ vst1.8 {q0, q1}, [r0], r1+ vst1.8 {q0, q1}, [r0], r1+ vst1.8 {q0, q1}, [r0], r1+ vst1.8 {q0, q1}, [r0], r1+ vst1.8 {q0, q1}, [r0], r1+ vst1.8 {q0, q1}, [r0], r1+ vst1.8 {q0, q1}, [r0], r1+ subs r2, r2, #1
+ bgt loop_v
+ bx lr
+ ENDP ; |vp9_v_predictor_32x32_neon|
+
+;void vp9_h_predictor_4x4_neon(uint8_t *dst, ptrdiff_t y_stride,
+; const uint8_t *above,
+; const uint8_t *left)
+; r0 uint8_t *dst
+; r1 ptrdiff_t y_stride
+; r2 const uint8_t *above
+; r3 const uint8_t *left
+
+|vp9_h_predictor_4x4_neon| PROC
+ vld1.32 {d1[0]}, [r3]+ vdup.8 d0, d1[0]
+ vst1.32 {d0[0]}, [r0], r1+ vdup.8 d0, d1[1]
+ vst1.32 {d0[0]}, [r0], r1+ vdup.8 d0, d1[2]
+ vst1.32 {d0[0]}, [r0], r1+ vdup.8 d0, d1[3]
+ vst1.32 {d0[0]}, [r0], r1+ bx lr
+ ENDP ; |vp9_h_predictor_4x4_neon|
+
+;void vp9_h_predictor_8x8_neon(uint8_t *dst, ptrdiff_t y_stride,
+; const uint8_t *above,
+; const uint8_t *left)
+; r0 uint8_t *dst
+; r1 ptrdiff_t y_stride
+; r2 const uint8_t *above
+; r3 const uint8_t *left
+
+|vp9_h_predictor_8x8_neon| PROC
+ vld1.64 {d1}, [r3]+ vdup.8 d0, d1[0]
+ vst1.64 {d0}, [r0], r1+ vdup.8 d0, d1[1]
+ vst1.64 {d0}, [r0], r1+ vdup.8 d0, d1[2]
+ vst1.64 {d0}, [r0], r1+ vdup.8 d0, d1[3]
+ vst1.64 {d0}, [r0], r1+ vdup.8 d0, d1[4]
+ vst1.64 {d0}, [r0], r1+ vdup.8 d0, d1[5]
+ vst1.64 {d0}, [r0], r1+ vdup.8 d0, d1[6]
+ vst1.64 {d0}, [r0], r1+ vdup.8 d0, d1[7]
+ vst1.64 {d0}, [r0], r1+ bx lr
+ ENDP ; |vp9_h_predictor_8x8_neon|
+
+;void vp9_h_predictor_16x16_neon(uint8_t *dst, ptrdiff_t y_stride,
+; const uint8_t *above,
+; const uint8_t *left)
+; r0 uint8_t *dst
+; r1 ptrdiff_t y_stride
+; r2 const uint8_t *above
+; r3 const uint8_t *left
+
+|vp9_h_predictor_16x16_neon| PROC
+ vld1.8 {q1}, [r3]+ vdup.8 q0, d2[0]
+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d2[1]
+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d2[2]
+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d2[3]
+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d2[4]
+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d2[5]
+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d2[6]
+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d2[7]
+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d3[0]
+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d3[1]
+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d3[2]
+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d3[3]
+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d3[4]
+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d3[5]
+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d3[6]
+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d3[7]
+ vst1.8 {q0}, [r0], r1+ bx lr
+ ENDP ; |vp9_h_predictor_16x16_neon|
+
+;void vp9_h_predictor_32x32_neon(uint8_t *dst, ptrdiff_t y_stride,
+; const uint8_t *above,
+; const uint8_t *left)
+; r0 uint8_t *dst
+; r1 ptrdiff_t y_stride
+; r2 const uint8_t *above
+; r3 const uint8_t *left
+
+|vp9_h_predictor_32x32_neon| PROC
+ sub r1, r1, #16
+ mov r2, #2
+loop_h
+ vld1.8 {q1}, [r3]!+ vdup.8 q0, d2[0]
+ vst1.8 {q0}, [r0]!+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d2[1]
+ vst1.8 {q0}, [r0]!+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d2[2]
+ vst1.8 {q0}, [r0]!+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d2[3]
+ vst1.8 {q0}, [r0]!+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d2[4]
+ vst1.8 {q0}, [r0]!+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d2[5]
+ vst1.8 {q0}, [r0]!+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d2[6]
+ vst1.8 {q0}, [r0]!+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d2[7]
+ vst1.8 {q0}, [r0]!+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d3[0]
+ vst1.8 {q0}, [r0]!+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d3[1]
+ vst1.8 {q0}, [r0]!+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d3[2]
+ vst1.8 {q0}, [r0]!+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d3[3]
+ vst1.8 {q0}, [r0]!+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d3[4]
+ vst1.8 {q0}, [r0]!+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d3[5]
+ vst1.8 {q0}, [r0]!+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d3[6]
+ vst1.8 {q0}, [r0]!+ vst1.8 {q0}, [r0], r1+ vdup.8 q0, d3[7]
+ vst1.8 {q0}, [r0]!+ vst1.8 {q0}, [r0], r1+ subs r2, r2, #1
+ bgt loop_h
+ bx lr
+ ENDP ; |vp9_h_predictor_32x32_neon|
+
+;void vp9_tm_predictor_4x4_neon (uint8_t *dst, ptrdiff_t y_stride,
+; const uint8_t *above,
+; const uint8_t *left)
+; r0 uint8_t *dst
+; r1 ptrdiff_t y_stride
+; r2 const uint8_t *above
+; r3 const uint8_t *left
+
+|vp9_tm_predictor_4x4_neon| PROC
+ ; Load ytop_left = above[-1];
+ sub r12, r2, #1
+ vld1.u8 {d0[]}, [r12]+
+ ; Load above 4 pixels
+ vld1.32 {d2[0]}, [r2]+
+ ; Compute above - ytop_left
+ vsubl.u8 q3, d2, d0
+
+ ; Load left row by row and compute left + (above - ytop_left)
+ ; 1st row and 2nd row
+ vld1.u8 {d2[]}, [r3]!+ vld1.u8 {d4[]}, [r3]!+ vmovl.u8 q1, d2
+ vmovl.u8 q2, d4
+ vadd.s16 q1, q1, q3
+ vadd.s16 q2, q2, q3
+ vqmovun.s16 d0, q1
+ vqmovun.s16 d1, q2
+ vst1.32 {d0[0]}, [r0], r1+ vst1.32 {d1[0]}, [r0], r1+
+ ; 3rd row and 4th row
+ vld1.u8 {d2[]}, [r3]!+ vld1.u8 {d4[]}, [r3]+ vmovl.u8 q1, d2
+ vmovl.u8 q2, d4
+ vadd.s16 q1, q1, q3
+ vadd.s16 q2, q2, q3
+ vqmovun.s16 d0, q1
+ vqmovun.s16 d1, q2
+ vst1.32 {d0[0]}, [r0], r1+ vst1.32 {d1[0]}, [r0], r1+ bx lr
+ ENDP ; |vp9_tm_predictor_4x4_neon|
+
+;void vp9_tm_predictor_8x8_neon (uint8_t *dst, ptrdiff_t y_stride,
+; const uint8_t *above,
+; const uint8_t *left)
+; r0 uint8_t *dst
+; r1 ptrdiff_t y_stride
+; r2 const uint8_t *above
+; r3 const uint8_t *left
+
+|vp9_tm_predictor_8x8_neon| PROC
+ ; Load ytop_left = above[-1];
+ sub r12, r2, #1
+ vld1.8 {d0[]}, [r12]+
+ ; preload 8 left
+ vld1.8 {d30}, [r3]+
+ ; Load above 8 pixels
+ vld1.64 {d2}, [r2]+
+ vmovl.u8 q10, d30
+
+ ; Compute above - ytop_left
+ vsubl.u8 q3, d2, d0
+
+ ; Load left row by row and compute left + (above - ytop_left)
+ ; 1st row and 2nd row
+ vdup.16 q0, d20[0]
+ vdup.16 q1, d20[1]
+ vadd.s16 q0, q3, q0
+ vadd.s16 q1, q3, q1
+
+ ; 3rd row and 4th row
+ vdup.16 q8, d20[2]
+ vdup.16 q9, d20[3]
+ vadd.s16 q8, q3, q8
+ vadd.s16 q9, q3, q9
+
+ vqmovun.s16 d0, q0
+ vqmovun.s16 d1, q1
+ vqmovun.s16 d2, q8
+ vqmovun.s16 d3, q9
+
+ vst1.64 {d0}, [r0], r1+ vst1.64 {d1}, [r0], r1+ vst1.64 {d2}, [r0], r1+ vst1.64 {d3}, [r0], r1+
+ ; 5th row and 6th row
+ vdup.16 q0, d21[0]
+ vdup.16 q1, d21[1]
+ vadd.s16 q0, q3, q0
+ vadd.s16 q1, q3, q1
+
+ ; 7th row and 8th row
+ vdup.16 q8, d21[2]
+ vdup.16 q9, d21[3]
+ vadd.s16 q8, q3, q8
+ vadd.s16 q9, q3, q9
+
+ vqmovun.s16 d0, q0
+ vqmovun.s16 d1, q1
+ vqmovun.s16 d2, q8
+ vqmovun.s16 d3, q9
+
+ vst1.64 {d0}, [r0], r1+ vst1.64 {d1}, [r0], r1+ vst1.64 {d2}, [r0], r1+ vst1.64 {d3}, [r0], r1+
+ bx lr
+ ENDP ; |vp9_tm_predictor_8x8_neon|
+
+;void vp9_tm_predictor_16x16_neon (uint8_t *dst, ptrdiff_t y_stride,
+; const uint8_t *above,
+; const uint8_t *left)
+; r0 uint8_t *dst
+; r1 ptrdiff_t y_stride
+; r2 const uint8_t *above
+; r3 const uint8_t *left
+
+|vp9_tm_predictor_16x16_neon| PROC
+ ; Load ytop_left = above[-1];
+ sub r12, r2, #1
+ vld1.8 {d0[]}, [r12]+
+ ; Load above 8 pixels
+ vld1.8 {q1}, [r2]+
+ ; preload 8 left into r12
+ vld1.8 {d18}, [r3]!+
+ ; Compute above - ytop_left
+ vsubl.u8 q2, d2, d0
+ vsubl.u8 q3, d3, d0
+
+ vmovl.u8 q10, d18
+
+ ; Load left row by row and compute left + (above - ytop_left)
+ ; Process 8 rows in each single loop and loop 2 times to process 16 rows.
+ mov r2, #2
+
+loop_16x16_neon
+ ; Process two rows.
+ vdup.16 q0, d20[0]
+ vdup.16 q8, d20[1]
+ vadd.s16 q1, q0, q2
+ vadd.s16 q0, q0, q3
+ vadd.s16 q11, q8, q2
+ vadd.s16 q8, q8, q3
+ vqmovun.s16 d2, q1
+ vqmovun.s16 d3, q0
+ vqmovun.s16 d22, q11
+ vqmovun.s16 d23, q8
+ vdup.16 q0, d20[2] ; proload next 2 rows data
+ vdup.16 q8, d20[3]
+ vst1.64 {d2,d3}, [r0], r1+ vst1.64 {d22,d23}, [r0], r1+
+ ; Process two rows.
+ vadd.s16 q1, q0, q2
+ vadd.s16 q0, q0, q3
+ vadd.s16 q11, q8, q2
+ vadd.s16 q8, q8, q3
+ vqmovun.s16 d2, q1
+ vqmovun.s16 d3, q0
+ vqmovun.s16 d22, q11
+ vqmovun.s16 d23, q8
+ vdup.16 q0, d21[0] ; proload next 2 rows data
+ vdup.16 q8, d21[1]
+ vst1.64 {d2,d3}, [r0], r1+ vst1.64 {d22,d23}, [r0], r1+
+ vadd.s16 q1, q0, q2
+ vadd.s16 q0, q0, q3
+ vadd.s16 q11, q8, q2
+ vadd.s16 q8, q8, q3
+ vqmovun.s16 d2, q1
+ vqmovun.s16 d3, q0
+ vqmovun.s16 d22, q11
+ vqmovun.s16 d23, q8
+ vdup.16 q0, d21[2] ; proload next 2 rows data
+ vdup.16 q8, d21[3]
+ vst1.64 {d2,d3}, [r0], r1+ vst1.64 {d22,d23}, [r0], r1+
+
+ vadd.s16 q1, q0, q2
+ vadd.s16 q0, q0, q3
+ vadd.s16 q11, q8, q2
+ vadd.s16 q8, q8, q3
+ vqmovun.s16 d2, q1
+ vqmovun.s16 d3, q0
+ vqmovun.s16 d22, q11
+ vqmovun.s16 d23, q8
+ vld1.8 {d18}, [r3]! ; preload 8 left into r12+ vmovl.u8 q10, d18
+ vst1.64 {d2,d3}, [r0], r1+ vst1.64 {d22,d23}, [r0], r1+
+ subs r2, r2, #1
+ bgt loop_16x16_neon
+
+ bx lr
+ ENDP ; |vp9_tm_predictor_16x16_neon|
+
+;void vp9_tm_predictor_32x32_neon (uint8_t *dst, ptrdiff_t y_stride,
+; const uint8_t *above,
+; const uint8_t *left)
+; r0 uint8_t *dst
+; r1 ptrdiff_t y_stride
+; r2 const uint8_t *above
+; r3 const uint8_t *left
+
+|vp9_tm_predictor_32x32_neon| PROC
+ ; Load ytop_left = above[-1];
+ sub r12, r2, #1
+ vld1.8 {d0[]}, [r12]+
+ ; Load above 32 pixels
+ vld1.8 {q1}, [r2]!+ vld1.8 {q2}, [r2]+
+ ; preload 8 left pixels
+ vld1.8 {d26}, [r3]!+
+ ; Compute above - ytop_left
+ vsubl.u8 q8, d2, d0
+ vsubl.u8 q9, d3, d0
+ vsubl.u8 q10, d4, d0
+ vsubl.u8 q11, d5, d0
+
+ vmovl.u8 q3, d26
+
+ ; Load left row by row and compute left + (above - ytop_left)
+ ; Process 8 rows in each single loop and loop 4 times to process 32 rows.
+ mov r2, #4
+
+loop_32x32_neon
+ ; Process two rows.
+ vdup.16 q0, d6[0]
+ vdup.16 q2, d6[1]
+ vadd.s16 q12, q0, q8
+ vadd.s16 q13, q0, q9
+ vadd.s16 q14, q0, q10
+ vadd.s16 q15, q0, q11
+ vqmovun.s16 d0, q12
+ vqmovun.s16 d1, q13
+ vadd.s16 q12, q2, q8
+ vadd.s16 q13, q2, q9
+ vqmovun.s16 d2, q14
+ vqmovun.s16 d3, q15
+ vadd.s16 q14, q2, q10
+ vadd.s16 q15, q2, q11
+ vst1.64 {d0-d3}, [r0], r1+ vqmovun.s16 d24, q12
+ vqmovun.s16 d25, q13
+ vqmovun.s16 d26, q14
+ vqmovun.s16 d27, q15
+ vdup.16 q1, d6[2]
+ vdup.16 q2, d6[3]
+ vst1.64 {d24-d27}, [r0], r1+
+ ; Process two rows.
+ vadd.s16 q12, q1, q8
+ vadd.s16 q13, q1, q9
+ vadd.s16 q14, q1, q10
+ vadd.s16 q15, q1, q11
+ vqmovun.s16 d0, q12
+ vqmovun.s16 d1, q13
+ vadd.s16 q12, q2, q8
+ vadd.s16 q13, q2, q9
+ vqmovun.s16 d2, q14
+ vqmovun.s16 d3, q15
+ vadd.s16 q14, q2, q10
+ vadd.s16 q15, q2, q11
+ vst1.64 {d0-d3}, [r0], r1+ vqmovun.s16 d24, q12
+ vqmovun.s16 d25, q13
+ vqmovun.s16 d26, q14
+ vqmovun.s16 d27, q15
+ vdup.16 q0, d7[0]
+ vdup.16 q2, d7[1]
+ vst1.64 {d24-d27}, [r0], r1+
+ ; Process two rows.
+ vadd.s16 q12, q0, q8
+ vadd.s16 q13, q0, q9
+ vadd.s16 q14, q0, q10
+ vadd.s16 q15, q0, q11
+ vqmovun.s16 d0, q12
+ vqmovun.s16 d1, q13
+ vadd.s16 q12, q2, q8
+ vadd.s16 q13, q2, q9
+ vqmovun.s16 d2, q14
+ vqmovun.s16 d3, q15
+ vadd.s16 q14, q2, q10
+ vadd.s16 q15, q2, q11
+ vst1.64 {d0-d3}, [r0], r1+ vqmovun.s16 d24, q12
+ vqmovun.s16 d25, q13
+ vqmovun.s16 d26, q14
+ vqmovun.s16 d27, q15
+ vdup.16 q0, d7[2]
+ vdup.16 q2, d7[3]
+ vst1.64 {d24-d27}, [r0], r1+
+ ; Process two rows.
+ vadd.s16 q12, q0, q8
+ vadd.s16 q13, q0, q9
+ vadd.s16 q14, q0, q10
+ vadd.s16 q15, q0, q11
+ vqmovun.s16 d0, q12
+ vqmovun.s16 d1, q13
+ vadd.s16 q12, q2, q8
+ vadd.s16 q13, q2, q9
+ vqmovun.s16 d2, q14
+ vqmovun.s16 d3, q15
+ vadd.s16 q14, q2, q10
+ vadd.s16 q15, q2, q11
+ vst1.64 {d0-d3}, [r0], r1+ vqmovun.s16 d24, q12
+ vqmovun.s16 d25, q13
+ vld1.8 {d0}, [r3]! ; preload 8 left pixels+ vqmovun.s16 d26, q14
+ vqmovun.s16 d27, q15
+ vmovl.u8 q3, d0
+ vst1.64 {d24-d27}, [r0], r1+
+ subs r2, r2, #1
+ bgt loop_32x32_neon
+
+ bx lr
+ ENDP ; |vp9_tm_predictor_32x32_neon|
+
+ END
--- /dev/null
+++ b/vpx_dsp/intrapred.c
@@ -1,0 +1,692 @@
+/*
+ * Copyright (c) 2015 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "./vpx_config.h"
+#include "./vpx_dsp_rtcd.h"
+
+#include "vpx_dsp/vpx_dsp_common.h"
+#include "vpx_mem/vpx_mem.h"
+
+#define DST(x, y) dst[(x) + (y) * stride]
+#define AVG3(a, b, c) (((a) + 2 * (b) + (c) + 2) >> 2)
+#define AVG2(a, b) (((a) + (b) + 1) >> 1)
+
+static INLINE void d207_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
+ const uint8_t *above, const uint8_t *left) {+ int r, c;
+ (void) above;
+ // first column
+ for (r = 0; r < bs - 1; ++r)
+ dst[r * stride] = AVG2(left[r], left[r + 1]);
+ dst[(bs - 1) * stride] = left[bs - 1];
+ dst++;
+
+ // second column
+ for (r = 0; r < bs - 2; ++r)
+ dst[r * stride] = AVG3(left[r], left[r + 1], left[r + 2]);
+ dst[(bs - 2) * stride] = AVG3(left[bs - 2], left[bs - 1], left[bs - 1]);
+ dst[(bs - 1) * stride] = left[bs - 1];
+ dst++;
+
+ // rest of last row
+ for (c = 0; c < bs - 2; ++c)
+ dst[(bs - 1) * stride + c] = left[bs - 1];
+
+ for (r = bs - 2; r >= 0; --r)
+ for (c = 0; c < bs - 2; ++c)
+ dst[r * stride + c] = dst[(r + 1) * stride + c - 2];
+}
+
+static INLINE void d63_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
+ const uint8_t *above, const uint8_t *left) {+ int r, c;
+ int size;
+ (void)left;
+ for (c = 0; c < bs; ++c) {+ dst[c] = AVG2(above[c], above[c + 1]);
+ dst[stride + c] = AVG3(above[c], above[c + 1], above[c + 2]);
+ }
+ for (r = 2, size = bs - 2; r < bs; r += 2, --size) {+ memcpy(dst + (r + 0) * stride, dst + (r >> 1), size);
+ memset(dst + (r + 0) * stride + size, above[bs - 1], bs - size);
+ memcpy(dst + (r + 1) * stride, dst + stride + (r >> 1), size);
+ memset(dst + (r + 1) * stride + size, above[bs - 1], bs - size);
+ }
+}
+
+static INLINE void d45_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
+ const uint8_t *above, const uint8_t *left) {+ const uint8_t above_right = above[bs - 1];
+ const uint8_t *const dst_row0 = dst;
+ int x, size;
+ (void)left;
+
+ for (x = 0; x < bs - 1; ++x) {+ dst[x] = AVG3(above[x], above[x + 1], above[x + 2]);
+ }
+ dst[bs - 1] = above_right;
+ dst += stride;
+ for (x = 1, size = bs - 2; x < bs; ++x, --size) {+ memcpy(dst, dst_row0 + x, size);
+ memset(dst + size, above_right, x + 1);
+ dst += stride;
+ }
+}
+
+static INLINE void d117_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
+ const uint8_t *above, const uint8_t *left) {+ int r, c;
+
+ // first row
+ for (c = 0; c < bs; c++)
+ dst[c] = AVG2(above[c - 1], above[c]);
+ dst += stride;
+
+ // second row
+ dst[0] = AVG3(left[0], above[-1], above[0]);
+ for (c = 1; c < bs; c++)
+ dst[c] = AVG3(above[c - 2], above[c - 1], above[c]);
+ dst += stride;
+
+ // the rest of first col
+ dst[0] = AVG3(above[-1], left[0], left[1]);
+ for (r = 3; r < bs; ++r)
+ dst[(r - 2) * stride] = AVG3(left[r - 3], left[r - 2], left[r - 1]);
+
+ // the rest of the block
+ for (r = 2; r < bs; ++r) {+ for (c = 1; c < bs; c++)
+ dst[c] = dst[-2 * stride + c - 1];
+ dst += stride;
+ }
+}
+
+static INLINE void d135_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
+ const uint8_t *above, const uint8_t *left) {+ int r, c;
+ dst[0] = AVG3(left[0], above[-1], above[0]);
+ for (c = 1; c < bs; c++)
+ dst[c] = AVG3(above[c - 2], above[c - 1], above[c]);
+
+ dst[stride] = AVG3(above[-1], left[0], left[1]);
+ for (r = 2; r < bs; ++r)
+ dst[r * stride] = AVG3(left[r - 2], left[r - 1], left[r]);
+
+ dst += stride;
+ for (r = 1; r < bs; ++r) {+ for (c = 1; c < bs; c++)
+ dst[c] = dst[-stride + c - 1];
+ dst += stride;
+ }
+}
+
+static INLINE void d153_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
+ const uint8_t *above, const uint8_t *left) {+ int r, c;
+ dst[0] = AVG2(above[-1], left[0]);
+ for (r = 1; r < bs; r++)
+ dst[r * stride] = AVG2(left[r - 1], left[r]);
+ dst++;
+
+ dst[0] = AVG3(left[0], above[-1], above[0]);
+ dst[stride] = AVG3(above[-1], left[0], left[1]);
+ for (r = 2; r < bs; r++)
+ dst[r * stride] = AVG3(left[r - 2], left[r - 1], left[r]);
+ dst++;
+
+ for (c = 0; c < bs - 2; c++)
+ dst[c] = AVG3(above[c - 1], above[c], above[c + 1]);
+ dst += stride;
+
+ for (r = 1; r < bs; ++r) {+ for (c = 0; c < bs - 2; c++)
+ dst[c] = dst[-stride + c - 2];
+ dst += stride;
+ }
+}
+
+static INLINE void v_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
+ const uint8_t *above, const uint8_t *left) {+ int r;
+ (void) left;
+
+ for (r = 0; r < bs; r++) {+ memcpy(dst, above, bs);
+ dst += stride;
+ }
+}
+
+static INLINE void h_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
+ const uint8_t *above, const uint8_t *left) {+ int r;
+ (void) above;
+
+ for (r = 0; r < bs; r++) {+ memset(dst, left[r], bs);
+ dst += stride;
+ }
+}
+
+static INLINE void tm_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
+ const uint8_t *above, const uint8_t *left) {+ int r, c;
+ int ytop_left = above[-1];
+
+ for (r = 0; r < bs; r++) {+ for (c = 0; c < bs; c++)
+ dst[c] = clip_pixel(left[r] + above[c] - ytop_left);
+ dst += stride;
+ }
+}
+
+static INLINE void dc_128_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
+ const uint8_t *above, const uint8_t *left) {+ int r;
+ (void) above;
+ (void) left;
+
+ for (r = 0; r < bs; r++) {+ memset(dst, 128, bs);
+ dst += stride;
+ }
+}
+
+static INLINE void dc_left_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
+ const uint8_t *above,
+ const uint8_t *left) {+ int i, r, expected_dc, sum = 0;
+ (void) above;
+
+ for (i = 0; i < bs; i++)
+ sum += left[i];
+ expected_dc = (sum + (bs >> 1)) / bs;
+
+ for (r = 0; r < bs; r++) {+ memset(dst, expected_dc, bs);
+ dst += stride;
+ }
+}
+
+static INLINE void dc_top_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
+ const uint8_t *above, const uint8_t *left) {+ int i, r, expected_dc, sum = 0;
+ (void) left;
+
+ for (i = 0; i < bs; i++)
+ sum += above[i];
+ expected_dc = (sum + (bs >> 1)) / bs;
+
+ for (r = 0; r < bs; r++) {+ memset(dst, expected_dc, bs);
+ dst += stride;
+ }
+}
+
+static INLINE void dc_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
+ const uint8_t *above, const uint8_t *left) {+ int i, r, expected_dc, sum = 0;
+ const int count = 2 * bs;
+
+ for (i = 0; i < bs; i++) {+ sum += above[i];
+ sum += left[i];
+ }
+
+ expected_dc = (sum + (count >> 1)) / count;
+
+ for (r = 0; r < bs; r++) {+ memset(dst, expected_dc, bs);
+ dst += stride;
+ }
+}
+
+void vp9_d207_predictor_4x4_c(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left) {+ const int I = left[0];
+ const int J = left[1];
+ const int K = left[2];
+ const int L = left[3];
+ (void)above;
+ DST(0, 0) = AVG2(I, J);
+ DST(2, 0) = DST(0, 1) = AVG2(J, K);
+ DST(2, 1) = DST(0, 2) = AVG2(K, L);
+ DST(1, 0) = AVG3(I, J, K);
+ DST(3, 0) = DST(1, 1) = AVG3(J, K, L);
+ DST(3, 1) = DST(1, 2) = AVG3(K, L, L);
+ DST(3, 2) = DST(2, 2) =
+ DST(0, 3) = DST(1, 3) = DST(2, 3) = DST(3, 3) = L;
+}
+
+void vp9_d63_predictor_4x4_c(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left) {+ const int A = above[0];
+ const int B = above[1];
+ const int C = above[2];
+ const int D = above[3];
+ const int E = above[4];
+ const int F = above[5];
+ const int G = above[6];
+ (void)left;
+ DST(0, 0) = AVG2(A, B);
+ DST(1, 0) = DST(0, 2) = AVG2(B, C);
+ DST(2, 0) = DST(1, 2) = AVG2(C, D);
+ DST(3, 0) = DST(2, 2) = AVG2(D, E);
+ DST(3, 2) = AVG2(E, F); // differs from vp8
+
+ DST(0, 1) = AVG3(A, B, C);
+ DST(1, 1) = DST(0, 3) = AVG3(B, C, D);
+ DST(2, 1) = DST(1, 3) = AVG3(C, D, E);
+ DST(3, 1) = DST(2, 3) = AVG3(D, E, F);
+ DST(3, 3) = AVG3(E, F, G); // differs from vp8
+}
+
+void vp9_d45_predictor_4x4_c(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left) {+ const int A = above[0];
+ const int B = above[1];
+ const int C = above[2];
+ const int D = above[3];
+ const int E = above[4];
+ const int F = above[5];
+ const int G = above[6];
+ const int H = above[7];
+ (void)stride;
+ (void)left;
+ DST(0, 0) = AVG3(A, B, C);
+ DST(1, 0) = DST(0, 1) = AVG3(B, C, D);
+ DST(2, 0) = DST(1, 1) = DST(0, 2) = AVG3(C, D, E);
+ DST(3, 0) = DST(2, 1) = DST(1, 2) = DST(0, 3) = AVG3(D, E, F);
+ DST(3, 1) = DST(2, 2) = DST(1, 3) = AVG3(E, F, G);
+ DST(3, 2) = DST(2, 3) = AVG3(F, G, H);
+ DST(3, 3) = H; // differs from vp8
+}
+
+void vp9_d117_predictor_4x4_c(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left) {+ const int I = left[0];
+ const int J = left[1];
+ const int K = left[2];
+ const int X = above[-1];
+ const int A = above[0];
+ const int B = above[1];
+ const int C = above[2];
+ const int D = above[3];
+ DST(0, 0) = DST(1, 2) = AVG2(X, A);
+ DST(1, 0) = DST(2, 2) = AVG2(A, B);
+ DST(2, 0) = DST(3, 2) = AVG2(B, C);
+ DST(3, 0) = AVG2(C, D);
+
+ DST(0, 3) = AVG3(K, J, I);
+ DST(0, 2) = AVG3(J, I, X);
+ DST(0, 1) = DST(1, 3) = AVG3(I, X, A);
+ DST(1, 1) = DST(2, 3) = AVG3(X, A, B);
+ DST(2, 1) = DST(3, 3) = AVG3(A, B, C);
+ DST(3, 1) = AVG3(B, C, D);
+}
+
+void vp9_d135_predictor_4x4_c(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left) {+ const int I = left[0];
+ const int J = left[1];
+ const int K = left[2];
+ const int L = left[3];
+ const int X = above[-1];
+ const int A = above[0];
+ const int B = above[1];
+ const int C = above[2];
+ const int D = above[3];
+ (void)stride;
+ DST(0, 3) = AVG3(J, K, L);
+ DST(1, 3) = DST(0, 2) = AVG3(I, J, K);
+ DST(2, 3) = DST(1, 2) = DST(0, 1) = AVG3(X, I, J);
+ DST(3, 3) = DST(2, 2) = DST(1, 1) = DST(0, 0) = AVG3(A, X, I);
+ DST(3, 2) = DST(2, 1) = DST(1, 0) = AVG3(B, A, X);
+ DST(3, 1) = DST(2, 0) = AVG3(C, B, A);
+ DST(3, 0) = AVG3(D, C, B);
+}
+
+void vp9_d153_predictor_4x4_c(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left) {+ const int I = left[0];
+ const int J = left[1];
+ const int K = left[2];
+ const int L = left[3];
+ const int X = above[-1];
+ const int A = above[0];
+ const int B = above[1];
+ const int C = above[2];
+
+ DST(0, 0) = DST(2, 1) = AVG2(I, X);
+ DST(0, 1) = DST(2, 2) = AVG2(J, I);
+ DST(0, 2) = DST(2, 3) = AVG2(K, J);
+ DST(0, 3) = AVG2(L, K);
+
+ DST(3, 0) = AVG3(A, B, C);
+ DST(2, 0) = AVG3(X, A, B);
+ DST(1, 0) = DST(3, 1) = AVG3(I, X, A);
+ DST(1, 1) = DST(3, 2) = AVG3(J, I, X);
+ DST(1, 2) = DST(3, 3) = AVG3(K, J, I);
+ DST(1, 3) = AVG3(L, K, J);
+}
+
+#if CONFIG_VP9_HIGHBITDEPTH
+static INLINE void highbd_d207_predictor(uint16_t *dst, ptrdiff_t stride,
+ int bs, const uint16_t *above,
+ const uint16_t *left, int bd) {+ int r, c;
+ (void) above;
+ (void) bd;
+
+ // First column.
