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
1466 lines
64 KiB
LLVM
1466 lines
64 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
|
|
; RUN: llc -mtriple=aarch64-linux-unknown-gnu %s -o - -mattr=+sve | FileCheck %s --check-prefix=CHECK-SVE
|
|
; RUN: llc -mtriple=aarch64-linux-unknown-gnu %s -o - -mattr=+sve,sve-aes | FileCheck %s --check-prefix=CHECK-SVE-AES
|
|
; RUN: llc -mtriple=aarch64-linux-unknown-gnu %s -o - -mattr=+sme -force-streaming | FileCheck %s --check-prefix=CHECK-SME-STREAMING
|
|
; RUN: llc -mtriple=aarch64-linux-unknown-gnu %s -o - -mattr=+sme,ssve-aes -force-streaming | FileCheck %s --check-prefix=CHECK-SME-STREAMING-SSVE-AES
|
|
; RUN: llc -mtriple=aarch64-linux-unknown-gnu %s -o - -mattr=+sve2 | FileCheck %s --check-prefix=CHECK-SVE2
|
|
; RUN: llc -mtriple=aarch64-linux-unknown-gnu %s -o - -mattr=+sve2-aes | FileCheck %s --check-prefix=CHECK-SVE2-AES
|
|
|
|
define <vscale x 16 x i8> @clmul_nxv16i8(<vscale x 16 x i8> %x, <vscale x 16 x i8> %y) {
|
|
; CHECK-SVE-LABEL: clmul_nxv16i8:
|
|
; CHECK-SVE: // %bb.0:
|
|
; CHECK-SVE-NEXT: movprfx z2, z1
|
|
; CHECK-SVE-NEXT: and z2.b, z2.b, #0x2
|
|
; CHECK-SVE-NEXT: movprfx z3, z1
|
|
; CHECK-SVE-NEXT: and z3.b, z3.b, #0x1
|
|
; CHECK-SVE-NEXT: movprfx z4, z1
|
|
; CHECK-SVE-NEXT: and z4.b, z4.b, #0x4
|
|
; CHECK-SVE-NEXT: movprfx z5, z1
|
|
; CHECK-SVE-NEXT: and z5.b, z5.b, #0x8
|
|
; CHECK-SVE-NEXT: movprfx z6, z1
|
|
; CHECK-SVE-NEXT: and z6.b, z6.b, #0x10
|
|
; CHECK-SVE-NEXT: movprfx z7, z1
|
|
; CHECK-SVE-NEXT: and z7.b, z7.b, #0x20
|
|
; CHECK-SVE-NEXT: ptrue p0.b
|
|
; CHECK-SVE-NEXT: movprfx z24, z1
|
|
; CHECK-SVE-NEXT: and z24.b, z24.b, #0x40
|
|
; CHECK-SVE-NEXT: and z1.b, z1.b, #0x80
|
|
; CHECK-SVE-NEXT: mul z2.b, p0/m, z2.b, z0.b
|
|
; CHECK-SVE-NEXT: mul z3.b, p0/m, z3.b, z0.b
|
|
; CHECK-SVE-NEXT: mul z4.b, p0/m, z4.b, z0.b
|
|
; CHECK-SVE-NEXT: mul z5.b, p0/m, z5.b, z0.b
|
|
; CHECK-SVE-NEXT: mul z6.b, p0/m, z6.b, z0.b
|
|
; CHECK-SVE-NEXT: mul z7.b, p0/m, z7.b, z0.b
|
|
; CHECK-SVE-NEXT: mul z24.b, p0/m, z24.b, z0.b
|
|
; CHECK-SVE-NEXT: mul z0.b, p0/m, z0.b, z1.b
|
|
; CHECK-SVE-NEXT: eor z2.d, z3.d, z2.d
|
|
; CHECK-SVE-NEXT: eor z3.d, z4.d, z5.d
|
|
; CHECK-SVE-NEXT: eor z4.d, z6.d, z7.d
|
|
; CHECK-SVE-NEXT: eor z2.d, z2.d, z3.d
|
|
; CHECK-SVE-NEXT: eor z3.d, z4.d, z24.d
|
|
; CHECK-SVE-NEXT: eor z1.d, z2.d, z3.d
|
|
; CHECK-SVE-NEXT: eor z0.d, z1.d, z0.d
|
|
; CHECK-SVE-NEXT: ret
|
|
;
|
|
; CHECK-SVE-AES-LABEL: clmul_nxv16i8:
|
|
; CHECK-SVE-AES: // %bb.0:
|
|
; CHECK-SVE-AES-NEXT: movprfx z2, z1
|
|
; CHECK-SVE-AES-NEXT: and z2.b, z2.b, #0x2
|
|
; CHECK-SVE-AES-NEXT: movprfx z3, z1
|
|
; CHECK-SVE-AES-NEXT: and z3.b, z3.b, #0x1
|
|
; CHECK-SVE-AES-NEXT: movprfx z4, z1
|
|
; CHECK-SVE-AES-NEXT: and z4.b, z4.b, #0x4
|
|
; CHECK-SVE-AES-NEXT: movprfx z5, z1
|
|
; CHECK-SVE-AES-NEXT: and z5.b, z5.b, #0x8
|
|
; CHECK-SVE-AES-NEXT: movprfx z6, z1
|
|
; CHECK-SVE-AES-NEXT: and z6.b, z6.b, #0x10
|
|
; CHECK-SVE-AES-NEXT: movprfx z7, z1
|
|
; CHECK-SVE-AES-NEXT: and z7.b, z7.b, #0x20
|
|
; CHECK-SVE-AES-NEXT: ptrue p0.b
|
|
; CHECK-SVE-AES-NEXT: movprfx z24, z1
|
|
; CHECK-SVE-AES-NEXT: and z24.b, z24.b, #0x40
|
|
; CHECK-SVE-AES-NEXT: and z1.b, z1.b, #0x80
|
|
; CHECK-SVE-AES-NEXT: mul z2.b, p0/m, z2.b, z0.b
|
|
; CHECK-SVE-AES-NEXT: mul z3.b, p0/m, z3.b, z0.b
|
|
; CHECK-SVE-AES-NEXT: mul z4.b, p0/m, z4.b, z0.b
|
|
; CHECK-SVE-AES-NEXT: mul z5.b, p0/m, z5.b, z0.b
|
|
; CHECK-SVE-AES-NEXT: mul z6.b, p0/m, z6.b, z0.b
|
|
; CHECK-SVE-AES-NEXT: mul z7.b, p0/m, z7.b, z0.b
|
|
; CHECK-SVE-AES-NEXT: mul z24.b, p0/m, z24.b, z0.b
|
|
; CHECK-SVE-AES-NEXT: mul z0.b, p0/m, z0.b, z1.b
|
|
; CHECK-SVE-AES-NEXT: eor z2.d, z3.d, z2.d
|
|
; CHECK-SVE-AES-NEXT: eor z3.d, z4.d, z5.d
|
|
; CHECK-SVE-AES-NEXT: eor z4.d, z6.d, z7.d
|
|
; CHECK-SVE-AES-NEXT: eor z2.d, z2.d, z3.d
|
|
; CHECK-SVE-AES-NEXT: eor z3.d, z4.d, z24.d
|
|
; CHECK-SVE-AES-NEXT: eor z1.d, z2.d, z3.d
|
|
; CHECK-SVE-AES-NEXT: eor z0.d, z1.d, z0.d
|
|
; CHECK-SVE-AES-NEXT: ret
|
|
;
|
|
; CHECK-SME-STREAMING-LABEL: clmul_nxv16i8:
|
|
; CHECK-SME-STREAMING: // %bb.0:
|
|
; CHECK-SME-STREAMING-NEXT: pmul z0.b, z0.b, z1.b
|
|
; CHECK-SME-STREAMING-NEXT: ret
|
|
;
|
|
; CHECK-SME-STREAMING-SSVE-AES-LABEL: clmul_nxv16i8:
|
|
; CHECK-SME-STREAMING-SSVE-AES: // %bb.0:
|
|
; CHECK-SME-STREAMING-SSVE-AES-NEXT: pmul z0.b, z0.b, z1.b
|
|
; CHECK-SME-STREAMING-SSVE-AES-NEXT: ret
|
|
;
|
|
; CHECK-SVE2-LABEL: clmul_nxv16i8:
|
|
; CHECK-SVE2: // %bb.0:
|
|
; CHECK-SVE2-NEXT: pmul z0.b, z0.b, z1.b
|
|
; CHECK-SVE2-NEXT: ret
|
|
;
|
|
; CHECK-SVE2-AES-LABEL: clmul_nxv16i8:
|
|
; CHECK-SVE2-AES: // %bb.0:
|
|
; CHECK-SVE2-AES-NEXT: pmul z0.b, z0.b, z1.b
|
|
; CHECK-SVE2-AES-NEXT: ret
|
|
%a = call <vscale x 16 x i8> @llvm.clmul.nxv16i8(<vscale x 16 x i8> %x, <vscale x 16 x i8> %y)
|
|
ret <vscale x 16 x i8> %a
|
|
}
|
|
|
|
define <vscale x 8 x i16> @clmul_nxv8i16(<vscale x 8 x i16> %x, <vscale x 8 x i16> %y) {
|
|
; CHECK-SVE-LABEL: clmul_nxv8i16:
|
|
; CHECK-SVE: // %bb.0:
|
|
; CHECK-SVE-NEXT: movprfx z2, z1
|
|
; CHECK-SVE-NEXT: and z2.h, z2.h, #0x2
|
|
; CHECK-SVE-NEXT: movprfx z3, z1
|
|
; CHECK-SVE-NEXT: and z3.h, z3.h, #0x1
|
|
; CHECK-SVE-NEXT: movprfx z4, z1
|
|
; CHECK-SVE-NEXT: and z4.h, z4.h, #0x4
|
|
; CHECK-SVE-NEXT: movprfx z5, z1
|
|
; CHECK-SVE-NEXT: and z5.h, z5.h, #0x8
|
|
; CHECK-SVE-NEXT: ptrue p0.h
|
|
; CHECK-SVE-NEXT: movprfx z6, z1
|
|
; CHECK-SVE-NEXT: and z6.h, z6.h, #0x10
|
|
; CHECK-SVE-NEXT: movprfx z7, z1
|
|
; CHECK-SVE-NEXT: and z7.h, z7.h, #0x20
|
|
; CHECK-SVE-NEXT: movprfx z24, z1
|
|
; CHECK-SVE-NEXT: and z24.h, z24.h, #0x80
|
|
; CHECK-SVE-NEXT: movprfx z25, z1
|
|
; CHECK-SVE-NEXT: and z25.h, z25.h, #0x100
|
|
; CHECK-SVE-NEXT: movprfx z26, z1
|
|
; CHECK-SVE-NEXT: and z26.h, z26.h, #0x800
|
|
; CHECK-SVE-NEXT: movprfx z27, z1
|
|
; CHECK-SVE-NEXT: and z27.h, z27.h, #0x1000
|
|
; CHECK-SVE-NEXT: mul z2.h, p0/m, z2.h, z0.h
|
|
; CHECK-SVE-NEXT: mul z3.h, p0/m, z3.h, z0.h
|
|
; CHECK-SVE-NEXT: mul z4.h, p0/m, z4.h, z0.h
|
|
; CHECK-SVE-NEXT: movprfx z28, z1
|
|
; CHECK-SVE-NEXT: and z28.h, z28.h, #0x40
|
|
; CHECK-SVE-NEXT: mul z5.h, p0/m, z5.h, z0.h
|
|
; CHECK-SVE-NEXT: mul z6.h, p0/m, z6.h, z0.h
|
|
; CHECK-SVE-NEXT: movprfx z29, z1
|
|
; CHECK-SVE-NEXT: and z29.h, z29.h, #0x200
|
|
; CHECK-SVE-NEXT: mul z7.h, p0/m, z7.h, z0.h
|
|
; CHECK-SVE-NEXT: mul z24.h, p0/m, z24.h, z0.h
|
|
; CHECK-SVE-NEXT: movprfx z30, z1
|
|
; CHECK-SVE-NEXT: and z30.h, z30.h, #0x2000
|
|
; CHECK-SVE-NEXT: mul z25.h, p0/m, z25.h, z0.h
|
|
; CHECK-SVE-NEXT: mul z26.h, p0/m, z26.h, z0.h
|
|
; CHECK-SVE-NEXT: mul z27.h, p0/m, z27.h, z0.h
|
|
; CHECK-SVE-NEXT: eor z2.d, z3.d, z2.d
|
|
; CHECK-SVE-NEXT: mul z28.h, p0/m, z28.h, z0.h
|
|
; CHECK-SVE-NEXT: eor z3.d, z4.d, z5.d
|
|
; CHECK-SVE-NEXT: mul z29.h, p0/m, z29.h, z0.h
|
|
; CHECK-SVE-NEXT: movprfx z4, z1
|
|
; CHECK-SVE-NEXT: and z4.h, z4.h, #0x400
|
|
; CHECK-SVE-NEXT: mul z30.h, p0/m, z30.h, z0.h
|
|
; CHECK-SVE-NEXT: movprfx z5, z1
|
|
; CHECK-SVE-NEXT: and z5.h, z5.h, #0x4000
|
|
; CHECK-SVE-NEXT: eor z6.d, z6.d, z7.d
|
|
; CHECK-SVE-NEXT: eor z7.d, z24.d, z25.d
|
|
; CHECK-SVE-NEXT: and z1.h, z1.h, #0x8000
|
|
; CHECK-SVE-NEXT: eor z24.d, z26.d, z27.d
|
|
; CHECK-SVE-NEXT: mul z4.h, p0/m, z4.h, z0.h
|
|
; CHECK-SVE-NEXT: eor z2.d, z2.d, z3.d
|
|
; CHECK-SVE-NEXT: mul z5.h, p0/m, z5.h, z0.h
|
|
; CHECK-SVE-NEXT: eor z3.d, z6.d, z28.d
|
|
; CHECK-SVE-NEXT: eor z6.d, z7.d, z29.d
|
|
; CHECK-SVE-NEXT: mul z0.h, p0/m, z0.h, z1.h
|
|
; CHECK-SVE-NEXT: eor z7.d, z24.d, z30.d
|
|
; CHECK-SVE-NEXT: eor z1.d, z2.d, z3.d
|
|
; CHECK-SVE-NEXT: eor z2.d, z6.d, z4.d
|
|
; CHECK-SVE-NEXT: eor z3.d, z7.d, z5.d
|
|
; CHECK-SVE-NEXT: eor z1.d, z1.d, z2.d
|
|
; CHECK-SVE-NEXT: eor z0.d, z3.d, z0.d
|
|
; CHECK-SVE-NEXT: eor z0.d, z1.d, z0.d
|
|
; CHECK-SVE-NEXT: ret
|
|
;
|
|
; CHECK-SVE-AES-LABEL: clmul_nxv8i16:
|
|
; CHECK-SVE-AES: // %bb.0:
|
|
; CHECK-SVE-AES-NEXT: movprfx z2, z1
|
|
; CHECK-SVE-AES-NEXT: and z2.h, z2.h, #0x2
|
|
; CHECK-SVE-AES-NEXT: movprfx z3, z1
|
|
; CHECK-SVE-AES-NEXT: and z3.h, z3.h, #0x1
|
|
; CHECK-SVE-AES-NEXT: movprfx z4, z1
|
|
; CHECK-SVE-AES-NEXT: and z4.h, z4.h, #0x4
|
|
; CHECK-SVE-AES-NEXT: movprfx z5, z1
|
|
; CHECK-SVE-AES-NEXT: and z5.h, z5.h, #0x8
|
|
; CHECK-SVE-AES-NEXT: ptrue p0.h
|
|
; CHECK-SVE-AES-NEXT: movprfx z6, z1
|
|
; CHECK-SVE-AES-NEXT: and z6.h, z6.h, #0x10
|
|
; CHECK-SVE-AES-NEXT: movprfx z7, z1
|
|
; CHECK-SVE-AES-NEXT: and z7.h, z7.h, #0x20
|
|
; CHECK-SVE-AES-NEXT: movprfx z24, z1
|
|
