I411ToARGB AVX2 version
BUG=403
TESTED=I411ToARGB unittest
R=harryjin@google.com
Review URL: https://webrtc-codereview.appspot.com/42689004
git-svn-id: http://libyuv.googlecode.com/svn/trunk@1321 16f28f9a-4ce2-e073-06de-1de4eb20be90
diff --git a/README.chromium b/README.chromium
index e5bb1d0..9b1a417 100644
--- a/README.chromium
+++ b/README.chromium
@@ -1,6 +1,6 @@
Name: libyuv
URL: http://code.google.com/p/libyuv/
-Version: 1320
+Version: 1321
License: BSD
License File: LICENSE
diff --git a/include/libyuv/row.h b/include/libyuv/row.h
index cbd0200..be8cf61 100644
--- a/include/libyuv/row.h
+++ b/include/libyuv/row.h
@@ -199,6 +199,7 @@
#define HAS_I422TOARGB1555ROW_AVX2
#define HAS_I422TOARGB4444ROW_AVX2
#define HAS_I444TOARGBROW_AVX2
+#define HAS_I411TOARGBROW_AVX2
// TODO(fbarchard): Port to Neon
#define HAS_ARGBTORGB565DITHERROW_SSE2
#define HAS_ARGBTORGB565DITHERROW_AVX2
@@ -1061,6 +1062,11 @@
const uint8* src_v,
uint8* dst_argb,
int width);
+void I411ToARGBRow_AVX2(const uint8* src_y,
+ const uint8* src_u,
+ const uint8* src_v,
+ uint8* dst_argb,
+ int width);
void NV12ToARGBRow_SSSE3(const uint8* src_y,
const uint8* src_uv,
uint8* dst_argb,
@@ -1205,6 +1211,11 @@
const uint8* src_v,
uint8* dst_argb,
int width);
+void I411ToARGBRow_Any_AVX2(const uint8* src_y,
+ const uint8* src_u,
+ const uint8* src_v,
+ uint8* dst_argb,
+ int width);
void NV12ToARGBRow_Any_SSSE3(const uint8* src_y,
const uint8* src_uv,
uint8* dst_argb,
diff --git a/include/libyuv/version.h b/include/libyuv/version.h
index 44dfe37..3557bc6 100644
--- a/include/libyuv/version.h
+++ b/include/libyuv/version.h
@@ -11,6 +11,6 @@
#ifndef INCLUDE_LIBYUV_VERSION_H_ // NOLINT
#define INCLUDE_LIBYUV_VERSION_H_
-#define LIBYUV_VERSION 1320
+#define LIBYUV_VERSION 1321
#endif // INCLUDE_LIBYUV_VERSION_H_ NOLINT
diff --git a/source/convert_argb.cc b/source/convert_argb.cc
index beb6543..5ce2fd5 100644
--- a/source/convert_argb.cc
+++ b/source/convert_argb.cc
@@ -230,6 +230,14 @@
}
}
#endif
+#if defined(HAS_I411TOARGBROW_AVX2)
+ if (TestCpuFlag(kCpuHasAVX2)) {
+ I411ToARGBRow = I411ToARGBRow_Any_AVX2;
+ if (IS_ALIGNED(width, 16)) {
+ I411ToARGBRow = I411ToARGBRow_AVX2;
+ }
+ }
+#endif
#if defined(HAS_I411TOARGBROW_NEON)
if (TestCpuFlag(kCpuHasNEON)) {
I411ToARGBRow = I411ToARGBRow_Any_NEON;
diff --git a/source/row_any.cc b/source/row_any.cc
index 84d8d06..216c995 100644
--- a/source/row_any.cc
+++ b/source/row_any.cc
@@ -76,6 +76,9 @@
#ifdef HAS_I444TOARGBROW_AVX2
YANY(I444ToARGBRow_Any_AVX2, I444ToARGBRow_AVX2, I444ToARGBRow_C, 0, 4, 15)
#endif
+#ifdef HAS_I411TOARGBROW_AVX2
+YANY(I411ToARGBRow_Any_AVX2, I411ToARGBRow_AVX2, I411ToARGBRow_C, 2, 4, 15)
+#endif
#ifdef HAS_I422TOARGB4444ROW_AVX2
YANY(I422ToARGB4444Row_Any_AVX2, I422ToARGB4444Row_AVX2, I422ToARGB4444Row_C,
1, 2, 7)
diff --git a/source/row_win.cc b/source/row_win.cc
index 93ee582..a98728d 100644
--- a/source/row_win.cc
+++ b/source/row_win.cc
@@ -1698,6 +1698,17 @@
__asm vpunpcklwd ymm0, ymm0, ymm0 /* UVUV (upsample) */ \
}
+// Read 4 UV from 411, upsample to 16 UV.
+#define READYUV411_AVX2 __asm { \
+ __asm vmovd xmm0, dword ptr [esi] /* U */ /* NOLINT */ \
+ __asm vmovd xmm1, dword ptr [esi + edi] /* V */ /* NOLINT */ \
+ __asm lea esi, [esi + 4] \
+ __asm vpunpcklbw ymm0, ymm0, ymm1 /* UV */ \
+ __asm vpunpcklwd ymm0, ymm0, ymm0 /* UVUV (upsample) */ \
+ __asm vpermq ymm0, ymm0, 0xd8 \
+ __asm vpunpckldq ymm0, ymm0, ymm0 /* UVUVUVUV (upsample) */ \
+ }
+
// Read 8 UV from NV12, upsample to 16 UV.
#define READNV12_AVX2 __asm { \
__asm vmovdqu xmm0, [esi] /* UV */ \
@@ -1821,7 +1832,41 @@
}
#endif // HAS_I444TOARGBROW_AVX2
+#ifdef HAS_I411TOARGBROW_AVX2
+// 16 pixels
+// 4 UV values upsampled to 16 UV, mixed with 16 Y producing 16 ARGB (64 bytes).
+__declspec(naked) __declspec(align(16))
+void I411ToARGBRow_AVX2(const uint8* y_buf,
+ const uint8* u_buf,
+ const uint8* v_buf,
+ uint8* dst_argb,
+ int width) {
+ __asm {
+ push esi
+ push edi
+ mov eax, [esp + 8 + 4] // Y
+ mov esi, [esp + 8 + 8] // U
+ mov edi, [esp + 8 + 12] // V
+ mov edx, [esp + 8 + 16] // argb
+ mov ecx, [esp + 8 + 20] // width
+ sub edi, esi
+ vpcmpeqb ymm5, ymm5, ymm5 // generate 0xffffffffffffffff for alpha
+ convertloop:
+ READYUV411_AVX2
+ YUVTORGB_AVX2(kYuvConstants)
+ STOREARGB_AVX2
+
+ sub ecx, 16
+ jg convertloop
+
+ pop edi
+ pop esi
+ vzeroupper
+ ret
+ }
+}
+#endif // HAS_I411TOARGBROW_AVX2
#ifdef HAS_NV12TOARGBROW_AVX2
// 16 pixels.
@@ -2050,7 +2095,7 @@
__asm lea esi, [esi + 2] \
__asm punpcklbw xmm0, xmm1 /* UV */ \
__asm punpcklwd xmm0, xmm0 /* UVUV (upsample) */ \
- __asm punpckldq xmm0, xmm0 /* UVUV (upsample) */ \
+ __asm punpckldq xmm0, xmm0 /* UVUVUVUV (upsample) */ \
}
// Read 4 UV from NV12, upsample to 8 UV.
