[X86] Add tests showing failure to concat icmp instructions together.

[RKSimon]

No description provided.

[llvmbot]

@llvm/pr-subscribers-backend-x86

Author: Simon Pilgrim (RKSimon)

Changes

---

Patch is 36.92 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/170210.diff

1 Files Affected:

• (added) llvm/test/CodeGen/X86/combine-icmp.ll (+846)

diff --git a/llvm/test/CodeGen/X86/combine-icmp.ll b/llvm/test/CodeGen/X86/combine-icmp.ll
new file mode 100644
index 0000000000000..ea1ab15f6e9ba
--- /dev/null
+++ b/llvm/test/CodeGen/X86/combine-icmp.ll
@@ -0,0 +1,846 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64    | FileCheck %s --check-prefixes=SSE,SSE2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64-v2 | FileCheck %s --check-prefixes=SSE,SSE42
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=sandybridge | FileCheck %s --check-prefixes=AVX,AVX1OR2,AVX1
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64-v3 | FileCheck %s --check-prefixes=AVX,AVX1OR2,AVX2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64-v4 | FileCheck %s --check-prefixes=AVX,AVX512
+
+define i4 @concat_icmp_v4i64_v2i64(<2 x i64> %a0, <2 x i64> %a1) {
+; SSE2-LABEL: concat_icmp_v4i64_v2i64:
+; SSE2:       # %bb.0:
+; SSE2-NEXT:    pxor %xmm2, %xmm2
+; SSE2-NEXT:    pcmpeqd %xmm2, %xmm0
+; SSE2-NEXT:    pcmpeqd %xmm2, %xmm1
+; SSE2-NEXT:    movdqa %xmm0, %xmm2
+; SSE2-NEXT:    shufps {{.*#+}} xmm2 = xmm2[1,3],xmm1[1,3]
+; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
+; SSE2-NEXT:    andps %xmm2, %xmm0
+; SSE2-NEXT:    movmskps %xmm0, %eax
+; SSE2-NEXT:    xorl $15, %eax
+; SSE2-NEXT:    # kill: def $al killed $al killed $eax
+; SSE2-NEXT:    retq
+;
+; SSE42-LABEL: concat_icmp_v4i64_v2i64:
+; SSE42:       # %bb.0:
+; SSE42-NEXT:    pxor %xmm2, %xmm2
+; SSE42-NEXT:    pcmpeqq %xmm2, %xmm0
+; SSE42-NEXT:    pcmpeqq %xmm2, %xmm1
+; SSE42-NEXT:    packssdw %xmm1, %xmm0
+; SSE42-NEXT:    movmskps %xmm0, %eax
+; SSE42-NEXT:    xorl $15, %eax
+; SSE42-NEXT:    # kill: def $al killed $al killed $eax
+; SSE42-NEXT:    retq
+;
+; AVX1OR2-LABEL: concat_icmp_v4i64_v2i64:
+; AVX1OR2:       # %bb.0:
+; AVX1OR2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX1OR2-NEXT:    vpcmpeqq %xmm2, %xmm0, %xmm0
+; AVX1OR2-NEXT:    vpcmpeqq %xmm2, %xmm1, %xmm1
+; AVX1OR2-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
+; AVX1OR2-NEXT:    vmovmskps %xmm0, %eax
+; AVX1OR2-NEXT:    xorl $15, %eax
+; AVX1OR2-NEXT:    # kill: def $al killed $al killed $eax
+; AVX1OR2-NEXT:    retq
+;
+; AVX512-LABEL: concat_icmp_v4i64_v2i64:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vptestmq %xmm0, %xmm0, %k0
+; AVX512-NEXT:    vptestmq %xmm1, %xmm1, %k1
+; AVX512-NEXT:    kshiftlb $2, %k1, %k1
+; AVX512-NEXT:    korw %k1, %k0, %k0
+; AVX512-NEXT:    kmovd %k0, %eax
+; AVX512-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512-NEXT:    retq
+  %v0 = icmp ne <2 x i64> %a0, zeroinitializer
