summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorGeorgi Gerganov <ggerganov@gmail.com>2023-10-09 14:28:27 +0300
committerGitHub <noreply@github.com>2023-10-09 14:28:27 +0300
commitdcc09d25961c5d0626bc148e558ee841141748f7 (patch)
tree0961390c68ea32c1502dc6e96d6ad27917ead3c8
parentdb3abcc114d5d1790ba814aa1a80ac673d4ccc3e (diff)
metal : do not use mul_mm kernels when ne00 < 64 (#3542)
-rw-r--r--ggml-metal.m2
1 files changed, 1 insertions, 1 deletions
diff --git a/ggml-metal.m b/ggml-metal.m
index 29cb3c92..7d67db90 100644
--- a/ggml-metal.m
+++ b/ggml-metal.m
@@ -1040,7 +1040,7 @@ void ggml_metal_graph_compute(
!ggml_is_transposed(src0) &&
!ggml_is_transposed(src1) &&
src1t == GGML_TYPE_F32 &&
- ne00 % 32 == 0 &&
+ ne00 % 32 == 0 && ne00 >= 64 &&
ne11 > ne11_mm_min) {
//printf("matrix: ne00 = %6d, ne01 = %6d, ne02 = %6d, ne11 = %6d, ne12 = %6d\n", ne00, ne01, ne02, ne11, ne12);
switch (src0->type) {