Skip to content

Commit 093395c

Browse files
AmrDeveloperAmrDeveloper
authored
[CIR] Mul CompoundAssignment support for ComplexType (llvm#152354)
This change adds support for Mul CompoundAssignment for ComplexType llvm#141365
1 parent 193995d commit 093395c

File tree

2 files changed

+213
-1
lines changed

2 files changed

+213
-1
lines changed

‎clang/lib/CIR/CodeGen/CIRGenExprComplex.cpp‎

Lines changed: 5 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -171,6 +171,10 @@ class ComplexExprEmitter : public StmtVisitor<ComplexExprEmitter, mlir::Value> {
171171
mlir::Value VisitBinSubAssign(const CompoundAssignOperator *e) {
172172
return emitCompoundAssign(e, &ComplexExprEmitter::emitBinSub);
173173
}
174+
175+
mlir::Value VisitBinMulAssign(const CompoundAssignOperator *e) {
176+
return emitCompoundAssign(e, &ComplexExprEmitter::emitBinMul);
177+
}
174178
};
175179
} // namespace
176180

@@ -813,7 +817,7 @@ using CompoundFunc =
813817
static CompoundFunc getComplexOp(BinaryOperatorKind op) {
814818
switch (op) {
815819
case BO_MulAssign:
816-
llvm_unreachable("getComplexOp: BO_MulAssign");
820+
return &ComplexExprEmitter::emitBinMul;
817821
case BO_DivAssign:
818822
llvm_unreachable("getComplexOp: BO_DivAssign");
819823
case BO_SubAssign:

‎clang/test/CIR/CodeGen/complex-compound-assignment.cpp‎

Lines changed: 208 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -286,3 +286,211 @@ void foo4() {
286286
// CXX_OGCG: %[[C_IMAG_PTR:.*]] = getelementptr inbounds nuw { i32, i32 }, ptr %[[C_ADDR]], i32 0, i32 1
287287
// CXX_OGCG: store i32 %[[B_REAL]], ptr %[[C_REAL_PTR]], align 4
288288
// CXX_OGCG: store i32 %[[B_IMAG]], ptr %[[C_IMAG_PTR]], align 4
289+
290+
void foo5() {
291+
float _Complex a;
292+
float b;
293+
a += b;
294+
}
295+
296+
// CIR: %[[A_ADDR:.*]] = cir.alloca !cir.complex<!cir.float>, !cir.ptr<!cir.complex<!cir.float>>, ["a"]
297+
// CIR: %[[B_ADDR:.*]] = cir.alloca !cir.float, !cir.ptr<!cir.float>, ["b"]
298+
// CIR: %[[TMP_B:.*]] = cir.load{{.*}} %[[B_ADDR]] : !cir.ptr<!cir.float>, !cir.float
299+
// CIR: %[[CONST_ZERO:.*]] = cir.const #cir.fp<0.000000e+00> : !cir.float
300+
// CIR: %[[COMPLEX_B:.*]] = cir.complex.create %[[TMP_B]], %[[CONST_ZERO]] : !cir.float -> !cir.complex<!cir.float>
301+
// CIR: %[[TMP_A:.*]] = cir.load{{.*}} %[[A_ADDR]] : !cir.ptr<!cir.complex<!cir.float>>, !cir.complex<!cir.float>
302+
// CIR: %[[RESULT:.*]] = cir.complex.add %[[TMP_A]], %[[COMPLEX_B]] : !cir.complex<!cir.float>
303+
// CIR: cir.store{{.*}} %[[RESULT]], %[[A_ADDR]] : !cir.complex<!cir.float>, !cir.ptr<!cir.complex<!cir.float>>
304+
305+
// LLVM: %[[A_ADDR:.*]] = alloca { float, float }, i64 1, align 4
306+
// LLVM: %[[B_ADDR:.*]] = alloca float, i64 1, align 4
307+
// LLVM: %[[TMP_B:.*]] = load float, ptr %[[B_ADDR]], align 4
308+
// LLVM: %[[TMP_COMPLEX_B:.*]] = insertvalue { float, float } {{.*}}, float %[[TMP_B]], 0
309+
// LLVM: %[[COMPLEX_B:.*]] = insertvalue { float, float } %[[TMP_COMPLEX_B]], float 0.000000e+00, 1
310+
// LLVM: %[[TMP_A:.*]] = load { float, float }, ptr %[[A_ADDR]], align 4
311+
// LLVM: %[[A_REAL:.*]] = extractvalue { float, float } %[[TMP_A]], 0
312+
// LLVM: %[[A_IMAG:.*]] = extractvalue { float, float } %[[TMP_A]], 1
313+
// LLVM: %[[B_REAL:.*]] = extractvalue { float, float } %[[COMPLEX_B]], 0
314+
// LLVM: %[[B_IMAG:.*]] = extractvalue { float, float } %[[COMPLEX_B]], 1
315+
// LLVM: %[[ADD_REAL:.*]] = fadd float %[[A_REAL]], %[[B_REAL]]
316+
// LLVM: %[[ADD_IMAG:.*]] = fadd float %[[A_IMAG]], %[[B_IMAG]]
317+
// LLVM: %[[TMP_RESULT:.*]] = insertvalue { float, float } poison, float %[[ADD_REAL]], 0
318+
// LLVM: %[[RESULT:.*]] = insertvalue { float, float } %[[TMP_RESULT]], float %[[ADD_IMAG]], 1
319+
// LLVM: store { float, float } %[[RESULT]], ptr %[[A_ADDR]], align 4
320+
321+
// OGCG: %[[A_ADDR:.*]] = alloca { float, float }, align 4
322+
// OGCG: %[[B_ADDR:.*]] = alloca float, align 4
323+
// OGCG: %[[TMP_B:.*]] = load float, ptr %[[B_ADDR]], align 4
324+
// OGCG: %[[A_REAL_PTR:.*]] = getelementptr inbounds nuw { float, float }, ptr %[[A_ADDR]], i32 0, i32 0
325+
// OGCG: %[[A_REAL:.*]] = load float, ptr %[[A_REAL_PTR]], align 4
326+
// OGCG: %[[A_IMAG_PTR:.*]] = getelementptr inbounds nuw { float, float }, ptr %[[A_ADDR]], i32 0, i32 1
327+
// OGCG: %[[A_IMAG:.*]] = load float, ptr %[[A_IMAG_PTR]], align 4
328+
// OGCG: %[[ADD_REAL:.*]] = fadd float %[[A_REAL]], %[[TMP_B]]
329+
// OGCG: %[[A_REAL_PTR:.*]] = getelementptr inbounds nuw { float, float }, ptr %[[A_ADDR]], i32 0, i32 0
330+
// OGCG: %[[A_IMAG_PTR:.*]] = getelementptr inbounds nuw { float, float }, ptr %[[A_ADDR]], i32 0, i32 1
331+
// OGCG: store float %[[ADD_REAL]], ptr %[[A_REAL_PTR]], align 4
332+
// OGCG: store float %[[A_IMAG]], ptr %[[A_IMAG_PTR]], align 4
333+
334+
void foo6() {
335+
int _Complex a;
336+
int _Complex b;
337+
b *= a;
338+
}
339+
340+
// CIR: %[[A_ADDR:.*]] = cir.alloca !cir.complex<!s32i>, !cir.ptr<!cir.complex<!s32i>>, ["a"]
341+
// CIR: %[[B_ADDR:.*]] = cir.alloca !cir.complex<!s32i>, !cir.ptr<!cir.complex<!s32i>>, ["b"]
342+
// CIR: %[[TMP_A:.*]] = cir.load{{.*}} %[[A_ADDR]] : !cir.ptr<!cir.complex<!s32i>>, !cir.complex<!s32i>
343+
// CIR: %[[TMP_B:.*]] = cir.load{{.*}} %[[B_ADDR]] : !cir.ptr<!cir.complex<!s32i>>, !cir.complex<!s32i>
344+
// CIR: %[[B_REAL:.*]] = cir.complex.real %[[TMP_B]] : !cir.complex<!s32i> -> !s32i
345+
// CIR: %[[B_IMAG:.*]] = cir.complex.imag %[[TMP_B]] : !cir.complex<!s32i> -> !s32i
346+
// CIR: %[[A_REAL:.*]] = cir.complex.real %[[TMP_A]] : !cir.complex<!s32i> -> !s32i
347+
// CIR: %[[A_IMAG:.*]] = cir.complex.imag %[[TMP_A]] : !cir.complex<!s32i> -> !s32i
348+
// CIR: %[[MUL_BR_AR:.*]] = cir.binop(mul, %[[B_REAL]], %[[A_REAL]]) : !s32i
349+
// CIR: %[[MUL_BI_AI:.*]] = cir.binop(mul, %[[B_IMAG]], %[[A_IMAG]]) : !s32i
350+
// CIR: %[[MUL_BR_AI:.*]] = cir.binop(mul, %[[B_REAL]], %[[A_IMAG]]) : !s32i
351+
// CIR: %[[MUL_BI_AR:.*]] = cir.binop(mul, %[[B_IMAG]], %[[A_REAL]]) : !s32i
352+
// CIR: %[[RESULT_REAL:.*]] = cir.binop(sub, %[[MUL_BR_AR]], %[[MUL_BI_AI]]) : !s32i
353+
// CIR: %[[RESULT_IMAG:.*]] = cir.binop(add, %[[MUL_BR_AI]], %[[MUL_BI_AR]]) : !s32i
354+
// CIR: %[[RESULT:.*]] = cir.complex.create %[[RESULT_REAL]], %[[RESULT_IMAG]] : !s32i -> !cir.complex<!s32i>
355+
// CIR: cir.store{{.*}} %[[RESULT]], %[[B_ADDR]] : !cir.complex<!s32i>, !cir.ptr<!cir.complex<!s32i>>
356+
357+
// LLVM: %[[A_ADDR:.*]] = alloca { i32, i32 }, i64 1, align 4
358+
// LLVM: %[[B_ADDR:.*]] = alloca { i32, i32 }, i64 1, align 4
359+
// LLVM: %[[TMP_A:.*]] = load { i32, i32 }, ptr %[[A_ADDR]], align 4
360+
