[PowerPC] Fix STRICT_FRINT/STRICT_FNEARBYINT lowering

[qiucf]

Summary

In standard C library, both rint and nearbyint returns rounding result in current rounding mode. But nearbyint never raises inexact exception. On PowerPC, x(v|s)r(d|s)pic may modify FPSCR XX, raising inexact exception. So we can't select constrained fnearbyint into xvrdpic.

One exception here is xsrqpi, which will not raise inexact exception, so fnearbyint f128 is okay here.

[qiucf]

Fix clang to align the change

[uweigand]

Makes sense to me. As an aside, this code:

// The nearbyint variants are not allowed to raise the inexact exception
// so we can only code-gen them with unsafe math.
if (TM.Options.UnsafeFPMath) {
  setOperationAction(ISD::FNEARBYINT, MVT::f64, Legal);
  setOperationAction(ISD::FNEARBYINT, MVT::f32, Legal);
}

is probably overly strict: FNEARBYINT (as opposed to STRICT_FNEARBYINT) is defined to ignore exceptions, whether or not we have UnsafeFPMath or not, so this can probably be enabled unconditionally.

But that can be done independently of this patch. LGTM.

clang/lib/CodeGen/CGBuiltin.cpp

case PPC::BI__builtin_vsx_xvrspiz: {
     else if (BuiltinID == PPC::BI__builtin_vsx_xvrdpi ||
              BuiltinID == PPC::BI__builtin_vsx_xvrspi)
       ID = Builder.getIsFPConstrained()
                ? Intrinsic::experimental_constrained_round
                : Intrinsic::round;
     else if (BuiltinID == PPC::BI__builtin_vsx_xvrdpic ||
              BuiltinID == PPC::BI__builtin_vsx_xvrspic)
       ID = Builder.getIsFPConstrained()
-               ? Intrinsic::experimental_constrained_nearbyint
-               : Intrinsic::nearbyint;
+               ? Intrinsic::experimental_constrained_rint
+               : Intrinsic::rint;
     else if (BuiltinID == PPC::BI__builtin_vsx_xvrdpip ||
              BuiltinID == PPC::BI__builtin_vsx_xvrspip)
       ID = Builder.getIsFPConstrained()
                ? Intrinsic::experimental_constrained_ceil
                : Intrinsic::ceil;
     else if (BuiltinID == PPC::BI__builtin_vsx_xvrdpiz ||
              BuiltinID == PPC::BI__builtin_vsx_xvrspiz)
       ID = Builder.getIsFPConstrained()

clang/test/CodeGen/builtins-ppc-fpconstrained.c

void test_float(void) {
   vd = __builtin_vsx_xvrdpi(vd);
   // CHECK-LABEL: try-xvrdpi
   // CHECK-UNCONSTRAINED: @llvm.round.v2f64(<2 x double> %{{.*}})
   // CHECK-CONSTRAINED: @llvm.experimental.constrained.round.v2f64(<2 x double> %{{.*}}, metadata !"fpexcept.strict")
   // CHECK-ASM: xvrdpi
 
   vf = __builtin_vsx_xvrspic(vf);
   // CHECK-LABEL: try-xvrspic
-  // CHECK-UNCONSTRAINED: @llvm.nearbyint.v4f32(<4 x float> %{{.*}})
-  // CHECK-CONSTRAINED: @llvm.experimental.constrained.nearbyint.v4f32(<4 x float> %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK-UNCONSTRAINED: @llvm.rint.v4f32(<4 x float> %{{.*}})
+  // CHECK-CONSTRAINED: @llvm.experimental.constrained.rint.v4f32(<4 x float> %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
   // CHECK-ASM: xvrspic
 
   vd = __builtin_vsx_xvrdpic(vd);
   // CHECK-LABEL: try-xvrdpic
-  // CHECK-UNCONSTRAINED: @llvm.nearbyint.v2f64(<2 x double> %{{.*}})
-  // CHECK-CONSTRAINED: @llvm.experimental.constrained.nearbyint.v2f64(<2 x double> %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  // CHECK-UNCONSTRAINED: @llvm.rint.v2f64(<2 x double> %{{.*}})
+  // CHECK-CONSTRAINED: @llvm.experimental.constrained.rint.v2f64(<2 x double> %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
   // CHECK-ASM: xvrdpic
 
   vf = __builtin_vsx_xvrspip(vf);
   // CHECK-LABEL: try-xvrspip
   // CHECK-UNCONSTRAINED: @llvm.ceil.v4f32(<4 x float> %{{.*}})
   // CHECK-CONSTRAINED: @llvm.experimental.constrained.ceil.v4f32(<4 x float> %{{.*}}, metadata !"fpexcept.strict")
   // CHECK-ASM: xvrspip
 

clang/test/CodeGen/builtins-ppc-vsx.c

 // CHECK: call <4 x float> @llvm.ppc.vsx.xvresp(<4 x float>
 // CHECK-LE: call <4 x float> @llvm.ppc.vsx.xvresp(<4 x float>
 
   res_vd = vec_re(vd);
 // CHECK: call <2 x double> @llvm.ppc.vsx.xvredp(<2 x double>
 // CHECK-LE: call <2 x double> @llvm.ppc.vsx.xvredp(<2 x double>
 
   res_vf = vec_rint(vf);
-// CHECK: call <4 x float> @llvm.nearbyint.v4f32(<4 x float> %{{[0-9]+}})
-// CHECK-LE: call <4 x float> @llvm.nearbyint.v4f32(<4 x float> %{{[0-9]+}})
+// CHECK: call <4 x float> @llvm.rint.v4f32(<4 x float> %{{[0-9]+}})
+// CHECK-LE: call <4 x float> @llvm.rint.v4f32(<4 x float> %{{[0-9]+}})
 
