mirror of
https://chromium.googlesource.com/libyuv/libyuv
synced 2026-10-28 18:07:12 +08:00
STOREAR30 remove initial shift by 4
- reduced shifts to isolate 10 bits - permi2w instead of packlwd to combine words (ARG) (GB) into ARGB - vpternlogd to combine A,R,G into 1 register Saphhire Rapids 10000 iterations of 1280x720P: Conversion Kernel │ Was │ Now │ Speedup ──────────────────────┼────────────────────────┼──────────────── I420ToAR30_Opt │ 1675 ms │ 1568 ms │ +7.1% H420ToAR30_Opt │ 1632 ms │ 1555 ms │ +5.2% I210ToAR30_Opt │ 2459 ms │ 2167 ms │ +13.5% P010ToAR30_Opt │ 2054 ms │ 1926 ms │ +6.6% P016ToAR30_Opt │ 2043 ms │ 1938 ms │ +5.4% Total Test Suite Time │ 9,866 ms │ 9,155 ms │ +7.8% overall Bug: None Change-Id: I7972f98408a65a2bb51c506145e4c0c63824c424 Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/8259796 Commit-Queue: Frank Barchard <fbarchard@google.com> Reviewed-by: richard winterton <rrwinterton@gmail.com>
This commit is contained in:
Frank Barchard
libyuv-scoped@luci-project-accounts.iam.gserviceaccount.com
parent
8d7ed8af35
commit
44d9bc68f2
@ -1,6 +1,6 @@
|
|||||||
Name: libyuv
|
Name: libyuv
|
||||||
URL: https://chromium.googlesource.com/libyuv/libyuv/
|
URL: https://chromium.googlesource.com/libyuv/libyuv/
|
||||||
Version: 1961
|
Version: 1962
|
||||||
Revision: DEPS
|
Revision: DEPS
|
||||||
License: BSD-3-Clause
|
License: BSD-3-Clause
|
||||||
License File: LICENSE
|
License File: LICENSE
|
||||||
|
|||||||
@ -339,6 +339,7 @@ extern "C" {
|
|||||||
defined(LIBYUV_ENABLE_ROWWIN))
|
defined(LIBYUV_ENABLE_ROWWIN))
|
||||||
#define HAS_RAWTOARGBROW_AVX2
|
#define HAS_RAWTOARGBROW_AVX2
|
||||||
#define HAS_I422TORGB24ROW_AVX2
|
#define HAS_I422TORGB24ROW_AVX2
|
||||||
|
#define HAS_I422TOAR30ROW_AVX2
|
||||||
#define HAS_RGB24TOARGBROW_AVX2
|
#define HAS_RGB24TOARGBROW_AVX2
|
||||||
#define HAS_RGB565TOARGBROW_AVX2
|
#define HAS_RGB565TOARGBROW_AVX2
|
||||||
#define HAS_ARGB1555TOARGBROW_AVX2
|
#define HAS_ARGB1555TOARGBROW_AVX2
|
||||||
@ -350,6 +351,7 @@ extern "C" {
|
|||||||
#define HAS_ARGBSHUFFLEROW_AVX512BW
|
#define HAS_ARGBSHUFFLEROW_AVX512BW
|
||||||
#define HAS_I422TORGB24ROW_AVX512VBMI
|
#define HAS_I422TORGB24ROW_AVX512VBMI
|
||||||
#define HAS_I422TORGB24ROW_AVX512BW
|
#define HAS_I422TORGB24ROW_AVX512BW
|
||||||
|
#define HAS_I422TOAR30ROW_AVX512BW
|
||||||
#define HAS_RGBTOYMATRIXROW_AVX512BW
|
#define HAS_RGBTOYMATRIXROW_AVX512BW
|
||||||
#endif
|
#endif
|
||||||
#define HAS_ARGBTOYMATRIXROW_AVX2
|
#define HAS_ARGBTOYMATRIXROW_AVX2
|
||||||
|
|||||||
@ -11,6 +11,6 @@
|
|||||||
#ifndef INCLUDE_LIBYUV_VERSION_H_
|
#ifndef INCLUDE_LIBYUV_VERSION_H_
|
||||||
#define INCLUDE_LIBYUV_VERSION_H_
|
#define INCLUDE_LIBYUV_VERSION_H_
|
||||||
|
|
||||||
#define LIBYUV_VERSION 1961
|
#define LIBYUV_VERSION 1962
|
||||||
|
|
||||||
#endif // INCLUDE_LIBYUV_VERSION_H_
|
#endif // INCLUDE_LIBYUV_VERSION_H_
|
||||||
|
|||||||
@ -2362,9 +2362,6 @@ void ARGBToUVMatrixRow_AVX512BW(const uint8_t* src_argb,
|
|||||||
|
|
||||||
// Store 8 AR30 values.
|
// Store 8 AR30 values.
