Skip to content

[flang][fir] always use memcpy for fir.box #113949

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

Merged
merged 5 commits into from
Oct 30, 2024
Merged

[flang][fir] always use memcpy for fir.box #113949

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
58bc994
[fir] Always use memcpy for fir.load/store
ashermancinelli Oct 23, 2024
b8b403b
[fir][test] Update box.fir tests
ashermancinelli Oct 25, 2024
0d276b6
[tests] Update lit tests wrt memcpy change
ashermancinelli Oct 27, 2024
1de3879
update tests
ashermancinelli Oct 27, 2024
57965ad
update tests (all passing now)
ashermancinelli Oct 27, 2024
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
59 changes: 20 additions & 39 deletions flang/lib/Optimizer/CodeGen/CodeGen.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2949,9 +2949,10 @@ struct LoadOpConversion : public fir::FIROpConversion<fir::LoadOp> {
llvm::LogicalResult
matchAndRewrite(fir::LoadOp load, OpAdaptor adaptor,
mlir::ConversionPatternRewriter &rewriter) const override {

mlir::Type llvmLoadTy = convertObjectType(load.getType());
if (auto boxTy = mlir::dyn_cast<fir::BaseBoxType>(load.getType())) {
// fir.box is a special case because it is considered as an ssa values in
// fir.box is a special case because it is considered an ssa value in
// fir, but it is lowered as a pointer to a descriptor. So
// fir.ref<fir.box> and fir.box end up being the same llvm types and
// loading a fir.ref<fir.box> is implemented as taking a snapshot of the
Expand All @@ -2960,30 +2961,17 @@ struct LoadOpConversion : public fir::FIROpConversion<fir::LoadOp> {
mlir::Location loc = load.getLoc();
auto newBoxStorage =
genAllocaAndAddrCastWithType(loc, llvmLoadTy, defaultAlign, rewriter);
// TODO: always generate llvm.memcpy, LLVM is better at optimizing it than
// aggregate loads + stores.
if (boxTy.isAssumedRank()) {

TypePair boxTypePair{boxTy, llvmLoadTy};
mlir::Value boxSize =
computeBoxSize(loc, boxTypePair, inputBoxStorage, rewriter);
auto memcpy = rewriter.create<mlir::LLVM::MemcpyOp>(
loc, newBoxStorage, inputBoxStorage, boxSize, /*isVolatile=*/false);
if (std::optional<mlir::ArrayAttr> optionalTag = load.getTbaa())
memcpy.setTBAATags(*optionalTag);
else
attachTBAATag(memcpy, boxTy, boxTy, nullptr);
} else {
auto boxValue = rewriter.create<mlir::LLVM::LoadOp>(loc, llvmLoadTy,
inputBoxStorage);
if (std::optional<mlir::ArrayAttr> optionalTag = load.getTbaa())
boxValue.setTBAATags(*optionalTag);
else
attachTBAATag(boxValue, boxTy, boxTy, nullptr);
auto storeOp =
rewriter.create<mlir::LLVM::StoreOp>(loc, boxValue, newBoxStorage);
attachTBAATag(storeOp, boxTy, boxTy, nullptr);
}

TypePair boxTypePair{boxTy, llvmLoadTy};
mlir::Value boxSize =
computeBoxSize(loc, boxTypePair, inputBoxStorage, rewriter);
auto memcpy = rewriter.create<mlir::LLVM::MemcpyOp>(
loc, newBoxStorage, inputBoxStorage, boxSize, /*isVolatile=*/false);

