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
1028 lines
37 KiB
LLVM
1028 lines
37 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
|
|
; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=knl | FileCheck %s --check-prefix=ALL --check-prefix=KNL
|
|
; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=skx | FileCheck %s --check-prefix=ALL --check-prefix=SKX
|
|
|
|
|
|
define <16 x i32> @vpandd(<16 x i32> %a, <16 x i32> %b) nounwind uwtable readnone ssp {
|
|
; ALL-LABEL: vpandd:
|
|
; ALL: ## %bb.0: ## %entry
|
|
; ALL-NEXT: vpaddd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to16}, %zmm0, %zmm0
|
|
; ALL-NEXT: vpandd %zmm1, %zmm0, %zmm0
|
|
; ALL-NEXT: retq
|
|
entry:
|
|
; Force the execution domain with an add.
|
|
%a2 = add <16 x i32> %a, <i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2,
|
|
i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2>
|
|
%x = and <16 x i32> %a2, %b
|
|
ret <16 x i32> %x
|
|
}
|
|
|
|
define <16 x i32> @vpandnd(<16 x i32> %a, <16 x i32> %b) nounwind uwtable readnone ssp {
|
|
; ALL-LABEL: vpandnd:
|
|
; ALL: ## %bb.0: ## %entry
|
|
; ALL-NEXT: vpaddd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to16}, %zmm0, %zmm0
|
|
; ALL-NEXT: vpandnd %zmm0, %zmm1, %zmm0
|
|
; ALL-NEXT: retq
|
|
entry:
|
|
; Force the execution domain with an add.
|
|
%a2 = add <16 x i32> %a, <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3,
|
|
i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
|
|
%b2 = xor <16 x i32> %b, <i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1,
|
|
i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1>
|
|
%x = and <16 x i32> %a2, %b2
|
|
ret <16 x i32> %x
|
|
}
|
|
|
|
define <16 x i32> @vpord(<16 x i32> %a, <16 x i32> %b) nounwind uwtable readnone ssp {
|
|
; ALL-LABEL: vpord:
|
|
; ALL: ## %bb.0: ## %entry
|
|
; ALL-NEXT: vpaddd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to16}, %zmm0, %zmm0
|
|
; ALL-NEXT: vpord %zmm1, %zmm0, %zmm0
|
|
; ALL-NEXT: retq
|
|
entry:
|
|
; Force the execution domain with an add.
|
|
%a2 = add <16 x i32> %a, <i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4,
|
|
i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4>
|
|
%x = or <16 x i32> %a2, %b
|
|
ret <16 x i32> %x
|
|
}
|
|
|
|
define <16 x i32> @vpxord(<16 x i32> %a, <16 x i32> %b) nounwind uwtable readnone ssp {
|
|
; ALL-LABEL: vpxord:
|
|
; ALL: ## %bb.0: ## %entry
|
|
; ALL-NEXT: vpaddd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to16}, %zmm0, %zmm0
|
|
; ALL-NEXT: vpxord %zmm1, %zmm0, %zmm0
|
|
; ALL-NEXT: retq
|
|
entry:
|
|
; Force the execution domain with an add.
|
|
%a2 = add <16 x i32> %a, <i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5,
|
|
i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5>
|
|
%x = xor <16 x i32> %a2, %b
|
|
ret <16 x i32> %x
|
|
}
|
|
|
|
define <8 x i64> @vpandq(<8 x i64> %a, <8 x i64> %b) nounwind uwtable readnone ssp {
|
|
; ALL-LABEL: vpandq:
|
|
; ALL: ## %bb.0: ## %entry
|
|
; ALL-NEXT: vpaddq {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm0, %zmm0
|
|
; ALL-NEXT: vpandq %zmm1, %zmm0, %zmm0
|
|
; ALL-NEXT: retq
|
|
entry:
|
|
; Force the execution domain with an add.
|
|
%a2 = add <8 x i64> %a, <i64 6, i64 6, i64 6, i64 6, i64 6, i64 6, i64 6, i64 6>
|
|
%x = and <8 x i64> %a2, %b
|
|
ret <8 x i64> %x
|
|
}
|
|
|
|
define <8 x i64> @vpandnq(<8 x i64> %a, <8 x i64> %b) nounwind uwtable readnone ssp {
|
|
; ALL-LABEL: vpandnq:
|
|
; ALL: ## %bb.0: ## %entry
|
|
; ALL-NEXT: vpaddq {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm0, %zmm0
|
|
; ALL-NEXT: vpandnq %zmm0, %zmm1, %zmm0
|
|
; ALL-NEXT: retq
|
|
entry:
|
|
; Force the execution domain with an add.
|
|
%a2 = add <8 x i64> %a, <i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7>
|
|
%b2 = xor <8 x i64> %b, <i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 -1>
|
|
%x = and <8 x i64> %a2, %b2
|
|
ret <8 x i64> %x
|
|
}
|
|
|
|
define <8 x i64> @vporq(<8 x i64> %a, <8 x i64> %b) nounwind uwtable readnone ssp {
|
|
; ALL-LABEL: vporq:
|
|
; ALL: ## %bb.0: ## %entry
|
|
; ALL-NEXT: vpaddq {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm0, %zmm0
|
|
; ALL-NEXT: vporq %zmm1, %zmm0, %zmm0
|
|
; ALL-NEXT: retq
|
|
entry:
|
|
; Force the execution domain with an add.
|
|
%a2 = add <8 x i64> %a, <i64 8, i64 8, i64 8, i64 8, i64 8, i64 8, i64 8, i64 8>
|
|
%x = or <8 x i64> %a2, %b
|
|
ret <8 x i64> %x
|
|
}
|
|
|
|
define <8 x i64> @vpxorq(<8 x i64> %a, <8 x i64> %b) nounwind uwtable readnone ssp {
|
|
; ALL-LABEL: vpxorq:
|
|
; ALL: ## %bb.0: ## %entry
|
|
; ALL-NEXT: vpaddq {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm0, %zmm0
|
|
; ALL-NEXT: vpxorq %zmm1, %zmm0, %zmm0
|
|
; ALL-NEXT: retq
|
|
entry:
|
|
; Force the execution domain with an add.
|
|
%a2 = add <8 x i64> %a, <i64 9, i64 9, i64 9, i64 9, i64 9, i64 9, i64 9, i64 9>
|
|
%x = xor <8 x i64> %a2, %b
|
|
ret <8 x i64> %x
|
|
}
|
|
|
|
|
|
define <8 x i64> @orq_broadcast(<8 x i64> %a) nounwind {
|
|
; KNL-LABEL: orq_broadcast:
|
|
; KNL: ## %bb.0:
|
|
; KNL-NEXT: vporq {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm0, %zmm0
|
|
; KNL-NEXT: retq
|
|
;
|
|
; SKX-LABEL: orq_broadcast:
|
|
; SKX: ## %bb.0:
|
|
; SKX-NEXT: vorpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm0, %zmm0
|
|
; SKX-NEXT: retq
|
|
%b = or <8 x i64> %a, <i64 2, i64 2, i64 2, i64 2, i64 2, i64 2, i64 2, i64 2>
|
|
ret <8 x i64> %b
|
|
}
|
|
|
|
define <16 x i32> @andd512fold(<16 x i32> %y, ptr %x) {
|
|
; KNL-LABEL: andd512fold:
|
|
; KNL: ## %bb.0: ## %entry
|
|
; KNL-NEXT: vpandd (%rdi), %zmm0, %zmm0
|
|
; KNL-NEXT: retq
|
|
;
|
|
; SKX-LABEL: andd512fold:
|
|
; SKX: ## %bb.0: ## %entry
|
|
; SKX-NEXT: vandps (%rdi), %zmm0, %zmm0
|
|
; SKX-NEXT: retq
|
|
entry:
|
|
%a = load <16 x i32>, ptr %x, align 4
|
|
%b = and <16 x i32> %y, %a
|
|
ret <16 x i32> %b
|
|
}
|
|
|
|
define <8 x i64> @andqbrst(<8 x i64> %p1, ptr %ap) {
|
|
; KNL-LABEL: andqbrst:
|
|
; KNL: ## %bb.0: ## %entry
|
|
; KNL-NEXT: vpandq (%rdi){1to8}, %zmm0, %zmm0
|
|
; KNL-NEXT: retq
|
|
;
|
|
; SKX-LABEL: andqbrst:
|
|
; SKX: ## %bb.0: ## %entry
|
|
; SKX-NEXT: vandpd (%rdi){1to8}, %zmm0, %zmm0
|
|
; SKX-NEXT: retq
|
|
entry:
|
|
%a = load i64, ptr %ap, align 8
|
|
%b = insertelement <8 x i64> undef, i64 %a, i32 0
|
|
%c = shufflevector <8 x i64> %b, <8 x i64> undef, <8 x i32> zeroinitializer
|
|
%d = and <8 x i64> %p1, %c
|
|
ret <8 x i64>%d
|
|
}
|
|
|
|
define <64 x i8> @and_v64i8(<64 x i8> %a, <64 x i8> %b) {
|
|
; KNL-LABEL: and_v64i8:
|
|
; KNL: ## %bb.0:
|
|
; KNL-NEXT: vpandq %zmm1, %zmm0, %zmm0
|
|
; KNL-NEXT: retq
|
|
;
|
|
; SKX-LABEL: and_v64i8:
|
|
; SKX: ## %bb.0:
|
|
; SKX-NEXT: vandps %zmm1, %zmm0, %zmm0
|
|
; SKX-NEXT: retq
|
|
%res = and <64 x i8> %a, %b
|
|
ret <64 x i8> %res
|
|
}
|
|
|
|
define <64 x i8> @andn_v64i8(<64 x i8> %a, <64 x i8> %b) {
|
|
; KNL-LABEL: andn_v64i8:
|
|
; KNL: ## %bb.0:
|
|
; KNL-NEXT: vpandnq %zmm0, %zmm1, %zmm0
|
|
; KNL-NEXT: retq
|
|
;
|
|
; SKX-LABEL: andn_v64i8:
|
|
; SKX: ## %bb.0:
|
|
; SKX-NEXT: vandnps %zmm0, %zmm1, %zmm0
|
|
; SKX-NEXT: retq
|
|
%b2 = xor <64 x i8> %b, <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,
|
|
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,
|
|
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,
|
|
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>
|
|
%res = and <64 x i8> %a, %b2
|
|
ret <64 x i8> %res
|
|
}
|
|
|
|
define <64 x i8> @or_v64i8(<64 x i8> %a, <64 x i8> %b) {
|
|
; KNL-LABEL: or_v64i8:
|
|
; KNL: ## %bb.0:
|
|
; KNL-NEXT: vporq %zmm1, %zmm0, %zmm0
|
|
; KNL-NEXT: retq
|
|
;
|
|
; SKX-LABEL: or_v64i8:
|
|
; SKX: ## %bb.0:
|
|
; SKX-NEXT: vorps %zmm1, %zmm0, %zmm0
|
|
; SKX-NEXT: retq
|
|
%res = or <64 x i8> %a, %b
|
|
ret <64 x i8> %res
|
|
}
|
|
|
|
define <64 x i8> @xor_v64i8(<64 x i8> %a, <64 x i8> %b) {
|
|
; KNL-LABEL: xor_v64i8:
|
|
; KNL: ## %bb.0:
|
|
; KNL-NEXT: vpxorq %zmm1, %zmm0, %zmm0
|
|
; KNL-NEXT: retq
|
|
;
|
|
; SKX-LABEL: xor_v64i8:
|
|
; SKX: ## %bb.0:
|
|
; SKX-NEXT: vxorps %zmm1, %zmm0, %zmm0
|
|
; SKX-NEXT: retq
|
|
%res = xor <64 x i8> %a, %b
|
|
ret <64 x i8> %res
|
|
}
|
|
|
|
define <32 x i16> @and_v32i16(<32 x i16> %a, <32 x i16> %b) {
|
|
; KNL-LABEL: and_v32i16:
|
|
; KNL: ## %bb.0:
|
|
; KNL-NEXT: vpandq %zmm1, %zmm0, %zmm0
|
|
; KNL-NEXT: retq
|
|
;
|
|
; SKX-LABEL: and_v32i16:
|
|
; SKX: ## %bb.0:
|
|
; SKX-NEXT: vandps %zmm1, %zmm0, %zmm0
|
|
; SKX-NEXT: retq
|
|
%res = and <32 x i16> %a, %b
|
|
ret <32 x i16> %res
|
|
}
|
|
|
|
define <32 x i16> @andn_v32i16(<32 x i16> %a, <32 x i16> %b) {
|
|
; KNL-LABEL: andn_v32i16:
|
|
; KNL: ## %bb.0:
|
|
; KNL-NEXT: vpandnq %zmm0, %zmm1, %zmm0
|
|
; KNL-NEXT: retq
|
|
;
|
|
; SKX-LABEL: andn_v32i16:
|
|
; SKX: ## %bb.0:
|
|
; SKX-NEXT: vandnps %zmm0, %zmm1, %zmm0
|
|
; SKX-NEXT: retq
|
|
%b2 = xor <32 x i16> %b, <i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1,
|
|
i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1>
|
|
%res = and <32 x i16> %a, %b2
|
|
ret <32 x i16> %res
|
|
}
|
|
|
|
define <32 x i16> @or_v32i16(<32 x i16> %a, <32 x i16> %b) {
|
|
; KNL-LABEL: or_v32i16:
|
|
; KNL: ## %bb.0:
|
|
; KNL-NEXT: vporq %zmm1, %zmm0, %zmm0
|
|
; KNL-NEXT: retq
|
|
;
|
|
; SKX-LABEL: or_v32i16:
|
|
; SKX: ## %bb.0:
|
|
; SKX-NEXT: vorps %zmm1, %zmm0, %zmm0
|
|
; SKX-NEXT: retq
|
|
%res = or <32 x i16> %a, %b
|
|
ret <32 x i16> %res
|
|
}
|
|
|
|
define <32 x i16> @xor_v32i16(<32 x i16> %a, <32 x i16> %b) {
|
|
; KNL-LABEL: xor_v32i16:
|
|
; KNL: ## %bb.0:
|
|
; KNL-NEXT: vpxorq %zmm1, %zmm0, %zmm0
|
|
; KNL-NEXT: retq
|
|
;
|
|
; SKX-LABEL: xor_v32i16:
|
|
; SKX: ## %bb.0:
|
|
; SKX-NEXT: vxorps %zmm1, %zmm0, %zmm0
