Skip to content

Commit 647f02a

Browse files
AmrDeveloperAmrDeveloper
authored
[CIR] Implement SubOp for ComplexType (#148025)
This change adds support for SubOp for ComplexType #141365
1 parent d06e9ce commit 647f02a

File tree

5 files changed

+252
-0
lines changed

5 files changed

+252
-0
lines changed

‎clang/include/clang/CIR/Dialect/IR/CIROps.td‎

Lines changed: 27 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -2668,6 +2668,33 @@ def ComplexAddOp : CIR_Op<"complex.add", [Pure, SameOperandsAndResultType]> {
26682668
}];
26692669
}
26702670

2671+
//===----------------------------------------------------------------------===//
2672+
// ComplexSubOp
2673+
//===----------------------------------------------------------------------===//
2674+
2675+
def ComplexSubOp : CIR_Op<"complex.sub", [Pure, SameOperandsAndResultType]> {
2676+
let summary = "Complex subtraction";
2677+
let description = [{
2678+
The `cir.complex.sub` operation takes two complex numbers and returns
2679+
their difference.
2680+
2681+
Example:
2682+
2683+
```mlir
2684+
%2 = cir.complex.sub %0, %1 : !cir.complex<!cir.float>
2685+
```
2686+
}];
2687+
2688+
let arguments = (ins CIR_ComplexType:$lhs, CIR_ComplexType:$rhs);
2689+
2690+
let results = (outs CIR_ComplexType:$result);
2691+
2692+
let assemblyFormat = [{
2693+
$lhs `,` $rhs `:` qualified(type($result)) attr-dict
2694+
}];
2695+
}
2696+
2697+
26712698
//===----------------------------------------------------------------------===//
26722699
// Bit Manipulation Operations
26732700
//===----------------------------------------------------------------------===//

‎clang/lib/CIR/CodeGen/CIRGenExprComplex.cpp‎

Lines changed: 9 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -74,6 +74,7 @@ class ComplexExprEmitter : public StmtVisitor<ComplexExprEmitter, mlir::Value> {
7474
mlir::Value emitPromotedComplexOperand(const Expr *e, QualType promotionTy);
7575

7676
mlir::Value emitBinAdd(const BinOpInfo &op);
77+
mlir::Value emitBinSub(const BinOpInfo &op);
7778

7879
QualType getPromotionType(QualType ty, bool isDivOpCode = false) {
7980
if (auto *complexTy = ty->getAs<ComplexType>()) {
@@ -105,6 +106,7 @@ class ComplexExprEmitter : public StmtVisitor<ComplexExprEmitter, mlir::Value> {
105106
}
106107

107108
HANDLEBINOP(Add)
109+
HANDLEBINOP(Sub)
108110
#undef HANDLEBINOP
109111
};
110112
} // namespace
@@ -345,6 +347,7 @@ mlir::Value ComplexExprEmitter::emitPromoted(const Expr *e,
345347
case BO_##OP: \
346348
return emitBin##OP(emitBinOps(bo, promotionTy));
347349
HANDLE_BINOP(Add)
350+
HANDLE_BINOP(Sub)
348351
#undef HANDLE_BINOP
349352
default:
350353
break;
@@ -390,6 +393,12 @@ mlir::Value ComplexExprEmitter::emitBinAdd(const BinOpInfo &op) {
390393
return builder.create<cir::ComplexAddOp>(op.loc, op.lhs, op.rhs);
391394
}
392395

396+
mlir::Value ComplexExprEmitter::emitBinSub(const BinOpInfo &op) {
397+
assert(!cir::MissingFeatures::fastMathFlags());
398+
assert(!cir::MissingFeatures::cgFPOptionsRAII());
399+
return builder.create<cir::ComplexSubOp>(op.loc, op.lhs, op.rhs);
400+
}
401+
393402
LValue CIRGenFunction::emitComplexAssignmentLValue(const BinaryOperator *e) {
394403
assert(e->getOpcode() == BO_Assign && "Expected assign op");
395404

