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
4387 lines
157 KiB
LLVM
4387 lines
157 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
|
|
; RUN: llc < %s -mtriple=x86_64-- -mcpu=x86-64 | FileCheck %s -check-prefixes=SSE,SSE2
|
|
; RUN: llc < %s -mtriple=x86_64-- -mcpu=x86-64-v2 | FileCheck %s -check-prefixes=SSE,SSE42
|
|
; RUN: llc < %s -mtriple=x86_64-- -mcpu=x86-64-v3 | FileCheck %s -check-prefixes=AVX2
|
|
; RUN: llc < %s -mtriple=x86_64-- -mcpu=knl | FileCheck %s -check-prefixes=AVX512,AVX512F
|
|
; RUN: llc < %s -mtriple=x86_64-- -mcpu=x86-64-v4 | FileCheck %s -check-prefixes=AVX512,AVX512VL
|
|
; RUN: llc < %s -mtriple=x86_64-- -mcpu=x86-64-v4 -mattr=+avx512vbmi2 | FileCheck %s -check-prefixes=AVX512,AVX512VBMI
|
|
|
|
define i512 @fshl_i512(i512 %a0, i512 %a1, i512 %a2) nounwind {
|
|
; SSE2-LABEL: fshl_i512:
|
|
; SSE2: # %bb.0:
|
|
; SSE2-NEXT: pushq %r15
|
|
; SSE2-NEXT: pushq %r14
|
|
; SSE2-NEXT: pushq %r13
|
|
; SSE2-NEXT: pushq %r12
|
|
; SSE2-NEXT: pushq %rbx
|
|
; SSE2-NEXT: movq %rcx, %rax
|
|
; SSE2-NEXT: movq {{[0-9]+}}(%rsp), %rbx
|
|
; SSE2-NEXT: movq {{[0-9]+}}(%rsp), %r11
|
|
; SSE2-NEXT: movq {{[0-9]+}}(%rsp), %rcx
|
|
; SSE2-NEXT: testl $256, %ecx # imm = 0x100
|
|
; SSE2-NEXT: movq %r8, %r10
|
|
; SSE2-NEXT: cmovneq %r11, %r10
|
|
; SSE2-NEXT: movq {{[0-9]+}}(%rsp), %r15
|
|
; SSE2-NEXT: movq %rdx, %r13
|
|
; SSE2-NEXT: cmovneq %r15, %r13
|
|
; SSE2-NEXT: movq {{[0-9]+}}(%rsp), %r12
|
|
; SSE2-NEXT: movq %rax, %r14
|
|
; SSE2-NEXT: cmovneq %rbx, %r14
|
|
; SSE2-NEXT: cmovneq %rsi, %r9
|
|
; SSE2-NEXT: cmovneq %r12, %rsi
|
|
; SSE2-NEXT: cmoveq {{[0-9]+}}(%rsp), %rax
|
|
; SSE2-NEXT: cmoveq {{[0-9]+}}(%rsp), %r8
|
|
; SSE2-NEXT: cmoveq {{[0-9]+}}(%rsp), %rdx
|
|
; SSE2-NEXT: cmovneq {{[0-9]+}}(%rsp), %r11
|
|
; SSE2-NEXT: cmovneq {{[0-9]+}}(%rsp), %r15
|
|
; SSE2-NEXT: cmovneq {{[0-9]+}}(%rsp), %rbx
|
|
; SSE2-NEXT: cmovneq {{[0-9]+}}(%rsp), %r12
|
|
; SSE2-NEXT: testb $-128, %cl
|
|
; SSE2-NEXT: cmovneq %r9, %rax
|
|
; SSE2-NEXT: cmovneq %r14, %r9
|
|
; SSE2-NEXT: cmovneq %rsi, %r14
|
|
; SSE2-NEXT: cmoveq %rbx, %r12
|
|
; SSE2-NEXT: cmoveq %rsi, %rbx
|
|
; SSE2-NEXT: movq %r10, %rsi
|
|
; SSE2-NEXT: cmovneq %r13, %rsi
|
|
; SSE2-NEXT: cmoveq %r11, %r15
|
|
; SSE2-NEXT: cmoveq %r13, %r11
|
|
; SSE2-NEXT: cmoveq %rdx, %r10
|
|
; SSE2-NEXT: cmoveq %r8, %rdx
|
|
; SSE2-NEXT: testb $64, %cl
|
|
; SSE2-NEXT: cmovneq %rax, %rdx
|
|
; SSE2-NEXT: cmovneq %r10, %rax
|
|
; SSE2-NEXT: cmovneq %r9, %r10
|
|
; SSE2-NEXT: movq %r9, %r8
|
|
; SSE2-NEXT: cmovneq %rsi, %r8
|
|
; SSE2-NEXT: cmovneq %r14, %rsi
|
|
; SSE2-NEXT: cmovneq %r11, %r14
|
|
; SSE2-NEXT: cmovneq %rbx, %r11
|
|
; SSE2-NEXT: cmovneq %r15, %rbx
|
|
; SSE2-NEXT: cmovneq %r12, %r15
|
|
; SSE2-NEXT: movq %r8, %r9
|
|
; SSE2-NEXT: shldq %cl, %rsi, %r9
|
|
; SSE2-NEXT: movq %rbx, %r12
|
|
; SSE2-NEXT: shldq %cl, %r15, %r12
|
|
; SSE2-NEXT: movq %r11, %r15
|
|
; SSE2-NEXT: shldq %cl, %rbx, %r15
|
|
; SSE2-NEXT: movq %r14, %rbx
|
|
; SSE2-NEXT: shldq %cl, %r11, %rbx
|
|
; SSE2-NEXT: shldq %cl, %r14, %rsi
|
|
; SSE2-NEXT: movq %r10, %r11
|
|
; SSE2-NEXT: shldq %cl, %r8, %r11
|
|
; SSE2-NEXT: movq %rax, %r8
|
|
; SSE2-NEXT: shldq %cl, %r10, %r8
|
|
; SSE2-NEXT: # kill: def $cl killed $cl killed $rcx
|
|
; SSE2-NEXT: shldq %cl, %rax, %rdx
|
|
; SSE2-NEXT: movq %rdi, %rax
|
|
; SSE2-NEXT: movq %rdx, 56(%rdi)
|
|
; SSE2-NEXT: movq %r8, 48(%rdi)
|
|
; SSE2-NEXT: movq %r11, 40(%rdi)
|
|
; SSE2-NEXT: movq %r9, 32(%rdi)
|
|
; SSE2-NEXT: movq %rsi, 24(%rdi)
|
|
; SSE2-NEXT: movq %rbx, 16(%rdi)
|
|
; SSE2-NEXT: movq %r15, 8(%rdi)
|
|
; SSE2-NEXT: movq %r12, (%rdi)
|
|
; SSE2-NEXT: popq %rbx
|
|
; SSE2-NEXT: popq %r12
|
|
; SSE2-NEXT: popq %r13
|
|
; SSE2-NEXT: popq %r14
|
|
; SSE2-NEXT: popq %r15
|
|
; SSE2-NEXT: retq
|
|
;
|
|
; SSE42-LABEL: fshl_i512:
|
|
; SSE42: # %bb.0:
|
|
; SSE42-NEXT: pushq %r15
|
|
; SSE42-NEXT: pushq %r14
|
|
; SSE42-NEXT: pushq %r13
|
|
; SSE42-NEXT: pushq %r12
|
|
; SSE42-NEXT: pushq %rbx
|
|
; SSE42-NEXT: movq %rcx, %rax
|
|
; SSE42-NEXT: movq {{[0-9]+}}(%rsp), %r14
|
|
; SSE42-NEXT: movq {{[0-9]+}}(%rsp), %r11
|
|
; SSE42-NEXT: movq {{[0-9]+}}(%rsp), %rcx
|
|
; SSE42-NEXT: testl $256, %ecx # imm = 0x100
|
|
; SSE42-NEXT: movq %r8, %r10
|
|
; SSE42-NEXT: cmovneq %r11, %r10
|
|
; SSE42-NEXT: movq {{[0-9]+}}(%rsp), %r12
|
|
; SSE42-NEXT: movq %rdx, %r13
|
|
; SSE42-NEXT: cmovneq %r12, %r13
|
|
; SSE42-NEXT: movq {{[0-9]+}}(%rsp), %r15
|
|
; SSE42-NEXT: movq %rax, %rbx
|
|
; SSE42-NEXT: cmovneq %r15, %rbx
|
|
; SSE42-NEXT: cmovneq %rsi, %r9
|
|
; SSE42-NEXT: cmovneq %r14, %rsi
|
|
; SSE42-NEXT: cmoveq {{[0-9]+}}(%rsp), %rax
|
|
; SSE42-NEXT: cmoveq {{[0-9]+}}(%rsp), %r8
|
|
; SSE42-NEXT: cmoveq {{[0-9]+}}(%rsp), %rdx
|
|
; SSE42-NEXT: cmovneq {{[0-9]+}}(%rsp), %r11
|
|
; SSE42-NEXT: cmovneq {{[0-9]+}}(%rsp), %r12
|
|
; SSE42-NEXT: cmovneq {{[0-9]+}}(%rsp), %r15
|
|
; SSE42-NEXT: cmovneq {{[0-9]+}}(%rsp), %r14
|
|
; SSE42-NEXT: testb $-128, %cl
|
|
; SSE42-NEXT: cmovneq %r9, %rax
|
|
; SSE42-NEXT: cmovneq %rbx, %r9
|
|
; SSE42-NEXT: cmovneq %rsi, %rbx
|
|
; SSE42-NEXT: cmoveq %r15, %r14
|
|
; SSE42-NEXT: cmoveq %rsi, %r15
|
|
; SSE42-NEXT: movq %r10, %rsi
|
|
; SSE42-NEXT: cmovneq %r13, %rsi
|
|
; SSE42-NEXT: cmoveq %r11, %r12
|
|
; SSE42-NEXT: cmoveq %r13, %r11
|
|
; SSE42-NEXT: cmoveq %rdx, %r10
|
|
; SSE42-NEXT: cmoveq %r8, %rdx
|
|
; SSE42-NEXT: testb $64, %cl
|
|
; SSE42-NEXT: cmovneq %rax, %rdx
|
|
; SSE42-NEXT: cmovneq %r10, %rax
|
|
; SSE42-NEXT: cmovneq %r9, %r10
|
|
; SSE42-NEXT: movq %r9, %r8
|
|
; SSE42-NEXT: cmovneq %rsi, %r8
|
|
; SSE42-NEXT: cmovneq %rbx, %rsi
|
|
; SSE42-NEXT: cmovneq %r11, %rbx
|
|
; SSE42-NEXT: cmovneq %r15, %r11
|
|
; SSE42-NEXT: cmovneq %r12, %r15
|
|
; SSE42-NEXT: cmovneq %r14, %r12
|
|
; SSE42-NEXT: movq %r8, %r9
|
|
; SSE42-NEXT: shldq %cl, %rsi, %r9
|
|
; SSE42-NEXT: movq %r15, %r14
|
|
; SSE42-NEXT: shldq %cl, %r12, %r14
|
|
; SSE42-NEXT: movq %r11, %r12
|
|
; SSE42-NEXT: shldq %cl, %r15, %r12
|
|
; SSE42-NEXT: movq %rbx, %r15
|
|
; SSE42-NEXT: shldq %cl, %r11, %r15
|
|
; SSE42-NEXT: shldq %cl, %rbx, %rsi
|
|
; SSE42-NEXT: movq %r10, %r11
|
|
; SSE42-NEXT: shldq %cl, %r8, %r11
|
|
; SSE42-NEXT: movq %rax, %r8
|
|
; SSE42-NEXT: shldq %cl, %r10, %r8
|
|
; SSE42-NEXT: # kill: def $cl killed $cl killed $rcx
|
|
; SSE42-NEXT: shldq %cl, %rax, %rdx
|
|
; SSE42-NEXT: movq %rdi, %rax
|
|
; SSE42-NEXT: movq %rdx, 56(%rdi)
|
|
; SSE42-NEXT: movq %r8, 48(%rdi)
|
|
; SSE42-NEXT: movq %r11, 40(%rdi)
|
|
; SSE42-NEXT: movq %r9, 32(%rdi)
|
|
; SSE42-NEXT: movq %rsi, 24(%rdi)
|
|
; SSE42-NEXT: movq %r15, 16(%rdi)
|
|
; SSE42-NEXT: movq %r12, 8(%rdi)
|
|
; SSE42-NEXT: movq %r14, (%rdi)
|
|
; SSE42-NEXT: popq %rbx
|
|
; SSE42-NEXT: popq %r12
|
|
; SSE42-NEXT: popq %r13
|
|
; SSE42-NEXT: popq %r14
|
|
; SSE42-NEXT: popq %r15
|
|
; SSE42-NEXT: retq
|
|
;
|
|
; AVX2-LABEL: fshl_i512:
|
|
; AVX2: # %bb.0:
|
|
; AVX2-NEXT: pushq %r15
|
|
; AVX2-NEXT: pushq %r14
|
|
; AVX2-NEXT: pushq %r13
|
|
; AVX2-NEXT: pushq %r12
|
|
; AVX2-NEXT: pushq %rbx
|
|
; AVX2-NEXT: movq %rcx, %rax
|
|
; AVX2-NEXT: movq {{[0-9]+}}(%rsp), %rbx
|
|
; AVX2-NEXT: movq {{[0-9]+}}(%rsp), %r12
|
|
; AVX2-NEXT: movq {{[0-9]+}}(%rsp), %r15
|
|
; AVX2-NEXT: movq {{[0-9]+}}(%rsp), %r11
|
|
; AVX2-NEXT: movq {{[0-9]+}}(%rsp), %rcx
|
|
; AVX2-NEXT: testl $256, %ecx # imm = 0x100
|
|
; AVX2-NEXT: movq %r8, %r10
|
|
; AVX2-NEXT: cmovneq %r11, %r10
|
|
; AVX2-NEXT: movq %rdx, %r13
|
|
; AVX2-NEXT: cmovneq %r15, %r13
|
|
; AVX2-NEXT: movq %rax, %r14
|
|
; AVX2-NEXT: cmovneq %rbx, %r14
|
|
; AVX2-NEXT: cmovneq %rsi, %r9
|
|
; AVX2-NEXT: cmovneq %r12, %rsi
|
|
; AVX2-NEXT: cmoveq {{[0-9]+}}(%rsp), %rax
|
|
; AVX2-NEXT: cmoveq {{[0-9]+}}(%rsp), %r8
|
|
; AVX2-NEXT: cmoveq {{[0-9]+}}(%rsp), %rdx
|
|
; AVX2-NEXT: cmovneq {{[0-9]+}}(%rsp), %r11
|
|
; AVX2-NEXT: cmovneq {{[0-9]+}}(%rsp), %r15
|
|
; AVX2-NEXT: cmovneq {{[0-9]+}}(%rsp), %rbx
|
|
; AVX2-NEXT: cmovneq {{[0-9]+}}(%rsp), %r12
|
|
; AVX2-NEXT: testb $-128, %cl
|
|
; AVX2-NEXT: cmovneq %r9, %rax
|
|
; AVX2-NEXT: cmovneq %r14, %r9
|
|
; AVX2-NEXT: cmovneq %rsi, %r14
|
|
; AVX2-NEXT: cmoveq %rbx, %r12
|
|
; AVX2-NEXT: cmoveq %rsi, %rbx
|
|
; AVX2-NEXT: movq %r10, %rsi
|
|
; AVX2-NEXT: cmovneq %r13, %rsi
|
|
; AVX2-NEXT: cmoveq %r11, %r15
|
|
; AVX2-NEXT: cmoveq %r13, %r11
|
|
; AVX2-NEXT: cmoveq %rdx, %r10
|
|
; AVX2-NEXT: cmoveq %r8, %rdx
|
|
; AVX2-NEXT: testb $64, %cl
|
|
; AVX2-NEXT: cmovneq %rax, %rdx
|
|
; AVX2-NEXT: cmovneq %r10, %rax
|
|
; AVX2-NEXT: cmovneq %r9, %r10
|
|
; AVX2-NEXT: movq %r9, %r8
|
|
; AVX2-NEXT: cmovneq %rsi, %r8
|
|
; AVX2-NEXT: cmovneq %r14, %rsi
|
|
; AVX2-NEXT: cmovneq %r11, %r14
|
|
; AVX2-NEXT: cmovneq %rbx, %r11
|
|
; AVX2-NEXT: cmovneq %r15, %rbx
|
|
; AVX2-NEXT: cmovneq %r12, %r15
|
|
; AVX2-NEXT: movq %r8, %r9
|
|
; AVX2-NEXT: shldq %cl, %rsi, %r9
|
|
; AVX2-NEXT: movq %rbx, %r12
|
|
; AVX2-NEXT: shldq %cl, %r15, %r12
|
|
; AVX2-NEXT: movq %r11, %r15
|
|
; AVX2-NEXT: shldq %cl, %rbx, %r15
|
|
; AVX2-NEXT: movq %r14, %rbx
|
|
; AVX2-NEXT: shldq %cl, %r11, %rbx
|
|
; AVX2-NEXT: shldq %cl, %r14, %rsi
|
|
; AVX2-NEXT: movq %r10, %r11
|
|
; AVX2-NEXT: shldq %cl, %r8, %r11
|
|
; AVX2-NEXT: movq %rax, %r8
|
|
; AVX2-NEXT: shldq %cl, %r10, %r8
|
|
; AVX2-NEXT: # kill: def $cl killed $cl killed $rcx
|
|
; AVX2-NEXT: shldq %cl, %rax, %rdx
|
|
; AVX2-NEXT: movq %rdx, 56(%rdi)
|
|
; AVX2-NEXT: movq %r8, 48(%rdi)
|
|
; AVX2-NEXT: movq %r11, 40(%rdi)
|
|
; AVX2-NEXT: movq %r9, 32(%rdi)
|
|
; AVX2-NEXT: movq %rsi, 24(%rdi)
|
|
; AVX2-NEXT: movq %rbx, 16(%rdi)
|
|
; AVX2-NEXT: movq %r15, 8(%rdi)
|
|
; AVX2-NEXT: movq %r12, (%rdi)
|
|
; AVX2-NEXT: movq %rdi, %rax
|
|
; AVX2-NEXT: popq %rbx
|
|
; AVX2-NEXT: popq %r12
|
|
; AVX2-NEXT: popq %r13
|
|
; AVX2-NEXT: popq %r14
|
|
; AVX2-NEXT: popq %r15
|
|
; AVX2-NEXT: retq
|
|
;
|
|
; AVX512F-LABEL: fshl_i512:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: pushq %r15
|
|
; AVX512F-NEXT: pushq %r14
|
|
; AVX512F-NEXT: pushq %r13
|
|
; AVX512F-NEXT: pushq %r12
|
|
; AVX512F-NEXT: pushq %rbx
|
|
; AVX512F-NEXT: movq %rcx, %rax
|
|
; AVX512F-NEXT: movq {{[0-9]+}}(%rsp), %rbx
|
|
; AVX512F-NEXT: movq {{[0-9]+}}(%rsp), %r12
|
|
; AVX512F-NEXT: movq {{[0-9]+}}(%rsp), %r15
|
|
; AVX512F-NEXT: movq {{[0-9]+}}(%rsp), %r11
|
|
; AVX512F-NEXT: movq {{[0-9]+}}(%rsp), %rcx
|
|
; AVX512F-NEXT: testl $256, %ecx # imm = 0x100
|
|
; AVX512F-NEXT: movq %r8, %r10
|
|
; AVX512F-NEXT: cmovneq %r11, %r10
|
|
; AVX512F-NEXT: movq %rdx, %r13
|
|
; AVX512F-NEXT: cmovneq %r15, %r13
|
|
; AVX512F-NEXT: movq %rax, %r14
|
|
; AVX512F-NEXT: cmovneq %rbx, %r14
|
|
; AVX512F-NEXT: cmovneq %rsi, %r9
|
|
; AVX512F-NEXT: cmovneq %r12, %rsi
|
|
; AVX512F-NEXT: cmoveq {{[0-9]+}}(%rsp), %rax
|
|
; AVX512F-NEXT: cmoveq {{[0-9]+}}(%rsp), %r8
|
|
; AVX512F-NEXT: cmoveq {{[0-9]+}}(%rsp), %rdx
|
|
; AVX512F-NEXT: cmovneq {{[0-9]+}}(%rsp), %r11
|
|
; AVX512F-NEXT: cmovneq {{[0-9]+}}(%rsp), %r15
|
|
; AVX512F-NEXT: cmovneq {{[0-9]+}}(%rsp), %rbx
|
|
; AVX512F-NEXT: cmovneq {{[0-9]+}}(%rsp), %r12
|
|
; AVX512F-NEXT: testb $-128, %cl
|
|
; AVX512F-NEXT: cmovneq %r9, %rax
|
|
; AVX512F-NEXT: cmovneq %r14, %r9
|
|
; AVX512F-NEXT: cmovneq %rsi, %r14
|
|
; AVX512F-NEXT: cmoveq %rbx, %r12
|
|
; AVX512F-NEXT: cmoveq %rsi, %rbx
|
|
; AVX512F-NEXT: movq %r10, %rsi
|
|
; AVX512F-NEXT: cmovneq %r13, %rsi
|
|
; AVX512F-NEXT: cmoveq %r11, %r15
|
|
; AVX512F-NEXT: cmoveq %r13, %r11
|
|
; AVX512F-NEXT: cmoveq %rdx, %r10
|
|
; AVX512F-NEXT: cmoveq %r8, %rdx
|
|
; AVX512F-NEXT: testb $64, %cl
|
|
; AVX512F-NEXT: cmovneq %rax, %rdx
|
|
; AVX512F-NEXT: cmovneq %r10, %rax
|
|
; AVX512F-NEXT: cmovneq %r9, %r10
|
|
; AVX512F-NEXT: movq %r9, %r8
|
|
; AVX512F-NEXT: cmovneq %rsi, %r8
|
|
; AVX512F-NEXT: cmovneq %r14, %rsi
|
|
; AVX512F-NEXT: cmovneq %r11, %r14
|
|
; AVX512F-NEXT: cmovneq %rbx, %r11
|
|
; AVX512F-NEXT: cmovneq %r15, %rbx
|
|
; AVX512F-NEXT: cmovneq %r12, %r15
|
|
; AVX512F-NEXT: movq %r8, %r9
|
|
; AVX512F-NEXT: shldq %cl, %rsi, %r9
|
|
; AVX512F-NEXT: movq %rbx, %r12
|
|
; AVX512F-NEXT: shldq %cl, %r15, %r12
|
|
; AVX512F-NEXT: movq %r11, %r15
|
|
; AVX512F-NEXT: shldq %cl, %rbx, %r15
|
|
; AVX512F-NEXT: movq %r14, %rbx
|
|
; AVX512F-NEXT: shldq %cl, %r11, %rbx
|
|
; AVX512F-NEXT: shldq %cl, %r14, %rsi
|
|
; AVX512F-NEXT: movq %r10, %r11
|
|
; AVX512F-NEXT: shldq %cl, %r8, %r11
|
|
; AVX512F-NEXT: movq %rax, %r8
|
|
; AVX512F-NEXT: shldq %cl, %r10, %r8
|
|
; AVX512F-NEXT: # kill: def $cl killed $cl killed $rcx
|
|
; AVX512F-NEXT: shldq %cl, %rax, %rdx
|
|
; AVX512F-NEXT: movq %rdx, 56(%rdi)
|
|
; AVX512F-NEXT: movq %r8, 48(%rdi)
|
|
; AVX512F-NEXT: movq %r11, 40(%rdi)
|
|
; AVX512F-NEXT: movq %r9, 32(%rdi)
|
|
; AVX512F-NEXT: movq %rsi, 24(%rdi)
|
|
; AVX512F-NEXT: movq %rbx, 16(%rdi)
|
|
; AVX512F-NEXT: movq %r15, 8(%rdi)
|
|
; AVX512F-NEXT: movq %r12, (%rdi)
|
|
; AVX512F-NEXT: movq %rdi, %rax
|
|
; AVX512F-NEXT: popq %rbx
|
|
; AVX512F-NEXT: popq %r12
|
|
; AVX512F-NEXT: popq %r13
|
|
; AVX512F-NEXT: popq %r14
|
|
; AVX512F-NEXT: popq %r15
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: fshl_i512:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: pushq %r15
|
|
; AVX512VL-NEXT: pushq %r14
|
|
; AVX512VL-NEXT: pushq %r13
|
|
; AVX512VL-NEXT: pushq %r12
|
|
; AVX512VL-NEXT: pushq %rbx
|
|
; AVX512VL-NEXT: movq %rcx, %rax
|
|
; AVX512VL-NEXT: movq {{[0-9]+}}(%rsp), %rbx
|
|
; AVX512VL-NEXT: movq {{[0-9]+}}(%rsp), %r12
|
|
; AVX512VL-NEXT: movq {{[0-9]+}}(%rsp), %r15
|
|
; AVX512VL-NEXT: movq {{[0-9]+}}(%rsp), %r11
|
|
; AVX512VL-NEXT: movq {{[0-9]+}}(%rsp), %rcx
|
|
; AVX512VL-NEXT: testl $256, %ecx # imm = 0x100
|
|
; AVX512VL-NEXT: movq %r8, %r10
|
|
; AVX512VL-NEXT: cmovneq %r11, %r10
|
|
; AVX512VL-NEXT: movq %rdx, %r13
|
|
; AVX512VL-NEXT: cmovneq %r15, %r13
|
|
; AVX512VL-NEXT: movq %rax, %r14
|
|
; AVX512VL-NEXT: cmovneq %rbx, %r14
|
|
; AVX512VL-NEXT: cmovneq %rsi, %r9
|
|
; AVX512VL-NEXT: cmoveq {{[0-9]+}}(%rsp), %rax
|
|
; AVX512VL-NEXT: cmoveq {{[0-9]+}}(%rsp), %r8
|
|
; AVX512VL-NEXT: cmoveq {{[0-9]+}}(%rsp), %rdx
|
|
; AVX512VL-NEXT: cmovneq {{[0-9]+}}(%rsp), %r11
|
|
; AVX512VL-NEXT: cmovneq {{[0-9]+}}(%rsp), %r15
|
|
; AVX512VL-NEXT: cmovneq %r12, %rsi
|
|
; AVX512VL-NEXT: cmovneq {{[0-9]+}}(%rsp), %rbx
|
|
; AVX512VL-NEXT: cmovneq {{[0-9]+}}(%rsp), %r12
|
|
; AVX512VL-NEXT: testb $-128, %cl
|
|
; AVX512VL-NEXT: cmovneq %r9, %rax
|
|
; AVX512VL-NEXT: cmovneq %r14, %r9
|
|
; AVX512VL-NEXT: cmovneq %rsi, %r14
|
|
; AVX512VL-NEXT: cmoveq %rbx, %r12
|
|
; AVX512VL-NEXT: cmoveq %rsi, %rbx
|
|
; AVX512VL-NEXT: movq %r10, %rsi
|
|
; AVX512VL-NEXT: cmovneq %r13, %rsi
|
|
; AVX512VL-NEXT: cmoveq %r11, %r15
|
|
; AVX512VL-NEXT: cmoveq %r13, %r11
|
|
; AVX512VL-NEXT: cmoveq %rdx, %r10
|
|
; AVX512VL-NEXT: cmoveq %r8, %rdx
|
|
; AVX512VL-NEXT: testb $64, %cl
|
|
; AVX512VL-NEXT: cmovneq %rax, %rdx
|
|
; AVX512VL-NEXT: cmovneq %r10, %rax
|
|
; AVX512VL-NEXT: cmovneq %r9, %r10
|
|
; AVX512VL-NEXT: movq %r9, %r8
|
|
; AVX512VL-NEXT: cmovneq %rsi, %r8
|
|
; AVX512VL-NEXT: cmovneq %r14, %rsi
|
|
; AVX512VL-NEXT: cmovneq %r11, %r14
|
|
; AVX512VL-NEXT: cmovneq %rbx, %r11
|
|
; AVX512VL-NEXT: cmovneq %r15, %rbx
|
|
; AVX512VL-NEXT: cmovneq %r12, %r15
|
|
; AVX512VL-NEXT: movq %r8, %r9
|
|
; AVX512VL-NEXT: shldq %cl, %rsi, %r9
|
|
; AVX512VL-NEXT: movq %rbx, %r12
|
|
; AVX512VL-NEXT: shldq %cl, %r15, %r12
|
|
; AVX512VL-NEXT: movq %r11, %r15
|
|
; AVX512VL-NEXT: shldq %cl, %rbx, %r15
|
|
; AVX512VL-NEXT: movq %r14, %rbx
|
|
; AVX512VL-NEXT: shldq %cl, %r11, %rbx
|
|
; AVX512VL-NEXT: shldq %cl, %r14, %rsi
|
|
; AVX512VL-NEXT: movq %r10, %r11
|
|
; AVX512VL-NEXT: shldq %cl, %r8, %r11
|
|
; AVX512VL-NEXT: movq %rax, %r8
|
|
; AVX512VL-NEXT: shldq %cl, %r10, %r8
|
|
; AVX512VL-NEXT: # kill: def $cl killed $cl killed $rcx
|
|
; AVX512VL-NEXT: shldq %cl, %rax, %rdx
|
|
; AVX512VL-NEXT: movq %rdx, 56(%rdi)
|
|
; AVX512VL-NEXT: movq %r8, 48(%rdi)
|
|
; AVX512VL-NEXT: movq %r11, 40(%rdi)
|
|
; AVX512VL-NEXT: movq %r9, 32(%rdi)
|
|
; AVX512VL-NEXT: movq %rsi, 24(%rdi)
|
|
; AVX512VL-NEXT: movq %rbx, 16(%rdi)
|
|
; AVX512VL-NEXT: movq %r15, 8(%rdi)
|
|
; AVX512VL-NEXT: movq %r12, (%rdi)
|
|
; AVX512VL-NEXT: movq %rdi, %rax
|
|
; AVX512VL-NEXT: popq %rbx
|
|
; AVX512VL-NEXT: popq %r12
|
|
; AVX512VL-NEXT: popq %r13
|
|
; AVX512VL-NEXT: popq %r14
|
|
; AVX512VL-NEXT: popq %r15
|
|
; AVX512VL-NEXT: retq
|
|
;
|
|
; AVX512VBMI-LABEL: fshl_i512:
|
|
; AVX512VBMI: # %bb.0:
|
|
; AVX512VBMI-NEXT: pushq %r15
|
|
; AVX512VBMI-NEXT: pushq %r14
|
|
; AVX512VBMI-NEXT: pushq %r13
|
|
; AVX512VBMI-NEXT: pushq %r12
|
|
; AVX512VBMI-NEXT: pushq %rbx
|
|
; AVX512VBMI-NEXT: movq %rcx, %rax
|
|
; AVX512VBMI-NEXT: movq {{[0-9]+}}(%rsp), %rbx
|
|
; AVX512VBMI-NEXT: movq {{[0-9]+}}(%rsp), %r12
|
|
; AVX512VBMI-NEXT: movq {{[0-9]+}}(%rsp), %r15
|
|
; AVX512VBMI-NEXT: movq {{[0-9]+}}(%rsp), %r11
|
|
; AVX512VBMI-NEXT: movq {{[0-9]+}}(%rsp), %rcx
|
|
; AVX512VBMI-NEXT: testl $256, %ecx # imm = 0x100
|
|
; AVX512VBMI-NEXT: movq %r8, %r10
|
|
; AVX512VBMI-NEXT: cmovneq %r11, %r10
|
|
; AVX512VBMI-NEXT: movq %rdx, %r13
|
|
; AVX512VBMI-NEXT: cmovneq %r15, %r13
|
|
; AVX512VBMI-NEXT: movq %rax, %r14
|
|
; AVX512VBMI-NEXT: cmovneq %rbx, %r14
|
|
; AVX512VBMI-NEXT: cmovneq %rsi, %r9
|
|
; AVX512VBMI-NEXT: cmoveq {{[0-9]+}}(%rsp), %rax
|
|
; AVX512VBMI-NEXT: cmoveq {{[0-9]+}}(%rsp), %r8
|
|
; AVX512VBMI-NEXT: cmoveq {{[0-9]+}}(%rsp), %rdx
|
|
; AVX512VBMI-NEXT: cmovneq {{[0-9]+}}(%rsp), %r11
|
|
; AVX512VBMI-NEXT: cmovneq {{[0-9]+}}(%rsp), %r15
|
|
; AVX512VBMI-NEXT: cmovneq %r12, %rsi
|
|
; AVX512VBMI-NEXT: cmovneq {{[0-9]+}}(%rsp), %rbx
|
|
; AVX512VBMI-NEXT: cmovneq {{[0-9]+}}(%rsp), %r12
|
|
; AVX512VBMI-NEXT: testb $-128, %cl
|
|
; AVX512VBMI-NEXT: cmovneq %r9, %rax
|
|
; AVX512VBMI-NEXT: cmovneq %r14, %r9
|
|
; AVX512VBMI-NEXT: cmovneq %rsi, %r14
|
|
; AVX512VBMI-NEXT: cmoveq %rbx, %r12
|
|
; AVX512VBMI-NEXT: cmoveq %rsi, %rbx
|
|
; AVX512VBMI-NEXT: movq %r10, %rsi
|
|
; AVX512VBMI-NEXT: cmovneq %r13, %rsi
|
|
; AVX512VBMI-NEXT: cmoveq %r11, %r15
|
|
; AVX512VBMI-NEXT: cmoveq %r13, %r11
|
|
; AVX512VBMI-NEXT: cmoveq %rdx, %r10
|
|
; AVX512VBMI-NEXT: cmoveq %r8, %rdx
|
|
; AVX512VBMI-NEXT: testb $64, %cl
|
|
; AVX512VBMI-NEXT: cmovneq %rax, %rdx
|
|
; AVX512VBMI-NEXT: cmovneq %r10, %rax
|
|
; AVX512VBMI-NEXT: cmovneq %r9, %r10
|
|
; AVX512VBMI-NEXT: movq %r9, %r8
|
|
; AVX512VBMI-NEXT: cmovneq %rsi, %r8
|
|
; AVX512VBMI-NEXT: cmovneq %r14, %rsi
|
|
; AVX512VBMI-NEXT: cmovneq %r11, %r14
|
|
; AVX512VBMI-NEXT: cmovneq %rbx, %r11
|
|
; AVX512VBMI-NEXT: cmovneq %r15, %rbx
|
|
; AVX512VBMI-NEXT: cmovneq %r12, %r15
|
|
; AVX512VBMI-NEXT: movq %r8, %r9
|
|
; AVX512VBMI-NEXT: shldq %cl, %rsi, %r9
|
|
; AVX512VBMI-NEXT: movq %rbx, %r12
|
|
; AVX512VBMI-NEXT: shldq %cl, %r15, %r12
|
|
; AVX512VBMI-NEXT: movq %r11, %r15
|
|
; AVX512VBMI-NEXT: shldq %cl, %rbx, %r15
|
|
; AVX512VBMI-NEXT: movq %r14, %rbx
|
|
; AVX512VBMI-NEXT: shldq %cl, %r11, %rbx
|
|
; AVX512VBMI-NEXT: shldq %cl, %r14, %rsi
|
|
; AVX512VBMI-NEXT: movq %r10, %r11
|
|
; AVX512VBMI-NEXT: shldq %cl, %r8, %r11
|
|
; AVX512VBMI-NEXT: movq %rax, %r8
|
|
; AVX512VBMI-NEXT: shldq %cl, %r10, %r8
|
|
; AVX512VBMI-NEXT: # kill: def $cl killed $cl killed $rcx
|
|
; AVX512VBMI-NEXT: shldq %cl, %rax, %rdx
|
|
; AVX512VBMI-NEXT: movq %rdx, 56(%rdi)
|
|
; AVX512VBMI-NEXT: movq %r8, 48(%rdi)
|
|
; AVX512VBMI-NEXT: movq %r11, 40(%rdi)
|
|
; AVX512VBMI-NEXT: movq %r9, 32(%rdi)
|
|
; AVX512VBMI-NEXT: movq %rsi, 24(%rdi)
|
|
; AVX512VBMI-NEXT: movq %rbx, 16(%rdi)
|
|
; AVX512VBMI-NEXT: movq %r15, 8(%rdi)
|
|
; AVX512VBMI-NEXT: movq %r12, (%rdi)
|
|
; AVX512VBMI-NEXT: movq %rdi, %rax
|
|
; AVX512VBMI-NEXT: popq %rbx
|
|
; AVX512VBMI-NEXT: popq %r12
|
|
; AVX512VBMI-NEXT: popq %r13
|
|
; AVX512VBMI-NEXT: popq %r14
|
|
; AVX512VBMI-NEXT: popq %r15
|
|
; AVX512VBMI-NEXT: retq
|
|
%r = call i512 @llvm.fshl.i512(i512 %a0, i512 %a1, i512 %a2)
|
|
ret i512 %r
|
|
}
|
|
|
|
define i512 @fshr_i512(i512 %a0, i512 %a1, i512 %a2) nounwind {
|
|
; SSE-LABEL: fshr_i512:
|
|
; SSE: # %bb.0:
|
|
; SSE-NEXT: pushq %r15
|
|
; SSE-NEXT: pushq %r14
|
|
; SSE-NEXT: pushq %r13
|
|
; SSE-NEXT: pushq %r12
|
|
; SSE-NEXT: pushq %rbx
|
|
; SSE-NEXT: movq %rcx, %rax
|
|
; SSE-NEXT: movq {{[0-9]+}}(%rsp), %r10
|
|
; SSE-NEXT: movq {{[0-9]+}}(%rsp), %r12
|
|
; SSE-NEXT: movq {{[0-9]+}}(%rsp), %r11
|
|
; SSE-NEXT: movq {{[0-9]+}}(%rsp), %rcx
|
|
; SSE-NEXT: testl $256, %ecx # imm = 0x100
|
|
; SSE-NEXT: cmoveq %rsi, %r9
|
|
; SSE-NEXT: movq %r8, %rbx
|
|
; SSE-NEXT: cmoveq %r11, %rbx
|
|
; SSE-NEXT: movq %rdx, %r13
|
|
; SSE-NEXT: cmoveq %r12, %r13
|
|
; SSE-NEXT: cmoveq %r10, %rsi
|
|
; SSE-NEXT: movq {{[0-9]+}}(%rsp), %r14
|
|
; SSE-NEXT: movq %rax, %r15
|
|
; SSE-NEXT: cmoveq %r14, %r15
|
|
; SSE-NEXT: cmovneq {{[0-9]+}}(%rsp), %rax
|
|
; SSE-NEXT: cmovneq {{[0-9]+}}(%rsp), %r8
|
|
; SSE-NEXT: cmovneq {{[0-9]+}}(%rsp), %rdx
|
|
; SSE-NEXT: cmoveq {{[0-9]+}}(%rsp), %r14
|
|