|
|
; SKX-NEXT: retq
|
|
%res = xor <32 x i16> %a, %b
|
|
ret <32 x i16> %res
|
|
}
|
|
|
|
define <16 x float> @masked_and_v16f32(<16 x float> %a, <16 x float> %b, <16 x float> %passThru, i16 %mask, <16 x float> %c) {
|
|
; KNL-LABEL: masked_and_v16f32:
|
|
; KNL: ## %bb.0:
|
|
; KNL-NEXT: kmovw %edi, %k1
|
|
; KNL-NEXT: vpandd %zmm1, %zmm0, %zmm2 {%k1}
|
|
; KNL-NEXT: vaddps %zmm2, %zmm3, %zmm0
|
|
; KNL-NEXT: retq
|
|
;
|
|
; SKX-LABEL: masked_and_v16f32:
|
|
; SKX: ## %bb.0:
|
|
; SKX-NEXT: kmovd %edi, %k1
|
|
; SKX-NEXT: vandps %zmm1, %zmm0, %zmm2 {%k1}
|
|
; SKX-NEXT: vaddps %zmm2, %zmm3, %zmm0
|
|
; SKX-NEXT: retq
|
|
%a1 = bitcast <16 x float> %a to <16 x i32>
|
|
%b1 = bitcast <16 x float> %b to <16 x i32>
|
|
%passThru1 = bitcast <16 x float> %passThru to <16 x i32>
|
|
%mask1 = bitcast i16 %mask to <16 x i1>
|
|
%op = and <16 x i32> %a1, %b1
|
|
%select = select <16 x i1> %mask1, <16 x i32> %op, <16 x i32> %passThru1
|
|
%cast = bitcast <16 x i32> %select to <16 x float>
|
|
%add = fadd <16 x float> %c, %cast
|
|
ret <16 x float> %add
|
|
}
|
|
|
|
define <16 x float> @masked_or_v16f32(<16 x float> %a, <16 x float> %b, <16 x float> %passThru, i16 %mask, <16 x float> %c) {
|
|
; KNL-LABEL: masked_or_v16f32:
|
|
; KNL: ## %bb.0:
|
|
; KNL-NEXT: kmovw %edi, %k1
|
|
; KNL-NEXT: vpandd %zmm1, %zmm0, %zmm2 {%k1}
|
|
; KNL-NEXT: vaddps %zmm2, %zmm3, %zmm0
|
|
; KNL-NEXT: retq
|
|
;
|
|
; SKX-LABEL: masked_or_v16f32:
|
|
; SKX: ## %bb.0:
|
|
; SKX-NEXT: kmovd %edi, %k1
|
|
; SKX-NEXT: vandps %zmm1, %zmm0, %zmm2 {%k1}
|
|
; SKX-NEXT: vaddps %zmm2, %zmm3, %zmm0
|
|
; SKX-NEXT: retq
|
|
%a1 = bitcast <16 x float> %a to <16 x i32>
|
|
%b1 = bitcast <16 x float> %b to <16 x i32>
|
|
%passThru1 = bitcast <16 x float> %passThru to <16 x i32>
|
|
%mask1 = bitcast i16 %mask to <16 x i1>
|
|
%op = and <16 x i32> %a1, %b1
|
|
%select = select <16 x i1> %mask1, <16 x i32> %op, <16 x i32> %passThru1
|
|
%cast = bitcast <16 x i32> %select to <16 x float>
|
|
%add = fadd <16 x float> %c, %cast
|
|
ret <16 x float> %add
|
|
}
|
|
|
|
define <16 x float> @masked_xor_v16f32(<16 x float> %a, <16 x float> %b, <16 x float> %passThru, i16 %mask, <16 x float> %c) {
|
|
; KNL-LABEL: masked_xor_v16f32:
|
|
; KNL: ## %bb.0:
|
|
; KNL-NEXT: kmovw %edi, %k1
|
|
; KNL-NEXT: vpandd %zmm1, %zmm0, %zmm2 {%k1}
|
|
; KNL-NEXT: vaddps %zmm2, %zmm3, %zmm0
|
|
; KNL-NEXT: retq
|
|
;
|
|
; SKX-LABEL: masked_xor_v16f32:
|
|
; SKX: ## %bb.0:
|
|
; SKX-NEXT: kmovd %edi, %k1
|
|
; SKX-NEXT: vandps %zmm1, %zmm0, %zmm2 {%k1}
|
|
; SKX-NEXT: vaddps %zmm2, %zmm3, %zmm0
|
|
; SKX-NEXT: retq
|
|
%a1 = bitcast <16 x float> %a to <16 x i32>
|
|
%b1 = bitcast <16 x float> %b to <16 x i32>
|
|
%passThru1 = bitcast <16 x float> %passThru to <16 x i32>
|
|
%mask1 = bitcast i16 %mask to <16 x i1>
|
|
%op = and <16 x i32> %a1, %b1
|
|
%select = select <16 x i1> %mask1, <16 x i32> %op, <16 x i32> %passThru1
|
|
%cast = bitcast <16 x i32> %select to <16 x float>
|
|
%add = fadd <16 x float> %c, %cast
|
|
ret <16 x float> %add
|
|
}
|
|
|
|
define <8 x double> @masked_and_v8f64(<8 x double> %a, <8 x double> %b, <8 x double> %passThru, i8 %mask, <8 x double> %c) {
|
|
; KNL-LABEL: masked_and_v8f64:
|
|
; KNL: ## %bb.0:
|
|
; KNL-NEXT: kmovw %edi, %k1
|
|
; KNL-NEXT: vpandq %zmm1, %zmm0, %zmm2 {%k1}
|
|
; KNL-NEXT: vaddpd %zmm2, %zmm3, %zmm0
|
|
; KNL-NEXT: retq
|
|
;
|
|
; SKX-LABEL: masked_and_v8f64:
|
|
; SKX: ## %bb.0:
|
|
; SKX-NEXT: kmovd %edi, %k1
|
|
; SKX-NEXT: vandpd %zmm1, %zmm0, %zmm2 {%k1}
|
|
; SKX-NEXT: vaddpd %zmm2, %zmm3, %zmm0
|
|
; SKX-NEXT: retq
|
|
%a1 = bitcast <8 x double> %a to <8 x i64>
|
|
%b1 = bitcast <8 x double> %b to <8 x i64>
|
|
%passThru1 = bitcast <8 x double> %passThru to <8 x i64>
|
|
%mask1 = bitcast i8 %mask to <8 x i1>
|
|
%op = and <8 x i64> %a1, %b1
|
|
%select = select <8 x i1> %mask1, <8 x i64> %op, <8 x i64> %passThru1
|
|
%cast = bitcast <8 x i64> %select to <8 x double>
|
|
%add = fadd <8 x double> %c, %cast
|
|
ret <8 x double> %add
|
|
}
|
|
|
|
define <8 x double> @masked_or_v8f64(<8 x double> %a, <8 x double> %b, <8 x double> %passThru, i8 %mask, <8 x double> %c) {
|
|
; KNL-LABEL: masked_or_v8f64:
|
|
; KNL: ## %bb.0:
|
|
; KNL-NEXT: kmovw %edi, %k1
|
|
; KNL-NEXT: vpandq %zmm1, %zmm0, %zmm2 {%k1}
|
|
; KNL-NEXT: vaddpd %zmm2, %zmm3, %zmm0
|
|
; KNL-NEXT: retq
|
|
;
|
|
; SKX-LABEL: masked_or_v8f64:
|
|
; SKX: ## %bb.0:
|
|
; SKX-NEXT: kmovd %edi, %k1
|
|
; SKX-NEXT: vandpd %zmm1, %zmm0, %zmm2 {%k1}
|
|
; SKX-NEXT: vaddpd %zmm2, %zmm3, %zmm0
|
|
; SKX-NEXT: retq
|
|
%a1 = bitcast <8 x double> %a to <8 x i64>
|
|
%b1 = bitcast <8 x double> %b to <8 x i64>
|
|
%passThru1 = bitcast <8 x double> %passThru to <8 x i64>
|
|
%mask1 = bitcast i8 %mask to <8 x i1>
|
|
%op = and <8 x i64> %a1, %b1
|
|
%select = select <8 x i1> %mask1, <8 x i64> %op, <8 x i64> %passThru1
|
|
%cast = bitcast <8 x i64> %select to <8 x double>
|
|
%add = fadd <8 x double> %c, %cast
|
|
ret <8 x double> %add
|
|
}
|
|
|
|
define <8 x double> @masked_xor_v8f64(<8 x double> %a, <8 x double> %b, <8 x double> %passThru, i8 %mask, <8 x double> %c) {
|
|
; KNL-LABEL: masked_xor_v8f64:
|
|
; KNL: ## %bb.0:
|
|
; KNL-NEXT: kmovw %edi, %k1
|
|
; KNL-NEXT: vpandq %zmm1, %zmm0, %zmm2 {%k1}
|
|
; KNL-NEXT: vaddpd %zmm2, %zmm3, %zmm0
|
|
; KNL-NEXT: retq
|
|
;
|
|
; SKX-LABEL: masked_xor_v8f64:
|
|
; SKX: ## %bb.0:
|
|
; SKX-NEXT: kmovd %edi, %k1
|
|
; SKX-NEXT: vandpd %zmm1, %zmm0, %zmm2 {%k1}
|
|
; SKX-NEXT: vaddpd %zmm2, %zmm3, %zmm0
|
|
; SKX-NEXT: retq
|
|
%a1 = bitcast <8 x double> %a to <8 x i64>
|
|
%b1 = bitcast <8 x double> %b to <8 x i64>
|
|
%passThru1 = bitcast <8 x double> %passThru to <8 x i64>
|
|
%mask1 = bitcast i8 %mask to <8 x i1>
|
|
%op = and <8 x i64> %a1, %b1
|
|
%select = select <8 x i1> %mask1, <8 x i64> %op, <8 x i64> %passThru1
|
|
%cast = bitcast <8 x i64> %select to <8 x double>
|
|
%add = fadd <8 x double> %c, %cast
|
|
ret <8 x double> %add
|
|
}
|
|
|
|
define <8 x i64> @test_mm512_mask_and_epi32(<8 x i64> %__src, i16 zeroext %__k, <8 x i64> %__a, <8 x i64> %__b) {
|
|
; KNL-LABEL: test_mm512_mask_and_epi32:
|
|
; KNL: ## %bb.0: ## %entry
|
|
; KNL-NEXT: kmovw %edi, %k1
|
|
; KNL-NEXT: vpandd %zmm2, %zmm1, %zmm0 {%k1}
|
|
; KNL-NEXT: retq
|
|
;
|
|
; SKX-LABEL: test_mm512_mask_and_epi32:
|
|
; SKX: ## %bb.0: ## %entry
|
|
; SKX-NEXT: kmovd %edi, %k1
|
|
; SKX-NEXT: vandps %zmm2, %zmm1, %zmm0 {%k1}
|
|
; SKX-NEXT: retq
|
|
entry:
|
|
%and1.i.i = and <8 x i64> %__a, %__b
|
|
%0 = bitcast <8 x i64> %and1.i.i to <16 x i32>
|
|
%1 = bitcast <8 x i64> %__src to <16 x i32>
|
|
%2 = bitcast i16 %__k to <16 x i1>
|
|
%3 = select <16 x i1> %2, <16 x i32> %0, <16 x i32> %1
|
|
%4 = bitcast <16 x i32> %3 to <8 x i64>
|
|
ret <8 x i64> %4
|
|
}
|
|
|
|
define <8 x i64> @test_mm512_mask_or_epi32(<8 x i64> %__src, i16 zeroext %__k, <8 x i64> %__a, <8 x i64> %__b) {
|
|
; KNL-LABEL: test_mm512_mask_or_epi32:
|
|
; KNL: ## %bb.0: ## %entry
|
|
; KNL-NEXT: kmovw %edi, %k1
|
|
; KNL-NEXT: vpord %zmm2, %zmm1, %zmm0 {%k1}
|
|
; KNL-NEXT: retq
|
|
;
|
|
; SKX-LABEL: test_mm512_mask_or_epi32:
|
|
; SKX: ## %bb.0: ## %entry
|
|
; SKX-NEXT: kmovd %edi, %k1
|
|
; SKX-NEXT: vorps %zmm2, %zmm1, %zmm0 {%k1}
|
|
; SKX-NEXT: retq
|
|
entry:
|
|
%or1.i.i = or <8 x i64> %__a, %__b
|
|
%0 = bitcast <8 x i64> %or1.i.i to <16 x i32>
|
|
%1 = bitcast <8 x i64> %__src to <16 x i32>
|
|
%2 = bitcast i16 %__k to <16 x i1>
|
|
%3 = select <16 x i1> %2, <16 x i32> %0, <16 x i32> %1
|
|
%4 = bitcast <16 x i32> %3 to <8 x i64>
|
|
ret <8 x i64> %4
|
|
}
|
|
|
|
define <8 x i64> @test_mm512_mask_xor_epi32(<8 x i64> %__src, i16 zeroext %__k, <8 x i64> %__a, <8 x i64> %__b) {
|
|
; KNL-LABEL: test_mm512_mask_xor_epi32:
|
|
; KNL: ## %bb.0: ## %entry
|
|
; KNL-NEXT: kmovw %edi, %k1
|
|
; KNL-NEXT: vpxord %zmm2, %zmm1, %zmm0 {%k1}
|
|
; KNL-NEXT: retq
|
|
;
|
|
; SKX-LABEL: test_mm512_mask_xor_epi32:
|
|
; SKX: ## %bb.0: ## %entry
|
|
; SKX-NEXT: kmovd %edi, %k1
|
|
; SKX-NEXT: vxorps %zmm2, %zmm1, %zmm0 {%k1}
|
|
; SKX-NEXT: retq
|
|
entry:
|
|
%xor1.i.i = xor <8 x i64> %__a, %__b
|
|
%0 = bitcast <8 x i64> %xor1.i.i to <16 x i32>
|
|
%1 = bitcast <8 x i64> %__src to <16 x i32>
|
|
%2 = bitcast i16 %__k to <16 x i1>
|
|
%3 = select <16 x i1> %2, <16 x i32> %0, <16 x i32> %1
|
|
%4 = bitcast <16 x i32> %3 to <8 x i64>
|
|
ret <8 x i64> %4
|
|
}
|
|
|
|
define <8 x double> @test_mm512_mask_xor_pd(<8 x double> %__W, i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B) {
|
|
; KNL-LABEL: test_mm512_mask_xor_pd:
|
|
; KNL: ## %bb.0: ## %entry
|
|
; KNL-NEXT: kmovw %edi, %k1
|
|
; KNL-NEXT: vpxorq %zmm2, %zmm1, %zmm0 {%k1}
|
|
; KNL-NEXT: retq
|
|
;
|
|
; SKX-LABEL: test_mm512_mask_xor_pd:
|
|
; SKX: ## %bb.0: ## %entry
|
|
; SKX-NEXT: kmovd %edi, %k1
|
|
; SKX-NEXT: vxorpd %zmm2, %zmm1, %zmm0 {%k1}
|
|
; SKX-NEXT: retq
|
|
entry:
|
|
%0 = bitcast <8 x double> %__A to <8 x i64>
|
|
%1 = bitcast <8 x double> %__B to <8 x i64>
|
|
%xor.i.i = xor <8 x i64> %0, %1
|
|
%2 = bitcast <8 x i64> %xor.i.i to <8 x double>
|
|
%3 = bitcast i8 %__U to <8 x i1>
|
|
%4 = select <8 x i1> %3, <8 x double> %2, <8 x double> %__W
|
|
ret <8 x double> %4
|
|
}
|
|
|
|
define <8 x double> @test_mm512_maskz_xor_pd(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B) {
|
|
; KNL-LABEL: test_mm512_maskz_xor_pd:
|
|
; KNL: ## %bb.0: ## %entry
|
|
; KNL-NEXT: kmovw %edi, %k1
|
|
; KNL-NEXT: vpxorq %zmm1, %zmm0, %zmm0 {%k1} {z}
|
|
; KNL-NEXT: retq
|
|
;
|
|
; SKX-LABEL: test_mm512_maskz_xor_pd:
|
|
; SKX: ## %bb.0: ## %entry
|
|
; SKX-NEXT: kmovd %edi, %k1
|
|
; SKX-NEXT: vxorpd %zmm1, %zmm0, %zmm0 {%k1} {z}
|
|
; SKX-NEXT: retq
|
|
entry:
|
|
%0 = bitcast <8 x double> %__A to <8 x i64>
|
|
%1 = bitcast <8 x double> %__B to <8 x i64>
|
|