   res_vd = vec_rint(vd);
-// CHECK: call <2 x double> @llvm.nearbyint.v2f64(<2 x double> %{{[0-9]+}})
-// CHECK-LE: call <2 x double> @llvm.nearbyint.v2f64(<2 x double> %{{[0-9]+}})
+// CHECK: call <2 x double> @llvm.rint.v2f64(<2 x double> %{{[0-9]+}})
+// CHECK-LE: call <2 x double> @llvm.rint.v2f64(<2 x double> %{{[0-9]+}})
 
   res_vf = vec_rsqrte(vf);
 // CHECK: call <4 x float> @llvm.ppc.vsx.xvrsqrtesp(<4 x float> %{{[0-9]+}})
 // CHECK-LE: call <4 x float> @llvm.ppc.vsx.xvrsqrtesp(<4 x float> %{{[0-9]+}})
 
   res_vd = vec_rsqrte(vd);
 // CHECK: call <2 x double> @llvm.ppc.vsx.xvrsqrtedp(<2 x double> %{{[0-9]+}})
 // CHECK-LE: call <2 x double> @llvm.ppc.vsx.xvrsqrtedp(<2 x double> %{{[0-9]+}})

llvm/lib/Target/PowerPC/PPCISelLowering.cpp

PPCTargetLowering::PPCTargetLowering(const PPCTargetMachine &TM,
   setOperationAction(ISD::STRICT_FMA, MVT::f32, Legal);
   setOperationAction(ISD::STRICT_FP_ROUND, MVT::f32, Legal);
 
   setOperationAction(ISD::STRICT_FADD, MVT::f64, Legal);
   setOperationAction(ISD::STRICT_FSUB, MVT::f64, Legal);
   setOperationAction(ISD::STRICT_FMUL, MVT::f64, Legal);
   setOperationAction(ISD::STRICT_FDIV, MVT::f64, Legal);
   setOperationAction(ISD::STRICT_FMA, MVT::f64, Legal);
-  if (Subtarget.hasVSX())
-    setOperationAction(ISD::STRICT_FNEARBYINT, MVT::f64, Legal);
+  if (Subtarget.hasVSX()) {
+    setOperationAction(ISD::STRICT_FRINT, MVT::f32, Legal);
+    setOperationAction(ISD::STRICT_FRINT, MVT::f64, Legal);
+  }
 
   if (Subtarget.hasFSQRT()) {
     setOperationAction(ISD::STRICT_FSQRT, MVT::f32, Legal);
     setOperationAction(ISD::STRICT_FSQRT, MVT::f64, Legal);
   }
 
   if (Subtarget.hasFPRND()) {
     setOperationAction(ISD::STRICT_FFLOOR, MVT::f32, Legal);
if (Subtarget.hasVSX()) {
       setOperationAction(ISD::STRICT_FADD, MVT::v4f32, Legal);
       setOperationAction(ISD::STRICT_FSUB, MVT::v4f32, Legal);
       setOperationAction(ISD::STRICT_FMUL, MVT::v4f32, Legal);
       setOperationAction(ISD::STRICT_FDIV, MVT::v4f32, Legal);
       setOperationAction(ISD::STRICT_FMA, MVT::v4f32, Legal);
       setOperationAction(ISD::STRICT_FSQRT, MVT::v4f32, Legal);
       setOperationAction(ISD::STRICT_FMAXNUM, MVT::v4f32, Legal);
       setOperationAction(ISD::STRICT_FMINNUM, MVT::v4f32, Legal);
-      setOperationAction(ISD::STRICT_FNEARBYINT, MVT::v4f32, Legal);
+      setOperationAction(ISD::STRICT_FRINT, MVT::v4f32, Legal);
       setOperationAction(ISD::STRICT_FFLOOR, MVT::v4f32, Legal);
       setOperationAction(ISD::STRICT_FCEIL,  MVT::v4f32, Legal);
       setOperationAction(ISD::STRICT_FTRUNC, MVT::v4f32, Legal);
       setOperationAction(ISD::STRICT_FROUND, MVT::v4f32, Legal);
 
       setOperationAction(ISD::STRICT_FADD, MVT::v2f64, Legal);
       setOperationAction(ISD::STRICT_FSUB, MVT::v2f64, Legal);
       setOperationAction(ISD::STRICT_FMUL, MVT::v2f64, Legal);
       setOperationAction(ISD::STRICT_FDIV, MVT::v2f64, Legal);
       setOperationAction(ISD::STRICT_FMA, MVT::v2f64, Legal);
       setOperationAction(ISD::STRICT_FSQRT, MVT::v2f64, Legal);
       setOperationAction(ISD::STRICT_FMAXNUM, MVT::v2f64, Legal);
       setOperationAction(ISD::STRICT_FMINNUM, MVT::v2f64, Legal);
-      setOperationAction(ISD::STRICT_FNEARBYINT, MVT::v2f64, Legal);
+      setOperationAction(ISD::STRICT_FRINT, MVT::v2f64, Legal);
       setOperationAction(ISD::STRICT_FFLOOR, MVT::v2f64, Legal);
       setOperationAction(ISD::STRICT_FCEIL,  MVT::v2f64, Legal);
       setOperationAction(ISD::STRICT_FTRUNC, MVT::v2f64, Legal);
       setOperationAction(ISD::STRICT_FROUND, MVT::v2f64, Legal);
 
       addRegisterClass(MVT::v2i64, &PPC::VSRCRegClass);
     }
 

llvm/lib/Target/PowerPC/PPCInstrVSX.td

def XVCVUXWSP : XX2Form<60, 168,
                       (outs vsrc:$XT), (ins vsrc:$XB),
                       "xvcvuxwsp $XT, $XB", IIC_VecFP,
                       [(set v4f32:$XT, (any_uint_to_fp v4i32:$XB))]>;
 