; CHECK-SVE-AES-NEXT: and z24.h, z24.h, #0x80
|
|
; CHECK-SVE-AES-NEXT: movprfx z25, z1
|
|
; CHECK-SVE-AES-NEXT: and z25.h, z25.h, #0x100
|
|
; CHECK-SVE-AES-NEXT: movprfx z26, z1
|
|
; CHECK-SVE-AES-NEXT: and z26.h, z26.h, #0x800
|
|
; CHECK-SVE-AES-NEXT: movprfx z27, z1
|
|
; CHECK-SVE-AES-NEXT: and z27.h, z27.h, #0x1000
|
|
; CHECK-SVE-AES-NEXT: mul z2.h, p0/m, z2.h, z0.h
|
|
; CHECK-SVE-AES-NEXT: mul z3.h, p0/m, z3.h, z0.h
|
|
; CHECK-SVE-AES-NEXT: mul z4.h, p0/m, z4.h, z0.h
|
|
; CHECK-SVE-AES-NEXT: movprfx z28, z1
|
|
; CHECK-SVE-AES-NEXT: and z28.h, z28.h, #0x40
|
|
; CHECK-SVE-AES-NEXT: mul z5.h, p0/m, z5.h, z0.h
|
|
; CHECK-SVE-AES-NEXT: mul z6.h, p0/m, z6.h, z0.h
|
|
; CHECK-SVE-AES-NEXT: movprfx z29, z1
|
|
; CHECK-SVE-AES-NEXT: and z29.h, z29.h, #0x200
|
|
; CHECK-SVE-AES-NEXT: mul z7.h, p0/m, z7.h, z0.h
|
|
; CHECK-SVE-AES-NEXT: mul z24.h, p0/m, z24.h, z0.h
|
|
; CHECK-SVE-AES-NEXT: movprfx z30, z1
|
|
; CHECK-SVE-AES-NEXT: and z30.h, z30.h, #0x2000
|
|
; CHECK-SVE-AES-NEXT: mul z25.h, p0/m, z25.h, z0.h
|
|
; CHECK-SVE-AES-NEXT: mul z26.h, p0/m, z26.h, z0.h
|
|
; CHECK-SVE-AES-NEXT: mul z27.h, p0/m, z27.h, z0.h
|
|
; CHECK-SVE-AES-NEXT: eor z2.d, z3.d, z2.d
|
|
; CHECK-SVE-AES-NEXT: mul z28.h, p0/m, z28.h, z0.h
|
|
; CHECK-SVE-AES-NEXT: eor z3.d, z4.d, z5.d
|
|
; CHECK-SVE-AES-NEXT: mul z29.h, p0/m, z29.h, z0.h
|
|
; CHECK-SVE-AES-NEXT: movprfx z4, z1
|
|
; CHECK-SVE-AES-NEXT: and z4.h, z4.h, #0x400
|
|
; CHECK-SVE-AES-NEXT: mul z30.h, p0/m, z30.h, z0.h
|
|
; CHECK-SVE-AES-NEXT: movprfx z5, z1
|
|
; CHECK-SVE-AES-NEXT: and z5.h, z5.h, #0x4000
|
|
; CHECK-SVE-AES-NEXT: eor z6.d, z6.d, z7.d
|
|
; CHECK-SVE-AES-NEXT: eor z7.d, z24.d, z25.d
|
|
; CHECK-SVE-AES-NEXT: and z1.h, z1.h, #0x8000
|
|
; CHECK-SVE-AES-NEXT: eor z24.d, z26.d, z27.d
|
|
; CHECK-SVE-AES-NEXT: mul z4.h, p0/m, z4.h, z0.h
|
|
; CHECK-SVE-AES-NEXT: eor z2.d, z2.d, z3.d
|
|
; CHECK-SVE-AES-NEXT: mul z5.h, p0/m, z5.h, z0.h
|
|
; CHECK-SVE-AES-NEXT: eor z3.d, z6.d, z28.d
|
|
; CHECK-SVE-AES-NEXT: eor z6.d, z7.d, z29.d
|
|
; CHECK-SVE-AES-NEXT: mul z0.h, p0/m, z0.h, z1.h
|
|
; CHECK-SVE-AES-NEXT: eor z7.d, z24.d, z30.d
|
|
; CHECK-SVE-AES-NEXT: eor z1.d, z2.d, z3.d
|
|
; CHECK-SVE-AES-NEXT: eor z2.d, z6.d, z4.d
|
|
; CHECK-SVE-AES-NEXT: eor z3.d, z7.d, z5.d
|
|
; CHECK-SVE-AES-NEXT: eor z1.d, z1.d, z2.d
|
|
; CHECK-SVE-AES-NEXT: eor z0.d, z3.d, z0.d
|
|
; CHECK-SVE-AES-NEXT: eor z0.d, z1.d, z0.d
|
|
; CHECK-SVE-AES-NEXT: ret
|
|
;
|
|
; CHECK-SME-STREAMING-LABEL: clmul_nxv8i16:
|
|
; CHECK-SME-STREAMING: // %bb.0:
|
|
; CHECK-SME-STREAMING-NEXT: trn1 z2.b, z1.b, z0.b
|
|
; CHECK-SME-STREAMING-NEXT: trn2 z3.b, z0.b, z1.b
|
|
; CHECK-SME-STREAMING-NEXT: pmullb z0.h, z0.b, z1.b
|
|
; CHECK-SME-STREAMING-NEXT: pmul z1.b, z3.b, z2.b
|
|
; CHECK-SME-STREAMING-NEXT: eorbt z1.b, z1.b, z1.b
|
|
; CHECK-SME-STREAMING-NEXT: eortb z0.b, z0.b, z1.b
|
|
; CHECK-SME-STREAMING-NEXT: ret
|
|
;
|
|
; CHECK-SME-STREAMING-SSVE-AES-LABEL: clmul_nxv8i16:
|
|
; CHECK-SME-STREAMING-SSVE-AES: // %bb.0:
|
|
; CHECK-SME-STREAMING-SSVE-AES-NEXT: trn1 z2.b, z1.b, z0.b
|
|
; CHECK-SME-STREAMING-SSVE-AES-NEXT: trn2 z3.b, z0.b, z1.b
|
|
; CHECK-SME-STREAMING-SSVE-AES-NEXT: pmullb z0.h, z0.b, z1.b
|
|
; CHECK-SME-STREAMING-SSVE-AES-NEXT: pmul z1.b, z3.b, z2.b
|
|
; CHECK-SME-STREAMING-SSVE-AES-NEXT: eorbt z1.b, z1.b, z1.b
|
|
; CHECK-SME-STREAMING-SSVE-AES-NEXT: eortb z0.b, z0.b, z1.b
|
|
; CHECK-SME-STREAMING-SSVE-AES-NEXT: ret
|
|
;
|
|
; CHECK-SVE2-LABEL: clmul_nxv8i16:
|
|
; CHECK-SVE2: // %bb.0:
|
|
; CHECK-SVE2-NEXT: trn1 z2.b, z1.b, z0.b
|
|
; CHECK-SVE2-NEXT: trn2 z3.b, z0.b, z1.b
|
|
; CHECK-SVE2-NEXT: pmullb z0.h, z0.b, z1.b
|
|
; CHECK-SVE2-NEXT: pmul z1.b, z3.b, z2.b
|
|
; CHECK-SVE2-NEXT: eorbt z1.b, z1.b, z1.b
|
|
; CHECK-SVE2-NEXT: eortb z0.b, z0.b, z1.b
|
|
; CHECK-SVE2-NEXT: ret
|
|
;
|
|
; CHECK-SVE2-AES-LABEL: clmul_nxv8i16:
|
|
; CHECK-SVE2-AES: // %bb.0:
|
|
; CHECK-SVE2-AES-NEXT: trn1 z2.b, z1.b, z0.b
|
|
; CHECK-SVE2-AES-NEXT: trn2 z3.b, z0.b, z1.b
|
|
; CHECK-SVE2-AES-NEXT: pmullb z0.h, z0.b, z1.b
|
|
; CHECK-SVE2-AES-NEXT: pmul z1.b, z3.b, z2.b
|
|
; CHECK-SVE2-AES-NEXT: eorbt z1.b, z1.b, z1.b
|
|
; CHECK-SVE2-AES-NEXT: eortb z0.b, z0.b, z1.b
|
|
; CHECK-SVE2-AES-NEXT: ret
|
|
%a = call <vscale x 8 x i16> @llvm.clmul.nxv8i16(<vscale x 8 x i16> %x, <vscale x 8 x i16> %y)
|
|
ret <vscale x 8 x i16> %a
|
|
}
|
|
|
|
define <vscale x 4 x i32> @clmul_nxv4i32(<vscale x 4 x i32> %x, <vscale x 4 x i32> %y) {
|
|
; CHECK-SVE-LABEL: clmul_nxv4i32:
|
|
; CHECK-SVE: // %bb.0:
|
|
; CHECK-SVE-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill
|
|
; CHECK-SVE-NEXT: addvl sp, sp, #-1
|
|
; CHECK-SVE-NEXT: str z8, [sp] // 16-byte Folded Spill
|
|
; CHECK-SVE-NEXT: .cfi_escape 0x0f, 0x08, 0x8f, 0x10, 0x92, 0x2e, 0x00, 0x38, 0x1e, 0x22 // sp + 16 + 8 * VG
|
|
; CHECK-SVE-NEXT: .cfi_offset w29, -16
|
|
; CHECK-SVE-NEXT: .cfi_escape 0x10, 0x48, 0x09, 0x92, 0x2e, 0x00, 0x11, 0x78, 0x1e, 0x22, 0x40, 0x1c // $d8 @ cfa - 8 * VG - 16
|
|
; CHECK-SVE-NEXT: movprfx z2, z1
|
|
; CHECK-SVE-NEXT: and z2.s, z2.s, #0x2
|
|
; CHECK-SVE-NEXT: movprfx z3, z1
|
|
; CHECK-SVE-NEXT: and z3.s, z3.s, #0x1
|
|
; CHECK-SVE-NEXT: movprfx z4, z1
|
|
; CHECK-SVE-NEXT: and z4.s, z4.s, #0x4
|
|
; CHECK-SVE-NEXT: movprfx z5, z1
|
|
; CHECK-SVE-NEXT: and z5.s, z5.s, #0x8
|
|
; CHECK-SVE-NEXT: movprfx z6, z1
|
|
; CHECK-SVE-NEXT: and z6.s, z6.s, #0x10
|
|
; CHECK-SVE-NEXT: movprfx z7, z1
|
|
; CHECK-SVE-NEXT: and z7.s, z7.s, #0x20
|
|
; CHECK-SVE-NEXT: ptrue p0.s
|
|
; CHECK-SVE-NEXT: movprfx z24, z1
|
|
; CHECK-SVE-NEXT: and z24.s, z24.s, #0x40
|
|
; CHECK-SVE-NEXT: movprfx z25, z1
|
|
; CHECK-SVE-NEXT: and z25.s, z25.s, #0x80
|
|
; CHECK-SVE-NEXT: movprfx z26, z1
|
|
; CHECK-SVE-NEXT: and z26.s, z26.s, #0x100
|
|
; CHECK-SVE-NEXT: movprfx z27, z1
|
|
; CHECK-SVE-NEXT: and z27.s, z27.s, #0x200
|
|
; CHECK-SVE-NEXT: movprfx z29, z1
|
|
; CHECK-SVE-NEXT: and z29.s, z29.s, #0x800
|
|
; CHECK-SVE-NEXT: mul z2.s, p0/m, z2.s, z0.s
|
|
; CHECK-SVE-NEXT: mul z3.s, p0/m, z3.s, z0.s
|
|
; CHECK-SVE-NEXT: movprfx z30, z1
|
|
; CHECK-SVE-NEXT: and z30.s, z30.s, #0x400000
|
|
; CHECK-SVE-NEXT: mul z4.s, p0/m, z4.s, z0.s
|
|
; CHECK-SVE-NEXT: mul z5.s, p0/m, z5.s, z0.s
|
|
; CHECK-SVE-NEXT: movprfx z31, z1
|
|
; CHECK-SVE-NEXT: and z31.s, z31.s, #0x800000
|
|
; CHECK-SVE-NEXT: mul z6.s, p0/m, z6.s, z0.s
|
|
; CHECK-SVE-NEXT: mul z7.s, p0/m, z7.s, z0.s
|
|
; CHECK-SVE-NEXT: movprfx z28, z1
|
|
; CHECK-SVE-NEXT: and z28.s, z28.s, #0x400
|
|
; CHECK-SVE-NEXT: mul z24.s, p0/m, z24.s, z0.s
|
|
; CHECK-SVE-NEXT: mul z25.s, p0/m, z25.s, z0.s
|
|
; CHECK-SVE-NEXT: movprfx z8, z1
|
|
; CHECK-SVE-NEXT: and z8.s, z8.s, #0x2000000
|
|
; CHECK-SVE-NEXT: mul z26.s, p0/m, z26.s, z0.s
|
|
; CHECK-SVE-NEXT: eor z2.d, z3.d, z2.d
|
|
; CHECK-SVE-NEXT: mul z27.s, p0/m, z27.s, z0.s
|
|
; CHECK-SVE-NEXT: eor z3.d, z4.d, z5.d
|
|
; CHECK-SVE-NEXT: movprfx z5, z1
|
|
; CHECK-SVE-NEXT: and z5.s, z5.s, #0x1000
|
|
; CHECK-SVE-NEXT: mul z29.s, p0/m, z29.s, z0.s
|
|
; CHECK-SVE-NEXT: eor z4.d, z6.d, z7.d
|
|
; CHECK-SVE-NEXT: movprfx z6, z1
|
|
; CHECK-SVE-NEXT: and z6.s, z6.s, #0x10000
|
|
; CHECK-SVE-NEXT: movprfx z7, z1
|
|
; CHECK-SVE-NEXT: and z7.s, z7.s, #0x20000
|
|
; CHECK-SVE-NEXT: mul z30.s, p0/m, z30.s, z0.s
|
|
; CHECK-SVE-NEXT: mul z31.s, p0/m, z31.s, z0.s
|
|
; CHECK-SVE-NEXT: eor z25.d, z25.d, z26.d
|
|
; CHECK-SVE-NEXT: eor z2.d, z2.d, z3.d
|
|
; CHECK-SVE-NEXT: mul z5.s, p0/m, z5.s, z0.s
|
|
; CHECK-SVE-NEXT: eor z3.d, z4.d, z24.d
|
|
; CHECK-SVE-NEXT: movprfx z4, z1
|
|
; CHECK-SVE-NEXT: and z4.s, z4.s, #0x2000
|
|
; CHECK-SVE-NEXT: mul z6.s, p0/m, z6.s, z0.s
|
|
; CHECK-SVE-NEXT: mul z7.s, p0/m, z7.s, z0.s
|
|
; CHECK-SVE-NEXT: movprfx z24, z1
|
|
; CHECK-SVE-NEXT: and z24.s, z24.s, #0x40000
|
|
; CHECK-SVE-NEXT: movprfx z26, z1
|
|
; CHECK-SVE-NEXT: and z26.s, z26.s, #0x1000000
|
|
; CHECK-SVE-NEXT: mul z28.s, p0/m, z28.s, z0.s
|
|
; CHECK-SVE-NEXT: mul z8.s, p0/m, z8.s, z0.s
|
|
; CHECK-SVE-NEXT: eor z2.d, z2.d, z3.d
|
|
; CHECK-SVE-NEXT: eor z3.d, z25.d, z27.d
|
|
; CHECK-SVE-NEXT: mul z4.s, p0/m, z4.s, z0.s
|
|
; CHECK-SVE-NEXT: movprfx z25, z1
|
|
; CHECK-SVE-NEXT: and z25.s, z25.s, #0x4000
|
|
; CHECK-SVE-NEXT: movprfx z27, z1
|
|
; CHECK-SVE-NEXT: and z27.s, z27.s, #0x80000
|
|
; CHECK-SVE-NEXT: mul z24.s, p0/m, z24.s, z0.s
|
|
; CHECK-SVE-NEXT: mul z26.s, p0/m, z26.s, z0.s
|
|
; CHECK-SVE-NEXT: eor z5.d, z29.d, z5.d
|
|
; CHECK-SVE-NEXT: eor z6.d, z6.d, z7.d
|
|
; CHECK-SVE-NEXT: eor z7.d, z30.d, z31.d
|
|
; CHECK-SVE-NEXT: movprfx z29, z1
|
|
