Commit 79e2e74eb for llama.cpp
commit 79e2e74eb11022c1ba2e438df7f0ca2d4c10f8b6
Author: Dante <144588016+KafKafrnZ@users.noreply.github.com>
Date: Fri Oct 9 23:29:50 2026 +0530
CUDA: fix round issue, under MSVC the CPU and GPU agree (#30229)
diff --git a/ggml/src/ggml-cuda/unary.cu b/ggml/src/ggml-cuda/unary.cu
index 436fdfffb..af839acd7 100644
--- a/ggml/src/ggml-cuda/unary.cu
+++ b/ggml/src/ggml-cuda/unary.cu
@@ -107,7 +107,7 @@ static __device__ __forceinline__ float op_ceil(float x) {
}
static __device__ __forceinline__ float op_round(float x) {
- return round(x);
+ return roundf(x);
}
static __device__ __forceinline__ float op_trunc(float x) {