DAG: Combine AssertZext with and AssertSext

arsenm · arsenm · commit 4ed546462dac · 2025-06-06T16:18:27.000+09:00
#142898 (comment)
diff --git a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -14933,6 +14933,25 @@ SDValue DAGCombiner::visitAssertExt(SDNode *N) {
     }
   }
 
+  // If we have (AssertZext (and (AssertSext X, iX), M), iY) and Y is smaller
+  // than X, and the And doesn't change the lower iY bits, we can move the
+  // AssertZext in front of the And and drop the AssertSext.
+  if (Opcode == ISD::AssertZext && N0.getOpcode() == ISD::AND &&
+      N0.hasOneUse() && N0.getOperand(0).getOpcode() == ISD::AssertSext &&
+      isa<ConstantSDNode>(N0.getOperand(1))) {
+    SDValue BigA = N0.getOperand(0);
+    EVT BigA_AssertVT = cast<VTSDNode>(BigA.getOperand(1))->getVT();
+    const APInt &Mask = N0.getConstantOperandAPInt(1);
+    if (AssertVT.bitsLT(BigA_AssertVT) &&
+        Mask.countr_one() >= AssertVT.getScalarSizeInBits()) {
+      SDLoc DL(N);
+      SDValue NewAssert =
+          DAG.getNode(Opcode, DL, N->getValueType(0), BigA.getOperand(0), N1);
+      return DAG.getNode(ISD::AND, DL, N->getValueType(0), NewAssert,
+                         N0.getOperand(1));
+    }
+  }
+
   return SDValue();
 }
 
diff --git a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
@@ -434,11 +434,7 @@ void TargetLowering::softenSetCCOperands(SelectionDAG &DAG, EVT VT,
 
