[mlir][Vector] Remove 'lower-permutation-maps' option from VectorToSCF

[dcaballe]

Summary

This patch is part of a larger simplification effort of vector transfer
operations. It removes the flag lower-permutation-maps from
VectorToSCF conversion and enables the lowering of permutation maps
by default. This means that VectorToSCF will always lower permutation
maps to independent broadcast/transpose operations before lowering
vector operations to SCF.

mlir/include/mlir/Conversion/Passes.td

let dependentDialects = [
     "memref::MemRefDialect",
     "scf::SCFDialect"
   ];
   let options = [
     Option<"fullUnroll", "full-unroll", "bool", /*default=*/"false",
            "Perform full unrolling when converting vector transfers to SCF">,
     Option<"targetRank", "target-rank", "unsigned", /*default=*/"1",
            "Target vector rank to which transfer ops should be lowered">,
-    Option<"lowerPermutationMaps", "lower-permutation-maps", "bool",
-           /*default=*/"false", "Replace permutation maps with vector "
-           "transposes/broadcasts before lowering transfer ops">,
     Option<"lowerTensors", "lower-tensors", "bool", /*default=*/"false",
            "Lower transfer ops that operate on tensors">
   ];
 }
 
 //===----------------------------------------------------------------------===//
 // VectorToLLVM
 //===----------------------------------------------------------------------===//

mlir/include/mlir/Conversion/VectorToSCF/VectorToSCF.h

 /// is reused and only a second vector.type_cast is added.
 struct VectorTransferToSCFOptions {
   /// Minimal rank to which vector transfer are lowered.
   unsigned targetRank = 1;
   VectorTransferToSCFOptions &setTargetRank(unsigned r) {
     targetRank = r;
     return *this;
   }
-  ///
-  bool lowerPermutationMaps = false;
-  VectorTransferToSCFOptions &enableLowerPermutationMaps(bool l = true) {
-    lowerPermutationMaps = l;
-    return *this;
-  }
   /// Allows vector transfers that operated on tensors to be lowered (this is an
   /// uncommon alternative).
   bool lowerTensors = false;
   VectorTransferToSCFOptions &enableLowerTensors(bool l = true) {
     lowerTensors = l;
     return *this;
   }
   /// Triggers full unrolling (vs iterating with a loop) during transfer to scf.

mlir/lib/Conversion/VectorToSCF/VectorToSCF.cpp

 namespace {
 
 struct ConvertVectorToSCFPass
     : public impl::ConvertVectorToSCFBase<ConvertVectorToSCFPass> {
   ConvertVectorToSCFPass() = default;
   ConvertVectorToSCFPass(const VectorTransferToSCFOptions &options) {
     this->fullUnroll = options.unroll;
     this->targetRank = options.targetRank;
-    this->lowerPermutationMaps = options.lowerPermutationMaps;
     this->lowerTensors = options.lowerTensors;
   }
 
   void runOnOperation() override {
     VectorTransferToSCFOptions options;
     options.unroll = fullUnroll;
     options.targetRank = targetRank;
-    options.lowerPermutationMaps = lowerPermutationMaps;
     options.lowerTensors = lowerTensors;
 
     // Lower permutation maps first.
-    if (lowerPermutationMaps) {
-      RewritePatternSet lowerTransferPatterns(&getContext());
-      mlir::vector::populateVectorTransferPermutationMapLoweringPatterns(
-          lowerTransferPatterns);
-      (void)applyPatternsAndFoldGreedily(getOperation(),
-                                         std::move(lowerTransferPatterns));
-    }
 
     RewritePatternSet patterns(&getContext());
     populateVectorToSCFConversionPatterns(patterns, options);
     (void)applyPatternsAndFoldGreedily(getOperation(), std::move(patterns));
   }
 };
 
 } // namespace
 
 std::unique_ptr<Pass>
 mlir::createConvertVectorToSCFPass(const VectorTransferToSCFOptions &options) {
   return std::make_unique<ConvertVectorToSCFPass>(options);
 }

mlir/lib/Dialect/Vector/TransformOps/VectorTransformOps.cpp

for (Operation *target : payloadOps) {
 
     vector::VectorTransformsOptions vectorTransformOptions;
     vectorTransformOptions.setVectorTransformsOptions(vectorContractLowering)
         .setVectorMultiReductionLowering(vectorMultiReductionLowering)
         .setVectorTransposeLowering(vectorTransposeLowering)
         .setVectorTransferSplit(vectorTransferSplit);
 
     VectorTransferToSCFOptions vectorTransferToSCFOptions =
-        VectorTransferToSCFOptions()
-            .enableFullUnroll(getUnrollVectorTransfers())
-            .enableLowerPermutationMaps();
+        VectorTransferToSCFOptions().enableFullUnroll(
+            getUnrollVectorTransfers());
 
     int maxTransferRank = 1;
 
     auto avx2LoweringOptions =
         x86vector::avx2::LoweringOptions().setTransposeOptions(
             x86vector::avx2::TransposeLoweringOptions()
                 .lower4x8xf32(getTransposeAvx2Lowering())
                 .lower8x8xf32(getTransposeAvx2Lowering()));

mlir/test/Conversion/VectorToSCF/vector-to-scf-mask-and-permutation-map.mlir

-// RUN: mlir-opt %s -pass-pipeline="builtin.module(func.func(convert-vector-to-scf{lower-permutation-maps=true}))" -split-input-file | FileCheck %s
+// RUN: mlir-opt %s -pass-pipeline="builtin.module(func.func(convert-vector-to-scf))" -split-input-file | FileCheck %s
 
 // Ensure that the permutation map is lowered (by inserting a transpose op)
 // before lowering the vector.transfer_read.
 
 // CHECK-LABEL: func @transfer_read_2d_mask_transposed(
 //   CHECK-DAG:   %[[PADDING:.*]] = arith.constant dense<-4.200000e+01> : vector<9xf32>
 //   CHECK-DAG:   %[[MASK:.*]] = arith.constant dense<{{.*}}> : vector<9x4xi1>
 //       CHECK:   %[[MASK_MEM:.*]] = memref.alloca() : memref<vector<4x9xi1>>

mlir/test/Integration/Dialect/Vector/CPU/test-transfer-read-1d.mlir

 // RUN: mlir-opt %s -pass-pipeline="builtin.module(func.func(convert-vector-to-scf,lower-affine,convert-scf-to-cf),convert-vector-to-llvm,convert-memref-to-llvm,convert-func-to-llvm,reconcile-unrealized-casts)" | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
 // RUN:   -shared-libs=%mlir_lib_dir/libmlir_c_runner_utils%shlibext | \
 // RUN: FileCheck %s
 
-// RUN: mlir-opt %s -pass-pipeline="builtin.module(func.func(convert-vector-to-scf{lower-permutation-maps=true},lower-affine,convert-scf-to-cf),convert-vector-to-llvm,convert-memref-to-llvm,convert-func-to-llvm,reconcile-unrealized-casts)" | \
-// RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
-// RUN:   -shared-libs=%mlir_lib_dir/libmlir_c_runner_utils%shlibext | \
-// RUN: FileCheck %s
-
 // RUN: mlir-opt %s -pass-pipeline="builtin.module(func.func(convert-vector-to-scf{full-unroll=true},lower-affine,convert-scf-to-cf),convert-vector-to-llvm,convert-memref-to-llvm,convert-func-to-llvm,reconcile-unrealized-casts)" | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
 // RUN:   -shared-libs=%mlir_lib_dir/libmlir_c_runner_utils%shlibext | \
 // RUN: FileCheck %s
 
-// RUN: mlir-opt %s -pass-pipeline="builtin.module(func.func(convert-vector-to-scf{full-unroll=true lower-permutation-maps=true},lower-affine,convert-scf-to-cf),convert-vector-to-llvm,convert-memref-to-llvm,convert-func-to-llvm,reconcile-unrealized-casts)" | \
-// RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
-// RUN:   -shared-libs=%mlir_lib_dir/libmlir_c_runner_utils%shlibext | \
-// RUN: FileCheck %s
-
 // Test for special cases of 1D vector transfer ops.
 
 memref.global "private" @gv : memref<5x6xf32> =
     dense<[[0. , 1. , 2. , 3. , 4. , 5. ],
            [10., 11., 12., 13., 14., 15.],
            [20., 21., 22., 23., 24., 25.],
            [30., 31., 32., 33., 34., 35.],
            [40., 41., 42., 43., 44., 45.]]>

mlir/test/Integration/Dialect/Vector/CPU/test-transfer-read-2d.mlir

 // RUN: mlir-opt %s -pass-pipeline="builtin.module(func.func(convert-vector-to-scf,lower-affine,convert-scf-to-cf),convert-vector-to-llvm,convert-memref-to-llvm,convert-func-to-llvm,reconcile-unrealized-casts)" | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
 // RUN:   -shared-libs=%mlir_lib_dir/libmlir_c_runner_utils%shlibext | \
 // RUN: FileCheck %s
 