+ for (r = 0; r < bs - 1; ++r) {+ dst[r * stride] = AVG2(left[r], left[r + 1]);
+ }
+ dst[(bs - 1) * stride] = left[bs - 1];
+ dst++;
+
+ // Second column.
+ for (r = 0; r < bs - 2; ++r) {+ dst[r * stride] = AVG3(left[r], left[r + 1], left[r + 2]);
+ }
+ dst[(bs - 2) * stride] = AVG3(left[bs - 2], left[bs - 1], left[bs - 1]);
+ dst[(bs - 1) * stride] = left[bs - 1];
+ dst++;
+
+ // Rest of last row.
+ for (c = 0; c < bs - 2; ++c)
+ dst[(bs - 1) * stride + c] = left[bs - 1];
+
+ for (r = bs - 2; r >= 0; --r) {+ for (c = 0; c < bs - 2; ++c)
+ dst[r * stride + c] = dst[(r + 1) * stride + c - 2];
+ }
+}
+
+static INLINE void highbd_d63_predictor(uint16_t *dst, ptrdiff_t stride,
+ int bs, const uint16_t *above,
+ const uint16_t *left, int bd) {+ int r, c;
+ (void) left;
+ (void) bd;
+ for (r = 0; r < bs; ++r) {+ for (c = 0; c < bs; ++c) {+ dst[c] = r & 1 ? AVG3(above[(r >> 1) + c], above[(r >> 1) + c + 1],
+ above[(r >> 1) + c + 2])
+ : AVG2(above[(r >> 1) + c], above[(r >> 1) + c + 1]);
+ }
+ dst += stride;
+ }
+}
+
+static INLINE void highbd_d45_predictor(uint16_t *dst, ptrdiff_t stride, int bs,
+ const uint16_t *above,
+ const uint16_t *left, int bd) {+ int r, c;
+ (void) left;
+ (void) bd;
+ for (r = 0; r < bs; ++r) {+ for (c = 0; c < bs; ++c) {+ dst[c] = r + c + 2 < bs * 2 ? AVG3(above[r + c], above[r + c + 1],
+ above[r + c + 2])
+ : above[bs * 2 - 1];
+ }
+ dst += stride;
+ }
+}
+
+static INLINE void highbd_d117_predictor(uint16_t *dst, ptrdiff_t stride,
+ int bs, const uint16_t *above,
+ const uint16_t *left, int bd) {+ int r, c;
+ (void) bd;
+
+ // first row
+ for (c = 0; c < bs; c++)
+ dst[c] = AVG2(above[c - 1], above[c]);
+ dst += stride;
+
+ // second row
+ dst[0] = AVG3(left[0], above[-1], above[0]);
+ for (c = 1; c < bs; c++)
+ dst[c] = AVG3(above[c - 2], above[c - 1], above[c]);
+ dst += stride;
+
+ // the rest of first col
+ dst[0] = AVG3(above[-1], left[0], left[1]);
+ for (r = 3; r < bs; ++r)
+ dst[(r - 2) * stride] = AVG3(left[r - 3], left[r - 2], left[r - 1]);
+
+ // the rest of the block
+ for (r = 2; r < bs; ++r) {+ for (c = 1; c < bs; c++)
+ dst[c] = dst[-2 * stride + c - 1];
+ dst += stride;
+ }
+}
+
+static INLINE void highbd_d135_predictor(uint16_t *dst, ptrdiff_t stride,
+ int bs, const uint16_t *above,
+ const uint16_t *left, int bd) {+ int r, c;
+ (void) bd;
+ dst[0] = AVG3(left[0], above[-1], above[0]);
+ for (c = 1; c < bs; c++)
+ dst[c] = AVG3(above[c - 2], above[c - 1], above[c]);
+
+ dst[stride] = AVG3(above[-1], left[0], left[1]);
+ for (r = 2; r < bs; ++r)
+ dst[r * stride] = AVG3(left[r - 2], left[r - 1], left[r]);
+
+ dst += stride;
+ for (r = 1; r < bs; ++r) {+ for (c = 1; c < bs; c++)
+ dst[c] = dst[-stride + c - 1];
+ dst += stride;
+ }
+}
+
+static INLINE void highbd_d153_predictor(uint16_t *dst, ptrdiff_t stride,
+ int bs, const uint16_t *above,
+ const uint16_t *left, int bd) {+ int r, c;
+ (void) bd;
+ dst[0] = AVG2(above[-1], left[0]);
+ for (r = 1; r < bs; r++)
+ dst[r * stride] = AVG2(left[r - 1], left[r]);
+ dst++;
+
+ dst[0] = AVG3(left[0], above[-1], above[0]);
+ dst[stride] = AVG3(above[-1], left[0], left[1]);
+ for (r = 2; r < bs; r++)
+ dst[r * stride] = AVG3(left[r - 2], left[r - 1], left[r]);
+ dst++;
+
+ for (c = 0; c < bs - 2; c++)
+ dst[c] = AVG3(above[c - 1], above[c], above[c + 1]);
+ dst += stride;
+
+ for (r = 1; r < bs; ++r) {+ for (c = 0; c < bs - 2; c++)
+ dst[c] = dst[-stride + c - 2];
+ dst += stride;
+ }
+}
+
+static INLINE void highbd_v_predictor(uint16_t *dst, ptrdiff_t stride,
+ int bs, const uint16_t *above,
+ const uint16_t *left, int bd) {+ int r;
+ (void) left;
+ (void) bd;
+ for (r = 0; r < bs; r++) {+ memcpy(dst, above, bs * sizeof(uint16_t));
+ dst += stride;
+ }
+}
+
+static INLINE void highbd_h_predictor(uint16_t *dst, ptrdiff_t stride,
+ int bs, const uint16_t *above,
+ const uint16_t *left, int bd) {+ int r;
+ (void) above;
+ (void) bd;
+ for (r = 0; r < bs; r++) {+ vpx_memset16(dst, left[r], bs);
+ dst += stride;
+ }
+}
+
+static INLINE void highbd_tm_predictor(uint16_t *dst, ptrdiff_t stride,
+ int bs, const uint16_t *above,
+ const uint16_t *left, int bd) {+ int r, c;
+ int ytop_left = above[-1];
+ (void) bd;
+
+ for (r = 0; r < bs; r++) {+ for (c = 0; c < bs; c++)
+ dst[c] = clip_pixel_highbd(left[r] + above[c] - ytop_left, bd);
+ dst += stride;
+ }
+}
+
+static INLINE void highbd_dc_128_predictor(uint16_t *dst, ptrdiff_t stride,
+ int bs, const uint16_t *above,
+ const uint16_t *left, int bd) {+ int r;
+ (void) above;
+ (void) left;
+
+ for (r = 0; r < bs; r++) {+ vpx_memset16(dst, 128 << (bd - 8), bs);
+ dst += stride;
+ }
+}
+
+static INLINE void highbd_dc_left_predictor(uint16_t *dst, ptrdiff_t stride,
+ int bs, const uint16_t *above,
+ const uint16_t *left, int bd) {+ int i, r, expected_dc, sum = 0;
+ (void) above;
+ (void) bd;
+
+ for (i = 0; i < bs; i++)
+ sum += left[i];
+ expected_dc = (sum + (bs >> 1)) / bs;
+
+ for (r = 0; r < bs; r++) {+ vpx_memset16(dst, expected_dc, bs);
+ dst += stride;
+ }
+}
+
+static INLINE void highbd_dc_top_predictor(uint16_t *dst, ptrdiff_t stride,
+ int bs, const uint16_t *above,
+ const uint16_t *left, int bd) {+ int i, r, expected_dc, sum = 0;
+ (void) left;
+ (void) bd;
+
+ for (i = 0; i < bs; i++)
+ sum += above[i];
+ expected_dc = (sum + (bs >> 1)) / bs;
+
+ for (r = 0; r < bs; r++) {+ vpx_memset16(dst, expected_dc, bs);
+ dst += stride;
+ }
+}
+
+static INLINE void highbd_dc_predictor(uint16_t *dst, ptrdiff_t stride,
+ int bs, const uint16_t *above,
+ const uint16_t *left, int bd) {+ int i, r, expected_dc, sum = 0;
+ const int count = 2 * bs;
+ (void) bd;
+
+ for (i = 0; i < bs; i++) {+ sum += above[i];
+ sum += left[i];
+ }
+
+ expected_dc = (sum + (count >> 1)) / count;
+
+ for (r = 0; r < bs; r++) {+ vpx_memset16(dst, expected_dc, bs);
+ dst += stride;
+ }
+}
+#endif // CONFIG_VP9_HIGHBITDEPTH
+
+// This serves as a wrapper function, so that all the prediction functions
+// can be unified and accessed as a pointer array. Note that the boundary
+// above and left are not necessarily used all the time.
+#define intra_pred_sized(type, size) \
+ void vp9_##type##_predictor_##size##x##size##_c(uint8_t *dst, \
+ ptrdiff_t stride, \
+ const uint8_t *above, \
+ const uint8_t *left) { \+ type##_predictor(dst, stride, size, above, left); \
+ }
+
+#if CONFIG_VP9_HIGHBITDEPTH
+#define intra_pred_highbd_sized(type, size) \
+ void vp9_highbd_##type##_predictor_##size##x##size##_c( \
+ uint16_t *dst, ptrdiff_t stride, const uint16_t *above, \
+ const uint16_t *left, int bd) { \+ highbd_##type##_predictor(dst, stride, size, above, left, bd); \
+ }
+
+#define intra_pred_allsizes(type) \
+ intra_pred_sized(type, 4) \
+ intra_pred_sized(type, 8) \
+ intra_pred_sized(type, 16) \
+ intra_pred_sized(type, 32) \
+ intra_pred_highbd_sized(type, 4) \
+ intra_pred_highbd_sized(type, 8) \
+ intra_pred_highbd_sized(type, 16) \
+ intra_pred_highbd_sized(type, 32)
+
+#define intra_pred_no_4x4(type) \
+ intra_pred_sized(type, 8) \
+ intra_pred_sized(type, 16) \
+ intra_pred_sized(type, 32) \
+ intra_pred_highbd_sized(type, 4) \
+ intra_pred_highbd_sized(type, 8) \
+ intra_pred_highbd_sized(type, 16) \
+ intra_pred_highbd_sized(type, 32)
+
+#else
+#define intra_pred_allsizes(type) \
+ intra_pred_sized(type, 4) \
+ intra_pred_sized(type, 8) \
+ intra_pred_sized(type, 16) \
+ intra_pred_sized(type, 32)
+
+#define intra_pred_no_4x4(type) \
+ intra_pred_sized(type, 8) \
+ intra_pred_sized(type, 16) \
+ intra_pred_sized(type, 32)
+#endif // CONFIG_VP9_HIGHBITDEPTH
+
+intra_pred_no_4x4(d207)
+intra_pred_no_4x4(d63)
+intra_pred_no_4x4(d45)
+intra_pred_no_4x4(d117)
+intra_pred_no_4x4(d135)
+intra_pred_no_4x4(d153)
+intra_pred_allsizes(v)
+intra_pred_allsizes(h)
+intra_pred_allsizes(tm)
+intra_pred_allsizes(dc_128)
+intra_pred_allsizes(dc_left)
+intra_pred_allsizes(dc_top)
+intra_pred_allsizes(dc)
+#undef intra_pred_allsizes
--- a/vpx_dsp/mips/common_dspr2.h
+++ b/vpx_dsp/mips/common_dspr2.h
@@ -21,6 +21,8 @@
#if HAVE_DSPR2
#define CROP_WIDTH 512
+extern uint8_t *vpx_ff_cropTbl; // From "vpx_dsp/mips/intrapred4_dspr2.c"
+
static INLINE void prefetch_load(const unsigned char *src) {__asm__ __volatile__ (
"pref 0, 0(%[src]) \n\t"
--- /dev/null
+++ b/vpx_dsp/mips/intrapred16_dspr2.c
@@ -1,0 +1,329 @@
+/*
+ * Copyright (c) 2015 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "vpx_dsp/mips/common_dspr2.h"
+
+#if HAVE_DSPR2
+void vp9_h_predictor_16x16_dspr2(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left) {+ int32_t tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp8;
+ int32_t tmp9, tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, tmp16;
+
+ __asm__ __volatile__ (
+ "lb %[tmp1], (%[left]) \n\t"
+ "lb %[tmp2], 1(%[left]) \n\t"
+ "lb %[tmp3], 2(%[left]) \n\t"
+ "lb %[tmp4], 3(%[left]) \n\t"
+ "lb %[tmp5], 4(%[left]) \n\t"
+ "lb %[tmp6], 5(%[left]) \n\t"
+ "lb %[tmp7], 6(%[left]) \n\t"
+ "lb %[tmp8], 7(%[left]) \n\t"
+ "lb %[tmp9], 8(%[left]) \n\t"
+ "lb %[tmp10], 9(%[left]) \n\t"
+ "lb %[tmp11], 10(%[left]) \n\t"
+ "lb %[tmp12], 11(%[left]) \n\t"
+ "lb %[tmp13], 12(%[left]) \n\t"
+ "lb %[tmp14], 13(%[left]) \n\t"
+ "lb %[tmp15], 14(%[left]) \n\t"
+ "lb %[tmp16], 15(%[left]) \n\t"
+
+ "replv.qb %[tmp1], %[tmp1] \n\t"
+ "replv.qb %[tmp2], %[tmp2] \n\t"
+ "replv.qb %[tmp3], %[tmp3] \n\t"
+ "replv.qb %[tmp4], %[tmp4] \n\t"
+ "replv.qb %[tmp5], %[tmp5] \n\t"
+ "replv.qb %[tmp6], %[tmp6] \n\t"
+ "replv.qb %[tmp7], %[tmp7] \n\t"
+ "replv.qb %[tmp8], %[tmp8] \n\t"
+ "replv.qb %[tmp9], %[tmp9] \n\t"
+ "replv.qb %[tmp10], %[tmp10] \n\t"
+ "replv.qb %[tmp11], %[tmp11] \n\t"
+ "replv.qb %[tmp12], %[tmp12] \n\t"
+ "replv.qb %[tmp13], %[tmp13] \n\t"
+ "replv.qb %[tmp14], %[tmp14] \n\t"
+ "replv.qb %[tmp15], %[tmp15] \n\t"
+ "replv.qb %[tmp16], %[tmp16] \n\t"
+
+ "sw %[tmp1], (%[dst]) \n\t"
+ "sw %[tmp1], 4(%[dst]) \n\t"
+ "sw %[tmp1], 8(%[dst]) \n\t"
+ "sw %[tmp1], 12(%[dst]) \n\t"
+
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[tmp2], (%[dst]) \n\t"
+ "sw %[tmp2], 4(%[dst]) \n\t"
+ "sw %[tmp2], 8(%[dst]) \n\t"
+ "sw %[tmp2], 12(%[dst]) \n\t"
+
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[tmp3], (%[dst]) \n\t"
+ "sw %[tmp3], 4(%[dst]) \n\t"
+ "sw %[tmp3], 8(%[dst]) \n\t"
+ "sw %[tmp3], 12(%[dst]) \n\t"
+
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[tmp4], (%[dst]) \n\t"
+ "sw %[tmp4], 4(%[dst]) \n\t"
+ "sw %[tmp4], 8(%[dst]) \n\t"
+ "sw %[tmp4], 12(%[dst]) \n\t"
+
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[tmp5], (%[dst]) \n\t"
+ "sw %[tmp5], 4(%[dst]) \n\t"
+ "sw %[tmp5], 8(%[dst]) \n\t"
+ "sw %[tmp5], 12(%[dst]) \n\t"
+
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[tmp6], (%[dst]) \n\t"
+ "sw %[tmp6], 4(%[dst]) \n\t"
+ "sw %[tmp6], 8(%[dst]) \n\t"
+ "sw %[tmp6], 12(%[dst]) \n\t"
+
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[tmp7], (%[dst]) \n\t"
+ "sw %[tmp7], 4(%[dst]) \n\t"
+ "sw %[tmp7], 8(%[dst]) \n\t"
+ "sw %[tmp7], 12(%[dst]) \n\t"
+
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[tmp8], (%[dst]) \n\t"
+ "sw %[tmp8], 4(%[dst]) \n\t"
+ "sw %[tmp8], 8(%[dst]) \n\t"
+ "sw %[tmp8], 12(%[dst]) \n\t"
+
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[tmp9], (%[dst]) \n\t"
+ "sw %[tmp9], 4(%[dst]) \n\t"
+ "sw %[tmp9], 8(%[dst]) \n\t"
+ "sw %[tmp9], 12(%[dst]) \n\t"
+
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[tmp10], (%[dst]) \n\t"
+ "sw %[tmp10], 4(%[dst]) \n\t"
+ "sw %[tmp10], 8(%[dst]) \n\t"
+ "sw %[tmp10], 12(%[dst]) \n\t"
+
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[tmp11], (%[dst]) \n\t"
+ "sw %[tmp11], 4(%[dst]) \n\t"
+ "sw %[tmp11], 8(%[dst]) \n\t"
+ "sw %[tmp11], 12(%[dst]) \n\t"
+
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[tmp12], (%[dst]) \n\t"
+ "sw %[tmp12], 4(%[dst]) \n\t"
+ "sw %[tmp12], 8(%[dst]) \n\t"
+ "sw %[tmp12], 12(%[dst]) \n\t"
+
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[tmp13], (%[dst]) \n\t"
+ "sw %[tmp13], 4(%[dst]) \n\t"
+ "sw %[tmp13], 8(%[dst]) \n\t"
+ "sw %[tmp13], 12(%[dst]) \n\t"
+
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[tmp14], (%[dst]) \n\t"
+ "sw %[tmp14], 4(%[dst]) \n\t"
+ "sw %[tmp14], 8(%[dst]) \n\t"
+ "sw %[tmp14], 12(%[dst]) \n\t"
+
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[tmp15], (%[dst]) \n\t"
+ "sw %[tmp15], 4(%[dst]) \n\t"
+ "sw %[tmp15], 8(%[dst]) \n\t"
+ "sw %[tmp15], 12(%[dst]) \n\t"
+
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[tmp16], (%[dst]) \n\t"
+ "sw %[tmp16], 4(%[dst]) \n\t"
+ "sw %[tmp16], 8(%[dst]) \n\t"
+ "sw %[tmp16], 12(%[dst]) \n\t"
+
+ : [tmp1] "=&r" (tmp1), [tmp2] "=&r" (tmp2),
+ [tmp3] "=&r" (tmp3), [tmp4] "=&r" (tmp4),
+ [tmp5] "=&r" (tmp5), [tmp7] "=&r" (tmp7),
+ [tmp6] "=&r" (tmp6), [tmp8] "=&r" (tmp8),
+ [tmp9] "=&r" (tmp9), [tmp10] "=&r" (tmp10),
+ [tmp11] "=&r" (tmp11), [tmp12] "=&r" (tmp12),
+ [tmp13] "=&r" (tmp13), [tmp14] "=&r" (tmp14),
+ [tmp15] "=&r" (tmp15), [tmp16] "=&r" (tmp16)
+ : [left] "r" (left), [dst] "r" (dst), [stride] "r" (stride)
+ );
+}
+
+void vp9_dc_predictor_16x16_dspr2(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left) {+ int32_t expected_dc;
+ int32_t average;
+ int32_t tmp, above1, above_l1, above_r1, left1, left_r1, left_l1;
+ int32_t above2, left2;
+
+ __asm__ __volatile__ (
+ "lw %[above1], (%[above]) \n\t"
+ "lw %[above2], 4(%[above]) \n\t"
+ "lw %[left1], (%[left]) \n\t"
+ "lw %[left2], 4(%[left]) \n\t"
+
+ "preceu.ph.qbl %[above_l1], %[above1] \n\t"
+ "preceu.ph.qbr %[above_r1], %[above1] \n\t"
+ "preceu.ph.qbl %[left_l1], %[left1] \n\t"
+ "preceu.ph.qbr %[left_r1], %[left1] \n\t"
+
+ "addu.ph %[average], %[above_r1], %[above_l1] \n\t"
+ "addu.ph %[average], %[average], %[left_l1] \n\t"
+ "addu.ph %[average], %[average], %[left_r1] \n\t"
+
+ "preceu.ph.qbl %[above_l1], %[above2] \n\t"
+ "preceu.ph.qbr %[above_r1], %[above2] \n\t"
+ "preceu.ph.qbl %[left_l1], %[left2] \n\t"
+ "preceu.ph.qbr %[left_r1], %[left2] \n\t"
+
+ "addu.ph %[average], %[average], %[above_l1] \n\t"
+ "addu.ph %[average], %[average], %[above_r1] \n\t"
+ "addu.ph %[average], %[average], %[left_l1] \n\t"
+ "addu.ph %[average], %[average], %[left_r1] \n\t"
+
+ "lw %[above1], 8(%[above]) \n\t"
+ "lw %[above2], 12(%[above]) \n\t"
+ "lw %[left1], 8(%[left]) \n\t"
+ "lw %[left2], 12(%[left]) \n\t"
+
+ "preceu.ph.qbl %[above_l1], %[above1] \n\t"
+ "preceu.ph.qbr %[above_r1], %[above1] \n\t"
+ "preceu.ph.qbl %[left_l1], %[left1] \n\t"
+ "preceu.ph.qbr %[left_r1], %[left1] \n\t"
+
+ "addu.ph %[average], %[average], %[above_l1] \n\t"
+ "addu.ph %[average], %[average], %[above_r1] \n\t"
+ "addu.ph %[average], %[average], %[left_l1] \n\t"
+ "addu.ph %[average], %[average], %[left_r1] \n\t"
+
+ "preceu.ph.qbl %[above_l1], %[above2] \n\t"
+ "preceu.ph.qbr %[above_r1], %[above2] \n\t"
+ "preceu.ph.qbl %[left_l1], %[left2] \n\t"
+ "preceu.ph.qbr %[left_r1], %[left2] \n\t"
+
+ "addu.ph %[average], %[average], %[above_l1] \n\t"
+ "addu.ph %[average], %[average], %[above_r1] \n\t"
+ "addu.ph %[average], %[average], %[left_l1] \n\t"
+ "addu.ph %[average], %[average], %[left_r1] \n\t"
+
+ "addiu %[average], %[average], 16 \n\t"
+ "srl %[tmp], %[average], 16 \n\t"
+ "addu.ph %[average], %[tmp], %[average] \n\t"
+ "srl %[expected_dc], %[average], 5 \n\t"
+ "replv.qb %[expected_dc], %[expected_dc] \n\t"
+
+ "sw %[expected_dc], (%[dst]) \n\t"
+ "sw %[expected_dc], 4(%[dst]) \n\t"
+ "sw %[expected_dc], 8(%[dst]) \n\t"
+ "sw %[expected_dc], 12(%[dst]) \n\t"
+
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[expected_dc], (%[dst]) \n\t"
+ "sw %[expected_dc], 4(%[dst]) \n\t"
+ "sw %[expected_dc], 8(%[dst]) \n\t"
+ "sw %[expected_dc], 12(%[dst]) \n\t"
+
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[expected_dc], (%[dst]) \n\t"
+ "sw %[expected_dc], 4(%[dst]) \n\t"
+ "sw %[expected_dc], 8(%[dst]) \n\t"
+ "sw %[expected_dc], 12(%[dst]) \n\t"
+
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[expected_dc], (%[dst]) \n\t"
+ "sw %[expected_dc], 4(%[dst]) \n\t"
+ "sw %[expected_dc], 8(%[dst]) \n\t"
+ "sw %[expected_dc], 12(%[dst]) \n\t"
+
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[expected_dc], (%[dst]) \n\t"
+ "sw %[expected_dc], 4(%[dst]) \n\t"
+ "sw %[expected_dc], 8(%[dst]) \n\t"
+ "sw %[expected_dc], 12(%[dst]) \n\t"
+
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[expected_dc], (%[dst]) \n\t"
+ "sw %[expected_dc], 4(%[dst]) \n\t"
+ "sw %[expected_dc], 8(%[dst]) \n\t"
+ "sw %[expected_dc], 12(%[dst]) \n\t"
+
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[expected_dc], (%[dst]) \n\t"
+ "sw %[expected_dc], 4(%[dst]) \n\t"
+ "sw %[expected_dc], 8(%[dst]) \n\t"
+ "sw %[expected_dc], 12(%[dst]) \n\t"
+
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[expected_dc], (%[dst]) \n\t"
+ "sw %[expected_dc], 4(%[dst]) \n\t"
+ "sw %[expected_dc], 8(%[dst]) \n\t"
+ "sw %[expected_dc], 12(%[dst]) \n\t"
+
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[expected_dc], (%[dst]) \n\t"
+ "sw %[expected_dc], 4(%[dst]) \n\t"
+ "sw %[expected_dc], 8(%[dst]) \n\t"
+ "sw %[expected_dc], 12(%[dst]) \n\t"
+
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[expected_dc], (%[dst]) \n\t"
+ "sw %[expected_dc], 4(%[dst]) \n\t"
+ "sw %[expected_dc], 8(%[dst]) \n\t"
+ "sw %[expected_dc], 12(%[dst]) \n\t"
+
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[expected_dc], (%[dst]) \n\t"
+ "sw %[expected_dc], 4(%[dst]) \n\t"
+ "sw %[expected_dc], 8(%[dst]) \n\t"
+ "sw %[expected_dc], 12(%[dst]) \n\t"
+
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[expected_dc], (%[dst]) \n\t"
+ "sw %[expected_dc], 4(%[dst]) \n\t"
+ "sw %[expected_dc], 8(%[dst]) \n\t"
+ "sw %[expected_dc], 12(%[dst]) \n\t"
+
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[expected_dc], (%[dst]) \n\t"
+ "sw %[expected_dc], 4(%[dst]) \n\t"
+ "sw %[expected_dc], 8(%[dst]) \n\t"
+ "sw %[expected_dc], 12(%[dst]) \n\t"
+
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[expected_dc], (%[dst]) \n\t"
+ "sw %[expected_dc], 4(%[dst]) \n\t"
+ "sw %[expected_dc], 8(%[dst]) \n\t"
+ "sw %[expected_dc], 12(%[dst]) \n\t"
+
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[expected_dc], (%[dst]) \n\t"
+ "sw %[expected_dc], 4(%[dst]) \n\t"
+ "sw %[expected_dc], 8(%[dst]) \n\t"
+ "sw %[expected_dc], 12(%[dst]) \n\t"
+
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[expected_dc], (%[dst]) \n\t"
+ "sw %[expected_dc], 4(%[dst]) \n\t"
+ "sw %[expected_dc], 8(%[dst]) \n\t"
+ "sw %[expected_dc], 12(%[dst]) \n\t"
+
+ : [left1] "=&r" (left1), [above1] "=&r" (above1),
+ [left_l1] "=&r" (left_l1), [above_l1] "=&r" (above_l1),
+ [left_r1] "=&r" (left_r1), [above_r1] "=&r" (above_r1),
+ [above2] "=&r" (above2), [left2] "=&r" (left2),
+ [average] "=&r" (average), [tmp] "=&r" (tmp),
+ [expected_dc] "=&r" (expected_dc)
+ : [above] "r" (above), [left] "r" (left),
+ [dst] "r" (dst), [stride] "r" (stride)
+ );
+}
+#endif // #if HAVE_DSPR2
--- /dev/null
+++ b/vpx_dsp/mips/intrapred4_dspr2.c
@@ -1,0 +1,245 @@
+/*
+ * Copyright (c) 2015 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "vpx_dsp/mips/common_dspr2.h"
+
+#if HAVE_DSPR2
+uint8_t vpx_ff_cropTbl_a[256 + 2 * CROP_WIDTH];
+uint8_t *vpx_ff_cropTbl;
+
+void vpx_dsputil_static_init(void) {+ int i;
+
+ for (i = 0; i < 256; i++) vpx_ff_cropTbl_a[i + CROP_WIDTH] = i;
+
+ for (i = 0; i < CROP_WIDTH; i++) {+ vpx_ff_cropTbl_a[i] = 0;
+ vpx_ff_cropTbl_a[i + CROP_WIDTH + 256] = 255;
+ }
+
+ vpx_ff_cropTbl = &vpx_ff_cropTbl_a[CROP_WIDTH];
+}
+
+void vp9_h_predictor_4x4_dspr2(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left) {+ int32_t tmp1, tmp2, tmp3, tmp4;
+
+ __asm__ __volatile__ (
+ "lb %[tmp1], (%[left]) \n\t"
+ "lb %[tmp2], 1(%[left]) \n\t"
+ "lb %[tmp3], 2(%[left]) \n\t"
+ "lb %[tmp4], 3(%[left]) \n\t"
+ "replv.qb %[tmp1], %[tmp1] \n\t"
+ "replv.qb %[tmp2], %[tmp2] \n\t"
+ "replv.qb %[tmp3], %[tmp3] \n\t"
+ "replv.qb %[tmp4], %[tmp4] \n\t"
+ "sw %[tmp1], (%[dst]) \n\t"
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[tmp2], (%[dst]) \n\t"
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[tmp3], (%[dst]) \n\t"
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[tmp4], (%[dst]) \n\t"
+