|
||||||
#define STOREAR30 \
|
#define STOREAR30 \
|
||||||
"psraw $0x4,%%xmm0 \n" \
|
|
||||||
"psraw $0x4,%%xmm1 \n" \
|
|
||||||
"psraw $0x4,%%xmm2 \n" \
|
|
||||||
"pminsw %%xmm7,%%xmm0 \n" \
|
"pminsw %%xmm7,%%xmm0 \n" \
|
||||||
"pminsw %%xmm7,%%xmm1 \n" \
|
"pminsw %%xmm7,%%xmm1 \n" \
|
||||||
"pminsw %%xmm7,%%xmm2 \n" \
|
"pminsw %%xmm7,%%xmm2 \n" \
|
||||||
@ -2372,11 +2369,13 @@ void ARGBToUVMatrixRow_AVX512BW(const uint8_t* src_argb,
|
|||||||
"pmaxsw %%xmm6,%%xmm1 \n" \
|
"pmaxsw %%xmm6,%%xmm1 \n" \
|
||||||
"pmaxsw %%xmm6,%%xmm2 \n" \
|
"pmaxsw %%xmm6,%%xmm2 \n" \
|
||||||
"movdqa %%xmm1,%%xmm3 \n" \
|
"movdqa %%xmm1,%%xmm3 \n" \
|
||||||
"psllw $10,%%xmm3 \n" \
|
"pand %%xmm7,%%xmm3 \n" \
|
||||||
"psrlw $6,%%xmm1 \n" \
|
"pand %%xmm7,%%xmm2 \n" \
|
||||||
"psllw $4,%%xmm2 \n" \
|
"psrlw $4,%%xmm0 \n" \
|
||||||
|
"psrlw $10,%%xmm1 \n" \
|
||||||
|
"psllw $6,%%xmm3 \n" \
|
||||||
"por %%xmm3,%%xmm0 \n" \
|
"por %%xmm3,%%xmm0 \n" \
|
||||||
"por %%xmm5,%%xmm1 \n" \
|
"por %%xmm5,%%xmm2 \n" \
|
||||||
"por %%xmm1,%%xmm2 \n" \
|
"por %%xmm1,%%xmm2 \n" \
|
||||||
"movdqa %%xmm0,%%xmm3 \n" \
|
"movdqa %%xmm0,%%xmm3 \n" \
|
||||||
"punpcklwd %%xmm2,%%xmm0 \n" \
|
"punpcklwd %%xmm2,%%xmm0 \n" \
|
||||||
@ -2557,7 +2556,8 @@ void OMITFP I422ToAR30Row_SSSE3(const uint8_t* y_buf,
|
|||||||
"sub %[u_buf],%[v_buf] \n"
|
"sub %[u_buf],%[v_buf] \n"
|
||||||
"pcmpeqb %%xmm5,%%xmm5 \n" // all 1s
|
"pcmpeqb %%xmm5,%%xmm5 \n" // all 1s
|
||||||
"movdqa %%xmm5,%%xmm7 \n"
|
"movdqa %%xmm5,%%xmm7 \n"
|
||||||
"psrlw $6,%%xmm7 \n" // 1023 for max
|
"psrlw $6,%%xmm7 \n"
|
||||||
|
"psllw $4,%%xmm7 \n" // 1023 * 16 for max
|
||||||
"psllw $14,%%xmm5 \n" // 0xc000 for 2 alpha bits
|
"psllw $14,%%xmm5 \n" // 0xc000 for 2 alpha bits
|
||||||
"pxor %%xmm6,%%xmm6 \n" // 0 for min
|
"pxor %%xmm6,%%xmm6 \n" // 0 for min
|
||||||
|
|
||||||
@ -2651,7 +2651,8 @@ void OMITFP I210ToAR30Row_SSSE3(const uint16_t* y_buf,
|
|||||||
"sub %[u_buf],%[v_buf] \n"
|
"sub %[u_buf],%[v_buf] \n"
|
||||||
"pcmpeqb %%xmm5,%%xmm5 \n" // all 1s
|
"pcmpeqb %%xmm5,%%xmm5 \n" // all 1s
|
||||||
"movdqa %%xmm5,%%xmm7 \n"
|
"movdqa %%xmm5,%%xmm7 \n"
|
||||||
"psrlw $6,%%xmm7 \n" // 1023 for max
|
"psrlw $6,%%xmm7 \n"
|
||||||
|
"psllw $4,%%xmm7 \n" // 1023 * 16 for max
|
||||||
"psllw $14,%%xmm5 \n" // 0xc000 for 2 alpha bits
|
"psllw $14,%%xmm5 \n" // 0xc000 for 2 alpha bits
|
||||||
"pxor %%xmm6,%%xmm6 \n" // 0 for min
|
"pxor %%xmm6,%%xmm6 \n" // 0 for min
|
||||||
|
|
||||||
@ -2685,7 +2686,8 @@ void OMITFP I212ToAR30Row_SSSE3(const uint16_t* y_buf,
|
|||||||
"sub %[u_buf],%[v_buf] \n"
|
"sub %[u_buf],%[v_buf] \n"
|
||||||
"pcmpeqb %%xmm5,%%xmm5 \n" // all 1s
|
"pcmpeqb %%xmm5,%%xmm5 \n" // all 1s
|
||||||
"movdqa %%xmm5,%%xmm7 \n"
|
"movdqa %%xmm5,%%xmm7 \n"
|
||||||
"psrlw $6,%%xmm7 \n" // 1023 for max
|
"psrlw $6,%%xmm7 \n"
|
||||||
|
"psllw $4,%%xmm7 \n" // 1023 * 16 for max
|
||||||
"psllw $14,%%xmm5 \n" // 0xc000 for 2 alpha bits
|
"psllw $14,%%xmm5 \n" // 0xc000 for 2 alpha bits
|
||||||
"pxor %%xmm6,%%xmm6 \n" // 0 for min
|
"pxor %%xmm6,%%xmm6 \n" // 0 for min
|
||||||
|
|
||||||
@ -2812,7 +2814,8 @@ void OMITFP I410ToAR30Row_SSSE3(const uint16_t* y_buf,
|
|||||||
"sub %[u_buf],%[v_buf] \n"
|
"sub %[u_buf],%[v_buf] \n"
|
||||||
"pcmpeqb %%xmm5,%%xmm5 \n" // all 1s
|
"pcmpeqb %%xmm5,%%xmm5 \n" // all 1s
|
||||||
"movdqa %%xmm5,%%xmm7 \n"
|
"movdqa %%xmm5,%%xmm7 \n"
|
||||||
"psrlw $6,%%xmm7 \n" // 1023 for max
|
"psrlw $6,%%xmm7 \n"
|
||||||
|
"psllw $4,%%xmm7 \n" // 1023 * 16 for max
|
||||||
"psllw $14,%%xmm5 \n" // 0xc000 for 2 alpha bits
|
"psllw $14,%%xmm5 \n" // 0xc000 for 2 alpha bits
|
||||||
"pxor %%xmm6,%%xmm6 \n" // 0 for min
|
"pxor %%xmm6,%%xmm6 \n" // 0 for min
|
||||||
|
|
||||||
@ -3001,7 +3004,8 @@ void OMITFP P210ToAR30Row_SSSE3(const uint16_t* y_buf,
|
|||||||
YUVTORGB_SETUP(yuvconstants)
|
YUVTORGB_SETUP(yuvconstants)
|
||||||
"pcmpeqb %%xmm5,%%xmm5 \n" // all 1s
|
"pcmpeqb %%xmm5,%%xmm5 \n" // all 1s
|
||||||
"movdqa %%xmm5,%%xmm7 \n"
|
"movdqa %%xmm5,%%xmm7 \n"
|
||||||
"psrlw $6,%%xmm7 \n" // 1023 for max
|
"psrlw $6,%%xmm7 \n"
|
||||||
|
"psllw $4,%%xmm7 \n" // 1023 * 16 for max
|
||||||
"psllw $14,%%xmm5 \n" // 0xc000 for 2 alpha bits
|
"psllw $14,%%xmm5 \n" // 0xc000 for 2 alpha bits
|
||||||
"pxor %%xmm6,%%xmm6 \n" // 0 for min
|
"pxor %%xmm6,%%xmm6 \n" // 0 for min
|
||||||
|
|
||||||
@ -3031,7 +3035,8 @@ void OMITFP P410ToAR30Row_SSSE3(const uint16_t* y_buf,
|
|||||||
YUVTORGB_SETUP(yuvconstants)
|
YUVTORGB_SETUP(yuvconstants)
|
||||||
"pcmpeqb %%xmm5,%%xmm5 \n" // all 1s
|
"pcmpeqb %%xmm5,%%xmm5 \n" // all 1s
|
||||||
"movdqa %%xmm5,%%xmm7 \n"
|
"movdqa %%xmm5,%%xmm7 \n"
|
||||||
"psrlw $6,%%xmm7 \n" // 1023 for max
|
"psrlw $6,%%xmm7 \n"
|
||||||
|
"psllw $4,%%xmm7 \n" // 1023 * 16 for max
|
||||||
"psllw $14,%%xmm5 \n" // 0xc000 for 2 alpha bits
|
"psllw $14,%%xmm5 \n" // 0xc000 for 2 alpha bits
|
||||||
"pxor %%xmm6,%%xmm6 \n" // 0 for min
|
"pxor %%xmm6,%%xmm6 \n" // 0 for min
|
||||||
|
|
||||||
@ -3476,52 +3481,48 @@ void OMITFP I422ToRGBARow_SSSE3(const uint8_t* y_buf,
|
|||||||
|
|
||||||
// Store 32 AR30 values.
|
// Store 32 AR30 values.
|
||||||
#define STOREAR30_AVX512BW \
|
#define STOREAR30_AVX512BW \
|
||||||
"vpsraw $0x4,%%zmm0,%%zmm0 \n" \
|
"vpminsw %%zmm7,%%zmm0,%%zmm0 \n" \
|
||||||
"vpsraw $0x4,%%zmm1,%%zmm1 \n" \
|
"vpminsw %%zmm7,%%zmm1,%%zmm1 \n" \
|
||||||
"vpsraw $0x4,%%zmm2,%%zmm2 \n" \
|
"vpminsw %%zmm7,%%zmm2,%%zmm2 \n" \
|
||||||
"vpminsw %%zmm7,%%zmm0,%%zmm0 \n" \
|
"vpmaxsw %%zmm6,%%zmm0,%%zmm0 \n" \
|
||||||
"vpminsw %%zmm7,%%zmm1,%%zmm1 \n" \
|
"vpmaxsw %%zmm6,%%zmm1,%%zmm1 \n" \
|
||||||
"vpminsw %%zmm7,%%zmm2,%%zmm2 \n" \
|
"vpmaxsw %%zmm6,%%zmm2,%%zmm2 \n" \
|
||||||
"vpmaxsw %%zmm6,%%zmm0,%%zmm0 \n" \
|
"vpsrlw $4,%%zmm0,%%zmm0 \n" \
|
||||||
"vpmaxsw %%zmm6,%%zmm1,%%zmm1 \n" \
|
"vpsrlw $10,%%zmm1,%%zmm3 \n" \
|
||||||
"vpmaxsw %%zmm6,%%zmm2,%%zmm2 \n" \
|
"vpsllw $6,%%zmm1,%%zmm1 \n" \
|
||||||
"vpsrlw $6,%%zmm1,%%zmm3 \n" \
|
"vpternlogd $0xf8,%%zmm19,%%zmm1,%%zmm0 \n" \
|
||||||
"vpsllw $4,%%zmm2,%%zmm2 \n" \
|
"vpternlogd $0xe4,%%zmm7,%%zmm3,%%zmm2 \n" \
|
||||||
"vpsllw $10,%%zmm1,%%zmm1 \n" \
|
"vpord %%zmm5,%%zmm2,%%zmm2 \n" \
|
||||||
"vpternlogd $0xfe,%%zmm5,%%zmm3,%%zmm2 \n" \
|
"vmovdqa64 %%zmm0,%%zmm3 \n" \
|
||||||
"vpord %%zmm1,%%zmm0,%%zmm0 \n" \
|
"vpermt2w %%zmm2,%%zmm20,%%zmm0 \n" \
|
||||||
"vpermq %%zmm2,%%zmm18,%%zmm2 \n" \
|
"vpermt2w %%zmm2,%%zmm21,%%zmm3 \n" \
|
||||||
"vpermq %%zmm0,%%zmm18,%%zmm0 \n" \
|
"vmovdqu32 %%zmm0,(%[dst_ar30]) \n" \
|
||||||
"vpunpcklwd %%zmm2,%%zmm0,%%zmm1 \n" \
|
"vmovdqu32 %%zmm3,0x40(%[dst_ar30]) \n" \
|
||||||
"vpunpckhwd %%zmm2,%%zmm0,%%zmm3 \n" \
|
"lea 0x80(%[dst_ar30]), %[dst_ar30] \n"
|
||||||
"vmovdqu32 %%zmm1,(%[dst_ar30]) \n" \
|
|
||||||
"vmovdqu32 %%zmm3,0x40(%[dst_ar30]) \n" \
|
|
||||||
"lea 0x80(%[dst_ar30]), %[dst_ar30] \n"
|
|
||||||
|
|
||||||
// Store 16 AR30 values.
|
// Store 16 AR30 values.
