aboutsummaryrefslogtreecommitdiff
path: root/src/gpu
diff options
context:
space:
mode:
authorGian Marco Iodice <gianmarco.iodice@arm.com>2023-04-28 10:40:07 +0100
committerGian Marco Iodice <gianmarco.iodice@arm.com>2023-05-02 15:53:28 +0000
commit60ab4e66ea3cb85042035fd1aafbfea666bb4ea7 (patch)
tree05818f7fafb0cf02d337b201756548152090436f /src/gpu
parentd7113e4af5b5497d3a3a62dc9cf6b147e2a024cd (diff)
downloadComputeLibrary-60ab4e66ea3cb85042035fd1aafbfea666bb4ea7.tar.gz
Fix export_to_cl_image issue in the fp16 GeMM implementation
- The issue affects Fp16 GeMM on Arm® Mali™-G78 - The issue was caused by a missing fallback implementation for the case when export_to_cl_image cannot be used - The new implementation fixes this issues and make the GeMM implementation for M=1 also faster (4-5% on various networks with fully connected at the end of the model) - This patch also enables the H0=0 case in the GeMM examples Resolves COMPMID-5812, COMPMID-5688, and COMPMID-6147 Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Change-Id: Ib7b355ae25337962598dd2ba21665b1a6b48686f Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/c/VisualCompute/ComputeLibrary/+/514664 Tested-by: bsgcomp <bsgcomp@arm.com> Reviewed-by: Viet-Hoa Do <viet-hoa.do@arm.com> Comments-Addressed: bsgcomp <bsgcomp@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/9526 Benchmark: Arm Jenkins <bsgcomp@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/gpu')
-rw-r--r--src/gpu/cl/kernels/gemm/reshaped_only_rhs/ClGemmDefaultConfigReshapedRhsOnlyValhall.cpp26
1 files changed, 12 insertions, 14 deletions
diff --git a/src/gpu/cl/kernels/gemm/reshaped_only_rhs/ClGemmDefaultConfigReshapedRhsOnlyValhall.cpp b/src/gpu/cl/kernels/gemm/reshaped_only_rhs/ClGemmDefaultConfigReshapedRhsOnlyValhall.cpp
index 76551b076a..d08bf84c72 100644
--- a/src/gpu/cl/kernels/gemm/reshaped_only_rhs/ClGemmDefaultConfigReshapedRhsOnlyValhall.cpp
+++ b/src/gpu/cl/kernels/gemm/reshaped_only_rhs/ClGemmDefaultConfigReshapedRhsOnlyValhall.cpp
@@ -562,21 +562,19 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedRhsOn
if(m == 1)
{
- if(r_mn <= 0.0045f)
+ const GeMMConfigsMatrix configs_mnkb_best =
{
- if(workload <= 278.7000f)
- {
- return configure_lhs_rhs_info(m, n, 1, 2, 16, 1, 8, 0, 0, 0, 1, 1);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 1, 4, 8, 1, 32, 0, 0, 1, 0, 0);
- }
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 1, 2, 16, 1, 8, 0, 0, 1, 0, 0);
- }
+ { 1, 8984, 640, 1, 1, 4, 2, 1, 0, 1, 0, 1, 1, 0 },
+ { 1, 420, 392, 1, 1, 2, 4, 1, 0, 1, 0, 1, 0, 0 },
+ { 1, 644, 5288, 1, 1, 2, 4, 1, 0, 1, 0, 1, 0, 0 },
+ { 1, 6512, 6404, 1, 1, 2, 2, 1, 0, 1, 0, 1, 1, 0 },
+ { 1, 5304, 640, 1, 1, 2, 2, 1, 0, 1, 0, 1, 0, 0 },
+ { 1, 1352, 1520, 1, 1, 2, 4, 1, 0, 1, 0, 1, 0, 0 },
+ { 1, 4096, 25088, 1, 1, 2, 4, 1, 0, 1, 0, 1, 0, 0 },
+ { 1, 732, 8988, 1, 1, 2, 4, 1, 0, 1, 0, 1, 0, 0 }
+ };
+
+ return find_lhs_rhs_info(configs_mnkb_best, m, n, k, b);
}
else
{