rm GREATER_EQUAL

cmake/cudnn.cmake

@@ -35,7 +35,7 @@ list(APPEND CUDNN_CHECK_LIBRARY_DIRS
 	${CUDA_TOOLKIT_ROOT_DIR}/lib/x64
 	)
 
-if (${CUDA_VERSION} GREATER_EQUAL 10.0)
+if((${CUDA_VERSION} GREATER 10.0) OR (${CUDA_VERSION} EQUAL 10.0))
     find_library(CUBLAS_LIBRARY  NAMES libcublas.so PATHS ${CUDNN_CHECK_LIBRARY_DIRS} NO_DEFAULT_PATH)
     set(CUBLAS_LIBRARIES ${CUBLAS_LIBRARY})
 else()

lite/backends/cuda/math/scale.cu

@@ -16,24 +16,12 @@
 #include "lite/backends/cuda/cuda_utils.h"
 #include "lite/backends/cuda/math/scale.h"
 #include "lite/backends/cuda/math/utils.h"
+
 namespace paddle {
 namespace lite {
 namespace cuda {
 namespace math {
-/*
-template <typename T>
-__global__ void scale_kernel(int num, const T* in, T* out, const float scale,
-const float bias) {
-  int tid = blockIdx.x * blockDim.x + threadIdx.x;
-  if (tid < num) {
-#if __CUDA_ARCH__ >= 350
-    out[tid] = __ldg(in + tid) * scale + bias;
-#else
-    out[tid] = in[tid] * scale;
-#endif
-  }
-}
-*/
+
 #define CUDA_KERNEL_LOOP(i, n)                                 \
   for (int i = blockIdx.x * blockDim.x + threadIdx.x; i < (n); \
        i += blockDim.x * gridDim.x)
@@ -61,9 +49,6 @@ template <typename T>
 __global__ void scale_kernel(
     int count, const T* in_data, T* out_data, const T scale, const T bias) {
   int tid = blockIdx.x * blockDim.x + threadIdx.x;
-  // if (tid < count){
-  // out_data[tid] = scale * in_data[tid] + bias;
-  //}
   CUDA_KERNEL_LOOP(tid, count) { out_data[tid] = scale * in_data[tid] + bias; }
 }