[Mlir-commits] [llvm] [mlir] [nvvm] Include `llvm-optimize-for-nvvm-target` pass in `gpu-lower-to-nvvm-pipeline` (PR #107102)

[Christian Sigg]

https://github.com/chsigg created https://github.com/llvm/llvm-project/pull/107102

The generic pass pipeline to lower to NVVM should apply these optimizations.

>From 0a4d80abf120d70e1d4712c25b9b3a812fadd24c Mon Sep 17 00:00:00 2001
From: Christian Sigg <csigg at google.com>
Date: Tue, 3 Sep 2024 14:46:24 +0200
Subject: [PATCH] Add `llvm-optimize-for-nvvm-target` pass to
 `gpu-lower-to-nvvm-pipeline`.

---
 mlir/lib/Dialect/GPU/Pipelines/CMakeLists.txt        | 1 +
 mlir/lib/Dialect/GPU/Pipelines/GPUToNVVMPipeline.cpp | 2 ++
 utils/bazel/llvm-project-overlay/mlir/BUILD.bazel    | 1 +
 3 files changed, 4 insertions(+)

diff --git a/mlir/lib/Dialect/GPU/Pipelines/CMakeLists.txt b/mlir/lib/Dialect/GPU/Pipelines/CMakeLists.txt
index 70a9c77a6d796a..5e5003e0e4a329 100644
--- a/mlir/lib/Dialect/GPU/Pipelines/CMakeLists.txt
+++ b/mlir/lib/Dialect/GPU/Pipelines/CMakeLists.txt
@@ -11,6 +11,7 @@ add_mlir_dialect_library(MLIRGPUPipelines
   MLIRTransforms
   MLIRLinalgTransforms
   MLIRAffineToStandard
+  MLIRLLVMIRTransforms
   MLIRGPUToNVVMTransforms
   MLIRIndexToLLVM
   MLIRMathToLLVM
diff --git a/mlir/lib/Dialect/GPU/Pipelines/GPUToNVVMPipeline.cpp b/mlir/lib/Dialect/GPU/Pipelines/GPUToNVVMPipeline.cpp
index fb440756e0c1d5..6051d232b2ddc1 100644
--- a/mlir/lib/Dialect/GPU/Pipelines/GPUToNVVMPipeline.cpp
+++ b/mlir/lib/Dialect/GPU/Pipelines/GPUToNVVMPipeline.cpp
@@ -30,6 +30,7 @@
 #include "mlir/Dialect/GPU/Pipelines/Passes.h"
 #include "mlir/Dialect/GPU/Transforms/Passes.h"
 #include "mlir/Dialect/LLVMIR/LLVMDialect.h"
+#include "mlir/Dialect/LLVMIR/Transforms/OptimizeForNVVM.h"
 #include "mlir/Dialect/Linalg/Passes.h"
 #include "mlir/Dialect/MemRef/Transforms/Passes.h"
 #include "mlir/Pass/PassManager.h"
@@ -78,6 +79,7 @@ void buildGpuPassPipeline(OpPassManager &pm,
   opt.useBarePtrCallConv = options.kernelUseBarePtrCallConv;
   opt.indexBitwidth = options.indexBitWidth;
   pm.addNestedPass<gpu::GPUModuleOp>(createConvertGpuOpsToNVVMOps(opt));
+  pm.addNestedPass<gpu::GPUModuleOp>(NVVM::createOptimizeForTargetPass());
   pm.addNestedPass<gpu::GPUModuleOp>(createCanonicalizerPass());
   pm.addNestedPass<gpu::GPUModuleOp>(createCSEPass());
   pm.addNestedPass<gpu::GPUModuleOp>(createReconcileUnrealizedCastsPass());
diff --git a/utils/bazel/llvm-project-overlay/mlir/BUILD.bazel b/utils/bazel/llvm-project-overlay/mlir/BUILD.bazel
index ddb08f12f04976..c19345ea378a00 100644
--- a/utils/bazel/llvm-project-overlay/mlir/BUILD.bazel
+++ b/utils/bazel/llvm-project-overlay/mlir/BUILD.bazel
@@ -5636,6 +5636,7 @@ cc_library(
         ":GPUTransforms",
         ":IndexToLLVM",
         ":LLVMDialect",
+        ":LLVMIRTransforms",
         ":LinalgTransforms",
         ":MathToLLVM",
         ":MemRefToLLVM",