; SSE-NEXT: cmoveq {{[0-9]+}}(%rsp), %r10
|
|
; SSE-NEXT: cmoveq {{[0-9]+}}(%rsp), %r11
|
|
; SSE-NEXT: cmoveq {{[0-9]+}}(%rsp), %r12
|
|
; SSE-NEXT: testb $-128, %cl
|
|
; SSE-NEXT: cmoveq %r9, %rax
|
|
; SSE-NEXT: cmoveq %r15, %r9
|
|
; SSE-NEXT: cmoveq %rsi, %r15
|
|
; SSE-NEXT: cmovneq %r14, %r10
|
|
; SSE-NEXT: cmovneq %rsi, %r14
|
|
; SSE-NEXT: movq %rbx, %rsi
|
|
; SSE-NEXT: cmoveq %r13, %rsi
|
|
; SSE-NEXT: cmovneq %r11, %r12
|
|
; SSE-NEXT: cmovneq %r13, %r11
|
|
; SSE-NEXT: cmovneq %rdx, %rbx
|
|
; SSE-NEXT: cmovneq %r8, %rdx
|
|
; SSE-NEXT: testb $64, %cl
|
|
; SSE-NEXT: cmoveq %rax, %rdx
|
|
; SSE-NEXT: cmoveq %rbx, %rax
|
|
; SSE-NEXT: cmoveq %r9, %rbx
|
|
; SSE-NEXT: movq %r9, %r8
|
|
; SSE-NEXT: cmoveq %rsi, %r8
|
|
; SSE-NEXT: cmoveq %r15, %rsi
|
|
; SSE-NEXT: cmoveq %r11, %r15
|
|
; SSE-NEXT: cmoveq %r14, %r11
|
|
; SSE-NEXT: cmovneq %r12, %r10
|
|
; SSE-NEXT: cmoveq %r12, %r14
|
|
; SSE-NEXT: movq %rsi, %r9
|
|
; SSE-NEXT: shrdq %cl, %r8, %r9
|
|
; SSE-NEXT: shrdq %cl, %r14, %r10
|
|
; SSE-NEXT: shrdq %cl, %r11, %r14
|
|
; SSE-NEXT: shrdq %cl, %r15, %r11
|
|
; SSE-NEXT: shrdq %cl, %rsi, %r15
|
|
; SSE-NEXT: shrdq %cl, %rbx, %r8
|
|
; SSE-NEXT: shrdq %cl, %rax, %rbx
|
|
; SSE-NEXT: # kill: def $cl killed $cl killed $rcx
|
|
; SSE-NEXT: shrdq %cl, %rdx, %rax
|
|
; SSE-NEXT: movq %rax, 56(%rdi)
|
|
; SSE-NEXT: movq %rbx, 48(%rdi)
|
|
; SSE-NEXT: movq %r8, 40(%rdi)
|
|
; SSE-NEXT: movq %r9, 32(%rdi)
|
|
; SSE-NEXT: movq %r15, 24(%rdi)
|
|
; SSE-NEXT: movq %r11, 16(%rdi)
|
|
; SSE-NEXT: movq %r14, 8(%rdi)
|
|
; SSE-NEXT: movq %r10, (%rdi)
|
|
; SSE-NEXT: movq %rdi, %rax
|
|
; SSE-NEXT: popq %rbx
|
|
; SSE-NEXT: popq %r12
|
|
; SSE-NEXT: popq %r13
|
|
; SSE-NEXT: popq %r14
|
|
; SSE-NEXT: popq %r15
|
|
; SSE-NEXT: retq
|
|
;
|
|
; AVX2-LABEL: fshr_i512:
|
|
; AVX2: # %bb.0:
|
|
; AVX2-NEXT: pushq %r15
|
|
; AVX2-NEXT: pushq %r14
|
|
; AVX2-NEXT: pushq %r13
|
|
; AVX2-NEXT: pushq %r12
|
|
; AVX2-NEXT: pushq %rbx
|
|
; AVX2-NEXT: movq %rcx, %rax
|
|
; AVX2-NEXT: movq {{[0-9]+}}(%rsp), %r11
|
|
; AVX2-NEXT: movq {{[0-9]+}}(%rsp), %r10
|
|
; AVX2-NEXT: movq {{[0-9]+}}(%rsp), %r12
|
|
; AVX2-NEXT: movq {{[0-9]+}}(%rsp), %rbx
|
|
; AVX2-NEXT: movq {{[0-9]+}}(%rsp), %rcx
|
|
; AVX2-NEXT: testl $256, %ecx # imm = 0x100
|
|
; AVX2-NEXT: cmoveq %rsi, %r9
|
|
; AVX2-NEXT: movq %r8, %r14
|
|
; AVX2-NEXT: cmoveq %rbx, %r14
|
|
; AVX2-NEXT: movq %rdx, %r13
|
|
; AVX2-NEXT: cmoveq %r12, %r13
|
|
; AVX2-NEXT: cmoveq %r10, %rsi
|
|
; AVX2-NEXT: movq %rax, %r15
|
|
; AVX2-NEXT: cmoveq %r11, %r15
|
|
; AVX2-NEXT: cmovneq {{[0-9]+}}(%rsp), %rax
|
|
; AVX2-NEXT: cmovneq {{[0-9]+}}(%rsp), %r8
|
|
; AVX2-NEXT: cmovneq {{[0-9]+}}(%rsp), %rdx
|
|
; AVX2-NEXT: cmoveq {{[0-9]+}}(%rsp), %r11
|
|
; AVX2-NEXT: cmoveq {{[0-9]+}}(%rsp), %r10
|
|
; AVX2-NEXT: cmoveq {{[0-9]+}}(%rsp), %rbx
|
|
; AVX2-NEXT: cmoveq {{[0-9]+}}(%rsp), %r12
|
|
; AVX2-NEXT: testb $-128, %cl
|
|
; AVX2-NEXT: cmoveq %r9, %rax
|
|
; AVX2-NEXT: cmoveq %r15, %r9
|
|
; AVX2-NEXT: cmoveq %rsi, %r15
|
|
; AVX2-NEXT: cmovneq %r11, %r10
|
|
; AVX2-NEXT: cmovneq %rsi, %r11
|
|
; AVX2-NEXT: movq %r14, %rsi
|
|
; AVX2-NEXT: cmoveq %r13, %rsi
|
|
; AVX2-NEXT: cmovneq %rbx, %r12
|
|
; AVX2-NEXT: cmovneq %r13, %rbx
|
|
; AVX2-NEXT: cmovneq %rdx, %r14
|
|
; AVX2-NEXT: cmovneq %r8, %rdx
|
|
; AVX2-NEXT: testb $64, %cl
|
|
; AVX2-NEXT: cmoveq %rax, %rdx
|
|
; AVX2-NEXT: cmoveq %r14, %rax
|
|
; AVX2-NEXT: cmoveq %r9, %r14
|
|
; AVX2-NEXT: movq %r9, %r8
|
|
; AVX2-NEXT: cmoveq %rsi, %r8
|
|
; AVX2-NEXT: cmoveq %r15, %rsi
|
|
; AVX2-NEXT: cmoveq %rbx, %r15
|
|
; AVX2-NEXT: cmoveq %r11, %rbx
|
|
; AVX2-NEXT: cmovneq %r12, %r10
|
|
; AVX2-NEXT: cmoveq %r12, %r11
|
|
; AVX2-NEXT: movq %rsi, %r9
|
|
; AVX2-NEXT: shrdq %cl, %r8, %r9
|
|
; AVX2-NEXT: shrdq %cl, %r11, %r10
|
|
; AVX2-NEXT: shrdq %cl, %rbx, %r11
|
|
; AVX2-NEXT: shrdq %cl, %r15, %rbx
|
|
; AVX2-NEXT: shrdq %cl, %rsi, %r15
|
|
; AVX2-NEXT: shrdq %cl, %r14, %r8
|
|
; AVX2-NEXT: shrdq %cl, %rax, %r14
|
|
; AVX2-NEXT: # kill: def $cl killed $cl killed $rcx
|
|
; AVX2-NEXT: shrdq %cl, %rdx, %rax
|
|
; AVX2-NEXT: movq %rax, 56(%rdi)
|
|
; AVX2-NEXT: movq %r14, 48(%rdi)
|
|
; AVX2-NEXT: movq %r8, 40(%rdi)
|
|
; AVX2-NEXT: movq %r9, 32(%rdi)
|
|
; AVX2-NEXT: movq %r15, 24(%rdi)
|
|
; AVX2-NEXT: movq %rbx, 16(%rdi)
|
|
; AVX2-NEXT: movq %r11, 8(%rdi)
|
|
; AVX2-NEXT: movq %r10, (%rdi)
|
|
; AVX2-NEXT: movq %rdi, %rax
|
|
; AVX2-NEXT: popq %rbx
|
|
; AVX2-NEXT: popq %r12
|
|
; AVX2-NEXT: popq %r13
|
|
; AVX2-NEXT: popq %r14
|
|
; AVX2-NEXT: popq %r15
|
|
; AVX2-NEXT: retq
|
|
;
|
|
; AVX512F-LABEL: fshr_i512:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: pushq %r15
|
|
; AVX512F-NEXT: pushq %r14
|
|
; AVX512F-NEXT: pushq %r13
|
|
; AVX512F-NEXT: pushq %r12
|
|
; AVX512F-NEXT: pushq %rbx
|
|
; AVX512F-NEXT: movq %rcx, %rax
|
|
; AVX512F-NEXT: movq {{[0-9]+}}(%rsp), %r11
|
|
; AVX512F-NEXT: movq {{[0-9]+}}(%rsp), %r10
|
|
; AVX512F-NEXT: movq {{[0-9]+}}(%rsp), %r12
|
|
; AVX512F-NEXT: movq {{[0-9]+}}(%rsp), %rbx
|
|
; AVX512F-NEXT: movq {{[0-9]+}}(%rsp), %rcx
|
|
; AVX512F-NEXT: testl $256, %ecx # imm = 0x100
|
|
; AVX512F-NEXT: cmoveq %rsi, %r9
|
|
; AVX512F-NEXT: movq %r8, %r14
|
|
; AVX512F-NEXT: cmoveq %rbx, %r14
|
|
; AVX512F-NEXT: movq %rdx, %r13
|
|
; AVX512F-NEXT: cmoveq %r12, %r13
|
|
; AVX512F-NEXT: cmoveq %r10, %rsi
|
|
; AVX512F-NEXT: movq %rax, %r15
|
|
; AVX512F-NEXT: cmoveq %r11, %r15
|
|
; AVX512F-NEXT: cmovneq {{[0-9]+}}(%rsp), %rax
|
|
; AVX512F-NEXT: cmovneq {{[0-9]+}}(%rsp), %r8
|
|
; AVX512F-NEXT: cmovneq {{[0-9]+}}(%rsp), %rdx
|
|
; AVX512F-NEXT: cmoveq {{[0-9]+}}(%rsp), %r11
|
|
; AVX512F-NEXT: cmoveq {{[0-9]+}}(%rsp), %r10
|
|
; AVX512F-NEXT: cmoveq {{[0-9]+}}(%rsp), %rbx
|
|
; AVX512F-NEXT: cmoveq {{[0-9]+}}(%rsp), %r12
|
|
; AVX512F-NEXT: testb $-128, %cl
|
|
; AVX512F-NEXT: cmoveq %r9, %rax
|
|
; AVX512F-NEXT: cmoveq %r15, %r9
|
|
; AVX512F-NEXT: cmoveq %rsi, %r15
|
|
; AVX512F-NEXT: cmovneq %r11, %r10
|
|
; AVX512F-NEXT: cmovneq %rsi, %r11
|
|
; AVX512F-NEXT: movq %r14, %rsi
|
|
; AVX512F-NEXT: cmoveq %r13, %rsi
|
|
; AVX512F-NEXT: cmovneq %rbx, %r12
|
|
; AVX512F-NEXT: cmovneq %r13, %rbx
|
|
; AVX512F-NEXT: cmovneq %rdx, %r14
|
|
; AVX512F-NEXT: cmovneq %r8, %rdx
|
|
; AVX512F-NEXT: testb $64, %cl
|
|
; AVX512F-NEXT: cmoveq %rax, %rdx
|
|
; AVX512F-NEXT: cmoveq %r14, %rax
|
|
; AVX512F-NEXT: cmoveq %r9, %r14
|
|
; AVX512F-NEXT: movq %r9, %r8
|
|
; AVX512F-NEXT: cmoveq %rsi, %r8
|
|
; AVX512F-NEXT: cmoveq %r15, %rsi
|
|
; AVX512F-NEXT: cmoveq %rbx, %r15
|
|
; AVX512F-NEXT: cmoveq %r11, %rbx
|
|
; AVX512F-NEXT: cmovneq %r12, %r10
|
|
; AVX512F-NEXT: cmoveq %r12, %r11
|
|
; AVX512F-NEXT: movq %rsi, %r9
|
|
; AVX512F-NEXT: shrdq %cl, %r8, %r9
|
|
; AVX512F-NEXT: shrdq %cl, %r11, %r10
|
|
; AVX512F-NEXT: shrdq %cl, %rbx, %r11
|
|
; AVX512F-NEXT: shrdq %cl, %r15, %rbx
|
|
; AVX512F-NEXT: shrdq %cl, %rsi, %r15
|
|
; AVX512F-NEXT: shrdq %cl, %r14, %r8
|
|
; AVX512F-NEXT: shrdq %cl, %rax, %r14
|
|
; AVX512F-NEXT: # kill: def $cl killed $cl killed $rcx
|
|
; AVX512F-NEXT: shrdq %cl, %rdx, %rax
|
|
; AVX512F-NEXT: movq %rax, 56(%rdi)
|
|
; AVX512F-NEXT: movq %r14, 48(%rdi)
|
|
; AVX512F-NEXT: movq %r8, 40(%rdi)
|
|
; AVX512F-NEXT: movq %r9, 32(%rdi)
|
|
; AVX512F-NEXT: movq %r15, 24(%rdi)
|
|
; AVX512F-NEXT: movq %rbx, 16(%rdi)
|
|
; AVX512F-NEXT: movq %r11, 8(%rdi)
|
|
; AVX512F-NEXT: movq %r10, (%rdi)
|
|
; AVX512F-NEXT: movq %rdi, %rax
|
|
; AVX512F-NEXT: popq %rbx
|
|
; AVX512F-NEXT: popq %r12
|
|
; AVX512F-NEXT: popq %r13
|
|
; AVX512F-NEXT: popq %r14
|
|
; AVX512F-NEXT: popq %r15
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: fshr_i512:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: pushq %r15
|
|
; AVX512VL-NEXT: pushq %r14
|
|
; AVX512VL-NEXT: pushq %r13
|
|
; AVX512VL-NEXT: pushq %r12
|
|
; AVX512VL-NEXT: pushq %rbx
|
|
; AVX512VL-NEXT: movq %rcx, %rax
|
|
; AVX512VL-NEXT: movq {{[0-9]+}}(%rsp), %r11
|
|
; AVX512VL-NEXT: movq {{[0-9]+}}(%rsp), %r10
|
|
; AVX512VL-NEXT: movq {{[0-9]+}}(%rsp), %r12
|
|
; AVX512VL-NEXT: movq {{[0-9]+}}(%rsp), %rbx
|
|
; AVX512VL-NEXT: movq {{[0-9]+}}(%rsp), %rcx
|
|
; AVX512VL-NEXT: testl $256, %ecx # imm = 0x100
|
|
; AVX512VL-NEXT: cmoveq %rsi, %r9
|
|
; AVX512VL-NEXT: movq %r8, %r14
|
|
; AVX512VL-NEXT: cmoveq %rbx, %r14
|
|
; AVX512VL-NEXT: movq %rdx, %r13
|
|
; AVX512VL-NEXT: cmoveq %r10, %rsi
|
|
; AVX512VL-NEXT: movq %rax, %r15
|
|
; AVX512VL-NEXT: cmoveq %r11, %r15
|
|
; AVX512VL-NEXT: cmovneq {{[0-9]+}}(%rsp), %rax
|
|
; AVX512VL-NEXT: cmovneq {{[0-9]+}}(%rsp), %r8
|
|
; AVX512VL-NEXT: cmovneq {{[0-9]+}}(%rsp), %rdx
|
|
; AVX512VL-NEXT: cmoveq {{[0-9]+}}(%rsp), %r11
|
|
; AVX512VL-NEXT: cmoveq {{[0-9]+}}(%rsp), %r10
|
|
; AVX512VL-NEXT: cmoveq %r12, %r13
|
|
; AVX512VL-NEXT: cmoveq {{[0-9]+}}(%rsp), %rbx
|
|
; AVX512VL-NEXT: cmoveq {{[0-9]+}}(%rsp), %r12
|
|
; AVX512VL-NEXT: testb $-128, %cl
|
|
; AVX512VL-NEXT: cmoveq %r9, %rax
|
|
; AVX512VL-NEXT: cmoveq %r15, %r9
|
|
; AVX512VL-NEXT: cmoveq %rsi, %r15
|
|
; AVX512VL-NEXT: cmovneq %r11, %r10
|
|
; AVX512VL-NEXT: cmovneq %rsi, %r11
|
|
; AVX512VL-NEXT: movq %r14, %rsi
|
|
; AVX512VL-NEXT: cmoveq %r13, %rsi
|
|
; AVX512VL-NEXT: cmovneq %rbx, %r12
|
|
; AVX512VL-NEXT: cmovneq %r13, %rbx
|
|
; AVX512VL-NEXT: cmovneq %rdx, %r14
|
|
; AVX512VL-NEXT: cmovneq %r8, %rdx
|
|
; AVX512VL-NEXT: testb $64, %cl
|
|
; AVX512VL-NEXT: cmoveq %rax, %rdx
|
|
; AVX512VL-NEXT: cmoveq %r14, %rax
|
|
; AVX512VL-NEXT: cmoveq %r9, %r14
|
|
; AVX512VL-NEXT: movq %r9, %r8
|
|
; AVX512VL-NEXT: cmoveq %rsi, %r8
|
|
; AVX512VL-NEXT: cmoveq %r15, %rsi
|
|
; AVX512VL-NEXT: cmoveq %rbx, %r15
|
|
; AVX512VL-NEXT: cmoveq %r11, %rbx
|
|
; AVX512VL-NEXT: cmovneq %r12, %r10
|
|
; AVX512VL-NEXT: cmoveq %r12, %r11
|
|
; AVX512VL-NEXT: movq %rsi, %r9
|
|
; AVX512VL-NEXT: shrdq %cl, %r8, %r9
|
|
; AVX512VL-NEXT: shrdq %cl, %r11, %r10
|
|
; AVX512VL-NEXT: shrdq %cl, %rbx, %r11
|
|
; AVX512VL-NEXT: shrdq %cl, %r15, %rbx
|
|
; AVX512VL-NEXT: shrdq %cl, %rsi, %r15
|
|
; AVX512VL-NEXT: shrdq %cl, %r14, %r8
|
|
; AVX512VL-NEXT: shrdq %cl, %rax, %r14
|
|
; AVX512VL-NEXT: # kill: def $cl killed $cl killed $rcx
|
|
; AVX512VL-NEXT: shrdq %cl, %rdx, %rax
|
|
; AVX512VL-NEXT: movq %rax, 56(%rdi)
|
|
; AVX512VL-NEXT: movq %r14, 48(%rdi)
|
|
; AVX512VL-NEXT: movq %r8, 40(%rdi)
|
|
; AVX512VL-NEXT: movq %r9, 32(%rdi)
|
|
; AVX512VL-NEXT: movq %r15, 24(%rdi)
|
|
; AVX512VL-NEXT: movq %rbx, 16(%rdi)
|
|
; AVX512VL-NEXT: movq %r11, 8(%rdi)
|
|
; AVX512VL-NEXT: movq %r10, (%rdi)
|
|
; AVX512VL-NEXT: movq %rdi, %rax
|
|
; AVX512VL-NEXT: popq %rbx
|
|
; AVX512VL-NEXT: popq %r12
|
|
; AVX512VL-NEXT: popq %r13
|
|
; AVX512VL-NEXT: popq %r14
|
|
; AVX512VL-NEXT: popq %r15
|
|
; AVX512VL-NEXT: retq
|
|
;
|
|
; AVX512VBMI-LABEL: fshr_i512:
|
|
; AVX512VBMI: # %bb.0:
|
|
; AVX512VBMI-NEXT: pushq %r15
|
|
; AVX512VBMI-NEXT: pushq %r14
|
|
; AVX512VBMI-NEXT: pushq %r13
|
|
; AVX512VBMI-NEXT: pushq %r12
|
|
; AVX512VBMI-NEXT: pushq %rbx
|
|
; AVX512VBMI-NEXT: movq %rcx, %rax
|
|
; AVX512VBMI-NEXT: movq {{[0-9]+}}(%rsp), %r11
|
|
; AVX512VBMI-NEXT: movq {{[0-9]+}}(%rsp), %r10
|
|
; AVX512VBMI-NEXT: movq {{[0-9]+}}(%rsp), %r12
|
|
; AVX512VBMI-NEXT: movq {{[0-9]+}}(%rsp), %rbx
|
|
; AVX512VBMI-NEXT: movq {{[0-9]+}}(%rsp), %rcx
|
|
; AVX512VBMI-NEXT: testl $256, %ecx # imm = 0x100
|
|
; AVX512VBMI-NEXT: cmoveq %rsi, %r9
|
|
; AVX512VBMI-NEXT: movq %r8, %r14
|
|
; AVX512VBMI-NEXT: cmoveq %rbx, %r14
|
|
; AVX512VBMI-NEXT: movq %rdx, %r13
|
|
; AVX512VBMI-NEXT: cmoveq %r10, %rsi
|
|
; AVX512VBMI-NEXT: movq %rax, %r15
|
|
; AVX512VBMI-NEXT: cmoveq %r11, %r15
|
|
; AVX512VBMI-NEXT: cmovneq {{[0-9]+}}(%rsp), %rax
|
|
; AVX512VBMI-NEXT: cmovneq {{[0-9]+}}(%rsp), %r8
|
|
; AVX512VBMI-NEXT: cmovneq {{[0-9]+}}(%rsp), %rdx
|
|
; AVX512VBMI-NEXT: cmoveq {{[0-9]+}}(%rsp), %r11
|
|
; AVX512VBMI-NEXT: cmoveq {{[0-9]+}}(%rsp), %r10
|
|
; AVX512VBMI-NEXT: cmoveq %r12, %r13
|
|
; AVX512VBMI-NEXT: cmoveq {{[0-9]+}}(%rsp), %rbx
|
|
; AVX512VBMI-NEXT: cmoveq {{[0-9]+}}(%rsp), %r12
|
|
; AVX512VBMI-NEXT: testb $-128, %cl
|
|
; AVX512VBMI-NEXT: cmoveq %r9, %rax
|
|
; AVX512VBMI-NEXT: cmoveq %r15, %r9
|
|
; AVX512VBMI-NEXT: cmoveq %rsi, %r15
|
|
; AVX512VBMI-NEXT: cmovneq %r11, %r10
|
|
; AVX512VBMI-NEXT: cmovneq %rsi, %r11
|
|
; AVX512VBMI-NEXT: movq %r14, %rsi
|
|
; AVX512VBMI-NEXT: cmoveq %r13, %rsi
|
|
; AVX512VBMI-NEXT: cmovneq %rbx, %r12
|
|
; AVX512VBMI-NEXT: cmovneq %r13, %rbx
|
|
; AVX512VBMI-NEXT: cmovneq %rdx, %r14
|
|
; AVX512VBMI-NEXT: cmovneq %r8, %rdx
|
|
; AVX512VBMI-NEXT: testb $64, %cl
|
|
; AVX512VBMI-NEXT: cmoveq %rax, %rdx
|
|
; AVX512VBMI-NEXT: cmoveq %r14, %rax
|
|
; AVX512VBMI-NEXT: cmoveq %r9, %r14
|
|
; AVX512VBMI-NEXT: movq %r9, %r8
|
|
; AVX512VBMI-NEXT: cmoveq %rsi, %r8
|
|
; AVX512VBMI-NEXT: cmoveq %r15, %rsi
|
|
; AVX512VBMI-NEXT: cmoveq %rbx, %r15
|
|
; AVX512VBMI-NEXT: cmoveq %r11, %rbx
|
|
; AVX512VBMI-NEXT: cmovneq %r12, %r10
|
|
; AVX512VBMI-NEXT: cmoveq %r12, %r11
|
|
; AVX512VBMI-NEXT: movq %rsi, %r9
|
|
; AVX512VBMI-NEXT: shrdq %cl, %r8, %r9
|
|
; AVX512VBMI-NEXT: shrdq %cl, %r11, %r10
|
|
; AVX512VBMI-NEXT: shrdq %cl, %rbx, %r11
|
|
; AVX512VBMI-NEXT: shrdq %cl, %r15, %rbx
|
|
; AVX512VBMI-NEXT: shrdq %cl, %rsi, %r15
|
|
; AVX512VBMI-NEXT: shrdq %cl, %r14, %r8
|
|
; AVX512VBMI-NEXT: shrdq %cl, %rax, %r14
|
|
; AVX512VBMI-NEXT: # kill: def $cl killed $cl killed $rcx
|
|
; AVX512VBMI-NEXT: shrdq %cl, %rdx, %rax
|
|
; AVX512VBMI-NEXT: movq %rax, 56(%rdi)
|
|
; AVX512VBMI-NEXT: movq %r14, 48(%rdi)
|
|
; AVX512VBMI-NEXT: movq %r8, 40(%rdi)
|
|
; AVX512VBMI-NEXT: movq %r9, 32(%rdi)
|
|
; AVX512VBMI-NEXT: movq %r15, 24(%rdi)
|
|
; AVX512VBMI-NEXT: movq %rbx, 16(%rdi)
|
|
; AVX512VBMI-NEXT: movq %r11, 8(%rdi)
|
|
; AVX512VBMI-NEXT: movq %r10, (%rdi)
|
|
; AVX512VBMI-NEXT: movq %rdi, %rax
|
|
; AVX512VBMI-NEXT: popq %rbx
|
|
; AVX512VBMI-NEXT: popq %r12
|
|
; AVX512VBMI-NEXT: popq %r13
|
|
; AVX512VBMI-NEXT: popq %r14
|
|
; AVX512VBMI-NEXT: popq %r15
|
|
; AVX512VBMI-NEXT: retq
|
|
%r = call i512 @llvm.fshr.i512(i512 %a0, i512 %a1, i512 %a2)
|
|
ret i512 %r
|
|
}
|
|
|
|
define i512 @fshl_rot_i512(i512 %a0, i512 %a1) nounwind {
|
|
; SSE-LABEL: fshl_rot_i512:
|
|
; SSE: # %bb.0:
|
|
; SSE-NEXT: pushq %r15
|
|
; SSE-NEXT: pushq %r14
|
|
; SSE-NEXT: pushq %r13
|
|
; SSE-NEXT: pushq %r12
|
|
; SSE-NEXT: pushq %rbx
|
|
; SSE-NEXT: movq %rcx, %rax
|
|
; SSE-NEXT: movq {{[0-9]+}}(%rsp), %rcx
|
|
; SSE-NEXT: testl $256, %ecx # imm = 0x100
|
|
; SSE-NEXT: movq %r9, %r12
|
|
; SSE-NEXT: cmovneq %rsi, %r12
|
|
; SSE-NEXT: movq {{[0-9]+}}(%rsp), %r15
|
|
; SSE-NEXT: movq %r8, %r13
|
|
; SSE-NEXT: cmovneq %r15, %r13
|
|
; SSE-NEXT: movq {{[0-9]+}}(%rsp), %r10
|
|
; SSE-NEXT: movq %rax, %rbx
|
|
; SSE-NEXT: cmovneq %r10, %rbx
|
|
; SSE-NEXT: movq {{[0-9]+}}(%rsp), %r11
|
|
; SSE-NEXT: movq %rdx, %r14
|
|
; SSE-NEXT: cmovneq %r11, %r14
|
|
; SSE-NEXT: cmovneq %r9, %rsi
|
|
; SSE-NEXT: cmovneq %rax, %r10
|
|
; SSE-NEXT: cmovneq %rdx, %r11
|
|
; SSE-NEXT: cmovneq %r8, %r15
|
|
; SSE-NEXT: testb $-128, %cl
|
|
; SSE-NEXT: movq %r14, %r8
|
|
; SSE-NEXT: cmovneq %r15, %r8
|
|
; SSE-NEXT: cmovneq %r11, %r15
|
|
; SSE-NEXT: movq %rbx, %rdx
|
|
; SSE-NEXT: cmovneq %rsi, %rdx
|
|
; SSE-NEXT: cmovneq %r10, %rsi
|
|
; SSE-NEXT: cmoveq %r13, %r14
|
|
; SSE-NEXT: cmoveq %r12, %rbx
|
|
; SSE-NEXT: cmovneq %r13, %r11
|
|
; SSE-NEXT: cmovneq %r12, %r10
|
|
; SSE-NEXT: testb $64, %cl
|
|
; SSE-NEXT: movq %r10, %rax
|
|
; SSE-NEXT: cmovneq %r11, %rax
|
|
; SSE-NEXT: cmovneq %rbx, %r11
|
|
; SSE-NEXT: cmovneq %r14, %rbx
|
|
; SSE-NEXT: cmovneq %rdx, %r14
|
|
; SSE-NEXT: cmovneq %r8, %rdx
|
|
; SSE-NEXT: cmovneq %rsi, %r8
|
|
; SSE-NEXT: cmoveq %r15, %r10
|
|
; SSE-NEXT: cmovneq %r15, %rsi
|
|
; SSE-NEXT: movq %rsi, %r9
|
|
; SSE-NEXT: shldq %cl, %r10, %r9
|
|
; SSE-NEXT: movq %r8, %r15
|
|
; SSE-NEXT: shldq %cl, %rsi, %r15
|
|
; SSE-NEXT: movq %rdx, %rsi
|
|
; SSE-NEXT: shldq %cl, %r8, %rsi
|
|
; SSE-NEXT: movq %r14, %r8
|
|
; SSE-NEXT: shldq %cl, %rdx, %r8
|
|
; SSE-NEXT: movq %rbx, %rdx
|
|
; SSE-NEXT: shldq %cl, %r14, %rdx
|
|
; SSE-NEXT: movq %r11, %r14
|
|
; SSE-NEXT: shldq %cl, %rbx, %r14
|
|
; SSE-NEXT: movq %rax, %rbx
|
|
; SSE-NEXT: shldq %cl, %r11, %rbx
|
|
; SSE-NEXT: # kill: def $cl killed $cl killed $rcx
|
|
; SSE-NEXT: shldq %cl, %rax, %r10
|
|
; SSE-NEXT: movq %rdi, %rax
|
|
; SSE-NEXT: movq %r10, 56(%rdi)
|
|
; SSE-NEXT: movq %rbx, 48(%rdi)
|
|
; SSE-NEXT: movq %r14, 40(%rdi)
|
|
; SSE-NEXT: movq %rdx, 32(%rdi)
|
|
; SSE-NEXT: movq %r8, 24(%rdi)
|
|
; SSE-NEXT: movq %rsi, 16(%rdi)
|
|
; SSE-NEXT: movq %r15, 8(%rdi)
|
|
; SSE-NEXT: movq %r9, (%rdi)
|
|
; SSE-NEXT: popq %rbx
|
|
; SSE-NEXT: popq %r12
|
|
; SSE-NEXT: popq %r13
|
|
; SSE-NEXT: popq %r14
|
|
; SSE-NEXT: popq %r15
|
|
; SSE-NEXT: retq
|
|
;
|
|
; AVX2-LABEL: fshl_rot_i512:
|
|
; AVX2: # %bb.0:
|
|
; AVX2-NEXT: pushq %r15
|
|
; AVX2-NEXT: pushq %r14
|
|
; AVX2-NEXT: pushq %r13
|
|
; AVX2-NEXT: pushq %r12
|
|
; AVX2-NEXT: pushq %rbx
|
|
; AVX2-NEXT: movq %rcx, %rax
|
|
; AVX2-NEXT: movq {{[0-9]+}}(%rsp), %r11
|
|
; AVX2-NEXT: movq {{[0-9]+}}(%rsp), %r10
|
|
; AVX2-NEXT: movq {{[0-9]+}}(%rsp), %r15
|
|
; AVX2-NEXT: movq {{[0-9]+}}(%rsp), %rcx
|
|
; AVX2-NEXT: testl $256, %ecx # imm = 0x100
|
|
; AVX2-NEXT: movq %r9, %r12
|
|
; AVX2-NEXT: cmovneq %rsi, %r12
|
|
; AVX2-NEXT: movq %r8, %r13
|
|
; AVX2-NEXT: cmovneq %r15, %r13
|
|
; AVX2-NEXT: movq %rax, %rbx
|
|
; AVX2-NEXT: cmovneq %r10, %rbx
|
|
; AVX2-NEXT: movq %rdx, %r14
|
|
; AVX2-NEXT: cmovneq %r11, %r14
|
|
; AVX2-NEXT: cmovneq %r9, %rsi
|
|
; AVX2-NEXT: cmovneq %rax, %r10
|
|
; AVX2-NEXT: cmovneq %rdx, %r11
|
|
; AVX2-NEXT: cmovneq %r8, %r15
|
|
; AVX2-NEXT: testb $-128, %cl
|
|
; AVX2-NEXT: movq %r14, %r9
|
|
; AVX2-NEXT: cmovneq %r15, %r9
|
|
; AVX2-NEXT: cmovneq %r11, %r15
|
|
; AVX2-NEXT: movq %rbx, %r8
|
|
; AVX2-NEXT: cmovneq %rsi, %r8
|
|
; AVX2-NEXT: cmovneq %r10, %rsi
|
|
; AVX2-NEXT: cmoveq %r13, %r14
|
|
; AVX2-NEXT: cmoveq %r12, %rbx
|
|
; AVX2-NEXT: cmovneq %r13, %r11
|
|
; AVX2-NEXT: movq %rdi, %rax
|
|
; AVX2-NEXT: cmovneq %r12, %r10
|
|
; AVX2-NEXT: testb $64, %cl
|
|
; AVX2-NEXT: movq %r10, %rdx
|
|
; AVX2-NEXT: cmovneq %r11, %rdx
|
|
; AVX2-NEXT: cmovneq %rbx, %r11
|
|
; AVX2-NEXT: cmovneq %r14, %rbx
|
|
; AVX2-NEXT: cmovneq %r8, %r14
|
|
; AVX2-NEXT: cmovneq %r9, %r8
|
|
; AVX2-NEXT: cmovneq %rsi, %r9
|
|
; AVX2-NEXT: cmoveq %r15, %r10
|
|
; AVX2-NEXT: cmovneq %r15, %rsi
|
|
; AVX2-NEXT: movq %rsi, %rdi
|
|
; AVX2-NEXT: shldq %cl, %r10, %rdi
|
|
; AVX2-NEXT: movq %r9, %r15
|
|
; AVX2-NEXT: shldq %cl, %rsi, %r15
|
|
; AVX2-NEXT: movq %r8, %rsi
|
|
; AVX2-NEXT: shldq %cl, %r9, %rsi
|
|
; AVX2-NEXT: movq %r14, %r9
|
|
; AVX2-NEXT: shldq %cl, %r8, %r9
|
|
; AVX2-NEXT: movq %rbx, %r8
|
|
; AVX2-NEXT: shldq %cl, %r14, %r8
|
|
; AVX2-NEXT: movq %r11, %r14
|
|
; AVX2-NEXT: shldq %cl, %rbx, %r14
|
|
; AVX2-NEXT: movq %rdx, %rbx
|
|
; AVX2-NEXT: shldq %cl, %r11, %rbx
|
|
; AVX2-NEXT: # kill: def $cl killed $cl killed $rcx
|
|
; AVX2-NEXT: shldq %cl, %rdx, %r10
|
|
; AVX2-NEXT: movq %r10, 56(%rax)
|
|
; AVX2-NEXT: movq %rbx, 48(%rax)
|
|
; AVX2-NEXT: movq %r14, 40(%rax)
|
|
; AVX2-NEXT: movq %r8, 32(%rax)
|
|
; AVX2-NEXT: movq %r9, 24(%rax)
|
|
; AVX2-NEXT: movq %rsi, 16(%rax)
|
|
; AVX2-NEXT: movq %r15, 8(%rax)
|
|
; AVX2-NEXT: movq %rdi, (%rax)
|
|
; AVX2-NEXT: popq %rbx
|
|
; AVX2-NEXT: popq %r12
|
|
; AVX2-NEXT: popq %r13
|
|
; AVX2-NEXT: popq %r14
|
|
; AVX2-NEXT: popq %r15
|
|
; AVX2-NEXT: retq
|
|
;
|
|
; AVX512F-LABEL: fshl_rot_i512:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: pushq %r15
|
|
; AVX512F-NEXT: pushq %r14
|
|
; AVX512F-NEXT: pushq %r13
|
|
; AVX512F-NEXT: pushq %r12
|
|
; AVX512F-NEXT: pushq %rbx
|
|
; AVX512F-NEXT: movq %rcx, %rax
|
|
; AVX512F-NEXT: movq {{[0-9]+}}(%rsp), %r11
|
|
; AVX512F-NEXT: movq {{[0-9]+}}(%rsp), %r10
|
|
; AVX512F-NEXT: movq {{[0-9]+}}(%rsp), %r15
|
|
; AVX512F-NEXT: movq {{[0-9]+}}(%rsp), %rcx
|
|
; AVX512F-NEXT: testl $256, %ecx # imm = 0x100
|
|
; AVX512F-NEXT: movq %r9, %r12
|
|
; AVX512F-NEXT: cmovneq %rsi, %r12
|
|
; AVX512F-NEXT: movq %r8, %r13
|
|
; AVX512F-NEXT: cmovneq %r15, %r13
|
|
; AVX512F-NEXT: movq %rax, %rbx
|
|
; AVX512F-NEXT: cmovneq %r10, %rbx
|
|
; AVX512F-NEXT: movq %rdx, %r14
|
|
; AVX512F-NEXT: cmovneq %r11, %r14
|
|
; AVX512F-NEXT: cmovneq %r9, %rsi
|
|
; AVX512F-NEXT: cmovneq %rax, %r10
|
|
; AVX512F-NEXT: cmovneq %rdx, %r11
|
|
; AVX512F-NEXT: cmovneq %r8, %r15
|
|
; AVX512F-NEXT: testb $-128, %cl
|
|
; AVX512F-NEXT: movq %r14, %r9
|
|
; AVX512F-NEXT: cmovneq %r15, %r9
|
|
; AVX512F-NEXT: cmovneq %r11, %r15
|
|
; AVX512F-NEXT: movq %rbx, %r8
|
|
; AVX512F-NEXT: cmovneq %rsi, %r8
|
|
; AVX512F-NEXT: cmovneq %r10, %rsi
|
|
; AVX512F-NEXT: cmoveq %r13, %r14
|
|
; AVX512F-NEXT: cmoveq %r12, %rbx
|
|
; AVX512F-NEXT: cmovneq %r13, %r11
|
|
; AVX512F-NEXT: movq %rdi, %rax
|
|
; AVX512F-NEXT: cmovneq %r12, %r10
|
|
; AVX512F-NEXT: testb $64, %cl
|
|
; AVX512F-NEXT: movq %r10, %rdx
|
|
; AVX512F-NEXT: cmovneq %r11, %rdx
|
|
; AVX512F-NEXT: cmovneq %rbx, %r11
|
|
; AVX512F-NEXT: cmovneq %r14, %rbx
|
|
; AVX512F-NEXT: cmovneq %r8, %r14
|
|
; AVX512F-NEXT: cmovneq %r9, %r8
|
|
; AVX512F-NEXT: cmovneq %rsi, %r9
|
|
; AVX512F-NEXT: cmoveq %r15, %r10
|
|
; AVX512F-NEXT: cmovneq %r15, %rsi
|
|
; AVX512F-NEXT: movq %rsi, %rdi
|
|
; AVX512F-NEXT: shldq %cl, %r10, %rdi
|
|
; AVX512F-NEXT: movq %r9, %r15
|
|
; AVX512F-NEXT: shldq %cl, %rsi, %r15
|
|
; AVX512F-NEXT: movq %r8, %rsi
|
|
; AVX512F-NEXT: shldq %cl, %r9, %rsi
|
|
; AVX512F-NEXT: movq %r14, %r9
|
|
; AVX512F-NEXT: shldq %cl, %r8, %r9
|
|
; AVX512F-NEXT: movq %rbx, %r8
|
|
; AVX512F-NEXT: shldq %cl, %r14, %r8
|
|
; AVX512F-NEXT: movq %r11, %r14
|
|
; AVX512F-NEXT: shldq %cl, %rbx, %r14
|
|
; AVX512F-NEXT: movq %rdx, %rbx
|
|
; AVX512F-NEXT: shldq %cl, %r11, %rbx
|
|
; AVX512F-NEXT: # kill: def $cl killed $cl killed $rcx
|
|
; AVX512F-NEXT: shldq %cl, %rdx, %r10
|
|
; AVX512F-NEXT: movq %r10, 56(%rax)
|
|
; AVX512F-NEXT: movq %rbx, 48(%rax)
|
|
; AVX512F-NEXT: movq %r14, 40(%rax)
|
|
; AVX512F-NEXT: movq %r8, 32(%rax)
|
|
; AVX512F-NEXT: movq %r9, 24(%rax)
|
|
; AVX512F-NEXT: movq %rsi, 16(%rax)
|
|
; AVX512F-NEXT: movq %r15, 8(%rax)
|
|
; AVX512F-NEXT: movq %rdi, (%rax)
|
|
; AVX512F-NEXT: popq %rbx
|
|
; AVX512F-NEXT: popq %r12
|
|
; AVX512F-NEXT: popq %r13
|
|
; AVX512F-NEXT: popq %r14
|
|
; AVX512F-NEXT: popq %r15
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: fshl_rot_i512:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: pushq %r15
|
|
; AVX512VL-NEXT: pushq %r14
|
|
; AVX512VL-NEXT: pushq %r13
|
|
; AVX512VL-NEXT: pushq %r12
|
|
; AVX512VL-NEXT: pushq %rbx
|
|
; AVX512VL-NEXT: movq %rcx, %rax
|
|
; AVX512VL-NEXT: movq {{[0-9]+}}(%rsp), %r11
|
|
; AVX512VL-NEXT: movq {{[0-9]+}}(%rsp), %r10
|
|
; AVX512VL-NEXT: movq {{[0-9]+}}(%rsp), %r15
|
|
; AVX512VL-NEXT: movq {{[0-9]+}}(%rsp), %rcx
|
|
; AVX512VL-NEXT: testl $256, %ecx # imm = 0x100
|
|
; AVX512VL-NEXT: movq %r9, %r12
|
|
; AVX512VL-NEXT: cmovneq %rsi, %r12
|
|
; AVX512VL-NEXT: movq %r8, %r13
|
|
; AVX512VL-NEXT: cmovneq %r15, %r13
|
|
; AVX512VL-NEXT: movq %rax, %rbx
|
|
; AVX512VL-NEXT: cmovneq %r10, %rbx
|
|
; AVX512VL-NEXT: movq %rdx, %r14
|
|
; AVX512VL-NEXT: cmovneq %r11, %r14
|
|
; AVX512VL-NEXT: cmovneq %r9, %rsi
|
|
; AVX512VL-NEXT: cmovneq %rax, %r10
|
|
; AVX512VL-NEXT: cmovneq %rdx, %r11
|
|
; AVX512VL-NEXT: cmovneq %r8, %r15
|
|
; AVX512VL-NEXT: testb $-128, %cl
|
|
; AVX512VL-NEXT: movq %r14, %r8
|
|
; AVX512VL-NEXT: cmovneq %r15, %r8
|
|
; AVX512VL-NEXT: cmovneq %r11, %r15
|
|
; AVX512VL-NEXT: movq %rbx, %rax
|
|
; AVX512VL-NEXT: cmovneq %rsi, %rax
|
|
; AVX512VL-NEXT: cmovneq %r10, %rsi
|
|
; AVX512VL-NEXT: cmoveq %r13, %r14
|
|
; AVX512VL-NEXT: cmoveq %r12, %rbx
|
|
; AVX512VL-NEXT: cmovneq %r13, %r11
|
|
; AVX512VL-NEXT: cmovneq %r12, %r10
|
|
; AVX512VL-NEXT: testb $64, %cl
|
|
; AVX512VL-NEXT: movq %r10, %rdx
|
|
; AVX512VL-NEXT: cmovneq %r11, %rdx
|
|
; AVX512VL-NEXT: cmovneq %rbx, %r11
|
|
; AVX512VL-NEXT: cmovneq %r14, %rbx
|
|
; AVX512VL-NEXT: cmovneq %rax, %r14
|
|
; AVX512VL-NEXT: cmovneq %r8, %rax
|
|
; AVX512VL-NEXT: cmovneq %rsi, %r8
|
|
; AVX512VL-NEXT: cmoveq %r15, %r10
|
|
; AVX512VL-NEXT: cmovneq %r15, %rsi
|
|
; AVX512VL-NEXT: movq %rsi, %r9
|
|
; AVX512VL-NEXT: shldq %cl, %r10, %r9
|
|
; AVX512VL-NEXT: movq %r8, %r15
|
|
; AVX512VL-NEXT: shldq %cl, %rsi, %r15
|
|
; AVX512VL-NEXT: movq %rax, %rsi
|
|
; AVX512VL-NEXT: shldq %cl, %r8, %rsi
|
|
; AVX512VL-NEXT: movq %r14, %r8
|
|
; AVX512VL-NEXT: shldq %cl, %rax, %r8
|
|
; AVX512VL-NEXT: movq %rbx, %r12
|
|
; AVX512VL-NEXT: shldq %cl, %r14, %r12
|
|
; AVX512VL-NEXT: movq %r11, %r14
|
|
; AVX512VL-NEXT: shldq %cl, %rbx, %r14
|
|
; AVX512VL-NEXT: movq %rdi, %rax
|
|
; AVX512VL-NEXT: movq %rdx, %rdi
|
|
; AVX512VL-NEXT: shldq %cl, %r11, %rdi
|
|
; AVX512VL-NEXT: # kill: def $cl killed $cl killed $rcx
|
|
; AVX512VL-NEXT: shldq %cl, %rdx, %r10
|
|
; AVX512VL-NEXT: movq %r10, 56(%rax)
|
|
; AVX512VL-NEXT: movq %rdi, 48(%rax)
|
|
; AVX512VL-NEXT: movq %r14, 40(%rax)
|
|
; AVX512VL-NEXT: movq %r12, 32(%rax)
|
|
; AVX512VL-NEXT: movq %r8, 24(%rax)
|
|
; AVX512VL-NEXT: movq %rsi, 16(%rax)
|
|
; AVX512VL-NEXT: movq %r15, 8(%rax)
|
|
; AVX512VL-NEXT: movq %r9, (%rax)
|
|
; AVX512VL-NEXT: popq %rbx
|
|
; AVX512VL-NEXT: popq %r12
|
|
; AVX512VL-NEXT: popq %r13
|
|
; AVX512VL-NEXT: popq %r14
|
|
; AVX512VL-NEXT: popq %r15
|
|
; AVX512VL-NEXT: retq
|
|
;
|
|
; AVX512VBMI-LABEL: fshl_rot_i512:
|
|
; AVX512VBMI: # %bb.0:
|
|
; AVX512VBMI-NEXT: pushq %r15
|
|
; AVX512VBMI-NEXT: pushq %r14
|
|
; AVX512VBMI-NEXT: pushq %r13
|
|
; AVX512VBMI-NEXT: pushq %r12
|
|
; AVX512VBMI-NEXT: pushq %rbx
|
|
; AVX512VBMI-NEXT: movq %rcx, %rax
|
|
; AVX512VBMI-NEXT: movq {{[0-9]+}}(%rsp), %r11
|
|
; AVX512VBMI-NEXT: movq {{[0-9]+}}(%rsp), %r10
|
|
; AVX512VBMI-NEXT: movq {{[0-9]+}}(%rsp), %r15
|
|
; AVX512VBMI-NEXT: movq {{[0-9]+}}(%rsp), %rcx
|
|
; AVX512VBMI-NEXT: testl $256, %ecx # imm = 0x100
|
|
; AVX512VBMI-NEXT: movq %r9, %r12
|
|
; AVX512VBMI-NEXT: cmovneq %rsi, %r12
|
|
; AVX512VBMI-NEXT: movq %r8, %r13
|
|
; AVX512VBMI-NEXT: cmovneq %r15, %r13
|
|
; AVX512VBMI-NEXT: movq %rax, %rbx
|
|
; AVX512VBMI-NEXT: cmovneq %r10, %rbx
|
|
; AVX512VBMI-NEXT: movq %rdx, %r14
|
|
; AVX512VBMI-NEXT: cmovneq %r11, %r14
|
|
; AVX512VBMI-NEXT: cmovneq %r9, %rsi
|
|
; AVX512VBMI-NEXT: cmovneq %rax, %r10
|
|
; AVX512VBMI-NEXT: cmovneq %rdx, %r11
|
|
; AVX512VBMI-NEXT: cmovneq %r8, %r15
|
|
; AVX512VBMI-NEXT: testb $-128, %cl
|
|
; AVX512VBMI-NEXT: movq %r14, %r8
|
|
; AVX512VBMI-NEXT: cmovneq %r15, %r8
|
|
; AVX512VBMI-NEXT: cmovneq %r11, %r15
|
|
; AVX512VBMI-NEXT: movq %rbx, %rax
|
|
; AVX512VBMI-NEXT: cmovneq %rsi, %rax
|
|
; AVX512VBMI-NEXT: cmovneq %r10, %rsi
|
|
; AVX512VBMI-NEXT: cmoveq %r13, %r14
|
|
; AVX512VBMI-NEXT: cmoveq %r12, %rbx
|
|
; AVX512VBMI-NEXT: cmovneq %r13, %r11
|
|
; AVX512VBMI-NEXT: cmovneq %r12, %r10
|
|
; AVX512VBMI-NEXT: testb $64, %cl
|
|
; AVX512VBMI-NEXT: movq %r10, %rdx
|
|
; AVX512VBMI-NEXT: cmovneq %r11, %rdx
|
|
; AVX512VBMI-NEXT: cmovneq %rbx, %r11
|
|
; AVX512VBMI-NEXT: cmovneq %r14, %rbx
|
|
; AVX512VBMI-NEXT: cmovneq %rax, %r14
|
|
; AVX512VBMI-NEXT: cmovneq %r8, %rax
|
|
; AVX512VBMI-NEXT: cmovneq %rsi, %r8
|
|
; AVX512VBMI-NEXT: cmoveq %r15, %r10
|
|
; AVX512VBMI-NEXT: cmovneq %r15, %rsi
|
|
; AVX512VBMI-NEXT: movq %rsi, %r9
|
|
; AVX512VBMI-NEXT: shldq %cl, %r10, %r9
|
|
; AVX512VBMI-NEXT: movq %r8, %r15
|
|
; AVX512VBMI-NEXT: shldq %cl, %rsi, %r15
|
|
; AVX512VBMI-NEXT: movq %rax, %rsi
|
|
; AVX512VBMI-NEXT: shldq %cl, %r8, %rsi
|
|
; AVX512VBMI-NEXT: movq %r14, %r8
|
|
; AVX512VBMI-NEXT: shldq %cl, %rax, %r8
|
|
; AVX512VBMI-NEXT: movq %rbx, %r12
|
|
; AVX512VBMI-NEXT: shldq %cl, %r14, %r12
|
|
; AVX512VBMI-NEXT: movq %r11, %r14
|
|
; AVX512VBMI-NEXT: shldq %cl, %rbx, %r14
|
|
; AVX512VBMI-NEXT: movq %rdi, %rax
|
|
; AVX512VBMI-NEXT: movq %rdx, %rdi
|
|
; AVX512VBMI-NEXT: shldq %cl, %r11, %rdi
|
|
; AVX512VBMI-NEXT: # kill: def $cl killed $cl killed $rcx
|
|
; AVX512VBMI-NEXT: shldq %cl, %rdx, %r10
|
|
; AVX512VBMI-NEXT: movq %r10, 56(%rax)
|
|
; AVX512VBMI-NEXT: movq %rdi, 48(%rax)
|
|
; AVX512VBMI-NEXT: movq %r14, 40(%rax)
|
|
; AVX512VBMI-NEXT: movq %r12, 32(%rax)
|
|
; AVX512VBMI-NEXT: movq %r8, 24(%rax)
|
|
; AVX512VBMI-NEXT: movq %rsi, 16(%rax)
|
|
; AVX512VBMI-NEXT: movq %r15, 8(%rax)
|
|
; AVX512VBMI-NEXT: movq %r9, (%rax)
|
|
; AVX512VBMI-NEXT: popq %rbx
|
|
; AVX512VBMI-NEXT: popq %r12
|
|
; AVX512VBMI-NEXT: popq %r13
|
|
; AVX512VBMI-NEXT: popq %r14
|
|
; AVX512VBMI-NEXT: popq %r15
|
|
; AVX512VBMI-NEXT: retq
|
|
%r = call i512 @llvm.fshl.i512(i512 %a0, i512 %a0, i512 %a1)
|
|
ret i512 %r
|
|
}
|
|
|
|
define i512 @fshr_rot_i512(i512 %a0, i512 %a1) nounwind {
|
|
; SSE-LABEL: fshr_rot_i512:
|
|
; SSE: # %bb.0:
|
|
; SSE-NEXT: pushq %r15
|
|
; SSE-NEXT: pushq %r14
|
|
; SSE-NEXT: pushq %r13
|
|
; SSE-NEXT: pushq %r12
|
|
; SSE-NEXT: pushq %rbx
|
|
; SSE-NEXT: movq %rcx, %rax
|
|
; SSE-NEXT: movq {{[0-9]+}}(%rsp), %rcx
|
|
; SSE-NEXT: testl $256, %ecx # imm = 0x100
|
|
; SSE-NEXT: movq %r9, %r12
|
|
; SSE-NEXT: cmoveq %rsi, %r12
|
|
; SSE-NEXT: movq {{[0-9]+}}(%rsp), %r15
|
|
; SSE-NEXT: movq %r8, %r13
|
|
; SSE-NEXT: cmoveq %r15, %r13
|
|
; SSE-NEXT: movq {{[0-9]+}}(%rsp), %r14
|
|
; SSE-NEXT: movq %rax, %r10
|
|
; SSE-NEXT: cmoveq %r14, %r10
|
|
; SSE-NEXT: movq {{[0-9]+}}(%rsp), %rbx
|
|
; SSE-NEXT: movq %rdx, %r11
|
|
; SSE-NEXT: cmoveq %rbx, %r11
|
|
; SSE-NEXT: cmoveq %r9, %rsi
|
|
; SSE-NEXT: cmoveq %rax, %r14
|
|
; SSE-NEXT: cmoveq %rdx, %rbx
|
|
; SSE-NEXT: cmoveq %r8, %r15
|
|
; SSE-NEXT: testb $-128, %cl
|
|
; SSE-NEXT: movq %r11, %rdx
|
|
; SSE-NEXT: cmoveq %r15, %rdx
|
|
; SSE-NEXT: cmoveq %rbx, %r15
|
|
; SSE-NEXT: movq %r10, %r8
|
|
; SSE-NEXT: cmoveq %rsi, %r8
|
|
; SSE-NEXT: cmoveq %r14, %rsi
|
|
; SSE-NEXT: cmovneq %r13, %r11
|
|
; SSE-NEXT: cmovneq %r12, %r10
|
|
; SSE-NEXT: cmoveq %r13, %rbx
|
|
; SSE-NEXT: cmoveq %r12, %r14
|
|
; SSE-NEXT: testb $64, %cl
|
|
; SSE-NEXT: movq %r14, %r9
|
|
; SSE-NEXT: cmoveq %rbx, %r9
|
|
; SSE-NEXT: cmoveq %r10, %rbx
|
|
; SSE-NEXT: cmoveq %r11, %r10
|
|
; SSE-NEXT: cmoveq %r8, %r11
|
|
; SSE-NEXT: cmoveq %rdx, %r8
|
|
; SSE-NEXT: cmoveq %rsi, %rdx
|
|
; SSE-NEXT: cmoveq %r15, %rsi
|
|
; SSE-NEXT: cmovneq %r15, %r14
|
|
; SSE-NEXT: movq %r14, %r15
|
|
; SSE-NEXT: shrdq %cl, %rsi, %r15
|
|
; SSE-NEXT: shrdq %cl, %rdx, %rsi
|
|
; SSE-NEXT: shrdq %cl, %r8, %rdx
|
|
; SSE-NEXT: shrdq %cl, %r11, %r8
|
|
; SSE-NEXT: shrdq %cl, %r10, %r11
|
|
; SSE-NEXT: shrdq %cl, %rbx, %r10
|
|
; SSE-NEXT: shrdq %cl, %r9, %rbx
|
|
; SSE-NEXT: # kill: def $cl killed $cl killed $rcx
|
|
; SSE-NEXT: shrdq %cl, %r14, %r9
|
|
; SSE-NEXT: movq %rdi, %rax
|
|
; SSE-NEXT: movq %r9, 56(%rdi)
|
|
; SSE-NEXT: movq %rbx, 48(%rdi)
|
|
; SSE-NEXT: movq %r10, 40(%rdi)
|
|
; SSE-NEXT: movq %r11, 32(%rdi)
|
|
; SSE-NEXT: movq %r8, 24(%rdi)
|
|
; SSE-NEXT: movq %rdx, 16(%rdi)
|
|
; SSE-NEXT: movq %rsi, 8(%rdi)
|
|
; SSE-NEXT: movq %r15, (%rdi)
|
|
; SSE-NEXT: popq %rbx
|
|
; SSE-NEXT: popq %r12
|
|
; SSE-NEXT: popq %r13
|
|
; SSE-NEXT: popq %r14
|
|
; SSE-NEXT: popq %r15
|
|
; SSE-NEXT: retq
|
|
;
|
|
; AVX2-LABEL: fshr_rot_i512:
|
|
; AVX2: # %bb.0:
|
|
; AVX2-NEXT: pushq %r15
|
|
; AVX2-NEXT: pushq %r14
|
|
; AVX2-NEXT: pushq %r13
|
|
; AVX2-NEXT: pushq %r12
|
|
; AVX2-NEXT: pushq %rbx
|
|
; AVX2-NEXT: movq %rcx, %rax
|
|
; AVX2-NEXT: movq {{[0-9]+}}(%rsp), %r10
|
|
; AVX2-NEXT: movq {{[0-9]+}}(%rsp), %r11
|
|
; AVX2-NEXT: movq {{[0-9]+}}(%rsp), %r15
|
|
; AVX2-NEXT: movq {{[0-9]+}}(%rsp), %rcx
|
|
; AVX2-NEXT: testl $256, %ecx # imm = 0x100
|
|
; AVX2-NEXT: movq %r9, %r12
|
|
; AVX2-NEXT: cmoveq %rsi, %r12
|
|
; AVX2-NEXT: movq %r8, %r13
|
|
; AVX2-NEXT: cmoveq %r15, %r13
|
|
; AVX2-NEXT: movq %rax, %rbx
|
|
; AVX2-NEXT: cmoveq %r11, %rbx
|
|
; AVX2-NEXT: movq %rdx, %r14
|
|
; AVX2-NEXT: cmoveq %r10, %r14
|
|
; AVX2-NEXT: cmoveq %r9, %rsi
|
|
; AVX2-NEXT: cmoveq %rax, %r11
|
|
; AVX2-NEXT: cmoveq %rdx, %r10
|
|
; AVX2-NEXT: cmoveq %r8, %r15
|
|
; AVX2-NEXT: testb $-128, %cl
|
|
; AVX2-NEXT: movq %r14, %rdx
|
|
; AVX2-NEXT: cmoveq %r15, %rdx
|
|
; AVX2-NEXT: cmoveq %r10, %r15
|
|
; AVX2-NEXT: movq %rbx, %r8
|
|
; AVX2-NEXT: cmoveq %rsi, %r8
|
|
; AVX2-NEXT: cmoveq %r11, %rsi
|
|
; AVX2-NEXT: cmovneq %r13, %r14
|
|
; AVX2-NEXT: cmovneq %r12, %rbx
|
|
; AVX2-NEXT: cmoveq %r13, %r10
|
|
; AVX2-NEXT: movq %rdi, %rax
|
|
; AVX2-NEXT: cmoveq %r12, %r11
|
|
; AVX2-NEXT: testb $64, %cl
|
|
; AVX2-NEXT: movq %r11, %rdi
|
|
; AVX2-NEXT: cmoveq %r10, %rdi
|
|
; AVX2-NEXT: cmoveq %rbx, %r10
|
|
; AVX2-NEXT: cmoveq %r14, %rbx
|
|
; AVX2-NEXT: cmoveq %r8, %r14
|
|
; AVX2-NEXT: cmoveq %rdx, %r8
|
|
; AVX2-NEXT: cmoveq %rsi, %rdx
|
|
; AVX2-NEXT: cmoveq %r15, %rsi
|
|
; AVX2-NEXT: cmovneq %r15, %r11
|
|
; AVX2-NEXT: movq %r11, %r9
|
|
; AVX2-NEXT: shrdq %cl, %rsi, %r9
|
|
; AVX2-NEXT: shrdq %cl, %rdx, %rsi
|
|
; AVX2-NEXT: shrdq %cl, %r8, %rdx
|
|
; AVX2-NEXT: shrdq %cl, %r14, %r8
|
|
; AVX2-NEXT: shrdq %cl, %rbx, %r14
|
|
; AVX2-NEXT: shrdq %cl, %r10, %rbx
|
|
; AVX2-NEXT: shrdq %cl, %rdi, %r10
|
|
; AVX2-NEXT: # kill: def $cl killed $cl killed $rcx
|
|
; AVX2-NEXT: shrdq %cl, %r11, %rdi
|
|
; AVX2-NEXT: movq %rdi, 56(%rax)
|
|
; AVX2-NEXT: movq %r10, 48(%rax)
|
|
; AVX2-NEXT: movq %rbx, 40(%rax)
|
|
; AVX2-NEXT: movq %r14, 32(%rax)
|
|
; AVX2-NEXT: movq %r8, 24(%rax)
|
|
; AVX2-NEXT: movq %rdx, 16(%rax)
|
|
; AVX2-NEXT: movq %rsi, 8(%rax)
|
|
; AVX2-NEXT: movq %r9, (%rax)
|
|
; AVX2-NEXT: popq %rbx
|
|
; AVX2-NEXT: popq %r12
|
|
; AVX2-NEXT: popq %r13
|
|
; AVX2-NEXT: popq %r14
|
|
; AVX2-NEXT: popq %r15
|
|
; AVX2-NEXT: retq
|
|
;
|
|
; AVX512F-LABEL: fshr_rot_i512:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: pushq %r15
|
|
; AVX512F-NEXT: pushq %r14
|
|
; AVX512F-NEXT: pushq %r13
|
|
; AVX512F-NEXT: pushq %r12
|
|
; AVX512F-NEXT: pushq %rbx
|
|
; AVX512F-NEXT: movq %rcx, %rax
|
|
; AVX512F-NEXT: movq {{[0-9]+}}(%rsp), %r10
|
|
; AVX512F-NEXT: movq {{[0-9]+}}(%rsp), %r11
|
|
; AVX512F-NEXT: movq {{[0-9]+}}(%rsp), %r15
|
|
; AVX512F-NEXT: movq {{[0-9]+}}(%rsp), %rcx
|
|
; AVX512F-NEXT: testl $256, %ecx # imm = 0x100
|
|
; AVX512F-NEXT: movq %r9, %r12
|
|
; AVX512F-NEXT: cmoveq %rsi, %r12
|
|
; AVX512F-NEXT: movq %r8, %r13
|
|
; AVX512F-NEXT: cmoveq %r15, %r13
|
|
; AVX512F-NEXT: movq %rax, %rbx
|
|
; AVX512F-NEXT: cmoveq %r11, %rbx
|
|
; AVX512F-NEXT: movq %rdx, %r14
|
|
; AVX512F-NEXT: cmoveq %r10, %r14
|
|
; AVX512F-NEXT: cmoveq %r9, %rsi
|
|
; AVX512F-NEXT: cmoveq %rax, %r11
|
|
; AVX512F-NEXT: cmoveq %rdx, %r10
|
|
; AVX512F-NEXT: cmoveq %r8, %r15
|
|
; AVX512F-NEXT: testb $-128, %cl
|
|
; AVX512F-NEXT: movq %r14, %rdx
|
|
; AVX512F-NEXT: cmoveq %r15, %rdx
|
|
; AVX512F-NEXT: cmoveq %r10, %r15
|
|
; AVX512F-NEXT: movq %rbx, %r8
|
|
; AVX512F-NEXT: cmoveq %rsi, %r8
|
|
; AVX512F-NEXT: cmoveq %r11, %rsi
|
|
; AVX512F-NEXT: cmovneq %r13, %r14
|
|
; AVX512F-NEXT: cmovneq %r12, %rbx
|
|
; AVX512F-NEXT: cmoveq %r13, %r10
|
|
; AVX512F-NEXT: movq %rdi, %rax
|
|
; AVX512F-NEXT: cmoveq %r12, %r11
|
|
; AVX512F-NEXT: testb $64, %cl
|
|
; AVX512F-NEXT: movq %r11, %rdi
|
|
; AVX512F-NEXT: cmoveq %r10, %rdi
|
|
; AVX512F-NEXT: cmoveq %rbx, %r10
|
|
; AVX512F-NEXT: cmoveq %r14, %rbx
|
|
; AVX512F-NEXT: cmoveq %r8, %r14
|
|
; AVX512F-NEXT: cmoveq %rdx, %r8
|
|
; AVX512F-NEXT: cmoveq %rsi, %rdx
|
|
; AVX512F-NEXT: cmoveq %r15, %rsi
|
|
; AVX512F-NEXT: cmovneq %r15, %r11
|
|
; AVX512F-NEXT: movq %r11, %r9
|
|
; AVX512F-NEXT: shrdq %cl, %rsi, %r9
|
|
; AVX512F-NEXT: shrdq %cl, %rdx, %rsi
|
|
; AVX512F-NEXT: shrdq %cl, %r8, %rdx
|
|
; AVX512F-NEXT: shrdq %cl, %r14, %r8
|
|
; AVX512F-NEXT: shrdq %cl, %rbx, %r14
|
|
; AVX512F-NEXT: shrdq %cl, %r10, %rbx
|
|
; AVX512F-NEXT: shrdq %cl, %rdi, %r10
|
|
; AVX512F-NEXT: # kill: def $cl killed $cl killed $rcx
|
|
; AVX512F-NEXT: shrdq %cl, %r11, %rdi
|
|
; AVX512F-NEXT: movq %rdi, 56(%rax)
|
|
; AVX512F-NEXT: movq %r10, 48(%rax)
|
|
; AVX512F-NEXT: movq %rbx, 40(%rax)
|
|
; AVX512F-NEXT: movq %r14, 32(%rax)
|
|
; AVX512F-NEXT: movq %r8, 24(%rax)
|
|
; AVX512F-NEXT: movq %rdx, 16(%rax)
|
|
; AVX512F-NEXT: movq %rsi, 8(%rax)
|
|
; AVX512F-NEXT: movq %r9, (%rax)
|
|
; AVX512F-NEXT: popq %rbx
|
|
; AVX512F-NEXT: popq %r12
|
|
; AVX512F-NEXT: popq %r13
|
|
; AVX512F-NEXT: popq %r14
|
|
; AVX512F-NEXT: popq %r15
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: fshr_rot_i512:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: pushq %r15
|
|
; AVX512VL-NEXT: pushq %r14
|
|
; AVX512VL-NEXT: pushq %r13
|
|
; AVX512VL-NEXT: pushq %r12
|
|
; AVX512VL-NEXT: pushq %rbx
|
|
; AVX512VL-NEXT: movq %rcx, %rax
|
|
; AVX512VL-NEXT: movq {{[0-9]+}}(%rsp), %r10
|
|
; AVX512VL-NEXT: movq {{[0-9]+}}(%rsp), %r11
|
|
; AVX512VL-NEXT: movq {{[0-9]+}}(%rsp), %r15
|
|
; AVX512VL-NEXT: movq {{[0-9]+}}(%rsp), %rcx
|
|
; AVX512VL-NEXT: testl $256, %ecx # imm = 0x100
|
|
; AVX512VL-NEXT: movq %r9, %r12
|
|
; AVX512VL-NEXT: cmoveq %rsi, %r12
|
|
; AVX512VL-NEXT: movq %r8, %r13
|
|
; AVX512VL-NEXT: cmoveq %r15, %r13
|
|
; AVX512VL-NEXT: movq %rax, %rbx
|
|
; AVX512VL-NEXT: cmoveq %r11, %rbx
|
|
; AVX512VL-NEXT: movq %rdx, %r14
|
|
; AVX512VL-NEXT: cmoveq %r10, %r14
|
|
; AVX512VL-NEXT: cmoveq %r9, %rsi
|
|
; AVX512VL-NEXT: cmoveq %rax, %r11
|
|
; AVX512VL-NEXT: cmoveq %rdx, %r10
|
|
; AVX512VL-NEXT: cmoveq %r8, %r15
|
|
; AVX512VL-NEXT: testb $-128, %cl
|
|
; AVX512VL-NEXT: movq %r14, %rdx
|
|
; AVX512VL-NEXT: cmoveq %r15, %rdx
|
|
; AVX512VL-NEXT: cmoveq %r10, %r15
|
|
; AVX512VL-NEXT: movq %rbx, %r8
|
|
; AVX512VL-NEXT: cmoveq %rsi, %r8
|
|
; AVX512VL-NEXT: cmoveq %r11, %rsi
|
|
; AVX512VL-NEXT: cmovneq %r13, %r14
|
|
; AVX512VL-NEXT: cmovneq %r12, %rbx
|
|
; AVX512VL-NEXT: cmoveq %r13, %r10
|
|
; AVX512VL-NEXT: cmoveq %r12, %r11
|
|
; AVX512VL-NEXT: testb $64, %cl
|
|
; AVX512VL-NEXT: movq %r11, %r9
|
|
; AVX512VL-NEXT: cmoveq %r10, %r9
|
|
; AVX512VL-NEXT: cmoveq %rbx, %r10
|
|
; AVX512VL-NEXT: cmoveq %r14, %rbx
|
|
; AVX512VL-NEXT: cmoveq %r8, %r14
|
|
; AVX512VL-NEXT: cmoveq %rdx, %r8
|
|
; AVX512VL-NEXT: cmoveq %rsi, %rdx
|
|
; AVX512VL-NEXT: cmoveq %r15, %rsi
|
|
; AVX512VL-NEXT: cmovneq %r15, %r11
|
|
; AVX512VL-NEXT: movq %r11, %r15
|
|
; AVX512VL-NEXT: shrdq %cl, %rsi, %r15
|
|
; AVX512VL-NEXT: shrdq %cl, %rdx, %rsi
|
|
; AVX512VL-NEXT: shrdq %cl, %r8, %rdx
|
|
; AVX512VL-NEXT: shrdq %cl, %r14, %r8
|
|
; AVX512VL-NEXT: shrdq %cl, %rbx, %r14
|
|
; AVX512VL-NEXT: shrdq %cl, %r10, %rbx
|
|
; AVX512VL-NEXT: movq %rdi, %rax
|
|
; AVX512VL-NEXT: shrdq %cl, %r9, %r10
|
|
; AVX512VL-NEXT: # kill: def $cl killed $cl killed $rcx
|
|
; AVX512VL-NEXT: shrdq %cl, %r11, %r9
|
|
; AVX512VL-NEXT: movq %r9, 56(%rdi)
|
|
; AVX512VL-NEXT: movq %r10, 48(%rdi)
|
|
; AVX512VL-NEXT: movq %rbx, 40(%rdi)
|
|
; AVX512VL-NEXT: movq %r14, 32(%rdi)
|
|
; AVX512VL-NEXT: movq %r8, 24(%rdi)
|
|
; AVX512VL-NEXT: movq %rdx, 16(%rdi)
|
|
; AVX512VL-NEXT: movq %rsi, 8(%rdi)
|
|
; AVX512VL-NEXT: movq %r15, (%rdi)
|
|
; AVX512VL-NEXT: popq %rbx
|
|
; AVX512VL-NEXT: popq %r12
|
|
; AVX512VL-NEXT: popq %r13
|
|
; AVX512VL-NEXT: popq %r14
|
|
; AVX512VL-NEXT: popq %r15
|
|
; AVX512VL-NEXT: retq
|
|
;
|
|
; AVX512VBMI-LABEL: fshr_rot_i512:
|
|
; AVX512VBMI: # %bb.0:
|
|
; AVX512VBMI-NEXT: pushq %r15
|
|
; AVX512VBMI-NEXT: pushq %r14
|
|
; AVX512VBMI-NEXT: pushq %r13
|
|
; AVX512VBMI-NEXT: pushq %r12
|
|
; AVX512VBMI-NEXT: pushq %rbx
|
|
; AVX512VBMI-NEXT: movq %rcx, %rax
|
|
; AVX512VBMI-NEXT: movq {{[0-9]+}}(%rsp), %r10
|
|
; AVX512VBMI-NEXT: movq {{[0-9]+}}(%rsp), %r11
|
|
; AVX512VBMI-NEXT: movq {{[0-9]+}}(%rsp), %r15
|
|
; AVX512VBMI-NEXT: movq {{[0-9]+}}(%rsp), %rcx
|
|
; AVX512VBMI-NEXT: testl $256, %ecx # imm = 0x100
|
|
; AVX512VBMI-NEXT: movq %r9, %r12
|
|
; AVX512VBMI-NEXT: cmoveq %rsi, %r12
|
|
; AVX512VBMI-NEXT: movq %r8, %r13
|
|
; AVX512VBMI-NEXT: cmoveq %r15, %r13
|
|
; AVX512VBMI-NEXT: movq %rax, %rbx
|
|
; AVX512VBMI-NEXT: cmoveq %r11, %rbx
|
|
; AVX512VBMI-NEXT: movq %rdx, %r14
|
|
; AVX512VBMI-NEXT: cmoveq %r10, %r14
|
|
; AVX512VBMI-NEXT: cmoveq %r9, %rsi
|
|
; AVX512VBMI-NEXT: cmoveq %rax, %r11
|
|
; AVX512VBMI-NEXT: cmoveq %rdx, %r10
|
|
; AVX512VBMI-NEXT: cmoveq %r8, %r15
|
|
; AVX512VBMI-NEXT: testb $-128, %cl
|
|
; AVX512VBMI-NEXT: movq %r14, %rdx
|
|
; AVX512VBMI-NEXT: cmoveq %r15, %rdx
|
|
; AVX512VBMI-NEXT: cmoveq %r10, %r15
|
|
; AVX512VBMI-NEXT: movq %rbx, %r8
|
|
; AVX512VBMI-NEXT: cmoveq %rsi, %r8
|
|
; AVX512VBMI-NEXT: cmoveq %r11, %rsi
|
|
; AVX512VBMI-NEXT: cmovneq %r13, %r14
|
|
; AVX512VBMI-NEXT: cmovneq %r12, %rbx
|
|
; AVX512VBMI-NEXT: cmoveq %r13, %r10
|
|
; AVX512VBMI-NEXT: cmoveq %r12, %r11
|
|
; AVX512VBMI-NEXT: testb $64, %cl
|
|
; AVX512VBMI-NEXT: movq %r11, %r9
|
|
; AVX512VBMI-NEXT: cmoveq %r10, %r9
|
|
; AVX512VBMI-NEXT: cmoveq %rbx, %r10
|
|
; AVX512VBMI-NEXT: cmoveq %r14, %rbx
|
|
; AVX512VBMI-NEXT: cmoveq %r8, %r14
|
|
; AVX512VBMI-NEXT: cmoveq %rdx, %r8
|
|
; AVX512VBMI-NEXT: cmoveq %rsi, %rdx
|
|
; AVX512VBMI-NEXT: cmoveq %r15, %rsi
|
|
; AVX512VBMI-NEXT: cmovneq %r15, %r11
|
|
; AVX512VBMI-NEXT: movq %r11, %r15
|
|
; AVX512VBMI-NEXT: shrdq %cl, %rsi, %r15
|
|
; AVX512VBMI-NEXT: shrdq %cl, %rdx, %rsi
|
|
; AVX512VBMI-NEXT: shrdq %cl, %r8, %rdx
|
|
; AVX512VBMI-NEXT: shrdq %cl, %r14, %r8
|
|
; AVX512VBMI-NEXT: shrdq %cl, %rbx, %r14
|
|
; AVX512VBMI-NEXT: shrdq %cl, %r10, %rbx
|
|
; AVX512VBMI-NEXT: movq %rdi, %rax
|
|
; AVX512VBMI-NEXT: shrdq %cl, %r9, %r10
|
|
; AVX512VBMI-NEXT: # kill: def $cl killed $cl killed $rcx
|
|
; AVX512VBMI-NEXT: shrdq %cl, %r11, %r9
|
|
; AVX512VBMI-NEXT: movq %r9, 56(%rdi)
|
|
; AVX512VBMI-NEXT: movq %r10, 48(%rdi)
|
|
; AVX512VBMI-NEXT: movq %rbx, 40(%rdi)
|
|
; AVX512VBMI-NEXT: movq %r14, 32(%rdi)
|
|
; AVX512VBMI-NEXT: movq %r8, 24(%rdi)
|
|
; AVX512VBMI-NEXT: movq %rdx, 16(%rdi)
|
|
; AVX512VBMI-NEXT: movq %rsi, 8(%rdi)
|
|
; AVX512VBMI-NEXT: movq %r15, (%rdi)
|
|
; AVX512VBMI-NEXT: popq %rbx
|
|
