coffee/libyuv
1
0
Fork 0
mirror of https://chromium.googlesource.com/libyuv/libyuv synced 2025-12-06 08:46:47 +08:00
Code Issues Packages Projects Releases Wiki Activity

Add supports for loongarch LSX and LASX.

Browse Source 
1. Add supports for LSX and LASX.
2. Three optimization functions are added in loongarch/row_lasx.cc file:
   I422ToARGBRow_LASX,I422ToRGBARow_LASX,I422AlphaToARGBRow_LASX.

Bug: libyuv:912, Bug: libyuv:913
Change-Id: I043c2704f99a5215724b5c0b7f97e6bf5f7a199b
Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/3329189
Reviewed-by: Frank Barchard <fbarchard@chromium.org>
Commit-Queue: Frank Barchard <fbarchard@chromium.org>
This commit is contained in:
Hao Chen 2020-11-06 14:35:20 +08:00 committed by libyuv LUCI CQ
parent 90ffd5cba9
commit 51de1e16f2
12 changed files with 2351 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
docs/environment_variables.md
View File

@ -37,6 +37,10 @@ By default the cpu is detected and the most advanced form of SIMD is used. But
LIBYUV_DISABLE_MSA
LIBYUV_DISABLE_MMI
## LOONGARCH CPUs
LIBYUV_DISABLE_LSX
LIBYUV_DISABLE_LASX
# Test Width/Height/Repeat
The unittests default to a small image (128x72) to run fast. This can be set by environment variable to test a specific resolutions.

7
include/libyuv/cpu_id.h
View File

@ -51,6 +51,11 @@ static const int kCpuHasMIPS = 0x400000;
static const int kCpuHasMSA = 0x800000;
static const int kCpuHasMMI = 0x1000000;
// These flags are only valid on LOONGARCH processors.
static const int kCpuHasLOONGARCH = 0x2000000;
static const int kCpuHasLSX = 0x4000000;
static const int kCpuHasLASX = 0x8000000;
// Optional init function. TestCpuFlag does an auto-init.
// Returns cpu_info flags.
LIBYUV_API
@ -74,6 +79,8 @@ LIBYUV_API
int ArmCpuCaps(const char* cpuinfo_name);
LIBYUV_API
int MipsCpuCaps(const char* cpuinfo_name);
LIBYUV_API
int LoongarchCpuCaps(void);
// For testing, allow CPU flags to be disabled.
// ie MaskCpuFlags(~kCpuHasSSSE3) to disable SSSE3.

