diff options
Diffstat (limited to 'src/core/CL/kernels/CLDepthwiseIm2ColKernel.cpp')
-rw-r--r-- | src/core/CL/kernels/CLDepthwiseIm2ColKernel.cpp | 42 |
1 files changed, 22 insertions, 20 deletions
diff --git a/src/core/CL/kernels/CLDepthwiseIm2ColKernel.cpp b/src/core/CL/kernels/CLDepthwiseIm2ColKernel.cpp index ad9ac0ecd6..8467b39910 100644 --- a/src/core/CL/kernels/CLDepthwiseIm2ColKernel.cpp +++ b/src/core/CL/kernels/CLDepthwiseIm2ColKernel.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2017 ARM Limited. + * Copyright (c) 2017-2018 ARM Limited. * * SPDX-License-Identifier: MIT * @@ -44,9 +44,10 @@ CLDepthwiseIm2ColKernel::CLDepthwiseIm2ColKernel() void CLDepthwiseIm2ColKernel::configure(const ICLTensor *input, ICLTensor *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info, bool has_bias) { - ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::F16, DataType::F32); + ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8, DataType::F16, DataType::F32); ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(input, output); ARM_COMPUTE_ERROR_ON_MISMATCHING_FIXED_POINT(input, output); + ARM_COMPUTE_ERROR_ON(is_data_type_quantized_asymmetric(input->info()->data_type()) && has_bias); ARM_COMPUTE_ERROR_ON(input->info()->dimension(2) != output->info()->dimension(2)); ARM_COMPUTE_ERROR_ON(output->info()->dimension(0) != (kernel_dims.width * kernel_dims.height + ((has_bias) ? 1 : 0))); @@ -54,24 +55,25 @@ void CLDepthwiseIm2ColKernel::configure(const ICLTensor *input, ICLTensor *outpu _output = output; // Create kernel - std::set<std::string> build_opts; - - build_opts.emplace("-DDATA_TYPE=" + get_cl_type_from_data_type(input->info()->data_type())); - build_opts.emplace("-DSTRIDE_X=" + support::cpp11::to_string(conv_info.stride().first)); - build_opts.emplace("-DSTRIDE_Y=" + support::cpp11::to_string(conv_info.stride().second)); - build_opts.emplace("-DPAD_LEFT=" + support::cpp11::to_string(conv_info.pad_left())); - build_opts.emplace("-DPAD_TOP=" + support::cpp11::to_string(conv_info.pad_top())); - build_opts.emplace("-DPAD_RIGHT=" + support::cpp11::to_string(conv_info.pad_right())); - build_opts.emplace("-DPAD_BOTTOM=" + support::cpp11::to_string(conv_info.pad_bottom())); - build_opts.emplace("-DSRC_WIDTH=" + support::cpp11::to_string(input->info()->dimension(0))); - build_opts.emplace("-DSRC_HEIGHT=" + support::cpp11::to_string(input->info()->dimension(1))); - build_opts.emplace("-DKERNEL_WIDTH=" + support::cpp11::to_string(kernel_dims.width)); - build_opts.emplace("-DKERNEL_HEIGHT=" + support::cpp11::to_string(kernel_dims.height)); - if(has_bias) - { - build_opts.emplace("-DHAS_BIAS"); - } - _kernel = static_cast<cl::Kernel>(CLKernelLibrary::get().create_kernel("depthwise_im2col", build_opts)); + CLBuildOptions build_opts; + + build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(input->info()->data_type())); + build_opts.add_option("-DSTRIDE_X=" + support::cpp11::to_string(conv_info.stride().first)); + build_opts.add_option("-DSTRIDE_Y=" + support::cpp11::to_string(conv_info.stride().second)); + build_opts.add_option("-DPAD_LEFT=" + support::cpp11::to_string(conv_info.pad_left())); + build_opts.add_option("-DPAD_TOP=" + support::cpp11::to_string(conv_info.pad_top())); + build_opts.add_option("-DPAD_RIGHT=" + support::cpp11::to_string(conv_info.pad_right())); + build_opts.add_option("-DPAD_BOTTOM=" + support::cpp11::to_string(conv_info.pad_bottom())); + build_opts.add_option("-DSRC_WIDTH=" + support::cpp11::to_string(input->info()->dimension(0))); + build_opts.add_option("-DSRC_HEIGHT=" + support::cpp11::to_string(input->info()->dimension(1))); + build_opts.add_option("-DKERNEL_WIDTH=" + support::cpp11::to_string(kernel_dims.width)); + build_opts.add_option("-DKERNEL_HEIGHT=" + support::cpp11::to_string(kernel_dims.height)); + build_opts.add_option_if(has_bias, "-DHAS_BIAS"); + build_opts.add_option_if_else(is_data_type_quantized(input->info()->data_type()), + "-DPAD_VALUE=" + support::cpp11::to_string(input->info()->quantization_info().offset), + "-DPAD_VALUE=0"); + + _kernel = static_cast<cl::Kernel>(CLKernelLibrary::get().create_kernel("depthwise_im2col", build_opts.options())); // Configure the local work size for Bifrost with a value obtained // via exhaustive autotuning for the MobileNets tensor shapes. |