Split fpscr into two registers: FPSCR and FPSCR_NZCV.

The fpscr register contains both flags (set by FP operations/comparisons) and
control bits. The control bits (FPSCR) should be reserved, since they're always
available and needn't be defined before use. The flag bits (FPSCR_NZCV) should
like to be unreserved so they can be hoisted by MachineCSE. This fixes PR12165.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@152076 91177308-0d34-0410-b5e6-96231b3b80d8

lib/Target/ARM/ARMBaseRegisterInfo.cpp

@@ -79,6 +79,7 @@ getReservedRegs(const MachineFunction &MF) const {
   BitVector Reserved(getNumRegs());
   Reserved.set(ARM::SP);
   Reserved.set(ARM::PC);
+  Reserved.set(ARM::FPSCR);
   if (TFI->hasFP(MF))
     Reserved.set(FramePtr);
   if (hasBasePointer(MF))

lib/Target/ARM/ARMFastISel.cpp

@@ -1496,7 +1496,7 @@ bool ARMFastISel::SelectCmp(const Instruction *I) {
   Constant *Zero = ConstantInt::get(Type::getInt32Ty(*Context), 0);
   unsigned ZeroReg = TargetMaterializeConstant(Zero);
   bool isFloat = (Ty->isFloatTy() || Ty->isDoubleTy());
-  unsigned CondReg = isFloat ? ARM::FPSCR : ARM::CPSR;
+  unsigned CondReg = isFloat ? ARM::FPSCR_NZCV : ARM::CPSR;
   BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(MovCCOpc), DestReg)
           .addReg(ZeroReg).addImm(1)
           .addImm(ARMPred).addReg(CondReg);

lib/Target/ARM/ARMInstrVFP.td

@@ -294,7 +294,7 @@ def : Pat<(fmul (fneg SPR:$a), SPR:$b),
           (VNMULS SPR:$a, SPR:$b)>, Requires<[NoHonorSignDependentRounding]>;
 
 // These are encoded as unary instructions.
-let Defs = [FPSCR] in {
+let Defs = [FPSCR_NZCV] in {
 def VCMPED : ADuI<0b11101, 0b11, 0b0100, 0b11, 0,
                   (outs), (ins DPR:$Dd, DPR:$Dm),
                   IIC_fpCMP64, "vcmpe", ".f64\t$Dd, $Dm",
@@ -323,7 +323,7 @@ def VCMPS  : ASuI<0b11101, 0b11, 0b0100, 0b01, 0,
   // VFP pipelines on A8.
   let D = VFPNeonA8Domain;
 }
-} // Defs = [FPSCR]
+} // Defs = [FPSCR_NZCV]
 
 //===----------------------------------------------------------------------===//
 // FP Unary Operations.
@@ -343,7 +343,7 @@ def VABSS  : ASuIn<0b11101, 0b11, 0b0000, 0b11, 0,
   let D = VFPNeonA8Domain;
 }
 
-let Defs = [FPSCR] in {
+let Defs = [FPSCR_NZCV] in {
 def VCMPEZD : ADuI<0b11101, 0b11, 0b0101, 0b11, 0,
                    (outs), (ins DPR:$Dd),
                    IIC_fpCMP64, "vcmpe", ".f64\t$Dd, #0",
@@ -384,7 +384,7 @@ def VCMPZS  : ASuI<0b11101, 0b11, 0b0101, 0b01, 0,
   // VFP pipelines on A8.
   let D = VFPNeonA8Domain;
 }
-} // Defs = [FPSCR]
+} // Defs = [FPSCR_NZCV]
 
 def VCVTDS  : ASuI<0b11101, 0b11, 0b0111, 0b11, 0,
                    (outs DPR:$Dd), (ins SPR:$Sm),
@@ -1174,7 +1174,7 @@ class MovFromVFP<bits<4> opc19_16, dag oops, dag iops, string opc, string asm,
 
 // APSR is the application level alias of CPSR. This FPSCR N, Z, C, V flags
 // to APSR.
-let Defs = [CPSR], Uses = [FPSCR], Rt = 0b1111 /* apsr_nzcv */ in
+let Defs = [CPSR], Uses = [FPSCR_NZCV], Rt = 0b1111 /* apsr_nzcv */ in
 def FMSTAT : MovFromVFP<0b0001 /* fpscr */, (outs), (ins),
                         "vmrs", "\tapsr_nzcv, fpscr", [(arm_fmstat)]>;
 

lib/Target/ARM/ARMRegisterInfo.td

@@ -153,11 +153,16 @@ def Q15 : ARMReg<15, "q15", [D30, D31]>;
 }
 
 // Current Program Status Register.
-def CPSR    : ARMReg<0, "cpsr">;
-def APSR    : ARMReg<1, "apsr">;
-def SPSR    : ARMReg<2, "spsr">;
-def FPSCR   : ARMReg<3, "fpscr">;
-def ITSTATE : ARMReg<4, "itstate">;
+// We model fpscr with two registers: FPSCR models the control bits and will be
+// reserved. FPSCR_NZCV models the flag bits and will be unreserved. 
+def CPSR       : ARMReg<0, "cpsr">;
+def APSR       : ARMReg<1, "apsr">;
+def SPSR       : ARMReg<2, "spsr">;
+def FPSCR      : ARMReg<3, "fpscr">;
+def FPSCR_NZCV : ARMReg<3, "fpscr_nzcv"> {
+  let Aliases = [FPSCR];
+}
+def ITSTATE    : ARMReg<4, "itstate">;
 
 // Special Registers - only available in privileged mode.
 def FPSID   : ARMReg<0, "fpsid">;

test/CodeGen/ARM/2012-03-05-FPSCR-bug.ll

@@ -0,0 +1,36 @@
+; RUN: llc -march=arm -mcpu=cortex-a8 -verify-machineinstrs < %s
+; PR12165
+target datalayout = "e-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-p:32:32:32-v128:32:32"
+target triple = "arm-none-linux"
+
+define hidden void @_strtod_r() nounwind {
+  br i1 undef, label %1, label %2
+
+; <label>:1                                       ; preds = %0
+  br label %2
+
+; <label>:2                                       ; preds = %1, %0
+  br i1 undef, label %3, label %8
+
+; <label>:3                                       ; preds = %2
+  br i1 undef, label %4, label %7
+
+; <label>:4                                       ; preds = %3
+  %5 = call i32 @llvm.flt.rounds()
+  %6 = icmp eq i32 %5, 1
+  br i1 %6, label %8, label %7
+
+; <label>:7                                       ; preds = %4, %3
+  unreachable
+
+; <label>:8                                       ; preds = %4, %2
+  br i1 undef, label %9, label %10
+
+; <label>:9                                       ; preds = %8
+  br label %10
+
+; <label>:10                                      ; preds = %9, %8
+  ret void
+}
+
+declare i32 @llvm.flt.rounds() nounwind