[RISCV] Pass f32/f64 directly without a bitcast for Zfinx/Zdinx. (llvm#107464)

With Zfinx/Zdinx, f32/f64 are legal types for a GPR, we don't need a
bitcast.

This avoids turning fneg/fabs into bitwise operations purely because of
these bitcasts. If the bitwise operations are faster for some reason on
a Zfinx CPU, then that seems like it should be done for all fneg/fabs,
not just ones near function arguments/returns.

I don't have much interest in Zfinx, this just makes the code more
similar to what I proposed for Zhinx in llvm#107446.

Author: topperc

llvm/lib/Target/RISCV/RISCVCallingConv.cpp

@@ -312,6 +312,14 @@ bool llvm::CC_RISCV(unsigned ValNo, MVT ValVT, MVT LocVT,
 
   ArrayRef<MCPhysReg> ArgGPRs = RISCV::getArgGPRs(ABI);
 
+  if ((ValVT == MVT::f32 && XLen == 32 && Subtarget.hasStdExtZfinx()) ||
+      (ValVT == MVT::f64 && XLen == 64 && Subtarget.hasStdExtZdinx())) {
+    if (MCRegister Reg = State.AllocateReg(ArgGPRs)) {
+      State.addLoc(CCValAssign::getReg(ValNo, ValVT, Reg, LocVT, LocInfo));
+      return false;
+    }
+  }
+
   if (UseGPRForF16_F32 && (ValVT == MVT::f16 || ValVT == MVT::bf16 ||
                            (ValVT == MVT::f32 && XLen == 64))) {
     MCRegister Reg = State.AllocateReg(ArgGPRs);
@@ -571,8 +579,6 @@ bool llvm::CC_RISCV_FastCC(unsigned ValNo, MVT ValVT, MVT LocVT,
             CCValAssign::getCustomReg(ValNo, ValVT, Reg, LocVT, LocInfo));
         return false;
       }
-      LocVT = Subtarget.getXLenVT();
-      LocInfo = CCValAssign::BCvt;
       State.addLoc(CCValAssign::getReg(ValNo, ValVT, Reg, LocVT, LocInfo));
       return false;
     }

llvm/test/CodeGen/RISCV/double-arith.ll

@@ -320,8 +320,7 @@ define double @fsgnjn_d(double %a, double %b) nounwind {
 ;
 ; RV64IZFINXZDINX-LABEL: fsgnjn_d:
 ; RV64IZFINXZDINX:       # %bb.0:
-; RV64IZFINXZDINX-NEXT:    not a1, a1
-; RV64IZFINXZDINX-NEXT:    fsgnj.d a0, a0, a1
+; RV64IZFINXZDINX-NEXT:    fsgnjn.d a0, a0, a1
 ; RV64IZFINXZDINX-NEXT:    ret
 ;
 ; RV32I-LABEL: fsgnjn_d:
@@ -852,9 +851,7 @@ define double @fnmadd_d_3(double %a, double %b, double %c) nounwind {
 ; RV64IZFINXZDINX-LABEL: fnmadd_d_3:
 ; RV64IZFINXZDINX:       # %bb.0:
 ; RV64IZFINXZDINX-NEXT:    fmadd.d a0, a0, a1, a2
-; RV64IZFINXZDINX-NEXT:    li a1, -1
-; RV64IZFINXZDINX-NEXT:    slli a1, a1, 63
-; RV64IZFINXZDINX-NEXT:    xor a0, a0, a1
+; RV64IZFINXZDINX-NEXT:    fneg.d a0, a0
 ; RV64IZFINXZDINX-NEXT:    ret
 ;
 ; RV32I-LABEL: fnmadd_d_3:
@@ -900,10 +897,7 @@ define double @fnmadd_nsz(double %a, double %b, double %c) nounwind {
 ;
 ; RV64IZFINXZDINX-LABEL: fnmadd_nsz:
 ; RV64IZFINXZDINX:       # %bb.0:
-; RV64IZFINXZDINX-NEXT:    fmadd.d a0, a0, a1, a2
-; RV64IZFINXZDINX-NEXT:    li a1, -1
-; RV64IZFINXZDINX-NEXT:    slli a1, a1, 63
-; RV64IZFINXZDINX-NEXT:    xor a0, a0, a1
+; RV64IZFINXZDINX-NEXT:    fnmadd.d a0, a0, a1, a2
 ; RV64IZFINXZDINX-NEXT:    ret
 ;
 ; RV32I-LABEL: fnmadd_nsz:

llvm/test/CodeGen/RISCV/double-bitmanip-dagcombines.ll

@@ -56,9 +56,7 @@ define double @fneg(double %a) nounwind {
 ;
 ; RV64IZFINXZDINX-LABEL: fneg:
 ; RV64IZFINXZDINX:       # %bb.0:
-; RV64IZFINXZDINX-NEXT:    li a1, -1
-; RV64IZFINXZDINX-NEXT:    slli a1, a1, 63
-; RV64IZFINXZDINX-NEXT:    xor a0, a0, a1
+; RV64IZFINXZDINX-NEXT:    fneg.d a0, a0
 ; RV64IZFINXZDINX-NEXT:    ret
   %1 = fneg double %a
   ret double %1
@@ -99,8 +97,7 @@ define double @fabs(double %a) nounwind {
 ;
 ; RV64IZFINXZDINX-LABEL: fabs:
 ; RV64IZFINXZDINX:       # %bb.0:
-; RV64IZFINXZDINX-NEXT:    slli a0, a0, 1
-; RV64IZFINXZDINX-NEXT:    srli a0, a0, 1
+; RV64IZFINXZDINX-NEXT:    fabs.d a0, a0
 ; RV64IZFINXZDINX-NEXT:    ret
   %1 = call double @llvm.fabs.f64(double %a)
   ret double %1
@@ -165,8 +162,7 @@ define double @fcopysign_fneg(double %a, double %b) nounwind {
 ;
 ; RV64IZFINXZDINX-LABEL: fcopysign_fneg:
 ; RV64IZFINXZDINX:       # %bb.0:
-; RV64IZFINXZDINX-NEXT:    not a1, a1
-; RV64IZFINXZDINX-NEXT:    fsgnj.d a0, a0, a1
+; RV64IZFINXZDINX-NEXT:    fsgnjn.d a0, a0, a1
 ; RV64IZFINXZDINX-NEXT:    ret
   %1 = fneg double %b
   %2 = call double @llvm.copysign.f64(double %a, double %1)

llvm/test/CodeGen/RISCV/double-imm.ll

@@ -115,8 +115,7 @@ define double @double_negative_zero(ptr %pd) nounwind {
 ;
 ; CHECKRV64ZDINX-LABEL: double_negative_zero:
 ; CHECKRV64ZDINX:       # %bb.0:
-; CHECKRV64ZDINX-NEXT:    li a0, -1
-; CHECKRV64ZDINX-NEXT:    slli a0, a0, 63
+; CHECKRV64ZDINX-NEXT:    fneg.d a0, zero
 ; CHECKRV64ZDINX-NEXT:    ret
   ret double -0.0
 }
@@ -160,12 +159,11 @@ define dso_local double @negzero_sel(i16 noundef %a, double noundef %d) nounwind
 ; CHECKRV64ZDINX-LABEL: negzero_sel:
 ; CHECKRV64ZDINX:       # %bb.0: # %entry
 ; CHECKRV64ZDINX-NEXT:    slli a2, a0, 48
+; CHECKRV64ZDINX-NEXT:    mv a0, a1
 ; CHECKRV64ZDINX-NEXT:    beqz a2, .LBB4_2
 ; CHECKRV64ZDINX-NEXT:  # %bb.1: # %entry
 ; CHECKRV64ZDINX-NEXT:    fneg.d a0, zero
-; CHECKRV64ZDINX-NEXT:    ret
-; CHECKRV64ZDINX-NEXT:  .LBB4_2:
-; CHECKRV64ZDINX-NEXT:    mv a0, a1
+; CHECKRV64ZDINX-NEXT:  .LBB4_2: # %entry
 ; CHECKRV64ZDINX-NEXT:    ret
 entry:
   %tobool.not = icmp eq i16 %a, 0

llvm/test/CodeGen/RISCV/double-intrinsics.ll

@@ -684,8 +684,7 @@ define double @fabs_f64(double %a) nounwind {
 ;
 ; RV64IZFINXZDINX-LABEL: fabs_f64:
 ; RV64IZFINXZDINX:       # %bb.0:
-; RV64IZFINXZDINX-NEXT:    slli a0, a0, 1
-; RV64IZFINXZDINX-NEXT:    srli a0, a0, 1
+; RV64IZFINXZDINX-NEXT:    fabs.d a0, a0
 ; RV64IZFINXZDINX-NEXT:    ret
 ;
 ; RV32I-LABEL: fabs_f64: