diff options
| author | Sheri Zhang <sheri.zhang@arm.com> | 2021-11-10 13:08:40 +0000 |
|---|---|---|
| committer | Sheri Zhang <sheri.zhang@arm.com> | 2021-11-17 10:31:59 +0000 |
| commit | e3b197410842652f0a78d04fe7b2c333cbeabab6 (patch) | |
| tree | 128d7b40239886d71a84605a8eb8f17c1d36e561 /src/core/CL/kernels/CLDepthwiseConvolutionLayerNativeKernel.cpp | |
| parent | d7154dbf0f4a347f2f35f2475a893f1631c5ee1a (diff) | |
| download | ComputeLibrary-e3b197410842652f0a78d04fe7b2c333cbeabab6.tar.gz | |
Improve start-up time for depthwise convolution
- Pass source and destination tensor dimension info at runtime
Resolves: COMPMID-4887
Signed-off-by: Sheri Zhang <sheri.zhang@arm.com>
Change-Id: Ib7c9f3ce6fb7cef600f7b0cd0fadafa4fa6888a1
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/6635
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Diffstat (limited to 'src/core/CL/kernels/CLDepthwiseConvolutionLayerNativeKernel.cpp')
| -rw-r--r-- | src/core/CL/kernels/CLDepthwiseConvolutionLayerNativeKernel.cpp | 10 |
1 files changed, 3 insertions, 7 deletions
diff --git a/src/core/CL/kernels/CLDepthwiseConvolutionLayerNativeKernel.cpp b/src/core/CL/kernels/CLDepthwiseConvolutionLayerNativeKernel.cpp index 2b74f91a05..61c8d90f78 100644 --- a/src/core/CL/kernels/CLDepthwiseConvolutionLayerNativeKernel.cpp +++ b/src/core/CL/kernels/CLDepthwiseConvolutionLayerNativeKernel.cpp @@ -215,15 +215,11 @@ void CLDepthwiseConvolutionLayerNativeKernel::configure(const CLCompileContext & build_opts.add_option("-DACTIVATION_TYPE=" + lower_string(string_from_activation_func(conv_info.act_info.activation()))); build_opts.add_option("-DDEPTH_MULTIPLIER=" + support::cpp11::to_string(conv_info.depth_multiplier)); build_opts.add_option("-DSRC_TENSOR_TYPE=BUFFER"); - build_opts.add_option("-DSRC_WIDTH=" + support::cpp11::to_string(_input->info()->dimension(1))); - build_opts.add_option("-DSRC_HEIGHT=" + support::cpp11::to_string(_input->info()->dimension(2))); // Note: SRC_DATA_TYPE must have the same data type of WEI_DATA_TYPE. In quantized, we could // have a case where the data types for the activation and weights are different. However, since the implementation // only works when both have same data type, we have to change the offset to take into account this aspect build_opts.add_option("-DSRC_DATA_TYPE=" + get_cl_type_from_data_type(_input->info()->data_type())); build_opts.add_option("-DDST_TENSOR_TYPE=BUFFER"); - build_opts.add_option("-DDST_WIDTH=" + support::cpp11::to_string(_output->info()->dimension(1))); - build_opts.add_option("-DDST_HEIGHT=" + support::cpp11::to_string(_output->info()->dimension(2))); build_opts.add_option("-DDST_DATA_TYPE=" + get_cl_type_from_data_type(_output->info()->data_type())); build_opts.add_option_if_else(_export_to_cl_image, "-DWEI_TENSOR_TYPE=IMAGE", "-DWEI_TENSOR_TYPE=BUFFER"); build_opts.add_option("-DWEI_WIDTH=" + support::cpp11::to_string(_weights->info()->dimension(1))); @@ -290,7 +286,6 @@ void CLDepthwiseConvolutionLayerNativeKernel::configure(const CLCompileContext & { kernel_name = "dwc_native_fp_nhwc"; build_opts.add_option("-DACC_DATA_TYPE=" + get_cl_type_from_data_type(input->info()->data_type())); - build_opts.add_option("-DZERO_VALUE=" + support::cpp11::to_string(0)); build_opts.add_option_if(conv_info.act_info.enabled(), "-DA_VAL=" + float_to_string_with_full_precision(conv_info.act_info.a())); build_opts.add_option_if(conv_info.act_info.enabled(), "-DB_VAL=" + float_to_string_with_full_precision(conv_info.act_info.b())); } @@ -358,8 +353,9 @@ void CLDepthwiseConvolutionLayerNativeKernel::run(const Window &window, cl::Comm } unsigned int idx = 0; - add_4D_tensor_argument(idx, _input, slice); - add_4D_tensor_argument(idx, _output, slice); + add_4d_tensor_nhwc_argument(idx, _input); + add_4d_tensor_nhwc_argument(idx, _output); + if(_export_to_cl_image) { _kernel.setArg(idx++, weights_cl_image); |