; CHECK-SVE-NEXT: and z29.s, z29.s, #0x8000
|
|
; CHECK-SVE-NEXT: movprfx z30, z1
|
|
; CHECK-SVE-NEXT: and z30.s, z30.s, #0x100000
|
|
; CHECK-SVE-NEXT: mul z25.s, p0/m, z25.s, z0.s
|
|
; CHECK-SVE-NEXT: mul z27.s, p0/m, z27.s, z0.s
|
|
; CHECK-SVE-NEXT: movprfx z31, z1
|
|
; CHECK-SVE-NEXT: and z31.s, z31.s, #0x4000000
|
|
; CHECK-SVE-NEXT: eor z4.d, z5.d, z4.d
|
|
; CHECK-SVE-NEXT: eor z5.d, z6.d, z24.d
|
|
; CHECK-SVE-NEXT: eor z3.d, z3.d, z28.d
|
|
; CHECK-SVE-NEXT: eor z6.d, z7.d, z26.d
|
|
; CHECK-SVE-NEXT: mul z29.s, p0/m, z29.s, z0.s
|
|
; CHECK-SVE-NEXT: movprfx z7, z1
|
|
; CHECK-SVE-NEXT: and z7.s, z7.s, #0x200000
|
|
; CHECK-SVE-NEXT: mul z30.s, p0/m, z30.s, z0.s
|
|
; CHECK-SVE-NEXT: mul z31.s, p0/m, z31.s, z0.s
|
|
; CHECK-SVE-NEXT: movprfx z24, z1
|
|
; CHECK-SVE-NEXT: and z24.s, z24.s, #0x8000000
|
|
; CHECK-SVE-NEXT: eor z4.d, z4.d, z25.d
|
|
; CHECK-SVE-NEXT: eor z5.d, z5.d, z27.d
|
|
; CHECK-SVE-NEXT: eor z2.d, z2.d, z3.d
|
|
; CHECK-SVE-NEXT: eor z6.d, z6.d, z8.d
|
|
; CHECK-SVE-NEXT: mul z7.s, p0/m, z7.s, z0.s
|
|
; CHECK-SVE-NEXT: movprfx z3, z1
|
|
; CHECK-SVE-NEXT: and z3.s, z3.s, #0x10000000
|
|
; CHECK-SVE-NEXT: mul z24.s, p0/m, z24.s, z0.s
|
|
; CHECK-SVE-NEXT: movprfx z25, z1
|
|
; CHECK-SVE-NEXT: and z25.s, z25.s, #0x20000000
|
|
; CHECK-SVE-NEXT: movprfx z26, z1
|
|
; CHECK-SVE-NEXT: and z26.s, z26.s, #0x40000000
|
|
; CHECK-SVE-NEXT: eor z4.d, z4.d, z29.d
|
|
; CHECK-SVE-NEXT: eor z5.d, z5.d, z30.d
|
|
; CHECK-SVE-NEXT: and z1.s, z1.s, #0x80000000
|
|
; CHECK-SVE-NEXT: eor z6.d, z6.d, z31.d
|
|
; CHECK-SVE-NEXT: mul z3.s, p0/m, z3.s, z0.s
|
|
; CHECK-SVE-NEXT: ldr z8, [sp] // 16-byte Folded Reload
|
|
; CHECK-SVE-NEXT: mul z25.s, p0/m, z25.s, z0.s
|
|
; CHECK-SVE-NEXT: mul z26.s, p0/m, z26.s, z0.s
|
|
; CHECK-SVE-NEXT: eor z2.d, z2.d, z4.d
|
|
; CHECK-SVE-NEXT: eor z4.d, z5.d, z7.d
|
|
; CHECK-SVE-NEXT: mul z0.s, p0/m, z0.s, z1.s
|
|
; CHECK-SVE-NEXT: eor z5.d, z6.d, z24.d
|
|
; CHECK-SVE-NEXT: eor z1.d, z2.d, z4.d
|
|
; CHECK-SVE-NEXT: eor z2.d, z5.d, z3.d
|
|
; CHECK-SVE-NEXT: eor z3.d, z25.d, z26.d
|
|
; CHECK-SVE-NEXT: eor z1.d, z1.d, z2.d
|
|
; CHECK-SVE-NEXT: eor z0.d, z3.d, z0.d
|
|
; CHECK-SVE-NEXT: eor z0.d, z1.d, z0.d
|
|
; CHECK-SVE-NEXT: addvl sp, sp, #1
|
|
; CHECK-SVE-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload
|
|
; CHECK-SVE-NEXT: ret
|
|
;
|
|
; CHECK-SVE-AES-LABEL: clmul_nxv4i32:
|
|
; CHECK-SVE-AES: // %bb.0:
|
|
; CHECK-SVE-AES-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill
|
|
; CHECK-SVE-AES-NEXT: addvl sp, sp, #-1
|
|
; CHECK-SVE-AES-NEXT: str z8, [sp] // 16-byte Folded Spill
|
|
; CHECK-SVE-AES-NEXT: .cfi_escape 0x0f, 0x08, 0x8f, 0x10, 0x92, 0x2e, 0x00, 0x38, 0x1e, 0x22 // sp + 16 + 8 * VG
|
|
; CHECK-SVE-AES-NEXT: .cfi_offset w29, -16
|
|
; CHECK-SVE-AES-NEXT: .cfi_escape 0x10, 0x48, 0x09, 0x92, 0x2e, 0x00, 0x11, 0x78, 0x1e, 0x22, 0x40, 0x1c // $d8 @ cfa - 8 * VG - 16
|
|
; CHECK-SVE-AES-NEXT: movprfx z2, z1
|
|
; CHECK-SVE-AES-NEXT: and z2.s, z2.s, #0x2
|
|
; CHECK-SVE-AES-NEXT: movprfx z3, z1
|
|
; CHECK-SVE-AES-NEXT: and z3.s, z3.s, #0x1
|
|
; CHECK-SVE-AES-NEXT: movprfx z4, z1
|
|
; CHECK-SVE-AES-NEXT: and z4.s, z4.s, #0x4
|
|
; CHECK-SVE-AES-NEXT: movprfx z5, z1
|
|
; CHECK-SVE-AES-NEXT: and z5.s, z5.s, #0x8
|
|
; CHECK-SVE-AES-NEXT: movprfx z6, z1
|
|
; CHECK-SVE-AES-NEXT: and z6.s, z6.s, #0x10
|
|
; CHECK-SVE-AES-NEXT: movprfx z7, z1
|
|
; CHECK-SVE-AES-NEXT: and z7.s, z7.s, #0x20
|
|
; CHECK-SVE-AES-NEXT: ptrue p0.s
|
|
; CHECK-SVE-AES-NEXT: movprfx z24, z1
|
|
; CHECK-SVE-AES-NEXT: and z24.s, z24.s, #0x40
|
|
; CHECK-SVE-AES-NEXT: movprfx z25, z1
|
|
; CHECK-SVE-AES-NEXT: and z25.s, z25.s, #0x80
|
|
; CHECK-SVE-AES-NEXT: movprfx z26, z1
|
|
; CHECK-SVE-AES-NEXT: and z26.s, z26.s, #0x100
|
|
; CHECK-SVE-AES-NEXT: movprfx z27, z1
|
|
; CHECK-SVE-AES-NEXT: and z27.s, z27.s, #0x200
|
|
; CHECK-SVE-AES-NEXT: movprfx z29, z1
|
|
; CHECK-SVE-AES-NEXT: and z29.s, z29.s, #0x800
|
|
; CHECK-SVE-AES-NEXT: mul z2.s, p0/m, z2.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: mul z3.s, p0/m, z3.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: movprfx z30, z1
|
|
; CHECK-SVE-AES-NEXT: and z30.s, z30.s, #0x400000
|
|
; CHECK-SVE-AES-NEXT: mul z4.s, p0/m, z4.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: mul z5.s, p0/m, z5.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: movprfx z31, z1
|
|
; CHECK-SVE-AES-NEXT: and z31.s, z31.s, #0x800000
|
|
; CHECK-SVE-AES-NEXT: mul z6.s, p0/m, z6.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: mul z7.s, p0/m, z7.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: movprfx z28, z1
|
|
; CHECK-SVE-AES-NEXT: and z28.s, z28.s, #0x400
|
|
; CHECK-SVE-AES-NEXT: mul z24.s, p0/m, z24.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: mul z25.s, p0/m, z25.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: movprfx z8, z1
|
|
; CHECK-SVE-AES-NEXT: and z8.s, z8.s, #0x2000000
|
|
; CHECK-SVE-AES-NEXT: mul z26.s, p0/m, z26.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: eor z2.d, z3.d, z2.d
|
|
; CHECK-SVE-AES-NEXT: mul z27.s, p0/m, z27.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: eor z3.d, z4.d, z5.d
|
|
; CHECK-SVE-AES-NEXT: movprfx z5, z1
|
|
; CHECK-SVE-AES-NEXT: and z5.s, z5.s, #0x1000
|
|
; CHECK-SVE-AES-NEXT: mul z29.s, p0/m, z29.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: eor z4.d, z6.d, z7.d
|
|
; CHECK-SVE-AES-NEXT: movprfx z6, z1
|
|
; CHECK-SVE-AES-NEXT: and z6.s, z6.s, #0x10000
|
|
; CHECK-SVE-AES-NEXT: movprfx z7, z1
|
|
; CHECK-SVE-AES-NEXT: and z7.s, z7.s, #0x20000
|
|
; CHECK-SVE-AES-NEXT: mul z30.s, p0/m, z30.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: mul z31.s, p0/m, z31.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: eor z25.d, z25.d, z26.d
|
|
; CHECK-SVE-AES-NEXT: eor z2.d, z2.d, z3.d
|
|
; CHECK-SVE-AES-NEXT: mul z5.s, p0/m, z5.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: eor z3.d, z4.d, z24.d
|
|
; CHECK-SVE-AES-NEXT: movprfx z4, z1
|
|
; CHECK-SVE-AES-NEXT: and z4.s, z4.s, #0x2000
|
|
; CHECK-SVE-AES-NEXT: mul z6.s, p0/m, z6.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: mul z7.s, p0/m, z7.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: movprfx z24, z1
|
|
; CHECK-SVE-AES-NEXT: and z24.s, z24.s, #0x40000
|
|
; CHECK-SVE-AES-NEXT: movprfx z26, z1
|
|
; CHECK-SVE-AES-NEXT: and z26.s, z26.s, #0x1000000
|
|
; CHECK-SVE-AES-NEXT: mul z28.s, p0/m, z28.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: mul z8.s, p0/m, z8.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: eor z2.d, z2.d, z3.d
|
|
; CHECK-SVE-AES-NEXT: eor z3.d, z25.d, z27.d
|
|
; CHECK-SVE-AES-NEXT: mul z4.s, p0/m, z4.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: movprfx z25, z1
|
|
; CHECK-SVE-AES-NEXT: and z25.s, z25.s, #0x4000
|
|
; CHECK-SVE-AES-NEXT: movprfx z27, z1
|
|
; CHECK-SVE-AES-NEXT: and z27.s, z27.s, #0x80000
|
|
; CHECK-SVE-AES-NEXT: mul z24.s, p0/m, z24.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: mul z26.s, p0/m, z26.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: eor z5.d, z29.d, z5.d
|
|
; CHECK-SVE-AES-NEXT: eor z6.d, z6.d, z7.d
|
|
; CHECK-SVE-AES-NEXT: eor z7.d, z30.d, z31.d
|
|
; CHECK-SVE-AES-NEXT: movprfx z29, z1
|
|
; CHECK-SVE-AES-NEXT: and z29.s, z29.s, #0x8000
|
|
; CHECK-SVE-AES-NEXT: movprfx z30, z1
|
|
; CHECK-SVE-AES-NEXT: and z30.s, z30.s, #0x100000
|
|
; CHECK-SVE-AES-NEXT: mul z25.s, p0/m, z25.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: mul z27.s, p0/m, z27.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: movprfx z31, z1
|
|
; CHECK-SVE-AES-NEXT: and z31.s, z31.s, #0x4000000
|
|
; CHECK-SVE-AES-NEXT: eor z4.d, z5.d, z4.d
|
|
; CHECK-SVE-AES-NEXT: eor z5.d, z6.d, z24.d
|
|
; CHECK-SVE-AES-NEXT: eor z3.d, z3.d, z28.d
|
|
; CHECK-SVE-AES-NEXT: eor z6.d, z7.d, z26.d
|
|
; CHECK-SVE-AES-NEXT: mul z29.s, p0/m, z29.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: movprfx z7, z1
|
|
; CHECK-SVE-AES-NEXT: and z7.s, z7.s, #0x200000
|
|
; CHECK-SVE-AES-NEXT: mul z30.s, p0/m, z30.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: mul z31.s, p0/m, z31.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: movprfx z24, z1
|
|
; CHECK-SVE-AES-NEXT: and z24.s, z24.s, #0x8000000
|
|
; CHECK-SVE-AES-NEXT: eor z4.d, z4.d, z25.d
|
|
; CHECK-SVE-AES-NEXT: eor z5.d, z5.d, z27.d
|
|
; CHECK-SVE-AES-NEXT: eor z2.d, z2.d, z3.d
|
|
; CHECK-SVE-AES-NEXT: eor z6.d, z6.d, z8.d
|
|
; CHECK-SVE-AES-NEXT: mul z7.s, p0/m, z7.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: movprfx z3, z1
