PowerPC: Fix SPE extloadf32 handling.

[jhibbits]

Summary

The patterns were incorrect copies from the FPU code, and are
unnecessary, since there's no extended load for SPE. Just let LLVM
itself do the work by marking it expand.

[jhibbits]

Ping?

[jhibbits]

Ping again? Since it's a trivial change, and isolated at that, I'm not opposed to post-commit review instead.

[lkail]

LGTM. Might need to rebase.

llvm/lib/Target/PowerPC/PPCISelLowering.cpp

PPCTargetLowering::PPCTargetLowering(const PPCTargetMachine &TM,
   if (Subtarget.hasSPE()) {
     setOperationAction(ISD::FMA  , MVT::f64, Expand);
     setOperationAction(ISD::FMA  , MVT::f32, Expand);
   } else {
     setOperationAction(ISD::FMA  , MVT::f64, Legal);
     setOperationAction(ISD::FMA  , MVT::f32, Legal);
   }
 
+  if (Subtarget.hasSPE())
+    setLoadExtAction(ISD::EXTLOAD, MVT::f64, MVT::f32, Expand);
+
   setOperationAction(ISD::FLT_ROUNDS_, MVT::i32, Custom);
 
   // If we're enabling GP optimizations, use hardware square root
   if (!Subtarget.hasFSQRT() &&
       !(TM.Options.UnsafeFPMath && Subtarget.hasFRSQRTE() &&
         Subtarget.hasFRE()))
     setOperationAction(ISD::FSQRT, MVT::f64, Expand);
 

llvm/lib/Target/PowerPC/PPCInstrSPE.td

 def SPESTWX       : XForm_8<31, 151, (outs), (ins spe4rc:$rS, memrr:$dst),
                            "stwx $rS, $dst", IIC_LdStStore,
                            [(store f32:$rS, xaddr:$dst)]>;
 }
 
 } // HasSPE
 
 let Predicates = [HasSPE] in {
-def : Pat<(f64 (extloadf32 iaddr:$src)),
-          (COPY_TO_REGCLASS (SPELWZ iaddr:$src), SPERC)>;
-def : Pat<(f64 (extloadf32 xaddr:$src)),
-          (COPY_TO_REGCLASS (SPELWZX xaddr:$src), SPERC)>;
-
-def : Pat<(f64 (fpextend f32:$src)),
-          (COPY_TO_REGCLASS $src, SPERC)>;
-}
-
-let Predicates = [HasSPE] in {
 def SELECT_CC_SPE4 : PPCCustomInserterPseudo<(outs spe4rc:$dst),
                             (ins crrc:$cond, spe4rc:$T, spe4rc:$F,
                             i32imm:$BROPC), "#SELECT_CC_SPE4",
                             []>;
 def SELECT_CC_SPE  : PPCCustomInserterPseudo<(outs sperc:$dst),
                             (ins crrc:$cond, sperc:$T, sperc:$F, i32imm:$BROPC),
                             "#SELECT_CC_SPE",
                             []>;

llvm/test/CodeGen/PowerPC/spe.ll

for.body:                                         ; preds = %for.body, %entry
   %exitcond = icmp eq i32 %inc, %d
   br i1 %exitcond, label %for.cond.cleanup, label %for.body
 }
 
 ; Function Attrs: nounwind readnone speculatable willreturn
 declare float @llvm.fma.f32(float, float, float) #1
 
 attributes #1 = { nounwind readnone speculatable willreturn }
+
+%struct.a = type { float, float }
+
+define void @d(%struct.a* %e, %struct.a* %f) {
+; CHECK-LABEL: d:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    mflr 0
+; CHECK-NEXT:    stw 0, 4(1)
+; CHECK-NEXT:    stwu 1, -48(1)
+; CHECK-NEXT:    .cfi_def_cfa_offset 48
+; CHECK-NEXT:    .cfi_offset lr, 4
+; CHECK-NEXT:    .cfi_offset r29, -12
+; CHECK-NEXT:    .cfi_offset r30, -8
+; CHECK-NEXT:    .cfi_offset r29, -40
+; CHECK-NEXT:    .cfi_offset r30, -32
+; CHECK-NEXT:    lwz 4, 0(4)
+; CHECK-NEXT:    lwz 3, 0(3)
+; CHECK-NEXT:    stw 29, 36(1) # 4-byte Folded Spill
+; CHECK-NEXT:    evstdd 29, 8(1) # 8-byte Folded Spill
+; CHECK-NEXT:    efdcfs 29, 4
+; CHECK-NEXT:    stw 30, 40(1) # 4-byte Folded Spill
+; CHECK-NEXT:    mr 4, 29
+; CHECK-NEXT:    evstdd 30, 16(1) # 8-byte Folded Spill
+; CHECK-NEXT:    efdcfs 30, 3
+; CHECK-NEXT:    evmergehi 3, 29, 29
+; CHECK-NEXT:    mtctr 3
+; CHECK-NEXT:    # kill: def $r3 killed $r3 killed $s3
+; CHECK-NEXT:    bctrl
+; CHECK-NEXT:    evmergehi 3, 30, 30
+; CHECK-NEXT:    mr 4, 30
+; CHECK-NEXT:    mtctr 3
+; CHECK-NEXT:    # kill: def $r3 killed $r3 killed $s3
+; CHECK-NEXT:    bctrl
+; CHECK-NEXT:    li 3, .LCPI58_0@l
+; CHECK-NEXT:    lis 4, .LCPI58_0@ha
+; CHECK-NEXT:    evlddx 3, 4, 3
+; CHECK-NEXT:    evldd 30, 16(1) # 8-byte Folded Reload
+; CHECK-NEXT:    efdmul 3, 29, 3
+; CHECK-NEXT:    evldd 29, 8(1) # 8-byte Folded Reload
+; CHECK-NEXT:    efscfd 3, 3
+; CHECK-NEXT:    stw 3, 0(3)
+; CHECK-NEXT:    lwz 30, 40(1) # 4-byte Folded Reload
+; CHECK-NEXT:    lwz 29, 36(1) # 4-byte Folded Reload
+; CHECK-NEXT:    lwz 0, 52(1)
+; CHECK-NEXT:    addi 1, 1, 48
+; CHECK-NEXT:    mtlr 0
+; CHECK-NEXT:    blr
+entry:
+  %0 = getelementptr %struct.a, %struct.a* %f, i32 0, i32 0
+  %1 = load float, float* undef
+  %conv = fpext float %1 to double
+  %2 = load float, float* %0
+  %g = fpext float %2 to double
+  %3 = call i32 undef(double %g)
+  %h = call i32 undef(double %conv)
+  %n = sitofp i32 %3 to double
+  %k = fmul double %g, %n
+  %l = fptrunc double %k to float
+  store float %l, float* undef
+  ret void
+}