|
||||||
#define STOREAR30_AVX2 \
|
#define STOREAR30_AVX2 \
|
||||||
"vpsraw $0x4,%%ymm0,%%ymm0 \n" \
|
"vpminsw %%ymm7,%%ymm0,%%ymm0 \n" \
|
||||||
"vpsraw $0x4,%%ymm1,%%ymm1 \n" \
|
"vpminsw %%ymm7,%%ymm1,%%ymm1 \n" \
|
||||||
"vpsraw $0x4,%%ymm2,%%ymm2 \n" \
|
"vpminsw %%ymm7,%%ymm2,%%ymm2 \n" \
|
||||||
"vpminsw %%ymm7,%%ymm0,%%ymm0 \n" \
|
"vpmaxsw %%ymm6,%%ymm0,%%ymm0 \n" \
|
||||||
"vpminsw %%ymm7,%%ymm1,%%ymm1 \n" \
|
"vpmaxsw %%ymm6,%%ymm1,%%ymm1 \n" \
|
||||||
"vpminsw %%ymm7,%%ymm2,%%ymm2 \n" \
|
"vpmaxsw %%ymm6,%%ymm2,%%ymm2 \n" \
|
||||||
"vpmaxsw %%ymm6,%%ymm0,%%ymm0 \n" \
|
"vpand %%ymm7,%%ymm1,%%ymm3 \n" \
|
||||||
"vpmaxsw %%ymm6,%%ymm1,%%ymm1 \n" \
|
"vpand %%ymm7,%%ymm2,%%ymm2 \n" \
|
||||||
"vpmaxsw %%ymm6,%%ymm2,%%ymm2 \n" \
|
"vpsrlw $4,%%ymm0,%%ymm0 \n" \
|
||||||
"vpsrlw $6,%%ymm1,%%ymm3 \n" \
|
"vpsrlw $10,%%ymm1,%%ymm1 \n" \
|
||||||
"vpsllw $4,%%ymm2,%%ymm2 \n" \
|
"vpsllw $6,%%ymm3,%%ymm3 \n" \
|
||||||
"vpsllw $10,%%ymm1,%%ymm1 \n" \
|
"vpor %%ymm3,%%ymm0,%%ymm0 \n" \
|
||||||
"vpor %%ymm5,%%ymm3,%%ymm3 \n" \
|
"vpor %%ymm5,%%ymm2,%%ymm2 \n" \
|
||||||
"vpor %%ymm3,%%ymm2,%%ymm2 \n" \
|
"vpor %%ymm1,%%ymm2,%%ymm2 \n" \
|
||||||
"vpor %%ymm1,%%ymm0,%%ymm0 \n" \
|
"vpermq $0xd8,%%ymm2,%%ymm2 \n" \
|
||||||
"vpermq $0xd8,%%ymm2,%%ymm2 \n" \
|
"vpermq $0xd8,%%ymm0,%%ymm0 \n" \
|
||||||
"vpermq $0xd8,%%ymm0,%%ymm0 \n" \
|
"vpunpcklwd %%ymm2,%%ymm0,%%ymm1 \n" \
|
||||||
"vpunpcklwd %%ymm2,%%ymm0,%%ymm1 \n" \
|
"vpunpckhwd %%ymm2,%%ymm0,%%ymm3 \n" \
|
||||||
"vpunpckhwd %%ymm2,%%ymm0,%%ymm3 \n" \
|
"vmovdqu %%ymm1,(%[dst_ar30]) \n" \
|
||||||
"vmovdqu %%ymm1,(%[dst_ar30]) \n" \
|
"vmovdqu %%ymm3,0x20(%[dst_ar30]) \n" \
|
||||||
"vmovdqu %%ymm3,0x20(%[dst_ar30]) \n" \
|
"lea 0x40(%[dst_ar30]), %[dst_ar30] \n"
|
||||||
"lea 0x40(%[dst_ar30]), %[dst_ar30] \n"
|
|
||||||
|
|
||||||
#ifdef HAS_I444TOARGBROW_AVX2
|
#ifdef HAS_I444TOARGBROW_AVX2
|
||||||
// 16 pixels
|
// 16 pixels
|
||||||
@ -3644,6 +3645,15 @@ static const uint64_t kSplitDoubleQuadWords[8] = {0, 1, 4, 4, 2, 3, 4, 4};
|
|||||||
static const uint64_t kUnpermuteAVX512[8] = {0, 4, 1, 5, 2, 6, 3, 7};
|
static const uint64_t kUnpermuteAVX512[8] = {0, 4, 1, 5, 2, 6, 3, 7};
|
||||||
#endif
|
#endif
|
||||||
|
|
||||||
|
#if defined(HAS_I422TOAR30ROW_AVX512BW)
|
||||||
|
static const uint16_t kPermAR30_0[32] = {
|
||||||
|
0, 32, 1, 33, 2, 34, 3, 35, 4, 36, 5, 37, 6, 38, 7, 39,
|
||||||
|
8, 40, 9, 41, 10, 42, 11, 43, 12, 44, 13, 45, 14, 46, 15, 47};
|
||||||
|
static const uint16_t kPermAR30_1[32] = {
|
||||||
|
16, 48, 17, 49, 18, 50, 19, 51, 20, 52, 21, 53, 22, 54, 23, 55,
|
||||||
|
24, 56, 25, 57, 26, 58, 27, 59, 28, 60, 29, 61, 30, 62, 31, 63};
|
||||||
|
#endif
|
||||||
|
|
||||||
#if defined(HAS_I422TOARGBROW_AVX512BW)
|
#if defined(HAS_I422TOARGBROW_AVX512BW)
|
||||||
// 32 pixels
|
// 32 pixels
|
||||||
// 16 UV values upsampled to 32 UV, mixed with 32 Y producing 32 ARGB (128
|
// 16 UV values upsampled to 32 UV, mixed with 32 Y producing 32 ARGB (128
|
||||||
@ -3811,9 +3821,13 @@ void OMITFP I422ToAR30Row_AVX512BW(const uint8_t* y_buf,
|
|||||||
YUVTORGB_SETUP_AVX512BW(yuvconstants)
|
YUVTORGB_SETUP_AVX512BW(yuvconstants)
|
||||||
"sub %[u_buf],%[v_buf] \n"
|
"sub %[u_buf],%[v_buf] \n"
|
||||||
"vpternlogd $0xff,%%zmm5,%%zmm5,%%zmm5 \n" // all 1s
|
"vpternlogd $0xff,%%zmm5,%%zmm5,%%zmm5 \n" // all 1s
|
||||||
"vpsrlw $6,%%zmm5,%%zmm7 \n" // 1023 for max
|
"vpsrlw $6,%%zmm5,%%zmm7 \n"
|
||||||
|
"vpsllw $4,%%zmm7,%%zmm7 \n" // 1023 * 16 for max
|
||||||
|
"vpsllw $10,%%zmm5,%%zmm19 \n" // 0xfc00 for G mask
|
||||||
"vpsllw $14,%%zmm5,%%zmm5 \n" // 0xc000 for 2 alpha bits
|
"vpsllw $14,%%zmm5,%%zmm5 \n" // 0xc000 for 2 alpha bits
|
||||||
"vpxord %%zmm6,%%zmm6,%%zmm6 \n" // 0 for min
|
"vpxord %%zmm6,%%zmm6,%%zmm6 \n" // 0 for min
|
||||||
|
"vmovdqu16 (%[permar30_0]),%%zmm20 \n"
|
||||||
|
"vmovdqu16 (%[permar30_1]),%%zmm21 \n"
|
||||||
|
|
||||||
LABELALIGN
|
LABELALIGN
|
||||||
"1: \n"
|
"1: \n"
|
||||||
@ -3832,9 +3846,12 @@ void OMITFP I422ToAR30Row_AVX512BW(const uint8_t* y_buf,
|
|||||||
: [yuvconstants]"r"(yuvconstants), // %[yuvconstants]
|
: [yuvconstants]"r"(yuvconstants), // %[yuvconstants]
|
||||||
[quadsplitperm]"r"(kSplitQuadWords), // %[quadsplitperm]
|
[quadsplitperm]"r"(kSplitQuadWords), // %[quadsplitperm]
|
||||||
[dquadsplitperm]"r"(kSplitDoubleQuadWords), // %[dquadsplitperm]
|
[dquadsplitperm]"r"(kSplitDoubleQuadWords), // %[dquadsplitperm]
|
||||||
[unperm]"r"(kUnpermuteAVX512) // %[unperm]
|
[unperm]"r"(kUnpermuteAVX512), // %[unperm]
|
||||||
|
[permar30_0]"r"(kPermAR30_0), // %[permar30_0]
|
||||||
|
[permar30_1]"r"(kPermAR30_1) // %[permar30_1]
|
||||||
: "memory", "cc", YUVTORGB_REGS_AVX512BW
|
: "memory", "cc", YUVTORGB_REGS_AVX512BW
|
||||||
"xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7"
|
"xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7",
|
||||||
|
"xmm19", "xmm20", "xmm21"
|
||||||
);
|
);
|
||||||
}
|
}
|
||||||
#endif // HAS_I422TOAR30ROW_AVX512BW
|
#endif // HAS_I422TOAR30ROW_AVX512BW
|
||||||
@ -3852,7 +3869,8 @@ void OMITFP I422ToAR30Row_AVX2(const uint8_t* y_buf,
|
|||||||
YUVTORGB_SETUP_AVX2(yuvconstants)
|
YUVTORGB_SETUP_AVX2(yuvconstants)
|
||||||
"sub %[u_buf],%[v_buf] \n"
|
"sub %[u_buf],%[v_buf] \n"
|
||||||
"vpcmpeqb %%ymm5,%%ymm5,%%ymm5 \n" // all 1s
|
"vpcmpeqb %%ymm5,%%ymm5,%%ymm5 \n" // all 1s
|
||||||
"vpsrlw $6,%%ymm5,%%ymm7 \n" // 1023 for max
|
"vpsrlw $6,%%ymm5,%%ymm7 \n"
|
||||||
|
"vpsllw $4,%%ymm7,%%ymm7 \n" // 1023 * 16 for max
|
||||||
"vpsllw $14,%%ymm5,%%ymm5 \n" // 0xc000 for 2 alpha bits
|
"vpsllw $14,%%ymm5,%%ymm5 \n" // 0xc000 for 2 alpha bits
|
||||||
"vpxor %%ymm6,%%ymm6,%%ymm6 \n" // 0 for min
|
"vpxor %%ymm6,%%ymm6,%%ymm6 \n" // 0 for min
|
||||||
|
|
||||||
@ -3960,7 +3978,8 @@ void OMITFP I210ToAR30Row_AVX2(const uint16_t* y_buf,
|
|||||||
YUVTORGB_SETUP_AVX2(yuvconstants)
|
YUVTORGB_SETUP_AVX2(yuvconstants)
|
||||||
"sub %[u_buf],%[v_buf] \n"
|
"sub %[u_buf],%[v_buf] \n"
|
||||||
"vpcmpeqb %%ymm5,%%ymm5,%%ymm5 \n" // all 1s
|
"vpcmpeqb %%ymm5,%%ymm5,%%ymm5 \n" // all 1s
|
||||||
"vpsrlw $6,%%ymm5,%%ymm7 \n" // 1023 for max
|
"vpsrlw $6,%%ymm5,%%ymm7 \n"
|
||||||
|
"vpsllw $4,%%ymm7,%%ymm7 \n" // 1023 * 16 for max
|
||||||
"vpsllw $14,%%ymm5,%%ymm5 \n" // 0xc000 for 2 alpha bits
|
"vpsllw $14,%%ymm5,%%ymm5 \n" // 0xc000 for 2 alpha bits
|
||||||
"vpxor %%ymm6,%%ymm6,%%ymm6 \n" // 0 for min
|
"vpxor %%ymm6,%%ymm6,%%ymm6 \n" // 0 for min
|
||||||
|
|
||||||
@ -3998,7 +4017,8 @@ void OMITFP I212ToAR30Row_AVX2(const uint16_t* y_buf,
|
|||||||
YUVTORGB_SETUP_AVX2(yuvconstants)
|
YUVTORGB_SETUP_AVX2(yuvconstants)
|
||||||
"sub %[u_buf],%[v_buf] \n"
|
"sub %[u_buf],%[v_buf] \n"
|
||||||
"vpcmpeqb %%ymm5,%%ymm5,%%ymm5 \n" // all 1s
|
"vpcmpeqb %%ymm5,%%ymm5,%%ymm5 \n" // all 1s
|
||||||
"vpsrlw $6,%%ymm5,%%ymm7 \n" // 1023 for max
|
"vpsrlw $6,%%ymm5,%%ymm7 \n"
|
||||||
|
"vpsllw $4,%%ymm7,%%ymm7 \n" // 1023 * 16 for max
|
||||||
"vpsllw $14,%%ymm5,%%ymm5 \n" // 0xc000 for 2 alpha bits
|
"vpsllw $14,%%ymm5,%%ymm5 \n" // 0xc000 for 2 alpha bits
|
||||||
"vpxor %%ymm6,%%ymm6,%%ymm6 \n" // 0 for min
|
"vpxor %%ymm6,%%ymm6,%%ymm6 \n" // 0 for min
|
||||||
|
|
||||||
@ -4141,7 +4161,8 @@ void OMITFP I410ToAR30Row_AVX2(const uint16_t* y_buf,
|
|||||||
YUVTORGB_SETUP_AVX2(yuvconstants)
|
YUVTORGB_SETUP_AVX2(yuvconstants)
|
||||||
"sub %[u_buf],%[v_buf] \n"
|
"sub %[u_buf],%[v_buf] \n"
|
||||||
"vpcmpeqb %%ymm5,%%ymm5,%%ymm5 \n" // all 1s
|
"vpcmpeqb %%ymm5,%%ymm5,%%ymm5 \n" // all 1s
|
||||||
"vpsrlw $6,%%ymm5,%%ymm7 \n" // 1023 for max
|
"vpsrlw $6,%%ymm5,%%ymm7 \n"
|
||||||
|
"vpsllw $4,%%ymm7,%%ymm7 \n" // 1023 * 16 for max
|
||||||
"vpsllw $14,%%ymm5,%%ymm5 \n" // 0xc000 for 2 alpha bits
|
"vpsllw $14,%%ymm5,%%ymm5 \n" // 0xc000 for 2 alpha bits
|
||||||
"vpxor %%ymm6,%%ymm6,%%ymm6 \n" // 0 for min
|
"vpxor %%ymm6,%%ymm6,%%ymm6 \n" // 0 for min
|
||||||
|
|
||||||
@ -4448,7 +4469,8 @@ void OMITFP P210ToAR30Row_AVX2(const uint16_t* y_buf,
