[X86] Add tests for optimizing out zero-check in is_pow2 setcc pattern; NFC

[RKSimon]

LGTM

llvm/test/CodeGen/X86/ispow2.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown | FileCheck %s --check-prefixes=CHECK,CHECK-NOBMI
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+bmi2,+bmi | FileCheck %s --check-prefixes=CHECK,CHECK-BMI2,CHECK-AVX2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vl,+avx512vpopcntdq,+bmi2,+bmi | FileCheck %s --check-prefixes=CHECK,CHECK-BMI2,CHECK-AVX512
+
+declare i32 @llvm.ctpop.i32(i32)
+declare <4 x i64> @llvm.ctpop.v4i64(<4 x i64>)
+define i1 @is_pow2_non_zero(i32 %xin) {
+; CHECK-NOBMI-LABEL: is_pow2_non_zero:
+; CHECK-NOBMI:       # %bb.0:
+; CHECK-NOBMI-NEXT:    # kill: def $edi killed $edi def $rdi
+; CHECK-NOBMI-NEXT:    orl $256, %edi # imm = 0x100
+; CHECK-NOBMI-NEXT:    leal -1(%rdi), %eax
+; CHECK-NOBMI-NEXT:    testl %eax, %edi
+; CHECK-NOBMI-NEXT:    sete %cl
+; CHECK-NOBMI-NEXT:    testl %edi, %edi
+; CHECK-NOBMI-NEXT:    setne %al
+; CHECK-NOBMI-NEXT:    andb %cl, %al
+; CHECK-NOBMI-NEXT:    retq
+;
+; CHECK-BMI2-LABEL: is_pow2_non_zero:
+; CHECK-BMI2:       # %bb.0:
+; CHECK-BMI2-NEXT:    orl $256, %edi # imm = 0x100
+; CHECK-BMI2-NEXT:    setne %cl
+; CHECK-BMI2-NEXT:    blsrl %edi, %eax
+; CHECK-BMI2-NEXT:    sete %al
+; CHECK-BMI2-NEXT:    andb %cl, %al
+; CHECK-BMI2-NEXT:    retq
+  %x = or i32 %xin, 256
+  %cnt = call i32 @llvm.ctpop.i32(i32 %x)
+  %r = icmp eq i32 %cnt, 1
+  ret i1 %r
+}
+
+define i1 @is_pow2_non_zero_x_maybe_z(i32 %x) {
+; CHECK-NOBMI-LABEL: is_pow2_non_zero_x_maybe_z:
+; CHECK-NOBMI:       # %bb.0:
+; CHECK-NOBMI-NEXT:    # kill: def $edi killed $edi def $rdi
+; CHECK-NOBMI-NEXT:    leal -1(%rdi), %eax
+; CHECK-NOBMI-NEXT:    testl %eax, %edi
+; CHECK-NOBMI-NEXT:    sete %cl
+; CHECK-NOBMI-NEXT:    testl %edi, %edi
+; CHECK-NOBMI-NEXT:    setne %al
+; CHECK-NOBMI-NEXT:    andb %cl, %al
+; CHECK-NOBMI-NEXT:    retq
+;
+; CHECK-BMI2-LABEL: is_pow2_non_zero_x_maybe_z:
+; CHECK-BMI2:       # %bb.0:
+; CHECK-BMI2-NEXT:    testl %edi, %edi
+; CHECK-BMI2-NEXT:    setne %cl
+; CHECK-BMI2-NEXT:    blsrl %edi, %eax
+; CHECK-BMI2-NEXT:    sete %al
+; CHECK-BMI2-NEXT:    andb %cl, %al
+; CHECK-BMI2-NEXT:    retq
+  %cnt = call i32 @llvm.ctpop.i32(i32 %x)
+  %r = icmp eq i32 %cnt, 1
+  ret i1 %r
+}
+