if (std::optional<mlir::ArrayAttr> optionalTag = load.getTbaa())
memcpy.setTBAATags(*optionalTag);
else
attachTBAATag(memcpy, boxTy, boxTy, nullptr);
rewriter.replaceOp(load, newBoxStorage);
} else {
auto loadOp = rewriter.create<mlir::LLVM::LoadOp>(
Expand Down Expand Up @@ -3227,20 +3215,13 @@ struct StoreOpConversion : public fir::FIROpConversion<fir::StoreOp> {
mlir::LLVM::AliasAnalysisOpInterface newOp;
if (auto boxTy = mlir::dyn_cast<fir::BaseBoxType>(storeTy)) {
mlir::Type llvmBoxTy = lowerTy().convertBoxTypeAsStruct(boxTy);
// fir.box value is actually in memory, load it first before storing it,
// or do a memcopy for assumed-rank descriptors.
if (boxTy.isAssumedRank()) {
TypePair boxTypePair{boxTy, llvmBoxTy};
mlir::Value boxSize =
computeBoxSize(loc, boxTypePair, llvmValue, rewriter);
newOp = rewriter.create<mlir::LLVM::MemcpyOp>(
loc, llvmMemref, llvmValue, boxSize, /*isVolatile=*/false);
} else {
auto val =
rewriter.create<mlir::LLVM::LoadOp>(loc, llvmBoxTy, llvmValue);
attachTBAATag(val, boxTy, boxTy, nullptr);
newOp = rewriter.create<mlir::LLVM::StoreOp>(loc, val, llvmMemref);
}
// Always use memcpy because LLVM is not as effective at optimizing
// aggregate loads/stores as it is optimizing memcpy.
TypePair boxTypePair{boxTy, llvmBoxTy};
mlir::Value boxSize =
computeBoxSize(loc, boxTypePair, llvmValue, rewriter);
newOp = rewriter.create<mlir::LLVM::MemcpyOp>(
loc, llvmMemref, llvmValue, boxSize, /*isVolatile=*/false);
} else {
newOp = rewriter.create<mlir::LLVM::StoreOp>(loc, llvmValue, llvmMemref);
}
Expand Down
19 changes: 13 additions & 6 deletions flang/test/Fir/box.fir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -56,12 +56,14 @@ func.func @fa(%a : !fir.ref<!fir.array<100xf32>>) {
// CHECK-LABEL: define void @b1(
// CHECK-SAME: ptr %[[res:.*]], ptr %[[arg0:.*]], i64 %[[arg1:.*]])
func.func @b1(%arg0 : !fir.ref<!fir.char<1,?>>, %arg1 : index) -> !fir.box<!fir.char<1,?>> {
// CHECK: %[[alloca:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8 }
// CHECK: %[[size:.*]] = mul i64 ptrtoint (ptr getelementptr (i8, ptr null, i32 1) to i64), %[[arg1]]
// CHECK: insertvalue {{.*}} undef, i64 %[[size]], 1
// CHECK: insertvalue {{.*}} i32 20240719, 2
// CHECK: insertvalue {{.*}} ptr %[[arg0]], 0
%x = fir.embox %arg0 typeparams %arg1 : (!fir.ref<!fir.char<1,?>>, index) -> !fir.box<!fir.char<1,?>>
// CHECK: store {{.*}}, ptr %[[res]]
// CHECK: store {{.*}}, ptr %[[alloca]]
// CHECK: call void @llvm.memcpy.p0.p0.i32(ptr %[[res]], ptr %[[alloca]], i32 24, i1 false)
return %x : !fir.box<!fir.char<1,?>>
}

Expand All @@ -71,11 +73,13 @@ func.func @b1(%arg0 : !fir.ref<!fir.char<1,?>>, %arg1 : index) -> !fir.box<!fir.
// CHECK-SAME: ptr %[[arg0:.*]], i64 %[[arg1:.*]])
func.func @b2(%arg0 : !fir.ref<!fir.array<?x!fir.char<1,5>>>, %arg1 : index) -> !fir.box<!fir.array<?x!fir.char<1,5>>> {
%1 = fir.shape %arg1 : (index) -> !fir.shape<1>
// CHECK: %[[alloca:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] }
// CHECK: insertvalue {{.*}} { ptr undef, i64 ptrtoint (ptr getelementptr ([5 x i8], ptr null, i32 1) to i64), i32 20240719, i8 1, i8 40, i8 0, i8 0, {{.*}} }, i64 %[[arg1]], 7, 0, 1
// CHECK: insertvalue {{.*}} %{{.*}}, i64 ptrtoint (ptr getelementptr ([5 x i8], ptr null, i32 1) to i64), 7, 0, 2
// CHECK: insertvalue {{.*}} ptr %[[arg0]], 0
%2 = fir.embox %arg0(%1) : (!fir.ref<!fir.array<?x!fir.char<1,5>>>, !fir.shape<1>) -> !fir.box<!fir.array<?x!fir.char<1,5>>>
// CHECK: store {{.*}}, ptr %[[res]]
// CHECK: store {{.*}}, ptr %[[alloca]]
// CHECK: call void @llvm.memcpy.p0.p0.i32(ptr %[[res]], ptr %[[alloca]], i32 48, i1 false)
return %2 : !fir.box<!fir.array<?x!fir.char<1,5>>>
}