|
|||||||
asm volatile (
|
asm volatile (
|
||||||
YUVTORGB_SETUP_AVX2(yuvconstants)
|
YUVTORGB_SETUP_AVX2(yuvconstants)
|
||||||
"vpcmpeqb %%ymm5,%%ymm5,%%ymm5 \n" // all 1s
|
"vpcmpeqb %%ymm5,%%ymm5,%%ymm5 \n" // all 1s
|
||||||
"vpsrlw $6,%%ymm5,%%ymm7 \n" // 1023 for max
|
"vpsrlw $6,%%ymm5,%%ymm7 \n"
|
||||||
|
"vpsllw $4,%%ymm7,%%ymm7 \n" // 1023 * 16 for max
|
||||||
"vpsllw $14,%%ymm5,%%ymm5 \n" // 0xc000 for 2 alpha bits
|
"vpsllw $14,%%ymm5,%%ymm5 \n" // 0xc000 for 2 alpha bits
|
||||||
"vpxor %%ymm6,%%ymm6,%%ymm6 \n" // 0 for min
|
"vpxor %%ymm6,%%ymm6,%%ymm6 \n" // 0 for min
|
||||||
|
|
||||||
@ -4483,7 +4505,8 @@ void OMITFP P410ToAR30Row_AVX2(const uint16_t* y_buf,
|
|||||||
asm volatile (
|
asm volatile (
|
||||||
YUVTORGB_SETUP_AVX2(yuvconstants)
|
YUVTORGB_SETUP_AVX2(yuvconstants)
|
||||||
"vpcmpeqb %%ymm5,%%ymm5,%%ymm5 \n" // all 1s
|
"vpcmpeqb %%ymm5,%%ymm5,%%ymm5 \n" // all 1s
|
||||||
"vpsrlw $6,%%ymm5,%%ymm7 \n" // 1023 for max
|
"vpsrlw $6,%%ymm5,%%ymm7 \n"
|
||||||
|
"vpsllw $4,%%ymm7,%%ymm7 \n" // 1023 * 16 for max
|
||||||
"vpsllw $14,%%ymm5,%%ymm5 \n" // 0xc000 for 2 alpha bits
|
"vpsllw $14,%%ymm5,%%ymm5 \n" // 0xc000 for 2 alpha bits
|
||||||
"vpxor %%ymm6,%%ymm6,%%ymm6 \n" // 0 for min
|
"vpxor %%ymm6,%%ymm6,%%ymm6 \n" // 0 for min
|
||||||
|
|
||||||
|
|||||||
@ -102,6 +102,22 @@ extern "C" {
|
|||||||
_mm_storeu_si128((__m128i*)(dst_argb + 16), xmm1); \
|
_mm_storeu_si128((__m128i*)(dst_argb + 16), xmm1); \
|
||||||
dst_argb += 32;
|
dst_argb += 32;
|
||||||
|
|
||||||
|
// Store 8 AR30 values.
|
||||||
|
#define STOREAR30 \
|
||||||
|
xmm1 = _mm_and_si128(xmm1, xmm7); \
|
||||||
|
xmm2 = _mm_and_si128(xmm2, xmm7); \
|
||||||
|
xmm0 = _mm_srli_epi16(xmm0, 4); \
|
||||||
|
xmm3 = _mm_slli_epi16(xmm1, 6); \
|
||||||
|
xmm0 = _mm_or_si128(xmm0, xmm3); \
|
||||||
|
xmm1 = _mm_srli_epi16(xmm1, 10); \
|
||||||
|
xmm2 = _mm_or_si128(xmm2, xmm6); \
|
||||||
|
xmm2 = _mm_or_si128(xmm2, xmm1); \
|
||||||
|
xmm1 = _mm_unpackhi_epi16(xmm0, xmm2); \
|
||||||
|
xmm0 = _mm_unpacklo_epi16(xmm0, xmm2); \
|
||||||
|
_mm_storeu_si128((__m128i*)dst_ar30, xmm0); \
|
||||||
|
_mm_storeu_si128((__m128i*)(dst_ar30 + 16), xmm1); \
|
||||||
|
dst_ar30 += 32;
|
||||||
|
|
||||||
#if defined(HAS_ARGBTOYMATRIXROW_AVX2)
|
#if defined(HAS_ARGBTOYMATRIXROW_AVX2)
|
||||||
|
|
||||||
#if defined(__clang__) || defined(__GNUC__)
|
#if defined(__clang__) || defined(__GNUC__)
|
||||||
@ -1146,6 +1162,89 @@ void I422ToARGBRow_AVX2(const uint8_t* src_y,
|
|||||||
}
|
}
|
||||||
#endif
|
#endif
|
||||||
|
|
||||||
|
#ifdef HAS_I422TOAR30ROW_AVX2
|
||||||
|
LIBYUV_TARGET_AVX2
|
||||||
|
void I422ToAR30Row_AVX2(const uint8_t* src_y,
|
||||||
|
const uint8_t* src_u,
|
||||||
|
const uint8_t* src_v,
|
||||||
|
uint8_t* dst_ar30,
|
||||||
|
const struct YuvConstants* yuvconstants,
|
||||||
|
int width) {
|
||||||
|
__m256i ymm_kUVToB = _mm256_loadu_si256((const __m256i*)yuvconstants->kUVToB);
|
||||||
|
__m256i ymm_kUVToG = _mm256_loadu_si256((const __m256i*)yuvconstants->kUVToG);
|
||||||
|
__m256i ymm_kUVToR = _mm256_loadu_si256((const __m256i*)yuvconstants->kUVToR);
|
||||||
|
__m256i ymm_kYToRgb = _mm256_loadu_si256((const __m256i*)yuvconstants->kYToRgb);
|
||||||
|
__m256i ymm_kYBiasToRgb = _mm256_loadu_si256((const __m256i*)yuvconstants->kYBiasToRgb);
|
||||||
|
__m256i ymm_128 = _mm256_set1_epi8((char)0x80);
|
||||||
|
__m256i ymm_3ff0 = _mm256_set1_epi16((short)0x3ff0);
|
||||||
|
__m256i ymm_c000 = _mm256_set1_epi16((short)0xc000);
|
||||||
|
__m256i ymm_u_zero = _mm256_setzero_si256();
|
||||||
|
|
||||||
|
ptrdiff_t offset = src_v - src_u;
|
||||||
|
|
||||||
|
while (width >= 16) {
|
||||||
|
__m128i xmm_u = _mm_loadl_epi64((const __m128i*)src_u);
|
||||||
|
__m128i xmm_v = _mm_loadl_epi64((const __m128i*)(src_u + offset));
|
||||||
|
src_u += 8;
|
||||||
|
|
||||||
|
__m256i ymm3 = _mm256_insertf128_si256(ymm_u_zero, xmm_u, 0);
|
||||||
|
__m256i ymm1 = _mm256_insertf128_si256(ymm_u_zero, xmm_v, 0);
|
