llvm/test/CodeGen/X86/combine-movmsk.ll

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefixes=SSE,SSE2
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.2 | FileCheck %s --check-prefixes=SSE,SSE42
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=AVX
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=AVX
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+prefer-movmsk-over-vtest | FileCheck %s --check-prefixes=ADL

declare i32 @llvm.x86.sse.movmsk.ps(<4 x float>)
declare i32 @llvm.x86.sse2.movmsk.pd(<2 x double>)
declare i32 @llvm.x86.sse2.pmovmskb.128(<16 x i8>)

; Use widest possible vector for movmsk comparisons (PR37087)

define i1 @movmskps_noneof_bitcast_v2f64(<2 x double> %a0) {
; SSE-LABEL: movmskps_noneof_bitcast_v2f64:
; SSE:       # %bb.0:
; SSE-NEXT:    xorpd %xmm1, %xmm1
; SSE-NEXT:    cmpeqpd %xmm0, %xmm1
; SSE-NEXT:    movmskpd %xmm1, %eax
; SSE-NEXT:    testl %eax, %eax
; SSE-NEXT:    sete %al
; SSE-NEXT:    retq
;
; AVX-LABEL: movmskps_noneof_bitcast_v2f64:
; AVX:       # %bb.0:
; AVX-NEXT:    vxorpd %xmm1, %xmm1, %xmm1
; AVX-NEXT:    vcmpeqpd %xmm0, %xmm1, %xmm0
; AVX-NEXT:    vtestpd %xmm0, %xmm0
; AVX-NEXT:    sete %al
; AVX-NEXT:    retq
;
; ADL-LABEL: movmskps_noneof_bitcast_v2f64:
; ADL:       # %bb.0:
; ADL-NEXT:    vxorpd %xmm1, %xmm1, %xmm1
; ADL-NEXT:    vcmpeqpd %xmm0, %xmm1, %xmm0
; ADL-NEXT:    vmovmskpd %xmm0, %eax
; ADL-NEXT:    testl %eax, %eax
; ADL-NEXT:    sete %al
; ADL-NEXT:    retq
  %1 = fcmp oeq <2 x double> zeroinitializer, %a0
  %2 = sext <2 x i1> %1 to <2 x i64>
  %3 = bitcast <2 x i64> %2 to <4 x float>
  %4 = tail call i32 @llvm.x86.sse.movmsk.ps(<4 x float> %3)
  %5 = icmp eq i32 %4, 0
  ret i1 %5
}

define i1 @movmskps_allof_bitcast_v2f64(<2 x double> %a0) {
; SSE-LABEL: movmskps_allof_bitcast_v2f64:
; SSE:       # %bb.0:
; SSE-NEXT:    xorpd %xmm1, %xmm1
; SSE-NEXT:    cmpeqpd %xmm0, %xmm1
; SSE-NEXT:    movmskpd %xmm1, %eax
; SSE-NEXT:    cmpl $3, %eax
; SSE-NEXT:    sete %al
; SSE-NEXT:    retq
;
; AVX-LABEL: movmskps_allof_bitcast_v2f64:
; AVX:       # %bb.0:
; AVX-NEXT:    vxorpd %xmm1, %xmm1, %xmm1
; AVX-NEXT:    vcmpeqpd %xmm0, %xmm1, %xmm0
; AVX-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX-NEXT:    vtestpd %xmm1, %xmm0
; AVX-NEXT:    setb %al
; AVX-NEXT:    retq
;
; ADL-LABEL: movmskps_allof_bitcast_v2f64:
; ADL:       # %bb.0:
; ADL-NEXT:    vxorpd %xmm1, %xmm1, %xmm1
; ADL-NEXT:    vcmpeqpd %xmm0, %xmm1, %xmm0
; ADL-NEXT:    vmovmskpd %xmm0, %eax
; ADL-NEXT:    cmpl $3, %eax
; ADL-NEXT:    sete %al
; ADL-NEXT:    retq
  %1 = fcmp oeq <2 x double> zeroinitializer, %a0
  %2 = sext <2 x i1> %1 to <2 x i64>
  %3 = bitcast <2 x i64> %2 to <4 x float>
  %4 = tail call i32 @llvm.x86.sse.movmsk.ps(<4 x float> %3)
  %5 = icmp eq i32 %4, 15
  ret i1 %5
}