; AVX512VBMI-NEXT: popq %r12
|
|
; AVX512VBMI-NEXT: popq %r13
|
|
; AVX512VBMI-NEXT: popq %r14
|
|
; AVX512VBMI-NEXT: popq %r15
|
|
; AVX512VBMI-NEXT: retq
|
|
%r = call i512 @llvm.fshr.i512(i512 %a0, i512 %a0, i512 %a1)
|
|
ret i512 %r
|
|
}
|
|
|
|
define i512 @fshl_i512_load(ptr %p0, ptr %p1, i512 %a2) nounwind {
|
|
; SSE-LABEL: fshl_i512_load:
|
|
; SSE: # %bb.0:
|
|
; SSE-NEXT: pushq %rbp
|
|
; SSE-NEXT: pushq %r15
|
|
; SSE-NEXT: pushq %r14
|
|
; SSE-NEXT: pushq %r13
|
|
; SSE-NEXT: pushq %r12
|
|
; SSE-NEXT: pushq %rbx
|
|
; SSE-NEXT: movq (%rsi), %r10
|
|
; SSE-NEXT: movq 8(%rsi), %r9
|
|
; SSE-NEXT: movq 16(%rsi), %r8
|
|
; SSE-NEXT: movq 24(%rsi), %rax
|
|
; SSE-NEXT: movq 32(%rdx), %rbx
|
|
; SSE-NEXT: testl $256, %ecx # imm = 0x100
|
|
; SSE-NEXT: movq 56(%rsi), %r13
|
|
; SSE-NEXT: cmovneq %rax, %r13
|
|
; SSE-NEXT: movq 40(%rdx), %r12
|
|
; SSE-NEXT: movq 48(%rsi), %rbp
|
|
; SSE-NEXT: cmovneq %r8, %rbp
|
|
; SSE-NEXT: movq 48(%rdx), %r14
|
|
; SSE-NEXT: movq 40(%rsi), %r11
|
|
; SSE-NEXT: cmovneq %r9, %r11
|
|
; SSE-NEXT: movq 56(%rdx), %r15
|
|
; SSE-NEXT: movq 32(%rsi), %rsi
|
|
; SSE-NEXT: cmovneq %r10, %rsi
|
|
; SSE-NEXT: cmovneq %r15, %rax
|
|
; SSE-NEXT: cmovneq %r14, %r8
|
|
; SSE-NEXT: cmovneq %r12, %r9
|
|
; SSE-NEXT: cmovneq %rbx, %r10
|
|
; SSE-NEXT: cmovneq 16(%rdx), %r14
|
|
; SSE-NEXT: cmovneq (%rdx), %rbx
|
|
; SSE-NEXT: cmovneq 8(%rdx), %r12
|
|
; SSE-NEXT: cmovneq 24(%rdx), %r15
|
|
; SSE-NEXT: testb $-128, %cl
|
|
; SSE-NEXT: cmoveq %r15, %r12
|
|
; SSE-NEXT: cmoveq %r14, %rbx
|
|
; SSE-NEXT: cmoveq %r10, %r14
|
|
; SSE-NEXT: cmoveq %r9, %r15
|
|
; SSE-NEXT: cmoveq %r8, %r10
|
|
; SSE-NEXT: cmoveq %rax, %r9
|
|
; SSE-NEXT: cmoveq %rsi, %r8
|
|
; SSE-NEXT: cmoveq %r11, %rax
|
|
; SSE-NEXT: cmoveq %rbp, %rsi
|
|
; SSE-NEXT: cmoveq %r13, %r11
|
|
; SSE-NEXT: testb $64, %cl
|
|
; SSE-NEXT: cmovneq %rsi, %r11
|
|
; SSE-NEXT: cmovneq %rax, %rsi
|
|
; SSE-NEXT: cmovneq %r8, %rax
|
|
; SSE-NEXT: cmovneq %r9, %r8
|
|
; SSE-NEXT: cmovneq %r10, %r9
|
|
; SSE-NEXT: cmovneq %r15, %r10
|
|
; SSE-NEXT: cmovneq %r14, %r15
|
|
; SSE-NEXT: cmoveq %r12, %rbx
|
|
; SSE-NEXT: cmovneq %r12, %r14
|
|
; SSE-NEXT: movq %r14, %rdx
|
|
; SSE-NEXT: shldq %cl, %rbx, %rdx
|
|
; SSE-NEXT: movq %r15, %rbx
|
|
; SSE-NEXT: shldq %cl, %r14, %rbx
|
|
; SSE-NEXT: movq %r10, %r14
|
|
; SSE-NEXT: shldq %cl, %r15, %r14
|
|
; SSE-NEXT: movq %r9, %r15
|
|
; SSE-NEXT: shldq %cl, %r10, %r15
|
|
; SSE-NEXT: movq %r8, %r10
|
|
; SSE-NEXT: shldq %cl, %r9, %r10
|
|
; SSE-NEXT: movq %rax, %r9
|
|
; SSE-NEXT: shldq %cl, %r8, %r9
|
|
; SSE-NEXT: movq %rsi, %r8
|
|
; SSE-NEXT: shldq %cl, %rax, %r8
|
|
; SSE-NEXT: # kill: def $cl killed $cl killed $rcx
|
|
; SSE-NEXT: shldq %cl, %rsi, %r11
|
|
; SSE-NEXT: movq %rdi, %rax
|
|
; SSE-NEXT: movq %r11, 56(%rdi)
|
|
; SSE-NEXT: movq %r8, 48(%rdi)
|
|
; SSE-NEXT: movq %r9, 40(%rdi)
|
|
; SSE-NEXT: movq %r10, 32(%rdi)
|
|
; SSE-NEXT: movq %r15, 24(%rdi)
|
|
; SSE-NEXT: movq %r14, 16(%rdi)
|
|
; SSE-NEXT: movq %rbx, 8(%rdi)
|
|
; SSE-NEXT: movq %rdx, (%rdi)
|
|
; SSE-NEXT: popq %rbx
|
|
; SSE-NEXT: popq %r12
|
|
; SSE-NEXT: popq %r13
|
|
; SSE-NEXT: popq %r14
|
|
; SSE-NEXT: popq %r15
|
|
; SSE-NEXT: popq %rbp
|
|
; SSE-NEXT: retq
|
|
;
|
|
; AVX2-LABEL: fshl_i512_load:
|
|
; AVX2: # %bb.0:
|
|
; AVX2-NEXT: pushq %rbp
|
|
; AVX2-NEXT: pushq %r15
|
|
; AVX2-NEXT: pushq %r14
|
|
; AVX2-NEXT: pushq %r13
|
|
; AVX2-NEXT: pushq %r12
|
|
; AVX2-NEXT: pushq %rbx
|
|
; AVX2-NEXT: movq (%rsi), %r11
|
|
; AVX2-NEXT: movq 8(%rsi), %r10
|
|
; AVX2-NEXT: movq 16(%rsi), %r9
|
|
; AVX2-NEXT: movq 24(%rsi), %r8
|
|
; AVX2-NEXT: movq 32(%rdx), %r14
|
|
; AVX2-NEXT: testl $256, %ecx # imm = 0x100
|
|
; AVX2-NEXT: movq 56(%rsi), %rax
|
|
; AVX2-NEXT: cmovneq %r8, %rax
|
|
; AVX2-NEXT: movq 40(%rdx), %r13
|
|
; AVX2-NEXT: movq 48(%rsi), %rbp
|
|
; AVX2-NEXT: cmovneq %r9, %rbp
|
|
; AVX2-NEXT: movq 48(%rdx), %r15
|
|
; AVX2-NEXT: movq 40(%rsi), %rbx
|
|
; AVX2-NEXT: cmovneq %r10, %rbx
|
|
; AVX2-NEXT: movq 56(%rdx), %r12
|
|
; AVX2-NEXT: movq 32(%rsi), %rsi
|
|
; AVX2-NEXT: cmovneq %r11, %rsi
|
|
; AVX2-NEXT: cmovneq %r12, %r8
|
|
; AVX2-NEXT: cmovneq %r15, %r9
|
|
; AVX2-NEXT: cmovneq %r13, %r10
|
|
; AVX2-NEXT: cmovneq %r14, %r11
|
|
; AVX2-NEXT: cmovneq 16(%rdx), %r15
|
|
; AVX2-NEXT: cmovneq (%rdx), %r14
|
|
; AVX2-NEXT: cmovneq 8(%rdx), %r13
|
|
; AVX2-NEXT: cmovneq 24(%rdx), %r12
|
|
; AVX2-NEXT: testb $-128, %cl
|
|
; AVX2-NEXT: cmoveq %r12, %r13
|
|
; AVX2-NEXT: cmoveq %r15, %r14
|
|
; AVX2-NEXT: cmoveq %r11, %r15
|
|
; AVX2-NEXT: cmoveq %r10, %r12
|
|
; AVX2-NEXT: cmoveq %r9, %r11
|
|
; AVX2-NEXT: cmoveq %r8, %r10
|
|
; AVX2-NEXT: cmoveq %rsi, %r9
|
|
; AVX2-NEXT: cmoveq %rbx, %r8
|
|
; AVX2-NEXT: cmoveq %rbp, %rsi
|
|
; AVX2-NEXT: cmoveq %rax, %rbx
|
|
; AVX2-NEXT: movq %rdi, %rax
|
|
; AVX2-NEXT: testb $64, %cl
|
|
; AVX2-NEXT: cmovneq %rsi, %rbx
|
|
; AVX2-NEXT: cmovneq %r8, %rsi
|
|
; AVX2-NEXT: cmovneq %r9, %r8
|
|
; AVX2-NEXT: cmovneq %r10, %r9
|
|
; AVX2-NEXT: cmovneq %r11, %r10
|
|
; AVX2-NEXT: cmovneq %r12, %r11
|
|
; AVX2-NEXT: cmovneq %r15, %r12
|
|
; AVX2-NEXT: cmoveq %r13, %r14
|
|
; AVX2-NEXT: cmovneq %r13, %r15
|
|
; AVX2-NEXT: movq %r15, %rdx
|
|
; AVX2-NEXT: shldq %cl, %r14, %rdx
|
|
; AVX2-NEXT: movq %r12, %rdi
|
|
; AVX2-NEXT: shldq %cl, %r15, %rdi
|
|
; AVX2-NEXT: movq %r11, %r14
|
|
; AVX2-NEXT: shldq %cl, %r12, %r14
|
|
; AVX2-NEXT: movq %r10, %r15
|
|
; AVX2-NEXT: shldq %cl, %r11, %r15
|
|
; AVX2-NEXT: movq %r9, %r11
|
|
; AVX2-NEXT: shldq %cl, %r10, %r11
|
|
; AVX2-NEXT: movq %r8, %r10
|
|
; AVX2-NEXT: shldq %cl, %r9, %r10
|
|
; AVX2-NEXT: movq %rsi, %r9
|
|
; AVX2-NEXT: shldq %cl, %r8, %r9
|
|
; AVX2-NEXT: # kill: def $cl killed $cl killed $rcx
|
|
; AVX2-NEXT: shldq %cl, %rsi, %rbx
|
|
; AVX2-NEXT: movq %rbx, 56(%rax)
|
|
; AVX2-NEXT: movq %r9, 48(%rax)
|
|
; AVX2-NEXT: movq %r10, 40(%rax)
|
|
; AVX2-NEXT: movq %r11, 32(%rax)
|
|
; AVX2-NEXT: movq %r15, 24(%rax)
|
|
; AVX2-NEXT: movq %r14, 16(%rax)
|
|
; AVX2-NEXT: movq %rdi, 8(%rax)
|
|
; AVX2-NEXT: movq %rdx, (%rax)
|
|
; AVX2-NEXT: popq %rbx
|
|
; AVX2-NEXT: popq %r12
|
|
; AVX2-NEXT: popq %r13
|
|
; AVX2-NEXT: popq %r14
|
|
; AVX2-NEXT: popq %r15
|
|
; AVX2-NEXT: popq %rbp
|
|
; AVX2-NEXT: retq
|
|
;
|
|
; AVX512F-LABEL: fshl_i512_load:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: movq %rdi, %rax
|
|
; AVX512F-NEXT: vmovdqu64 (%rsi), %zmm1
|
|
; AVX512F-NEXT: vmovdqu64 (%rdx), %zmm2
|
|
; AVX512F-NEXT: andl $511, %ecx # imm = 0x1FF
|
|
; AVX512F-NEXT: vmovq %rcx, %xmm0
|
|
; AVX512F-NEXT: vpbroadcastq %xmm0, %xmm3
|
|
; AVX512F-NEXT: vpbroadcastq {{.*#+}} xmm0 = [63,63]
|
|
; AVX512F-NEXT: vpand %xmm0, %xmm3, %xmm4
|
|
; AVX512F-NEXT: movl %ecx, %edx
|
|
; AVX512F-NEXT: shrl $6, %edx
|
|
; AVX512F-NEXT: movl $-1, %esi
|
|
; AVX512F-NEXT: shlxl %edx, %esi, %edx
|
|
; AVX512F-NEXT: kmovw %edx, %k1
|
|
; AVX512F-NEXT: vpexpandq %zmm1, %zmm5 {%k1} {z}
|
|
; AVX512F-NEXT: vpsllq %xmm4, %zmm5, %zmm4
|
|
; AVX512F-NEXT: vpandn %xmm0, %xmm3, %xmm3
|
|
; AVX512F-NEXT: vpxor %xmm6, %xmm6, %xmm6
|
|
; AVX512F-NEXT: valignq {{.*#+}} zmm5 = zmm6[7],zmm5[0,1,2,3,4,5,6]
|
|
; AVX512F-NEXT: vpsrlq $1, %zmm5, %zmm5
|
|
; AVX512F-NEXT: vpsrlq %xmm3, %zmm5, %zmm3
|
|
; AVX512F-NEXT: vporq %zmm3, %zmm4, %zmm3
|
|
; AVX512F-NEXT: movl $512, %edx # imm = 0x200
|
|
; AVX512F-NEXT: subq %rcx, %rdx
|
|
; AVX512F-NEXT: vmovq %rdx, %xmm4
|
|
; AVX512F-NEXT: vpbroadcastq %xmm4, %xmm4
|
|
; AVX512F-NEXT: vpand %xmm0, %xmm4, %xmm5
|
|
; AVX512F-NEXT: shrl $6, %edx
|
|
; AVX512F-NEXT: shlxl %edx, %esi, %edx
|
|
; AVX512F-NEXT: kmovw %edx, %k1
|
|
; AVX512F-NEXT: vpcompressq %zmm2, %zmm2 {%k1} {z}
|
|
; AVX512F-NEXT: vpsrlq %xmm5, %zmm2, %zmm5
|
|
; AVX512F-NEXT: vpandn %xmm0, %xmm4, %xmm0
|
|
; AVX512F-NEXT: valignq {{.*#+}} zmm2 = zmm2[1,2,3,4,5,6,7],zmm6[0]
|
|
; AVX512F-NEXT: vpaddq %zmm2, %zmm2, %zmm2
|
|
; AVX512F-NEXT: vpsllq %xmm0, %zmm2, %zmm0
|
|
; AVX512F-NEXT: xorl %edx, %edx
|
|
; AVX512F-NEXT: negq %rcx
|
|
; AVX512F-NEXT: sbbl %edx, %edx
|
|
; AVX512F-NEXT: vporq %zmm5, %zmm0, %zmm0
|
|
; AVX512F-NEXT: kmovw %edx, %k1
|
|
; AVX512F-NEXT: vporq %zmm0, %zmm3, %zmm1 {%k1}
|
|
; AVX512F-NEXT: vmovdqu64 %zmm1, (%rdi)
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: fshl_i512_load:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: vmovdqu64 (%rsi), %zmm0
|
|
; AVX512VL-NEXT: vmovdqu64 (%rdx), %zmm1
|
|
; AVX512VL-NEXT: andl $511, %ecx # imm = 0x1FF
|
|
; AVX512VL-NEXT: vpbroadcastq %rcx, %xmm2
|
|
; AVX512VL-NEXT: vpbroadcastq {{.*#+}} xmm3 = [63,63]
|
|
; AVX512VL-NEXT: vpand %xmm3, %xmm2, %xmm4
|
|
; AVX512VL-NEXT: movl %ecx, %eax
|
|
; AVX512VL-NEXT: shrl $6, %eax
|
|
; AVX512VL-NEXT: movl $-1, %edx
|
|
; AVX512VL-NEXT: shlxl %eax, %edx, %eax
|
|
; AVX512VL-NEXT: kmovd %eax, %k1
|
|
; AVX512VL-NEXT: vpexpandq %zmm0, %zmm5 {%k1} {z}
|
|
; AVX512VL-NEXT: vpsllq %xmm4, %zmm5, %zmm4
|
|
; AVX512VL-NEXT: vpandn %xmm3, %xmm2, %xmm2
|
|
; AVX512VL-NEXT: vpxor %xmm6, %xmm6, %xmm6
|
|
; AVX512VL-NEXT: valignq {{.*#+}} zmm5 = zmm6[7],zmm5[0,1,2,3,4,5,6]
|
|
; AVX512VL-NEXT: vpsrlq $1, %zmm5, %zmm5
|
|
; AVX512VL-NEXT: vpsrlq %xmm2, %zmm5, %zmm2
|
|
; AVX512VL-NEXT: movq %rdi, %rax
|
|
; AVX512VL-NEXT: vporq %zmm2, %zmm4, %zmm2
|
|
; AVX512VL-NEXT: movl $512, %esi # imm = 0x200
|
|
; AVX512VL-NEXT: subq %rcx, %rsi
|
|
; AVX512VL-NEXT: vpbroadcastq %rsi, %xmm4
|
|
; AVX512VL-NEXT: vpand %xmm3, %xmm4, %xmm5
|
|
; AVX512VL-NEXT: shrl $6, %esi
|
|
; AVX512VL-NEXT: shlxl %esi, %edx, %edx
|
|
; AVX512VL-NEXT: kmovd %edx, %k1
|
|
; AVX512VL-NEXT: vpcompressq %zmm1, %zmm1 {%k1} {z}
|
|
; AVX512VL-NEXT: vpsrlq %xmm5, %zmm1, %zmm5
|
|
; AVX512VL-NEXT: vpandn %xmm3, %xmm4, %xmm3
|
|
; AVX512VL-NEXT: valignq {{.*#+}} zmm1 = zmm1[1,2,3,4,5,6,7],zmm6[0]
|
|
; AVX512VL-NEXT: vpaddq %zmm1, %zmm1, %zmm1
|
|
; AVX512VL-NEXT: vpsllq %xmm3, %zmm1, %zmm1
|
|
; AVX512VL-NEXT: vporq %zmm5, %zmm1, %zmm1
|
|
; AVX512VL-NEXT: xorl %edx, %edx
|
|
; AVX512VL-NEXT: negq %rcx
|
|
; AVX512VL-NEXT: sbbl %edx, %edx
|
|
; AVX512VL-NEXT: kmovd %edx, %k1
|
|
; AVX512VL-NEXT: vporq %zmm1, %zmm2, %zmm0 {%k1}
|
|
; AVX512VL-NEXT: vmovdqu64 %zmm0, (%rdi)
|
|
; AVX512VL-NEXT: vzeroupper
|
|
; AVX512VL-NEXT: retq
|
|
;
|
|
; AVX512VBMI-LABEL: fshl_i512_load:
|
|
; AVX512VBMI: # %bb.0:
|
|
; AVX512VBMI-NEXT: movq %rdi, %rax
|
|
; AVX512VBMI-NEXT: vmovdqu64 (%rsi), %zmm0
|
|
; AVX512VBMI-NEXT: vmovdqu64 (%rdx), %zmm1
|
|
; AVX512VBMI-NEXT: andl $511, %ecx # imm = 0x1FF
|
|
; AVX512VBMI-NEXT: movl $512, %edx # imm = 0x200
|
|
; AVX512VBMI-NEXT: subq %rcx, %rdx
|
|
; AVX512VBMI-NEXT: vpbroadcastq %rdx, %zmm2
|
|
; AVX512VBMI-NEXT: # kill: def $edx killed $edx killed $rdx
|
|
; AVX512VBMI-NEXT: shrl $6, %edx
|
|
; AVX512VBMI-NEXT: movl $-1, %esi
|
|
; AVX512VBMI-NEXT: shlxl %edx, %esi, %edx
|
|
; AVX512VBMI-NEXT: kmovd %edx, %k1
|
|
; AVX512VBMI-NEXT: vpcompressq %zmm1, %zmm1 {%k1} {z}
|
|
; AVX512VBMI-NEXT: vpxor %xmm3, %xmm3, %xmm3
|
|
; AVX512VBMI-NEXT: valignq {{.*#+}} zmm4 = zmm1[1,2,3,4,5,6,7],zmm3[0]
|
|
; AVX512VBMI-NEXT: vpshrdvq %zmm2, %zmm4, %zmm1
|
|
; AVX512VBMI-NEXT: movl %ecx, %edx
|
|
; AVX512VBMI-NEXT: shrl $6, %edx
|
|
; AVX512VBMI-NEXT: shlxl %edx, %esi, %edx
|
|
; AVX512VBMI-NEXT: kmovd %edx, %k1
|
|
; AVX512VBMI-NEXT: vpexpandq %zmm0, %zmm2 {%k1} {z}
|
|
; AVX512VBMI-NEXT: valignq {{.*#+}} zmm3 = zmm3[7],zmm2[0,1,2,3,4,5,6]
|
|
; AVX512VBMI-NEXT: vpbroadcastq %rcx, %zmm4
|
|
; AVX512VBMI-NEXT: vpshldvq %zmm4, %zmm3, %zmm2
|
|
; AVX512VBMI-NEXT: xorl %edx, %edx
|
|
; AVX512VBMI-NEXT: negq %rcx
|
|
; AVX512VBMI-NEXT: sbbl %edx, %edx
|
|
; AVX512VBMI-NEXT: kmovd %edx, %k1
|
|
; AVX512VBMI-NEXT: vporq %zmm1, %zmm2, %zmm0 {%k1}
|
|
; AVX512VBMI-NEXT: vmovdqu64 %zmm0, (%rdi)
|
|
; AVX512VBMI-NEXT: vzeroupper
|
|
; AVX512VBMI-NEXT: retq
|
|
%a0 = load i512, ptr %p0
|
|
%a1 = load i512, ptr %p1
|
|
%r = call i512 @llvm.fshl.i512(i512 %a0, i512 %a1, i512 %a2)
|
|
ret i512 %r
|
|
}
|
|
|
|
define i512 @fshr_i512_load(ptr %p0, ptr %p1, i512 %a2) nounwind {
|
|
; SSE-LABEL: fshr_i512_load:
|
|
; SSE: # %bb.0:
|
|
; SSE-NEXT: pushq %rbp
|
|
; SSE-NEXT: pushq %r15
|
|
; SSE-NEXT: pushq %r14
|
|
; SSE-NEXT: pushq %r13
|
|
; SSE-NEXT: pushq %r12
|
|
; SSE-NEXT: pushq %rbx
|
|
; SSE-NEXT: movq (%rsi), %r8
|
|
; SSE-NEXT: movq 8(%rsi), %r9
|
|
; SSE-NEXT: movq 16(%rsi), %r10
|
|
; SSE-NEXT: movq 24(%rsi), %r11
|
|
; SSE-NEXT: movq 32(%rdx), %r12
|
|
; SSE-NEXT: testl $256, %ecx # imm = 0x100
|
|
; SSE-NEXT: movq 56(%rsi), %r13
|
|
; SSE-NEXT: cmoveq %r11, %r13
|
|
; SSE-NEXT: movq 40(%rdx), %rbx
|
|
; SSE-NEXT: movq 48(%rsi), %rbp
|
|
; SSE-NEXT: cmoveq %r10, %rbp
|
|
; SSE-NEXT: movq 48(%rdx), %r14
|
|
; SSE-NEXT: movq 40(%rsi), %rax
|
|
; SSE-NEXT: cmoveq %r9, %rax
|
|
; SSE-NEXT: movq 56(%rdx), %r15
|
|
; SSE-NEXT: movq 32(%rsi), %rsi
|
|
; SSE-NEXT: cmoveq %r8, %rsi
|
|
; SSE-NEXT: cmoveq %r15, %r11
|
|
; SSE-NEXT: cmoveq %r14, %r10
|
|
; SSE-NEXT: cmoveq %rbx, %r9
|
|
; SSE-NEXT: cmoveq %r12, %r8
|
|
; SSE-NEXT: cmoveq 24(%rdx), %r15
|
|
; SSE-NEXT: cmoveq 8(%rdx), %rbx
|
|
; SSE-NEXT: cmoveq (%rdx), %r12
|
|
; SSE-NEXT: cmoveq 16(%rdx), %r14
|
|
; SSE-NEXT: testb $-128, %cl
|
|
; SSE-NEXT: cmovneq %r14, %r12
|
|
; SSE-NEXT: cmovneq %r15, %rbx
|
|
; SSE-NEXT: cmovneq %r8, %r14
|
|
; SSE-NEXT: cmovneq %r9, %r15
|
|
; SSE-NEXT: cmovneq %r10, %r8
|
|
; SSE-NEXT: cmovneq %r11, %r9
|
|
; SSE-NEXT: cmovneq %rsi, %r10
|
|
; SSE-NEXT: cmovneq %rax, %r11
|
|
; SSE-NEXT: cmovneq %rbp, %rsi
|
|
; SSE-NEXT: cmovneq %r13, %rax
|
|
; SSE-NEXT: testb $64, %cl
|
|
; SSE-NEXT: cmoveq %rsi, %rax
|
|
; SSE-NEXT: cmoveq %r11, %rsi
|
|
; SSE-NEXT: cmoveq %r10, %r11
|
|
; SSE-NEXT: cmoveq %r9, %r10
|
|
; SSE-NEXT: cmoveq %r8, %r9
|
|
; SSE-NEXT: cmoveq %r15, %r8
|
|
; SSE-NEXT: cmoveq %r14, %r15
|
|
; SSE-NEXT: cmoveq %rbx, %r14
|
|
; SSE-NEXT: cmoveq %r12, %rbx
|
|
; SSE-NEXT: shrdq %cl, %r14, %rbx
|
|
; SSE-NEXT: shrdq %cl, %r15, %r14
|
|
; SSE-NEXT: shrdq %cl, %r8, %r15
|
|
; SSE-NEXT: shrdq %cl, %r9, %r8
|
|
; SSE-NEXT: shrdq %cl, %r10, %r9
|
|
; SSE-NEXT: shrdq %cl, %r11, %r10
|
|
; SSE-NEXT: shrdq %cl, %rsi, %r11
|
|
; SSE-NEXT: # kill: def $cl killed $cl killed $rcx
|
|
; SSE-NEXT: shrdq %cl, %rax, %rsi
|
|
; SSE-NEXT: movq %rdi, %rax
|
|
; SSE-NEXT: movq %rsi, 56(%rdi)
|
|
; SSE-NEXT: movq %r11, 48(%rdi)
|
|
; SSE-NEXT: movq %r10, 40(%rdi)
|
|
; SSE-NEXT: movq %r9, 32(%rdi)
|
|
; SSE-NEXT: movq %r8, 24(%rdi)
|
|
; SSE-NEXT: movq %r15, 16(%rdi)
|
|
; SSE-NEXT: movq %r14, 8(%rdi)
|
|
; SSE-NEXT: movq %rbx, (%rdi)
|
|
; SSE-NEXT: popq %rbx
|
|
; SSE-NEXT: popq %r12
|
|
; SSE-NEXT: popq %r13
|
|
; SSE-NEXT: popq %r14
|
|
; SSE-NEXT: popq %r15
|
|
; SSE-NEXT: popq %rbp
|
|
; SSE-NEXT: retq
|
|
;
|
|
; AVX2-LABEL: fshr_i512_load:
|
|
; AVX2: # %bb.0:
|
|
; AVX2-NEXT: pushq %rbp
|
|
; AVX2-NEXT: pushq %r15
|
|
; AVX2-NEXT: pushq %r14
|
|
; AVX2-NEXT: pushq %r13
|
|
; AVX2-NEXT: pushq %r12
|
|
; AVX2-NEXT: pushq %rbx
|
|
; AVX2-NEXT: movq (%rsi), %r8
|
|
; AVX2-NEXT: movq 8(%rsi), %r9
|
|
; AVX2-NEXT: movq 16(%rsi), %r10
|
|
; AVX2-NEXT: movq 24(%rsi), %r11
|
|
; AVX2-NEXT: movq 32(%rdx), %r13
|
|
; AVX2-NEXT: testl $256, %ecx # imm = 0x100
|
|
; AVX2-NEXT: movq 56(%rsi), %rax
|
|
; AVX2-NEXT: cmoveq %r11, %rax
|
|
; AVX2-NEXT: movq 40(%rdx), %rbx
|
|
; AVX2-NEXT: movq 48(%rsi), %rbp
|
|
; AVX2-NEXT: cmoveq %r10, %rbp
|
|
; AVX2-NEXT: movq 48(%rdx), %r14
|
|
; AVX2-NEXT: movq 40(%rsi), %r12
|
|
; AVX2-NEXT: cmoveq %r9, %r12
|
|
; AVX2-NEXT: movq 56(%rdx), %r15
|
|
; AVX2-NEXT: movq 32(%rsi), %rsi
|
|
; AVX2-NEXT: cmoveq %r8, %rsi
|
|
; AVX2-NEXT: cmoveq %r15, %r11
|
|
; AVX2-NEXT: cmoveq %r14, %r10
|
|
; AVX2-NEXT: cmoveq %rbx, %r9
|
|
; AVX2-NEXT: cmoveq %r13, %r8
|
|
; AVX2-NEXT: cmoveq 24(%rdx), %r15
|
|
; AVX2-NEXT: cmoveq 8(%rdx), %rbx
|
|
; AVX2-NEXT: cmoveq (%rdx), %r13
|
|
; AVX2-NEXT: cmoveq 16(%rdx), %r14
|
|
; AVX2-NEXT: testb $-128, %cl
|
|
; AVX2-NEXT: cmovneq %r14, %r13
|
|
; AVX2-NEXT: cmovneq %r15, %rbx
|
|
; AVX2-NEXT: cmovneq %r8, %r14
|
|
; AVX2-NEXT: cmovneq %r9, %r15
|
|
; AVX2-NEXT: cmovneq %r10, %r8
|
|
; AVX2-NEXT: cmovneq %r11, %r9
|
|
; AVX2-NEXT: cmovneq %rsi, %r10
|
|
; AVX2-NEXT: cmovneq %r12, %r11
|
|
; AVX2-NEXT: cmovneq %rbp, %rsi
|
|
; AVX2-NEXT: cmovneq %rax, %r12
|
|
; AVX2-NEXT: movq %rdi, %rax
|
|
; AVX2-NEXT: testb $64, %cl
|
|
; AVX2-NEXT: cmoveq %rsi, %r12
|
|
; AVX2-NEXT: cmoveq %r11, %rsi
|
|
; AVX2-NEXT: cmoveq %r10, %r11
|
|
; AVX2-NEXT: cmoveq %r9, %r10
|
|
; AVX2-NEXT: cmoveq %r8, %r9
|
|
; AVX2-NEXT: cmoveq %r15, %r8
|
|
; AVX2-NEXT: cmoveq %r14, %r15
|
|
; AVX2-NEXT: cmoveq %rbx, %r14
|
|
; AVX2-NEXT: cmoveq %r13, %rbx
|
|
; AVX2-NEXT: shrdq %cl, %r14, %rbx
|
|
; AVX2-NEXT: shrdq %cl, %r15, %r14
|
|
; AVX2-NEXT: shrdq %cl, %r8, %r15
|
|
; AVX2-NEXT: shrdq %cl, %r9, %r8
|
|
; AVX2-NEXT: shrdq %cl, %r10, %r9
|
|
; AVX2-NEXT: shrdq %cl, %r11, %r10
|
|
; AVX2-NEXT: shrdq %cl, %rsi, %r11
|
|
; AVX2-NEXT: # kill: def $cl killed $cl killed $rcx
|
|
; AVX2-NEXT: shrdq %cl, %r12, %rsi
|
|
; AVX2-NEXT: movq %rsi, 56(%rdi)
|
|
; AVX2-NEXT: movq %r11, 48(%rdi)
|
|
; AVX2-NEXT: movq %r10, 40(%rdi)
|
|
; AVX2-NEXT: movq %r9, 32(%rdi)
|
|
; AVX2-NEXT: movq %r8, 24(%rdi)
|
|
; AVX2-NEXT: movq %r15, 16(%rdi)
|
|
; AVX2-NEXT: movq %r14, 8(%rdi)
|
|
; AVX2-NEXT: movq %rbx, (%rdi)
|
|
; AVX2-NEXT: popq %rbx
|
|
; AVX2-NEXT: popq %r12
|
|
; AVX2-NEXT: popq %r13
|
|
; AVX2-NEXT: popq %r14
|
|
; AVX2-NEXT: popq %r15
|
|
; AVX2-NEXT: popq %rbp
|
|
; AVX2-NEXT: retq
|
|
;
|
|
; AVX512F-LABEL: fshr_i512_load:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: vmovdqu64 (%rdx), %zmm1
|
|
; AVX512F-NEXT: andl $511, %ecx # imm = 0x1FF
|
|
; AVX512F-NEXT: vmovq %rcx, %xmm0
|
|
; AVX512F-NEXT: vpbroadcastq %xmm0, %xmm2
|
|
; AVX512F-NEXT: vpbroadcastq {{.*#+}} xmm0 = [63,63]
|
|
; AVX512F-NEXT: vpand %xmm0, %xmm2, %xmm3
|
|
; AVX512F-NEXT: movl %ecx, %eax
|
|
; AVX512F-NEXT: shrl $6, %eax
|
|
; AVX512F-NEXT: movl $-1, %edx
|
|
; AVX512F-NEXT: shlxl %eax, %edx, %eax
|
|
; AVX512F-NEXT: kmovw %eax, %k1
|
|
; AVX512F-NEXT: vpcompressq %zmm1, %zmm4 {%k1} {z}
|
|
; AVX512F-NEXT: vpsrlq %xmm3, %zmm4, %zmm3
|
|
; AVX512F-NEXT: movq %rdi, %rax
|
|
; AVX512F-NEXT: vpandn %xmm0, %xmm2, %xmm2
|
|
; AVX512F-NEXT: vpxor %xmm5, %xmm5, %xmm5
|
|
; AVX512F-NEXT: valignq {{.*#+}} zmm4 = zmm4[1,2,3,4,5,6,7],zmm5[0]
|
|
; AVX512F-NEXT: vpaddq %zmm4, %zmm4, %zmm4
|
|
; AVX512F-NEXT: vpsllq %xmm2, %zmm4, %zmm2
|
|
; AVX512F-NEXT: vporq %zmm3, %zmm2, %zmm2
|
|
; AVX512F-NEXT: movl $512, %edi # imm = 0x200
|
|
; AVX512F-NEXT: subq %rcx, %rdi
|
|
; AVX512F-NEXT: vmovq %rdi, %xmm3
|
|
; AVX512F-NEXT: vpbroadcastq %xmm3, %xmm3
|
|
; AVX512F-NEXT: shrl $6, %edi
|
|
; AVX512F-NEXT: shlxl %edi, %edx, %edx
|
|
; AVX512F-NEXT: kmovw %edx, %k1
|
|
; AVX512F-NEXT: vpexpandq (%rsi), %zmm4 {%k1} {z}
|
|
; AVX512F-NEXT: vpand %xmm0, %xmm3, %xmm6
|
|
; AVX512F-NEXT: vpsllq %xmm6, %zmm4, %zmm6
|
|
; AVX512F-NEXT: vpandn %xmm0, %xmm3, %xmm0
|
|
; AVX512F-NEXT: valignq {{.*#+}} zmm3 = zmm5[7],zmm4[0,1,2,3,4,5,6]
|
|
; AVX512F-NEXT: vpsrlq $1, %zmm3, %zmm3
|
|
; AVX512F-NEXT: vpsrlq %xmm0, %zmm3, %zmm0
|
|
; AVX512F-NEXT: xorl %edx, %edx
|
|
; AVX512F-NEXT: negq %rcx
|
|
; AVX512F-NEXT: sbbl %edx, %edx
|
|
; AVX512F-NEXT: vporq %zmm0, %zmm6, %zmm0
|
|
; AVX512F-NEXT: kmovw %edx, %k1
|
|
; AVX512F-NEXT: vporq %zmm2, %zmm0, %zmm1 {%k1}
|
|
; AVX512F-NEXT: vmovdqu64 %zmm1, (%rax)
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: fshr_i512_load:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: movq %rdi, %rax
|
|
; AVX512VL-NEXT: vmovdqu64 (%rdx), %zmm0
|
|
; AVX512VL-NEXT: andl $511, %ecx # imm = 0x1FF
|
|
; AVX512VL-NEXT: vpbroadcastq %rcx, %xmm1
|
|
; AVX512VL-NEXT: vpbroadcastq {{.*#+}} xmm2 = [63,63]
|
|
; AVX512VL-NEXT: vpand %xmm2, %xmm1, %xmm3
|
|
; AVX512VL-NEXT: movl %ecx, %edx
|
|
; AVX512VL-NEXT: shrl $6, %edx
|
|
; AVX512VL-NEXT: movl $-1, %edi
|
|
; AVX512VL-NEXT: shlxl %edx, %edi, %edx
|
|
; AVX512VL-NEXT: kmovd %edx, %k1
|
|
; AVX512VL-NEXT: vpcompressq %zmm0, %zmm4 {%k1} {z}
|
|
; AVX512VL-NEXT: vpsrlq %xmm3, %zmm4, %zmm3
|
|
; AVX512VL-NEXT: vpandn %xmm2, %xmm1, %xmm1
|
|
; AVX512VL-NEXT: vpxor %xmm5, %xmm5, %xmm5
|
|
; AVX512VL-NEXT: valignq {{.*#+}} zmm4 = zmm4[1,2,3,4,5,6,7],zmm5[0]
|
|
; AVX512VL-NEXT: vpaddq %zmm4, %zmm4, %zmm4
|
|
; AVX512VL-NEXT: vpsllq %xmm1, %zmm4, %zmm1
|
|
; AVX512VL-NEXT: vporq %zmm3, %zmm1, %zmm1
|
|
; AVX512VL-NEXT: movl $512, %edx # imm = 0x200
|
|
; AVX512VL-NEXT: subq %rcx, %rdx
|
|
; AVX512VL-NEXT: vpbroadcastq %rdx, %xmm3
|
|
; AVX512VL-NEXT: vpand %xmm2, %xmm3, %xmm4
|
|
; AVX512VL-NEXT: shrl $6, %edx
|
|
; AVX512VL-NEXT: shlxl %edx, %edi, %edx
|
|
; AVX512VL-NEXT: kmovd %edx, %k1
|
|
; AVX512VL-NEXT: vpexpandq (%rsi), %zmm6 {%k1} {z}
|
|
; AVX512VL-NEXT: vpsllq %xmm4, %zmm6, %zmm4
|
|
; AVX512VL-NEXT: vpandn %xmm2, %xmm3, %xmm2
|
|
; AVX512VL-NEXT: valignq {{.*#+}} zmm3 = zmm5[7],zmm6[0,1,2,3,4,5,6]
|
|
; AVX512VL-NEXT: vpsrlq $1, %zmm3, %zmm3
|
|