||||||
|
|
||||||
|
ymm3 = _mm256_unpacklo_epi8(ymm3, ymm1);
|
||||||
|
ymm3 = _mm256_permute4x64_epi64(ymm3, 0xd8);
|
||||||
|
ymm3 = _mm256_unpacklo_epi16(ymm3, ymm3);
|
||||||
|
|
||||||
|
__m128i xmm_y = _mm_loadu_si128((const __m128i*)src_y);
|
||||||
|
src_y += 16;
|
||||||
|
__m256i ymm4 = _mm256_insertf128_si256(ymm_u_zero, xmm_y, 0);
|
||||||
|
ymm4 = _mm256_permute4x64_epi64(ymm4, 0xd8);
|
||||||
|
ymm4 = _mm256_unpacklo_epi8(ymm4, ymm4);
|
||||||
|
|
||||||
|
ymm3 = _mm256_sub_epi8(ymm3, ymm_128);
|
||||||
|
ymm4 = _mm256_mulhi_epu16(ymm4, ymm_kYToRgb);
|
||||||
|
|
||||||
|
__m256i ymm0 = _mm256_maddubs_epi16(ymm_kUVToB, ymm3);
|
||||||
|
ymm1 = _mm256_maddubs_epi16(ymm_kUVToG, ymm3);
|
||||||
|
__m256i ymm2 = _mm256_maddubs_epi16(ymm_kUVToR, ymm3);
|
||||||
|
|
||||||
|
ymm4 = _mm256_add_epi16(ymm4, ymm_kYBiasToRgb);
|
||||||
|
|
||||||
|
ymm0 = _mm256_adds_epi16(ymm0, ymm4);
|
||||||
|
ymm1 = _mm256_subs_epi16(ymm4, ymm1);
|
||||||
|
ymm2 = _mm256_adds_epi16(ymm2, ymm4);
|
||||||
|
|
||||||
|
ymm0 = _mm256_min_epi16(ymm0, ymm_3ff0);
|
||||||
|
ymm1 = _mm256_min_epi16(ymm1, ymm_3ff0);
|
||||||
|
ymm2 = _mm256_min_epi16(ymm2, ymm_3ff0);
|
||||||
|
ymm0 = _mm256_max_epi16(ymm0, ymm_u_zero);
|
||||||
|
ymm1 = _mm256_max_epi16(ymm1, ymm_u_zero);
|
||||||
|
ymm2 = _mm256_max_epi16(ymm2, ymm_u_zero);
|
||||||
|
|
||||||
|
__m256i ymm_g_masked = _mm256_and_si256(ymm1, ymm_3ff0);
|
||||||
|
__m256i ymm_r_masked = _mm256_and_si256(ymm2, ymm_3ff0);
|
||||||
|
ymm0 = _mm256_srli_epi16(ymm0, 4);
|
||||||
|
ymm3 = _mm256_slli_epi16(ymm_g_masked, 6);
|
||||||
|
ymm0 = _mm256_or_si256(ymm0, ymm3);
|
||||||
|
ymm1 = _mm256_srli_epi16(ymm1, 10);
|
||||||
|
ymm2 = _mm256_or_si256(ymm_r_masked, ymm_c000);
|
||||||
|
ymm2 = _mm256_or_si256(ymm2, ymm1);
|
||||||
|
|
||||||
|
ymm0 = _mm256_permute4x64_epi64(ymm0, 0xd8);
|
||||||
|
ymm2 = _mm256_permute4x64_epi64(ymm2, 0xd8);
|
||||||
|
|
||||||
|
__m256i ar30_lo = _mm256_unpacklo_epi16(ymm0, ymm2);
|
||||||
|
__m256i ar30_hi = _mm256_unpackhi_epi16(ymm0, ymm2);
|
||||||
|
|
||||||
|
_mm256_storeu_si256((__m256i*)dst_ar30, ar30_lo);
|
||||||
|
_mm256_storeu_si256((__m256i*)(dst_ar30 + 32), ar30_hi);
|
||||||
|
|
||||||
|
dst_ar30 += 64;
|
||||||
|
width -= 16;
|
||||||
|
}
|
||||||
|
_mm256_zeroupper();
|
||||||
|
}
|
||||||
|
#endif
|
||||||
|
|
||||||
#ifdef HAS_NV12TOARGBROW_AVX2
|
#ifdef HAS_NV12TOARGBROW_AVX2
|
||||||
LIBYUV_TARGET_AVX2
|
LIBYUV_TARGET_AVX2
|
||||||
void NV12ToARGBRow_AVX2(const uint8_t* src_y,
|
void NV12ToARGBRow_AVX2(const uint8_t* src_y,
|
||||||
@ -1602,6 +1701,108 @@ void I422ToRGB24Row_AVX512BW(const uint8_t* src_y,
|
|||||||
}
|
}
|
||||||
#endif
|
#endif
|
||||||
|
|
||||||
|
#ifdef HAS_I422TOAR30ROW_AVX512BW
|
||||||
|
LIBYUV_TARGET_AVX512BW
|
||||||
|
void I422ToAR30Row_AVX512BW(const uint8_t* src_y,
|
||||||
|
const uint8_t* src_u,
|
||||||
|
const uint8_t* src_v,
|
||||||
|
uint8_t* dst_ar30,
|
||||||
|
const struct YuvConstants* yuvconstants,
|
||||||
|
int width) {
|
||||||
|
static const uint64_t kSplitQuadWords[8] = {0, 2, 2, 2, 1, 2, 2, 2};
|
||||||
|
static const uint64_t kSplitDoubleQuadWords[8] = {0, 1, 4, 4, 2, 3, 4, 4};
|
||||||
|
static const uint16_t kPermAR30_0[32] = {
|
||||||
|
0, 32, 1, 33, 2, 34, 3, 35, 4, 36, 5, 37, 6, 38, 7, 39,
|
||||||
|
8, 40, 9, 41, 10, 42, 11, 43, 12, 44, 13, 45, 14, 46, 15, 47};
|
||||||
|
static const uint16_t kPermAR30_1[32] = {
|
||||||
|
16, 48, 17, 49, 18, 50, 19, 51, 20, 52, 21, 53, 22, 54, 23, 55,
|
||||||
|
24, 56, 25, 57, 26, 58, 27, 59, 28, 60, 29, 61, 30, 62, 31, 63};
|
||||||
|
|
||||||
|
// Constants
|
||||||
|
__m512i zmm_kUVToB = _mm512_broadcast_i32x4(_mm_loadu_si128((const __m128i*)yuvconstants->kUVToB));
|
||||||
|
__m512i zmm_kUVToG = _mm512_broadcast_i32x4(_mm_loadu_si128((const __m128i*)yuvconstants->kUVToG));
|
||||||
|
__m512i zmm_kUVToR = _mm512_broadcast_i32x4(_mm_loadu_si128((const __m128i*)yuvconstants->kUVToR));
|
||||||
|
