sgemm_impl: Fix wrong next block_m logic for DMA

2024-10-29 22:35:56 -07:00
parent 21b6655c10
commit c001618fb9
1 changed files with 2 additions and 1 deletions
--- a/tests/regression/sgemm_tcore/sgemm_impl.hpp
+++ b/tests/regression/sgemm_tcore/sgemm_impl.hpp
@@ -1237,7 +1237,8 @@ inline void thread_block_gemm(const T *A, const T *B, float *C,
                  : block_n;
          const uint32_t next_block_m =
              (next_block_n == 0)
-                  ? ((block_m == block_m_end) ? 0 : block_n + 1)
+                  ? (((block_m + 1) == block_m_end) ? block_m_start /*unused*/
+                                                    : block_m + 1)
                  : block_m;

          asm volatile("next_index_end_%=:" ::);