// LLVM: %[[TMP_B:.*]] = load { i32, i32 }, ptr %[[B_ADDR]], align 4
361+
// LLVM: %[[B_REAL:.*]] = extractvalue { i32, i32 } %[[TMP_B]], 0
362+
// LLVM: %[[B_IMAG:.*]] = extractvalue { i32, i32 } %[[TMP_B]], 1
363+
// LLVM: %[[A_REAL:.*]] = extractvalue { i32, i32 } %[[TMP_A]], 0
364+
// LLVM: %[[A_IMAG:.*]] = extractvalue { i32, i32 } %[[TMP_A]], 1
365+
// LLVM: %[[MUL_BR_AR:.*]] = mul i32 %[[B_REAL]], %[[A_REAL]]
366+
// LLVM: %[[MUL_BI_AI:.*]] = mul i32 %[[B_IMAG]], %[[A_IMAG]]
367+
// LLVM: %[[MUL_BR_AI:.*]] = mul i32 %[[B_REAL]], %[[A_IMAG]]
368+
// LLVM: %[[MUL_BI_AR:.*]] = mul i32 %[[B_IMAG]], %[[A_REAL]]
369+
// LLVM: %[[RESULT_REAL:.*]] = sub i32 %[[MUL_BR_AR]], %[[MUL_BI_AI]]
370+
// LLVM: %[[RESULT_IMAG:.*]] = add i32 %[[MUL_BR_AI]], %[[MUL_BI_AR]]
371+
// LLVM: %[[MUL_A_B:.*]] = insertvalue { i32, i32 } {{.*}}, i32 %[[RESULT_REAL]], 0
372+
// LLVM: %[[RESULT:.*]] = insertvalue { i32, i32 } %[[MUL_A_B]], i32 %[[RESULT_IMAG]], 1
373+
// LLVM: store { i32, i32 } %[[RESULT]], ptr %[[B_ADDR]], align 4
374+
375+
// OGCG: %[[A_ADDR:.*]] = alloca { i32, i32 }, align 4
376+
// OGCG: %[[B_ADDR:.*]] = alloca { i32, i32 }, align 4
377+
// OGCG: %[[A_REAL_PTR:.*]] = getelementptr inbounds nuw { i32, i32 }, ptr %[[A_ADDR]], i32 0, i32 0
378+
// OGCG: %[[A_REAL:.*]] = load i32, ptr %[[A_REAL_PTR]], align 4
379+
// OGCG: %[[A_IMAG_PTR:.*]] = getelementptr inbounds nuw { i32, i32 }, ptr %[[A_ADDR]], i32 0, i32 1
380+
// OGCG: %[[A_IMAG:.*]] = load i32, ptr %[[A_IMAG_PTR]], align 4
381+
// OGCG: %[[B_REAL_PTR:.*]] = getelementptr inbounds nuw { i32, i32 }, ptr %[[B_ADDR]], i32 0, i32 0
382+
// OGCG: %[[B_REAL:.*]] = load i32, ptr %[[B_REAL_PTR]], align 4
383+
// OGCG: %[[B_IMAG_PTR:.*]] = getelementptr inbounds nuw { i32, i32 }, ptr %[[B_ADDR]], i32 0, i32 1
384+
// OGCG: %[[B_IMAG:.*]] = load i32, ptr %[[B_IMAG_PTR]], align 4
385+
// OGCG: %[[MUL_BR_AR:.*]] = mul i32 %[[B_REAL]], %[[A_REAL]]
386+
// OGCG: %[[MUL_BI_AI:.*]] = mul i32 %[[B_IMAG]], %[[A_IMAG]]
387+
// OGCG: %[[RESULT_REAL:.*]] = sub i32 %[[MUL_BR_AR]], %[[MUL_BI_AI]]
388+
// OGCG: %[[MUL_BI_AR:.*]] = mul i32 %[[B_IMAG]], %[[A_REAL]]
389+
// OGCG: %[[MUL_BR_AI:.*]] = mul i32 %[[B_REAL]], %[[A_IMAG]]
390+
// OGCG: %[[RESULT_IMAG:.*]] = add i32 %[[MUL_BI_AR]], %[[MUL_BR_AI]]
391+
// OGCG: %[[B_REAL_PTR:.*]] = getelementptr inbounds nuw { i32, i32 }, ptr %[[B_ADDR]], i32 0, i32 0
392+
// OGCG: %[[B_IMAG_PTR:.*]] = getelementptr inbounds nuw { i32, i32 }, ptr %[[B_ADDR]], i32 0, i32 1
393+
// OGCG: store i32 %[[RESULT_REAL]], ptr %[[B_REAL_PTR]], align 4
394+
// OGCG: store i32 %[[RESULT_IMAG]], ptr %[[B_IMAG_PTR]], align 4
395+
396+
void foo7() {
397+
float _Complex a;
398+
float _Complex b;