   // Rounding Instructions respecting current rounding mode
   def XSRDPIC : XX2Form<60, 107,
                       (outs vsfrc:$XT), (ins vsfrc:$XB),
                       "xsrdpic $XT, $XB", IIC_VecFP,
-                      [(set f64:$XT, (any_fnearbyint f64:$XB))]>;
+                      [(set f64:$XT, (fnearbyint f64:$XB))]>;
   def XVRDPIC : XX2Form<60, 235,
                       (outs vsrc:$XT), (ins vsrc:$XB),
                       "xvrdpic $XT, $XB", IIC_VecFP,
-                      [(set v2f64:$XT, (any_fnearbyint v2f64:$XB))]>;
+                      [(set v2f64:$XT, (fnearbyint v2f64:$XB))]>;
   def XVRSPIC : XX2Form<60, 171,
                       (outs vsrc:$XT), (ins vsrc:$XB),
                       "xvrspic $XT, $XB", IIC_VecFP,
-                      [(set v4f32:$XT, (any_fnearbyint v4f32:$XB))]>;
+                      [(set v4f32:$XT, (fnearbyint v4f32:$XB))]>;
   // Max/Min Instructions
   let isCommutable = 1 in {
   def XSMAXDP : XX3Form<60, 160,
                         (outs vsfrc:$XT), (ins vsfrc:$XA, vsfrc:$XB),
                         "xsmaxdp $XT, $XA, $XB", IIC_VecFP,
                         [(set vsfrc:$XT,
                               (int_ppc_vsx_xsmaxdp vsfrc:$XA, vsfrc:$XB))]>;
   def XSMINDP : XX3Form<60, 168,
def : Pat<(int_ppc_vsx_stxvw4x_be v4i32:$rS, xoaddr:$dst),
             (STXVW4X $rS, xoaddr:$dst)>;
 def : Pat<(v4i32 (int_ppc_vsx_lxvw4x_be xoaddr:$src)), (LXVW4X xoaddr:$src)>;
 def : Pat<(v2f64 (int_ppc_vsx_lxvd2x_be xoaddr:$src)), (LXVD2X xoaddr:$src)>;
 
 // Rounding for single precision.
 def : Pat<(f32 (any_fround f32:$S)),
           (f32 (COPY_TO_REGCLASS (XSRDPI
                                    (COPY_TO_REGCLASS $S, VSFRC)), VSSRC))>;
-def : Pat<(f32 (any_fnearbyint f32:$S)),
+def : Pat<(f32 (fnearbyint f32:$S)),
           (f32 (COPY_TO_REGCLASS (XSRDPIC
                                    (COPY_TO_REGCLASS $S, VSFRC)), VSSRC))>;
 def : Pat<(f32 (any_ffloor f32:$S)),
           (f32 (COPY_TO_REGCLASS (XSRDPIM
                                    (COPY_TO_REGCLASS $S, VSFRC)), VSSRC))>;
 def : Pat<(f32 (any_fceil f32:$S)),
           (f32 (COPY_TO_REGCLASS (XSRDPIP
                                    (COPY_TO_REGCLASS $S, VSFRC)), VSSRC))>;
 def : Pat<(f32 (any_ftrunc f32:$S)),
           (f32 (COPY_TO_REGCLASS (XSRDPIZ
                                    (COPY_TO_REGCLASS $S, VSFRC)), VSSRC))>;
 def : Pat<(f32 (any_frint f32:$S)),
           (f32 (COPY_TO_REGCLASS (XSRDPIC
                                    (COPY_TO_REGCLASS $S, VSFRC)), VSSRC))>;
-def : Pat<(v4f32 (frint v4f32:$S)), (v4f32 (XVRSPIC $S))>;
+def : Pat<(v4f32 (any_frint v4f32:$S)), (v4f32 (XVRSPIC $S))>;
 
 // Rounding for double precision.
-def : Pat<(f64 (frint f64:$S)), (f64 (XSRDPIC $S))>;
-def : Pat<(v2f64 (frint v2f64:$S)), (v2f64 (XVRDPIC $S))>;
+def : Pat<(f64 (any_frint f64:$S)), (f64 (XSRDPIC $S))>;
+def : Pat<(v2f64 (any_frint v2f64:$S)), (v2f64 (XVRDPIC $S))>;
 
 // Materialize a zero-vector of long long
 def : Pat<(v2i64 immAllZerosV),
           (v2i64 (XXLXORz))>;
 
 // Build vectors of floating point converted to i32.
 def : Pat<(v4i32 (build_vector DblToInt.A, DblToInt.A,
                                DblToInt.A, DblToInt.A)),

llvm/test/CodeGen/PowerPC/fp-strict-round.ll

%res = call <2 x double> @llvm.experimental.constrained.floor.v2f64(
                         <2 x double> %vf1,
                         metadata !"fpexcept.strict")
   ret <2 x double> %res;
 }
 
 define double @nearbyint_f64(double %f1, double %f2) {
 ; P8-LABEL: nearbyint_f64:
 ; P8:       # %bb.0:
-; P8-NEXT:    xsrdpic f1, f1
+; P8-NEXT:    mflr r0
+; P8-NEXT:    std r0, 16(r1)
+; P8-NEXT:    stdu r1, -112(r1)
+; P8-NEXT:    .cfi_def_cfa_offset 112
+; P8-NEXT:    .cfi_offset lr, 16
+; P8-NEXT:    bl nearbyint
+; P8-NEXT:    nop
+; P8-NEXT:    addi r1, r1, 112
+; P8-NEXT:    ld r0, 16(r1)
+; P8-NEXT:    mtlr r0
 ; P8-NEXT:    blr
 ;
 ; P9-LABEL: nearbyint_f64:
 ; P9:       # %bb.0:
-; P9-NEXT:    xsrdpic f1, f1
+; P9-NEXT:    mflr r0
+; P9-NEXT:    std r0, 16(r1)
+; P9-NEXT:    stdu r1, -32(r1)
+; P9-NEXT:    .cfi_def_cfa_offset 32
+; P9-NEXT:    .cfi_offset lr, 16
+; P9-NEXT:    bl nearbyint
+; P9-NEXT:    nop
+; P9-NEXT:    addi r1, r1, 32
+; P9-NEXT:    ld r0, 16(r1)
+; P9-NEXT:    mtlr r0
 ; P9-NEXT:    blr
   %res = call double @llvm.experimental.constrained.nearbyint.f64(
                         double %f1,
                         metadata !"round.dynamic",
                         metadata !"fpexcept.strict")
   ret double %res
 }
 