+ : [tmp1] "=&r" (tmp1), [tmp2] "=&r" (tmp2),
+ [tmp3] "=&r" (tmp3), [tmp4] "=&r" (tmp4)
+ : [left] "r" (left), [dst] "r" (dst), [stride] "r" (stride)
+ );
+}
+
+void vp9_dc_predictor_4x4_dspr2(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left) {+ int32_t expected_dc;
+ int32_t average;
+ int32_t tmp, above_c, above_l, above_r, left_c, left_r, left_l;
+
+ __asm__ __volatile__ (
+ "lw %[above_c], (%[above]) \n\t"
+ "lw %[left_c], (%[left]) \n\t"
+
+ "preceu.ph.qbl %[above_l], %[above_c] \n\t"
+ "preceu.ph.qbr %[above_r], %[above_c] \n\t"
+ "preceu.ph.qbl %[left_l], %[left_c] \n\t"
+ "preceu.ph.qbr %[left_r], %[left_c] \n\t"
+
+ "addu.ph %[average], %[above_r], %[above_l] \n\t"
+ "addu.ph %[average], %[average], %[left_l] \n\t"
+ "addu.ph %[average], %[average], %[left_r] \n\t"
+ "addiu %[average], %[average], 4 \n\t"
+ "srl %[tmp], %[average], 16 \n\t"
+ "addu.ph %[average], %[tmp], %[average] \n\t"
+ "srl %[expected_dc], %[average], 3 \n\t"
+ "replv.qb %[expected_dc], %[expected_dc] \n\t"
+
+ "sw %[expected_dc], (%[dst]) \n\t"
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[expected_dc], (%[dst]) \n\t"
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[expected_dc], (%[dst]) \n\t"
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[expected_dc], (%[dst]) \n\t"
+
+ : [above_c] "=&r" (above_c), [above_l] "=&r" (above_l),
+ [above_r] "=&r" (above_r), [left_c] "=&r" (left_c),
+ [left_l] "=&r" (left_l), [left_r] "=&r" (left_r),
+ [average] "=&r" (average), [tmp] "=&r" (tmp),
+ [expected_dc] "=&r" (expected_dc)
+ : [above] "r" (above), [left] "r" (left),
+ [dst] "r" (dst), [stride] "r" (stride)
+ );
+}
+
+void vp9_tm_predictor_4x4_dspr2(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left) {+ int32_t abovel, abover;
+ int32_t left0, left1, left2, left3;
+ int32_t res0, res1;
+ int32_t resl;
+ int32_t resr;
+ int32_t top_left;
+ uint8_t *cm = vpx_ff_cropTbl;
+
+ __asm__ __volatile__ (
+ "ulw %[resl], (%[above]) \n\t"
+
+ "lbu %[left0], (%[left]) \n\t"
+ "lbu %[left1], 1(%[left]) \n\t"
+ "lbu %[left2], 2(%[left]) \n\t"
+ "lbu %[left3], 3(%[left]) \n\t"
+
+ "lbu %[top_left], -1(%[above]) \n\t"
+
+ "preceu.ph.qbl %[abovel], %[resl] \n\t"
+ "preceu.ph.qbr %[abover], %[resl] \n\t"
+
+ "replv.ph %[left0], %[left0] \n\t"
+ "replv.ph %[left1], %[left1] \n\t"
+ "replv.ph %[left2], %[left2] \n\t"
+ "replv.ph %[left3], %[left3] \n\t"
+
+ "replv.ph %[top_left], %[top_left] \n\t"
+
+ "addu.ph %[resl], %[abovel], %[left0] \n\t"
+ "subu.ph %[resl], %[resl], %[top_left] \n\t"
+
+ "addu.ph %[resr], %[abover], %[left0] \n\t"
+ "subu.ph %[resr], %[resr], %[top_left] \n\t"
+
+ "sll %[res0], %[resr], 16 \n\t"
+ "sra %[res0], %[res0], 16 \n\t"
+ "lbux %[res0], %[res0](%[cm]) \n\t"
+
+ "sra %[res1], %[resr], 16 \n\t"
+ "lbux %[res1], %[res1](%[cm]) \n\t"
+ "sb %[res0], (%[dst]) \n\t"
+
+ "sll %[res0], %[resl], 16 \n\t"
+ "sra %[res0], %[res0], 16 \n\t"
+ "lbux %[res0], %[res0](%[cm]) \n\t"
+ "sb %[res1], 1(%[dst]) \n\t"
+
+ "sra %[res1], %[resl], 16 \n\t"
+ "lbux %[res1], %[res1](%[cm]) \n\t"
+
+ "addu.ph %[resl], %[abovel], %[left1] \n\t"
+ "subu.ph %[resl], %[resl], %[top_left] \n\t"
+
+ "addu.ph %[resr], %[abover], %[left1] \n\t"
+ "subu.ph %[resr], %[resr], %[top_left] \n\t"
+
+ "sb %[res0], 2(%[dst]) \n\t"
+ "sb %[res1], 3(%[dst]) \n\t"
+
+ "add %[dst], %[dst], %[stride] \n\t"
+
+ "sll %[res0], %[resr], 16 \n\t"
+ "sra %[res0], %[res0], 16 \n\t"
+ "lbux %[res0], %[res0](%[cm]) \n\t"
+
+ "sra %[res1], %[resr], 16 \n\t"
+ "lbux %[res1], %[res1](%[cm]) \n\t"
+ "sb %[res0], (%[dst]) \n\t"
+
+ "sll %[res0], %[resl], 16 \n\t"
+ "sra %[res0], %[res0], 16 \n\t"
+ "lbux %[res0], %[res0](%[cm]) \n\t"
+
+ "sb %[res1], 1(%[dst]) \n\t"
+ "sra %[res1], %[resl], 16 \n\t"
+ "lbux %[res1], %[res1](%[cm]) \n\t"
+
+ "addu.ph %[resl], %[abovel], %[left2] \n\t"
+ "subu.ph %[resl], %[resl], %[top_left] \n\t"
+
+ "addu.ph %[resr], %[abover], %[left2] \n\t"
+ "subu.ph %[resr], %[resr], %[top_left] \n\t"
+
+ "sb %[res0], 2(%[dst]) \n\t"
+ "sb %[res1], 3(%[dst]) \n\t"
+
+ "add %[dst], %[dst], %[stride] \n\t"
+
+ "sll %[res0], %[resr], 16 \n\t"
+ "sra %[res0], %[res0], 16 \n\t"
+ "lbux %[res0], %[res0](%[cm]) \n\t"
+
+
+ "sra %[res1], %[resr], 16 \n\t"
+ "lbux %[res1], %[res1](%[cm]) \n\t"
+ "sb %[res0], (%[dst]) \n\t"
+
+ "sll %[res0], %[resl], 16 \n\t"
+ "sra %[res0], %[res0], 16 \n\t"
+ "lbux %[res0], %[res0](%[cm]) \n\t"
+
+
+ "sb %[res1], 1(%[dst]) \n\t"
+ "sra %[res1], %[resl], 16 \n\t"
+ "lbux %[res1], %[res1](%[cm]) \n\t"
+
+ "addu.ph %[resl], %[abovel], %[left3] \n\t"
+ "subu.ph %[resl], %[resl], %[top_left] \n\t"
+
+ "addu.ph %[resr], %[abover], %[left3] \n\t"
+ "subu.ph %[resr], %[resr], %[top_left] \n\t"
+
+ "sb %[res0], 2(%[dst]) \n\t"
+ "sb %[res1], 3(%[dst]) \n\t"
+
+ "add %[dst], %[dst], %[stride] \n\t"
+
+ "sll %[res0], %[resr], 16 \n\t"
+ "sra %[res0], %[res0], 16 \n\t"
+ "lbux %[res0], %[res0](%[cm]) \n\t"
+
+ "sra %[res1], %[resr], 16 \n\t"
+ "lbux %[res1], %[res1](%[cm]) \n\t"
+ "sb %[res0], (%[dst]) \n\t"
+
+ "sll %[res0], %[resl], 16 \n\t"
+ "sra %[res0], %[res0], 16 \n\t"
+ "lbux %[res0], %[res0](%[cm]) \n\t"
+ "sb %[res1], 1(%[dst]) \n\t"
+
+ "sra %[res1], %[resl], 16 \n\t"
+ "lbux %[res1], %[res1](%[cm]) \n\t"
+
+ "sb %[res0], 2(%[dst]) \n\t"
+ "sb %[res1], 3(%[dst]) \n\t"
+
+ : [abovel] "=&r" (abovel), [abover] "=&r" (abover),
+ [left0] "=&r" (left0), [left1] "=&r" (left1), [left2] "=&r" (left2),
+ [res0] "=&r" (res0), [res1] "=&r" (res1), [left3] "=&r" (left3),
+ [resl] "=&r" (resl), [resr] "=&r" (resr), [top_left] "=&r" (top_left)
+ : [above] "r" (above), [left] "r" (left),
+ [dst] "r" (dst), [stride] "r" (stride), [cm] "r" (cm)
+ );
+}
+#endif // #if HAVE_DSPR2
--- /dev/null
+++ b/vpx_dsp/mips/intrapred8_dspr2.c
@@ -1,0 +1,607 @@
+/*
+ * Copyright (c) 2015 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "vpx_dsp/mips/common_dspr2.h"
+
+#if HAVE_DSPR2
+void vp9_h_predictor_8x8_dspr2(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left) {+ int32_t tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp8;
+
+ __asm__ __volatile__ (
+ "lb %[tmp1], (%[left]) \n\t"
+ "lb %[tmp2], 1(%[left]) \n\t"
+ "lb %[tmp3], 2(%[left]) \n\t"
+ "lb %[tmp4], 3(%[left]) \n\t"
+ "lb %[tmp5], 4(%[left]) \n\t"
+ "lb %[tmp6], 5(%[left]) \n\t"
+ "lb %[tmp7], 6(%[left]) \n\t"
+ "lb %[tmp8], 7(%[left]) \n\t"
+
+ "replv.qb %[tmp1], %[tmp1] \n\t"
+ "replv.qb %[tmp2], %[tmp2] \n\t"
+ "replv.qb %[tmp3], %[tmp3] \n\t"
+ "replv.qb %[tmp4], %[tmp4] \n\t"
+ "replv.qb %[tmp5], %[tmp5] \n\t"
+ "replv.qb %[tmp6], %[tmp6] \n\t"
+ "replv.qb %[tmp7], %[tmp7] \n\t"
+ "replv.qb %[tmp8], %[tmp8] \n\t"
+
+ "sw %[tmp1], (%[dst]) \n\t"
+ "sw %[tmp1], 4(%[dst]) \n\t"
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[tmp2], (%[dst]) \n\t"
+ "sw %[tmp2], 4(%[dst]) \n\t"
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[tmp3], (%[dst]) \n\t"
+ "sw %[tmp3], 4(%[dst]) \n\t"
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[tmp4], (%[dst]) \n\t"
+ "sw %[tmp4], 4(%[dst]) \n\t"
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[tmp5], (%[dst]) \n\t"
+ "sw %[tmp5], 4(%[dst]) \n\t"
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[tmp6], (%[dst]) \n\t"
+ "sw %[tmp6], 4(%[dst]) \n\t"
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[tmp7], (%[dst]) \n\t"
+ "sw %[tmp7], 4(%[dst]) \n\t"
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[tmp8], (%[dst]) \n\t"
+ "sw %[tmp8], 4(%[dst]) \n\t"
+
+ : [tmp1] "=&r" (tmp1), [tmp2] "=&r" (tmp2),
+ [tmp3] "=&r" (tmp3), [tmp4] "=&r" (tmp4),
+ [tmp5] "=&r" (tmp5), [tmp7] "=&r" (tmp7),
+ [tmp6] "=&r" (tmp6), [tmp8] "=&r" (tmp8)
+ : [left] "r" (left), [dst] "r" (dst),
+ [stride] "r" (stride)
+ );
+}
+
+void vp9_dc_predictor_8x8_dspr2(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left) {+ int32_t expected_dc;
+ int32_t average;
+ int32_t tmp, above1, above_l1, above_r1, left1, left_r1, left_l1;
+ int32_t above2, above_l2, above_r2, left2, left_r2, left_l2;
+
+ __asm__ __volatile__ (
+ "lw %[above1], (%[above]) \n\t"
+ "lw %[above2], 4(%[above]) \n\t"
+ "lw %[left1], (%[left]) \n\t"
+ "lw %[left2], 4(%[left]) \n\t"
+
+ "preceu.ph.qbl %[above_l1], %[above1] \n\t"
+ "preceu.ph.qbr %[above_r1], %[above1] \n\t"
+ "preceu.ph.qbl %[left_l1], %[left1] \n\t"
+ "preceu.ph.qbr %[left_r1], %[left1] \n\t"
+
+ "preceu.ph.qbl %[above_l2], %[above2] \n\t"
+ "preceu.ph.qbr %[above_r2], %[above2] \n\t"
+ "preceu.ph.qbl %[left_l2], %[left2] \n\t"
+ "preceu.ph.qbr %[left_r2], %[left2] \n\t"
+
+ "addu.ph %[average], %[above_r1], %[above_l1] \n\t"
+ "addu.ph %[average], %[average], %[left_l1] \n\t"
+ "addu.ph %[average], %[average], %[left_r1] \n\t"
+
+ "addu.ph %[average], %[average], %[above_l2] \n\t"
+ "addu.ph %[average], %[average], %[above_r2] \n\t"
+ "addu.ph %[average], %[average], %[left_l2] \n\t"
+ "addu.ph %[average], %[average], %[left_r2] \n\t"
+
+ "addiu %[average], %[average], 8 \n\t"
+
+ "srl %[tmp], %[average], 16 \n\t"
+ "addu.ph %[average], %[tmp], %[average] \n\t"
+ "srl %[expected_dc], %[average], 4 \n\t"
+ "replv.qb %[expected_dc], %[expected_dc] \n\t"
+
+ "sw %[expected_dc], (%[dst]) \n\t"
+ "sw %[expected_dc], 4(%[dst]) \n\t"
+
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[expected_dc], (%[dst]) \n\t"
+ "sw %[expected_dc], 4(%[dst]) \n\t"
+
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[expected_dc], (%[dst]) \n\t"
+ "sw %[expected_dc], 4(%[dst]) \n\t"
+
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[expected_dc], (%[dst]) \n\t"
+ "sw %[expected_dc], 4(%[dst]) \n\t"
+
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[expected_dc], (%[dst]) \n\t"
+ "sw %[expected_dc], 4(%[dst]) \n\t"
+
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[expected_dc], (%[dst]) \n\t"
+ "sw %[expected_dc], 4(%[dst]) \n\t"
+
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[expected_dc], (%[dst]) \n\t"
+ "sw %[expected_dc], 4(%[dst]) \n\t"
+
+ "add %[dst], %[dst], %[stride] \n\t"
+ "sw %[expected_dc], (%[dst]) \n\t"
+ "sw %[expected_dc], 4(%[dst]) \n\t"
+
+ : [above1] "=&r" (above1), [above_l1] "=&r" (above_l1),
+ [above_r1] "=&r" (above_r1), [left1] "=&r" (left1),
+ [left_l1] "=&r" (left_l1), [left_r1] "=&r" (left_r1),
+ [above2] "=&r" (above2), [above_l2] "=&r" (above_l2),
+ [above_r2] "=&r" (above_r2), [left2] "=&r" (left2),
+ [left_l2] "=&r" (left_l2), [left_r2] "=&r" (left_r2),
+ [average] "=&r" (average), [tmp] "=&r" (tmp),
+ [expected_dc] "=&r" (expected_dc)
+ : [above] "r" (above), [left] "r" (left), [dst] "r" (dst),
+ [stride] "r" (stride)
+ );
+}
+
+void vp9_tm_predictor_8x8_dspr2(uint8_t *dst, ptrdiff_t stride,
+ const uint8_t *above, const uint8_t *left) {+ int32_t abovel, abover;
+ int32_t abovel_1, abover_1;
+ int32_t left0;
+ int32_t res0, res1, res2, res3;
+ int32_t reshw;
+ int32_t top_left;
+ uint8_t *cm = vpx_ff_cropTbl;
+
+ __asm__ __volatile__ (
+ "ulw %[reshw], (%[above]) \n\t"
+ "ulw %[top_left], 4(%[above]) \n\t"
+
+ "lbu %[left0], (%[left]) \n\t"
+
+ "preceu.ph.qbl %[abovel], %[reshw] \n\t"
+ "preceu.ph.qbr %[abover], %[reshw] \n\t"
+ "preceu.ph.qbl %[abovel_1], %[top_left] \n\t"
+ "preceu.ph.qbr %[abover_1], %[top_left] \n\t"
+
+ "lbu %[top_left], -1(%[above]) \n\t"
+ "replv.ph %[left0], %[left0] \n\t"
+
+ "replv.ph %[top_left], %[top_left] \n\t"
+
+ "addu.ph %[reshw], %[abovel], %[left0] \n\t"
+ "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
+
+ "sll %[res2], %[reshw], 16 \n\t"
+ "sra %[res2], %[res2], 16 \n\t"
+ "sra %[res3], %[reshw], 16 \n\t"
+
+ "addu.ph %[reshw], %[abover], %[left0] \n\t"
+ "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
+
+ "sll %[res0], %[reshw], 16 \n\t"
+ "sra %[res0], %[res0], 16 \n\t"
+ "sra %[res1], %[reshw], 16 \n\t"
+
+ "lbux %[res0], %[res0](%[cm]) \n\t"
+ "lbux %[res1], %[res1](%[cm]) \n\t"
+ "lbux %[res2], %[res2](%[cm]) \n\t"
+ "lbux %[res3], %[res3](%[cm]) \n\t"
+
+ "sb %[res0], (%[dst]) \n\t"
+ "sb %[res1], 1(%[dst]) \n\t"
+ "sb %[res2], 2(%[dst]) \n\t"
+ "sb %[res3], 3(%[dst]) \n\t"
+
+ "addu.ph %[reshw], %[abovel_1], %[left0] \n\t"
+ "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
+
+ "sll %[res2], %[reshw], 16 \n\t"
+ "sra %[res2], %[res2], 16 \n\t"
+ "sra %[res3], %[reshw], 16 \n\t"
+
+ "addu.ph %[reshw], %[abover_1], %[left0] \n\t"
+ "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
+
+ "sll %[res0], %[reshw], 16 \n\t"
+ "sra %[res0], %[res0], 16 \n\t"
+ "sra %[res1], %[reshw], 16 \n\t"
+
+ "lbu %[left0], 1(%[left]) \n\t"
+
+ "lbux %[res0], %[res0](%[cm]) \n\t"
+ "lbux %[res1], %[res1](%[cm]) \n\t"
+ "lbux %[res2], %[res2](%[cm]) \n\t"
+ "lbux %[res3], %[res3](%[cm]) \n\t"
+
+ "sb %[res0], 4(%[dst]) \n\t"
+ "sb %[res1], 5(%[dst]) \n\t"
+ "sb %[res2], 6(%[dst]) \n\t"
+ "sb %[res3], 7(%[dst]) \n\t"
+
+ "replv.ph %[left0], %[left0] \n\t"
+ "add %[dst], %[dst], %[stride] \n\t"
+
+ "addu.ph %[reshw], %[abovel], %[left0] \n\t"
+ "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
+
+ "sll %[res2], %[reshw], 16 \n\t"
+ "sra %[res2], %[res2], 16 \n\t"
+ "sra %[res3], %[reshw], 16 \n\t"
+
+ "addu.ph %[reshw], %[abover], %[left0] \n\t"
+ "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
+
+ "sll %[res0], %[reshw], 16 \n\t"
+ "sra %[res0], %[res0], 16 \n\t"
+ "sra %[res1], %[reshw], 16 \n\t"
+
+ "lbux %[res0], %[res0](%[cm]) \n\t"
+ "lbux %[res1], %[res1](%[cm]) \n\t"
+ "lbux %[res2], %[res2](%[cm]) \n\t"
+ "lbux %[res3], %[res3](%[cm]) \n\t"
+
+ "sb %[res0], (%[dst]) \n\t"
+ "sb %[res1], 1(%[dst]) \n\t"
+ "sb %[res2], 2(%[dst]) \n\t"
+ "sb %[res3], 3(%[dst]) \n\t"
+
+ "addu.ph %[reshw], %[abovel_1], %[left0] \n\t"
+ "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
+
+ "sll %[res2], %[reshw], 16 \n\t"
+ "sra %[res2], %[res2], 16 \n\t"
+ "sra %[res3], %[reshw], 16 \n\t"
+
+ "addu.ph %[reshw], %[abover_1], %[left0] \n\t"
+ "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
+
+ "sll %[res0], %[reshw], 16 \n\t"
+ "sra %[res0], %[res0], 16 \n\t"
+ "sra %[res1], %[reshw], 16 \n\t"
+
+ "lbu %[left0], 2(%[left]) \n\t"
+
+ "lbux %[res0], %[res0](%[cm]) \n\t"
+ "lbux %[res1], %[res1](%[cm]) \n\t"
+ "lbux %[res2], %[res2](%[cm]) \n\t"
+ "lbux %[res3], %[res3](%[cm]) \n\t"
+
+ "sb %[res0], 4(%[dst]) \n\t"
+ "sb %[res1], 5(%[dst]) \n\t"
+ "sb %[res2], 6(%[dst]) \n\t"
+ "sb %[res3], 7(%[dst]) \n\t"
+
+ "replv.ph %[left0], %[left0] \n\t"
+ "add %[dst], %[dst], %[stride] \n\t"
+
+ "addu.ph %[reshw], %[abovel], %[left0] \n\t"
+ "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
+
+ "sll %[res2], %[reshw], 16 \n\t"
+ "sra %[res2], %[res2], 16 \n\t"
+ "sra %[res3], %[reshw], 16 \n\t"
+
+ "addu.ph %[reshw], %[abover], %[left0] \n\t"
+ "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
+
+ "sll %[res0], %[reshw], 16 \n\t"
+ "sra %[res0], %[res0], 16 \n\t"
+ "sra %[res1], %[reshw], 16 \n\t"
+
+ "lbux %[res0], %[res0](%[cm]) \n\t"
+ "lbux %[res1], %[res1](%[cm]) \n\t"
+ "lbux %[res2], %[res2](%[cm]) \n\t"
+ "lbux %[res3], %[res3](%[cm]) \n\t"
+
+ "sb %[res0], (%[dst]) \n\t"
+ "sb %[res1], 1(%[dst]) \n\t"
+ "sb %[res2], 2(%[dst]) \n\t"
+ "sb %[res3], 3(%[dst]) \n\t"
+
+ "addu.ph %[reshw], %[abovel_1], %[left0] \n\t"
+ "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
+
+ "sll %[res2], %[reshw], 16 \n\t"
+ "sra %[res2], %[res2], 16 \n\t"
+ "sra %[res3], %[reshw], 16 \n\t"
+
+ "addu.ph %[reshw], %[abover_1], %[left0] \n\t"
+ "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
+
+ "sll %[res0], %[reshw], 16 \n\t"
+ "sra %[res0], %[res0], 16 \n\t"
+ "sra %[res1], %[reshw], 16 \n\t"
+
+ "lbu %[left0], 3(%[left]) \n\t"
+
+ "lbux %[res0], %[res0](%[cm]) \n\t"
+ "lbux %[res1], %[res1](%[cm]) \n\t"
+ "lbux %[res2], %[res2](%[cm]) \n\t"
+ "lbux %[res3], %[res3](%[cm]) \n\t"
+
+ "sb %[res0], 4(%[dst]) \n\t"
+ "sb %[res1], 5(%[dst]) \n\t"
+ "sb %[res2], 6(%[dst]) \n\t"
+ "sb %[res3], 7(%[dst]) \n\t"
+
+ "replv.ph %[left0], %[left0] \n\t"
+ "add %[dst], %[dst], %[stride] \n\t"
+
+ "addu.ph %[reshw], %[abovel], %[left0] \n\t"
+ "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
+
+ "sll %[res2], %[reshw], 16 \n\t"
+ "sra %[res2], %[res2], 16 \n\t"
+ "sra %[res3], %[reshw], 16 \n\t"
+
+ "addu.ph %[reshw], %[abover], %[left0] \n\t"
+ "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
+
+ "sll %[res0], %[reshw], 16 \n\t"
+ "sra %[res0], %[res0], 16 \n\t"
+ "sra %[res1], %[reshw], 16 \n\t"
+
+ "lbux %[res0], %[res0](%[cm]) \n\t"
+ "lbux %[res1], %[res1](%[cm]) \n\t"
+ "lbux %[res2], %[res2](%[cm]) \n\t"
+ "lbux %[res3], %[res3](%[cm]) \n\t"
+
+ "sb %[res0], (%[dst]) \n\t"
+ "sb %[res1], 1(%[dst]) \n\t"
+ "sb %[res2], 2(%[dst]) \n\t"
+ "sb %[res3], 3(%[dst]) \n\t"
+
+ "addu.ph %[reshw], %[abovel_1], %[left0] \n\t"
+ "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
+
+ "sll %[res2], %[reshw], 16 \n\t"
+ "sra %[res2], %[res2], 16 \n\t"
+ "sra %[res3], %[reshw], 16 \n\t"
+
+ "addu.ph %[reshw], %[abover_1], %[left0] \n\t"
+ "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
+
+ "sll %[res0], %[reshw], 16 \n\t"
+ "sra %[res0], %[res0], 16 \n\t"
+ "sra %[res1], %[reshw], 16 \n\t"
+
+ "lbu %[left0], 4(%[left]) \n\t"
+
+ "lbux %[res0], %[res0](%[cm]) \n\t"
+ "lbux %[res1], %[res1](%[cm]) \n\t"
+ "lbux %[res2], %[res2](%[cm]) \n\t"
+ "lbux %[res3], %[res3](%[cm]) \n\t"
+
+ "sb %[res0], 4(%[dst]) \n\t"
+ "sb %[res1], 5(%[dst]) \n\t"
+ "sb %[res2], 6(%[dst]) \n\t"
+ "sb %[res3], 7(%[dst]) \n\t"
+
+ "replv.ph %[left0], %[left0] \n\t"
+ "add %[dst], %[dst], %[stride] \n\t"
+
+ "addu.ph %[reshw], %[abovel], %[left0] \n\t"
+ "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
+
+ "sll %[res2], %[reshw], 16 \n\t"
+ "sra %[res2], %[res2], 16 \n\t"
+ "sra %[res3], %[reshw], 16 \n\t"
+
+ "addu.ph %[reshw], %[abover], %[left0] \n\t"
+ "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
+
+ "sll %[res0], %[reshw], 16 \n\t"
+ "sra %[res0], %[res0], 16 \n\t"
+ "sra %[res1], %[reshw], 16 \n\t"
+
+ "lbux %[res0], %[res0](%[cm]) \n\t"
+ "lbux %[res1], %[res1](%[cm]) \n\t"
+ "lbux %[res2], %[res2](%[cm]) \n\t"
+ "lbux %[res3], %[res3](%[cm]) \n\t"
+
+ "sb %[res0], (%[dst]) \n\t"
+ "sb %[res1], 1(%[dst]) \n\t"
+ "sb %[res2], 2(%[dst]) \n\t"
+ "sb %[res3], 3(%[dst]) \n\t"
+
+ "addu.ph %[reshw], %[abovel_1], %[left0] \n\t"
+ "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
+
+ "sll %[res2], %[reshw], 16 \n\t"
+ "sra %[res2], %[res2], 16 \n\t"
+ "sra %[res3], %[reshw], 16 \n\t"
+
+ "addu.ph %[reshw], %[abover_1], %[left0] \n\t"
+ "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
+
+ "sll %[res0], %[reshw], 16 \n\t"
+ "sra %[res0], %[res0], 16 \n\t"
+ "sra %[res1], %[reshw], 16 \n\t"
+
+ "lbu %[left0], 5(%[left]) \n\t"
+
+ "lbux %[res0], %[res0](%[cm]) \n\t"
+ "lbux %[res1], %[res1](%[cm]) \n\t"
+ "lbux %[res2], %[res2](%[cm]) \n\t"
+ "lbux %[res3], %[res3](%[cm]) \n\t"
+
+ "sb %[res0], 4(%[dst]) \n\t"
+ "sb %[res1], 5(%[dst]) \n\t"
+ "sb %[res2], 6(%[dst]) \n\t"
+ "sb %[res3], 7(%[dst]) \n\t"
+
+ "replv.ph %[left0], %[left0] \n\t"
+ "add %[dst], %[dst], %[stride] \n\t"
+
+ "addu.ph %[reshw], %[abovel], %[left0] \n\t"
+ "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
+
+ "sll %[res2], %[reshw], 16 \n\t"
+ "sra %[res2], %[res2], 16 \n\t"
+ "sra %[res3], %[reshw], 16 \n\t"
+
+ "addu.ph %[reshw], %[abover], %[left0] \n\t"
+ "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
+
+ "sll %[res0], %[reshw], 16 \n\t"
+ "sra %[res0], %[res0], 16 \n\t"
+ "sra %[res1], %[reshw], 16 \n\t"
+
+ "lbux %[res0], %[res0](%[cm]) \n\t"
+ "lbux %[res1], %[res1](%[cm]) \n\t"
+ "lbux %[res2], %[res2](%[cm]) \n\t"
+ "lbux %[res3], %[res3](%[cm]) \n\t"
+
+ "sb %[res0], (%[dst]) \n\t"
+ "sb %[res1], 1(%[dst]) \n\t"
+ "sb %[res2], 2(%[dst]) \n\t"
+ "sb %[res3], 3(%[dst]) \n\t"
+
+ "addu.ph %[reshw], %[abovel_1], %[left0] \n\t"
+ "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
+
+ "sll %[res2], %[reshw], 16 \n\t"
+ "sra %[res2], %[res2], 16 \n\t"
+ "sra %[res3], %[reshw], 16 \n\t"
+
+ "addu.ph %[reshw], %[abover_1], %[left0] \n\t"
+ "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
+
+ "sll %[res0], %[reshw], 16 \n\t"
+ "sra %[res0], %[res0], 16 \n\t"
+ "sra %[res1], %[reshw], 16 \n\t"
+
+ "lbu %[left0], 6(%[left]) \n\t"
+
+ "lbux %[res0], %[res0](%[cm]) \n\t"
