Merge pull request #4920 from martin-frbg/issue4917

Fix potential inaccuracy in multithreaded level3 related to SWITCH_RATIO
OpenMathLib · Oct 7, 2024 · bc0691a · bc0691a
2 parents 0dfe42d + 8a1710d
commit bc0691a
Showing 1 changed file with 1 addition and 1 deletion.
diff --git a/driver/level3/level3_thread.c b/driver/level3/level3_thread.c
@@ -742,7 +742,7 @@ static int gemm_driver(blas_arg_t *args, BLASLONG *range_m, BLASLONG
     num_parts  = 0;
     while (n > 0){
       width = blas_quickdivide(n + nthreads - num_parts - 1, nthreads - num_parts);
-      if (width < switch_ratio) {
+      if (width < switch_ratio && width > 1) {
         width = switch_ratio;
       }
       width = round_up(n, width, GEMM_PREFERED_SIZE);