ref: 9114f0afdb23d98ac0704832db43a88b4ca4af01
parent: 2d8f581330114f7fa73ce714b827724c19ff0465
author: James Yu <james.yu@linaro.org>
date: Thu Jan 30 06:54:35 EST 2014
VP9 common for ARMv8 by using NEON intrinsics 08 Add vp9_idct4x4_1_add_neon.c - vp9_idct4x4_1_add_neon Change-Id: Ieab9af107dbd07a4f9503bc945890c90faccb8ac Signed-off-by: James Yu <james.yu@linaro.org>
--- a/vp9/common/arm/neon/vp9_idct4x4_1_add_neon.asm
+++ /dev/null
@@ -1,68 +1,0 @@
-;
-; Copyright (c) 2013 The WebM project authors. All Rights Reserved.
-;
-; Use of this source code is governed by a BSD-style license and patent
-; grant that can be found in the LICENSE file in the root of the source
-; tree. All contributing project authors may be found in the AUTHORS
-; file in the root of the source tree.
-;
-
-
- EXPORT |vp9_idct4x4_1_add_neon|
- ARM
- REQUIRE8
- PRESERVE8
-
- AREA ||.text||, CODE, READONLY, ALIGN=2
-
-;void vp9_idct4x4_1_add_neon(int16_t *input, uint8_t *dest,
-; int dest_stride)
-;
-; r0 int16_t input
-; r1 uint8_t *dest
-; r2 int dest_stride)
-
-|vp9_idct4x4_1_add_neon| PROC
- ldrsh r0, [r0]
-
- ; generate cospi_16_64 = 11585
- mov r12, #0x2d00
- add r12, #0x41
-
- ; out = dct_const_round_shift(input[0] * cospi_16_64)
- mul r0, r0, r12 ; input[0] * cospi_16_64
- add r0, r0, #0x2000 ; +(1 << ((DCT_CONST_BITS) - 1))
- asr r0, r0, #14 ; >> DCT_CONST_BITS
-
- ; out = dct_const_round_shift(out * cospi_16_64)
- mul r0, r0, r12 ; out * cospi_16_64
- mov r12, r1 ; save dest
- add r0, r0, #0x2000 ; +(1 << ((DCT_CONST_BITS) - 1))
- asr r0, r0, #14 ; >> DCT_CONST_BITS
-
- ; a1 = ROUND_POWER_OF_TWO(out, 4)
- add r0, r0, #8 ; + (1 <<((4) - 1))
- asr r0, r0, #4 ; >> 4
-
- vdup.s16 q0, r0 ; duplicate a1
-
- vld1.32 {d2[0]}, [r1], r2
- vld1.32 {d2[1]}, [r1], r2
- vld1.32 {d4[0]}, [r1], r2
- vld1.32 {d4[1]}, [r1]
-
- vaddw.u8 q8, q0, d2 ; dest[x] + a1
- vaddw.u8 q9, q0, d4
-
- vqmovun.s16 d6, q8 ; clip_pixel
- vqmovun.s16 d7, q9
-
- vst1.32 {d6[0]}, [r12], r2
- vst1.32 {d6[1]}, [r12], r2
- vst1.32 {d7[0]}, [r12], r2
- vst1.32 {d7[1]}, [r12]
-
- bx lr
- ENDP ; |vp9_idct4x4_1_add_neon|
-
- END
--- /dev/null
+++ b/vp9/common/arm/neon/vp9_idct4x4_1_add_neon.c
@@ -1,0 +1,48 @@
+/*
+ * 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/common/vp9_idct.h"
+
+void vp9_idct4x4_1_add_neon(
+ int16_t *input,
+ uint8_t *dest,
+ int dest_stride) {
+ uint8x8_t d6u8;
+ uint32x2_t d2u32 = vdup_n_u32(0);
+ uint16x8_t q8u16;
+ int16x8_t q0s16;
+ uint8_t *d1, *d2;
+ int16_t i, a1, cospi_16_64 = 11585;
+ int16_t out = dct_const_round_shift(input[0] * cospi_16_64);
+ out = dct_const_round_shift(out * cospi_16_64);
+ a1 = ROUND_POWER_OF_TWO(out, 4);
+
+ q0s16 = vdupq_n_s16(a1);
+
+ // dc_only_idct_add
+ d1 = d2 = dest;
+ for (i = 0; i < 2; i++) {
+ d2u32 = vld1_lane_u32((const uint32_t *)d1, d2u32, 0);
+ d1 += dest_stride;
+ d2u32 = vld1_lane_u32((const uint32_t *)d1, d2u32, 1);
+ d1 += dest_stride;
+
+ q8u16 = vaddw_u8(vreinterpretq_u16_s16(q0s16),
+ vreinterpret_u8_u32(d2u32));
+ d6u8 = vqmovun_s16(vreinterpretq_s16_u16(q8u16));
+
+ vst1_lane_u32((uint32_t *)d2, vreinterpret_u32_u8(d6u8), 0);
+ d2 += dest_stride;
+ vst1_lane_u32((uint32_t *)d2, vreinterpret_u32_u8(d6u8), 1);
+ d2 += dest_stride;
+ }
+ return;
+}
--- /dev/null
+++ b/vp9/common/arm/neon/vp9_idct4x4_1_add_neon_asm.asm
@@ -1,0 +1,68 @@
+;
+; Copyright (c) 2013 The WebM project authors. All Rights Reserved.
+;
+; Use of this source code is governed by a BSD-style license and patent
+; grant that can be found in the LICENSE file in the root of the source
+; tree. All contributing project authors may be found in the AUTHORS
+; file in the root of the source tree.
