commit 96fa7637c0d7bffa4a64bfffdadb7aef36838bc0
author Adnan AlSinan <adnan.alsinan@arm.com> Wed Aug 10 18:14:33 2022 +0100
committer Ramy Elgammal <ramy.elgammal@arm.com> Fri Aug 12 15:15:02 2022 +0100
tree 40ba61a4eddbaa4e87a81133b53edc678d67a26f
parent d4f162827c017939fe4faef4ef0062f58bcae10c

Fix performance regression in Conv2D on OpenCL

- Affect CL backend.
- For FP32 datatype, affect different platforms.

Resolves COMPMID-5479
Signed-off-by: Adnan AlSinan <adnan.alsinan@arm.com>
Change-Id: I705d718bc9b7def218034958f7ef86f2c2abe06d
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8064
Benchmark: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Comments-Addressed: 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 db193ce..8119fc8 100644
--- a/src/gpu/cl/operators/ClConv2d.cpp
+++ b/src/gpu/cl/operators/ClConv2d.cpp
@@ -281,7 +281,7 @@
                 }
 
                 // Direct convolution case
-                if(is_direct_valid && workload_gte_8192)
+                if(is_direct_valid)
                 {
                     if((gpu_target == arm_compute::GPUTarget::G71 || gpu_target == arm_compute::GPUTarget::G72 || get_arch_from_target(gpu_target) == arm_compute::GPUTarget::MIDGARD))
                     {
@@ -292,14 +292,14 @@
                     }
                     else if(gpu_target == arm_compute::GPUTarget::G76)
                     {
-                        if((is_large_kernel_sz && is_ifm_ge_16) || (is_ofm_lte_8 && is_ifm_ge_16))
+                        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)
+                        if( ((is_large_kernel_sz || is_m_one) && workload_gte_8192) || is_ofm_lte_8 )
                         {
                             return ConvolutionMethod::DIRECT;
                         }