 define <4 x float> @nearbyint_v4f32(<4 x float> %vf1, <4 x float> %vf2) {
 ; P8-LABEL: nearbyint_v4f32:
 ; P8:       # %bb.0:
-; P8-NEXT:    xvrspic v2, v2
+; P8-NEXT:    mflr r0
+; P8-NEXT:    std r0, 16(r1)
+; P8-NEXT:    stdu r1, -176(r1)
+; P8-NEXT:    .cfi_def_cfa_offset 176
+; P8-NEXT:    .cfi_offset lr, 16
+; P8-NEXT:    .cfi_offset v30, -32
+; P8-NEXT:    .cfi_offset v31, -16
+; P8-NEXT:    xxsldwi vs0, v2, v2, 3
+; P8-NEXT:    li r3, 144
+; P8-NEXT:    stxvd2x v30, r1, r3 # 16-byte Folded Spill
+; P8-NEXT:    li r3, 160
+; P8-NEXT:    stxvd2x v31, r1, r3 # 16-byte Folded Spill
+; P8-NEXT:    vmr v31, v2
+; P8-NEXT:    xscvspdpn f1, vs0
+; P8-NEXT:    bl nearbyintf
+; P8-NEXT:    nop
+; P8-NEXT:    xxsldwi vs0, v31, v31, 1
+; P8-NEXT:    # kill: def $f1 killed $f1 def $vsl1
+; P8-NEXT:    li r3, 128
+; P8-NEXT:    stxvd2x vs1, r1, r3 # 16-byte Folded Spill
+; P8-NEXT:    xscvspdpn f1, vs0
+; P8-NEXT:    bl nearbyintf
+; P8-NEXT:    nop
+; P8-NEXT:    li r3, 128
+; P8-NEXT:    # kill: def $f1 killed $f1 def $vsl1
+; P8-NEXT:    lxvd2x vs0, r1, r3 # 16-byte Folded Reload
+; P8-NEXT:    xxmrghd vs0, vs1, vs0
+; P8-NEXT:    xscvspdpn f1, v31
+; P8-NEXT:    xvcvdpsp v30, vs0
+; P8-NEXT:    bl nearbyintf
+; P8-NEXT:    nop
+; P8-NEXT:    xxswapd vs0, v31
+; P8-NEXT:    # kill: def $f1 killed $f1 def $vsl1
+; P8-NEXT:    li r3, 128
+; P8-NEXT:    stxvd2x vs1, r1, r3 # 16-byte Folded Spill
+; P8-NEXT:    xscvspdpn f1, vs0
+; P8-NEXT:    bl nearbyintf
+; P8-NEXT:    nop
+; P8-NEXT:    li r3, 128
+; P8-NEXT:    # kill: def $f1 killed $f1 def $vsl1
+; P8-NEXT:    lxvd2x vs0, r1, r3 # 16-byte Folded Reload
+; P8-NEXT:    li r3, 160
+; P8-NEXT:    lxvd2x v31, r1, r3 # 16-byte Folded Reload
+; P8-NEXT:    li r3, 144
+; P8-NEXT:    xxmrghd vs0, vs0, vs1
+; P8-NEXT:    xvcvdpsp v2, vs0
+; P8-NEXT:    vmrgew v2, v2, v30
+; P8-NEXT:    lxvd2x v30, r1, r3 # 16-byte Folded Reload
+; P8-NEXT:    addi r1, r1, 176
+; P8-NEXT:    ld r0, 16(r1)
+; P8-NEXT:    mtlr r0
 ; P8-NEXT:    blr
 ;
 ; P9-LABEL: nearbyint_v4f32:
 ; P9:       # %bb.0:
-; P9-NEXT:    xvrspic v2, v2
+; P9-NEXT:    mflr r0
+; P9-NEXT:    std r0, 16(r1)
+; P9-NEXT:    stdu r1, -80(r1)
+; P9-NEXT:    .cfi_def_cfa_offset 80
+; P9-NEXT:    .cfi_offset lr, 16
+; P9-NEXT:    .cfi_offset v30, -32
+; P9-NEXT:    .cfi_offset v31, -16
+; P9-NEXT:    xxsldwi vs0, v2, v2, 3
+; P9-NEXT:    stxv v30, 48(r1) # 16-byte Folded Spill
+; P9-NEXT:    xscvspdpn f1, vs0
+; P9-NEXT:    stxv v31, 64(r1) # 16-byte Folded Spill
+; P9-NEXT:    vmr v31, v2
+; P9-NEXT:    bl nearbyintf
+; P9-NEXT:    nop
+; P9-NEXT:    xxsldwi vs0, v31, v31, 1
+; P9-NEXT:    # kill: def $f1 killed $f1 def $vsl1
+; P9-NEXT:    stxv vs1, 32(r1) # 16-byte Folded Spill
+; P9-NEXT:    xscvspdpn f1, vs0
+; P9-NEXT:    bl nearbyintf
+; P9-NEXT:    nop
+; P9-NEXT:    lxv vs0, 32(r1) # 16-byte Folded Reload
+; P9-NEXT:    # kill: def $f1 killed $f1 def $vsl1
+; P9-NEXT:    xxmrghd vs0, vs1, vs0
+; P9-NEXT:    xscvspdpn f1, v31
+; P9-NEXT:    xvcvdpsp v30, vs0
+; P9-NEXT:    bl nearbyintf
+; P9-NEXT:    nop
+; P9-NEXT:    xxswapd vs0, v31
+; P9-NEXT:    # kill: def $f1 killed $f1 def $vsl1
+; P9-NEXT:    stxv vs1, 32(r1) # 16-byte Folded Spill
+; P9-NEXT:    xscvspdpn f1, vs0
+; P9-NEXT:    bl nearbyintf
+; P9-NEXT:    nop
+; P9-NEXT:    lxv vs0, 32(r1) # 16-byte Folded Reload
+; P9-NEXT:    # kill: def $f1 killed $f1 def $vsl1
+; P9-NEXT:    lxv v31, 64(r1) # 16-byte Folded Reload
+; P9-NEXT:    xxmrghd vs0, vs0, vs1
+; P9-NEXT:    xvcvdpsp v2, vs0
+; P9-NEXT:    vmrgew v2, v2, v30
+; P9-NEXT:    lxv v30, 48(r1) # 16-byte Folded Reload
+; P9-NEXT:    addi r1, r1, 80
+; P9-NEXT:    ld r0, 16(r1)
+; P9-NEXT:    mtlr r0
 ; P9-NEXT:    blr
   %res = call <4 x float> @llvm.experimental.constrained.nearbyint.v4f32(
                         <4 x float> %vf1,
                         metadata !"round.dynamic",
                         metadata !"fpexcept.strict")
   ret <4 x float> %res
 }
 