%xor.i.i = xor <8 x i64> %0, %1
|
|
%2 = bitcast <8 x i64> %xor.i.i to <8 x double>
|
|
%3 = bitcast i8 %__U to <8 x i1>
|
|
%4 = select <8 x i1> %3, <8 x double> %2, <8 x double> zeroinitializer
|
|
ret <8 x double> %4
|
|
}
|
|
|
|
define <16 x float> @test_mm512_mask_xor_ps(<16 x float> %__W, i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B) {
|
|
; KNL-LABEL: test_mm512_mask_xor_ps:
|
|
; KNL: ## %bb.0: ## %entry
|
|
; KNL-NEXT: kmovw %edi, %k1
|
|
; KNL-NEXT: vpxord %zmm2, %zmm1, %zmm0 {%k1}
|
|
; KNL-NEXT: retq
|
|
;
|
|
; SKX-LABEL: test_mm512_mask_xor_ps:
|
|
; SKX: ## %bb.0: ## %entry
|
|
; SKX-NEXT: kmovd %edi, %k1
|
|
; SKX-NEXT: vxorps %zmm2, %zmm1, %zmm0 {%k1}
|
|
; SKX-NEXT: retq
|
|
entry:
|
|
%0 = bitcast <16 x float> %__A to <16 x i32>
|
|
%1 = bitcast <16 x float> %__B to <16 x i32>
|
|
%xor.i.i = xor <16 x i32> %0, %1
|
|
%2 = bitcast <16 x i32> %xor.i.i to <16 x float>
|
|
%3 = bitcast i16 %__U to <16 x i1>
|
|
%4 = select <16 x i1> %3, <16 x float> %2, <16 x float> %__W
|
|
ret <16 x float> %4
|
|
}
|
|
|
|
define <16 x float> @test_mm512_maskz_xor_ps(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B) {
|
|
; KNL-LABEL: test_mm512_maskz_xor_ps:
|
|
; KNL: ## %bb.0: ## %entry
|
|
; KNL-NEXT: kmovw %edi, %k1
|
|
; KNL-NEXT: vpxord %zmm1, %zmm0, %zmm0 {%k1} {z}
|
|
; KNL-NEXT: retq
|
|
;
|
|
; SKX-LABEL: test_mm512_maskz_xor_ps:
|
|
; SKX: ## %bb.0: ## %entry
|
|
; SKX-NEXT: kmovd %edi, %k1
|
|
; SKX-NEXT: vxorps %zmm1, %zmm0, %zmm0 {%k1} {z}
|
|
; SKX-NEXT: retq
|
|
entry:
|
|
%0 = bitcast <16 x float> %__A to <16 x i32>
|
|
%1 = bitcast <16 x float> %__B to <16 x i32>
|
|
%xor.i.i = xor <16 x i32> %0, %1
|
|
%2 = bitcast <16 x i32> %xor.i.i to <16 x float>
|
|
%3 = bitcast i16 %__U to <16 x i1>
|
|
%4 = select <16 x i1> %3, <16 x float> %2, <16 x float> zeroinitializer
|
|
ret <16 x float> %4
|
|
}
|
|
|
|
define <8 x double> @test_mm512_mask_or_pd(<8 x double> %__W, i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B) {
|
|
; KNL-LABEL: test_mm512_mask_or_pd:
|
|
; KNL: ## %bb.0: ## %entry
|
|
; KNL-NEXT: kmovw %edi, %k1
|
|
; KNL-NEXT: vporq %zmm1, %zmm2, %zmm0 {%k1}
|
|
; KNL-NEXT: retq
|
|
;
|
|
; SKX-LABEL: test_mm512_mask_or_pd:
|
|
; SKX: ## %bb.0: ## %entry
|
|
; SKX-NEXT: kmovd %edi, %k1
|
|
; SKX-NEXT: vorpd %zmm1, %zmm2, %zmm0 {%k1}
|
|
; SKX-NEXT: retq
|
|
entry:
|
|
%0 = bitcast <8 x double> %__A to <8 x i64>
|
|
%1 = bitcast <8 x double> %__B to <8 x i64>
|
|
%or.i.i = or <8 x i64> %1, %0
|
|
%2 = bitcast <8 x i64> %or.i.i to <8 x double>
|
|
%3 = bitcast i8 %__U to <8 x i1>
|
|
%4 = select <8 x i1> %3, <8 x double> %2, <8 x double> %__W
|
|
ret <8 x double> %4
|
|
}
|
|
|
|
define <8 x double> @test_mm512_maskz_or_pd(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B) {
|
|
; KNL-LABEL: test_mm512_maskz_or_pd:
|
|
; KNL: ## %bb.0: ## %entry
|
|
; KNL-NEXT: kmovw %edi, %k1
|
|
; KNL-NEXT: vporq %zmm0, %zmm1, %zmm0 {%k1} {z}
|
|
; KNL-NEXT: retq
|
|
;
|
|
; SKX-LABEL: test_mm512_maskz_or_pd:
|
|
; SKX: ## %bb.0: ## %entry
|
|
; SKX-NEXT: kmovd %edi, %k1
|
|
; SKX-NEXT: vorpd %zmm0, %zmm1, %zmm0 {%k1} {z}
|
|
; SKX-NEXT: retq
|
|
entry:
|
|
%0 = bitcast <8 x double> %__A to <8 x i64>
|
|
%1 = bitcast <8 x double> %__B to <8 x i64>
|
|
%or.i.i = or <8 x i64> %1, %0
|
|
%2 = bitcast <8 x i64> %or.i.i to <8 x double>
|
|
%3 = bitcast i8 %__U to <8 x i1>
|
|
%4 = select <8 x i1> %3, <8 x double> %2, <8 x double> zeroinitializer
|
|
ret <8 x double> %4
|
|
}
|
|
|
|
define <16 x float> @test_mm512_mask_or_ps(<16 x float> %__W, i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B) {
|
|
; KNL-LABEL: test_mm512_mask_or_ps:
|
|
; KNL: ## %bb.0: ## %entry
|
|
; KNL-NEXT: kmovw %edi, %k1
|
|
; KNL-NEXT: vpord %zmm1, %zmm2, %zmm0 {%k1}
|
|
; KNL-NEXT: retq
|
|
;
|
|
; SKX-LABEL: test_mm512_mask_or_ps:
|
|
; SKX: ## %bb.0: ## %entry
|
|
; SKX-NEXT: kmovd %edi, %k1
|
|
; SKX-NEXT: vorps %zmm1, %zmm2, %zmm0 {%k1}
|
|
; SKX-NEXT: retq
|
|
entry:
|
|
%0 = bitcast <16 x float> %__A to <16 x i32>
|
|
%1 = bitcast <16 x float> %__B to <16 x i32>
|
|
%or.i.i = or <16 x i32> %1, %0
|
|
%2 = bitcast <16 x i32> %or.i.i to <16 x float>
|
|
%3 = bitcast i16 %__U to <16 x i1>
|
|
%4 = select <16 x i1> %3, <16 x float> %2, <16 x float> %__W
|
|
ret <16 x float> %4
|
|
}
|
|
|
|
define <16 x float> @test_mm512_maskz_or_ps(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B) {
|
|
; KNL-LABEL: test_mm512_maskz_or_ps:
|
|
; KNL: ## %bb.0: ## %entry
|
|
; KNL-NEXT: kmovw %edi, %k1
|
|
; KNL-NEXT: vpord %zmm0, %zmm1, %zmm0 {%k1} {z}
|
|
; KNL-NEXT: retq
|
|
;
|
|
; SKX-LABEL: test_mm512_maskz_or_ps:
|
|
; SKX: ## %bb.0: ## %entry
|
|
; SKX-NEXT: kmovd %edi, %k1
|
|
; SKX-NEXT: vorps %zmm0, %zmm1, %zmm0 {%k1} {z}
|
|
; SKX-NEXT: retq
|
|
entry:
|
|
%0 = bitcast <16 x float> %__A to <16 x i32>
|
|
%1 = bitcast <16 x float> %__B to <16 x i32>
|
|
%or.i.i = or <16 x i32> %1, %0