+  %v1 = icmp ne <2 x i64> %a1, zeroinitializer
+  %v = shufflevector <2 x i1> %v0, <2 x i1> %v1, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %r = bitcast <4 x i1> %v to i4
+  ret i4 %r
+}
+
+define i8 @concat_icmp_v8i32_v4i32(<4 x i32> %a0, <4 x i32> %a1) {
+; SSE-LABEL: concat_icmp_v8i32_v4i32:
+; SSE:       # %bb.0:
+; SSE-NEXT:    pxor %xmm2, %xmm2
+; SSE-NEXT:    pcmpeqd %xmm2, %xmm0
+; SSE-NEXT:    pcmpeqd %xmm2, %xmm1
+; SSE-NEXT:    packssdw %xmm1, %xmm0
+; SSE-NEXT:    packsswb %xmm0, %xmm0
+; SSE-NEXT:    pmovmskb %xmm0, %eax
+; SSE-NEXT:    # kill: def $al killed $al killed $eax
+; SSE-NEXT:    retq
+;
+; AVX1OR2-LABEL: concat_icmp_v8i32_v4i32:
+; AVX1OR2:       # %bb.0:
+; AVX1OR2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX1OR2-NEXT:    vpcmpeqd %xmm2, %xmm0, %xmm0
+; AVX1OR2-NEXT:    vpcmpeqd %xmm2, %xmm1, %xmm1
+; AVX1OR2-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
+; AVX1OR2-NEXT:    vpacksswb %xmm0, %xmm0, %xmm0
+; AVX1OR2-NEXT:    vpmovmskb %xmm0, %eax
+; AVX1OR2-NEXT:    # kill: def $al killed $al killed $eax
+; AVX1OR2-NEXT:    retq
+;
+; AVX512-LABEL: concat_icmp_v8i32_v4i32:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vptestnmd %xmm0, %xmm0, %k0
+; AVX512-NEXT:    vptestnmd %xmm1, %xmm1, %k1
+; AVX512-NEXT:    kshiftlb $4, %k1, %k1
+; AVX512-NEXT:    korb %k1, %k0, %k0
+; AVX512-NEXT:    kmovd %k0, %eax
+; AVX512-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512-NEXT:    retq
+  %v0 = icmp eq <4 x i32> %a0, zeroinitializer
+  %v1 = icmp eq <4 x i32> %a1, zeroinitializer
+  %v = shufflevector <4 x i1> %v0, <4 x i1> %v1, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+  %r = bitcast <8 x i1> %v to i8
+  ret i8 %r
+}
+
+define i16 @concat_icmp_v16i16_v8i16(<8 x i16> %a0, <8 x i16> %a1) {
+; SSE2-LABEL: concat_icmp_v16i16_v8i16:
+; SSE2:       # %bb.0:
+; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [2,2,2,2,2,2,2,2]
+; SSE2-NEXT:    movdqa %xmm2, %xmm3
+; SSE2-NEXT:    psubusw %xmm0, %xmm3
+; SSE2-NEXT:    pxor %xmm0, %xmm0
+; SSE2-NEXT:    pcmpeqw %xmm0, %xmm3
+; SSE2-NEXT:    psubusw %xmm1, %xmm2
+; SSE2-NEXT:    pcmpeqw %xmm0, %xmm2
+; SSE2-NEXT:    packsswb %xmm2, %xmm3
+; SSE2-NEXT:    pmovmskb %xmm3, %eax
+; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
+; SSE2-NEXT:    retq
+;
+; SSE42-LABEL: concat_icmp_v16i16_v8i16:
+; SSE42:       # %bb.0:
+; SSE42-NEXT:    movdqa {{.*#+}} xmm2 = [2,2,2,2,2,2,2,2]
+; SSE42-NEXT:    movdqa %xmm0, %xmm3
+; SSE42-NEXT:    pmaxuw %xmm2, %xmm3
+; SSE42-NEXT:    pcmpeqw %xmm0, %xmm3
+; SSE42-NEXT:    pmaxuw %xmm1, %xmm2
+; SSE42-NEXT:    pcmpeqw %xmm1, %xmm2
+; SSE42-NEXT:    packsswb %xmm2, %xmm3
+; SSE42-NEXT:    pmovmskb %xmm3, %eax
+; SSE42-NEXT:    # kill: def $ax killed $ax killed $eax
+; SSE42-NEXT:    retq
+;
+; AVX1-LABEL: concat_icmp_v16i16_v8i16:
+; AVX1:       # %bb.0:
+; AVX1-NEXT:    vbroadcastss {{.*#+}} xmm2 = [2,2,2,2,2,2,2,2]
