Skip to content

Commit

Permalink
[CIR][CIRGen][Builtin][NFC] Refactor IntrinsicCallOp (#1056)
Browse files Browse the repository at this point in the history 
`IntrinsicCallOp` is now named `LLVMIntrinsicCallOp` to better reflect
its purpose. And now In CIR, we do not have "llvm" prefix which will be
added later during LLVMLowering.
  • Loading branch information
@ghehg
ghehg authored Nov 5, 2024
1 parent ea733a4 commit 5f1afad
Show file tree
Hide file tree
Showing 9 changed files with 207 additions and 227 deletions.
7 changes: 3 additions & 4 deletions clang/include/clang/CIR/Dialect/IR/CIROps.td
View file
Original file line number Diff line number Diff line change
Expand Up @@ -3445,11 +3445,11 @@ def FuncOp : CIR_Op<"func", [
}

//===----------------------------------------------------------------------===//
// IntrinsicCallOp
// LLVMIntrinsicCallOp
//===----------------------------------------------------------------------===//

def IntrinsicCallOp : CIR_Op<"llvm.intrinsic"> {
let summary = "Call to intrinsic functions that is not defined in CIR";
def LLVMIntrinsicCallOp : CIR_Op<"llvm.intrinsic"> {
let summary = "Call to llvm intrinsic functions that is not defined in CIR";
let description = [{
`cir.llvm.intrinsic` operation represents a call-like expression which has
return type and arguments that maps directly to a llvm intrinsic.
Expand All @@ -3476,7 +3476,6 @@ def IntrinsicCallOp : CIR_Op<"llvm.intrinsic"> {
}]>,
];

let hasVerifier = 1;
}

//===----------------------------------------------------------------------===//
Expand Down
77 changes: 37 additions & 40 deletions clang/lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -2127,7 +2127,7 @@ static mlir::Value buildArmLdrexNon128Intrinsic(unsigned int builtinID,
CIRGenFunction &cgf) {
StringRef intrinsicName;
if (builtinID == clang::AArch64::BI__builtin_arm_ldrex) {
intrinsicName = "llvm.aarch64.ldxr";
intrinsicName = "aarch64.ldxr";
} else {
llvm_unreachable("Unknown builtinID");
}
Expand All @@ -2141,8 +2141,9 @@ static mlir::Value buildArmLdrexNon128Intrinsic(unsigned int builtinID,
// which can be found under LLVM IR directory.
mlir::Type funcResTy = builder.getSInt64Ty();
mlir::Location loc = cgf.getLoc(clangCallExpr->getExprLoc());
mlir::cir::IntrinsicCallOp op = builder.create<mlir::cir::IntrinsicCallOp>(
loc, builder.getStringAttr(intrinsicName), funcResTy, loadAddr);
mlir::cir::LLVMIntrinsicCallOp op =
builder.create<mlir::cir::LLVMIntrinsicCallOp>(
loc, builder.getStringAttr(intrinsicName), funcResTy, loadAddr);
mlir::Value res = op.getResult();

