commit | 3357d9de85270a83a0cc8dff8a15e0be43a4603b | [log] [tgz] |
---|---|---|
author | Marat Dukhan <maratek@google.com> | Fri Jun 25 15:22:30 2021 -0700 |
committer | XNNPACK Team <xnnpack-github-robot@google.com> | Fri Jun 25 15:23:03 2021 -0700 |
tree | 82082d69ac4a6325e499b532f6edd6a73114a549 | |
parent | e742d2a5b5771197b91999e4854e37277a37cb29 [diff] [blame] |
Minor optimizations in NEON QS8 GEMM/IGEMM microkernels PiperOrigin-RevId: 381558969
diff --git a/src/qs8-gemm/gen/2x8c2-minmax-gemmlowp-neon-mull-padal-dup.c b/src/qs8-gemm/gen/2x8c2-minmax-gemmlowp-neon-mull-padal-dup.c index 9741b73..a290314 100644 --- a/src/qs8-gemm/gen/2x8c2-minmax-gemmlowp-neon-mull-padal-dup.c +++ b/src/qs8-gemm/gen/2x8c2-minmax-gemmlowp-neon-mull-padal-dup.c
@@ -36,7 +36,7 @@ assert(w != NULL); assert(c != NULL); - kc = round_up_po2(kc, 2); + kc = round_up_po2(kc, 2 * sizeof(int8_t)); const int8_t* a0 = a; int8_t* c0 = c; const int8_t* a1 = (const int8_t*) ((uintptr_t) a0 + a_stride);