399+
b *= a;
400+
}
401+
402+
// CIR: %[[A_ADDR:.*]] = cir.alloca !cir.complex<!cir.float>, !cir.ptr<!cir.complex<!cir.float>>, ["a"]
403+
// CIR: %[[B_ADDR:.*]] = cir.alloca !cir.complex<!cir.float>, !cir.ptr<!cir.complex<!cir.float>>, ["b"]
404+
// CIR: %[[TMP_A:.*]] = cir.load{{.*}} %[[A_ADDR]] : !cir.ptr<!cir.complex<!cir.float>>, !cir.complex<!cir.float>
405+
// CIR: %[[TMP_B:.*]] = cir.load{{.*}} %[[B_ADDR]] : !cir.ptr<!cir.complex<!cir.float>>, !cir.complex<!cir.float>
406+
// CIR: %[[B_REAL:.*]] = cir.complex.real %[[TMP_B]] : !cir.complex<!cir.float> -> !cir.float
407+
// CIR: %[[B_IMAG:.*]] = cir.complex.imag %[[TMP_B]] : !cir.complex<!cir.float> -> !cir.float
408+
// CIR: %[[A_REAL:.*]] = cir.complex.real %[[TMP_A]] : !cir.complex<!cir.float> -> !cir.float
409+
// CIR: %[[A_IMAG:.*]] = cir.complex.imag %[[TMP_A]] : !cir.complex<!cir.float> -> !cir.float
410+
// CIR: %[[MUL_BR_AR:.*]] = cir.binop(mul, %[[B_REAL]], %[[A_REAL]]) : !cir.float
411+
// CIR: %[[MUL_BI_AI:.*]] = cir.binop(mul, %[[B_IMAG]], %[[A_IMAG]]) : !cir.float
412+
// CIR: %[[MUL_BR_AI:.*]] = cir.binop(mul, %[[B_REAL]], %[[A_IMAG]]) : !cir.float
413+
// CIR: %[[MUL_BI_AR:.*]] = cir.binop(mul, %[[B_IMAG]], %[[A_REAL]]) : !cir.float
414+
// CIR: %[[C_REAL:.*]] = cir.binop(sub, %[[MUL_BR_AR]], %[[MUL_BI_AI]]) : !cir.float
415+
// CIR: %[[C_IMAG:.*]] = cir.binop(add, %[[MUL_BR_AI]], %[[MUL_BI_AR]]) : !cir.float
416+
// CIR: %[[COMPLEX:.*]] = cir.complex.create %[[C_REAL]], %[[C_IMAG]] : !cir.float -> !cir.complex<!cir.float>
417+
// CIR: %[[IS_C_REAL_NAN:.*]] = cir.cmp(ne, %[[C_REAL]], %[[C_REAL]]) : !cir.float, !cir.bool
418+
// CIR: %[[IS_C_IMAG_NAN:.*]] = cir.cmp(ne, %[[C_IMAG]], %[[C_IMAG]]) : !cir.float, !cir.bool
419+
// CIR: %[[CONST_FALSE:.*]] = cir.const #false
420+
// CIR: %[[SELECT_CONDITION:.*]] = cir.select if %[[IS_C_REAL_NAN]] then %[[IS_C_IMAG_NAN]] else %[[CONST_FALSE]] : (!cir.bool, !cir.bool, !cir.bool) -> !cir.bool
421+
// CIR: %[[RESULT:.*]] = cir.ternary(%[[SELECT_CONDITION]], true {
422+
// CIR: %[[LIBC_COMPLEX:.*]] = cir.call @__mulsc3(%[[B_REAL]], %[[B_IMAG]], %[[A_REAL]], %[[A_IMAG]]) : (!cir.float, !cir.float, !cir.float, !cir.float) -> !cir.complex<!cir.float>
423+
// CIR: cir.yield %[[LIBC_COMPLEX]] : !cir.complex<!cir.float>
424+
// CIR: }, false {
425+
// CIR: cir.yield %[[COMPLEX]] : !cir.complex<!cir.float>
426+
// CIR: }) : (!cir.bool) -> !cir.complex<!cir.float>
427+
// CIR: cir.store{{.*}} %[[RESULT]], %[[B_ADDR]] : !cir.complex<!cir.float>, !cir.ptr<!cir.complex<!cir.float>>
428+
429+
// LLVM: %[[A_ADDR:.*]] = alloca { float, float }, i64 1, align 4
430+
// LLVM: %[[B_ADDR:.*]] = alloca { float, float }, i64 1, align 4