|
|
%2 = bitcast <16 x i32> %or.i.i to <16 x float>
|
|
%3 = bitcast i16 %__U to <16 x i1>
|
|
%4 = select <16 x i1> %3, <16 x float> %2, <16 x float> zeroinitializer
|
|
ret <16 x float> %4
|
|
}
|
|
|
|
define <8 x double> @test_mm512_mask_and_pd(<8 x double> %__W, i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B) {
|
|
; KNL-LABEL: test_mm512_mask_and_pd:
|
|
; KNL: ## %bb.0: ## %entry
|
|
; KNL-NEXT: kmovw %edi, %k1
|
|
; KNL-NEXT: vpandq %zmm1, %zmm2, %zmm0 {%k1}
|
|
; KNL-NEXT: retq
|
|
;
|
|
; SKX-LABEL: test_mm512_mask_and_pd:
|
|
; SKX: ## %bb.0: ## %entry
|
|
; SKX-NEXT: kmovd %edi, %k1
|
|
; SKX-NEXT: vandpd %zmm1, %zmm2, %zmm0 {%k1}
|
|
; SKX-NEXT: retq
|
|
entry:
|
|
%0 = bitcast <8 x double> %__A to <8 x i64>
|
|
%1 = bitcast <8 x double> %__B to <8 x i64>
|
|
%and.i.i = and <8 x i64> %1, %0
|
|
%2 = bitcast <8 x i64> %and.i.i to <8 x double>
|
|
%3 = bitcast i8 %__U to <8 x i1>
|
|
%4 = select <8 x i1> %3, <8 x double> %2, <8 x double> %__W
|
|
ret <8 x double> %4
|
|
}
|
|
|
|
define <8 x double> @test_mm512_maskz_and_pd(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B) {
|
|
; KNL-LABEL: test_mm512_maskz_and_pd:
|
|
; KNL: ## %bb.0: ## %entry
|
|
; KNL-NEXT: kmovw %edi, %k1
|
|
; KNL-NEXT: vpandq %zmm0, %zmm1, %zmm0 {%k1} {z}
|
|
; KNL-NEXT: retq
|
|
;
|
|
; SKX-LABEL: test_mm512_maskz_and_pd:
|
|
; SKX: ## %bb.0: ## %entry
|
|
; SKX-NEXT: kmovd %edi, %k1
|
|
; SKX-NEXT: vandpd %zmm0, %zmm1, %zmm0 {%k1} {z}
|
|
; SKX-NEXT: retq
|
|
entry:
|
|
%0 = bitcast <8 x double> %__A to <8 x i64>
|
|
%1 = bitcast <8 x double> %__B to <8 x i64>
|
|
%and.i.i = and <8 x i64> %1, %0
|
|
%2 = bitcast <8 x i64> %and.i.i to <8 x double>
|
|
%3 = bitcast i8 %__U to <8 x i1>
|
|
%4 = select <8 x i1> %3, <8 x double> %2, <8 x double> zeroinitializer
|
|
ret <8 x double> %4
|
|
}
|
|
|
|
define <16 x float> @test_mm512_mask_and_ps(<16 x float> %__W, i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B) {
|
|
; KNL-LABEL: test_mm512_mask_and_ps:
|
|
; KNL: ## %bb.0: ## %entry
|
|
; KNL-NEXT: kmovw %edi, %k1
|
|
; KNL-NEXT: vpandd %zmm1, %zmm2, %zmm0 {%k1}
|
|
; KNL-NEXT: retq
|
|
;
|
|
; SKX-LABEL: test_mm512_mask_and_ps:
|
|
; SKX: ## %bb.0: ## %entry
|
|
; SKX-NEXT: kmovd %edi, %k1
|
|
; SKX-NEXT: vandps %zmm1, %zmm2, %zmm0 {%k1}
|
|
; SKX-NEXT: retq
|
|
entry:
|
|
%0 = bitcast <16 x float> %__A to <16 x i32>
|
|
%1 = bitcast <16 x float> %__B to <16 x i32>
|
|
%and.i.i = and <16 x i32> %1, %0
|
|
%2 = bitcast <16 x i32> %and.i.i to <16 x float>
|
|
%3 = bitcast i16 %__U to <16 x i1>
|
|
%4 = select <16 x i1> %3, <16 x float> %2, <16 x float> %__W
|
|
ret <16 x float> %4
|
|
}
|
|
|
|
define <16 x float> @test_mm512_maskz_and_ps(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B) {
|
|
; KNL-LABEL: test_mm512_maskz_and_ps:
|
|
; KNL: ## %bb.0: ## %entry
|
|
; KNL-NEXT: kmovw %edi, %k1
|
|
; KNL-NEXT: vpandd %zmm0, %zmm1, %zmm0 {%k1} {z}
|
|
; KNL-NEXT: retq
|
|
;
|
|
; SKX-LABEL: test_mm512_maskz_and_ps:
|
|
; SKX: ## %bb.0: ## %entry
|
|
; SKX-NEXT: kmovd %edi, %k1
|
|
; SKX-NEXT: vandps %zmm0, %zmm1, %zmm0 {%k1} {z}
|
|
; SKX-NEXT: retq
|
|
entry:
|
|
%0 = bitcast <16 x float> %__A to <16 x i32>
|
|
%1 = bitcast <16 x float> %__B to <16 x i32>
|
|
%and.i.i = and <16 x i32> %1, %0
|
|
%2 = bitcast <16 x i32> %and.i.i to <16 x float>
|
|
%3 = bitcast i16 %__U to <16 x i1>
|
|
%4 = select <16 x i1> %3, <16 x float> %2, <16 x float> zeroinitializer
|
|
ret <16 x float> %4
|
|
}
|
|
|
|
define <8 x double> @test_mm512_mask_andnot_pd(<8 x double> %__W, i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B) {
|
|
; KNL-LABEL: test_mm512_mask_andnot_pd:
|
|
; KNL: ## %bb.0: ## %entry
|
|
; KNL-NEXT: kmovw %edi, %k1
|
|
; KNL-NEXT: vpandnq %zmm2, %zmm1, %zmm0 {%k1}
|
|
; KNL-NEXT: retq
|
|
;
|
|
; SKX-LABEL: test_mm512_mask_andnot_pd:
|
|
; SKX: ## %bb.0: ## %entry
|
|
; SKX-NEXT: kmovd %edi, %k1
|
|
; SKX-NEXT: vandnpd %zmm2, %zmm1, %zmm0 {%k1}
|
|
; SKX-NEXT: retq
|
|
entry:
|
|
%0 = bitcast <8 x double> %__A to <8 x i64>
|
|
%neg.i.i = xor <8 x i64> %0, <i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 -1>
|
|
%1 = bitcast <8 x double> %__B to <8 x i64>
|
|
%and.i.i = and <8 x i64> %1, %neg.i.i
|
|
%2 = bitcast <8 x i64> %and.i.i to <8 x double>
|
|
%3 = bitcast i8 %__U to <8 x i1>
|
|
%4 = select <8 x i1> %3, <8 x double> %2, <8 x double> %__W
|
|
ret <8 x double> %4
|
|
}
|
|
|
|
define <8 x double> @test_mm512_maskz_andnot_pd(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B) {
|
|
; KNL-LABEL: test_mm512_maskz_andnot_pd:
|
|
; KNL: ## %bb.0: ## %entry
|
|
; KNL-NEXT: kmovw %edi, %k1
|
|
; KNL-NEXT: vpandnq %zmm1, %zmm0, %zmm0 {%k1} {z}
|
|
; KNL-NEXT: retq
|
|
;
|
|
; SKX-LABEL: test_mm512_maskz_andnot_pd:
|
|
; SKX: ## %bb.0: ## %entry
|
|
; SKX-NEXT: kmovd %edi, %k1
|
|
; SKX-NEXT: vandnpd %zmm1, %zmm0, %zmm0 {%k1} {z}
|
|
; SKX-NEXT: retq
