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: update GF_MUL_XOR macro to support VEX encoding

Browse Source 
Signed-off-by: Marcel Cornu <marcel.d.cornu@intel.com>
This commit is contained in:
Marcel Cornu 2023-11-29 16:19:34 +00:00 committed by Pablo de Lara
parent c8dd92f04a
commit bc34d87427
13 changed files with 38 additions and 33 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
erasure_code/gf_2vect_dot_prod_avx512_gfni.asm
View File

@ -152,7 +152,7 @@ section .text
vbroadcastf32x2 xgft2, [tmp + vec]
add tmp, 8
GF_MUL_XOR x0, xgft1, xgft1, xp1, xgft2, xgft2, xp2
GF_MUL_XOR EVEX, x0, xgft1, xgft1, xp1, xgft2, xgft2, xp2
cmp vec_i, vec
jl %%next_vect

2
erasure_code/gf_2vect_mad_avx512_gfni.asm
View File

@ -133,7 +133,7 @@ section .text
XLDR xd2, [dest2 + pos] ;Get next dest vector
%endif
GF_MUL_XOR x0, xgft1, xret1, xd1, xgft2, xret2, xd2
GF_MUL_XOR EVEX, x0, xgft1, xret1, xd1, xgft2, xret2, xd2
%if %0 == 1
vmovdqu8 [dest1 + pos]{%%KMASK}, xd1

2
erasure_code/gf_3vect_dot_prod_avx512_gfni.asm
View File

@ -165,7 +165,7 @@ section .text
vbroadcastf32x2 xgft3, [tmp + vec*2]
add tmp, 8
GF_MUL_XOR x0, xgft1, xgft1, xp1, xgft2, xgft2, xp2, xgft3, xgft3, xp3
GF_MUL_XOR EVEX, x0, xgft1, xgft1, xp1, xgft2, xgft2, xp2, xgft3, xgft3, xp3
cmp vec_i, vec
jl %%next_vect

2
erasure_code/gf_3vect_mad_avx512_gfni.asm
View File

@ -144,7 +144,7 @@ section .text
XLDR xd3, [dest3 + pos] ;Get next dest vector
%endif
GF_MUL_XOR x0, xgft1, xret1, xd1, xgft2, xret2, xd2, xgft3, xret3, xd3
GF_MUL_XOR EVEX, x0, xgft1, xret1, xd1, xgft2, xret2, xd2, xgft3, xret3, xd3
%if %0 == 1
vmovdqu8 [dest1 + pos]{%%KMASK}, xd1

2
erasure_code/gf_4vect_dot_prod_avx512_gfni.asm
View File

@ -187,7 +187,7 @@ section .text
vbroadcastf32x2 xgft4, [tmp + vskip3]
add tmp, 8
GF_MUL_XOR x0, xgft1, xgft1, xp1, xgft2, xgft2, xp2, xgft3, xgft3, xp3, \
GF_MUL_XOR EVEX, x0, xgft1, xgft1, xp1, xgft2, xgft2, xp2, xgft3, xgft3, xp3, \
xgft4, xgft4, xp4
cmp vec_i, vec

2
erasure_code/gf_4vect_mad_avx512_gfni.asm
View File

@ -157,7 +157,7 @@ section .text
XLDR xd4, [dest4 + pos] ;Get next dest vector
%endif
GF_MUL_XOR x0, xgft1, xret1, xd1, xgft2, xret2, xd2, xgft3, xret3, xd3, \
GF_MUL_XOR EVEX, x0, xgft1, xret1, xd1, xgft2, xret2, xd2, xgft3, xret3, xd3, \
xgft4, xret4, xd4
%if %0 == 1

2
erasure_code/gf_5vect_dot_prod_avx512_gfni.asm
View File

@ -202,7 +202,7 @@ section .text
vbroadcastf32x2 xgft5, [tmp + vec*4]
add tmp, 8
GF_MUL_XOR x0, xgft1, xgft1, xp1, xgft2, xgft2, xp2, xgft3, xgft3, xp3, \
GF_MUL_XOR EVEX, x0, xgft1, xgft1, xp1, xgft2, xgft2, xp2, xgft3, xgft3, xp3, \
xgft4, xgft4, xp4, xgft5, xgft5, xp5
cmp vec_i, vec

2
erasure_code/gf_5vect_mad_avx512_gfni.asm
View File

@ -171,7 +171,7 @@ section .text
XLDR xd5, [dest5 + pos] ;Get next dest vector
%endif
GF_MUL_XOR x0, xgft1, xret1, xd1, xgft2, xret2, xd2, xgft3, xret3, xd3, \
GF_MUL_XOR EVEX, x0, xgft1, xret1, xd1, xgft2, xret2, xd2, xgft3, xret3, xd3, \
xgft4, xret4, xd4, xgft5, xret5, xd5
%if %0 == 1

