commit e87120731ca65c54b082734af07f748ac9651427
author Adnan AlSinan <adnan.alsinan@arm.com> Thu Jul 21 16:34:49 2022 +0100
committer Adnan AlSinan <adnan.alsinan@arm.com> Fri Jul 22 09:59:40 2022 +0000
tree 46ea0317661060b70c74955ae495894fa374260b
parent c6b7913b894f8191878c7b73626fbeb65850131c

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>

src/gpu/cl/operators/ClConv2d.cpp

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