[X86][BMI] Fold cmpeq/ne(or(X,Y),X) --> cmpeq/ne(and(~X,Y),0) (PR44136)

I've initially just enabled this for BMI which has the ANDN instruction for i32/i64 - the i16/i8 cases give an idea of what'd we get when we enable it in all cases (I'll do this as a later commit).

Additionally, the i16/i8 cases could be freely promoted to i32 (as the args are already zeroext) and we could then make use of ANDN + the free cmp0 there as well - this has come up in PR48768 and PR49028 so I'm going to look at this soon.

https://alive2.llvm.org/ce/z/QVWHP_
https://alive2.llvm.org/ce/z/pLngT-

Vector cases do not appear to benefit from this as we end up with having to generate the zero vector as well - this is one of the reasons I didn't try to tie this into hasAndNot/hasAndNotCompare.

Differential Revision: https://reviews.llvm.org/D100177

llvm/lib/Target/X86/X86ISelLowering.cpp

@@ -48162,6 +48162,26 @@ static SDValue combineSetCC(SDNode *N, SelectionDAG &DAG,
         return DAG.getNode(ISD::TRUNCATE, DL, VT,
                            DAG.getNode(X86ISD::SETCC, DL, MVT::i8, X86CC, V));
     }
+
+    // cmpeq(or(X,Y),X) --> cmpeq(and(~X,Y),0)
+    // cmpne(or(X,Y),X) --> cmpne(and(~X,Y),0)
+    if (OpVT.isScalarInteger() && Subtarget.hasBMI()) {
+      auto MatchOrCmpEq = [&](SDValue N0, SDValue N1) {
+        if (N0.getOpcode() == ISD::OR && N0->hasOneUse()) {
+          if (N0.getOperand(0) == N1)
+            return DAG.getNode(ISD::AND, DL, OpVT, DAG.getNOT(DL, N1, OpVT),
+                               N0.getOperand(1));
+          if (N0.getOperand(1) == N1)
+            return DAG.getNode(ISD::AND, DL, OpVT, DAG.getNOT(DL, N1, OpVT),
+                               N0.getOperand(0));
+        }
+        return SDValue();
+      };
+      if (SDValue AndN = MatchOrCmpEq(LHS, RHS))
+        return DAG.getSetCC(DL, VT, AndN, DAG.getConstant(0, DL, OpVT), CC);
+      if (SDValue AndN = MatchOrCmpEq(RHS, LHS))
+        return DAG.getSetCC(DL, VT, AndN, DAG.getConstant(0, DL, OpVT), CC);
+    }
   }
 
   if (VT.isVector() && VT.getVectorElementType() == MVT::i1 &&

llvm/test/CodeGen/X86/setcc-logic.ll

@@ -1,6 +1,6 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown | FileCheck %s
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown | FileCheck %s --check-prefixes=CHECK,NOBMI
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+bmi | FileCheck %s
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+bmi | FileCheck %s --check-prefixes=CHECK,BMI
 
 define zeroext i1 @all_bits_clear(i32 %P, i32 %Q) nounwind {
 ; CHECK-LABEL: all_bits_clear:
@@ -607,51 +607,77 @@ define i1 @and_icmps_const_1bit_diff_common_op(i32 %x, i32 %y) {
   ret i1 %r
 }
 
-; TODO: PR44136 - fold cmpeq(or(X,Y),X) --> cmpeq(and(~X,Y),0)
+; PR44136 - fold cmpeq(or(X,Y),X) --> cmpeq(and(~X,Y),0)
 