; AVX512VL-NEXT: vpsrlq %xmm2, %zmm3, %zmm2
|
|
; AVX512VL-NEXT: vporq %zmm2, %zmm4, %zmm2
|
|
; AVX512VL-NEXT: xorl %edx, %edx
|
|
; AVX512VL-NEXT: negq %rcx
|
|
; AVX512VL-NEXT: sbbl %edx, %edx
|
|
; AVX512VL-NEXT: kmovd %edx, %k1
|
|
; AVX512VL-NEXT: vporq %zmm1, %zmm2, %zmm0 {%k1}
|
|
; AVX512VL-NEXT: vmovdqu64 %zmm0, (%rax)
|
|
; AVX512VL-NEXT: vzeroupper
|
|
; AVX512VL-NEXT: retq
|
|
;
|
|
; AVX512VBMI-LABEL: fshr_i512_load:
|
|
; AVX512VBMI: # %bb.0:
|
|
; AVX512VBMI-NEXT: movq %rdi, %rax
|
|
; AVX512VBMI-NEXT: andl $511, %ecx # imm = 0x1FF
|
|
; AVX512VBMI-NEXT: movl $512, %edi # imm = 0x200
|
|
; AVX512VBMI-NEXT: subq %rcx, %rdi
|
|
; AVX512VBMI-NEXT: vpbroadcastq %rdi, %zmm0
|
|
; AVX512VBMI-NEXT: # kill: def $edi killed $edi killed $rdi
|
|
; AVX512VBMI-NEXT: shrl $6, %edi
|
|
; AVX512VBMI-NEXT: movl $-1, %r8d
|
|
; AVX512VBMI-NEXT: shlxl %edi, %r8d, %edi
|
|
; AVX512VBMI-NEXT: kmovd %edi, %k1
|
|
; AVX512VBMI-NEXT: vpexpandq (%rsi), %zmm1 {%k1} {z}
|
|
; AVX512VBMI-NEXT: vmovdqu64 (%rdx), %zmm2
|
|
; AVX512VBMI-NEXT: vpxor %xmm3, %xmm3, %xmm3
|
|
; AVX512VBMI-NEXT: valignq {{.*#+}} zmm4 = zmm3[7],zmm1[0,1,2,3,4,5,6]
|
|
; AVX512VBMI-NEXT: vpshldvq %zmm0, %zmm4, %zmm1
|
|
; AVX512VBMI-NEXT: movl %ecx, %edx
|
|
; AVX512VBMI-NEXT: shrl $6, %edx
|
|
; AVX512VBMI-NEXT: shlxl %edx, %r8d, %edx
|
|
; AVX512VBMI-NEXT: kmovd %edx, %k1
|
|
; AVX512VBMI-NEXT: vpcompressq %zmm2, %zmm0 {%k1} {z}
|
|
; AVX512VBMI-NEXT: valignq {{.*#+}} zmm3 = zmm0[1,2,3,4,5,6,7],zmm3[0]
|
|
; AVX512VBMI-NEXT: vpbroadcastq %rcx, %zmm4
|
|
; AVX512VBMI-NEXT: vpshrdvq %zmm4, %zmm3, %zmm0
|
|
; AVX512VBMI-NEXT: xorl %edx, %edx
|
|
; AVX512VBMI-NEXT: negq %rcx
|
|
; AVX512VBMI-NEXT: sbbl %edx, %edx
|
|
; AVX512VBMI-NEXT: kmovd %edx, %k1
|
|
; AVX512VBMI-NEXT: vporq %zmm0, %zmm1, %zmm2 {%k1}
|
|
; AVX512VBMI-NEXT: vmovdqu64 %zmm2, (%rax)
|
|
; AVX512VBMI-NEXT: vzeroupper
|
|
; AVX512VBMI-NEXT: retq
|
|
%a0 = load i512, ptr %p0
|
|
%a1 = load i512, ptr %p1
|
|
%r = call i512 @llvm.fshr.i512(i512 %a0, i512 %a1, i512 %a2)
|
|
ret i512 %r
|
|
}
|
|
|
|
define i512 @fshl_rot_i512_load(ptr %p0, i512 %a2) nounwind {
|
|
; SSE-LABEL: fshl_rot_i512_load:
|
|
; SSE: # %bb.0:
|
|
; SSE-NEXT: pushq %r15
|
|
; SSE-NEXT: pushq %r14
|
|
; SSE-NEXT: pushq %r13
|
|
; SSE-NEXT: pushq %r12
|
|
; SSE-NEXT: pushq %rbx
|
|
; SSE-NEXT: movq %rdx, %rcx
|
|
; SSE-NEXT: movq 40(%rsi), %rax
|
|
; SSE-NEXT: movq 48(%rsi), %rdx
|
|
; SSE-NEXT: movq 56(%rsi), %r10
|
|
; SSE-NEXT: movq (%rsi), %r9
|
|
; SSE-NEXT: movq 32(%rsi), %rbx
|
|
; SSE-NEXT: testl $256, %ecx # imm = 0x100
|
|
; SSE-NEXT: movq %rbx, %r11
|
|
; SSE-NEXT: cmovneq %r9, %r11
|
|
; SSE-NEXT: movq 24(%rsi), %r14
|
|
; SSE-NEXT: movq %r14, %r15
|
|
; SSE-NEXT: cmovneq %r10, %r15
|
|
; SSE-NEXT: movq 16(%rsi), %r12
|
|
; SSE-NEXT: movq %r12, %r8
|
|
; SSE-NEXT: cmovneq %rdx, %r8
|
|
; SSE-NEXT: movq 8(%rsi), %r13
|
|
; SSE-NEXT: movq %r13, %rsi
|
|
; SSE-NEXT: cmovneq %rax, %rsi
|
|
; SSE-NEXT: cmovneq %rbx, %r9
|
|
; SSE-NEXT: cmovneq %r12, %rdx
|
|
; SSE-NEXT: cmovneq %r13, %rax
|
|
; SSE-NEXT: cmovneq %r14, %r10
|
|
; SSE-NEXT: testb $-128, %cl
|
|
; SSE-NEXT: movq %rsi, %r14
|
|
; SSE-NEXT: cmovneq %r10, %r14
|
|
; SSE-NEXT: cmovneq %rax, %r10
|
|
; SSE-NEXT: movq %r8, %rbx
|
|
; SSE-NEXT: cmovneq %r9, %rbx
|
|
; SSE-NEXT: cmovneq %rdx, %r9
|
|
; SSE-NEXT: cmoveq %r15, %rsi
|
|
; SSE-NEXT: cmoveq %r11, %r8
|
|
; SSE-NEXT: cmovneq %r15, %rax
|
|
; SSE-NEXT: cmovneq %r11, %rdx
|
|
; SSE-NEXT: testb $64, %cl
|
|
; SSE-NEXT: movq %rdx, %r11
|
|
; SSE-NEXT: cmovneq %rax, %r11
|
|
; SSE-NEXT: cmovneq %r8, %rax
|
|
; SSE-NEXT: cmovneq %rsi, %r8
|
|
; SSE-NEXT: cmovneq %rbx, %rsi
|
|
; SSE-NEXT: cmovneq %r14, %rbx
|
|
; SSE-NEXT: cmovneq %r9, %r14
|
|
; SSE-NEXT: cmoveq %r10, %rdx
|
|
; SSE-NEXT: cmovneq %r10, %r9
|
|
; SSE-NEXT: movq %r9, %r10
|
|
; SSE-NEXT: shldq %cl, %rdx, %r10
|
|
; SSE-NEXT: movq %r14, %r15
|
|
; SSE-NEXT: shldq %cl, %r9, %r15
|
|
; SSE-NEXT: movq %rbx, %r9
|
|
; SSE-NEXT: shldq %cl, %r14, %r9
|
|
; SSE-NEXT: movq %rsi, %r14
|
|
; SSE-NEXT: shldq %cl, %rbx, %r14
|
|
; SSE-NEXT: movq %r8, %rbx
|
|
; SSE-NEXT: shldq %cl, %rsi, %rbx
|
|
; SSE-NEXT: movq %rax, %rsi
|
|
; SSE-NEXT: shldq %cl, %r8, %rsi
|
|
; SSE-NEXT: movq %r11, %r8
|
|
; SSE-NEXT: shldq %cl, %rax, %r8
|
|
; SSE-NEXT: # kill: def $cl killed $cl killed $rcx
|
|
; SSE-NEXT: shldq %cl, %r11, %rdx
|
|
; SSE-NEXT: movq %rdi, %rax
|
|
; SSE-NEXT: movq %rdx, 56(%rdi)
|
|
; SSE-NEXT: movq %r8, 48(%rdi)
|
|
; SSE-NEXT: movq %rsi, 40(%rdi)
|
|
; SSE-NEXT: movq %rbx, 32(%rdi)
|
|
; SSE-NEXT: movq %r14, 24(%rdi)
|
|
; SSE-NEXT: movq %r9, 16(%rdi)
|
|
; SSE-NEXT: movq %r15, 8(%rdi)
|
|
; SSE-NEXT: movq %r10, (%rdi)
|
|
; SSE-NEXT: popq %rbx
|
|
; SSE-NEXT: popq %r12
|
|
; SSE-NEXT: popq %r13
|
|
; SSE-NEXT: popq %r14
|
|
; SSE-NEXT: popq %r15
|
|
; SSE-NEXT: retq
|
|
;
|
|
; AVX2-LABEL: fshl_rot_i512_load:
|
|
; AVX2: # %bb.0:
|
|
; AVX2-NEXT: pushq %r15
|
|
; AVX2-NEXT: pushq %r14
|
|
; AVX2-NEXT: pushq %r13
|
|
; AVX2-NEXT: pushq %r12
|
|
; AVX2-NEXT: pushq %rbx
|
|
; AVX2-NEXT: movq %rdx, %rcx
|
|
; AVX2-NEXT: movq 40(%rsi), %r8
|
|
; AVX2-NEXT: movq 48(%rsi), %rdx
|
|
; AVX2-NEXT: movq 16(%rsi), %rax
|
|
; AVX2-NEXT: movq 56(%rsi), %r11
|
|
; AVX2-NEXT: movq 24(%rsi), %rbx
|
|
; AVX2-NEXT: movq (%rsi), %r10
|
|
; AVX2-NEXT: movq 8(%rsi), %r14
|
|
; AVX2-NEXT: movq 32(%rsi), %r12
|
|
; AVX2-NEXT: testl $256, %ecx # imm = 0x100
|
|
; AVX2-NEXT: movq %r12, %r15
|
|
; AVX2-NEXT: cmovneq %r10, %r15
|
|
; AVX2-NEXT: movq %rbx, %r13
|
|
; AVX2-NEXT: cmovneq %r11, %r13
|
|
; AVX2-NEXT: movq %rax, %rsi
|
|
; AVX2-NEXT: cmovneq %rdx, %rsi
|
|
; AVX2-NEXT: movq %r14, %r9
|
|
; AVX2-NEXT: cmovneq %r8, %r9
|
|
; AVX2-NEXT: cmovneq %r12, %r10
|
|
; AVX2-NEXT: cmovneq %rax, %rdx
|
|
; AVX2-NEXT: cmovneq %r14, %r8
|
|
; AVX2-NEXT: cmovneq %rbx, %r11
|
|
; AVX2-NEXT: testb $-128, %cl
|
|
; AVX2-NEXT: movq %r9, %r14
|
|
; AVX2-NEXT: cmovneq %r11, %r14
|
|
; AVX2-NEXT: cmovneq %r8, %r11
|
|
; AVX2-NEXT: movq %rsi, %rbx
|
|
; AVX2-NEXT: cmovneq %r10, %rbx
|
|
; AVX2-NEXT: cmovneq %rdx, %r10
|
|
; AVX2-NEXT: cmoveq %r13, %r9
|
|
; AVX2-NEXT: cmoveq %r15, %rsi
|
|
; AVX2-NEXT: cmovneq %r13, %r8
|
|
; AVX2-NEXT: movq %rdi, %rax
|
|
; AVX2-NEXT: cmovneq %r15, %rdx
|
|
; AVX2-NEXT: testb $64, %cl
|
|
; AVX2-NEXT: movq %rdx, %rdi
|
|
; AVX2-NEXT: cmovneq %r8, %rdi
|
|
; AVX2-NEXT: cmovneq %rsi, %r8
|
|
; AVX2-NEXT: cmovneq %r9, %rsi
|
|
; AVX2-NEXT: cmovneq %rbx, %r9
|
|
; AVX2-NEXT: cmovneq %r14, %rbx
|
|
; AVX2-NEXT: cmovneq %r10, %r14
|
|
; AVX2-NEXT: cmoveq %r11, %rdx
|
|
; AVX2-NEXT: cmovneq %r11, %r10
|
|
; AVX2-NEXT: movq %r10, %r11
|
|
; AVX2-NEXT: shldq %cl, %rdx, %r11
|
|
; AVX2-NEXT: movq %r14, %r15
|
|
; AVX2-NEXT: shldq %cl, %r10, %r15
|
|
; AVX2-NEXT: movq %rbx, %r10
|
|
; AVX2-NEXT: shldq %cl, %r14, %r10
|
|
; AVX2-NEXT: movq %r9, %r14
|
|
; AVX2-NEXT: shldq %cl, %rbx, %r14
|
|
; AVX2-NEXT: movq %rsi, %rbx
|
|
; AVX2-NEXT: shldq %cl, %r9, %rbx
|
|
; AVX2-NEXT: movq %r8, %r9
|
|
; AVX2-NEXT: shldq %cl, %rsi, %r9
|
|
; AVX2-NEXT: movq %rdi, %rsi
|
|
; AVX2-NEXT: shldq %cl, %r8, %rsi
|
|
; AVX2-NEXT: # kill: def $cl killed $cl killed $rcx
|
|
; AVX2-NEXT: shldq %cl, %rdi, %rdx
|
|
; AVX2-NEXT: movq %rdx, 56(%rax)
|
|
; AVX2-NEXT: movq %rsi, 48(%rax)
|
|
; AVX2-NEXT: movq %r9, 40(%rax)
|
|
; AVX2-NEXT: movq %rbx, 32(%rax)
|
|
; AVX2-NEXT: movq %r14, 24(%rax)
|
|
; AVX2-NEXT: movq %r10, 16(%rax)
|
|
; AVX2-NEXT: movq %r15, 8(%rax)
|
|
; AVX2-NEXT: movq %r11, (%rax)
|
|
; AVX2-NEXT: popq %rbx
|
|
; AVX2-NEXT: popq %r12
|
|
; AVX2-NEXT: popq %r13
|
|
; AVX2-NEXT: popq %r14
|
|
; AVX2-NEXT: popq %r15
|
|
; AVX2-NEXT: retq
|
|
;
|
|
; AVX512F-LABEL: fshl_rot_i512_load:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: movq %rdi, %rax
|
|
; AVX512F-NEXT: vmovdqu64 (%rsi), %zmm1
|
|
; AVX512F-NEXT: andl $511, %edx # imm = 0x1FF
|
|
; AVX512F-NEXT: vmovq %rdx, %xmm0
|
|
; AVX512F-NEXT: vpbroadcastq %xmm0, %xmm2
|
|
; AVX512F-NEXT: vpbroadcastq {{.*#+}} xmm0 = [63,63]
|
|
; AVX512F-NEXT: vpand %xmm0, %xmm2, %xmm3
|
|
; AVX512F-NEXT: movl %edx, %ecx
|
|
; AVX512F-NEXT: shrl $6, %ecx
|
|
; AVX512F-NEXT: movl $-1, %esi
|
|
; AVX512F-NEXT: shlxl %ecx, %esi, %ecx
|
|
; AVX512F-NEXT: kmovw %ecx, %k1
|
|
; AVX512F-NEXT: vpexpandq %zmm1, %zmm4 {%k1} {z}
|
|
; AVX512F-NEXT: vpsllq %xmm3, %zmm4, %zmm3
|
|
; AVX512F-NEXT: vpandn %xmm0, %xmm2, %xmm2
|
|
; AVX512F-NEXT: vpxor %xmm5, %xmm5, %xmm5
|
|
; AVX512F-NEXT: valignq {{.*#+}} zmm4 = zmm5[7],zmm4[0,1,2,3,4,5,6]
|
|
; AVX512F-NEXT: vpsrlq $1, %zmm4, %zmm4
|
|
; AVX512F-NEXT: vpsrlq %xmm2, %zmm4, %zmm2
|
|
; AVX512F-NEXT: vporq %zmm2, %zmm3, %zmm2
|
|
; AVX512F-NEXT: movl $512, %ecx # imm = 0x200
|
|
; AVX512F-NEXT: subq %rdx, %rcx
|
|
; AVX512F-NEXT: vmovq %rcx, %xmm3
|
|
; AVX512F-NEXT: vpbroadcastq %xmm3, %xmm3
|
|
; AVX512F-NEXT: vpand %xmm0, %xmm3, %xmm4
|
|
; AVX512F-NEXT: shrl $6, %ecx
|
|
; AVX512F-NEXT: shlxl %ecx, %esi, %ecx
|
|
; AVX512F-NEXT: kmovw %ecx, %k1
|
|
; AVX512F-NEXT: vpcompressq %zmm1, %zmm6 {%k1} {z}
|
|
; AVX512F-NEXT: vpsrlq %xmm4, %zmm6, %zmm4
|
|
; AVX512F-NEXT: vpandn %xmm0, %xmm3, %xmm0
|
|
; AVX512F-NEXT: valignq {{.*#+}} zmm3 = zmm6[1,2,3,4,5,6,7],zmm5[0]
|
|
; AVX512F-NEXT: vpaddq %zmm3, %zmm3, %zmm3
|
|
; AVX512F-NEXT: vpsllq %xmm0, %zmm3, %zmm0
|
|
; AVX512F-NEXT: xorl %ecx, %ecx
|
|
; AVX512F-NEXT: negq %rdx
|
|
; AVX512F-NEXT: sbbl %ecx, %ecx
|
|
; AVX512F-NEXT: vporq %zmm4, %zmm0, %zmm0
|
|
; AVX512F-NEXT: kmovw %ecx, %k1
|
|
; AVX512F-NEXT: vporq %zmm0, %zmm2, %zmm1 {%k1}
|
|
; AVX512F-NEXT: vmovdqu64 %zmm1, (%rdi)
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: fshl_rot_i512_load:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: movq %rdi, %rax
|
|
; AVX512VL-NEXT: vmovdqu64 (%rsi), %zmm0
|
|
; AVX512VL-NEXT: andl $511, %edx # imm = 0x1FF
|
|
; AVX512VL-NEXT: vpbroadcastq %rdx, %xmm1
|
|
; AVX512VL-NEXT: vpbroadcastq {{.*#+}} xmm2 = [63,63]
|
|
; AVX512VL-NEXT: vpand %xmm2, %xmm1, %xmm3
|
|
; AVX512VL-NEXT: movl %edx, %ecx
|
|
; AVX512VL-NEXT: shrl $6, %ecx
|
|
; AVX512VL-NEXT: movl $-1, %esi
|
|
; AVX512VL-NEXT: shlxl %ecx, %esi, %ecx
|
|
; AVX512VL-NEXT: kmovd %ecx, %k1
|
|
; AVX512VL-NEXT: vpexpandq %zmm0, %zmm4 {%k1} {z}
|
|
; AVX512VL-NEXT: vpsllq %xmm3, %zmm4, %zmm3
|
|
; AVX512VL-NEXT: vpandn %xmm2, %xmm1, %xmm1
|
|
; AVX512VL-NEXT: vpxor %xmm5, %xmm5, %xmm5
|
|
; AVX512VL-NEXT: valignq {{.*#+}} zmm4 = zmm5[7],zmm4[0,1,2,3,4,5,6]
|
|
; AVX512VL-NEXT: vpsrlq $1, %zmm4, %zmm4
|
|
; AVX512VL-NEXT: vpsrlq %xmm1, %zmm4, %zmm1
|
|
; AVX512VL-NEXT: vporq %zmm1, %zmm3, %zmm1
|
|
; AVX512VL-NEXT: movl $512, %ecx # imm = 0x200
|
|
; AVX512VL-NEXT: subq %rdx, %rcx
|
|
; AVX512VL-NEXT: vpbroadcastq %rcx, %xmm3
|
|
; AVX512VL-NEXT: vpand %xmm2, %xmm3, %xmm4
|
|
; AVX512VL-NEXT: shrl $6, %ecx
|
|
; AVX512VL-NEXT: shlxl %ecx, %esi, %ecx
|
|
; AVX512VL-NEXT: kmovd %ecx, %k1
|
|
; AVX512VL-NEXT: vpcompressq %zmm0, %zmm6 {%k1} {z}
|
|
; AVX512VL-NEXT: vpsrlq %xmm4, %zmm6, %zmm4
|
|
; AVX512VL-NEXT: vpandn %xmm2, %xmm3, %xmm2
|
|
; AVX512VL-NEXT: valignq {{.*#+}} zmm3 = zmm6[1,2,3,4,5,6,7],zmm5[0]
|
|
; AVX512VL-NEXT: vpaddq %zmm3, %zmm3, %zmm3
|
|
; AVX512VL-NEXT: vpsllq %xmm2, %zmm3, %zmm2
|
|
; AVX512VL-NEXT: vporq %zmm4, %zmm2, %zmm2
|
|
; AVX512VL-NEXT: xorl %ecx, %ecx
|
|
; AVX512VL-NEXT: negq %rdx
|
|
; AVX512VL-NEXT: sbbl %ecx, %ecx
|
|
; AVX512VL-NEXT: kmovd %ecx, %k1
|
|
; AVX512VL-NEXT: vporq %zmm2, %zmm1, %zmm0 {%k1}
|
|
; AVX512VL-NEXT: vextracti64x4 $1, %zmm0, 32(%rdi)
|
|
; AVX512VL-NEXT: vmovdqu %ymm0, (%rdi)
|
|
; AVX512VL-NEXT: vzeroupper
|
|
; AVX512VL-NEXT: retq
|
|
;
|
|
; AVX512VBMI-LABEL: fshl_rot_i512_load:
|
|
; AVX512VBMI: # %bb.0:
|
|
; AVX512VBMI-NEXT: movq %rdi, %rax
|
|
; AVX512VBMI-NEXT: vmovdqu64 (%rsi), %zmm0
|
|
; AVX512VBMI-NEXT: andl $511, %edx # imm = 0x1FF
|
|
; AVX512VBMI-NEXT: movl $512, %ecx # imm = 0x200
|
|
; AVX512VBMI-NEXT: subq %rdx, %rcx
|
|
; AVX512VBMI-NEXT: vpbroadcastq %rcx, %zmm1
|
|
; AVX512VBMI-NEXT: # kill: def $ecx killed $ecx killed $rcx
|
|
; AVX512VBMI-NEXT: shrl $6, %ecx
|
|
; AVX512VBMI-NEXT: movl $-1, %esi
|
|
; AVX512VBMI-NEXT: shlxl %ecx, %esi, %ecx
|
|
; AVX512VBMI-NEXT: kmovd %ecx, %k1
|
|
; AVX512VBMI-NEXT: vpcompressq %zmm0, %zmm2 {%k1} {z}
|
|
; AVX512VBMI-NEXT: vpxor %xmm3, %xmm3, %xmm3
|
|
; AVX512VBMI-NEXT: valignq {{.*#+}} zmm4 = zmm2[1,2,3,4,5,6,7],zmm3[0]
|
|
; AVX512VBMI-NEXT: movl %edx, %ecx
|
|
; AVX512VBMI-NEXT: shrl $6, %ecx
|
|
; AVX512VBMI-NEXT: shlxl %ecx, %esi, %ecx
|
|
; AVX512VBMI-NEXT: kmovd %ecx, %k1
|
|
; AVX512VBMI-NEXT: vpexpandq %zmm0, %zmm5 {%k1} {z}
|
|
; AVX512VBMI-NEXT: vpshrdvq %zmm1, %zmm4, %zmm2
|
|
; AVX512VBMI-NEXT: valignq {{.*#+}} zmm1 = zmm3[7],zmm5[0,1,2,3,4,5,6]
|
|
; AVX512VBMI-NEXT: vpbroadcastq %rdx, %zmm3
|
|
; AVX512VBMI-NEXT: vpshldvq %zmm3, %zmm1, %zmm5
|
|
; AVX512VBMI-NEXT: xorl %ecx, %ecx
|
|
; AVX512VBMI-NEXT: negq %rdx
|
|
; AVX512VBMI-NEXT: sbbl %ecx, %ecx
|
|
; AVX512VBMI-NEXT: kmovd %ecx, %k1
|
|
; AVX512VBMI-NEXT: vporq %zmm2, %zmm5, %zmm0 {%k1}
|
|
; AVX512VBMI-NEXT: vextracti64x4 $1, %zmm0, 32(%rdi)
|
|
; AVX512VBMI-NEXT: vmovdqu %ymm0, (%rdi)
|
|
; AVX512VBMI-NEXT: vzeroupper
|
|
; AVX512VBMI-NEXT: retq
|
|
%a0 = load i512, ptr %p0
|
|
%r = call i512 @llvm.fshl.i512(i512 %a0, i512 %a0, i512 %a2)
|
|
ret i512 %r
|
|
}
|
|
|
|
define i512 @fshr_rot_i512_load(ptr %p0, i512 %a2) nounwind {
|
|
; SSE-LABEL: fshr_rot_i512_load:
|
|
; SSE: # %bb.0:
|
|
; SSE-NEXT: pushq %r15
|
|
; SSE-NEXT: pushq %r14
|
|
; SSE-NEXT: pushq %r13
|
|
; SSE-NEXT: pushq %r12
|
|
; SSE-NEXT: pushq %rbx
|
|
; SSE-NEXT: movq %rdx, %rcx
|
|
; SSE-NEXT: movq 40(%rsi), %r8
|
|
; SSE-NEXT: movq 48(%rsi), %rax
|
|
; SSE-NEXT: movq 56(%rsi), %r11
|
|
; SSE-NEXT: movq (%rsi), %rdx
|
|
; SSE-NEXT: movq 32(%rsi), %r10
|
|
; SSE-NEXT: testl $256, %ecx # imm = 0x100
|
|
; SSE-NEXT: movq %r10, %r14
|
|
; SSE-NEXT: cmoveq %rdx, %r14
|
|
; SSE-NEXT: movq 24(%rsi), %rbx
|
|
; SSE-NEXT: movq %rbx, %r15
|
|
; SSE-NEXT: cmoveq %r11, %r15
|
|
; SSE-NEXT: movq 16(%rsi), %r12
|
|
; SSE-NEXT: movq %r12, %r9
|
|
; SSE-NEXT: cmoveq %rax, %r9
|
|
; SSE-NEXT: movq 8(%rsi), %r13
|
|
; SSE-NEXT: movq %r13, %rsi
|
|
; SSE-NEXT: cmoveq %r8, %rsi
|
|
; SSE-NEXT: cmoveq %r10, %rdx
|
|
; SSE-NEXT: cmoveq %r12, %rax
|
|
; SSE-NEXT: cmoveq %r13, %r8
|
|
; SSE-NEXT: cmoveq %rbx, %r11
|
|
; SSE-NEXT: testb $-128, %cl
|
|
; SSE-NEXT: movq %rsi, %r10
|
|
; SSE-NEXT: cmoveq %r11, %r10
|
|
; SSE-NEXT: cmoveq %r8, %r11
|
|
; SSE-NEXT: movq %r9, %rbx
|
|
; SSE-NEXT: cmoveq %rdx, %rbx
|
|
; SSE-NEXT: cmoveq %rax, %rdx
|
|
; SSE-NEXT: cmovneq %r15, %rsi
|
|
; SSE-NEXT: cmovneq %r14, %r9
|
|
; SSE-NEXT: cmoveq %r15, %r8
|
|
; SSE-NEXT: cmoveq %r14, %rax
|
|
; SSE-NEXT: testb $64, %cl
|
|
; SSE-NEXT: movq %rax, %r14
|
|
; SSE-NEXT: cmoveq %r8, %r14
|
|
; SSE-NEXT: cmoveq %r9, %r8
|
|
; SSE-NEXT: cmoveq %rsi, %r9
|
|
; SSE-NEXT: cmoveq %rbx, %rsi
|
|
; SSE-NEXT: cmoveq %r10, %rbx
|
|
; SSE-NEXT: cmoveq %rdx, %r10
|
|
; SSE-NEXT: cmoveq %r11, %rdx
|
|
; SSE-NEXT: cmovneq %r11, %rax
|
|
; SSE-NEXT: movq %rax, %r11
|
|
; SSE-NEXT: shrdq %cl, %rdx, %r11
|
|
; SSE-NEXT: shrdq %cl, %r10, %rdx
|
|
; SSE-NEXT: shrdq %cl, %rbx, %r10
|
|
; SSE-NEXT: shrdq %cl, %rsi, %rbx
|
|
; SSE-NEXT: shrdq %cl, %r9, %rsi
|
|
; SSE-NEXT: shrdq %cl, %r8, %r9
|
|
; SSE-NEXT: shrdq %cl, %r14, %r8
|
|
; SSE-NEXT: # kill: def $cl killed $cl killed $rcx
|
|
; SSE-NEXT: shrdq %cl, %rax, %r14
|
|
; SSE-NEXT: movq %rdi, %rax
|
|
; SSE-NEXT: movq %r14, 56(%rdi)
|
|
; SSE-NEXT: movq %r8, 48(%rdi)
|
|
; SSE-NEXT: movq %r9, 40(%rdi)
|
|
; SSE-NEXT: movq %rsi, 32(%rdi)
|
|
; SSE-NEXT: movq %rbx, 24(%rdi)
|
|
; SSE-NEXT: movq %r10, 16(%rdi)
|
|
; SSE-NEXT: movq %rdx, 8(%rdi)
|
|
; SSE-NEXT: movq %r11, (%rdi)
|
|
; SSE-NEXT: popq %rbx
|
|
; SSE-NEXT: popq %r12
|
|
; SSE-NEXT: popq %r13
|
|
; SSE-NEXT: popq %r14
|
|
; SSE-NEXT: popq %r15
|
|
; SSE-NEXT: retq
|
|
;
|
|
; AVX2-LABEL: fshr_rot_i512_load:
|
|
; AVX2: # %bb.0:
|
|
; AVX2-NEXT: pushq %r15
|
|
; AVX2-NEXT: pushq %r14
|
|
; AVX2-NEXT: pushq %r13
|
|
; AVX2-NEXT: pushq %r12
|
|
; AVX2-NEXT: pushq %rbx
|
|
; AVX2-NEXT: movq %rdx, %rcx
|
|
; AVX2-NEXT: movq 40(%rsi), %r8
|
|
; AVX2-NEXT: movq 48(%rsi), %r9
|
|
; AVX2-NEXT: movq 16(%rsi), %rax
|
|
; AVX2-NEXT: movq 56(%rsi), %r14
|
|
; AVX2-NEXT: movq 24(%rsi), %r11
|
|
; AVX2-NEXT: movq (%rsi), %rdx
|
|
; AVX2-NEXT: movq 8(%rsi), %rbx
|
|
; AVX2-NEXT: movq 32(%rsi), %r12
|
|
; AVX2-NEXT: testl $256, %ecx # imm = 0x100
|
|
; AVX2-NEXT: movq %r12, %r15
|
|
; AVX2-NEXT: cmoveq %rdx, %r15
|
|
; AVX2-NEXT: movq %r11, %r13
|
|
; AVX2-NEXT: cmoveq %r14, %r13
|
|
; AVX2-NEXT: movq %rax, %rsi
|
|
; AVX2-NEXT: cmoveq %r9, %rsi
|
|
; AVX2-NEXT: movq %rbx, %r10
|
|
; AVX2-NEXT: cmoveq %r8, %r10
|
|
; AVX2-NEXT: cmoveq %r12, %rdx
|
|
; AVX2-NEXT: cmoveq %rax, %r9
|
|
; AVX2-NEXT: cmoveq %rbx, %r8
|
|
; AVX2-NEXT: cmoveq %r11, %r14
|
|
; AVX2-NEXT: testb $-128, %cl
|
|
; AVX2-NEXT: movq %r10, %r11
|
|
; AVX2-NEXT: cmoveq %r14, %r11
|
|
; AVX2-NEXT: cmoveq %r8, %r14
|
|
; AVX2-NEXT: movq %rsi, %rbx
|
|
; AVX2-NEXT: cmoveq %rdx, %rbx
|
|
; AVX2-NEXT: cmoveq %r9, %rdx
|
|
; AVX2-NEXT: cmovneq %r13, %r10
|
|
; AVX2-NEXT: cmovneq %r15, %rsi
|
|
; AVX2-NEXT: cmoveq %r13, %r8
|
|
; AVX2-NEXT: movq %rdi, %rax
|
|
; AVX2-NEXT: cmoveq %r15, %r9
|
|
; AVX2-NEXT: testb $64, %cl
|
|
; AVX2-NEXT: movq %r9, %rdi
|
|
; AVX2-NEXT: cmoveq %r8, %rdi
|
|
; AVX2-NEXT: cmoveq %rsi, %r8
|
|
; AVX2-NEXT: cmoveq %r10, %rsi
|
|
; AVX2-NEXT: cmoveq %rbx, %r10
|
|
; AVX2-NEXT: cmoveq %r11, %rbx
|
|
; AVX2-NEXT: cmoveq %rdx, %r11
|
|
; AVX2-NEXT: cmoveq %r14, %rdx
|
|
; AVX2-NEXT: cmovneq %r14, %r9
|
|
; AVX2-NEXT: movq %r9, %r14
|
|
; AVX2-NEXT: shrdq %cl, %rdx, %r14
|
|
; AVX2-NEXT: shrdq %cl, %r11, %rdx
|
|
; AVX2-NEXT: shrdq %cl, %rbx, %r11
|
|
; AVX2-NEXT: shrdq %cl, %r10, %rbx
|
|
; AVX2-NEXT: shrdq %cl, %rsi, %r10
|
|
; AVX2-NEXT: shrdq %cl, %r8, %rsi
|
|
; AVX2-NEXT: shrdq %cl, %rdi, %r8
|
|
; AVX2-NEXT: # kill: def $cl killed $cl killed $rcx
|
|
; AVX2-NEXT: shrdq %cl, %r9, %rdi
|
|
; AVX2-NEXT: movq %rdi, 56(%rax)
|
|
; AVX2-NEXT: movq %r8, 48(%rax)
|
|
; AVX2-NEXT: movq %rsi, 40(%rax)
|
|
; AVX2-NEXT: movq %r10, 32(%rax)
|
|
; AVX2-NEXT: movq %rbx, 24(%rax)
|
|
; AVX2-NEXT: movq %r11, 16(%rax)
|
|
; AVX2-NEXT: movq %rdx, 8(%rax)
|
|
; AVX2-NEXT: movq %r14, (%rax)
|
|
; AVX2-NEXT: popq %rbx
|
|
; AVX2-NEXT: popq %r12
|
|
; AVX2-NEXT: popq %r13
|
|
; AVX2-NEXT: popq %r14
|
|
; AVX2-NEXT: popq %r15
|
|
; AVX2-NEXT: retq
|
|
;
|
|
; AVX512F-LABEL: fshr_rot_i512_load:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: movq %rdi, %rax
|
|
; AVX512F-NEXT: vmovdqu64 (%rsi), %zmm1
|
|
; AVX512F-NEXT: andl $511, %edx # imm = 0x1FF
|
|
; AVX512F-NEXT: vmovq %rdx, %xmm0
|
|
; AVX512F-NEXT: vpbroadcastq %xmm0, %xmm2
|
|
; AVX512F-NEXT: vpbroadcastq {{.*#+}} xmm0 = [63,63]
|
|
; AVX512F-NEXT: vpand %xmm0, %xmm2, %xmm3
|
|
; AVX512F-NEXT: movl %edx, %ecx
|
|
; AVX512F-NEXT: shrl $6, %ecx
|
|
; AVX512F-NEXT: movl $-1, %esi
|
|
; AVX512F-NEXT: shlxl %ecx, %esi, %ecx
|
|
; AVX512F-NEXT: kmovw %ecx, %k1
|
|
; AVX512F-NEXT: vpcompressq %zmm1, %zmm4 {%k1} {z}
|
|
; AVX512F-NEXT: vpsrlq %xmm3, %zmm4, %zmm3
|
|
; AVX512F-NEXT: vpandn %xmm0, %xmm2, %xmm2
|
|
; AVX512F-NEXT: vpxor %xmm5, %xmm5, %xmm5
|
|
; AVX512F-NEXT: valignq {{.*#+}} zmm4 = zmm4[1,2,3,4,5,6,7],zmm5[0]
|
|
; AVX512F-NEXT: vpaddq %zmm4, %zmm4, %zmm4
|
|
; AVX512F-NEXT: vpsllq %xmm2, %zmm4, %zmm2
|
|
; AVX512F-NEXT: vporq %zmm3, %zmm2, %zmm2
|
|
; AVX512F-NEXT: movl $512, %ecx # imm = 0x200
|
|
; AVX512F-NEXT: subq %rdx, %rcx
|
|
; AVX512F-NEXT: vmovq %rcx, %xmm3
|
|
; AVX512F-NEXT: vpbroadcastq %xmm3, %xmm3
|
|
; AVX512F-NEXT: vpand %xmm0, %xmm3, %xmm4
|
|
; AVX512F-NEXT: shrl $6, %ecx
|
|
; AVX512F-NEXT: shlxl %ecx, %esi, %ecx
|
|
; AVX512F-NEXT: kmovw %ecx, %k1
|
|
; AVX512F-NEXT: vpexpandq %zmm1, %zmm6 {%k1} {z}
|
|
; AVX512F-NEXT: vpsllq %xmm4, %zmm6, %zmm4
|
|
; AVX512F-NEXT: vpandn %xmm0, %xmm3, %xmm0
|
|
; AVX512F-NEXT: valignq {{.*#+}} zmm3 = zmm5[7],zmm6[0,1,2,3,4,5,6]
|
|
; AVX512F-NEXT: vpsrlq $1, %zmm3, %zmm3
|
|
; AVX512F-NEXT: vpsrlq %xmm0, %zmm3, %zmm0
|
|
; AVX512F-NEXT: xorl %ecx, %ecx
|
|
; AVX512F-NEXT: negq %rdx
|
|
; AVX512F-NEXT: sbbl %ecx, %ecx
|
|
; AVX512F-NEXT: vporq %zmm0, %zmm4, %zmm0
|
|
; AVX512F-NEXT: kmovw %ecx, %k1
|
|
; AVX512F-NEXT: vporq %zmm2, %zmm0, %zmm1 {%k1}
|
|
; AVX512F-NEXT: vmovdqu64 %zmm1, (%rdi)
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: fshr_rot_i512_load:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: movq %rdi, %rax
|
|
; AVX512VL-NEXT: vmovdqu64 (%rsi), %zmm0
|
|
; AVX512VL-NEXT: andl $511, %edx # imm = 0x1FF
|
|
; AVX512VL-NEXT: vpbroadcastq %rdx, %xmm1
|
|
; AVX512VL-NEXT: vpbroadcastq {{.*#+}} xmm2 = [63,63]
|
|
; AVX512VL-NEXT: vpand %xmm2, %xmm1, %xmm3
|
|
; AVX512VL-NEXT: movl %edx, %ecx
|
|
; AVX512VL-NEXT: shrl $6, %ecx
|
|
; AVX512VL-NEXT: movl $-1, %esi
|
|