1862
include/libyuv/loongson_intrinsics.h Normal file
View File

File diff suppressed because it is too large Load Diff

45
include/libyuv/row.h
View File

@ -684,6 +684,13 @@ extern "C" {
#define HAS_YUY2TOYROW_MMI
#endif
#if !defined(LIBYUV_DISABLE_LASX) && defined(__loongarch_asx)
#define HAS_I422TOARGBROW_LASX
#define HAS_I422TORGBAROW_LASX
#define HAS_I422ALPHATOARGBROW_LASX
#endif
#if defined(_MSC_VER) && !defined(__CLR_VER) && !defined(__clang__)
#if defined(VISUALC_HAS_AVX2)
#define SIMD_ALIGNED(var) __declspec(align(32)) var
@ -954,12 +961,24 @@ void I422ToARGBRow_MSA(const uint8_t* src_y,
uint8_t* dst_argb,
const struct YuvConstants* yuvconstants,
int width);
void I422ToARGBRow_LASX(const uint8_t* src_y,
const uint8_t* src_u,
const uint8_t* src_v,
uint8_t* dst_argb,
const struct YuvConstants* yuvconstants,
int width);
void I422ToRGBARow_MSA(const uint8_t* src_y,
const uint8_t* src_u,
const uint8_t* src_v,
uint8_t* dst_argb,
const struct YuvConstants* yuvconstants,
int width);
void I422ToRGBARow_LASX(const uint8_t* src_y,
const uint8_t* src_u,
const uint8_t* src_v,
uint8_t* dst_argb,
const struct YuvConstants* yuvconstants,
int width);
void I422ToARGBRow_MMI(const uint8_t* src_y,
const uint8_t* src_u,
const uint8_t* src_v,
@ -973,6 +992,13 @@ void I422AlphaToARGBRow_MSA(const uint8_t* src_y,
uint8_t* dst_argb,
const struct YuvConstants* yuvconstants,
int width);
void I422AlphaToARGBRow_LASX(const uint8_t* src_y,
const uint8_t* src_u,
const uint8_t* src_v,
const uint8_t* src_a,
uint8_t* dst_argb,
const struct YuvConstants* yuvconstants,
int width);
void I422ToRGB24Row_MSA(const uint8_t* src_y,
const uint8_t* src_u,
const uint8_t* src_v,
@ -4221,6 +4247,12 @@ void I422ToARGBRow_Any_MSA(const uint8_t* y_buf,
uint8_t* dst_ptr,
const struct YuvConstants* yuvconstants,
int width);
void I422ToARGBRow_Any_LASX(const uint8_t* y_buf,
const uint8_t* u_buf,
const uint8_t* v_buf,
uint8_t* dst_ptr,
const struct YuvConstants* yuvconstants,
int width);
void I422ToARGBRow_Any_MMI(const uint8_t* y_buf,
const uint8_t* u_buf,
const uint8_t* v_buf,
@ -4233,6 +4265,12 @@ void I422ToRGBARow_Any_MSA(const uint8_t* y_buf,
uint8_t* dst_ptr,
const struct YuvConstants* yuvconstants,
int width);
void I422ToRGBARow_Any_LASX(const uint8_t* y_buf,
const uint8_t* u_buf,
const uint8_t* v_buf,
uint8_t* dst_ptr,
const struct YuvConstants* yuvconstants,
int width);
void I422AlphaToARGBRow_Any_MSA(const uint8_t* y_buf,
const uint8_t* u_buf,
const uint8_t* v_buf,
@ -4240,6 +4278,13 @@ void I422AlphaToARGBRow_Any_MSA(const uint8_t* y_buf,
uint8_t* dst_ptr,
const struct YuvConstants* yuvconstants,
int width);
void I422AlphaToARGBRow_Any_LASX(const uint8_t* y_buf,
const uint8_t* u_buf,
const uint8_t* v_buf,
const uint8_t* a_buf,
uint8_t* dst_ptr,
const struct YuvConstants* yuvconstants,
int width);
void I422ToRGB24Row_Any_MSA(const uint8_t* y_buf,
const uint8_t* u_buf,
const uint8_t* v_buf,

