[X86] Add vector signbit parity checks for non-popcnt targets

Noticed while looking at D117983 - we miss some parity patterns with/without popcnt

llvm/test/CodeGen/X86/parity-vec.ll

@@ -1,27 +1,44 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+popcnt,+sse2 | FileCheck %s
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=-popcnt,+sse2 | FileCheck %s --check-prefix=NOPOPCNT
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+popcnt,+sse2 | FileCheck %s --check-prefix=POPCNT
 
 define i1 @noncanonical_parity(<16 x i1> %x) {
-; CHECK-LABEL: noncanonical_parity:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    psllw $7, %xmm0
-; CHECK-NEXT:    pmovmskb %xmm0, %eax
-; CHECK-NEXT:    popcntl %eax, %eax
-; CHECK-NEXT:    andl $1, %eax
-; CHECK-NEXT:    # kill: def $al killed $al killed $eax
-; CHECK-NEXT:    retq
+; NOPOPCNT-LABEL: noncanonical_parity:
+; NOPOPCNT:       # %bb.0:
+; NOPOPCNT-NEXT:    psllw $7, %xmm0
+; NOPOPCNT-NEXT:    pmovmskb %xmm0, %eax
+; NOPOPCNT-NEXT:    xorb %ah, %al
+; NOPOPCNT-NEXT:    setnp %al
+; NOPOPCNT-NEXT:    retq
+;
+; POPCNT-LABEL: noncanonical_parity:
+; POPCNT:       # %bb.0:
+; POPCNT-NEXT:    psllw $7, %xmm0
+; POPCNT-NEXT:    pmovmskb %xmm0, %eax
+; POPCNT-NEXT:    popcntl %eax, %eax
+; POPCNT-NEXT:    andl $1, %eax
+; POPCNT-NEXT:    # kill: def $al killed $al killed $eax
+; POPCNT-NEXT:    retq
   %r = call i1 @llvm.vector.reduce.xor.v16i1(<16 x i1> %x)
   ret i1 %r
 }
 define i1 @canonical_parity(<16 x i1> %x) {
-; CHECK-LABEL: canonical_parity:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    psllw $7, %xmm0
-; CHECK-NEXT:    pmovmskb %xmm0, %eax
-; CHECK-NEXT:    popcntl %eax, %eax
-; CHECK-NEXT:    testb $1, %al
-; CHECK-NEXT:    setne %al
-; CHECK-NEXT:    retq
+; NOPOPCNT-LABEL: canonical_parity:
+; NOPOPCNT:       # %bb.0:
+; NOPOPCNT-NEXT:    psllw $7, %xmm0
+; NOPOPCNT-NEXT:    pmovmskb %xmm0, %eax
+; NOPOPCNT-NEXT:    xorb %ah, %al
+; NOPOPCNT-NEXT:    setnp %al
+; NOPOPCNT-NEXT:    retq
+;
+; POPCNT-LABEL: canonical_parity:
+; POPCNT:       # %bb.0:
+; POPCNT-NEXT:    psllw $7, %xmm0