define i1 @pmovmskb_noneof_bitcast_v2i64(<2 x i64> %a0) {
; SSE2-LABEL: pmovmskb_noneof_bitcast_v2i64:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
; SSE2-NEXT:    movmskps %xmm0, %eax
; SSE2-NEXT:    testl %eax, %eax
; SSE2-NEXT:    sete %al
; SSE2-NEXT:    retq
;
; SSE42-LABEL: pmovmskb_noneof_bitcast_v2i64:
; SSE42:       # %bb.0:
; SSE42-NEXT:    movmskpd %xmm0, %eax
; SSE42-NEXT:    testl %eax, %eax
; SSE42-NEXT:    sete %al
; SSE42-NEXT:    retq
;
; AVX-LABEL: pmovmskb_noneof_bitcast_v2i64:
; AVX:       # %bb.0:
; AVX-NEXT:    vtestpd %xmm0, %xmm0
; AVX-NEXT:    sete %al
; AVX-NEXT:    retq
;
; ADL-LABEL: pmovmskb_noneof_bitcast_v2i64:
; ADL:       # %bb.0:
; ADL-NEXT:    vmovmskpd %xmm0, %eax
; ADL-NEXT:    testl %eax, %eax
; ADL-NEXT:    sete %al
; ADL-NEXT:    retq
  %1 = icmp sgt <2 x i64> zeroinitializer, %a0
  %2 = sext <2 x i1> %1 to <2 x i64>
  %3 = bitcast <2 x i64> %2 to <16 x i8>
  %4 = tail call i32 @llvm.x86.sse2.pmovmskb.128(<16 x i8> %3)
  %5 = icmp eq i32 %4, 0
  ret i1 %5
}

define i1 @pmovmskb_allof_bitcast_v2i64(<2 x i64> %a0) {
; SSE2-LABEL: pmovmskb_allof_bitcast_v2i64:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
; SSE2-NEXT:    movmskps %xmm0, %eax
; SSE2-NEXT:    cmpl $15, %eax
; SSE2-NEXT:    sete %al
; SSE2-NEXT:    retq
;
; SSE42-LABEL: pmovmskb_allof_bitcast_v2i64:
; SSE42:       # %bb.0:
; SSE42-NEXT:    movmskpd %xmm0, %eax
; SSE42-NEXT:    cmpl $3, %eax
; SSE42-NEXT:    sete %al
; SSE42-NEXT:    retq
;
; AVX-LABEL: pmovmskb_allof_bitcast_v2i64:
; AVX:       # %bb.0:
; AVX-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX-NEXT:    vtestpd %xmm1, %xmm0
; AVX-NEXT:    setb %al
; AVX-NEXT:    retq
;
; ADL-LABEL: pmovmskb_allof_bitcast_v2i64:
; ADL:       # %bb.0:
; ADL-NEXT:    vmovmskpd %xmm0, %eax
; ADL-NEXT:    cmpl $3, %eax
; ADL-NEXT:    sete %al
; ADL-NEXT:    retq
  %1 = icmp sgt <2 x i64> zeroinitializer, %a0
  %2 = sext <2 x i1> %1 to <2 x i64>
  %3 = bitcast <2 x i64> %2 to <16 x i8>
  %4 = tail call i32 @llvm.x86.sse2.pmovmskb.128(<16 x i8> %3)
  %5 = icmp eq i32 %4, 65535
  ret i1 %5
}

