Files
TheRock/compiler/amd-llvm/llvm/test/CodeGen/X86/avx512-shuffles/broadcast-vector-int.ll
T
derek.holloway 7502bf64a4
CodeQL / Analyze (actions) (push) Canceled after 0s
CodeQL / Analyze (python) (push) Canceled after 0s
pre-commit / pre-commit (push) Canceled after 0s
Multi-Arch CI / setup (push) Canceled after 0s
Publish build_manylinux_x86_64 images / publish_build_manylinux_x86_64 (push) Canceled after 0s
Setup (Multi-Arch) / setup (push) Canceled after 0s
Multi-Arch CI / Linux::${{ fromJSON(needs.setup.outputs.linux_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish no_rocm_image_ubi10 image / publish_no_rocm_image_ubi10 (push) Canceled after 0s
Multi-Arch CI / Windows::${{ fromJSON(needs.setup.outputs.windows_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Multi-Arch CI / Manifest Diff (push) Canceled after 0s
TheRock Manifest Diff Report / Generate Manifest Diff Report (push) Canceled after 0s
Multi-Arch CI / CI Summary (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04 (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04_ocl_rt images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04_rocgdb images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Unit Tests / Unit Tests :: ubuntu-24.04 (push) Canceled after 0s
Unit Tests / Unit Tests :: windows-2022 (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_media (push) Canceled after 0s
Unit Tests / Unit Tests Summary (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_openmpi (push) Canceled after 0s
Flatten subprojects into a single base monorepo
2026-09-02 16:34:47 -07:00

1298 lines
64 KiB
LLVM

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+avx512f,+avx512vl,+avx512dq,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle %s -o - | FileCheck %s
; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+avx512f,+avx512vl,+avx512dq,+fast-variable-perlane-shuffle %s -o - | FileCheck %s
define <4 x i32> @test_2xi32_to_4xi32(<4 x i32> %vec) {
; CHECK-LABEL: test_2xi32_to_4xi32:
; CHECK: # %bb.0:
; CHECK-NEXT: vmovddup {{.*#+}} xmm0 = xmm0[0,0]
; CHECK-NEXT: retq
%res = shufflevector <4 x i32> %vec, <4 x i32> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
ret <4 x i32> %res
}
define <4 x i32> @test_masked_2xi32_to_4xi32_mask0(<4 x i32> %vec, <4 x i32> %default, <4 x i32> %mask) {
; CHECK-LABEL: test_masked_2xi32_to_4xi32_mask0:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %xmm2, %xmm2, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} xmm1 {%k1} = xmm0[0,1,0,1]
; CHECK-NEXT: vmovdqa %xmm1, %xmm0
; CHECK-NEXT: retq
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <4 x i32> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i32> %shuf, <4 x i32> %default
ret <4 x i32> %res
}
define <4 x i32> @test_masked_z_2xi32_to_4xi32_mask0(<4 x i32> %vec, <4 x i32> %mask) {
; CHECK-LABEL: test_masked_z_2xi32_to_4xi32_mask0:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %xmm1, %xmm1, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} xmm0 {%k1} {z} = xmm0[0,1,0,1]
; CHECK-NEXT: retq
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <4 x i32> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i32> %shuf, <4 x i32> zeroinitializer
ret <4 x i32> %res
}
define <4 x i32> @test_masked_2xi32_to_4xi32_mask1(<4 x i32> %vec, <4 x i32> %default, <4 x i32> %mask) {
; CHECK-LABEL: test_masked_2xi32_to_4xi32_mask1:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %xmm2, %xmm2, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} xmm1 {%k1} = xmm0[0,1,0,1]
; CHECK-NEXT: vmovdqa %xmm1, %xmm0
; CHECK-NEXT: retq
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <4 x i32> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i32> %shuf, <4 x i32> %default
ret <4 x i32> %res
}
define <4 x i32> @test_masked_z_2xi32_to_4xi32_mask1(<4 x i32> %vec, <4 x i32> %mask) {
; CHECK-LABEL: test_masked_z_2xi32_to_4xi32_mask1:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %xmm1, %xmm1, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} xmm0 {%k1} {z} = xmm0[0,1,0,1]
; CHECK-NEXT: retq
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <4 x i32> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i32> %shuf, <4 x i32> zeroinitializer
ret <4 x i32> %res
}
define <4 x i32> @test_masked_2xi32_to_4xi32_mask2(<4 x i32> %vec, <4 x i32> %default, <4 x i32> %mask) {
; CHECK-LABEL: test_masked_2xi32_to_4xi32_mask2:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %xmm2, %xmm2, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} xmm1 {%k1} = xmm0[0,1,0,1]
; CHECK-NEXT: vmovdqa %xmm1, %xmm0
; CHECK-NEXT: retq
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <4 x i32> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i32> %shuf, <4 x i32> %default
ret <4 x i32> %res
}
define <4 x i32> @test_masked_z_2xi32_to_4xi32_mask2(<4 x i32> %vec, <4 x i32> %mask) {
; CHECK-LABEL: test_masked_z_2xi32_to_4xi32_mask2:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %xmm1, %xmm1, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} xmm0 {%k1} {z} = xmm0[0,1,0,1]
; CHECK-NEXT: retq
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <4 x i32> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i32> %shuf, <4 x i32> zeroinitializer
ret <4 x i32> %res
}
define <4 x i32> @test_masked_2xi32_to_4xi32_mask3(<4 x i32> %vec, <4 x i32> %default, <4 x i32> %mask) {
; CHECK-LABEL: test_masked_2xi32_to_4xi32_mask3:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %xmm2, %xmm2, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} xmm1 {%k1} = xmm0[0,1,0,1]
; CHECK-NEXT: vmovdqa %xmm1, %xmm0
; CHECK-NEXT: retq
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <4 x i32> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i32> %shuf, <4 x i32> %default
ret <4 x i32> %res
}
define <4 x i32> @test_masked_z_2xi32_to_4xi32_mask3(<4 x i32> %vec, <4 x i32> %mask) {
; CHECK-LABEL: test_masked_z_2xi32_to_4xi32_mask3:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %xmm1, %xmm1, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} xmm0 {%k1} {z} = xmm0[0,1,0,1]
; CHECK-NEXT: retq
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <4 x i32> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i32> %shuf, <4 x i32> zeroinitializer
ret <4 x i32> %res
}
define <8 x i32> @test_2xi32_to_8xi32(<8 x i32> %vec) {
; CHECK-LABEL: test_2xi32_to_8xi32:
; CHECK: # %bb.0:
; CHECK-NEXT: vbroadcastsd %xmm0, %ymm0
; CHECK-NEXT: retq
%res = shufflevector <8 x i32> %vec, <8 x i32> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
ret <8 x i32> %res
}
define <8 x i32> @test_masked_2xi32_to_8xi32_mask0(<8 x i32> %vec, <8 x i32> %default, <8 x i32> %mask) {
