From: Aman Gupta <redacted>
Date: Tue, 14 Oct 2025 14:48:08 +0000 (+0800)
Subject: CUDA + openCL: fix bug in accessing rms_norm->src while doing fusion (llama/16577)
X-Git-Tag: upstream/1.8.2~4
X-Git-Url: https://git.djapps.eu/?a=commitdiff_plain;h=2eb9119754efe2a7a7ece560fd87d0b42044f262;p=pkg%2Fggml%2Fsources%2Fwhisper.cpp

CUDA + openCL: fix bug in accessing rms_norm->src while doing fusion (llama/16577)
---

diff --git a/ggml/src/ggml-cuda/ggml-cuda.cu b/ggml/src/ggml-cuda/ggml-cuda.cu
index 83b82c1a..da312992 100644
--- a/ggml/src/ggml-cuda/ggml-cuda.cu
+++ b/ggml/src/ggml-cuda/ggml-cuda.cu
@@ -2876,7 +2876,7 @@ static bool ggml_cuda_can_fuse(const struct ggml_cgraph * cgraph, int node_idx,
         }
 
         //if rms norm is the B operand, then we don't handle broadcast
-        if (rms_norm == mul->src[1] && !ggml_are_same_shape(mul->src[0], rms_norm->src[1])) {
+        if (rms_norm == mul->src[1] && !ggml_are_same_shape(mul->src[0], rms_norm)) {
             return false;
         }
 
diff --git a/ggml/src/ggml-opencl/ggml-opencl.cpp b/ggml/src/ggml-opencl/ggml-opencl.cpp
index d2759069..0693d38d 100644
--- a/ggml/src/ggml-opencl/ggml-opencl.cpp
+++ b/ggml/src/ggml-opencl/ggml-opencl.cpp
@@ -2686,7 +2686,7 @@ static bool ggml_opencl_can_fuse(const struct ggml_cgraph * cgraph, int node_idx
 
         // if rms_norm is the B operand, then we don't handle broadcast
         if (rms_norm == mul->src[1] &&
-            !ggml_are_same_shape(mul->src[0], rms_norm->src[1])) {
+            !ggml_are_same_shape(mul->src[0], rms_norm)) {
             return false;
         }