From 6e815e708d23483467aec6be06f9d9a5f25b62a6 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Martin=20Storsj=C3=B6?= <martin@martin.st>
Date: Sun, 29 Jun 2014 00:51:58 +0300
Subject: [PATCH] Use ENFORCE_STACK_ALIGN_1D instead of manually doing stack
 buffer alignment

---
 test/decoder/DecUT_IntraPrediction.cpp | 14 ++++----------
 test/encoder/EncUT_EncoderMbAux.cpp    | 13 +++++--------
 2 files changed, 9 insertions(+), 18 deletions(-)

diff --git a/test/decoder/DecUT_IntraPrediction.cpp b/test/decoder/DecUT_IntraPrediction.cpp
index 6af45dce..75c56340 100644
--- a/test/decoder/DecUT_IntraPrediction.cpp
+++ b/test/decoder/DecUT_IntraPrediction.cpp
@@ -382,9 +382,8 @@ GENERATE_4x4_UT (WelsI4x4LumaPredHD_c, WelsI4x4LumaPredHD_ref, 0, 0)
 TEST(DecoderIntraPredictionTest, pred) {\
 const int32_t kiStride = 32; \
 int iRunTimes = 1000; \
-uint8_t _pRefBuffer[18 * kiStride + 64]; \
-uint8_t _pPredBuffer[18 * kiStride + 64]; \
-uint8_t *pRefBuffer, *pPredBuffer; \
+ENFORCE_STACK_ALIGN_1D (uint8_t, pRefBuffer, 18 * kiStride, 16); \
+ENFORCE_STACK_ALIGN_1D (uint8_t, pPredBuffer, 18 * kiStride, 16); \
 if (ASM) { \
   int32_t iTmp = 1; \
   uint32_t uiCPUFlags = WelsCPUFeatureDetect(&iTmp); \
@@ -392,8 +391,6 @@ if (ASM) { \
     return; \
   } \
 } \
-pRefBuffer = (uint8_t*)((((intptr_t)(&_pRefBuffer[31])) >> 4) << 4); \
-pPredBuffer = (uint8_t*)((((intptr_t)(&_pPredBuffer[31])) >> 4) << 4); \
 srand((unsigned int)time(NULL)); \
 while(iRunTimes--) {\
 for (int i = 0; i < 17; i ++) {\
@@ -523,9 +520,8 @@ GENERATE_8x8_UT (WelsIChromaPredV_c, LumaI8x8PredV, 0, 0)
 TEST(DecoderIntraPredictionTest, pred) {\
 const int32_t kiStride = 32; \
 int32_t iRunTimes = 1000; \
-uint8_t _pRefBuffer[18 * kiStride + 64]; \
-uint8_t _pPredBuffer[18 * kiStride + 64]; \
-uint8_t *pRefBuffer, *pPredBuffer; \
+ENFORCE_STACK_ALIGN_1D (uint8_t, pRefBuffer, 18 * kiStride, 16); \
+ENFORCE_STACK_ALIGN_1D (uint8_t, pPredBuffer, 18 * kiStride, 16); \
 if (ASM) { \
   int32_t iTmp = 1; \
   uint32_t uiCPUFlags = WelsCPUFeatureDetect( &iTmp); \
@@ -533,8 +529,6 @@ if (ASM) { \
     return ; \
   } \
 }\
-pRefBuffer = (uint8_t*)((((intptr_t)(&_pRefBuffer[31])) >> 4) << 4); \
-pPredBuffer = (uint8_t*)((((intptr_t)(&_pPredBuffer[31])) >> 4) << 4); \
 srand((unsigned int)time(NULL)); \
 while(iRunTimes--) {\
 for (int i = 0; i < 17; i ++) {\
diff --git a/test/encoder/EncUT_EncoderMbAux.cpp b/test/encoder/EncUT_EncoderMbAux.cpp
index 3ae3e409..3a30558f 100644
--- a/test/encoder/EncUT_EncoderMbAux.cpp
+++ b/test/encoder/EncUT_EncoderMbAux.cpp
@@ -239,10 +239,9 @@ void copy (uint8_t* pDst, int32_t iDStride, uint8_t* pSrc, int32_t iSStride, int
 TEST(EncodeMbAuxTest, function) { \
   const int iSStride = 64;  \
   const int iDStride = 64;  \
-  uint8_t _ref_src[64*64], _ref_dst[64*64], _dst[64*64]; \
-  uint8_t *ref_src = (uint8_t*)((((uintptr_t)(_ref_src + 15)) >> 4) << 4); \
-  uint8_t *ref_dst = (uint8_t*)((((uintptr_t)(_ref_dst + 15)) >> 4) << 4); \
-  uint8_t *dst = (uint8_t*)((((uintptr_t)(_dst + 15)) >> 4) << 4); \
+  ENFORCE_STACK_ALIGN_1D (uint8_t, ref_src, iSStride*height, 16); \
+  ENFORCE_STACK_ALIGN_1D (uint8_t, ref_dst, iDStride*height, 16); \
+  ENFORCE_STACK_ALIGN_1D (uint8_t, dst, iDStride*height, 16); \
   srand((unsigned int)time(NULL)); \
   for(int i = 0; i < height; i++) \
     for(int j = 0; j < width; j++) \
@@ -265,8 +264,7 @@ GENERATE_UT_FOR_COPY (16, 16, WelsCopy16x16NotAligned_sse2);
 GENERATE_UT_FOR_COPY (16, 16, WelsCopy16x16_sse2);
 #endif
 TEST (EncodeMbAuxTest, WelsGetNoneZeroCount_c) {
-  int16_t _iLevel[32];
-  int16_t* pLevel = (int16_t*) (((((uintptr_t)_iLevel) + 15) >> 4) << 4);
+  ENFORCE_STACK_ALIGN_1D (int16_t, pLevel, 16, 16);
   srand ((unsigned int)time (NULL));
   int32_t result = 0;
   for (int i = 0; i < 16; i++) {
@@ -278,8 +276,7 @@ TEST (EncodeMbAuxTest, WelsGetNoneZeroCount_c) {
 }
 #ifdef X86_ASM
 TEST (EncodeMbAuxTest, WelsGetNoneZeroCount_sse2) {
-  int16_t _iLevel[32];
-  int16_t* pLevel = (int16_t*) (((((uintptr_t)_iLevel) + 15) >> 4) << 4);
+  ENFORCE_STACK_ALIGN_1D (int16_t, pLevel, 16, 16);
   srand ((unsigned int)time (NULL));
   int32_t result = 0;
   for (int i = 0; i < 16; i++) {