define i1 @pmovmskb_noneof_bitcast_v4f32(<4 x float> %a0) {
; SSE-LABEL: pmovmskb_noneof_bitcast_v4f32:
; SSE:       # %bb.0:
; SSE-NEXT:    xorps %xmm1, %xmm1
; SSE-NEXT:    cmpeqps %xmm0, %xmm1
; SSE-NEXT:    movmskps %xmm1, %eax
; SSE-NEXT:    testl %eax, %eax
; SSE-NEXT:    sete %al
; SSE-NEXT:    retq
;
; AVX-LABEL: pmovmskb_noneof_bitcast_v4f32:
; AVX:       # %bb.0:
; AVX-NEXT:    vxorps %xmm1, %xmm1, %xmm1
; AVX-NEXT:    vcmpeqps %xmm1, %xmm0, %xmm0
; AVX-NEXT:    vtestps %xmm0, %xmm0
; AVX-NEXT:    sete %al
; AVX-NEXT:    retq
;
; ADL-LABEL: pmovmskb_noneof_bitcast_v4f32:
; ADL:       # %bb.0:
; ADL-NEXT:    vxorps %xmm1, %xmm1, %xmm1
; ADL-NEXT:    vcmpeqps %xmm1, %xmm0, %xmm0
; ADL-NEXT:    vmovmskps %xmm0, %eax
; ADL-NEXT:    testl %eax, %eax
; ADL-NEXT:    sete %al
; ADL-NEXT:    retq
  %1 = fcmp oeq <4 x float> %a0, zeroinitializer
  %2 = sext <4 x i1> %1 to <4 x i32>
  %3 = bitcast <4 x i32> %2 to <16 x i8>
  %4 = tail call i32 @llvm.x86.sse2.pmovmskb.128(<16 x i8> %3)
  %5 = icmp eq i32 %4, 0
  ret i1 %5
}

define i1 @pmovmskb_allof_bitcast_v4f32(<4 x float> %a0) {
; SSE-LABEL: pmovmskb_allof_bitcast_v4f32:
; SSE:       # %bb.0:
; SSE-NEXT:    xorps %xmm1, %xmm1
; SSE-NEXT:    cmpeqps %xmm0, %xmm1
; SSE-NEXT:    movmskps %xmm1, %eax
; SSE-NEXT:    cmpl $15, %eax
; SSE-NEXT:    sete %al
; SSE-NEXT:    retq
;
; AVX-LABEL: pmovmskb_allof_bitcast_v4f32:
; AVX:       # %bb.0:
; AVX-NEXT:    vxorps %xmm1, %xmm1, %xmm1
; AVX-NEXT:    vcmpeqps %xmm1, %xmm0, %xmm0
; AVX-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX-NEXT:    vtestps %xmm1, %xmm0
; AVX-NEXT:    setb %al
; AVX-NEXT:    retq
;
; ADL-LABEL: pmovmskb_allof_bitcast_v4f32:
; ADL:       # %bb.0:
; ADL-NEXT:    vxorps %xmm1, %xmm1, %xmm1
; ADL-NEXT:    vcmpeqps %xmm1, %xmm0, %xmm0
; ADL-NEXT:    vmovmskps %xmm0, %eax
; ADL-NEXT:    cmpl $15, %eax
; ADL-NEXT:    sete %al
; ADL-NEXT:    retq
  %1 = fcmp oeq <4 x float> %a0, zeroinitializer
  %2 = sext <4 x i1> %1 to <4 x i32>
  %3 = bitcast <4 x i32> %2 to <16 x i8>
  %4 = tail call i32 @llvm.x86.sse2.pmovmskb.128(<16 x i8> %3)
  %5 = icmp eq i32 %4, 65535
  ret i1 %5
}

; MOVMSK(ICMP_SGT(X,-1)) -> NOT(MOVMSK(X)))
define i1 @movmskps_allof_v4i32_positive(<4 x i32> %a0) {
; SSE-LABEL: movmskps_allof_v4i32_positive:
; SSE:       # %bb.0:
; SSE-NEXT:    movmskps %xmm0, %eax
; SSE-NEXT:    xorl $15, %eax
; SSE-NEXT:    cmpl $15, %eax
; SSE-NEXT:    sete %al
; SSE-NEXT:    retq
;
; AVX-LABEL: movmskps_allof_v4i32_positive:
; AVX:       # %bb.0:
; AVX-NEXT:    vmovmskps %xmm0, %eax
; AVX-NEXT:    xorl $15, %eax
; AVX-NEXT:    cmpl $15, %eax
; AVX-NEXT:    sete %al
; AVX-NEXT:    retq
;
; ADL-LABEL: movmskps_allof_v4i32_positive:
; ADL:       # %bb.0:
; ADL-NEXT:    vmovmskps %xmm0, %eax
; ADL-NEXT:    xorl $15, %eax
; ADL-NEXT:    cmpl $15, %eax
; ADL-NEXT:    sete %al
; ADL-NEXT:    retq
  %1 = icmp sgt <4 x i32> %a0, <i32 -1, i32 -1, i32 -1, i32 -1>
  %2 = sext <4 x i1> %1 to <4 x i32>
  %3 = bitcast <4 x i32> %2 to <4 x float>
  %4 = tail call i32 @llvm.x86.sse.movmsk.ps(<4 x float> %3)
  %5 = icmp eq i32 %4, 15
  ret i1 %5
}

