From 1f86449ef4be37c2e0e8e6c90d4bcb6307dca1fb Mon Sep 17 00:00:00 2001 From: ramelg01 Date: Thu, 7 Jul 2022 15:12:20 +0100 Subject: Fix Neoverse V1 heuristics for FP32 fast mode Resolves: COMPMID-5401 Signed-off-by: Ramy Elgammal Change-Id: I432b304483236efd392dfc47d541e6759c135104 Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/7934 Benchmark: Arm Jenkins Tested-by: Arm Jenkins Reviewed-by: Pablo Marquez Tello --- src/core/NEON/kernels/arm_gemm/gemm_interleaved.hpp | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'src/core/NEON/kernels/arm_gemm/gemm_interleaved.hpp') 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(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(args._nbatches) * args._nmulti * roundup(args._Msize, strategy::out_height()) * get_ktotal(args) * sizeof(Toi); - uint64_t merge_bytes = static_cast(args._nbatches) * args._nmulti * k_blocks * args._Msize * roundup(args._Nsize, strategy::out_width()) * sizeof(Tr); + uint64_t merge_bytes = static_cast(args._nbatches) * args._nmulti * k_blocks * args._Msize * roundup(args._Nsize, strategy::out_width()) * sizeof(Tr); float mac_cycles = static_cast(total_macs) / params.kernel_macs_cycle; float prepare_cycles = static_cast(prepare_bytes) / params.prepare_bytes_cycle; -- cgit v1.2.1