libyuv: update to c244a3e9

Fixes color issue when scaling without breaking mingw.

BUG=https://bugs.chromium.org/p/libyuv/issues/detail?id=605
BUG=https://bugs.chromium.org/p/webm/issues/detail?id=1252

Change-Id: I09437d93fd65964ad57113274d8c819f3eaf2e57

1 files changed, 40 insertions, 47 deletions

diff --git a/third_party/libyuv/source/compare_win.cc b/third_party/libyuv/source/compare_win.cc
index 19806f275..dc86fe25b 100644
--- a/third_party/libyuv/source/compare_win.cc
+++ b/third_party/libyuv/source/compare_win.cc
@@ -9,6 +9,8 @@
  */
 
 #include "libyuv/basic_types.h"
+
+#include "libyuv/compare_row.h"
 #include "libyuv/row.h"
 
 #ifdef __cplusplus
@@ -16,9 +18,8 @@ namespace libyuv {
 extern "C" {
 #endif
 
-// This module is for Visual C x86.
-#if !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86) && \
-    defined(_MSC_VER) && !defined(__clang__)
+// This module is for 32 bit Visual C x86 and clangcl
+#if !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86)
 
 __declspec(naked)
 uint32 SumSquareError_SSE2(const uint8* src_a, const uint8* src_b, int count) {
@@ -100,41 +101,32 @@ uint32 SumSquareError_AVX2(const uint8* src_a, const uint8* src_b, int count) {
 }
 #endif  // _MSC_VER >= 1700
 
-#define HAS_HASHDJB2_SSE41
-static uvec32 kHash16x33 = { 0x92d9e201, 0, 0, 0 };  // 33 ^ 16
-static uvec32 kHashMul0 = {
+uvec32 kHash16x33 = { 0x92d9e201, 0, 0, 0 };  // 33 ^ 16
+uvec32 kHashMul0 = {
   0x0c3525e1,  // 33 ^ 15
   0xa3476dc1,  // 33 ^ 14
   0x3b4039a1,  // 33 ^ 13
   0x4f5f0981,  // 33 ^ 12
 };
-static uvec32 kHashMul1 = {
+uvec32 kHashMul1 = {
   0x30f35d61,  // 33 ^ 11
   0x855cb541,  // 33 ^ 10
   0x040a9121,  // 33 ^ 9
   0x747c7101,  // 33 ^ 8
 };
-static uvec32 kHashMul2 = {
+uvec32 kHashMul2 = {
   0xec41d4e1,  // 33 ^ 7
   0x4cfa3cc1,  // 33 ^ 6
   0x025528a1,  // 33 ^ 5
   0x00121881,  // 33 ^ 4
 };
-static uvec32 kHashMul3 = {
+uvec32 kHashMul3 = {
   0x00008c61,  // 33 ^ 3
   0x00000441,  // 33 ^ 2
   0x00000021,  // 33 ^ 1
   0x00000001,  // 33 ^ 0
 };
 
-// 27: 66 0F 38 40 C6     pmulld      xmm0,xmm6
-// 44: 66 0F 38 40 DD     pmulld      xmm3,xmm5
-// 59: 66 0F 38 40 E5     pmulld      xmm4,xmm5
-// 72: 66 0F 38 40 D5     pmulld      xmm2,xmm5
-// 83: 66 0F 38 40 CD     pmulld      xmm1,xmm5
-#define pmulld(reg) _asm _emit 0x66 _asm _emit 0x0F _asm _emit 0x38 \
-    _asm _emit 0x40 _asm _emit reg
-
 __declspec(naked)
 uint32 HashDjb2_SSE41(const uint8* src, int count, uint32 seed) {
   __asm {
@@ -143,30 +135,30 @@ uint32 HashDjb2_SSE41(const uint8* src, int count, uint32 seed) {
     movd       xmm0, [esp + 12]  // seed
 
     pxor       xmm7, xmm7        // constant 0 for unpck
-    movdqa     xmm6, kHash16x33
+    movdqa     xmm6, xmmword ptr kHash16x33
 
