* moved LOCAL_LOAD cluster

triton-lang · Oct 15, 2024 · d56a84c · d56a84c
1 parent efdbbe3
commit d56a84c
Showing 1 changed file with 1 addition and 1 deletion.
diff --git 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]};