‎clang/lib/CIR/Lowering/DirectToLLVM/LowerToLLVM.cpp‎

Lines changed: 50 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -2069,6 +2069,7 @@ void ConvertCIRToLLVMPass::runOnOperation() {
20692069
CIRToLLVMComplexImagOpLowering,
20702070
CIRToLLVMComplexRealOpLowering,
20712071
CIRToLLVMComplexRealPtrOpLowering,
2072+
CIRToLLVMComplexSubOpLowering,
20722073
CIRToLLVMConstantOpLowering,
20732074
CIRToLLVMExpectOpLowering,
20742075
CIRToLLVMFuncOpLowering,
@@ -2452,6 +2453,55 @@ mlir::LogicalResult CIRToLLVMComplexRealOpLowering::matchAndRewrite(
24522453
return mlir::success();
24532454
}
24542455

2456+
mlir::LogicalResult CIRToLLVMComplexSubOpLowering::matchAndRewrite(
2457+
cir::ComplexSubOp op, OpAdaptor adaptor,
2458+
mlir::ConversionPatternRewriter &rewriter) const {
2459+
mlir::Value lhs = adaptor.getLhs();
2460+
mlir::Value rhs = adaptor.getRhs();
2461+
mlir::Location loc = op.getLoc();
2462+
2463+
auto complexType = mlir::cast<cir::ComplexType>(op.getLhs().getType());
2464+
mlir::Type complexElemTy =
2465+
getTypeConverter()->convertType(complexType.getElementType());
2466+
auto lhsReal =
2467+
rewriter.create<mlir::LLVM::ExtractValueOp>(loc, complexElemTy, lhs, 0);
2468+
auto lhsImag =
2469+
rewriter.create<mlir::LLVM::ExtractValueOp>(loc, complexElemTy, lhs, 1);
2470+
auto rhsReal =
2471+
rewriter.create<mlir::LLVM::ExtractValueOp>(loc, complexElemTy, rhs, 0);
2472+
auto rhsImag =
2473+
rewriter.create<mlir::LLVM::ExtractValueOp>(loc, complexElemTy, rhs, 1);
2474+
2475+
mlir::Value newReal;
2476+
mlir::Value newImag;
2477+
if (complexElemTy.isInteger()) {
2478+
newReal = rewriter.create<mlir::LLVM::SubOp>(loc, complexElemTy, lhsReal,
2479+
rhsReal);
2480+
newImag = rewriter.create<mlir::LLVM::SubOp>(loc, complexElemTy, lhsImag,
2481+
rhsImag);
2482+
} else {
2483+
assert(!cir::MissingFeatures::fastMathFlags());
2484+
assert(!cir::MissingFeatures::fpConstraints());
2485+
newReal = rewriter.create<mlir::LLVM::FSubOp>(loc, complexElemTy, lhsReal,
2486+
rhsReal);
2487+
newImag = rewriter.create<mlir::LLVM::FSubOp>(loc, complexElemTy, lhsImag,
2488+
rhsImag);
2489+
}
2490+
2491+
mlir::Type complexLLVMTy =
2492+
getTypeConverter()->convertType(op.getResult().getType());
2493+
auto initialComplex =
2494+
rewriter.create<mlir::LLVM::PoisonOp>(op->getLoc(), complexLLVMTy);
2495+
2496+
auto realComplex = rewriter.create<mlir::LLVM::InsertValueOp>(
2497+
op->getLoc(), initialComplex, newReal, 0);
2498+
2499+
rewriter.replaceOpWithNewOp<mlir::LLVM::InsertValueOp>(op, realComplex,
2500+
newImag, 1);
2501+
2502+
return mlir::success();
2503+
}
2504+
24552505
mlir::LogicalResult CIRToLLVMComplexImagOpLowering::matchAndRewrite(
24562506
cir::ComplexImagOp op, OpAdaptor adaptor,
24572507
mlir::ConversionPatternRewriter &rewriter) const {

‎clang/lib/CIR/Lowering/DirectToLLVM/LowerToLLVM.h‎

Lines changed: 10 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -553,6 +553,16 @@ class CIRToLLVMComplexAddOpLowering
553553
mlir::ConversionPatternRewriter &) const override;
554554
};
555555