|
|
; CHECK-SVE-AES-NEXT: and z3.s, z3.s, #0x10000000
|
|
; CHECK-SVE-AES-NEXT: mul z24.s, p0/m, z24.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: movprfx z25, z1
|
|
; CHECK-SVE-AES-NEXT: and z25.s, z25.s, #0x20000000
|
|
; CHECK-SVE-AES-NEXT: movprfx z26, z1
|
|
; CHECK-SVE-AES-NEXT: and z26.s, z26.s, #0x40000000
|
|
; CHECK-SVE-AES-NEXT: eor z4.d, z4.d, z29.d
|
|
; CHECK-SVE-AES-NEXT: eor z5.d, z5.d, z30.d
|
|
; CHECK-SVE-AES-NEXT: and z1.s, z1.s, #0x80000000
|
|
; CHECK-SVE-AES-NEXT: eor z6.d, z6.d, z31.d
|
|
; CHECK-SVE-AES-NEXT: mul z3.s, p0/m, z3.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: ldr z8, [sp] // 16-byte Folded Reload
|
|
; CHECK-SVE-AES-NEXT: mul z25.s, p0/m, z25.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: mul z26.s, p0/m, z26.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: eor z2.d, z2.d, z4.d
|
|
; CHECK-SVE-AES-NEXT: eor z4.d, z5.d, z7.d
|
|
; CHECK-SVE-AES-NEXT: mul z0.s, p0/m, z0.s, z1.s
|
|
; CHECK-SVE-AES-NEXT: eor z5.d, z6.d, z24.d
|
|
; CHECK-SVE-AES-NEXT: eor z1.d, z2.d, z4.d
|
|
; CHECK-SVE-AES-NEXT: eor z2.d, z5.d, z3.d
|
|
; CHECK-SVE-AES-NEXT: eor z3.d, z25.d, z26.d
|
|
; CHECK-SVE-AES-NEXT: eor z1.d, z1.d, z2.d
|
|
; CHECK-SVE-AES-NEXT: eor z0.d, z3.d, z0.d
|
|
; CHECK-SVE-AES-NEXT: eor z0.d, z1.d, z0.d
|
|
; CHECK-SVE-AES-NEXT: addvl sp, sp, #1
|
|
; CHECK-SVE-AES-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload
|
|
; CHECK-SVE-AES-NEXT: ret
|
|
;
|
|
; CHECK-SME-STREAMING-LABEL: clmul_nxv4i32:
|
|
; CHECK-SME-STREAMING: // %bb.0:
|
|
; CHECK-SME-STREAMING-NEXT: pmullt z2.d, z0.s, z1.s
|
|
; CHECK-SME-STREAMING-NEXT: pmullb z0.d, z0.s, z1.s
|
|
; CHECK-SME-STREAMING-NEXT: trn1 z0.s, z0.s, z2.s
|
|
; CHECK-SME-STREAMING-NEXT: ret
|
|
;
|
|
; CHECK-SME-STREAMING-SSVE-AES-LABEL: clmul_nxv4i32:
|
|
; CHECK-SME-STREAMING-SSVE-AES: // %bb.0:
|
|
; CHECK-SME-STREAMING-SSVE-AES-NEXT: pmullt z2.d, z0.s, z1.s
|
|
; CHECK-SME-STREAMING-SSVE-AES-NEXT: pmullb z0.d, z0.s, z1.s
|
|
; CHECK-SME-STREAMING-SSVE-AES-NEXT: trn1 z0.s, z0.s, z2.s
|
|
; CHECK-SME-STREAMING-SSVE-AES-NEXT: ret
|
|
;
|
|
; CHECK-SVE2-LABEL: clmul_nxv4i32:
|
|
; CHECK-SVE2: // %bb.0:
|
|
; CHECK-SVE2-NEXT: pmullt z2.d, z0.s, z1.s
|
|
; CHECK-SVE2-NEXT: pmullb z0.d, z0.s, z1.s
|
|
; CHECK-SVE2-NEXT: trn1 z0.s, z0.s, z2.s
|
|
; CHECK-SVE2-NEXT: ret
|
|
;
|
|
; CHECK-SVE2-AES-LABEL: clmul_nxv4i32:
|
|
; CHECK-SVE2-AES: // %bb.0:
|
|
; CHECK-SVE2-AES-NEXT: pmullt z2.d, z0.s, z1.s
|
|
; CHECK-SVE2-AES-NEXT: pmullb z0.d, z0.s, z1.s
|
|
; CHECK-SVE2-AES-NEXT: trn1 z0.s, z0.s, z2.s
|
|
; CHECK-SVE2-AES-NEXT: ret
|
|
%a = call <vscale x 4 x i32> @llvm.clmul.nxv4i32(<vscale x 4 x i32> %x, <vscale x 4 x i32> %y)
|
|
ret <vscale x 4 x i32> %a
|
|
}
|
|
|
|
define <vscale x 2 x i64> @clmul_nxv2i64(<vscale x 2 x i64> %x, <vscale x 2 x i64> %y) {
|
|
; CHECK-SVE-LABEL: clmul_nxv2i64:
|
|
; CHECK-SVE: // %bb.0:
|
|
; CHECK-SVE-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill
|
|
; CHECK-SVE-NEXT: addvl sp, sp, #-1
|
|
; CHECK-SVE-NEXT: str z8, [sp] // 16-byte Folded Spill
|
|
; CHECK-SVE-NEXT: .cfi_escape 0x0f, 0x08, 0x8f, 0x10, 0x92, 0x2e, 0x00, 0x38, 0x1e, 0x22 // sp + 16 + 8 * VG
|
|
; CHECK-SVE-NEXT: .cfi_offset w29, -16
|
|
; CHECK-SVE-NEXT: .cfi_escape 0x10, 0x48, 0x09, 0x92, 0x2e, 0x00, 0x11, 0x78, 0x1e, 0x22, 0x40, 0x1c // $d8 @ cfa - 8 * VG - 16
|
|
; CHECK-SVE-NEXT: movprfx z2, z1
|
|
; CHECK-SVE-NEXT: and z2.d, z2.d, #0x2
|
|
; CHECK-SVE-NEXT: movprfx z3, z1
|
|
; CHECK-SVE-NEXT: and z3.d, z3.d, #0x1
|
|
; CHECK-SVE-NEXT: movprfx z4, z1
|
|
; CHECK-SVE-NEXT: and z4.d, z4.d, #0x4
|
|
; CHECK-SVE-NEXT: movprfx z5, z1
|
|
; CHECK-SVE-NEXT: and z5.d, z5.d, #0x8
|
|
; CHECK-SVE-NEXT: ptrue p0.d
|
|
; CHECK-SVE-NEXT: movprfx z6, z1
|
|
; CHECK-SVE-NEXT: and z6.d, z6.d, #0x10
|
|
; CHECK-SVE-NEXT: movprfx z7, z1
|
|
; CHECK-SVE-NEXT: and z7.d, z7.d, #0x20
|
|
; CHECK-SVE-NEXT: movprfx z24, z1
|
|
; CHECK-SVE-NEXT: and z24.d, z24.d, #0x80
|
|
; CHECK-SVE-NEXT: movprfx z25, z1
|
|
; CHECK-SVE-NEXT: and z25.d, z25.d, #0x100
|
|
; CHECK-SVE-NEXT: mul z2.d, p0/m, z2.d, z0.d
|
|
; CHECK-SVE-NEXT: mul z3.d, p0/m, z3.d, z0.d
|
|
; CHECK-SVE-NEXT: movprfx z26, z1
|
|
; CHECK-SVE-NEXT: and z26.d, z26.d, #0x40
|
|
; CHECK-SVE-NEXT: mul z4.d, p0/m, z4.d, z0.d
|
|
; CHECK-SVE-NEXT: mul z5.d, p0/m, z5.d, z0.d
|
|
; CHECK-SVE-NEXT: movprfx z27, z1
|
|
; CHECK-SVE-NEXT: and z27.d, z27.d, #0x200
|
|
; CHECK-SVE-NEXT: mul z6.d, p0/m, z6.d, z0.d
|
|
; CHECK-SVE-NEXT: mul z7.d, p0/m, z7.d, z0.d
|
|
; CHECK-SVE-NEXT: movprfx z28, z1
|
|
; CHECK-SVE-NEXT: and z28.d, z28.d, #0x800
|
|
; CHECK-SVE-NEXT: mul z24.d, p0/m, z24.d, z0.d
|
|
; CHECK-SVE-NEXT: mul z25.d, p0/m, z25.d, z0.d
|
|
; CHECK-SVE-NEXT: movprfx z29, z1
|
|
; CHECK-SVE-NEXT: and z29.d, z29.d, #0x1000
|
|
; CHECK-SVE-NEXT: mul z26.d, p0/m, z26.d, z0.d
|
|
; CHECK-SVE-NEXT: mul z27.d, p0/m, z27.d, z0.d
|
|
; CHECK-SVE-NEXT: eor z2.d, z3.d, z2.d
|
|
; CHECK-SVE-NEXT: eor z3.d, z4.d, z5.d
|
|
; CHECK-SVE-NEXT: movprfx z4, z1
|
|
; CHECK-SVE-NEXT: and z4.d, z4.d, #0x400
|
|
; CHECK-SVE-NEXT: mul z28.d, p0/m, z28.d, z0.d
|
|
; CHECK-SVE-NEXT: mul z29.d, p0/m, z29.d, z0.d
|
|
; CHECK-SVE-NEXT: movprfx z5, z1
|
|
; CHECK-SVE-NEXT: and z5.d, z5.d, #0x2000
|
|
; CHECK-SVE-NEXT: eor z6.d, z6.d, z7.d
|
|
; CHECK-SVE-NEXT: eor z7.d, z24.d, z25.d
|
|
; CHECK-SVE-NEXT: movprfx z25, z1
|
|
; CHECK-SVE-NEXT: and z25.d, z25.d, #0x10000
|
|
; CHECK-SVE-NEXT: movprfx z30, z1
|
|
; CHECK-SVE-NEXT: and z30.d, z30.d, #0x80000000
|
|
; CHECK-SVE-NEXT: mul z4.d, p0/m, z4.d, z0.d
|
|
; CHECK-SVE-NEXT: eor z2.d, z2.d, z3.d
|
|
; CHECK-SVE-NEXT: movprfx z3, z1
|
|
; CHECK-SVE-NEXT: and z3.d, z3.d, #0x4000
|
|
; CHECK-SVE-NEXT: mul z5.d, p0/m, z5.d, z0.d
|
|
; CHECK-SVE-NEXT: eor z6.d, z6.d, z26.d
|
|
; CHECK-SVE-NEXT: movprfx z26, z1
|
|
; CHECK-SVE-NEXT: and z26.d, z26.d, #0x20000
|
|
; CHECK-SVE-NEXT: eor z7.d, z7.d, z27.d
|
|
; CHECK-SVE-NEXT: eor z24.d, z28.d, z29.d
|
|
; CHECK-SVE-NEXT: movprfx z27, z1
|
|
; CHECK-SVE-NEXT: and z27.d, z27.d, #0x400000
|
|
; CHECK-SVE-NEXT: mul z3.d, p0/m, z3.d, z0.d
|
|
; CHECK-SVE-NEXT: mul z25.d, p0/m, z25.d, z0.d
|
|
; CHECK-SVE-NEXT: movprfx z29, z1
|
|
; CHECK-SVE-NEXT: and z29.d, z29.d, #0x80000
|
|
; CHECK-SVE-NEXT: eor z2.d, z2.d, z6.d
|
|
; CHECK-SVE-NEXT: movprfx z6, z1
|
|
; CHECK-SVE-NEXT: and z6.d, z6.d, #0x800000
|
|
; CHECK-SVE-NEXT: mul z26.d, p0/m, z26.d, z0.d
|
|
; CHECK-SVE-NEXT: eor z4.d, z7.d, z4.d
|
|
; CHECK-SVE-NEXT: movprfx z7, z1
|
|
; CHECK-SVE-NEXT: and z7.d, z7.d, #0x40000
|
|
; CHECK-SVE-NEXT: eor z5.d, z24.d, z5.d
|
|
; CHECK-SVE-NEXT: mul z27.d, p0/m, z27.d, z0.d
|
|
; CHECK-SVE-NEXT: movprfx z24, z1
|
|
; CHECK-SVE-NEXT: and z24.d, z24.d, #0x40000000
|
|
; CHECK-SVE-NEXT: mul z29.d, p0/m, z29.d, z0.d
|
|
; CHECK-SVE-NEXT: mul z6.d, p0/m, z6.d, z0.d
|
|
; CHECK-SVE-NEXT: movprfx z28, z1
|
|
; CHECK-SVE-NEXT: and z28.d, z28.d, #0x8000
|
|
; CHECK-SVE-NEXT: movprfx z31, z1
|
|
; CHECK-SVE-NEXT: and z31.d, z31.d, #0x100000
|
|
; CHECK-SVE-NEXT: eor z2.d, z2.d, z4.d
|
|
; CHECK-SVE-NEXT: movprfx z4, z1
|
|
; CHECK-SVE-NEXT: and z4.d, z4.d, #0x1000000
|
|
; CHECK-SVE-NEXT: mul z7.d, p0/m, z7.d, z0.d
|
|
; CHECK-SVE-NEXT: eor z3.d, z5.d, z3.d
|
|
; CHECK-SVE-NEXT: movprfx z5, z1
|
|
; CHECK-SVE-NEXT: and z5.d, z5.d, #0x20000000
|
|
; CHECK-SVE-NEXT: eor z25.d, z25.d, z26.d
|
|
; CHECK-SVE-NEXT: movprfx z26, z1
|
|
; CHECK-SVE-NEXT: and z26.d, z26.d, #0x2000000
|
|
; CHECK-SVE-NEXT: mul z24.d, p0/m, z24.d, z0.d
|
|
; CHECK-SVE-NEXT: mul z28.d, p0/m, z28.d, z0.d
|
|
; CHECK-SVE-NEXT: mul z4.d, p0/m, z4.d, z0.d
|
|
; CHECK-SVE-NEXT: eor z6.d, z27.d, z6.d
|
|
; CHECK-SVE-NEXT: mul z30.d, p0/m, z30.d, z0.d
|
|
; CHECK-SVE-NEXT: mul z5.d, p0/m, z5.d, z0.d
|
|
; CHECK-SVE-NEXT: eor z7.d, z25.d, z7.d
|
|
; CHECK-SVE-NEXT: movprfx z25, z1
|
|
; CHECK-SVE-NEXT: and z25.d, z25.d, #0x4000000
|
|
; CHECK-SVE-NEXT: mul z26.d, p0/m, z26.d, z0.d
|
|
; CHECK-SVE-NEXT: movprfx z27, z1
|
|
; CHECK-SVE-NEXT: and z27.d, z27.d, #0x100000000
|
|
; CHECK-SVE-NEXT: mul z31.d, p0/m, z31.d, z0.d
|
|
; CHECK-SVE-NEXT: eor z3.d, z3.d, z28.d
|
|
; CHECK-SVE-NEXT: movprfx z28, z1
|
|
; CHECK-SVE-NEXT: and z28.d, z28.d, #0x400000000
|
|
; CHECK-SVE-NEXT: movprfx z8, z1
|
|
; CHECK-SVE-NEXT: and z8.d, z8.d, #0x800000000000000
|
|
; CHECK-SVE-NEXT: eor z4.d, z6.d, z4.d
|
|
; CHECK-SVE-NEXT: eor z6.d, z7.d, z29.d
|
|
; CHECK-SVE-NEXT: movprfx z7, z1
|
|
; CHECK-SVE-NEXT: and z7.d, z7.d, #0x200000
|
|
; CHECK-SVE-NEXT: eor z5.d, z5.d, z24.d
|
|
; CHECK-SVE-NEXT: movprfx z24, z1