+define i1 @neither_pow2_non_zero(i32 %xin) {
+; CHECK-NOBMI-LABEL: neither_pow2_non_zero:
+; CHECK-NOBMI:       # %bb.0:
+; CHECK-NOBMI-NEXT:    # kill: def $edi killed $edi def $rdi
+; CHECK-NOBMI-NEXT:    orl $256, %edi # imm = 0x100
+; CHECK-NOBMI-NEXT:    leal -1(%rdi), %eax
+; CHECK-NOBMI-NEXT:    testl %eax, %edi
+; CHECK-NOBMI-NEXT:    setne %cl
+; CHECK-NOBMI-NEXT:    testl %edi, %edi
+; CHECK-NOBMI-NEXT:    sete %al
+; CHECK-NOBMI-NEXT:    orb %cl, %al
+; CHECK-NOBMI-NEXT:    retq
+;
+; CHECK-BMI2-LABEL: neither_pow2_non_zero:
+; CHECK-BMI2:       # %bb.0:
+; CHECK-BMI2-NEXT:    orl $256, %edi # imm = 0x100
+; CHECK-BMI2-NEXT:    sete %cl
+; CHECK-BMI2-NEXT:    blsrl %edi, %eax
+; CHECK-BMI2-NEXT:    setne %al
+; CHECK-BMI2-NEXT:    orb %cl, %al
+; CHECK-BMI2-NEXT:    retq
+  %x = or i32 %xin, 256
+  %cnt = call i32 @llvm.ctpop.i32(i32 %x)
+  %r = icmp ne i32 %cnt, 1
+  ret i1 %r
+}
+
+define <4 x i1> @is_pow2_non_zero_4xv64(<4 x i64> %xin) {
+; CHECK-NOBMI-LABEL: is_pow2_non_zero_4xv64:
+; CHECK-NOBMI:       # %bb.0:
+; CHECK-NOBMI-NEXT:    movdqa {{.*#+}} xmm2 = [256,256]
+; CHECK-NOBMI-NEXT:    por %xmm2, %xmm0
+; CHECK-NOBMI-NEXT:    por %xmm2, %xmm1
+; CHECK-NOBMI-NEXT:    pcmpeqd %xmm2, %xmm2
+; CHECK-NOBMI-NEXT:    movdqa %xmm1, %xmm3
+; CHECK-NOBMI-NEXT:    paddq %xmm2, %xmm3
+; CHECK-NOBMI-NEXT:    pand %xmm1, %xmm3
+; CHECK-NOBMI-NEXT:    pxor %xmm4, %xmm4
+; CHECK-NOBMI-NEXT:    pcmpeqd %xmm4, %xmm3
+; CHECK-NOBMI-NEXT:    pshufd {{.*#+}} xmm5 = xmm3[1,0,3,2]
+; CHECK-NOBMI-NEXT:    pand %xmm3, %xmm5
+; CHECK-NOBMI-NEXT:    pcmpeqd %xmm4, %xmm1
+; CHECK-NOBMI-NEXT:    pshufd {{.*#+}} xmm3 = xmm1[1,0,3,2]
+; CHECK-NOBMI-NEXT:    pand %xmm1, %xmm3
+; CHECK-NOBMI-NEXT:    pandn %xmm5, %xmm3
+; CHECK-NOBMI-NEXT:    paddq %xmm0, %xmm2
+; CHECK-NOBMI-NEXT:    pand %xmm0, %xmm2
+; CHECK-NOBMI-NEXT:    pcmpeqd %xmm4, %xmm2
+; CHECK-NOBMI-NEXT:    pshufd {{.*#+}} xmm1 = xmm2[1,0,3,2]
+; CHECK-NOBMI-NEXT:    pand %xmm2, %xmm1
+; CHECK-NOBMI-NEXT:    pcmpeqd %xmm4, %xmm0
+; CHECK-NOBMI-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[1,0,3,2]
+; CHECK-NOBMI-NEXT:    pand %xmm2, %xmm0
+; CHECK-NOBMI-NEXT:    pandn %xmm1, %xmm0
+; CHECK-NOBMI-NEXT:    packssdw %xmm3, %xmm0
+; CHECK-NOBMI-NEXT:    retq
+;
+; CHECK-AVX2-LABEL: is_pow2_non_zero_4xv64:
+; CHECK-AVX2:       # %bb.0:
