deb-libisal/include/multibinary.asm
Greg Tucker a5b324d2cd Add avx512 versions of ec_encode_data
- Includes gf_nvect_dot_prod, gf_nvect_mad functions
 - Change ec multibinary to use common macros
 - Autoconf checks for nasm or yasm support and picks if available
 - Leave out compile of any avx512 code if assembler not available

Signed-off-by: Greg Tucker <greg.b.tucker@intel.com>
2016-02-25 17:32:34 -07:00

272 lines
7.0 KiB
NASM

;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
; Copyright(c) 2011-2015 Intel Corporation All rights reserved.
;
; Redistribution and use in source and binary forms, with or without
; modification, are permitted provided that the following conditions
; are met:
; * Redistributions of source code must retain the above copyright
; notice, this list of conditions and the following disclaimer.
; * Redistributions in binary form must reproduce the above copyright
; notice, this list of conditions and the following disclaimer in
; the documentation and/or other materials provided with the
; distribution.
; * Neither the name of Intel Corporation nor the names of its
; contributors may be used to endorse or promote products derived
; from this software without specific prior written permission.
;
; THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
; "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
; LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
; A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
; OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
; SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
; LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
; DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
; THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
; (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
; OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
%ifndef _MULTIBINARY_ASM_
%define _MULTIBINARY_ASM_
%ifidn __OUTPUT_FORMAT__, elf32
%define mbin_def_ptr dd
%define mbin_ptr_sz dword
%define mbin_rdi edi
%define mbin_rsi esi
%define mbin_rax eax
%define mbin_rbx ebx
%define mbin_rcx ecx
%define mbin_rdx edx
%else
%define mbin_def_ptr dq
%define mbin_ptr_sz qword
%define mbin_rdi rdi
%define mbin_rsi rsi
%define mbin_rax rax
%define mbin_rbx rbx
%define mbin_rcx rcx
%define mbin_rdx rdx
%endif
;;;;
; multibinary macro:
; creates the visable entry point that uses HW optimized call pointer
; creates the init of the HW optimized call pointer
;;;;
%macro mbin_interface 1
;;;;
; *_dispatched is defaulted to *_mbinit and replaced on first call.
; Therefore, *_dispatch_init is only executed on first call.
;;;;
section .data
%1_dispatched:
mbin_def_ptr %1_mbinit
section .text
global %1:function
%1_mbinit:
;;; only called the first time to setup hardware match
call %1_dispatch_init
;;; falls thru to execute the hw optimized code
%1:
jmp mbin_ptr_sz [%1_dispatched]
%endmacro
;;;;;
; mbin_dispatch_init parameters
; Use this function when SSE/00/01 is a minimum requirement
; 1-> function name
; 2-> SSE/00/01 optimized function used as base
; 3-> AVX or AVX/02 opt func
; 4-> AVX2 or AVX/04 opt func
;;;;;
%macro mbin_dispatch_init 4
section .text
%1_dispatch_init:
push mbin_rsi
push mbin_rax
push mbin_rbx
push mbin_rcx
push mbin_rdx
lea mbin_rsi, [%2 WRT_OPT] ; Default to SSE 00/01
mov eax, 1
cpuid
and ecx, (FLAG_CPUID1_ECX_AVX | FLAG_CPUID1_ECX_OSXSAVE)
cmp ecx, (FLAG_CPUID1_ECX_AVX | FLAG_CPUID1_ECX_OSXSAVE)
lea mbin_rbx, [%3 WRT_OPT] ; AVX (gen2) opt func