     EVT SetCCVT =
         getSetCCResultType(DAG.getDataLayout(), *DAG.getContext(), RetVT);
-    if (RetVT == SetCCVT &&
-        getBooleanContents(RetVT) == ZeroOrOneBooleanContent) {
-      // FIXME: Checking the type matches is a hack in case the calling
-      // convention lowering inserted some instructions after the
-      // CopyFromReg. Combines fail to look through the AssertZext.
+    if (getBooleanContents(RetVT) == ZeroOrOneBooleanContent) {
       NewLHS = DAG.getNode(ISD::AssertZext, dl, RetVT, Call.first,
                            DAG.getValueType(MVT::i1));
     }
diff --git a/llvm/test/CodeGen/RISCV/double-fcmp-strict.ll b/llvm/test/CodeGen/RISCV/double-fcmp-strict.ll
@@ -498,13 +498,13 @@ define i32 @fcmp_ueq(double %a, double %b) nounwind strictfp {
 ; RV64I-NEXT:    sd s2, 0(sp) # 8-byte Folded Spill
 ; RV64I-NEXT:    mv s0, a1
 ; RV64I-NEXT:    mv s1, a0
-; RV64I-NEXT:    call __eqdf2
-; RV64I-NEXT:    seqz s2, a0
+; RV64I-NEXT:    call __unorddf2
+; RV64I-NEXT:    mv s2, a0
 ; RV64I-NEXT:    mv a0, s1
 ; RV64I-NEXT:    mv a1, s0
-; RV64I-NEXT:    call __unorddf2
-; RV64I-NEXT:    snez a0, a0
-; RV64I-NEXT:    or a0, a0, s2
+; RV64I-NEXT:    call __eqdf2
+; RV64I-NEXT:    seqz a0, a0
+; RV64I-NEXT:    or a0, s2, a0
 ; RV64I-NEXT:    ld ra, 24(sp) # 8-byte Folded Reload
 ; RV64I-NEXT:    ld s0, 16(sp) # 8-byte Folded Reload
 ; RV64I-NEXT:    ld s1, 8(sp) # 8-byte Folded Reload
@@ -1226,13 +1226,13 @@ define i32 @fcmps_ueq(double %a, double %b) nounwind strictfp {
 ; RV64I-NEXT:    sd s2, 0(sp) # 8-byte Folded Spill
 ; RV64I-NEXT:    mv s0, a1
 ; RV64I-NEXT:    mv s1, a0
-; RV64I-NEXT:    call __eqdf2
-; RV64I-NEXT:    seqz s2, a0
+; RV64I-NEXT:    call __unorddf2
+; RV64I-NEXT:    mv s2, a0
 ; RV64I-NEXT:    mv a0, s1
 ; RV64I-NEXT:    mv a1, s0
-; RV64I-NEXT:    call __unorddf2
-; RV64I-NEXT:    snez a0, a0
-; RV64I-NEXT:    or a0, a0, s2
+; RV64I-NEXT:    call __eqdf2
+; RV64I-NEXT:    seqz a0, a0
+; RV64I-NEXT:    or a0, s2, a0
 ; RV64I-NEXT:    ld ra, 24(sp) # 8-byte Folded Reload
 ; RV64I-NEXT:    ld s0, 16(sp) # 8-byte Folded Reload
 ; RV64I-NEXT:    ld s1, 8(sp) # 8-byte Folded Reload
diff --git a/llvm/test/CodeGen/RISCV/double-fcmp.ll b/llvm/test/CodeGen/RISCV/double-fcmp.ll
@@ -430,13 +430,13 @@ define i32 @fcmp_ueq(double %a, double %b) nounwind {
 ; RV64I-NEXT:    sd s2, 0(sp) # 8-byte Folded Spill
 ; RV64I-NEXT:    mv s0, a1
 ; RV64I-NEXT:    mv s1, a0
-; RV64I-NEXT:    call __eqdf2
-; RV64I-NEXT:    seqz s2, a0
+; RV64I-NEXT:    call __unorddf2
+; RV64I-NEXT:    mv s2, a0
 ; RV64I-NEXT:    mv a0, s1
 ; RV64I-NEXT:    mv a1, s0
-; RV64I-NEXT:    call __unorddf2
-; RV64I-NEXT:    snez a0, a0
-; RV64I-NEXT:    or a0, a0, s2
+; RV64I-NEXT:    call __eqdf2
+; RV64I-NEXT:    seqz a0, a0
+; RV64I-NEXT:    or a0, s2, a0
 ; RV64I-NEXT:    ld ra, 24(sp) # 8-byte Folded Reload
 ; RV64I-NEXT:    ld s0, 16(sp) # 8-byte Folded Reload
 ; RV64I-NEXT:    ld s1, 8(sp) # 8-byte Folded Reload
diff --git a/llvm/test/CodeGen/RISCV/float-fcmp-strict.ll b/llvm/test/CodeGen/RISCV/float-fcmp-strict.ll
@@ -405,13 +405,13 @@ define i32 @fcmp_ueq(float %a, float %b) nounwind strictfp {
 ; RV64I-NEXT:    sd s2, 0(sp) # 8-byte Folded Spill