+ "lbux %[res1], %[res1](%[cm]) \n\t"
+ "lbux %[res2], %[res2](%[cm]) \n\t"
+ "lbux %[res3], %[res3](%[cm]) \n\t"
+
+ "sb %[res0], 4(%[dst]) \n\t"
+ "sb %[res1], 5(%[dst]) \n\t"
+ "sb %[res2], 6(%[dst]) \n\t"
+ "sb %[res3], 7(%[dst]) \n\t"
+
+ "replv.ph %[left0], %[left0] \n\t"
+ "add %[dst], %[dst], %[stride] \n\t"
+
+ "addu.ph %[reshw], %[abovel], %[left0] \n\t"
+ "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
+
+ "sll %[res2], %[reshw], 16 \n\t"
+ "sra %[res2], %[res2], 16 \n\t"
+ "sra %[res3], %[reshw], 16 \n\t"
+
+ "addu.ph %[reshw], %[abover], %[left0] \n\t"
+ "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
+
+ "sll %[res0], %[reshw], 16 \n\t"
+ "sra %[res0], %[res0], 16 \n\t"
+ "sra %[res1], %[reshw], 16 \n\t"
+
+ "lbux %[res0], %[res0](%[cm]) \n\t"
+ "lbux %[res1], %[res1](%[cm]) \n\t"
+ "lbux %[res2], %[res2](%[cm]) \n\t"
+ "lbux %[res3], %[res3](%[cm]) \n\t"
+
+ "sb %[res0], (%[dst]) \n\t"
+ "sb %[res1], 1(%[dst]) \n\t"
+ "sb %[res2], 2(%[dst]) \n\t"
+ "sb %[res3], 3(%[dst]) \n\t"
+
+ "addu.ph %[reshw], %[abovel_1], %[left0] \n\t"
+ "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
+
+ "sll %[res2], %[reshw], 16 \n\t"
+ "sra %[res2], %[res2], 16 \n\t"
+ "sra %[res3], %[reshw], 16 \n\t"
+
+ "addu.ph %[reshw], %[abover_1], %[left0] \n\t"
+ "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
+
+ "sll %[res0], %[reshw], 16 \n\t"
+ "sra %[res0], %[res0], 16 \n\t"
+ "sra %[res1], %[reshw], 16 \n\t"
+
+ "lbu %[left0], 7(%[left]) \n\t"
+
+ "lbux %[res0], %[res0](%[cm]) \n\t"
+ "lbux %[res1], %[res1](%[cm]) \n\t"
+ "lbux %[res2], %[res2](%[cm]) \n\t"
+ "lbux %[res3], %[res3](%[cm]) \n\t"
+
+ "sb %[res0], 4(%[dst]) \n\t"
+ "sb %[res1], 5(%[dst]) \n\t"
+ "sb %[res2], 6(%[dst]) \n\t"
+ "sb %[res3], 7(%[dst]) \n\t"
+
+ "replv.ph %[left0], %[left0] \n\t"
+ "add %[dst], %[dst], %[stride] \n\t"
+
+ "addu.ph %[reshw], %[abovel], %[left0] \n\t"
+ "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
+
+ "sll %[res2], %[reshw], 16 \n\t"
+ "sra %[res2], %[res2], 16 \n\t"
+ "sra %[res3], %[reshw], 16 \n\t"
+
+ "addu.ph %[reshw], %[abover], %[left0] \n\t"
+ "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
+
+ "sll %[res0], %[reshw], 16 \n\t"
+ "sra %[res0], %[res0], 16 \n\t"
+ "sra %[res1], %[reshw], 16 \n\t"
+
+ "lbux %[res0], %[res0](%[cm]) \n\t"
+ "lbux %[res1], %[res1](%[cm]) \n\t"
+ "lbux %[res2], %[res2](%[cm]) \n\t"
+ "lbux %[res3], %[res3](%[cm]) \n\t"
+
+ "sb %[res0], (%[dst]) \n\t"
+ "sb %[res1], 1(%[dst]) \n\t"
+ "sb %[res2], 2(%[dst]) \n\t"
+ "sb %[res3], 3(%[dst]) \n\t"
+
+ "addu.ph %[reshw], %[abovel_1], %[left0] \n\t"
+ "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
+
+ "sll %[res2], %[reshw], 16 \n\t"
+ "sra %[res2], %[res2], 16 \n\t"
+ "sra %[res3], %[reshw], 16 \n\t"
+
+ "addu.ph %[reshw], %[abover_1], %[left0] \n\t"
+ "subu.ph %[reshw], %[reshw], %[top_left] \n\t"
+
+ "sll %[res0], %[reshw], 16 \n\t"
+ "sra %[res0], %[res0], 16 \n\t"
+ "sra %[res1], %[reshw], 16 \n\t"
+
+ "lbux %[res0], %[res0](%[cm]) \n\t"
+ "lbux %[res1], %[res1](%[cm]) \n\t"
+ "lbux %[res2], %[res2](%[cm]) \n\t"
+ "lbux %[res3], %[res3](%[cm]) \n\t"
+
+ "sb %[res0], 4(%[dst]) \n\t"
+ "sb %[res1], 5(%[dst]) \n\t"
+ "sb %[res2], 6(%[dst]) \n\t"
+ "sb %[res3], 7(%[dst]) \n\t"
+
+ : [abovel] "=&r" (abovel), [abover] "=&r" (abover),
+ [abovel_1] "=&r" (abovel_1), [abover_1] "=&r" (abover_1),
+ [left0] "=&r" (left0), [res2] "=&r" (res2), [res3] "=&r" (res3),
+ [res0] "=&r" (res0), [res1] "=&r" (res1),
+ [reshw] "=&r" (reshw), [top_left] "=&r" (top_left)
+ : [above] "r" (above), [left] "r" (left),
+ [dst] "r" (dst), [stride] "r" (stride), [cm] "r" (cm)
+ );
+}
+#endif // #if HAVE_DSPR2
--- /dev/null
+++ b/vpx_dsp/mips/intrapred_msa.c
@@ -1,0 +1,737 @@
+/*
+ * Copyright (c) 2015 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "./vpx_dsp_rtcd.h"
+#include "vpx_dsp/mips/macros_msa.h"
+
+#define IPRED_SUBS_UH2_UH(in0, in1, out0, out1) { \+ out0 = __msa_subs_u_h(out0, in0); \
+ out1 = __msa_subs_u_h(out1, in1); \
+}
+
+static void intra_predict_vert_4x4_msa(const uint8_t *src, uint8_t *dst,
+ int32_t dst_stride) {+ uint32_t src_data;
+
+ src_data = LW(src);
+
+ SW4(src_data, src_data, src_data, src_data, dst, dst_stride);
+}
+
+static void intra_predict_vert_8x8_msa(const uint8_t *src, uint8_t *dst,
+ int32_t dst_stride) {+ uint32_t row;
+ uint32_t src_data1, src_data2;
+
+ src_data1 = LW(src);
+ src_data2 = LW(src + 4);
+
+ for (row = 8; row--;) {+ SW(src_data1, dst);
+ SW(src_data2, (dst + 4));
+ dst += dst_stride;
+ }
+}
+
+static void intra_predict_vert_16x16_msa(const uint8_t *src, uint8_t *dst,
+ int32_t dst_stride) {+ uint32_t row;
+ v16u8 src0;
+
+ src0 = LD_UB(src);
+
+ for (row = 16; row--;) {+ ST_UB(src0, dst);
+ dst += dst_stride;
+ }
+}
+
+static void intra_predict_vert_32x32_msa(const uint8_t *src, uint8_t *dst,
+ int32_t dst_stride) {+ uint32_t row;
+ v16u8 src1, src2;
+
+ src1 = LD_UB(src);
+ src2 = LD_UB(src + 16);
+
+ for (row = 32; row--;) {+ ST_UB2(src1, src2, dst, 16);
+ dst += dst_stride;
+ }
+}
+
+static void intra_predict_horiz_4x4_msa(const uint8_t *src, uint8_t *dst,
+ int32_t dst_stride) {+ uint32_t out0, out1, out2, out3;
+
+ out0 = src[0] * 0x01010101;
+ out1 = src[1] * 0x01010101;
+ out2 = src[2] * 0x01010101;
+ out3 = src[3] * 0x01010101;
+
+ SW4(out0, out1, out2, out3, dst, dst_stride);
+}
+
+static void intra_predict_horiz_8x8_msa(const uint8_t *src, uint8_t *dst,
+ int32_t dst_stride) {+ uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+
+ out0 = src[0] * 0x0101010101010101ull;
+ out1 = src[1] * 0x0101010101010101ull;
+ out2 = src[2] * 0x0101010101010101ull;
+ out3 = src[3] * 0x0101010101010101ull;
+ out4 = src[4] * 0x0101010101010101ull;
+ out5 = src[5] * 0x0101010101010101ull;
+ out6 = src[6] * 0x0101010101010101ull;
+ out7 = src[7] * 0x0101010101010101ull;
+
+ SD4(out0, out1, out2, out3, dst, dst_stride);
+ dst += (4 * dst_stride);
+ SD4(out4, out5, out6, out7, dst, dst_stride);
+}
+
+static void intra_predict_horiz_16x16_msa(const uint8_t *src, uint8_t *dst,
+ int32_t dst_stride) {+ uint32_t row;
+ uint8_t inp0, inp1, inp2, inp3;
+ v16u8 src0, src1, src2, src3;
+
+ for (row = 4; row--;) {+ inp0 = src[0];
+ inp1 = src[1];
+ inp2 = src[2];
+ inp3 = src[3];
+ src += 4;
+
+ src0 = (v16u8)__msa_fill_b(inp0);
+ src1 = (v16u8)__msa_fill_b(inp1);
+ src2 = (v16u8)__msa_fill_b(inp2);
+ src3 = (v16u8)__msa_fill_b(inp3);
+
+ ST_UB4(src0, src1, src2, src3, dst, dst_stride);
+ dst += (4 * dst_stride);
+ }
+}
+
+static void intra_predict_horiz_32x32_msa(const uint8_t *src, uint8_t *dst,
+ int32_t dst_stride) {+ uint32_t row;
+ uint8_t inp0, inp1, inp2, inp3;
+ v16u8 src0, src1, src2, src3;
+
+ for (row = 8; row--;) {+ inp0 = src[0];
+ inp1 = src[1];
+ inp2 = src[2];
+ inp3 = src[3];
+ src += 4;
+
+ src0 = (v16u8)__msa_fill_b(inp0);
+ src1 = (v16u8)__msa_fill_b(inp1);
+ src2 = (v16u8)__msa_fill_b(inp2);
+ src3 = (v16u8)__msa_fill_b(inp3);
+
+ ST_UB2(src0, src0, dst, 16);
+ dst += dst_stride;
+ ST_UB2(src1, src1, dst, 16);
+ dst += dst_stride;
+ ST_UB2(src2, src2, dst, 16);
+ dst += dst_stride;
+ ST_UB2(src3, src3, dst, 16);
+ dst += dst_stride;
+ }
+}
+
+static void intra_predict_dc_4x4_msa(const uint8_t *src_top,
+ const uint8_t *src_left,
+ uint8_t *dst, int32_t dst_stride) {+ uint32_t val0, val1;
+ v16i8 store, src = { 0 };+ v8u16 sum_h;
+ v4u32 sum_w;
+ v2u64 sum_d;
+
+ val0 = LW(src_top);
+ val1 = LW(src_left);
+ INSERT_W2_SB(val0, val1, src);
+ sum_h = __msa_hadd_u_h((v16u8)src, (v16u8)src);
+ sum_w = __msa_hadd_u_w(sum_h, sum_h);
+ sum_d = __msa_hadd_u_d(sum_w, sum_w);
+ sum_w = (v4u32)__msa_srari_w((v4i32)sum_d, 3);
+ store = __msa_splati_b((v16i8)sum_w, 0);
+ val0 = __msa_copy_u_w((v4i32)store, 0);
+
+ SW4(val0, val0, val0, val0, dst, dst_stride);
+}
+
+static void intra_predict_dc_tl_4x4_msa(const uint8_t *src, uint8_t *dst,
+ int32_t dst_stride) {+ uint32_t val0;
+ v16i8 store, data = { 0 };+ v8u16 sum_h;
+ v4u32 sum_w;
+
+ val0 = LW(src);
+ data = (v16i8)__msa_insert_w((v4i32)data, 0, val0);
+ sum_h = __msa_hadd_u_h((v16u8)data, (v16u8)data);
+ sum_w = __msa_hadd_u_w(sum_h, sum_h);
+ sum_w = (v4u32)__msa_srari_w((v4i32)sum_w, 2);
+ store = __msa_splati_b((v16i8)sum_w, 0);
+ val0 = __msa_copy_u_w((v4i32)store, 0);
+
+ SW4(val0, val0, val0, val0, dst, dst_stride);
+}
+
+static void intra_predict_128dc_4x4_msa(uint8_t *dst, int32_t dst_stride) {+ uint32_t out;
+ const v16i8 store = __msa_ldi_b(128);
+
+ out = __msa_copy_u_w((v4i32)store, 0);
+
+ SW4(out, out, out, out, dst, dst_stride);
+}
+
+static void intra_predict_dc_8x8_msa(const uint8_t *src_top,
+ const uint8_t *src_left,
+ uint8_t *dst, int32_t dst_stride) {+ uint64_t val0, val1;
+ v16i8 store;
+ v16u8 src = { 0 };+ v8u16 sum_h;
+ v4u32 sum_w;
+ v2u64 sum_d;
+
+ val0 = LD(src_top);
+ val1 = LD(src_left);
+ INSERT_D2_UB(val0, val1, src);
+ sum_h = __msa_hadd_u_h(src, src);
+ sum_w = __msa_hadd_u_w(sum_h, sum_h);
+ sum_d = __msa_hadd_u_d(sum_w, sum_w);
+ sum_w = (v4u32)__msa_pckev_w((v4i32)sum_d, (v4i32)sum_d);
+ sum_d = __msa_hadd_u_d(sum_w, sum_w);
+ sum_w = (v4u32)__msa_srari_w((v4i32)sum_d, 4);
+ store = __msa_splati_b((v16i8)sum_w, 0);
+ val0 = __msa_copy_u_d((v2i64)store, 0);
+
+ SD4(val0, val0, val0, val0, dst, dst_stride);
+ dst += (4 * dst_stride);
+ SD4(val0, val0, val0, val0, dst, dst_stride);
+}
+
+static void intra_predict_dc_tl_8x8_msa(const uint8_t *src, uint8_t *dst,
+ int32_t dst_stride) {+ uint64_t val0;
+ v16i8 store;
+ v16u8 data = { 0 };+ v8u16 sum_h;
+ v4u32 sum_w;
+ v2u64 sum_d;
+
+ val0 = LD(src);
+ data = (v16u8)__msa_insert_d((v2i64)data, 0, val0);
+ sum_h = __msa_hadd_u_h(data, data);
+ sum_w = __msa_hadd_u_w(sum_h, sum_h);
+ sum_d = __msa_hadd_u_d(sum_w, sum_w);
+ sum_w = (v4u32)__msa_srari_w((v4i32)sum_d, 3);
+ store = __msa_splati_b((v16i8)sum_w, 0);
+ val0 = __msa_copy_u_d((v2i64)store, 0);
+
+ SD4(val0, val0, val0, val0, dst, dst_stride);
+ dst += (4 * dst_stride);
+ SD4(val0, val0, val0, val0, dst, dst_stride);
+}
+
+static void intra_predict_128dc_8x8_msa(uint8_t *dst, int32_t dst_stride) {+ uint64_t out;
+ const v16i8 store = __msa_ldi_b(128);
+
+ out = __msa_copy_u_d((v2i64)store, 0);
+
+ SD4(out, out, out, out, dst, dst_stride);
+ dst += (4 * dst_stride);
+ SD4(out, out, out, out, dst, dst_stride);
+}
+
+static void intra_predict_dc_16x16_msa(const uint8_t *src_top,
+ const uint8_t *src_left,
+ uint8_t *dst, int32_t dst_stride) {+ v16u8 top, left, out;
+ v8u16 sum_h, sum_top, sum_left;
+ v4u32 sum_w;
+ v2u64 sum_d;
+
+ top = LD_UB(src_top);
+ left = LD_UB(src_left);
+ HADD_UB2_UH(top, left, sum_top, sum_left);
+ sum_h = sum_top + sum_left;
+ sum_w = __msa_hadd_u_w(sum_h, sum_h);
+ sum_d = __msa_hadd_u_d(sum_w, sum_w);
+ sum_w = (v4u32)__msa_pckev_w((v4i32)sum_d, (v4i32)sum_d);
+ sum_d = __msa_hadd_u_d(sum_w, sum_w);
+ sum_w = (v4u32)__msa_srari_w((v4i32)sum_d, 5);
+ out = (v16u8)__msa_splati_b((v16i8)sum_w, 0);
+
+ ST_UB8(out, out, out, out, out, out, out, out, dst, dst_stride);
+ dst += (8 * dst_stride);
+ ST_UB8(out, out, out, out, out, out, out, out, dst, dst_stride);
+}
+
+static void intra_predict_dc_tl_16x16_msa(const uint8_t *src, uint8_t *dst,
+ int32_t dst_stride) {+ v16u8 data, out;
+ v8u16 sum_h;
+ v4u32 sum_w;
+ v2u64 sum_d;
+
+ data = LD_UB(src);
+ sum_h = __msa_hadd_u_h(data, data);
+ sum_w = __msa_hadd_u_w(sum_h, sum_h);
+ sum_d = __msa_hadd_u_d(sum_w, sum_w);
+ sum_w = (v4u32)__msa_pckev_w((v4i32)sum_d, (v4i32)sum_d);
+ sum_d = __msa_hadd_u_d(sum_w, sum_w);
+ sum_w = (v4u32)__msa_srari_w((v4i32)sum_d, 4);
+ out = (v16u8)__msa_splati_b((v16i8)sum_w, 0);
+
+ ST_UB8(out, out, out, out, out, out, out, out, dst, dst_stride);
+ dst += (8 * dst_stride);
+ ST_UB8(out, out, out, out, out, out, out, out, dst, dst_stride);
+}
+
+static void intra_predict_128dc_16x16_msa(uint8_t *dst, int32_t dst_stride) {+ const v16u8 out = (v16u8)__msa_ldi_b(128);
+
+ ST_UB8(out, out, out, out, out, out, out, out, dst, dst_stride);
+ dst += (8 * dst_stride);
+ ST_UB8(out, out, out, out, out, out, out, out, dst, dst_stride);
+}
+
+static void intra_predict_dc_32x32_msa(const uint8_t *src_top,
+ const uint8_t *src_left,
+ uint8_t *dst, int32_t dst_stride) {+ uint32_t row;
+ v16u8 top0, top1, left0, left1, out;
+ v8u16 sum_h, sum_top0, sum_top1, sum_left0, sum_left1;
+ v4u32 sum_w;
+ v2u64 sum_d;
+
+ LD_UB2(src_top, 16, top0, top1);
+ LD_UB2(src_left, 16, left0, left1);
+ HADD_UB2_UH(top0, top1, sum_top0, sum_top1);
+ HADD_UB2_UH(left0, left1, sum_left0, sum_left1);
+ sum_h = sum_top0 + sum_top1;
+ sum_h += sum_left0 + sum_left1;
+ sum_w = __msa_hadd_u_w(sum_h, sum_h);
+ sum_d = __msa_hadd_u_d(sum_w, sum_w);
+ sum_w = (v4u32)__msa_pckev_w((v4i32)sum_d, (v4i32)sum_d);
+ sum_d = __msa_hadd_u_d(sum_w, sum_w);
+ sum_w = (v4u32)__msa_srari_w((v4i32)sum_d, 6);
+ out = (v16u8)__msa_splati_b((v16i8)sum_w, 0);
+
+ for (row = 16; row--;) {+ ST_UB2(out, out, dst, 16);
+ dst += dst_stride;
+ ST_UB2(out, out, dst, 16);
+ dst += dst_stride;
+ }
+}
+
+static void intra_predict_dc_tl_32x32_msa(const uint8_t *src, uint8_t *dst,
+ int32_t dst_stride) {+ uint32_t row;
+ v16u8 data0, data1, out;
+ v8u16 sum_h, sum_data0, sum_data1;
+ v4u32 sum_w;
+ v2u64 sum_d;
+
+ LD_UB2(src, 16, data0, data1);
+ HADD_UB2_UH(data0, data1, sum_data0, sum_data1);
+ sum_h = sum_data0 + sum_data1;
+ sum_w = __msa_hadd_u_w(sum_h, sum_h);
+ sum_d = __msa_hadd_u_d(sum_w, sum_w);
+ sum_w = (v4u32)__msa_pckev_w((v4i32)sum_d, (v4i32)sum_d);
+ sum_d = __msa_hadd_u_d(sum_w, sum_w);
+ sum_w = (v4u32)__msa_srari_w((v4i32)sum_d, 5);
+ out = (v16u8)__msa_splati_b((v16i8)sum_w, 0);
+
+ for (row = 16; row--;) {+ ST_UB2(out, out, dst, 16);
+ dst += dst_stride;
+ ST_UB2(out, out, dst, 16);
+ dst += dst_stride;
+ }
+}
+
+static void intra_predict_128dc_32x32_msa(uint8_t *dst, int32_t dst_stride) {+ uint32_t row;
+ const v16u8 out = (v16u8)__msa_ldi_b(128);
+
+ for (row = 16; row--;) {+ ST_UB2(out, out, dst, 16);
+ dst += dst_stride;
+ ST_UB2(out, out, dst, 16);
+ dst += dst_stride;
+ }
+}
+
+static void intra_predict_tm_4x4_msa(const uint8_t *src_top_ptr,
+ const uint8_t *src_left,
+ uint8_t *dst, int32_t dst_stride) {+ uint32_t val;
+ uint8_t top_left = src_top_ptr[-1];
+ v16i8 src_left0, src_left1, src_left2, src_left3, tmp0, tmp1, src_top = { 0 };+ v16u8 src0, src1, src2, src3;
+ v8u16 src_top_left, vec0, vec1, vec2, vec3;
+
+ src_top_left = (v8u16)__msa_fill_h(top_left);
+ val = LW(src_top_ptr);
+ src_top = (v16i8)__msa_insert_w((v4i32)src_top, 0, val);
+
+ src_left0 = __msa_fill_b(src_left[0]);
+ src_left1 = __msa_fill_b(src_left[1]);
+ src_left2 = __msa_fill_b(src_left[2]);
+ src_left3 = __msa_fill_b(src_left[3]);
+
+ ILVR_B4_UB(src_left0, src_top, src_left1, src_top, src_left2, src_top,
+ src_left3, src_top, src0, src1, src2, src3);
+ HADD_UB4_UH(src0, src1, src2, src3, vec0, vec1, vec2, vec3);
+ IPRED_SUBS_UH2_UH(src_top_left, src_top_left, vec0, vec1);
+ IPRED_SUBS_UH2_UH(src_top_left, src_top_left, vec2, vec3);
+ SAT_UH4_UH(vec0, vec1, vec2, vec3, 7);
+ PCKEV_B2_SB(vec1, vec0, vec3, vec2, tmp0, tmp1);
+ ST4x4_UB(tmp0, tmp1, 0, 2, 0, 2, dst, dst_stride);
+}
+
+static void intra_predict_tm_8x8_msa(const uint8_t *src_top_ptr,
+ const uint8_t *src_left,
+ uint8_t *dst, int32_t dst_stride) {+ uint64_t val;
+ uint8_t top_left = src_top_ptr[-1];
+ uint32_t loop_cnt;
+ v16i8 src_left0, src_left1, src_left2, src_left3, tmp0, tmp1, src_top = { 0 };+ v8u16 src_top_left, vec0, vec1, vec2, vec3;
+ v16u8 src0, src1, src2, src3;
+
+ val = LD(src_top_ptr);
+ src_top = (v16i8)__msa_insert_d((v2i64)src_top, 0, val);
+ src_top_left = (v8u16)__msa_fill_h(top_left);
+
+ for (loop_cnt = 2; loop_cnt--;) {+ src_left0 = __msa_fill_b(src_left[0]);
+ src_left1 = __msa_fill_b(src_left[1]);
+ src_left2 = __msa_fill_b(src_left[2]);
+ src_left3 = __msa_fill_b(src_left[3]);
+ src_left += 4;
+
+ ILVR_B4_UB(src_left0, src_top, src_left1, src_top, src_left2, src_top,
+ src_left3, src_top, src0, src1, src2, src3);
+ HADD_UB4_UH(src0, src1, src2, src3, vec0, vec1, vec2, vec3);
+ IPRED_SUBS_UH2_UH(src_top_left, src_top_left, vec0, vec1);
+ IPRED_SUBS_UH2_UH(src_top_left, src_top_left, vec2, vec3);
+ SAT_UH4_UH(vec0, vec1, vec2, vec3, 7);
+ PCKEV_B2_SB(vec1, vec0, vec3, vec2, tmp0, tmp1);
+ ST8x4_UB(tmp0, tmp1, dst, dst_stride);
+ dst += (4 * dst_stride);
+ }
+}
+
+static void intra_predict_tm_16x16_msa(const uint8_t *src_top_ptr,
+ const uint8_t *src_left,
+ uint8_t *dst, int32_t dst_stride) {+ uint8_t top_left = src_top_ptr[-1];
+ uint32_t loop_cnt;
+ v16i8 src_top, src_left0, src_left1, src_left2, src_left3;
+ v8u16 src_top_left, res_r, res_l;
+
+ src_top = LD_SB(src_top_ptr);
+ src_top_left = (v8u16)__msa_fill_h(top_left);
+
+ for (loop_cnt = 4; loop_cnt--;) {+ src_left0 = __msa_fill_b(src_left[0]);
+ src_left1 = __msa_fill_b(src_left[1]);
+ src_left2 = __msa_fill_b(src_left[2]);
+ src_left3 = __msa_fill_b(src_left[3]);
+ src_left += 4;
+
+ ILVRL_B2_UH(src_left0, src_top, res_r, res_l);
+ HADD_UB2_UH(res_r, res_l, res_r, res_l);
+ IPRED_SUBS_UH2_UH(src_top_left, src_top_left, res_r, res_l);
+
+ SAT_UH2_UH(res_r, res_l, 7);
+ PCKEV_ST_SB(res_r, res_l, dst);
+ dst += dst_stride;
+
+ ILVRL_B2_UH(src_left1, src_top, res_r, res_l);
+ HADD_UB2_UH(res_r, res_l, res_r, res_l);
+ IPRED_SUBS_UH2_UH(src_top_left, src_top_left, res_r, res_l);
+ SAT_UH2_UH(res_r, res_l, 7);
+ PCKEV_ST_SB(res_r, res_l, dst);
+ dst += dst_stride;
+
+ ILVRL_B2_UH(src_left2, src_top, res_r, res_l);
+ HADD_UB2_UH(res_r, res_l, res_r, res_l);
+ IPRED_SUBS_UH2_UH(src_top_left, src_top_left, res_r, res_l);
+ SAT_UH2_UH(res_r, res_l, 7);
+ PCKEV_ST_SB(res_r, res_l, dst);
+ dst += dst_stride;
+
+ ILVRL_B2_UH(src_left3, src_top, res_r, res_l);
+ HADD_UB2_UH(res_r, res_l, res_r, res_l);
+ IPRED_SUBS_UH2_UH(src_top_left, src_top_left, res_r, res_l);
+ SAT_UH2_UH(res_r, res_l, 7);
+ PCKEV_ST_SB(res_r, res_l, dst);
+ dst += dst_stride;
+ }
+}
+
+static void intra_predict_tm_32x32_msa(const uint8_t *src_top,
+ const uint8_t *src_left,
+ uint8_t *dst, int32_t dst_stride) {+ uint8_t top_left = src_top[-1];
+ uint32_t loop_cnt;
+ v16i8 src_top0, src_top1, src_left0, src_left1, src_left2, src_left3;
+ v8u16 src_top_left, res_r0, res_r1, res_l0, res_l1;
+
+ LD_SB2(src_top, 16, src_top0, src_top1);
+ src_top_left = (v8u16)__msa_fill_h(top_left);
+
+ for (loop_cnt = 8; loop_cnt--;) {+ src_left0 = __msa_fill_b(src_left[0]);
+ src_left1 = __msa_fill_b(src_left[1]);
+ src_left2 = __msa_fill_b(src_left[2]);
+ src_left3 = __msa_fill_b(src_left[3]);
+ src_left += 4;
+
+ ILVR_B2_UH(src_left0, src_top0, src_left0, src_top1, res_r0, res_r1);
+ ILVL_B2_UH(src_left0, src_top0, src_left0, src_top1, res_l0, res_l1);
+ HADD_UB4_UH(res_r0, res_l0, res_r1, res_l1, res_r0, res_l0, res_r1, res_l1);
+ IPRED_SUBS_UH2_UH(src_top_left, src_top_left, res_r0, res_l0);
+ IPRED_SUBS_UH2_UH(src_top_left, src_top_left, res_r1, res_l1);
+ SAT_UH4_UH(res_r0, res_l0, res_r1, res_l1, 7);
+ PCKEV_ST_SB(res_r0, res_l0, dst);
+ PCKEV_ST_SB(res_r1, res_l1, dst + 16);
+ dst += dst_stride;
+
+ ILVR_B2_UH(src_left1, src_top0, src_left1, src_top1, res_r0, res_r1);
+ ILVL_B2_UH(src_left1, src_top0, src_left1, src_top1, res_l0, res_l1);
+ HADD_UB4_UH(res_r0, res_l0, res_r1, res_l1, res_r0, res_l0, res_r1, res_l1);
+ IPRED_SUBS_UH2_UH(src_top_left, src_top_left, res_r0, res_l0);
+ IPRED_SUBS_UH2_UH(src_top_left, src_top_left, res_r1, res_l1);
+ SAT_UH4_UH(res_r0, res_l0, res_r1, res_l1, 7);
+ PCKEV_ST_SB(res_r0, res_l0, dst);
+ PCKEV_ST_SB(res_r1, res_l1, dst + 16);
+ dst += dst_stride;
+
+ ILVR_B2_UH(src_left2, src_top0, src_left2, src_top1, res_r0, res_r1);
+ ILVL_B2_UH(src_left2, src_top0, src_left2, src_top1, res_l0, res_l1);
+ HADD_UB4_UH(res_r0, res_l0, res_r1, res_l1, res_r0, res_l0, res_r1, res_l1);
+ IPRED_SUBS_UH2_UH(src_top_left, src_top_left, res_r0, res_l0);
+ IPRED_SUBS_UH2_UH(src_top_left, src_top_left, res_r1, res_l1);
+ SAT_UH4_UH(res_r0, res_l0, res_r1, res_l1, 7);
+ PCKEV_ST_SB(res_r0, res_l0, dst);
+ PCKEV_ST_SB(res_r1, res_l1, dst + 16);
+ dst += dst_stride;
+
+ ILVR_B2_UH(src_left3, src_top0, src_left3, src_top1, res_r0, res_r1);
