add 3.12 inductor CI tests

ghstack-source-id: 54279665250323134ec3ba7f724f89a34833cce6
Pull Request resolved: #126218

.ci/docker/build.sh

@@ -149,6 +149,21 @@ case "$image" in
     TRITON=yes
     INDUCTOR_BENCHMARKS=yes
     ;;
+  pytorch-linux-focal-cuda12.1-cudnn8-py3.12-gcc9-inductor-benchmarks)
+    CUDA_VERSION=12.1.1
+    CUDNN_VERSION=8
+    ANACONDA_PYTHON_VERSION=3.12
+    GCC_VERSION=9
+    PROTOBUF=yes
+    DB=yes
+    VISION=yes
+    KATEX=yes
+    UCX_COMMIT=${_UCX_COMMIT}
+    UCC_COMMIT=${_UCC_COMMIT}
+    CONDA_CMAKE=yes
+    TRITON=yes
+    INDUCTOR_BENCHMARKS=yes
+    ;;
   pytorch-linux-focal-cuda11.8-cudnn8-py3-gcc9)
     CUDA_VERSION=11.8.0
     CUDNN_VERSION=8

.github/workflows/docker-builds.yml

@@ -42,6 +42,7 @@ jobs:
           pytorch-linux-focal-cuda12.4-cudnn8-py3-gcc9-inductor-benchmarks,
           pytorch-linux-focal-cuda12.1-cudnn8-py3-gcc9,
           pytorch-linux-focal-cuda12.1-cudnn8-py3-gcc9-inductor-benchmarks,
+          pytorch-linux-focal-cuda12.1-cudnn8-py3.12-gcc9-inductor-benchmarks,
           pytorch-linux-focal-cuda11.8-cudnn8-py3-gcc9,
           pytorch-linux-focal-py3.8-clang10,
           pytorch-linux-focal-py3.11-clang10,

.github/workflows/inductor.yml

@@ -107,6 +107,27 @@ jobs:
     secrets:
       HUGGING_FACE_HUB_TOKEN: ${{ secrets.HUGGING_FACE_HUB_TOKEN }}
 
+  linux-focal-cuda12_1-py3_12-gcc9-inductor-build:
+    name: cuda12.1-py3.12-gcc9-sm86
+    uses: ./.github/workflows/_linux-build.yml
+    with:
+      build-environment: linux-focal-cuda12.1-py3.12-gcc9-sm86
+      docker-image-name: pytorch-linux-focal-cuda12.1-cudnn8-py3.12-gcc9-inductor-benchmarks
+      cuda-arch-list: '8.6'
+      test-matrix: |
+        { include: [
+          { config: "inductor", shard: 1, num_shards: 1, runner: "linux.g5.4xlarge.nvidia.gpu" },
+        ]}
+
+  linux-focal-cuda12_1-py3_12-gcc9-inductor-test:
+    name: cuda12.1-py3.12-gcc9-sm86
+    uses: ./.github/workflows/_linux-test.yml
+    needs: linux-focal-cuda12_1-py3_12-gcc9-inductor-build
+    with:
+      build-environment: linux-focal-cuda12.1-py3.12-gcc9-sm86
+      docker-image: ${{ needs.linux-focal-cuda12_1-py3_12-gcc9-inductor-build.outputs.docker-image }}
+      test-matrix: ${{ needs.linux-focal-cuda12_1-py3_12-gcc9-inductor-build.outputs.test-matrix }}
+
   linux-jammy-cpu-py3_8-gcc11-inductor-build:
     name: linux-jammy-cpu-py3.8-gcc11-inductor
     uses: ./.github/workflows/_linux-build.yml

test/inductor/test_torchinductor.py

@@ -36,6 +36,8 @@
     expectedFailureCodegenDynamic,
     rand_strided,
     same,
+    skipIfPy312,
+    xfailIfPy312,
 )
 from torch._inductor.codegen.common import DataTypePropagation, OptimizationContext
 from torch._inductor.fx_passes import pad_mm
@@ -2743,6 +2745,7 @@ def fn(a, b):
             check_lowp=False,
         )
 
+    @skipIfPy312  # segfaults
     @config.patch(force_mixed_mm=True)
     def test_mixed_mm(self):
         def fn(a, b):
@@ -2757,6 +2760,7 @@ def fn(a, b):
             check_lowp=True,
         )
 
+    @skipIfPy312  # segfaults
     @config.patch(force_mixed_mm=True)
     def test_mixed_mm2(self):
         def fn(a, b, scale, bias):
@@ -9448,6 +9452,7 @@ def fn(inp, offsets):
 
         self.common(fn, (inp, offsets), check_lowp=False)
 
+    @xfailIfPy312
     @requires_gpu()
     @config.patch(assume_aligned_inputs=False)
     def test_config_option_dont_assume_alignment(self):

torch/_dynamo/testing.py

@@ -349,6 +349,12 @@ def xfailIfPy312(fn):
     return fn
 
 
+def skipIfPy312(fn):
+    if sys.version_info >= (3, 12):
+        return unittest.skip(fn)
+    return fn
+
+
 # Controls tests generated in test/inductor/test_torchinductor_dynamic_shapes.py
 # and test/dynamo/test_dynamic_shapes.py
 def expectedFailureDynamic(fn):