[X86] Promote cttz_i32(x) -> cttz_i64((i64)x | (1 << 32)) (#102900)

On 64bit targets we can promote i32 CTTZ nodes to i64 CTTZ_ZERO_UNDEF by setting the 32nd bit.

#57811 also queried about whether we should use BTS instead of MOVABS+OR to avoid a i64 immediate - I'm willing to tweak the DAGToDAG isel peephole for these cases if reviewers think it worthwhile. But most recent CPUs can actually handle MOVABS faster than BTS/C/R.......

Fixes #57811

Author: RKSimon

llvm/lib/Target/X86/X86ISelLowering.cpp

@@ -412,6 +412,7 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM,
     setOperationAction(ISD::CTTZ           , MVT::i32  , Custom);
     setOperationAction(ISD::CTTZ_ZERO_UNDEF, MVT::i32  , Legal);
     if (Subtarget.is64Bit()) {
+      setOperationPromotedToType(ISD::CTTZ , MVT::i32, MVT::i64);
       setOperationAction(ISD::CTTZ         , MVT::i64  , Custom);
       setOperationAction(ISD::CTTZ_ZERO_UNDEF, MVT::i64, Legal);
     }
@@ -3237,9 +3238,10 @@ bool X86TargetLowering::shouldFormOverflowOp(unsigned Opcode, EVT VT,
 }
 
 bool X86TargetLowering::isCheapToSpeculateCttz(Type *Ty) const {
-  // Speculate cttz only if we can directly use TZCNT or can promote to i32.
+  // Speculate cttz only if we can directly use TZCNT or can promote to i32/i64.
   return Subtarget.hasBMI() ||
-         (!Ty->isVectorTy() && Ty->getScalarSizeInBits() < 32);
+         (!Ty->isVectorTy() &&
+          Ty->getScalarSizeInBits() < (Subtarget.is64Bit() ? 64u : 32u));
 }
 
 bool X86TargetLowering::isCheapToSpeculateCtlz(Type *Ty) const {

llvm/test/CodeGen/X86/cttz.ll

@@ -317,13 +317,11 @@ define i32 @cttz_i32_zero_test(i32 %n) {
 ;
 ; X64-LABEL: cttz_i32_zero_test:
 ; X64:       # %bb.0:
-; X64-NEXT:    testl %edi, %edi
-; X64-NEXT:    je .LBB6_1
-; X64-NEXT:  # %bb.2: # %cond.false
-; X64-NEXT:    rep bsfl %edi, %eax
-; X64-NEXT:    retq
-; X64-NEXT:  .LBB6_1:
-; X64-NEXT:    movl $32, %eax
+; X64-NEXT:    # kill: def $edi killed $edi def $rdi
+; X64-NEXT:    movabsq $4294967296, %rax # imm = 0x100000000
+; X64-NEXT:    orq %rdi, %rax
+; X64-NEXT:    rep bsfq %rax, %rax
+; X64-NEXT:    # kill: def $eax killed $eax killed $rax
 ; X64-NEXT:    retq
 ;
 ; X86-CLZ-LABEL: cttz_i32_zero_test: