[Triton-IR] Added type inference and verifier for Triton-IR operations (#767)

2022-10-11 18:16:41 -07:00
parent b6e5a231e5
commit 623c99609f
27 changed files with 494 additions and 348 deletions
--- a/python/src/triton.cc
+++ b/python/src/triton.cc
@@ -1182,10 +1182,6 @@ void init_triton_ir(py::module &&m) {
           [](mlir::PassManager &self) {
             self.addPass(mlir::createTritonGPUCombineOpsPass());
           })
-      .def("add_triton_gpu_verifier_pass",
-           [](mlir::PassManager &self) {
-             self.addPass(mlir::createTritonGPUVerifier());
-           })
      .def("add_triton_gpu_to_llvm",
           [](mlir::PassManager &self) {
             self.addPass(mlir::triton::createConvertTritonGPUToLLVMPass());
--- a/python/triton/compiler.py
+++ b/python/triton/compiler.py
@@ -861,7 +861,6 @@ def optimize_tritongpu_ir(mod, num_stages):
    pm.add_cse_pass()
    pm.add_coalesce_pass()
    pm.add_triton_gpu_combine_pass()
-    pm.add_triton_gpu_verifier_pass()
    pm.run(mod)
    return mod