+ ILVL_B2_UH(src_left3, src_top0, src_left3, src_top1, res_l0, res_l1);
+ HADD_UB4_UH(res_r0, res_l0, res_r1, res_l1, res_r0, res_l0, res_r1, res_l1);
+ IPRED_SUBS_UH2_UH(src_top_left, src_top_left, res_r0, res_l0);
+ IPRED_SUBS_UH2_UH(src_top_left, src_top_left, res_r1, res_l1);
+ SAT_UH4_UH(res_r0, res_l0, res_r1, res_l1, 7);
+ PCKEV_ST_SB(res_r0, res_l0, dst);
+ PCKEV_ST_SB(res_r1, res_l1, dst + 16);
+ dst += dst_stride;
+ }
+}
+
+void vp9_v_predictor_4x4_msa(uint8_t *dst, ptrdiff_t y_stride,
+ const uint8_t *above, const uint8_t *left) {+ (void)left;
+
+ intra_predict_vert_4x4_msa(above, dst, y_stride);
+}
+
+void vp9_v_predictor_8x8_msa(uint8_t *dst, ptrdiff_t y_stride,
+ const uint8_t *above, const uint8_t *left) {+ (void)left;
+
+ intra_predict_vert_8x8_msa(above, dst, y_stride);
+}
+
+void vp9_v_predictor_16x16_msa(uint8_t *dst, ptrdiff_t y_stride,
+ const uint8_t *above, const uint8_t *left) {+ (void)left;
+
+ intra_predict_vert_16x16_msa(above, dst, y_stride);
+}
+
+void vp9_v_predictor_32x32_msa(uint8_t *dst, ptrdiff_t y_stride,
+ const uint8_t *above, const uint8_t *left) {+ (void)left;
+
+ intra_predict_vert_32x32_msa(above, dst, y_stride);
+}
+
+void vp9_h_predictor_4x4_msa(uint8_t *dst, ptrdiff_t y_stride,
+ const uint8_t *above, const uint8_t *left) {+ (void)above;
+
+ intra_predict_horiz_4x4_msa(left, dst, y_stride);
+}
+
+void vp9_h_predictor_8x8_msa(uint8_t *dst, ptrdiff_t y_stride,
+ const uint8_t *above, const uint8_t *left) {+ (void)above;
+
+ intra_predict_horiz_8x8_msa(left, dst, y_stride);
+}
+
+void vp9_h_predictor_16x16_msa(uint8_t *dst, ptrdiff_t y_stride,
+ const uint8_t *above, const uint8_t *left) {+ (void)above;
+
+ intra_predict_horiz_16x16_msa(left, dst, y_stride);
+}
+
+void vp9_h_predictor_32x32_msa(uint8_t *dst, ptrdiff_t y_stride,
+ const uint8_t *above, const uint8_t *left) {+ (void)above;
+
+ intra_predict_horiz_32x32_msa(left, dst, y_stride);
+}
+
+void vp9_dc_predictor_4x4_msa(uint8_t *dst, ptrdiff_t y_stride,
+ const uint8_t *above, const uint8_t *left) {+ intra_predict_dc_4x4_msa(above, left, dst, y_stride);
+}
+
+void vp9_dc_predictor_8x8_msa(uint8_t *dst, ptrdiff_t y_stride,
+ const uint8_t *above, const uint8_t *left) {+ intra_predict_dc_8x8_msa(above, left, dst, y_stride);
+}
+
+void vp9_dc_predictor_16x16_msa(uint8_t *dst, ptrdiff_t y_stride,
+ const uint8_t *above, const uint8_t *left) {+ intra_predict_dc_16x16_msa(above, left, dst, y_stride);
+}
+
+void vp9_dc_predictor_32x32_msa(uint8_t *dst, ptrdiff_t y_stride,
+ const uint8_t *above, const uint8_t *left) {+ intra_predict_dc_32x32_msa(above, left, dst, y_stride);
+}
+
+void vp9_dc_top_predictor_4x4_msa(uint8_t *dst, ptrdiff_t y_stride,
+ const uint8_t *above, const uint8_t *left) {+ (void)left;
+
+ intra_predict_dc_tl_4x4_msa(above, dst, y_stride);
+}
+
+void vp9_dc_top_predictor_8x8_msa(uint8_t *dst, ptrdiff_t y_stride,
+ const uint8_t *above, const uint8_t *left) {+ (void)left;
+
+ intra_predict_dc_tl_8x8_msa(above, dst, y_stride);
+}
+
+void vp9_dc_top_predictor_16x16_msa(uint8_t *dst, ptrdiff_t y_stride,
+ const uint8_t *above, const uint8_t *left) {+ (void)left;
+
+ intra_predict_dc_tl_16x16_msa(above, dst, y_stride);
+}
+
+void vp9_dc_top_predictor_32x32_msa(uint8_t *dst, ptrdiff_t y_stride,
+ const uint8_t *above, const uint8_t *left) {+ (void)left;
+
+ intra_predict_dc_tl_32x32_msa(above, dst, y_stride);
+}
+
+void vp9_dc_left_predictor_4x4_msa(uint8_t *dst, ptrdiff_t y_stride,
+ const uint8_t *above, const uint8_t *left) {+ (void)above;
+
+ intra_predict_dc_tl_4x4_msa(left, dst, y_stride);
+}
+
+void vp9_dc_left_predictor_8x8_msa(uint8_t *dst, ptrdiff_t y_stride,
+ const uint8_t *above, const uint8_t *left) {+ (void)above;
+
+ intra_predict_dc_tl_8x8_msa(left, dst, y_stride);
+}
+
+void vp9_dc_left_predictor_16x16_msa(uint8_t *dst, ptrdiff_t y_stride,
+ const uint8_t *above,
+ const uint8_t *left) {+ (void)above;
+
+ intra_predict_dc_tl_16x16_msa(left, dst, y_stride);
+}
+
+void vp9_dc_left_predictor_32x32_msa(uint8_t *dst, ptrdiff_t y_stride,
+ const uint8_t *above,
+ const uint8_t *left) {+ (void)above;
+
+ intra_predict_dc_tl_32x32_msa(left, dst, y_stride);
+}
+
+void vp9_dc_128_predictor_4x4_msa(uint8_t *dst, ptrdiff_t y_stride,
+ const uint8_t *above, const uint8_t *left) {+ (void)above;
+ (void)left;
+
+ intra_predict_128dc_4x4_msa(dst, y_stride);
+}
+
+void vp9_dc_128_predictor_8x8_msa(uint8_t *dst, ptrdiff_t y_stride,
+ const uint8_t *above, const uint8_t *left) {+ (void)above;
+ (void)left;
+
+ intra_predict_128dc_8x8_msa(dst, y_stride);
+}
+
+void vp9_dc_128_predictor_16x16_msa(uint8_t *dst, ptrdiff_t y_stride,
+ const uint8_t *above, const uint8_t *left) {+ (void)above;
+ (void)left;
+
+ intra_predict_128dc_16x16_msa(dst, y_stride);
+}
+
+void vp9_dc_128_predictor_32x32_msa(uint8_t *dst, ptrdiff_t y_stride,
+ const uint8_t *above, const uint8_t *left) {+ (void)above;
+ (void)left;
+
+ intra_predict_128dc_32x32_msa(dst, y_stride);
+}
+
+void vp9_tm_predictor_4x4_msa(uint8_t *dst, ptrdiff_t y_stride,
+ const uint8_t *above, const uint8_t *left) {+ intra_predict_tm_4x4_msa(above, left, dst, y_stride);
+}
+
+void vp9_tm_predictor_8x8_msa(uint8_t *dst, ptrdiff_t y_stride,
+ const uint8_t *above, const uint8_t *left) {+ intra_predict_tm_8x8_msa(above, left, dst, y_stride);
+}
+
+void vp9_tm_predictor_16x16_msa(uint8_t *dst, ptrdiff_t y_stride,
+ const uint8_t *above, const uint8_t *left) {+ intra_predict_tm_16x16_msa(above, left, dst, y_stride);
+}
+
+void vp9_tm_predictor_32x32_msa(uint8_t *dst, ptrdiff_t y_stride,
+ const uint8_t *above, const uint8_t *left) {+ intra_predict_tm_32x32_msa(above, left, dst, y_stride);
+}
--- a/vpx_dsp/vpx_dsp.mk
+++ b/vpx_dsp/vpx_dsp.mk
@@ -31,6 +31,27 @@
DSP_SRCS-yes += bitreader_buffer.h
endif
+# intra predictions
+DSP_SRCS-yes += intrapred.c
+
+ifeq ($(CONFIG_USE_X86INC),yes)
+DSP_SRCS-$(HAVE_SSE2) += x86/intrapred_sse2.asm
+DSP_SRCS-$(HAVE_SSSE3) += x86/intrapred_ssse3.asm
+endif
+
+ifeq ($(CONFIG_VP9_HIGHBITDEPTH),yes)
+ifeq ($(CONFIG_USE_X86INC),yes)
+DSP_SRCS-$(HAVE_SSE2) += x86/highbd_intrapred_sse2.asm
+endif
+endif
+
+DSP_SRCS-$(HAVE_NEON_ASM) += arm/intrapred_neon_asm$(ASM)
+DSP_SRCS-$(HAVE_NEON) += arm/intrapred_neon.c
+DSP_SRCS-$(HAVE_MSA) += mips/intrapred_msa.c
+DSP_SRCS-$(HAVE_DSPR2) += mips/intrapred4_dspr2.c
+DSP_SRCS-$(HAVE_DSPR2) += mips/intrapred8_dspr2.c
+DSP_SRCS-$(HAVE_DSPR2) += mips/intrapred16_dspr2.c
+
# loop filters
DSP_SRCS-yes += loopfilter.c
--- a/vpx_dsp/vpx_dsp_common.h
+++ b/vpx_dsp/vpx_dsp_common.h
@@ -50,6 +50,20 @@
return value < low ? low : (value > high ? high : value);
}
+#if CONFIG_VP9_HIGHBITDEPTH
+static INLINE uint16_t clip_pixel_highbd(int val, int bd) {+ switch (bd) {+ case 8:
+ default:
+ return (uint16_t)clamp(val, 0, 255);
+ case 10:
+ return (uint16_t)clamp(val, 0, 1023);
+ case 12:
+ return (uint16_t)clamp(val, 0, 4095);
+ }
+}
+#endif // CONFIG_VP9_HIGHBITDEPTH
+
#ifdef __cplusplus
} // extern "C"
#endif
--- a/vpx_dsp/vpx_dsp_rtcd_defs.pl
+++ b/vpx_dsp/vpx_dsp_rtcd_defs.pl
@@ -45,6 +45,324 @@
}
#
+# Intra prediction
+#
+add_proto qw/void vp9_d207_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_d207_predictor_4x4/, "$ssse3_x86inc";
+
+add_proto qw/void vp9_d45_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_d45_predictor_4x4 neon/, "$ssse3_x86inc";
+
+add_proto qw/void vp9_d63_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_d63_predictor_4x4/, "$ssse3_x86inc";
+
+add_proto qw/void vp9_h_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_h_predictor_4x4 neon dspr2 msa/, "$ssse3_x86inc";
+
+add_proto qw/void vp9_d117_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_d117_predictor_4x4/;
+
+add_proto qw/void vp9_d135_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_d135_predictor_4x4 neon/;
+
+add_proto qw/void vp9_d153_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_d153_predictor_4x4/, "$ssse3_x86inc";
+
+add_proto qw/void vp9_v_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_v_predictor_4x4 neon msa/, "$sse_x86inc";
+
+add_proto qw/void vp9_tm_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_tm_predictor_4x4 neon dspr2 msa/, "$sse_x86inc";
+
+add_proto qw/void vp9_dc_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_dc_predictor_4x4 dspr2 msa neon/, "$sse_x86inc";
+
+add_proto qw/void vp9_dc_top_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_dc_top_predictor_4x4 msa neon/, "$sse_x86inc";
+
+add_proto qw/void vp9_dc_left_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_dc_left_predictor_4x4 msa neon/, "$sse_x86inc";
+
+add_proto qw/void vp9_dc_128_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_dc_128_predictor_4x4 msa neon/, "$sse_x86inc";
+
+add_proto qw/void vp9_d207_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_d207_predictor_8x8/, "$ssse3_x86inc";
+
+add_proto qw/void vp9_d45_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_d45_predictor_8x8 neon/, "$ssse3_x86inc";
+
+add_proto qw/void vp9_d63_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_d63_predictor_8x8/, "$ssse3_x86inc";
+
+add_proto qw/void vp9_h_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_h_predictor_8x8 neon dspr2 msa/, "$ssse3_x86inc";
+
+add_proto qw/void vp9_d117_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_d117_predictor_8x8/;
+
+add_proto qw/void vp9_d135_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_d135_predictor_8x8/;
+
+add_proto qw/void vp9_d153_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_d153_predictor_8x8/, "$ssse3_x86inc";
+
+add_proto qw/void vp9_v_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_v_predictor_8x8 neon msa/, "$sse_x86inc";
+
+add_proto qw/void vp9_tm_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_tm_predictor_8x8 neon dspr2 msa/, "$sse2_x86inc";
+
+add_proto qw/void vp9_dc_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_dc_predictor_8x8 dspr2 neon msa/, "$sse_x86inc";
+
+add_proto qw/void vp9_dc_top_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_dc_top_predictor_8x8 neon msa/, "$sse_x86inc";
+
+add_proto qw/void vp9_dc_left_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_dc_left_predictor_8x8 neon msa/, "$sse_x86inc";
+
+add_proto qw/void vp9_dc_128_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_dc_128_predictor_8x8 neon msa/, "$sse_x86inc";
+
+add_proto qw/void vp9_d207_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_d207_predictor_16x16/, "$ssse3_x86inc";
+
+add_proto qw/void vp9_d45_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_d45_predictor_16x16 neon/, "$ssse3_x86inc";
+
+add_proto qw/void vp9_d63_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_d63_predictor_16x16/, "$ssse3_x86inc";
+
+add_proto qw/void vp9_h_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_h_predictor_16x16 neon dspr2 msa/, "$ssse3_x86inc";
+
+add_proto qw/void vp9_d117_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_d117_predictor_16x16/;
+
+add_proto qw/void vp9_d135_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_d135_predictor_16x16/;
+
+add_proto qw/void vp9_d153_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_d153_predictor_16x16/, "$ssse3_x86inc";
+
+add_proto qw/void vp9_v_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_v_predictor_16x16 neon msa/, "$sse2_x86inc";
+
+add_proto qw/void vp9_tm_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_tm_predictor_16x16 neon msa/, "$sse2_x86inc";
+
+add_proto qw/void vp9_dc_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_dc_predictor_16x16 dspr2 neon msa/, "$sse2_x86inc";
+
+add_proto qw/void vp9_dc_top_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_dc_top_predictor_16x16 neon msa/, "$sse2_x86inc";
+
+add_proto qw/void vp9_dc_left_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_dc_left_predictor_16x16 neon msa/, "$sse2_x86inc";
+
+add_proto qw/void vp9_dc_128_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_dc_128_predictor_16x16 neon msa/, "$sse2_x86inc";
+
+add_proto qw/void vp9_d207_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_d207_predictor_32x32/, "$ssse3_x86inc";
+
+add_proto qw/void vp9_d45_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_d45_predictor_32x32/, "$ssse3_x86inc";
+
+add_proto qw/void vp9_d63_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_d63_predictor_32x32/, "$ssse3_x86inc";
+
+add_proto qw/void vp9_h_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_h_predictor_32x32 neon msa/, "$ssse3_x86inc";
+
+add_proto qw/void vp9_d117_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_d117_predictor_32x32/;
+
+add_proto qw/void vp9_d135_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_d135_predictor_32x32/;
+
+add_proto qw/void vp9_d153_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_d153_predictor_32x32/, "$ssse3_x86inc";
+
+add_proto qw/void vp9_v_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_v_predictor_32x32 neon msa/, "$sse2_x86inc";
+
+add_proto qw/void vp9_tm_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_tm_predictor_32x32 neon msa/, "$sse2_x86_64_x86inc";
+
+add_proto qw/void vp9_dc_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_dc_predictor_32x32 msa neon/, "$sse2_x86inc";
+
+add_proto qw/void vp9_dc_top_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_dc_top_predictor_32x32 msa neon/, "$sse2_x86inc";
+
+add_proto qw/void vp9_dc_left_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_dc_left_predictor_32x32 msa neon/, "$sse2_x86inc";
+
+add_proto qw/void vp9_dc_128_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
+specialize qw/vp9_dc_128_predictor_32x32 msa neon/, "$sse2_x86inc";
+
+# High bitdepth functions
+if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {+ add_proto qw/void vp9_highbd_d207_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_d207_predictor_4x4/;
+
+ add_proto qw/void vp9_highbd_d45_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_d45_predictor_4x4/;
+
+ add_proto qw/void vp9_highbd_d63_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_d63_predictor_4x4/;
+
+ add_proto qw/void vp9_highbd_h_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_h_predictor_4x4/;
+
+ add_proto qw/void vp9_highbd_d117_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_d117_predictor_4x4/;
+
+ add_proto qw/void vp9_highbd_d135_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_d135_predictor_4x4/;
+
+ add_proto qw/void vp9_highbd_d153_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_d153_predictor_4x4/;
+
+ add_proto qw/void vp9_highbd_v_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_v_predictor_4x4/, "$sse_x86inc";
+
+ add_proto qw/void vp9_highbd_tm_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_tm_predictor_4x4/, "$sse_x86inc";
+
+ add_proto qw/void vp9_highbd_dc_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_dc_predictor_4x4/, "$sse_x86inc";
+
+ add_proto qw/void vp9_highbd_dc_top_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_dc_top_predictor_4x4/;
+
+ add_proto qw/void vp9_highbd_dc_left_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_dc_left_predictor_4x4/;
+
+ add_proto qw/void vp9_highbd_dc_128_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_dc_128_predictor_4x4/;
+
+ add_proto qw/void vp9_highbd_d207_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_d207_predictor_8x8/;
+
+ add_proto qw/void vp9_highbd_d45_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_d45_predictor_8x8/;
+
+ add_proto qw/void vp9_highbd_d63_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_d63_predictor_8x8/;
+
+ add_proto qw/void vp9_highbd_h_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_h_predictor_8x8/;
+
+ add_proto qw/void vp9_highbd_d117_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_d117_predictor_8x8/;
+
+ add_proto qw/void vp9_highbd_d135_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_d135_predictor_8x8/;
+
+ add_proto qw/void vp9_highbd_d153_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_d153_predictor_8x8/;
+
+ add_proto qw/void vp9_highbd_v_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_v_predictor_8x8/, "$sse2_x86inc";
+
+ add_proto qw/void vp9_highbd_tm_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_tm_predictor_8x8/, "$sse2_x86inc";
+
+ add_proto qw/void vp9_highbd_dc_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_dc_predictor_8x8/, "$sse2_x86inc";;
+
+ add_proto qw/void vp9_highbd_dc_top_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_dc_top_predictor_8x8/;
+
+ add_proto qw/void vp9_highbd_dc_left_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_dc_left_predictor_8x8/;
+
+ add_proto qw/void vp9_highbd_dc_128_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_dc_128_predictor_8x8/;
+
+ add_proto qw/void vp9_highbd_d207_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_d207_predictor_16x16/;
+
+ add_proto qw/void vp9_highbd_d45_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_d45_predictor_16x16/;
+
+ add_proto qw/void vp9_highbd_d63_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_d63_predictor_16x16/;
+
+ add_proto qw/void vp9_highbd_h_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_h_predictor_16x16/;
+
+ add_proto qw/void vp9_highbd_d117_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_d117_predictor_16x16/;
+
+ add_proto qw/void vp9_highbd_d135_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_d135_predictor_16x16/;
+
+ add_proto qw/void vp9_highbd_d153_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_d153_predictor_16x16/;
+
+ add_proto qw/void vp9_highbd_v_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_v_predictor_16x16/, "$sse2_x86inc";
+
+ add_proto qw/void vp9_highbd_tm_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_tm_predictor_16x16/, "$sse2_x86_64_x86inc";
+
+ add_proto qw/void vp9_highbd_dc_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_dc_predictor_16x16/, "$sse2_x86inc";
+
+ add_proto qw/void vp9_highbd_dc_top_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_dc_top_predictor_16x16/;
+
+ add_proto qw/void vp9_highbd_dc_left_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_dc_left_predictor_16x16/;
+
+ add_proto qw/void vp9_highbd_dc_128_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_dc_128_predictor_16x16/;
+
+ add_proto qw/void vp9_highbd_d207_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_d207_predictor_32x32/;
+
+ add_proto qw/void vp9_highbd_d45_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_d45_predictor_32x32/;
+
+ add_proto qw/void vp9_highbd_d63_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_d63_predictor_32x32/;
+
+ add_proto qw/void vp9_highbd_h_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_h_predictor_32x32/;
+
+ add_proto qw/void vp9_highbd_d117_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_d117_predictor_32x32/;
+
+ add_proto qw/void vp9_highbd_d135_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_d135_predictor_32x32/;
+
+ add_proto qw/void vp9_highbd_d153_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_d153_predictor_32x32/;
+
+ add_proto qw/void vp9_highbd_v_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_v_predictor_32x32/, "$sse2_x86inc";
+
+ add_proto qw/void vp9_highbd_tm_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_tm_predictor_32x32/, "$sse2_x86_64_x86inc";
+
+ add_proto qw/void vp9_highbd_dc_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_dc_predictor_32x32/, "$sse2_x86_64_x86inc";
+
+ add_proto qw/void vp9_highbd_dc_top_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_dc_top_predictor_32x32/;
+
+ add_proto qw/void vp9_highbd_dc_left_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_dc_left_predictor_32x32/;
+
+ add_proto qw/void vp9_highbd_dc_128_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vp9_highbd_dc_128_predictor_32x32/;
+}
+
+#
# Loopfilter
#
add_proto qw/void vpx_lpf_vertical_16/, "uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh";
--- /dev/null
+++ b/vpx_dsp/x86/highbd_intrapred_sse2.asm
@@ -1,0 +1,476 @@
+;
+; Copyright (c) 2014 The WebM project authors. All Rights Reserved.