+; AVX1-NEXT:    vpmaxuw %xmm2, %xmm0, %xmm3
+; AVX1-NEXT:    vpcmpeqw %xmm3, %xmm0, %xmm0
+; AVX1-NEXT:    vpmaxuw %xmm2, %xmm1, %xmm2
+; AVX1-NEXT:    vpcmpeqw %xmm2, %xmm1, %xmm1
+; AVX1-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpmovmskb %xmm0, %eax
+; AVX1-NEXT:    # kill: def $ax killed $ax killed $eax
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: concat_icmp_v16i16_v8i16:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [2,2,2,2,2,2,2,2]
+; AVX2-NEXT:    vpmaxuw %xmm2, %xmm0, %xmm3
+; AVX2-NEXT:    vpcmpeqw %xmm3, %xmm0, %xmm0
+; AVX2-NEXT:    vpmaxuw %xmm2, %xmm1, %xmm2
+; AVX2-NEXT:    vpcmpeqw %xmm2, %xmm1, %xmm1
+; AVX2-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpmovmskb %xmm0, %eax
+; AVX2-NEXT:    # kill: def $ax killed $ax killed $eax
+; AVX2-NEXT:    retq
+;
+; AVX512-LABEL: concat_icmp_v16i16_v8i16:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [1,1,1,1,1,1,1,1]
+; AVX512-NEXT:    vpcmpnleuw %xmm2, %xmm0, %k0
+; AVX512-NEXT:    vpcmpnleuw %xmm2, %xmm1, %k1
+; AVX512-NEXT:    kunpckbw %k0, %k1, %k0
+; AVX512-NEXT:    kmovd %k0, %eax
+; AVX512-NEXT:    # kill: def $ax killed $ax killed $eax
+; AVX512-NEXT:    retq
+  %v0 = icmp ugt <8 x i16> %a0, splat (i16 1)
+  %v1 = icmp ugt <8 x i16> %a1, splat (i16 1)
+  %v = shufflevector <8 x i1> %v0, <8 x i1> %v1, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
+  %r = bitcast <16 x i1> %v to i16
+  ret i16 %r
+}
+
+define i32 @concat_icmp_v32i8_v16i8(<16 x i8> %a0, <16 x i8> %a1) {
+; SSE-LABEL: concat_icmp_v32i8_v16i8:
+; SSE:       # %bb.0:
+; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5]
+; SSE-NEXT:    pcmpgtb %xmm2, %xmm0
+; SSE-NEXT:    pcmpgtb %xmm2, %xmm1
+; SSE-NEXT:    pmovmskb %xmm0, %ecx
+; SSE-NEXT:    pmovmskb %xmm1, %eax
+; SSE-NEXT:    shll $16, %eax
+; SSE-NEXT:    orl %ecx, %eax
+; SSE-NEXT:    retq
+;
+; AVX1-LABEL: concat_icmp_v32i8_v16i8:
+; AVX1:       # %bb.0:
+; AVX1-NEXT:    vbroadcastss {{.*#+}} xmm2 = [5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5]
+; AVX1-NEXT:    vpcmpgtb %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpcmpgtb %xmm2, %xmm1, %xmm1
+; AVX1-NEXT:    vpmovmskb %xmm0, %ecx
+; AVX1-NEXT:    vpmovmskb %xmm1, %eax
+; AVX1-NEXT:    shll $16, %eax
+; AVX1-NEXT:    orl %ecx, %eax
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: concat_icmp_v32i8_v16i8:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
+; AVX2-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-NEXT:    vpcmpgtb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
+; AVX2-NEXT:    vpmovmskb %ymm0, %eax
+; AVX2-NEXT:    vzeroupper
+; AVX2-NEXT:    retq
+;
+; AVX512-LABEL: concat_icmp_v32i8_v16i8:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5]
+; AVX512-NEXT:    vpcmpgtb %xmm2, %xmm0, %k0
+; AVX512-NEXT:    vpcmpgtb %xmm2, %xmm1, %k1
+; AVX512-NEXT:    kunpckwd %k0, %k1, %k0
+; AVX512-NEXT:    kmovd %k0, %eax
