Fix cmake error of failing to find UINT64_MAX.

CMakeLists.txt

@@ -25,6 +25,7 @@ find_package(Git REQUIRED)
 find_package(Threads REQUIRED)
 
 include(system)
+include(simd)
 
 ################################ Configurations #######################################
 option(WITH_GPU         "Compile PaddlePaddle with NVIDIA GPU"          ${CUDA_FOUND})
@@ -64,7 +65,6 @@ include(external/openblas)  # download, build, install openblas
 include(external/swig)      # download, build, install swig
 include(external/warpctc)   # download, build, install warpctc
 
-include(simd)               # set simd flag
 include(package)            # set paddle packages
 include(cpplint)            # set paddle c++ style
 include(ccache)             # set ccache for compilation

cmake/configure.cmake

@@ -32,6 +32,16 @@ if(NOT WITH_PROFILER)
     add_definitions(-DPADDLE_DISABLE_PROFILER)
 endif(NOT WITH_PROFILER)
 
+if(NEON_FOUND)
+    set(SIMD_FLAG ${NEON_FLAG})
+else(NEON_FOUND)
+    if(WITH_AVX)
+        set(SIMD_FLAG ${AVX_FLAG})
+    else(WITH_AVX)
+        set(SIMD_FLAG ${SSE3_FLAG})
+    endif(WITH_AVX)
+endif(NEON_FOUND)
+
 if(NOT WITH_GPU)
     add_definitions(-DPADDLE_ONLY_CPU)
     add_definitions(-DHPPL_STUB_FUNC)
@@ -48,11 +58,7 @@ else()
         message(FATAL_ERROR "Paddle need cudnn to compile")
     endif()
 
-    if(WITH_AVX)
-        set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} "-Xcompiler ${AVX_FLAG}")
-    else(WITH_AVX)
-        set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} "-Xcompiler ${SSE3_FLAG}")
-    endif(WITH_AVX)
+    set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} "-Xcompiler ${SIMD_FLAG}")
 
     # Include cuda and cudnn
     include_directories(${CUDNN_INCLUDE_DIR})

cmake/simd.cmake

@@ -88,14 +88,5 @@ int main()
     return 0;
 }" NEON_FOUND)
 
-if(NEON_FOUND)
-    set(SIMD_FLAG ${NEON_FLAG})
-else(NEON_FOUND)
-    if(WITH_AVX)
-        set(SIMD_FLAG ${AVX_FLAG})
-    else(WITH_AVX)
-        set(SIMD_FLAG ${SSE3_FLAG})
-    endif(WITH_AVX)
-endif(NEON_FOUND)
-
+set(CMAKE_REQUIRED_FLAGS "")
 mark_as_advanced(MMX_FOUND SSE2_FOUND SSE3_FOUND AVX_FOUND AVX2_FOUND NEON_FOUND)

paddle/math/SIMDFunctions.h

@@ -128,7 +128,7 @@ void decayL1AvxImpl(
 
 template <>
 inline void addTo(float* a, const float* b, size_t len) {
-#ifdef __SSE__
+#ifdef __SSE3__
   internal::addToImpl(a, b, len);
 #else
   naive::addTo(a, b, len);
@@ -137,7 +137,7 @@ inline void addTo(float* a, const float* b, size_t len) {
 
 template <>
 inline void batchAddTo(float* a, const float* b[], int batch, size_t len) {
-#ifdef __SSE__
+#ifdef __SSE3__
   internal::batchAddToImpl(a, b, batch, len);
 #else
   naive::batchAddTo(a, b, batch, len);
@@ -146,7 +146,7 @@ inline void batchAddTo(float* a, const float* b[], int batch, size_t len) {
 
 template <>
 inline void colMax(float* result, const float* data, int dim, int numSamples) {
-#ifdef __SSE__
+#ifdef __SSE3__
   internal::colMaxImpl(result, data, dim, numSamples);
 #else
   naive::colMax(result, data, dim, numSamples);