[DAGCombiner] Fold subtraction if above a constant threshold to umin

Like llvm#134235, but with a constant.
It's a pattern in Adler-32 checksum calculation in zlib.

Example:

    unsigned adler32_mod(unsigned x) {
      return x >= 65521u ? x - 65521u : x;
    }

Before, on RISC-V:

    lui     a1, 16
    lui     a2, 1048560
    addiw   a1, a1, -16
    sltu    a1, a1, a0
    negw    a1, a1
    addi    a2, a2, 15
    and     a1, a1, a2
    addw    a0, a0, a1

Or, with Zicond:

    lui     a1, 16
    lui     a2, 1048560
    addiw   a1, a1, -16
    sltu    a1, a1, a0
    addi    a2, a2, 15
    czero.eqz  a1, a2, a1
    addw    a0, a0, a1

After, with Zbb:

    lui a1, 1048560
    addi a1, a1, 15
    addw a1, a0, a1
    minu a0, a1, a0

Author: pfusik

llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp

@@ -12078,6 +12078,12 @@ SDValue DAGCombiner::visitSELECT(SDNode *N) {
 
     if (SDValue NewSel = SimplifySelect(DL, N0, N1, N2))
       return NewSel;
+
+    // (select (ugt x, C), (add x, ~C), x) -> (umin (add x, ~C), x)
+    APInt C;
+    if (CC == ISD::SETUGT && Cond0 == N2 && sd_match(Cond1, m_ConstInt(C)) &&
+        sd_match(N1, m_Add(m_Specific(N2), m_SpecificInt(~C))) && hasUMin(VT))
+      return DAG.getNode(ISD::UMIN, DL, VT, N1, N2);
   }
 
   if (!VT.isVector())

llvm/test/CodeGen/RISCV/rv32zbb.ll

@@ -1720,45 +1720,67 @@ define i32 @sub_if_uge_multiuse_cmp_store_i32(i32 %x, i32 %y, ptr %z) {
 }
 
 define i8 @sub_if_uge_C_i8(i8 zeroext %x) {
-; CHECK-LABEL: sub_if_uge_C_i8:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    sltiu a1, a0, 13
-; CHECK-NEXT:    addi a1, a1, -1
-; CHECK-NEXT:    andi a1, a1, -13
-; CHECK-NEXT:    add a0, a0, a1
-; CHECK-NEXT:    ret
+; RV32I-LABEL: sub_if_uge_C_i8:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    sltiu a1, a0, 13
+; RV32I-NEXT:    addi a1, a1, -1
+; RV32I-NEXT:    andi a1, a1, -13
+; RV32I-NEXT:    add a0, a0, a1
+; RV32I-NEXT:    ret
+;
+; RV32ZBB-LABEL: sub_if_uge_C_i8:
+; RV32ZBB:       # %bb.0:
+; RV32ZBB-NEXT:    addi a1, a0, -13
+; RV32ZBB-NEXT:    zext.b a1, a1
+; RV32ZBB-NEXT:    minu a0, a1, a0
+; RV32ZBB-NEXT:    ret
   %cmp = icmp ugt i8 %x, 12
   %sub = add i8 %x, -13
   %conv4 = select i1 %cmp, i8 %sub, i8 %x
   ret i8 %conv4
 }
 
 define i16 @sub_if_uge_C_i16(i16 zeroext %x) {
-; CHECK-LABEL: sub_if_uge_C_i16:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    sltiu a1, a0, 251
-; CHECK-NEXT:    addi a1, a1, -1
-; CHECK-NEXT:    andi a1, a1, -251
-; CHECK-NEXT:    add a0, a0, a1
-; CHECK-NEXT:    ret
+; RV32I-LABEL: sub_if_uge_C_i16:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    sltiu a1, a0, 251
+; RV32I-NEXT:    addi a1, a1, -1
+; RV32I-NEXT:    andi a1, a1, -251
+; RV32I-NEXT:    add a0, a0, a1
+; RV32I-NEXT:    ret
+;
+; RV32ZBB-LABEL: sub_if_uge_C_i16:
+; RV32ZBB:       # %bb.0:
+; RV32ZBB-NEXT:    addi a1, a0, -251
+; RV32ZBB-NEXT:    zext.h a1, a1
+; RV32ZBB-NEXT:    minu a0, a1, a0
+; RV32ZBB-NEXT:    ret
   %cmp = icmp ugt i16 %x, 250
   %sub = add i16 %x, -251
   %conv4 = select i1 %cmp, i16 %sub, i16 %x
   ret i16 %conv4
 }
 
 define i32 @sub_if_uge_C_i32(i32 signext %x) {
-; CHECK-LABEL: sub_if_uge_C_i32:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    lui a1, 16
-; CHECK-NEXT:    lui a2, 1048560
-; CHECK-NEXT:    addi a1, a1, -16
-; CHECK-NEXT:    sltu a1, a1, a0
-; CHECK-NEXT:    neg a1, a1
-; CHECK-NEXT:    addi a2, a2, 15
-; CHECK-NEXT:    and a1, a1, a2
-; CHECK-NEXT:    add a0, a0, a1
-; CHECK-NEXT:    ret
+; RV32I-LABEL: sub_if_uge_C_i32:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    lui a1, 16
+; RV32I-NEXT:    lui a2, 1048560
+; RV32I-NEXT:    addi a1, a1, -16
+; RV32I-NEXT:    sltu a1, a1, a0
+; RV32I-NEXT:    neg a1, a1
+; RV32I-NEXT:    addi a2, a2, 15
+; RV32I-NEXT:    and a1, a1, a2
+; RV32I-NEXT:    add a0, a0, a1
+; RV32I-NEXT:    ret
+;
+; RV32ZBB-LABEL: sub_if_uge_C_i32:
+; RV32ZBB:       # %bb.0:
+; RV32ZBB-NEXT:    lui a1, 1048560
+; RV32ZBB-NEXT:    addi a1, a1, 15
+; RV32ZBB-NEXT:    add a1, a0, a1
+; RV32ZBB-NEXT:    minu a0, a1, a0
+; RV32ZBB-NEXT:    ret
   %cmp = icmp ugt i32 %x, 65520
   %sub = add i32 %x, -65521
   %cond = select i1 %cmp, i32 %sub, i32 %x
@@ -1797,18 +1819,30 @@ define i64 @sub_if_uge_C_i64(i64 %x) {
 }
 