; CHECK-LABEL: test_masked_2xi32_to_8xi32_mask0:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %ymm2, %ymm2, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} ymm1 {%k1} = xmm0[0,1,0,1,0,1,0,1]
; CHECK-NEXT: vmovdqa %ymm1, %ymm0
; CHECK-NEXT: retq
%shuf = shufflevector <8 x i32> %vec, <8 x i32> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> %default
ret <8 x i32> %res
}
define <8 x i32> @test_masked_z_2xi32_to_8xi32_mask0(<8 x i32> %vec, <8 x i32> %mask) {
; CHECK-LABEL: test_masked_z_2xi32_to_8xi32_mask0:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %ymm1, %ymm1, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} ymm0 {%k1} {z} = xmm0[0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%shuf = shufflevector <8 x i32> %vec, <8 x i32> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> zeroinitializer
ret <8 x i32> %res
}
define <8 x i32> @test_masked_2xi32_to_8xi32_mask1(<8 x i32> %vec, <8 x i32> %default, <8 x i32> %mask) {
; CHECK-LABEL: test_masked_2xi32_to_8xi32_mask1:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %ymm2, %ymm2, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} ymm1 {%k1} = xmm0[0,1,0,1,0,1,0,1]
; CHECK-NEXT: vmovdqa %ymm1, %ymm0
; CHECK-NEXT: retq
%shuf = shufflevector <8 x i32> %vec, <8 x i32> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> %default
ret <8 x i32> %res
}
define <8 x i32> @test_masked_z_2xi32_to_8xi32_mask1(<8 x i32> %vec, <8 x i32> %mask) {
; CHECK-LABEL: test_masked_z_2xi32_to_8xi32_mask1:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %ymm1, %ymm1, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} ymm0 {%k1} {z} = xmm0[0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%shuf = shufflevector <8 x i32> %vec, <8 x i32> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> zeroinitializer
ret <8 x i32> %res
}
define <8 x i32> @test_masked_2xi32_to_8xi32_mask2(<8 x i32> %vec, <8 x i32> %default, <8 x i32> %mask) {
; CHECK-LABEL: test_masked_2xi32_to_8xi32_mask2:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %ymm2, %ymm2, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} ymm1 {%k1} = xmm0[0,1,0,1,0,1,0,1]
; CHECK-NEXT: vmovdqa %ymm1, %ymm0
; CHECK-NEXT: retq
%shuf = shufflevector <8 x i32> %vec, <8 x i32> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> %default
ret <8 x i32> %res
}
define <8 x i32> @test_masked_z_2xi32_to_8xi32_mask2(<8 x i32> %vec, <8 x i32> %mask) {
; CHECK-LABEL: test_masked_z_2xi32_to_8xi32_mask2:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %ymm1, %ymm1, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} ymm0 {%k1} {z} = xmm0[0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%shuf = shufflevector <8 x i32> %vec, <8 x i32> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> zeroinitializer
ret <8 x i32> %res
}
define <8 x i32> @test_masked_2xi32_to_8xi32_mask3(<8 x i32> %vec, <8 x i32> %default, <8 x i32> %mask) {
; CHECK-LABEL: test_masked_2xi32_to_8xi32_mask3:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %ymm2, %ymm2, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} ymm1 {%k1} = xmm0[0,1,0,1,0,1,0,1]
; CHECK-NEXT: vmovdqa %ymm1, %ymm0
; CHECK-NEXT: retq
%shuf = shufflevector <8 x i32> %vec, <8 x i32> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> %default
ret <8 x i32> %res
}
define <8 x i32> @test_masked_z_2xi32_to_8xi32_mask3(<8 x i32> %vec, <8 x i32> %mask) {
; CHECK-LABEL: test_masked_z_2xi32_to_8xi32_mask3:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %ymm1, %ymm1, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} ymm0 {%k1} {z} = xmm0[0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%shuf = shufflevector <8 x i32> %vec, <8 x i32> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> zeroinitializer
ret <8 x i32> %res
}
define <16 x i32> @test_2xi32_to_16xi32(<16 x i32> %vec) {
; CHECK-LABEL: test_2xi32_to_16xi32:
; CHECK: # %bb.0:
; CHECK-NEXT: vbroadcastsd %xmm0, %zmm0
; CHECK-NEXT: retq
%res = shufflevector <16 x i32> %vec, <16 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
ret <16 x i32> %res
}
define <16 x i32> @test_masked_2xi32_to_16xi32_mask0(<16 x i32> %vec, <16 x i32> %default, <16 x i32> %mask) {
; CHECK-LABEL: test_masked_2xi32_to_16xi32_mask0:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %zmm2, %zmm2, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} zmm1 {%k1} = xmm0[0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1]
; CHECK-NEXT: vmovdqa64 %zmm1, %zmm0
; CHECK-NEXT: retq
%shuf = shufflevector <16 x i32> %vec, <16 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> %default
ret <16 x i32> %res
}
define <16 x i32> @test_masked_z_2xi32_to_16xi32_mask0(<16 x i32> %vec, <16 x i32> %mask) {
; CHECK-LABEL: test_masked_z_2xi32_to_16xi32_mask0:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %zmm1, %zmm1, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} zmm0 {%k1} {z} = xmm0[0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%shuf = shufflevector <16 x i32> %vec, <16 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> zeroinitializer
ret <16 x i32> %res
}
define <16 x i32> @test_masked_2xi32_to_16xi32_mask1(<16 x i32> %vec, <16 x i32> %default, <16 x i32> %mask) {
; CHECK-LABEL: test_masked_2xi32_to_16xi32_mask1:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %zmm2, %zmm2, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} zmm1 {%k1} = xmm0[0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1]
; CHECK-NEXT: vmovdqa64 %zmm1, %zmm0
; CHECK-NEXT: retq
%shuf = shufflevector <16 x i32> %vec, <16 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> %default
ret <16 x i32> %res
}
define <16 x i32> @test_masked_z_2xi32_to_16xi32_mask1(<16 x i32> %vec, <16 x i32> %mask) {
; CHECK-LABEL: test_masked_z_2xi32_to_16xi32_mask1:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %zmm1, %zmm1, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} zmm0 {%k1} {z} = xmm0[0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%shuf = shufflevector <16 x i32> %vec, <16 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> zeroinitializer
ret <16 x i32> %res
}
define <16 x i32> @test_masked_2xi32_to_16xi32_mask2(<16 x i32> %vec, <16 x i32> %default, <16 x i32> %mask) {
; CHECK-LABEL: test_masked_2xi32_to_16xi32_mask2:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %zmm2, %zmm2, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} zmm1 {%k1} = xmm0[0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1]
; CHECK-NEXT: vmovdqa64 %zmm1, %zmm0
; CHECK-NEXT: retq
%shuf = shufflevector <16 x i32> %vec, <16 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> %default
ret <16 x i32> %res
}
define <16 x i32> @test_masked_z_2xi32_to_16xi32_mask2(<16 x i32> %vec, <16 x i32> %mask) {
