AcoSail/llvm-project
8
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

[mlir] Add test-convergence option to Canonicalizer tests

Browse Source 
This new option is set to `false` by default. It should  be set only in Canonicalizer tests to detect faulty canonicalization patterns. I.e., patterns that prevent the canonicalizer from converging. The canonicalizer should always convergence on such small unit tests that we have in `canonicalize.mlir`.

Two faulty canonicalization patterns were detected and fixed with this change.

Differential Revision: https://reviews.llvm.org/D140873
This commit is contained in:
Matthias Springer 2023-01-04 11:39:41 +01:00
parent 8a06b2362a
commit e7790fbed3
28 changed files with 45 additions and 43 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
mlir/include/mlir/Transforms/Passes.td
View File

@ -39,7 +39,9 @@ def Canonicalizer : Pass<"canonicalize"> {
/*default=*/"10",
"Max. iterations between applying patterns / simplifying regions">,
Option<"maxNumRewrites", "max-num-rewrites", "int64_t", /*default=*/"-1",
"Max. number of pattern rewrites within an iteration">
"Max. number of pattern rewrites within an iteration">,
Option<"testConvergence", "test-convergence", "bool", /*default=*/"false",
"Test only: Fail pass on non-convergence to detect cyclic pattern">
] # RewritePassUtils.options;
}

4
mlir/lib/Dialect/GPU/IR/GPUDialect.cpp
View File

@ -687,6 +687,8 @@ struct FoldLaunchArguments : public OpRewritePattern<LaunchOp> {
// Check if size is trivially one.
if (!matchPattern(size, m_One()))
return;
if (id.getUses().empty())
return;
if (!simplified) {
// Create a zero value the first time.
OpBuilder::InsertionGuard guard(rewriter);
@ -694,7 +696,7 @@ struct FoldLaunchArguments : public OpRewritePattern<LaunchOp> {
zero =
rewriter.create<arith::ConstantIndexOp>(op.getLoc(), /*value=*/0);
}
id.replaceAllUsesWith(zero);
rewriter.replaceAllUsesWith(id, zero);
simplified = true;
};
constPropIdUses(op.getBlockIds().x, op.getGridSizeX());

15
mlir/lib/Dialect/OpenACC/IR/OpenACC.cpp
View File

@ -178,16 +178,15 @@ struct RemoveConstantIfCondition : public OpRewritePattern<OpTy> {
// Early return if there is no condition.
Value ifCond = op.getIfCond();
if (!ifCond)
return success();
return failure();
IntegerAttr constAttr;
if (matchPattern(ifCond, m_Constant(&constAttr))) {
if (constAttr.getInt())
rewriter.updateRootInPlace(op,
[&]() { op.getIfCondMutable().erase(0); });
else
rewriter.eraseOp(op);
}
if (!matchPattern(ifCond, m_Constant(&constAttr)))
return failure();
if (constAttr.getInt())
rewriter.updateRootInPlace(op, [&]() { op.getIfCondMutable().erase(0); });
else
rewriter.eraseOp(op);
return success();
}

5
mlir/lib/Transforms/Canonicalizer.cpp
View File

@ -57,8 +57,11 @@ struct Canonicalizer : public impl::CanonicalizerBase<Canonicalizer> {
config.enableRegionSimplification = enableRegionSimplification;
config.maxIterations = maxIterations;
config.maxNumRewrites = maxNumRewrites;
LogicalResult converged =
applyPatternsAndFoldGreedily(getOperation(), patterns, config);
// Canonicalization is best-effort. Non-convergence is not a pass failure.
(void)applyPatternsAndFoldGreedily(getOperation(), patterns, config);
if (testConvergence && failed(converged))
signalPassFailure();
}
FrozenRewritePatternSet patterns;