define i1 @pmovmskb_noneof_v16i8_positive(<16 x i8> %a0) {
; SSE-LABEL: pmovmskb_noneof_v16i8_positive:
; SSE:       # %bb.0:
; SSE-NEXT:    pmovmskb %xmm0, %eax
; SSE-NEXT:    xorl $65535, %eax # imm = 0xFFFF
; SSE-NEXT:    sete %al
; SSE-NEXT:    retq
;
; AVX-LABEL: pmovmskb_noneof_v16i8_positive:
; AVX:       # %bb.0:
; AVX-NEXT:    vpmovmskb %xmm0, %eax
; AVX-NEXT:    xorl $65535, %eax # imm = 0xFFFF
; AVX-NEXT:    sete %al
; AVX-NEXT:    retq
;
; ADL-LABEL: pmovmskb_noneof_v16i8_positive:
; ADL:       # %bb.0:
; ADL-NEXT:    vpmovmskb %xmm0, %eax
; ADL-NEXT:    xorl $65535, %eax # imm = 0xFFFF
; ADL-NEXT:    sete %al
; ADL-NEXT:    retq
  %1 = icmp sgt <16 x i8> %a0, <i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1>
  %2 = sext <16 x i1> %1 to <16 x i8>
  %3 = tail call i32 @llvm.x86.sse2.pmovmskb.128(<16 x i8> %2)
  %4 = icmp eq i32 %3, 0
  ret i1 %4
}

; MOVMSK(CMPEQ(AND(X,C1),0)) -> MOVMSK(NOT(SHL(X,C2)))
define i32 @movmskpd_pow2_mask(<2 x i64> %a0) {
; SSE2-LABEL: movmskpd_pow2_mask:
; SSE2:       # %bb.0:
; SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
; SSE2-NEXT:    pxor %xmm1, %xmm1
; SSE2-NEXT:    pcmpeqd %xmm0, %xmm1
; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,0,3,2]
; SSE2-NEXT:    pand %xmm1, %xmm0
; SSE2-NEXT:    movmskpd %xmm0, %eax
; SSE2-NEXT:    retq
;
; SSE42-LABEL: movmskpd_pow2_mask:
; SSE42:       # %bb.0:
; SSE42-NEXT:    movmskpd %xmm0, %eax
; SSE42-NEXT:    xorl $3, %eax
; SSE42-NEXT:    retq
;
; AVX-LABEL: movmskpd_pow2_mask:
; AVX:       # %bb.0:
; AVX-NEXT:    vmovmskpd %xmm0, %eax
; AVX-NEXT:    xorl $3, %eax
; AVX-NEXT:    retq
;
; ADL-LABEL: movmskpd_pow2_mask:
; ADL:       # %bb.0:
; ADL-NEXT:    vmovmskpd %xmm0, %eax
; ADL-NEXT:    xorl $3, %eax
; ADL-NEXT:    retq
  %1 = and <2 x i64> %a0, <i64 -9223372036854775808, i64 -9223372036854775808>
  %2 = icmp eq <2 x i64> %1, zeroinitializer
  %3 = sext <2 x i1> %2 to <2 x i64>
  %4 = bitcast <2 x i64> %3 to <2 x double>
  %5 = tail call i32 @llvm.x86.sse2.movmsk.pd(<2 x double> %4)
  ret i32 %5
}

