Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[CombToArith] Fix lowering of concat with single operand #6505

Merged
merged 1 commit into from
Dec 8, 2023
Merged

[CombToArith] Fix lowering of concat with single operand #6505

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
35 changes: 22 additions & 13 deletions lib/Conversion/CombToArith/CombToArith.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -139,21 +139,30 @@ struct ConcatOpConversion : OpConversionPattern<ConcatOp> {
ConversionPatternRewriter &rewriter) const override {
Type type = op.getResult().getType();
Location loc = op.getLoc();
unsigned nextInsertion = type.getIntOrFloatBitWidth();

// Handle the trivial case where we have only one operand. The concat is a
// no-op in this case.
if (op.getNumOperands() == 1) {
rewriter.replaceOp(op, adaptor.getOperands().back());
return success();
}

// The operand at the least significant bit position (the one all the way on
// the right at the highest index) does not need to be shifted and can just
// be zero-extended to the final bit width.
Value aggregate =
rewriter.create<arith::ConstantOp>(loc, IntegerAttr::get(type, 0));

for (unsigned i = 0, e = op.getNumOperands(); i < e; i++) {
nextInsertion -=
adaptor.getOperands()[i].getType().getIntOrFloatBitWidth();

Value nextInsValue = rewriter.create<arith::ConstantOp>(
loc, IntegerAttr::get(type, nextInsertion));
Value extended =
rewriter.create<ExtUIOp>(loc, type, adaptor.getOperands()[i]);
Value shifted = rewriter.create<ShLIOp>(loc, extended, nextInsValue);
aggregate = rewriter.create<OrIOp>(loc, aggregate, shifted);
rewriter.createOrFold<ExtUIOp>(loc, type, adaptor.getOperands().back());

// Shift and OR all the other operands onto the aggregate. Skip the last
// operand because it has already been incorporated into the aggregate.
unsigned offset = type.getIntOrFloatBitWidth();
for (auto operand : adaptor.getOperands().drop_back()) {
offset -= operand.getType().getIntOrFloatBitWidth();
auto offsetConst = rewriter.create<arith::ConstantOp>(
loc, IntegerAttr::get(type, offset));
auto extended = rewriter.createOrFold<ExtUIOp>(loc, type, operand);
auto shifted = rewriter.createOrFold<ShLIOp>(loc, extended, offsetConst);
aggregate = rewriter.createOrFold<OrIOp>(loc, aggregate, shifted);
}

rewriter.replaceOp(op, aggregate);
Expand Down
36 changes: 18 additions & 18 deletions test/Conversion/CombToArith/comb-to-arith.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -75,30 +75,30 @@ hw.module @test(in %arg0: i32, in %arg1: i32, in %arg2: i32, in %arg3: i32, in %
// CHECK-NEXT: arith.trunci [[V0]] : i32 to i16
%28 = comb.extract %arg0 from 5 : (i32) -> i16

// CHECK-NEXT: %c0_i64 = arith.constant 0 : i64
// CHECK-NEXT: %c32_i64 = arith.constant 32 : i64
// CHECK-NEXT: [[V1:%.+]] = arith.extui %arg0 : i32 to i64
// CHECK-NEXT: [[V2:%.+]] = arith.shli [[V1]], %c32_i64 : i64
// CHECK-NEXT: [[V3:%.+]] = arith.ori %c0_i64, [[V2]] : i64
// CHECK-NEXT: %c0_i64_0 = arith.constant 0 : i64
// CHECK-NEXT: [[V4:%.+]] = arith.extui %arg1 : i32 to i64
// CHECK-NEXT: [[V5:%.+]] = arith.shli [[V4]], %c0_i64_0 : i64
// CHECK-NEXT: arith.ori [[V3]], [[V5]] : i64
// CHECK-NEXT: [[AGG0:%.+]] = arith.extui %arg1 : i32 to i64
// CHECK-NEXT: [[C32:%.+]] = arith.constant 32 : i64
// CHECK-NEXT: [[V0:%.+]] = arith.extui %arg0 : i32 to i64
// CHECK-NEXT: [[V1:%.+]] = arith.shli [[V0]], [[C32]] : i64
// CHECK-NEXT: [[AGG1:%.+]] = arith.ori [[AGG0]], [[V1]] : i64
%29 = comb.concat %arg0, %arg1 : i32, i32

// CHECK-NEXT: arith.extsi %arg4 : i1 to i32
%30 = comb.replicate %arg4 : (i1) -> i32

// CHECK-NEXT: [[C0:%.+]] = arith.constant 0 : i64
// CHECK-NEXT: [[C1:%.+]] = arith.constant 32 : i64
// CHECK-NEXT: [[V1:%.+]] = arith.extui %arg0 : i32 to i64
// CHECK-NEXT: [[V2:%.+]] = arith.shli [[V1]], [[C1]] : i64
// CHECK-NEXT: [[V3:%.+]] = arith.ori [[C0]], [[V2]] : i64
// CHECK-NEXT: [[C2:%.+]] = arith.constant 0 : i64
// CHECK-NEXT: [[V4:%.+]] = arith.extui %arg0 : i32 to i64
// CHECK-NEXT: [[V5:%.+]] = arith.shli [[V4]], [[C2]] : i64
// CHECK-NEXT: arith.ori [[V3]], [[V5]] : i64
// CHECK-NEXT: [[AGG0:%.+]] = arith.extui %arg0 : i32 to i64
// CHECK-NEXT: [[C32:%.+]] = arith.constant 32 : i64
// CHECK-NEXT: [[V0:%.+]] = arith.extui %arg0 : i32 to i64
// CHECK-NEXT: [[V1:%.+]] = arith.shli [[V0]], [[C32]] : i64
// CHECK-NEXT: [[AGG1:%.+]] = arith.ori [[AGG0]], [[V1]] : i64
%31 = comb.replicate %arg0 : (i32) -> i64

// The following used to trigger the arith.extui verifier.
// CHECK-NEXT: scf.execute_region
// CHECK-NEXT: scf.yield %arg0 : i32
scf.execute_region -> i32 {
%32 = comb.concat %arg0 : i32
scf.yield %32 : i32
}
}

// CHECK-LABEL: @shlTest
Expand Down