431+
// LLVM: %[[TMP_A:.*]] = load { float, float }, ptr %[[A_ADDR]], align 4
432+
// LLVM: %[[TMP_B:.*]] = load { float, float }, ptr %[[B_ADDR]], align 4
433+
// LLVM: %[[B_REAL:.*]] = extractvalue { float, float } %[[TMP_B]], 0
434+
// LLVM: %[[B_IMAG:.*]] = extractvalue { float, float } %[[TMP_B]], 1
435+
// LLVM: %[[A_REAL:.*]] = extractvalue { float, float } %[[TMP_A]], 0
436+
// LLVM: %[[A_IMAG:.*]] = extractvalue { float, float } %[[TMP_A]], 1
437+
// LLVM: %[[MUL_BR_AR:.*]] = fmul float %[[B_REAL]], %[[A_REAL]]
438+
// LLVM: %[[MUL_BI_AI:.*]] = fmul float %[[B_IMAG]], %[[A_IMAG]]
439+
// LLVM: %[[MUL_BR_AI:.*]] = fmul float %[[B_REAL]], %[[A_IMAG]]
440+
// LLVM: %[[MUL_BI_AR:.*]] = fmul float %[[B_IMAG]], %[[A_REAL]]
441+
// LLVM: %[[C_REAL:.*]] = fsub float %[[MUL_BR_AR]], %[[MUL_BI_AI]]
442+
// LLVM: %[[C_IMAG:.*]] = fadd float %[[MUL_BR_AI]], %[[MUL_BI_AR]]
443+
// LLVM: %[[MUL_A_B:.*]] = insertvalue { float, float } {{.*}}, float %[[C_REAL]], 0
444+
// LLVM: %[[COMPLEX:.*]] = insertvalue { float, float } %[[MUL_A_B]], float %[[C_IMAG]], 1
445+
// LLVM: %[[IS_C_REAL_NAN:.*]] = fcmp une float %[[C_REAL]], %[[C_REAL]]
446+
// LLVM: %[[IS_C_IMAG_NAN:.*]] = fcmp une float %[[C_IMAG]], %[[C_IMAG]]
447+
// LLVM: %[[SELECT_CONDITION:.*]] = and i1 %[[IS_C_REAL_NAN]], %[[IS_C_IMAG_NAN]]
448+
// LLVM: br i1 %[[SELECT_CONDITION]], label %[[THEN_LABEL:.*]], label %[[ELSE_LABEL:.*]]
449+
// LLVM: [[THEN_LABEL]]:
450+
// LLVM: %[[LIBC_COMPLEX:.*]] = call { float, float } @__mulsc3(float %[[B_REAL]], float %[[B_IMAG]], float %[[A_REAL]], float %[[A_IMAG]])
451+
// LLVM: br label %[[PHI_BRANCH:.*]]
452+
// LLVM: [[ELSE_LABEL]]:
453+
// LLVM: br label %[[PHI_BRANCH:]]
454+
// LLVM: [[PHI_BRANCH:]]:
455+
// LLVM: %[[RESULT:.*]] = phi { float, float } [ %[[COMPLEX]], %[[ELSE_LABEL]] ], [ %[[LIBC_COMPLEX]], %[[THEN_LABEL]] ]
456+
// LLVM: br label %[[END_LABEL:.*]]
457+
// LLVM: [[END_LABEL]]:
458+
// LLVM: store { float, float } %[[RESULT]], ptr %[[B_ADDR]], align 4
459+
460+
// OGCG: %[[A_ADDR:.*]] = alloca { float, float }, align 4
461+
// OGCG: %[[B_ADDR:.*]] = alloca { float, float }, align 4
462+
// OGCG: %[[COMPLEX_CALL_ADDR:.*]] = alloca { float, float }, align 4
463+
// OGCG: %[[A_REAL_PTR:.*]] = getelementptr inbounds nuw { float, float }, ptr %[[A_ADDR]], i32 0, i32 0
464+
// OGCG: %[[A_REAL:.*]] = load float, ptr %[[A_REAL_PTR]], align 4
465+
// OGCG: %[[A_IMAG_PTR:.*]] = getelementptr inbounds nuw { float, float }, ptr %[[A_ADDR]], i32 0, i32 1
466+
// OGCG: %[[A_IMAG:.*]] = load float, ptr %[[A_IMAG_PTR]], align 4
467+
// OGCG: %[[B_REAL_PTR:.*]] = getelementptr inbounds nuw { float, float }, ptr %[[B_ADDR]], i32 0, i32 0