+; POPCNT-NEXT:    pmovmskb %xmm0, %eax
+; POPCNT-NEXT:    popcntl %eax, %eax
+; POPCNT-NEXT:    testb $1, %al
+; POPCNT-NEXT:    setne %al
+; POPCNT-NEXT:    retq
   %i1 = bitcast <16 x i1> %x to i16
   %i2 = call i16 @llvm.ctpop.i16(i16 %i1)
   %i3 = and i16 %i2, 1
@@ -29,13 +46,37 @@ define i1 @canonical_parity(<16 x i1> %x) {
   ret i1 %i4
 }
 define i1 @canonical_parity_noncanonical_pred(<16 x i1> %x) {
-; CHECK-LABEL: canonical_parity_noncanonical_pred:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    psllw $7, %xmm0
-; CHECK-NEXT:    pmovmskb %xmm0, %eax
-; CHECK-NEXT:    popcntl %eax, %eax
-; CHECK-NEXT:    # kill: def $al killed $al killed $eax
-; CHECK-NEXT:    retq
+; NOPOPCNT-LABEL: canonical_parity_noncanonical_pred:
+; NOPOPCNT:       # %bb.0:
+; NOPOPCNT-NEXT:    psllw $7, %xmm0
+; NOPOPCNT-NEXT:    pmovmskb %xmm0, %eax
+; NOPOPCNT-NEXT:    movl %eax, %ecx
+; NOPOPCNT-NEXT:    shrl %ecx
+; NOPOPCNT-NEXT:    andl $21845, %ecx # imm = 0x5555
+; NOPOPCNT-NEXT:    subl %ecx, %eax
+; NOPOPCNT-NEXT:    movl %eax, %ecx
+; NOPOPCNT-NEXT:    andl $13107, %ecx # imm = 0x3333
+; NOPOPCNT-NEXT:    shrl $2, %eax
+; NOPOPCNT-NEXT:    andl $13107, %eax # imm = 0x3333
+; NOPOPCNT-NEXT:    addl %ecx, %eax
+; NOPOPCNT-NEXT:    movl %eax, %ecx
+; NOPOPCNT-NEXT:    shrl $4, %ecx
+; NOPOPCNT-NEXT:    addl %eax, %ecx
+; NOPOPCNT-NEXT:    andl $3855, %ecx # imm = 0xF0F
+; NOPOPCNT-NEXT:    movl %ecx, %eax
+; NOPOPCNT-NEXT:    shll $8, %eax
+; NOPOPCNT-NEXT:    addl %ecx, %eax
+; NOPOPCNT-NEXT:    shrl $8, %eax
+; NOPOPCNT-NEXT:    # kill: def $al killed $al killed $eax
+; NOPOPCNT-NEXT:    retq
+;
+; POPCNT-LABEL: canonical_parity_noncanonical_pred:
+; POPCNT:       # %bb.0:
+; POPCNT-NEXT:    psllw $7, %xmm0
+; POPCNT-NEXT:    pmovmskb %xmm0, %eax
+; POPCNT-NEXT:    popcntl %eax, %eax
+; POPCNT-NEXT:    # kill: def $al killed $al killed $eax
+; POPCNT-NEXT:    retq
   %i1 = bitcast <16 x i1> %x to i16
   %i2 = call i16 @llvm.ctpop.i16(i16 %i1)
   %i3 = and i16 %i2, 1
@@ -44,28 +85,44 @@ define i1 @canonical_parity_noncanonical_pred(<16 x i1> %x) {
 }
 