56
source/convert_argb.cc
View File

@ -121,6 +121,14 @@ int I420ToARGBMatrix(const uint8_t* src_y,
}
}
#endif
#if defined(HAS_I422TOARGBROW_LASX)
if (TestCpuFlag(kCpuHasLASX)) {
I422ToARGBRow = I422ToARGBRow_Any_LASX;
if (IS_ALIGNED(width, 32)) {
I422ToARGBRow = I422ToARGBRow_LASX;
}
}
#endif
for (y = 0; y < height; ++y) {
I422ToARGBRow(src_y, src_u, src_v, dst_argb, yuvconstants, width);
@ -360,6 +368,14 @@ int I422ToARGBMatrix(const uint8_t* src_y,
}
}
#endif
#if defined(HAS_I422TOARGBROW_LASX)
if (TestCpuFlag(kCpuHasLASX)) {
I422ToARGBRow = I422ToARGBRow_Any_LASX;
if (IS_ALIGNED(width, 32)) {
I422ToARGBRow = I422ToARGBRow_LASX;
}
}
#endif
for (y = 0; y < height; ++y) {
I422ToARGBRow(src_y, src_u, src_v, dst_argb, yuvconstants, width);
@ -1869,6 +1885,14 @@ int I420AlphaToARGBMatrix(const uint8_t* src_y,
}
}
#endif
#if defined(HAS_I422ALPHATOARGBROW_LASX)
if (TestCpuFlag(kCpuHasLASX)) {
I422AlphaToARGBRow = I422AlphaToARGBRow_Any_LASX;
if (IS_ALIGNED(width, 16)) {
I422AlphaToARGBRow = I422AlphaToARGBRow_LASX;
}
}
#endif
#if defined(HAS_ARGBATTENUATEROW_SSSE3)
if (TestCpuFlag(kCpuHasSSSE3)) {
ARGBAttenuateRow = ARGBAttenuateRow_Any_SSSE3;
@ -2000,6 +2024,14 @@ int I422AlphaToARGBMatrix(const uint8_t* src_y,
}
}
#endif
#if defined(HAS_I422ALPHATOARGBROW_LASX)
if (TestCpuFlag(kCpuHasLASX)) {
I422AlphaToARGBRow = I422AlphaToARGBRow_Any_LASX;
if (IS_ALIGNED(width, 16)) {
I422AlphaToARGBRow = I422AlphaToARGBRow_LASX;
}
}
#endif
#if defined(HAS_ARGBATTENUATEROW_SSSE3)
if (TestCpuFlag(kCpuHasSSSE3)) {
ARGBAttenuateRow = ARGBAttenuateRow_Any_SSSE3;
@ -4299,6 +4331,14 @@ int I422ToRGBAMatrix(const uint8_t* src_y,
}
}
#endif
#if defined(HAS_I422TORGBAROW_LASX)
if (TestCpuFlag(kCpuHasLASX)) {
I422ToRGBARow = I422ToRGBARow_Any_LASX;
if (IS_ALIGNED(width, 32)) {
I422ToRGBARow = I422ToRGBARow_LASX;
}
}
#endif
for (y = 0; y < height; ++y) {
I422ToRGBARow(src_y, src_u, src_v, dst_rgba, yuvconstants, width);
@ -4504,6 +4544,14 @@ int I420ToRGBAMatrix(const uint8_t* src_y,
}
}
#endif
#if defined(HAS_I422TORGBAROW_LASX)
if (TestCpuFlag(kCpuHasLASX)) {
I422ToRGBARow = I422ToRGBARow_Any_LASX;
if (IS_ALIGNED(width, 32)) {
I422ToRGBARow = I422ToRGBARow_LASX;
}
}
#endif
for (y = 0; y < height; ++y) {
I422ToRGBARow(src_y, src_u, src_v, dst_rgba, yuvconstants, width);
@ -5189,6 +5237,14 @@ int I420ToRGB565Dither(const uint8_t* src_y,
}
}
#endif
#if defined(HAS_I422TOARGBROW_LASX)
if (TestCpuFlag(kCpuHasLASX)) {
I422ToARGBRow = I422ToARGBRow_Any_LASX;
if (IS_ALIGNED(width, 32)) {
I422ToARGBRow = I422ToARGBRow_LASX;
}
}
#endif
#if defined(HAS_ARGBTORGB565DITHERROW_SSE2)
if (TestCpuFlag(kCpuHasSSE2)) {
ARGBToRGB565DitherRow = ARGBToRGB565DitherRow_Any_SSE2;

29
source/cpu_id.cc
View File

@ -197,6 +197,31 @@ LIBYUV_API SAFEBUFFERS int MipsCpuCaps(const char* cpuinfo_name) {
return flag;
}
// TODO(fbarchard): Consider read_loongarch_ir().
#define LOONGARCH_CFG2 0x2
#define LOONGARCH_CFG2_LSX (1 << 6)
#define LOONGARCH_CFG2_LASX (1 << 7)
#if defined(__loongarch__) && defined(__linux__)
LIBYUV_API SAFEBUFFERS int LoongarchCpuCaps(void) {
int flag = 0x0;
uint32_t cfg2 = 0;
__asm__ volatile(
"cpucfg %0, %1 \n\t"
: "+&r"(cfg2)
: "r"(LOONGARCH_CFG2)
);
if (cfg2 & LOONGARCH_CFG2_LSX)
flag |= kCpuHasLSX;
if (cfg2 & LOONGARCH_CFG2_LASX)
flag |= kCpuHasLASX;
return flag;
}
#endif
static SAFEBUFFERS int GetCpuFlags(void) {
int cpu_info = 0;
#if !defined(__pnacl__) && !defined(__CLR_VER) && \
@ -240,6 +265,10 @@ static SAFEBUFFERS int GetCpuFlags(void) {
cpu_info = MipsCpuCaps("/proc/cpuinfo");
cpu_info |= kCpuHasMIPS;
#endif
#if defined(__loongarch__) && defined(__linux__)
cpu_info = LoongarchCpuCaps();
cpu_info |= kCpuHasLOONGARCH;
#endif
#if defined(__arm__) || defined(__aarch64__)
// gcc -mfpu=neon defines __ARM_NEON__
// __ARM_NEON__ generates code that requires Neon. NaCL also requires Neon.