// Convert result type to the expected type.
Expand Down Expand Up @@ -2269,7 +2270,7 @@ mlir::Value buildNeonCall(CIRGenBuilderTy &builder,
return nullptr;
}
return builder
.create<mlir::cir::IntrinsicCallOp>(
.create<mlir::cir::LLVMIntrinsicCallOp>(
loc, builder.getStringAttr(intrinsicName), funcResTy, args)
.getResult();
}
Expand Down Expand Up @@ -2379,27 +2380,27 @@ mlir::Value CIRGenFunction::buildCommonNeonBuiltinExpr(
return buildNeonCall(builder, {resTy, mulVecT, SInt32Ty}, ops,
(builtinID == NEON::BI__builtin_neon_vqdmulhq_lane_v ||
builtinID == NEON::BI__builtin_neon_vqdmulh_lane_v)
? "llvm.aarch64.neon.sqdmulh.lane"
: "llvm.aarch64.neon.sqrdmulh.lane",
? "aarch64.neon.sqdmulh.lane"
: "aarch64.neon.sqrdmulh.lane",
resTy, getLoc(e->getExprLoc()));
}
case NEON::BI__builtin_neon_vqshlu_n_v:
case NEON::BI__builtin_neon_vqshluq_n_v: {
// These intrinsics expect signed vector type as input, but
// return unsigned vector type.
mlir::cir::VectorType srcTy = getSignChangedVectorType(builder, vTy);
return buildNeonCall(
builder, {srcTy, srcTy}, ops, "llvm.aarch64.neon.sqshlu", vTy,
getLoc(e->getExprLoc()), false, /* not fp constrained op */
1, /* second arg is shift amount */
false /* leftshift */);
return buildNeonCall(builder, {srcTy, srcTy}, ops, "aarch64.neon.sqshlu",
vTy, getLoc(e->getExprLoc()),
false, /* not fp constrained op */
1, /* second arg is shift amount */
false /* leftshift */);
}
case NEON::BI__builtin_neon_vrshr_n_v:
case NEON::BI__builtin_neon_vrshrq_n_v: {
return buildNeonCall(
builder,
{vTy, isUnsigned ? getSignChangedVectorType(builder, vTy) : vTy}, ops,
isUnsigned ? "llvm.aarch64.neon.urshl" : "llvm.aarch64.neon.srshl", vTy,
isUnsigned ? "aarch64.neon.urshl" : "aarch64.neon.srshl", vTy,
getLoc(e->getExprLoc()), false, /* not fp constrained op*/
1, /* second arg is shift amount */
true /* rightshift */);
Expand Down Expand Up @@ -2440,7 +2441,7 @@ mlir::Value CIRGenFunction::buildCommonNeonBuiltinExpr(
ops[0] = builder.createBitcast(ops[0], ty);
ops[1] = builder.createBitcast(ops[1], ty);
ops[0] = builder.createAnd(ops[0], ops[1]);
// Note that during LLVM Lowering, result of `VecCmpOp` is sign extended,
// Note that during vmVM Lowering, result of `VecCmpOp` is sign extended,
// matching traditional codegen behavior.
return builder.create<mlir::cir::VecCmpOp>(
loc, ty, mlir::cir::CmpOpKind::ne, ops[0], builder.getZero(loc, ty));
Expand All @@ -2459,42 +2460,41 @@ mlir::Value CIRGenFunction::buildCommonNeonBuiltinExpr(
case NEON::BI__builtin_neon_vpadd_v:
case NEON::BI__builtin_neon_vpaddq_v: {
intrincsName = mlir::isa<mlir::FloatType>(vTy.getEltType())
? "llvm.aarch64.neon.faddp"
: "llvm.aarch64.neon.addp";
? "aarch64.neon.faddp"
: "aarch64.neon.addp";
break;
}
case NEON::BI__builtin_neon_vqadd_v:
case NEON::BI__builtin_neon_vqaddq_v: {
intrincsName = (intrinicId != altLLVMIntrinsic) ? "llvm.aarch64.neon.uqadd"
: "llvm.aarch64.neon.sqadd";
intrincsName = (intrinicId != altLLVMIntrinsic) ? "aarch64.neon.uqadd"
: "aarch64.neon.sqadd";
break;
}
case NEON::BI__builtin_neon_vqsub_v:
case NEON::BI__builtin_neon_vqsubq_v: {
intrincsName = (intrinicId != altLLVMIntrinsic) ? "llvm.aarch64.neon.uqsub"
: "llvm.aarch64.neon.sqsub";
intrincsName = (intrinicId != altLLVMIntrinsic) ? "aarch64.neon.uqsub"
: "aarch64.neon.sqsub";
break;
}
case NEON::BI__builtin_neon_vrhadd_v:
case NEON::BI__builtin_neon_vrhaddq_v: {
intrincsName = (intrinicId != altLLVMIntrinsic)
? "llvm.aarch64.neon.urhadd"
: "llvm.aarch64.neon.srhadd";
intrincsName = (intrinicId != altLLVMIntrinsic) ? "aarch64.neon.urhadd"
: "aarch64.neon.srhadd";
break;
}
case NEON::BI__builtin_neon_vshlq_v: {
intrincsName = (intrinicId != altLLVMIntrinsic) ? "llvm.aarch64.neon.ushl"
: "llvm.aarch64.neon.sshl";
intrincsName = (intrinicId != altLLVMIntrinsic) ? "aarch64.neon.ushl"
: "aarch64.neon.sshl";
break;
}
case NEON::BI__builtin_neon_vhadd_v:
case NEON::BI__builtin_neon_vhaddq_v: {
intrincsName = (intrinicId != altLLVMIntrinsic) ? "llvm.aarch64.neon.uhadd"
