diff options
Diffstat (limited to 'src/gpu/cl/kernels/ClQuantizeKernel.cpp')
-rw-r--r-- | src/gpu/cl/kernels/ClQuantizeKernel.cpp | 27 |
1 files changed, 15 insertions, 12 deletions
diff --git a/src/gpu/cl/kernels/ClQuantizeKernel.cpp b/src/gpu/cl/kernels/ClQuantizeKernel.cpp index 5c8bf97f0f..e8df420f67 100644 --- a/src/gpu/cl/kernels/ClQuantizeKernel.cpp +++ b/src/gpu/cl/kernels/ClQuantizeKernel.cpp @@ -29,13 +29,12 @@ #include "arm_compute/core/Error.h" #include "arm_compute/core/TensorInfo.h" #include "arm_compute/core/Utils.h" +#include "arm_compute/core/utils/quantization/AsymmHelpers.h" #include "arm_compute/core/utils/StringUtils.h" #include "arm_compute/core/Validate.h" -#include "arm_compute/core/utils/quantization/AsymmHelpers.h" #include "src/core/CL/CLValidate.h" #include "src/core/helpers/WindowHelpers.h" - #include "support/Cast.h" #include "support/StringSupport.h" @@ -50,12 +49,14 @@ namespace Status validate_arguments(const ITensorInfo *src, const ITensorInfo *dst) { ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(src, dst); - ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(src, 1, DataType::QASYMM8, DataType::QASYMM8_SIGNED, DataType::F32, DataType::F16); + ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(src, 1, DataType::QASYMM8, DataType::QASYMM8_SIGNED, + DataType::F32, DataType::F16); ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(src); // Output must always be initialized ARM_COMPUTE_RETURN_ERROR_ON(dst->tensor_shape().total_size() == 0); - ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(dst, 1, DataType::QASYMM8, DataType::QASYMM8_SIGNED, DataType::QASYMM16); + ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(dst, 1, DataType::QASYMM8, DataType::QASYMM8_SIGNED, + DataType::QASYMM16); ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(src, dst); return Status{}; @@ -71,7 +72,7 @@ void ClQuantizeKernel::configure(const CLCompileContext &compile_context, const { ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst); - auto padding_info = get_padding_info({ src, dst }); + auto padding_info = get_padding_info({src, dst}); ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(src, dst)); @@ -84,7 +85,7 @@ void ClQuantizeKernel::configure(const CLCompileContext &compile_context, const float scale_to_apply = qinfo.scale; int32_t offset_to_apply = qinfo.offset; - if(is_data_type_quantized_asymmetric(src->data_type())) + if (is_data_type_quantized_asymmetric(src->data_type())) { /* * In case of requantization of a quantized input tensor to an output tensor with another quantization @@ -132,8 +133,10 @@ void ClQuantizeKernel::configure(const CLCompileContext &compile_context, const build_opts.add_option("-DVEC_SIZE=" + support::cpp11::to_string(vec_size_x)); build_opts.add_option("-DDATA_TYPE_IN=" + get_cl_type_from_data_type(src->data_type())); build_opts.add_option("-DDATA_TYPE_OUT=" + get_cl_type_from_data_type(output_data_type)); - build_opts.add_option_if(multi_access_x, "-DLAST_ACCESSED_X=" + support::cpp11::to_string(std::max<int>(input_width_x - vec_size_x, 0))); - std::pair<int, int> min_max_quant_values = quantization::get_min_max_values_from_quantized_data_type(output_data_type); + build_opts.add_option_if( + multi_access_x, "-DLAST_ACCESSED_X=" + support::cpp11::to_string(std::max<int>(input_width_x - vec_size_x, 0))); + std::pair<int, int> min_max_quant_values = + quantization::get_min_max_values_from_quantized_data_type(output_data_type); build_opts.add_option("-DMIN_QUANT_VAL=" + support::cpp11::to_string(min_max_quant_values.first)); build_opts.add_option("-DMAX_QUANT_VAL=" + support::cpp11::to_string(min_max_quant_values.second)); @@ -141,9 +144,10 @@ void ClQuantizeKernel::configure(const CLCompileContext &compile_context, const // Configure kernel window Window win = calculate_max_window(*src, Steps()); - if(multi_access_x) + if (multi_access_x) { - win.set(Window::DimX, Window::Dimension(win.x().start(), ceil_to_multiple(win.x().end(), vec_size_x), vec_size_x)); + win.set(Window::DimX, + Window::Dimension(win.x().start(), ceil_to_multiple(win.x().end(), vec_size_x), vec_size_x)); } ICLKernel::configure_internal(win); @@ -173,8 +177,7 @@ void ClQuantizeKernel::run_op(ITensorPack &tensors, const Window &window, cl::Co add_3D_tensor_argument(idx, src, slice); add_3D_tensor_argument(idx, dst, slice); enqueue(queue, *this, slice, lws_hint()); - } - while(window_collapsed.slide_window_slice_3D(slice)); + } while (window_collapsed.slide_window_slice_3D(slice)); } } // namespace kernels } // namespace opencl |