Expand All @@ -84,14 +88,16 @@ func.func @b2(%arg0 : !fir.ref<!fir.array<?x!fir.char<1,5>>>, %arg1 : index) ->
// CHECK-SAME: ptr %[[res:.*]], ptr %[[arg0:.*]], i64 %[[arg1:.*]], i64 %[[arg2:.*]])
func.func @b3(%arg0 : !fir.ref<!fir.array<?x!fir.char<1,?>>>, %arg1 : index, %arg2 : index) -> !fir.box<!fir.array<?x!fir.char<1,?>>> {
%1 = fir.shape %arg2 : (index) -> !fir.shape<1>
// CHECK: %[[alloca:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] }
// CHECK: %[[size:.*]] = mul i64 ptrtoint (ptr getelementptr (i8, ptr null, i32 1) to i64), %[[arg1]]
// CHECK: insertvalue {{.*}} i64 %[[size]], 1
// CHECK: insertvalue {{.*}} i32 20240719, 2
// CHECK: insertvalue {{.*}} i64 %[[arg2]], 7, 0, 1
// CHECK: insertvalue {{.*}} i64 %[[size]], 7, 0, 2
// CHECK: insertvalue {{.*}} ptr %[[arg0]], 0
%2 = fir.embox %arg0(%1) typeparams %arg1 : (!fir.ref<!fir.array<?x!fir.char<1,?>>>, !fir.shape<1>, index) -> !fir.box<!fir.array<?x!fir.char<1,?>>>
// CHECK: store {{.*}}, ptr %[[res]]
// CHECK: store {{.*}}, ptr %[[alloca]]
// CHECK: call void @llvm.memcpy.p0.p0.i32(ptr %[[res]], ptr %[[alloca]], i32 48, i1 false)
return %2 : !fir.box<!fir.array<?x!fir.char<1,?>>>
}

Expand All @@ -101,14 +107,16 @@ func.func @b3(%arg0 : !fir.ref<!fir.array<?x!fir.char<1,?>>>, %arg1 : index, %ar
func.func @b4(%arg0 : !fir.ref<!fir.array<7x!fir.char<1,?>>>, %arg1 : index) -> !fir.box<!fir.array<7x!fir.char<1,?>>> {
%c_7 = arith.constant 7 : index
%1 = fir.shape %c_7 : (index) -> !fir.shape<1>
// CHECK: %[[alloca:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] }
// CHECK: %[[size:.*]] = mul i64 ptrtoint (ptr getelementptr (i8, ptr null, i32 1) to i64), %[[arg1]]
// CHECK: insertvalue {{.*}} i64 %[[size]], 1
// CHECK: insertvalue {{.*}} i32 20240719, 2
// CHECK: insertvalue {{.*}} i64 7, 7, 0, 1
// CHECK: insertvalue {{.*}} i64 %[[size]], 7, 0, 2
// CHECK: insertvalue {{.*}} ptr %[[arg0]], 0
%x = fir.embox %arg0(%1) typeparams %arg1 : (!fir.ref<!fir.array<7x!fir.char<1,?>>>, !fir.shape<1>, index) -> !fir.box<!fir.array<7x!fir.char<1,?>>>
// CHECK: store {{.*}}, ptr %[[res]]
// CHECK: store {{.*}}, ptr %[[alloca]]
// CHECK: call void @llvm.memcpy.p0.p0.i32(ptr %[[res]], ptr %[[alloca]], i32 48, i1 false)
return %x : !fir.box<!fir.array<7x!fir.char<1,?>>>
}

Expand All @@ -117,8 +125,7 @@ func.func @b4(%arg0 : !fir.ref<!fir.array<7x!fir.char<1,?>>>, %arg1 : index) ->
// CHECK-SAME: ptr %[[arg0:.*]], ptr %[[arg1:.*]])
func.func @b5(%arg0 : !fir.ref<!fir.box<!fir.heap<!fir.array<?x?xf32>>>>, %arg1 : !fir.box<!fir.heap<!fir.array<?x?xf32>>>) {
fir.store %arg1 to %arg0 : !fir.ref<!fir.box<!fir.heap<!fir.array<?x?xf32>>>>
// CHECK: %[[boxLoad:.*]] = load { ptr, i64, i32, i8, i8, i8, i8, [2 x [3 x i64]] }, ptr %[[arg1]]
// CHECK: store { ptr, i64, i32, i8, i8, i8, i8, [2 x [3 x i64]] } %[[boxLoad]], ptr %[[arg0]]
// CHECK: call void @llvm.memcpy.p0.p0.i32(ptr %0, ptr %1, i32 72, i1 false)
return
}

