diff options
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/gemm_interleaved.hpp')
| -rw-r--r-- | src/core/NEON/kernels/arm_gemm/gemm_interleaved.hpp | 4 |
1 files changed, 2 insertions, 2 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/gemm_interleaved.hpp b/src/core/NEON/kernels/arm_gemm/gemm_interleaved.hpp index 92c1086a5f..7f870b83d7 100644 --- a/src/core/NEON/kernels/arm_gemm/gemm_interleaved.hpp +++ b/src/core/NEON/kernels/arm_gemm/gemm_interleaved.hpp @@ -1022,8 +1022,8 @@ public: static uint64_t estimate_cycles(const GemmArgs &args, const PerformanceParameters ¶ms) { unsigned int k_blocks = iceildiv(args._Ksize, get_k_block_size(args)); - uint64_t total_macs = static_cast<uint64_t>(args._nbatches) * args._nmulti * roundup(args._Msize, strategy::out_height()) * roundup(args._Nsize, strategy::out_width()) * roundup(args._Ksize, strategy::k_unroll()); - uint64_t prepare_bytes = static_cast<uint64_t>(args._nbatches) * args._nmulti * roundup(args._Msize, strategy::out_height()) * roundup(args._Ksize, strategy::k_unroll()) * sizeof(Toi); + 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 * roundup(args._Msize, strategy::out_height()) * roundup(args._Nsize, strategy::out_width()) * sizeof(Tr); float mac_cycles = static_cast<float>(total_macs) / params.kernel_macs_cycle; |