aboutsummaryrefslogtreecommitdiff
path: root/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_u8_8x12.hpp
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/core/NEON/kernels/arm_gemm/kernels/a64_gemm_u8_8x12.hpp
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/core/NEON/kernels/arm_gemm/kernels/a64_gemm_u8_8x12.hpp')
-rw-r--r--src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_u8_8x12.hpp36
1 files changed, 30 insertions, 6 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_u8_8x12.hpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_u8_8x12.hpp
index 00ed5d03bf..0329f57615 100644
--- a/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_u8_8x12.hpp
+++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_u8_8x12.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2017-2021 Arm Limited.
+ * Copyright (c) 2017-2018,2021 Arm Limited.
*
* SPDX-License-Identifier: MIT
*
@@ -69,14 +69,38 @@ public:
StdTransformsFixed<operand_type, result_type, 8, 12, 4> transforms = {};
StdTransformsFixed<operand_type, result_type, 8, 12, 4, true> transforms_quantized = {};
+ template<typename T>
static PerformanceParameters get_performance_parameters(const CPUInfo *ci) {
- switch (ci->get_cpu_model()) {
- case CPUModel::A55r1:
- return { 15.361, 0.9341, 0.1636 };
+ if (std::is_same<T, int8_t>::value) {
+ switch (ci->get_cpu_model()) {
+ case CPUModel::A510:
+ return { 19.73, 3.38, 0.27 };
- default:
- return { 29.0698, 3.9793, 0.4003 };
+ case CPUModel::A55r1:
+ return { 15.361, 0.9341, 0.1636 };
+
+ case CPUModel::V1:
+ return { 62.40, 4.71, 0.67 };
+
+ default:
+ return { 29.0698, 3.9793, 0.4003 };
+ }
}
+
+ if (std::is_same<T, int32_t>::value) {
+ switch (ci->get_cpu_model()) {
+ case CPUModel::A510:
+ return { 19.73, 3.38, 3.70 };
+
+ case CPUModel::V1:
+ return { 61.58, 4.78, 10.83 };
+
+ default:
+ return { 31.82, 3.51, 8.03 };
+ }
+ }
+
+ return { 0.0 };
}
kern_type kernel = a64_gemm_u8_8x12;