|
|
; CHECK-SVE-NEXT: and z24.d, z24.d, #0x8000000
|
|
; CHECK-SVE-NEXT: mul z25.d, p0/m, z25.d, z0.d
|
|
; CHECK-SVE-NEXT: mul z27.d, p0/m, z27.d, z0.d
|
|
; CHECK-SVE-NEXT: movprfx z29, z1
|
|
; CHECK-SVE-NEXT: and z29.d, z29.d, #0x200000000
|
|
; CHECK-SVE-NEXT: eor z2.d, z2.d, z3.d
|
|
; CHECK-SVE-NEXT: mul z7.d, p0/m, z7.d, z0.d
|
|
; CHECK-SVE-NEXT: eor z4.d, z4.d, z26.d
|
|
; CHECK-SVE-NEXT: eor z6.d, z6.d, z31.d
|
|
; CHECK-SVE-NEXT: eor z5.d, z5.d, z30.d
|
|
; CHECK-SVE-NEXT: mul z24.d, p0/m, z24.d, z0.d
|
|
; CHECK-SVE-NEXT: movprfx z26, z1
|
|
; CHECK-SVE-NEXT: and z26.d, z26.d, #0x10000000
|
|
; CHECK-SVE-NEXT: mul z29.d, p0/m, z29.d, z0.d
|
|
; CHECK-SVE-NEXT: mul z28.d, p0/m, z28.d, z0.d
|
|
; CHECK-SVE-NEXT: movprfx z30, z1
|
|
; CHECK-SVE-NEXT: and z30.d, z30.d, #0x200000000000000
|
|
; CHECK-SVE-NEXT: eor z4.d, z4.d, z25.d
|
|
; CHECK-SVE-NEXT: movprfx z25, z1
|
|
; CHECK-SVE-NEXT: and z25.d, z25.d, #0x800000000000
|
|
; CHECK-SVE-NEXT: movprfx z31, z1
|
|
; CHECK-SVE-NEXT: and z31.d, z31.d, #0x400000000000000
|
|
; CHECK-SVE-NEXT: eor z5.d, z5.d, z27.d
|
|
; CHECK-SVE-NEXT: eor z3.d, z6.d, z7.d
|
|
; CHECK-SVE-NEXT: mul z26.d, p0/m, z26.d, z0.d
|
|
; CHECK-SVE-NEXT: movprfx z6, z1
|
|
; CHECK-SVE-NEXT: and z6.d, z6.d, #0x800000000
|
|
; CHECK-SVE-NEXT: movprfx z7, z1
|
|
; CHECK-SVE-NEXT: and z7.d, z7.d, #0x2000000000
|
|
; CHECK-SVE-NEXT: mul z30.d, p0/m, z30.d, z0.d
|
|
; CHECK-SVE-NEXT: eor z4.d, z4.d, z24.d
|
|
; CHECK-SVE-NEXT: movprfx z24, z1
|
|
; CHECK-SVE-NEXT: and z24.d, z24.d, #0x4000000000
|
|
; CHECK-SVE-NEXT: mul z25.d, p0/m, z25.d, z0.d
|
|
; CHECK-SVE-NEXT: eor z5.d, z5.d, z29.d
|
|
; CHECK-SVE-NEXT: eor z2.d, z2.d, z3.d
|
|
; CHECK-SVE-NEXT: movprfx z3, z1
|
|
; CHECK-SVE-NEXT: and z3.d, z3.d, #0x400000000000
|
|
; CHECK-SVE-NEXT: mul z6.d, p0/m, z6.d, z0.d
|
|
; CHECK-SVE-NEXT: mul z7.d, p0/m, z7.d, z0.d
|
|
; CHECK-SVE-NEXT: movprfx z29, z1
|
|
; CHECK-SVE-NEXT: and z29.d, z29.d, #0x100000000000000
|
|
; CHECK-SVE-NEXT: eor z4.d, z4.d, z26.d
|
|
; CHECK-SVE-NEXT: mul z24.d, p0/m, z24.d, z0.d
|
|
; CHECK-SVE-NEXT: movprfx z26, z1
|
|
; CHECK-SVE-NEXT: and z26.d, z26.d, #0x8000000000
|
|
; CHECK-SVE-NEXT: eor z5.d, z5.d, z28.d
|
|
; CHECK-SVE-NEXT: movprfx z27, z0
|
|
; CHECK-SVE-NEXT: mul z27.d, p0/m, z27.d, z3.d
|
|
; CHECK-SVE-NEXT: movprfx z28, z1
|
|
; CHECK-SVE-NEXT: and z28.d, z28.d, #0x1000000000000
|
|
; CHECK-SVE-NEXT: mul z29.d, p0/m, z29.d, z0.d
|
|
; CHECK-SVE-NEXT: mul z31.d, p0/m, z31.d, z0.d
|
|
; CHECK-SVE-NEXT: eor z2.d, z2.d, z4.d
|
|
; CHECK-SVE-NEXT: movprfx z4, z1
|
|
; CHECK-SVE-NEXT: and z4.d, z4.d, #0x10000000000
|
|
; CHECK-SVE-NEXT: mul z26.d, p0/m, z26.d, z0.d
|
|
; CHECK-SVE-NEXT: eor z3.d, z5.d, z6.d
|
|
; CHECK-SVE-NEXT: mul z28.d, p0/m, z28.d, z0.d
|
|
; CHECK-SVE-NEXT: movprfx z5, z1
|
|
; CHECK-SVE-NEXT: and z5.d, z5.d, #0x2000000000000
|
|
; CHECK-SVE-NEXT: eor z6.d, z7.d, z24.d
|
|
; CHECK-SVE-NEXT: eor z24.d, z27.d, z25.d
|
|
; CHECK-SVE-NEXT: movprfx z7, z1
|
|
; CHECK-SVE-NEXT: and z7.d, z7.d, #0x20000000000
|
|
; CHECK-SVE-NEXT: mul z4.d, p0/m, z4.d, z0.d
|
|
; CHECK-SVE-NEXT: movprfx z27, z1
|
|
; CHECK-SVE-NEXT: and z27.d, z27.d, #0x4000000000000
|
|
; CHECK-SVE-NEXT: movprfx z25, z1
|
|
; CHECK-SVE-NEXT: and z25.d, z25.d, #0x1000000000
|
|
; CHECK-SVE-NEXT: mul z5.d, p0/m, z5.d, z0.d
|
|
; CHECK-SVE-NEXT: mul z8.d, p0/m, z8.d, z0.d
|
|
; CHECK-SVE-NEXT: eor z6.d, z6.d, z26.d
|
|
; CHECK-SVE-NEXT: eor z24.d, z24.d, z28.d
|
|
; CHECK-SVE-NEXT: mul z7.d, p0/m, z7.d, z0.d
|
|
; CHECK-SVE-NEXT: movprfx z26, z1
|
|
; CHECK-SVE-NEXT: and z26.d, z26.d, #0x40000000000
|
|
; CHECK-SVE-NEXT: mul z27.d, p0/m, z27.d, z0.d
|
|
; CHECK-SVE-NEXT: movprfx z28, z1
|
|
; CHECK-SVE-NEXT: and z28.d, z28.d, #0x8000000000000
|
|
; CHECK-SVE-NEXT: mul z25.d, p0/m, z25.d, z0.d
|
|
; CHECK-SVE-NEXT: eor z4.d, z6.d, z4.d
|
|
; CHECK-SVE-NEXT: eor z5.d, z24.d, z5.d
|
|
; CHECK-SVE-NEXT: movprfx z6, z1
|
|
; CHECK-SVE-NEXT: and z6.d, z6.d, #0x80000000000
|
|
; CHECK-SVE-NEXT: mul z26.d, p0/m, z26.d, z0.d
|
|
; CHECK-SVE-NEXT: movprfx z24, z1
|
|
; CHECK-SVE-NEXT: and z24.d, z24.d, #0x10000000000000
|
|
; CHECK-SVE-NEXT: mul z28.d, p0/m, z28.d, z0.d
|
|
; CHECK-SVE-NEXT: eor z4.d, z4.d, z7.d
|
|
; CHECK-SVE-NEXT: eor z5.d, z5.d, z27.d
|
|
; CHECK-SVE-NEXT: eor z7.d, z29.d, z30.d
|
|
; CHECK-SVE-NEXT: mul z6.d, p0/m, z6.d, z0.d
|
|
; CHECK-SVE-NEXT: movprfx z27, z1
|
|
; CHECK-SVE-NEXT: and z27.d, z27.d, #0x100000000000
|
|
; CHECK-SVE-NEXT: mul z24.d, p0/m, z24.d, z0.d
|
|
; CHECK-SVE-NEXT: movprfx z29, z1
|
|
; CHECK-SVE-NEXT: and z29.d, z29.d, #0x20000000000000
|
|
; CHECK-SVE-NEXT: movprfx z30, z1
|
|
; CHECK-SVE-NEXT: and z30.d, z30.d, #0x1000000000000000
|
|
; CHECK-SVE-NEXT: eor z3.d, z3.d, z25.d
|
|
; CHECK-SVE-NEXT: eor z4.d, z4.d, z26.d
|
|
; CHECK-SVE-NEXT: eor z5.d, z5.d, z28.d
|
|
; CHECK-SVE-NEXT: eor z7.d, z7.d, z31.d
|
|
; CHECK-SVE-NEXT: mul z27.d, p0/m, z27.d, z0.d
|
|
; CHECK-SVE-NEXT: movprfx z26, z1
|
|
; CHECK-SVE-NEXT: and z26.d, z26.d, #0x200000000000
|
|
; CHECK-SVE-NEXT: movprfx z28, z1
|
|
; CHECK-SVE-NEXT: and z28.d, z28.d, #0x40000000000000
|
|
; CHECK-SVE-NEXT: mul z29.d, p0/m, z29.d, z0.d
|
|
; CHECK-SVE-NEXT: mul z30.d, p0/m, z30.d, z0.d
|
|
; CHECK-SVE-NEXT: movprfx z31, z1
|
|
; CHECK-SVE-NEXT: and z31.d, z31.d, #0x2000000000000000
|
|
; CHECK-SVE-NEXT: eor z4.d, z4.d, z6.d
|
|
; CHECK-SVE-NEXT: eor z5.d, z5.d, z24.d
|
|
; CHECK-SVE-NEXT: eor z6.d, z7.d, z8.d
|
|
; CHECK-SVE-NEXT: mul z26.d, p0/m, z26.d, z0.d
|
|
; CHECK-SVE-NEXT: mul z28.d, p0/m, z28.d, z0.d
|
|
; CHECK-SVE-NEXT: movprfx z7, z1
|
|
; CHECK-SVE-NEXT: and z7.d, z7.d, #0x80000000000000
|
|
; CHECK-SVE-NEXT: mul z31.d, p0/m, z31.d, z0.d
|
|
; CHECK-SVE-NEXT: movprfx z24, z1
|
|
; CHECK-SVE-NEXT: and z24.d, z24.d, #0x4000000000000000
|
|
; CHECK-SVE-NEXT: and z1.d, z1.d, #0x8000000000000000
|
|
; CHECK-SVE-NEXT: eor z4.d, z4.d, z27.d
|
|
; CHECK-SVE-NEXT: eor z5.d, z5.d, z29.d
|
|
; CHECK-SVE-NEXT: eor z6.d, z6.d, z30.d
|
|
; CHECK-SVE-NEXT: mul z7.d, p0/m, z7.d, z0.d
|
|
; CHECK-SVE-NEXT: eor z2.d, z2.d, z3.d
|
|
; CHECK-SVE-NEXT: ldr z8, [sp] // 16-byte Folded Reload
|
|
; CHECK-SVE-NEXT: mul z24.d, p0/m, z24.d, z0.d
|
|
; CHECK-SVE-NEXT: mul z0.d, p0/m, z0.d, z1.d
|
|
; CHECK-SVE-NEXT: eor z3.d, z4.d, z26.d
|
|
; CHECK-SVE-NEXT: eor z4.d, z5.d, z28.d
|
|
; CHECK-SVE-NEXT: eor z5.d, z6.d, z31.d
|
|
; CHECK-SVE-NEXT: eor z1.d, z2.d, z3.d
|
|
; CHECK-SVE-NEXT: eor z2.d, z4.d, z7.d
|
|
; CHECK-SVE-NEXT: eor z3.d, z5.d, z24.d
|
|
; CHECK-SVE-NEXT: eor z1.d, z1.d, z2.d
|
|
; CHECK-SVE-NEXT: eor z0.d, z3.d, z0.d
|
|
; CHECK-SVE-NEXT: eor z0.d, z1.d, z0.d
|
|
; CHECK-SVE-NEXT: addvl sp, sp, #1
|
|
; CHECK-SVE-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload
|
|
; CHECK-SVE-NEXT: ret
|
|
;
|
|
; CHECK-SVE-AES-LABEL: clmul_nxv2i64:
|
|
; CHECK-SVE-AES: // %bb.0:
|
|
; CHECK-SVE-AES-NEXT: pmullt z2.q, z0.d, z1.d
|
|
; CHECK-SVE-AES-NEXT: pmullb z0.q, z0.d, z1.d
|
|
; CHECK-SVE-AES-NEXT: trn1 z0.d, z0.d, z2.d
|
|
; CHECK-SVE-AES-NEXT: ret
|
|
;
|
|
; CHECK-SME-STREAMING-LABEL: clmul_nxv2i64:
|
|
; CHECK-SME-STREAMING: // %bb.0:
|
|
; CHECK-SME-STREAMING-NEXT: trn1 z2.s, z1.s, z0.s
|
|
; CHECK-SME-STREAMING-NEXT: trn2 z3.s, z0.s, z1.s
|
|
; CHECK-SME-STREAMING-NEXT: pmullb z0.d, z0.s, z1.s
|
|
; CHECK-SME-STREAMING-NEXT: pmullt z4.d, z3.s, z2.s
|
|
; CHECK-SME-STREAMING-NEXT: pmullb z2.d, z3.s, z2.s
|
|
; CHECK-SME-STREAMING-NEXT: trn1 z1.s, z2.s, z4.s
|
|
; CHECK-SME-STREAMING-NEXT: eorbt z1.s, z1.s, z1.s
|
|
; CHECK-SME-STREAMING-NEXT: eortb z0.s, z0.s, z1.s
|
|
; CHECK-SME-STREAMING-NEXT: ret
|
|
;
|
|
; CHECK-SME-STREAMING-SSVE-AES-LABEL: clmul_nxv2i64:
|
|
; CHECK-SME-STREAMING-SSVE-AES: // %bb.0:
|
|
; CHECK-SME-STREAMING-SSVE-AES-NEXT: pmullt z2.q, z0.d, z1.d
|
|
; CHECK-SME-STREAMING-SSVE-AES-NEXT: pmullb z0.q, z0.d, z1.d
|
|
; CHECK-SME-STREAMING-SSVE-AES-NEXT: trn1 z0.d, z0.d, z2.d
|
|
; CHECK-SME-STREAMING-SSVE-AES-NEXT: ret
|
|
;
|
|
; CHECK-SVE2-LABEL: clmul_nxv2i64:
|
|
; CHECK-SVE2: // %bb.0:
|
|
; CHECK-SVE2-NEXT: trn1 z2.s, z1.s, z0.s
|
|
; CHECK-SVE2-NEXT: trn2 z3.s, z0.s, z1.s
|
|
; CHECK-SVE2-NEXT: pmullb z0.d, z0.s, z1.s
|
|
; CHECK-SVE2-NEXT: pmullt z4.d, z3.s, z2.s
|
|
; CHECK-SVE2-NEXT: pmullb z2.d, z3.s, z2.s
|
|
; CHECK-SVE2-NEXT: trn1 z1.s, z2.s, z4.s
|
|
; CHECK-SVE2-NEXT: eorbt z1.s, z1.s, z1.s
|
|
; CHECK-SVE2-NEXT: eortb z0.s, z0.s, z1.s
|
|
; CHECK-SVE2-NEXT: ret
|
|
;
|
|
; CHECK-SVE2-AES-LABEL: clmul_nxv2i64:
|
|
; CHECK-SVE2-AES: // %bb.0:
|
|
