Skip to content

[MLIR][MemRefToLLVM] Fix atomic_rmw lowering for floats #2

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

Closed
google-labs-jules wants to merge 1 commit into from
Closed

[MLIR][MemRefToLLVM] Fix atomic_rmw lowering for floats #2

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
23 changes: 19 additions & 4 deletions mlir/lib/Conversion/MemRefToLLVM/MemRefToLLVM.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -747,21 +747,36 @@ struct GenericAtomicRMWOpLowering
// Append the cmpxchg op to the end of the loop block.
auto successOrdering = LLVM::AtomicOrdering::acq_rel;
auto failureOrdering = LLVM::AtomicOrdering::monotonic;

Value ptr = dataPtr;
Value cmp = loopArgument;
Value val = result;
if (auto floatType = dyn_cast<FloatType>(valueType)) {
unsigned width = floatType.getWidth();
Type intType = rewriter.getIntegerType(width);
cmp = rewriter.create<LLVM::BitcastOp>(loc, intType, cmp);
val = rewriter.create<LLVM::BitcastOp>(loc, intType, val);
}

auto cmpxchg =
LLVM::AtomicCmpXchgOp::create(rewriter, loc, dataPtr, loopArgument,
result, successOrdering, failureOrdering);
LLVM::AtomicCmpXchgOp::create(rewriter, loc, ptr, cmp,
val, successOrdering, failureOrdering);
// Extract the %new_loaded and %ok values from the pair.
Value newLoaded = LLVM::ExtractValueOp::create(rewriter, loc, cmpxchg, 0);
Value ok = LLVM::ExtractValueOp::create(rewriter, loc, cmpxchg, 1);

Value nextLoopArg = newLoaded;
if (isa<FloatType>(valueType))
nextLoopArg = rewriter.create<LLVM::BitcastOp>(loc, valueType, newLoaded);

// Conditionally branch to the end or back to the loop depending on %ok.
LLVM::CondBrOp::create(rewriter, loc, ok, endBlock, ArrayRef<Value>(),
loopBlock, newLoaded);
loopBlock, nextLoopArg);

rewriter.setInsertionPointToEnd(endBlock);

// The 'result' of the atomic_rmw op is the newly loaded value.
rewriter.replaceOp(atomicOp, {newLoaded});
rewriter.replaceOp(atomicOp, {nextLoopArg});

return success();
}
Expand Down
22 changes: 22 additions & 0 deletions mlir/test/Conversion/MemRefToLLVM/generic-atomic-rmw-flt.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,22 @@
// RUN: mlir-opt -convert-memref-to-llvm %s | FileCheck %s

// CHECK-LABEL: func @atomic_rmw_f32
func.func @atomic_rmw_f32(%mem : memref<f32>, %val : f32) {
// CHECK: %[[LOADED_F32:.*]] = llvm.load %{{.*}} : !llvm.ptr -> f32
// CHECK: llvm.br ^[[LOOP:.*]](%[[LOADED_F32]] : f32)

// CHECK: ^[[LOOP]](%[[ITER_F32:.*]]: f32):
// CHECK-NEXT: %[[ITER_I32:.*]] = llvm.bitcast %[[ITER_F32]] : f32 to i32
// CHECK-NEXT: %[[NEW_VAL_I32:.*]] = llvm.bitcast %{{.*}} : f32 to i32
// CHECK-NEXT: %[[RES:.*]] = llvm.cmpxchg %{{.*}}, %[[ITER_I32]], %[[NEW_VAL_I32]] acq_rel monotonic : !llvm.ptr, i32
// CHECK-NEXT: %[[NEW_LOADED_I32:.*]] = llvm.extractvalue %[[RES]][0]
// CHECK-NEXT: %[[OK:.*]] = llvm.extractvalue %[[RES]][1]
// CHECK-NEXT: %[[NEW_LOADED_F32:.*]] = llvm.bitcast %[[NEW_LOADED_I32]] : i32 to f32
// CHECK-NEXT: llvm.cond_br %[[OK]], ^[[END:.*]], ^[[LOOP]](%[[NEW_LOADED_F32]] : f32)

%x = memref.generic_atomic_rmw %mem[] : memref<f32> {
^bb0(%current_val: f32):
memref.atomic_yield %val : f32
}
return
}