[DAGCombiner][RISCV][AArch64][PowerPC] Restrict foldAndOrOfSETCC from using SMIN/SMAX where and OR/AND would do.

[craig.topper]

Summary

This removes some diffs created by D153502.

I'm assuming an AND/OR won't be worse than an SMIN/SMAX. For
RISC-V at least, AND/OR can be a shorter encoding than SMIN/SMAX.

It's weird that we have two different functions responsible for
folding logic of setccs, but I'm not ready to try to untangle that.

I'm unclear if the PowerPC chang is a regression or not. It looks
like it might use more registers, but I don't understand PowerPC
register so I'm not sure.

[arsenm]

I'm assuming the register numbers on ppc don't matter because it's not a gpu

llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp

 //===- DAGCombiner.cpp - Implement a DAG node combiner --------------------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
if (CCL == CCR) {
       } else if (RHS0 == LHS1) {
         CommonValue = LHS1;
         Operand1 = LHS0;
         Operand2 = RHS1;
         CC = CCL;
       }
     }
 
+    // Don't do this transform for sign bit tests. Let foldLogicOfSetCCs
+    // handle it using OR/AND.
+    if (CC == ISD::SETLT && isNullOrNullSplat(CommonValue))
+      CC = ISD::SETCC_INVALID;
+    else if (CC == ISD::SETGT && isAllOnesOrAllOnesSplat(CommonValue))
+      CC = ISD::SETCC_INVALID;
+
     if (CC != ISD::SETCC_INVALID) {
       unsigned NewOpcode;
       bool IsSigned = isSignedIntSetCC(CC);
       bool IsLess = (CC == ISD::SETLE || CC == ISD::SETULE ||
                      CC == ISD::SETLT || CC == ISD::SETULT);
       bool IsOr = (LogicOp->getOpcode() == ISD::OR);
       if (IsLess == IsOr)
         NewOpcode = IsSigned ? ISD::SMIN : ISD::UMIN;

llvm/test/CodeGen/AArch64/vecreduce-bool.ll

; CHECK-NEXT:    ret
   %y = call i1 @llvm.vector.reduce.and.v16i1(<16 x i1> %x)
   %z = select i1 %y, i32 %a1, i32 %a2
   ret i32 %z
 }
 
 define i32 @reduce_and_v32(<32 x i8> %a0, i32 %a1, i32 %a2) nounwind {
 ; CHECK-LABEL: reduce_and_v32:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    smax v0.16b, v0.16b, v1.16b
+; CHECK-NEXT:    and v0.16b, v0.16b, v1.16b
 ; CHECK-NEXT:    cmlt v0.16b, v0.16b, #0
 ; CHECK-NEXT:    uminv b0, v0.16b
 ; CHECK-NEXT:    fmov w8, s0
 ; CHECK-NEXT:    tst w8, #0x1
 ; CHECK-NEXT:    csel w0, w0, w1, ne
 ; CHECK-NEXT:    ret
   %x = icmp slt <32 x i8> %a0, zeroinitializer
   %y = call i1 @llvm.vector.reduce.and.v32i1(<32 x i1> %x)
; CHECK-NEXT:    ret
   %y = call i1 @llvm.vector.reduce.or.v16i1(<16 x i1> %x)
   %z = select i1 %y, i32 %a1, i32 %a2
   ret i32 %z
 }
 
 define i32 @reduce_or_v32(<32 x i8> %a0, i32 %a1, i32 %a2) nounwind {
 ; CHECK-LABEL: reduce_or_v32:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    smin v0.16b, v0.16b, v1.16b
+; CHECK-NEXT:    orr v0.16b, v0.16b, v1.16b
 ; CHECK-NEXT:    cmlt v0.16b, v0.16b, #0
 ; CHECK-NEXT:    umaxv b0, v0.16b
 ; CHECK-NEXT:    fmov w8, s0
 ; CHECK-NEXT:    tst w8, #0x1
 ; CHECK-NEXT:    csel w0, w0, w1, ne
 ; CHECK-NEXT:    ret
   %x = icmp slt <32 x i8> %a0, zeroinitializer
   %y = call i1 @llvm.vector.reduce.or.v32i1(<32 x i1> %x)
   %z = select i1 %y, i32 %a1, i32 %a2
   ret i32 %z
 }