   wloop:
     movdqu     xmm1, [eax]       // src[0-15]
     lea        eax, [eax + 16]
-    pmulld(0xc6)                 // pmulld      xmm0,xmm6  hash *= 33 ^ 16
-    movdqa     xmm5, kHashMul0
+    pmulld     xmm0, xmm6        // hash *= 33 ^ 16
+    movdqa     xmm5, xmmword ptr kHashMul0
     movdqa     xmm2, xmm1
     punpcklbw  xmm2, xmm7        // src[0-7]
     movdqa     xmm3, xmm2
     punpcklwd  xmm3, xmm7        // src[0-3]
-    pmulld(0xdd)                 // pmulld     xmm3, xmm5
-    movdqa     xmm5, kHashMul1
+    pmulld     xmm3, xmm5
+    movdqa     xmm5, xmmword ptr kHashMul1
     movdqa     xmm4, xmm2
     punpckhwd  xmm4, xmm7        // src[4-7]
-    pmulld(0xe5)                 // pmulld     xmm4, xmm5
-    movdqa     xmm5, kHashMul2
+    pmulld     xmm4, xmm5
+    movdqa     xmm5, xmmword ptr kHashMul2
     punpckhbw  xmm1, xmm7        // src[8-15]
     movdqa     xmm2, xmm1
     punpcklwd  xmm2, xmm7        // src[8-11]
-    pmulld(0xd5)                 // pmulld     xmm2, xmm5
-    movdqa     xmm5, kHashMul3
+    pmulld     xmm2, xmm5
+    movdqa     xmm5, xmmword ptr kHashMul3
     punpckhwd  xmm1, xmm7        // src[12-15]
-    pmulld(0xcd)                 // pmulld     xmm1, xmm5
+    pmulld     xmm1, xmm5
     paddd      xmm3, xmm4        // add 16 results
     paddd      xmm1, xmm2
     paddd      xmm1, xmm3
@@ -191,36 +183,37 @@ uint32 HashDjb2_AVX2(const uint8* src, int count, uint32 seed) {
   __asm {
     mov        eax, [esp + 4]    // src
     mov        ecx, [esp + 8]    // count
-    movd       xmm0, [esp + 12]  // seed
-    movdqa     xmm6, kHash16x33
+    vmovd      xmm0, [esp + 12]  // seed
 
   wloop:
-    vpmovzxbd  xmm3, dword ptr [eax]  // src[0-3]
-    pmulld     xmm0, xmm6  // hash *= 33 ^ 16
-    vpmovzxbd  xmm4, dword ptr [eax + 4]  // src[4-7]
-    pmulld     xmm3, kHashMul0
-    vpmovzxbd  xmm2, dword ptr [eax + 8]  // src[8-11]
-    pmulld     xmm4, kHashMul1
-    vpmovzxbd  xmm1, dword ptr [eax + 12]  // src[12-15]
-    pmulld     xmm2, kHashMul2
+    vpmovzxbd  xmm3, [eax]  // src[0-3]
+    vpmulld    xmm0, xmm0, xmmword ptr kHash16x33  // hash *= 33 ^ 16
+    vpmovzxbd  xmm4, [eax + 4]  // src[4-7]
+    vpmulld    xmm3, xmm3, xmmword ptr kHashMul0
+    vpmovzxbd  xmm2, [eax + 8]  // src[8-11]
+    vpmulld    xmm4, xmm4, xmmword ptr kHashMul1
+    vpmovzxbd  xmm1, [eax + 12]  // src[12-15]
+    vpmulld    xmm2, xmm2, xmmword ptr kHashMul2
     lea        eax, [eax + 16]
-    pmulld     xmm1, kHashMul3
-    paddd      xmm3, xmm4        // add 16 results
-    paddd      xmm1, xmm2
-    paddd      xmm1, xmm3
-    pshufd     xmm2, xmm1, 0x0e  // upper 2 dwords
-    paddd      xmm1, xmm2
-    pshufd     xmm2, xmm1, 0x01
-    paddd      xmm1, xmm2
-    paddd      xmm0, xmm1
+    vpmulld    xmm1, xmm1, xmmword ptr kHashMul3
+    vpaddd     xmm3, xmm3, xmm4        // add 16 results
+    vpaddd     xmm1, xmm1, xmm2
+    vpaddd     xmm1, xmm1, xmm3
+    vpshufd    xmm2, xmm1, 0x0e  // upper 2 dwords
+    vpaddd     xmm1, xmm1,xmm2
+    vpshufd    xmm2, xmm1, 0x01
+    vpaddd     xmm1, xmm1, xmm2
+    vpaddd     xmm0, xmm0, xmm1
     sub        ecx, 16
     jg         wloop
 
-    movd       eax, xmm0         // return hash
+    vmovd      eax, xmm0         // return hash
+    vzeroupper
     ret
   }
 }
 #endif  // _MSC_VER >= 1700
+
 #endif  // !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86)
 
 #ifdef __cplusplus