diff options
| author | ramelg01 <ramy.elgammal@arm.com> | 2022-07-07 15:12:20 +0100 |
|---|---|---|
| committer | Pablo Marquez Tello <pablo.tello@arm.com> | 2022-07-18 05:29:41 +0000 |
| commit | 1f86449ef4be37c2e0e8e6c90d4bcb6307dca1fb (patch) | |
| tree | ea63e45bcfe2b556580b9a515fe16762c4b14722 /src/core/NEON/kernels/arm_gemm/gemm_interleaved.hpp | |
| parent | a1f7851e2f776610019db8725c2963c36b0c85eb (diff) | |
| download | ComputeLibrary-1f86449ef4be37c2e0e8e6c90d4bcb6307dca1fb.tar.gz | |
Fix Neoverse V1 heuristics for FP32 fast mode
Resolves: COMPMID-5401
Signed-off-by: Ramy Elgammal <ramy.elgammal@arm.com>
Change-Id: I432b304483236efd392dfc47d541e6759c135104
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/7934
Benchmark: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Pablo Marquez Tello <pablo.tello@arm.com>
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/gemm_interleaved.hpp')
| -rw-r--r-- | src/core/NEON/kernels/arm_gemm/gemm_interleaved.hpp | 2 |
1 files changed, 1 insertions, 1 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/gemm_interleaved.hpp b/src/core/NEON/kernels/arm_gemm/gemm_interleaved.hpp index 4ad54426e9..6e14a68438 100644 --- a/src/core/NEON/kernels/arm_gemm/gemm_interleaved.hpp +++ b/src/core/NEON/kernels/arm_gemm/gemm_interleaved.hpp @@ -1128,7 +1128,7 @@ public: uint64_t total_macs = static_cast<uint64_t>(args._nbatches) * args._nmulti * roundup(args._Msize, strategy::out_height()) * roundup(args._Nsize, strategy::out_width()) * get_ktotal(args); uint64_t prepare_bytes = static_cast<uint64_t>(args._nbatches) * args._nmulti * roundup(args._Msize, strategy::out_height()) * get_ktotal(args) * sizeof(Toi); - uint64_t merge_bytes = static_cast<uint16_t>(args._nbatches) * args._nmulti * k_blocks * args._Msize * roundup(args._Nsize, strategy::out_width()) * sizeof(Tr); + uint64_t merge_bytes = static_cast<uint64_t>(args._nbatches) * args._nmulti * k_blocks * args._Msize * roundup(args._Nsize, strategy::out_width()) * sizeof(Tr); float mac_cycles = static_cast<float>(total_macs) / params.kernel_macs_cycle; float prepare_cycles = static_cast<float>(prepare_bytes) / params.prepare_bytes_cycle; |