 define <2 x double> @nearbyint_v2f64(<2 x double> %vf1, <2 x double> %vf2) {
 ; P8-LABEL: nearbyint_v2f64:
 ; P8:       # %bb.0:
-; P8-NEXT:    xvrdpic v2, v2
+; P8-NEXT:    mflr r0
+; P8-NEXT:    std r0, 16(r1)
+; P8-NEXT:    stdu r1, -160(r1)
+; P8-NEXT:    .cfi_def_cfa_offset 160
+; P8-NEXT:    .cfi_offset lr, 16
+; P8-NEXT:    .cfi_offset v31, -16
+; P8-NEXT:    li r3, 144
+; P8-NEXT:    stxvd2x v31, r1, r3 # 16-byte Folded Spill
+; P8-NEXT:    vmr v31, v2
+; P8-NEXT:    xxlor f1, v31, v31
+; P8-NEXT:    bl nearbyint
+; P8-NEXT:    nop
+; P8-NEXT:    # kill: def $f1 killed $f1 def $vsl1
+; P8-NEXT:    li r3, 128
+; P8-NEXT:    stxvd2x vs1, r1, r3 # 16-byte Folded Spill
+; P8-NEXT:    xxswapd vs1, v31
+; P8-NEXT:    # kill: def $f1 killed $f1 killed $vsl1
+; P8-NEXT:    bl nearbyint
+; P8-NEXT:    nop
+; P8-NEXT:    li r3, 128
+; P8-NEXT:    # kill: def $f1 killed $f1 def $vsl1
+; P8-NEXT:    lxvd2x vs0, r1, r3 # 16-byte Folded Reload
+; P8-NEXT:    li r3, 144
+; P8-NEXT:    lxvd2x v31, r1, r3 # 16-byte Folded Reload
+; P8-NEXT:    xxmrghd v2, vs0, vs1
+; P8-NEXT:    addi r1, r1, 160
+; P8-NEXT:    ld r0, 16(r1)
+; P8-NEXT:    mtlr r0
 ; P8-NEXT:    blr
 ;
 ; P9-LABEL: nearbyint_v2f64:
 ; P9:       # %bb.0:
-; P9-NEXT:    xvrdpic v2, v2
+; P9-NEXT:    mflr r0
+; P9-NEXT:    std r0, 16(r1)
+; P9-NEXT:    stdu r1, -64(r1)
+; P9-NEXT:    .cfi_def_cfa_offset 64
+; P9-NEXT:    .cfi_offset lr, 16
+; P9-NEXT:    .cfi_offset v31, -16
+; P9-NEXT:    stxv v31, 48(r1) # 16-byte Folded Spill
+; P9-NEXT:    vmr v31, v2
+; P9-NEXT:    xscpsgndp f1, v31, v31
+; P9-NEXT:    bl nearbyint
+; P9-NEXT:    nop
+; P9-NEXT:    # kill: def $f1 killed $f1 def $vsl1
+; P9-NEXT:    stxv vs1, 32(r1) # 16-byte Folded Spill
+; P9-NEXT:    xxswapd vs1, v31
+; P9-NEXT:    # kill: def $f1 killed $f1 killed $vsl1
+; P9-NEXT:    bl nearbyint
+; P9-NEXT:    nop
+; P9-NEXT:    lxv vs0, 32(r1) # 16-byte Folded Reload
+; P9-NEXT:    lxv v31, 48(r1) # 16-byte Folded Reload
+; P9-NEXT:    # kill: def $f1 killed $f1 def $vsl1
+; P9-NEXT:    xxmrghd v2, vs0, vs1
+; P9-NEXT:    addi r1, r1, 64
+; P9-NEXT:    ld r0, 16(r1)
+; P9-NEXT:    mtlr r0
 ; P9-NEXT:    blr
   %res = call <2 x double> @llvm.experimental.constrained.nearbyint.v2f64(
                         <2 x double> %vf1,
                         metadata !"round.dynamic",
                         metadata !"fpexcept.strict")
   ret <2 x double> %res
 }
 

llvm/test/CodeGen/PowerPC/vector-constrained-fp-intrinsics.ll

%nearby = call <1 x float> @llvm.experimental.constrained.nearbyint.v1f32(
                                metadata !"round.dynamic",
                                metadata !"fpexcept.strict") #1
   ret <1 x float> %nearby
 }
 