2
mlir/test/Dialect/AMDGPU/canonicalize.mlir
View File

@ -1,4 +1,4 @@
// RUN: mlir-opt %s -split-input-file -canonicalize | FileCheck %s
// RUN: mlir-opt %s -split-input-file -canonicalize="test-convergence" | FileCheck %s
// CHECK-LABEL: func @known_oob_load
func.func @known_oob_load(%arg0: memref<4xf32>) -> f32 {

4
mlir/test/Dialect/Affine/canonicalize.mlir
View File

@ -1,5 +1,5 @@
// RUN: mlir-opt -allow-unregistered-dialect %s -split-input-file -canonicalize | FileCheck %s
// RUN: mlir-opt -allow-unregistered-dialect %s -split-input-file -canonicalize="top-down=0" | FileCheck %s --check-prefix=CHECK-BOTTOM-UP
// RUN: mlir-opt -allow-unregistered-dialect %s -split-input-file -canonicalize="test-convergence" | FileCheck %s
// RUN: mlir-opt -allow-unregistered-dialect %s -split-input-file -canonicalize="test-convergence top-down=0" | FileCheck %s --check-prefix=CHECK-BOTTOM-UP
// -----

2
mlir/test/Dialect/Arith/canonicalize.mlir
View File

@ -1,4 +1,4 @@
// RUN: mlir-opt %s -canonicalize --split-input-file | FileCheck %s
// RUN: mlir-opt %s -canonicalize="test-convergence" --split-input-file | FileCheck %s
// CHECK-LABEL: @select_same_val
// CHECK: return %arg1

5
mlir/test/Dialect/Bufferization/canonicalize.mlir
View File

@ -1,5 +1,6 @@
// RUN: mlir-opt %s -canonicalize --split-input-file \
// RUN: -allow-unregistered-dialect |\
// RUN: mlir-opt %s \
// RUN: -canonicalize="test-convergence" \
// RUN: --split-input-file -allow-unregistered-dialect | \
// RUN: FileCheck %s
// Basic folding of to_tensor(to_memref(t)) -> t

2
mlir/test/Dialect/Builtin/canonicalize.mlir
View File

@ -1,4 +1,4 @@
// RUN: mlir-opt %s -canonicalize | FileCheck %s
// RUN: mlir-opt %s -canonicalize="test-convergence" | FileCheck %s
//===----------------------------------------------------------------------===//
// UnrealizedConversionCastOp

4
mlir/test/Dialect/Complex/canonicalize.mlir
View File

@ -1,4 +1,4 @@
// RUN: mlir-opt %s -canonicalize | FileCheck %s
// RUN: mlir-opt %s -canonicalize="test-convergence" | FileCheck %s
// CHECK-LABEL: func @create_of_real_and_imag
// CHECK-SAME: (%[[CPLX:.*]]: complex<f32>)
@ -154,4 +154,4 @@ func.func @complex_sub_zero() -> complex<f32> {
// CHECK-NEXT: return %[[CPLX:.*]] : complex<f32>
%sub = complex.sub %complex1, %complex2 : complex<f32>
return %sub : complex<f32>
}
}

2
mlir/test/Dialect/ControlFlow/canonicalize.mlir
View File

@ -1,4 +1,4 @@
// RUN: mlir-opt %s -allow-unregistered-dialect -pass-pipeline='builtin.module(func.func(canonicalize))' -split-input-file | FileCheck --dump-input-context 20 %s
// RUN: mlir-opt %s -allow-unregistered-dialect -pass-pipeline='builtin.module(func.func(canonicalize{test-convergence}))' -split-input-file | FileCheck --dump-input-context 20 %s
/// Test the folding of BranchOp.