 ; RV64I-NEXT:    mv s0, a1
 ; RV64I-NEXT:    mv s1, a0
-; RV64I-NEXT:    call __eqsf2
-; RV64I-NEXT:    seqz s2, a0
+; RV64I-NEXT:    call __unordsf2
+; RV64I-NEXT:    mv s2, a0
 ; RV64I-NEXT:    mv a0, s1
 ; RV64I-NEXT:    mv a1, s0
-; RV64I-NEXT:    call __unordsf2
-; RV64I-NEXT:    snez a0, a0
-; RV64I-NEXT:    or a0, a0, s2
+; RV64I-NEXT:    call __eqsf2
+; RV64I-NEXT:    seqz a0, a0
+; RV64I-NEXT:    or a0, s2, a0
 ; RV64I-NEXT:    ld ra, 24(sp) # 8-byte Folded Reload
 ; RV64I-NEXT:    ld s0, 16(sp) # 8-byte Folded Reload
 ; RV64I-NEXT:    ld s1, 8(sp) # 8-byte Folded Reload
@@ -1014,13 +1014,13 @@ define i32 @fcmps_ueq(float %a, float %b) nounwind strictfp {
 ; RV64I-NEXT:    sd s2, 0(sp) # 8-byte Folded Spill
 ; RV64I-NEXT:    mv s0, a1
 ; RV64I-NEXT:    mv s1, a0
-; RV64I-NEXT:    call __eqsf2
-; RV64I-NEXT:    seqz s2, a0
+; RV64I-NEXT:    call __unordsf2
+; RV64I-NEXT:    mv s2, a0
 ; RV64I-NEXT:    mv a0, s1
 ; RV64I-NEXT:    mv a1, s0
-; RV64I-NEXT:    call __unordsf2
-; RV64I-NEXT:    snez a0, a0
-; RV64I-NEXT:    or a0, a0, s2
+; RV64I-NEXT:    call __eqsf2
+; RV64I-NEXT:    seqz a0, a0
+; RV64I-NEXT:    or a0, s2, a0
 ; RV64I-NEXT:    ld ra, 24(sp) # 8-byte Folded Reload
 ; RV64I-NEXT:    ld s0, 16(sp) # 8-byte Folded Reload
 ; RV64I-NEXT:    ld s1, 8(sp) # 8-byte Folded Reload
diff --git a/llvm/test/CodeGen/RISCV/float-fcmp.ll b/llvm/test/CodeGen/RISCV/float-fcmp.ll
@@ -367,13 +367,13 @@ define i32 @fcmp_ueq(float %a, float %b) nounwind {
 ; RV64I-NEXT:    sd s2, 0(sp) # 8-byte Folded Spill
 ; RV64I-NEXT:    mv s0, a1
 ; RV64I-NEXT:    mv s1, a0
-; RV64I-NEXT:    call __eqsf2
-; RV64I-NEXT:    seqz s2, a0
+; RV64I-NEXT:    call __unordsf2
+; RV64I-NEXT:    mv s2, a0
 ; RV64I-NEXT:    mv a0, s1
 ; RV64I-NEXT:    mv a1, s0
-; RV64I-NEXT:    call __unordsf2
-; RV64I-NEXT:    snez a0, a0
-; RV64I-NEXT:    or a0, a0, s2
+; RV64I-NEXT:    call __eqsf2
+; RV64I-NEXT:    seqz a0, a0
+; RV64I-NEXT:    or a0, s2, a0
 ; RV64I-NEXT:    ld ra, 24(sp) # 8-byte Folded Reload
 ; RV64I-NEXT:    ld s0, 16(sp) # 8-byte Folded Reload
 ; RV64I-NEXT:    ld s1, 8(sp) # 8-byte Folded Reload
diff --git a/llvm/test/CodeGen/X86/fp128-libcalls-strict.ll b/llvm/test/CodeGen/X86/fp128-libcalls-strict.ll
@@ -3443,8 +3443,6 @@ define i64 @cmp_ueq_q(i64 %a, i64 %b, fp128 %x, fp128 %y) #0 {
 ; X86-NEXT:    pushl {{[0-9]+}}(%esp)
 ; X86-NEXT:    calll __unordtf2
 ; X86-NEXT:    addl $32, %esp
-; X86-NEXT:    testl %eax, %eax
-; X86-NEXT:    setne %al
 ; X86-NEXT:    orb %bl, %al
 ; X86-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X86-NEXT:    leal {{[0-9]+}}(%esp), %ecx
@@ -3526,8 +3524,6 @@ define i64 @cmp_ueq_q(i64 %a, i64 %b, fp128 %x, fp128 %y) #0 {
 ; WIN-X86-NEXT:    pushl {{[0-9]+}}(%esp)
 ; WIN-X86-NEXT:    calll ___unordtf2
 ; WIN-X86-NEXT:    addl $32, %esp
-; WIN-X86-NEXT:    testl %eax, %eax
-; WIN-X86-NEXT:    setne %al
 ; WIN-X86-NEXT:    orb %bl, %al
 ; WIN-X86-NEXT:    jne LBB39_1
 ; WIN-X86-NEXT:  # %bb.2:
diff --git a/llvm/test/CodeGen/X86/fpcmp-soft-fp.ll b/llvm/test/CodeGen/X86/fpcmp-soft-fp.ll
@@ -99,8 +99,6 @@ entry:
 ; CHECK: calll __eqdf2
 ; CHECK: sete
 ; CHECK: calll __unorddf2
-; CHECK: setne
-; CHECK: or
 ; CHECK: retl
 
 define i1 @test11(double %d) #0 {