+;
+
+
+ EXPORT |vp9_idct4x4_1_add_neon|
+ ARM
+ REQUIRE8
+ PRESERVE8
+
+ AREA ||.text||, CODE, READONLY, ALIGN=2
+
+;void vp9_idct4x4_1_add_neon(int16_t *input, uint8_t *dest,
+; int dest_stride)
+;
+; r0 int16_t input
+; r1 uint8_t *dest
+; r2 int dest_stride)
+
+|vp9_idct4x4_1_add_neon| PROC
+ ldrsh r0, [r0]
+
+ ; generate cospi_16_64 = 11585
+ mov r12, #0x2d00
+ add r12, #0x41
+
+ ; out = dct_const_round_shift(input[0] * cospi_16_64)
+ mul r0, r0, r12 ; input[0] * cospi_16_64
+ add r0, r0, #0x2000 ; +(1 << ((DCT_CONST_BITS) - 1))
+ asr r0, r0, #14 ; >> DCT_CONST_BITS
+
+ ; out = dct_const_round_shift(out * cospi_16_64)
+ mul r0, r0, r12 ; out * cospi_16_64
+ mov r12, r1 ; save dest
+ add r0, r0, #0x2000 ; +(1 << ((DCT_CONST_BITS) - 1))
+ asr r0, r0, #14 ; >> DCT_CONST_BITS
+
+ ; a1 = ROUND_POWER_OF_TWO(out, 4)
+ add r0, r0, #8 ; + (1 <<((4) - 1))
+ asr r0, r0, #4 ; >> 4
+
+ vdup.s16 q0, r0 ; duplicate a1
+
+ vld1.32 {d2[0]}, [r1], r2
+ vld1.32 {d2[1]}, [r1], r2
+ vld1.32 {d4[0]}, [r1], r2
+ vld1.32 {d4[1]}, [r1]
+
+ vaddw.u8 q8, q0, d2 ; dest[x] + a1
+ vaddw.u8 q9, q0, d4
+
+ vqmovun.s16 d6, q8 ; clip_pixel
+ vqmovun.s16 d7, q9
+
+ vst1.32 {d6[0]}, [r12], r2
+ vst1.32 {d6[1]}, [r12], r2
+ vst1.32 {d7[0]}, [r12], r2
+ vst1.32 {d7[1]}, [r12]
+
+ bx lr
+ ENDP ; |vp9_idct4x4_1_add_neon|
+
+ END
--- a/vp9/common/vp9_rtcd_defs.pl
+++ b/vp9/common/vp9_rtcd_defs.pl
@@ -422,8 +422,7 @@
specialize qw/vp9_iwht4x4_16_add/;
} else {
add_proto qw/void vp9_idct4x4_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
- specialize qw/vp9_idct4x4_1_add sse2 neon_asm dspr2/;
- $vp9_idct4x4_1_add_neon_asm=vp9_idct4x4_1_add_neon;
+ specialize qw/vp9_idct4x4_1_add sse2 neon dspr2/;
add_proto qw/void vp9_idct4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
specialize qw/vp9_idct4x4_16_add sse2 neon_asm dspr2/;
--- a/vp9/vp9_common.mk
+++ b/vp9/vp9_common.mk
@@ -134,7 +134,6 @@
VP9_COMMON_SRCS-$(HAVE_NEON_ASM) += common/arm/neon/vp9_idct16x16_neon.c
VP9_COMMON_SRCS-$(HAVE_NEON_ASM) += common/arm/neon/vp9_loopfilter_16_neon_asm$(ASM)
VP9_COMMON_SRCS-$(HAVE_NEON_ASM) += common/arm/neon/vp9_dc_only_idct_add_neon$(ASM)
-VP9_COMMON_SRCS-$(HAVE_NEON_ASM) += common/arm/neon/vp9_idct4x4_1_add_neon$(ASM)
VP9_COMMON_SRCS-$(HAVE_NEON_ASM) += common/arm/neon/vp9_idct4x4_add_neon$(ASM)
VP9_COMMON_SRCS-$(HAVE_NEON_ASM) += common/arm/neon/vp9_idct8x8_1_add_neon$(ASM)
VP9_COMMON_SRCS-$(HAVE_NEON_ASM) += common/arm/neon/vp9_idct8x8_add_neon$(ASM)
@@ -156,6 +155,7 @@
VP9_COMMON_SRCS-yes += common/arm/neon/vp9_convolve8_neon_asm$(ASM)
VP9_COMMON_SRCS-yes += common/arm/neon/vp9_convolve_neon.c
VP9_COMMON_SRCS-yes += common/arm/neon/vp9_copy_neon_asm$(ASM)
+VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct4x4_1_add_neon_asm$(ASM)
VP9_COMMON_SRCS-yes += common/arm/neon/vp9_loopfilter_neon_asm$(ASM)
VP9_COMMON_SRCS-yes += common/arm/neon/vp9_loopfilter_16_neon.c
else
@@ -165,6 +165,7 @@
VP9_COMMON_SRCS-yes += common/arm/neon/vp9_convolve8_neon.c
VP9_COMMON_SRCS-yes += common/arm/neon/vp9_convolve_neon.c
VP9_COMMON_SRCS-yes += common/arm/neon/vp9_copy_neon.c
+VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct4x4_1_add_neon.c
VP9_COMMON_SRCS-yes += common/arm/neon/vp9_loopfilter_neon.c
VP9_COMMON_SRCS-yes += common/arm/neon/vp9_loopfilter_16_neon.c
endif # HAVE_NEON