Expand Down
4 changes: 2 additions & 2 deletions flang/test/Fir/convert-to-llvm-openmp-and-fir.fir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -799,8 +799,8 @@ func.func @_QPs(%arg0: !fir.ref<complex<f32>> {fir.bindc_name = "x"}) {
//CHECK: omp.parallel {
//CHECK: %[[CONST_1:.*]] = llvm.mlir.constant(1 : i32) : i32
//CHECK: %[[ALLOCA_1:.*]] = llvm.alloca %[[CONST_1:.*]] x !llvm.struct<(ptr, i64, i32, i8, i8, i8, i8)> {alignment = 8 : i64} : (i32) -> !llvm.ptr
//CHECK: %[[LOAD:.*]] = llvm.load %[[ALLOCA]] : !llvm.ptr -> !llvm.struct<(ptr, i64, i32, i8, i8, i8, i8)>
//CHECK: llvm.store %[[LOAD]], %[[ALLOCA_1]] : !llvm.struct<(ptr, i64, i32, i8, i8, i8, i8)>, !llvm.ptr
//CHECK: %[[SIZE:.*]] = llvm.mlir.constant(24 : i32) : i32
//CHECK: "llvm.intr.memcpy"(%[[ALLOCA_1]], %[[ALLOCA]], %[[SIZE]]) <{isVolatile = false}> : (!llvm.ptr, !llvm.ptr, i32) -> ()
//CHECK: %[[GEP:.*]] = llvm.getelementptr %[[ALLOCA_1]][0, 0] : (!llvm.ptr) -> !llvm.ptr
//CHECK: %[[LOAD_2:.*]] = llvm.load %[[GEP]] : !llvm.ptr -> !llvm.ptr
//CHECK: omp.terminator
Expand Down
28 changes: 15 additions & 13 deletions flang/test/Fir/convert-to-llvm.fir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -862,8 +862,8 @@ func.func @test_store_box(%array : !fir.ref<!fir.box<!fir.array<?x?xf32>>>, %box
// CHECK-LABEL: llvm.func @test_store_box
// CHECK-SAME: (%[[arg0:.*]]: !llvm.ptr,
// CHECK-SAME: %[[arg1:.*]]: !llvm.ptr) {
// CHECK-NEXT: %[[box_to_store:.*]] = llvm.load %arg1 : !llvm.ptr -> !llvm.struct<(ptr, i{{.*}}, i{{.*}}, i{{.*}}, i{{.*}}, i{{.*}}, i{{.*}}, array<2 x array<3 x i{{.*}}>>)>
// CHECK-NEXT: llvm.store %[[box_to_store]], %[[arg0]] : !llvm.struct<(ptr, i{{.*}}, i{{.*}}, i{{.*}}, i{{.*}}, i{{.*}}, i{{.*}}, array<2 x array<3 x i{{.*}}>>)>, !llvm.ptr
// CHECK-NEXT: %[[size:.*]] = llvm.mlir.constant(72 : i32) : i32
// CHECK-NEXT: "llvm.intr.memcpy"(%[[arg0]], %[[arg1]], %[[size]]) <{isVolatile = false}> : (!llvm.ptr, !llvm.ptr, i32) -> ()
// CHECK-NEXT: llvm.return
// CHECK-NEXT: }