556+
class CIRToLLVMComplexSubOpLowering
557+
: public mlir::OpConversionPattern<cir::ComplexSubOp> {
558+
public:
559+
using mlir::OpConversionPattern<cir::ComplexSubOp>::OpConversionPattern;
560+
561+
mlir::LogicalResult
562+
matchAndRewrite(cir::ComplexSubOp op, OpAdaptor,
563+
mlir::ConversionPatternRewriter &) const override;
564+
};
565+
556566
class CIRToLLVMSetBitfieldOpLowering
557567
: public mlir::OpConversionPattern<cir::SetBitfieldOp> {
558568
public:

‎clang/test/CIR/CodeGen/complex-arithmetic.cpp‎

Lines changed: 156 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -158,3 +158,159 @@ void foo3() {
158158
// OGCG: %[[RESULT_IMAG_PTR:.*]] = getelementptr inbounds nuw { float, float }, ptr %[[RESULT]], i32 0, i32 1
159159
// OGCG: store float %[[ADD_REAL_A_B_C]], ptr %[[RESULT_REAL_PTR]], align 4
160160
// OGCG: store float %[[ADD_IMAG_A_B_C]], ptr %[[RESULT_IMAG_PTR]], align 4
161+
162+
void foo4() {
163+
int _Complex a;
164+
int _Complex b;
165+
int _Complex c = a - b;
166+
}
167+
168+
// CIR: %[[COMPLEX_A:.*]] = cir.alloca !cir.complex<!s32i>, !cir.ptr<!cir.complex<!s32i>>, ["a"]
169+
// CIR: %[[COMPLEX_B:.*]] = cir.alloca !cir.complex<!s32i>, !cir.ptr<!cir.complex<!s32i>>, ["b"]
170+
// CIR: %[[TMP_A:.*]] = cir.load{{.*}} %[[COMPLEX_A]] : !cir.ptr<!cir.complex<!s32i>>, !cir.complex<!s32i>
171+
// CIR: %[[TMP_B:.*]] = cir.load{{.*}} %[[COMPLEX_B]] : !cir.ptr<!cir.complex<!s32i>>, !cir.complex<!s32i>
172+
// CIR: %[[SUB:.*]] = cir.complex.sub %[[TMP_A]], %[[TMP_B]] : !cir.complex<!s32i>
173+
174+
// LLVM: %[[COMPLEX_A:.*]] = alloca { i32, i32 }, i64 1, align 4
175+
// LLVM: %[[COMPLEX_B:.*]] = alloca { i32, i32 }, i64 1, align 4
176+
// LLVM: %[[COMPLEX_C:.*]] = alloca { i32, i32 }, i64 1, align 4
177+
// LLVM: %[[TMP_A:.*]] = load { i32, i32 }, ptr %[[COMPLEX_A]], align 4
178+
// LLVM: %[[TMP_B:.*]] = load { i32, i32 }, ptr %[[COMPLEX_B]], align 4
179+
// LLVM: %[[A_REAL:.*]] = extractvalue { i32, i32 } %[[TMP_A]], 0
180+
// LLVM: %[[A_IMAG:.*]] = extractvalue { i32, i32 } %[[TMP_A]], 1
181+
// LLVM: %[[B_REAL:.*]] = extractvalue { i32, i32 } %[[TMP_B]], 0
182+
// LLVM: %[[B_IMAG:.*]] = extractvalue { i32, i32 } %[[TMP_B]], 1
183+
// LLVM: %[[SUB_REAL:.*]] = sub i32 %[[A_REAL]], %[[B_REAL]]
184+
// LLVM: %[[SUB_IMAG:.*]] = sub i32 %[[A_IMAG]], %[[B_IMAG]]
185+
// LLVM: %[[RESULT:.*]] = insertvalue { i32, i32 } poison, i32 %[[SUB_REAL]], 0
186+
// LLVM: %[[RESULT_2:.*]] = insertvalue { i32, i32 } %[[RESULT]], i32 %[[SUB_IMAG]], 1
