generic-library/isa-l
1
0
Fork 0
mirror of https://github.com/intel/isa-l.git synced 2024-12-12 09:23:50 +01:00
Code Issues Projects Releases Wiki Activity

erasure_code: add initial AVX2 mad with GFNI implementation

Browse Source 
Signed-off-by: Marcel Cornu <marcel.d.cornu@intel.com>
This commit is contained in:
Marcel Cornu 2023-11-30 16:24:55 +00:00 committed by Pablo de Lara
parent 447d9af75b
commit 5f23c03415
5 changed files with 194 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
Makefile.nmake
View File

@ -105,6 +105,7 @@ objs = \
bin\gf_5vect_mad_avx512.obj \
bin\gf_6vect_mad_avx512.obj \
bin\gf_vect_mad_avx512_gfni.obj \
bin\gf_vect_mad_avx2_gfni.obj \
bin\gf_2vect_mad_avx512_gfni.obj \
bin\gf_3vect_mad_avx512_gfni.obj \
bin\gf_4vect_mad_avx512_gfni.obj \

1
erasure_code/Makefile.am
View File

@ -98,6 +98,7 @@ lsrc_x86_64 += \
erasure_code/gf_5vect_mad_avx512.asm \
erasure_code/gf_6vect_mad_avx512.asm \
erasure_code/gf_vect_mad_avx512_gfni.asm \
erasure_code/gf_vect_mad_avx2_gfni.asm \
erasure_code/gf_2vect_mad_avx512_gfni.asm \
erasure_code/gf_3vect_mad_avx512_gfni.asm \
erasure_code/gf_4vect_mad_avx512_gfni.asm \

14
erasure_code/ec_highlevel_func.c
View File

@ -269,6 +269,8 @@ extern void gf_6vect_mad_avx512_gfni(int len, int vec, int vec_i, unsigned char
extern void gf_vect_dot_prod_avx2_gfni(int len, int k, unsigned char *g_tbls,
unsigned char **data, unsigned char *dest);
extern void gf_vect_mad_avx2_gfni(int len, int vec, int vec_i, unsigned char *gftbls,
unsigned char *src, unsigned char *dest);
void ec_init_tables_gfni(int k, int rows, unsigned char *a, unsigned char *g_tbls)
{
@ -358,6 +360,18 @@ void ec_encode_data_update_avx512_gfni(int len, int k, int rows, int vec_i,
}
}
void ec_encode_data_update_avx2_gfni(int len, int k, int rows, int vec_i,
unsigned char *g_tbls, unsigned char *data,
unsigned char **coding)
{
while (rows) {
gf_vect_mad_avx2_gfni(len, k, vec_i, g_tbls, data, *coding);
g_tbls += k * 8;
coding++;
rows--;
}
}
#endif // AS_FEATURE_LEVEL >= 10
#endif // HAVE_AS_KNOWS_AVX512

3
erasure_code/ec_multibinary.asm
View File

@ -58,6 +58,7 @@
extern ec_encode_data_avx512_gfni
extern ec_encode_data_avx2_gfni
extern ec_encode_data_update_avx512_gfni
extern ec_encode_data_update_avx2_gfni
%endif
extern ec_init_tables_base
@ -93,7 +94,7 @@ mbin_interface ec_init_tables
mbin_dispatch_init5 gf_vect_mul, gf_vect_mul_base, gf_vect_mul_sse, gf_vect_mul_avx, gf_vect_mul_avx
mbin_dispatch_init8 ec_encode_data, ec_encode_data_base, ec_encode_data_sse, ec_encode_data_avx, ec_encode_data_avx2, ec_encode_data_avx512, ec_encode_data_avx2_gfni, ec_encode_data_avx512_gfni
mbin_dispatch_init7 ec_encode_data_update, ec_encode_data_update_base, ec_encode_data_update_sse, ec_encode_data_update_avx, ec_encode_data_update_avx2, ec_encode_data_update_avx512, ec_encode_data_update_avx512_gfni
mbin_dispatch_init8 ec_encode_data_update, ec_encode_data_update_base, ec_encode_data_update_sse, ec_encode_data_update_avx, ec_encode_data_update_avx2, ec_encode_data_update_avx512, ec_encode_data_update_avx2_gfni, ec_encode_data_update_avx512_gfni
mbin_dispatch_init6 gf_vect_mad, gf_vect_mad_base, gf_vect_mad_sse, gf_vect_mad_avx, gf_vect_mad_avx2, gf_vect_mad_avx512
mbin_dispatch_init6 gf_vect_dot_prod, gf_vect_dot_prod_base, gf_vect_dot_prod_sse, gf_vect_dot_prod_avx, gf_vect_dot_prod_avx2, gf_vect_dot_prod_avx512
mbin_dispatch_init8 ec_init_tables, ec_init_tables_base, ec_init_tables_base, ec_init_tables_base, ec_init_tables_base, ec_init_tables_base, ec_init_tables_gfni, ec_init_tables_gfni