7
source/row_any.cc
View File

@ -119,6 +119,9 @@ ANY41C(I444AlphaToARGBRow_Any_MMI, I444AlphaToARGBRow_MMI, 0, 0, 4, 7)
#ifdef HAS_I422ALPHATOARGBROW_MMI
ANY41C(I422AlphaToARGBRow_Any_MMI, I422AlphaToARGBRow_MMI, 1, 0, 4, 7)
#endif
#ifdef HAS_I422ALPHATOARGBROW_LASX
ANY41C(I422AlphaToARGBRow_Any_LASX, I422AlphaToARGBRow_LASX, 1, 0, 4, 15)
#endif
#undef ANY41C
// Any 4 planes to 1 plane of 8 bit with yuvconstants
@ -419,6 +422,10 @@ ANY31C(I422ToARGB1555Row_Any_MMI, I422ToARGB1555Row_MMI, 1, 0, 2, 7)
ANY31C(I422ToRGB565Row_Any_MMI, I422ToRGB565Row_MMI, 1, 0, 2, 7)
ANY31C(I422ToRGBARow_Any_MMI, I422ToRGBARow_MMI, 1, 0, 4, 7)
#endif
#ifdef HAS_I422TOARGBROW_LASX
ANY31C(I422ToARGBRow_Any_LASX, I422ToARGBRow_LASX, 1, 0, 4, 31)
ANY31C(I422ToRGBARow_Any_LASX, I422ToRGBARow_LASX, 1, 0, 4, 31)
#endif
#undef ANY31C
// Any 3 planes of 16 bit to 1 with yuvconstants

