ref: 6d6fdd9c3d763c6fbcd4f79a07ed2ec131500bfd
parent: 8013aaa10bf7b44cc85381ef31105cb1ee3ddbfd
author: James Yu <james.yu@linaro.org>
date: Mon Jan 6 12:56:49 EST 2014
VP8 encoder for ARMv8 by using NEON intrinsics 3 Add subtract_neon.c - vp8_subtract_b_neon - vp8_subtract_mby_neon - vp8_subtract_mbuv_neon Change-Id: If9a17a093478552e3e3276eeaa3f098b9021d08c Signed-off-by: James Yu <james.yu@linaro.org>
--- a/test/subtract_test.cc
+++ b/test/subtract_test.cc
@@ -105,7 +105,7 @@
INSTANTIATE_TEST_CASE_P(C, SubtractBlockTest,
::testing::Values(vp8_subtract_b_c));
-#if HAVE_NEON_ASM
+#if HAVE_NEON
INSTANTIATE_TEST_CASE_P(NEON, SubtractBlockTest,
::testing::Values(vp8_subtract_b_neon));
#endif
--- a/vp8/common/rtcd_defs.pl
+++ b/vp8/common/rtcd_defs.pl
@@ -502,19 +502,16 @@
$vp8_mbuverror_sse2=vp8_mbuverror_xmm;
add_proto qw/void vp8_subtract_b/, "struct block *be, struct blockd *bd, int pitch";
-specialize qw/vp8_subtract_b mmx sse2 media neon_asm/;
+specialize qw/vp8_subtract_b mmx sse2 media neon/;
$vp8_subtract_b_media=vp8_subtract_b_armv6;
-$vp8_subtract_b_neon_asm=vp8_subtract_b_neon;
add_proto qw/void vp8_subtract_mby/, "short *diff, unsigned char *src, int src_stride, unsigned char *pred, int pred_stride";
-specialize qw/vp8_subtract_mby mmx sse2 media neon_asm/;
+specialize qw/vp8_subtract_mby mmx sse2 media neon/;
$vp8_subtract_mby_media=vp8_subtract_mby_armv6;
-$vp8_subtract_mby_neon_asm=vp8_subtract_mby_neon;
add_proto qw/void vp8_subtract_mbuv/, "short *diff, unsigned char *usrc, unsigned char *vsrc, int src_stride, unsigned char *upred, unsigned char *vpred, int pred_stride";
-specialize qw/vp8_subtract_mbuv mmx sse2 media neon_asm/;
+specialize qw/vp8_subtract_mbuv mmx sse2 media neon/;
$vp8_subtract_mbuv_media=vp8_subtract_mbuv_armv6;
-$vp8_subtract_mbuv_neon_asm=vp8_subtract_mbuv_neon;
#
# Motion search
--- a/vp8/encoder/arm/neon/subtract_neon.asm
+++ /dev/null
@@ -1,205 +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.
-;
-
- EXPORT |vp8_subtract_b_neon|
- EXPORT |vp8_subtract_mby_neon|
- EXPORT |vp8_subtract_mbuv_neon|
-
- INCLUDE vp8_asm_enc_offsets.asm
-
- ARM
- REQUIRE8
- PRESERVE8
-
- AREA ||.text||, CODE, READONLY, ALIGN=2
-
-;void vp8_subtract_b_neon(BLOCK *be, BLOCKD *bd, int pitch)
-|vp8_subtract_b_neon| PROC
-
- stmfd sp!, {r4-r7}-
- ldr r3, [r0, #vp8_block_base_src]
- ldr r4, [r0, #vp8_block_src]
- ldr r5, [r0, #vp8_block_src_diff]
- ldr r3, [r3]
- ldr r6, [r0, #vp8_block_src_stride]
- add r3, r3, r4 ; src = *base_src + src
- ldr r7, [r1, #vp8_blockd_predictor]
-