define i32 @movmskps_pow2_mask(<4 x i32> %a0) {
; SSE-LABEL: movmskps_pow2_mask:
; SSE:       # %bb.0:
; SSE-NEXT:    pslld $29, %xmm0
; SSE-NEXT:    movmskps %xmm0, %eax
; SSE-NEXT:    xorl $15, %eax
; SSE-NEXT:    retq
;
; AVX-LABEL: movmskps_pow2_mask:
; AVX:       # %bb.0:
; AVX-NEXT:    vpslld $29, %xmm0, %xmm0
; AVX-NEXT:    vmovmskps %xmm0, %eax
; AVX-NEXT:    xorl $15, %eax
; AVX-NEXT:    retq
;
; ADL-LABEL: movmskps_pow2_mask:
; ADL:       # %bb.0:
; ADL-NEXT:    vpslld $29, %xmm0, %xmm0
; ADL-NEXT:    vmovmskps %xmm0, %eax
; ADL-NEXT:    xorl $15, %eax
; ADL-NEXT:    retq
  %1 = and <4 x i32> %a0, <i32 4, i32 4, i32 4, i32 4>
  %2 = icmp eq <4 x i32> %1, zeroinitializer
  %3 = sext <4 x i1> %2 to <4 x i32>
  %4 = bitcast <4 x i32> %3 to <4 x float>
  %5 = tail call i32 @llvm.x86.sse.movmsk.ps(<4 x float> %4)
  ret i32 %5
}

define i32 @pmovmskb_pow2_mask(<16 x i8> %a0) {
; SSE-LABEL: pmovmskb_pow2_mask:
; SSE:       # %bb.0:
; SSE-NEXT:    psllw $7, %xmm0
; SSE-NEXT:    pmovmskb %xmm0, %eax
; SSE-NEXT:    xorl $65535, %eax # imm = 0xFFFF
; SSE-NEXT:    retq
;
; AVX-LABEL: pmovmskb_pow2_mask:
; AVX:       # %bb.0:
; AVX-NEXT:    vpsllw $7, %xmm0, %xmm0
; AVX-NEXT:    vpmovmskb %xmm0, %eax
; AVX-NEXT:    xorl $65535, %eax # imm = 0xFFFF
; AVX-NEXT:    retq
;
; ADL-LABEL: pmovmskb_pow2_mask:
; ADL:       # %bb.0:
; ADL-NEXT:    vpsllw $7, %xmm0, %xmm0
; ADL-NEXT:    vpmovmskb %xmm0, %eax
; ADL-NEXT:    xorl $65535, %eax # imm = 0xFFFF
; ADL-NEXT:    retq
  %1 = and <16 x i8> %a0, <i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1>
  %2 = icmp eq <16 x i8> %1, zeroinitializer
  %3 = sext <16 x i1> %2 to <16 x i8>
  %4 = tail call i32 @llvm.x86.sse2.pmovmskb.128(<16 x i8> %3)
  ret i32 %4
}

; AND(MOVMSK(X),MOVMSK(Y)) -> MOVMSK(AND(X,Y))
; XOR(MOVMSK(X),MOVMSK(Y)) -> MOVMSK(XOR(X,Y))
; OR(MOVMSK(X),MOVMSK(Y)) -> MOVMSK(OR(X,Y))
; if the elements are the same width.

