Merge commit 'e67b525388a5ae11ed243e94bbc25b4934b03a66'

taey16 · Jan 13, 2017 · b8a5b1e · b8a5b1e
2 parents 69d8331 + e67b525
commit b8a5b1e
Show file tree

Hide file tree

Showing 5 changed files with 18 additions and 10 deletions.
diff --git a/torch/lib/TH/THGeneral.c b/torch/lib/TH/THGeneral.c
@@ -109,7 +109,7 @@ void _THArgCheck(const char *file, int line, int condition, int argNumber, const
       snprintf(msg + n, 2048 - n, " at %s:%d", file, line);
     }
 
-    if (threadArgErrorHandlerData)
+    if (threadArgErrorHandler)
       (*threadArgErrorHandler)(argNumber, msg, threadArgErrorHandlerData);
     else
       (*defaultArgErrorHandler)(argNumber, msg, defaultArgErrorHandlerData);

diff --git a/torch/lib/TH/THHalf.c b/torch/lib/TH/THHalf.c
@@ -77,15 +77,17 @@ float TH_half2float(THHalf h)
     }
 
     int temp = ((sign << 31) | (exponent << 23) | mantissa);
-
-    return *((float*)((void*)&temp));
+    float x;
+    memcpy(&x,&temp,sizeof(float));
+    return x;
 }
 
 THHalf TH_float2half(float f)
 {
     THHalf ret;
 
-    unsigned x = *((int*)(void*)(&f));
+    unsigned x;
+    memcpy(&x,&f,sizeof(f));
     unsigned u = (x & 0x7fffffff), remainder, shift, lsb, lsb_s1, lsb_m1;
     unsigned sign, exponent, mantissa;
 

diff --git a/torch/lib/TH/cmake/FindARM.cmake b/torch/lib/TH/cmake/FindARM.cmake
@@ -68,9 +68,9 @@ if(NOT NEON_FOUND)
       MESSAGE(STATUS "Could not find hardware support for NEON on this machine.")
 endif(NOT NEON_FOUND)
 if(NOT CORTEXA8_FOUND)
-      MESSAGE(STATUS "No OMAP3 processor on this on this machine.")
+      MESSAGE(STATUS "No OMAP3 processor on this machine.")
 endif(NOT CORTEXA8_FOUND)
 if(NOT CORTEXA9_FOUND)
-      MESSAGE(STATUS "No OMAP4 processor on this on this machine.")
+      MESSAGE(STATUS "No OMAP4 processor on this machine.")
 endif(NOT CORTEXA9_FOUND)
 mark_as_advanced(NEON_FOUND)
diff --git a/torch/lib/TH/generic/THTensorCopy.c b/torch/lib/TH/generic/THTensorCopy.c
@@ -4,7 +4,7 @@
 
 void THTensor_(copy)(THTensor *tensor, THTensor *src)
 {
-  TH_TENSOR_APPLY2(real, tensor, real, src, *tensor_data = (real)(*src_data);)
+  TH_TENSOR_APPLY2(real, tensor, real, src, *tensor_data = *src_data;)
 }
 
 #define IMPLEMENT_THTensor_COPY(TYPENAMESRC, TYPE_SRC) \
@@ -25,6 +25,12 @@ void THTensor_(copy##TYPENAMESRC)(THTensor *tensor, TH##TYPENAMESRC##Tensor *src
  TH_TENSOR_APPLY2(real, tensor, TYPE_SRC, src, *tensor_data = (real)TH_half2float(*src_data);) \
 }
 
+#define IMPLEMENT_THTensor_COPY_TO_FROM_HALF(TYPENAMESRC, TYPE_SRC) \
+void THTensor_(copy##TYPENAMESRC)(THTensor *tensor, TH##TYPENAMESRC##Tensor *src) \
+{ \
+ TH_TENSOR_APPLY2(real, tensor, TYPE_SRC, src, *tensor_data = *src_data;) \
+}
+
 #ifndef TH_REAL_IS_HALF
 IMPLEMENT_THTensor_COPY(Byte, unsigned char)
 IMPLEMENT_THTensor_COPY(Char, char)
@@ -36,7 +42,7 @@ IMPLEMENT_THTensor_COPY(Double, double)
 IMPLEMENT_THTensor_COPY_FROM_HALF(Half, THHalf)
 #else
 /* only allow pass-through for Half */
-IMPLEMENT_THTensor_COPY(Half, THHalf)
+IMPLEMENT_THTensor_COPY_TO_FROM_HALF(Half, THHalf)
 IMPLEMENT_THTensor_COPY_TO_HALF(Byte, unsigned char)
 IMPLEMENT_THTensor_COPY_TO_HALF(Char, char)
 IMPLEMENT_THTensor_COPY_TO_HALF(Short, short)

diff --git a/torch/lib/TH/generic/simd/simd.h b/torch/lib/TH/generic/simd/simd.h
@@ -53,7 +53,7 @@ enum SIMDExtensions
 };
 
 
-#if defined(__arm__)
+#if defined(__arm__) || defined(__aarch64__) // incl. armel, armhf, arm64
 
  #if defined(__NEON__)
 
@@ -80,7 +80,7 @@ static inline uint32_t detectHostSIMDExtensions()
   return SIMDExtension_VSX;
 }
 
- #else
+ #else //PPC64 without VSX
 
 static inline uint32_t detectHostSIMDExtensions()
 {