303
source/row_lasx.cc Normal file
View File

@ -0,0 +1,303 @@
/*
* Copyright 2022 The LibYuv Project Authors. All rights reserved.
*
* Copyright (c) 2022 Loongson Technology Corporation Limited
*
* 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 "libyuv/row.h"
#if !defined(LIBYUV_DISABLE_LASX) && defined(__loongarch_asx)
#include "libyuv/loongson_intrinsics.h"
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
#define ALPHA_VAL (-1)
// Fill YUV -> RGB conversion constants into vectors
#define YUVTORGB_SETUP(yuvconst, ubvr, ugvg, yg, yb) \
{ \
__m256i ub, vr, ug, vg; \
\
ub = __lasx_xvreplgr2vr_h(yuvconst->kUVToB[0]); \
vr = __lasx_xvreplgr2vr_h(yuvconst->kUVToR[1]); \
ug = __lasx_xvreplgr2vr_h(yuvconst->kUVToG[0]); \
vg = __lasx_xvreplgr2vr_h(yuvconst->kUVToG[1]); \
yg = __lasx_xvreplgr2vr_h(yuvconst->kYToRgb[0]); \
yb = __lasx_xvreplgr2vr_w(yuvconst->kYBiasToRgb[0]); \
ubvr = __lasx_xvilvl_h(ub, vr); \
ugvg = __lasx_xvilvl_h(ug, vg); \
}
// Load 32 YUV422 pixel data
#define READYUV422_D(psrc_y, psrc_u, psrc_v, out_y, uv_l, uv_h) \
{ \
__m256i temp0, temp1; \
\
DUP2_ARG2(__lasx_xvld, psrc_y, 0, psrc_u, 0, out_y, temp0); \
temp1 = __lasx_xvld(psrc_v, 0); \
temp0 = __lasx_xvsub_b(temp0, const_0x80); \
temp1 = __lasx_xvsub_b(temp1, const_0x80); \
temp0 = __lasx_vext2xv_h_b(temp0); \
temp1 = __lasx_vext2xv_h_b(temp1); \
uv_l = __lasx_xvilvl_h(temp0, temp1); \
uv_h = __lasx_xvilvh_h(temp0, temp1); \
}
// Load 16 YUV422 pixel data
#define READYUV422(psrc_y, psrc_u, psrc_v, out_y, uv) \
{ \
__m256i temp0, temp1; \
\
out_y = __lasx_xvld(psrc_y, 0); \
temp0 = __lasx_xvldrepl_d(psrc_u, 0); \
temp1 = __lasx_xvldrepl_d(psrc_v, 0); \
uv = __lasx_xvilvl_b(temp0, temp1); \
uv = __lasx_xvsub_b(uv, const_0x80); \
uv = __lasx_vext2xv_h_b(uv); \
}
// Convert 16 pixels of YUV420 to RGB.
#define YUVTORGB_D(in_y, in_uvl, in_uvh, ubvr, ugvg, \
yg, yb, b_l, b_h, g_l, g_h, r_l, r_h) \
{ \
__m256i u_l, u_h, v_l, v_h; \
__m256i yl_ev, yl_od, yh_ev, yh_od; \
__m256i temp0, temp1, temp2, temp3; \
\
temp0 = __lasx_xvilvl_b(in_y, in_y); \
temp1 = __lasx_xvilvh_b(in_y, in_y); \
yl_ev = __lasx_xvmulwev_w_hu_h(temp0, yg); \
yl_od = __lasx_xvmulwod_w_hu_h(temp0, yg); \
yh_ev = __lasx_xvmulwev_w_hu_h(temp1, yg); \
yh_od = __lasx_xvmulwod_w_hu_h(temp1, yg); \
DUP4_ARG2(__lasx_xvsrai_w, yl_ev, 16, yl_od, 16, yh_ev, 16, yh_od, 16, \
yl_ev, yl_od, yh_ev, yh_od); \
yl_ev = __lasx_xvadd_w(yl_ev, yb); \
yl_od = __lasx_xvadd_w(yl_od, yb); \
yh_ev = __lasx_xvadd_w(yh_ev, yb); \
yh_od = __lasx_xvadd_w(yh_od, yb); \
v_l = __lasx_xvmulwev_w_h(in_uvl, ubvr); \
u_l = __lasx_xvmulwod_w_h(in_uvl, ubvr); \
v_h = __lasx_xvmulwev_w_h(in_uvh, ubvr); \
u_h = __lasx_xvmulwod_w_h(in_uvh, ubvr); \
temp0 = __lasx_xvadd_w(yl_ev, u_l); \
temp1 = __lasx_xvadd_w(yl_od, u_l); \
temp2 = __lasx_xvadd_w(yh_ev, u_h); \
temp3 = __lasx_xvadd_w(yh_od, u_h); \
DUP4_ARG2(__lasx_xvsrai_w, temp0, 6, temp1, 6, temp2, 6, temp3, 6, \
temp0, temp1, temp2, temp3); \
DUP4_ARG1(__lasx_xvclip255_w, temp0, temp1, temp2, temp3, \
temp0, temp1, temp2, temp3); \
b_l = __lasx_xvpackev_h(temp1, temp0); \
b_h = __lasx_xvpackev_h(temp3, temp2); \
