diff options
Diffstat (limited to 'src/gpu/cl/kernels/ClCol2ImKernel.cpp')
| -rw-r--r-- | src/gpu/cl/kernels/ClCol2ImKernel.cpp | 47 |
1 files changed, 32 insertions, 15 deletions
diff --git a/src/gpu/cl/kernels/ClCol2ImKernel.cpp b/src/gpu/cl/kernels/ClCol2ImKernel.cpp index 3316742912..9972e07f05 100644 --- a/src/gpu/cl/kernels/ClCol2ImKernel.cpp +++ b/src/gpu/cl/kernels/ClCol2ImKernel.cpp @@ -30,6 +30,7 @@ #include "arm_compute/core/Helpers.h" #include "arm_compute/core/utils/misc/ShapeCalculator.h" #include "arm_compute/core/utils/StringUtils.h" + #include "src/core/CL/CLValidate.h" #include "src/core/helpers/AutoConfiguration.h" #include "src/core/helpers/WindowHelpers.h" @@ -47,29 +48,38 @@ namespace kernels { namespace { -Status validate_arguments(const ITensorInfo *src, const ITensorInfo *dst, const Size2D &convolved_dims, unsigned int num_groups) +Status validate_arguments(const ITensorInfo *src, + const ITensorInfo *dst, + const Size2D &convolved_dims, + unsigned int num_groups) { ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(src, dst); ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(src); - ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(src, 1, DataType::QASYMM8, DataType::QASYMM8_SIGNED, DataType::F16, DataType::F32); + ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(src, 1, DataType::QASYMM8, DataType::QASYMM8_SIGNED, + DataType::F16, DataType::F32); // Checks performed when output is configured - if(dst->total_size() != 0) + if (dst->total_size() != 0) { - ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(dst->tensor_shape(), compute_col2im_shape(*src, convolved_dims, true, num_groups)); + ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS( + dst->tensor_shape(), compute_col2im_shape(*src, convolved_dims, true, num_groups)); ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(src, dst); ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_QUANTIZATION_INFO(src, dst); - ARM_COMPUTE_RETURN_ERROR_ON_MSG(dst->data_layout() != DataLayout::NCHW, "Col2Im output's data layout must always be NCHW"); + ARM_COMPUTE_RETURN_ERROR_ON_MSG(dst->data_layout() != DataLayout::NCHW, + "Col2Im output's data layout must always be NCHW"); } return Status{}; } -std::pair<Status, Window> validate_and_configure_window(ITensorInfo *src, ITensorInfo *dst, const Size2D &convolved_dims, unsigned int num_groups) +std::pair<Status, Window> +validate_and_configure_window(ITensorInfo *src, ITensorInfo *dst, const Size2D &convolved_dims, unsigned int num_groups) { ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst); // Output auto inizialitation if not yet initialized - auto_init_if_empty(*dst, src->clone()->set_tensor_shape(compute_col2im_shape(*src, convolved_dims, true, num_groups)).set_data_layout(DataLayout::NCHW)); + auto_init_if_empty(*dst, src->clone() + ->set_tensor_shape(compute_col2im_shape(*src, convolved_dims, true, num_groups)) + .set_data_layout(DataLayout::NCHW)); constexpr unsigned int num_elems_read_per_iteration = 8; @@ -80,18 +90,22 @@ std::pair<Status, Window> validate_and_configure_window(ITensorInfo *src, ITenso AccessWindowHorizontal input_access(src, 0, num_elems_read_per_iteration); bool window_changed = update_window_and_padding(win, input_access); - Status err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{}; + Status err = + (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{}; return std::make_pair(err, win); } } // namespace -ClCol2ImKernel::ClCol2ImKernel() - : _convolved_dims() +ClCol2ImKernel::ClCol2ImKernel() : _convolved_dims() { _type = CLKernelType::ELEMENTWISE; } -void ClCol2ImKernel::configure(const CLCompileContext &compile_context, ITensorInfo *src, ITensorInfo *dst, const Size2D &convolved_dims, unsigned int num_groups) +void ClCol2ImKernel::configure(const CLCompileContext &compile_context, + ITensorInfo *src, + ITensorInfo *dst, + const Size2D &convolved_dims, + unsigned int num_groups) { ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst); @@ -132,11 +146,15 @@ void ClCol2ImKernel::configure(const CLCompileContext &compile_context, ITensorI _config_id += support::cpp11::to_string(dst->dimension(1)); } -Status ClCol2ImKernel::validate(const ITensorInfo *src, const ITensorInfo *dst, const Size2D &convolved_dims, unsigned int num_groups) +Status ClCol2ImKernel::validate(const ITensorInfo *src, + const ITensorInfo *dst, + const Size2D &convolved_dims, + unsigned int num_groups) { ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst); ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(src, dst, convolved_dims, num_groups)); - ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(src->clone().get(), dst->clone().get(), convolved_dims, num_groups).first); + ARM_COMPUTE_RETURN_ON_ERROR( + validate_and_configure_window(src->clone().get(), dst->clone().get(), convolved_dims, num_groups).first); return Status{}; } @@ -168,8 +186,7 @@ void ClCol2ImKernel::run_op(ITensorPack &tensors, const Window &window, cl::Comm add_3D_tensor_argument(idx, src, slice); add_4D_tensor_argument(idx, dst, slice_out); enqueue(queue, *this, slice, lws_hint()); - } - while(collapsed.slide_window_slice_3D(slice) && collapsed_out.slide_window_slice_4D(slice_out)); + } while (collapsed.slide_window_slice_3D(slice) && collapsed_out.slide_window_slice_4D(slice_out)); } } // namespace kernels } // namespace opencl |