Support string instructions for NaCl

BUG=253
TEST=validator
R=fbarchard@google.com

Review URL: https://webrtc-codereview.appspot.com/1990004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@750 16f28f9a-4ce2-e073-06de-1de4eb20be90

README.chromium

 Name: libyuv
 URL: http://code.google.com/p/libyuv/
-Version: 749
+Version: 750
 License: BSD
 License File: LICENSE
 

include/libyuv/row.h

@@ -55,10 +55,13 @@ extern "C" {
 // Conversions:
 #define HAS_ARGBTOBAYERROW_SSSE3
 #define HAS_ARGBSHUFFLEROW_SSSE3
+#define HAS_COPYROW_X86
+#define HAS_COPYROW_ERMS
 #define HAS_FIXEDDIV_X86
 #define HAS_ARGBTOYROW_SSSE3
 #define HAS_ARGBTOYJROW_SSSE3
 #define HAS_I400TOARGBROW_SSE2
+#define HAS_SETROW_X86
 #endif
 
 // The following are available on all x86 platforms except NaCL x64:
@@ -83,8 +86,6 @@ extern "C" {
 #define HAS_BGRATOUVROW_SSSE3
 #define HAS_BGRATOYROW_SSSE3
 #define HAS_COPYROW_SSE2
-#define HAS_COPYROW_X86
-#define HAS_COPYROW_ERMS
 #define HAS_HALFROW_SSE2
 #define HAS_I411TOARGBROW_SSSE3
 #define HAS_I422TOABGRROW_SSSE3
@@ -113,7 +114,6 @@ extern "C" {
 #define HAS_RGB565TOARGBROW_SSE2
 #define HAS_RGBATOUVROW_SSSE3
 #define HAS_RGBATOYROW_SSSE3
-#define HAS_SETROW_X86
 #define HAS_SPLITUVROW_SSE2
 #define HAS_UYVYTOARGBROW_SSSE3
 #define HAS_UYVYTOUV422ROW_SSE2

include/libyuv/version.h

@@ -11,6 +11,6 @@
 #ifndef INCLUDE_LIBYUV_VERSION_H_  // NOLINT
 #define INCLUDE_LIBYUV_VERSION_H_
 
-#define LIBYUV_VERSION 749
+#define LIBYUV_VERSION 750
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_  NOLINT

source/row_posix.cc

@@ -26,12 +26,16 @@ extern "C" {
 #define MEMLEA(offset, base) #offset "(%q" #base ")"
 #define MEMLEA4(offset, base, index, scale) \
     #offset "(%q" #base ",%q" #index "," #scale ")"
+#define MEMMOVESTRING(s, d) "%%nacl:(%q" #s "),%%nacl:(%q" #d "), %%r15"
+#define MEMSTORESTRING(d) "%%nacl:(%q" #d "), %%r15"
 #else
 #define MEMACCESS(base) "(%" #base ")"
 #define MEMACCESS2(offset, base) #offset "(%" #base ")"
 #define MEMLEA(offset, base) #offset "(%" #base ")"
 #define MEMLEA4(offset, base, index, scale) \
     #offset "(%" #base ",%" #index "," #scale ")"
+#define MEMMOVESTRING(s, d)
+#define MEMSTORESTRING(d)
 #endif
 
 #if defined(HAS_ARGBTOYROW_SSSE3) || defined(HAS_ARGBGRAYROW_SSSE3)
@@ -3040,7 +3044,7 @@ void CopyRow_X86(const uint8* src, uint8* dst, int width) {
   size_t width_tmp = static_cast<size_t>(width);
   asm volatile (
     "shr       $0x2,%2                         \n"
-    "rep movsl                                 \n"
+    "rep movsl "MEMMOVESTRING(0,1)"            \n"
   : "+S"(src),  // %0
     "+D"(dst),  // %1
     "+c"(width_tmp) // %2
@@ -3055,7 +3059,7 @@ void CopyRow_X86(const uint8* src, uint8* dst, int width) {
 void CopyRow_ERMS(const uint8* src, uint8* dst, int width) {
   size_t width_tmp = static_cast<size_t>(width);
   asm volatile (
-    "rep movsb                                 \n"
+    "rep movsb "MEMMOVESTRING(0,1)"           \n"
   : "+S"(src),  // %0
     "+D"(dst),  // %1
     "+c"(width_tmp) // %2
@@ -3070,7 +3074,7 @@ void SetRow_X86(uint8* dst, uint32 v32, int width) {
   size_t width_tmp = static_cast<size_t>(width);
   asm volatile (
     "shr       $0x2,%1                         \n"
-    "rep stosl                                 \n"
+    "rep stosl "MEMSTORESTRING(0)"             \n"
     : "+D"(dst),       // %0
       "+c"(width_tmp)  // %1
     : "a"(v32)         // %2
@@ -3083,7 +3087,7 @@ void ARGBSetRows_X86(uint8* dst, uint32 v32, int width,
     size_t width_tmp = static_cast<size_t>(width);
     uint32* d = reinterpret_cast<uint32*>(dst);
     asm volatile (
-      "rep stosl                               \n"
+      "rep stosl "MEMSTORESTRING(0)"           \n"
       : "+D"(d),         // %0
         "+c"(width_tmp)  // %1
       : "a"(v32)         // %2