diff options
Diffstat (limited to 'src/core/CL/kernels/CLFFTDigitReverseKernel.cpp')
| -rw-r--r-- | src/core/CL/kernels/CLFFTDigitReverseKernel.cpp | 36 |
1 files changed, 22 insertions, 14 deletions
diff --git a/src/core/CL/kernels/CLFFTDigitReverseKernel.cpp b/src/core/CL/kernels/CLFFTDigitReverseKernel.cpp index d72647c3c9..b04293db5b 100644 --- a/src/core/CL/kernels/CLFFTDigitReverseKernel.cpp +++ b/src/core/CL/kernels/CLFFTDigitReverseKernel.cpp @@ -34,16 +34,19 @@ namespace arm_compute { namespace { -Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, const ITensorInfo *idx, unsigned int axis) +Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, const ITensorInfo *idx, const FFTDigitReverseKernelInfo &config) { ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(input); - ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 2, DataType::F32); + ARM_COMPUTE_RETURN_ERROR_ON(input->data_type() != DataType::F32); + ARM_COMPUTE_RETURN_ERROR_ON(input->num_channels() != 1 && input->num_channels() != 2); ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(idx, 1, DataType::U32); - ARM_COMPUTE_RETURN_ERROR_ON(axis != 0); + ARM_COMPUTE_RETURN_ERROR_ON(std::set<unsigned int>({ 0, 1 }).count(config.axis) == 0); + ARM_COMPUTE_RETURN_ERROR_ON(input->tensor_shape()[config.axis] != idx->tensor_shape().x()); // Checks performed when output is configured if((output != nullptr) && (output->total_size() != 0)) { + ARM_COMPUTE_RETURN_ERROR_ON(output->num_channels() != 2); ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input, output); ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output); } @@ -51,11 +54,11 @@ Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, c return Status{}; } -std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, ITensorInfo *output, ITensorInfo *idx, unsigned int axis) +std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, ITensorInfo *output, ITensorInfo *idx, const FFTDigitReverseKernelInfo &config) { - ARM_COMPUTE_UNUSED(idx, axis); + ARM_COMPUTE_UNUSED(idx, config); - auto_init_if_empty(*output, *input); + auto_init_if_empty(*output, input->clone()->set_num_channels(2)); Window win = calculate_max_window(*output, Steps()); output->set_valid_region(ValidRegion(Coordinates(), output->tensor_shape())); @@ -69,25 +72,30 @@ CLFFTDigitReverseKernel::CLFFTDigitReverseKernel() { } -void CLFFTDigitReverseKernel::configure(const ICLTensor *input, ICLTensor *output, const ICLTensor *idx, unsigned int axis) +void CLFFTDigitReverseKernel::configure(const ICLTensor *input, ICLTensor *output, const ICLTensor *idx, const FFTDigitReverseKernelInfo &config) { ARM_COMPUTE_ERROR_ON_NULLPTR(input, output, idx); - ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info(), idx->info(), axis)); + ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info(), idx->info(), config)); _input = input; _output = output; _idx = idx; // Create kernel - _kernel = static_cast<cl::Kernel>(CLKernelLibrary::get().create_kernel("digit_reverse")); + CLBuildOptions build_opts; + build_opts.add_option("-DVEC_SIZE=" + support::cpp11::to_string(input->info()->num_channels())); + build_opts.add_option_if(config.conjugate, "-DCONJ"); + std::string kernel_name = "fft_digit_reverse_axis_" + support::cpp11::to_string(config.axis); + _kernel = static_cast<cl::Kernel>(CLKernelLibrary::get().create_kernel(kernel_name, build_opts.options())); // Configure kernel window - auto win_config = validate_and_configure_window(input->info(), output->info(), idx->info(), axis); + auto win_config = validate_and_configure_window(input->info(), output->info(), idx->info(), config); ARM_COMPUTE_ERROR_THROW_ON(win_config.first); ICLKernel::configure_internal(win_config.second); // Set config_id for enabling LWS tuning - _config_id = "digit_reverse_"; + _config_id = kernel_name; + _config_id += "_"; _config_id += lower_string(string_from_data_type(input->info()->data_type())); _config_id += "_"; _config_id += support::cpp11::to_string(input->info()->dimension(0)); @@ -95,10 +103,10 @@ void CLFFTDigitReverseKernel::configure(const ICLTensor *input, ICLTensor *outpu _config_id += support::cpp11::to_string(input->info()->dimension(1)); } -Status CLFFTDigitReverseKernel::validate(const ITensorInfo *input, const ITensorInfo *output, const ITensorInfo *idx, unsigned int axis) +Status CLFFTDigitReverseKernel::validate(const ITensorInfo *input, const ITensorInfo *output, const ITensorInfo *idx, const FFTDigitReverseKernelInfo &config) { - ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, idx, axis)); - ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input->clone().get(), output->clone().get(), idx->clone().get(), axis).first); + ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, idx, config)); + ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input->clone().get(), output->clone().get(), idx->clone().get(), config).first); return Status{}; } |