; CHECK-LABEL: test_masked_z_2xi32_to_16xi32_mask2:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %zmm1, %zmm1, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} zmm0 {%k1} {z} = xmm0[0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%shuf = shufflevector <16 x i32> %vec, <16 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> zeroinitializer
ret <16 x i32> %res
}
define <16 x i32> @test_masked_2xi32_to_16xi32_mask3(<16 x i32> %vec, <16 x i32> %default, <16 x i32> %mask) {
; CHECK-LABEL: test_masked_2xi32_to_16xi32_mask3:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %zmm2, %zmm2, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} zmm1 {%k1} = xmm0[0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1]
; CHECK-NEXT: vmovdqa64 %zmm1, %zmm0
; CHECK-NEXT: retq
%shuf = shufflevector <16 x i32> %vec, <16 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> %default
ret <16 x i32> %res
}
define <16 x i32> @test_masked_z_2xi32_to_16xi32_mask3(<16 x i32> %vec, <16 x i32> %mask) {
; CHECK-LABEL: test_masked_z_2xi32_to_16xi32_mask3:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %zmm1, %zmm1, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} zmm0 {%k1} {z} = xmm0[0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%shuf = shufflevector <16 x i32> %vec, <16 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> zeroinitializer
ret <16 x i32> %res
}
define <4 x i32> @test_2xi32_to_4xi32_mem(ptr %vp) {
; CHECK-LABEL: test_2xi32_to_4xi32_mem:
; CHECK: # %bb.0:
; CHECK-NEXT: vmovddup {{.*#+}} xmm0 = mem[0,0]
; CHECK-NEXT: retq
%vec = load <2 x i32>, ptr %vp
%res = shufflevector <2 x i32> %vec, <2 x i32> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
ret <4 x i32> %res
}
define <4 x i32> @test_masked_2xi32_to_4xi32_mem_mask0(ptr %vp, <4 x i32> %default, <4 x i32> %mask) {
; CHECK-LABEL: test_masked_2xi32_to_4xi32_mem_mask0:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %xmm1, %xmm1, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} xmm0 {%k1} = mem[0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i32>, ptr %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <4 x i32> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i32> %shuf, <4 x i32> %default
ret <4 x i32> %res
}
define <4 x i32> @test_masked_z_2xi32_to_4xi32_mem_mask0(ptr %vp, <4 x i32> %mask) {
; CHECK-LABEL: test_masked_z_2xi32_to_4xi32_mem_mask0:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %xmm0, %xmm0, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} xmm0 {%k1} {z} = mem[0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i32>, ptr %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <4 x i32> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i32> %shuf, <4 x i32> zeroinitializer
ret <4 x i32> %res
}
define <4 x i32> @test_masked_2xi32_to_4xi32_mem_mask1(ptr %vp, <4 x i32> %default, <4 x i32> %mask) {
; CHECK-LABEL: test_masked_2xi32_to_4xi32_mem_mask1:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %xmm1, %xmm1, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} xmm0 {%k1} = mem[0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i32>, ptr %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <4 x i32> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i32> %shuf, <4 x i32> %default
ret <4 x i32> %res
}
define <4 x i32> @test_masked_z_2xi32_to_4xi32_mem_mask1(ptr %vp, <4 x i32> %mask) {
; CHECK-LABEL: test_masked_z_2xi32_to_4xi32_mem_mask1:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %xmm0, %xmm0, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} xmm0 {%k1} {z} = mem[0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i32>, ptr %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <4 x i32> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i32> %shuf, <4 x i32> zeroinitializer
ret <4 x i32> %res
}
define <4 x i32> @test_masked_2xi32_to_4xi32_mem_mask2(ptr %vp, <4 x i32> %default, <4 x i32> %mask) {
; CHECK-LABEL: test_masked_2xi32_to_4xi32_mem_mask2:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %xmm1, %xmm1, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} xmm0 {%k1} = mem[0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i32>, ptr %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <4 x i32> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i32> %shuf, <4 x i32> %default
ret <4 x i32> %res
}
define <4 x i32> @test_masked_z_2xi32_to_4xi32_mem_mask2(ptr %vp, <4 x i32> %mask) {
; CHECK-LABEL: test_masked_z_2xi32_to_4xi32_mem_mask2:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %xmm0, %xmm0, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} xmm0 {%k1} {z} = mem[0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i32>, ptr %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <4 x i32> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i32> %shuf, <4 x i32> zeroinitializer
ret <4 x i32> %res
}
define <4 x i32> @test_masked_2xi32_to_4xi32_mem_mask3(ptr %vp, <4 x i32> %default, <4 x i32> %mask) {
; CHECK-LABEL: test_masked_2xi32_to_4xi32_mem_mask3:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %xmm1, %xmm1, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} xmm0 {%k1} = mem[0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i32>, ptr %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <4 x i32> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i32> %shuf, <4 x i32> %default
ret <4 x i32> %res
}
define <4 x i32> @test_masked_z_2xi32_to_4xi32_mem_mask3(ptr %vp, <4 x i32> %mask) {
; CHECK-LABEL: test_masked_z_2xi32_to_4xi32_mem_mask3:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %xmm0, %xmm0, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} xmm0 {%k1} {z} = mem[0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i32>, ptr %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <4 x i32> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i32> %shuf, <4 x i32> zeroinitializer
ret <4 x i32> %res
}
define <8 x i32> @test_2xi32_to_8xi32_mem(ptr %vp) {
; CHECK-LABEL: test_2xi32_to_8xi32_mem:
; CHECK: # %bb.0:
; CHECK-NEXT: vbroadcastsd (%rdi), %ymm0
; CHECK-NEXT: retq
%vec = load <2 x i32>, ptr %vp
%res = shufflevector <2 x i32> %vec, <2 x i32> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
ret <8 x i32> %res
}
define <8 x i32> @test_masked_2xi32_to_8xi32_mem_mask0(ptr %vp, <8 x i32> %default, <8 x i32> %mask) {
; CHECK-LABEL: test_masked_2xi32_to_8xi32_mem_mask0:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %ymm1, %ymm1, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} ymm0 {%k1} = mem[0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i32>, ptr %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> %default
ret <8 x i32> %res
}
define <8 x i32> @test_masked_z_2xi32_to_8xi32_mem_mask0(ptr %vp, <8 x i32> %mask) {
