[PowerPC] Optimize compares fed by ANDISo

[nemanjai]

Summary

Both ANDIo and ANDISo (and the 64-bit versions) are record-form instructions. When optimizing compares, we handle the former in order to eliminate the compare instruction but not the latter. This patch just adds the latter to the set of instructions we optimize.
The reason these instructions need to be handled separately is that they are not part of the RecFormRel map (since they don't have a non-record-form). The missing "and-immediate-shifted" is just an oversight in the initial implementation.

[nemanjai]

@inouehrs Can you please review/comment/approve this as I've separated this out per your request.

[inouehrs]

LGTM with nitpicking.

I feel that the current test case is somewhat fragile for minor change in inst. scheduling or register allocation.
Checking there is no cmpdi between andis. and isel may be enough.

[nemanjai]

In D51353#1220682, @inouehrs wrote:

LGTM with nitpicking.

I feel that the current test case is somewhat fragile for minor change in inst. scheduling or register allocation.
Checking there is no cmpdi between andis. and isel may be enough.

Yeah, I usually don't like to modify the checks produced by the automated tool. However, I think your suggestion is reasonable - having the andis. and no cmp before its use should sufficiently test this patch.

lib/Target/PowerPC/PPCInstrInfo.cpp

bool PPCInstrInfo::optimizeCompareInstr(MachineInstr &CmpInstr, unsigned SrcReg,
   if (!MI && !Sub)
     return false;
 
   // The single candidate is called MI.
   if (!MI) MI = Sub;
 
   int NewOpC = -1;
   int MIOpC = MI->getOpcode();
-  if (MIOpC == PPC::ANDIo || MIOpC == PPC::ANDIo8)
+  if (MIOpC == PPC::ANDIo || MIOpC == PPC::ANDIo8 ||
+      MIOpC == PPC::ANDISo || MIOpC == PPC::ANDISo8)
     NewOpC = MIOpC;
   else {
     NewOpC = PPC::getRecordFormOpcode(MIOpC);
     if (NewOpC == -1 && PPC::getNonRecordFormOpcode(MIOpC) != -1)
       NewOpC = MIOpC;
   }
 
   // FIXME: On the non-embedded POWER architectures, only some of the record

test/CodeGen/PowerPC/optimize-andiso.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -verify-machineinstrs -mtriple=powerpc64-unknown-linux-gnu \
+; RUN:   -mcpu=pwr8 -ppc-asm-full-reg-names < %s | FileCheck %s
+; RUN: llc -verify-machineinstrs -mtriple=powerpc64le-unknown-linux-gnu \
+; RUN:   -mcpu=pwr8 -ppc-asm-full-reg-names < %s | FileCheck %s
+define float @floatundisf(i64 %a) {
+; CHECK-LABEL: floatundisf:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    bc 4, 4*cr5+lt, .LBB0_2
+; CHECK-NEXT:  # %bb.1:
+; CHECK-NEXT:    xxlxor f1, f1, f1
+; CHECK-NEXT:    blr
+; CHECK-NEXT:  .LBB0_2: # %sw.epilog
+; CHECK-NEXT:    addi r3, r3, 1
+; CHECK-NEXT:    li r5, 2
+; CHECK-NEXT:    andis. r4, r3, 1024
+; CHECK-NEXT:    li r4, 3
+; CHECK-NEXT:    isel r4, r5, r4, eq
+; CHECK-NEXT:    srd r3, r3, r4
+; CHECK-NEXT:    rlwinm r3, r3, 0, 9, 31
+; CHECK-NEXT:    mtvsrd f0, r3
+; CHECK-NEXT:    xxsldwi vs0, vs0, vs0, 1
+; CHECK-NEXT:    xscvspdpn f1, vs0
+; CHECK-NEXT:    blr
+entry:
+  br i1 undef, label %return, label %sw.epilog
+
+sw.epilog:                                        ; preds = %entry
+  %or14 = or i64 0, %a
+  %inc = add i64 %or14, 1
+  %and16 = and i64 %inc, 67108864
+  %tobool = icmp eq i64 %and16, 0
+  %tmp.select.v = select i1 %tobool, i64 2, i64 3
+  %tmp.select = lshr i64 %inc, %tmp.select.v
+  %conv26 = trunc i64 %tmp.select to i32
+  %and27 = and i32 %conv26, 8388607
+  %or28 = or i32 0, %and27
+  %0 = bitcast i32 %or28 to float
+  br label %return
+
+return:                                           ; preds = %sw.epilog, %entry
+  %retval.0 = phi float [ %0, %sw.epilog ], [ 0.000000e+00, %entry ]
+  ret float %retval.0
+}