From 01e994d74e4e3937ee1a3efdc048320a1e51f818 Mon Sep 17 00:00:00 2001
From: Frank Barchard <fbarchard@chromium.org>
Date: Tue, 7 Nov 2017 18:17:33 -0800
Subject: [PATCH] Enable SSE2 code without -msse

Bug: libyuv:754
Test: CC=clang CXX=clang++ CFLAGS="-m32" CXXFLAGS="-m32 -mno-sse -O2" make -f linux.mk
Change-Id: I74bf8d032013694e65ea7637bc38d3253db53ff2
Reviewed-on: https://chromium-review.googlesource.com/758043
Reviewed-by: Frank Barchard <fbarchard@google.com>
---
 include/libyuv/compare_row.h      | 2 +-
 include/libyuv/planar_functions.h | 2 +-
 include/libyuv/rotate_row.h       | 2 +-
 include/libyuv/row.h              | 2 +-
 include/libyuv/scale_row.h        | 2 +-
 linux.mk                          | 2 +-
 source/row_gcc.cc                 | 6 +++---
 7 files changed, 9 insertions(+), 9 deletions(-)

diff --git a/include/libyuv/compare_row.h b/include/libyuv/compare_row.h
index 9b7013a24..a34f9ad4a 100644
--- a/include/libyuv/compare_row.h
+++ b/include/libyuv/compare_row.h
@@ -19,7 +19,7 @@ extern "C" {
 #endif
 
 #if defined(__pnacl__) || defined(__CLR_VER) || \
-    (defined(__i386__) && !defined(__SSE2__))
+    (defined(__i386__) && !defined(__SSE__) && !defined(__clang__))
 #define LIBYUV_DISABLE_X86
 #endif
 // MemorySanitizer does not support assembly code yet. http://crbug.com/344505
diff --git a/include/libyuv/planar_functions.h b/include/libyuv/planar_functions.h
index d97965cb8..c91501a9c 100644
--- a/include/libyuv/planar_functions.h
+++ b/include/libyuv/planar_functions.h
@@ -746,7 +746,7 @@ int I420Interpolate(const uint8* src0_y,
                     int interpolation);
 
 #if defined(__pnacl__) || defined(__CLR_VER) || \
-    (defined(__i386__) && !defined(__SSE2__))
+    (defined(__i386__) && !defined(__SSE__) && !defined(__clang__))
 #define LIBYUV_DISABLE_X86
 #endif
 // MemorySanitizer does not support assembly code yet. http://crbug.com/344505
diff --git a/include/libyuv/rotate_row.h b/include/libyuv/rotate_row.h
index 60ac55ef9..973fc1528 100644
--- a/include/libyuv/rotate_row.h
+++ b/include/libyuv/rotate_row.h
@@ -19,7 +19,7 @@ extern "C" {
 #endif
 
 #if defined(__pnacl__) || defined(__CLR_VER) || \
-    (defined(__i386__) && !defined(__SSE2__))
+    (defined(__i386__) && !defined(__SSE__) && !defined(__clang__))
 #define LIBYUV_DISABLE_X86
 #endif
 // MemorySanitizer does not support assembly code yet. http://crbug.com/344505
diff --git a/include/libyuv/row.h b/include/libyuv/row.h
index 5ccf94d9d..6f0cd500e 100644
--- a/include/libyuv/row.h
+++ b/include/libyuv/row.h
@@ -31,7 +31,7 @@ extern "C" {
   var = 0
 
 #if defined(__pnacl__) || defined(__CLR_VER) || \
-    (defined(__i386__) && !defined(__SSE2__))
+    (defined(__i386__) && !defined(__SSE__) && !defined(__clang__))
 #define LIBYUV_DISABLE_X86
 #endif
 // MemorySanitizer does not support assembly code yet. http://crbug.com/344505
diff --git a/include/libyuv/scale_row.h b/include/libyuv/scale_row.h
index ebafac4f6..b97f3e789 100644
--- a/include/libyuv/scale_row.h
+++ b/include/libyuv/scale_row.h
@@ -20,7 +20,7 @@ extern "C" {
 #endif
 
 #if defined(__pnacl__) || defined(__CLR_VER) || \
-    (defined(__i386__) && !defined(__SSE2__))
+    (defined(__i386__) && !defined(__SSE__) && !defined(__clang__))
 #define LIBYUV_DISABLE_X86
 #endif
 // MemorySanitizer does not support assembly code yet. http://crbug.com/344505
diff --git a/linux.mk b/linux.mk
index 1dd527c75..7e9aa5e4e 100644
--- a/linux.mk
+++ b/linux.mk
@@ -80,4 +80,4 @@ cpuid: util/cpuid.c libyuv.a
 	$(CC) $(CFLAGS) -o $@ util/cpuid.c libyuv.a
 
 clean:
-	/bin/rm -f source/*.o *.ii *.s libyuv.a convert cpuid psnr
+	/bin/rm -f source/*.o *.ii *.s libyuv.a yuvconvert cpuid psnr
diff --git a/source/row_gcc.cc b/source/row_gcc.cc
index 3af320454..93399f0b9 100644
--- a/source/row_gcc.cc
+++ b/source/row_gcc.cc
@@ -5713,7 +5713,7 @@ void HalfFloatRow_SSE2(const uint16* src, uint16* dst, float scale, int width) {
   : "+r"(src),    // %0
     "+r"(dst),    // %1
     "+r"(width)   // %2
-  : "x"(scale * kScaleBias)   // %3
+  : "mx"(scale * kScaleBias)   // %3
   : "memory", "cc",
     "xmm2", "xmm3", "xmm4", "xmm5"
   );
@@ -5749,7 +5749,7 @@ void HalfFloatRow_AVX2(const uint16* src, uint16* dst, float scale, int width) {
   : "+r"(src),    // %0
     "+r"(dst),    // %1
     "+r"(width)   // %2
-  : "x"(scale * kScaleBias)   // %3
+  : "mx"(scale * kScaleBias)   // %3
   : "memory", "cc",
     "xmm2", "xmm3", "xmm4", "xmm5"
   );
@@ -5782,7 +5782,7 @@ void HalfFloatRow_F16C(const uint16* src, uint16* dst, float scale, int width) {
   : "+r"(src),   // %0
     "+r"(dst),   // %1
     "+r"(width)  // %2
-  : "x"(scale)   // %3
+  : "mx"(scale)  // %3
   : "memory", "cc",
     "xmm2", "xmm3", "xmm4"
   );