diff options
Diffstat (limited to 'src/core/CL/cl_kernels')
-rw-r--r-- | src/core/CL/cl_kernels/direct_convolution1x1.cl | 2 | ||||
-rw-r--r-- | src/core/CL/cl_kernels/direct_convolution3x3.cl | 2 | ||||
-rw-r--r-- | src/core/CL/cl_kernels/direct_convolution5x5.cl | 2 |
3 files changed, 3 insertions, 3 deletions
diff --git a/src/core/CL/cl_kernels/direct_convolution1x1.cl b/src/core/CL/cl_kernels/direct_convolution1x1.cl index fb516ddd6e..7b73b85eac 100644 --- a/src/core/CL/cl_kernels/direct_convolution1x1.cl +++ b/src/core/CL/cl_kernels/direct_convolution1x1.cl @@ -191,7 +191,7 @@ __kernel void direct_convolution1x1( weights.ptr += z_index * weights_stride_w; - for(int d = 0; d < WEIGHTS_DEPTH; ++d) + for(volatile int d = 0; d < WEIGHTS_DEPTH; ++d) { DATA_TYPE weight = *(__global DATA_TYPE *)weights.ptr; VEC_DATA_TYPE(DATA_TYPE, 8) diff --git a/src/core/CL/cl_kernels/direct_convolution3x3.cl b/src/core/CL/cl_kernels/direct_convolution3x3.cl index d094eca115..1420d7c873 100644 --- a/src/core/CL/cl_kernels/direct_convolution3x3.cl +++ b/src/core/CL/cl_kernels/direct_convolution3x3.cl @@ -138,7 +138,7 @@ __kernel void direct_convolution3x3( const int kernel_index = get_global_id(2); weights_addr += kernel_index * weights_stride_w; - for(int d = 0; d < WEIGHTS_DEPTH; ++d) + for(volatile int d = 0; d < WEIGHTS_DEPTH; ++d) { CONVOLUTION1x3(pixels0, (__global DATA_TYPE *)(src_addr + 0 * src_stride_y), (__global DATA_TYPE *)(weights_addr + 0 * weights_stride_y)); CONVOLUTION1x3(pixels0, (__global DATA_TYPE *)(src_addr + 1 * src_stride_y), (__global DATA_TYPE *)(weights_addr + 1 * weights_stride_y)); diff --git a/src/core/CL/cl_kernels/direct_convolution5x5.cl b/src/core/CL/cl_kernels/direct_convolution5x5.cl index 496da97a09..6fdd019a14 100644 --- a/src/core/CL/cl_kernels/direct_convolution5x5.cl +++ b/src/core/CL/cl_kernels/direct_convolution5x5.cl @@ -127,7 +127,7 @@ __kernel void direct_convolution5x5( const int kernel_index = get_global_id(2); weights_addr += kernel_index * weights_stride_w; - for(int d = 0; d < WEIGHTS_DEPTH; ++d) + for(volatile int d = 0; d < WEIGHTS_DEPTH; ++d) { CONVOLUTION1x5(pixels0, (__global DATA_TYPE *)src_addr, (__global DATA_TYPE *)weights_addr); CONVOLUTION1x5(pixels0, (__global DATA_TYPE *)(src_addr + 1 * src_stride_y), (__global DATA_TYPE *)(weights_addr + 1 * weights_stride_y)); |