[DAG] visitSELECT - add "select usubo(x, y).overflow, (sub y, x), (usubo x, y) -> abdu(x, y)" fold (and neg equivalent)

RKSimon · RKSimon · commit b25b9a7d6c87 · 2024-09-04T11:59:10.000+01:00
Handle cases where CGP has merged the CMP+SUB into a USUBO node - improves a few outstanding niggles from #100810
diff --git a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -11719,6 +11719,24 @@ SDValue DAGCombiner::visitSELECT(SDNode *N) {
                              N2_2, Flags);
       }
     }
+
+    // select usubo(x, y).overflow, (sub y, x), (usubo x, y) -> abdu(x, y)
+    if (N0.getOpcode() == ISD::USUBO && N0.getResNo() == 1 &&
+        N2.getNode() == N0.getNode() && N2.getResNo() == 0 &&
+        N1.getOpcode() == ISD::SUB && N2.getOperand(0) == N1.getOperand(1) &&
+        N2.getOperand(1) == N1.getOperand(0) &&
+        (!LegalOperations || TLI.isOperationLegal(ISD::ABDU, VT)))
+      return DAG.getNode(ISD::ABDU, DL, VT, N0.getOperand(0), N0.getOperand(1));
+
+    // select usubo(x, y).overflow, (usubo x, y), (sub y, x) -> neg (abdu x, y)
+    if (N0.getOpcode() == ISD::USUBO && N0.getResNo() == 1 &&
+        N1.getNode() == N0.getNode() && N1.getResNo() == 0 &&
+        N2.getOpcode() == ISD::SUB && N2.getOperand(0) == N1.getOperand(1) &&
+        N2.getOperand(1) == N1.getOperand(0) &&
+        (!LegalOperations || TLI.isOperationLegal(ISD::ABDU, VT)))
+      return DAG.getNegative(
+          DAG.getNode(ISD::ABDU, DL, VT, N0.getOperand(0), N0.getOperand(1)),
+          DL, VT);
   }
 
   // Fold selects based on a setcc into other things, such as min/max/abs.
@@ -11776,6 +11794,9 @@ SDValue DAGCombiner::visitSELECT(SDNode *N) {
       return SelectNode;
     }
 
+    if (SDValue ABD = foldSelectToABD(Cond0, Cond1, N1, N2, CC, DL))
+      return ABD;
+
     if (SDValue NewSel = SimplifySelect(DL, N0, N1, N2))
       return NewSel;
   }
diff --git a/llvm/test/CodeGen/X86/abdu-neg.ll b/llvm/test/CodeGen/X86/abdu-neg.ll
@@ -751,27 +751,23 @@ define i8 @abd_cmp_i8(i8 %a, i8 %b) nounwind {
 define i16 @abd_cmp_i16(i16 %a, i16 %b) nounwind {
 ; X86-LABEL: abd_cmp_i16:
 ; X86:       # %bb.0:
-; X86-NEXT:    pushl %esi
+; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
 ; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %edx
-; X86-NEXT:    movl %ecx, %esi
-; X86-NEXT:    subw %dx, %si
-; X86-NEXT:    movl %esi, %eax
+; X86-NEXT:    subl %eax, %ecx
+; X86-NEXT:    movl %ecx, %eax
 ; X86-NEXT:    negl %eax
-; X86-NEXT:    cmpw %dx, %cx
-; X86-NEXT:    cmovbl %esi, %eax
+; X86-NEXT:    cmovnsl %ecx, %eax
 ; X86-NEXT:    # kill: def $ax killed $ax killed $eax
-; X86-NEXT:    popl %esi
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: abd_cmp_i16:
 ; X64:       # %bb.0:
-; X64-NEXT:    movl %edi, %ecx
-; X64-NEXT:    subw %si, %cx
+; X64-NEXT:    movzwl %si, %eax
+; X64-NEXT:    movzwl %di, %ecx
+; X64-NEXT:    subl %eax, %ecx
 ; X64-NEXT:    movl %ecx, %eax
 ; X64-NEXT:    negl %eax
-; X64-NEXT:    cmpw %si, %di
-; X64-NEXT:    cmovbl %ecx, %eax
+; X64-NEXT:    cmovnsl %ecx, %eax
 ; X64-NEXT:    # kill: def $ax killed $ax killed $eax
 ; X64-NEXT:    retq