; CHECK-LABEL: test_masked_z_2xi32_to_8xi32_mem_mask0:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %ymm0, %ymm0, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} ymm0 {%k1} {z} = mem[0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i32>, ptr %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> zeroinitializer
ret <8 x i32> %res
}
define <8 x i32> @test_masked_2xi32_to_8xi32_mem_mask1(ptr %vp, <8 x i32> %default, <8 x i32> %mask) {
; CHECK-LABEL: test_masked_2xi32_to_8xi32_mem_mask1:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %ymm1, %ymm1, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} ymm0 {%k1} = mem[0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i32>, ptr %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> %default
ret <8 x i32> %res
}
define <8 x i32> @test_masked_z_2xi32_to_8xi32_mem_mask1(ptr %vp, <8 x i32> %mask) {
; CHECK-LABEL: test_masked_z_2xi32_to_8xi32_mem_mask1:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %ymm0, %ymm0, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} ymm0 {%k1} {z} = mem[0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i32>, ptr %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> zeroinitializer
ret <8 x i32> %res
}
define <8 x i32> @test_masked_2xi32_to_8xi32_mem_mask2(ptr %vp, <8 x i32> %default, <8 x i32> %mask) {
; CHECK-LABEL: test_masked_2xi32_to_8xi32_mem_mask2:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %ymm1, %ymm1, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} ymm0 {%k1} = mem[0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i32>, ptr %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> %default
ret <8 x i32> %res
}
define <8 x i32> @test_masked_z_2xi32_to_8xi32_mem_mask2(ptr %vp, <8 x i32> %mask) {
; CHECK-LABEL: test_masked_z_2xi32_to_8xi32_mem_mask2:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %ymm0, %ymm0, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} ymm0 {%k1} {z} = mem[0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i32>, ptr %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> zeroinitializer
ret <8 x i32> %res
}
define <8 x i32> @test_masked_2xi32_to_8xi32_mem_mask3(ptr %vp, <8 x i32> %default, <8 x i32> %mask) {
; CHECK-LABEL: test_masked_2xi32_to_8xi32_mem_mask3:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %ymm1, %ymm1, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} ymm0 {%k1} = mem[0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i32>, ptr %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> %default
ret <8 x i32> %res
}
define <8 x i32> @test_masked_z_2xi32_to_8xi32_mem_mask3(ptr %vp, <8 x i32> %mask) {
; CHECK-LABEL: test_masked_z_2xi32_to_8xi32_mem_mask3:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %ymm0, %ymm0, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} ymm0 {%k1} {z} = mem[0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i32>, ptr %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> zeroinitializer
ret <8 x i32> %res
}
define <16 x i32> @test_2xi32_to_16xi32_mem(ptr %vp) {
; CHECK-LABEL: test_2xi32_to_16xi32_mem:
; CHECK: # %bb.0:
; CHECK-NEXT: vbroadcastsd (%rdi), %zmm0
; CHECK-NEXT: retq
%vec = load <2 x i32>, ptr %vp
%res = shufflevector <2 x i32> %vec, <2 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
ret <16 x i32> %res
}
define <16 x i32> @test_masked_2xi32_to_16xi32_mem_mask0(ptr %vp, <16 x i32> %default, <16 x i32> %mask) {
; CHECK-LABEL: test_masked_2xi32_to_16xi32_mem_mask0:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %zmm1, %zmm1, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} zmm0 {%k1} = mem[0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i32>, ptr %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> %default
ret <16 x i32> %res
}
define <16 x i32> @test_masked_z_2xi32_to_16xi32_mem_mask0(ptr %vp, <16 x i32> %mask) {
; CHECK-LABEL: test_masked_z_2xi32_to_16xi32_mem_mask0:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %zmm0, %zmm0, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} zmm0 {%k1} {z} = mem[0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i32>, ptr %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> zeroinitializer
ret <16 x i32> %res
}
define <16 x i32> @test_masked_2xi32_to_16xi32_mem_mask1(ptr %vp, <16 x i32> %default, <16 x i32> %mask) {
; CHECK-LABEL: test_masked_2xi32_to_16xi32_mem_mask1:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %zmm1, %zmm1, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} zmm0 {%k1} = mem[0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i32>, ptr %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> %default
ret <16 x i32> %res
}
define <16 x i32> @test_masked_z_2xi32_to_16xi32_mem_mask1(ptr %vp, <16 x i32> %mask) {
; CHECK-LABEL: test_masked_z_2xi32_to_16xi32_mem_mask1:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %zmm0, %zmm0, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} zmm0 {%k1} {z} = mem[0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i32>, ptr %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> zeroinitializer
ret <16 x i32> %res
}
define <16 x i32> @test_masked_2xi32_to_16xi32_mem_mask2(ptr %vp, <16 x i32> %default, <16 x i32> %mask) {
; CHECK-LABEL: test_masked_2xi32_to_16xi32_mem_mask2:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %zmm1, %zmm1, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} zmm0 {%k1} = mem[0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i32>, ptr %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> %default
ret <16 x i32> %res
}
define <16 x i32> @test_masked_z_2xi32_to_16xi32_mem_mask2(ptr %vp, <16 x i32> %mask) {
; CHECK-LABEL: test_masked_z_2xi32_to_16xi32_mem_mask2:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %zmm0, %zmm0, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} zmm0 {%k1} {z} = mem[0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i32>, ptr %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> zeroinitializer
ret <16 x i32> %res
}
define <16 x i32> @test_masked_2xi32_to_16xi32_mem_mask3(ptr %vp, <16 x i32> %default, <16 x i32> %mask) {
; CHECK-LABEL: test_masked_2xi32_to_16xi32_mem_mask3:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %zmm1, %zmm1, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} zmm0 {%k1} = mem[0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i32>, ptr %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> %default
ret <16 x i32> %res
}
define <16 x i32> @test_masked_z_2xi32_to_16xi32_mem_mask3(ptr %vp, <16 x i32> %mask) {
; CHECK-LABEL: test_masked_z_2xi32_to_16xi32_mem_mask3:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %zmm0, %zmm0, %k1
; CHECK-NEXT: vbroadcasti32x2 {{.*#+}} zmm0 {%k1} {z} = mem[0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i32>, ptr %vp
%shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> zeroinitializer
ret <16 x i32> %res
}
define <8 x i32> @test_4xi32_to_8xi32_mem(ptr %vp) {
; CHECK-LABEL: test_4xi32_to_8xi32_mem:
; CHECK: # %bb.0:
; CHECK-NEXT: vbroadcasti128 {{.*#+}} ymm0 = mem[0,1,0,1]
; CHECK-NEXT: retq
%vec = load <4 x i32>, ptr %vp
%res = shufflevector <4 x i32> %vec, <4 x i32> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
ret <8 x i32> %res
}
define <8 x i32> @test_masked_4xi32_to_8xi32_mem_mask0(ptr %vp, <8 x i32> %default, <8 x i32> %mask) {
; CHECK-LABEL: test_masked_4xi32_to_8xi32_mem_mask0:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %ymm1, %ymm1, %k1
; CHECK-NEXT: vbroadcasti32x4 {{.*#+}} ymm0 {%k1} = mem[0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i32>, ptr %vp
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> %default
ret <8 x i32> %res
}
define <8 x i32> @test_masked_z_4xi32_to_8xi32_mem_mask0(ptr %vp, <8 x i32> %mask) {
; CHECK-LABEL: test_masked_z_4xi32_to_8xi32_mem_mask0:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %ymm0, %ymm0, %k1
; CHECK-NEXT: vbroadcasti32x4 {{.*#+}} ymm0 {%k1} {z} = mem[0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i32>, ptr %vp
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> zeroinitializer
ret <8 x i32> %res
}
define <8 x i32> @test_masked_4xi32_to_8xi32_mem_mask1(ptr %vp, <8 x i32> %default, <8 x i32> %mask) {
; CHECK-LABEL: test_masked_4xi32_to_8xi32_mem_mask1:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %ymm1, %ymm1, %k1
; CHECK-NEXT: vbroadcasti32x4 {{.*#+}} ymm0 {%k1} = mem[0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i32>, ptr %vp
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> %default
ret <8 x i32> %res
}
define <8 x i32> @test_masked_z_4xi32_to_8xi32_mem_mask1(ptr %vp, <8 x i32> %mask) {
; CHECK-LABEL: test_masked_z_4xi32_to_8xi32_mem_mask1:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %ymm0, %ymm0, %k1
; CHECK-NEXT: vbroadcasti32x4 {{.*#+}} ymm0 {%k1} {z} = mem[0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i32>, ptr %vp
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> zeroinitializer
ret <8 x i32> %res
}
define <8 x i32> @test_masked_4xi32_to_8xi32_mem_mask2(ptr %vp, <8 x i32> %default, <8 x i32> %mask) {
; CHECK-LABEL: test_masked_4xi32_to_8xi32_mem_mask2:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %ymm1, %ymm1, %k1
; CHECK-NEXT: vbroadcasti32x4 {{.*#+}} ymm0 {%k1} = mem[0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i32>, ptr %vp
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> %default
ret <8 x i32> %res
}
define <8 x i32> @test_masked_z_4xi32_to_8xi32_mem_mask2(ptr %vp, <8 x i32> %mask) {
; CHECK-LABEL: test_masked_z_4xi32_to_8xi32_mem_mask2:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %ymm0, %ymm0, %k1
; CHECK-NEXT: vbroadcasti32x4 {{.*#+}} ymm0 {%k1} {z} = mem[0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i32>, ptr %vp
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> zeroinitializer
ret <8 x i32> %res
}
define <8 x i32> @test_masked_4xi32_to_8xi32_mem_mask3(ptr %vp, <8 x i32> %default, <8 x i32> %mask) {
; CHECK-LABEL: test_masked_4xi32_to_8xi32_mem_mask3:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %ymm1, %ymm1, %k1
; CHECK-NEXT: vbroadcasti32x4 {{.*#+}} ymm0 {%k1} = mem[0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i32>, ptr %vp
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> %default
ret <8 x i32> %res
}
define <8 x i32> @test_masked_z_4xi32_to_8xi32_mem_mask3(ptr %vp, <8 x i32> %mask) {
; CHECK-LABEL: test_masked_z_4xi32_to_8xi32_mem_mask3:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %ymm0, %ymm0, %k1
; CHECK-NEXT: vbroadcasti32x4 {{.*#+}} ymm0 {%k1} {z} = mem[0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i32>, ptr %vp
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
%cmp = icmp eq <8 x i32> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i32> %shuf, <8 x i32> zeroinitializer
ret <8 x i32> %res
}
define <16 x i32> @test_4xi32_to_16xi32_mem(ptr %vp) {
; CHECK-LABEL: test_4xi32_to_16xi32_mem:
; CHECK: # %bb.0:
; CHECK-NEXT: vbroadcasti32x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i32>, ptr %vp
%res = shufflevector <4 x i32> %vec, <4 x i32> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
ret <16 x i32> %res
}
define <16 x i32> @test_masked_4xi32_to_16xi32_mem_mask0(ptr %vp, <16 x i32> %default, <16 x i32> %mask) {
; CHECK-LABEL: test_masked_4xi32_to_16xi32_mem_mask0:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %zmm1, %zmm1, %k1
; CHECK-NEXT: vbroadcasti32x4 {{.*#+}} zmm0 {%k1} = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i32>, ptr %vp
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> %default
ret <16 x i32> %res
}
define <16 x i32> @test_masked_z_4xi32_to_16xi32_mem_mask0(ptr %vp, <16 x i32> %mask) {
; CHECK-LABEL: test_masked_z_4xi32_to_16xi32_mem_mask0:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %zmm0, %zmm0, %k1
; CHECK-NEXT: vbroadcasti32x4 {{.*#+}} zmm0 {%k1} {z} = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i32>, ptr %vp
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> zeroinitializer
ret <16 x i32> %res
}
define <16 x i32> @test_masked_4xi32_to_16xi32_mem_mask1(ptr %vp, <16 x i32> %default, <16 x i32> %mask) {
; CHECK-LABEL: test_masked_4xi32_to_16xi32_mem_mask1:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %zmm1, %zmm1, %k1
; CHECK-NEXT: vbroadcasti32x4 {{.*#+}} zmm0 {%k1} = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i32>, ptr %vp
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> %default
ret <16 x i32> %res
}
define <16 x i32> @test_masked_z_4xi32_to_16xi32_mem_mask1(ptr %vp, <16 x i32> %mask) {
; CHECK-LABEL: test_masked_z_4xi32_to_16xi32_mem_mask1:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %zmm0, %zmm0, %k1
; CHECK-NEXT: vbroadcasti32x4 {{.*#+}} zmm0 {%k1} {z} = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i32>, ptr %vp
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> zeroinitializer
ret <16 x i32> %res
}
define <16 x i32> @test_masked_4xi32_to_16xi32_mem_mask2(ptr %vp, <16 x i32> %default, <16 x i32> %mask) {
; CHECK-LABEL: test_masked_4xi32_to_16xi32_mem_mask2:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %zmm1, %zmm1, %k1
; CHECK-NEXT: vbroadcasti32x4 {{.*#+}} zmm0 {%k1} = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i32>, ptr %vp
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> %default
ret <16 x i32> %res
}
define <16 x i32> @test_masked_z_4xi32_to_16xi32_mem_mask2(ptr %vp, <16 x i32> %mask) {
; CHECK-LABEL: test_masked_z_4xi32_to_16xi32_mem_mask2:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %zmm0, %zmm0, %k1
; CHECK-NEXT: vbroadcasti32x4 {{.*#+}} zmm0 {%k1} {z} = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i32>, ptr %vp