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

; CHECK-NEXT:    blr
   %b = icmp eq <4 x i32> %Q, zeroinitializer
   %c = and <4 x i1> %a, %b
   ret <4 x i1> %c
 }
 
 define <4 x i1> @all_sign_bits_clear_vec(<4 x i32> %P, <4 x i32> %Q) {
 ; CHECK-LABEL: all_sign_bits_clear_vec:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    vminsw 2, 2, 3
-; CHECK-NEXT:    xxleqv 35, 35, 35
-; CHECK-NEXT:    vcmpgtsw 2, 2, 3
+; CHECK-NEXT:    xxleqv 36, 36, 36
+; CHECK-NEXT:    xxlor 34, 34, 35
+; CHECK-NEXT:    vcmpgtsw 2, 2, 4
 ; CHECK-NEXT:    blr
   %a = icmp sgt <4 x i32> %P, <i32 -1, i32 -1, i32 -1, i32 -1>
   %b = icmp sgt <4 x i32> %Q, <i32 -1, i32 -1, i32 -1, i32 -1>
   %c = and <4 x i1> %a, %b
   ret <4 x i1> %c
 }
 
 define <4 x i1> @all_bits_set_vec(<4 x i32> %P, <4 x i32> %Q) {
 ; CHECK-LABEL: all_bits_set_vec:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    xxleqv 36, 36, 36
 ; CHECK-NEXT:    xxland 34, 34, 35
 ; CHECK-NEXT:    vcmpequw 2, 2, 4
 ; CHECK-NEXT:    blr
   %a = icmp eq <4 x i32> %P, <i32 -1, i32 -1, i32 -1, i32 -1>
   %b = icmp eq <4 x i32> %Q, <i32 -1, i32 -1, i32 -1, i32 -1>
   %c = and <4 x i1> %a, %b
   ret <4 x i1> %c
 }
 
 define <4 x i1> @all_sign_bits_set_vec(<4 x i32> %P, <4 x i32> %Q) {
 ; CHECK-LABEL: all_sign_bits_set_vec:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmaxsw 2, 2, 3
-; CHECK-NEXT:    xxlxor 35, 35, 35
-; CHECK-NEXT:    vcmpgtsw 2, 3, 2
+; CHECK-NEXT:    xxlxor 36, 36, 36
+; CHECK-NEXT:    xxland 34, 34, 35
+; CHECK-NEXT:    vcmpgtsw 2, 4, 2
 ; CHECK-NEXT:    blr
   %a = icmp slt <4 x i32> %P, zeroinitializer
   %b = icmp slt <4 x i32> %Q, zeroinitializer
   %c = and <4 x i1> %a, %b
   ret <4 x i1> %c
 }
 
 define <4 x i1> @any_bits_set_vec(<4 x i32> %P, <4 x i32> %Q) {
 ; CHECK-LABEL: any_bits_set_vec:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    xxlxor 36, 36, 36
 ; CHECK-NEXT:    xxlor 34, 34, 35
 ; CHECK-NEXT:    vcmpequw 2, 2, 4
 ; CHECK-NEXT:    xxlnor 34, 34, 34
 ; CHECK-NEXT:    blr
   %a = icmp ne <4 x i32> %P, zeroinitializer
   %b = icmp ne <4 x i32> %Q, zeroinitializer
   %c = or <4 x i1> %a, %b
   ret <4 x i1> %c
 }
 
 define <4 x i1> @any_sign_bits_set_vec(<4 x i32> %P, <4 x i32> %Q) {
 ; CHECK-LABEL: any_sign_bits_set_vec:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    vminsw 2, 2, 3
-; CHECK-NEXT:    xxlxor 35, 35, 35
-; CHECK-NEXT:    vcmpgtsw 2, 3, 2
+; CHECK-NEXT:    xxlxor 36, 36, 36
+; CHECK-NEXT:    xxlor 34, 34, 35
+; CHECK-NEXT:    vcmpgtsw 2, 4, 2
 ; CHECK-NEXT:    blr
   %a = icmp slt <4 x i32> %P, zeroinitializer
   %b = icmp slt <4 x i32> %Q, zeroinitializer
   %c = or <4 x i1> %a, %b
   ret <4 x i1> %c
 }
 