   %cmp = icmp ult i16 %a, %b
@@ -811,33 +807,30 @@ define i32 @abd_cmp_i32(i32 %a, i32 %b) nounwind {
 define i64 @abd_cmp_i64(i64 %a, i64 %b) nounwind {
 ; X86-LABEL: abd_cmp_i64:
 ; X86:       # %bb.0:
-; X86-NEXT:    pushl %ebx
-; X86-NEXT:    pushl %edi
 ; X86-NEXT:    pushl %esi
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-NEXT:    movl %ecx, %edi
-; X86-NEXT:    subl %eax, %edi
-; X86-NEXT:    movl %esi, %ebx
-; X86-NEXT:    sbbl %edx, %ebx
-; X86-NEXT:    subl %ecx, %eax
-; X86-NEXT:    sbbl %esi, %edx
-; X86-NEXT:    cmovael %edi, %eax
-; X86-NEXT:    cmovael %ebx, %edx
+; X86-NEXT:    xorl %edx, %edx
+; X86-NEXT:    subl {{[0-9]+}}(%esp), %eax
+; X86-NEXT:    sbbl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    movl $0, %esi
+; X86-NEXT:    sbbl %esi, %esi
+; X86-NEXT:    xorl %esi, %ecx
+; X86-NEXT:    xorl %esi, %eax
+; X86-NEXT:    subl %esi, %eax
+; X86-NEXT:    sbbl %esi, %ecx
+; X86-NEXT:    negl %eax
+; X86-NEXT:    sbbl %ecx, %edx
 ; X86-NEXT:    popl %esi
-; X86-NEXT:    popl %edi
-; X86-NEXT:    popl %ebx
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: abd_cmp_i64:
 ; X64:       # %bb.0:
 ; X64-NEXT:    movq %rdi, %rax
 ; X64-NEXT:    subq %rsi, %rax
+; X64-NEXT:    subq %rdi, %rsi
+; X64-NEXT:    cmovaeq %rsi, %rax
 ; X64-NEXT:    negq %rax
-; X64-NEXT:    subq %rsi, %rdi
-; X64-NEXT:    cmovbq %rdi, %rax
 ; X64-NEXT:    retq
   %cmp = icmp ult i64 %a, %b
   %ab = sub i64 %a, %b
@@ -853,34 +846,36 @@ define i128 @abd_cmp_i128(i128 %a, i128 %b) nounwind {
 ; X86-NEXT:    pushl %ebx
 ; X86-NEXT:    pushl %edi
 ; X86-NEXT:    pushl %esi
-; X86-NEXT:    pushl %eax
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %esi
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %ebx
-; X86-NEXT:    subl %edx, %eax
-; X86-NEXT:    movl %eax, (%esp) # 4-byte Spill
-; X86-NEXT:    sbbl %esi, %ebx
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %ebp
-; X86-NEXT:    sbbl %ecx, %ebp
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %edi
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    sbbl %edi, %eax
+; X86-NEXT:    xorl %edi, %edi
 ; X86-NEXT:    subl {{[0-9]+}}(%esp), %edx
+; X86-NEXT:    sbbl {{[0-9]+}}(%esp), %ebx
 ; X86-NEXT:    sbbl {{[0-9]+}}(%esp), %esi
 ; X86-NEXT:    sbbl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    sbbl {{[0-9]+}}(%esp), %edi
-; X86-NEXT:    cmovael (%esp), %edx # 4-byte Folded Reload
-; X86-NEXT:    cmovael %ebx, %esi
-; X86-NEXT:    cmovael %ebp, %ecx
-; X86-NEXT:    cmovael %eax, %edi
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    movl %edi, 12(%eax)
-; X86-NEXT:    movl %ecx, 8(%eax)
-; X86-NEXT:    movl %esi, 4(%eax)
+; X86-NEXT:    movl $0, %ebp
+; X86-NEXT:    sbbl %ebp, %ebp
+; X86-NEXT:    xorl %ebp, %ecx
+; X86-NEXT:    xorl %ebp, %esi
+; X86-NEXT:    xorl %ebp, %ebx
+; X86-NEXT:    xorl %ebp, %edx
+; X86-NEXT:    subl %ebp, %edx
+; X86-NEXT:    sbbl %ebp, %ebx
+; X86-NEXT:    sbbl %ebp, %esi
+; X86-NEXT:    sbbl %ebp, %ecx
+; X86-NEXT:    negl %edx
+; X86-NEXT:    movl $0, %ebp
+; X86-NEXT:    sbbl %ebx, %ebp
+; X86-NEXT:    movl $0, %ebx
+; X86-NEXT:    sbbl %esi, %ebx
+; X86-NEXT:    sbbl %ecx, %edi
