diff options
Diffstat (limited to 'src/runtime/CL/functions/CLInstanceNormalizationLayer.cpp')
| -rw-r--r-- | src/runtime/CL/functions/CLInstanceNormalizationLayer.cpp | 59 |
1 files changed, 48 insertions, 11 deletions
diff --git a/src/runtime/CL/functions/CLInstanceNormalizationLayer.cpp b/src/runtime/CL/functions/CLInstanceNormalizationLayer.cpp index bd680f448d..0e994e1aee 100644 --- a/src/runtime/CL/functions/CLInstanceNormalizationLayer.cpp +++ b/src/runtime/CL/functions/CLInstanceNormalizationLayer.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2019-2020 Arm Limited. + * Copyright (c) 2019-2021 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -23,31 +23,68 @@ */ #include "arm_compute/runtime/CL/functions/CLInstanceNormalizationLayer.h" +#include "arm_compute/core/Error.h" #include "arm_compute/core/Types.h" +#include "arm_compute/runtime/CL/CLHelpers.h" +#include "arm_compute/runtime/CL/CLScheduler.h" + +#include "src/common/utils/Log.h" +#include "src/core/CL/ICLKernel.h" #include "src/core/CL/kernels/CLFillBorderKernel.h" #include "src/core/CL/kernels/CLInstanceNormalizationLayerKernel.h" -#include "support/MemorySupport.h" namespace arm_compute { -CLInstanceNormalizationLayer::CLInstanceNormalizationLayer() +CLInstanceNormalizationLayer::CLInstanceNormalizationLayer(CLRuntimeContext *ctx) // NOLINT + : _inst_norm_kernel(), _mean_var_kernel(), _mean_var_tensor(), _ctx(ctx) +{ +} +CLInstanceNormalizationLayer::~CLInstanceNormalizationLayer() { } -void CLInstanceNormalizationLayer::configure(ICLTensor *input, ICLTensor *output, float gamma, float beta, float epsilon, bool use_mixed_precision) +void CLInstanceNormalizationLayer::configure( + ICLTensor *input, ICLTensor *output, float gamma, float beta, float epsilon, bool use_mixed_precision) { configure(CLKernelLibrary::get().get_compile_context(), input, output, gamma, beta, epsilon, use_mixed_precision); } -void CLInstanceNormalizationLayer::configure(const CLCompileContext &compile_context, ICLTensor *input, ICLTensor *output, float gamma, float beta, float epsilon, bool use_mixed_precision) +void CLInstanceNormalizationLayer::configure(const CLCompileContext &compile_context, + ICLTensor *input, + ICLTensor *output, + float gamma, + float beta, + float epsilon, + bool use_mixed_precision) { - auto k = arm_compute::support::cpp14::make_unique<CLInstanceNormalizationLayerKernel>(); - k->configure(compile_context, input, output, InstanceNormalizationLayerKernelInfo(gamma, beta, epsilon, use_mixed_precision)); - _kernel = std::move(k); + ARM_COMPUTE_LOG_PARAMS(input, output, gamma, beta, epsilon, use_mixed_precision); + auto w = std::make_unique<CLComputeMeanVariance>(); + w->configure(compile_context, input, &_mean_var_tensor, use_mixed_precision); + _mean_var_kernel = std::move(w); + auto k = std::make_unique<CLInstanceNormalizationLayerKernel>(); + k->configure(compile_context, input, &_mean_var_tensor, output, + InstanceNormalizationLayerKernelInfo(gamma, beta, epsilon, use_mixed_precision)); + _inst_norm_kernel = std::move(k); + _mean_var_tensor.allocator()->allocate(); } -Status CLInstanceNormalizationLayer::validate(const ITensorInfo *input, const ITensorInfo *output, float gamma, float beta, float epsilon, bool use_mixed_precision) +Status CLInstanceNormalizationLayer::validate(const ITensorInfo *input, + const ITensorInfo *output, + float gamma, + float beta, + float epsilon, + bool use_mixed_precision) { - return CLInstanceNormalizationLayerKernel::validate(input, output, InstanceNormalizationLayerKernelInfo(gamma, beta, epsilon, use_mixed_precision)); + return CLInstanceNormalizationLayerKernel::validate( + input, output, InstanceNormalizationLayerKernelInfo(gamma, beta, epsilon, use_mixed_precision)); } -} // namespace arm_compute
\ No newline at end of file + +void CLInstanceNormalizationLayer::run() +{ + ARM_COMPUTE_ERROR_ON_MSG(!_inst_norm_kernel, + "The child class didn't set the CL kernel or function isn't configured"); + schedule_kernel_on_ctx(_ctx, _mean_var_kernel.get()); + schedule_kernel_on_ctx(_ctx, _inst_norm_kernel.get()); +} + +} // namespace arm_compute |