commit 76335eb8d8733b0bbc0110546797211540870c50
author Gian Marco Iodice <gianmarco.iodice@arm.com> Thu Nov 17 11:03:39 2022 +0000
committer Gian Marco Iodice <gianmarco.iodice@arm.com> Fri Dec 09 11:06:23 2022 +0000
tree 812fc44de593c9e1e45ac8b534094511b06163bf
parent f16973b8b4605f12608bffa9f0ca6ed590202d41

Implement the OpenCL kernel to compute the indirect convolution

- Implement indirect convolution kernel
- Add operator support
- Add test

Resolves COMPMID-5709

Change-Id: I9272304163471a5a40da7fdec204599f3c1d8e32
Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8701
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Gunes Bayir <gunes.bayir@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>

src/gpu/cl/kernels/ClIndirectConv2dAddressPrecalculationKernel.cpp

diff --git a/src/gpu/cl/kernels/ClIndirectConv2dAddressPrecalculationKernel.cpp b/src/gpu/cl/kernels/ClIndirectConv2dAddressPrecalculationKernel.cpp
index 95186fe..8e12f23 100644
--- a/src/gpu/cl/kernels/ClIndirectConv2dAddressPrecalculationKernel.cpp
+++ b/src/gpu/cl/kernels/ClIndirectConv2dAddressPrecalculationKernel.cpp
@@ -80,8 +80,8 @@
     ARM_COMPUTE_ERROR_ON_NULLPTR(src, weights, dst);
     ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(src, weights, dst, conv_info, desc));
 
-    constexpr unsigned int width_idx   = 1;
-    constexpr unsigned int height_idx  = 2;
+    constexpr unsigned int width_idx  = 1;
+    constexpr unsigned int height_idx = 2;
 
     // Get dst shape
     TensorShape output_shape = misc::shape_calculator::compute_indirect_buffer_shape(src->tensor_shape(),
@@ -127,6 +127,9 @@
     build_options.add_option("-DPAD_TOP=" + support::cpp11::to_string(pad_top));
     build_options.add_option("-DM0=" + support::cpp11::to_string(desc.m0));
 
+    // A macro guard to compile ONLY the kernel of interest
+    build_options.add_option("-D" + upper_string(kernel_name.str()));
+
     _kernel = create_kernel(compile_context, kernel_name.str(), build_options.options());
 
     // Since this kernel should be called only once, we do not need to set the config_id for tuning