define i32 @and_movmskpd_movmskpd(<2 x double> %a0, <2 x i64> %a1) {
; SSE-LABEL: and_movmskpd_movmskpd:
; SSE:       # %bb.0:
; SSE-NEXT:    xorpd %xmm2, %xmm2
; SSE-NEXT:    cmpeqpd %xmm0, %xmm2
; SSE-NEXT:    andpd %xmm1, %xmm2
; SSE-NEXT:    movmskpd %xmm2, %eax
; SSE-NEXT:    retq
;
; AVX-LABEL: and_movmskpd_movmskpd:
; AVX:       # %bb.0:
; AVX-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; AVX-NEXT:    vcmpeqpd %xmm0, %xmm2, %xmm0
; AVX-NEXT:    vandpd %xmm1, %xmm0, %xmm0
; AVX-NEXT:    vmovmskpd %xmm0, %eax
; AVX-NEXT:    retq
;
; ADL-LABEL: and_movmskpd_movmskpd:
; ADL:       # %bb.0:
; ADL-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; ADL-NEXT:    vcmpeqpd %xmm0, %xmm2, %xmm0
; ADL-NEXT:    vandpd %xmm1, %xmm0, %xmm0
; ADL-NEXT:    vmovmskpd %xmm0, %eax
; ADL-NEXT:    retq
  %1 = fcmp oeq <2 x double> zeroinitializer, %a0
  %2 = sext <2 x i1> %1 to <2 x i64>
  %3 = bitcast <2 x i64> %2 to <2 x double>
  %4 = tail call i32 @llvm.x86.sse2.movmsk.pd(<2 x double> %3)
  %5 = icmp sgt <2 x i64> zeroinitializer, %a1
  %6 = bitcast <2 x i1> %5 to i2
  %7 = zext i2 %6 to i32
  %8 = and i32 %4, %7
  ret i32 %8
}

define i32 @xor_movmskps_movmskps(<4 x float> %a0, <4 x i32> %a1) {
; SSE-LABEL: xor_movmskps_movmskps:
; SSE:       # %bb.0:
; SSE-NEXT:    xorps %xmm2, %xmm2
; SSE-NEXT:    cmpeqps %xmm0, %xmm2
; SSE-NEXT:    xorps %xmm1, %xmm2
; SSE-NEXT:    movmskps %xmm2, %eax
; SSE-NEXT:    retq
;
; AVX-LABEL: xor_movmskps_movmskps:
; AVX:       # %bb.0:
; AVX-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; AVX-NEXT:    vcmpeqps %xmm0, %xmm2, %xmm0
; AVX-NEXT:    vxorps %xmm1, %xmm0, %xmm0
; AVX-NEXT:    vmovmskps %xmm0, %eax
; AVX-NEXT:    retq
;
; ADL-LABEL: xor_movmskps_movmskps:
; ADL:       # %bb.0:
; ADL-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; ADL-NEXT:    vcmpeqps %xmm0, %xmm2, %xmm0
; ADL-NEXT:    vxorps %xmm1, %xmm0, %xmm0
; ADL-NEXT:    vmovmskps %xmm0, %eax
; ADL-NEXT:    retq
  %1 = fcmp oeq <4 x float> zeroinitializer, %a0
  %2 = sext <4 x i1> %1 to <4 x i32>
  %3 = bitcast <4 x i32> %2 to <4 x float>
  %4 = tail call i32 @llvm.x86.sse.movmsk.ps(<4 x float> %3)
  %5 = ashr <4 x i32> %a1, <i32 31, i32 31, i32 31, i32 31>
  %6 = bitcast <4 x i32> %5 to <4 x float>
  %7 = tail call i32 @llvm.x86.sse.movmsk.ps(<4 x float> %6)
  %8 = xor i32 %4, %7
  ret i32 %8
}

define i32 @or_pmovmskb_pmovmskb(<16 x i8> %a0, <8 x i16> %a1) {
; SSE-LABEL: or_pmovmskb_pmovmskb:
; SSE:       # %bb.0:
; SSE-NEXT:    pxor %xmm2, %xmm2
; SSE-NEXT:    pcmpeqb %xmm0, %xmm2
; SSE-NEXT:    psraw $15, %xmm1
; SSE-NEXT:    por %xmm2, %xmm1
; SSE-NEXT:    pmovmskb %xmm1, %eax
; SSE-NEXT:    retq
;
; AVX-LABEL: or_pmovmskb_pmovmskb:
; AVX:       # %bb.0:
; AVX-NEXT:    vpxor %xmm2, %xmm2, %xmm2
; AVX-NEXT:    vpcmpeqb %xmm2, %xmm0, %xmm0
; AVX-NEXT:    vpsraw $15, %xmm1, %xmm1
; AVX-NEXT:    vpor %xmm1, %xmm0, %xmm0
; AVX-NEXT:    vpmovmskb %xmm0, %eax
; AVX-NEXT:    retq
;
; ADL-LABEL: or_pmovmskb_pmovmskb:
; ADL:       # %bb.0:
; ADL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
; ADL-NEXT:    vpcmpeqb %xmm2, %xmm0, %xmm0
; ADL-NEXT:    vpsraw $15, %xmm1, %xmm1
; ADL-NEXT:    vpor %xmm1, %xmm0, %xmm0
; ADL-NEXT:    vpmovmskb %xmm0, %eax
; ADL-NEXT:    retq
  %1 = icmp eq <16 x i8> zeroinitializer, %a0
  %2 = sext <16 x i1> %1 to <16 x i8>
  %3 = tail call i32 @llvm.x86.sse2.pmovmskb.128(<16 x i8> %2)
  %4 = ashr <8 x i16> %a1, <i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15>
  %5 = bitcast <8 x i16> %4 to <16 x i8>
  %6 = tail call i32 @llvm.x86.sse2.pmovmskb.128(<16 x i8> %5)
  %7 = or i32 %3, %6
  ret i32 %7
}

; FREEZE(MOVMSK(X)) -> MOVMSK(FREEZE(X))
define i32 @movmskps_freeze(<4 x i32> %a0) {
; SSE-LABEL: movmskps_freeze:
; SSE:       # %bb.0:
; SSE-NEXT:    movmskps %xmm0, %eax
; SSE-NEXT:    retq
;
; AVX-LABEL: movmskps_freeze:
; AVX:       # %bb.0:
; AVX-NEXT:    vmovmskps %xmm0, %eax
; AVX-NEXT:    retq
;
; ADL-LABEL: movmskps_freeze:
; ADL:       # %bb.0:
; ADL-NEXT:    vmovmskps %xmm0, %eax
; ADL-NEXT:    retq
  %1 = icmp slt <4 x i32> %a0, zeroinitializer
  %2 = sext <4 x i1> %1 to <4 x i32>
  %3 = bitcast <4 x i32> %2 to <4 x float>
  %4 = tail call i32 @llvm.x86.sse.movmsk.ps(<4 x float> %3)
  %5 = freeze i32 %4
  %6 = and i32 %5, 15
  ret i32 %6
}

; We can't fold to ptest if we're not checking every pcmpeq result
define i32 @movmskps_ptest_numelts_mismatch(<16 x i8> %a0) {
; SSE-LABEL: movmskps_ptest_numelts_mismatch:
; SSE:       # %bb.0:
; SSE-NEXT:    pxor %xmm1, %xmm1
; SSE-NEXT:    pcmpeqb %xmm0, %xmm1
; SSE-NEXT:    movmskps %xmm1, %ecx
; SSE-NEXT:    xorl %eax, %eax
; SSE-NEXT:    cmpl $15, %ecx
; SSE-NEXT:    sete %al
; SSE-NEXT:    negl %eax
; SSE-NEXT:    retq
;
; AVX-LABEL: movmskps_ptest_numelts_mismatch:
; AVX:       # %bb.0:
; AVX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
; AVX-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
; AVX-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
; AVX-NEXT:    xorl %eax, %eax
; AVX-NEXT:    vtestps %xmm1, %xmm0
; AVX-NEXT:    sbbl %eax, %eax
; AVX-NEXT:    retq
;
; ADL-LABEL: movmskps_ptest_numelts_mismatch:
; ADL:       # %bb.0:
; ADL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
; ADL-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
; ADL-NEXT:    vmovmskps %xmm0, %ecx
; ADL-NEXT:    xorl %eax, %eax
; ADL-NEXT:    cmpl $15, %ecx
; ADL-NEXT:    sete %al
; ADL-NEXT:    negl %eax
; ADL-NEXT:    retq
  %1 = icmp eq <16 x i8> %a0, zeroinitializer
  %2 = sext <16 x i1> %1 to <16 x i8>
  %3 = bitcast <16 x i8> %2 to <4 x float>
  %4 = tail call i32 @llvm.x86.sse.movmsk.ps(<4 x float> %3)
  %5 = icmp eq i32 %4, 15
  %6 = sext i1 %5 to i32
  ret i32 %6
}