Fix typos so to call correct vnni functions under vnni condition (#21625)

liqunfu · rohan11235813 · commit 099ba67ac9b3 · 2025-09-15T10:40:27.000-07:00
### Description
Fix 2 typos in mlas avx 4bit gemm implementation to call correct vnni
functions under vnni condition



### Motivation and Context
needed for 1.19.0 release

Signed-off-by: liqunfu &lt;liqun.fu@microsoft.com&gt;
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;

Original file line number	Diff line number	Diff line change
`@@ -721,7 +721,7 @@ Q4Int8GemmR1xC1BlkLen32Avx512(`
`721`	`721`	`accumulate_blklen32_r1c1blk4_avx512vnni(av_00_epi8, av_01_epi8, QuantBDataPtr, QuantAScalePtr, QuantBScalePtr, acc0);`
`722`	`722`	`}`
`723`	`723`	`else {`
`724`		`- accumulate_blklen32_r1c1blk4_avx512vnni(av_00_epi8, av_01_epi8, QuantBDataPtr, QuantAScalePtr, QuantBScalePtr, acc0);`
	`724`	`+ accumulate_blklen32_r1c1blk4_avx512(av_00_epi8, av_01_epi8, QuantBDataPtr, QuantAScalePtr, QuantBScalePtr, acc0);`
`725`	`725`	`}`
`726`	`726`
`727`	`727`	`QuantAPtr += BlkLen32 * PerAccuBlk4;`