[flang,nfc] Initial changes needed to use llvm intrinsics instead of regular calls #134170
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
ashermancinelli
requested review from
razvanlupusoru,
tblah,
kiranchandramohan and
vzakhari
llvmbot
added
flang
|
@llvm/pr-subscribers-flang-fir-hlfir Author: Asher Mancinelli (ashermancinelli) ChangesFlang uses Full diff: https://github.com/llvm/llvm-project/pull/134170.diff 4 Files Affected:
diff --git a/flang/include/flang/Optimizer/Builder/LowLevelIntrinsics.h b/flang/include/flang/Optimizer/Builder/LowLevelIntrinsics.h
index 9be051632f93d..be106f7ea33b7 100644
--- a/flang/include/flang/Optimizer/Builder/LowLevelIntrinsics.h
+++ b/flang/include/flang/Optimizer/Builder/LowLevelIntrinsics.h
@@ -24,9 +24,6 @@ class FirOpBuilder;
namespace fir::factory {
-/// Get the LLVM intrinsic for `memcpy`. Use the 64 bit version.
-mlir::func::FuncOp getLlvmMemcpy(FirOpBuilder &builder);
-
/// Get the LLVM intrinsic for `memmove`. Use the 64 bit version.
mlir::func::FuncOp getLlvmMemmove(FirOpBuilder &builder);
diff --git a/flang/lib/Lower/ConvertExpr.cpp b/flang/lib/Lower/ConvertExpr.cpp
index b677a136a74aa..2d61c2ee8dd8e 100644
--- a/flang/lib/Lower/ConvertExpr.cpp
+++ b/flang/lib/Lower/ConvertExpr.cpp
@@ -6184,17 +6184,16 @@ class ArrayExprLowering {
/// Get the function signature of the LLVM memcpy intrinsic.
mlir::FunctionType memcpyType() {
- return fir::factory::getLlvmMemcpy(builder).getFunctionType();
+ auto ptrTy = mlir::LLVM::LLVMPointerType::get(builder.getContext());
+ llvm::SmallVector<mlir::Type> args = {ptrTy, ptrTy, builder.getI64Type()};
+ return mlir::FunctionType::get(builder.getContext(), args, std::nullopt);
}
/// Create a call to the LLVM memcpy intrinsic.
- void createCallMemcpy(llvm::ArrayRef<mlir::Value> args) {
+ void createCallMemcpy(llvm::ArrayRef<mlir::Value> args, bool isVolatile) {
mlir::Location loc = getLoc();
- mlir::func::FuncOp memcpyFunc = fir::factory::getLlvmMemcpy(builder);
- mlir::SymbolRefAttr funcSymAttr =
- builder.getSymbolRefAttr(memcpyFunc.getName());
- mlir::FunctionType funcTy = memcpyFunc.getFunctionType();
- builder.create<fir::CallOp>(loc, funcSymAttr, funcTy.getResults(), args);
+ builder.create<mlir::LLVM::MemcpyOp>(loc, args[0], args[1], args[2],
+ isVolatile);
}
// Construct code to check for a buffer overrun and realloc the buffer when
@@ -6306,9 +6305,8 @@ class ArrayExprLowering {
auto buff = builder.createConvert(loc, fir::HeapType::get(resTy), mem);
mlir::Value buffi = computeCoordinate(buff, off);
llvm::SmallVector<mlir::Value> args = fir::runtime::createArguments(
- builder, loc, memcpyType(), buffi, v.getAddr(), byteSz,
- /*volatile=*/builder.createBool(loc, false));
- createCallMemcpy(args);
+ builder, loc, memcpyType(), buffi, v.getAddr(), byteSz);
+ createCallMemcpy(args, /*isVolatile=*/false);
// Save the incremented buffer position.