__m512i zmm_kYToRgb = _mm512_broadcast_i32x4(_mm_loadu_si128((const __m128i*)yuvconstants->kYToRgb));
|
||||||
|
__m512i zmm_kYBiasToRgb = _mm512_broadcast_i32x4(_mm_loadu_si128((const __m128i*)yuvconstants->kYBiasToRgb));
|
||||||
|
__m512i zmm_128 = _mm512_set1_epi8((char)0x80);
|
||||||
|
__m512i zmm_3ff0 = _mm512_set1_epi16((short)0x3ff0);
|
||||||
|
__m512i zmm_fc00 = _mm512_set1_epi16((short)0xfc00);
|
||||||
|
__m512i zmm_c000 = _mm512_set1_epi16((short)0xc000);
|
||||||
|
__m512i zmm_zero = _mm512_setzero_si512();
|
||||||
|
|
||||||
|
__m512i zmm_perm0 = _mm512_loadu_si512((const __m512i*)kPermAR30_0);
|
||||||
|
__m512i zmm_perm1 = _mm512_loadu_si512((const __m512i*)kPermAR30_1);
|
||||||
|
__m512i zmm_split = _mm512_loadu_si512((const __m512i*)kSplitQuadWords);
|
||||||
|
__m512i zmm_split_y = _mm512_loadu_si512((const __m512i*)kSplitDoubleQuadWords);
|
||||||
|
|
||||||
|
ptrdiff_t offset = src_v - src_u;
|
||||||
|
|
||||||
|
while (width >= 32) {
|
||||||
|
// READYUV422_AVX512BW
|
||||||
|
__m128i xmm_u = _mm_loadu_si128((const __m128i*)src_u);
|
||||||
|
__m128i xmm_v = _mm_loadu_si128((const __m128i*)(src_u + offset));
|
||||||
|
src_u += 16;
|
||||||
|
|
||||||
|
__m512i zmm_u_val = _mm512_castsi128_si512(xmm_u);
|
||||||
|
__m512i zmm_v_val = _mm512_castsi128_si512(xmm_v);
|
||||||
|
|
||||||
|
zmm_u_val = _mm512_permutexvar_epi64(zmm_split, zmm_u_val);
|
||||||
|
zmm_v_val = _mm512_permutexvar_epi64(zmm_split, zmm_v_val);
|
||||||
|
|
||||||
|
__m512i zmm3 = _mm512_unpacklo_epi8(zmm_u_val, zmm_v_val);
|
||||||
|
zmm3 = _mm512_permutex_epi64(zmm3, 0xd8);
|
||||||
|
zmm3 = _mm512_unpacklo_epi16(zmm3, zmm3);
|
||||||
|
|
||||||
|
__m256i ymm_y = _mm256_loadu_si256((const __m256i*)src_y);
|
||||||
|
src_y += 32;
|
||||||
|
__m512i zmm4 = _mm512_castsi256_si512(ymm_y);
|
||||||
|
zmm4 = _mm512_permutexvar_epi64(zmm_split_y, zmm4);
|
||||||
|
zmm4 = _mm512_permutex_epi64(zmm4, 0xd8);
|
||||||
|
zmm4 = _mm512_unpacklo_epi8(zmm4, zmm4);
|
||||||
|
|
||||||
|
// YUVTORGB_AVX512BW
|
||||||
|
zmm3 = _mm512_sub_epi8(zmm3, zmm_128);
|
||||||
|
zmm4 = _mm512_mulhi_epu16(zmm4, zmm_kYToRgb);
|
||||||
|
|
||||||
|
__m512i zmm0 = _mm512_maddubs_epi16(zmm_kUVToB, zmm3);
|
||||||
|
__m512i zmm1 = _mm512_maddubs_epi16(zmm_kUVToG, zmm3);
|
||||||
|
__m512i zmm2 = _mm512_maddubs_epi16(zmm_kUVToR, zmm3);
|
||||||
|
|
||||||
|
zmm4 = _mm512_add_epi16(zmm4, zmm_kYBiasToRgb);
|
||||||
|
|
||||||
|
zmm0 = _mm512_adds_epi16(zmm0, zmm4);
|
||||||
|
zmm1 = _mm512_subs_epi16(zmm4, zmm1);
|
||||||
|
zmm2 = _mm512_adds_epi16(zmm2, zmm4);
|
||||||
|
|
||||||
|
// Clamp to [0, 0x3ff0] (1023 * 16)
|
||||||
|
zmm0 = _mm512_min_epi16(zmm0, zmm_3ff0);
|
||||||
|
zmm1 = _mm512_min_epi16(zmm1, zmm_3ff0);
|
||||||
|
zmm2 = _mm512_min_epi16(zmm2, zmm_3ff0);
|
||||||
|
zmm0 = _mm512_max_epi16(zmm0, zmm_zero);
|
||||||
|
zmm1 = _mm512_max_epi16(zmm1, zmm_zero);
|
||||||
|
zmm2 = _mm512_max_epi16(zmm2, zmm_zero);
|
||||||
|
|
||||||
|
// STOREAR30_AVX512BW:
|
||||||
|
zmm0 = _mm512_srli_epi16(zmm0, 4);
|
||||||
|
zmm3 = _mm512_srli_epi16(zmm1, 10);
|
||||||
|
zmm1 = _mm512_slli_epi16(zmm1, 6);
|
||||||
|
zmm0 = _mm512_ternarylogic_epi32(zmm0, zmm1, zmm_fc00, 0xf8);
|
||||||
|
zmm2 = _mm512_ternarylogic_epi32(zmm2, zmm3, zmm_3ff0, 0xe4);
|
||||||
|
zmm2 = _mm512_or_si512(zmm2, zmm_c000);
|
||||||
|
|
||||||
|
__m512i zmm_dst0 = _mm512_permutex2var_epi16(zmm0, zmm_perm0, zmm2);
|
||||||
|
__m512i zmm_dst1 = _mm512_permutex2var_epi16(zmm0, zmm_perm1, zmm2);
|
||||||
|
|
||||||
|
_mm512_storeu_si512((__m512i*)dst_ar30, zmm_dst0);
|
||||||
|
_mm512_storeu_si512((__m512i*)(dst_ar30 + 64), zmm_dst1);
|
||||||
|
|
||||||
|
dst_ar30 += 128;
|
||||||
|
width -= 32;
|
||||||
|
}
|
||||||
|
_mm256_zeroupper();
|
||||||
|
}
|
||||||
|
#endif
|
||||||
|
|
||||||
#ifdef __cplusplus
|
#ifdef __cplusplus
|
||||||
} // extern "C"
|
} // extern "C"
|
||||||
} // namespace libyuv
|
} // namespace libyuv
|
||||||
|
|||||||
Loading…
x
Reference in New Issue
Block a user