: "llvm.aarch64.neon.shadd";
intrincsName = (intrinicId != altLLVMIntrinsic) ? "aarch64.neon.uhadd"
: "aarch64.neon.shadd";
break;
}
case NEON::BI__builtin_neon_vqmovun_v: {
intrincsName = "llvm.aarch64.neon.sqxtun";
intrincsName = "aarch64.neon.sqxtun";
argTypes.push_back(builder.getExtendedOrTruncatedElementVectorType(
vTy, true /* extended */, true /* signed */));
break;
Expand Down Expand Up @@ -3259,10 +3259,9 @@ CIRGenFunction::buildAArch64BuiltinExpr(unsigned BuiltinID, const CallExpr *E,
llvm_unreachable("NYI");
}
case NEON::BI__builtin_neon_vmull_v: {
llvm::StringRef name =
usgn ? "llvm.aarch64.neon.umull" : "llvm.aarch64.neon.smull";
llvm::StringRef name = usgn ? "aarch64.neon.umull" : "aarch64.neon.smull";
if (Type.isPoly())
name = "llvm.aarch64.neon.pmull";
name = "aarch64.neon.pmull";
mlir::cir::VectorType argTy =
builder.getExtendedOrTruncatedElementVectorType(
ty, false /* truncated */, !usgn);
Expand All @@ -3277,10 +3276,9 @@ CIRGenFunction::buildAArch64BuiltinExpr(unsigned BuiltinID, const CallExpr *E,
}
case NEON::BI__builtin_neon_vmin_v:
case NEON::BI__builtin_neon_vminq_v: {
llvm::StringRef name =
usgn ? "llvm.aarch64.neon.umin" : "llvm.aarch64.neon.smin";
llvm::StringRef name = usgn ? "aarch64.neon.umin" : "aarch64.neon.smin";
if (mlir::cir::isFPOrFPVectorTy(ty))
name = "llvm.aarch64.neon.fmin";
name = "aarch64.neon.fmin";
return buildNeonCall(builder, {ty, ty}, Ops, name, ty,
getLoc(E->getExprLoc()));
}
Expand All @@ -3289,10 +3287,9 @@ CIRGenFunction::buildAArch64BuiltinExpr(unsigned BuiltinID, const CallExpr *E,
}
case NEON::BI__builtin_neon_vabd_v:
case NEON::BI__builtin_neon_vabdq_v: {
llvm::StringRef name =
usgn ? "llvm.aarch64.neon.uabd" : "llvm.aarch64.neon.sabd";
llvm::StringRef name = usgn ? "aarch64.neon.uabd" : "aarch64.neon.sabd";
if (mlir::cir::isFPOrFPVectorTy(ty))
name = "llvm.aarch64.neon.fabd";
name = "aarch64.neon.fabd";
return buildNeonCall(builder, {ty, ty}, Ops, name, ty,
getLoc(E->getExprLoc()));
}
Expand Down Expand Up @@ -3332,7 +3329,7 @@ CIRGenFunction::buildAArch64BuiltinExpr(unsigned BuiltinID, const CallExpr *E,
builder,
{builder.getExtendedOrTruncatedElementVectorType(ty, true, true),
SInt32Ty},
Ops, "llvm.aarch64.neon.sqrshrun", ty, getLoc(E->getExprLoc()));
Ops, "aarch64.neon.sqrshrun", ty, getLoc(E->getExprLoc()));
case NEON::BI__builtin_neon_vqshrn_n_v:
llvm_unreachable("NYI");
case NEON::BI__builtin_neon_vrshrn_n_v:
Expand All @@ -3342,7 +3339,7 @@ CIRGenFunction::buildAArch64BuiltinExpr(unsigned BuiltinID, const CallExpr *E,
vTy, true /* extend */,
mlir::cast<mlir::cir::IntType>(vTy.getEltType()).isSigned()),
SInt32Ty},
Ops, "llvm.aarch64.neon.rshrn", ty, getLoc(E->getExprLoc()));
Ops, "aarch64.neon.rshrn", ty, getLoc(E->getExprLoc()));
case NEON::BI__builtin_neon_vqrshrn_n_v:
llvm_unreachable("NYI");
case NEON::BI__builtin_neon_vrndah_f16: {
Expand All @@ -3351,7 +3348,7 @@ CIRGenFunction::buildAArch64BuiltinExpr(unsigned BuiltinID, const CallExpr *E,
case NEON::BI__builtin_neon_vrnda_v:
case NEON::BI__builtin_neon_vrndaq_v: {
assert(!MissingFeatures::buildConstrainedFPCall());
return buildNeonCall(builder, {ty}, Ops, "llvm.round", ty,
return buildNeonCall(builder, {ty}, Ops, "round", ty,
getLoc(E->getExprLoc()));
}
case NEON::BI__builtin_neon_vrndih_f16: {
Expand All @@ -3374,7 +3371,7 @@ CIRGenFunction::buildAArch64BuiltinExpr(unsigned BuiltinID, const CallExpr *E,
case NEON::BI__builtin_neon_vrndns_f32: {
mlir::Value arg0 = buildScalarExpr(E->getArg(0));
args.push_back(arg0);
return buildNeonCall(builder, {arg0.getType()}, args, "llvm.roundeven.f32",
return buildNeonCall(builder, {arg0.getType()}, args, "roundeven.f32",
getCIRGenModule().FloatTy, getLoc(E->getExprLoc()));
}
case NEON::BI__builtin_neon_vrndph_f16: {
Expand Down
6 changes: 0 additions & 6 deletions clang/lib/CIR/Dialect/IR/CIRDialect.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -2561,12 +2561,6 @@ LogicalResult mlir::cir::FuncOp::verifyType() {
return success();
}