jne _%1_init_done ; AVX is not available so end
mov mbin_rsi, mbin_rbx
;; Try for AVX2
xor ecx, ecx
mov eax, 7
cpuid
test ebx, FLAG_CPUID7_EBX_AVX2
lea mbin_rbx, [%4 WRT_OPT] ; AVX (gen4) opt func
cmovne mbin_rsi, mbin_rbx
;; Does it have xmm and ymm support
xor ecx, ecx
xgetbv
and eax, FLAG_XGETBV_EAX_XMM_YMM
cmp eax, FLAG_XGETBV_EAX_XMM_YMM
je _%1_init_done
lea mbin_rsi, [%2 WRT_OPT]
_%1_init_done:
pop mbin_rdx
pop mbin_rcx
pop mbin_rbx
pop mbin_rax
mov [%1_dispatched], mbin_rsi
pop mbin_rsi
ret
%endmacro
;;;;;
; mbin_dispatch_init2 parameters
; Cases where only base functions are available
; 1-> function name
; 2-> base function
;;;;;
%macro mbin_dispatch_init2 2
section .text
%1_dispatch_init:
push mbin_rsi
lea mbin_rsi, [%2 WRT_OPT] ; Default
mov [%1_dispatched], mbin_rsi
pop mbin_rsi
ret
%endmacro
;;;;;
; mbin_dispatch_init5 parameters
; 1-> function name
; 2-> base function
; 3-> SSE4_1 or 00/01 optimized function
; 4-> AVX/02 opt func
; 5-> AVX2/04 opt func
;;;;;
%macro mbin_dispatch_init5 5
section .text
%1_dispatch_init:
push mbin_rsi
push mbin_rax
push mbin_rbx
push mbin_rcx
push mbin_rdx
lea mbin_rsi, [%2 WRT_OPT] ; Default - use base function
mov eax, 1
cpuid
; Test for SSE4.1
test ecx, FLAG_CPUID1_ECX_SSE4_1
lea mbin_rbx, [%3 WRT_OPT] ; SSE opt func
cmovne mbin_rsi, mbin_rbx
and ecx, (FLAG_CPUID1_ECX_AVX | FLAG_CPUID1_ECX_OSXSAVE)
cmp ecx, (FLAG_CPUID1_ECX_AVX | FLAG_CPUID1_ECX_OSXSAVE)
lea mbin_rbx, [%4 WRT_OPT] ; AVX (gen2) opt func
jne _%1_init_done ; AVX is not available so end
mov mbin_rsi, mbin_rbx
;; Try for AVX2
xor ecx, ecx
mov eax, 7
cpuid
test ebx, FLAG_CPUID7_EBX_AVX2
lea mbin_rbx, [%5 WRT_OPT] ; AVX (gen4) opt func
cmovne mbin_rsi, mbin_rbx
;; Does it have xmm and ymm support
xor ecx, ecx
xgetbv
and eax, FLAG_XGETBV_EAX_XMM_YMM
cmp eax, FLAG_XGETBV_EAX_XMM_YMM
je _%1_init_done
lea mbin_rsi, [%3 WRT_OPT]
_%1_init_done:
pop mbin_rdx
pop mbin_rcx
pop mbin_rbx
pop mbin_rax
mov [%1_dispatched], mbin_rsi
pop mbin_rsi
ret
%endmacro
;;;;;
; mbin_dispatch_init6 parameters
; 1-> function name
; 2-> base function
; 3-> SSE4_1 or 00/01 optimized function
; 4-> AVX/02 opt func
; 5-> AVX2/04 opt func
; 6-> AVX512/06 opt func
;;;;;
%macro mbin_dispatch_init6 6
section .text
%1_dispatch_init:
push mbin_rsi
push mbin_rax
push mbin_rbx
push mbin_rcx
push mbin_rdx
push mbin_rdi
lea mbin_rsi, [%2 WRT_OPT] ; Default - use base function
mov eax, 1
cpuid
mov ebx, ecx ; save cpuid1.ecx
test ecx, FLAG_CPUID1_ECX_SSE4_1
je _%1_init_done ; Use base function if no SSE4_1
lea mbin_rsi, [%3 WRT_OPT] ; SSE possible so use 00/01 opt
;; Test for XMM_YMM support/AVX
test ecx, FLAG_CPUID1_ECX_OSXSAVE
je _%1_init_done
xor ecx, ecx
xgetbv ; xcr -> edx:eax
mov edi, eax ; save xgetvb.eax
and eax, FLAG_XGETBV_EAX_XMM_YMM
cmp eax, FLAG_XGETBV_EAX_XMM_YMM
jne _%1_init_done
test ebx, FLAG_CPUID1_ECX_AVX
je _%1_init_done
lea mbin_rsi, [%4 WRT_OPT] ; AVX/02 opt
;; Test for AVX2
xor ecx, ecx
mov eax, 7
cpuid
test ebx, FLAG_CPUID7_EBX_AVX2
je _%1_init_done ; No AVX2 possible
lea mbin_rsi, [%5 WRT_OPT] ; AVX2/04 opt func
;; Test for AVX512
and edi, FLAG_XGETBV_EAX_ZMM_OPM
cmp edi, FLAG_XGETBV_EAX_ZMM_OPM
jne _%1_init_done ; No AVX512 possible
and ebx, FLAGS_CPUID7_ECX_AVX512_G1
cmp ebx, FLAGS_CPUID7_ECX_AVX512_G1
lea mbin_rbx, [%6 WRT_OPT] ; AVX512/06 opt
cmove mbin_rsi, mbin_rbx
_%1_init_done:
pop mbin_rdi
pop mbin_rdx
pop mbin_rcx
pop mbin_rbx
pop mbin_rax
mov [%1_dispatched], mbin_rsi
pop mbin_rsi
ret
%endmacro
%endif ; ifndef _MULTIBINARY_ASM_