diff options
author | Gian Marco Iodice <gianmarco.iodice@arm.com> | 2021-07-06 13:19:41 +0100 |
---|---|---|
committer | Georgios Pinitas <georgios.pinitas@arm.com> | 2021-07-06 16:38:19 +0000 |
commit | 5e281814c5110724d99fe8ee64bdf42ef2c31bce (patch) | |
tree | 008a57c80f5b846265b0339f6e3a9f7876fa8922 /src/core/CL/kernels | |
parent | 900289936c458eff95499e0a0eaba989a27aaa4d (diff) | |
download | ComputeLibrary-5e281814c5110724d99fe8ee64bdf42ef2c31bce.tar.gz |
Fix manual LOOP_UNROLLING
The issue is caused by the number of iterations passed to
LOOP_UNROLLING. When we use the manual LOOP_UNROLLING, the number of
iterations must be less than or equal to 128.
To overcome this problem, we create a utility function to check if
any of the critical iterations (kernel dimensions) are beyond that
limit. If so, the utility function, disable the manual loop unrolling.
Resolves COMPMID-4609
Change-Id: I7221c967609e462a5abd1cbb74e2a120f344fcb3
Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5913
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/core/CL/kernels')
-rw-r--r-- | src/core/CL/kernels/CLDepthwiseConvolutionLayerNativeKernel.cpp | 11 |
1 files changed, 9 insertions, 2 deletions
diff --git a/src/core/CL/kernels/CLDepthwiseConvolutionLayerNativeKernel.cpp b/src/core/CL/kernels/CLDepthwiseConvolutionLayerNativeKernel.cpp index 4bde303f1e..1437b5bebb 100644 --- a/src/core/CL/kernels/CLDepthwiseConvolutionLayerNativeKernel.cpp +++ b/src/core/CL/kernels/CLDepthwiseConvolutionLayerNativeKernel.cpp @@ -83,7 +83,7 @@ Status validate_arguments(const ITensorInfo *input, const ITensorInfo *weights, } const ConvolutionInfo info{ conv_info.pad_stride_info, conv_info.depth_multiplier, ActivationLayerInfo(), conv_info.dilation }; - const TensorShape output_shape = arm_compute::misc::shape_calculator::compute_depthwise_convolution_shape(*input, *weights, conv_info); + const TensorShape output_shape = arm_compute::misc::shape_calculator::compute_depthwise_convolution_shape(*input, *weights, conv_info); const bool is_quantized = is_data_type_quantized(input->data_type()); @@ -237,9 +237,16 @@ void CLDepthwiseConvolutionLayerNativeKernel::configure(const CLCompileContext & build_opts.add_option("-DDILATION_Y=" + support::cpp11::to_string(conv_info.dilation.y())); build_opts.add_option("-DN0=" + support::cpp11::to_string(n0)); build_opts.add_option("-DM0=" + support::cpp11::to_string(m0)); - build_opts.add_option("-DM0_A=" + support::cpp11::to_string(weights->info()->dimension(1) + m0 - 1)); + build_opts.add_option("-DM0_A=" + support::cpp11::to_string(_weights->info()->dimension(1) + m0 - 1)); build_opts.add_option("-DPARTIAL_N0=" + support::cpp11::to_string(_input->info()->dimension(0) % n0)); build_opts.add_option_if(_input->info()->num_dimensions() > 3, "-DBATCHED_EXECUTION"); + + // Force unroll with pragma when any of the following values exceed the maximum number of manual unroll + set_unroll_with_pragma(build_opts, { static_cast<int>(_weights->info()->dimension(1) + m0 - 1), + static_cast<int>(_weights->info()->dimension(1)), + static_cast<int>(_weights->info()->dimension(2)) + }); + if(biases != nullptr) { build_opts.add_option(std::string("-DHAS_BIAS")); |