2
mlir/test/Dialect/GPU/canonicalize.mlir
View File

@ -1,4 +1,4 @@
// RUN: mlir-opt %s -canonicalize --split-input-file -allow-unregistered-dialect | FileCheck %s
// RUN: mlir-opt %s -canonicalize="test-convergence" --split-input-file -allow-unregistered-dialect | FileCheck %s
// Fold all the gpu.wait ops as they are redundant.
// CHECK-LABEL: func @fold_wait_op_test1

2
mlir/test/Dialect/LLVMIR/canonicalize.mlir
View File

@ -1,4 +1,4 @@
// RUN: mlir-opt -canonicalize %s -split-input-file | FileCheck %s
// RUN: mlir-opt -canonicalize="test-convergence" %s -split-input-file | FileCheck %s
// CHECK-LABEL: fold_extractvalue
llvm.func @fold_extractvalue() -> i32 {

2
mlir/test/Dialect/Linalg/canonicalize.mlir
View File

@ -1,4 +1,4 @@
// RUN: mlir-opt %s -canonicalize -split-input-file | FileCheck %s
// RUN: mlir-opt %s -canonicalize="test-convergence" -split-input-file | FileCheck %s
// CHECK-LABEL: func @memref_cast(
func.func @memref_cast(%a: index, %b: index) -> memref<?x?xf32> {

2
mlir/test/Dialect/Math/canonicalize.mlir
View File

@ -1,4 +1,4 @@
// RUN: mlir-opt %s -canonicalize | FileCheck %s
// RUN: mlir-opt %s -canonicalize="test-convergence" | FileCheck %s
// CHECK-LABEL: @ceil_fold
// CHECK: %[[cst:.+]] = arith.constant 1.000000e+00 : f32

2
mlir/test/Dialect/MemRef/canonicalize.mlir
View File

@ -1,4 +1,4 @@
// RUN: mlir-opt %s -canonicalize --split-input-file -allow-unregistered-dialect | FileCheck %s
// RUN: mlir-opt %s -canonicalize="test-convergence" --split-input-file -allow-unregistered-dialect | FileCheck %s
// CHECK-LABEL: func @subview_of_size_memcast
// CHECK-SAME: %[[ARG0:.[a-z0-9A-Z_]+]]: memref<4x6x16x32xi8>

2
mlir/test/Dialect/OpenACC/canonicalize.mlir
View File

@ -1,4 +1,4 @@
// RUN: mlir-opt %s -canonicalize -split-input-file | FileCheck %s
// RUN: mlir-opt %s -canonicalize="test-convergence" -split-input-file | FileCheck %s
func.func @testenterdataop(%a: memref<10xf32>) -> () {
%ifCond = arith.constant true

2
mlir/test/Dialect/OpenMP/canonicalize.mlir
View File

@ -1,4 +1,4 @@
// RUN: mlir-opt %s -canonicalize -split-input-file | FileCheck %s
// RUN: mlir-opt %s -canonicalize="test-convergence" -split-input-file | FileCheck %s
func.func @update_no_op(%x : memref<i32>) {
omp.atomic.update %x : memref<i32> {

2
mlir/test/Dialect/PDL/canonicalize.mlir
View File

@ -1,4 +1,4 @@
// RUN: mlir-opt -canonicalize %s | FileCheck %s
// RUN: mlir-opt -canonicalize="test-convergence" %s | FileCheck %s
pdl.pattern @operation_op : benefit(1) {
%root = operation "foo.op"

2
mlir/test/Dialect/Quant/canonicalize.mlir
View File

@ -1,4 +1,4 @@
// RUN: mlir-opt %s -split-input-file -pass-pipeline='builtin.module(func.func(canonicalize))' | FileCheck %s
// RUN: mlir-opt %s -split-input-file -pass-pipeline='builtin.module(func.func(canonicalize{test-convergence}))' | FileCheck %s
// -----
// CHECK-LABEL: redundant_scast

5
mlir/test/Dialect/SCF/canonicalize.mlir
View File

@ -1,7 +1,4 @@
// RUN: mlir-opt %s -pass-pipeline='builtin.module(func.func(canonicalize))' -split-input-file | FileCheck %s
// -----
// RUN: mlir-opt %s -pass-pipeline='builtin.module(func.func(canonicalize{test-convergence}))' -split-input-file | FileCheck %s
func.func @single_iteration_some(%A: memref<?x?x?xi32>) {
%c0 = arith.constant 0 : index

2
mlir/test/Dialect/SPIRV/Transforms/canonicalize.mlir
View File

@ -1,4 +1,4 @@
// RUN: mlir-opt %s -split-input-file -pass-pipeline='builtin.module(func.func(canonicalize))' | FileCheck %s
// RUN: mlir-opt %s -split-input-file -pass-pipeline='builtin.module(func.func(canonicalize{test-convergence}))' | FileCheck %s
//===----------------------------------------------------------------------===//
// spirv.AccessChain

2
mlir/test/Dialect/Shape/canonicalize.mlir
View File

@ -1,4 +1,4 @@
// RUN: mlir-opt -split-input-file -allow-unregistered-dialect -canonicalize %s | FileCheck %s
// RUN: mlir-opt -split-input-file -allow-unregistered-dialect -canonicalize="test-convergence" %s | FileCheck %s
// CHECK-LABEL: func @f
func.func @f(%arg0: tensor<2x3x4xf32>) -> tensor<3xindex> {

2
mlir/test/Dialect/Tensor/canonicalize.mlir
View File

@ -1,4 +1,4 @@
// RUN: mlir-opt %s -split-input-file -canonicalize | FileCheck %s
// RUN: mlir-opt %s -split-input-file -canonicalize="test-convergence" | FileCheck %s
// Checks that NOP casts are removed.
// CHECK-LABEL: cast_values

2
mlir/test/Dialect/Tosa/canonicalize.mlir
View File

@ -1,4 +1,4 @@
// RUN: mlir-opt --canonicalize %s | FileCheck %s
// RUN: mlir-opt -canonicalize="test-convergence" %s | FileCheck %s
// CHECK-LABEL: @argmax_nofold
func.func @argmax_nofold(%arg0: tensor<?x1xf32>) -> tensor<?x1xf32> {

4
mlir/test/Dialect/Vector/canonicalize.mlir
View File

@ -1,6 +1,4 @@
// RUN: mlir-opt %s -pass-pipeline='builtin.module(func.func(canonicalize))' -split-input-file -allow-unregistered-dialect | FileCheck %s
// -----
// RUN: mlir-opt %s -canonicalize="test-convergence" -split-input-file -allow-unregistered-dialect | FileCheck %s
// CHECK-LABEL: create_vector_mask_to_constant_mask
func.func @create_vector_mask_to_constant_mask() -> (vector<4x3xi1>) {

2
mlir/test/Pass/run-reproducer.mlir
View File

@ -14,7 +14,7 @@ func.func @bar() {
external_resources: {
mlir_reproducer: {
verify_each: true,
// CHECK: builtin.module(func.func(cse,canonicalize{ max-iterations=1 max-num-rewrites=-1 region-simplify=false top-down=false}))
// CHECK: builtin.module(func.func(cse,canonicalize{ max-iterations=1 max-num-rewrites=-1 region-simplify=false test-convergence=false top-down=false}))
pipeline: "builtin.module(func.func(cse,canonicalize{max-iterations=1 max-num-rewrites=-1 region-simplify=false top-down=false}))",
disable_threading: true
}

2
mlir/test/Transforms/canonicalize.mlir
View File

@ -1,4 +1,4 @@
// RUN: mlir-opt -allow-unregistered-dialect %s -pass-pipeline='builtin.module(func.func(canonicalize))' -split-input-file | FileCheck %s
// RUN: mlir-opt -allow-unregistered-dialect %s -pass-pipeline='builtin.module(func.func(canonicalize{test-convergence}))' -split-input-file | FileCheck %s
// CHECK-LABEL: func @test_subi_zero
func.func @test_subi_zero(%arg0: i32) -> i32 {