+;
+; Use of this source code is governed by a BSD-style license
+; that can be found in the LICENSE file in the root of the source
+; tree. An additional intellectual property rights grant can be found
+; in the file PATENTS. All contributing project authors may
+; be found in the AUTHORS file in the root of the source tree.
+;
+
+%include "third_party/x86inc/x86inc.asm"
+
+SECTION_RODATA
+pw_4: times 8 dw 4
+pw_8: times 8 dw 8
+pw_16: times 4 dd 16
+pw_32: times 4 dd 32
+
+SECTION .text
+INIT_MMX sse
+cglobal highbd_dc_predictor_4x4, 4, 5, 4, dst, stride, above, left, goffset
+ GET_GOT goffsetq
+
+ movq m0, [aboveq]
+ movq m2, [leftq]
+ DEFINE_ARGS dst, stride, one
+ mov oned, 0x0001
+ pxor m1, m1
+ movd m3, oned
+ pshufw m3, m3, 0x0
+ paddw m0, m2
+ pmaddwd m0, m3
+ packssdw m0, m1
+ pmaddwd m0, m3
+ paddw m0, [GLOBAL(pw_4)]
+ psraw m0, 3
+ pshufw m0, m0, 0x0
+ movq [dstq ], m0
+ movq [dstq+strideq*2], m0
+ lea dstq, [dstq+strideq*4]
+ movq [dstq ], m0
+ movq [dstq+strideq*2], m0
+
+ RESTORE_GOT
+ RET
+
+INIT_XMM sse2
+cglobal highbd_dc_predictor_8x8, 4, 5, 4, dst, stride, above, left, goffset
+ GET_GOT goffsetq
+
+ pxor m1, m1
+ mova m0, [aboveq]
+ mova m2, [leftq]
+ DEFINE_ARGS dst, stride, stride3, one
+ mov oned, 0x00010001
+ lea stride3q, [strideq*3]
+ movd m3, oned
+ pshufd m3, m3, 0x0
+ paddw m0, m2
+ pmaddwd m0, m3
+ packssdw m0, m1
+ pmaddwd m0, m3
+ packssdw m0, m1
+ pmaddwd m0, m3
+ paddw m0, [GLOBAL(pw_8)]
+ psrlw m0, 4
+ pshuflw m0, m0, 0x0
+ punpcklqdq m0, m0
+ mova [dstq ], m0
+ mova [dstq+strideq*2 ], m0
+ mova [dstq+strideq*4 ], m0
+ mova [dstq+stride3q*2], m0
+ lea dstq, [dstq+strideq*8]
+ mova [dstq ], m0
+ mova [dstq+strideq*2 ], m0
+ mova [dstq+strideq*4 ], m0
+ mova [dstq+stride3q*2], m0
+
+ RESTORE_GOT
+ RET
+
+INIT_XMM sse2
+cglobal highbd_dc_predictor_16x16, 4, 5, 5, dst, stride, above, left, goffset
+ GET_GOT goffsetq
+
+ pxor m1, m1
+ mova m0, [aboveq]
+ mova m3, [aboveq+16]
+ mova m2, [leftq]
+ mova m4, [leftq+16]
+ DEFINE_ARGS dst, stride, stride3, lines4
+ lea stride3q, [strideq*3]
+ mov lines4d, 4
+ paddw m0, m2
+ paddw m0, m3
+ paddw m0, m4
+ movhlps m2, m0
+ paddw m0, m2
+ punpcklwd m0, m1
+ movhlps m2, m0
+ paddd m0, m2
+ punpckldq m0, m1
+ movhlps m2, m0
+ paddd m0, m2
+ paddd m0, [GLOBAL(pw_16)]
+ psrad m0, 5
+ pshuflw m0, m0, 0x0
+ punpcklqdq m0, m0
+.loop:
+ mova [dstq ], m0
+ mova [dstq +16], m0
+ mova [dstq+strideq*2 ], m0
+ mova [dstq+strideq*2 +16], m0
+ mova [dstq+strideq*4 ], m0
+ mova [dstq+strideq*4 +16], m0
+ mova [dstq+stride3q*2 ], m0
+ mova [dstq+stride3q*2+16], m0
+ lea dstq, [dstq+strideq*8]
+ dec lines4d
+ jnz .loop
+
+ RESTORE_GOT
+ REP_RET
+
+%if ARCH_X86_64
+INIT_XMM sse2
+cglobal highbd_dc_predictor_32x32, 4, 5, 9, dst, stride, above, left, goffset
+ GET_GOT goffsetq
+
+ pxor m1, m1
+ mova m0, [aboveq]
+ mova m2, [aboveq+16]
+ mova m3, [aboveq+32]
+ mova m4, [aboveq+48]
+ mova m5, [leftq]
+ mova m6, [leftq+16]
+ mova m7, [leftq+32]
+ mova m8, [leftq+48]
+ DEFINE_ARGS dst, stride, stride3, lines4
+ lea stride3q, [strideq*3]
+ mov lines4d, 8
+ paddw m0, m2
+ paddw m0, m3
+ paddw m0, m4
+ paddw m0, m5
+ paddw m0, m6
+ paddw m0, m7
+ paddw m0, m8
+ movhlps m2, m0
+ paddw m0, m2
+ punpcklwd m0, m1
+ movhlps m2, m0
+ paddd m0, m2
+ punpckldq m0, m1
+ movhlps m2, m0
+ paddd m0, m2
+ paddd m0, [GLOBAL(pw_32)]
+ psrad m0, 6
+ pshuflw m0, m0, 0x0
+ punpcklqdq m0, m0
+.loop:
+ mova [dstq ], m0
+ mova [dstq +16 ], m0
+ mova [dstq +32 ], m0
+ mova [dstq +48 ], m0
+ mova [dstq+strideq*2 ], m0
+ mova [dstq+strideq*2+16 ], m0
+ mova [dstq+strideq*2+32 ], m0
+ mova [dstq+strideq*2+48 ], m0
+ mova [dstq+strideq*4 ], m0
+ mova [dstq+strideq*4+16 ], m0
+ mova [dstq+strideq*4+32 ], m0
+ mova [dstq+strideq*4+48 ], m0
+ mova [dstq+stride3q*2 ], m0
+ mova [dstq+stride3q*2 +16], m0
+ mova [dstq+stride3q*2 +32], m0
+ mova [dstq+stride3q*2 +48], m0
+ lea dstq, [dstq+strideq*8]
+ dec lines4d
+ jnz .loop
+
+ RESTORE_GOT
+ REP_RET
+%endif
+
+INIT_MMX sse
+cglobal highbd_v_predictor_4x4, 3, 3, 1, dst, stride, above
+ movq m0, [aboveq]
+ movq [dstq ], m0
+ movq [dstq+strideq*2], m0
+ lea dstq, [dstq+strideq*4]
+ movq [dstq ], m0
+ movq [dstq+strideq*2], m0
+ RET
+
+INIT_XMM sse2
+cglobal highbd_v_predictor_8x8, 3, 3, 1, dst, stride, above
+ mova m0, [aboveq]
+ DEFINE_ARGS dst, stride, stride3
+ lea stride3q, [strideq*3]
+ mova [dstq ], m0
+ mova [dstq+strideq*2 ], m0
+ mova [dstq+strideq*4 ], m0
+ mova [dstq+stride3q*2], m0
+ lea dstq, [dstq+strideq*8]
+ mova [dstq ], m0
+ mova [dstq+strideq*2 ], m0
+ mova [dstq+strideq*4 ], m0
+ mova [dstq+stride3q*2], m0
+ RET
+
+INIT_XMM sse2
+cglobal highbd_v_predictor_16x16, 3, 4, 2, dst, stride, above
+ mova m0, [aboveq]
+ mova m1, [aboveq+16]
+ DEFINE_ARGS dst, stride, stride3, nlines4
+ lea stride3q, [strideq*3]
+ mov nlines4d, 4
+.loop:
+ mova [dstq ], m0
+ mova [dstq +16], m1
+ mova [dstq+strideq*2 ], m0
+ mova [dstq+strideq*2 +16], m1
+ mova [dstq+strideq*4 ], m0
+ mova [dstq+strideq*4 +16], m1
+ mova [dstq+stride3q*2 ], m0
+ mova [dstq+stride3q*2+16], m1
+ lea dstq, [dstq+strideq*8]
+ dec nlines4d
+ jnz .loop
+ REP_RET
+
+INIT_XMM sse2
+cglobal highbd_v_predictor_32x32, 3, 4, 4, dst, stride, above
+ mova m0, [aboveq]
+ mova m1, [aboveq+16]
+ mova m2, [aboveq+32]
+ mova m3, [aboveq+48]
+ DEFINE_ARGS dst, stride, stride3, nlines4
+ lea stride3q, [strideq*3]
+ mov nlines4d, 8
+.loop:
+ mova [dstq ], m0
+ mova [dstq +16], m1
+ mova [dstq +32], m2
+ mova [dstq +48], m3
+ mova [dstq+strideq*2 ], m0
+ mova [dstq+strideq*2 +16], m1
+ mova [dstq+strideq*2 +32], m2
+ mova [dstq+strideq*2 +48], m3
+ mova [dstq+strideq*4 ], m0
+ mova [dstq+strideq*4 +16], m1
+ mova [dstq+strideq*4 +32], m2
+ mova [dstq+strideq*4 +48], m3
+ mova [dstq+stride3q*2 ], m0
+ mova [dstq+stride3q*2 +16], m1
+ mova [dstq+stride3q*2 +32], m2
+ mova [dstq+stride3q*2 +48], m3
+ lea dstq, [dstq+strideq*8]
+ dec nlines4d
+ jnz .loop
+ REP_RET
+
+INIT_MMX sse
+cglobal highbd_tm_predictor_4x4, 5, 6, 5, dst, stride, above, left, bps, one
+ movd m1, [aboveq-2]
+ movq m0, [aboveq]
+ pshufw m1, m1, 0x0
+ ; Get the values to compute the maximum value at this bit depth
+ mov oned, 1
+ movd m3, oned
+ movd m4, bpsd
+ pshufw m3, m3, 0x0
+ DEFINE_ARGS dst, stride, line, left
+ mov lineq, -2
+ mova m2, m3
+ psllw m3, m4
+ add leftq, 8
+ psubw m3, m2 ; max possible value
+ pxor m4, m4 ; min possible value
+ psubw m0, m1
+.loop:
+ movq m1, [leftq+lineq*4]
+ movq m2, [leftq+lineq*4+2]
+ pshufw m1, m1, 0x0
+ pshufw m2, m2, 0x0
+ paddw m1, m0
+ paddw m2, m0
+ ;Clamp to the bit-depth
+ pminsw m1, m3
+ pminsw m2, m3
+ pmaxsw m1, m4
+ pmaxsw m2, m4
+ ;Store the values
+ movq [dstq ], m1
+ movq [dstq+strideq*2], m2
+ lea dstq, [dstq+strideq*4]
+ inc lineq
+ jnz .loop
+ REP_RET
+
+INIT_XMM sse2
+cglobal highbd_tm_predictor_8x8, 5, 6, 5, dst, stride, above, left, bps, one
+ movd m1, [aboveq-2]
+ mova m0, [aboveq]
+ pshuflw m1, m1, 0x0
+ ; Get the values to compute the maximum value at this bit depth
+ mov oned, 1
+ pxor m3, m3
+ pxor m4, m4
+ pinsrw m3, oned, 0
+ pinsrw m4, bpsd, 0
+ pshuflw m3, m3, 0x0
+ DEFINE_ARGS dst, stride, line, left
+ punpcklqdq m3, m3
+ mov lineq, -4
+ mova m2, m3
+ punpcklqdq m1, m1
+ psllw m3, m4
+ add leftq, 16
+ psubw m3, m2 ; max possible value
+ pxor m4, m4 ; min possible value
+ psubw m0, m1
+.loop:
+ movd m1, [leftq+lineq*4]
+ movd m2, [leftq+lineq*4+2]
+ pshuflw m1, m1, 0x0
+ pshuflw m2, m2, 0x0
+ punpcklqdq m1, m1
+ punpcklqdq m2, m2
+ paddw m1, m0
+ paddw m2, m0
+ ;Clamp to the bit-depth
+ pminsw m1, m3
+ pminsw m2, m3
+ pmaxsw m1, m4
+ pmaxsw m2, m4
+ ;Store the values
+ mova [dstq ], m1
+ mova [dstq+strideq*2], m2
+ lea dstq, [dstq+strideq*4]
+ inc lineq
+ jnz .loop
+ REP_RET
+
+%if ARCH_X86_64
+INIT_XMM sse2
+cglobal highbd_tm_predictor_16x16, 5, 6, 9, dst, stride, above, left, bps, one
+ movd m2, [aboveq-2]
+ mova m0, [aboveq]
+ mova m1, [aboveq+16]
+ pshuflw m2, m2, 0x0
+ ; Get the values to compute the maximum value at this bit depth
+ mov oned, 1
+ pxor m7, m7
+ pxor m8, m8
+ pinsrw m7, oned, 0
+ pinsrw m8, bpsd, 0
+ pshuflw m7, m7, 0x0
+ DEFINE_ARGS dst, stride, line, left
+ punpcklqdq m7, m7
+ mov lineq, -8
+ mova m5, m7
+ punpcklqdq m2, m2
+ psllw m7, m8
+ add leftq, 32
+ psubw m7, m5 ; max possible value
+ pxor m8, m8 ; min possible value
+ psubw m0, m2
+ psubw m1, m2
+.loop:
+ movd m2, [leftq+lineq*4]
+ movd m3, [leftq+lineq*4+2]
+ pshuflw m2, m2, 0x0
+ pshuflw m3, m3, 0x0
+ punpcklqdq m2, m2
+ punpcklqdq m3, m3
+ paddw m4, m2, m0
+ paddw m5, m3, m0
+ paddw m2, m1
+ paddw m3, m1
+ ;Clamp to the bit-depth
+ pminsw m4, m7
+ pminsw m5, m7
+ pminsw m2, m7
+ pminsw m3, m7
+ pmaxsw m4, m8
+ pmaxsw m5, m8
+ pmaxsw m2, m8
+ pmaxsw m3, m8
+ ;Store the values
+ mova [dstq ], m4
+ mova [dstq+strideq*2 ], m5
+ mova [dstq +16], m2
+ mova [dstq+strideq*2+16], m3
+ lea dstq, [dstq+strideq*4]
+ inc lineq
+ jnz .loop
+ REP_RET
+
+INIT_XMM sse2
+cglobal highbd_tm_predictor_32x32, 5, 6, 12, dst, stride, above, left, bps, one
+ movd m0, [aboveq-2]
+ mova m1, [aboveq]
+ mova m2, [aboveq+16]
+ mova m3, [aboveq+32]
+ mova m4, [aboveq+48]
+ pshuflw m0, m0, 0x0
+ ; Get the values to compute the maximum value at this bit depth
+ mov oned, 1
+ pxor m10, m10
+ pxor m11, m11
+ pinsrw m10, oned, 0
+ pinsrw m11, bpsd, 0
+ pshuflw m10, m10, 0x0
+ DEFINE_ARGS dst, stride, line, left
+ punpcklqdq m10, m10
+ mov lineq, -16
+ mova m5, m10
+ punpcklqdq m0, m0
+ psllw m10, m11
+ add leftq, 64
+ psubw m10, m5 ; max possible value
+ pxor m11, m11 ; min possible value
+ psubw m1, m0
+ psubw m2, m0
+ psubw m3, m0
+ psubw m4, m0
+.loop:
+ movd m5, [leftq+lineq*4]
+ movd m6, [leftq+lineq*4+2]
+ pshuflw m5, m5, 0x0
+ pshuflw m6, m6, 0x0
+ punpcklqdq m5, m5
+ punpcklqdq m6, m6
+ paddw m7, m5, m1
+ paddw m8, m5, m2
+ paddw m9, m5, m3
+ paddw m5, m4
+ ;Clamp these values to the bit-depth
+ pminsw m7, m10
+ pminsw m8, m10
+ pminsw m9, m10
+ pminsw m5, m10
+ pmaxsw m7, m11
+ pmaxsw m8, m11
+ pmaxsw m9, m11
+ pmaxsw m5, m11
+ ;Store these values
+ mova [dstq ], m7
+ mova [dstq +16], m8
+ mova [dstq +32], m9
+ mova [dstq +48], m5
+ paddw m7, m6, m1
+ paddw m8, m6, m2
+ paddw m9, m6, m3
+ paddw m6, m4
+ ;Clamp these values to the bit-depth
+ pminsw m7, m10
+ pminsw m8, m10
+ pminsw m9, m10
+ pminsw m6, m10
+ pmaxsw m7, m11
+ pmaxsw m8, m11
+ pmaxsw m9, m11
+ pmaxsw m6, m11
+ ;Store these values
+ mova [dstq+strideq*2 ], m7
+ mova [dstq+strideq*2+16], m8
+ mova [dstq+strideq*2+32], m9
+ mova [dstq+strideq*2+48], m6
+ lea dstq, [dstq+strideq*4]
+ inc lineq
+ jnz .loop
+ REP_RET
+%endif
--- /dev/null
+++ b/vpx_dsp/x86/intrapred_sse2.asm
@@ -1,0 +1,667 @@
+;
+; Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+;
+; Use of this source code is governed by a BSD-style license
+; that can be found in the LICENSE file in the root of the source
+; tree. An additional intellectual property rights grant can be found
+; in the file PATENTS. All contributing project authors may
+; be found in the AUTHORS file in the root of the source tree.
+;
+
+%include "third_party/x86inc/x86inc.asm"
+
+SECTION_RODATA
+pw_4: times 8 dw 4
+pw_8: times 8 dw 8
+pw_16: times 8 dw 16
+pw_32: times 8 dw 32
+dc_128: times 16 db 128
+pw2_4: times 8 dw 2
+pw2_8: times 8 dw 4
+pw2_16: times 8 dw 8
+pw2_32: times 8 dw 16
+
+SECTION .text
+
+INIT_MMX sse
+cglobal dc_predictor_4x4, 4, 5, 2, dst, stride, above, left, goffset
+ GET_GOT goffsetq
+
+ pxor m1, m1
+ movd m0, [aboveq]
+ punpckldq m0, [leftq]
+ psadbw m0, m1
+ paddw m0, [GLOBAL(pw_4)]
+ psraw m0, 3
+ pshufw m0, m0, 0x0
+ packuswb m0, m0
+ movd [dstq ], m0
+ movd [dstq+strideq], m0
+ lea dstq, [dstq+strideq*2]
+ movd [dstq ], m0
+ movd [dstq+strideq], m0
+
+ RESTORE_GOT
+ RET
+
+INIT_MMX sse
+cglobal dc_left_predictor_4x4, 4, 5, 2, dst, stride, above, left, goffset
+ GET_GOT goffsetq
+
+ pxor m1, m1
+ movd m0, [leftq]
+ psadbw m0, m1
+ paddw m0, [GLOBAL(pw2_4)]
+ psraw m0, 2
+ pshufw m0, m0, 0x0
+ packuswb m0, m0
+ movd [dstq ], m0
+ movd [dstq+strideq], m0
+ lea dstq, [dstq+strideq*2]
+ movd [dstq ], m0
+ movd [dstq+strideq], m0
+
+ RESTORE_GOT
+ RET
+
+INIT_MMX sse
+cglobal dc_top_predictor_4x4, 4, 5, 2, dst, stride, above, left, goffset
+ GET_GOT goffsetq
+
+ pxor m1, m1
+ movd m0, [aboveq]
+ psadbw m0, m1
+ paddw m0, [GLOBAL(pw2_4)]
+ psraw m0, 2
+ pshufw m0, m0, 0x0
+ packuswb m0, m0
+ movd [dstq ], m0
+ movd [dstq+strideq], m0
+ lea dstq, [dstq+strideq*2]
+ movd [dstq ], m0
+ movd [dstq+strideq], m0
+
+ RESTORE_GOT
+ RET
+
+INIT_MMX sse
+cglobal dc_predictor_8x8, 4, 5, 3, dst, stride, above, left, goffset
+ GET_GOT goffsetq
+
+ pxor m1, m1
+ movq m0, [aboveq]
+ movq m2, [leftq]
+ DEFINE_ARGS dst, stride, stride3
+ lea stride3q, [strideq*3]
+ psadbw m0, m1
+ psadbw m2, m1
+ paddw m0, m2
+ paddw m0, [GLOBAL(pw_8)]
+ psraw m0, 4
+ pshufw m0, m0, 0x0
+ packuswb m0, m0
+ movq [dstq ], m0
+ movq [dstq+strideq ], m0
+ movq [dstq+strideq*2], m0
+ movq [dstq+stride3q ], m0
+ lea dstq, [dstq+strideq*4]
+ movq [dstq ], m0
+ movq [dstq+strideq ], m0
+ movq [dstq+strideq*2], m0
+ movq [dstq+stride3q ], m0
+
+ RESTORE_GOT
+ RET
+
+INIT_MMX sse
+cglobal dc_top_predictor_8x8, 4, 5, 3, dst, stride, above, left, goffset
+ GET_GOT goffsetq
+
+ pxor m1, m1
+ movq m0, [aboveq]
+ DEFINE_ARGS dst, stride, stride3
+ lea stride3q, [strideq*3]
+ psadbw m0, m1
+ paddw m0, [GLOBAL(pw2_8)]
+ psraw m0, 3
+ pshufw m0, m0, 0x0
+ packuswb m0, m0
+ movq [dstq ], m0
+ movq [dstq+strideq ], m0
+ movq [dstq+strideq*2], m0
+ movq [dstq+stride3q ], m0
+ lea dstq, [dstq+strideq*4]
+ movq [dstq ], m0
+ movq [dstq+strideq ], m0
+ movq [dstq+strideq*2], m0
+ movq [dstq+stride3q ], m0
+
+ RESTORE_GOT
+ RET
+
+INIT_MMX sse
+cglobal dc_left_predictor_8x8, 4, 5, 3, dst, stride, above, left, goffset
+ GET_GOT goffsetq
+
+ pxor m1, m1
+ movq m0, [leftq]
+ DEFINE_ARGS dst, stride, stride3
+ lea stride3q, [strideq*3]
+ psadbw m0, m1
+ paddw m0, [GLOBAL(pw2_8)]
+ psraw m0, 3
+ pshufw m0, m0, 0x0
+ packuswb m0, m0
+ movq [dstq ], m0
+ movq [dstq+strideq ], m0
+ movq [dstq+strideq*2], m0
+ movq [dstq+stride3q ], m0
+ lea dstq, [dstq+strideq*4]
+ movq [dstq ], m0
+ movq [dstq+strideq ], m0
+ movq [dstq+strideq*2], m0
+ movq [dstq+stride3q ], m0
+
+ RESTORE_GOT
+ RET
+
+INIT_MMX sse
+cglobal dc_128_predictor_4x4, 4, 5, 3, dst, stride, above, left, goffset
+ GET_GOT goffsetq
+
+ DEFINE_ARGS dst, stride, stride3
+ lea stride3q, [strideq*3]
+ movd m0, [GLOBAL(dc_128)]
+ movd [dstq ], m0
+ movd [dstq+strideq ], m0
+ movd [dstq+strideq*2], m0
+ movd [dstq+stride3q ], m0
+ RESTORE_GOT
+ RET
+
+INIT_MMX sse
+cglobal dc_128_predictor_8x8, 4, 5, 3, dst, stride, above, left, goffset
+ GET_GOT goffsetq
+
+ DEFINE_ARGS dst, stride, stride3
+ lea stride3q, [strideq*3]
+ movq m0, [GLOBAL(dc_128)]
+ movq [dstq ], m0
+ movq [dstq+strideq ], m0
+ movq [dstq+strideq*2], m0
+ movq [dstq+stride3q ], m0
+ lea dstq, [dstq+strideq*4]
+ movq [dstq ], m0
+ movq [dstq+strideq ], m0
+ movq [dstq+strideq*2], m0
+ movq [dstq+stride3q ], m0
+ RESTORE_GOT
+ RET
+
+INIT_XMM sse2
+cglobal dc_predictor_16x16, 4, 5, 3, dst, stride, above, left, goffset
+ GET_GOT goffsetq
+
+ pxor m1, m1
+ mova m0, [aboveq]
+ mova m2, [leftq]
+ DEFINE_ARGS dst, stride, stride3, lines4
+ lea stride3q, [strideq*3]
+ mov lines4d, 4
+ psadbw m0, m1
+ psadbw m2, m1
+ paddw m0, m2
+ movhlps m2, m0
+ paddw m0, m2
+ paddw m0, [GLOBAL(pw_16)]
+ psraw m0, 5
+ pshuflw m0, m0, 0x0
+ punpcklqdq m0, m0
+ packuswb m0, m0
+.loop:
+ mova [dstq ], m0
+ mova [dstq+strideq ], m0
+ mova [dstq+strideq*2], m0
+ mova [dstq+stride3q ], m0
+ lea dstq, [dstq+strideq*4]
+ dec lines4d
+ jnz .loop
+
+ RESTORE_GOT
+ REP_RET
+
+
+INIT_XMM sse2
+cglobal dc_top_predictor_16x16, 4, 5, 3, dst, stride, above, left, goffset
+ GET_GOT goffsetq
+
+ pxor m1, m1
+ pxor m2, m2
+ mova m0, [aboveq]
+ DEFINE_ARGS dst, stride, stride3, lines4
+ lea stride3q, [strideq*3]
+ mov lines4d, 4
+ psadbw m0, m1
+ psadbw m2, m1
+ paddw m0, m2
+ movhlps m2, m0
+ paddw m0, m2
+ paddw m0, [GLOBAL(pw2_16)]
+ psraw m0, 4
+ pshuflw m0, m0, 0x0
+ punpcklqdq m0, m0
+ packuswb m0, m0
+.loop:
+ mova [dstq ], m0
+ mova [dstq+strideq ], m0
+ mova [dstq+strideq*2], m0
+ mova [dstq+stride3q ], m0
+ lea dstq, [dstq+strideq*4]
+ dec lines4d
+ jnz .loop
+
+ RESTORE_GOT
+ REP_RET
+
+INIT_XMM sse2
+cglobal dc_left_predictor_16x16, 4, 5, 3, dst, stride, above, left, goffset
+ GET_GOT goffsetq
+
+ pxor m1, m1
+ pxor m2, m2
+ mova m0, [leftq]
+ DEFINE_ARGS dst, stride, stride3, lines4
+ lea stride3q, [strideq*3]
+ mov lines4d, 4
+ psadbw m0, m1
+ psadbw m2, m1
+ paddw m0, m2
+ movhlps m2, m0
+ paddw m0, m2
+ paddw m0, [GLOBAL(pw2_16)]
+ psraw m0, 4
+ pshuflw m0, m0, 0x0
+ punpcklqdq m0, m0
+ packuswb m0, m0
+.loop:
+ mova [dstq ], m0
+ mova [dstq+strideq ], m0
+ mova [dstq+strideq*2], m0
+ mova [dstq+stride3q ], m0
+ lea dstq, [dstq+strideq*4]
+ dec lines4d
+ jnz .loop
+
+ RESTORE_GOT
+ REP_RET
+
+INIT_XMM sse2
+cglobal dc_128_predictor_16x16, 4, 5, 3, dst, stride, above, left, goffset
+ GET_GOT goffsetq
+
+ DEFINE_ARGS dst, stride, stride3, lines4
+ lea stride3q, [strideq*3]
+ mov lines4d, 4
+ mova m0, [GLOBAL(dc_128)]
+.loop:
+ mova [dstq ], m0
+ mova [dstq+strideq ], m0
+ mova [dstq+strideq*2], m0
+ mova [dstq+stride3q ], m0
+ lea dstq, [dstq+strideq*4]
+ dec lines4d
+ jnz .loop
+ RESTORE_GOT
+ RET
+
+
+INIT_XMM sse2
+cglobal dc_predictor_32x32, 4, 5, 5, dst, stride, above, left, goffset
+ GET_GOT goffsetq
+
+ pxor m1, m1
+ mova m0, [aboveq]
+ mova m2, [aboveq+16]
+ mova m3, [leftq]
+ mova m4, [leftq+16]
+ DEFINE_ARGS dst, stride, stride3, lines4
+ lea stride3q, [strideq*3]
+ mov lines4d, 8
+ psadbw m0, m1
+ psadbw m2, m1
+ psadbw m3, m1
+ psadbw m4, m1
+ paddw m0, m2
+ paddw m0, m3
+ paddw m0, m4
+ movhlps m2, m0
+ paddw m0, m2
+ paddw m0, [GLOBAL(pw_32)]
+ psraw m0, 6
+ pshuflw m0, m0, 0x0
+ punpcklqdq m0, m0
+ packuswb m0, m0
+.loop:
+ mova [dstq ], m0
+ mova [dstq +16], m0
+ mova [dstq+strideq ], m0
+ mova [dstq+strideq +16], m0
+ mova [dstq+strideq*2 ], m0
+ mova [dstq+strideq*2+16], m0
+ mova [dstq+stride3q ], m0
+ mova [dstq+stride3q +16], m0
+ lea dstq, [dstq+strideq*4]
+ dec lines4d
+ jnz .loop
+
+ RESTORE_GOT
+ REP_RET
+
+INIT_XMM sse2
+cglobal dc_top_predictor_32x32, 4, 5, 5, dst, stride, above, left, goffset
+ GET_GOT goffsetq
+
+ pxor m1, m1
+ mova m0, [aboveq]
+ mova m2, [aboveq+16]
+ DEFINE_ARGS dst, stride, stride3, lines4
+ lea stride3q, [strideq*3]
+ mov lines4d, 8
+ psadbw m0, m1
+ psadbw m2, m1
+ paddw m0, m2
+ movhlps m2, m0
+ paddw m0, m2
+ paddw m0, [GLOBAL(pw2_32)]
+ psraw m0, 5
+ pshuflw m0, m0, 0x0
+ punpcklqdq m0, m0
+ packuswb m0, m0
+.loop:
+ mova [dstq ], m0
+ mova [dstq +16], m0
+ mova [dstq+strideq ], m0
+ mova [dstq+strideq +16], m0
+ mova [dstq+strideq*2 ], m0
+ mova [dstq+strideq*2+16], m0
+ mova [dstq+stride3q ], m0
+ mova [dstq+stride3q +16], m0
+ lea dstq, [dstq+strideq*4]
+ dec lines4d
+ jnz .loop
+
+ RESTORE_GOT
+ REP_RET
+
+INIT_XMM sse2
+cglobal dc_left_predictor_32x32, 4, 5, 5, dst, stride, above, left, goffset
+ GET_GOT goffsetq
+
+ pxor m1, m1
+ mova m0, [leftq]
+ mova m2, [leftq+16]
+ DEFINE_ARGS dst, stride, stride3, lines4
+ lea stride3q, [strideq*3]
+ mov lines4d, 8
+ psadbw m0, m1
+ psadbw m2, m1
+ paddw m0, m2
+ movhlps m2, m0
+ paddw m0, m2
+ paddw m0, [GLOBAL(pw2_32)]
+ psraw m0, 5
+ pshuflw m0, m0, 0x0
+ punpcklqdq m0, m0
+ packuswb m0, m0
+.loop:
+ mova [dstq ], m0
+ mova [dstq +16], m0
+ mova [dstq+strideq ], m0
+ mova [dstq+strideq +16], m0
+ mova [dstq+strideq*2 ], m0
+ mova [dstq+strideq*2+16], m0
+ mova [dstq+stride3q ], m0
+ mova [dstq+stride3q +16], m0
+ lea dstq, [dstq+strideq*4]
+ dec lines4d
+ jnz .loop
+
+ RESTORE_GOT
+ REP_RET
+
+INIT_XMM sse2
+cglobal dc_128_predictor_32x32, 4, 5, 3, dst, stride, above, left, goffset
+ GET_GOT goffsetq
+
+ DEFINE_ARGS dst, stride, stride3, lines4
+ lea stride3q, [strideq*3]
+ mov lines4d, 8
+ mova m0, [GLOBAL(dc_128)]
+.loop:
+ mova [dstq ], m0
+ mova [dstq +16], m0
+ mova [dstq+strideq ], m0
+ mova [dstq+strideq +16], m0
+ mova [dstq+strideq*2 ], m0
+ mova [dstq+strideq*2+16], m0
+ mova [dstq+stride3q ], m0
+ mova [dstq+stride3q +16], m0
+ lea dstq, [dstq+strideq*4]
+ dec lines4d
+ jnz .loop
+ RESTORE_GOT
+ RET
+
+INIT_MMX sse
+cglobal v_predictor_4x4, 3, 3, 1, dst, stride, above
+ movd m0, [aboveq]
+ movd [dstq ], m0
+ movd [dstq+strideq], m0
+ lea dstq, [dstq+strideq*2]
+ movd [dstq ], m0
+ movd [dstq+strideq], m0
+ RET
+
+INIT_MMX sse
+cglobal v_predictor_8x8, 3, 3, 1, dst, stride, above
+ movq m0, [aboveq]
+ DEFINE_ARGS dst, stride, stride3
+ lea stride3q, [strideq*3]
+ movq [dstq ], m0
+ movq [dstq+strideq ], m0
+ movq [dstq+strideq*2], m0
+ movq [dstq+stride3q ], m0
+ lea dstq, [dstq+strideq*4]
+ movq [dstq ], m0
+ movq [dstq+strideq ], m0
+ movq [dstq+strideq*2], m0
+ movq [dstq+stride3q ], m0
+ RET
+
+INIT_XMM sse2
+cglobal v_predictor_16x16, 3, 4, 1, dst, stride, above
+ mova m0, [aboveq]
+ DEFINE_ARGS dst, stride, stride3, nlines4
+ lea stride3q, [strideq*3]
+ mov nlines4d, 4
+.loop:
+ mova [dstq ], m0
+ mova [dstq+strideq ], m0
+ mova [dstq+strideq*2], m0
+ mova [dstq+stride3q ], m0
+ lea dstq, [dstq+strideq*4]
+ dec nlines4d
+ jnz .loop
+ REP_RET
+
+INIT_XMM sse2
+cglobal v_predictor_32x32, 3, 4, 2, dst, stride, above
+ mova m0, [aboveq]
+ mova m1, [aboveq+16]
+ DEFINE_ARGS dst, stride, stride3, nlines4
+ lea stride3q, [strideq*3]
+ mov nlines4d, 8
+.loop:
+ mova [dstq ], m0
+ mova [dstq +16], m1
+ mova [dstq+strideq ], m0
+ mova [dstq+strideq +16], m1
+ mova [dstq+strideq*2 ], m0
+ mova [dstq+strideq*2+16], m1
+ mova [dstq+stride3q ], m0
+ mova [dstq+stride3q +16], m1
+ lea dstq, [dstq+strideq*4]
+ dec nlines4d
+ jnz .loop
+ REP_RET
+
+INIT_MMX sse
+cglobal tm_predictor_4x4, 4, 4, 4, dst, stride, above, left
+ pxor m1, m1
+ movd m2, [aboveq-1]
+ movd m0, [aboveq]
+ punpcklbw m2, m1
+ punpcklbw m0, m1
+ pshufw m2, m2, 0x0
+ DEFINE_ARGS dst, stride, line, left
+ mov lineq, -2
+ add leftq, 4
+ psubw m0, m2
+.loop:
+ movd m2, [leftq+lineq*2]
+ movd m3, [leftq+lineq*2+1]
+ punpcklbw m2, m1
+ punpcklbw m3, m1
+ pshufw m2, m2, 0x0
+ pshufw m3, m3, 0x0
+ paddw m2, m0
+ paddw m3, m0
+ packuswb m2, m2
+ packuswb m3, m3
+ movd [dstq ], m2
+ movd [dstq+strideq], m3
+ lea dstq, [dstq+strideq*2]
+ inc lineq
+ jnz .loop
+ REP_RET
+
+INIT_XMM sse2
+cglobal tm_predictor_8x8, 4, 4, 4, dst, stride, above, left
+ pxor m1, m1
+ movd m2, [aboveq-1]
+ movq m0, [aboveq]
+ punpcklbw m2, m1
+ punpcklbw m0, m1
+ pshuflw m2, m2, 0x0
+ DEFINE_ARGS dst, stride, line, left
+ mov lineq, -4
+ punpcklqdq m2, m2
+ add leftq, 8
+ psubw m0, m2
+.loop:
+ movd m2, [leftq+lineq*2]
+ movd m3, [leftq+lineq*2+1]
+ punpcklbw m2, m1
+ punpcklbw m3, m1
+ pshuflw m2, m2, 0x0
+ pshuflw m3, m3, 0x0
+ punpcklqdq m2, m2
+ punpcklqdq m3, m3
+ paddw m2, m0
+ paddw m3, m0
+ packuswb m2, m3
+ movq [dstq ], m2
+ movhps [dstq+strideq], m2
+ lea dstq, [dstq+strideq*2]
+ inc lineq
+ jnz .loop
+ REP_RET
+
+INIT_XMM sse2
+cglobal tm_predictor_16x16, 4, 4, 7, dst, stride, above, left
+ pxor m1, m1
+ movd m2, [aboveq-1]
+ mova m0, [aboveq]
+ punpcklbw m2, m1
+ punpckhbw m4, m0, m1
+ punpcklbw m0, m1
+ pshuflw m2, m2, 0x0
+ DEFINE_ARGS dst, stride, line, left
+ mov lineq, -8
+ punpcklqdq m2, m2
+ add leftq, 16
+ psubw m0, m2
+ psubw m4, m2
+.loop:
+ movd m2, [leftq+lineq*2]
+ movd m3, [leftq+lineq*2+1]
+ punpcklbw m2, m1
+ punpcklbw m3, m1
+ pshuflw m2, m2, 0x0
+ pshuflw m3, m3, 0x0
+ punpcklqdq m2, m2
+ punpcklqdq m3, m3
+ paddw m5, m2, m0
+ paddw m6, m3, m0
+ paddw m2, m4
+ paddw m3, m4
+ packuswb m5, m2
+ packuswb m6, m3
+ mova [dstq ], m5
+ mova [dstq+strideq], m6
+ lea dstq, [dstq+strideq*2]
+ inc lineq
+ jnz .loop
+ REP_RET
+
+%if ARCH_X86_64
+INIT_XMM sse2
+cglobal tm_predictor_32x32, 4, 4, 10, dst, stride, above, left
+ pxor m1, m1
+ movd m2, [aboveq-1]
+ mova m0, [aboveq]
+ mova m4, [aboveq+16]
+ punpcklbw m2, m1
+ punpckhbw m3, m0, m1
+ punpckhbw m5, m4, m1
+ punpcklbw m0, m1
+ punpcklbw m4, m1
+ pshuflw m2, m2, 0x0
+ DEFINE_ARGS dst, stride, line, left
+ mov lineq, -16
+ punpcklqdq m2, m2
+ add leftq, 32
+ psubw m0, m2
+ psubw m3, m2
+ psubw m4, m2
+ psubw m5, m2
+.loop:
+ movd m2, [leftq+lineq*2]
+ movd m6, [leftq+lineq*2+1]
+ punpcklbw m2, m1
+ punpcklbw m6, m1
+ pshuflw m2, m2, 0x0
+ pshuflw m6, m6, 0x0
+ punpcklqdq m2, m2
+ punpcklqdq m6, m6
+ paddw m7, m2, m0
+ paddw m8, m2, m3
+ paddw m9, m2, m4
+ paddw m2, m5
+ packuswb m7, m8
+ packuswb m9, m2
+ paddw m2, m6, m0
+ paddw m8, m6, m3
+ mova [dstq ], m7
+ paddw m7, m6, m4
+ paddw m6, m5
+ mova [dstq +16], m9
+ packuswb m2, m8
+ packuswb m7, m6
+ mova [dstq+strideq ], m2
+ mova [dstq+strideq+16], m7
+ lea dstq, [dstq+strideq*2]
+ inc lineq
+ jnz .loop
+ REP_RET
+%endif
--- /dev/null
+++ b/vpx_dsp/x86/intrapred_ssse3.asm
@@ -1,0 +1,1036 @@
+;
+; Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+;
+; Use of this source code is governed by a BSD-style license
+; that can be found in the LICENSE file in the root of the source
+; tree. An additional intellectual property rights grant can be found
+; in the file PATENTS. All contributing project authors may
+; be found in the AUTHORS file in the root of the source tree.
+;
+
+%include "third_party/x86inc/x86inc.asm"
+
+SECTION_RODATA
+
+pb_1: times 16 db 1
+sh_b01234577: db 0, 1, 2, 3, 4, 5, 7, 7, 0, 0, 0, 0, 0, 0, 0, 0
+sh_b12345677: db 1, 2, 3, 4, 5, 6, 7, 7, 0, 0, 0, 0, 0, 0, 0, 0
+sh_b23456777: db 2, 3, 4, 5, 6, 7, 7, 7, 0, 0, 0, 0, 0, 0, 0, 0
+sh_b0123456777777777: db 0, 1, 2, 3, 4, 5, 6, 7, 7, 7, 7, 7, 7, 7, 7, 7
+sh_b1234567777777777: db 1, 2, 3, 4, 5, 6, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7
+sh_b2345677777777777: db 2, 3, 4, 5, 6, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7
+sh_b123456789abcdeff: db 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 15
+sh_b23456789abcdefff: db 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 15, 15
+sh_b32104567: db 3, 2, 1, 0, 4, 5, 6, 7, 0, 0, 0, 0, 0, 0, 0, 0
+sh_b8091a2b345: db 8, 0, 9, 1, 10, 2, 11, 3, 4, 5, 0, 0, 0, 0, 0, 0
+sh_b76543210: db 7, 6, 5, 4, 3, 2, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0
+sh_b65432108: db 6, 5, 4, 3, 2, 1, 0, 8, 0, 0, 0, 0, 0, 0, 0, 0
+sh_b54321089: db 5, 4, 3, 2, 1, 0, 8, 9, 0, 0, 0, 0, 0, 0, 0, 0
+sh_b89abcdef: db 8, 9, 10, 11, 12, 13, 14, 15, 0, 0, 0, 0, 0, 0, 0, 0
+sh_bfedcba9876543210: db 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0
+sh_b1233: db 1, 2, 3, 3, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
+sh_b2333: db 2, 3, 3, 3, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
+
+SECTION .text
+
+INIT_MMX ssse3
+cglobal h_predictor_4x4, 2, 4, 3, dst, stride, line, left
+ movifnidn leftq, leftmp
+ add leftq, 4
+ mov lineq, -2
+ pxor m0, m0
+.loop:
+ movd m1, [leftq+lineq*2 ]
+ movd m2, [leftq+lineq*2+1]
+ pshufb m1, m0
+ pshufb m2, m0
+ movd [dstq ], m1
+ movd [dstq+strideq], m2
+ lea dstq, [dstq+strideq*2]
+ inc lineq
+ jnz .loop
+ REP_RET
+
+INIT_MMX ssse3
+cglobal h_predictor_8x8, 2, 4, 3, dst, stride, line, left
+ movifnidn leftq, leftmp
+ add leftq, 8
+ mov lineq, -4
+ pxor m0, m0
+.loop:
+ movd m1, [leftq+lineq*2 ]
+ movd m2, [leftq+lineq*2+1]
+ pshufb m1, m0
+ pshufb m2, m0
+ movq [dstq ], m1
+ movq [dstq+strideq], m2
+ lea dstq, [dstq+strideq*2]
+ inc lineq
+ jnz .loop
+ REP_RET
+
+INIT_XMM ssse3
+cglobal h_predictor_16x16, 2, 4, 3, dst, stride, line, left
+ movifnidn leftq, leftmp
+ add leftq, 16
+ mov lineq, -8
+ pxor m0, m0
+.loop:
+ movd m1, [leftq+lineq*2 ]
+ movd m2, [leftq+lineq*2+1]
+ pshufb m1, m0
+ pshufb m2, m0
+ mova [dstq ], m1
+ mova [dstq+strideq], m2
+ lea dstq, [dstq+strideq*2]
+ inc lineq
+ jnz .loop
+ REP_RET
+
+INIT_XMM ssse3
+cglobal h_predictor_32x32, 2, 4, 3, dst, stride, line, left
+ movifnidn leftq, leftmp
+ add leftq, 32
+ mov lineq, -16
+ pxor m0, m0
+.loop:
+ movd m1, [leftq+lineq*2 ]
+ movd m2, [leftq+lineq*2+1]
+ pshufb m1, m0
+ pshufb m2, m0
+ mova [dstq ], m1
+ mova [dstq +16], m1
+ mova [dstq+strideq ], m2
+ mova [dstq+strideq+16], m2
+ lea dstq, [dstq+strideq*2]
+ inc lineq
+ jnz .loop
+ REP_RET
+
+INIT_MMX ssse3
+cglobal d45_predictor_4x4, 3, 4, 4, dst, stride, above, goffset
+ GET_GOT goffsetq
+
+ movq m0, [aboveq]
+ pshufb m2, m0, [GLOBAL(sh_b23456777)]
+ pshufb m1, m0, [GLOBAL(sh_b01234577)]
+ pshufb m0, [GLOBAL(sh_b12345677)]
+ pavgb m3, m2, m1
+ pxor m2, m1
+ pand m2, [GLOBAL(pb_1)]
+ psubb m3, m2
+ pavgb m0, m3
+
+ ; store 4 lines
+ movd [dstq ], m0
+ psrlq m0, 8
+ movd [dstq+strideq], m0
+ lea dstq, [dstq+strideq*2]
+ psrlq m0, 8
+ movd [dstq ], m0
+ psrlq m0, 8
+ movd [dstq+strideq], m0
+
+ RESTORE_GOT
+ RET
+
+INIT_MMX ssse3
+cglobal d45_predictor_8x8, 3, 4, 4, dst, stride, above, goffset
+ GET_GOT goffsetq
+
+ movq m0, [aboveq]
+ mova m1, [GLOBAL(sh_b12345677)]
+ DEFINE_ARGS dst, stride, stride3
+ lea stride3q, [strideq*3]
+ pshufb m2, m0, [GLOBAL(sh_b23456777)]
+ pavgb m3, m2, m0
+ pxor m2, m0
+ pshufb m0, m1
+ pand m2, [GLOBAL(pb_1)]
+ psubb m3, m2
+ pavgb m0, m3
+
+ ; store 4 lines
+ movq [dstq ], m0
+ pshufb m0, m1
+ movq [dstq+strideq ], m0
+ pshufb m0, m1
+ movq [dstq+strideq*2], m0
+ pshufb m0, m1
+ movq [dstq+stride3q ], m0
+ pshufb m0, m1
+ lea dstq, [dstq+strideq*4]
+
+ ; store next 4 lines
+ movq [dstq ], m0
+ pshufb m0, m1
+ movq [dstq+strideq ], m0
+ pshufb m0, m1
+ movq [dstq+strideq*2], m0
+ pshufb m0, m1
+ movq [dstq+stride3q ], m0
+
+ RESTORE_GOT
+ RET
+
+INIT_XMM ssse3
+cglobal d45_predictor_16x16, 3, 6, 4, dst, stride, above, dst8, line, goffset
+ GET_GOT goffsetq
+
+ mova m0, [aboveq]
+ DEFINE_ARGS dst, stride, stride3, dst8, line
+ lea stride3q, [strideq*3]
+ lea dst8q, [dstq+strideq*8]
+ mova m1, [GLOBAL(sh_b123456789abcdeff)]
+ pshufb m2, m0, [GLOBAL(sh_b23456789abcdefff)]
+ pavgb m3, m2, m0
+ pxor m2, m0
+ pshufb m0, m1
+ pand m2, [GLOBAL(pb_1)]
+ psubb m3, m2
+ pavgb m0, m3
+
+ ; first 4 lines and first half of 3rd 4 lines
+ mov lined, 2
+.loop:
+ mova [dstq ], m0
+ movhps [dst8q ], m0
+ pshufb m0, m1
+ mova [dstq +strideq ], m0
+ movhps [dst8q+strideq ], m0
+ pshufb m0, m1
+ mova [dstq +strideq*2 ], m0
+ movhps [dst8q+strideq*2 ], m0
+ pshufb m0, m1
+ mova [dstq +stride3q ], m0
+ movhps [dst8q+stride3q ], m0
+ pshufb m0, m1
+ lea dstq, [dstq +strideq*4]
+ lea dst8q, [dst8q+strideq*4]
+ dec lined
+ jnz .loop
+
+ ; bottom-right 8x8 block
+ movhps [dstq +8], m0
+ movhps [dstq+strideq +8], m0
+ movhps [dstq+strideq*2+8], m0
+ movhps [dstq+stride3q +8], m0
+ lea dstq, [dstq+strideq*4]
+ movhps [dstq +8], m0
+ movhps [dstq+strideq +8], m0
+ movhps [dstq+strideq*2+8], m0
+ movhps [dstq+stride3q +8], m0
+
+ RESTORE_GOT
+ RET
+
+INIT_XMM ssse3
+cglobal d45_predictor_32x32, 3, 6, 7, dst, stride, above, dst16, line, goffset
+ GET_GOT goffsetq
+
+ mova m0, [aboveq]
+ mova m4, [aboveq+16]
+ DEFINE_ARGS dst, stride, stride3, dst16, line
+ lea stride3q, [strideq*3]
+ lea dst16q, [dstq +strideq*8]
+ lea dst16q, [dst16q+strideq*8]
+ mova m1, [GLOBAL(sh_b123456789abcdeff)]
+ pshufb m2, m4, [GLOBAL(sh_b23456789abcdefff)]
+ pavgb m3, m2, m4
+ pxor m2, m4
+ palignr m5, m4, m0, 1
+ palignr m6, m4, m0, 2
+ pshufb m4, m1
+ pand m2, [GLOBAL(pb_1)]
+ psubb m3, m2
+ pavgb m4, m3
+ pavgb m3, m0, m6
+ pxor m0, m6
+ pand m0, [GLOBAL(pb_1)]
+ psubb m3, m0
+ pavgb m5, m3
+
+ ; write 4x4 lines (and the first half of the second 4x4 lines)
+ mov lined, 4
+.loop:
+ mova [dstq ], m5
+ mova [dstq +16], m4
+ mova [dst16q ], m4
+ palignr m3, m4, m5, 1
+ pshufb m4, m1
+ mova [dstq +strideq ], m3
+ mova [dstq +strideq +16], m4
+ mova [dst16q+strideq ], m4
+ palignr m5, m4, m3, 1
+ pshufb m4, m1
+ mova [dstq +strideq*2 ], m5
+ mova [dstq +strideq*2+16], m4
+ mova [dst16q+strideq*2 ], m4
+ palignr m3, m4, m5, 1
+ pshufb m4, m1
+ mova [dstq +stride3q ], m3
+ mova [dstq +stride3q +16], m4
+ mova [dst16q+stride3q ], m4
+ palignr m5, m4, m3, 1
+ pshufb m4, m1
+ lea dstq, [dstq +strideq*4]
+ lea dst16q, [dst16q+strideq*4]
+ dec lined
+ jnz .loop
+
+ ; write second half of second 4x4 lines
+ mova [dstq +16], m4
+ mova [dstq +strideq +16], m4
+ mova [dstq +strideq*2+16], m4
+ mova [dstq +stride3q +16], m4
+ lea dstq, [dstq +strideq*4]
+ mova [dstq +16], m4
+ mova [dstq +strideq +16], m4
+ mova [dstq +strideq*2+16], m4
+ mova [dstq +stride3q +16], m4
+ lea dstq, [dstq +strideq*4]
+ mova [dstq +16], m4
+ mova [dstq +strideq +16], m4
+ mova [dstq +strideq*2+16], m4
+ mova [dstq +stride3q +16], m4
+ lea dstq, [dstq +strideq*4]
+ mova [dstq +16], m4
+ mova [dstq +strideq +16], m4
+ mova [dstq +strideq*2+16], m4
+ mova [dstq +stride3q +16], m4
+
+ RESTORE_GOT
+ RET
+
+; ------------------------------------------
+; input: x, y, z, result
+;
+; trick from pascal
+; (x+2y+z+2)>>2 can be calculated as:
+; result = avg(x,z)
+; result -= xor(x,z) & 1
+; result = avg(result,y)
+; ------------------------------------------
+%macro X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 4
+ pavgb %4, %1, %3
+ pxor %3, %1
+ pand %3, [GLOBAL(pb_1)]
+ psubb %4, %3
+ pavgb %4, %2
+%endmacro
+
+INIT_XMM ssse3
+cglobal d63_predictor_4x4, 3, 4, 5, dst, stride, above, goffset
+ GET_GOT goffsetq
+
+ movq m3, [aboveq]
+ pshufb m1, m3, [GLOBAL(sh_b23456777)]
+ pshufb m2, m3, [GLOBAL(sh_b12345677)]
+
+ X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m3, m2, m1, m4
+ pavgb m3, m2
+
+ ; store 4 lines
+ movd [dstq ], m3
+ movd [dstq+strideq], m4
+ lea dstq, [dstq+strideq*2]
+ psrldq m3, 1
+ psrldq m4, 1
+ movd [dstq ], m3
+ movd [dstq+strideq], m4
+ RESTORE_GOT
+ RET
+
+INIT_XMM ssse3
+cglobal d63_predictor_8x8, 3, 4, 5, dst, stride, above, goffset
+ GET_GOT goffsetq
+
+ movq m3, [aboveq]
+ DEFINE_ARGS dst, stride, stride3
+ lea stride3q, [strideq*3]
+ pshufb m1, m3, [GLOBAL(sh_b2345677777777777)]
+ pshufb m0, m3, [GLOBAL(sh_b0123456777777777)]
+ pshufb m2, m3, [GLOBAL(sh_b1234567777777777)]
+ pshufb m3, [GLOBAL(sh_b0123456777777777)]
+
+ X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m0, m2, m1, m4
+ pavgb m3, m2
+
+ ; store 4 lines
+ movq [dstq ], m3
+ movq [dstq+strideq], m4
+ psrldq m3, 1
+ psrldq m4, 1
+ movq [dstq+strideq*2], m3
+ movq [dstq+stride3q ], m4
+ lea dstq, [dstq+strideq*4]
+ psrldq m3, 1
+ psrldq m4, 1
+
+ ; store 4 lines
+ movq [dstq ], m3
+ movq [dstq+strideq], m4
+ psrldq m3, 1
+ psrldq m4, 1
+ movq [dstq+strideq*2], m3
+ movq [dstq+stride3q ], m4
+ RESTORE_GOT
+ RET
+
+INIT_XMM ssse3
+cglobal d63_predictor_16x16, 3, 5, 5, dst, stride, above, line, goffset
+ GET_GOT goffsetq
+
+ mova m0, [aboveq]
+ DEFINE_ARGS dst, stride, stride3, line
+ lea stride3q, [strideq*3]
+ mova m1, [GLOBAL(sh_b123456789abcdeff)]
+ pshufb m2, m0, [GLOBAL(sh_b23456789abcdefff)]
+ pshufb m3, m0, m1
+
+ X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m0, m3, m2, m4
+ pavgb m0, m3
+
+ mov lined, 4
+.loop:
+ mova [dstq ], m0
+ mova [dstq+strideq ], m4
+ pshufb m0, m1
+ pshufb m4, m1
+ mova [dstq+strideq*2], m0
+ mova [dstq+stride3q ], m4
+ pshufb m0, m1
+ pshufb m4, m1
+ lea dstq, [dstq+strideq*4]
+ dec lined
+ jnz .loop
+ RESTORE_GOT
+ REP_RET
+
+INIT_XMM ssse3
+cglobal d63_predictor_32x32, 3, 5, 8, dst, stride, above, line, goffset
+ GET_GOT goffsetq
+
+ mova m0, [aboveq]
+ mova m7, [aboveq+16]
+ DEFINE_ARGS dst, stride, stride3, line
+ mova m1, [GLOBAL(sh_b123456789abcdeff)]
+ lea stride3q, [strideq*3]
+ pshufb m2, m7, [GLOBAL(sh_b23456789abcdefff)]
+ pshufb m3, m7, m1
+
+ X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m7, m3, m2, m4
+ palignr m6, m7, m0, 1
+ palignr m5, m7, m0, 2
+ pavgb m7, m3
+
+ X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m0, m6, m5, m2
+ pavgb m0, m6
+
+ mov lined, 8
+.loop:
+ mova [dstq ], m0
+ mova [dstq +16], m7
+ mova [dstq+strideq ], m2
+ mova [dstq+strideq +16], m4
+ palignr m3, m7, m0, 1
+ palignr m5, m4, m2, 1
+ pshufb m7, m1
+ pshufb m4, m1
+
+ mova [dstq+strideq*2 ], m3
+ mova [dstq+strideq*2+16], m7
+ mova [dstq+stride3q ], m5
+ mova [dstq+stride3q +16], m4
+ palignr m0, m7, m3, 1
+ palignr m2, m4, m5, 1
+ pshufb m7, m1
+ pshufb m4, m1
+ lea dstq, [dstq+strideq*4]
+ dec lined
+ jnz .loop
+ RESTORE_GOT
+ REP_RET
+
+INIT_XMM ssse3
+cglobal d153_predictor_4x4, 4, 5, 4, dst, stride, above, left, goffset
+ GET_GOT goffsetq
+ movd m0, [leftq] ; l1, l2, l3, l4
+ movd m1, [aboveq-1] ; tl, t1, t2, t3
+ punpckldq m0, m1 ; l1, l2, l3, l4, tl, t1, t2, t3
+ pshufb m0, [GLOBAL(sh_b32104567)]; l4, l3, l2, l1, tl, t1, t2, t3
+ psrldq m1, m0, 1 ; l3, l2, l1, tl, t1, t2, t3
+ psrldq m2, m0, 2 ; l2, l1, tl, t1, t2, t3
+ ; comments below are for a predictor like this
+ ; A1 B1 C1 D1
+ ; A2 B2 A1 B1
+ ; A3 B3 A2 B2
+ ; A4 B4 A3 B3
+ X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m0, m1, m2, m3 ; 3-tap avg B4 B3 B2 B1 C1 D1
+ pavgb m1, m0 ; 2-tap avg A4 A3 A2 A1
+
+ punpcklqdq m3, m1 ; B4 B3 B2 B1 C1 D1 x x A4 A3 A2 A1 ..