187+
// LLVM: store { i32, i32 } %[[RESULT_2]], ptr %[[COMPLEX_C]], align 4
188+
189+
// OGCG: %[[COMPLEX_A:.*]] = alloca { i32, i32 }, align 4
190+
// OGCG: %[[COMPLEX_B:.*]] = alloca { i32, i32 }, align 4
191+
// OGCG: %[[RESULT:.*]] = alloca { i32, i32 }, align 4
192+
// OGCG: %[[A_REAL_PTR:.*]] = getelementptr inbounds nuw { i32, i32 }, ptr %[[COMPLEX_A]], i32 0, i32 0
193+
// OGCG: %[[A_REAL:.*]] = load i32, ptr %[[A_REAL_PTR]], align 4
194+
// OGCG: %[[A_IMAG_PTR:.*]] = getelementptr inbounds nuw { i32, i32 }, ptr %[[COMPLEX_A]], i32 0, i32 1
195+
// OGCG: %[[A_IMAG:.*]] = load i32, ptr %[[A_IMAG_PTR]], align 4
196+
// OGCG: %[[B_REAL_PTR:.*]] = getelementptr inbounds nuw { i32, i32 }, ptr %[[COMPLEX_B]], i32 0, i32 0
197+
// OGCG: %[[B_REAL:.*]] = load i32, ptr %[[B_REAL_PTR]], align 4
198+
// OGCG: %[[B_IMAG_PTR:.*]] = getelementptr inbounds nuw { i32, i32 }, ptr %[[COMPLEX_B]], i32 0, i32 1
199+
// OGCG: %[[B_IMAG:.*]] = load i32, ptr %[[B_IMAG_PTR]], align 4
200+
// OGCG: %[[SUB_REAL:.*]] = sub i32 %[[A_REAL]], %[[B_REAL]]
201+
// OGCG: %[[SUB_IMAG:.*]] = sub i32 %[[A_IMAG]], %[[B_IMAG]]
202+
// OGCG: %[[RESULT_REAL_PTR:.*]] = getelementptr inbounds nuw { i32, i32 }, ptr %[[RESULT]], i32 0, i32 0
203+
// OGCG: %[[RESULT_IMAG_PTR:.*]] = getelementptr inbounds nuw { i32, i32 }, ptr %[[RESULT]], i32 0, i32 1
204+
// OGCG: store i32 %[[SUB_REAL]], ptr %[[RESULT_REAL_PTR]], align 4
205+
// OGCG: store i32 %[[SUB_IMAG]], ptr %[[RESULT_IMAG_PTR]], align 4
206+
207+
void foo5() {
208+
float _Complex a;
209+
float _Complex b;
210+
float _Complex c = a - b;
211+
}
212+
213+
// CIR: %[[COMPLEX_A:.*]] = cir.alloca !cir.complex<!cir.float>, !cir.ptr<!cir.complex<!cir.float>>, ["a"]
214+
// CIR: %[[COMPLEX_B:.*]] = cir.alloca !cir.complex<!cir.float>, !cir.ptr<!cir.complex<!cir.float>>, ["b"]
215+
// CIR: %[[TMP_A:.*]] = cir.load{{.*}} %[[COMPLEX_A]] : !cir.ptr<!cir.complex<!cir.float>>, !cir.complex<!cir.float>
216+
// CIR: %[[TMP_B:.*]] = cir.load{{.*}} %[[COMPLEX_B]] : !cir.ptr<!cir.complex<!cir.float>>, !cir.complex<!cir.float>
217+
// CIR: %[[SUB:.*]] = cir.complex.sub %[[TMP_A]], %[[TMP_B]] : !cir.complex<!cir.float>
218+
219+
// LLVM: %[[COMPLEX_A:.*]] = alloca { float, float }, i64 1, align 4
220+
// LLVM: %[[COMPLEX_B:.*]] = alloca { float, float }, i64 1, align 4
221+
// LLVM: %[[TMP_A:.*]] = load { float, float }, ptr %[[COMPLEX_A]], align 4
222+
// LLVM: %[[TMP_B:.*]] = load { float, float }, ptr %[[COMPLEX_B]], align 4
