shylie/llvm-project
1
0
Fork 0
Code Issues Pull Requests Actions 6 Packages Projects Releases Wiki Activity

[flang][acc] Handle fir.undefined with OutlineRematerializationOpInterface in OffloadLiveInValueCanonicalization (#188325)

Browse Source 
Example:
```fortran
!$ACC KERNELS PRESENT(CG, W1)
  CG(1:W1%WDES1%NPL, NN) = W1%CPTWFP(1:W1%WDES1%NPL)
  CPROJ(:, NN) = W1%CPROJ(1:SIZE(CPROJ,1))
!$ACC END KERNELS
```

When compiling OpenACC kernels containing array section assignments of
rank-2 arrays with a scalar index in one dimension (e.g. `CG(1:NPL,
NN)`), the Fortran lowering creates a `fir.slice` where collapsed
(scalar) dimensions use `fir.undefined index` as the stop/step values.
`SliceOp::getOutputRank()` relies on `getDefiningOp()` returning
`fir::UndefOp` to identify these collapsed dimensions and compute the
correct output rank.

When `fir.undefined` values defined outside an offload region are used
inside it, `gpu-kernel-outlining` turns them into function arguments.
Since function arguments have no defining op (`getDefiningOp()` returns
`nullptr`), `getOutputRank()` no longer recognizes the collapsed
dimensions, computing rank 2 instead of rank 1. This causes the
`fir.rebox` verifier to fail with:

```
'fir.rebox' op result type rank and rank after applying slice operand must match
```

Fix: Register `OutlineRematerializationOpInterface` for `fir::UndefOp`
(and `fir::SliceOp`) in `RegisterOpenACCExtensions.cpp`. This causes
`OffloadLiveInValueCanonicalization` to clone these operations inside
the offload region before outlining, preserving the `fir::UndefOp`
identity so that `getOutputRank()` correctly identifies collapsed
dimensions.
This commit is contained in:
khaki3 2026-03-24 18:03:57 -07:00 committed by GitHub
parent 408bb4d4b2
commit aecfaf1195
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 56 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
flang/lib/Optimizer/OpenACC/Support/RegisterOpenACCExtensions.cpp
View File

@ -81,6 +81,10 @@ void registerOpenACCExtensions(mlir::DialectRegistry &registry) {
OutlineRematerializationModel<fir::FieldIndexOp>>(*ctx);
fir::ConvertOp::attachInterface<
OutlineRematerializationModel<fir::ConvertOp>>(*ctx);
fir::UndefOp::attachInterface<OutlineRematerializationModel<fir::UndefOp>>(
*ctx);
fir::SliceOp::attachInterface<OutlineRematerializationModel<fir::SliceOp>>(
*ctx);
});
// Register HLFIR operation interfaces

52
flang/test/Fir/OpenACC/offload-livein-value-canonicalization.fir
View File

@ -371,3 +371,55 @@ func.func @test_convert_chain_and_direct(%arg0: !fir.boxchar<1>, %arg1: !fir.ref
// CHECK: %[[CVT_INT:.*]] = fir.convert %arg1
// CHECK: fir.call @use_ref(%[[DECL]])
// CHECK: fir.call @use_i64(%[[CVT_INT]])
// -----
// Test fir.undefined sinking into offload region.
// fir.undefined is used in fir.slice to mark collapsed dimensions.
// After gpu-kernel-outlining, function arguments lose their defining op,
// so SliceOp::getOutputRank() can no longer identify collapsed dims.
// The pass must sink fir.undefined to preserve this information.
func.func @test_undef_slice_sink(%arg0: !fir.box<!fir.array<?x?xf32>>) {
%c1 = arith.constant 1 : index
%c10 = arith.constant 10 : index
%undef = fir.undefined index
%slice = fir.slice %c1, %c10, %c1, %c1, %undef, %undef : (index, index, index, index, index, index) -> !fir.slice<2>
acc.serial {
%rebox = fir.rebox %arg0 [%slice] : (!fir.box<!fir.array<?x?xf32>>, !fir.slice<2>) -> !fir.box<!fir.array<?xf32>>
acc.yield
}
return
}
// CHECK-LABEL: @test_undef_slice_sink
// CHECK: acc.serial {
// CHECK: %[[UNDEF:.*]] = fir.undefined index
// CHECK: %[[SLICE:.*]] = fir.slice {{.*}}, %[[UNDEF]], %[[UNDEF]]
// CHECK: fir.rebox %arg0 [%[[SLICE]]]
// -----
// Test fir.undefined rematerialization (used both inside and outside region).
func.func @test_undef_slice_rematerialize(%arg0: !fir.box<!fir.array<?x?xf32>>) {
%c1 = arith.constant 1 : index
%c10 = arith.constant 10 : index
%undef = fir.undefined index
%slice = fir.slice %c1, %c10, %c1, %c1, %undef, %undef : (index, index, index, index, index, index) -> !fir.slice<2>
%rebox_outer = fir.rebox %arg0 [%slice] : (!fir.box<!fir.array<?x?xf32>>, !fir.slice<2>) -> !fir.box<!fir.array<?xf32>>
acc.serial {
%rebox_inner = fir.rebox %arg0 [%slice] : (!fir.box<!fir.array<?x?xf32>>, !fir.slice<2>) -> !fir.box<!fir.array<?xf32>>
acc.yield
}
return
}
// CHECK-LABEL: @test_undef_slice_rematerialize
// CHECK: %[[UNDEF_OUTER:.*]] = fir.undefined index
// CHECK: %[[SLICE_OUTER:.*]] = fir.slice {{.*}}, %[[UNDEF_OUTER]], %[[UNDEF_OUTER]]
// CHECK: fir.rebox %arg0 [%[[SLICE_OUTER]]]
// CHECK: acc.serial {
// CHECK: %[[UNDEF_INNER:.*]] = fir.undefined index
// CHECK: %[[SLICE_INNER:.*]] = fir.slice {{.*}}, %[[UNDEF_INNER]], %[[UNDEF_INNER]]
// CHECK: fir.rebox %arg0 [%[[SLICE_INNER]]]