From d56a84ccd6082ae248e781c310d7aa24eebc19d0 Mon Sep 17 00:00:00 2001
From: SJW <simon.waters@amd.com>
Date: Tue, 15 Oct 2024 20:07:30 +0000
Subject: [PATCH] * moved LOCAL_LOAD cluster

---
 third_party/amd/lib/TritonAMDGPUTransforms/StreamPipelineV2.cpp | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/third_party/amd/lib/TritonAMDGPUTransforms/StreamPipelineV2.cpp b/third_party/amd/lib/TritonAMDGPUTransforms/StreamPipelineV2.cpp
index 0d655349c696..d3d1d5d3cf93 100644
--- a/third_party/amd/lib/TritonAMDGPUTransforms/StreamPipelineV2.cpp
+++ b/third_party/amd/lib/TritonAMDGPUTransforms/StreamPipelineV2.cpp
@@ -82,7 +82,7 @@ class StreamPipeliner {
     config.resize(5);
     config[SCHED_GLOBAL_LOAD] = {0, clusters[prefetch]};
     config[SCHED_LOCAL_STORE] = {lastStage - 1, clusters[prefetch ? 0 : 2]};
-    config[SCHED_LOCAL_LOAD] = {lastStage - prefetch, clusters[1]};
+    config[SCHED_LOCAL_LOAD] = {lastStage - prefetch, clusters[prefetch + 1]};
     config[SCHED_COMPUTE] = {lastStage, clusters[prefetch]};
     config[SCHED_TAIL] = {lastStage, clusters[3 + prefetch]};