223+
// LLVM: %[[A_REAL:.*]] = extractvalue { float, float } %[[TMP_A]], 0
224+
// LLVM: %[[A_IMAG:.*]] = extractvalue { float, float } %[[TMP_A]], 1
225+
// LLVM: %[[B_REAL:.*]] = extractvalue { float, float } %[[TMP_B]], 0
226+
// LLVM: %[[B_IMAG:.*]] = extractvalue { float, float } %[[TMP_B]], 1
227+
// LLVM: %[[SUB_REAL:.*]] = fsub float %[[A_REAL]], %[[B_REAL]]
228+
// LLVM: %[[SUB_IMAG:.*]] = fsub float %[[A_IMAG]], %[[B_IMAG]]
229+
// LLVM: %[[RESULT:.*]] = insertvalue { float, float } poison, float %[[SUB_REAL]], 0
230+
// LLVM: %[[RESULT_2:.*]] = insertvalue { float, float } %[[RESULT]], float %[[SUB_IMAG]], 1
231+
232+
// OGCG: %[[COMPLEX_A:.*]] = alloca { float, float }, align 4
233+
// OGCG: %[[COMPLEX_B:.*]] = alloca { float, float }, align 4
234+
// OGCG: %[[RESULT:.*]] = alloca { float, float }, align 4
235+
// OGCG: %[[A_REAL_PTR:.*]] = getelementptr inbounds nuw { float, float }, ptr %[[COMPLEX_A]], i32 0, i32 0
236+
// OGCG: %[[A_REAL:.*]] = load float, ptr %[[A_REAL_PTR]], align 4
237+
// OGCG: %[[A_IMAG_PTR:.*]] = getelementptr inbounds nuw { float, float }, ptr %[[COMPLEX_A]], i32 0, i32 1
238+
// OGCG: %[[A_IMAG:.*]] = load float, ptr %[[A_IMAG_PTR]], align 4
239+
// OGCG: %[[B_REAL_PTR:.*]] = getelementptr inbounds nuw { float, float }, ptr %[[COMPLEX_B]], i32 0, i32 0
240+
// OGCG: %[[B_REAL:.*]] = load float, ptr %[[B_REAL_PTR]], align 4
241+
// OGCG: %[[B_IMAG_PTR:.*]] = getelementptr inbounds nuw { float, float }, ptr %[[COMPLEX_B]], i32 0, i32 1
242+
// OGCG: %[[B_IMAG:.*]] = load float, ptr %[[B_IMAG_PTR]], align 4
243+
// OGCG: %[[SUB_REAL:.*]] = fsub float %[[A_REAL]], %[[B_REAL]]
244+
// OGCG: %[[SUB_IMAG:.*]] = fsub float %[[A_IMAG]], %[[B_IMAG]]
245+
// OGCG: %[[RESULT_REAL_PTR:.*]] = getelementptr inbounds nuw { float, float }, ptr %[[RESULT]], i32 0, i32 0
246+
// OGCG: %[[RESULT_IMAG_PTR:.*]] = getelementptr inbounds nuw { float, float }, ptr %[[RESULT]], i32 0, i32 1
247+
// OGCG: store float %[[SUB_REAL]], ptr %[[RESULT_REAL_PTR]], align 4
248+
// OGCG: store float %[[SUB_IMAG]], ptr %[[RESULT_IMAG_PTR]], align 4
249+
250+
void foo6() {
251+
float _Complex a;
252+
float _Complex b;
253+
float _Complex c;
254+
float _Complex d = (a - b) - c;
255+
}
256+
257+
// CIR: %[[COMPLEX_A:.*]] = cir.alloca !cir.complex<!cir.float>, !cir.ptr<!cir.complex<!cir.float>>, ["a"]
258+
// CIR: %[[COMPLEX_B:.*]] = cir.alloca !cir.complex<!cir.float>, !cir.ptr<!cir.complex<!cir.float>>, ["b"]
