Skip to content

Commit

Permalink
cpu: x64: matmul_reorder: issue f16 instructions for f16 src_dt only
Browse files Browse the repository at this point in the history
  • Loading branch information
dzarukin committed Oct 25, 2024
1 parent a9864fb commit 442b5e9
Showing 1 changed file with 2 additions and 1 deletion.
3 changes: 2 additions & 1 deletion src/cpu/x64/matmul/brgemm_matmul_copy_utils.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -944,7 +944,8 @@ void jit_brgemm_matmul_copy_a_transposed_impl_t<Xbyak::Zmm>::transpose_f32(
? ptr[i % 2 == 0 ? reg_aux_src0 : reg_aux_src1]
: EVEX_compress_addr(src, i * src_stride);
if (i < nrows)
if (conf_->isa == avx512_core_fp16)
if (conf_->isa == avx512_core_fp16
&& conf_->src_dt == data_type::f16)
vcvtph2psx(src_zmm(i) | kTail | T_z, addr);
else
vmovups(src_zmm(i) | kTail | T_z, addr);
Expand Down

0 comments on commit 442b5e9

Please sign in to comment.