 define i32 @sub_if_uge_C_multiuse_cmp_i32(i32 signext %x, ptr %z) {
-; CHECK-LABEL: sub_if_uge_C_multiuse_cmp_i32:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    lui a2, 16
-; CHECK-NEXT:    lui a3, 1048560
-; CHECK-NEXT:    addi a2, a2, -16
-; CHECK-NEXT:    sltu a2, a2, a0
-; CHECK-NEXT:    neg a4, a2
-; CHECK-NEXT:    addi a3, a3, 15
-; CHECK-NEXT:    and a3, a4, a3
-; CHECK-NEXT:    add a0, a0, a3
-; CHECK-NEXT:    sw a2, 0(a1)
-; CHECK-NEXT:    ret
+; RV32I-LABEL: sub_if_uge_C_multiuse_cmp_i32:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    lui a2, 16
+; RV32I-NEXT:    lui a3, 1048560
+; RV32I-NEXT:    addi a2, a2, -16
+; RV32I-NEXT:    sltu a2, a2, a0
+; RV32I-NEXT:    neg a4, a2
+; RV32I-NEXT:    addi a3, a3, 15
+; RV32I-NEXT:    and a3, a4, a3
+; RV32I-NEXT:    add a0, a0, a3
+; RV32I-NEXT:    sw a2, 0(a1)
+; RV32I-NEXT:    ret
+;
+; RV32ZBB-LABEL: sub_if_uge_C_multiuse_cmp_i32:
+; RV32ZBB:       # %bb.0:
+; RV32ZBB-NEXT:    lui a2, 16
+; RV32ZBB-NEXT:    lui a3, 1048560
+; RV32ZBB-NEXT:    addi a2, a2, -16
+; RV32ZBB-NEXT:    addi a3, a3, 15
+; RV32ZBB-NEXT:    sltu a2, a2, a0
+; RV32ZBB-NEXT:    add a3, a0, a3
+; RV32ZBB-NEXT:    minu a0, a3, a0
+; RV32ZBB-NEXT:    sw a2, 0(a1)
+; RV32ZBB-NEXT:    ret
   %cmp = icmp ugt i32 %x, 65520
   %conv = zext i1 %cmp to i32
   store i32 %conv, ptr %z, align 4
@@ -1818,20 +1852,29 @@ define i32 @sub_if_uge_C_multiuse_cmp_i32(i32 signext %x, ptr %z) {
 }
 
 define i32 @sub_if_uge_C_multiuse_sub_i32(i32 signext %x, ptr %z) {
-; CHECK-LABEL: sub_if_uge_C_multiuse_sub_i32:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    lui a2, 1048560
-; CHECK-NEXT:    lui a3, 16
-; CHECK-NEXT:    addi a2, a2, 15
-; CHECK-NEXT:    add a2, a0, a2
-; CHECK-NEXT:    addi a3, a3, -16
-; CHECK-NEXT:    sw a2, 0(a1)
-; CHECK-NEXT:    bltu a3, a0, .LBB62_2
-; CHECK-NEXT:  # %bb.1:
-; CHECK-NEXT:    mv a2, a0
-; CHECK-NEXT:  .LBB62_2:
-; CHECK-NEXT:    mv a0, a2
-; CHECK-NEXT:    ret
+; RV32I-LABEL: sub_if_uge_C_multiuse_sub_i32:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    lui a2, 1048560
+; RV32I-NEXT:    lui a3, 16
+; RV32I-NEXT:    addi a2, a2, 15
+; RV32I-NEXT:    add a2, a0, a2
+; RV32I-NEXT:    addi a3, a3, -16
+; RV32I-NEXT:    sw a2, 0(a1)
+; RV32I-NEXT:    bltu a3, a0, .LBB62_2
+; RV32I-NEXT:  # %bb.1:
+; RV32I-NEXT:    mv a2, a0
+; RV32I-NEXT:  .LBB62_2:
+; RV32I-NEXT:    mv a0, a2
+; RV32I-NEXT:    ret
+;
+; RV32ZBB-LABEL: sub_if_uge_C_multiuse_sub_i32:
+; RV32ZBB:       # %bb.0:
+; RV32ZBB-NEXT:    lui a2, 1048560
+; RV32ZBB-NEXT:    addi a2, a2, 15
+; RV32ZBB-NEXT:    add a2, a0, a2
+; RV32ZBB-NEXT:    minu a0, a2, a0
+; RV32ZBB-NEXT:    sw a2, 0(a1)
+; RV32ZBB-NEXT:    ret
   %sub = add i32 %x, -65521
   store i32 %sub, ptr %z, align 4
   %cmp = icmp ugt i32 %x, 65520