+; CHECK-AVX2-NEXT:    vpbroadcastq {{.*#+}} ymm1 = [256,256,256,256]
+; CHECK-AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
+; CHECK-AVX2-NEXT:    vpcmpeqd %ymm1, %ymm1, %ymm1
+; CHECK-AVX2-NEXT:    vpaddq %ymm1, %ymm0, %ymm1
+; CHECK-AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0
+; CHECK-AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; CHECK-AVX2-NEXT:    vpcmpeqq %ymm1, %ymm0, %ymm0
+; CHECK-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; CHECK-AVX2-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
+; CHECK-AVX2-NEXT:    vzeroupper
+; CHECK-AVX2-NEXT:    retq
+;
+; CHECK-AVX512-LABEL: is_pow2_non_zero_4xv64:
+; CHECK-AVX512:       # %bb.0:
+; CHECK-AVX512-NEXT:    vporq {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %ymm0, %ymm0
+; CHECK-AVX512-NEXT:    vpopcntq %ymm0, %ymm0
+; CHECK-AVX512-NEXT:    vpcmpeqq {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %ymm0, %k1
+; CHECK-AVX512-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
+; CHECK-AVX512-NEXT:    vmovdqa32 %xmm0, %xmm0 {%k1} {z}
+; CHECK-AVX512-NEXT:    vzeroupper
+; CHECK-AVX512-NEXT:    retq
+  %x = or <4 x i64> %xin, <i64 256, i64 256, i64 256, i64 256>
+  %cnt = call <4 x i64> @llvm.ctpop.v4i64(<4 x i64> %x)
+  %r = icmp eq <4 x i64> %cnt, <i64 1, i64 1, i64 1, i64 1>
+  ret <4 x i1> %r
+}
+
+define <4 x i1> @neither_pow2_non_zero_4xv64(<4 x i64> %xin) {
+; CHECK-NOBMI-LABEL: neither_pow2_non_zero_4xv64:
+; CHECK-NOBMI:       # %bb.0:
+; CHECK-NOBMI-NEXT:    movdqa {{.*#+}} xmm2 = [256,256]
+; CHECK-NOBMI-NEXT:    por %xmm2, %xmm0
+; CHECK-NOBMI-NEXT:    por %xmm2, %xmm1
+; CHECK-NOBMI-NEXT:    pcmpeqd %xmm2, %xmm2
+; CHECK-NOBMI-NEXT:    movdqa %xmm1, %xmm3
+; CHECK-NOBMI-NEXT:    paddq %xmm2, %xmm3
+; CHECK-NOBMI-NEXT:    pand %xmm1, %xmm3
+; CHECK-NOBMI-NEXT:    pxor %xmm4, %xmm4
+; CHECK-NOBMI-NEXT:    pcmpeqd %xmm4, %xmm3
+; CHECK-NOBMI-NEXT:    pshufd {{.*#+}} xmm5 = xmm3[1,0,3,2]
+; CHECK-NOBMI-NEXT:    pand %xmm3, %xmm5
+; CHECK-NOBMI-NEXT:    pxor %xmm2, %xmm5
+; CHECK-NOBMI-NEXT:    pcmpeqd %xmm4, %xmm1
+; CHECK-NOBMI-NEXT:    pshufd {{.*#+}} xmm3 = xmm1[1,0,3,2]
+; CHECK-NOBMI-NEXT:    pand %xmm1, %xmm3
+; CHECK-NOBMI-NEXT:    por %xmm5, %xmm3
+; CHECK-NOBMI-NEXT:    movdqa %xmm0, %xmm1
+; CHECK-NOBMI-NEXT:    paddq %xmm2, %xmm1
+; CHECK-NOBMI-NEXT:    pand %xmm0, %xmm1