 define i1 @noncanonical_nonparity(<16 x i1> %x) {
-; CHECK-LABEL: noncanonical_nonparity:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    psllw $7, %xmm0
-; CHECK-NEXT:    pmovmskb %xmm0, %eax
-; CHECK-NEXT:    popcntl %eax, %eax
-; CHECK-NEXT:    andl $1, %eax
-; CHECK-NEXT:    xorb $1, %al
-; CHECK-NEXT:    # kill: def $al killed $al killed $eax
-; CHECK-NEXT:    retq
+; NOPOPCNT-LABEL: noncanonical_nonparity:
+; NOPOPCNT:       # %bb.0:
+; NOPOPCNT-NEXT:    psllw $7, %xmm0
+; NOPOPCNT-NEXT:    pmovmskb %xmm0, %eax
+; NOPOPCNT-NEXT:    xorb %ah, %al
+; NOPOPCNT-NEXT:    setp %al
+; NOPOPCNT-NEXT:    retq
+;
+; POPCNT-LABEL: noncanonical_nonparity:
+; POPCNT:       # %bb.0:
+; POPCNT-NEXT:    psllw $7, %xmm0
+; POPCNT-NEXT:    pmovmskb %xmm0, %eax
+; POPCNT-NEXT:    popcntl %eax, %eax
+; POPCNT-NEXT:    andl $1, %eax
+; POPCNT-NEXT:    xorb $1, %al
+; POPCNT-NEXT:    # kill: def $al killed $al killed $eax
+; POPCNT-NEXT:    retq
   %r.inv = call i1 @llvm.vector.reduce.xor.v16i1(<16 x i1> %x)
   %r = xor i1 %r.inv, -1
   ret i1 %r
 }
 define i1 @canonical_nonparity(<16 x i1> %x) {
-; CHECK-LABEL: canonical_nonparity:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    psllw $7, %xmm0
-; CHECK-NEXT:    pmovmskb %xmm0, %eax
-; CHECK-NEXT:    popcntl %eax, %eax
-; CHECK-NEXT:    testb $1, %al
-; CHECK-NEXT:    sete %al
-; CHECK-NEXT:    retq
+; NOPOPCNT-LABEL: canonical_nonparity:
+; NOPOPCNT:       # %bb.0:
+; NOPOPCNT-NEXT:    psllw $7, %xmm0
+; NOPOPCNT-NEXT:    pmovmskb %xmm0, %eax
+; NOPOPCNT-NEXT:    xorb %ah, %al
+; NOPOPCNT-NEXT:    setp %al
+; NOPOPCNT-NEXT:    retq
+;
+; POPCNT-LABEL: canonical_nonparity:
+; POPCNT:       # %bb.0:
+; POPCNT-NEXT:    psllw $7, %xmm0
+; POPCNT-NEXT:    pmovmskb %xmm0, %eax
+; POPCNT-NEXT:    popcntl %eax, %eax
+; POPCNT-NEXT:    testb $1, %al
+; POPCNT-NEXT:    sete %al
+; POPCNT-NEXT:    retq
   %i1 = bitcast <16 x i1> %x to i16
   %i2 = call i16 @llvm.ctpop.i16(i16 %i1)
   %i3 = and i16 %i2, 1
@@ -73,15 +130,23 @@ define i1 @canonical_nonparity(<16 x i1> %x) {
   ret i1 %i4
 }
 define i1 @canonical_nonparity_noncanonical_pred(<16 x i1> %x) {
-; CHECK-LABEL: canonical_nonparity_noncanonical_pred:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    psllw $7, %xmm0
-; CHECK-NEXT:    pmovmskb %xmm0, %eax
-; CHECK-NEXT:    popcntl %eax, %eax
-; CHECK-NEXT:    andl $1, %eax
-; CHECK-NEXT:    xorb $1, %al
-; CHECK-NEXT:    # kill: def $al killed $al killed $eax
-; CHECK-NEXT:    retq
+; NOPOPCNT-LABEL: canonical_nonparity_noncanonical_pred:
+; NOPOPCNT:       # %bb.0:
+; NOPOPCNT-NEXT:    psllw $7, %xmm0
+; NOPOPCNT-NEXT:    pmovmskb %xmm0, %eax
+; NOPOPCNT-NEXT:    xorb %ah, %al
+; NOPOPCNT-NEXT:    setp %al
+; NOPOPCNT-NEXT:    retq
+;
+; POPCNT-LABEL: canonical_nonparity_noncanonical_pred:
+; POPCNT:       # %bb.0:
+; POPCNT-NEXT:    psllw $7, %xmm0
+; POPCNT-NEXT:    pmovmskb %xmm0, %eax
+; POPCNT-NEXT:    popcntl %eax, %eax
+; POPCNT-NEXT:    andl $1, %eax
+; POPCNT-NEXT:    xorb $1, %al
+; POPCNT-NEXT:    # kill: def $al killed $al killed $eax
+; POPCNT-NEXT:    retq
   %i1 = bitcast <16 x i1> %x to i16
   %i2 = call i16 @llvm.ctpop.i16(i16 %i1)
   %i3 = and i16 %i2, 1