Fix typos so to call correct vnni functions under vnni condition

Signed-off-by: liqunfu <[email protected]>
microsoft · Aug 5, 2024 · fb68b9d · fb68b9d
1 parent 88c811b
commit fb68b9d
Show file tree

Hide file tree

Showing 2 changed files with 3 additions and 3 deletions.
diff --git a/onnxruntime/core/mlas/lib/sqnbitgemm_kernel_avx512_int8_blklen16.h b/onnxruntime/core/mlas/lib/sqnbitgemm_kernel_avx512_int8_blklen16.h
@@ -679,9 +679,9 @@ Q4Int8GemmR1xC1BlkLen16Avx512(
                 const __m512i av_01_epi8 = _mm512_loadu_si512((const __m512i*)(QuantAPtr + 64));
 
                 if constexpr (vnni) {
-                    accumulate_blklen16_r1c1blk8_avx512(av_00_epi8, av_01_epi8, QuantBDataPtr, QuantAScalePtr, QuantBScalePtr, acc0);
-                } else {
                     accumulate_blklen16_r1c1blk8_avx512vnni(av_00_epi8, av_01_epi8, QuantBDataPtr, QuantAScalePtr, QuantBScalePtr, acc0);
+                } else {
+                    accumulate_blklen16_r1c1blk8_avx512(av_00_epi8, av_01_epi8, QuantBDataPtr, QuantAScalePtr, QuantBScalePtr, acc0);
                 }
 
                 QuantAPtr += BlkLen16 * PerAccuBlk8;

diff --git a/onnxruntime/core/mlas/lib/sqnbitgemm_kernel_avx512_int8_blklen32.h b/onnxruntime/core/mlas/lib/sqnbitgemm_kernel_avx512_int8_blklen32.h
@@ -721,7 +721,7 @@ Q4Int8GemmR1xC1BlkLen32Avx512(
                     accumulate_blklen32_r1c1blk4_avx512vnni(av_00_epi8, av_01_epi8, QuantBDataPtr, QuantAScalePtr, QuantBScalePtr, acc0);
                 }
                 else {
-                    accumulate_blklen32_r1c1blk4_avx512vnni(av_00_epi8, av_01_epi8, QuantBDataPtr, QuantAScalePtr, QuantBScalePtr, acc0);
+                    accumulate_blklen32_r1c1blk4_avx512(av_00_epi8, av_01_epi8, QuantBDataPtr, QuantAScalePtr, QuantBScalePtr, acc0);
                 }
 
                 QuantAPtr += BlkLen32 * PerAccuBlk4;