259+
// CIR: %[[COMPLEX_C:.*]] = cir.alloca !cir.complex<!cir.float>, !cir.ptr<!cir.complex<!cir.float>>, ["c"]
260+
// CIR: %[[RESULT:.*]] = cir.alloca !cir.complex<!cir.float>, !cir.ptr<!cir.complex<!cir.float>>, ["d", init]
261+
// CIR: %[[TMP_A:.*]] = cir.load{{.*}} %[[COMPLEX_A]] : !cir.ptr<!cir.complex<!cir.float>>, !cir.complex<!cir.float>
262+
// CIR: %[[TMP_B:.*]] = cir.load{{.*}} %[[COMPLEX_B]] : !cir.ptr<!cir.complex<!cir.float>>, !cir.complex<!cir.float>
263+
// CIR: %[[SUB_A_B:.*]] = cir.complex.sub %[[TMP_A]], %[[TMP_B]] : !cir.complex<!cir.float>
264+
// CIR: %[[TMP_C:.*]] = cir.load{{.*}} %[[COMPLEX_C]] : !cir.ptr<!cir.complex<!cir.float>>, !cir.complex<!cir.float>
265+
// CIR: %[[SUB_A_B_C:.*]] = cir.complex.sub %[[SUB_A_B]], %[[TMP_C]] : !cir.complex<!cir.float>
266+
// CIR: cir.store{{.*}} %[[SUB_A_B_C]], %[[RESULT]] : !cir.complex<!cir.float>, !cir.ptr<!cir.complex<!cir.float>>
267+
268+
// LLVM: %[[COMPLEX_A:.*]] = alloca { float, float }, i64 1, align 4
269+
// LLVM: %[[COMPLEX_B:.*]] = alloca { float, float }, i64 1, align 4
270+
// LLVM: %[[COMPLEX_C:.*]] = alloca { float, float }, i64 1, align 4
271+
// LLVM: %[[RESULT:.*]] = alloca { float, float }, i64 1, align 4
272+
// LLVM: %[[TMP_A:.*]] = load { float, float }, ptr %[[COMPLEX_A]], align 4
273+
// LLVM: %[[TMP_B:.*]] = load { float, float }, ptr %[[COMPLEX_B]], align 4
274+
// LLVM: %[[A_REAL:.*]] = extractvalue { float, float } %[[TMP_A]], 0
275+
// LLVM: %[[A_IMAG:.*]] = extractvalue { float, float } %[[TMP_A]], 1
276+
// LLVM: %[[B_REAL:.*]] = extractvalue { float, float } %[[TMP_B]], 0
277+
// LLVM: %[[B_IMAG:.*]] = extractvalue { float, float } %[[TMP_B]], 1
278+
// LLVM: %[[SUB_REAL_A_B:.*]] = fsub float %[[A_REAL]], %[[B_REAL]]
279+
// LLVM: %[[SUB_IMAG_A_B:.*]] = fsub float %[[A_IMAG]], %[[B_IMAG]]
280+
// LLVM: %[[A_B:.*]] = insertvalue { float, float } poison, float %[[SUB_REAL_A_B]], 0
281+
// LLVM: %[[TMP_A_B:.*]] = insertvalue { float, float } %[[A_B]], float %[[SUB_IMAG_A_B]], 1
282+
// LLVM: %[[TMP_C:.*]] = load { float, float }, ptr %[[COMPLEX_C]], align 4
283+
// LLVM: %[[A_B_REAL:.*]] = extractvalue { float, float } %[[TMP_A_B]], 0
284+
// LLVM: %[[A_B_IMAG:.*]] = extractvalue { float, float } %[[TMP_A_B]], 1
285+
// LLVM: %[[C_REAL:.*]] = extractvalue { float, float } %[[TMP_C]], 0
286+
// LLVM: %[[C_IMAG:.*]] = extractvalue { float, float } %[[TMP_C]], 1
287+
// LLVM: %[[SUB_REAL_A_B_C:.*]] = fsub float %[[A_B_REAL]], %[[C_REAL]]
