Update ClConv2D heuristic to use direct convolution
Resolves COMPMID-5298
Change-Id: I4a7d788bc1f5f568bedcc22e7aca47ede6de71bf
Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Signed-off-by: Adnan AlSinan <adnan.alsinan@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/7891
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Gunes Bayir <gunes.bayir@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>
diff --git a/src/gpu/cl/operators/ClConv2d.cpp b/src/gpu/cl/operators/ClConv2d.cpp
index 23c1b8a..09beee7 100644
--- a/src/gpu/cl/operators/ClConv2d.cpp
+++ b/src/gpu/cl/operators/ClConv2d.cpp
@@ -261,33 +261,49 @@
const bool is_ifm_ge_16 = src->dimension(idx_c) >= 16;
const bool is_ofm_lte_8 = weights->dimension(3U) <= 8;
const bool workload_gte_8192 = (output_shape[0] * output_shape[1] * output_shape[2]) / 16 >= 8192;
- const bool is_ifm_gt_ofm = src->dimension(idx_c) > weights->dimension(3U);
+ const bool is_ifm_gt_ofm = src->dimension(idx_c) > weights->dimension(3U);
+ const bool is_m_one = output_shape[1] * output_shape[2] == 1;
// Run Winograd if valid and IFM >= 16
if(is_wino_valid && is_ifm_ge_16)
{
- return ConvolutionMethod::WINOGRAD;
+ if(is_ofm_lte_8)
+ {
+ if(gpu_target == arm_compute::GPUTarget::G71 || gpu_target == arm_compute::GPUTarget::G72 || get_arch_from_target(gpu_target) == arm_compute::GPUTarget::MIDGARD)
+ {
+ return ConvolutionMethod::WINOGRAD;
+ }
+ }
+ else
+ {
+ return ConvolutionMethod::WINOGRAD;
+ }
}
// Direct convolution case
if(is_direct_valid)
{
- if((gpu_target == arm_compute::GPUTarget::G71 ||
- gpu_target == arm_compute::GPUTarget::G72 ||
- gpu_target == arm_compute::GPUTarget::MIDGARD))
+ if((gpu_target == arm_compute::GPUTarget::G71 || gpu_target == arm_compute::GPUTarget::G72 || get_arch_from_target(gpu_target) == arm_compute::GPUTarget::MIDGARD))
{
if(is_large_kernel_sz && is_ifm_ge_16 && is_ifm_gt_ofm)
{
return ConvolutionMethod::DIRECT;
}
}
- else
+ else if(gpu_target == arm_compute::GPUTarget::G76)
{
if((is_large_kernel_sz && workload_gte_8192 && is_ifm_ge_16) || (is_ofm_lte_8 && is_ifm_ge_16))
{
return ConvolutionMethod::DIRECT;
}
}
+ else
+ {
+ if(is_large_kernel_sz || is_ofm_lte_8 || is_m_one)
+ {
+ return ConvolutionMethod::DIRECT;
+ }
+ }
}
// Default case