+
+ DEFINE_ARGS dst, stride, stride3
+ lea stride3q, [strideq*3]
+ pshufb m3, [GLOBAL(sh_b8091a2b345)] ; A4 B4 A3 B3 A2 B2 A1 B1 C1 D1 ..
+ movd [dstq+stride3q ], m3
+ psrldq m3, 2 ; A3 B3 A2 B2 A1 B1 C1 D1 ..
+ movd [dstq+strideq*2], m3
+ psrldq m3, 2 ; A2 B2 A1 B1 C1 D1 ..
+ movd [dstq+strideq ], m3
+ psrldq m3, 2 ; A1 B1 C1 D1 ..
+ movd [dstq ], m3
+ RESTORE_GOT
+ RET
+
+INIT_XMM ssse3
+cglobal d153_predictor_8x8, 4, 5, 8, dst, stride, above, left, goffset
+ GET_GOT goffsetq
+ movq m0, [leftq] ; [0- 7] l1-8 [byte]
+ movhps m0, [aboveq-1] ; [8-15] tl, t1-7 [byte]
+ pshufb m1, m0, [GLOBAL(sh_b76543210)] ; l8-1 [word]
+ pshufb m2, m0, [GLOBAL(sh_b65432108)] ; l7-1,tl [word]
+ pshufb m3, m0, [GLOBAL(sh_b54321089)] ; l6-1,tl,t1 [word]
+ pshufb m0, [GLOBAL(sh_b89abcdef)] ; tl,t1-7 [word]
+ psrldq m4, m0, 1 ; t1-7 [word]
+ psrldq m5, m0, 2 ; t2-7 [word]
+ ; comments below are for a predictor like this
+ ; A1 B1 C1 D1 E1 F1 G1 H1
+ ; A2 B2 A1 B1 C1 D1 E1 F1
+ ; A3 B3 A2 B2 A1 B1 C1 D1
+ ; A4 B4 A3 B3 A2 B2 A1 B1
+ ; A5 B5 A4 B4 A3 B3 A2 B2
+ ; A6 B6 A5 B5 A4 B4 A3 B3
+ ; A7 B7 A6 B6 A5 B5 A4 B4
+ ; A8 B8 A7 B7 A6 B6 A5 B5
+ pavgb m6, m1, m2 ; 2-tap avg A8-A1
+
+ X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m0, m4, m5, m7 ; 3-tap avg C-H1
+
+ X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m1, m2, m3, m0 ; 3-tap avg B8-1
+
+ punpcklbw m6, m0 ; A-B8, A-B7 ... A-B2, A-B1
+
+ DEFINE_ARGS dst, stride, stride3
+ lea stride3q, [strideq*3]
+
+ movhps [dstq+stride3q], m6 ; A-B4, A-B3, A-B2, A-B1
+ palignr m0, m7, m6, 10 ; A-B3, A-B2, A-B1, C-H1
+ movq [dstq+strideq*2], m0
+ psrldq m0, 2 ; A-B2, A-B1, C-H1
+ movq [dstq+strideq ], m0
+ psrldq m0, 2 ; A-H1
+ movq [dstq ], m0
+ lea dstq, [dstq+strideq*4]
+ movq [dstq+stride3q ], m6 ; A-B8, A-B7, A-B6, A-B5
+ psrldq m6, 2 ; A-B7, A-B6, A-B5, A-B4
+ movq [dstq+strideq*2], m6
+ psrldq m6, 2 ; A-B6, A-B5, A-B4, A-B3
+ movq [dstq+strideq ], m6
+ psrldq m6, 2 ; A-B5, A-B4, A-B3, A-B2
+ movq [dstq ], m6
+ RESTORE_GOT
+ RET
+
+INIT_XMM ssse3
+cglobal d153_predictor_16x16, 4, 5, 8, dst, stride, above, left, goffset
+ GET_GOT goffsetq
+ mova m0, [leftq]
+ movu m7, [aboveq-1]
+ ; comments below are for a predictor like this
+ ; A1 B1 C1 D1 E1 F1 G1 H1 I1 J1 K1 L1 M1 N1 O1 P1
+ ; A2 B2 A1 B1 C1 D1 E1 F1 G1 H1 I1 J1 K1 L1 M1 N1
+ ; A3 B3 A2 B2 A1 B1 C1 D1 E1 F1 G1 H1 I1 J1 K1 L1
+ ; A4 B4 A3 B3 A2 B2 A1 B1 C1 D1 E1 F1 G1 H1 I1 J1
+ ; A5 B5 A4 B4 A3 B3 A2 B2 A1 B1 C1 D1 E1 F1 G1 H1
+ ; A6 B6 A5 B5 A4 B4 A3 B3 A2 B2 A1 B1 C1 D1 E1 F1
+ ; A7 B7 A6 B6 A5 B5 A4 B4 A3 B3 A2 B2 A1 B1 C1 D1
+ ; A8 B8 A7 B7 A6 B6 A5 B5 A4 B4 A3 B3 A2 B2 A1 B1
+ ; A9 B9 A8 B8 A7 B7 A6 B6 A5 B5 A4 B4 A3 B3 A2 B2
+ ; Aa Ba A9 B9 A8 B8 A7 B7 A6 B6 A5 B5 A4 B4 A3 B3
+ ; Ab Bb Aa Ba A9 B9 A8 B8 A7 B7 A6 B6 A5 B5 A4 B4
+ ; Ac Bc Ab Bb Aa Ba A9 B9 A8 B8 A7 B7 A6 B6 A5 B5
+ ; Ad Bd Ac Bc Ab Bb Aa Ba A9 B9 A8 B8 A7 B7 A6 B6
+ ; Ae Be Ad Bd Ac Bc Ab Bb Aa Ba A9 B9 A8 B8 A7 B7
+ ; Af Bf Ae Be Ad Bd Ac Bc Ab Bb Aa Ba A9 B9 A8 B8
+ ; Ag Bg Af Bf Ae Be Ad Bd Ac Bc Ab Bb Aa Ba A9 B9
+ pshufb m6, m7, [GLOBAL(sh_bfedcba9876543210)]
+ palignr m5, m0, m6, 15
+ palignr m3, m0, m6, 14
+
+ X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m0, m5, m3, m4 ; 3-tap avg B3-Bg
+ pshufb m1, m0, [GLOBAL(sh_b123456789abcdeff)]
+ pavgb m5, m0 ; A1 - Ag
+
+ punpcklbw m0, m4, m5 ; A-B8 ... A-B1
+ punpckhbw m4, m5 ; A-B9 ... A-Bg
+
+ pshufb m3, m7, [GLOBAL(sh_b123456789abcdeff)]
+ pshufb m5, m7, [GLOBAL(sh_b23456789abcdefff)]
+
+ X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m7, m3, m5, m1 ; 3-tap avg C1-P1
+
+ pshufb m6, m0, [GLOBAL(sh_bfedcba9876543210)]
+ DEFINE_ARGS dst, stride, stride3
+ lea stride3q, [strideq*3]
+ palignr m2, m1, m6, 14
+ mova [dstq ], m2
+ palignr m2, m1, m6, 12
+ mova [dstq+strideq ], m2
+ palignr m2, m1, m6, 10
+ mova [dstq+strideq*2], m2
+ palignr m2, m1, m6, 8
+ mova [dstq+stride3q ], m2
+ lea dstq, [dstq+strideq*4]
+ palignr m2, m1, m6, 6
+ mova [dstq ], m2
+ palignr m2, m1, m6, 4
+ mova [dstq+strideq ], m2
+ palignr m2, m1, m6, 2
+ mova [dstq+strideq*2], m2
+ pshufb m4, [GLOBAL(sh_bfedcba9876543210)]
+ mova [dstq+stride3q ], m6
+ lea dstq, [dstq+strideq*4]
+
+ palignr m2, m6, m4, 14
+ mova [dstq ], m2
+ palignr m2, m6, m4, 12
+ mova [dstq+strideq ], m2
+ palignr m2, m6, m4, 10
+ mova [dstq+strideq*2], m2
+ palignr m2, m6, m4, 8
+ mova [dstq+stride3q ], m2
+ lea dstq, [dstq+strideq*4]
+ palignr m2, m6, m4, 6
+ mova [dstq ], m2
+ palignr m2, m6, m4, 4
+ mova [dstq+strideq ], m2
+ palignr m2, m6, m4, 2
+ mova [dstq+strideq*2], m2
+ mova [dstq+stride3q ], m4
+ RESTORE_GOT
+ RET
+
+INIT_XMM ssse3
+cglobal d153_predictor_32x32, 4, 5, 8, dst, stride, above, left, goffset
+ GET_GOT goffsetq
+ mova m0, [leftq]
+ movu m7, [aboveq-1]
+ movu m1, [aboveq+15]
+
+ pshufb m4, m1, [GLOBAL(sh_b123456789abcdeff)]
+ pshufb m6, m1, [GLOBAL(sh_b23456789abcdefff)]
+
+ X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m1, m4, m6, m2 ; 3-tap avg above [high]
+
+ palignr m3, m1, m7, 1
+ palignr m5, m1, m7, 2
+
+ X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m7, m3, m5, m1 ; 3-tap avg above [low]
+
+ pshufb m7, [GLOBAL(sh_bfedcba9876543210)]
+ palignr m5, m0, m7, 15
+ palignr m3, m0, m7, 14
+
+ X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m0, m5, m3, m4 ; 3-tap avg B3-Bg
+ pavgb m5, m0 ; A1 - Ag
+ punpcklbw m6, m4, m5 ; A-B8 ... A-B1
+ punpckhbw m4, m5 ; A-B9 ... A-Bg
+ pshufb m6, [GLOBAL(sh_bfedcba9876543210)]
+ pshufb m4, [GLOBAL(sh_bfedcba9876543210)]
+
+ DEFINE_ARGS dst, stride, stride3, left, line
+ lea stride3q, [strideq*3]
+
+ palignr m5, m2, m1, 14
+ palignr m7, m1, m6, 14
+ mova [dstq ], m7
+ mova [dstq+16 ], m5
+ palignr m5, m2, m1, 12
+ palignr m7, m1, m6, 12
+ mova [dstq+strideq ], m7
+ mova [dstq+strideq+16 ], m5
+ palignr m5, m2, m1, 10
+ palignr m7, m1, m6, 10
+ mova [dstq+strideq*2 ], m7
+ mova [dstq+strideq*2+16], m5
+ palignr m5, m2, m1, 8
+ palignr m7, m1, m6, 8
+ mova [dstq+stride3q ], m7
+ mova [dstq+stride3q+16 ], m5
+ lea dstq, [dstq+strideq*4]
+ palignr m5, m2, m1, 6
+ palignr m7, m1, m6, 6
+ mova [dstq ], m7
+ mova [dstq+16 ], m5
+ palignr m5, m2, m1, 4
+ palignr m7, m1, m6, 4
+ mova [dstq+strideq ], m7
+ mova [dstq+strideq+16 ], m5
+ palignr m5, m2, m1, 2
+ palignr m7, m1, m6, 2
+ mova [dstq+strideq*2 ], m7
+ mova [dstq+strideq*2+16], m5
+ mova [dstq+stride3q ], m6
+ mova [dstq+stride3q+16 ], m1
+ lea dstq, [dstq+strideq*4]
+
+ palignr m5, m1, m6, 14
+ palignr m3, m6, m4, 14
+ mova [dstq ], m3
+ mova [dstq+16 ], m5
+ palignr m5, m1, m6, 12
+ palignr m3, m6, m4, 12
+ mova [dstq+strideq ], m3
+ mova [dstq+strideq+16 ], m5
+ palignr m5, m1, m6, 10
+ palignr m3, m6, m4, 10
+ mova [dstq+strideq*2 ], m3
+ mova [dstq+strideq*2+16], m5
+ palignr m5, m1, m6, 8
+ palignr m3, m6, m4, 8
+ mova [dstq+stride3q ], m3
+ mova [dstq+stride3q+16 ], m5
+ lea dstq, [dstq+strideq*4]
+ palignr m5, m1, m6, 6
+ palignr m3, m6, m4, 6
+ mova [dstq ], m3
+ mova [dstq+16 ], m5
+ palignr m5, m1, m6, 4
+ palignr m3, m6, m4, 4
+ mova [dstq+strideq ], m3
+ mova [dstq+strideq+16 ], m5
+ palignr m5, m1, m6, 2
+ palignr m3, m6, m4, 2
+ mova [dstq+strideq*2 ], m3
+ mova [dstq+strideq*2+16], m5
+ mova [dstq+stride3q ], m4
+ mova [dstq+stride3q+16 ], m6
+ lea dstq, [dstq+strideq*4]
+
+ mova m7, [leftq]
+ mova m3, [leftq+16]
+ palignr m5, m3, m7, 15
+ palignr m0, m3, m7, 14
+
+ X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m3, m5, m0, m2 ; 3-tap avg Bh -
+ pavgb m5, m3 ; Ah -
+ punpcklbw m3, m2, m5 ; A-B8 ... A-B1
+ punpckhbw m2, m5 ; A-B9 ... A-Bg
+ pshufb m3, [GLOBAL(sh_bfedcba9876543210)]
+ pshufb m2, [GLOBAL(sh_bfedcba9876543210)]
+
+ palignr m7, m6, m4, 14
+ palignr m0, m4, m3, 14
+ mova [dstq ], m0
+ mova [dstq+16 ], m7
+ palignr m7, m6, m4, 12
+ palignr m0, m4, m3, 12
+ mova [dstq+strideq ], m0
+ mova [dstq+strideq+16 ], m7
+ palignr m7, m6, m4, 10
+ palignr m0, m4, m3, 10
+ mova [dstq+strideq*2 ], m0
+ mova [dstq+strideq*2+16], m7
+ palignr m7, m6, m4, 8
+ palignr m0, m4, m3, 8
+ mova [dstq+stride3q ], m0
+ mova [dstq+stride3q+16 ], m7
+ lea dstq, [dstq+strideq*4]
+ palignr m7, m6, m4, 6
+ palignr m0, m4, m3, 6
+ mova [dstq ], m0
+ mova [dstq+16 ], m7
+ palignr m7, m6, m4, 4
+ palignr m0, m4, m3, 4
+ mova [dstq+strideq ], m0
+ mova [dstq+strideq+16 ], m7
+ palignr m7, m6, m4, 2
+ palignr m0, m4, m3, 2
+ mova [dstq+strideq*2 ], m0
+ mova [dstq+strideq*2+16], m7
+ mova [dstq+stride3q ], m3
+ mova [dstq+stride3q+16 ], m4
+ lea dstq, [dstq+strideq*4]
+
+ palignr m7, m4, m3, 14
+ palignr m0, m3, m2, 14
+ mova [dstq ], m0
+ mova [dstq+16 ], m7
+ palignr m7, m4, m3, 12
+ palignr m0, m3, m2, 12
+ mova [dstq+strideq ], m0
+ mova [dstq+strideq+16 ], m7
+ palignr m7, m4, m3, 10
+ palignr m0, m3, m2, 10
+ mova [dstq+strideq*2 ], m0
+ mova [dstq+strideq*2+16], m7
+ palignr m7, m4, m3, 8
+ palignr m0, m3, m2, 8
+ mova [dstq+stride3q ], m0
+ mova [dstq+stride3q+16 ], m7
+ lea dstq, [dstq+strideq*4]
+ palignr m7, m4, m3, 6
+ palignr m0, m3, m2, 6
+ mova [dstq ], m0
+ mova [dstq+16 ], m7
+ palignr m7, m4, m3, 4
+ palignr m0, m3, m2, 4
+ mova [dstq+strideq ], m0
+ mova [dstq+strideq+16 ], m7
+ palignr m7, m4, m3, 2
+ palignr m0, m3, m2, 2
+ mova [dstq+strideq*2 ], m0
+ mova [dstq+strideq*2+16], m7
+ mova [dstq+stride3q ], m2
+ mova [dstq+stride3q+16 ], m3
+
+ RESTORE_GOT
+ RET
+
+INIT_MMX ssse3
+cglobal d207_predictor_4x4, 4, 5, 4, dst, stride, unused, left, goffset
+ GET_GOT goffsetq
+ movd m0, [leftq] ; abcd [byte]
+ pshufb m1, m0, [GLOBAL(sh_b1233)] ; bcdd [byte]
+ pshufb m3, m0, [GLOBAL(sh_b2333)] ; cddd
+
+ X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m0, m1, m3, m2
+ pavgb m1, m0 ; ab, bc, cd, d [byte]
+
+ punpcklbw m1, m2 ; ab, a2bc, bc, b2cd, cd, c3d, d, d
+ movd [dstq ], m1
+ psrlq m1, 16 ; bc, b2cd, cd, c3d, d, d
+ movd [dstq+strideq], m1
+ lea dstq, [dstq+strideq*2]
+ psrlq m1, 16 ; cd, c3d, d, d
+ movd [dstq ], m1
+ pshufw m1, m1, q1111 ; d, d, d, d
+ movd [dstq+strideq], m1
+ RESTORE_GOT
+ RET
+
+INIT_XMM ssse3
+cglobal d207_predictor_8x8, 4, 5, 4, dst, stride, stride3, left, goffset
+ GET_GOT goffsetq
+ movq m3, [leftq] ; abcdefgh [byte]
+ lea stride3q, [strideq*3]
+
+ pshufb m1, m3, [GLOBAL(sh_b2345677777777777)]
+ pshufb m0, m3, [GLOBAL(sh_b0123456777777777)]
+ pshufb m2, m3, [GLOBAL(sh_b1234567777777777)]
+
+ X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m0, m2, m1, m3
+ pavgb m0, m2
+ punpcklbw m0, m3 ; interleaved output
+
+ movq [dstq ], m0
+ psrldq m0, 2
+ movq [dstq+strideq ], m0
+ psrldq m0, 2
+ movq [dstq+strideq*2], m0
+ psrldq m0, 2
+ movq [dstq+stride3q ], m0
+ lea dstq, [dstq+strideq*4]
+ pshufhw m0, m0, q0000 ; de, d2ef, ef, e2fg, fg, f2gh, gh, g3h, 8xh
+ psrldq m0, 2
+ movq [dstq ], m0
+ psrldq m0, 2
+ movq [dstq+strideq ], m0
+ psrldq m0, 2
+ movq [dstq+strideq*2], m0
+ psrldq m0, 2
+ movq [dstq+stride3q ], m0
+ RESTORE_GOT
+ RET
+
+INIT_XMM ssse3
+cglobal d207_predictor_16x16, 4, 5, 5, dst, stride, stride3, left, goffset
+ GET_GOT goffsetq
+ lea stride3q, [strideq*3]
+ mova m0, [leftq] ; abcdefghijklmnop [byte]
+ pshufb m1, m0, [GLOBAL(sh_b123456789abcdeff)] ; bcdefghijklmnopp
+ pshufb m2, m0, [GLOBAL(sh_b23456789abcdefff)]
+
+ X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m0, m1, m2, m3
+ pavgb m1, m0 ; ab, bc, cd .. no, op, pp [byte]
+
+ punpckhbw m4, m1, m3 ; interleaved input
+ punpcklbw m1, m3 ; interleaved output
+ mova [dstq ], m1
+ palignr m3, m4, m1, 2
+ mova [dstq+strideq ], m3
+ palignr m3, m4, m1, 4
+ mova [dstq+strideq*2], m3
+ palignr m3, m4, m1, 6
+ mova [dstq+stride3q ], m3
+ lea dstq, [dstq+strideq*4]
+ palignr m3, m4, m1, 8
+ mova [dstq ], m3
+ palignr m3, m4, m1, 10
+ mova [dstq+strideq ], m3
+ palignr m3, m4, m1, 12
+ mova [dstq+strideq*2], m3
+ palignr m3, m4, m1, 14
+ mova [dstq+stride3q ], m3
+ DEFINE_ARGS dst, stride, stride3, line
+ mov lined, 2
+ mova m0, [GLOBAL(sh_b23456789abcdefff)]
+.loop:
+ lea dstq, [dstq+strideq*4]
+ mova [dstq ], m4
+ pshufb m4, m0
+ mova [dstq+strideq ], m4
+ pshufb m4, m0
+ mova [dstq+strideq*2], m4
+ pshufb m4, m0
+ mova [dstq+stride3q ], m4
+ pshufb m4, m0
+ dec lined
+ jnz .loop
+ RESTORE_GOT
+ REP_RET
+
+INIT_XMM ssse3
+cglobal d207_predictor_32x32, 4, 5, 8, dst, stride, stride3, left, goffset
+ GET_GOT goffsetq
+ lea stride3q, [strideq*3]
+ mova m1, [leftq] ; 0-15 [byte]
+ mova m2, [leftq+16] ; 16-31 [byte]
+ pshufb m0, m2, [GLOBAL(sh_b23456789abcdefff)]
+ pshufb m4, m2, [GLOBAL(sh_b123456789abcdeff)]
+
+ X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m2, m4, m0, m3
+ palignr m6, m2, m1, 1
+ palignr m5, m2, m1, 2
+ pavgb m2, m4 ; high 16px even lines
+
+ X_PLUS_2Y_PLUS_Z_PLUS_2_RSH_2 m1, m6, m5, m0
+ pavgb m1, m6 ; low 16px even lines
+
+ punpckhbw m6, m1, m0 ; interleaved output 2
+ punpcklbw m1, m0 ; interleaved output 1
+
+ punpckhbw m7, m2, m3 ; interleaved output 4
+ punpcklbw m2, m3 ; interleaved output 3
+
+ ; output 1st 8 lines (and half of 2nd 8 lines)
+ DEFINE_ARGS dst, stride, stride3, dst8
+ lea dst8q, [dstq+strideq*8]
+ mova [dstq ], m1
+ mova [dstq +16], m6
+ mova [dst8q ], m6
+ palignr m0, m6, m1, 2
+ palignr m4, m2, m6, 2
+ mova [dstq +strideq ], m0
+ mova [dstq +strideq +16], m4
+ mova [dst8q+strideq ], m4
+ palignr m0, m6, m1, 4
+ palignr m4, m2, m6, 4
+ mova [dstq +strideq*2 ], m0
+ mova [dstq +strideq*2+16], m4
+ mova [dst8q+strideq*2 ], m4
+ palignr m0, m6, m1, 6
+ palignr m4, m2, m6, 6
+ mova [dstq +stride3q ], m0
+ mova [dstq +stride3q +16], m4
+ mova [dst8q+stride3q ], m4
+ lea dstq, [dstq +strideq*4]
+ lea dst8q, [dst8q+strideq*4]
+ palignr m0, m6, m1, 8
+ palignr m4, m2, m6, 8
+ mova [dstq ], m0
+ mova [dstq +16], m4
+ mova [dst8q ], m4
+ palignr m0, m6, m1, 10
+ palignr m4, m2, m6, 10
+ mova [dstq +strideq ], m0
+ mova [dstq +strideq +16], m4
+ mova [dst8q+strideq ], m4
+ palignr m0, m6, m1, 12
+ palignr m4, m2, m6, 12
+ mova [dstq +strideq*2 ], m0
+ mova [dstq +strideq*2+16], m4
+ mova [dst8q+strideq*2 ], m4
+ palignr m0, m6, m1, 14
+ palignr m4, m2, m6, 14
+ mova [dstq +stride3q ], m0
+ mova [dstq +stride3q +16], m4
+ mova [dst8q+stride3q ], m4
+ lea dstq, [dstq+strideq*4]
+ lea dst8q, [dst8q+strideq*4]
+
+ ; output 2nd half of 2nd 8 lines and half of 3rd 8 lines
+ mova [dstq +16], m2
+ mova [dst8q ], m2
+ palignr m4, m7, m2, 2
+ mova [dstq +strideq +16], m4
+ mova [dst8q+strideq ], m4
+ palignr m4, m7, m2, 4
+ mova [dstq +strideq*2+16], m4
+ mova [dst8q+strideq*2 ], m4
+ palignr m4, m7, m2, 6
+ mova [dstq +stride3q +16], m4
+ mova [dst8q+stride3q ], m4
+ lea dstq, [dstq+strideq*4]
+ lea dst8q, [dst8q+strideq*4]
+ palignr m4, m7, m2, 8
+ mova [dstq +16], m4
+ mova [dst8q ], m4
+ palignr m4, m7, m2, 10
+ mova [dstq +strideq +16], m4
+ mova [dst8q+strideq ], m4
+ palignr m4, m7, m2, 12
+ mova [dstq +strideq*2+16], m4
+ mova [dst8q+strideq*2 ], m4
+ palignr m4, m7, m2, 14
+ mova [dstq +stride3q +16], m4
+ mova [dst8q+stride3q ], m4
+ lea dstq, [dstq+strideq*4]
+ lea dst8q, [dst8q+strideq*4]
+
+ ; output 2nd half of 3rd 8 lines and half of 4th 8 lines
+ mova m0, [GLOBAL(sh_b23456789abcdefff)]
+ mova [dstq +16], m7
+ mova [dst8q ], m7
+ pshufb m7, m0
+ mova [dstq +strideq +16], m7
+ mova [dst8q+strideq ], m7
+ pshufb m7, m0
+ mova [dstq +strideq*2+16], m7
+ mova [dst8q+strideq*2 ], m7
+ pshufb m7, m0
+ mova [dstq +stride3q +16], m7
+ mova [dst8q+stride3q ], m7
+ pshufb m7, m0
+ lea dstq, [dstq+strideq*4]
+ lea dst8q, [dst8q+strideq*4]
+ mova [dstq +16], m7
+ mova [dst8q ], m7
+ pshufb m7, m0
+ mova [dstq +strideq +16], m7
+ mova [dst8q+strideq ], m7
+ pshufb m7, m0
+ mova [dstq +strideq*2+16], m7
+ mova [dst8q+strideq*2 ], m7
+ pshufb m7, m0
+ mova [dstq +stride3q +16], m7
+ mova [dst8q+stride3q ], m7
+ pshufb m7, m0
+ lea dstq, [dstq+strideq*4]
+
+ ; output last half of 4th 8 lines
+ mova [dstq +16], m7
+ mova [dstq +strideq +16], m7
+ mova [dstq +strideq*2+16], m7
+ mova [dstq +stride3q +16], m7
+ lea dstq, [dstq+strideq*4]
+ mova [dstq +16], m7
+ mova [dstq +strideq +16], m7
+ mova [dstq +strideq*2+16], m7
+ mova [dstq +stride3q +16], m7
+
+ ; done!
+ RESTORE_GOT
+ RET