Update on "Make CI error on inductor fallback when decomp is available"

Fixes #99446 

Remove the warning, as that annoyed end-users who don't know what to do about it.

Instead, try to hold the line by preventing any decomp from being added without making
the corresponding change to inductor's fallbacks.

Note: we probably still need to better document how to update inductor's decomps,
for now it's pretty much "go ask the inductor team for advice"

cc soumith voznesenskym penguinwu anijain2305 EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng Xia-Weiwen wenzhe-nrv jiayisunx peterbell10 desertfire

[ghstack-poisoned]

.github/scripts/generate_binary_build_matrix.py

@@ -189,6 +189,7 @@ def generate_wheels_matrix(
     os: str,
     arches: Optional[List[str]] = None,
     python_versions: Optional[List[str]] = None,
+    gen_special_an_non_special_wheel: bool = True,
 ) -> List[Dict[str, str]]:
     package_type = "wheel"
     if os == "linux":
@@ -249,6 +250,8 @@ def generate_wheels_matrix(
                         ),
                     }
                 )
+                if not gen_special_an_non_special_wheel:
+                    continue
 
             ret.append(
                 {

.github/scripts/generate_ci_workflows.py

@@ -146,15 +146,10 @@ class OperatingSystem:
         os=OperatingSystem.LINUX,
         package_type="manywheel",
         build_configs=generate_binary_build_matrix.generate_wheels_matrix(
-            OperatingSystem.LINUX, arches=["11.8"], python_versions=["3.8"]
-        ),
-        branches="main",
-    ),
-    BinaryBuildWorkflow(
-        os=OperatingSystem.LINUX,
-        package_type="manywheel",
-        build_configs=generate_binary_build_matrix.generate_wheels_matrix(
-            OperatingSystem.LINUX, arches=["11.7"], python_versions=["3.8"]
+            OperatingSystem.LINUX,
+            arches=["11.7", "12.1"],
+            python_versions=["3.8"],
+            gen_special_an_non_special_wheel=False,
         ),
         branches="main",
     ),

.lintrunner.toml

@@ -1,4 +1,4 @@
-merge_base_with = "origin/master"
+merge_base_with = "origin/main"
 
 [[linter]]
 code = 'FLAKE8'
@@ -174,6 +174,8 @@ include_patterns = [
     'torch/_dynamo/optimizations/__init__.py',
     'torch/_dynamo/optimizations/backends.py',
     'torch/_dynamo/optimizations/training.py',
+    'torch/_dynamo/debug_utils.py',
+    'torch/_dynamo/repro/**/*.py',
     'torch/_inductor/graph.py',
     'torch/_C/_dynamo/**/*.py',
     'test/test_utils.py',  # used to by in MYPY but after importing op_db it took 10+ minutes

README.md

@@ -195,6 +195,10 @@ pip install -r requirements.txt
 conda install mkl mkl-include
 # CUDA only: Add LAPACK support for the GPU if needed
 conda install -c pytorch magma-cuda110  # or the magma-cuda* that matches your CUDA version from https://anaconda.org/pytorch/repo
+
+# (optional) If using torch.compile with inductor/triton, install the matching version of triton
+# Run from the pytorch directory after cloning
+make triton
 ```
 
 **On MacOS**

aten/src/ATen/CPUGeneratorImpl.cpp

@@ -94,21 +94,6 @@ void CPUGeneratorImpl::set_current_seed(uint64_t seed) {
   engine_ = mt19937(seed);
 }
 
-/**
- * Sets the offset of RNG state.
- * See Note [Acquire lock when using random generators]
- */
-void CPUGeneratorImpl::set_offset(uint64_t offset) {
-  TORCH_CHECK(false, "CPU Generator does not use offset");
-}
-
-/**
- * Gets the current offset of CPUGeneratorImpl.
- */
-uint64_t CPUGeneratorImpl::get_offset() const {
-  TORCH_CHECK(false, "CPU Generator does not use offset");
-}
-
 /**
  * Gets the current seed of CPUGeneratorImpl.
  */

aten/src/ATen/CPUGeneratorImpl.h

@@ -15,8 +15,6 @@ struct TORCH_API CPUGeneratorImpl : public c10::GeneratorImpl {
   // CPUGeneratorImpl methods
   std::shared_ptr<CPUGeneratorImpl> clone() const;
   void set_current_seed(uint64_t seed) override;
-  void set_offset(uint64_t offset) override;
-  uint64_t get_offset() const override;
   uint64_t current_seed() const override;
   uint64_t seed() override;
   void set_state(const c10::TensorImpl& new_state) override;