sbgemm: cooperlake: fix bug in m64n12
authorWangyang Guo <wangyang.guo@intel.com>
Tue, 17 Aug 2021 13:13:29 +0000 (21:13 +0800)
committerWangyang Guo <wangyang.guo@intel.com>
Tue, 7 Sep 2021 13:30:46 +0000 (21:30 +0800)
kernel/x86_64/sbgemm_kernel_16x4_cooperlake.c

index d604235..c257a3f 100644 (file)
@@ -306,9 +306,8 @@ int CNAME (BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha, IFLOAT * A, IFLOAT *
                        STORE_4X(0, 0, 1); STORE_4X(1, 0, 1);
                        STORE_4X(0, 0, 2); STORE_4X(1, 0, 2);
                        ptr_c += 16 * 2;
-               }
-               if (m > 31) {
                        ptr_a0 = ptr_a1;
+                       ptr_a1 = ptr_a0 + 16 * k;
                }
                for (; m_count > 15; m_count -= 16) {
                        ptr_b0 = ptr_b00;