[RISCV] Fix passing two floating-point values in complex separately by two GPRs on RV64

[Jim]

Summary

This patch fixed the error of counting the remaining FPRs. Complex floating-point values should be passed by two FPRs for the hard-float ABI. If no two FPRs are available, it should be passed via a 64-bit GPR (fp+fp). ArgFPRsLeft is only decreased one while the type is complex floating-point. It causes two floating-point values in the complex are passed separately by two GPRs.

[luismarques]

clang/lib/CodeGen/TargetInfo.cpp

10241–10243

Do you have tests that show the impact of the added FLen >= Size && ArgFPRsLeft conditions for other values besides complex floats?

[asb]

Good catch, thanks for the fix! The logic was incorrectly written assuming isFloatingType would return false for complex values which is of course incorrect.

clang/lib/CodeGen/TargetInfo.cpp

10241–10243

That's actually not an added clause - it's just been clang-formatted onto a new line.

clang/lib/CodeGen/TargetInfo.cpp

ABIArgInfo RISCVABIInfo::classifyArgumentType(QualType Ty, bool IsFixed,
 
   // Ignore empty structs/unions.
   if (isEmptyRecord(getContext(), Ty, true))
     return ABIArgInfo::getIgnore();
 
   uint64_t Size = getContext().getTypeSize(Ty);
 
   // Pass floating point values via FPRs if possible.
-  if (IsFixed && Ty->isFloatingType() && FLen >= Size && ArgFPRsLeft) {
+  if (IsFixed && Ty->isFloatingType() && !Ty->isComplexType() &&
+      FLen >= Size && ArgFPRsLeft) {
     ArgFPRsLeft--;
     return ABIArgInfo::getDirect();
   }
 
   // Complex types for the hard float ABI must be passed direct rather than
   // using CoerceAndExpand.
   if (IsFixed && Ty->isComplexType() && FLen && ArgFPRsLeft >= 2) {
     QualType EltTy = Ty->castAs<ComplexType>()->getElementType();
     if (getContext().getTypeSize(EltTy) <= FLen) {

clang/test/CodeGen/riscv64-lp64-abi.c

 // the presence of large return values that consume a register due to the need
 // to pass a pointer.
 
 // CHECK-LABEL: define void @f_scalar_stack_2(%struct.large* noalias sret align 8 %agg.result, double %a, i128 %b, fp128 %c, <32 x i8>* %0, i8 zeroext %e, i8 %f, i8 %g)
 struct large f_scalar_stack_2(double a, __int128_t b, long double c, v32i8 d,
                               uint8_t e, int8_t f, uint8_t g) {
   return (struct large){a, e, f, g};
 }
+
+// Complex floating-point values or structs containing a single complex
+// floating-point value should be passed in a GPR.
+
+// CHECK: define void @f_floatcomplex(i64 %a.coerce)
+void f_floatcomplex(float __complex__ a) {}
+
+// CHECK: define i64 @f_ret_floatcomplex()
+float __complex__ f_ret_floatcomplex() {
+  return 1.0;
+}
+
+struct floatcomplex_s { float __complex__ c; };
+
+// CHECK: define void @f_floatcomplex_s_arg(i64 %a.coerce)
+void f_floatcomplex_s_arg(struct floatcomplex_s a) {}
+
+// CHECK: define i64 @f_ret_floatcomplex_s()
+struct floatcomplex_s f_ret_floatcomplex_s() {
+  return (struct floatcomplex_s){1.0};
+}

clang/test/CodeGen/riscv64-lp64f-lp64d-abi.c

 // CHECK: define void @f_floatcomplex_s_arg(float %0, float %1)
 void f_floatcomplex_s_arg(struct floatcomplex_s a) {}
 
 // CHECK: define { float, float } @f_ret_floatcomplex_s()
 struct floatcomplex_s f_ret_floatcomplex_s() {
   return (struct floatcomplex_s){1.0};
 }
 
+// Complex floating-point values or structs containing a single complex
+// floating-point value should be passed in GPRs if no two FPRs is available.
+
+// CHECK: define void @f_floatcomplex_insufficient_fprs1(float %a.coerce0, float %a.coerce1, float %b.coerce0, float %b.coerce1, float %c.coerce0, float %c.coerce1, float %d.coerce0, float %d.coerce1, i64 %e.coerce)
+void f_floatcomplex_insufficient_fprs1(float __complex__ a, float __complex__ b,
+                                       float __complex__ c, float __complex__ d,
+                                       float __complex__ e) {}
+
+
+// CHECK: define void @f_floatcomplex_s_arg_insufficient_fprs1(float %0, float %1, float %2, float %3, float %4, float %5, float %6, float %7, i64 %e.coerce)
+void f_floatcomplex_s_arg_insufficient_fprs1(struct floatcomplex_s a,
+                                             struct floatcomplex_s b,
+                                             struct floatcomplex_s c,
+                                             struct floatcomplex_s d,
+                                             struct floatcomplex_s e) {}
+
+// CHECK: define void @f_floatcomplex_insufficient_fprs2(float %a, float %b.coerce0, float %b.coerce1, float %c.coerce0, float %c.coerce1, float %d.coerce0, float %d.coerce1, i64 %e.coerce)
+void f_floatcomplex_insufficient_fprs2(float a,
+                                       float __complex__ b, float __complex__ c,
+                                       float __complex__ d, float __complex__ e) {}
+
+
+// CHECK: define void @f_floatcomplex_s_arg_insufficient_fprs2(float %a, float %0, float %1, float %2, float %3, float %4, float %5, i64 %e.coerce)
+void f_floatcomplex_s_arg_insufficient_fprs2(float a,
+                                             struct floatcomplex_s b,
+                                             struct floatcomplex_s c,
+                                             struct floatcomplex_s d,
+                                             struct floatcomplex_s e) {}
+
 // Test single or two-element structs that need flattening. e.g. those
 // containing nested structs, floats in small arrays, zero-length structs etc.
 
 struct floatarr1_s { float a[1]; };
 
 // CHECK: define void @f_floatarr1_s_arg(float %0)
 void f_floatarr1_s_arg(struct floatarr1_s a) {}