; AVX512VL-NEXT: shlxl %ecx, %esi, %ecx
|
|
; AVX512VL-NEXT: kmovd %ecx, %k1
|
|
; AVX512VL-NEXT: vpcompressq %zmm0, %zmm4 {%k1} {z}
|
|
; AVX512VL-NEXT: vpsrlq %xmm3, %zmm4, %zmm3
|
|
; AVX512VL-NEXT: vpandn %xmm2, %xmm1, %xmm1
|
|
; AVX512VL-NEXT: vpxor %xmm5, %xmm5, %xmm5
|
|
; AVX512VL-NEXT: valignq {{.*#+}} zmm4 = zmm4[1,2,3,4,5,6,7],zmm5[0]
|
|
; AVX512VL-NEXT: vpaddq %zmm4, %zmm4, %zmm4
|
|
; AVX512VL-NEXT: vpsllq %xmm1, %zmm4, %zmm1
|
|
; AVX512VL-NEXT: vporq %zmm3, %zmm1, %zmm1
|
|
; AVX512VL-NEXT: movl $512, %ecx # imm = 0x200
|
|
; AVX512VL-NEXT: subq %rdx, %rcx
|
|
; AVX512VL-NEXT: vpbroadcastq %rcx, %xmm3
|
|
; AVX512VL-NEXT: vpand %xmm2, %xmm3, %xmm4
|
|
; AVX512VL-NEXT: shrl $6, %ecx
|
|
; AVX512VL-NEXT: shlxl %ecx, %esi, %ecx
|
|
; AVX512VL-NEXT: kmovd %ecx, %k1
|
|
; AVX512VL-NEXT: vpexpandq %zmm0, %zmm6 {%k1} {z}
|
|
; AVX512VL-NEXT: vpsllq %xmm4, %zmm6, %zmm4
|
|
; AVX512VL-NEXT: vpandn %xmm2, %xmm3, %xmm2
|
|
; AVX512VL-NEXT: valignq {{.*#+}} zmm3 = zmm5[7],zmm6[0,1,2,3,4,5,6]
|
|
; AVX512VL-NEXT: vpsrlq $1, %zmm3, %zmm3
|
|
; AVX512VL-NEXT: vpsrlq %xmm2, %zmm3, %zmm2
|
|
; AVX512VL-NEXT: vporq %zmm2, %zmm4, %zmm2
|
|
; AVX512VL-NEXT: xorl %ecx, %ecx
|
|
; AVX512VL-NEXT: negq %rdx
|
|
; AVX512VL-NEXT: sbbl %ecx, %ecx
|
|
; AVX512VL-NEXT: kmovd %ecx, %k1
|
|
; AVX512VL-NEXT: vporq %zmm1, %zmm2, %zmm0 {%k1}
|
|
; AVX512VL-NEXT: vextracti64x4 $1, %zmm0, 32(%rdi)
|
|
; AVX512VL-NEXT: vmovdqu %ymm0, (%rdi)
|
|
; AVX512VL-NEXT: vzeroupper
|
|
; AVX512VL-NEXT: retq
|
|
;
|
|
; AVX512VBMI-LABEL: fshr_rot_i512_load:
|
|
; AVX512VBMI: # %bb.0:
|
|
; AVX512VBMI-NEXT: movq %rdi, %rax
|
|
; AVX512VBMI-NEXT: vmovdqu64 (%rsi), %zmm0
|
|
; AVX512VBMI-NEXT: andl $511, %edx # imm = 0x1FF
|
|
; AVX512VBMI-NEXT: movl $512, %ecx # imm = 0x200
|
|
; AVX512VBMI-NEXT: subq %rdx, %rcx
|
|
; AVX512VBMI-NEXT: vpbroadcastq %rcx, %zmm1
|
|
; AVX512VBMI-NEXT: # kill: def $ecx killed $ecx killed $rcx
|
|
; AVX512VBMI-NEXT: shrl $6, %ecx
|
|
; AVX512VBMI-NEXT: movl $-1, %esi
|
|
; AVX512VBMI-NEXT: shlxl %ecx, %esi, %ecx
|
|
; AVX512VBMI-NEXT: kmovd %ecx, %k1
|
|
; AVX512VBMI-NEXT: vpexpandq %zmm0, %zmm2 {%k1} {z}
|
|
; AVX512VBMI-NEXT: vpxor %xmm3, %xmm3, %xmm3
|
|
; AVX512VBMI-NEXT: valignq {{.*#+}} zmm4 = zmm3[7],zmm2[0,1,2,3,4,5,6]
|
|
; AVX512VBMI-NEXT: movl %edx, %ecx
|
|
; AVX512VBMI-NEXT: shrl $6, %ecx
|
|
; AVX512VBMI-NEXT: shlxl %ecx, %esi, %ecx
|
|
; AVX512VBMI-NEXT: kmovd %ecx, %k1
|
|
; AVX512VBMI-NEXT: vpcompressq %zmm0, %zmm5 {%k1} {z}
|
|
; AVX512VBMI-NEXT: vpshldvq %zmm1, %zmm4, %zmm2
|
|
; AVX512VBMI-NEXT: valignq {{.*#+}} zmm1 = zmm5[1,2,3,4,5,6,7],zmm3[0]
|
|
; AVX512VBMI-NEXT: vpbroadcastq %rdx, %zmm3
|
|
; AVX512VBMI-NEXT: vpshrdvq %zmm3, %zmm1, %zmm5
|
|
; AVX512VBMI-NEXT: xorl %ecx, %ecx
|
|
; AVX512VBMI-NEXT: negq %rdx
|
|
; AVX512VBMI-NEXT: sbbl %ecx, %ecx
|
|
; AVX512VBMI-NEXT: kmovd %ecx, %k1
|
|
; AVX512VBMI-NEXT: vporq %zmm5, %zmm2, %zmm0 {%k1}
|
|
; AVX512VBMI-NEXT: vextracti64x4 $1, %zmm0, 32(%rdi)
|
|
; AVX512VBMI-NEXT: vmovdqu %ymm0, (%rdi)
|
|
; AVX512VBMI-NEXT: vzeroupper
|
|
; AVX512VBMI-NEXT: retq
|
|
%a0 = load i512, ptr %p0
|
|
%r = call i512 @llvm.fshr.i512(i512 %a0, i512 %a0, i512 %a2)
|
|
ret i512 %r
|
|
}
|
|
|
|
define i512 @fshl_i512_vector(<8 x i64> %v0, <8 x i64> %v1, i512 %a2) nounwind {
|
|
; SSE2-LABEL: fshl_i512_vector:
|
|
; SSE2: # %bb.0:
|
|
; SSE2-NEXT: pushq %rbp
|
|
; SSE2-NEXT: pushq %r15
|
|
; SSE2-NEXT: pushq %r14
|
|
; SSE2-NEXT: pushq %r13
|
|
; SSE2-NEXT: pushq %r12
|
|
; SSE2-NEXT: pushq %rbx
|
|
; SSE2-NEXT: movq %rsi, %rcx
|
|
; SSE2-NEXT: movq %xmm0, %r10
|
|
; SSE2-NEXT: movq %xmm2, %rax
|
|
; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[2,3,2,3]
|
|
; SSE2-NEXT: movq %xmm0, %r9
|
|
; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm2[2,3,2,3]
|
|
; SSE2-NEXT: movq %xmm0, %rdx
|
|
; SSE2-NEXT: movq %xmm1, %r8
|
|
; SSE2-NEXT: movq %xmm3, %r12
|
|
; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm1[2,3,2,3]
|
|
; SSE2-NEXT: movq %xmm0, %rsi
|
|
; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm3[2,3,2,3]
|
|
; SSE2-NEXT: movq %xmm0, %r15
|
|
; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm5[2,3,2,3]
|
|
; SSE2-NEXT: movq %xmm0, %r11
|
|
; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm4[2,3,2,3]
|
|
; SSE2-NEXT: movq %xmm0, %r14
|
|
; SSE2-NEXT: movq %xmm5, %rbx
|
|
; SSE2-NEXT: movq %xmm7, %r13
|
|
; SSE2-NEXT: testl $256, %ecx # imm = 0x100
|
|
; SSE2-NEXT: cmovneq %r8, %r12
|
|
; SSE2-NEXT: cmovneq %r13, %r8
|
|
; SSE2-NEXT: cmoveq %r13, %rbx
|
|
; SSE2-NEXT: movq %xmm4, %r13
|
|
; SSE2-NEXT: movq %xmm6, %rbp
|
|
; SSE2-NEXT: cmovneq %r10, %rax
|
|
; SSE2-NEXT: cmovneq %rbp, %r10
|
|
; SSE2-NEXT: cmoveq %rbp, %r13
|
|
; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm6[2,3,2,3]
|
|
; SSE2-NEXT: movq %xmm0, %rbp
|
|
; SSE2-NEXT: cmovneq %r9, %rdx
|
|
; SSE2-NEXT: cmovneq %rbp, %r9
|
|
; SSE2-NEXT: cmoveq %rbp, %r14
|
|
; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm7[2,3,2,3]
|
|
; SSE2-NEXT: movq %xmm0, %rbp
|
|
; SSE2-NEXT: cmovneq %rsi, %r15
|
|
; SSE2-NEXT: cmovneq %rbp, %rsi
|
|
; SSE2-NEXT: cmoveq %rbp, %r11
|
|
; SSE2-NEXT: testb $-128, %cl
|
|
; SSE2-NEXT: cmoveq %rbx, %r13
|
|
; SSE2-NEXT: cmoveq %r10, %rbx
|
|
; SSE2-NEXT: cmoveq %r8, %r10
|
|
; SSE2-NEXT: cmoveq %rax, %r8
|
|
; SSE2-NEXT: cmoveq %r12, %rax
|
|
; SSE2-NEXT: cmoveq %r11, %r14
|
|
; SSE2-NEXT: cmoveq %r9, %r11
|
|
; SSE2-NEXT: cmoveq %rsi, %r9
|
|
; SSE2-NEXT: cmoveq %rdx, %rsi
|
|
; SSE2-NEXT: cmoveq %r15, %rdx
|
|
; SSE2-NEXT: testb $64, %cl
|
|
; SSE2-NEXT: cmovneq %rax, %rdx
|
|
; SSE2-NEXT: cmovneq %rsi, %rax
|
|
; SSE2-NEXT: cmovneq %r8, %rsi
|
|
; SSE2-NEXT: cmovneq %r9, %r8
|
|
; SSE2-NEXT: cmovneq %r10, %r9
|
|
; SSE2-NEXT: cmovneq %r11, %r10
|
|
; SSE2-NEXT: cmovneq %rbx, %r11
|
|
; SSE2-NEXT: cmoveq %r14, %r13
|
|
; SSE2-NEXT: cmovneq %r14, %rbx
|
|
; SSE2-NEXT: movq %rbx, %r14
|
|
; SSE2-NEXT: shldq %cl, %r13, %r14
|
|
; SSE2-NEXT: movq %r11, %r15
|
|
; SSE2-NEXT: shldq %cl, %rbx, %r15
|
|
; SSE2-NEXT: movq %r10, %rbx
|
|
; SSE2-NEXT: shldq %cl, %r11, %rbx
|
|
; SSE2-NEXT: movq %r9, %r11
|
|
; SSE2-NEXT: shldq %cl, %r10, %r11
|
|
; SSE2-NEXT: movq %r8, %r10
|
|
; SSE2-NEXT: shldq %cl, %r9, %r10
|
|
; SSE2-NEXT: movq %rsi, %r9
|
|
; SSE2-NEXT: shldq %cl, %r8, %r9
|
|
; SSE2-NEXT: movq %rax, %r8
|
|
; SSE2-NEXT: shldq %cl, %rsi, %r8
|
|
; SSE2-NEXT: # kill: def $cl killed $cl killed $rcx
|
|
; SSE2-NEXT: shldq %cl, %rax, %rdx
|
|
; SSE2-NEXT: movq %rdi, %rax
|
|
; SSE2-NEXT: movq %rdx, 56(%rdi)
|
|
; SSE2-NEXT: movq %r8, 48(%rdi)
|
|
; SSE2-NEXT: movq %r9, 40(%rdi)
|
|
; SSE2-NEXT: movq %r10, 32(%rdi)
|
|
; SSE2-NEXT: movq %r11, 24(%rdi)
|
|
; SSE2-NEXT: movq %rbx, 16(%rdi)
|
|
; SSE2-NEXT: movq %r15, 8(%rdi)
|
|
; SSE2-NEXT: movq %r14, (%rdi)
|
|
; SSE2-NEXT: popq %rbx
|
|
; SSE2-NEXT: popq %r12
|
|
; SSE2-NEXT: popq %r13
|
|
; SSE2-NEXT: popq %r14
|
|
; SSE2-NEXT: popq %r15
|
|
; SSE2-NEXT: popq %rbp
|
|
; SSE2-NEXT: retq
|
|
;
|
|
; SSE42-LABEL: fshl_i512_vector:
|
|
; SSE42: # %bb.0:
|
|
; SSE42-NEXT: pushq %rbp
|
|
; SSE42-NEXT: pushq %r15
|
|
; SSE42-NEXT: pushq %r14
|
|
; SSE42-NEXT: pushq %r13
|
|
; SSE42-NEXT: pushq %r12
|
|
; SSE42-NEXT: pushq %rbx
|
|
; SSE42-NEXT: movq %rsi, %rcx
|
|
; SSE42-NEXT: movq %xmm0, %r10
|
|
; SSE42-NEXT: pextrq $1, %xmm0, %r9
|
|
; SSE42-NEXT: pextrq $1, %xmm2, %rdx
|
|
; SSE42-NEXT: movq %xmm2, %rax
|
|
; SSE42-NEXT: movq %xmm1, %r8
|
|
; SSE42-NEXT: pextrq $1, %xmm1, %rsi
|
|
; SSE42-NEXT: pextrq $1, %xmm3, %r15
|
|
; SSE42-NEXT: pextrq $1, %xmm5, %r11
|
|
; SSE42-NEXT: movq %xmm3, %r13
|
|
; SSE42-NEXT: pextrq $1, %xmm4, %r14
|
|
; SSE42-NEXT: movq %xmm5, %rbx
|
|
; SSE42-NEXT: movq %xmm7, %r12
|
|
; SSE42-NEXT: testl $256, %ecx # imm = 0x100
|
|
; SSE42-NEXT: cmovneq %r8, %r13
|
|
; SSE42-NEXT: cmovneq %r12, %r8
|
|
; SSE42-NEXT: cmoveq %r12, %rbx
|
|
; SSE42-NEXT: movq %xmm4, %r12
|
|
; SSE42-NEXT: movq %xmm6, %rbp
|
|
; SSE42-NEXT: cmovneq %r10, %rax
|
|
; SSE42-NEXT: cmovneq %rbp, %r10
|
|
; SSE42-NEXT: cmoveq %rbp, %r12
|
|
; SSE42-NEXT: pextrq $1, %xmm6, %rbp
|
|
; SSE42-NEXT: cmovneq %r9, %rdx
|
|
; SSE42-NEXT: cmovneq %rbp, %r9
|
|
; SSE42-NEXT: cmoveq %rbp, %r14
|
|
; SSE42-NEXT: pextrq $1, %xmm7, %rbp
|
|
; SSE42-NEXT: cmovneq %rsi, %r15
|
|
; SSE42-NEXT: cmovneq %rbp, %rsi
|
|
; SSE42-NEXT: cmoveq %rbp, %r11
|
|
; SSE42-NEXT: testb $-128, %cl
|
|
; SSE42-NEXT: cmoveq %rbx, %r12
|
|
; SSE42-NEXT: cmoveq %r10, %rbx
|
|
; SSE42-NEXT: cmoveq %r8, %r10
|
|
; SSE42-NEXT: cmoveq %rax, %r8
|
|
; SSE42-NEXT: cmoveq %r13, %rax
|
|
; SSE42-NEXT: cmoveq %r11, %r14
|
|
; SSE42-NEXT: cmoveq %r9, %r11
|
|
; SSE42-NEXT: cmoveq %rsi, %r9
|
|
; SSE42-NEXT: cmoveq %rdx, %rsi
|
|
; SSE42-NEXT: cmoveq %r15, %rdx
|
|
; SSE42-NEXT: testb $64, %cl
|
|
; SSE42-NEXT: cmovneq %rax, %rdx
|
|
; SSE42-NEXT: cmovneq %rsi, %rax
|
|
; SSE42-NEXT: cmovneq %r8, %rsi
|
|
; SSE42-NEXT: cmovneq %r9, %r8
|
|
; SSE42-NEXT: cmovneq %r10, %r9
|
|
; SSE42-NEXT: cmovneq %r11, %r10
|
|
; SSE42-NEXT: cmovneq %rbx, %r11
|
|
; SSE42-NEXT: cmoveq %r14, %r12
|
|
; SSE42-NEXT: cmovneq %r14, %rbx
|
|
; SSE42-NEXT: movq %rbx, %r14
|
|
; SSE42-NEXT: shldq %cl, %r12, %r14
|
|
; SSE42-NEXT: movq %r11, %r15
|
|
; SSE42-NEXT: shldq %cl, %rbx, %r15
|
|
; SSE42-NEXT: movq %r10, %rbx
|
|
; SSE42-NEXT: shldq %cl, %r11, %rbx
|
|
; SSE42-NEXT: movq %r9, %r11
|
|
; SSE42-NEXT: shldq %cl, %r10, %r11
|
|
; SSE42-NEXT: movq %r8, %r10
|
|
; SSE42-NEXT: shldq %cl, %r9, %r10
|
|
; SSE42-NEXT: movq %rsi, %r9
|
|
; SSE42-NEXT: shldq %cl, %r8, %r9
|
|
; SSE42-NEXT: movq %rax, %r8
|
|
; SSE42-NEXT: shldq %cl, %rsi, %r8
|
|
; SSE42-NEXT: # kill: def $cl killed $cl killed $rcx
|
|
; SSE42-NEXT: shldq %cl, %rax, %rdx
|
|
; SSE42-NEXT: movq %rdi, %rax
|
|
; SSE42-NEXT: movq %rdx, 56(%rdi)
|
|
; SSE42-NEXT: movq %r8, 48(%rdi)
|
|
; SSE42-NEXT: movq %r9, 40(%rdi)
|
|
; SSE42-NEXT: movq %r10, 32(%rdi)
|
|
; SSE42-NEXT: movq %r11, 24(%rdi)
|
|
; SSE42-NEXT: movq %rbx, 16(%rdi)
|
|
; SSE42-NEXT: movq %r15, 8(%rdi)
|
|
; SSE42-NEXT: movq %r14, (%rdi)
|
|
; SSE42-NEXT: popq %rbx
|
|
; SSE42-NEXT: popq %r12
|
|
; SSE42-NEXT: popq %r13
|
|
; SSE42-NEXT: popq %r14
|
|
; SSE42-NEXT: popq %r15
|
|
; SSE42-NEXT: popq %rbp
|
|
; SSE42-NEXT: retq
|
|
;
|
|
; AVX2-LABEL: fshl_i512_vector:
|
|
; AVX2: # %bb.0:
|
|
; AVX2-NEXT: pushq %rbp
|
|
; AVX2-NEXT: pushq %r15
|
|
; AVX2-NEXT: pushq %r14
|
|
; AVX2-NEXT: pushq %r13
|
|
; AVX2-NEXT: pushq %r12
|
|
; AVX2-NEXT: pushq %rbx
|
|
; AVX2-NEXT: movq %rsi, %rcx
|
|
; AVX2-NEXT: vmovq %xmm0, %r11
|
|
; AVX2-NEXT: vpextrq $1, %xmm0, %r9
|
|
; AVX2-NEXT: vmovq %xmm1, %rdx
|
|
; AVX2-NEXT: vpextrq $1, %xmm1, %rsi
|
|
; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm0
|
|
; AVX2-NEXT: vmovq %xmm0, %r10
|
|
; AVX2-NEXT: vextracti128 $1, %ymm1, %xmm1
|
|
; AVX2-NEXT: vpextrq $1, %xmm0, %r8
|
|
; AVX2-NEXT: vmovq %xmm1, %r13
|
|
; AVX2-NEXT: vpextrq $1, %xmm1, %rax
|
|
; AVX2-NEXT: vextracti128 $1, %ymm2, %xmm1
|
|
; AVX2-NEXT: vpextrq $1, %xmm1, %rbx
|
|
; AVX2-NEXT: vpextrq $1, %xmm2, %r15
|
|
; AVX2-NEXT: vmovq %xmm2, %r14
|
|
; AVX2-NEXT: vextracti128 $1, %ymm3, %xmm0
|
|
; AVX2-NEXT: vmovq %xmm0, %rbp
|
|
; AVX2-NEXT: testl $256, %ecx # imm = 0x100
|
|
; AVX2-NEXT: cmovneq %r10, %r13
|
|
; AVX2-NEXT: cmovneq %rbp, %r10
|
|
; AVX2-NEXT: vmovq %xmm1, %r12
|
|
; AVX2-NEXT: cmoveq %rbp, %r12
|
|
; AVX2-NEXT: vmovq %xmm3, %rbp
|
|
; AVX2-NEXT: cmovneq %r11, %rdx
|
|
; AVX2-NEXT: cmovneq %rbp, %r11
|
|
; AVX2-NEXT: cmoveq %rbp, %r14
|
|
; AVX2-NEXT: vpextrq $1, %xmm3, %rbp
|
|
; AVX2-NEXT: cmovneq %r9, %rsi
|
|
; AVX2-NEXT: cmovneq %rbp, %r9
|
|
; AVX2-NEXT: cmoveq %rbp, %r15
|
|
; AVX2-NEXT: vpextrq $1, %xmm0, %rbp
|
|
; AVX2-NEXT: cmovneq %r8, %rax
|
|
; AVX2-NEXT: cmovneq %rbp, %r8
|
|
; AVX2-NEXT: cmoveq %rbp, %rbx
|
|
; AVX2-NEXT: testb $-128, %cl
|
|
; AVX2-NEXT: cmoveq %r12, %r14
|
|
; AVX2-NEXT: cmoveq %r11, %r12
|
|
; AVX2-NEXT: cmoveq %r10, %r11
|
|
; AVX2-NEXT: cmoveq %rdx, %r10
|
|
; AVX2-NEXT: cmoveq %r13, %rdx
|
|
; AVX2-NEXT: cmoveq %rbx, %r15
|
|
; AVX2-NEXT: cmoveq %r9, %rbx
|
|
; AVX2-NEXT: cmoveq %r8, %r9
|
|
; AVX2-NEXT: cmoveq %rsi, %r8
|
|
; AVX2-NEXT: cmoveq %rax, %rsi
|
|
; AVX2-NEXT: movq %rdi, %rax
|
|
; AVX2-NEXT: testb $64, %cl
|
|
; AVX2-NEXT: cmovneq %rdx, %rsi
|
|
; AVX2-NEXT: cmovneq %r8, %rdx
|
|
; AVX2-NEXT: cmovneq %r10, %r8
|
|
; AVX2-NEXT: cmovneq %r9, %r10
|
|
; AVX2-NEXT: cmovneq %r11, %r9
|
|
; AVX2-NEXT: cmovneq %rbx, %r11
|
|
; AVX2-NEXT: cmovneq %r12, %rbx
|
|
; AVX2-NEXT: cmoveq %r15, %r14
|
|
; AVX2-NEXT: cmovneq %r15, %r12
|
|
; AVX2-NEXT: movq %r12, %rdi
|
|
; AVX2-NEXT: shldq %cl, %r14, %rdi
|
|
; AVX2-NEXT: movq %rbx, %r14
|
|
; AVX2-NEXT: shldq %cl, %r12, %r14
|
|
; AVX2-NEXT: movq %r11, %r15
|
|
; AVX2-NEXT: shldq %cl, %rbx, %r15
|
|
; AVX2-NEXT: movq %r9, %rbx
|
|
; AVX2-NEXT: shldq %cl, %r11, %rbx
|
|
; AVX2-NEXT: movq %r10, %r11
|
|
; AVX2-NEXT: shldq %cl, %r9, %r11
|
|
; AVX2-NEXT: movq %r8, %r9
|
|
; AVX2-NEXT: shldq %cl, %r10, %r9
|
|
; AVX2-NEXT: movq %rdx, %r10
|
|
; AVX2-NEXT: shldq %cl, %r8, %r10
|
|
; AVX2-NEXT: # kill: def $cl killed $cl killed $rcx
|
|
; AVX2-NEXT: shldq %cl, %rdx, %rsi
|
|
; AVX2-NEXT: movq %rsi, 56(%rax)
|
|
; AVX2-NEXT: movq %r10, 48(%rax)
|
|
; AVX2-NEXT: movq %r9, 40(%rax)
|
|
; AVX2-NEXT: movq %r11, 32(%rax)
|
|
; AVX2-NEXT: movq %rbx, 24(%rax)
|
|
; AVX2-NEXT: movq %r15, 16(%rax)
|
|
; AVX2-NEXT: movq %r14, 8(%rax)
|
|
; AVX2-NEXT: movq %rdi, (%rax)
|
|
; AVX2-NEXT: popq %rbx
|
|
; AVX2-NEXT: popq %r12
|
|
; AVX2-NEXT: popq %r13
|
|
; AVX2-NEXT: popq %r14
|
|
; AVX2-NEXT: popq %r15
|
|
; AVX2-NEXT: popq %rbp
|
|
; AVX2-NEXT: vzeroupper
|
|
; AVX2-NEXT: retq
|
|
;
|
|
; AVX512F-LABEL: fshl_i512_vector:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: movq %rdi, %rax
|
|
; AVX512F-NEXT: andl $511, %esi # imm = 0x1FF
|
|
; AVX512F-NEXT: vmovq %rsi, %xmm2
|
|
; AVX512F-NEXT: vpbroadcastq %xmm2, %xmm3
|
|
; AVX512F-NEXT: vpbroadcastq {{.*#+}} xmm2 = [63,63]
|
|
; AVX512F-NEXT: vpand %xmm2, %xmm3, %xmm4
|
|
; AVX512F-NEXT: movl %esi, %ecx
|
|
; AVX512F-NEXT: shrl $6, %ecx
|
|
; AVX512F-NEXT: movl $-1, %edx
|
|
; AVX512F-NEXT: shlxl %ecx, %edx, %ecx
|
|
; AVX512F-NEXT: kmovw %ecx, %k1
|
|
; AVX512F-NEXT: vpexpandq %zmm0, %zmm5 {%k1} {z}
|
|
; AVX512F-NEXT: vpsllq %xmm4, %zmm5, %zmm4
|
|
; AVX512F-NEXT: vpandn %xmm2, %xmm3, %xmm3
|
|
; AVX512F-NEXT: vpxor %xmm6, %xmm6, %xmm6
|
|
; AVX512F-NEXT: valignq {{.*#+}} zmm5 = zmm6[7],zmm5[0,1,2,3,4,5,6]
|
|
; AVX512F-NEXT: vpsrlq $1, %zmm5, %zmm5
|
|
; AVX512F-NEXT: vpsrlq %xmm3, %zmm5, %zmm3
|
|
; AVX512F-NEXT: vporq %zmm3, %zmm4, %zmm3
|
|
; AVX512F-NEXT: movl $512, %ecx # imm = 0x200
|
|
; AVX512F-NEXT: subq %rsi, %rcx
|
|
; AVX512F-NEXT: vmovq %rcx, %xmm4
|
|
; AVX512F-NEXT: vpbroadcastq %xmm4, %xmm4
|
|
; AVX512F-NEXT: vpand %xmm2, %xmm4, %xmm5
|
|
; AVX512F-NEXT: shrl $6, %ecx
|
|
; AVX512F-NEXT: shlxl %ecx, %edx, %ecx
|
|
; AVX512F-NEXT: kmovw %ecx, %k1
|
|
; AVX512F-NEXT: vpcompressq %zmm1, %zmm1 {%k1} {z}
|
|
; AVX512F-NEXT: vpsrlq %xmm5, %zmm1, %zmm5
|
|
; AVX512F-NEXT: vpandn %xmm2, %xmm4, %xmm2
|
|
; AVX512F-NEXT: valignq {{.*#+}} zmm1 = zmm1[1,2,3,4,5,6,7],zmm6[0]
|
|
; AVX512F-NEXT: vpaddq %zmm1, %zmm1, %zmm1
|
|
; AVX512F-NEXT: vpsllq %xmm2, %zmm1, %zmm1
|
|
; AVX512F-NEXT: xorl %ecx, %ecx
|
|
; AVX512F-NEXT: negq %rsi
|
|
; AVX512F-NEXT: sbbl %ecx, %ecx
|
|
; AVX512F-NEXT: vporq %zmm5, %zmm1, %zmm1
|
|
; AVX512F-NEXT: kmovw %ecx, %k1
|
|
; AVX512F-NEXT: vporq %zmm1, %zmm3, %zmm0 {%k1}
|
|
; AVX512F-NEXT: vmovdqu64 %zmm0, (%rdi)
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: fshl_i512_vector:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: andl $511, %esi # imm = 0x1FF
|
|
; AVX512VL-NEXT: vpbroadcastq %rsi, %xmm2
|
|
; AVX512VL-NEXT: vpbroadcastq {{.*#+}} xmm3 = [63,63]
|
|
; AVX512VL-NEXT: vpand %xmm3, %xmm2, %xmm4
|
|
; AVX512VL-NEXT: movl %esi, %eax
|
|
; AVX512VL-NEXT: shrl $6, %eax
|
|
; AVX512VL-NEXT: movl $-1, %ecx
|
|
; AVX512VL-NEXT: shlxl %eax, %ecx, %eax
|
|
; AVX512VL-NEXT: kmovd %eax, %k1
|
|
; AVX512VL-NEXT: vpexpandq %zmm0, %zmm5 {%k1} {z}
|
|
; AVX512VL-NEXT: vpsllq %xmm4, %zmm5, %zmm4
|
|
; AVX512VL-NEXT: vpandn %xmm3, %xmm2, %xmm2
|
|
; AVX512VL-NEXT: vpxor %xmm6, %xmm6, %xmm6
|
|
; AVX512VL-NEXT: valignq {{.*#+}} zmm5 = zmm6[7],zmm5[0,1,2,3,4,5,6]
|
|
; AVX512VL-NEXT: vpsrlq $1, %zmm5, %zmm5
|
|
; AVX512VL-NEXT: vpsrlq %xmm2, %zmm5, %zmm2
|
|
; AVX512VL-NEXT: movq %rdi, %rax
|
|
; AVX512VL-NEXT: vporq %zmm2, %zmm4, %zmm2
|
|
; AVX512VL-NEXT: movl $512, %edx # imm = 0x200
|
|
; AVX512VL-NEXT: subq %rsi, %rdx
|
|
; AVX512VL-NEXT: vpbroadcastq %rdx, %xmm4
|
|
; AVX512VL-NEXT: vpand %xmm3, %xmm4, %xmm5
|
|
; AVX512VL-NEXT: shrl $6, %edx
|
|
; AVX512VL-NEXT: shlxl %edx, %ecx, %ecx
|
|
; AVX512VL-NEXT: kmovd %ecx, %k1
|
|
; AVX512VL-NEXT: vpcompressq %zmm1, %zmm1 {%k1} {z}
|
|
; AVX512VL-NEXT: vpsrlq %xmm5, %zmm1, %zmm5
|
|
; AVX512VL-NEXT: vpandn %xmm3, %xmm4, %xmm3
|
|
; AVX512VL-NEXT: valignq {{.*#+}} zmm1 = zmm1[1,2,3,4,5,6,7],zmm6[0]
|
|
; AVX512VL-NEXT: vpaddq %zmm1, %zmm1, %zmm1
|
|
; AVX512VL-NEXT: vpsllq %xmm3, %zmm1, %zmm1
|
|
; AVX512VL-NEXT: vporq %zmm5, %zmm1, %zmm1
|
|
; AVX512VL-NEXT: xorl %ecx, %ecx
|
|
; AVX512VL-NEXT: negq %rsi
|
|
; AVX512VL-NEXT: sbbl %ecx, %ecx
|
|
; AVX512VL-NEXT: kmovd %ecx, %k1
|
|
; AVX512VL-NEXT: vporq %zmm1, %zmm2, %zmm0 {%k1}
|
|
; AVX512VL-NEXT: vmovdqu64 %zmm0, (%rdi)
|
|
; AVX512VL-NEXT: vzeroupper
|
|
; AVX512VL-NEXT: retq
|
|
;
|
|
; AVX512VBMI-LABEL: fshl_i512_vector:
|
|
; AVX512VBMI: # %bb.0:
|
|
; AVX512VBMI-NEXT: movq %rdi, %rax
|
|
; AVX512VBMI-NEXT: andl $511, %esi # imm = 0x1FF
|
|
; AVX512VBMI-NEXT: movl $512, %ecx # imm = 0x200
|
|
; AVX512VBMI-NEXT: subq %rsi, %rcx
|
|
; AVX512VBMI-NEXT: vpbroadcastq %rcx, %zmm2
|
|
; AVX512VBMI-NEXT: # kill: def $ecx killed $ecx killed $rcx
|
|
; AVX512VBMI-NEXT: shrl $6, %ecx
|
|
; AVX512VBMI-NEXT: movl $-1, %edx
|
|
; AVX512VBMI-NEXT: shlxl %ecx, %edx, %ecx
|
|
; AVX512VBMI-NEXT: kmovd %ecx, %k1
|
|
; AVX512VBMI-NEXT: vpcompressq %zmm1, %zmm1 {%k1} {z}
|
|
; AVX512VBMI-NEXT: vpxor %xmm3, %xmm3, %xmm3
|
|
; AVX512VBMI-NEXT: valignq {{.*#+}} zmm4 = zmm1[1,2,3,4,5,6,7],zmm3[0]
|
|
; AVX512VBMI-NEXT: vpshrdvq %zmm2, %zmm4, %zmm1
|
|
; AVX512VBMI-NEXT: movl %esi, %ecx
|
|
; AVX512VBMI-NEXT: shrl $6, %ecx
|
|
; AVX512VBMI-NEXT: shlxl %ecx, %edx, %ecx
|
|
; AVX512VBMI-NEXT: kmovd %ecx, %k1
|
|
; AVX512VBMI-NEXT: vpexpandq %zmm0, %zmm2 {%k1} {z}
|
|
; AVX512VBMI-NEXT: valignq {{.*#+}} zmm3 = zmm3[7],zmm2[0,1,2,3,4,5,6]
|
|
; AVX512VBMI-NEXT: vpbroadcastq %rsi, %zmm4
|
|
; AVX512VBMI-NEXT: vpshldvq %zmm4, %zmm3, %zmm2
|
|
; AVX512VBMI-NEXT: xorl %ecx, %ecx
|
|
; AVX512VBMI-NEXT: negq %rsi
|
|
; AVX512VBMI-NEXT: sbbl %ecx, %ecx
|
|
; AVX512VBMI-NEXT: kmovd %ecx, %k1
|
|
; AVX512VBMI-NEXT: vporq %zmm1, %zmm2, %zmm0 {%k1}
|
|
; AVX512VBMI-NEXT: vmovdqu64 %zmm0, (%rdi)
|
|
; AVX512VBMI-NEXT: vzeroupper
|
|
; AVX512VBMI-NEXT: retq
|
|
%a0 = bitcast <8 x i64> %v0 to i512
|
|
%a1 = bitcast <8 x i64> %v1 to i512
|
|
%r = call i512 @llvm.fshl.i512(i512 %a0, i512 %a1, i512 %a2)
|
|
ret i512 %r
|
|
}
|
|
|
|
define i512 @fshr_i512_vector(<8 x i64> %v0, <8 x i64> %v1, i512 %a2) nounwind {
|
|
; SSE2-LABEL: fshr_i512_vector:
|
|
; SSE2: # %bb.0:
|
|
; SSE2-NEXT: pushq %rbp
|
|
; SSE2-NEXT: pushq %r15
|
|
; SSE2-NEXT: pushq %r14
|
|
; SSE2-NEXT: pushq %r13
|
|
; SSE2-NEXT: pushq %r12
|
|
; SSE2-NEXT: pushq %rbx
|
|
; SSE2-NEXT: movq %rsi, %rcx
|
|
; SSE2-NEXT: movq %xmm0, %rdx
|
|
; SSE2-NEXT: movq %xmm2, %r8
|
|
; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[2,3,2,3]
|
|
; SSE2-NEXT: movq %xmm0, %rsi
|
|
; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm2[2,3,2,3]
|
|
; SSE2-NEXT: movq %xmm0, %rax
|
|
; SSE2-NEXT: movq %xmm1, %r9
|
|
; SSE2-NEXT: movq %xmm3, %r13
|
|
; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm1[2,3,2,3]
|
|
; SSE2-NEXT: movq %xmm0, %r11
|
|
; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm3[2,3,2,3]
|
|
; SSE2-NEXT: movq %xmm0, %r12
|
|
; SSE2-NEXT: movq %xmm4, %r14
|
|
; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm4[2,3,2,3]
|
|
; SSE2-NEXT: movq %xmm0, %r10
|
|
; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm5[2,3,2,3]
|
|
; SSE2-NEXT: movq %xmm0, %rbx
|
|
; SSE2-NEXT: movq %xmm6, %r15
|
|
; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm7[2,3,2,3]
|
|
; SSE2-NEXT: movq %xmm0, %rbp
|
|
; SSE2-NEXT: testl $256, %ecx # imm = 0x100
|
|
; SSE2-NEXT: cmoveq %r11, %r12
|
|
; SSE2-NEXT: cmoveq %rbp, %r11
|
|
; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm6[2,3,2,3]
|
|
; SSE2-NEXT: cmovneq %rbp, %rbx
|
|
; SSE2-NEXT: movq %xmm0, %rbp
|
|
; SSE2-NEXT: cmoveq %rsi, %rax
|
|
; SSE2-NEXT: cmoveq %rdx, %r8
|
|
; SSE2-NEXT: cmoveq %rbp, %rsi
|
|
; SSE2-NEXT: cmoveq %r15, %rdx
|
|
; SSE2-NEXT: cmovneq %rbp, %r10
|
|
; SSE2-NEXT: cmovneq %r15, %r14
|
|
; SSE2-NEXT: movq %xmm5, %r15
|
|
; SSE2-NEXT: movq %xmm7, %rbp
|
|