builder.create<fir::StoreOp>(loc, endOff, buffPos);
@@ -6357,9 +6355,8 @@ class ArrayExprLowering {
builder.createConvert(loc, fir::HeapType::get(resTy), mem);
mlir::Value buffi = computeCoordinate(buff, off);
llvm::SmallVector<mlir::Value> args = fir::runtime::createArguments(
- builder, loc, memcpyType(), buffi, v.getAddr(), eleSz,
- /*volatile=*/builder.createBool(loc, false));
- createCallMemcpy(args);
+ builder, loc, memcpyType(), buffi, v.getAddr(), eleSz);
+ createCallMemcpy(args, /*isVolatile=*/false);
builder.create<fir::StoreOp>(loc, plusOne, buffPos);
}
diff --git a/flang/lib/Optimizer/Builder/LowLevelIntrinsics.cpp b/flang/lib/Optimizer/Builder/LowLevelIntrinsics.cpp
index 411a48614af6c..e8547cf2b1e1b 100644
--- a/flang/lib/Optimizer/Builder/LowLevelIntrinsics.cpp
+++ b/flang/lib/Optimizer/Builder/LowLevelIntrinsics.cpp
@@ -21,16 +21,6 @@
#include "flang/Optimizer/Builder/LowLevelIntrinsics.h"
#include "flang/Optimizer/Builder/FIRBuilder.h"
-mlir::func::FuncOp fir::factory::getLlvmMemcpy(fir::FirOpBuilder &builder) {
- auto ptrTy = builder.getRefType(builder.getIntegerType(8));
- llvm::SmallVector<mlir::Type> args = {ptrTy, ptrTy, builder.getI64Type(),
- builder.getI1Type()};
- auto memcpyTy =
- mlir::FunctionType::get(builder.getContext(), args, std::nullopt);
- return builder.createFunction(builder.getUnknownLoc(),
- "llvm.memcpy.p0.p0.i64", memcpyTy);
-}
-
mlir::func::FuncOp fir::factory::getLlvmMemmove(fir::FirOpBuilder &builder) {
auto ptrTy = builder.getRefType(builder.getIntegerType(8));
llvm::SmallVector<mlir::Type> args = {ptrTy, ptrTy, builder.getI64Type(),
diff --git a/flang/test/Lower/array-constructor-2.f90 b/flang/test/Lower/array-constructor-2.f90
index ae75a3b425202..c026c0673fbbd 100644
--- a/flang/test/Lower/array-constructor-2.f90
+++ b/flang/test/Lower/array-constructor-2.f90
@@ -78,12 +78,12 @@ end function test3c
! CHECK-DAG: %[[rep:.*]] = fir.convert %{{.*}} : (!fir.heap<f32>) -> !fir.ref<i8>
! CHECK-DAG: %[[res:.*]] = fir.convert %{{.*}} : (index) -> i64
! CHECK: %{{.*}} = fir.call @realloc(%[[rep]], %[[res]]) {{.*}}: (!fir.ref<i8>, i64) -> !fir.ref<i8>
- ! CHECK: fir.call @llvm.memcpy.p0.p0.i64(%{{.*}}, %{{.*}}, %{{.*}}, %false{{.*}}) {{.*}}: (!fir.ref<i8>, !fir.ref<i8>, i64, i1) -> ()
+ ! CHECK: "llvm.intr.memcpy"(%{{.*}}, %{{.*}}, %{{.*}}) <{isVolatile = false}> : (!llvm.ptr, !llvm.ptr, i64) -> ()
! CHECK: fir.call @_QPtest3c
! CHECK: fir.save_result
! CHECK: %[[tmp2:.*]] = fir.allocmem !fir.array<?xf32>, %{{.*}}#1 {uniq_name = ".array.expr"}
! CHECK: fir.call @realloc
- ! CHECK: fir.call @llvm.memcpy.p0.p0.i64(%
+ ! CHECK: "llvm.intr.memcpy"(%{{.*}}, %{{.*}}, %{{.*}}) <{isVolatile = false}> : (!llvm.ptr, !llvm.ptr, i64) -> ()
! CHECK: fir.array_coor %[[tmp:.*]](%{{.*}}) %{{.*}} : (!fir.heap<!fir.array<?xf32>>, !fir.shape<1>, index) -> !fir.ref<f32>
! CHECK-NEXT: fir.load
! CHECK-NEXT: fir.array_coor %arg0 %{{.*}} : (!fir.box<!fir.array<?xf32>>, index) -> !fir.ref<f32>
@@ -130,11 +130,11 @@ subroutine test5(a, array2)
! CHECK: %[[res:.*]] = fir.allocmem !fir.array<4xf32>
! CHECK: fir.address_of(@_QQro.2xr4.2) : !fir.ref<!fir.array<2xf32>>
! CHECK: %[[tmp1:.*]] = fir.allocmem !fir.array<2xf32>
- ! CHECK: fir.call @llvm.memcpy.p0.p0.i64(%{{.*}}, %{{.*}}, %{{.*}}, %false{{.*}}) {{.*}}: (!fir.ref<i8>, !fir.ref<i8>, i64, i1) -> ()
+ ! CHECK: "llvm.intr.memcpy"(%{{.*}}, %{{.*}}, %{{.*}}) <{isVolatile = false}> : (!llvm.ptr, !llvm.ptr, i64) -> ()
! CHECK: %[[tmp2:.*]] = fir.allocmem !fir.array<2xf32>
! CHECK: = fir.array_coor %[[array2]](%{{.*}}) %{{.*}} : (!fir.ref<!fir.array<2xf32>>, !fir.shape<1>, index) -> !fir.ref<f32>
! CHECK: = fir.array_coor %[[tmp2]](%{{.*}}) %{{.*}} : (!fir.heap<!fir.array<2xf32>>, !fir.shape<1>, index) -> !fir.ref<f32>
- ! CHECK: fir.call @llvm.memcpy.p0.p0.i64(%{{.*}}, %{{.*}}, %{{.*}}, %false{{.*}}) {{.*}}: (!fir.ref<i8>, !fir.ref<i8>, i64, i1) -> ()
+ ! CHECK: "llvm.intr.memcpy"(%{{.*}}, %{{.*}}, %{{.*}}) <{isVolatile = false}> : (!llvm.ptr, !llvm.ptr, i64) -> ()
! CHECK: = fir.array_coor %{{.*}}(%{{.*}}) %{{.*}} : (!fir.heap<!fir.array<4xf32>>, !fir.shape<1>, index) -> !fir.ref<f32>
! CHECK: = fir.array_coor %[[a]] %{{.*}} : (!fir.box<!fir.array<?xf32>>, index) -> !fir.ref<f32>
! CHECK-DAG: fir.freemem %{{.*}} : !fir.heap<!fir.array<4xf32>>
@@ -151,12 +151,12 @@ subroutine test6(c, d, e)
! CHECK: = fir.allocmem !fir.array<2x!fir.char<1,5>>
! CHECK: fir.call @realloc
! CHECK: %[[t:.*]] = fir.coordinate_of %{{.*}}, %{{.*}} : (!fir.heap<!fir.array<2x!fir.char<1,5>>>, index) -> !fir.ref<!fir.char<1,5>>
- ! CHECK: %[[to:.*]] = fir.convert %[[t]] : (!fir.ref<!fir.char<1,5>>) -> !fir.ref<i8>
- ! CHECK: fir.call @llvm.memcpy.p0.p0.i64(%[[to]], %{{.*}}, %{{.*}}, %false) {{.*}}: (!fir.ref<i8>, !fir.ref<i8>, i64, i1) -> ()
+ ! CHECK: %[[to:.*]] = fir.convert %[[t]] : (!fir.ref<!fir.char<1,5>>) -> !llvm.ptr
+ ! CHECK: "llvm.intr.memcpy"(%[[to]], %{{.*}}, %{{.*}}) <{isVolatile = false}> : (!llvm.ptr, !llvm.ptr, i64) -> ()
! CHECK: fir.call @realloc
! CHECK: %[[t:.*]] = fir.coordinate_of %{{.*}}, %{{.*}} : (!fir.heap<!fir.array<2x!fir.char<1,5>>>, index) -> !fir.ref<!fir.char<1,5>>
- ! CHECK: %[[to:.*]] = fir.convert %[[t]] : (!fir.ref<!fir.char<1,5>>) -> !fir.ref<i8>
- ! CHECK: fir.call @llvm.memcpy.p0.p0.i64(%[[to]], %{{.*}}, %{{.*}}, %false) {{.*}}: (!fir.ref<i8>, !fir.ref<i8>, i64, i1) -> ()
+ ! CHECK: %[[to:.*]] = fir.convert %[[t]] : (!fir.ref<!fir.char<1,5>>) -> !llvm.ptr
+ ! CHECK: "llvm.intr.memcpy"(%[[to]], %{{.*}}, %{{.*}}) <{isVolatile = false}> : (!llvm.ptr, !llvm.ptr, i64) -> ()
! CHECK: fir.freemem %{{.*}} : !fir.heap<!fir.array<2x!fir.char<1,5>>>
c = (/ d, e /)
end subroutine test6
|
clementval
approved these changes
Apr 2, 2025
ashermancinelli
changed the title
ashermancinelli
added a commit
that referenced
this pull request
Apr 4, 2025
Follow up to #134170. We should be using the LLVM intrinsics instead of plain fir.calls when we can. Existing code creates a declaration for the llvm intrinsic and a regular fir.call, which makes it hard for consumers of the IR to find all the intrinsic calls.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Flang uses
fir.call <llvm intrinsic>in a few places. This means consumers of the IR need to strcmp every fir.call if they want to find a particular LLVM intrinsic, which does not seem ideal given we have intrinsics available to us. I'd like to do the same for the rest of the functions used in this source file if folks are ok with this.