temp0 = __lasx_xvadd_w(yl_ev, v_l); \
temp1 = __lasx_xvadd_w(yl_od, v_l); \
temp2 = __lasx_xvadd_w(yh_ev, v_h); \
temp3 = __lasx_xvadd_w(yh_od, v_h); \
DUP4_ARG2(__lasx_xvsrai_w, temp0, 6, temp1, 6, temp2, 6, temp3, 6, \
temp0, temp1, temp2, temp3); \
DUP4_ARG1(__lasx_xvclip255_w, temp0, temp1, temp2, temp3, \
temp0, temp1, temp2, temp3); \
r_l = __lasx_xvpackev_h(temp1, temp0); \
r_h = __lasx_xvpackev_h(temp3, temp2); \
DUP2_ARG2(__lasx_xvdp2_w_h, in_uvl, ugvg, in_uvh, ugvg, u_l, u_h); \
temp0 = __lasx_xvsub_w(yl_ev, u_l); \
temp1 = __lasx_xvsub_w(yl_od, u_l); \
temp2 = __lasx_xvsub_w(yh_ev, u_h); \
temp3 = __lasx_xvsub_w(yh_od, u_h); \
DUP4_ARG2(__lasx_xvsrai_w, temp0, 6, temp1, 6, temp2, 6, temp3, 6, \
temp0, temp1, temp2, temp3); \
DUP4_ARG1(__lasx_xvclip255_w, temp0, temp1, temp2, temp3, \
temp0, temp1, temp2, temp3); \
g_l = __lasx_xvpackev_h(temp1, temp0); \
g_h = __lasx_xvpackev_h(temp3, temp2); \
}
// Convert 8 pixels of YUV420 to RGB.
#define YUVTORGB(in_y, in_uv, ubvr, ugvg, \
yg, yb, out_b, out_g, out_r) \
{ \
__m256i u_l, v_l, yl_ev, yl_od; \
__m256i temp0, temp1; \
\
in_y = __lasx_xvpermi_d(in_y, 0xD8); \
temp0 = __lasx_xvilvl_b(in_y, in_y); \
yl_ev = __lasx_xvmulwev_w_hu_h(temp0, yg); \
yl_od = __lasx_xvmulwod_w_hu_h(temp0, yg); \
DUP2_ARG2(__lasx_xvsrai_w, yl_ev, 16, yl_od, 16, yl_ev, yl_od); \
yl_ev = __lasx_xvadd_w(yl_ev, yb); \
yl_od = __lasx_xvadd_w(yl_od, yb); \
v_l = __lasx_xvmulwev_w_h(in_uv, ubvr); \
u_l = __lasx_xvmulwod_w_h(in_uv, ubvr); \
temp0 = __lasx_xvadd_w(yl_ev, u_l); \
temp1 = __lasx_xvadd_w(yl_od, u_l); \
DUP2_ARG2(__lasx_xvsrai_w, temp0, 6, temp1, 6, temp0, temp1); \
DUP2_ARG1(__lasx_xvclip255_w, temp0, temp1, temp0, temp1); \
out_b = __lasx_xvpackev_h(temp1, temp0); \
temp0 = __lasx_xvadd_w(yl_ev, v_l); \
temp1 = __lasx_xvadd_w(yl_od, v_l); \
DUP2_ARG2(__lasx_xvsrai_w, temp0, 6, temp1, 6, temp0, temp1); \
DUP2_ARG1(__lasx_xvclip255_w, temp0, temp1, temp0, temp1); \
out_r = __lasx_xvpackev_h(temp1, temp0); \
u_l = __lasx_xvdp2_w_h(in_uv, ugvg); \
temp0 = __lasx_xvsub_w(yl_ev, u_l); \
temp1 = __lasx_xvsub_w(yl_od, u_l); \
DUP2_ARG2(__lasx_xvsrai_w, temp0, 6, temp1, 6, temp0, temp1); \
DUP2_ARG1(__lasx_xvclip255_w, temp0, temp1, temp0, temp1); \
out_g = __lasx_xvpackev_h(temp1, temp0); \
}
// Pack and Store 16 ARGB values.
#define STOREARGB_D(a_l, a_h, r_l, r_h, g_l, g_h, \
b_l, b_h, pdst_argb) \
{ \
__m256i temp0, temp1, temp2, temp3; \
\
temp0 = __lasx_xvpackev_b(g_l, b_l); \
temp1 = __lasx_xvpackev_b(a_l, r_l); \
temp2 = __lasx_xvpackev_b(g_h, b_h); \
temp3 = __lasx_xvpackev_b(a_h, r_h); \
r_l = __lasx_xvilvl_h(temp1, temp0); \
r_h = __lasx_xvilvh_h(temp1, temp0); \
g_l = __lasx_xvilvl_h(temp3, temp2); \
g_h = __lasx_xvilvh_h(temp3, temp2); \
temp0 = __lasx_xvpermi_q(r_h, r_l, 0x20); \
temp1 = __lasx_xvpermi_q(g_h, g_l, 0x20); \
temp2 = __lasx_xvpermi_q(r_h, r_l, 0x31); \
temp3 = __lasx_xvpermi_q(g_h, g_l, 0x31); \
__lasx_xvst(temp0, pdst_argb, 0); \
__lasx_xvst(temp1, pdst_argb, 32); \
__lasx_xvst(temp2, pdst_argb, 64); \
__lasx_xvst(temp3, pdst_argb, 96); \
pdst_argb += 128; \
}
// Pack and Store 8 ARGB values.
#define STOREARGB(in_a, in_r, in_g, in_b, pdst_argb) \
{ \
__m256i temp0, temp1; \
\
temp0 = __lasx_xvpackev_b(in_g, in_b); \
temp1 = __lasx_xvpackev_b(in_a, in_r); \
in_a = __lasx_xvilvl_h(temp1, temp0); \
in_r = __lasx_xvilvh_h(temp1, temp0); \
temp0 = __lasx_xvpermi_q(in_r, in_a, 0x20); \
temp1 = __lasx_xvpermi_q(in_r, in_a, 0x31); \
__lasx_xvst(temp0, pdst_argb, 0); \
__lasx_xvst(temp1, pdst_argb, 32); \
pdst_argb += 64; \
}
void I422ToARGBRow_LASX(const uint8_t* src_y,
const uint8_t* src_u,
const uint8_t* src_v,
uint8_t* dst_argb,
const struct YuvConstants* yuvconstants,
int width) {
int x;
int len = width / 32;
__m256i vec_yb, vec_yg;
__m256i vec_ubvr, vec_ugvg;
__m256i alpha = __lasx_xvldi(0xFF);
__m256i const_0x80 = __lasx_xvldi(0x80);
YUVTORGB_SETUP(yuvconstants, vec_ubvr, vec_ugvg, vec_yg, vec_yb);
for (x = 0; x < len; x++) {
__m256i y, uv_l, uv_h, b_l, b_h, g_l, g_h, r_l, r_h;
READYUV422_D(src_y, src_u, src_v, y, uv_l, uv_h);
YUVTORGB_D(y, uv_l, uv_h, vec_ubvr, vec_ugvg, vec_yg,
vec_yb, b_l, b_h, g_l, g_h, r_l, r_h);
STOREARGB_D(alpha, alpha, r_l, r_h, g_l, g_h, b_l, b_h, dst_argb);
src_y += 32;
src_u += 16;
src_v += 16;
}
}
void I422ToRGBARow_LASX(const uint8_t* src_y,
const uint8_t* src_u,
const uint8_t* src_v,
uint8_t* dst_argb,
const struct YuvConstants* yuvconstants,
int width) {
int x;
int len = width / 32;
__m256i vec_yb, vec_yg;
__m256i vec_ubvr, vec_ugvg;
__m256i alpha = __lasx_xvldi(0xFF);
__m256i const_0x80 = __lasx_xvldi(0x80);
YUVTORGB_SETUP(yuvconstants, vec_ubvr, vec_ugvg, vec_yg, vec_yb);
for (x = 0; x < len; x++) {
__m256i y, uv_l, uv_h, b_l, b_h, g_l, g_h, r_l, r_h;
READYUV422_D(src_y, src_u, src_v, y, uv_l, uv_h);
YUVTORGB_D(y, uv_l, uv_h, vec_ubvr, vec_ugvg, vec_yg,
vec_yb, b_l, b_h, g_l, g_h, r_l, r_h);
STOREARGB_D(r_l, r_h, g_l, g_h, b_l, b_h, alpha, alpha, dst_argb);
src_y += 32;
src_u += 16;
src_v += 16;
}
}
void I422AlphaToARGBRow_LASX(const uint8_t* src_y,
const uint8_t* src_u,
const uint8_t* src_v,
const uint8_t* src_a,
uint8_t* dst_argb,
const struct YuvConstants* yuvconstants,
int width) {
int x;
int len = width / 32;
int res = width & 31;
__m256i vec_yb, vec_yg;
__m256i vec_ubvr, vec_ugvg;
__m256i zero = __lasx_xvldi(0);
__m256i const_0x80 = __lasx_xvldi(0x80);
YUVTORGB_SETUP(yuvconstants, vec_ubvr, vec_ugvg, vec_yg, vec_yb);
for (x = 0; x < len; x++) {
__m256i y, uv_l, uv_h, b_l, b_h, g_l, g_h, r_l, r_h, a_l, a_h;
y = __lasx_xvld(src_a, 0);
a_l = __lasx_xvilvl_b(zero, y);
a_h = __lasx_xvilvh_b(zero, y);
READYUV422_D(src_y, src_u, src_v, y, uv_l, uv_h);
YUVTORGB_D(y, uv_l, uv_h, vec_ubvr, vec_ugvg, vec_yg,
vec_yb, b_l, b_h, g_l, g_h, r_l, r_h);
STOREARGB_D(a_l, a_h, r_l, r_h, g_l, g_h, b_l, b_h, dst_argb);
src_y += 32;
src_u += 16;
src_v += 16;
src_a += 32;
}
if (res) {
__m256i y, uv, r, g, b, a;
a = __lasx_xvld(src_a, 0);
a = __lasx_vext2xv_hu_bu(a);
READYUV422(src_y, src_u, src_v, y, uv);
YUVTORGB(y, uv, vec_ubvr, vec_ugvg, vec_yg, vec_yb, b, g, r);
STOREARGB(a, r, g, b, dst_argb);
}
}
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif
#endif // !defined(LIBYUV_DISABLE_LASX) && defined(__loongarch_asx)