; SSE2-NEXT: cmoveq %r9, %r13
|
|
; SSE2-NEXT: cmoveq %rbp, %r9
|
|
; SSE2-NEXT: cmovneq %rbp, %r15
|
|
; SSE2-NEXT: testb $-128, %cl
|
|
; SSE2-NEXT: cmovneq %r15, %r14
|
|
; SSE2-NEXT: cmovneq %rdx, %r15
|
|
; SSE2-NEXT: cmovneq %r9, %rdx
|
|
; SSE2-NEXT: cmovneq %r8, %r9
|
|
; SSE2-NEXT: cmovneq %r13, %r8
|
|
; SSE2-NEXT: cmovneq %rbx, %r10
|
|
; SSE2-NEXT: cmovneq %rsi, %rbx
|
|
; SSE2-NEXT: cmovneq %r11, %rsi
|
|
; SSE2-NEXT: cmovneq %rax, %r11
|
|
; SSE2-NEXT: cmovneq %r12, %rax
|
|
; SSE2-NEXT: testb $64, %cl
|
|
; SSE2-NEXT: cmoveq %r8, %rax
|
|
; SSE2-NEXT: cmoveq %r11, %r8
|
|
; SSE2-NEXT: cmoveq %r9, %r11
|
|
; SSE2-NEXT: cmoveq %rsi, %r9
|
|
; SSE2-NEXT: cmoveq %rdx, %rsi
|
|
; SSE2-NEXT: cmoveq %rbx, %rdx
|
|
; SSE2-NEXT: cmoveq %r15, %rbx
|
|
; SSE2-NEXT: cmoveq %r10, %r15
|
|
; SSE2-NEXT: cmoveq %r14, %r10
|
|
; SSE2-NEXT: shrdq %cl, %r15, %r10
|
|
; SSE2-NEXT: shrdq %cl, %rbx, %r15
|
|
; SSE2-NEXT: shrdq %cl, %rdx, %rbx
|
|
; SSE2-NEXT: shrdq %cl, %rsi, %rdx
|
|
; SSE2-NEXT: shrdq %cl, %r9, %rsi
|
|
; SSE2-NEXT: shrdq %cl, %r11, %r9
|
|
; SSE2-NEXT: shrdq %cl, %r8, %r11
|
|
; SSE2-NEXT: # kill: def $cl killed $cl killed $rcx
|
|
; SSE2-NEXT: shrdq %cl, %rax, %r8
|
|
; SSE2-NEXT: movq %rdi, %rax
|
|
; SSE2-NEXT: movq %r8, 56(%rdi)
|
|
; SSE2-NEXT: movq %r11, 48(%rdi)
|
|
; SSE2-NEXT: movq %r9, 40(%rdi)
|
|
; SSE2-NEXT: movq %rsi, 32(%rdi)
|
|
; SSE2-NEXT: movq %rdx, 24(%rdi)
|
|
; SSE2-NEXT: movq %rbx, 16(%rdi)
|
|
; SSE2-NEXT: movq %r15, 8(%rdi)
|
|
; SSE2-NEXT: movq %r10, (%rdi)
|
|
; SSE2-NEXT: popq %rbx
|
|
; SSE2-NEXT: popq %r12
|
|
; SSE2-NEXT: popq %r13
|
|
; SSE2-NEXT: popq %r14
|
|
; SSE2-NEXT: popq %r15
|
|
; SSE2-NEXT: popq %rbp
|
|
; SSE2-NEXT: retq
|
|
;
|
|
; SSE42-LABEL: fshr_i512_vector:
|
|
; SSE42: # %bb.0:
|
|
; SSE42-NEXT: pushq %rbp
|
|
; SSE42-NEXT: pushq %r15
|
|
; SSE42-NEXT: pushq %r14
|
|
; SSE42-NEXT: pushq %r13
|
|
; SSE42-NEXT: pushq %r12
|
|
; SSE42-NEXT: pushq %rbx
|
|
; SSE42-NEXT: movq %rsi, %rcx
|
|
; SSE42-NEXT: movq %xmm0, %rdx
|
|
; SSE42-NEXT: pextrq $1, %xmm0, %rsi
|
|
; SSE42-NEXT: pextrq $1, %xmm2, %rax
|
|
; SSE42-NEXT: movq %xmm2, %r9
|
|
; SSE42-NEXT: movq %xmm1, %r8
|
|
; SSE42-NEXT: pextrq $1, %xmm1, %r11
|
|
; SSE42-NEXT: movq %xmm3, %r13
|
|
; SSE42-NEXT: pextrq $1, %xmm3, %r12
|
|
; SSE42-NEXT: movq %xmm4, %r14
|
|
; SSE42-NEXT: pextrq $1, %xmm4, %r10
|
|
; SSE42-NEXT: pextrq $1, %xmm5, %rbx
|
|
; SSE42-NEXT: movq %xmm6, %r15
|
|
; SSE42-NEXT: pextrq $1, %xmm7, %rbp
|
|
; SSE42-NEXT: testl $256, %ecx # imm = 0x100
|
|
; SSE42-NEXT: cmoveq %r11, %r12
|
|
; SSE42-NEXT: cmoveq %rbp, %r11
|
|
; SSE42-NEXT: cmovneq %rbp, %rbx
|
|
; SSE42-NEXT: pextrq $1, %xmm6, %rbp
|
|
; SSE42-NEXT: cmoveq %rsi, %rax
|
|
; SSE42-NEXT: cmoveq %rdx, %r9
|
|
; SSE42-NEXT: cmoveq %rbp, %rsi
|
|
; SSE42-NEXT: cmoveq %r15, %rdx
|
|
; SSE42-NEXT: cmovneq %rbp, %r10
|
|
; SSE42-NEXT: cmovneq %r15, %r14
|
|
; SSE42-NEXT: movq %xmm5, %r15
|
|
; SSE42-NEXT: movq %xmm7, %rbp
|
|
; SSE42-NEXT: cmoveq %r8, %r13
|
|
; SSE42-NEXT: cmoveq %rbp, %r8
|
|
; SSE42-NEXT: cmovneq %rbp, %r15
|
|
; SSE42-NEXT: testb $-128, %cl
|
|
; SSE42-NEXT: cmovneq %r15, %r14
|
|
; SSE42-NEXT: cmovneq %rdx, %r15
|
|
; SSE42-NEXT: cmovneq %r8, %rdx
|
|
; SSE42-NEXT: cmovneq %r9, %r8
|
|
; SSE42-NEXT: cmovneq %r13, %r9
|
|
; SSE42-NEXT: cmovneq %rbx, %r10
|
|
; SSE42-NEXT: cmovneq %rsi, %rbx
|
|
; SSE42-NEXT: cmovneq %r11, %rsi
|
|
; SSE42-NEXT: cmovneq %rax, %r11
|
|
; SSE42-NEXT: cmovneq %r12, %rax
|
|
; SSE42-NEXT: testb $64, %cl
|
|
; SSE42-NEXT: cmoveq %r9, %rax
|
|
; SSE42-NEXT: cmoveq %r11, %r9
|
|
; SSE42-NEXT: cmoveq %r8, %r11
|
|
; SSE42-NEXT: cmoveq %rsi, %r8
|
|
; SSE42-NEXT: cmoveq %rdx, %rsi
|
|
; SSE42-NEXT: cmoveq %rbx, %rdx
|
|
; SSE42-NEXT: cmoveq %r15, %rbx
|
|
; SSE42-NEXT: cmoveq %r10, %r15
|
|
; SSE42-NEXT: cmoveq %r14, %r10
|
|
; SSE42-NEXT: shrdq %cl, %r15, %r10
|
|
; SSE42-NEXT: shrdq %cl, %rbx, %r15
|
|
; SSE42-NEXT: shrdq %cl, %rdx, %rbx
|
|
; SSE42-NEXT: shrdq %cl, %rsi, %rdx
|
|
; SSE42-NEXT: shrdq %cl, %r8, %rsi
|
|
; SSE42-NEXT: shrdq %cl, %r11, %r8
|
|
; SSE42-NEXT: shrdq %cl, %r9, %r11
|
|
; SSE42-NEXT: # kill: def $cl killed $cl killed $rcx
|
|
; SSE42-NEXT: shrdq %cl, %rax, %r9
|
|
; SSE42-NEXT: movq %rdi, %rax
|
|
; SSE42-NEXT: movq %r9, 56(%rdi)
|
|
; SSE42-NEXT: movq %r11, 48(%rdi)
|
|
; SSE42-NEXT: movq %r8, 40(%rdi)
|
|
; SSE42-NEXT: movq %rsi, 32(%rdi)
|
|
; SSE42-NEXT: movq %rdx, 24(%rdi)
|
|
; SSE42-NEXT: movq %rbx, 16(%rdi)
|
|
; SSE42-NEXT: movq %r15, 8(%rdi)
|
|
; SSE42-NEXT: movq %r10, (%rdi)
|
|
; SSE42-NEXT: popq %rbx
|
|
; SSE42-NEXT: popq %r12
|
|
; SSE42-NEXT: popq %r13
|
|
; SSE42-NEXT: popq %r14
|
|
; SSE42-NEXT: popq %r15
|
|
; SSE42-NEXT: popq %rbp
|
|
; SSE42-NEXT: retq
|
|
;
|
|
; AVX2-LABEL: fshr_i512_vector:
|
|
; AVX2: # %bb.0:
|
|
; AVX2-NEXT: pushq %rbp
|
|
; AVX2-NEXT: pushq %r15
|
|
; AVX2-NEXT: pushq %r14
|
|
; AVX2-NEXT: pushq %r13
|
|
; AVX2-NEXT: pushq %r12
|
|
; AVX2-NEXT: pushq %rbx
|
|
; AVX2-NEXT: movq %rsi, %rcx
|
|
; AVX2-NEXT: vmovq %xmm0, %rdx
|
|
; AVX2-NEXT: vmovq %xmm1, %r8
|
|
; AVX2-NEXT: vpextrq $1, %xmm0, %rsi
|
|
; AVX2-NEXT: vpextrq $1, %xmm1, %r10
|
|
; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm0
|
|
; AVX2-NEXT: vmovq %xmm0, %r9
|
|
; AVX2-NEXT: vextracti128 $1, %ymm1, %xmm1
|
|
; AVX2-NEXT: vmovq %xmm1, %r13
|
|
; AVX2-NEXT: vpextrq $1, %xmm0, %rbx
|
|
; AVX2-NEXT: vpextrq $1, %xmm1, %rax
|
|
; AVX2-NEXT: vextracti128 $1, %ymm2, %xmm0
|
|
; AVX2-NEXT: vmovq %xmm2, %r15
|
|
; AVX2-NEXT: vpextrq $1, %xmm2, %r11
|
|
; AVX2-NEXT: vpextrq $1, %xmm0, %r14
|
|
; AVX2-NEXT: vpextrq $1, %xmm3, %r12
|
|
; AVX2-NEXT: vextracti128 $1, %ymm3, %xmm1
|
|
; AVX2-NEXT: vpextrq $1, %xmm1, %rbp
|
|
; AVX2-NEXT: testl $256, %ecx # imm = 0x100
|
|
; AVX2-NEXT: cmoveq %rbx, %rax
|
|
; AVX2-NEXT: cmoveq %rsi, %r10
|
|
; AVX2-NEXT: cmoveq %rbp, %rbx
|
|
; AVX2-NEXT: cmoveq %r12, %rsi
|
|
; AVX2-NEXT: cmovneq %rbp, %r14
|
|
; AVX2-NEXT: cmovneq %r12, %r11
|
|
; AVX2-NEXT: vmovq %xmm3, %r12
|
|
; AVX2-NEXT: vmovq %xmm1, %rbp
|
|
; AVX2-NEXT: cmoveq %r9, %r13
|
|
; AVX2-NEXT: cmoveq %rdx, %r8
|
|
; AVX2-NEXT: cmoveq %rbp, %r9
|
|
; AVX2-NEXT: cmoveq %r12, %rdx
|
|
; AVX2-NEXT: cmovneq %r12, %r15
|
|
; AVX2-NEXT: vmovq %xmm0, %r12
|
|
; AVX2-NEXT: cmovneq %rbp, %r12
|
|
; AVX2-NEXT: testb $-128, %cl
|
|
; AVX2-NEXT: cmovneq %r12, %r15
|
|
; AVX2-NEXT: cmovneq %rdx, %r12
|
|
; AVX2-NEXT: cmovneq %r9, %rdx
|
|
; AVX2-NEXT: cmovneq %r8, %r9
|
|
; AVX2-NEXT: cmovneq %r13, %r8
|
|
; AVX2-NEXT: cmovneq %r14, %r11
|
|
; AVX2-NEXT: cmovneq %rsi, %r14
|
|
; AVX2-NEXT: cmovneq %rbx, %rsi
|
|
; AVX2-NEXT: cmovneq %r10, %rbx
|
|
; AVX2-NEXT: cmovneq %rax, %r10
|
|
; AVX2-NEXT: movq %rdi, %rax
|
|
; AVX2-NEXT: testb $64, %cl
|
|
; AVX2-NEXT: cmoveq %r8, %r10
|
|
; AVX2-NEXT: cmoveq %rbx, %r8
|
|
; AVX2-NEXT: cmoveq %r9, %rbx
|
|
; AVX2-NEXT: cmoveq %rsi, %r9
|
|
; AVX2-NEXT: cmoveq %rdx, %rsi
|
|
; AVX2-NEXT: cmoveq %r14, %rdx
|
|
; AVX2-NEXT: cmoveq %r12, %r14
|
|
; AVX2-NEXT: cmoveq %r11, %r12
|
|
; AVX2-NEXT: cmoveq %r15, %r11
|
|
; AVX2-NEXT: shrdq %cl, %r12, %r11
|
|
; AVX2-NEXT: shrdq %cl, %r14, %r12
|
|
; AVX2-NEXT: shrdq %cl, %rdx, %r14
|
|
; AVX2-NEXT: shrdq %cl, %rsi, %rdx
|
|
; AVX2-NEXT: shrdq %cl, %r9, %rsi
|
|
; AVX2-NEXT: shrdq %cl, %rbx, %r9
|
|
; AVX2-NEXT: shrdq %cl, %r8, %rbx
|
|
; AVX2-NEXT: # kill: def $cl killed $cl killed $rcx
|
|
; AVX2-NEXT: shrdq %cl, %r10, %r8
|
|
; AVX2-NEXT: movq %r8, 56(%rdi)
|
|
; AVX2-NEXT: movq %rbx, 48(%rdi)
|
|
; AVX2-NEXT: movq %r9, 40(%rdi)
|
|
; AVX2-NEXT: movq %rsi, 32(%rdi)
|
|
; AVX2-NEXT: movq %rdx, 24(%rdi)
|
|
; AVX2-NEXT: movq %r14, 16(%rdi)
|
|
; AVX2-NEXT: movq %r12, 8(%rdi)
|
|
; AVX2-NEXT: movq %r11, (%rdi)
|
|
; AVX2-NEXT: popq %rbx
|
|
; AVX2-NEXT: popq %r12
|
|
; AVX2-NEXT: popq %r13
|
|
; AVX2-NEXT: popq %r14
|
|
; AVX2-NEXT: popq %r15
|
|
; AVX2-NEXT: popq %rbp
|
|
; AVX2-NEXT: vzeroupper
|
|
; AVX2-NEXT: retq
|
|
;
|
|
; AVX512F-LABEL: fshr_i512_vector:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: movq %rdi, %rax
|
|
; AVX512F-NEXT: andl $511, %esi # imm = 0x1FF
|
|
; AVX512F-NEXT: vmovq %rsi, %xmm2
|
|
; AVX512F-NEXT: vpbroadcastq %xmm2, %xmm3
|
|
; AVX512F-NEXT: vpbroadcastq {{.*#+}} xmm2 = [63,63]
|
|
; AVX512F-NEXT: vpand %xmm2, %xmm3, %xmm4
|
|
; AVX512F-NEXT: movl %esi, %ecx
|
|
; AVX512F-NEXT: shrl $6, %ecx
|
|
; AVX512F-NEXT: movl $-1, %edx
|
|
; AVX512F-NEXT: shlxl %ecx, %edx, %ecx
|
|
; AVX512F-NEXT: kmovw %ecx, %k1
|
|
; AVX512F-NEXT: vpcompressq %zmm1, %zmm5 {%k1} {z}
|
|
; AVX512F-NEXT: vpsrlq %xmm4, %zmm5, %zmm4
|
|
; AVX512F-NEXT: vpandn %xmm2, %xmm3, %xmm3
|
|
; AVX512F-NEXT: vpxor %xmm6, %xmm6, %xmm6
|
|
; AVX512F-NEXT: valignq {{.*#+}} zmm5 = zmm5[1,2,3,4,5,6,7],zmm6[0]
|
|
; AVX512F-NEXT: vpaddq %zmm5, %zmm5, %zmm5
|
|
; AVX512F-NEXT: vpsllq %xmm3, %zmm5, %zmm3
|
|
; AVX512F-NEXT: vporq %zmm4, %zmm3, %zmm3
|
|
; AVX512F-NEXT: movl $512, %ecx # imm = 0x200
|
|
; AVX512F-NEXT: subq %rsi, %rcx
|
|
; AVX512F-NEXT: vmovq %rcx, %xmm4
|
|
; AVX512F-NEXT: vpbroadcastq %xmm4, %xmm4
|
|
; AVX512F-NEXT: vpand %xmm2, %xmm4, %xmm5
|
|
; AVX512F-NEXT: shrl $6, %ecx
|
|
; AVX512F-NEXT: shlxl %ecx, %edx, %ecx
|
|
; AVX512F-NEXT: kmovw %ecx, %k1
|
|
; AVX512F-NEXT: vpexpandq %zmm0, %zmm0 {%k1} {z}
|
|
; AVX512F-NEXT: vpsllq %xmm5, %zmm0, %zmm5
|
|
; AVX512F-NEXT: vpandn %xmm2, %xmm4, %xmm2
|
|
; AVX512F-NEXT: valignq {{.*#+}} zmm0 = zmm6[7],zmm0[0,1,2,3,4,5,6]
|
|
; AVX512F-NEXT: vpsrlq $1, %zmm0, %zmm0
|
|
; AVX512F-NEXT: vpsrlq %xmm2, %zmm0, %zmm0
|
|
; AVX512F-NEXT: xorl %ecx, %ecx
|
|
; AVX512F-NEXT: negq %rsi
|
|
; AVX512F-NEXT: sbbl %ecx, %ecx
|
|
; AVX512F-NEXT: vporq %zmm0, %zmm5, %zmm0
|
|
; AVX512F-NEXT: kmovw %ecx, %k1
|
|
; AVX512F-NEXT: vporq %zmm3, %zmm0, %zmm1 {%k1}
|
|
; AVX512F-NEXT: vmovdqu64 %zmm1, (%rdi)
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: fshr_i512_vector:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: andl $511, %esi # imm = 0x1FF
|
|
; AVX512VL-NEXT: vpbroadcastq %rsi, %xmm2
|
|
; AVX512VL-NEXT: vpbroadcastq {{.*#+}} xmm3 = [63,63]
|
|
; AVX512VL-NEXT: vpand %xmm3, %xmm2, %xmm4
|
|
; AVX512VL-NEXT: movl %esi, %eax
|
|
; AVX512VL-NEXT: shrl $6, %eax
|
|
; AVX512VL-NEXT: movl $-1, %ecx
|
|
; AVX512VL-NEXT: shlxl %eax, %ecx, %eax
|
|
; AVX512VL-NEXT: kmovd %eax, %k1
|
|
; AVX512VL-NEXT: vpcompressq %zmm1, %zmm5 {%k1} {z}
|
|
; AVX512VL-NEXT: vpsrlq %xmm4, %zmm5, %zmm4
|
|
; AVX512VL-NEXT: vpandn %xmm3, %xmm2, %xmm2
|
|
; AVX512VL-NEXT: vpxor %xmm6, %xmm6, %xmm6
|
|
; AVX512VL-NEXT: valignq {{.*#+}} zmm5 = zmm5[1,2,3,4,5,6,7],zmm6[0]
|
|
; AVX512VL-NEXT: vpaddq %zmm5, %zmm5, %zmm5
|
|
; AVX512VL-NEXT: vpsllq %xmm2, %zmm5, %zmm2
|
|
; AVX512VL-NEXT: movq %rdi, %rax
|
|
; AVX512VL-NEXT: vporq %zmm4, %zmm2, %zmm2
|
|
; AVX512VL-NEXT: movl $512, %edx # imm = 0x200
|
|
; AVX512VL-NEXT: subq %rsi, %rdx
|
|
; AVX512VL-NEXT: vpbroadcastq %rdx, %xmm4
|
|
; AVX512VL-NEXT: vpand %xmm3, %xmm4, %xmm5
|
|
; AVX512VL-NEXT: shrl $6, %edx
|
|
; AVX512VL-NEXT: shlxl %edx, %ecx, %ecx
|
|
; AVX512VL-NEXT: kmovd %ecx, %k1
|
|
; AVX512VL-NEXT: vpexpandq %zmm0, %zmm0 {%k1} {z}
|
|
; AVX512VL-NEXT: vpsllq %xmm5, %zmm0, %zmm5
|
|
; AVX512VL-NEXT: vpandn %xmm3, %xmm4, %xmm3
|
|
; AVX512VL-NEXT: valignq {{.*#+}} zmm0 = zmm6[7],zmm0[0,1,2,3,4,5,6]
|
|
; AVX512VL-NEXT: vpsrlq $1, %zmm0, %zmm0
|
|
; AVX512VL-NEXT: vpsrlq %xmm3, %zmm0, %zmm0
|
|
; AVX512VL-NEXT: vporq %zmm0, %zmm5, %zmm0
|
|
; AVX512VL-NEXT: xorl %ecx, %ecx
|
|
; AVX512VL-NEXT: negq %rsi
|
|
; AVX512VL-NEXT: sbbl %ecx, %ecx
|
|
; AVX512VL-NEXT: kmovd %ecx, %k1
|
|
; AVX512VL-NEXT: vporq %zmm2, %zmm0, %zmm1 {%k1}
|
|
; AVX512VL-NEXT: vmovdqu64 %zmm1, (%rdi)
|
|
; AVX512VL-NEXT: vzeroupper
|
|
; AVX512VL-NEXT: retq
|
|
;
|
|
; AVX512VBMI-LABEL: fshr_i512_vector:
|
|
; AVX512VBMI: # %bb.0:
|
|
; AVX512VBMI-NEXT: movq %rdi, %rax
|
|
; AVX512VBMI-NEXT: andl $511, %esi # imm = 0x1FF
|
|
; AVX512VBMI-NEXT: movl $512, %ecx # imm = 0x200
|
|
; AVX512VBMI-NEXT: subq %rsi, %rcx
|
|
; AVX512VBMI-NEXT: vpbroadcastq %rcx, %zmm2
|
|
; AVX512VBMI-NEXT: # kill: def $ecx killed $ecx killed $rcx
|
|
; AVX512VBMI-NEXT: shrl $6, %ecx
|
|
; AVX512VBMI-NEXT: movl $-1, %edx
|
|
; AVX512VBMI-NEXT: shlxl %ecx, %edx, %ecx
|
|
; AVX512VBMI-NEXT: kmovd %ecx, %k1
|
|
; AVX512VBMI-NEXT: vpexpandq %zmm0, %zmm0 {%k1} {z}
|
|
; AVX512VBMI-NEXT: vpxor %xmm3, %xmm3, %xmm3
|
|
; AVX512VBMI-NEXT: valignq {{.*#+}} zmm4 = zmm3[7],zmm0[0,1,2,3,4,5,6]
|
|
; AVX512VBMI-NEXT: vpshldvq %zmm2, %zmm4, %zmm0
|
|
; AVX512VBMI-NEXT: movl %esi, %ecx
|
|
; AVX512VBMI-NEXT: shrl $6, %ecx
|
|
; AVX512VBMI-NEXT: shlxl %ecx, %edx, %ecx
|
|
; AVX512VBMI-NEXT: kmovd %ecx, %k1
|
|
; AVX512VBMI-NEXT: vpcompressq %zmm1, %zmm2 {%k1} {z}
|
|
; AVX512VBMI-NEXT: valignq {{.*#+}} zmm3 = zmm2[1,2,3,4,5,6,7],zmm3[0]
|
|
; AVX512VBMI-NEXT: vpbroadcastq %rsi, %zmm4
|
|
; AVX512VBMI-NEXT: vpshrdvq %zmm4, %zmm3, %zmm2
|
|
; AVX512VBMI-NEXT: xorl %ecx, %ecx
|
|
; AVX512VBMI-NEXT: negq %rsi
|
|
; AVX512VBMI-NEXT: sbbl %ecx, %ecx
|
|
; AVX512VBMI-NEXT: kmovd %ecx, %k1
|
|
; AVX512VBMI-NEXT: vporq %zmm2, %zmm0, %zmm1 {%k1}
|
|
; AVX512VBMI-NEXT: vmovdqu64 %zmm1, (%rdi)
|
|
; AVX512VBMI-NEXT: vzeroupper
|
|
; AVX512VBMI-NEXT: retq
|
|
%a0 = bitcast <8 x i64> %v0 to i512
|
|
%a1 = bitcast <8 x i64> %v1 to i512
|
|
%r = call i512 @llvm.fshr.i512(i512 %a0, i512 %a1, i512 %a2)
|
|
ret i512 %r
|
|
}
|
|
|
|
define i512 @fshl_rot_i512_vector(<8 x i64> %v0, i512 %a2) nounwind {
|
|
; SSE2-LABEL: fshl_rot_i512_vector:
|
|
; SSE2: # %bb.0:
|
|
; SSE2-NEXT: pushq %r15
|
|
; SSE2-NEXT: pushq %r14
|
|
; SSE2-NEXT: pushq %r13
|
|
; SSE2-NEXT: pushq %r12
|
|
; SSE2-NEXT: pushq %rbx
|
|
; SSE2-NEXT: movq %rsi, %rcx
|
|
; SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm2[2,3,2,3]
|
|
; SSE2-NEXT: movq %xmm4, %rax
|
|
; SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm0[2,3,2,3]
|
|
; SSE2-NEXT: movq %xmm3, %rdx
|
|
; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm3[2,3,2,3]
|
|
; SSE2-NEXT: movq %xmm3, %r10
|
|
; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm1[2,3,2,3]
|
|
; SSE2-NEXT: movq %xmm0, %r8
|
|
; SSE2-NEXT: movq %xmm2, %rbx
|
|
; SSE2-NEXT: testl $256, %ecx # imm = 0x100
|
|
; SSE2-NEXT: movq %rbx, %r11
|
|
; SSE2-NEXT: cmovneq %r8, %r11
|
|
; SSE2-NEXT: movq %xmm3, %r14
|
|
; SSE2-NEXT: movq %r14, %r15
|
|
; SSE2-NEXT: cmovneq %r10, %r15
|
|
; SSE2-NEXT: movq %xmm1, %r12
|
|
; SSE2-NEXT: movq %r12, %rsi
|
|
; SSE2-NEXT: cmovneq %rdx, %rsi
|
|
; SSE2-NEXT: movq %xmm4, %r13
|
|
; SSE2-NEXT: movq %r13, %r9
|
|
; SSE2-NEXT: cmovneq %rax, %r9
|
|
; SSE2-NEXT: cmovneq %rbx, %r8
|
|
; SSE2-NEXT: cmovneq %r12, %rdx
|
|
; SSE2-NEXT: cmovneq %r13, %rax
|
|
; SSE2-NEXT: cmovneq %r14, %r10
|
|
; SSE2-NEXT: testb $-128, %cl
|
|
; SSE2-NEXT: movq %r9, %r14
|
|
; SSE2-NEXT: cmovneq %r10, %r14
|
|
; SSE2-NEXT: cmovneq %rax, %r10
|
|
; SSE2-NEXT: movq %rsi, %rbx
|
|
; SSE2-NEXT: cmovneq %r8, %rbx
|
|
; SSE2-NEXT: cmovneq %rdx, %r8
|
|
; SSE2-NEXT: cmoveq %r15, %r9
|
|
; SSE2-NEXT: cmoveq %r11, %rsi
|
|
; SSE2-NEXT: cmovneq %r15, %rax
|
|
; SSE2-NEXT: cmovneq %r11, %rdx
|
|
; SSE2-NEXT: testb $64, %cl
|
|
; SSE2-NEXT: movq %rdx, %r11
|
|
; SSE2-NEXT: cmovneq %rax, %r11
|
|
; SSE2-NEXT: cmovneq %rsi, %rax
|
|
; SSE2-NEXT: cmovneq %r9, %rsi
|
|
; SSE2-NEXT: cmovneq %rbx, %r9
|
|
; SSE2-NEXT: cmovneq %r14, %rbx
|
|
; SSE2-NEXT: cmovneq %r8, %r14
|
|
; SSE2-NEXT: cmoveq %r10, %rdx
|
|
; SSE2-NEXT: cmovneq %r10, %r8
|
|
; SSE2-NEXT: movq %r8, %r10
|
|
; SSE2-NEXT: shldq %cl, %rdx, %r10
|
|
; SSE2-NEXT: movq %r14, %r15
|
|
; SSE2-NEXT: shldq %cl, %r8, %r15
|
|
; SSE2-NEXT: movq %rbx, %r8
|
|
; SSE2-NEXT: shldq %cl, %r14, %r8
|
|
; SSE2-NEXT: movq %r9, %r14
|
|
; SSE2-NEXT: shldq %cl, %rbx, %r14
|
|
; SSE2-NEXT: movq %rsi, %rbx
|
|
; SSE2-NEXT: shldq %cl, %r9, %rbx
|
|
; SSE2-NEXT: movq %rax, %r9
|
|
; SSE2-NEXT: shldq %cl, %rsi, %r9
|
|
; SSE2-NEXT: movq %r11, %rsi
|
|
; SSE2-NEXT: shldq %cl, %rax, %rsi
|
|
; SSE2-NEXT: # kill: def $cl killed $cl killed $rcx
|
|
; SSE2-NEXT: shldq %cl, %r11, %rdx
|
|
; SSE2-NEXT: movq %rdi, %rax
|
|
; SSE2-NEXT: movq %rdx, 56(%rdi)
|
|
; SSE2-NEXT: movq %rsi, 48(%rdi)
|
|
; SSE2-NEXT: movq %r9, 40(%rdi)
|
|
; SSE2-NEXT: movq %rbx, 32(%rdi)
|
|
; SSE2-NEXT: movq %r14, 24(%rdi)
|
|
; SSE2-NEXT: movq %r8, 16(%rdi)
|
|
; SSE2-NEXT: movq %r15, 8(%rdi)
|
|
; SSE2-NEXT: movq %r10, (%rdi)
|
|
; SSE2-NEXT: popq %rbx
|
|
; SSE2-NEXT: popq %r12
|
|
; SSE2-NEXT: popq %r13
|
|
; SSE2-NEXT: popq %r14
|
|
; SSE2-NEXT: popq %r15
|
|
; SSE2-NEXT: retq
|
|
;
|
|
; SSE42-LABEL: fshl_rot_i512_vector:
|
|
; SSE42: # %bb.0:
|
|
; SSE42-NEXT: pushq %r15
|
|
; SSE42-NEXT: pushq %r14
|
|
; SSE42-NEXT: pushq %r13
|
|
; SSE42-NEXT: pushq %r12
|
|
; SSE42-NEXT: pushq %rbx
|
|
; SSE42-NEXT: movq %rsi, %rcx
|
|
; SSE42-NEXT: pextrq $1, %xmm2, %rax
|
|
; SSE42-NEXT: pextrq $1, %xmm0, %rbx
|
|
; SSE42-NEXT: pextrq $1, %xmm3, %r10
|
|
; SSE42-NEXT: pextrq $1, %xmm1, %r14
|
|
; SSE42-NEXT: movq %xmm0, %r8
|
|
; SSE42-NEXT: movq %xmm2, %r15
|
|
; SSE42-NEXT: testl $256, %ecx # imm = 0x100
|
|
; SSE42-NEXT: movq %r15, %r11
|
|
; SSE42-NEXT: cmovneq %r8, %r11
|
|
; SSE42-NEXT: movq %xmm3, %rdx
|
|
; SSE42-NEXT: movq %r14, %r12
|
|
; SSE42-NEXT: cmovneq %r10, %r12
|
|
; SSE42-NEXT: movq %xmm1, %r13
|
|
; SSE42-NEXT: movq %r13, %rsi
|
|
; SSE42-NEXT: cmovneq %rdx, %rsi
|
|
; SSE42-NEXT: movq %rbx, %r9
|
|
; SSE42-NEXT: cmovneq %rax, %r9
|
|
; SSE42-NEXT: cmovneq %r15, %r8
|
|
; SSE42-NEXT: cmovneq %r13, %rdx
|
|
; SSE42-NEXT: cmovneq %rbx, %rax
|
|
; SSE42-NEXT: cmovneq %r14, %r10
|
|
; SSE42-NEXT: testb $-128, %cl
|
|
; SSE42-NEXT: movq %r9, %r14
|
|
; SSE42-NEXT: cmovneq %r10, %r14
|
|
; SSE42-NEXT: cmovneq %rax, %r10
|
|
; SSE42-NEXT: movq %rsi, %rbx
|
|
; SSE42-NEXT: cmovneq %r8, %rbx
|
|
; SSE42-NEXT: cmovneq %rdx, %r8
|
|
; SSE42-NEXT: cmoveq %r12, %r9
|
|
; SSE42-NEXT: cmoveq %r11, %rsi
|
|
; SSE42-NEXT: cmovneq %r12, %rax
|
|
; SSE42-NEXT: cmovneq %r11, %rdx
|
|
; SSE42-NEXT: testb $64, %cl
|
|
; SSE42-NEXT: movq %rdx, %r11
|
|
; SSE42-NEXT: cmovneq %rax, %r11
|
|
; SSE42-NEXT: cmovneq %rsi, %rax
|
|
; SSE42-NEXT: cmovneq %r9, %rsi
|
|
; SSE42-NEXT: cmovneq %rbx, %r9
|
|
; SSE42-NEXT: cmovneq %r14, %rbx
|
|
; SSE42-NEXT: cmovneq %r8, %r14
|
|
; SSE42-NEXT: cmoveq %r10, %rdx
|
|
; SSE42-NEXT: cmovneq %r10, %r8
|
|
; SSE42-NEXT: movq %r8, %r10
|
|
; SSE42-NEXT: shldq %cl, %rdx, %r10
|
|
; SSE42-NEXT: movq %r14, %r15
|
|
; SSE42-NEXT: shldq %cl, %r8, %r15
|
|
; SSE42-NEXT: movq %rbx, %r8
|
|
; SSE42-NEXT: shldq %cl, %r14, %r8
|
|
; SSE42-NEXT: movq %r9, %r14
|
|
; SSE42-NEXT: shldq %cl, %rbx, %r14
|
|
; SSE42-NEXT: movq %rsi, %rbx
|
|
; SSE42-NEXT: shldq %cl, %r9, %rbx
|
|
; SSE42-NEXT: movq %rax, %r9
|
|
; SSE42-NEXT: shldq %cl, %rsi, %r9
|
|
; SSE42-NEXT: movq %r11, %rsi
|
|
; SSE42-NEXT: shldq %cl, %rax, %rsi
|
|
; SSE42-NEXT: # kill: def $cl killed $cl killed $rcx
|
|
; SSE42-NEXT: shldq %cl, %r11, %rdx
|
|
; SSE42-NEXT: movq %rdi, %rax
|
|
; SSE42-NEXT: movq %rdx, 56(%rdi)
|
|
; SSE42-NEXT: movq %rsi, 48(%rdi)
|
|
; SSE42-NEXT: movq %r9, 40(%rdi)
|
|
; SSE42-NEXT: movq %rbx, 32(%rdi)
|
|
; SSE42-NEXT: movq %r14, 24(%rdi)
|
|
; SSE42-NEXT: movq %r8, 16(%rdi)
|
|
; SSE42-NEXT: movq %r15, 8(%rdi)
|
|
; SSE42-NEXT: movq %r10, (%rdi)
|
|
; SSE42-NEXT: popq %rbx
|
|
; SSE42-NEXT: popq %r12
|
|
; SSE42-NEXT: popq %r13
|
|
; SSE42-NEXT: popq %r14
|
|
; SSE42-NEXT: popq %r15
|
|
; SSE42-NEXT: retq
|
|
;
|
|
; AVX2-LABEL: fshl_rot_i512_vector:
|
|
; AVX2: # %bb.0:
|
|
; AVX2-NEXT: pushq %r15
|
|
; AVX2-NEXT: pushq %r14
|
|
; AVX2-NEXT: pushq %r13
|
|
; AVX2-NEXT: pushq %r12
|
|
; AVX2-NEXT: pushq %rbx
|
|
; AVX2-NEXT: vpextrq $1, %xmm1, %rdx
|
|
; AVX2-NEXT: vpextrq $1, %xmm0, %rax
|
|
; AVX2-NEXT: movq %rsi, %rcx
|
|
; AVX2-NEXT: vextracti128 $1, %ymm1, %xmm2
|
|
; AVX2-NEXT: vmovq %xmm2, %rsi
|
|
; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm3
|
|
; AVX2-NEXT: vpextrq $1, %xmm2, %r11
|
|
; AVX2-NEXT: vpextrq $1, %xmm3, %rbx
|
|
; AVX2-NEXT: vmovq %xmm3, %r14
|
|
; AVX2-NEXT: vmovq %xmm0, %r10
|
|
; AVX2-NEXT: vmovq %xmm1, %r12
|
|
; AVX2-NEXT: testl $256, %ecx # imm = 0x100
|
|
; AVX2-NEXT: movq %r12, %r15
|
|
; AVX2-NEXT: cmovneq %r10, %r15
|
|
; AVX2-NEXT: movq %rbx, %r13
|
|
; AVX2-NEXT: cmovneq %r11, %r13
|
|
; AVX2-NEXT: movq %r14, %r8
|
|
; AVX2-NEXT: cmovneq %rsi, %r8
|
|
; AVX2-NEXT: movq %rax, %r9
|
|
; AVX2-NEXT: cmovneq %rdx, %r9
|
|
; AVX2-NEXT: cmovneq %r12, %r10
|
|
; AVX2-NEXT: cmovneq %r14, %rsi
|
|
; AVX2-NEXT: cmovneq %rax, %rdx
|
|
; AVX2-NEXT: cmovneq %rbx, %r11
|
|
; AVX2-NEXT: testb $-128, %cl
|
|
; AVX2-NEXT: movq %r9, %r14
|
|
; AVX2-NEXT: cmovneq %r11, %r14
|
|
; AVX2-NEXT: cmovneq %rdx, %r11
|
|
; AVX2-NEXT: movq %r8, %rbx
|
|
; AVX2-NEXT: cmovneq %r10, %rbx
|
|
; AVX2-NEXT: cmovneq %rsi, %r10
|
|
; AVX2-NEXT: cmoveq %r13, %r9
|
|
; AVX2-NEXT: cmoveq %r15, %r8
|
|
; AVX2-NEXT: cmovneq %r13, %rdx
|
|
; AVX2-NEXT: movq %rdi, %rax
|
|