+; AVX512-NEXT:    retq
+  %v0 = icmp sgt <16 x i8> %a0, splat (i8 5)
+  %v1 = icmp sgt <16 x i8> %a1, splat (i8 5)
+  %v = shufflevector <16 x i1> %v0, <16 x i1> %v1, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
+  %r = bitcast <32 x i1> %v to i32
+  ret i32 %r
+}
+
+define i8 @concat_icmp_v8i64_v2i64(<2 x i64> %a0, <2 x i64> %a1, <2 x i64> %a2, <2 x i64> %a3) {
+; SSE2-LABEL: concat_icmp_v8i64_v2i64:
+; SSE2:       # %bb.0:
+; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [9223372039002259456,9223372039002259456]
+; SSE2-NEXT:    pxor %xmm4, %xmm0
+; SSE2-NEXT:    pxor %xmm4, %xmm1
+; SSE2-NEXT:    pxor %xmm4, %xmm2
+; SSE2-NEXT:    pxor %xmm4, %xmm3
+; SSE2-NEXT:    pshufd {{.*#+}} xmm6 = xmm3[0,2,2,3]
+; SSE2-NEXT:    movdqa {{.*#+}} xmm5 = [2147483776,2147483776,2147483776,2147483648]
+; SSE2-NEXT:    movdqa %xmm5, %xmm7
+; SSE2-NEXT:    pcmpgtd %xmm6, %xmm7
+; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[1,3,3,3]
+; SSE2-NEXT:    pcmpeqd %xmm4, %xmm3
+; SSE2-NEXT:    pand %xmm7, %xmm3
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[0,1,0,2,4,5,6,7]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm6 = xmm2[0,2,2,3]
+; SSE2-NEXT:    movdqa %xmm5, %xmm7
+; SSE2-NEXT:    pcmpgtd %xmm6, %xmm7
+; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,3,3,3]
+; SSE2-NEXT:    pcmpeqd %xmm4, %xmm2
+; SSE2-NEXT:    pand %xmm7, %xmm2
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[0,1,0,2,4,5,6,7]
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm1[0,2,2,3]
+; SSE2-NEXT:    movdqa %xmm5, %xmm6
+; SSE2-NEXT:    pcmpgtd %xmm3, %xmm6
+; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,3,3,3]
+; SSE2-NEXT:    pcmpeqd %xmm4, %xmm1
+; SSE2-NEXT:    pand %xmm6, %xmm1
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[0,2,2,3]
+; SSE2-NEXT:    pcmpgtd %xmm3, %xmm5
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,3,3,3]
+; SSE2-NEXT:    pcmpeqd %xmm4, %xmm0
+; SSE2-NEXT:    pand %xmm5, %xmm0
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; SSE2-NEXT:    packsswb %xmm2, %xmm0
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,3,2,3]
+; SSE2-NEXT:    pmovmskb %xmm0, %eax
+; SSE2-NEXT:    # kill: def $al killed $al killed $eax
+; SSE2-NEXT:    retq
+;
+; SSE42-LABEL: concat_icmp_v8i64_v2i64:
+; SSE42:       # %bb.0:
+; SSE42-NEXT:    movdqa {{.*#+}} xmm4 = [9223372036854775808,9223372036854775808]
+; SSE42-NEXT:    pxor %xmm4, %xmm0
+; SSE42-NEXT:    movdqa {{.*#+}} xmm5 = [9223372036854775936,9223372036854775936]
+; SSE42-NEXT:    movdqa %xmm5, %xmm6
+; SSE42-NEXT:    pcmpgtq %xmm0, %xmm6
+; SSE42-NEXT:    pxor %xmm4, %xmm1
+; SSE42-NEXT:    movdqa %xmm5, %xmm0
+; SSE42-NEXT:    pcmpgtq %xmm1, %xmm0
+; SSE42-NEXT:    packssdw %xmm0, %xmm6
+; SSE42-NEXT:    pxor %xmm4, %xmm2
+; SSE42-NEXT:    movdqa %xmm5, %xmm0
+; SSE42-NEXT:    pcmpgtq %xmm2, %xmm0
+; SSE42-NEXT:    pxor %xmm4, %xmm3
+; SSE42-NEXT:    pcmpgtq %xmm3, %xmm5
+; SSE42-NEXT:    packssdw %xmm5, %xmm0
+; SSE42-NEXT:    packssdw %xmm6, %xmm6
+; SSE42-NEXT:    packssdw %xmm0, %xmm0
+; SSE42-NEXT:    packsswb %xmm0, %xmm6
+; SSE42-NEXT:    pshufd {{.*#+}} xmm0 = xmm6[0,3,2,3]
+; SSE42-NEXT:    pmovmskb %xmm0, %eax
+; SSE42-NEXT:    # kill: def $al killed $al killed $eax
+; SSE42-NEXT:    retq
+;
+; AVX1-LABEL: concat_icmp_v8i64_v2i64:
+; AVX1:       # %bb.0:
+; AVX1-NEXT:    vmovddup {{.*#+}} xmm4 = [9223372036854775808,9223372036854775808]
+; AVX1-NEXT:    # xmm4 = mem[0,0]
+; AVX1-NEXT:    vpxor %xmm4, %xmm0, %xmm0
+; AVX1-NEXT:    vmovddup {{.*#+}} xmm5 = [9223372036854775936,9223372036854775936]
+; AVX1-NEXT:    # xmm5 = mem[0,0]
+; AVX1-NEXT:    vpcmpgtq %xmm0, %xmm5, %xmm0
+; AVX1-NEXT:    vpxor %xmm4, %xmm1, %xmm1
+; AVX1-NEXT:    vpcmpgtq %xmm1, %xmm5, %xmm1
+; AVX1-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpxor %xmm4, %xmm2, %xmm1
+; AVX1-NEXT:    vpcmpgtq %xmm1, %xmm5, %xmm1
+; AVX1-NEXT:    vpxor %xmm4, %xmm3, %xmm2
+; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm5, %xmm2
+; AVX1-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
+; AVX1-NEXT:    vpackssdw %xmm1, %xmm1, %xmm1
+; AVX1-NEXT:    vpackssdw %xmm0, %xmm0, %xmm0
+; AVX1-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,3,0,3]
+; AVX1-NEXT:    vpmovmskb %xmm0, %eax
+; AVX1-NEXT:    # kill: def $al killed $al killed $eax
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: concat_icmp_v8i64_v2i64:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vpbroadcastq {{.*#+}} xmm4 = [9223372036854775808,9223372036854775808]
+; AVX2-NEXT:    vpxor %xmm4, %xmm0, %xmm0
+; AVX2-NEXT:    vpbroadcastq {{.*#+}} xmm5 = [9223372036854775936,9223372036854775936]
+; AVX2-NEXT:    vpcmpgtq %xmm0, %xmm5, %xmm0
+; AVX2-NEXT:    vpxor %xmm4, %xmm1, %xmm1
+; AVX2-NEXT:    vpcmpgtq %xmm1, %xmm5, %xmm1
+; AVX2-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpxor %xmm4, %xmm2, %xmm1
+; AVX2-NEXT:    vpcmpgtq %xmm1, %xmm5, %xmm1
+; AVX2-NEXT:    vpxor %xmm4, %xmm3, %xmm2
+; AVX2-NEXT:    vpcmpgtq %xmm2, %xmm5, %xmm2
+; AVX2-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
+; AVX2-NEXT:    vpackssdw %xmm1, %xmm1, %xmm1
+; AVX2-NEXT:    vpackssdw %xmm0, %xmm0, %xmm0
+; AVX2-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,3,0,3]
+; AVX2-NEXT:    vpmovmskb %xmm0, %eax
+; AVX2-NEXT:    # kill: def $al killed $al killed $eax
+; AVX2-NEXT:    retq
+;
+; AVX512-LABEL: concat_icmp_v8i64_v2i64:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vpbroadcastq {{.*#+}} xmm4 = [128,128]
+; AVX512-NEXT:    vpcmpltuq %xmm4, %xmm0, %k0
+; AVX512-NEXT:    vpcmpltuq %xmm4, %xmm1, %k1
+; AVX512-NEXT:    vpcmpltuq %xmm4, %xmm2, %k2
+; AVX512-NEXT:    vpcmpltuq %xmm4, %xmm3, %k3
+; AVX512-NEXT:    kshiftlb $2, %k3, %k3
+; AVX512-NEXT:    korb %k3, %k2, %k2
+; AVX512-NEXT:    kshiftlb $4, %k2, %k2
+; AVX512-NEXT:    kshiftlb $2, %k1, %k1
+; AVX512-NEXT:    korw %k1, %k0, %k0
+; AVX512-NEXT:    kshiftlb $4, %k0, %k0
+; AVX512-NEXT:    kshiftrb $4, %k0, %k0
+; AVX512-NEXT:    korb %k2, %k0, %k0
+; AVX512-NEXT:    kmovd %k0, %eax
+; AVX512-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512-NEXT:    retq
+  %v0 = icmp ult <2 x i64> %a0, splat (i64 128)
+  %v1 = icmp ult <2 x i64> %a1, splat (i64 128)
+  %v2 = icmp ult <2 x i64> %a2, splat (i64 128)
+  %v3 = icmp ult <2 x i64> %a3, splat (i64 128)
+  %v01 = shufflevector <2 x i1> %v0, <2 x i1> %v1, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %v23 = shufflevector <2 x i1> %v2, <2 x i1> %v3, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %v = shufflevector <4 x i1> %v01, <4 x i1> %v23, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+  %r = bitcast <8 x i1> %v to i8
+  ret i8 %r
+}
+
+define i16 @concat_icmp_v16i32_v4i32(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2, <4 x i32> %a3) {
+; SSE-LABEL: concat_icmp_v16i32_v4i32:
+; SSE:       # %bb.0:
+; SSE-NEXT:    pxor %xmm4, %xmm4
+; SSE-NEXT:    pcmpgtd %xmm4, %xmm0
+; SSE-NEXT:    pcmpgtd %xmm4, %xmm1
+; SSE-NEXT:    packssdw %xmm1, %xmm0
+; SSE-NEXT:    pcmpgtd %xmm4, %xmm2
+; SSE-NEXT:    pcmpgtd %xmm4, %xmm3
+; SSE-NEXT:    packssdw %xmm3, %xmm2
+; SSE-NEXT:    packsswb %xmm2, %xmm0
+; SSE-NEXT:    pmovmskb %xmm0, %eax
+; SSE-NEXT:    # kill: def $ax killed $ax killed $eax
+; SSE-NEXT:    retq
+;
+; AVX1OR2-LABEL: concat_icmp_v16i32_v4i32:
+; AVX1OR2:       # %bb.0:
+; AVX1OR2-NEXT:    vpxor %xmm4, %xmm4, %xmm4
+; AVX1OR2-NEXT:    vpcmpgtd %xmm4, %xmm0, %xmm0
+; AVX1OR2-NEXT:    vpcmpgtd %xmm4, %xmm1, %xmm1
+; AVX1OR2-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
+; AVX1OR2-NEXT:    vpcmpgtd %xmm4, %xmm2, %xmm1
+; AVX1OR2-NEXT:    vpcmpgtd %xmm4, %xmm3, %xmm2
+; AVX1OR2-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
+; AVX1OR2-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
+; AVX1OR2-NEXT:    vpmovmskb %xmm0, %eax
+; AVX1OR2-NEXT:    # kill: def $ax killed $ax killed $eax
+; AVX1OR2-NEXT:    retq
+;
+; AVX512-LABEL: concat_icmp_v16i32_v4i32:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vpxor %xmm4, %xmm4, %xmm4
+; AVX512-NEXT:    vpcmpgtd %xmm4, %xmm0, %k0
+; AVX512-NEXT:    vpcmpgtd %xmm4, %xmm1, %k1
+; AVX512-NEXT:    vpcmpgtd %xmm4, %xmm2, %k2
+; AVX512-NEXT:    vpcmpgtd %xmm4, %xmm3, %k3
+; AVX512-NEXT:    kshiftlb $4, %k1, %k1
+; AVX512-NEXT:    korb %k1, %k0, %k0
+; AVX512-NEXT:    kshiftlb $4, %k3, %k1
+; AVX512-NEXT:    korb %k1, %k2, %k1
+; AVX512-NEXT:    kunpckbw %k0, %k1, %k0
+; AVX512-NEXT:    kmovd %k0, %eax
+; AVX512-NEXT:    # kill: def $ax killed $ax killed $eax
+; AVX512-NEXT:    retq
+  %v0 = icmp sgt <4 x i32> %a0, zeroinitializer
+  %v1 = icmp sgt <4 x i32> %a1, zeroinitializer
+  %v2 = icmp sgt <4 x i32> %a2, zeroinitializer
+  %v3 = icmp sgt <4 x i32> %a3, zeroinitializer
+  %v01 = shufflevector <4 x i1> %v0, <4 x i1> %v1, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+  %v23 = shufflevector <4 x i1> %v2, <4 x i1> %v3, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+  %v = shufflevector <8 x i1> %v01, <8 x i1> %v23, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
+  %r = bitcast <16 x i1> %v to i16
+  ret i16 %r
+}
+
+define i32 @concat_icmp_v32i16_v8i16(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> %a2, <8 x i16> %a3) {
+; SSE-LABEL: concat_icmp_v32i16_v8i16:
+; SSE:       # %bb.0:
+; SSE-NEXT:    pxor %xmm4, %xmm4
+; SSE-NEXT:    pcmpeqw %xmm4, %xmm0
+; SSE-NEXT:    pcmpeqw %xmm4, %xmm1
+; SSE-NEXT:    packsswb %xmm1, %xmm0
+; SSE-NEXT:    pcmpeqw %xmm4, %xmm2
+; SSE-NEXT:    pcmpeqw %xmm4, %xmm3
+; SSE-NEXT:    packsswb %xmm3, %xmm2
+; SSE-NEXT:    pmovmskb %xmm0, %ecx
+; SSE-NEXT:    xorl $65535, %ecx # imm = 0xFFFF
+; SSE-NEXT:    pmovmskb %xmm2, %eax
+; SSE-NEXT:    notl %eax
+; SSE-NEXT:    shll $16, %eax
+; SSE-NEXT:    orl %ecx, %eax
+; SSE-NEXT:    retq
+;
+; AVX1-LABEL: concat_icmp_v32i16_v8i16:
+; AVX1:       # %bb.0:
+; AVX1-NEXT:    vpxor %xmm4, %xmm4, %xmm4
+; AVX1-NEXT:    vpcmpeqw %xmm4, %xmm0, %xmm0
+; AVX1-NEXT:    vpcmpeqw %xmm4, %xmm1, %xmm1
+; AVX1-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpcmpeqw %xmm4, %xmm2, %xmm1
+; AVX1-NEXT:    vpcmpeqw %xmm4, %xmm3, %xmm2
+; AVX1-NEXT:    vpacksswb %xmm2, %xmm1, %xmm1
+; AVX1-NEXT:    vpmovmskb %xmm0, %ecx
+; AVX1-NEXT:    xorl $65535, %ecx # imm = 0xFFFF
+; AVX1-NEXT:    vpmovmskb %xmm1, %eax
+; AVX1-NEXT:    notl %eax
+; AVX1-NEXT:    shll $16, %eax
+; AVX1-NEXT:    orl %ecx, %eax
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: concat_icmp_v32i16_v8i16:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    # kill: def $xmm2 killed $xmm2 def $ymm2
+; AVX2-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
+; AVX2-NEXT:    vinserti128 $1, %xmm3, %ymm2, %ymm2
+; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
+; AVX2-NEXT:    vpcmpeqw %ymm3, %ymm2, %ymm2
+; AVX2-NEXT:    vpacksswb %ymm2, %ymm2, %ymm2
+; AVX2-NEXT:    vpcmpeqd %ymm4, %ymm4, %ymm4
+; AVX2-NEXT:    vpxor %ymm4, %ymm2, %ymm2
+; AVX2-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-NEXT:    vpcmpeqw %ymm3, %ymm0, %ymm0
+; AVX2-NEXT:    vpacksswb %ymm0, %ymm0, %ymm0
+; AVX2-NEXT:    vpxor %ymm4, %ymm0, %ymm0
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm2[2,3],ymm0[4,5],ymm2[6,7]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
+; AVX2-NEXT:    vpmovmskb %ymm0, %eax
+; AVX2-NEXT:    vzeroupper
+; AVX2-NEXT:    retq
+;
+; AVX512-LABEL: concat_icmp_v32i16_v8i16:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vptestmw %xmm0, %xmm0, %k0
+; AVX512-NEXT:    vptestmw %xmm1, %xmm1, %k1
+; AVX512-NEXT:    vptestmw %xmm2, %xmm2, %k2
+; AVX512-NEXT:    vptestmw %xmm3, %xmm3, %k3
+; AVX512-NEXT:    kunpckbw %k0, %k1, %k0
+; AVX512-NEXT:    kunpckbw %k2, %k3, %k1
+; AVX512-NEXT:   ...
[truncated]