aboutsummaryrefslogtreecommitdiff
path: root/src/runtime/cpu/operators/internal/CpuGemmAssemblyDispatch.cpp
diff options
context:
space:
mode:
authorGeorgios Pinitas <georgios.pinitas@arm.com>2021-07-16 16:16:43 +0100
committerGeorgios Pinitas <georgios.pinitas@arm.com>2021-07-22 02:25:50 +0000
commit4ee8b1599dbaf7634d25607fa5ac96ba3dc6b0f2 (patch)
tree2f8362d33cdad4212f4b96995681c68184c759e1 /src/runtime/cpu/operators/internal/CpuGemmAssemblyDispatch.cpp
parent59fd7a722e5bc7e85309d6200bc37a772721a719 (diff)
downloadComputeLibrary-4ee8b1599dbaf7634d25607fa5ac96ba3dc6b0f2.tar.gz
Update GEMM assembly kernels
- Introduce Fp32 kernels with internal calculations in Bfloat16 when fast_mode is enabled - Improve kernel selection heuristics Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com> Change-Id: I68a9e7e862b6fd2721b46e0d7cc791091c4ab279 Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5965 Tested-by: Arm Jenkins <bsgcomp@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/runtime/cpu/operators/internal/CpuGemmAssemblyDispatch.cpp')
-rw-r--r--src/runtime/cpu/operators/internal/CpuGemmAssemblyDispatch.cpp10
1 files changed, 5 insertions, 5 deletions
diff --git a/src/runtime/cpu/operators/internal/CpuGemmAssemblyDispatch.cpp b/src/runtime/cpu/operators/internal/CpuGemmAssemblyDispatch.cpp
index 79ea1cb5a7..bbbd5ac458 100644
--- a/src/runtime/cpu/operators/internal/CpuGemmAssemblyDispatch.cpp
+++ b/src/runtime/cpu/operators/internal/CpuGemmAssemblyDispatch.cpp
@@ -542,7 +542,7 @@ void create_arm_gemm(std::unique_ptr<CpuGemmAssemblyDispatch::IFallback> &arm_ge
const CPUInfo &ci = NEScheduler::get().cpu_info();
unsigned int num_threads = NEScheduler::get().num_threads();
- arm_gemm::GemmArgs args(&ci, p.M, p.N, p.K, p.sections, p.batches, p.multis, p.indirect, activation, num_threads);
+ arm_gemm::GemmArgs args(&ci, p.M, p.N, p.K, p.sections, p.batches, p.multis, p.indirect, activation, num_threads, info.fast_mode);
// Create arm_gemm fallback
auto fallback = std::make_unique<Fallback<TypeInput, TypeOutput>>();
@@ -556,11 +556,11 @@ void create_arm_gemm_quant(std::unique_ptr<CpuGemmAssemblyDispatch::IFallback> &
arm_gemm::Activation activation, const AsmGemmInfo &info)
{
ARM_COMPUTE_UNUSED(activation);
- Params p = extract_parameters(a, b, d, info);
- const CPUInfo &ci = NEScheduler::get().cpu_info();
- unsigned int num_threads = NEScheduler::get().num_threads();
+ Params p = extract_parameters(a, b, d, info);
+ const CPUInfo &ci = NEScheduler::get().cpu_info();
+ const unsigned int num_threads = NEScheduler::get().num_threads();
- arm_gemm::GemmArgs args(&ci, p.M, p.N, p.K, p.sections, p.batches, p.multis, p.indirect, activation, num_threads);
+ arm_gemm::GemmArgs args(&ci, p.M, p.N, p.K, p.sections, p.batches, p.multis, p.indirect, activation, num_threads, info.fast_mode);
// Create arm_gemm fallback
auto fallback = std::make_unique<Fallback<TypeInput, TypeOutput, arm_gemm::Requantize32>>();