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> zeroinitializer
ret <16 x i32> %res
}
define <16 x i32> @test_masked_4xi32_to_16xi32_mem_mask3(ptr %vp, <16 x i32> %default, <16 x i32> %mask) {
; CHECK-LABEL: test_masked_4xi32_to_16xi32_mem_mask3:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %zmm1, %zmm1, %k1
; CHECK-NEXT: vbroadcasti32x4 {{.*#+}} zmm0 {%k1} = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i32>, ptr %vp
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> %default
ret <16 x i32> %res
}
define <16 x i32> @test_masked_z_4xi32_to_16xi32_mem_mask3(ptr %vp, <16 x i32> %mask) {
; CHECK-LABEL: test_masked_z_4xi32_to_16xi32_mem_mask3:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %zmm0, %zmm0, %k1
; CHECK-NEXT: vbroadcasti32x4 {{.*#+}} zmm0 {%k1} {z} = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i32>, ptr %vp
%shuf = shufflevector <4 x i32> %vec, <4 x i32> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> zeroinitializer
ret <16 x i32> %res
}
define <4 x i64> @test_2xi64_to_4xi64_mem(ptr %vp) {
; CHECK-LABEL: test_2xi64_to_4xi64_mem:
; CHECK: # %bb.0:
; CHECK-NEXT: vbroadcasti128 {{.*#+}} ymm0 = mem[0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i64>, ptr %vp
%res = shufflevector <2 x i64> %vec, <2 x i64> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
ret <4 x i64> %res
}
define <4 x i64> @test_masked_2xi64_to_4xi64_mem_mask0(ptr %vp, <4 x i64> %default, <4 x i64> %mask) {
; CHECK-LABEL: test_masked_2xi64_to_4xi64_mem_mask0:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmq %ymm1, %ymm1, %k1
; CHECK-NEXT: vbroadcasti64x2 {{.*#+}} ymm0 {%k1} = mem[0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i64>, ptr %vp
%shuf = shufflevector <2 x i64> %vec, <2 x i64> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <4 x i64> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i64> %shuf, <4 x i64> %default
ret <4 x i64> %res
}
define <4 x i64> @test_masked_z_2xi64_to_4xi64_mem_mask0(ptr %vp, <4 x i64> %mask) {
; CHECK-LABEL: test_masked_z_2xi64_to_4xi64_mem_mask0:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmq %ymm0, %ymm0, %k1
; CHECK-NEXT: vbroadcasti64x2 {{.*#+}} ymm0 {%k1} {z} = mem[0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i64>, ptr %vp
%shuf = shufflevector <2 x i64> %vec, <2 x i64> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <4 x i64> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i64> %shuf, <4 x i64> zeroinitializer
ret <4 x i64> %res
}
define <4 x i64> @test_masked_2xi64_to_4xi64_mem_mask1(ptr %vp, <4 x i64> %default, <4 x i64> %mask) {
; CHECK-LABEL: test_masked_2xi64_to_4xi64_mem_mask1:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmq %ymm1, %ymm1, %k1
; CHECK-NEXT: vbroadcasti64x2 {{.*#+}} ymm0 {%k1} = mem[0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i64>, ptr %vp
%shuf = shufflevector <2 x i64> %vec, <2 x i64> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <4 x i64> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i64> %shuf, <4 x i64> %default
ret <4 x i64> %res
}
define <4 x i64> @test_masked_z_2xi64_to_4xi64_mem_mask1(ptr %vp, <4 x i64> %mask) {
; CHECK-LABEL: test_masked_z_2xi64_to_4xi64_mem_mask1:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmq %ymm0, %ymm0, %k1
; CHECK-NEXT: vbroadcasti64x2 {{.*#+}} ymm0 {%k1} {z} = mem[0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i64>, ptr %vp
%shuf = shufflevector <2 x i64> %vec, <2 x i64> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <4 x i64> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i64> %shuf, <4 x i64> zeroinitializer
ret <4 x i64> %res
}
define <4 x i64> @test_masked_2xi64_to_4xi64_mem_mask2(ptr %vp, <4 x i64> %default, <4 x i64> %mask) {
; CHECK-LABEL: test_masked_2xi64_to_4xi64_mem_mask2:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmq %ymm1, %ymm1, %k1
; CHECK-NEXT: vbroadcasti64x2 {{.*#+}} ymm0 {%k1} = mem[0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i64>, ptr %vp
%shuf = shufflevector <2 x i64> %vec, <2 x i64> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <4 x i64> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i64> %shuf, <4 x i64> %default
ret <4 x i64> %res
}
define <4 x i64> @test_masked_z_2xi64_to_4xi64_mem_mask2(ptr %vp, <4 x i64> %mask) {
; CHECK-LABEL: test_masked_z_2xi64_to_4xi64_mem_mask2:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmq %ymm0, %ymm0, %k1
; CHECK-NEXT: vbroadcasti64x2 {{.*#+}} ymm0 {%k1} {z} = mem[0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i64>, ptr %vp
%shuf = shufflevector <2 x i64> %vec, <2 x i64> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <4 x i64> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i64> %shuf, <4 x i64> zeroinitializer
ret <4 x i64> %res
}
define <4 x i64> @test_masked_2xi64_to_4xi64_mem_mask3(ptr %vp, <4 x i64> %default, <4 x i64> %mask) {
; CHECK-LABEL: test_masked_2xi64_to_4xi64_mem_mask3:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmq %ymm1, %ymm1, %k1
; CHECK-NEXT: vbroadcasti64x2 {{.*#+}} ymm0 {%k1} = mem[0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i64>, ptr %vp
%shuf = shufflevector <2 x i64> %vec, <2 x i64> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <4 x i64> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i64> %shuf, <4 x i64> %default
ret <4 x i64> %res
}
define <4 x i64> @test_masked_z_2xi64_to_4xi64_mem_mask3(ptr %vp, <4 x i64> %mask) {
; CHECK-LABEL: test_masked_z_2xi64_to_4xi64_mem_mask3:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmq %ymm0, %ymm0, %k1
; CHECK-NEXT: vbroadcasti64x2 {{.*#+}} ymm0 {%k1} {z} = mem[0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i64>, ptr %vp
%shuf = shufflevector <2 x i64> %vec, <2 x i64> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <4 x i64> %mask, zeroinitializer
%res = select <4 x i1> %cmp, <4 x i64> %shuf, <4 x i64> zeroinitializer
ret <4 x i64> %res
}
define <8 x i64> @test_2xi64_to_8xi64_mem(ptr %vp) {
; CHECK-LABEL: test_2xi64_to_8xi64_mem:
; CHECK: # %bb.0:
; CHECK-NEXT: vbroadcasti32x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <2 x i64>, ptr %vp
%res = shufflevector <2 x i64> %vec, <2 x i64> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
ret <8 x i64> %res
}
define <8 x i64> @test_masked_2xi64_to_8xi64_mem_mask0(ptr %vp, <8 x i64> %default, <8 x i64> %mask) {
; CHECK-LABEL: test_masked_2xi64_to_8xi64_mem_mask0:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmq %zmm1, %zmm1, %k1
; CHECK-NEXT: vbroadcasti64x2 {{.*#+}} zmm0 {%k1} = mem[0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i64>, ptr %vp
%shuf = shufflevector <2 x i64> %vec, <2 x i64> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <8 x i64> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i64> %shuf, <8 x i64> %default
