glibc/sysdeps/x86_64/fpu/svml_sd_wrapper_impl.h

191 lines
4.9 KiB
C

/* Common float/double wrapper implementations of vector math
functions.
Copyright (C) 2022-2024 Free Software Foundation, Inc.
This file is part of the GNU C Library.
The GNU C Library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 2.1 of the License, or (at your option) any later version.
The GNU C Library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with the GNU C Library; if not, see
<https://www.gnu.org/licenses/>. */
/* AVX/AVX2 ISA version as wrapper to SSE ISA version. */
.macro WRAPPER_IMPL_AVX callee
pushq %rbp
cfi_adjust_cfa_offset (8)
cfi_rel_offset (%rbp, 0)
movq %rsp, %rbp
cfi_def_cfa_register (%rbp)
andq $-32, %rsp
subq $32, %rsp
vmovaps %ymm0, (%rsp)
vzeroupper
call HIDDEN_JUMPTARGET(\callee)
vmovaps %xmm0, (%rsp)
vmovaps 16(%rsp), %xmm0
call HIDDEN_JUMPTARGET(\callee)
/* combine xmm0 (return of second call) with result of first
call (saved on stack). Might be worth exploring logic that
uses `vpblend` and reads in ymm1 using -16(rsp). */
vmovaps (%rsp), %xmm1
vinsertf128 $1, %xmm0, %ymm1, %ymm0
movq %rbp, %rsp
cfi_def_cfa_register (%rsp)
popq %rbp
cfi_adjust_cfa_offset (-8)
cfi_restore (%rbp)
ret
.endm
/* 2 argument AVX/AVX2 ISA version as wrapper to SSE ISA version. */
.macro WRAPPER_IMPL_AVX_ff callee
pushq %rbp
cfi_adjust_cfa_offset (8)
cfi_rel_offset (%rbp, 0)
movq %rsp, %rbp
cfi_def_cfa_register (%rbp)
andq $-32, %rsp
subq $64, %rsp
vmovaps %ymm0, (%rsp)
vmovaps %ymm1, 32(%rsp)
vzeroupper
call HIDDEN_JUMPTARGET(\callee)
vmovaps 48(%rsp), %xmm1
vmovaps %xmm0, (%rsp)
vmovaps 16(%rsp), %xmm0
call HIDDEN_JUMPTARGET(\callee)
/* combine xmm0 (return of second call) with result of first
call (saved on stack). Might be worth exploring logic that
uses `vpblend` and reads in ymm1 using -16(rsp). */
vmovaps (%rsp), %xmm1
vinsertf128 $1, %xmm0, %ymm1, %ymm0
movq %rbp, %rsp
cfi_def_cfa_register (%rsp)
popq %rbp
cfi_adjust_cfa_offset (-8)
cfi_restore (%rbp)
ret
.endm
/* 3 argument AVX/AVX2 ISA version as wrapper to SSE ISA version. */
.macro WRAPPER_IMPL_AVX_fFF callee
pushq %rbp
cfi_adjust_cfa_offset (8)
cfi_rel_offset (%rbp, 0)
movq %rsp, %rbp
andq $-32, %rsp
subq $32, %rsp
vmovaps %ymm0, (%rsp)
pushq %rbx
pushq %r14
movq %rdi, %rbx
movq %rsi, %r14
vzeroupper
call HIDDEN_JUMPTARGET(\callee)
vmovaps 32(%rsp), %xmm0
leaq 16(%rbx), %rdi
leaq 16(%r14), %rsi
call HIDDEN_JUMPTARGET(\callee)
popq %r14
popq %rbx
movq %rbp, %rsp
cfi_def_cfa_register (%rsp)
popq %rbp
cfi_adjust_cfa_offset (-8)
cfi_restore (%rbp)
ret
.endm
/* AVX512 ISA version as wrapper to AVX2 ISA version. */
.macro WRAPPER_IMPL_AVX512 callee
pushq %rbp
cfi_adjust_cfa_offset (8)
cfi_rel_offset (%rbp, 0)
movq %rsp, %rbp
cfi_def_cfa_register (%rbp)
andq $-64, %rsp
subq $64, %rsp
vmovups %zmm0, (%rsp)
call HIDDEN_JUMPTARGET(\callee)
vmovupd %ymm0, (%rsp)
vmovupd 32(%rsp), %ymm0
call HIDDEN_JUMPTARGET(\callee)
/* combine ymm0 (return of second call) with result of first
call (saved on stack). */
vmovaps (%rsp), %ymm1
vinserti64x4 $0x1, %ymm0, %zmm1, %zmm0
movq %rbp, %rsp
cfi_def_cfa_register (%rsp)
popq %rbp
cfi_adjust_cfa_offset (-8)
cfi_restore (%rbp)
ret
.endm
/* 2 argument AVX512 ISA version as wrapper to AVX2 ISA version. */
.macro WRAPPER_IMPL_AVX512_ff callee
pushq %rbp
cfi_adjust_cfa_offset (8)
cfi_rel_offset (%rbp, 0)
movq %rsp, %rbp
cfi_def_cfa_register (%rbp)
andq $-64, %rsp
addq $-128, %rsp
vmovups %zmm0, (%rsp)
vmovups %zmm1, 64(%rsp)
/* ymm0 and ymm1 are already set. */
call HIDDEN_JUMPTARGET(\callee)
vmovups 96(%rsp), %ymm1
vmovaps %ymm0, (%rsp)
vmovups 32(%rsp), %ymm0
call HIDDEN_JUMPTARGET(\callee)
/* combine ymm0 (return of second call) with result of first
call (saved on stack). */
vmovaps (%rsp), %ymm1
vinserti64x4 $0x1, %ymm0, %zmm1, %zmm0
movq %rbp, %rsp
cfi_def_cfa_register (%rsp)
popq %rbp
cfi_adjust_cfa_offset (-8)
cfi_restore (%rbp)
ret
.endm
/* 3 argument AVX512 ISA version as wrapper to AVX2 ISA version. */
.macro WRAPPER_IMPL_AVX512_fFF callee
pushq %rbp
cfi_adjust_cfa_offset (8)
cfi_rel_offset (%rbp, 0)
movq %rsp, %rbp
cfi_def_cfa_register (%rbp)
andq $-64, %rsp
subq $64, %rsp
vmovaps %zmm0, (%rsp)
pushq %rbx
pushq %r14
movq %rdi, %rbx
movq %rsi, %r14
/* ymm0 is already set. */
call HIDDEN_JUMPTARGET(\callee)
vmovaps 48(%rsp), %ymm0
leaq 32(%rbx), %rdi
leaq 32(%r14), %rsi
call HIDDEN_JUMPTARGET(\callee)
popq %r14
popq %rbx
movq %rbp, %rsp
cfi_def_cfa_register (%rsp)
popq %rbp
cfi_adjust_cfa_offset (-8)
cfi_restore (%rbp)
ret
.endm