176
erasure_code/gf_vect_mad_avx2_gfni.asm Normal file
View File

@ -0,0 +1,176 @@
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
; Copyright(c) 2023 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.
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;;;
;;; gf_vect_mad_avx2_gfni(len, vec, vec_i, mul_array, src, dest);
;;;
%include "reg_sizes.asm"
%include "gf_vect_gfni.inc"
%include "memcpy.asm"
%if AS_FEATURE_LEVEL >= 10
%ifidn __OUTPUT_FORMAT__, elf64
%define arg0 rdi
%define arg1 rsi
%define arg2 rdx
%define arg3 rcx
%define arg4 r8
%define arg5 r9
%define tmp r11
%define tmp2 r10
%define func(x) x: endbranch
%define FUNC_SAVE
%define FUNC_RESTORE
%endif
%ifidn __OUTPUT_FORMAT__, win64
%define arg0 rcx
%define arg1 rdx
%define arg2 r8
%define arg3 r9
%define arg4 r12 ; must be saved and loaded
%define arg5 r13
%define tmp r11
%define tmp2 r10
%define stack_size 3*8
%define arg(x) [rsp + stack_size + 8 + 8*x]
%define func(x) proc_frame x
%macro FUNC_SAVE 0
sub rsp, stack_size
mov [rsp + 0*8], r12
mov [rsp + 1*8], r13
end_prolog
mov arg4, arg(4)
mov arg5, arg(5)
%endmacro
%macro FUNC_RESTORE 0
mov r12, [rsp + 0*8]
mov r13, [rsp + 1*8]
add rsp, stack_size
%endmacro
%endif
;;; gf_vect_mad_avx2_gfni(len, vec, vec_i, mul_array, src, dest)
%define len arg0
%define vec arg1
%define vec_i arg2
%define mul_array arg3
%define src arg4
%define dest arg5
%define pos rax
%ifndef EC_ALIGNED_ADDR
;;; Use Un-aligned load/store
%define XLDR vmovdqu
%define XSTR vmovdqu
%else
;;; Use Non-temporal load/stor
%ifdef NO_NT_LDST
%define XLDR vmovdqa
%define XSTR vmovdqa
%else
%define XLDR vmovntdqa
%define XSTR vmovntdq
%endif
%endif
default rel
[bits 64]
section .text
%define x0 ymm0
%define xd ymm1
%define xgft1 ymm2
%define xret1 ymm3
;;
;; Encodes 32 bytes of a single source and updates single parity disk
;;
%macro ENCODE_32B 0
XLDR x0, [src + pos] ;Get next source vector
XLDR xd, [dest + pos] ;Get next dest vector
GF_MUL_XOR VEX, x0, xgft1, xret1, xd
XSTR [dest + pos], xd
%endmacro
;;
;; Encodes less than 32 bytes of a single source and updates single parity disk
;;
%macro ENCODE_LT_32B 1
%define %%LEN %1
simd_load_avx2 x0, src + pos, %%LEN, tmp, tmp2 ;Get next source vector
simd_load_avx2 xd, dest + pos, %%LEN, tmp, tmp2 ;Get next dest vector
GF_MUL_XOR VEX, x0, xgft1, xret1, xd
lea tmp, [dest + pos]
simd_store_avx2 tmp, xd, %%LEN, tmp2, pos ;Store updated encoded data
%endmacro
align 16
mk_global gf_vect_mad_avx2_gfni, function
func(gf_vect_mad_avx2_gfni)
FUNC_SAVE
xor pos, pos
shl vec_i, 3 ;Multiply by 8
vbroadcastsd xgft1, [vec_i + mul_array]
.loop32:
ENCODE_32B
add pos, 32 ;Loop on 32 bytes at a time
sub len, 32
cmp len, 32
jge .loop32
.len_lt_32:
cmp len, 0
jle .exit
ENCODE_LT_32B len
.exit:
vzeroupper
FUNC_RESTORE
ret
endproc_frame
%endif ; if AS_FEATURE_LEVEL >= 10