|
|
entry:
|
|
%0 = bitcast <8 x double> %__A to <8 x i64>
|
|
%neg.i.i = xor <8 x i64> %0, <i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 -1>
|
|
%1 = bitcast <8 x double> %__B to <8 x i64>
|
|
%and.i.i = and <8 x i64> %1, %neg.i.i
|
|
%2 = bitcast <8 x i64> %and.i.i to <8 x double>
|
|
%3 = bitcast i8 %__U to <8 x i1>
|
|
%4 = select <8 x i1> %3, <8 x double> %2, <8 x double> zeroinitializer
|
|
ret <8 x double> %4
|
|
}
|
|
|
|
define <16 x float> @test_mm512_mask_andnot_ps(<16 x float> %__W, i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B) {
|
|
; KNL-LABEL: test_mm512_mask_andnot_ps:
|
|
; KNL: ## %bb.0: ## %entry
|
|
; KNL-NEXT: kmovw %edi, %k1
|
|
; KNL-NEXT: vpandnd %zmm2, %zmm1, %zmm0 {%k1}
|
|
; KNL-NEXT: retq
|
|
;
|
|
; SKX-LABEL: test_mm512_mask_andnot_ps:
|
|
; SKX: ## %bb.0: ## %entry
|
|
; SKX-NEXT: kmovd %edi, %k1
|
|
; SKX-NEXT: vandnps %zmm2, %zmm1, %zmm0 {%k1}
|
|
; SKX-NEXT: retq
|
|
entry:
|
|
%0 = bitcast <16 x float> %__A to <16 x i32>
|
|
%neg.i.i = xor <16 x i32> %0, <i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1>
|
|
%1 = bitcast <16 x float> %__B to <16 x i32>
|
|
%and.i.i = and <16 x i32> %1, %neg.i.i
|
|
%2 = bitcast <16 x i32> %and.i.i to <16 x float>
|
|
%3 = bitcast i16 %__U to <16 x i1>
|
|
%4 = select <16 x i1> %3, <16 x float> %2, <16 x float> %__W
|
|
ret <16 x float> %4
|
|
}
|
|
|
|
define <16 x float> @test_mm512_maskz_andnot_ps(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B) {
|
|
; KNL-LABEL: test_mm512_maskz_andnot_ps:
|
|
; KNL: ## %bb.0: ## %entry
|
|
; KNL-NEXT: kmovw %edi, %k1
|
|
; KNL-NEXT: vpandnd %zmm1, %zmm0, %zmm0 {%k1} {z}
|
|
; KNL-NEXT: retq
|
|
;
|
|
; SKX-LABEL: test_mm512_maskz_andnot_ps:
|
|
; SKX: ## %bb.0: ## %entry
|
|
; SKX-NEXT: kmovd %edi, %k1
|
|
; SKX-NEXT: vandnps %zmm1, %zmm0, %zmm0 {%k1} {z}
|
|
; SKX-NEXT: retq
|
|
entry:
|
|
%0 = bitcast <16 x float> %__A to <16 x i32>
|
|
%neg.i.i = xor <16 x i32> %0, <i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1>
|
|
%1 = bitcast <16 x float> %__B to <16 x i32>
|
|
%and.i.i = and <16 x i32> %1, %neg.i.i
|
|
%2 = bitcast <16 x i32> %and.i.i to <16 x float>
|
|
%3 = bitcast i16 %__U to <16 x i1>
|
|
%4 = select <16 x i1> %3, <16 x float> %2, <16 x float> zeroinitializer
|
|
ret <16 x float> %4
|
|
}
|
|
|
|
define <16 x i32> @ternlog_and_andn(<16 x i32> %x, <16 x i32> %y, <16 x i32> %z) {
|
|
; ALL-LABEL: ternlog_and_andn:
|
|
; ALL: ## %bb.0:
|
|
; ALL-NEXT: vpternlogd {{.*#+}} zmm0 = zmm2 & zmm1 & ~zmm0
|
|
; ALL-NEXT: retq
|
|
%a = xor <16 x i32> %x, <i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1>
|
|
%b = and <16 x i32> %y, %a
|
|
%c = and <16 x i32> %b, %z
|
|
ret <16 x i32> %c
|
|
}
|
|
|
|
define <16 x i32> @ternlog_or_andn(<16 x i32> %x, <16 x i32> %y, <16 x i32> %z) {
|
|
; ALL-LABEL: ternlog_or_andn:
|
|
; ALL: ## %bb.0:
|
|
; ALL-NEXT: vpternlogd {{.*#+}} zmm0 = (zmm1 & ~zmm0) | zmm2
|
|
; ALL-NEXT: retq
|
|
%a = xor <16 x i32> %x, <i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1>
|
|
%b = and <16 x i32> %y, %a
|
|
%c = or <16 x i32> %b, %z
|
|
ret <16 x i32> %c
|
|
}
|
|
|
|
define <16 x i32> @ternlog_xor_andn(<16 x i32> %x, <16 x i32> %y, <16 x i32> %z) {
|
|
; ALL-LABEL: ternlog_xor_andn:
|
|
; ALL: ## %bb.0:
|
|
; ALL-NEXT: vpternlogd {{.*#+}} zmm0 = zmm2 ^ (zmm1 & ~zmm0)
|
|
; ALL-NEXT: retq
|
|
%a = xor <16 x i32> %x, <i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1>
|
|
%b = and <16 x i32> %y, %a
|
|
%c = xor <16 x i32> %b, %z
|
|
ret <16 x i32> %c
|
|
}
|
|
|
|
define <16 x i32> @ternlog_or_and_mask(<16 x i32> %x, <16 x i32> %y) {
|
|
; ALL-LABEL: ternlog_or_and_mask:
|
|
; ALL: ## %bb.0:
|
|
; ALL-NEXT: vpternlogd {{.*#+}} zmm0 = (zmm0 & m32bcst) | zmm1
|
|
; ALL-NEXT: retq
|
|
%a = and <16 x i32> %x, <i32 255, i32 255, i32 255, i32 255, i32 255, i32 255, i32 255, i32 255, i32 255, i32 255, i32 255, i32 255, i32 255, i32 255, i32 255, i32 255>
|
|
%b = or <16 x i32> %a, %y
|
|
ret <16 x i32> %b
|
|
}
|
|
|
|
define <8 x i64> @ternlog_xor_and_mask(<8 x i64> %x, <8 x i64> %y) {
|
|
; ALL-LABEL: ternlog_xor_and_mask:
|
|
; ALL: ## %bb.0:
|
|
; ALL-NEXT: vpternlogq {{.*#+}} zmm0 = zmm1 ^ (zmm0 & m64bcst)
|
|
; ALL-NEXT: retq
|
|
%a = and <8 x i64> %x, <i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295>
|
|
%b = xor <8 x i64> %a, %y
|
|
ret <8 x i64> %b
|
|
}
|
|
|
|
define <16 x i32> @ternlog_maskz_or_and_mask(<16 x i32> %x, <16 x i32> %y, <16 x i32> %mask) {
|
|
; ALL-LABEL: ternlog_maskz_or_and_mask:
|
|
; ALL: ## %bb.0:
|
|
; ALL-NEXT: vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to16}, %zmm0, %zmm3
|
|
; ALL-NEXT: vpsrad $31, %zmm2, %zmm0
|
|
; ALL-NEXT: vpternlogd {{.*#+}} zmm0 = zmm0 & (zmm3 | zmm1)
|
|
; ALL-NEXT: retq
|
|
%m = icmp slt <16 x i32> %mask, zeroinitializer
|
|
%a = and <16 x i32> %x, <i32 255, i32 255, i32 255, i32 255, i32 255, i32 255, i32 255, i32 255, i32 255, i32 255, i32 255, i32 255, i32 255, i32 255, i32 255, i32 255>