 ; X86-NEXT:    movl %edx, (%eax)
-; X86-NEXT:    addl $4, %esp
+; X86-NEXT:    movl %ebp, 4(%eax)
+; X86-NEXT:    movl %ebx, 8(%eax)
+; X86-NEXT:    movl %edi, 12(%eax)
 ; X86-NEXT:    popl %esi
 ; X86-NEXT:    popl %edi
 ; X86-NEXT:    popl %ebx
@@ -889,15 +884,19 @@ define i128 @abd_cmp_i128(i128 %a, i128 %b) nounwind {
 ;
 ; X64-LABEL: abd_cmp_i128:
 ; X64:       # %bb.0:
-; X64-NEXT:    movq %rdx, %rax
-; X64-NEXT:    subq %rdi, %rax
-; X64-NEXT:    movq %rcx, %r8
-; X64-NEXT:    sbbq %rsi, %r8
-; X64-NEXT:    subq %rdx, %rdi
+; X64-NEXT:    movq %rdi, %rax
+; X64-NEXT:    xorl %edi, %edi
+; X64-NEXT:    subq %rdx, %rax
 ; X64-NEXT:    sbbq %rcx, %rsi
-; X64-NEXT:    cmovbq %rdi, %rax
-; X64-NEXT:    cmovbq %rsi, %r8
-; X64-NEXT:    movq %r8, %rdx
+; X64-NEXT:    movl $0, %ecx
+; X64-NEXT:    sbbq %rcx, %rcx
+; X64-NEXT:    xorq %rcx, %rsi
+; X64-NEXT:    xorq %rcx, %rax
+; X64-NEXT:    subq %rcx, %rax
+; X64-NEXT:    sbbq %rcx, %rsi
+; X64-NEXT:    negq %rax
+; X64-NEXT:    sbbq %rsi, %rdi
+; X64-NEXT:    movq %rdi, %rdx
 ; X64-NEXT:    retq
   %cmp = icmp ult i128 %a, %b
   %ab = sub i128 %a, %b
diff --git a/llvm/test/CodeGen/X86/abdu.ll b/llvm/test/CodeGen/X86/abdu.ll
@@ -608,25 +608,21 @@ define i8 @abd_cmp_i8(i8 %a, i8 %b) nounwind {
 ; X86:       # %bb.0:
 ; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
 ; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    movl %eax, %edx
-; X86-NEXT:    subb %cl, %dl
-; X86-NEXT:    negb %dl
-; X86-NEXT:    subb %cl, %al
-; X86-NEXT:    movzbl %al, %ecx
-; X86-NEXT:    movzbl %dl, %eax
-; X86-NEXT:    cmovael %ecx, %eax
+; X86-NEXT:    subl %eax, %ecx
+; X86-NEXT:    movl %ecx, %eax
+; X86-NEXT:    negl %eax
+; X86-NEXT:    cmovsl %ecx, %eax
 ; X86-NEXT:    # kill: def $al killed $al killed $eax
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: abd_cmp_i8:
 ; X64:       # %bb.0:
-; X64-NEXT:    movl %esi, %eax
-; X64-NEXT:    subb %dil, %al
-; X64-NEXT:    negb %al
-; X64-NEXT:    subb %dil, %sil
+; X64-NEXT:    movzbl %dil, %eax
 ; X64-NEXT:    movzbl %sil, %ecx
-; X64-NEXT:    movzbl %al, %eax
-; X64-NEXT:    cmovael %ecx, %eax
+; X64-NEXT:    subl %eax, %ecx
+; X64-NEXT:    movl %ecx, %eax
+; X64-NEXT:    negl %eax
+; X64-NEXT:    cmovsl %ecx, %eax
 ; X64-NEXT:    # kill: def $al killed $al killed $eax
 ; X64-NEXT:    retq
   %cmp = icmp ugt i8 %a, %b
@@ -670,9 +666,8 @@ define i32 @abd_cmp_i32(i32 %a, i32 %b) nounwind {
 ; X86:       # %bb.0:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    movl %eax, %edx
-; X86-NEXT:    subl %ecx, %edx
-; X86-NEXT:    negl %edx
+; X86-NEXT:    movl %ecx, %edx
+; X86-NEXT:    subl %eax, %edx
 ; X86-NEXT:    subl %ecx, %eax
 ; X86-NEXT:    cmovbl %edx, %eax
 ; X86-NEXT:    retl
@@ -681,9 +676,8 @@ define i32 @abd_cmp_i32(i32 %a, i32 %b) nounwind {
 ; X64:       # %bb.0:
 ; X64-NEXT:    movl %edi, %eax
 ; X64-NEXT:    subl %esi, %eax
-; X64-NEXT:    negl %eax
-; X64-NEXT:    subl %esi, %edi
-; X64-NEXT:    cmovael %edi, %eax
+; X64-NEXT:    subl %edi, %esi
+; X64-NEXT:    cmovael %esi, %eax
 ; X64-NEXT:    retq
   %cmp = icmp ult i32 %a, %b
   %ab = sub i32 %a, %b