+; CHECK-NOBMI-NEXT:    pcmpeqd %xmm4, %xmm1
+; CHECK-NOBMI-NEXT:    pshufd {{.*#+}} xmm5 = xmm1[1,0,3,2]
+; CHECK-NOBMI-NEXT:    pand %xmm1, %xmm5
+; CHECK-NOBMI-NEXT:    pxor %xmm2, %xmm5
+; CHECK-NOBMI-NEXT:    pcmpeqd %xmm4, %xmm0
+; CHECK-NOBMI-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,0,3,2]
+; CHECK-NOBMI-NEXT:    pand %xmm1, %xmm0
+; CHECK-NOBMI-NEXT:    por %xmm5, %xmm0
+; CHECK-NOBMI-NEXT:    packssdw %xmm3, %xmm0
+; CHECK-NOBMI-NEXT:    retq
+;
+; CHECK-AVX2-LABEL: neither_pow2_non_zero_4xv64:
+; CHECK-AVX2:       # %bb.0:
+; CHECK-AVX2-NEXT:    vpbroadcastq {{.*#+}} ymm1 = [256,256,256,256]
+; CHECK-AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
+; CHECK-AVX2-NEXT:    vpcmpeqd %ymm1, %ymm1, %ymm1
+; CHECK-AVX2-NEXT:    vpaddq %ymm1, %ymm0, %ymm2
+; CHECK-AVX2-NEXT:    vpand %ymm2, %ymm0, %ymm0
+; CHECK-AVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; CHECK-AVX2-NEXT:    vpcmpeqq %ymm2, %ymm0, %ymm0
+; CHECK-AVX2-NEXT:    vpxor %ymm1, %ymm0, %ymm0
+; CHECK-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; CHECK-AVX2-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
+; CHECK-AVX2-NEXT:    vzeroupper
+; CHECK-AVX2-NEXT:    retq
+;
+; CHECK-AVX512-LABEL: neither_pow2_non_zero_4xv64:
+; CHECK-AVX512:       # %bb.0:
+; CHECK-AVX512-NEXT:    vporq {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %ymm0, %ymm0
+; CHECK-AVX512-NEXT:    vpopcntq %ymm0, %ymm0
+; CHECK-AVX512-NEXT:    vpcmpneqq {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %ymm0, %k1
+; CHECK-AVX512-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
+; CHECK-AVX512-NEXT:    vmovdqa32 %xmm0, %xmm0 {%k1} {z}
+; CHECK-AVX512-NEXT:    vzeroupper
+; CHECK-AVX512-NEXT:    retq
+  %x = or <4 x i64> %xin, <i64 256, i64 256, i64 256, i64 256>
+  %cnt = call <4 x i64> @llvm.ctpop.v4i64(<4 x i64> %x)
+  %r = icmp ne <4 x i64> %cnt, <i64 1, i64 1, i64 1, i64 1>
+  ret <4 x i1> %r
+}
+
+define <4 x i1> @neither_pow2_non_zero_4xv64_x_maybe_z(<4 x i64> %x) {
+; CHECK-NOBMI-LABEL: neither_pow2_non_zero_4xv64_x_maybe_z:
+; CHECK-NOBMI:       # %bb.0:
+; CHECK-NOBMI-NEXT:    pxor %xmm2, %xmm2
+; CHECK-NOBMI-NEXT:    pcmpeqd %xmm3, %xmm3
+; CHECK-NOBMI-NEXT:    movdqa %xmm1, %xmm4
+; CHECK-NOBMI-NEXT:    paddq %xmm3, %xmm4
+; CHECK-NOBMI-NEXT:    pand %xmm1, %xmm4
+; CHECK-NOBMI-NEXT:    pcmpeqd %xmm2, %xmm1