; CHECK-SVE2-AES-NEXT: pmullt z2.q, z0.d, z1.d
|
|
; CHECK-SVE2-AES-NEXT: pmullb z0.q, z0.d, z1.d
|
|
; CHECK-SVE2-AES-NEXT: trn1 z0.d, z0.d, z2.d
|
|
; CHECK-SVE2-AES-NEXT: ret
|
|
%a = call <vscale x 2 x i64> @llvm.clmul.nxv2i64(<vscale x 2 x i64> %x, <vscale x 2 x i64> %y)
|
|
ret <vscale x 2 x i64> %a
|
|
}
|
|
|
|
; TODO: fix
|
|
; define <vscale x 1 x i128> @clmul_nxv1i128(<vscale x 1 x i128> %x, <vscale x 1 x i128> %y) {
|
|
; %a = call <vscale x 1 x i128> @llvm.clmul.nxv1i128(<vscale x 1 x i128> %x, <vscale x 1 x i128> %y)
|
|
; ret <vscale x 1 x i128> %a
|
|
; }
|
|
|
|
define <vscale x 16 x i8> @clmul_nxv16i8_zext(<vscale x 16 x i4> %x, <vscale x 16 x i4> %y) {
|
|
; CHECK-SVE-LABEL: clmul_nxv16i8_zext:
|
|
; CHECK-SVE: // %bb.0:
|
|
; CHECK-SVE-NEXT: and z0.b, z0.b, #0xf
|
|
; CHECK-SVE-NEXT: movprfx z2, z1
|
|
; CHECK-SVE-NEXT: and z2.b, z2.b, #0x2
|
|
; CHECK-SVE-NEXT: movprfx z3, z1
|
|
; CHECK-SVE-NEXT: and z3.b, z3.b, #0x1
|
|
; CHECK-SVE-NEXT: movprfx z4, z1
|
|
; CHECK-SVE-NEXT: and z4.b, z4.b, #0x4
|
|
; CHECK-SVE-NEXT: and z1.b, z1.b, #0x8
|
|
; CHECK-SVE-NEXT: ptrue p0.b
|
|
; CHECK-SVE-NEXT: mul z2.b, p0/m, z2.b, z0.b
|
|
; CHECK-SVE-NEXT: mul z3.b, p0/m, z3.b, z0.b
|
|
; CHECK-SVE-NEXT: mul z4.b, p0/m, z4.b, z0.b
|
|
; CHECK-SVE-NEXT: mul z0.b, p0/m, z0.b, z1.b
|
|
; CHECK-SVE-NEXT: eor z1.d, z3.d, z2.d
|
|
; CHECK-SVE-NEXT: eor z0.d, z4.d, z0.d
|
|
; CHECK-SVE-NEXT: eor z0.d, z1.d, z0.d
|
|
; CHECK-SVE-NEXT: ret
|
|
;
|
|
; CHECK-SVE-AES-LABEL: clmul_nxv16i8_zext:
|
|
; CHECK-SVE-AES: // %bb.0:
|
|
; CHECK-SVE-AES-NEXT: and z0.b, z0.b, #0xf
|
|
; CHECK-SVE-AES-NEXT: movprfx z2, z1
|
|
; CHECK-SVE-AES-NEXT: and z2.b, z2.b, #0x2
|
|
; CHECK-SVE-AES-NEXT: movprfx z3, z1
|
|
; CHECK-SVE-AES-NEXT: and z3.b, z3.b, #0x1
|
|
; CHECK-SVE-AES-NEXT: movprfx z4, z1
|
|
; CHECK-SVE-AES-NEXT: and z4.b, z4.b, #0x4
|
|
; CHECK-SVE-AES-NEXT: and z1.b, z1.b, #0x8
|
|
; CHECK-SVE-AES-NEXT: ptrue p0.b
|
|
; CHECK-SVE-AES-NEXT: mul z2.b, p0/m, z2.b, z0.b
|
|
; CHECK-SVE-AES-NEXT: mul z3.b, p0/m, z3.b, z0.b
|
|
; CHECK-SVE-AES-NEXT: mul z4.b, p0/m, z4.b, z0.b
|
|
; CHECK-SVE-AES-NEXT: mul z0.b, p0/m, z0.b, z1.b
|
|
; CHECK-SVE-AES-NEXT: eor z1.d, z3.d, z2.d
|
|
; CHECK-SVE-AES-NEXT: eor z0.d, z4.d, z0.d
|
|
; CHECK-SVE-AES-NEXT: eor z0.d, z1.d, z0.d
|
|
; CHECK-SVE-AES-NEXT: ret
|
|
;
|
|
; CHECK-SME-STREAMING-LABEL: clmul_nxv16i8_zext:
|
|
; CHECK-SME-STREAMING: // %bb.0:
|
|
; CHECK-SME-STREAMING-NEXT: and z0.b, z0.b, #0xf
|
|
; CHECK-SME-STREAMING-NEXT: and z1.b, z1.b, #0xf
|
|
; CHECK-SME-STREAMING-NEXT: pmul z0.b, z0.b, z1.b
|
|
; CHECK-SME-STREAMING-NEXT: ret
|
|
;
|
|
; CHECK-SME-STREAMING-SSVE-AES-LABEL: clmul_nxv16i8_zext:
|
|
; CHECK-SME-STREAMING-SSVE-AES: // %bb.0:
|
|
; CHECK-SME-STREAMING-SSVE-AES-NEXT: and z0.b, z0.b, #0xf
|
|
; CHECK-SME-STREAMING-SSVE-AES-NEXT: and z1.b, z1.b, #0xf
|
|
; CHECK-SME-STREAMING-SSVE-AES-NEXT: pmul z0.b, z0.b, z1.b
|
|
; CHECK-SME-STREAMING-SSVE-AES-NEXT: ret
|
|
;
|
|
; CHECK-SVE2-LABEL: clmul_nxv16i8_zext:
|
|
; CHECK-SVE2: // %bb.0:
|
|
; CHECK-SVE2-NEXT: and z0.b, z0.b, #0xf
|
|
; CHECK-SVE2-NEXT: and z1.b, z1.b, #0xf
|
|
; CHECK-SVE2-NEXT: pmul z0.b, z0.b, z1.b
|
|
; CHECK-SVE2-NEXT: ret
|
|
;
|
|
; CHECK-SVE2-AES-LABEL: clmul_nxv16i8_zext:
|
|
; CHECK-SVE2-AES: // %bb.0:
|
|
; CHECK-SVE2-AES-NEXT: and z0.b, z0.b, #0xf
|
|
; CHECK-SVE2-AES-NEXT: and z1.b, z1.b, #0xf
|
|
; CHECK-SVE2-AES-NEXT: pmul z0.b, z0.b, z1.b
|
|
; CHECK-SVE2-AES-NEXT: ret
|
|
%zextx = zext <vscale x 16 x i4> %x to <vscale x 16 x i8>
|
|
%zexty = zext <vscale x 16 x i4> %y to <vscale x 16 x i8>
|
|
%a = call <vscale x 16 x i8> @llvm.clmul.nxv16i8(<vscale x 16 x i8> %zextx, <vscale x 16 x i8> %zexty)
|
|
ret <vscale x 16 x i8> %a
|
|
}
|
|
|
|
define <vscale x 8 x i16> @clmul_nxv8i16_zext(<vscale x 8 x i8> %x, <vscale x 8 x i8> %y) {
|
|
; CHECK-SVE-LABEL: clmul_nxv8i16_zext:
|
|
; CHECK-SVE: // %bb.0:
|
|
; CHECK-SVE-NEXT: and z0.h, z0.h, #0xff
|
|
; CHECK-SVE-NEXT: movprfx z2, z1
|
|
; CHECK-SVE-NEXT: and z2.h, z2.h, #0x2
|
|
; CHECK-SVE-NEXT: movprfx z3, z1
|
|
; CHECK-SVE-NEXT: and z3.h, z3.h, #0x1
|
|
; CHECK-SVE-NEXT: movprfx z4, z1
|
|
; CHECK-SVE-NEXT: and z4.h, z4.h, #0x4
|
|
; CHECK-SVE-NEXT: movprfx z5, z1
|
|
; CHECK-SVE-NEXT: and z5.h, z5.h, #0x8
|
|
; CHECK-SVE-NEXT: movprfx z6, z1
|
|
; CHECK-SVE-NEXT: and z6.h, z6.h, #0x10
|
|
; CHECK-SVE-NEXT: movprfx z7, z1
|
|
; CHECK-SVE-NEXT: and z7.h, z7.h, #0x20
|
|
; CHECK-SVE-NEXT: ptrue p0.h
|
|
; CHECK-SVE-NEXT: movprfx z24, z1
|
|
; CHECK-SVE-NEXT: and z24.h, z24.h, #0x40
|
|
; CHECK-SVE-NEXT: and z1.h, z1.h, #0x80
|
|
; CHECK-SVE-NEXT: mul z2.h, p0/m, z2.h, z0.h
|
|
; CHECK-SVE-NEXT: mul z3.h, p0/m, z3.h, z0.h
|
|
; CHECK-SVE-NEXT: mul z4.h, p0/m, z4.h, z0.h
|
|
; CHECK-SVE-NEXT: mul z5.h, p0/m, z5.h, z0.h
|
|
; CHECK-SVE-NEXT: mul z6.h, p0/m, z6.h, z0.h
|
|
; CHECK-SVE-NEXT: mul z7.h, p0/m, z7.h, z0.h
|
|
; CHECK-SVE-NEXT: mul z24.h, p0/m, z24.h, z0.h
|
|
; CHECK-SVE-NEXT: mul z0.h, p0/m, z0.h, z1.h
|
|
; CHECK-SVE-NEXT: eor z2.d, z3.d, z2.d
|
|
; CHECK-SVE-NEXT: eor z3.d, z4.d, z5.d
|
|
; CHECK-SVE-NEXT: eor z4.d, z6.d, z7.d
|
|
; CHECK-SVE-NEXT: eor z2.d, z2.d, z3.d
|
|
; CHECK-SVE-NEXT: eor z3.d, z4.d, z24.d
|
|
; CHECK-SVE-NEXT: eor z1.d, z2.d, z3.d
|
|
; CHECK-SVE-NEXT: eor z0.d, z1.d, z0.d
|
|
; CHECK-SVE-NEXT: ret
|
|
;
|
|
; CHECK-SVE-AES-LABEL: clmul_nxv8i16_zext:
|
|
; CHECK-SVE-AES: // %bb.0:
|
|
; CHECK-SVE-AES-NEXT: and z0.h, z0.h, #0xff
|
|
; CHECK-SVE-AES-NEXT: movprfx z2, z1
|
|
; CHECK-SVE-AES-NEXT: and z2.h, z2.h, #0x2
|
|
; CHECK-SVE-AES-NEXT: movprfx z3, z1
|
|
; CHECK-SVE-AES-NEXT: and z3.h, z3.h, #0x1
|
|
; CHECK-SVE-AES-NEXT: movprfx z4, z1
|
|
; CHECK-SVE-AES-NEXT: and z4.h, z4.h, #0x4
|
|
; CHECK-SVE-AES-NEXT: movprfx z5, z1
|
|
; CHECK-SVE-AES-NEXT: and z5.h, z5.h, #0x8
|
|
; CHECK-SVE-AES-NEXT: movprfx z6, z1
|
|
; CHECK-SVE-AES-NEXT: and z6.h, z6.h, #0x10
|
|
; CHECK-SVE-AES-NEXT: movprfx z7, z1
|
|
; CHECK-SVE-AES-NEXT: and z7.h, z7.h, #0x20
|
|
; CHECK-SVE-AES-NEXT: ptrue p0.h
|
|
; CHECK-SVE-AES-NEXT: movprfx z24, z1
|
|
; CHECK-SVE-AES-NEXT: and z24.h, z24.h, #0x40
|
|
; CHECK-SVE-AES-NEXT: and z1.h, z1.h, #0x80
|
|
; CHECK-SVE-AES-NEXT: mul z2.h, p0/m, z2.h, z0.h
|
|
; CHECK-SVE-AES-NEXT: mul z3.h, p0/m, z3.h, z0.h
|
|
; CHECK-SVE-AES-NEXT: mul z4.h, p0/m, z4.h, z0.h
|
|
; CHECK-SVE-AES-NEXT: mul z5.h, p0/m, z5.h, z0.h
|
|
; CHECK-SVE-AES-NEXT: mul z6.h, p0/m, z6.h, z0.h
|
|
; CHECK-SVE-AES-NEXT: mul z7.h, p0/m, z7.h, z0.h
|
|
; CHECK-SVE-AES-NEXT: mul z24.h, p0/m, z24.h, z0.h
|
|
; CHECK-SVE-AES-NEXT: mul z0.h, p0/m, z0.h, z1.h
|
|
; CHECK-SVE-AES-NEXT: eor z2.d, z3.d, z2.d
|
|
; CHECK-SVE-AES-NEXT: eor z3.d, z4.d, z5.d
|
|
; CHECK-SVE-AES-NEXT: eor z4.d, z6.d, z7.d
|
|
; CHECK-SVE-AES-NEXT: eor z2.d, z2.d, z3.d
|
|
; CHECK-SVE-AES-NEXT: eor z3.d, z4.d, z24.d
|
|
; CHECK-SVE-AES-NEXT: eor z1.d, z2.d, z3.d
|
|
; CHECK-SVE-AES-NEXT: eor z0.d, z1.d, z0.d
|
|
; CHECK-SVE-AES-NEXT: ret
|
|
;
|
|
; CHECK-SME-STREAMING-LABEL: clmul_nxv8i16_zext:
|
|
; CHECK-SME-STREAMING: // %bb.0:
|
|
; CHECK-SME-STREAMING-NEXT: and z0.h, z0.h, #0xff
|
|
; CHECK-SME-STREAMING-NEXT: and z1.h, z1.h, #0xff
|
|
; CHECK-SME-STREAMING-NEXT: pmullb z0.h, z0.b, z1.b
|
|
; CHECK-SME-STREAMING-NEXT: ret
|
|
;
|
|
; CHECK-SME-STREAMING-SSVE-AES-LABEL: clmul_nxv8i16_zext:
|
|
; CHECK-SME-STREAMING-SSVE-AES: // %bb.0:
|
|
; CHECK-SME-STREAMING-SSVE-AES-NEXT: and z0.h, z0.h, #0xff
|
|
; CHECK-SME-STREAMING-SSVE-AES-NEXT: and z1.h, z1.h, #0xff
|
|
; CHECK-SME-STREAMING-SSVE-AES-NEXT: pmullb z0.h, z0.b, z1.b
|
|
; CHECK-SME-STREAMING-SSVE-AES-NEXT: ret
|
|
;
|
|
; CHECK-SVE2-LABEL: clmul_nxv8i16_zext:
|
|
; CHECK-SVE2: // %bb.0:
|
|
; CHECK-SVE2-NEXT: and z0.h, z0.h, #0xff
|
|
; CHECK-SVE2-NEXT: and z1.h, z1.h, #0xff
|
|
; CHECK-SVE2-NEXT: pmullb z0.h, z0.b, z1.b
|
|
; CHECK-SVE2-NEXT: ret
|
|
;
|
|
; CHECK-SVE2-AES-LABEL: clmul_nxv8i16_zext:
|
|
; CHECK-SVE2-AES: // %bb.0:
|
|
; CHECK-SVE2-AES-NEXT: and z0.h, z0.h, #0xff
|
|
; CHECK-SVE2-AES-NEXT: and z1.h, z1.h, #0xff
|
|
; CHECK-SVE2-AES-NEXT: pmullb z0.h, z0.b, z1.b
|
|
; CHECK-SVE2-AES-NEXT: ret
|
|
%zextx = zext <vscale x 8 x i8> %x to <vscale x 8 x i16>
|
|
%zexty = zext <vscale x 8 x i8> %y to <vscale x 8 x i16>
|
|
%a = call <vscale x 8 x i16> @llvm.clmul.nxv8i16(<vscale x 8 x i16> %zextx, <vscale x 8 x i16> %zexty)
|
|
ret <vscale x 8 x i16> %a
|
|
}
|
|
|
|
define <vscale x 4 x i32> @clmul_nxv4i32_zext(<vscale x 4 x i16> %x, <vscale x 4 x i16> %y) {
|
|
; CHECK-SVE-LABEL: clmul_nxv4i32_zext:
|
|
; CHECK-SVE: // %bb.0:
|
|
; CHECK-SVE-NEXT: and z0.s, z0.s, #0xffff
|
|
; CHECK-SVE-NEXT: movprfx z2, z1
|
|
; CHECK-SVE-NEXT: and z2.s, z2.s, #0x2
|
|
; CHECK-SVE-NEXT: movprfx z3, z1
|
|
; CHECK-SVE-NEXT: and z3.s, z3.s, #0x1
|
|
; CHECK-SVE-NEXT: movprfx z4, z1
|
|
; CHECK-SVE-NEXT: and z4.s, z4.s, #0x4
|
|
; CHECK-SVE-NEXT: movprfx z5, z1
|
|
; CHECK-SVE-NEXT: and z5.s, z5.s, #0x8
|
|
; CHECK-SVE-NEXT: ptrue p0.s
|
|
; CHECK-SVE-NEXT: movprfx z6, z1
|
|
; CHECK-SVE-NEXT: and z6.s, z6.s, #0x10
|
|
; CHECK-SVE-NEXT: movprfx z7, z1
|
|
; CHECK-SVE-NEXT: and z7.s, z7.s, #0x20
|
|
; CHECK-SVE-NEXT: movprfx z24, z1
|
|
; CHECK-SVE-NEXT: and z24.s, z24.s, #0x80
|
|
; CHECK-SVE-NEXT: movprfx z25, z1
|
|
; CHECK-SVE-NEXT: and z25.s, z25.s, #0x100
|
|
; CHECK-SVE-NEXT: movprfx z26, z1
|
|
; CHECK-SVE-NEXT: and z26.s, z26.s, #0x800
|
|
; CHECK-SVE-NEXT: movprfx z27, z1
|
|
; CHECK-SVE-NEXT: and z27.s, z27.s, #0x1000
|
|
; CHECK-SVE-NEXT: mul z2.s, p0/m, z2.s, z0.s
|
|
; CHECK-SVE-NEXT: mul z3.s, p0/m, z3.s, z0.s
|
|
; CHECK-SVE-NEXT: movprfx z28, z1
|
|
; CHECK-SVE-NEXT: and z28.s, z28.s, #0x40
|
|
; CHECK-SVE-NEXT: mul z4.s, p0/m, z4.s, z0.s
|
|
; CHECK-SVE-NEXT: mul z5.s, p0/m, z5.s, z0.s
|
|
; CHECK-SVE-NEXT: movprfx z29, z1
|
|
; CHECK-SVE-NEXT: and z29.s, z29.s, #0x200
|
|
; CHECK-SVE-NEXT: mul z6.s, p0/m, z6.s, z0.s
|
|
; CHECK-SVE-NEXT: mul z7.s, p0/m, z7.s, z0.s
|
|
; CHECK-SVE-NEXT: movprfx z30, z1
|
|
; CHECK-SVE-NEXT: and z30.s, z30.s, #0x2000
|
|
; CHECK-SVE-NEXT: mul z24.s, p0/m, z24.s, z0.s
|
|
; CHECK-SVE-NEXT: mul z25.s, p0/m, z25.s, z0.s
|
|
; CHECK-SVE-NEXT: mul z26.s, p0/m, z26.s, z0.s
|
|
; CHECK-SVE-NEXT: mul z27.s, p0/m, z27.s, z0.s
|
|
; CHECK-SVE-NEXT: eor z2.d, z3.d, z2.d
|
|
; CHECK-SVE-NEXT: eor z3.d, z4.d, z5.d
|
|
; CHECK-SVE-NEXT: mul z28.s, p0/m, z28.s, z0.s
|
|
; CHECK-SVE-NEXT: mul z29.s, p0/m, z29.s, z0.s
|
|
; CHECK-SVE-NEXT: movprfx z4, z1
|
|
; CHECK-SVE-NEXT: and z4.s, z4.s, #0x400
|
|
; CHECK-SVE-NEXT: mul z30.s, p0/m, z30.s, z0.s
|
|
; CHECK-SVE-NEXT: movprfx z5, z1
|
|
; CHECK-SVE-NEXT: and z5.s, z5.s, #0x4000
|
|
; CHECK-SVE-NEXT: eor z6.d, z6.d, z7.d
|
|
; CHECK-SVE-NEXT: eor z7.d, z24.d, z25.d
|
|
; CHECK-SVE-NEXT: and z1.s, z1.s, #0x8000
|
|
; CHECK-SVE-NEXT: eor z24.d, z26.d, z27.d
|
|
; CHECK-SVE-NEXT: eor z2.d, z2.d, z3.d
|
|
; CHECK-SVE-NEXT: mul z4.s, p0/m, z4.s, z0.s
|
|
; CHECK-SVE-NEXT: mul z5.s, p0/m, z5.s, z0.s
|
|
; CHECK-SVE-NEXT: eor z3.d, z6.d, z28.d
|
|
; CHECK-SVE-NEXT: eor z6.d, z7.d, z29.d
|
|
; CHECK-SVE-NEXT: mul z0.s, p0/m, z0.s, z1.s
|
|
; CHECK-SVE-NEXT: eor z7.d, z24.d, z30.d
|
|
; CHECK-SVE-NEXT: eor z1.d, z2.d, z3.d
|
|
; CHECK-SVE-NEXT: eor z2.d, z6.d, z4.d
|
|
; CHECK-SVE-NEXT: eor z3.d, z7.d, z5.d
|
|
; CHECK-SVE-NEXT: eor z1.d, z1.d, z2.d
|
|
; CHECK-SVE-NEXT: eor z0.d, z3.d, z0.d
|
|
; CHECK-SVE-NEXT: eor z0.d, z1.d, z0.d
|
|
; CHECK-SVE-NEXT: ret
|
|
;
|
|
; CHECK-SVE-AES-LABEL: clmul_nxv4i32_zext:
|
|
; CHECK-SVE-AES: // %bb.0:
|
|
; CHECK-SVE-AES-NEXT: and z0.s, z0.s, #0xffff
|
|
; CHECK-SVE-AES-NEXT: movprfx z2, z1
|
|
; CHECK-SVE-AES-NEXT: and z2.s, z2.s, #0x2
|
|
; CHECK-SVE-AES-NEXT: movprfx z3, z1
|
|
; CHECK-SVE-AES-NEXT: and z3.s, z3.s, #0x1
|
|
; CHECK-SVE-AES-NEXT: movprfx z4, z1
|
|
; CHECK-SVE-AES-NEXT: and z4.s, z4.s, #0x4
|
|
; CHECK-SVE-AES-NEXT: movprfx z5, z1
|
|
; CHECK-SVE-AES-NEXT: and z5.s, z5.s, #0x8
|
|
; CHECK-SVE-AES-NEXT: ptrue p0.s
|
|
; CHECK-SVE-AES-NEXT: movprfx z6, z1
|
|
; CHECK-SVE-AES-NEXT: and z6.s, z6.s, #0x10
|
|
; CHECK-SVE-AES-NEXT: movprfx z7, z1
|
|
; CHECK-SVE-AES-NEXT: and z7.s, z7.s, #0x20
|
|
; CHECK-SVE-AES-NEXT: movprfx z24, z1
|
|
; CHECK-SVE-AES-NEXT: and z24.s, z24.s, #0x80
|
|
; CHECK-SVE-AES-NEXT: movprfx z25, z1
|
|
; CHECK-SVE-AES-NEXT: and z25.s, z25.s, #0x100
|
|
; CHECK-SVE-AES-NEXT: movprfx z26, z1
|
|
; CHECK-SVE-AES-NEXT: and z26.s, z26.s, #0x800
|
|
; CHECK-SVE-AES-NEXT: movprfx z27, z1
|
|
; CHECK-SVE-AES-NEXT: and z27.s, z27.s, #0x1000
|
|
; CHECK-SVE-AES-NEXT: mul z2.s, p0/m, z2.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: mul z3.s, p0/m, z3.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: movprfx z28, z1
|
|
; CHECK-SVE-AES-NEXT: and z28.s, z28.s, #0x40
|
|
; CHECK-SVE-AES-NEXT: mul z4.s, p0/m, z4.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: mul z5.s, p0/m, z5.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: movprfx z29, z1
|
|
; CHECK-SVE-AES-NEXT: and z29.s, z29.s, #0x200
|
|
; CHECK-SVE-AES-NEXT: mul z6.s, p0/m, z6.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: mul z7.s, p0/m, z7.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: movprfx z30, z1
|
|
; CHECK-SVE-AES-NEXT: and z30.s, z30.s, #0x2000
|
|
; CHECK-SVE-AES-NEXT: mul z24.s, p0/m, z24.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: mul z25.s, p0/m, z25.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: mul z26.s, p0/m, z26.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: mul z27.s, p0/m, z27.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: eor z2.d, z3.d, z2.d
|
|
; CHECK-SVE-AES-NEXT: eor z3.d, z4.d, z5.d
|
|
; CHECK-SVE-AES-NEXT: mul z28.s, p0/m, z28.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: mul z29.s, p0/m, z29.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: movprfx z4, z1
|
|
; CHECK-SVE-AES-NEXT: and z4.s, z4.s, #0x400
|
|
; CHECK-SVE-AES-NEXT: mul z30.s, p0/m, z30.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: movprfx z5, z1
|
|
; CHECK-SVE-AES-NEXT: and z5.s, z5.s, #0x4000
|
|
; CHECK-SVE-AES-NEXT: eor z6.d, z6.d, z7.d
|
|
; CHECK-SVE-AES-NEXT: eor z7.d, z24.d, z25.d
|
|
; CHECK-SVE-AES-NEXT: and z1.s, z1.s, #0x8000
|
|
; CHECK-SVE-AES-NEXT: eor z24.d, z26.d, z27.d
|
|
; CHECK-SVE-AES-NEXT: eor z2.d, z2.d, z3.d
|
|
; CHECK-SVE-AES-NEXT: mul z4.s, p0/m, z4.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: mul z5.s, p0/m, z5.s, z0.s
|
|
; CHECK-SVE-AES-NEXT: eor z3.d, z6.d, z28.d
|
|
; CHECK-SVE-AES-NEXT: eor z6.d, z7.d, z29.d
|
|
; CHECK-SVE-AES-NEXT: mul z0.s, p0/m, z0.s, z1.s
|
|
; CHECK-SVE-AES-NEXT: eor z7.d, z24.d, z30.d
|
|
; CHECK-SVE-AES-NEXT: eor z1.d, z2.d, z3.d
|
|
; CHECK-SVE-AES-NEXT: eor z2.d, z6.d, z4.d
|
|
; CHECK-SVE-AES-NEXT: eor z3.d, z7.d, z5.d
|
|
; CHECK-SVE-AES-NEXT: eor z1.d, z1.d, z2.d
|
|
; CHECK-SVE-AES-NEXT: eor z0.d, z3.d, z0.d
|
|
; CHECK-SVE-AES-NEXT: eor z0.d, z1.d, z0.d
|
|
; CHECK-SVE-AES-NEXT: ret
|
|
;
|
|
; CHECK-SME-STREAMING-LABEL: clmul_nxv4i32_zext:
|
|
; CHECK-SME-STREAMING: // %bb.0:
|
|
; CHECK-SME-STREAMING-NEXT: and z0.s, z0.s, #0xffff
|
|
; CHECK-SME-STREAMING-NEXT: and z1.s, z1.s, #0xffff
|
|
; CHECK-SME-STREAMING-NEXT: pmullt z2.d, z0.s, z1.s
|
|
; CHECK-SME-STREAMING-NEXT: pmullb z0.d, z0.s, z1.s
|
|
; CHECK-SME-STREAMING-NEXT: trn1 z0.s, z0.s, z2.s
|
|
; CHECK-SME-STREAMING-NEXT: ret
|
|
;
|
|
; CHECK-SME-STREAMING-SSVE-AES-LABEL: clmul_nxv4i32_zext:
|
|
; CHECK-SME-STREAMING-SSVE-AES: // %bb.0:
|
|
; CHECK-SME-STREAMING-SSVE-AES-NEXT: and z0.s, z0.s, #0xffff
|
|
; CHECK-SME-STREAMING-SSVE-AES-NEXT: and z1.s, z1.s, #0xffff
|
|
; CHECK-SME-STREAMING-SSVE-AES-NEXT: pmullt z2.d, z0.s, z1.s
|
|
; CHECK-SME-STREAMING-SSVE-AES-NEXT: pmullb z0.d, z0.s, z1.s
|
|
; CHECK-SME-STREAMING-SSVE-AES-NEXT: trn1 z0.s, z0.s, z2.s
|
|
; CHECK-SME-STREAMING-SSVE-AES-NEXT: ret
|
|
;
|
|
; CHECK-SVE2-LABEL: clmul_nxv4i32_zext:
|
|
; CHECK-SVE2: // %bb.0:
|
|
; CHECK-SVE2-NEXT: and z0.s, z0.s, #0xffff
|
|
; CHECK-SVE2-NEXT: and z1.s, z1.s, #0xffff
|
|
; CHECK-SVE2-NEXT: pmullt z2.d, z0.s, z1.s
|
|
; CHECK-SVE2-NEXT: pmullb z0.d, z0.s, z1.s
|
|
; CHECK-SVE2-NEXT: trn1 z0.s, z0.s, z2.s
|
|
; CHECK-SVE2-NEXT: ret
|
|
;
|
|
; CHECK-SVE2-AES-LABEL: clmul_nxv4i32_zext:
|
|
; CHECK-SVE2-AES: // %bb.0:
|
|
; CHECK-SVE2-AES-NEXT: and z0.s, z0.s, #0xffff
|
|
; CHECK-SVE2-AES-NEXT: and z1.s, z1.s, #0xffff
|
|
; CHECK-SVE2-AES-NEXT: pmullt z2.d, z0.s, z1.s
|
|
; CHECK-SVE2-AES-NEXT: pmullb z0.d, z0.s, z1.s
|
|
; CHECK-SVE2-AES-NEXT: trn1 z0.s, z0.s, z2.s
|
|
; CHECK-SVE2-AES-NEXT: ret
|
|
%zextx = zext <vscale x 4 x i16> %x to <vscale x 4 x i32>
|
|
%zexty = zext <vscale x 4 x i16> %y to <vscale x 4 x i32>
|
|
%a = call <vscale x 4 x i32> @llvm.clmul.nxv4i32(<vscale x 4 x i32> %zextx, <vscale x 4 x i32> %zexty)
|
|
ret <vscale x 4 x i32> %a
|
|
}
|
|
|
|
define <vscale x 2 x i64> @clmul_nxv2i64_zext(<vscale x 2 x i32> %x, <vscale x 2 x i32> %y) {
|
|
; CHECK-SVE-LABEL: clmul_nxv2i64_zext:
|
|
; CHECK-SVE: // %bb.0:
|
|
; CHECK-SVE-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill
|
|
; CHECK-SVE-NEXT: addvl sp, sp, #-1
|
|