- vld1.8 {d0}, [r3], r6 ;load src- vld1.8 {d1}, [r7], r2 ;load pred- vld1.8 {d2}, [r3], r6- vld1.8 {d3}, [r7], r2- vld1.8 {d4}, [r3], r6- vld1.8 {d5}, [r7], r2- vld1.8 {d6}, [r3], r6- vld1.8 {d7}, [r7], r2-
- vsubl.u8 q10, d0, d1
- vsubl.u8 q11, d2, d3
- vsubl.u8 q12, d4, d5
- vsubl.u8 q13, d6, d7
-
- mov r2, r2, lsl #1
-
- vst1.16 {d20}, [r5], r2 ;store diff- vst1.16 {d22}, [r5], r2- vst1.16 {d24}, [r5], r2- vst1.16 {d26}, [r5], r2-
- ldmfd sp!, {r4-r7}- bx lr
-
- ENDP
-
-
-;==========================================
-;void vp8_subtract_mby_neon(short *diff, unsigned char *src, int src_stride
-; unsigned char *pred, int pred_stride)
-|vp8_subtract_mby_neon| PROC
- push {r4-r7}- vpush {d8-d15}-
- mov r12, #4
- ldr r4, [sp, #80] ; pred_stride
- mov r6, #32 ; "diff" stride x2
- add r5, r0, #16 ; second diff pointer
-
-subtract_mby_loop
- vld1.8 {q0}, [r1], r2 ;load src- vld1.8 {q1}, [r3], r4 ;load pred- vld1.8 {q2}, [r1], r2- vld1.8 {q3}, [r3], r4- vld1.8 {q4}, [r1], r2- vld1.8 {q5}, [r3], r4- vld1.8 {q6}, [r1], r2- vld1.8 {q7}, [r3], r4-
- vsubl.u8 q8, d0, d2
- vsubl.u8 q9, d1, d3
- vsubl.u8 q10, d4, d6
- vsubl.u8 q11, d5, d7
- vsubl.u8 q12, d8, d10
- vsubl.u8 q13, d9, d11
- vsubl.u8 q14, d12, d14
- vsubl.u8 q15, d13, d15
-
- vst1.16 {q8}, [r0], r6 ;store diff- vst1.16 {q9}, [r5], r6- vst1.16 {q10}, [r0], r6- vst1.16 {q11}, [r5], r6- vst1.16 {q12}, [r0], r6- vst1.16 {q13}, [r5], r6- vst1.16 {q14}, [r0], r6- vst1.16 {q15}, [r5], r6-
- subs r12, r12, #1
- bne subtract_mby_loop
-
- vpop {d8-d15}- pop {r4-r7}- bx lr
- ENDP
-
-;=================================
-;void vp8_subtract_mbuv_c(short *diff, unsigned char *usrc, unsigned char *vsrc,
-; int src_stride, unsigned char *upred,
-; unsigned char *vpred, int pred_stride)
-
-|vp8_subtract_mbuv_neon| PROC
- push {r4-r7}- vpush {d8-d15}-
- ldr r4, [sp, #80] ; upred
- ldr r5, [sp, #84] ; vpred
- ldr r6, [sp, #88] ; pred_stride
- add r0, r0, #512 ; short *udiff = diff + 256;
- mov r12, #32 ; "diff" stride x2
- add r7, r0, #16 ; second diff pointer
-
-;u
- vld1.8 {d0}, [r1], r3 ;load usrc- vld1.8 {d1}, [r4], r6 ;load upred- vld1.8 {d2}, [r1], r3- vld1.8 {d3}, [r4], r6- vld1.8 {d4}, [r1], r3- vld1.8 {d5}, [r4], r6- vld1.8 {d6}, [r1], r3- vld1.8 {d7}, [r4], r6- vld1.8 {d8}, [r1], r3- vld1.8 {d9}, [r4], r6- vld1.8 {d10}, [r1], r3- vld1.8 {d11}, [r4], r6- vld1.8 {d12}, [r1], r3- vld1.8 {d13}, [r4], r6- vld1.8 {d14}, [r1], r3- vld1.8 {d15}, [r4], r6-
- vsubl.u8 q8, d0, d1
- vsubl.u8 q9, d2, d3
- vsubl.u8 q10, d4, d5
- vsubl.u8 q11, d6, d7
- vsubl.u8 q12, d8, d9
- vsubl.u8 q13, d10, d11
- vsubl.u8 q14, d12, d13
- vsubl.u8 q15, d14, d15
-