 define <4 x i1> @any_bits_clear_vec(<4 x i32> %P, <4 x i32> %Q) {
 ; CHECK-LABEL: any_bits_clear_vec:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    xxleqv 36, 36, 36
 ; CHECK-NEXT:    xxland 34, 34, 35
 ; CHECK-NEXT:    vcmpequw 2, 2, 4
 ; CHECK-NEXT:    xxlnor 34, 34, 34
 ; CHECK-NEXT:    blr
   %a = icmp ne <4 x i32> %P, <i32 -1, i32 -1, i32 -1, i32 -1>
   %b = icmp ne <4 x i32> %Q, <i32 -1, i32 -1, i32 -1, i32 -1>
   %c = or <4 x i1> %a, %b
   ret <4 x i1> %c
 }
 
 define <4 x i1> @any_sign_bits_clear_vec(<4 x i32> %P, <4 x i32> %Q) {
 ; CHECK-LABEL: any_sign_bits_clear_vec:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmaxsw 2, 2, 3
-; CHECK-NEXT:    xxleqv 35, 35, 35
-; CHECK-NEXT:    vcmpgtsw 2, 2, 3
+; CHECK-NEXT:    xxleqv 36, 36, 36
+; CHECK-NEXT:    xxland 34, 34, 35
+; CHECK-NEXT:    vcmpgtsw 2, 2, 4
 ; CHECK-NEXT:    blr
   %a = icmp sgt <4 x i32> %P, <i32 -1, i32 -1, i32 -1, i32 -1>
   %b = icmp sgt <4 x i32> %Q, <i32 -1, i32 -1, i32 -1, i32 -1>
   %c = or <4 x i1> %a, %b
   ret <4 x i1> %c
 }
 
 define zeroext i1 @ne_neg1_and_ne_zero(i64 %x) {

llvm/test/CodeGen/RISCV/zbb-cmp-combine.ll

 ; CHECK-RV64I-NEXT:    mv s1, a0
 ; CHECK-RV64I-NEXT:    mv a0, a1
 ; CHECK-RV64I-NEXT:    mv a1, s1
 ; CHECK-RV64I-NEXT:    call __gesf2@plt
 ; CHECK-RV64I-NEXT:    mv s2, a0
 ; CHECK-RV64I-NEXT:    mv a0, s0
 ; CHECK-RV64I-NEXT:    mv a1, s1
 ; CHECK-RV64I-NEXT:    call __gesf2@plt
-; CHECK-RV64I-NEXT:    min a0, s2, a0
+; CHECK-RV64I-NEXT:    or a0, s2, a0
 ; CHECK-RV64I-NEXT:    slti a0, a0, 0
 ; CHECK-RV64I-NEXT:    ld ra, 24(sp) # 8-byte Folded Reload
 ; CHECK-RV64I-NEXT:    ld s0, 16(sp) # 8-byte Folded Reload
 ; CHECK-RV64I-NEXT:    ld s1, 8(sp) # 8-byte Folded Reload
 ; CHECK-RV64I-NEXT:    ld s2, 0(sp) # 8-byte Folded Reload
 ; CHECK-RV64I-NEXT:    addi sp, sp, 32
 ; CHECK-RV64I-NEXT:    ret
 ;

 ; CHECK-NEXT:    mv s1, a0
 ; CHECK-NEXT:    mv a0, a1
 ; CHECK-NEXT:    mv a1, s1
 ; CHECK-NEXT:    call __gedf2@plt
 ; CHECK-NEXT:    mv s2, a0
 ; CHECK-NEXT:    mv a0, s0
 ; CHECK-NEXT:    mv a1, s1
 ; CHECK-NEXT:    call __gedf2@plt
-; CHECK-NEXT:    min a0, s2, a0
+; CHECK-NEXT:    or a0, s2, a0
 ; CHECK-NEXT:    slti a0, a0, 0
 ; CHECK-NEXT:    ld ra, 24(sp) # 8-byte Folded Reload
 ; CHECK-NEXT:    ld s0, 16(sp) # 8-byte Folded Reload
 ; CHECK-NEXT:    ld s1, 8(sp) # 8-byte Folded Reload
 ; CHECK-NEXT:    ld s2, 0(sp) # 8-byte Folded Reload
 ; CHECK-NEXT:    addi sp, sp, 32
 ; CHECK-NEXT:    ret
   %l0 = fcmp ult double %a, %c