Expand All @@ -875,15 +875,17 @@ func.func @store_unlimited_polymorphic_box(%arg0 : !fir.class<none>, %arg1 : !fi
fir.store %arg3 to %arg3r : !fir.ref<!fir.box<!fir.array<?xnone>>>
return
}
// CHECK-LABEL: llvm.func @store_unlimited_polymorphic_box(
// CHECK: %[[VAL_8:.*]] = llvm.load %{{.*}} : !llvm.ptr -> !llvm.struct<(ptr, i{{.*}}, i{{.*}}, i{{.*}}, i{{.*}}, i{{.*}}, i{{.*}}, ptr, array<1 x i{{.*}}>)>
// CHECK: llvm.store %[[VAL_8]], %{{.*}} : !llvm.struct<(ptr, i{{.*}}, i{{.*}}, i{{.*}}, i{{.*}}, i{{.*}}, i{{.*}}, ptr, array<1 x i{{.*}}>)>, !llvm.ptr
// CHECK: %[[VAL_9:.*]] = llvm.load %{{.*}} : !llvm.ptr -> !llvm.struct<(ptr, i{{.*}}, i{{.*}}, i{{.*}}, i{{.*}}, i{{.*}}, i{{.*}}, array<1 x array<3 x i{{.*}}>>, ptr, array<1 x i{{.*}}>)>
// CHECK: llvm.store %[[VAL_9]], %{{.*}} : !llvm.struct<(ptr, i{{.*}}, i{{.*}}, i{{.*}}, i{{.*}}, i{{.*}}, i{{.*}}, array<1 x array<3 x i{{.*}}>>, ptr, array<1 x i{{.*}}>)>, !llvm.ptr
// CHECK: %[[VAL_10:.*]] = llvm.load %{{.*}} : !llvm.ptr -> !llvm.struct<(ptr, i{{.*}}, i{{.*}}, i{{.*}}, i{{.*}}, i{{.*}}, i{{.*}}, ptr, array<1 x i{{.*}}>)>
// CHECK: llvm.store %[[VAL_10]], %{{.*}} : !llvm.struct<(ptr, i{{.*}}, i{{.*}}, i{{.*}}, i{{.*}}, i{{.*}}, i{{.*}}, ptr, array<1 x i{{.*}}>)>, !llvm.ptr
// CHECK: %[[VAL_11:.*]] = llvm.load %{{.*}}: !llvm.ptr
// CHECK: llvm.store %[[VAL_11]], %{{.*}} : !llvm.struct<(ptr, i{{.*}}, i{{.*}}, i{{.*}}, i{{.*}}, i{{.*}}, i{{.*}}, array<1 x array<3 x i{{.*}}>>, ptr, array<1 x i{{.*}}>)>, !llvm.ptr
// CHECK: llvm.func @store_unlimited_polymorphic_box(%[[VAL_0:.*]]: !llvm.ptr, %[[VAL_1:.*]]: !llvm.ptr, %[[VAL_2:.*]]: !llvm.ptr, %[[VAL_3:.*]]: !llvm.ptr, %[[VAL_4:.*]]: !llvm.ptr, %[[VAL_5:.*]]: !llvm.ptr, %[[VAL_6:.*]]: !llvm.ptr, %[[VAL_7:.*]]: !llvm.ptr) {
// CHECK: %[[VAL_8:.*]] = llvm.mlir.constant(40 : i32) : i32
// CHECK: "llvm.intr.memcpy"(%[[VAL_4]], %[[VAL_0]], %[[VAL_8]]) <{isVolatile = false}> : (!llvm.ptr, !llvm.ptr, i32) -> ()
// CHECK: %[[VAL_9:.*]] = llvm.mlir.constant(64 : i32) : i32
// CHECK: "llvm.intr.memcpy"(%[[VAL_5]], %[[VAL_1]], %[[VAL_9]]) <{isVolatile = false}> : (!llvm.ptr, !llvm.ptr, i32) -> ()
// CHECK: %[[VAL_10:.*]] = llvm.mlir.constant(40 : i32) : i32
// CHECK: "llvm.intr.memcpy"(%[[VAL_6]], %[[VAL_2]], %[[VAL_10]]) <{isVolatile = false}> : (!llvm.ptr, !llvm.ptr, i32) -> ()
// CHECK: %[[VAL_11:.*]] = llvm.mlir.constant(64 : i32) : i32
// CHECK: "llvm.intr.memcpy"(%[[VAL_7]], %[[VAL_3]], %[[VAL_11]]) <{isVolatile = false}> : (!llvm.ptr, !llvm.ptr, i32) -> ()
// CHECK: llvm.return
// CHECK: }


// -----
Expand Down Expand Up @@ -935,8 +937,8 @@ func.func @test_load_box(%addr : !fir.ref<!fir.box<!fir.array<10xf32>>>) {
// GENERIC-NEXT: %[[box_copy:.*]] = llvm.alloca %[[c1]] x !llvm.struct<([[DESC_TYPE:.*]])>
// AMDGPU-NEXT: %[[alloca_box_copy:.*]] = llvm.alloca %[[c1]] x !llvm.struct<([[DESC_TYPE:.*]])>{{.*}} : (i32) -> !llvm.ptr<5>
// AMDGPU-NEXT: %[[box_copy:.*]] = llvm.addrspacecast %[[alloca_box_copy]] : !llvm.ptr<5> to !llvm.ptr
// CHECK-NEXT: %[[box_val:.*]] = llvm.load %[[arg0]] : !llvm.ptr -> !llvm.struct<([[DESC_TYPE]])>
// CHECK-NEXT: llvm.store %[[box_val]], %[[box_copy]] : !llvm.struct<([[DESC_TYPE]])>, !llvm.ptr
// CHECK-NEXT: %[[size:.*]] = llvm.mlir.constant(48 : i32) : i32
// CHECK-NEXT: "llvm.intr.memcpy"(%[[box_copy]], %[[arg0]], %[[size]]) <{isVolatile = false}> : (!llvm.ptr, !llvm.ptr, i32) -> ()
// CHECK-NEXT: llvm.call @takes_box(%[[box_copy]]) : (!llvm.ptr) -> ()
// CHECK-NEXT: llvm.return
// CHECK-NEXT: }
Expand Down
Loading
Loading