 define <2 x double> @constrained_vector_nearbyint_v2f64() #0 {
 ; PC64LE-LABEL: constrained_vector_nearbyint_v2f64:
 ; PC64LE:       # %bb.0: # %entry
+; PC64LE-NEXT:    mflr 0
+; PC64LE-NEXT:    std 0, 16(1)
+; PC64LE-NEXT:    stdu 1, -64(1)
 ; PC64LE-NEXT:    addis 3, 2, .LCPI81_0@toc@ha
-; PC64LE-NEXT:    addi 3, 3, .LCPI81_0@toc@l
-; PC64LE-NEXT:    lxvd2x 0, 0, 3
-; PC64LE-NEXT:    xxswapd 0, 0
-; PC64LE-NEXT:    xvrdpic 34, 0
+; PC64LE-NEXT:    lfd 1, .LCPI81_0@toc@l(3)
+; PC64LE-NEXT:    bl nearbyint
+; PC64LE-NEXT:    nop
+; PC64LE-NEXT:    li 3, 48
+; PC64LE-NEXT:    # kill: def $f1 killed $f1 def $vsl1
+; PC64LE-NEXT:    stxvd2x 1, 1, 3 # 16-byte Folded Spill
+; PC64LE-NEXT:    addis 3, 2, .LCPI81_1@toc@ha
+; PC64LE-NEXT:    lfs 1, .LCPI81_1@toc@l(3)
+; PC64LE-NEXT:    bl nearbyint
+; PC64LE-NEXT:    nop
+; PC64LE-NEXT:    li 3, 48
+; PC64LE-NEXT:    # kill: def $f1 killed $f1 def $vsl1
+; PC64LE-NEXT:    lxvd2x 0, 1, 3 # 16-byte Folded Reload
+; PC64LE-NEXT:    xxmrghd 34, 1, 0
+; PC64LE-NEXT:    addi 1, 1, 64
+; PC64LE-NEXT:    ld 0, 16(1)
+; PC64LE-NEXT:    mtlr 0
 ; PC64LE-NEXT:    blr
 ;
 ; PC64LE9-LABEL: constrained_vector_nearbyint_v2f64:
 ; PC64LE9:       # %bb.0: # %entry
+; PC64LE9-NEXT:    mflr 0
+; PC64LE9-NEXT:    std 0, 16(1)
+; PC64LE9-NEXT:    stdu 1, -48(1)
 ; PC64LE9-NEXT:    addis 3, 2, .LCPI81_0@toc@ha
-; PC64LE9-NEXT:    addi 3, 3, .LCPI81_0@toc@l
-; PC64LE9-NEXT:    lxvx 0, 0, 3
-; PC64LE9-NEXT:    xvrdpic 34, 0
+; PC64LE9-NEXT:    lfd 1, .LCPI81_0@toc@l(3)
+; PC64LE9-NEXT:    bl nearbyint
+; PC64LE9-NEXT:    nop
+; PC64LE9-NEXT:    addis 3, 2, .LCPI81_1@toc@ha
+; PC64LE9-NEXT:    # kill: def $f1 killed $f1 def $vsl1
+; PC64LE9-NEXT:    stxv 1, 32(1) # 16-byte Folded Spill
+; PC64LE9-NEXT:    lfs 1, .LCPI81_1@toc@l(3)
+; PC64LE9-NEXT:    bl nearbyint
+; PC64LE9-NEXT:    nop
+; PC64LE9-NEXT:    lxv 0, 32(1) # 16-byte Folded Reload
+; PC64LE9-NEXT:    # kill: def $f1 killed $f1 def $vsl1
+; PC64LE9-NEXT:    xxmrghd 34, 1, 0
+; PC64LE9-NEXT:    addi 1, 1, 48
+; PC64LE9-NEXT:    ld 0, 16(1)
+; PC64LE9-NEXT:    mtlr 0
 ; PC64LE9-NEXT:    blr
 entry:
   %nearby = call <2 x double> @llvm.experimental.constrained.nearbyint.v2f64(
                                 <2 x double> <double 42.1, double 42.0>,
                                 metadata !"round.dynamic",
                                 metadata !"fpexcept.strict") #1
   ret <2 x double> %nearby
 }
%nearby = call <3 x float> @llvm.experimental.constrained.nearbyint.v3f32(
                               metadata !"round.dynamic",
                               metadata !"fpexcept.strict") #1
   ret <3 x float> %nearby
 }
 