LogicalResult mlir::cir::IntrinsicCallOp::verify() {
if (!getIntrinsicName().starts_with("llvm."))
return emitOpError() << "intrinsic name must start with 'llvm.'";
return success();
}

// Verifies linkage types
// - functions don't have 'common' linkage
// - external functions have 'external' or 'extern_weak' linkage
Expand Down
9 changes: 5 additions & 4 deletions clang/lib/CIR/Lowering/DirectToLLVM/LowerToLLVM.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -2981,12 +2981,13 @@ static mlir::LLVM::CallIntrinsicOp replaceOpWithCallLLVMIntrinsicOp(
}

class CIRIntrinsicCallLowering
: public mlir::OpConversionPattern<mlir::cir::IntrinsicCallOp> {
: public mlir::OpConversionPattern<mlir::cir::LLVMIntrinsicCallOp> {
public:
using OpConversionPattern<mlir::cir::IntrinsicCallOp>::OpConversionPattern;
using OpConversionPattern<
mlir::cir::LLVMIntrinsicCallOp>::OpConversionPattern;

mlir::LogicalResult
matchAndRewrite(mlir::cir::IntrinsicCallOp op, OpAdaptor adaptor,
matchAndRewrite(mlir::cir::LLVMIntrinsicCallOp op, OpAdaptor adaptor,
mlir::ConversionPatternRewriter &rewriter) const override {
mlir::Type llvmResTy =
getTypeConverter()->convertType(op->getResultTypes()[0]);
Expand All @@ -3003,7 +3004,7 @@ class CIRIntrinsicCallLowering
// TODO(cir): MLIR LLVM dialect should handle this part as CIR has no way
// to set LLVM IR attribute.
assert(!::cir::MissingFeatures::llvmIntrinsicElementTypeSupport());
replaceOpWithCallLLVMIntrinsicOp(rewriter, op, name, llvmResTy,
replaceOpWithCallLLVMIntrinsicOp(rewriter, op, "llvm." + name, llvmResTy,
adaptor.getOperands());
return mlir::success();
}
Expand Down
Loading

0 comments on commit 5f1afad

Please sign in to comment.