; AVX2-NEXT: cmovneq %r15, %rsi
|
|
; AVX2-NEXT: testb $64, %cl
|
|
; AVX2-NEXT: movq %rsi, %rdi
|
|
; AVX2-NEXT: cmovneq %rdx, %rdi
|
|
; AVX2-NEXT: cmovneq %r8, %rdx
|
|
; AVX2-NEXT: cmovneq %r9, %r8
|
|
; AVX2-NEXT: cmovneq %rbx, %r9
|
|
; AVX2-NEXT: cmovneq %r14, %rbx
|
|
; AVX2-NEXT: cmovneq %r10, %r14
|
|
; AVX2-NEXT: cmoveq %r11, %rsi
|
|
; AVX2-NEXT: cmovneq %r11, %r10
|
|
; AVX2-NEXT: movq %r10, %r11
|
|
; AVX2-NEXT: shldq %cl, %rsi, %r11
|
|
; AVX2-NEXT: movq %r14, %r15
|
|
; AVX2-NEXT: shldq %cl, %r10, %r15
|
|
; AVX2-NEXT: movq %rbx, %r10
|
|
; AVX2-NEXT: shldq %cl, %r14, %r10
|
|
; AVX2-NEXT: movq %r9, %r14
|
|
; AVX2-NEXT: shldq %cl, %rbx, %r14
|
|
; AVX2-NEXT: movq %r8, %rbx
|
|
; AVX2-NEXT: shldq %cl, %r9, %rbx
|
|
; AVX2-NEXT: movq %rdx, %r9
|
|
; AVX2-NEXT: shldq %cl, %r8, %r9
|
|
; AVX2-NEXT: movq %rdi, %r8
|
|
; AVX2-NEXT: shldq %cl, %rdx, %r8
|
|
; AVX2-NEXT: # kill: def $cl killed $cl killed $rcx
|
|
; AVX2-NEXT: shldq %cl, %rdi, %rsi
|
|
; AVX2-NEXT: movq %rsi, 56(%rax)
|
|
; AVX2-NEXT: movq %r8, 48(%rax)
|
|
; AVX2-NEXT: movq %r9, 40(%rax)
|
|
; AVX2-NEXT: movq %rbx, 32(%rax)
|
|
; AVX2-NEXT: movq %r14, 24(%rax)
|
|
; AVX2-NEXT: movq %r10, 16(%rax)
|
|
; AVX2-NEXT: movq %r15, 8(%rax)
|
|
; AVX2-NEXT: movq %r11, (%rax)
|
|
; AVX2-NEXT: popq %rbx
|
|
; AVX2-NEXT: popq %r12
|
|
; AVX2-NEXT: popq %r13
|
|
; AVX2-NEXT: popq %r14
|
|
; AVX2-NEXT: popq %r15
|
|
; AVX2-NEXT: vzeroupper
|
|
; AVX2-NEXT: retq
|
|
;
|
|
; AVX512F-LABEL: fshl_rot_i512_vector:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: movq %rdi, %rax
|
|
; AVX512F-NEXT: andl $511, %esi # imm = 0x1FF
|
|
; AVX512F-NEXT: vmovq %rsi, %xmm1
|
|
; AVX512F-NEXT: vpbroadcastq %xmm1, %xmm2
|
|
; AVX512F-NEXT: vpbroadcastq {{.*#+}} xmm1 = [63,63]
|
|
; AVX512F-NEXT: vpand %xmm1, %xmm2, %xmm3
|
|
; AVX512F-NEXT: movl %esi, %ecx
|
|
; AVX512F-NEXT: shrl $6, %ecx
|
|
; AVX512F-NEXT: movl $-1, %edx
|
|
; AVX512F-NEXT: shlxl %ecx, %edx, %ecx
|
|
; AVX512F-NEXT: kmovw %ecx, %k1
|
|
; AVX512F-NEXT: vpexpandq %zmm0, %zmm4 {%k1} {z}
|
|
; AVX512F-NEXT: vpsllq %xmm3, %zmm4, %zmm3
|
|
; AVX512F-NEXT: vpandn %xmm1, %xmm2, %xmm2
|
|
; AVX512F-NEXT: vpxor %xmm5, %xmm5, %xmm5
|
|
; AVX512F-NEXT: valignq {{.*#+}} zmm4 = zmm5[7],zmm4[0,1,2,3,4,5,6]
|
|
; AVX512F-NEXT: vpsrlq $1, %zmm4, %zmm4
|
|
; AVX512F-NEXT: vpsrlq %xmm2, %zmm4, %zmm2
|
|
; AVX512F-NEXT: vporq %zmm2, %zmm3, %zmm2
|
|
; AVX512F-NEXT: movl $512, %ecx # imm = 0x200
|
|
; AVX512F-NEXT: subq %rsi, %rcx
|
|
; AVX512F-NEXT: vmovq %rcx, %xmm3
|
|
; AVX512F-NEXT: vpbroadcastq %xmm3, %xmm3
|
|
; AVX512F-NEXT: vpand %xmm1, %xmm3, %xmm4
|
|
; AVX512F-NEXT: shrl $6, %ecx
|
|
; AVX512F-NEXT: shlxl %ecx, %edx, %ecx
|
|
; AVX512F-NEXT: kmovw %ecx, %k1
|
|
; AVX512F-NEXT: vpcompressq %zmm0, %zmm6 {%k1} {z}
|
|
; AVX512F-NEXT: vpsrlq %xmm4, %zmm6, %zmm4
|
|
; AVX512F-NEXT: vpandn %xmm1, %xmm3, %xmm1
|
|
; AVX512F-NEXT: valignq {{.*#+}} zmm3 = zmm6[1,2,3,4,5,6,7],zmm5[0]
|
|
; AVX512F-NEXT: vpaddq %zmm3, %zmm3, %zmm3
|
|
; AVX512F-NEXT: vpsllq %xmm1, %zmm3, %zmm1
|
|
; AVX512F-NEXT: xorl %ecx, %ecx
|
|
; AVX512F-NEXT: negq %rsi
|
|
; AVX512F-NEXT: sbbl %ecx, %ecx
|
|
; AVX512F-NEXT: vporq %zmm4, %zmm1, %zmm1
|
|
; AVX512F-NEXT: kmovw %ecx, %k1
|
|
; AVX512F-NEXT: vporq %zmm1, %zmm2, %zmm0 {%k1}
|
|
; AVX512F-NEXT: vmovdqu64 %zmm0, (%rdi)
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: fshl_rot_i512_vector:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: movq %rdi, %rax
|
|
; AVX512VL-NEXT: andl $511, %esi # imm = 0x1FF
|
|
; AVX512VL-NEXT: vpbroadcastq %rsi, %xmm1
|
|
; AVX512VL-NEXT: vpbroadcastq {{.*#+}} xmm2 = [63,63]
|
|
; AVX512VL-NEXT: vpand %xmm2, %xmm1, %xmm3
|
|
; AVX512VL-NEXT: movl %esi, %ecx
|
|
; AVX512VL-NEXT: shrl $6, %ecx
|
|
; AVX512VL-NEXT: movl $-1, %edx
|
|
; AVX512VL-NEXT: shlxl %ecx, %edx, %ecx
|
|
; AVX512VL-NEXT: kmovd %ecx, %k1
|
|
; AVX512VL-NEXT: vpexpandq %zmm0, %zmm4 {%k1} {z}
|
|
; AVX512VL-NEXT: vpsllq %xmm3, %zmm4, %zmm3
|
|
; AVX512VL-NEXT: vpandn %xmm2, %xmm1, %xmm1
|
|
; AVX512VL-NEXT: vpxor %xmm5, %xmm5, %xmm5
|
|
; AVX512VL-NEXT: valignq {{.*#+}} zmm4 = zmm5[7],zmm4[0,1,2,3,4,5,6]
|
|
; AVX512VL-NEXT: vpsrlq $1, %zmm4, %zmm4
|
|
; AVX512VL-NEXT: vpsrlq %xmm1, %zmm4, %zmm1
|
|
; AVX512VL-NEXT: vporq %zmm1, %zmm3, %zmm1
|
|
; AVX512VL-NEXT: movl $512, %ecx # imm = 0x200
|
|
; AVX512VL-NEXT: subq %rsi, %rcx
|
|
; AVX512VL-NEXT: vpbroadcastq %rcx, %xmm3
|
|
; AVX512VL-NEXT: vpand %xmm2, %xmm3, %xmm4
|
|
; AVX512VL-NEXT: shrl $6, %ecx
|
|
; AVX512VL-NEXT: shlxl %ecx, %edx, %ecx
|
|
; AVX512VL-NEXT: kmovd %ecx, %k1
|
|
; AVX512VL-NEXT: vpcompressq %zmm0, %zmm6 {%k1} {z}
|
|
; AVX512VL-NEXT: vpsrlq %xmm4, %zmm6, %zmm4
|
|
; AVX512VL-NEXT: vpandn %xmm2, %xmm3, %xmm2
|
|
; AVX512VL-NEXT: valignq {{.*#+}} zmm3 = zmm6[1,2,3,4,5,6,7],zmm5[0]
|
|
; AVX512VL-NEXT: vpaddq %zmm3, %zmm3, %zmm3
|
|
; AVX512VL-NEXT: vpsllq %xmm2, %zmm3, %zmm2
|
|
; AVX512VL-NEXT: vporq %zmm4, %zmm2, %zmm2
|
|
; AVX512VL-NEXT: xorl %ecx, %ecx
|
|
; AVX512VL-NEXT: negq %rsi
|
|
; AVX512VL-NEXT: sbbl %ecx, %ecx
|
|
; AVX512VL-NEXT: kmovd %ecx, %k1
|
|
; AVX512VL-NEXT: vporq %zmm2, %zmm1, %zmm0 {%k1}
|
|
; AVX512VL-NEXT: vextracti64x4 $1, %zmm0, 32(%rdi)
|
|
; AVX512VL-NEXT: vmovdqu %ymm0, (%rdi)
|
|
; AVX512VL-NEXT: vzeroupper
|
|
; AVX512VL-NEXT: retq
|
|
;
|
|
; AVX512VBMI-LABEL: fshl_rot_i512_vector:
|
|
; AVX512VBMI: # %bb.0:
|
|
; AVX512VBMI-NEXT: movq %rdi, %rax
|
|
; AVX512VBMI-NEXT: andl $511, %esi # imm = 0x1FF
|
|
; AVX512VBMI-NEXT: movl $512, %ecx # imm = 0x200
|
|
; AVX512VBMI-NEXT: subq %rsi, %rcx
|
|
; AVX512VBMI-NEXT: vpbroadcastq %rcx, %zmm1
|
|
; AVX512VBMI-NEXT: # kill: def $ecx killed $ecx killed $rcx
|
|
; AVX512VBMI-NEXT: shrl $6, %ecx
|
|
; AVX512VBMI-NEXT: movl $-1, %edx
|
|
; AVX512VBMI-NEXT: shlxl %ecx, %edx, %ecx
|
|
; AVX512VBMI-NEXT: kmovd %ecx, %k1
|
|
; AVX512VBMI-NEXT: vpcompressq %zmm0, %zmm2 {%k1} {z}
|
|
; AVX512VBMI-NEXT: vpxor %xmm3, %xmm3, %xmm3
|
|
; AVX512VBMI-NEXT: valignq {{.*#+}} zmm4 = zmm2[1,2,3,4,5,6,7],zmm3[0]
|
|
; AVX512VBMI-NEXT: movl %esi, %ecx
|
|
; AVX512VBMI-NEXT: shrl $6, %ecx
|
|
; AVX512VBMI-NEXT: shlxl %ecx, %edx, %ecx
|
|
; AVX512VBMI-NEXT: kmovd %ecx, %k1
|
|
; AVX512VBMI-NEXT: vpexpandq %zmm0, %zmm5 {%k1} {z}
|
|
; AVX512VBMI-NEXT: vpshrdvq %zmm1, %zmm4, %zmm2
|
|
; AVX512VBMI-NEXT: valignq {{.*#+}} zmm1 = zmm3[7],zmm5[0,1,2,3,4,5,6]
|
|
; AVX512VBMI-NEXT: vpbroadcastq %rsi, %zmm3
|
|
; AVX512VBMI-NEXT: vpshldvq %zmm3, %zmm1, %zmm5
|
|
; AVX512VBMI-NEXT: xorl %ecx, %ecx
|
|
; AVX512VBMI-NEXT: negq %rsi
|
|
; AVX512VBMI-NEXT: sbbl %ecx, %ecx
|
|
; AVX512VBMI-NEXT: kmovd %ecx, %k1
|
|
; AVX512VBMI-NEXT: vporq %zmm2, %zmm5, %zmm0 {%k1}
|
|
; AVX512VBMI-NEXT: vextracti64x4 $1, %zmm0, 32(%rdi)
|
|
; AVX512VBMI-NEXT: vmovdqu %ymm0, (%rdi)
|
|
; AVX512VBMI-NEXT: vzeroupper
|
|
; AVX512VBMI-NEXT: retq
|
|
%a0 = bitcast <8 x i64> %v0 to i512
|
|
%r = call i512 @llvm.fshl.i512(i512 %a0, i512 %a0, i512 %a2)
|
|
ret i512 %r
|
|
}
|
|
|
|
define i512 @fshr_rot_i512_vector(<8 x i64> %v0, i512 %a2) nounwind {
|
|
; SSE2-LABEL: fshr_rot_i512_vector:
|
|
; SSE2: # %bb.0:
|
|
; SSE2-NEXT: pushq %r15
|
|
; SSE2-NEXT: pushq %r14
|
|
; SSE2-NEXT: pushq %r13
|
|
; SSE2-NEXT: pushq %r12
|
|
; SSE2-NEXT: pushq %rbx
|
|
; SSE2-NEXT: movq %rsi, %rcx
|
|
; SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm2[2,3,2,3]
|
|
; SSE2-NEXT: movq %xmm4, %rdx
|
|
; SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm0[2,3,2,3]
|
|
; SSE2-NEXT: movq %xmm3, %rax
|
|
; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm3[2,3,2,3]
|
|
; SSE2-NEXT: movq %xmm3, %r11
|
|
; SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm1[2,3,2,3]
|
|
; SSE2-NEXT: movq %xmm0, %rsi
|
|
; SSE2-NEXT: movq %xmm2, %r10
|
|
; SSE2-NEXT: testl $256, %ecx # imm = 0x100
|
|
; SSE2-NEXT: movq %r10, %r14
|
|
; SSE2-NEXT: cmoveq %rsi, %r14
|
|
; SSE2-NEXT: movq %xmm3, %rbx
|
|
; SSE2-NEXT: movq %rbx, %r15
|
|
; SSE2-NEXT: cmoveq %r11, %r15
|
|
; SSE2-NEXT: movq %xmm1, %r12
|
|
; SSE2-NEXT: movq %r12, %r8
|
|
; SSE2-NEXT: cmoveq %rax, %r8
|
|
; SSE2-NEXT: movq %xmm4, %r13
|
|
; SSE2-NEXT: movq %r13, %r9
|
|
; SSE2-NEXT: cmoveq %rdx, %r9
|
|
; SSE2-NEXT: cmoveq %r10, %rsi
|
|
; SSE2-NEXT: cmoveq %r12, %rax
|
|
; SSE2-NEXT: cmoveq %r13, %rdx
|
|
; SSE2-NEXT: cmoveq %rbx, %r11
|
|
; SSE2-NEXT: testb $-128, %cl
|
|
; SSE2-NEXT: movq %r9, %r10
|
|
; SSE2-NEXT: cmoveq %r11, %r10
|
|
; SSE2-NEXT: cmoveq %rdx, %r11
|
|
; SSE2-NEXT: movq %r8, %rbx
|
|
; SSE2-NEXT: cmoveq %rsi, %rbx
|
|
; SSE2-NEXT: cmoveq %rax, %rsi
|
|
; SSE2-NEXT: cmovneq %r15, %r9
|
|
; SSE2-NEXT: cmovneq %r14, %r8
|
|
; SSE2-NEXT: cmoveq %r15, %rdx
|
|
; SSE2-NEXT: cmoveq %r14, %rax
|
|
; SSE2-NEXT: testb $64, %cl
|
|
; SSE2-NEXT: movq %rax, %r14
|
|
; SSE2-NEXT: cmoveq %rdx, %r14
|
|
; SSE2-NEXT: cmoveq %r8, %rdx
|
|
; SSE2-NEXT: cmoveq %r9, %r8
|
|
; SSE2-NEXT: cmoveq %rbx, %r9
|
|
; SSE2-NEXT: cmoveq %r10, %rbx
|
|
; SSE2-NEXT: cmoveq %rsi, %r10
|
|
; SSE2-NEXT: cmoveq %r11, %rsi
|
|
; SSE2-NEXT: cmovneq %r11, %rax
|
|
; SSE2-NEXT: movq %rax, %r11
|
|
; SSE2-NEXT: shrdq %cl, %rsi, %r11
|
|
; SSE2-NEXT: shrdq %cl, %r10, %rsi
|
|
; SSE2-NEXT: shrdq %cl, %rbx, %r10
|
|
; SSE2-NEXT: shrdq %cl, %r9, %rbx
|
|
; SSE2-NEXT: shrdq %cl, %r8, %r9
|
|
; SSE2-NEXT: shrdq %cl, %rdx, %r8
|
|
; SSE2-NEXT: shrdq %cl, %r14, %rdx
|
|
; SSE2-NEXT: # kill: def $cl killed $cl killed $rcx
|
|
; SSE2-NEXT: shrdq %cl, %rax, %r14
|
|
; SSE2-NEXT: movq %rdi, %rax
|
|
; SSE2-NEXT: movq %r14, 56(%rdi)
|
|
; SSE2-NEXT: movq %rdx, 48(%rdi)
|
|
; SSE2-NEXT: movq %r8, 40(%rdi)
|
|
; SSE2-NEXT: movq %r9, 32(%rdi)
|
|
; SSE2-NEXT: movq %rbx, 24(%rdi)
|
|
; SSE2-NEXT: movq %r10, 16(%rdi)
|
|
; SSE2-NEXT: movq %rsi, 8(%rdi)
|
|
; SSE2-NEXT: movq %r11, (%rdi)
|
|
; SSE2-NEXT: popq %rbx
|
|
; SSE2-NEXT: popq %r12
|
|
; SSE2-NEXT: popq %r13
|
|
; SSE2-NEXT: popq %r14
|
|
; SSE2-NEXT: popq %r15
|
|
; SSE2-NEXT: retq
|
|
;
|
|
; SSE42-LABEL: fshr_rot_i512_vector:
|
|
; SSE42: # %bb.0:
|
|
; SSE42-NEXT: pushq %r15
|
|
; SSE42-NEXT: pushq %r14
|
|
; SSE42-NEXT: pushq %r13
|
|
; SSE42-NEXT: pushq %r12
|
|
; SSE42-NEXT: pushq %rbx
|
|
; SSE42-NEXT: movq %rsi, %rcx
|
|
; SSE42-NEXT: pextrq $1, %xmm2, %rdx
|
|
; SSE42-NEXT: pextrq $1, %xmm0, %r10
|
|
; SSE42-NEXT: pextrq $1, %xmm3, %rbx
|
|
; SSE42-NEXT: pextrq $1, %xmm1, %r11
|
|
; SSE42-NEXT: movq %xmm0, %rsi
|
|
; SSE42-NEXT: movq %xmm2, %r15
|
|
; SSE42-NEXT: testl $256, %ecx # imm = 0x100
|
|
; SSE42-NEXT: movq %r15, %r14
|
|
; SSE42-NEXT: cmoveq %rsi, %r14
|
|
; SSE42-NEXT: movq %xmm3, %rax
|
|
; SSE42-NEXT: movq %r11, %r12
|
|
; SSE42-NEXT: cmoveq %rbx, %r12
|
|
; SSE42-NEXT: movq %xmm1, %r13
|
|
; SSE42-NEXT: movq %r13, %r8
|
|
; SSE42-NEXT: cmoveq %rax, %r8
|
|
; SSE42-NEXT: movq %r10, %r9
|
|
; SSE42-NEXT: cmoveq %rdx, %r9
|
|
; SSE42-NEXT: cmoveq %r15, %rsi
|
|
; SSE42-NEXT: cmoveq %r13, %rax
|
|
; SSE42-NEXT: cmoveq %r10, %rdx
|
|
; SSE42-NEXT: cmoveq %r11, %rbx
|
|
; SSE42-NEXT: testb $-128, %cl
|
|
; SSE42-NEXT: movq %r9, %r10
|
|
; SSE42-NEXT: cmoveq %rbx, %r10
|
|
; SSE42-NEXT: cmoveq %rdx, %rbx
|
|
; SSE42-NEXT: movq %r8, %r11
|
|
; SSE42-NEXT: cmoveq %rsi, %r11
|
|
; SSE42-NEXT: cmoveq %rax, %rsi
|
|
; SSE42-NEXT: cmovneq %r12, %r9
|
|
; SSE42-NEXT: cmovneq %r14, %r8
|
|
; SSE42-NEXT: cmoveq %r12, %rdx
|
|
; SSE42-NEXT: cmoveq %r14, %rax
|
|
; SSE42-NEXT: testb $64, %cl
|
|
; SSE42-NEXT: movq %rax, %r14
|
|
; SSE42-NEXT: cmoveq %rdx, %r14
|
|
; SSE42-NEXT: cmoveq %r8, %rdx
|
|
; SSE42-NEXT: cmoveq %r9, %r8
|
|
; SSE42-NEXT: cmoveq %r11, %r9
|
|
; SSE42-NEXT: cmoveq %r10, %r11
|
|
; SSE42-NEXT: cmoveq %rsi, %r10
|
|
; SSE42-NEXT: cmoveq %rbx, %rsi
|
|
; SSE42-NEXT: cmovneq %rbx, %rax
|
|
; SSE42-NEXT: movq %rax, %rbx
|
|
; SSE42-NEXT: shrdq %cl, %rsi, %rbx
|
|
; SSE42-NEXT: shrdq %cl, %r10, %rsi
|
|
; SSE42-NEXT: shrdq %cl, %r11, %r10
|
|
; SSE42-NEXT: shrdq %cl, %r9, %r11
|
|
; SSE42-NEXT: shrdq %cl, %r8, %r9
|
|
; SSE42-NEXT: shrdq %cl, %rdx, %r8
|
|
; SSE42-NEXT: shrdq %cl, %r14, %rdx
|
|
; SSE42-NEXT: # kill: def $cl killed $cl killed $rcx
|
|
; SSE42-NEXT: shrdq %cl, %rax, %r14
|
|
; SSE42-NEXT: movq %rdi, %rax
|
|
; SSE42-NEXT: movq %r14, 56(%rdi)
|
|
; SSE42-NEXT: movq %rdx, 48(%rdi)
|
|
; SSE42-NEXT: movq %r8, 40(%rdi)
|
|
; SSE42-NEXT: movq %r9, 32(%rdi)
|
|
; SSE42-NEXT: movq %r11, 24(%rdi)
|
|
; SSE42-NEXT: movq %r10, 16(%rdi)
|
|
; SSE42-NEXT: movq %rsi, 8(%rdi)
|
|
; SSE42-NEXT: movq %rbx, (%rdi)
|
|
; SSE42-NEXT: popq %rbx
|
|
; SSE42-NEXT: popq %r12
|
|
; SSE42-NEXT: popq %r13
|
|
; SSE42-NEXT: popq %r14
|
|
; SSE42-NEXT: popq %r15
|
|
; SSE42-NEXT: retq
|
|
;
|
|
; AVX2-LABEL: fshr_rot_i512_vector:
|
|
; AVX2: # %bb.0:
|
|
; AVX2-NEXT: pushq %r15
|
|
; AVX2-NEXT: pushq %r14
|
|
; AVX2-NEXT: pushq %r13
|
|
; AVX2-NEXT: pushq %r12
|
|
; AVX2-NEXT: pushq %rbx
|
|
; AVX2-NEXT: vpextrq $1, %xmm1, %rdx
|
|
; AVX2-NEXT: vpextrq $1, %xmm0, %rax
|
|
; AVX2-NEXT: movq %rsi, %rcx
|
|
; AVX2-NEXT: vextracti128 $1, %ymm1, %xmm2
|
|
; AVX2-NEXT: vmovq %xmm2, %r8
|
|
; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm3
|
|
; AVX2-NEXT: vpextrq $1, %xmm2, %r14
|
|
; AVX2-NEXT: vpextrq $1, %xmm3, %r11
|
|
; AVX2-NEXT: vmovq %xmm3, %rbx
|
|
; AVX2-NEXT: vmovq %xmm0, %rsi
|
|
; AVX2-NEXT: vmovq %xmm1, %r12
|
|
; AVX2-NEXT: testl $256, %ecx # imm = 0x100
|
|
; AVX2-NEXT: movq %r12, %r15
|
|
; AVX2-NEXT: cmoveq %rsi, %r15
|
|
; AVX2-NEXT: movq %r11, %r13
|
|
; AVX2-NEXT: cmoveq %r14, %r13
|
|
; AVX2-NEXT: movq %rbx, %r9
|
|
; AVX2-NEXT: cmoveq %r8, %r9
|
|
; AVX2-NEXT: movq %rax, %r10
|
|
; AVX2-NEXT: cmoveq %rdx, %r10
|
|
; AVX2-NEXT: cmoveq %r12, %rsi
|
|
; AVX2-NEXT: cmoveq %rbx, %r8
|
|
; AVX2-NEXT: cmoveq %rax, %rdx
|
|
; AVX2-NEXT: cmoveq %r11, %r14
|
|
; AVX2-NEXT: testb $-128, %cl
|
|
; AVX2-NEXT: movq %r10, %r11
|
|
; AVX2-NEXT: cmoveq %r14, %r11
|
|
; AVX2-NEXT: cmoveq %rdx, %r14
|
|
; AVX2-NEXT: movq %r9, %rbx
|
|
; AVX2-NEXT: cmoveq %rsi, %rbx
|
|
; AVX2-NEXT: cmoveq %r8, %rsi
|
|
; AVX2-NEXT: cmovneq %r13, %r10
|
|
; AVX2-NEXT: cmovneq %r15, %r9
|
|
; AVX2-NEXT: cmoveq %r13, %rdx
|
|
; AVX2-NEXT: movq %rdi, %rax
|
|
; AVX2-NEXT: cmoveq %r15, %r8
|
|
; AVX2-NEXT: testb $64, %cl
|
|
; AVX2-NEXT: movq %r8, %rdi
|
|
; AVX2-NEXT: cmoveq %rdx, %rdi
|
|
; AVX2-NEXT: cmoveq %r9, %rdx
|
|
; AVX2-NEXT: cmoveq %r10, %r9
|
|
; AVX2-NEXT: cmoveq %rbx, %r10
|
|
; AVX2-NEXT: cmoveq %r11, %rbx
|
|
; AVX2-NEXT: cmoveq %rsi, %r11
|
|
; AVX2-NEXT: cmoveq %r14, %rsi
|
|
; AVX2-NEXT: cmovneq %r14, %r8
|
|
; AVX2-NEXT: movq %r8, %r14
|
|
; AVX2-NEXT: shrdq %cl, %rsi, %r14
|
|
; AVX2-NEXT: shrdq %cl, %r11, %rsi
|
|
; AVX2-NEXT: shrdq %cl, %rbx, %r11
|
|
; AVX2-NEXT: shrdq %cl, %r10, %rbx
|
|
; AVX2-NEXT: shrdq %cl, %r9, %r10
|
|
; AVX2-NEXT: shrdq %cl, %rdx, %r9
|
|
; AVX2-NEXT: shrdq %cl, %rdi, %rdx
|
|
; AVX2-NEXT: # kill: def $cl killed $cl killed $rcx
|
|
; AVX2-NEXT: shrdq %cl, %r8, %rdi
|
|
; AVX2-NEXT: movq %rdi, 56(%rax)
|
|
; AVX2-NEXT: movq %rdx, 48(%rax)
|
|
; AVX2-NEXT: movq %r9, 40(%rax)
|
|
; AVX2-NEXT: movq %r10, 32(%rax)
|
|
; AVX2-NEXT: movq %rbx, 24(%rax)
|
|
; AVX2-NEXT: movq %r11, 16(%rax)
|
|
; AVX2-NEXT: movq %rsi, 8(%rax)
|
|
; AVX2-NEXT: movq %r14, (%rax)
|
|
; AVX2-NEXT: popq %rbx
|
|
; AVX2-NEXT: popq %r12
|
|
; AVX2-NEXT: popq %r13
|
|
; AVX2-NEXT: popq %r14
|
|
; AVX2-NEXT: popq %r15
|
|
; AVX2-NEXT: vzeroupper
|
|
; AVX2-NEXT: retq
|
|
;
|
|
; AVX512F-LABEL: fshr_rot_i512_vector:
|
|
; AVX512F: # %bb.0:
|
|
; AVX512F-NEXT: movq %rdi, %rax
|
|
; AVX512F-NEXT: andl $511, %esi # imm = 0x1FF
|
|
; AVX512F-NEXT: vmovq %rsi, %xmm1
|
|
; AVX512F-NEXT: vpbroadcastq %xmm1, %xmm2
|
|
; AVX512F-NEXT: vpbroadcastq {{.*#+}} xmm1 = [63,63]
|
|
; AVX512F-NEXT: vpand %xmm1, %xmm2, %xmm3
|
|
; AVX512F-NEXT: movl %esi, %ecx
|
|
; AVX512F-NEXT: shrl $6, %ecx
|
|
; AVX512F-NEXT: movl $-1, %edx
|
|
; AVX512F-NEXT: shlxl %ecx, %edx, %ecx
|
|
; AVX512F-NEXT: kmovw %ecx, %k1
|
|
; AVX512F-NEXT: vpcompressq %zmm0, %zmm4 {%k1} {z}
|
|
; AVX512F-NEXT: vpsrlq %xmm3, %zmm4, %zmm3
|
|
; AVX512F-NEXT: vpandn %xmm1, %xmm2, %xmm2
|
|
; AVX512F-NEXT: vpxor %xmm5, %xmm5, %xmm5
|
|
; AVX512F-NEXT: valignq {{.*#+}} zmm4 = zmm4[1,2,3,4,5,6,7],zmm5[0]
|
|
; AVX512F-NEXT: vpaddq %zmm4, %zmm4, %zmm4
|
|
; AVX512F-NEXT: vpsllq %xmm2, %zmm4, %zmm2
|
|
; AVX512F-NEXT: vporq %zmm3, %zmm2, %zmm2
|
|
; AVX512F-NEXT: movl $512, %ecx # imm = 0x200
|
|
; AVX512F-NEXT: subq %rsi, %rcx
|
|
; AVX512F-NEXT: vmovq %rcx, %xmm3
|
|
; AVX512F-NEXT: vpbroadcastq %xmm3, %xmm3
|
|
; AVX512F-NEXT: vpand %xmm1, %xmm3, %xmm4
|
|
; AVX512F-NEXT: shrl $6, %ecx
|
|
; AVX512F-NEXT: shlxl %ecx, %edx, %ecx
|
|
; AVX512F-NEXT: kmovw %ecx, %k1
|
|
; AVX512F-NEXT: vpexpandq %zmm0, %zmm6 {%k1} {z}
|
|
; AVX512F-NEXT: vpsllq %xmm4, %zmm6, %zmm4
|
|
; AVX512F-NEXT: vpandn %xmm1, %xmm3, %xmm1
|
|
; AVX512F-NEXT: valignq {{.*#+}} zmm3 = zmm5[7],zmm6[0,1,2,3,4,5,6]
|
|
; AVX512F-NEXT: vpsrlq $1, %zmm3, %zmm3
|
|
; AVX512F-NEXT: vpsrlq %xmm1, %zmm3, %zmm1
|
|
; AVX512F-NEXT: xorl %ecx, %ecx
|
|
; AVX512F-NEXT: negq %rsi
|
|
; AVX512F-NEXT: sbbl %ecx, %ecx
|
|
; AVX512F-NEXT: vporq %zmm1, %zmm4, %zmm1
|
|
; AVX512F-NEXT: kmovw %ecx, %k1
|
|
; AVX512F-NEXT: vporq %zmm2, %zmm1, %zmm0 {%k1}
|
|
; AVX512F-NEXT: vmovdqu64 %zmm0, (%rdi)
|
|
; AVX512F-NEXT: retq
|
|
;
|
|
; AVX512VL-LABEL: fshr_rot_i512_vector:
|
|
; AVX512VL: # %bb.0:
|
|
; AVX512VL-NEXT: movq %rdi, %rax
|
|
; AVX512VL-NEXT: andl $511, %esi # imm = 0x1FF
|
|
; AVX512VL-NEXT: vpbroadcastq %rsi, %xmm1
|
|
; AVX512VL-NEXT: vpbroadcastq {{.*#+}} xmm2 = [63,63]
|
|
; AVX512VL-NEXT: vpand %xmm2, %xmm1, %xmm3
|
|
; AVX512VL-NEXT: movl %esi, %ecx
|
|
; AVX512VL-NEXT: shrl $6, %ecx
|
|
; AVX512VL-NEXT: movl $-1, %edx
|
|
; AVX512VL-NEXT: shlxl %ecx, %edx, %ecx
|
|
; AVX512VL-NEXT: kmovd %ecx, %k1
|
|
; AVX512VL-NEXT: vpcompressq %zmm0, %zmm4 {%k1} {z}
|
|
; AVX512VL-NEXT: vpsrlq %xmm3, %zmm4, %zmm3
|
|
; AVX512VL-NEXT: vpandn %xmm2, %xmm1, %xmm1
|
|
; AVX512VL-NEXT: vpxor %xmm5, %xmm5, %xmm5
|
|
; AVX512VL-NEXT: valignq {{.*#+}} zmm4 = zmm4[1,2,3,4,5,6,7],zmm5[0]
|
|
; AVX512VL-NEXT: vpaddq %zmm4, %zmm4, %zmm4
|
|
; AVX512VL-NEXT: vpsllq %xmm1, %zmm4, %zmm1
|
|
; AVX512VL-NEXT: vporq %zmm3, %zmm1, %zmm1
|
|
; AVX512VL-NEXT: movl $512, %ecx # imm = 0x200
|
|
; AVX512VL-NEXT: subq %rsi, %rcx
|
|
; AVX512VL-NEXT: vpbroadcastq %rcx, %xmm3
|
|
; AVX512VL-NEXT: vpand %xmm2, %xmm3, %xmm4
|
|
; AVX512VL-NEXT: shrl $6, %ecx
|
|
; AVX512VL-NEXT: shlxl %ecx, %edx, %ecx
|
|
; AVX512VL-NEXT: kmovd %ecx, %k1
|
|
; AVX512VL-NEXT: vpexpandq %zmm0, %zmm6 {%k1} {z}
|
|
; AVX512VL-NEXT: vpsllq %xmm4, %zmm6, %zmm4
|
|
; AVX512VL-NEXT: vpandn %xmm2, %xmm3, %xmm2
|
|
; AVX512VL-NEXT: valignq {{.*#+}} zmm3 = zmm5[7],zmm6[0,1,2,3,4,5,6]
|
|
; AVX512VL-NEXT: vpsrlq $1, %zmm3, %zmm3
|
|
; AVX512VL-NEXT: vpsrlq %xmm2, %zmm3, %zmm2
|
|
; AVX512VL-NEXT: vporq %zmm2, %zmm4, %zmm2
|
|
; AVX512VL-NEXT: xorl %ecx, %ecx
|
|
; AVX512VL-NEXT: negq %rsi
|
|
; AVX512VL-NEXT: sbbl %ecx, %ecx
|
|
; AVX512VL-NEXT: kmovd %ecx, %k1
|
|
; AVX512VL-NEXT: vporq %zmm1, %zmm2, %zmm0 {%k1}
|
|
; AVX512VL-NEXT: vextracti64x4 $1, %zmm0, 32(%rdi)
|
|
; AVX512VL-NEXT: vmovdqu %ymm0, (%rdi)
|
|
; AVX512VL-NEXT: vzeroupper
|
|
; AVX512VL-NEXT: retq
|
|
;
|
|
; AVX512VBMI-LABEL: fshr_rot_i512_vector:
|
|
; AVX512VBMI: # %bb.0:
|
|
; AVX512VBMI-NEXT: movq %rdi, %rax
|
|
; AVX512VBMI-NEXT: andl $511, %esi # imm = 0x1FF
|
|
; AVX512VBMI-NEXT: movl $512, %ecx # imm = 0x200
|
|
; AVX512VBMI-NEXT: subq %rsi, %rcx
|
|
; AVX512VBMI-NEXT: vpbroadcastq %rcx, %zmm1
|
|
; AVX512VBMI-NEXT: # kill: def $ecx killed $ecx killed $rcx
|
|
; AVX512VBMI-NEXT: shrl $6, %ecx
|
|
; AVX512VBMI-NEXT: movl $-1, %edx
|
|
; AVX512VBMI-NEXT: shlxl %ecx, %edx, %ecx
|
|
; AVX512VBMI-NEXT: kmovd %ecx, %k1
|
|
; AVX512VBMI-NEXT: vpexpandq %zmm0, %zmm2 {%k1} {z}
|
|
; AVX512VBMI-NEXT: vpxor %xmm3, %xmm3, %xmm3
|
|
; AVX512VBMI-NEXT: valignq {{.*#+}} zmm4 = zmm3[7],zmm2[0,1,2,3,4,5,6]
|
|
; AVX512VBMI-NEXT: movl %esi, %ecx
|
|
; AVX512VBMI-NEXT: shrl $6, %ecx
|
|
; AVX512VBMI-NEXT: shlxl %ecx, %edx, %ecx
|
|
; AVX512VBMI-NEXT: kmovd %ecx, %k1
|
|
; AVX512VBMI-NEXT: vpcompressq %zmm0, %zmm5 {%k1} {z}
|
|
; AVX512VBMI-NEXT: vpshldvq %zmm1, %zmm4, %zmm2
|
|
; AVX512VBMI-NEXT: valignq {{.*#+}} zmm1 = zmm5[1,2,3,4,5,6,7],zmm3[0]
|
|
; AVX512VBMI-NEXT: vpbroadcastq %rsi, %zmm3
|
|
; AVX512VBMI-NEXT: vpshrdvq %zmm3, %zmm1, %zmm5
|
|
; AVX512VBMI-NEXT: xorl %ecx, %ecx
|
|
; AVX512VBMI-NEXT: negq %rsi
|
|
; AVX512VBMI-NEXT: sbbl %ecx, %ecx
|
|
; AVX512VBMI-NEXT: kmovd %ecx, %k1
|
|
; AVX512VBMI-NEXT: vporq %zmm5, %zmm2, %zmm0 {%k1}
|
|
; AVX512VBMI-NEXT: vextracti64x4 $1, %zmm0, 32(%rdi)
|
|
; AVX512VBMI-NEXT: vmovdqu %ymm0, (%rdi)
|
|
; AVX512VBMI-NEXT: vzeroupper
|
|
; AVX512VBMI-NEXT: retq
|
|
%a0 = bitcast <8 x i64> %v0 to i512
|
|
%r = call i512 @llvm.fshr.i512(i512 %a0, i512 %a0, i512 %a2)
|
|
ret i512 %r
|
|
}
|
|
|
|
;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
|
|
; AVX512: {{.*}}
|