468+
// OGCG: %[[B_REAL:.*]] = load float, ptr %[[B_REAL_PTR]], align 4
469+
// OGCG: %[[B_IMAG_PTR:.*]] = getelementptr inbounds nuw { float, float }, ptr %[[B_ADDR]], i32 0, i32 1
470+
// OGCG: %[[B_IMAG:.*]] = load float, ptr %[[B_IMAG_PTR]], align 4
471+
// OGCG: %[[MUL_BR_AR:.*]] = fmul float %[[B_REAL]], %[[A_REAL]]
472+
// OGCG: %[[MUL_BI_AI:.*]] = fmul float %[[B_IMAG]], %[[A_IMAG]]
473+
// OGCG: %[[MUL_BR_AI:.*]] = fmul float %[[B_REAL]], %[[A_IMAG]]
474+
// OGCG: %[[MUL_BI_AR:.*]] = fmul float %[[B_IMAG]], %[[A_REAL]]
475+
// OGCG: %[[C_REAL:.*]] = fsub float %[[MUL_BR_AR]], %[[MUL_BI_AI]]
476+
// OGCG: %[[C_IMAG:.*]] = fadd float %[[MUL_BR_AI]], %[[MUL_BI_AR]]
477+
// OGCG: %[[IS_C_REAL_NAN:.*]] = fcmp uno float %[[C_REAL]], %[[C_REAL]]
478+
// OGCG: br i1 %[[IS_C_REAL_NAN]], label %[[COMPLEX_IS_IMAG_NAN:.*]], label %[[END_LABEL:.*]], !prof !2
479+
// OGCG: [[COMPLEX_IS_IMAG_NAN]]:
480+
// OGCG: %[[IS_C_IMAG_NAN:.*]] = fcmp uno float %[[C_IMAG]], %[[C_IMAG]]
481+
// OGCG: br i1 %[[IS_C_IMAG_NAN]], label %[[COMPLEX_LIB_CALL:.*]], label %[[END_LABEL]], !prof !2
482+
// OGCG: [[COMPLEX_LIB_CALL]]:
483+
// OGCG: %[[CALL_RESULT:.*]] = call{{.*}} <2 x float> @__mulsc3(float noundef %[[B_REAL]], float noundef %[[B_IMAG]], float noundef %[[A_REAL]], float noundef %[[A_IMAG]])
484+
// OGCG: store <2 x float> %[[CALL_RESULT]], ptr %[[COMPLEX_CALL_ADDR]], align 4
485+
// OGCG: %[[COMPLEX_CALL_REAL_PTR:.*]] = getelementptr inbounds nuw { float, float }, ptr %[[COMPLEX_CALL_ADDR]], i32 0, i32 0
486+
// OGCG: %[[COMPLEX_CALL_REAL:.*]] = load float, ptr %[[COMPLEX_CALL_REAL_PTR]], align 4
487+
// OGCG: %[[COMPLEX_CALL_IMAG_PTR:.*]] = getelementptr inbounds nuw { float, float }, ptr %[[COMPLEX_CALL_ADDR]], i32 0, i32 1
488+
// OGCG: %[[COMPLEX_CALL_IMAG:.*]] = load float, ptr %[[COMPLEX_CALL_IMAG_PTR]], align 4
489+
// OGCG: br label %[[END_LABEL]]
490+
// OGCG: [[END_LABEL]]:
491+
// OGCG: %[[FINAL_REAL:.*]] = phi float [ %[[C_REAL]], %[[ENTRY:.*]] ], [ %[[C_REAL]], %[[COMPLEX_IS_IMAG_NAN]] ], [ %[[COMPLEX_CALL_REAL]], %[[COMPLEX_LIB_CALL]] ]
492+
// OGCG: %[[FINAL_IMAG:.*]] = phi float [ %[[C_IMAG]], %[[ENTRY]] ], [ %[[C_IMAG]], %[[COMPLEX_IS_IMAG_NAN]] ], [ %[[COMPLEX_CALL_IMAG]], %[[COMPLEX_LIB_CALL]] ]
493+
// OGCG: %[[C_REAL_PTR:.*]] = getelementptr inbounds nuw { float, float }, ptr %[[B_ADDR]], i32 0, i32 0
494+
// OGCG: %[[C_IMAG_PTR:.*]] = getelementptr inbounds nuw { float, float }, ptr %[[B_ADDR]], i32 0, i32 1
495+
// OGCG: store float %[[FINAL_REAL]], ptr %[[C_REAL_PTR]], align 4
496+
// OGCG: store float %[[FINAL_IMAG]], ptr %[[C_IMAG_PTR]], align 4

0 commit comments

Comments
 (0)