Fix *tune_gemm* directory in code and documentation

ROCm · Jul 25, 2024 · 531ffb9 · 531ffb9
1 parent 7f1aff6
commit 531ffb9
Show file tree

Hide file tree

Showing 2 changed files with 2 additions and 2 deletions.
diff --git a/python/perf-kernels/tune_gemm/README.md b/python/perf-kernels/tune_gemm/README.md
@@ -2,7 +2,7 @@
 
 ## matmul kernel
 
-The matmul kernel implementation can be found as [matmul_kernel.py](https://github.com/ROCm/triton/blob/triton-mlir/scripts/amd/gemm/matmul_kernel.py), which includes the following features:
+The matmul kernel implementation can be found as [matmul_kernel.py](https://github.com/ROCm/triton/blob/main_perf/python/perf-kernels/tune_gemm/matmul_kernel.py), which includes the following features:
 - grouping order of workgroup id, which is controlled by `GROUP_SIZE_M`, that
 implements L2 cache optimization introduced in the [tutorial](https://triton-lang.org/main/getting-started/tutorials/03-matrix-multiplication.html#l2-cache-optimizations).
 - split-k algorithm, which is controlled by `SPLIT_K`.

diff --git a/python/perf-kernels/tune_gemm/matmul.py b/python/perf-kernels/tune_gemm/matmul.py
@@ -351,7 +351,7 @@ def main():
             driver = 'rocprof_gemm.py'
             TRITON_DIR = os.getenv('TRITON_DIR')
             if TRITON_DIR is not None:
-                driver = os.path.join(TRITON_DIR, 'scripts/amd/gemm', driver)
+                driver = os.path.join(TRITON_DIR, 'python/perf-kernels/tune_gemm', driver)
             run_cmd = f'python {driver} -m {m} -n {n} -k {k} \
                         -block_m {block_m} -block_n {block_n} -block_k {block_k} \
                         -group_m {group_m} -split_k {split_k} -num_warps {num_warps} \