2
erasure_code/gf_6vect_dot_prod_avx512_gfni.asm
View File

@ -217,7 +217,7 @@ section .text
vbroadcastf32x2 xgft6, [tmp + vskip5]
add tmp, 8
GF_MUL_XOR x0, xgft1, xgft1, xp1, xgft2, xgft2, xp2, xgft3, xgft3, xp3, \
GF_MUL_XOR EVEX, x0, xgft1, xgft1, xp1, xgft2, xgft2, xp2, xgft3, xgft3, xp3, \
xgft4, xgft4, xp4, xgft5, xgft5, xp5, xgft6, xgft6, xp6
cmp vec_i, vec

2
erasure_code/gf_6vect_mad_avx512_gfni.asm
View File

@ -185,7 +185,7 @@ section .text
XLDR xd6, [dest6 + pos] ;Get next dest vector
%endif
GF_MUL_XOR x0, xgft1, xret1, xd1, xgft2, xret2, xd2, xgft3, xret3, xd3, \
GF_MUL_XOR EVEX, x0, xgft1, xret1, xd1, xgft2, xret2, xd2, xgft3, xret3, xd3, \
xgft4, xret4, xd4, xgft5, xret5, xd5, xgft6, xret6, xd6
%if %0 == 1

2
erasure_code/gf_vect_dot_prod_avx512_gfni.asm
View File

@ -138,7 +138,7 @@ section .text
vbroadcastf32x2 xgft1, [tmp]
add tmp, 8
GF_MUL_XOR x0, xgft1, xgft1, xp1
GF_MUL_XOR EVEX, x0, xgft1, xgft1, xp1
cmp vec_i, vec
jl %%next_vect

47
erasure_code/gf_vect_gfni.inc
View File

@ -31,28 +31,29 @@
; Multiply 1 source register to up to 6 different GF table registers
; and XOR the results to partial registers
;
%macro GF_MUL_XOR 4-19
%define %%SRC %1
%define %%GFTABLE1 %2
%define %%TMP1 %3
%define %%PARTIAL1 %4
%define %%GFTABLE2 %5
%define %%TMP2 %6
%define %%PARTIAL2 %7
%define %%GFTABLE3 %8
%define %%TMP3 %9
%define %%PARTIAL3 %10
%define %%GFTABLE4 %11
%define %%TMP4 %12
%define %%PARTIAL4 %13
%define %%GFTABLE5 %14
%define %%TMP5 %15
%define %%PARTIAL5 %16
%define %%GFTABLE6 %17
%define %%TMP6 %18
%define %%PARTIAL6 %19
%macro GF_MUL_XOR 5-20
%define %%ENCODING %1
%define %%SRC %2
%define %%GFTABLE1 %3
%define %%TMP1 %4
%define %%PARTIAL1 %5
%define %%GFTABLE2 %6
%define %%TMP2 %7
%define %%PARTIAL2 %8
%define %%GFTABLE3 %9
%define %%TMP3 %10
%define %%PARTIAL3 %11
%define %%GFTABLE4 %12
%define %%TMP4 %13
%define %%PARTIAL4 %14
%define %%GFTABLE5 %15
%define %%TMP5 %16
%define %%PARTIAL5 %17
%define %%GFTABLE6 %18
%define %%TMP6 %19
%define %%PARTIAL6 %20
%define %%N_BLOCKS ((%0 - 1) / 3)
%define %%N_BLOCKS ((%0 - 2) / 3)
%assign %%I 1
%rep %%N_BLOCKS
@ -61,7 +62,11 @@
%endrep
%assign %%I 1
%rep %%N_BLOCKS
%ifidn %%ENCODING, VEX
vpxor %%PARTIAL %+ %%I, %%TMP %+ %%I
%else
vpxorq %%PARTIAL %+ %%I, %%TMP %+ %%I
%endif
%assign %%I (%%I + 1)
%endrep
%endmacro

2
erasure_code/gf_vect_mad_avx512_gfni.asm
View File

@ -126,7 +126,7 @@ section .text
XLDR xd, [dest + pos] ;Get next dest vector
%endif
GF_MUL_XOR x0, xgft1, xret1, xd
GF_MUL_XOR EVEX, x0, xgft1, xret1, xd
%if %0 == 1
vmovdqu8 [dest + pos]{%%KMASK}, xd