ret <8 x i64> %res
}
define <8 x i64> @test_masked_z_2xi64_to_8xi64_mem_mask0(ptr %vp, <8 x i64> %mask) {
; CHECK-LABEL: test_masked_z_2xi64_to_8xi64_mem_mask0:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmq %zmm0, %zmm0, %k1
; CHECK-NEXT: vbroadcasti64x2 {{.*#+}} zmm0 {%k1} {z} = mem[0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i64>, ptr %vp
%shuf = shufflevector <2 x i64> %vec, <2 x i64> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <8 x i64> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i64> %shuf, <8 x i64> zeroinitializer
ret <8 x i64> %res
}
define <8 x i64> @test_masked_2xi64_to_8xi64_mem_mask1(ptr %vp, <8 x i64> %default, <8 x i64> %mask) {
; CHECK-LABEL: test_masked_2xi64_to_8xi64_mem_mask1:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmq %zmm1, %zmm1, %k1
; CHECK-NEXT: vbroadcasti64x2 {{.*#+}} zmm0 {%k1} = mem[0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i64>, ptr %vp
%shuf = shufflevector <2 x i64> %vec, <2 x i64> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <8 x i64> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i64> %shuf, <8 x i64> %default
ret <8 x i64> %res
}
define <8 x i64> @test_masked_z_2xi64_to_8xi64_mem_mask1(ptr %vp, <8 x i64> %mask) {
; CHECK-LABEL: test_masked_z_2xi64_to_8xi64_mem_mask1:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmq %zmm0, %zmm0, %k1
; CHECK-NEXT: vbroadcasti64x2 {{.*#+}} zmm0 {%k1} {z} = mem[0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i64>, ptr %vp
%shuf = shufflevector <2 x i64> %vec, <2 x i64> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <8 x i64> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i64> %shuf, <8 x i64> zeroinitializer
ret <8 x i64> %res
}
define <8 x i64> @test_masked_2xi64_to_8xi64_mem_mask2(ptr %vp, <8 x i64> %default, <8 x i64> %mask) {
; CHECK-LABEL: test_masked_2xi64_to_8xi64_mem_mask2:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmq %zmm1, %zmm1, %k1
; CHECK-NEXT: vbroadcasti64x2 {{.*#+}} zmm0 {%k1} = mem[0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i64>, ptr %vp
%shuf = shufflevector <2 x i64> %vec, <2 x i64> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <8 x i64> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i64> %shuf, <8 x i64> %default
ret <8 x i64> %res
}
define <8 x i64> @test_masked_z_2xi64_to_8xi64_mem_mask2(ptr %vp, <8 x i64> %mask) {
; CHECK-LABEL: test_masked_z_2xi64_to_8xi64_mem_mask2:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmq %zmm0, %zmm0, %k1
; CHECK-NEXT: vbroadcasti64x2 {{.*#+}} zmm0 {%k1} {z} = mem[0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i64>, ptr %vp
%shuf = shufflevector <2 x i64> %vec, <2 x i64> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <8 x i64> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i64> %shuf, <8 x i64> zeroinitializer
ret <8 x i64> %res
}
define <8 x i64> @test_masked_2xi64_to_8xi64_mem_mask3(ptr %vp, <8 x i64> %default, <8 x i64> %mask) {
; CHECK-LABEL: test_masked_2xi64_to_8xi64_mem_mask3:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmq %zmm1, %zmm1, %k1
; CHECK-NEXT: vbroadcasti64x2 {{.*#+}} zmm0 {%k1} = mem[0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i64>, ptr %vp
%shuf = shufflevector <2 x i64> %vec, <2 x i64> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <8 x i64> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i64> %shuf, <8 x i64> %default
ret <8 x i64> %res
}
define <8 x i64> @test_masked_z_2xi64_to_8xi64_mem_mask3(ptr %vp, <8 x i64> %mask) {
; CHECK-LABEL: test_masked_z_2xi64_to_8xi64_mem_mask3:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmq %zmm0, %zmm0, %k1
; CHECK-NEXT: vbroadcasti64x2 {{.*#+}} zmm0 {%k1} {z} = mem[0,1,0,1,0,1,0,1]
; CHECK-NEXT: retq
%vec = load <2 x i64>, ptr %vp
%shuf = shufflevector <2 x i64> %vec, <2 x i64> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
%cmp = icmp eq <8 x i64> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i64> %shuf, <8 x i64> zeroinitializer
ret <8 x i64> %res
}
define <16 x i32> @test_8xi32_to_16xi32_mem(ptr %vp) {
; CHECK-LABEL: test_8xi32_to_16xi32_mem:
; CHECK: # %bb.0:
; CHECK-NEXT: vbroadcasti64x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <8 x i32>, ptr %vp
%res = shufflevector <8 x i32> %vec, <8 x i32> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
ret <16 x i32> %res
}
define <16 x i32> @test_masked_8xi32_to_16xi32_mem_mask0(ptr %vp, <16 x i32> %default, <16 x i32> %mask) {
; CHECK-LABEL: test_masked_8xi32_to_16xi32_mem_mask0:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %zmm1, %zmm1, %k1
; CHECK-NEXT: vbroadcasti32x8 {{.*#+}} zmm0 {%k1} = mem[0,1,2,3,4,5,6,7,0,1,2,3,4,5,6,7]
; CHECK-NEXT: retq
%vec = load <8 x i32>, ptr %vp
%shuf = shufflevector <8 x i32> %vec, <8 x i32> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> %default
ret <16 x i32> %res
}
define <16 x i32> @test_masked_z_8xi32_to_16xi32_mem_mask0(ptr %vp, <16 x i32> %mask) {
; CHECK-LABEL: test_masked_z_8xi32_to_16xi32_mem_mask0:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %zmm0, %zmm0, %k1
; CHECK-NEXT: vbroadcasti32x8 {{.*#+}} zmm0 {%k1} {z} = mem[0,1,2,3,4,5,6,7,0,1,2,3,4,5,6,7]
; CHECK-NEXT: retq
%vec = load <8 x i32>, ptr %vp
%shuf = shufflevector <8 x i32> %vec, <8 x i32> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> zeroinitializer
ret <16 x i32> %res
}
define <16 x i32> @test_masked_8xi32_to_16xi32_mem_mask1(ptr %vp, <16 x i32> %default, <16 x i32> %mask) {
; CHECK-LABEL: test_masked_8xi32_to_16xi32_mem_mask1:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %zmm1, %zmm1, %k1
; CHECK-NEXT: vbroadcasti32x8 {{.*#+}} zmm0 {%k1} = mem[0,1,2,3,4,5,6,7,0,1,2,3,4,5,6,7]
; CHECK-NEXT: retq
%vec = load <8 x i32>, ptr %vp
%shuf = shufflevector <8 x i32> %vec, <8 x i32> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> %default
ret <16 x i32> %res
}
define <16 x i32> @test_masked_z_8xi32_to_16xi32_mem_mask1(ptr %vp, <16 x i32> %mask) {
; CHECK-LABEL: test_masked_z_8xi32_to_16xi32_mem_mask1:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %zmm0, %zmm0, %k1
; CHECK-NEXT: vbroadcasti32x8 {{.*#+}} zmm0 {%k1} {z} = mem[0,1,2,3,4,5,6,7,0,1,2,3,4,5,6,7]
; CHECK-NEXT: retq
%vec = load <8 x i32>, ptr %vp
%shuf = shufflevector <8 x i32> %vec, <8 x i32> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> zeroinitializer
ret <16 x i32> %res
}
define <16 x i32> @test_masked_8xi32_to_16xi32_mem_mask2(ptr %vp, <16 x i32> %default, <16 x i32> %mask) {
; CHECK-LABEL: test_masked_8xi32_to_16xi32_mem_mask2:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %zmm1, %zmm1, %k1
; CHECK-NEXT: vbroadcasti32x8 {{.*#+}} zmm0 {%k1} = mem[0,1,2,3,4,5,6,7,0,1,2,3,4,5,6,7]
; CHECK-NEXT: retq