+; CHECK-NOBMI-NEXT:    pshufd {{.*#+}} xmm5 = xmm1[1,0,3,2]
+; CHECK-NOBMI-NEXT:    pand %xmm1, %xmm5
+; CHECK-NOBMI-NEXT:    pcmpeqd %xmm2, %xmm4
+; CHECK-NOBMI-NEXT:    pshufd {{.*#+}} xmm1 = xmm4[1,0,3,2]
+; CHECK-NOBMI-NEXT:    pand %xmm4, %xmm1
+; CHECK-NOBMI-NEXT:    pxor %xmm3, %xmm1
+; CHECK-NOBMI-NEXT:    por %xmm5, %xmm1
+; CHECK-NOBMI-NEXT:    movdqa %xmm0, %xmm4
+; CHECK-NOBMI-NEXT:    pcmpeqd %xmm2, %xmm4
+; CHECK-NOBMI-NEXT:    pshufd {{.*#+}} xmm5 = xmm4[1,0,3,2]
+; CHECK-NOBMI-NEXT:    pand %xmm4, %xmm5
+; CHECK-NOBMI-NEXT:    movdqa %xmm0, %xmm4
+; CHECK-NOBMI-NEXT:    paddq %xmm3, %xmm4
+; CHECK-NOBMI-NEXT:    pand %xmm4, %xmm0
+; CHECK-NOBMI-NEXT:    pcmpeqd %xmm2, %xmm0
+; CHECK-NOBMI-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[1,0,3,2]
+; CHECK-NOBMI-NEXT:    pand %xmm2, %xmm0
+; CHECK-NOBMI-NEXT:    pxor %xmm3, %xmm0
+; CHECK-NOBMI-NEXT:    por %xmm5, %xmm0
+; CHECK-NOBMI-NEXT:    packssdw %xmm1, %xmm0
+; CHECK-NOBMI-NEXT:    retq
+;
+; CHECK-AVX2-LABEL: neither_pow2_non_zero_4xv64_x_maybe_z:
+; CHECK-AVX2:       # %bb.0:
+; CHECK-AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; CHECK-AVX2-NEXT:    vpcmpeqq %ymm1, %ymm0, %ymm2
+; CHECK-AVX2-NEXT:    vpcmpeqd %ymm3, %ymm3, %ymm3
+; CHECK-AVX2-NEXT:    vpaddq %ymm3, %ymm0, %ymm4
+; CHECK-AVX2-NEXT:    vpand %ymm4, %ymm0, %ymm0
+; CHECK-AVX2-NEXT:    vpcmpeqq %ymm1, %ymm0, %ymm0
+; CHECK-AVX2-NEXT:    vpxor %ymm3, %ymm0, %ymm0
+; CHECK-AVX2-NEXT:    vpor %ymm0, %ymm2, %ymm0
+; CHECK-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; CHECK-AVX2-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
+; CHECK-AVX2-NEXT:    vzeroupper
+; CHECK-AVX2-NEXT:    retq
+;
+; CHECK-AVX512-LABEL: neither_pow2_non_zero_4xv64_x_maybe_z:
+; CHECK-AVX512:       # %bb.0:
+; CHECK-AVX512-NEXT:    vpopcntq %ymm0, %ymm0
+; CHECK-AVX512-NEXT:    vpcmpneqq {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %ymm0, %k1
+; CHECK-AVX512-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
+; CHECK-AVX512-NEXT:    vmovdqa32 %xmm0, %xmm0 {%k1} {z}
+; CHECK-AVX512-NEXT:    vzeroupper
+; CHECK-AVX512-NEXT:    retq
+  %cnt = call <4 x i64> @llvm.ctpop.v4i64(<4 x i64> %x)
+  %r = icmp ne <4 x i64> %cnt, <i64 1, i64 1, i64 1, i64 1>
+  ret <4 x i1> %r
+}
+;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
+; CHECK: {{.*}}