-// RUN: mlir-opt %s -pass-pipeline="builtin.module(func.func(convert-vector-to-scf{lower-permutation-maps=true},lower-affine,convert-scf-to-cf),convert-vector-to-llvm,convert-memref-to-llvm,convert-func-to-llvm,reconcile-unrealized-casts)" | \
-// RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
-// RUN:   -shared-libs=%mlir_lib_dir/libmlir_c_runner_utils%shlibext | \
-// RUN: FileCheck %s
-
 // RUN: mlir-opt %s -pass-pipeline="builtin.module(func.func(convert-vector-to-scf{full-unroll=true},lower-affine,convert-scf-to-cf),convert-vector-to-llvm,convert-memref-to-llvm,convert-func-to-llvm,reconcile-unrealized-casts)" | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
 // RUN:   -shared-libs=%mlir_lib_dir/libmlir_c_runner_utils%shlibext | \
 // RUN: FileCheck %s
 
-// RUN: mlir-opt %s -pass-pipeline="builtin.module(func.func(convert-vector-to-scf{full-unroll=true lower-permutation-maps=true},lower-affine,convert-scf-to-cf),convert-vector-to-llvm,convert-memref-to-llvm,convert-func-to-llvm,reconcile-unrealized-casts)" | \
-// RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
-// RUN:   -shared-libs=%mlir_lib_dir/libmlir_c_runner_utils%shlibext | \
-// RUN: FileCheck %s
-
 memref.global "private" @gv : memref<3x4xf32> = dense<[[0. , 1. , 2. , 3. ],
                                                        [10., 11., 12., 13.],
                                                        [20., 21., 22., 23.]]>
 
 // Vector load.
 func.func @transfer_read_2d(%A : memref<?x?xf32>, %base1: index, %base2: index) {
   %fm42 = arith.constant -42.0: f32
   %f = vector.transfer_read %A[%base1, %base2], %fm42

mlir/test/Integration/Dialect/Vector/CPU/test-transfer-read-3d.mlir

 // RUN: mlir-opt %s -pass-pipeline="builtin.module(func.func(convert-vector-to-scf,lower-affine,convert-scf-to-cf),convert-vector-to-llvm,convert-memref-to-llvm,convert-func-to-llvm,reconcile-unrealized-casts)" | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
 // RUN:   -shared-libs=%mlir_lib_dir/libmlir_c_runner_utils%shlibext | \
 // RUN: FileCheck %s
 
-// RUN: mlir-opt %s -pass-pipeline="builtin.module(func.func(convert-vector-to-scf{lower-permutation-maps=true},lower-affine,convert-scf-to-cf),convert-vector-to-llvm,convert-memref-to-llvm,convert-func-to-llvm,reconcile-unrealized-casts)" | \
-// RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
-// RUN:   -shared-libs=%mlir_lib_dir/libmlir_c_runner_utils%shlibext | \
-// RUN: FileCheck %s
-
 // RUN: mlir-opt %s -pass-pipeline="builtin.module(func.func(convert-vector-to-scf{full-unroll=true},lower-affine,convert-scf-to-cf),convert-vector-to-llvm,convert-memref-to-llvm,convert-func-to-llvm,reconcile-unrealized-casts)" | \
 // RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
 // RUN:   -shared-libs=%mlir_lib_dir/libmlir_c_runner_utils%shlibext | \
 // RUN: FileCheck %s
 
-// RUN: mlir-opt %s -pass-pipeline="builtin.module(func.func(convert-vector-to-scf{full-unroll=true lower-permutation-maps=true},lower-affine,convert-scf-to-cf),convert-vector-to-llvm,convert-memref-to-llvm,convert-func-to-llvm,reconcile-unrealized-casts)" | \
-// RUN: mlir-cpu-runner -e entry -entry-point-result=void  \
-// RUN:   -shared-libs=%mlir_lib_dir/libmlir_c_runner_utils%shlibext | \
-// RUN: FileCheck %s
-
 func.func @transfer_read_3d(%A : memref<?x?x?x?xf32>,
                        %o: index, %a: index, %b: index, %c: index) {
   %fm42 = arith.constant -42.0: f32
   %f = vector.transfer_read %A[%o, %a, %b, %c], %fm42
       : memref<?x?x?x?xf32>, vector<2x5x3xf32>
   vector.print %f: vector<2x5x3xf32>
   return
 }