|
|
%b = or <16 x i32> %a, %y
|
|
%c = select <16 x i1> %m, <16 x i32> %b, <16 x i32> zeroinitializer
|
|
ret <16 x i32> %c
|
|
}
|
|
|
|
define <8 x i64> @ternlog_maskz_xor_and_mask(<8 x i64> %x, <8 x i64> %y, <8 x i64> %mask) {
|
|
; ALL-LABEL: ternlog_maskz_xor_and_mask:
|
|
; ALL: ## %bb.0:
|
|
; ALL-NEXT: vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm0, %zmm3
|
|
; ALL-NEXT: vpsraq $63, %zmm2, %zmm0
|
|
; ALL-NEXT: vpternlogq {{.*#+}} zmm0 = zmm0 & (zmm3 ^ zmm1)
|
|
; ALL-NEXT: retq
|
|
%m = icmp slt <8 x i64> %mask, zeroinitializer
|
|
%a = and <8 x i64> %x, <i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295>
|
|
%b = xor <8 x i64> %a, %y
|
|
%c = select <8 x i1> %m, <8 x i64> %b, <8 x i64> zeroinitializer
|
|
ret <8 x i64> %c
|
|
}
|
|
|
|
define <16 x i32> @ternlog_maskx_or_and_mask(<16 x i32> %x, <16 x i32> %y, <16 x i32> %mask) {
|
|
; KNL-LABEL: ternlog_maskx_or_and_mask:
|
|
; KNL: ## %bb.0:
|
|
; KNL-NEXT: vpxor %xmm3, %xmm3, %xmm3
|
|
; KNL-NEXT: vpcmpgtd %zmm2, %zmm3, %k1
|
|
; KNL-NEXT: vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to16}, %zmm0, %zmm2
|
|
; KNL-NEXT: vpord %zmm1, %zmm2, %zmm0 {%k1}
|
|
; KNL-NEXT: retq
|
|
;
|
|
; SKX-LABEL: ternlog_maskx_or_and_mask:
|
|
; SKX: ## %bb.0:
|
|
; SKX-NEXT: vpmovd2m %zmm2, %k1
|
|
; SKX-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to16}, %zmm0, %zmm2
|
|
; SKX-NEXT: vorps %zmm1, %zmm2, %zmm0 {%k1}
|
|
; SKX-NEXT: retq
|
|
%m = icmp slt <16 x i32> %mask, zeroinitializer
|
|
%a = and <16 x i32> %x, <i32 255, i32 255, i32 255, i32 255, i32 255, i32 255, i32 255, i32 255, i32 255, i32 255, i32 255, i32 255, i32 255, i32 255, i32 255, i32 255>
|
|
%b = or <16 x i32> %a, %y
|
|
%c = select <16 x i1> %m, <16 x i32> %b, <16 x i32> %x
|
|
ret <16 x i32> %c
|
|
}
|
|
|
|
define <16 x i32> @ternlog_masky_or_and_mask(<16 x i32> %x, <16 x i32> %y, <16 x i32> %mask) {
|
|
; KNL-LABEL: ternlog_masky_or_and_mask:
|
|
; KNL: ## %bb.0:
|
|
; KNL-NEXT: vpxor %xmm3, %xmm3, %xmm3
|
|
; KNL-NEXT: vpcmpgtd %zmm2, %zmm3, %k1
|
|
; KNL-NEXT: vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to16}, %zmm0, %zmm0
|
|
; KNL-NEXT: vpord %zmm1, %zmm0, %zmm1 {%k1}
|
|
; KNL-NEXT: vmovdqa64 %zmm1, %zmm0
|
|
; KNL-NEXT: retq
|
|
;
|
|
; SKX-LABEL: ternlog_masky_or_and_mask:
|
|
; SKX: ## %bb.0:
|
|
; SKX-NEXT: vpmovd2m %zmm2, %k1
|
|
; SKX-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to16}, %zmm0, %zmm0
|
|
; SKX-NEXT: vorps %zmm1, %zmm0, %zmm1 {%k1}
|
|
; SKX-NEXT: vmovaps %zmm1, %zmm0
|
|
; SKX-NEXT: retq
|
|
%m = icmp slt <16 x i32> %mask, zeroinitializer
|
|
%a = and <16 x i32> %x, <i32 255, i32 255, i32 255, i32 255, i32 255, i32 255, i32 255, i32 255, i32 255, i32 255, i32 255, i32 255, i32 255, i32 255, i32 255, i32 255>
|
|
%b = or <16 x i32> %a, %y
|
|
%c = select <16 x i1> %m, <16 x i32> %b, <16 x i32> %y
|
|
ret <16 x i32> %c
|
|
}
|
|
|
|
define <8 x i64> @ternlog_maskx_xor_and_mask(<8 x i64> %x, <8 x i64> %y, <8 x i64> %mask) {
|
|
; KNL-LABEL: ternlog_maskx_xor_and_mask:
|
|
; KNL: ## %bb.0:
|
|
; KNL-NEXT: vpxor %xmm3, %xmm3, %xmm3
|
|
; KNL-NEXT: vpcmpgtq %zmm2, %zmm3, %k1
|
|
; KNL-NEXT: vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm0, %zmm2
|
|
; KNL-NEXT: vpxorq %zmm1, %zmm2, %zmm0 {%k1}
|
|
; KNL-NEXT: retq
|
|
;
|
|
; SKX-LABEL: ternlog_maskx_xor_and_mask:
|
|
; SKX: ## %bb.0:
|
|
; SKX-NEXT: vpmovq2m %zmm2, %k1
|
|
; SKX-NEXT: vandpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm0, %zmm2
|
|
; SKX-NEXT: vxorpd %zmm1, %zmm2, %zmm0 {%k1}
|
|
; SKX-NEXT: retq
|
|
%m = icmp slt <8 x i64> %mask, zeroinitializer
|
|
%a = and <8 x i64> %x, <i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295>
|
|
%b = xor <8 x i64> %a, %y
|
|
%c = select <8 x i1> %m, <8 x i64> %b, <8 x i64> %x
|
|
ret <8 x i64> %c
|
|
}
|
|
|
|
define <8 x i64> @ternlog_masky_xor_and_mask(<8 x i64> %x, <8 x i64> %y, <8 x i64> %mask) {
|
|
; KNL-LABEL: ternlog_masky_xor_and_mask:
|
|
; KNL: ## %bb.0:
|
|
; KNL-NEXT: vpxor %xmm3, %xmm3, %xmm3
|
|
; KNL-NEXT: vpcmpgtq %zmm2, %zmm3, %k1
|
|
; KNL-NEXT: vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm0, %zmm0
|
|
; KNL-NEXT: vpxorq %zmm1, %zmm0, %zmm1 {%k1}
|
|
; KNL-NEXT: vmovdqa64 %zmm1, %zmm0
|
|
; KNL-NEXT: retq
|
|
;
|
|
; SKX-LABEL: ternlog_masky_xor_and_mask:
|
|
; SKX: ## %bb.0:
|
|
; SKX-NEXT: vpmovq2m %zmm2, %k1
|
|
; SKX-NEXT: vandpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm0, %zmm0
|
|
; SKX-NEXT: vxorpd %zmm1, %zmm0, %zmm1 {%k1}
|
|
; SKX-NEXT: vmovapd %zmm1, %zmm0
|
|
; SKX-NEXT: retq
|
|
%m = icmp slt <8 x i64> %mask, zeroinitializer
|
|
%a = and <8 x i64> %x, <i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295, i64 4294967295>
|
|
%b = xor <8 x i64> %a, %y
|
|
%c = select <8 x i1> %m, <8 x i64> %b, <8 x i64> %y
|
|
ret <8 x i64> %c
|
|
}
|