 define i1 @or_cmp_eq_i64(i64 %x, i64 %y) {
-; CHECK-LABEL: or_cmp_eq_i64:
+; NOBMI-LABEL: or_cmp_eq_i64:
-; CHECK:       # %bb.0:
+; NOBMI:       # %bb.0:
-; CHECK-NEXT:    orq %rdi, %rsi
+; NOBMI-NEXT:    orq %rdi, %rsi
-; CHECK-NEXT:    cmpq %rdi, %rsi
+; NOBMI-NEXT:    cmpq %rdi, %rsi
-; CHECK-NEXT:    sete %al
+; NOBMI-NEXT:    sete %al
-; CHECK-NEXT:    retq
+; NOBMI-NEXT:    retq
+;
+; BMI-LABEL: or_cmp_eq_i64:
+; BMI:       # %bb.0:
+; BMI-NEXT:    andnq %rsi, %rdi, %rax
+; BMI-NEXT:    sete %al
+; BMI-NEXT:    retq
   %o = or i64 %x, %y
   %c = icmp eq i64 %o, %x
   ret i1 %c
 }
 
 define i1 @or_cmp_ne_i32(i32 %x, i32 %y) {
-; CHECK-LABEL: or_cmp_ne_i32:
+; NOBMI-LABEL: or_cmp_ne_i32:
-; CHECK:       # %bb.0:
+; NOBMI:       # %bb.0:
-; CHECK-NEXT:    orl %esi, %edi
+; NOBMI-NEXT:    orl %esi, %edi
-; CHECK-NEXT:    cmpl %esi, %edi
+; NOBMI-NEXT:    cmpl %esi, %edi
-; CHECK-NEXT:    setne %al
+; NOBMI-NEXT:    setne %al
-; CHECK-NEXT:    retq
+; NOBMI-NEXT:    retq
+;
+; BMI-LABEL: or_cmp_ne_i32:
+; BMI:       # %bb.0:
+; BMI-NEXT:    andnl %edi, %esi, %eax
+; BMI-NEXT:    setne %al
+; BMI-NEXT:    retq
   %o = or i32 %x, %y
   %c = icmp ne i32 %o, %y
   ret i1 %c
 }
 
 define i1 @or_cmp_eq_i16(i16 zeroext %x, i16 zeroext %y) {
-; CHECK-LABEL: or_cmp_eq_i16:
+; NOBMI-LABEL: or_cmp_eq_i16:
-; CHECK:       # %bb.0:
+; NOBMI:       # %bb.0:
-; CHECK-NEXT:    orl %edi, %esi
+; NOBMI-NEXT:    orl %edi, %esi
-; CHECK-NEXT:    cmpw %si, %di
+; NOBMI-NEXT:    cmpw %si, %di
-; CHECK-NEXT:    sete %al
+; NOBMI-NEXT:    sete %al
-; CHECK-NEXT:    retq
+; NOBMI-NEXT:    retq
+;
+; BMI-LABEL: or_cmp_eq_i16:
+; BMI:       # %bb.0:
+; BMI-NEXT:    notl %edi
+; BMI-NEXT:    testw %si, %di
+; BMI-NEXT:    sete %al
+; BMI-NEXT:    retq
   %o = or i16 %x, %y
   %c = icmp eq i16 %x, %o
   ret i1 %c
 }
 
 define i1 @or_cmp_ne_i8(i8 zeroext %x, i8 zeroext %y) {
-; CHECK-LABEL: or_cmp_ne_i8:
+; NOBMI-LABEL: or_cmp_ne_i8:
-; CHECK:       # %bb.0:
+; NOBMI:       # %bb.0:
-; CHECK-NEXT:    orl %esi, %edi
+; NOBMI-NEXT:    orl %esi, %edi
-; CHECK-NEXT:    cmpb %dil, %sil
+; NOBMI-NEXT:    cmpb %dil, %sil
-; CHECK-NEXT:    setne %al
+; NOBMI-NEXT:    setne %al
-; CHECK-NEXT:    retq
+; NOBMI-NEXT:    retq
+;
+; BMI-LABEL: or_cmp_ne_i8:
+; BMI:       # %bb.0:
+; BMI-NEXT:    notb %sil
+; BMI-NEXT:    testb %dil, %sil
+; BMI-NEXT:    setne %al
+; BMI-NEXT:    retq
   %o = or i8 %x, %y
   %c = icmp ne i8 %y, %o
   ret i1 %c