- vst1.16 {q8}, [r0], r12 ;store diff- vst1.16 {q9}, [r7], r12- vst1.16 {q10}, [r0], r12- vst1.16 {q11}, [r7], r12- vst1.16 {q12}, [r0], r12- vst1.16 {q13}, [r7], r12- vst1.16 {q14}, [r0], r12- vst1.16 {q15}, [r7], r12-
-;v
- vld1.8 {d0}, [r2], r3 ;load vsrc- vld1.8 {d1}, [r5], r6 ;load vpred- vld1.8 {d2}, [r2], r3- vld1.8 {d3}, [r5], r6- vld1.8 {d4}, [r2], r3- vld1.8 {d5}, [r5], r6- vld1.8 {d6}, [r2], r3- vld1.8 {d7}, [r5], r6- vld1.8 {d8}, [r2], r3- vld1.8 {d9}, [r5], r6- vld1.8 {d10}, [r2], r3- vld1.8 {d11}, [r5], r6- vld1.8 {d12}, [r2], r3- vld1.8 {d13}, [r5], r6- vld1.8 {d14}, [r2], r3- vld1.8 {d15}, [r5], r6-
- vsubl.u8 q8, d0, d1
- vsubl.u8 q9, d2, d3
- vsubl.u8 q10, d4, d5
- vsubl.u8 q11, d6, d7
- vsubl.u8 q12, d8, d9
- vsubl.u8 q13, d10, d11
- vsubl.u8 q14, d12, d13
- vsubl.u8 q15, d14, d15
-
- vst1.16 {q8}, [r0], r12 ;store diff- vst1.16 {q9}, [r7], r12- vst1.16 {q10}, [r0], r12- vst1.16 {q11}, [r7], r12- vst1.16 {q12}, [r0], r12- vst1.16 {q13}, [r7], r12- vst1.16 {q14}, [r0], r12- vst1.16 {q15}, [r7], r12-
- vpop {d8-d15}- pop {r4-r7}- bx lr
-
- ENDP
-
- END
--- /dev/null
+++ b/vp8/encoder/arm/neon/subtract_neon.c
@@ -1,0 +1,154 @@
+/*
+ * 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 "vp8/encoder/block.h"
+
+void vp8_subtract_b_neon(
+ BLOCK *be,
+ BLOCKD *bd,
+ int pitch) {+ unsigned char *src_ptr, *predictor;
+ int src_stride;
+ int16_t *src_diff;
+ uint8x8_t d0u8, d1u8, d2u8, d3u8, d4u8, d5u8, d6u8, d7u8;
+ uint16x8_t q10u16, q11u16, q12u16, q13u16;
+
+ src_ptr = *be->base_src + be->src;
+ src_stride = be->src_stride;
+ predictor = bd->predictor;
+
+ d0u8 = vld1_u8(src_ptr);
+ src_ptr += src_stride;
+ d2u8 = vld1_u8(src_ptr);
+ src_ptr += src_stride;
+ d4u8 = vld1_u8(src_ptr);
+ src_ptr += src_stride;
+ d6u8 = vld1_u8(src_ptr);
+
+ d1u8 = vld1_u8(predictor);
+ predictor += pitch;
+ d3u8 = vld1_u8(predictor);
+ predictor += pitch;
+ d5u8 = vld1_u8(predictor);
+ predictor += pitch;
+ d7u8 = vld1_u8(predictor);
+
+ q10u16 = vsubl_u8(d0u8, d1u8);
+ q11u16 = vsubl_u8(d2u8, d3u8);
+ q12u16 = vsubl_u8(d4u8, d5u8);
+ q13u16 = vsubl_u8(d6u8, d7u8);
+
+ src_diff = be->src_diff;
+ vst1_u16((uint16_t *)src_diff, vget_low_u16(q10u16));
+ src_diff += pitch;
+ vst1_u16((uint16_t *)src_diff, vget_low_u16(q11u16));
+ src_diff += pitch;
+ vst1_u16((uint16_t *)src_diff, vget_low_u16(q12u16));
+ src_diff += pitch;
+ vst1_u16((uint16_t *)src_diff, vget_low_u16(q13u16));
+ return;
+}
+
+void vp8_subtract_mby_neon(
+ int16_t *diff,
+ unsigned char *src,
+ int src_stride,
+ unsigned char *pred,
+ int pred_stride) {+ int i;
+ uint8x16_t q0u8, q1u8, q2u8, q3u8;
+ uint16x8_t q8u16, q9u16, q10u16, q11u16;
+
+ for (i = 0; i < 8; i++) { // subtract_mby_loop+ q0u8 = vld1q_u8(src);
+ src += src_stride;
+ q2u8 = vld1q_u8(src);
+ src += src_stride;
+ q1u8 = vld1q_u8(pred);
+ pred += pred_stride;
+ q3u8 = vld1q_u8(pred);
+ pred += pred_stride;
+
+ q8u16 = vsubl_u8(vget_low_u8(q0u8), vget_low_u8(q1u8));
+ q9u16 = vsubl_u8(vget_high_u8(q0u8), vget_high_u8(q1u8));
+ q10u16 = vsubl_u8(vget_low_u8(q2u8), vget_low_u8(q3u8));
+ q11u16 = vsubl_u8(vget_high_u8(q2u8), vget_high_u8(q3u8));
+
+ vst1q_u16((uint16_t *)diff, q8u16);
+ diff += 8;
+ vst1q_u16((uint16_t *)diff, q9u16);
+ diff += 8;
+ vst1q_u16((uint16_t *)diff, q10u16);
+ diff += 8;
+ vst1q_u16((uint16_t *)diff, q11u16);
+ diff += 8;
+ }
+ return;
+}
+
+void vp8_subtract_mbuv_neon(
+ int16_t *diff,
+ unsigned char *usrc,
+ unsigned char *vsrc,
+ int src_stride,
+ unsigned char *upred,
+ unsigned char *vpred,
+ int pred_stride) {+ int i, j;
+ unsigned char *src_ptr, *pred_ptr;
+ uint8x8_t d0u8, d1u8, d2u8, d3u8, d4u8, d5u8, d6u8, d7u8;
+ uint16x8_t q8u16, q9u16, q10u16, q11u16;
+
+ diff += 256;
+ for (i = 0; i < 2; i++) {+ if (i == 0) {+ src_ptr = usrc;
+ pred_ptr = upred;
+ } else if (i == 1) {+ src_ptr = vsrc;
+ pred_ptr = vpred;
+ }
+
+ for (j = 0; j < 2; j++) {+ d0u8 = vld1_u8(src_ptr);
+ src_ptr += src_stride;
+ d1u8 = vld1_u8(pred_ptr);
+ pred_ptr += pred_stride;
+ d2u8 = vld1_u8(src_ptr);
+ src_ptr += src_stride;
+ d3u8 = vld1_u8(pred_ptr);
+ pred_ptr += pred_stride;
+ d4u8 = vld1_u8(src_ptr);
+ src_ptr += src_stride;
+ d5u8 = vld1_u8(pred_ptr);
+ pred_ptr += pred_stride;
+ d6u8 = vld1_u8(src_ptr);
+ src_ptr += src_stride;
+ d7u8 = vld1_u8(pred_ptr);
+ pred_ptr += pred_stride;
+
+ q8u16 = vsubl_u8(d0u8, d1u8);
+ q9u16 = vsubl_u8(d2u8, d3u8);
+ q10u16 = vsubl_u8(d4u8, d5u8);
+ q11u16 = vsubl_u8(d6u8, d7u8);
+
+ vst1q_u16((uint16_t *)diff, q8u16);
+ diff += 8;
+ vst1q_u16((uint16_t *)diff, q9u16);
+ diff += 8;
+ vst1q_u16((uint16_t *)diff, q10u16);
+ diff += 8;
+ vst1q_u16((uint16_t *)diff, q11u16);
+ diff += 8;
+ }
+ }
+ return;
+}
--- a/vp8/vp8cx_arm.mk
+++ b/vp8/vp8cx_arm.mk
@@ -38,9 +38,9 @@
VP8_CX_SRCS-$(HAVE_NEON_ASM) += encoder/arm/neon/fastquantizeb_neon$(ASM)
VP8_CX_SRCS-$(HAVE_NEON_ASM) += encoder/arm/neon/picklpf_arm.c
VP8_CX_SRCS-$(HAVE_NEON_ASM) += encoder/arm/neon/shortfdct_neon$(ASM)
-VP8_CX_SRCS-$(HAVE_NEON_ASM) += encoder/arm/neon/subtract_neon$(ASM)
VP8_CX_SRCS-$(HAVE_NEON_ASM) += encoder/arm/neon/vp8_mse16x16_neon$(ASM)
VP8_CX_SRCS-$(HAVE_NEON_ASM) += encoder/arm/neon/vp8_memcpy_neon$(ASM)
VP8_CX_SRCS-$(HAVE_NEON) += encoder/arm/neon/denoising_neon.c
VP8_CX_SRCS-$(HAVE_NEON) += encoder/arm/neon/vp8_shortwalsh4x4_neon.c
+VP8_CX_SRCS-$(HAVE_NEON) += encoder/arm/neon/subtract_neon.c