From f535d8d421c079ebeffd924e7154685fe1eaed58 Mon Sep 17 00:00:00 2001
From: Kawrakow <iwankawrakow@gmail.com>
Date: Fri, 17 Jul 2026 15:05:15 +0000
Subject: [PATCH] Fix race in indexer topk on CUDA

---
 ggml/src/ggml-cuda/indexer_topk.cu | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/ggml/src/ggml-cuda/indexer_topk.cu b/ggml/src/ggml-cuda/indexer_topk.cu
index 14b12386b..a110a7ace 100644
--- a/ggml/src/ggml-cuda/indexer_topk.cu
+++ b/ggml/src/ggml-cuda/indexer_topk.cu
@@ -240,7 +240,9 @@ static __global__ void k_indexer_mask(int ne0, int ne1, int ne2, int ntopk, int
 
     if (i1 < ne11) {
         for (int j = threadIdx.x; j < ne0;   j += blockDim.x) d[j] = inf;
+        __syncthreads();
         for (int j = threadIdx.x; j < ntopk; j += blockDim.x) d[i[j]] = zero;
+        __syncthreads();
         for (int j = threadIdx.x; j < ne0;   j += blockDim.x) d[j] += m[j];
     } else {
         for (int j = threadIdx.x; j < ne0;   j += blockDim.x) d[j] = m[j];