288+
// LLVM: %[[SUB_IMAG_A_B_C:.*]] = fsub float %[[A_B_IMAG]], %[[C_IMAG]]
289+
// LLVM: %[[A_B_C:.*]] = insertvalue { float, float } poison, float %[[SUB_REAL_A_B_C]], 0
290+
// LLVM: %[[TMP_A_B_C:.*]] = insertvalue { float, float } %[[A_B_C]], float %[[SUB_IMAG_A_B_C]], 1
291+
// LLVM: store { float, float } %[[TMP_A_B_C]], ptr %[[RESULT]], align 4
292+
293+
// OGCG: %[[COMPLEX_A:.*]] = alloca { float, float }, align 4
294+
// OGCG: %[[COMPLEX_B:.*]] = alloca { float, float }, align 4
295+
// OGCG: %[[COMPLEX_C:.*]] = alloca { float, float }, align 4
296+
// OGCG: %[[RESULT:.*]] = alloca { float, float }, align 4
297+
// OGCG: %[[A_REAL_PTR:.*]] = getelementptr inbounds nuw { float, float }, ptr %[[COMPLEX_A]], i32 0, i32 0
298+
// OGCG: %[[A_REAL:.*]] = load float, ptr %[[A_REAL_PTR]], align 4
299+
// OGCG: %[[A_IMAG_PTR:.*]] = getelementptr inbounds nuw { float, float }, ptr %[[COMPLEX_A]], i32 0, i32 1
300+
// OGCG: %[[A_IMAG:.*]] = load float, ptr %[[A_IMAG_PTR]], align 4
301+
// OGCG: %[[B_REAL_PTR:.*]] = getelementptr inbounds nuw { float, float }, ptr %[[COMPLEX_B]], i32 0, i32 0
302+
// OGCG: %[[B_REAL:.*]] = load float, ptr %[[B_REAL_PTR]], align 4
303+
// OGCG: %[[B_IMAG_PTR:.*]] = getelementptr inbounds nuw { float, float }, ptr %[[COMPLEX_B]], i32 0, i32 1
304+
// OGCG: %[[B_IMAG:.*]] = load float, ptr %[[B_IMAG_PTR]], align 4
305+
// OGCG: %[[SUB_REAL_A_B:.*]] = fsub float %[[A_REAL]], %[[B_REAL]]
306+
// OGCG: %[[SUB_IMAG_A_B:.*]] = fsub float %[[A_IMAG]], %[[B_IMAG]]
307+
// OGCG: %[[C_REAL_PTR:.*]] = getelementptr inbounds nuw { float, float }, ptr %[[COMPLEX_C]], i32 0, i32 0
308+
// OGCG: %[[C_REAL:.*]] = load float, ptr %[[C_REAL_PTR]], align 4
309+
// OGCG: %[[C_IMAG_PTR:.*]] = getelementptr inbounds nuw { float, float }, ptr %[[COMPLEX_C]], i32 0, i32 1
310+
// OGCG: %[[C_IMAG:.*]] = load float, ptr %[[C_IMAG_PTR]], align 4
311+
// OGCG: %[[SUB_REAL_A_B_C:.*]] = fsub float %[[SUB_REAL_A_B]], %[[C_REAL]]
312+
// OGCG: %[[SUB_IMAG_A_B_C:.*]] = fsub float %[[SUB_IMAG_A_B]], %[[C_IMAG]]
313+
// OGCG: %[[RESULT_REAL_PTR:.*]] = getelementptr inbounds nuw { float, float }, ptr %[[RESULT]], i32 0, i32 0
314+
// OGCG: %[[RESULT_IMAG_PTR:.*]] = getelementptr inbounds nuw { float, float }, ptr %[[RESULT]], i32 0, i32 1
315+
// OGCG: store float %[[SUB_REAL_A_B_C]], ptr %[[RESULT_REAL_PTR]], align 4
316+
// OGCG: store float %[[SUB_IMAG_A_B_C]], ptr %[[RESULT_IMAG_PTR]], align 4

0 commit comments

Comments
 (0)