From: SXX <redacted>
Date: Sat, 9 Nov 2024 07:35:46 +0000 (+0800)
Subject: ggml: fix zero division in ‘dne’ calculation in CUDA COUNT_EQUAL operator when ‘ne... 
X-Git-Tag: upstream/0.0.1642~217
X-Git-Url: https://git.djapps.eu/?a=commitdiff_plain;h=77e2744b52d97ea2067d6be0ff3f380d78cd4687;p=pkg%2Fggml%2Fsources%2Fggml

ggml: fix zero division in ‘dne’ calculation in CUDA COUNT_EQUAL operator when ‘ne’ is small (#10213)
---

diff --git a/src/ggml-cuda/count-equal.cu b/src/ggml-cuda/count-equal.cu
index ffb053b1..08898115 100644
--- a/src/ggml-cuda/count-equal.cu
+++ b/src/ggml-cuda/count-equal.cu
@@ -44,7 +44,7 @@ void ggml_cuda_count_equal(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
 
     const int64_t ne = ggml_nelements(src0);
     GGML_ASSERT(ne < (1 << 30) && "atomicAdd implementation only supports int");
-    const int64_t dne = GGML_PAD(ne / (4*nsm), CUDA_COUNT_EQUAL_CHUNK_SIZE);
+    const int64_t dne = GGML_PAD((ne + 4*nsm - 1) / (4*nsm), CUDA_COUNT_EQUAL_CHUNK_SIZE);
 
     CUDA_CHECK(cudaMemsetAsync(dst_d, 0, ggml_nbytes(dst), stream));