 define <3 x double> @constrained_vector_nearby_v3f64() #0 {
 ; PC64LE-LABEL: constrained_vector_nearby_v3f64:
 ; PC64LE:       # %bb.0: # %entry
-; PC64LE-NEXT:    addis 3, 2, .LCPI83_1@toc@ha
-; PC64LE-NEXT:    addi 3, 3, .LCPI83_1@toc@l
-; PC64LE-NEXT:    lxvd2x 0, 0, 3
+; PC64LE-NEXT:    mflr 0
+; PC64LE-NEXT:    std 0, 16(1)
+; PC64LE-NEXT:    stdu 1, -80(1)
+; PC64LE-NEXT:    li 3, 64
+; PC64LE-NEXT:    stxvd2x 63, 1, 3 # 16-byte Folded Spill
 ; PC64LE-NEXT:    addis 3, 2, .LCPI83_0@toc@ha
 ; PC64LE-NEXT:    lfd 1, .LCPI83_0@toc@l(3)
-; PC64LE-NEXT:    xxswapd 0, 0
-; PC64LE-NEXT:    xsrdpic 3, 1
-; PC64LE-NEXT:    xvrdpic 2, 0
-; PC64LE-NEXT:    xxswapd 1, 2
-; PC64LE-NEXT:    # kill: def $f2 killed $f2 killed $vsl2
-; PC64LE-NEXT:    # kill: def $f1 killed $f1 killed $vsl1
+; PC64LE-NEXT:    bl nearbyint
+; PC64LE-NEXT:    nop
+; PC64LE-NEXT:    li 3, 48
+; PC64LE-NEXT:    # kill: def $f1 killed $f1 def $vsl1
+; PC64LE-NEXT:    stxvd2x 1, 1, 3 # 16-byte Folded Spill
+; PC64LE-NEXT:    addis 3, 2, .LCPI83_1@toc@ha
+; PC64LE-NEXT:    lfs 1, .LCPI83_1@toc@l(3)
+; PC64LE-NEXT:    bl nearbyint
+; PC64LE-NEXT:    nop
+; PC64LE-NEXT:    li 3, 48
+; PC64LE-NEXT:    # kill: def $f1 killed $f1 def $vsl1
+; PC64LE-NEXT:    lxvd2x 0, 1, 3 # 16-byte Folded Reload
+; PC64LE-NEXT:    addis 3, 2, .LCPI83_2@toc@ha
+; PC64LE-NEXT:    xxmrghd 63, 0, 1
+; PC64LE-NEXT:    lfd 1, .LCPI83_2@toc@l(3)
+; PC64LE-NEXT:    bl nearbyint
+; PC64LE-NEXT:    nop
+; PC64LE-NEXT:    xxswapd 0, 63
+; PC64LE-NEXT:    li 3, 64
+; PC64LE-NEXT:    xxlor 2, 63, 63
+; PC64LE-NEXT:    lxvd2x 63, 1, 3 # 16-byte Folded Reload
+; PC64LE-NEXT:    fmr 3, 1
+; PC64LE-NEXT:    fmr 1, 0
+; PC64LE-NEXT:    addi 1, 1, 80
+; PC64LE-NEXT:    ld 0, 16(1)
+; PC64LE-NEXT:    mtlr 0
 ; PC64LE-NEXT:    blr
 ;
 ; PC64LE9-LABEL: constrained_vector_nearby_v3f64:
 ; PC64LE9:       # %bb.0: # %entry
+; PC64LE9-NEXT:    mflr 0
+; PC64LE9-NEXT:    std 0, 16(1)
+; PC64LE9-NEXT:    stdu 1, -64(1)
 ; PC64LE9-NEXT:    addis 3, 2, .LCPI83_0@toc@ha
-; PC64LE9-NEXT:    lfd 0, .LCPI83_0@toc@l(3)
+; PC64LE9-NEXT:    stxv 63, 48(1) # 16-byte Folded Spill
+; PC64LE9-NEXT:    lfd 1, .LCPI83_0@toc@l(3)
+; PC64LE9-NEXT:    bl nearbyint
+; PC64LE9-NEXT:    nop
 ; PC64LE9-NEXT:    addis 3, 2, .LCPI83_1@toc@ha
-; PC64LE9-NEXT:    addi 3, 3, .LCPI83_1@toc@l
-; PC64LE9-NEXT:    xsrdpic 3, 0
-; PC64LE9-NEXT:    lxvx 0, 0, 3
-; PC64LE9-NEXT:    xvrdpic 2, 0
-; PC64LE9-NEXT:    xxswapd 1, 2
+; PC64LE9-NEXT:    # kill: def $f1 killed $f1 def $vsl1
+; PC64LE9-NEXT:    stxv 1, 32(1) # 16-byte Folded Spill
+; PC64LE9-NEXT:    lfs 1, .LCPI83_1@toc@l(3)
+; PC64LE9-NEXT:    bl nearbyint
+; PC64LE9-NEXT:    nop
+; PC64LE9-NEXT:    lxv 0, 32(1) # 16-byte Folded Reload
+; PC64LE9-NEXT:    addis 3, 2, .LCPI83_2@toc@ha
+; PC64LE9-NEXT:    # kill: def $f1 killed $f1 def $vsl1
+; PC64LE9-NEXT:    xxmrghd 63, 0, 1
+; PC64LE9-NEXT:    lfd 1, .LCPI83_2@toc@l(3)
+; PC64LE9-NEXT:    bl nearbyint
+; PC64LE9-NEXT:    nop
+; PC64LE9-NEXT:    fmr 3, 1
+; PC64LE9-NEXT:    xxswapd 1, 63
+; PC64LE9-NEXT:    xscpsgndp 2, 63, 63
+; PC64LE9-NEXT:    lxv 63, 48(1) # 16-byte Folded Reload
 ; PC64LE9-NEXT:    # kill: def $f1 killed $f1 killed $vsl1
-; PC64LE9-NEXT:    # kill: def $f2 killed $f2 killed $vsl2
+; PC64LE9-NEXT:    addi 1, 1, 64
+; PC64LE9-NEXT:    ld 0, 16(1)
+; PC64LE9-NEXT:    mtlr 0
 ; PC64LE9-NEXT:    blr
 entry:
   %nearby = call <3 x double> @llvm.experimental.constrained.nearbyint.v3f64(
                           <3 x double> <double 42.0, double 42.1, double 42.2>,
                           metadata !"round.dynamic",
                           metadata !"fpexcept.strict") #1
   ret <3 x double> %nearby
 }
 
 define <4 x double> @constrained_vector_nearbyint_v4f64() #0 {
 ; PC64LE-LABEL: constrained_vector_nearbyint_v4f64:
 ; PC64LE:       # %bb.0: # %entry
+; PC64LE-NEXT:    mflr 0
+; PC64LE-NEXT:    std 0, 16(1)
+; PC64LE-NEXT:    stdu 1, -80(1)
+; PC64LE-NEXT:    li 3, 64
+; PC64LE-NEXT:    stxvd2x 63, 1, 3 # 16-byte Folded Spill
 ; PC64LE-NEXT:    addis 3, 2, .LCPI84_0@toc@ha
-; PC64LE-NEXT:    addis 4, 2, .LCPI84_1@toc@ha
-; PC64LE-NEXT:    addi 3, 3, .LCPI84_0@toc@l
-; PC64LE-NEXT:    lxvd2x 0, 0, 3
-; PC64LE-NEXT:    addi 3, 4, .LCPI84_1@toc@l
-; PC64LE-NEXT:    lxvd2x 1, 0, 3
-; PC64LE-NEXT:    xxswapd 0, 0
-; PC64LE-NEXT:    xxswapd 1, 1
-; PC64LE-NEXT:    xvrdpic 35, 0
-; PC64LE-NEXT:    xvrdpic 34, 1
+; PC64LE-NEXT:    lfd 1, .LCPI84_0@toc@l(3)
+; PC64LE-NEXT:    bl nearbyint
+; PC64LE-NEXT:    nop
+; PC64LE-NEXT:    li 3, 48
+; PC64LE-NEXT:    # kill: def $f1 killed $f1 def $vsl1
+; PC64LE-NEXT:    stxvd2x 1, 1, 3 # 16-byte Folded Spill
+; PC64LE-NEXT:    addis 3, 2, .LCPI84_1@toc@ha
+; PC64LE-NEXT:    lfd 1, .LCPI84_1@toc@l(3)
+; PC64LE-NEXT:    bl nearbyint
+; PC64LE-NEXT:    nop
+; PC64LE-NEXT:    li 3, 48
+; PC64LE-NEXT:    # kill: def $f1 killed $f1 def $vsl1
+; PC64LE-NEXT:    lxvd2x 0, 1, 3 # 16-byte Folded Reload
+; PC64LE-NEXT:    addis 3, 2, .LCPI84_2@toc@ha
+; PC64LE-NEXT:    xxmrghd 63, 1, 0
+; PC64LE-NEXT:    lfd 1, .LCPI84_2@toc@l(3)
+; PC64LE-NEXT:    bl nearbyint
+; PC64LE-NEXT:    nop
+; PC64LE-NEXT:    li 3, 48
+; PC64LE-NEXT:    # kill: def $f1 killed $f1 def $vsl1
+; PC64LE-NEXT:    stxvd2x 1, 1, 3 # 16-byte Folded Spill
+; PC64LE-NEXT:    addis 3, 2, .LCPI84_3@toc@ha
+; PC64LE-NEXT:    lfd 1, .LCPI84_3@toc@l(3)
+; PC64LE-NEXT:    bl nearbyint
+; PC64LE-NEXT:    nop
+; PC64LE-NEXT:    li 3, 48
+; PC64LE-NEXT:    vmr 2, 31
+; PC64LE-NEXT:    # kill: def $f1 killed $f1 def $vsl1
+; PC64LE-NEXT:    lxvd2x 0, 1, 3 # 16-byte Folded Reload
+; PC64LE-NEXT:    li 3, 64
+; PC64LE-NEXT:    lxvd2x 63, 1, 3 # 16-byte Folded Reload
+; PC64LE-NEXT:    xxmrghd 35, 1, 0
+; PC64LE-NEXT:    addi 1, 1, 80
+; PC64LE-NEXT:    ld 0, 16(1)
+; PC64LE-NEXT:    mtlr 0
 ; PC64LE-NEXT:    blr
 ;
 ; PC64LE9-LABEL: constrained_vector_nearbyint_v4f64:
 ; PC64LE9:       # %bb.0: # %entry
+; PC64LE9-NEXT:    mflr 0
+; PC64LE9-NEXT:    std 0, 16(1)
+; PC64LE9-NEXT:    stdu 1, -64(1)
 ; PC64LE9-NEXT:    addis 3, 2, .LCPI84_0@toc@ha
-; PC64LE9-NEXT:    addi 3, 3, .LCPI84_0@toc@l
-; PC64LE9-NEXT:    lxvx 0, 0, 3
+; PC64LE9-NEXT:    stxv 63, 48(1) # 16-byte Folded Spill
+; PC64LE9-NEXT:    lfd 1, .LCPI84_0@toc@l(3)
+; PC64LE9-NEXT:    bl nearbyint
+; PC64LE9-NEXT:    nop
 ; PC64LE9-NEXT:    addis 3, 2, .LCPI84_1@toc@ha
-; PC64LE9-NEXT:    addi 3, 3, .LCPI84_1@toc@l
-; PC64LE9-NEXT:    xvrdpic 35, 0
-; PC64LE9-NEXT:    lxvx 0, 0, 3
-; PC64LE9-NEXT:    xvrdpic 34, 0
+; PC64LE9-NEXT:    # kill: def $f1 killed $f1 def $vsl1
+; PC64LE9-NEXT:    stxv 1, 32(1) # 16-byte Folded Spill
+; PC64LE9-NEXT:    lfd 1, .LCPI84_1@toc@l(3)
+; PC64LE9-NEXT:    bl nearbyint
+; PC64LE9-NEXT:    nop
+; PC64LE9-NEXT:    lxv 0, 32(1) # 16-byte Folded Reload
+; PC64LE9-NEXT:    addis 3, 2, .LCPI84_2@toc@ha
+; PC64LE9-NEXT:    # kill: def $f1 killed $f1 def $vsl1
+; PC64LE9-NEXT:    xxmrghd 63, 1, 0
+; PC64LE9-NEXT:    lfd 1, .LCPI84_2@toc@l(3)
+; PC64LE9-NEXT:    bl nearbyint
+; PC64LE9-NEXT:    nop
+; PC64LE9-NEXT:    addis 3, 2, .LCPI84_3@toc@ha
+; PC64LE9-NEXT:    # kill: def $f1 killed $f1 def $vsl1
+; PC64LE9-NEXT:    stxv 1, 32(1) # 16-byte Folded Spill
+; PC64LE9-NEXT:    lfd 1, .LCPI84_3@toc@l(3)
+; PC64LE9-NEXT:    bl nearbyint
+; PC64LE9-NEXT:    nop
+; PC64LE9-NEXT:    lxv 0, 32(1) # 16-byte Folded Reload
+; PC64LE9-NEXT:    vmr 2, 31
+; PC64LE9-NEXT:    lxv 63, 48(1) # 16-byte Folded Reload
+; PC64LE9-NEXT:    # kill: def $f1 killed $f1 def $vsl1
+; PC64LE9-NEXT:    xxmrghd 35, 1, 0
+; PC64LE9-NEXT:    addi 1, 1, 64
+; PC64LE9-NEXT:    ld 0, 16(1)
+; PC64LE9-NEXT:    mtlr 0
 ; PC64LE9-NEXT:    blr
 entry:
   %nearby = call <4 x double> @llvm.experimental.constrained.nearbyint.v4f64(
                                 <4 x double> <double 42.1, double 42.2,
                                               double 42.3, double 42.4>,
                                 metadata !"round.dynamic",
                                 metadata !"fpexcept.strict") #1
   ret <4 x double> %nearby