%vec = load <8 x i32>, ptr %vp
%shuf = shufflevector <8 x i32> %vec, <8 x i32> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> %default
ret <16 x i32> %res
}
define <16 x i32> @test_masked_z_8xi32_to_16xi32_mem_mask2(ptr %vp, <16 x i32> %mask) {
; CHECK-LABEL: test_masked_z_8xi32_to_16xi32_mem_mask2:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %zmm0, %zmm0, %k1
; CHECK-NEXT: vbroadcasti32x8 {{.*#+}} zmm0 {%k1} {z} = mem[0,1,2,3,4,5,6,7,0,1,2,3,4,5,6,7]
; CHECK-NEXT: retq
%vec = load <8 x i32>, ptr %vp
%shuf = shufflevector <8 x i32> %vec, <8 x i32> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> zeroinitializer
ret <16 x i32> %res
}
define <16 x i32> @test_masked_8xi32_to_16xi32_mem_mask3(ptr %vp, <16 x i32> %default, <16 x i32> %mask) {
; CHECK-LABEL: test_masked_8xi32_to_16xi32_mem_mask3:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %zmm1, %zmm1, %k1
; CHECK-NEXT: vbroadcasti32x8 {{.*#+}} zmm0 {%k1} = mem[0,1,2,3,4,5,6,7,0,1,2,3,4,5,6,7]
; CHECK-NEXT: retq
%vec = load <8 x i32>, ptr %vp
%shuf = shufflevector <8 x i32> %vec, <8 x i32> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> %default
ret <16 x i32> %res
}
define <16 x i32> @test_masked_z_8xi32_to_16xi32_mem_mask3(ptr %vp, <16 x i32> %mask) {
; CHECK-LABEL: test_masked_z_8xi32_to_16xi32_mem_mask3:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmd %zmm0, %zmm0, %k1
; CHECK-NEXT: vbroadcasti32x8 {{.*#+}} zmm0 {%k1} {z} = mem[0,1,2,3,4,5,6,7,0,1,2,3,4,5,6,7]
; CHECK-NEXT: retq
%vec = load <8 x i32>, ptr %vp
%shuf = shufflevector <8 x i32> %vec, <8 x i32> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
%cmp = icmp eq <16 x i32> %mask, zeroinitializer
%res = select <16 x i1> %cmp, <16 x i32> %shuf, <16 x i32> zeroinitializer
ret <16 x i32> %res
}
define <8 x i64> @test_4xi64_to_8xi64_mem(ptr %vp) {
; CHECK-LABEL: test_4xi64_to_8xi64_mem:
; CHECK: # %bb.0:
; CHECK-NEXT: vbroadcasti64x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i64>, ptr %vp
%res = shufflevector <4 x i64> %vec, <4 x i64> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
ret <8 x i64> %res
}
define <8 x i64> @test_masked_4xi64_to_8xi64_mem_mask0(ptr %vp, <8 x i64> %default, <8 x i64> %mask) {
; CHECK-LABEL: test_masked_4xi64_to_8xi64_mem_mask0:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmq %zmm1, %zmm1, %k1
; CHECK-NEXT: vbroadcasti64x4 {{.*#+}} zmm0 {%k1} = mem[0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i64>, ptr %vp
%shuf = shufflevector <4 x i64> %vec, <4 x i64> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
%cmp = icmp eq <8 x i64> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i64> %shuf, <8 x i64> %default
ret <8 x i64> %res
}
define <8 x i64> @test_masked_z_4xi64_to_8xi64_mem_mask0(ptr %vp, <8 x i64> %mask) {
; CHECK-LABEL: test_masked_z_4xi64_to_8xi64_mem_mask0:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmq %zmm0, %zmm0, %k1
; CHECK-NEXT: vbroadcasti64x4 {{.*#+}} zmm0 {%k1} {z} = mem[0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i64>, ptr %vp
%shuf = shufflevector <4 x i64> %vec, <4 x i64> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
%cmp = icmp eq <8 x i64> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i64> %shuf, <8 x i64> zeroinitializer
ret <8 x i64> %res
}
define <8 x i64> @test_masked_4xi64_to_8xi64_mem_mask1(ptr %vp, <8 x i64> %default, <8 x i64> %mask) {
; CHECK-LABEL: test_masked_4xi64_to_8xi64_mem_mask1:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmq %zmm1, %zmm1, %k1
; CHECK-NEXT: vbroadcasti64x4 {{.*#+}} zmm0 {%k1} = mem[0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i64>, ptr %vp
%shuf = shufflevector <4 x i64> %vec, <4 x i64> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
%cmp = icmp eq <8 x i64> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i64> %shuf, <8 x i64> %default
ret <8 x i64> %res
}
define <8 x i64> @test_masked_z_4xi64_to_8xi64_mem_mask1(ptr %vp, <8 x i64> %mask) {
; CHECK-LABEL: test_masked_z_4xi64_to_8xi64_mem_mask1:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmq %zmm0, %zmm0, %k1
; CHECK-NEXT: vbroadcasti64x4 {{.*#+}} zmm0 {%k1} {z} = mem[0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i64>, ptr %vp
%shuf = shufflevector <4 x i64> %vec, <4 x i64> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
%cmp = icmp eq <8 x i64> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i64> %shuf, <8 x i64> zeroinitializer
ret <8 x i64> %res
}
define <8 x i64> @test_masked_4xi64_to_8xi64_mem_mask2(ptr %vp, <8 x i64> %default, <8 x i64> %mask) {
; CHECK-LABEL: test_masked_4xi64_to_8xi64_mem_mask2:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmq %zmm1, %zmm1, %k1
; CHECK-NEXT: vbroadcasti64x4 {{.*#+}} zmm0 {%k1} = mem[0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i64>, ptr %vp
%shuf = shufflevector <4 x i64> %vec, <4 x i64> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
%cmp = icmp eq <8 x i64> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i64> %shuf, <8 x i64> %default
ret <8 x i64> %res
}
define <8 x i64> @test_masked_z_4xi64_to_8xi64_mem_mask2(ptr %vp, <8 x i64> %mask) {
; CHECK-LABEL: test_masked_z_4xi64_to_8xi64_mem_mask2:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmq %zmm0, %zmm0, %k1
; CHECK-NEXT: vbroadcasti64x4 {{.*#+}} zmm0 {%k1} {z} = mem[0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i64>, ptr %vp
%shuf = shufflevector <4 x i64> %vec, <4 x i64> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
%cmp = icmp eq <8 x i64> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i64> %shuf, <8 x i64> zeroinitializer
ret <8 x i64> %res
}
define <8 x i64> @test_masked_4xi64_to_8xi64_mem_mask3(ptr %vp, <8 x i64> %default, <8 x i64> %mask) {
; CHECK-LABEL: test_masked_4xi64_to_8xi64_mem_mask3:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmq %zmm1, %zmm1, %k1
; CHECK-NEXT: vbroadcasti64x4 {{.*#+}} zmm0 {%k1} = mem[0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i64>, ptr %vp
%shuf = shufflevector <4 x i64> %vec, <4 x i64> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
%cmp = icmp eq <8 x i64> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i64> %shuf, <8 x i64> %default
ret <8 x i64> %res
}
define <8 x i64> @test_masked_z_4xi64_to_8xi64_mem_mask3(ptr %vp, <8 x i64> %mask) {
; CHECK-LABEL: test_masked_z_4xi64_to_8xi64_mem_mask3:
; CHECK: # %bb.0:
; CHECK-NEXT: vptestnmq %zmm0, %zmm0, %k1
; CHECK-NEXT: vbroadcasti64x4 {{.*#+}} zmm0 {%k1} {z} = mem[0,1,2,3,0,1,2,3]
; CHECK-NEXT: retq
%vec = load <4 x i64>, ptr %vp
%shuf = shufflevector <4 x i64> %vec, <4 x i64> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
%cmp = icmp eq <8 x i64> %mask, zeroinitializer
%res = select <8 x i1> %cmp, <8 x i64> %shuf, <8 x i64> zeroinitializer
ret <8 x i64> %res
}