skip autotuner test on CPU for now

Author: mark14wu

.github/workflows/python-app.yml

@@ -48,12 +48,9 @@ jobs:
         pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu121
         pip uninstall pytorch-triton -y
 
-    - name: Clone Triton and Install
+    - name: Install Triton
       run: |
-        git clone https://github.com/openai/triton.git
-        cd triton
-        pip install -r python/requirements.txt
-        pip install -e .
+        pip install triton==3.1.0
 
     - name: Install Triton-Viz
       run: |

tests/test_autotune_add.py

@@ -1,3 +1,4 @@
+import pytest
 import torch
 import triton
 import triton.language as tl
@@ -45,6 +46,10 @@ def add_kernel_no_mask(x_ptr, y_ptr, out_ptr, n_elements, BLOCK_SIZE: tl.constex
     tl.store(out_ptr + offsets, x_val + y_val)
 
 
+@pytest.mark.skipif(
+    not torch.backends.cuda.is_built(),
+    reason="This test requires a CUDA-enabled environment.",
+)
 def test_autotune_add_inrange():
     """
     This test uses n_elements = 128, matching the size of the input tensors.
@@ -61,6 +66,10 @@ def test_autotune_add_inrange():
     print("test_autotune_add_inrange() passed: No out-of-bound access.")
 
 
+@pytest.mark.skipif(
+    not torch.backends.cuda.is_built(),
+    reason="This test requires a CUDA-enabled environment.",
+)
 def test_autotune_add_out_of_bound():
     """
     This test deliberately sets n_elements = 256, exceeding the actual buffer size (128).