aboutsummaryrefslogtreecommitdiff
path: root/src/core/CL/kernels/CLSoftmaxLayerKernel.cpp
diff options
context:
space:
mode:
Diffstat (limited to 'src/core/CL/kernels/CLSoftmaxLayerKernel.cpp')
-rw-r--r--src/core/CL/kernels/CLSoftmaxLayerKernel.cpp14
1 files changed, 12 insertions, 2 deletions
diff --git a/src/core/CL/kernels/CLSoftmaxLayerKernel.cpp b/src/core/CL/kernels/CLSoftmaxLayerKernel.cpp
index d7a1778e26..112d864827 100644
--- a/src/core/CL/kernels/CLSoftmaxLayerKernel.cpp
+++ b/src/core/CL/kernels/CLSoftmaxLayerKernel.cpp
@@ -220,6 +220,11 @@ CLLogits1DMaxShiftExpSumKernel::CLLogits1DMaxShiftExpSumKernel()
void CLLogits1DMaxShiftExpSumKernel::configure(const ICLTensor *input, ICLTensor *max, ICLTensor *output, ICLTensor *sum, const SoftmaxKernelInfo &info)
{
+ configure(CLKernelLibrary::get().get_compile_context(), input, max, output, sum, info);
+}
+
+void CLLogits1DMaxShiftExpSumKernel::configure(CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *max, ICLTensor *output, ICLTensor *sum, const SoftmaxKernelInfo &info)
+{
ARM_COMPUTE_ERROR_ON_NULLPTR(input, max, sum, output);
// Output auto initialization if not yet initialized
@@ -277,7 +282,7 @@ void CLLogits1DMaxShiftExpSumKernel::configure(const ICLTensor *input, ICLTensor
}
// Create kernel.
- _kernel = static_cast<cl::Kernel>(CLKernelLibrary::get().create_kernel(kernel_name, build_opts.options()));
+ _kernel = create_kernel(compile_context, kernel_name, build_opts.options());
// Set static arguments. Both the kernels use the same arguments
unsigned int idx = 4 * num_arguments_per_3D_tensor(); //Skip the input and output parameters
@@ -343,6 +348,11 @@ CLLogits1DNormKernel::CLLogits1DNormKernel()
void CLLogits1DNormKernel::configure(const ICLTensor *input, const ICLTensor *sum, ICLTensor *output, const SoftmaxKernelInfo &info)
{
+ configure(CLKernelLibrary::get().get_compile_context(), input, sum, output, info);
+}
+
+void CLLogits1DNormKernel::configure(CLCompileContext &compile_context, const ICLTensor *input, const ICLTensor *sum, ICLTensor *output, const SoftmaxKernelInfo &info)
+{
ARM_COMPUTE_ERROR_ON_NULLPTR(input, sum, output);
// Note: output should always have a scale of 1/256 and offset 0
@@ -376,7 +386,7 @@ void CLLogits1DNormKernel::configure(const ICLTensor *input, const ICLTensor *su
// Create kernel
std::string kernel_name = is_quantized_asymmetric ? "softmax_layer_norm_quantized" : "softmax_layer_norm";
- _kernel = static_cast<cl::Kernel>(CLKernelLibrary::get().create_kernel(kernel_name, build_opts.options()));
+ _kernel = create_kernel(compile_context, kernel_name, build_opts.options());
// Configure window
auto win_config = validate_and_configure_window_1DNorm(input->info(), output->info(), sum->info(), info);