8
source/scale_argb.cc
View File

@ -643,6 +643,14 @@ static void ScaleYUVToARGBBilinearUp(int src_width,
}
}
#endif
#if defined(HAS_I422TOARGBROW_LASX)
if (TestCpuFlag(kCpuHasLASX)) {
I422ToARGBRow = I422ToARGBRow_Any_LASX;
if (IS_ALIGNED(src_width, 32)) {
I422ToARGBRow = I422ToARGBRow_LASX;
}
}
#endif
void (*InterpolateRow)(uint8_t * dst_argb, const uint8_t* src_argb,
ptrdiff_t src_stride, int dst_width,

12
unit_test/cpu_test.cc
View File

@ -72,6 +72,15 @@ TEST_F(LibYUVBaseTest, TestCpuHas) {
int has_mmi = TestCpuFlag(kCpuHasMMI);
printf("Has MMI %d\n", has_mmi);
#endif
#if defined(__loongarch__)
int has_loongarch = TestCpuFlag(kCpuHasLOONGARCH);
printf("Has LOONGARCH %d\n", has_loongarch);
int has_lsx = TestCpuFlag(kCpuHasLSX);
printf("Has LSX %d\n", has_lsx);
int has_lasx = TestCpuFlag(kCpuHasLASX);
printf("Has LASX %d\n", has_lasx);
#endif
}
TEST_F(LibYUVBaseTest, TestCompilerMacros) {
@ -151,6 +160,9 @@ TEST_F(LibYUVBaseTest, TestCompilerMacros) {
#ifdef _MIPS_ARCH_LOONGSON3A
printf("_MIPS_ARCH_LOONGSON3A %d\n", _MIPS_ARCH_LOONGSON3A);
#endif
#ifdef __loongarch__
printf("__loongarch__ %d\n", __loongarch__);
#endif
#ifdef _WIN32
printf("_WIN32 %d\n", _WIN32);
#endif

10
unit_test/unit_test.cc
View File

@ -81,6 +81,16 @@ int TestCpuEnv(int cpu_info) {
cpu_info &= ~libyuv::kCpuHasMMI;
}
#endif
#if defined(__longarch__) && defined(__linux__)
if (TestEnv("LIBYUV_DISABLE_LSX")) {
cpu_info &= ~libyuv::kCpuHasLSX;
}
#endif
#if defined(__longarch__) && defined(__linux__)
if (TestEnv("LIBYUV_DISABLE_LASX")) {
cpu_info &= ~libyuv::kCpuHasLASX;
}
#endif
#if !defined(__pnacl__) && !defined(__CLR_VER) && \
(defined(__x86_64__) || defined(_M_X64) || defined(__i386__) || \
defined(_M_IX86))

8
util/cpuid.c
View File

@ -23,6 +23,7 @@ int main(int argc, const char* argv[]) {
int has_arm = TestCpuFlag(kCpuHasARM);
int has_mips = TestCpuFlag(kCpuHasMIPS);
int has_x86 = TestCpuFlag(kCpuHasX86);
int has_loongarch = TestCpuFlag(kCpuHasLOONGARCH);
(void)argc;
(void)argv;
@ -65,6 +66,7 @@ int main(int argc, const char* argv[]) {
printf("Has ARM %x\n", has_arm);
printf("Has MIPS %x\n", has_mips);
printf("Has X86 %x\n", has_x86);
printf("Has LOONGARCH %x\n", has_loongarch);
if (has_arm) {
int has_neon = TestCpuFlag(kCpuHasNEON);
printf("Has NEON %x\n", has_neon);
@ -75,6 +77,12 @@ int main(int argc, const char* argv[]) {
int has_mmi = TestCpuFlag(kCpuHasMMI);
printf("Has MMI %x\n", has_mmi);
}
if (has_loongarch) {
int has_lsx = TestCpuFlag(kCpuHasLSX);
printf("Has LSX %x\n", has_lsx);
int has_lasx = TestCpuFlag(kCpuHasLASX);
printf("Has LASX %x\n", has_lasx);
}
if (has_x86) {
int has_sse2 = TestCpuFlag(kCpuHasSSE2);
int has_ssse3 = TestCpuFlag(kCpuHasSSSE3);