diff options
Diffstat (limited to 'src/core')
| -rw-r--r-- | src/core/AccessWindowTranspose.cpp | 4 | ||||
| -rw-r--r-- | src/core/CL/cl_kernels/gemm.cl | 6 |
2 files changed, 5 insertions, 5 deletions
diff --git a/src/core/AccessWindowTranspose.cpp b/src/core/AccessWindowTranspose.cpp index b1043305b8..4506a0b44c 100644 --- a/src/core/AccessWindowTranspose.cpp +++ b/src/core/AccessWindowTranspose.cpp @@ -201,8 +201,8 @@ bool AccessWindowTranspose::update_padding_if_needed(const Window &window) const PaddingSize padding; padding.left = std::max(0, -min_x); padding.right = std::max<int>(0, max_x - shape[0]); - padding.top = shape.num_dimensions() == 1 ? 0 : std::max(0, -min_y); - padding.bottom = shape.num_dimensions() == 1 ? 0 : std::max<int>(0, max_y - shape[1]); + padding.top = std::max(0, -min_y); + padding.bottom = std::max<int>(0, max_y - shape[1]); // Update strides in tensor info return _info->extend_padding(padding); diff --git a/src/core/CL/cl_kernels/gemm.cl b/src/core/CL/cl_kernels/gemm.cl index 35a2e4704f..7f2a08bc2c 100644 --- a/src/core/CL/cl_kernels/gemm.cl +++ b/src/core/CL/cl_kernels/gemm.cl @@ -1040,7 +1040,7 @@ __kernel void gemm_mm_floating_point(IMAGE_DECLARATION(src0), VECTOR_TYPE acc3 = 0.0f; #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 - for(; src_addr.s0 <= (end_row_vec_a - 2 * sizeof(DATA_TYPE)); src_addr += (int2)(2 * sizeof(DATA_TYPE), 2 * src1_stride_y)) + for(; src_addr.s0 <= (end_row_vec_a - 2 * (int)sizeof(DATA_TYPE)); src_addr += (int2)(2 * sizeof(DATA_TYPE), 2 * src1_stride_y)) { // Load values from matrix A VEC_DATA_TYPE(DATA_TYPE, 2) @@ -1344,7 +1344,7 @@ __kernel void gemm_mm_qs16(IMAGE_DECLARATION(src0), #endif // NUM_ELEMS_PROCESSED_PER_THREAD_Y > 3 // This for loop performs 4 accumulations per iteration - for(; src_addr.s0 <= (end_row_vec_a - 2 * sizeof(short)); src_addr += (int2)(2 * sizeof(short), 2 * src1_stride_y)) + for(; src_addr.s0 <= (end_row_vec_a - 2 * (int)sizeof(short)); src_addr += (int2)(2 * sizeof(short), 2 * src1_stride_y)) { short2 a0 = vload2(0, (__global short *)(src0_ptr + src_addr.s0 + 0 * src0_stride_y)); #if NUM_ELEMS_PROCESSED_PER_THREAD_Y > 1 @@ -1629,7 +1629,7 @@ __kernel void gemm_lc_vm_f32(IMAGE_DECLARATION(src0), float4 acc = 0.0f; - for(; src_addr.s0 <= (end_row_vec_a - 2 * sizeof(float)); src_addr += (int2)(2 * sizeof(float), 2 * src1_stride_y)) + for(; src_addr.s0 <= (end_row_vec_a - 2 * (int)sizeof(float)); src_addr += (int2)(2 * sizeof(float), 2 * src1_stride_y)) { float2 a0 = vload2(0, (__global float *)(src0_ptr + src_addr.s0)); float4 b0 = vload4(0, (__global float *)(src1_ptr + src_addr.s1)); |