[X86] Add X86ISD::CMOV to computeKnownBitsForTargetNode and ComputeNumSignBitsForTargetNode.

[craig.topper]

Summary

Implementations based on ISD::SELECT.

[RKSimon]

LGTM - one minor

lib/Target/X86/X86ISelLowering.cpp

27195 ↗

(On Diff #118128)

if (Known.isUnknown())

llvm/trunk/lib/Target/X86/X86ISelLowering.cpp

case X86ISD::VZEXT: {
 
     Known = KnownBits(InBitWidth);
     APInt DemandedSrcElts = APInt::getLowBitsSet(InNumElts, NumElts);
     DAG.computeKnownBits(N0, Known, DemandedSrcElts, Depth + 1);
     Known = Known.zext(BitWidth);
     Known.Zero.setBitsFrom(InBitWidth);
     break;
   }
+  case X86ISD::CMOV: {
+    DAG.computeKnownBits(Op.getOperand(1), Known, Depth+1);
+    // If we don't know any bits, early out.
+    if (Known.isUnknown())
+      break;
+    KnownBits Known2;
+    DAG.computeKnownBits(Op.getOperand(0), Known2, Depth+1);
+
+    // Only known if known in both the LHS and RHS.
+    Known.One &= Known2.One;
+    Known.Zero &= Known2.Zero;
+    break;
+  }
   }
 }
 
 unsigned X86TargetLowering::ComputeNumSignBitsForTargetNode(
     SDValue Op, const APInt &DemandedElts, const SelectionDAG &DAG,
     unsigned Depth) const {
   unsigned VTBits = Op.getScalarValueSizeInBits();
   unsigned Opcode = Op.getOpcode();
unsigned X86TargetLowering::ComputeNumSignBitsForTargetNode(
 
   case X86ISD::PCMPGT:
   case X86ISD::PCMPEQ:
   case X86ISD::CMPP:
   case X86ISD::VPCOM:
   case X86ISD::VPCOMU:
     // Vector compares return zero/all-bits result values.
     return VTBits;
+
+  case X86ISD::CMOV: {
+    unsigned Tmp0 = DAG.ComputeNumSignBits(Op.getOperand(0), Depth+1);
+    if (Tmp0 == 1) return 1;  // Early out.
+    unsigned Tmp1 = DAG.ComputeNumSignBits(Op.getOperand(1), Depth+1);
+    return std::min(Tmp0, Tmp1);
+  }
   }
 
   // Fallback case.
   return 1;
 }
 
 SDValue X86TargetLowering::unwrapAddress(SDValue N) const {
   if (N->getOpcode() == X86ISD::Wrapper || N->getOpcode() == X86ISD::WrapperRIP)

llvm/trunk/test/CodeGen/X86/cmovcmov.ll

 ; NOCMOV-NEXT:   fnstsw  %ax
 ; NOCMOV-NEXT:   sahf
 ; NOCMOV-NEXT:   leal  20(%esp), %ecx
 ; NOCMOV-NEXT:   jne  [[TBB:.LBB[0-9_]+]]
 ; NOCMOV-NEXT:   jp  [[TBB]]
 ; NOCMOV-NEXT:   leal  12(%esp), %ecx
 ; NOCMOV-NEXT: [[TBB]]:
 ; NOCMOV-NEXT:   movl  (%ecx), %eax
-; NOCMOV-NEXT:   orl  $4, %ecx
-; NOCMOV-NEXT:   movl  (%ecx), %edx
+; NOCMOV-NEXT:   movl  4(%ecx), %edx
 ; NOCMOV-NEXT:   retl
 define i64 @test_select_fcmp_oeq_i64(float %a, float %b, i64 %c, i64 %d) #0 {
 entry:
   %cmp = fcmp oeq float %a, %b
   %r = select i1 %cmp, i64 %c, i64 %d
   ret i64 %r
 }
 

 ; NOCMOV-NEXT:   fnstsw  %ax
 ; NOCMOV-NEXT:   sahf
 ; NOCMOV-NEXT:   leal  12(%esp), %ecx
 ; NOCMOV-NEXT:   jne  [[TBB:.LBB[0-9_]+]]
 ; NOCMOV-NEXT:   jp  [[TBB]]
 ; NOCMOV-NEXT:   leal  20(%esp), %ecx
 ; NOCMOV-NEXT: [[TBB]]:
 ; NOCMOV-NEXT:   movl  (%ecx), %eax
-; NOCMOV-NEXT:   orl  $4, %ecx
-; NOCMOV-NEXT:   movl  (%ecx), %edx
+; NOCMOV-NEXT:   movl  4(%ecx), %edx
 ; NOCMOV-NEXT:   retl
 define i64 @test_select_fcmp_une_i64(float %a, float %b, i64 %c, i64 %d) #0 {
 entry:
   %cmp = fcmp une float %a, %b
   %r = select i1 %cmp, i64 %c, i64 %d
   ret i64 %r
 }
 

llvm/trunk/test/CodeGen/X86/pr32282.ll

 ; X86-NEXT:    andnl %edx, %eax, %edx
 ; X86-NEXT:    shrdl $21, %ecx, %edx
 ; X86-NEXT:    shrl $21, %ecx
 ; X86-NEXT:    xorl %eax, %eax
 ; X86-NEXT:    testb %al, %al
 ; X86-NEXT:    cmovnel %ecx, %edx
 ; X86-NEXT:    cmovnel %eax, %ecx
 ; X86-NEXT:    andl $-2, %edx
-; X86-NEXT:    andl $2147483647, %ecx # imm = 0x7FFFFFFF
 ; X86-NEXT:    addl $7, %edx
 ; X86-NEXT:    adcxl %eax, %ecx
 ; X86-NEXT:    pushl %ecx
 ; X86-NEXT:  .Lcfi1:
 ; X86-NEXT:    .cfi_adjust_cfa_offset 4
 ; X86-NEXT:    pushl %edx
 ; X86-NEXT:  .Lcfi2:
 ; X86-NEXT:    .cfi_adjust_cfa_offset 4