[X86] matchAddressRecursively - add XOR(X, MIN_SIGNED_VALUE) handling

[RKSimon]

Summary

Allows us to fold XOR(X, MIN_SIGNED_VALUE) == ADD(X, MIN_SIGNED_VALUE) into LEA patterns

As mentioned on PR52267.

[spatel]

LGTM

llvm/lib/Target/X86/X86ISelDAGToDAG.cpp

case ISD::OR:
     // An 'lea' can then be used to match the shift (multiply) and add:
     // and $1, %esi
     // lea (%rsi, %rdi, 8), %rax
     if (CurDAG->haveNoCommonBitsSet(N.getOperand(0), N.getOperand(1)) &&
         !matchAdd(N, AM, Depth))
       return false;
     break;
 
+  case ISD::XOR:
+    // We want to look through a transform in InstCombine that
+    // turns 'add' with min_signed_val into 'xor', so we can treat this 'xor'
+    // exactly like an 'add'.
+    if (auto *NC1 = dyn_cast<ConstantSDNode>(N.getOperand(1)))
+      if (NC1->isMinSignedValue() && !matchAdd(N, AM, Depth))
+        return false;
+    break;
+
   case ISD::AND: {
     // Perform some heroic transforms on an and of a constant-count shift
     // with a constant to enable use of the scaled offset field.
 
     // Scale must not be used already.
     if (AM.IndexReg.getNode() != nullptr || AM.Scale != 1) break;
 
     // We only handle up to 64-bit values here as those are what matter for

llvm/lib/Target/X86/X86InstrInfo.td

 
 //===----------------------------------------------------------------------===//
 // X86 Complex Pattern Definitions.
 //
 
 // Define X86-specific addressing mode.
 def addr      : ComplexPattern<iPTR, 5, "selectAddr", [], [SDNPWantParent]>;
 def lea32addr : ComplexPattern<i32, 5, "selectLEAAddr",
-                               [add, sub, mul, X86mul_imm, shl, or, frameindex],
+                               [add, sub, mul, X86mul_imm, shl, or, xor, frameindex],
                                []>;
 // In 64-bit mode 32-bit LEAs can use RIP-relative addressing.
 def lea64_32addr : ComplexPattern<i32, 5, "selectLEA64_32Addr",
-                                  [add, sub, mul, X86mul_imm, shl, or,
+                                  [add, sub, mul, X86mul_imm, shl, or, xor,
                                    frameindex, X86WrapperRIP],
                                   []>;
 
 def tls32addr : ComplexPattern<i32, 5, "selectTLSADDRAddr",
                                [tglobaltlsaddr], []>;
 
 def tls32baseaddr : ComplexPattern<i32, 5, "selectTLSADDRAddr",
                                [tglobaltlsaddr], []>;
 
 def lea64addr : ComplexPattern<i64, 5, "selectLEAAddr",
-                        [add, sub, mul, X86mul_imm, shl, or, frameindex,
+                        [add, sub, mul, X86mul_imm, shl, or, xor, frameindex,
                          X86WrapperRIP], []>;
 
 def tls64addr : ComplexPattern<i64, 5, "selectTLSADDRAddr",
                                [tglobaltlsaddr], []>;
 
 def tls64baseaddr : ComplexPattern<i64, 5, "selectTLSADDRAddr",
                                [tglobaltlsaddr], []>;
 

llvm/test/CodeGen/X86/xor-lea.ll

 ; X86-NEXT:    movl $-2147483648, %eax # imm = 0x80000000
 ; X86-NEXT:    xorl {{[0-9]+}}(%esp), %eax
 ; X86-NEXT:    addl $-512, %eax # imm = 0xFE00
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: sub_xor_sminval_i32:
 ; X64:       # %bb.0:
 ; X64-NEXT:    # kill: def $edi killed $edi def $rdi
-; X64-NEXT:    xorl $-2147483648, %edi # imm = 0x80000000
-; X64-NEXT:    leal -512(%rdi), %eax
+; X64-NEXT:    movl $-512, %eax # imm = 0xFE00
+; X64-NEXT:    leal -2147483648(%rdi,%rax), %eax
 ; X64-NEXT:    retq
   %r = xor i32 %x, 2147483648
   %s = sub i32 %r, 512
   ret i32 %s
 }
 
 define i64 @add_xor_sminval_i64(i64 %x, i64 %y) {
 ; X86-LABEL: add_xor_sminval_i64:
; X64-NEXT:    retq
   %r = xor i16 %s, 32768
   ret i16 %r
 }
 
 define i32 @xor_shl_sminval_i32(i32 %x) {
 ; X86-LABEL: xor_shl_sminval_i32:
 ; X86:       # %bb.0:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    shll $3, %eax
-; X86-NEXT:    xorl $-2147483648, %eax # imm = 0x80000000
+; X86-NEXT:    leal -2147483648(,%eax,8), %eax
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: xor_shl_sminval_i32:
 ; X64:       # %bb.0:
 ; X64-NEXT:    # kill: def $edi killed $edi def $rdi
-; X64-NEXT:    leal (,%rdi,8), %eax
-; X64-NEXT:    xorl $-2147483648, %eax # imm = 0x80000000
+; X64-NEXT:    leal -2147483648(,%rdi,8), %eax
 ; X64-NEXT:    retq
   %s = shl i32 %x, 3
   %r = xor i32 %s, 2147483648
   ret i32 %r
 }
 
 ; negative test
 define i32 @xor_bigshl_sminval_i32(i32 %x) {

 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X86-NEXT:    shldl $2, %eax, %edx
 ; X86-NEXT:    shll $2, %eax
 ; X86-NEXT:    xorl $-2147483648, %edx # imm = 0x80000000
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: xor_shl_sminval_i64:
 ; X64:       # %bb.0:
-; X64-NEXT:    leaq (,%rdi,4), %rcx
 ; X64-NEXT:    movabsq $-9223372036854775808, %rax # imm = 0x8000000000000000
-; X64-NEXT:    xorq %rcx, %rax
+; X64-NEXT:    leaq (%rax,%rdi,4), %rax
 ; X64-NEXT:    retq
   %s = shl i64 %x, 2
   %r = xor i64 %s, -9223372036854775808
   ret i64 %r
 }