Skip to content

Commit

Permalink
clang format correction
Browse files Browse the repository at this point in the history
  • Loading branch information
Shreyas-fuj committed Sep 20, 2024
1 parent b02490d commit 3f28386
Show file tree
Hide file tree
Showing 2 changed files with 5 additions and 5 deletions.
2 changes: 1 addition & 1 deletion src/cpu/aarch64/brgemm/jit_brgemm_kernel.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -770,7 +770,7 @@ void jit_brgemm_kernel_t::zero_accumulators(int bd_block2, bool is_bdb_tail,
for_(int bd = 0; bd < bd_block; bd++)
for (int ld = 0; ld < ld_block2; ld++) {
auto zmm = accm(ld_block2, bd, ld);
// This part is moved here from apply_alpha_beta function so that fadd instruction can be avoided.
// This part is moved here from apply_alpha_beta function so that fadd instruction can be avoided.
// This is also required only when K is blocked.
if (need_to_apply_beta && brg.K != brg.reduce_dim) {
const bool is_tail = is_ld_tail && ld + 1 == ld_block2;
Expand Down
8 changes: 4 additions & 4 deletions src/cpu/aarch64/matmul/brgemm_matmul_utils.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -629,10 +629,10 @@ float compute_blocking_heuristic_sve_256(brgemm_matmul_conf_t &bgmmc,
}
}
min_m_blk = nstl::min(matmul.M, best_m_blk);
// Here min_m_blk is set based on M value and no.of threads. Decreasing m_blk size will
// increase no.of m blocks which might make better utilisation of threads. But it is found
// that m_blk being a factor of M is more important than max thread utilisation.Therefore
// in scoring that has been given more weightage(0.7). This was experimentally verified to
// Here min_m_blk is set based on M value and no.of threads. Decreasing m_blk size will
// increase no.of m blocks which might make better utilisation of threads. But it is found
// that m_blk being a factor of M is more important than max thread utilisation.Therefore
// in scoring that has been given more weightage(0.7). This was experimentally verified to
// be the best hueristics with multiple shapes.

bool low_spatial_work = matmul.M <= 40;
Expand Down

0 comments on commit 3f28386

Please sign in to comment.