[Mlir-commits] [mlir] [mlir][x86] Add pattern to tile and unroll vector.contract (PR #229076)

[Julian Oppermann]

================
@@ -0,0 +1,809 @@
+//===- VectorContractMultiLevelUnroll.cpp ---------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#include "mlir/Dialect/SCF/IR/SCF.h"
+#include "mlir/Dialect/Utils/StaticValueUtils.h"
+#include "mlir/Dialect/Vector/IR/VectorOps.h"
+#include "mlir/Dialect/Vector/Transforms/VectorRewritePatterns.h"
+#include "mlir/Dialect/Vector/Utils/VectorUtils.h"
+#include "mlir/Dialect/X86/Transforms.h"
+#include "mlir/Dialect/X86/Utils/X86Utils.h"
+
+#include "mlir/IR/BuiltinAttributes.h"
+#include "mlir/IR/PatternMatch.h"
+#include "mlir/Transforms/LoopInvariantCodeMotionUtils.h"
+
+#define DEBUG_TYPE "x86-vector-contract-multi-level-unroll"
+
+using namespace mlir;
+
+using llvm::all_of;
+using llvm::count;
+
+// Discardable attribute to control vector unroll patterns.
+static constexpr auto nativeShapeAttrName = "x86_vcmlu_native_shape";
+
+namespace {
+// Shared state for the pattern match and IR transformation.
+struct MLUCandidate {
+  // The CPU feature flag to target with the transformation, in the spelling
+  // used by llvm/include/llvm/TargetParser/X86TargetParser.def, e.g.
----------------
jopperm wrote:

I didn't find a good universal point of reference, so I ended up introducing the enum and just spelling out the choices. I also documented which nanokernel pattern a given target corresponds to.

(Review note: The transform op still takes the target as a string attribute (but verifies it). Introducing a proper wrapper attribute for the enum would have made the transform ops dependent on the X86 dialect definition, which would have involved more refactoring that I didn't want to do in the PR.)

https://github.com/llvm/llvm-project/pull/229076