Files
clang-p2996/mlir/test/Transforms/test-operation-folder.mlir
Jacques Pienaar 09dfc5713d [mlir] Enable decoupling two kinds of greedy behavior. (#104649)
The greedy rewriter is used in many different flows and it has a lot of
convenience (work list management, debugging actions, tracing, etc). But
it combines two kinds of greedy behavior 1) how ops are matched, 2)
folding wherever it can.

These are independent forms of greedy and leads to inefficiency. E.g.,
cases where one need to create different phases in lowering and is
required to applying patterns in specific order split across different
passes. Using the driver one ends up needlessly retrying folding/having
multiple rounds of folding attempts, where one final run would have
sufficed.

Of course folks can locally avoid this behavior by just building their
own, but this is also a common requested feature that folks keep on
working around locally in suboptimal ways.

For downstream users, there should be no behavioral change. Updating
from the deprecated should just be a find and replace (e.g., `find ./
-type f -exec sed -i
's|applyPatternsAndFoldGreedily|applyPatternsGreedily|g' {} \;` variety)
as the API arguments hasn't changed between the two.
2024-12-20 08:15:48 -08:00

83 lines
3.1 KiB
MLIR

// RUN: mlir-opt -test-greedy-patterns='top-down=false' %s | FileCheck %s
// RUN: mlir-opt -test-greedy-patterns='top-down=true' %s | FileCheck %s
// RUN: mlir-opt -test-greedy-patterns='cse-constants=false' %s | FileCheck %s --check-prefix=NOCSE
// RUN: mlir-opt -test-greedy-patterns='fold=false' %s | FileCheck %s --check-prefix=NOFOLD
func.func @foo() -> i32 {
%c42 = arith.constant 42 : i32
// The new operation should be present in the output and contain an attribute
// with value "42" that results from folding.
// CHECK: "test.op_in_place_fold"(%{{.*}}) <{attr = 42 : i32}
%0 = "test.op_in_place_fold_anchor"(%c42) : (i32) -> (i32)
return %0 : i32
}
func.func @test_fold_before_previously_folded_op() -> (i32, i32) {
// When folding two constants will be generated and uniqued. Check that the
// uniqued constant properly dominates both uses.
// CHECK: %[[CST:.+]] = arith.constant true
// CHECK-NEXT: "test.cast"(%[[CST]]) : (i1) -> i32
// CHECK-NEXT: "test.cast"(%[[CST]]) : (i1) -> i32
%0 = "test.cast"() {test_fold_before_previously_folded_op} : () -> (i32)
%1 = "test.cast"() {test_fold_before_previously_folded_op} : () -> (i32)
return %0, %1 : i32, i32
}
func.func @test_dont_reorder_constants() -> (i32, i32, i32) {
// Test that we don't reorder existing constants during folding if it isn't
// necessary.
// CHECK: %[[CST:.+]] = arith.constant 1
// CHECK-NEXT: %[[CST:.+]] = arith.constant 2
// CHECK-NEXT: %[[CST:.+]] = arith.constant 3
%0 = arith.constant 1 : i32
%1 = arith.constant 2 : i32
%2 = arith.constant 3 : i32
return %0, %1, %2 : i32, i32, i32
}
// CHECK-LABEL: test_fold_nofold_nocse
// NOCSE-LABEL: test_fold_nofold_nocse
// NOFOLD-LABEL: test_fold_nofold_nocse
func.func @test_fold_nofold_nocse() -> (i32, i32, i32, i32, i32, i32) {
// Test either not folding or deduping constants.
// Testing folding. There should be only 4 constants here.
// CHECK-NOT: arith.constant
// CHECK-DAG: %[[CST:.+]] = arith.constant 0
// CHECK-DAG: %[[CST:.+]] = arith.constant 1
// CHECK-DAG: %[[CST:.+]] = arith.constant 2
// CHECK-DAG: %[[CST:.+]] = arith.constant 3
// CHECK-NOT: arith.constant
// CHECK-NEXT: return
// Testing not-CSE'ing. In this case we have the 3 original constants and 3
// produced by folding.
// NOCSE-DAG: arith.constant 0 : i32
// NOCSE-DAG: arith.constant 1 : i32
// NOCSE-DAG: arith.constant 2 : i32
// NOCSE-DAG: arith.constant 1 : i32
// NOCSE-DAG: arith.constant 2 : i32
// NOCSE-DAG: arith.constant 3 : i32
// NOCSE-NEXT: return
// Testing not folding. In this case we just have the original constants.
// NOFOLD-DAG: %[[CST:.+]] = arith.constant 0
// NOFOLD-DAG: %[[CST:.+]] = arith.constant 1
// NOFOLD-DAG: %[[CST:.+]] = arith.constant 2
// NOFOLD: arith.addi
// NOFOLD: arith.addi
// NOFOLD: arith.addi
%c0 = arith.constant 0 : i32
%c1 = arith.constant 1 : i32
%c2 = arith.constant 2 : i32
%0 = arith.addi %c0, %c1 : i32
%1 = arith.addi %0, %c1 : i32
%2 = arith.addi %c2, %c1 : i32
return %0, %1, %2, %c0, %c1, %c2 : i32, i32, i32, i32, i32, i32
}