Improve performance of shape 1024x1024x1024 out of box (#2839)

Close #2822 After: 90% -> 103%
intel · Nov 28, 2024 · 31fe770 · 31fe770
1 parent 4d3a94d
commit 31fe770
Showing 1 changed file with 1 addition and 1 deletion.
diff --git a/benchmarks/triton_kernels_benchmark/gemm_benchmark.py b/benchmarks/triton_kernels_benchmark/gemm_benchmark.py
@@ -32,7 +32,7 @@
     ] + [
         triton.Config(
             {'BLOCK_SIZE_M': 256, 'BLOCK_SIZE_N': 128, 'BLOCK_SIZE_K': 32, 'GROUP_SIZE_M': 4, 'grf_mode': 'large'},
-            num_stages=s, num_warps=32) for s in [2, 3]
+            num_stages=s, num_warps=32) for s in [2, 3, 4]
     ] + [
         triton.Config(
             {'BLOCK_SIZE_M': 64, 'BLOCK_SIZE_N': 128, 'BLOCK_SIZE_K': 32, 'GROUP_SIZE_M': 4, 'grf_mode': 'large'},