; CHECK-SVE-NEXT: str z8, [sp] // 16-byte Folded Spill
|
|
; CHECK-SVE-NEXT: .cfi_escape 0x0f, 0x08, 0x8f, 0x10, 0x92, 0x2e, 0x00, 0x38, 0x1e, 0x22 // sp + 16 + 8 * VG
|
|
; CHECK-SVE-NEXT: .cfi_offset w29, -16
|
|
; CHECK-SVE-NEXT: .cfi_escape 0x10, 0x48, 0x09, 0x92, 0x2e, 0x00, 0x11, 0x78, 0x1e, 0x22, 0x40, 0x1c // $d8 @ cfa - 8 * VG - 16
|
|
; CHECK-SVE-NEXT: and z0.d, z0.d, #0xffffffff
|
|
; CHECK-SVE-NEXT: movprfx z2, z1
|
|
; CHECK-SVE-NEXT: and z2.d, z2.d, #0x2
|
|
; CHECK-SVE-NEXT: movprfx z3, z1
|
|
; CHECK-SVE-NEXT: and z3.d, z3.d, #0x1
|
|
; CHECK-SVE-NEXT: movprfx z4, z1
|
|
; CHECK-SVE-NEXT: and z4.d, z4.d, #0x4
|
|
; CHECK-SVE-NEXT: movprfx z5, z1
|
|
; CHECK-SVE-NEXT: and z5.d, z5.d, #0x8
|
|
; CHECK-SVE-NEXT: movprfx z6, z1
|
|
; CHECK-SVE-NEXT: and z6.d, z6.d, #0x10
|
|
; CHECK-SVE-NEXT: movprfx z7, z1
|
|
; CHECK-SVE-NEXT: and z7.d, z7.d, #0x20
|
|
; CHECK-SVE-NEXT: ptrue p0.d
|
|
; CHECK-SVE-NEXT: movprfx z24, z1
|
|
; CHECK-SVE-NEXT: and z24.d, z24.d, #0x80
|
|
; CHECK-SVE-NEXT: movprfx z25, z1
|
|
; CHECK-SVE-NEXT: and z25.d, z25.d, #0x100
|
|
; CHECK-SVE-NEXT: movprfx z26, z1
|
|
; CHECK-SVE-NEXT: and z26.d, z26.d, #0x40
|
|
; CHECK-SVE-NEXT: movprfx z27, z1
|
|
; CHECK-SVE-NEXT: and z27.d, z27.d, #0x200
|
|
; CHECK-SVE-NEXT: mul z2.d, p0/m, z2.d, z0.d
|
|
; CHECK-SVE-NEXT: mul z3.d, p0/m, z3.d, z0.d
|
|
; CHECK-SVE-NEXT: movprfx z29, z1
|
|
; CHECK-SVE-NEXT: and z29.d, z29.d, #0x20000
|
|
; CHECK-SVE-NEXT: mul z4.d, p0/m, z4.d, z0.d
|
|
; CHECK-SVE-NEXT: mul z5.d, p0/m, z5.d, z0.d
|
|
; CHECK-SVE-NEXT: movprfx z30, z1
|
|
; CHECK-SVE-NEXT: and z30.d, z30.d, #0x40000
|
|
; CHECK-SVE-NEXT: mul z6.d, p0/m, z6.d, z0.d
|
|
; CHECK-SVE-NEXT: mul z7.d, p0/m, z7.d, z0.d
|
|
; CHECK-SVE-NEXT: movprfx z31, z1
|
|
; CHECK-SVE-NEXT: and z31.d, z31.d, #0x1000000
|
|
; CHECK-SVE-NEXT: mul z24.d, p0/m, z24.d, z0.d
|
|
; CHECK-SVE-NEXT: mul z25.d, p0/m, z25.d, z0.d
|
|
; CHECK-SVE-NEXT: movprfx z28, z1
|
|
; CHECK-SVE-NEXT: and z28.d, z28.d, #0x400
|
|
; CHECK-SVE-NEXT: mul z26.d, p0/m, z26.d, z0.d
|
|
; CHECK-SVE-NEXT: eor z2.d, z3.d, z2.d
|
|
; CHECK-SVE-NEXT: mul z27.d, p0/m, z27.d, z0.d
|
|
; CHECK-SVE-NEXT: eor z3.d, z4.d, z5.d
|
|
; CHECK-SVE-NEXT: movprfx z5, z1
|
|
; CHECK-SVE-NEXT: and z5.d, z5.d, #0x800
|
|
; CHECK-SVE-NEXT: mul z29.d, p0/m, z29.d, z0.d
|
|
; CHECK-SVE-NEXT: eor z4.d, z6.d, z7.d
|
|
; CHECK-SVE-NEXT: movprfx z6, z1
|
|
; CHECK-SVE-NEXT: and z6.d, z6.d, #0x1000
|
|
; CHECK-SVE-NEXT: movprfx z7, z1
|
|
; CHECK-SVE-NEXT: and z7.d, z7.d, #0x10000
|
|
; CHECK-SVE-NEXT: eor z24.d, z24.d, z25.d
|
|
; CHECK-SVE-NEXT: movprfx z25, z1
|
|
; CHECK-SVE-NEXT: and z25.d, z25.d, #0x400000
|
|
; CHECK-SVE-NEXT: mul z30.d, p0/m, z30.d, z0.d
|
|
; CHECK-SVE-NEXT: eor z2.d, z2.d, z3.d
|
|
; CHECK-SVE-NEXT: mul z5.d, p0/m, z5.d, z0.d
|
|
; CHECK-SVE-NEXT: mul z31.d, p0/m, z31.d, z0.d
|
|
; CHECK-SVE-NEXT: eor z3.d, z4.d, z26.d
|
|
; CHECK-SVE-NEXT: movprfx z4, z1
|
|
; CHECK-SVE-NEXT: and z4.d, z4.d, #0x800000
|
|
; CHECK-SVE-NEXT: mul z6.d, p0/m, z6.d, z0.d
|
|
; CHECK-SVE-NEXT: movprfx z26, z1
|
|
; CHECK-SVE-NEXT: and z26.d, z26.d, #0x2000
|
|
; CHECK-SVE-NEXT: mul z7.d, p0/m, z7.d, z0.d
|
|
; CHECK-SVE-NEXT: mul z25.d, p0/m, z25.d, z0.d
|
|
; CHECK-SVE-NEXT: movprfx z8, z1
|
|
; CHECK-SVE-NEXT: and z8.d, z8.d, #0x2000000
|
|
; CHECK-SVE-NEXT: mul z28.d, p0/m, z28.d, z0.d
|
|
; CHECK-SVE-NEXT: mul z4.d, p0/m, z4.d, z0.d
|
|
; CHECK-SVE-NEXT: eor z2.d, z2.d, z3.d
|
|
; CHECK-SVE-NEXT: eor z3.d, z24.d, z27.d
|
|
; CHECK-SVE-NEXT: mul z26.d, p0/m, z26.d, z0.d
|
|
; CHECK-SVE-NEXT: movprfx z24, z1
|
|
; CHECK-SVE-NEXT: and z24.d, z24.d, #0x4000
|
|
; CHECK-SVE-NEXT: movprfx z27, z1
|
|
; CHECK-SVE-NEXT: and z27.d, z27.d, #0x80000
|
|
; CHECK-SVE-NEXT: eor z5.d, z5.d, z6.d
|
|
; CHECK-SVE-NEXT: eor z6.d, z7.d, z29.d
|
|
; CHECK-SVE-NEXT: movprfx z7, z1
|
|
; CHECK-SVE-NEXT: and z7.d, z7.d, #0x8000
|
|
; CHECK-SVE-NEXT: mul z8.d, p0/m, z8.d, z0.d
|
|
; CHECK-SVE-NEXT: movprfx z29, z1
|
|
; CHECK-SVE-NEXT: and z29.d, z29.d, #0x4000000
|
|
; CHECK-SVE-NEXT: eor z3.d, z3.d, z28.d
|
|
; CHECK-SVE-NEXT: eor z4.d, z25.d, z4.d
|
|
; CHECK-SVE-NEXT: mul z24.d, p0/m, z24.d, z0.d
|
|
; CHECK-SVE-NEXT: mul z27.d, p0/m, z27.d, z0.d
|
|
; CHECK-SVE-NEXT: movprfx z25, z1
|
|
; CHECK-SVE-NEXT: and z25.d, z25.d, #0x100000
|
|
; CHECK-SVE-NEXT: eor z5.d, z5.d, z26.d
|
|
; CHECK-SVE-NEXT: eor z6.d, z6.d, z30.d
|
|
; CHECK-SVE-NEXT: mul z7.d, p0/m, z7.d, z0.d
|
|
; CHECK-SVE-NEXT: movprfx z26, z1
|
|
; CHECK-SVE-NEXT: and z26.d, z26.d, #0x200000
|
|
; CHECK-SVE-NEXT: mul z29.d, p0/m, z29.d, z0.d
|
|
; CHECK-SVE-NEXT: eor z4.d, z4.d, z31.d
|
|
; CHECK-SVE-NEXT: movprfx z28, z1
|
|
; CHECK-SVE-NEXT: and z28.d, z28.d, #0x8000000
|
|
; CHECK-SVE-NEXT: eor z2.d, z2.d, z3.d
|
|
; CHECK-SVE-NEXT: mul z25.d, p0/m, z25.d, z0.d
|
|
; CHECK-SVE-NEXT: eor z5.d, z5.d, z24.d
|
|
; CHECK-SVE-NEXT: eor z6.d, z6.d, z27.d
|
|
; CHECK-SVE-NEXT: mul z26.d, p0/m, z26.d, z0.d
|
|
; CHECK-SVE-NEXT: movprfx z3, z1
|
|
; CHECK-SVE-NEXT: and z3.d, z3.d, #0x10000000
|
|
; CHECK-SVE-NEXT: movprfx z24, z1
|
|
; CHECK-SVE-NEXT: and z24.d, z24.d, #0x20000000
|
|
; CHECK-SVE-NEXT: eor z4.d, z4.d, z8.d
|
|
; CHECK-SVE-NEXT: mul z28.d, p0/m, z28.d, z0.d
|
|
; CHECK-SVE-NEXT: movprfx z27, z1
|
|
; CHECK-SVE-NEXT: and z27.d, z27.d, #0x40000000
|
|
; CHECK-SVE-NEXT: eor z5.d, z5.d, z7.d
|
|
; CHECK-SVE-NEXT: and z1.d, z1.d, #0x80000000
|
|
; CHECK-SVE-NEXT: ldr z8, [sp] // 16-byte Folded Reload
|
|
; CHECK-SVE-NEXT: eor z6.d, z6.d, z25.d
|
|
; CHECK-SVE-NEXT: mul z3.d, p0/m, z3.d, z0.d
|
|
; CHECK-SVE-NEXT: mul z24.d, p0/m, z24.d, z0.d
|
|
; CHECK-SVE-NEXT: eor z4.d, z4.d, z29.d
|
|
; CHECK-SVE-NEXT: mul z27.d, p0/m, z27.d, z0.d
|
|
; CHECK-SVE-NEXT: eor z2.d, z2.d, z5.d
|
|
; CHECK-SVE-NEXT: mul z0.d, p0/m, z0.d, z1.d
|
|
; CHECK-SVE-NEXT: eor z5.d, z6.d, z26.d
|
|
; CHECK-SVE-NEXT: eor z4.d, z4.d, z28.d
|
|
; CHECK-SVE-NEXT: eor z1.d, z2.d, z5.d
|
|
; CHECK-SVE-NEXT: eor z2.d, z4.d, z3.d
|
|
; CHECK-SVE-NEXT: eor z3.d, z24.d, z27.d
|
|
; CHECK-SVE-NEXT: eor z1.d, z1.d, z2.d
|
|
; CHECK-SVE-NEXT: eor z0.d, z3.d, z0.d
|
|
; CHECK-SVE-NEXT: eor z0.d, z1.d, z0.d
|
|
; CHECK-SVE-NEXT: addvl sp, sp, #1
|
|
; CHECK-SVE-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload
|
|
; CHECK-SVE-NEXT: ret
|
|
;
|
|
; CHECK-SVE-AES-LABEL: clmul_nxv2i64_zext:
|
|
; CHECK-SVE-AES: // %bb.0:
|
|
; CHECK-SVE-AES-NEXT: and z0.d, z0.d, #0xffffffff
|
|
; CHECK-SVE-AES-NEXT: and z1.d, z1.d, #0xffffffff
|
|
; CHECK-SVE-AES-NEXT: pmullt z2.q, z0.d, z1.d
|
|
; CHECK-SVE-AES-NEXT: pmullb z0.q, z0.d, z1.d
|
|
; CHECK-SVE-AES-NEXT: trn1 z0.d, z0.d, z2.d
|
|
; CHECK-SVE-AES-NEXT: ret
|
|
;
|
|
; CHECK-SME-STREAMING-LABEL: clmul_nxv2i64_zext:
|
|
; CHECK-SME-STREAMING: // %bb.0:
|
|
; CHECK-SME-STREAMING-NEXT: and z0.d, z0.d, #0xffffffff
|
|
; CHECK-SME-STREAMING-NEXT: and z1.d, z1.d, #0xffffffff
|
|
; CHECK-SME-STREAMING-NEXT: pmullb z0.d, z0.s, z1.s
|
|
; CHECK-SME-STREAMING-NEXT: ret
|
|
;
|
|
; CHECK-SME-STREAMING-SSVE-AES-LABEL: clmul_nxv2i64_zext:
|
|
; CHECK-SME-STREAMING-SSVE-AES: // %bb.0:
|
|
; CHECK-SME-STREAMING-SSVE-AES-NEXT: and z0.d, z0.d, #0xffffffff
|
|
; CHECK-SME-STREAMING-SSVE-AES-NEXT: and z1.d, z1.d, #0xffffffff
|
|
; CHECK-SME-STREAMING-SSVE-AES-NEXT: pmullb z0.d, z0.s, z1.s
|
|
; CHECK-SME-STREAMING-SSVE-AES-NEXT: ret
|
|
;
|
|
; CHECK-SVE2-LABEL: clmul_nxv2i64_zext:
|
|
; CHECK-SVE2: // %bb.0:
|
|
; CHECK-SVE2-NEXT: and z0.d, z0.d, #0xffffffff
|
|
; CHECK-SVE2-NEXT: and z1.d, z1.d, #0xffffffff
|
|
; CHECK-SVE2-NEXT: pmullb z0.d, z0.s, z1.s
|
|
; CHECK-SVE2-NEXT: ret
|
|
;
|
|
; CHECK-SVE2-AES-LABEL: clmul_nxv2i64_zext:
|
|
; CHECK-SVE2-AES: // %bb.0:
|
|
; CHECK-SVE2-AES-NEXT: and z0.d, z0.d, #0xffffffff
|
|
; CHECK-SVE2-AES-NEXT: and z1.d, z1.d, #0xffffffff
|
|
; CHECK-SVE2-AES-NEXT: pmullb z0.d, z0.s, z1.s
|
|
; CHECK-SVE2-AES-NEXT: ret
|
|
%zextx = zext <vscale x 2 x i32> %x to <vscale x 2 x i64>
|
|
%zexty = zext <vscale x 2 x i32> %y to <vscale x 2 x i64>
|
|
%a = call <vscale x 2 x i64> @llvm.clmul.nxv2i64(<vscale x 2 x i64> %zextx, <vscale x 2 x i64> %zexty)
|
|
ret <vscale x 2 x i64> %a
|
|
}
|
|
|
|
; TODO: fix
|
|
; define <vscale x 1 x i128> @clmul_nxv1i128_zext(<vscale x 1 x i64> %x, <vscale x 1 x i64> %y) {
|
|
; %zextx = zext <vscale x 1 x i64> %x to <vscale x 1 x i128>
|
|
; %zexty = zext <vscale x 1 x i64> %y to <vscale x 1 x i128>
|
|
; %a = call <vscale x 1 x i128> @llvm.clmul.nxv1i128(<vscale x 1 x i128> %zextx, <vscale x 1 x i128> %zexty)
|
|
; ret <vscale x 1 x i128> %a
|
|
; }
|