[Gmp-commit] /var/hg/gmp: 4 new changesets
mercurial at gmplib.org
mercurial at gmplib.org
Thu Feb 16 20:09:47 UTC 2017
details: /var/hg/gmp/rev/ffecb89c37be
changeset: 17272:ffecb89c37be
user: Torbjorn Granlund <tg at gmplib.org>
date: Thu Feb 16 18:23:41 2017 +0100
description:
Rewrite to use 4x unrolling.
details: /var/hg/gmp/rev/a34c7f47b324
changeset: 17273:a34c7f47b324
user: Torbjorn Granlund <tg at gmplib.org>
date: Thu Feb 16 18:27:46 2017 +0100
description:
Remove file since default code now performs similarly.
details: /var/hg/gmp/rev/663f9ca1b033
changeset: 17274:663f9ca1b033
user: Torbjorn Granlund <tg at gmplib.org>
date: Thu Feb 16 19:10:47 2017 +0100
description:
Provide arm64 addlsh1_n, sublsh1_n, rsblsh1_n, addlsh2_n, sublsh2_n, rsblsh2_n.
details: /var/hg/gmp/rev/f1416254f510
changeset: 17275:f1416254f510
user: Torbjorn Granlund <tg at gmplib.org>
date: Thu Feb 16 21:09:42 2017 +0100
description:
Tweak new aorsorrlshC_n.asm for some gains on A57, X-Gene.
diffstat:
mpn/arm64/aors_n.asm | 85 +++++++++++++++++---------
mpn/arm64/aorsorrlsh1_n.asm | 43 +++++++++++++
mpn/arm64/aorsorrlsh2_n.asm | 43 +++++++++++++
mpn/arm64/aorsorrlshC_n.asm | 139 ++++++++++++++++++++++++++++++++++++++++++++
mpn/arm64/xgene1/aors_n.asm | 100 -------------------------------
5 files changed, 281 insertions(+), 129 deletions(-)
diffs (truncated from 459 to 300 lines):
diff -r 65c543170956 -r f1416254f510 mpn/arm64/aors_n.asm
--- a/mpn/arm64/aors_n.asm Wed Feb 15 23:23:41 2017 +0100
+++ b/mpn/arm64/aors_n.asm Thu Feb 16 21:09:42 2017 +0100
@@ -33,9 +33,9 @@
include(`../config.m4')
C cycles/limb
-C Cortex-A53 3-3.5
-C Cortex-A57 2
-C X-Gene 2.5
+C Cortex-A53 2.75-3.25
+C Cortex-A57 1.5
+C X-Gene 2.0
changecom(blah)
@@ -49,14 +49,14 @@
define(`CLRCY', `cmn xzr, xzr')
define(`SETCY', `cmp $1, #1')
define(`RETVAL', `cset x0, cs')
- define(`func', mpn_add_n)
+ define(`func_n', mpn_add_n)
define(`func_nc', mpn_add_nc)')
ifdef(`OPERATION_sub_n', `
define(`ADDSUBC', sbcs)
define(`CLRCY', `cmp xzr, xzr')
define(`SETCY', `cmp xzr, $1')
define(`RETVAL', `cset x0, cc')
- define(`func', mpn_sub_n)
+ define(`func_n', mpn_sub_n)
define(`func_nc', mpn_sub_nc)')
MULFUNC_PROLOGUE(mpn_add_n mpn_add_nc mpn_sub_n mpn_sub_nc)
@@ -66,33 +66,60 @@
SETCY( x4)
b L(ent)
EPILOGUE()
-PROLOGUE(func)
+PROLOGUE(func_n)
CLRCY
-L(ent): tbz n, #0, L(b0)
-
- ldr x4, [up],#8
- ldr x6, [vp],#8
- sub n, n, #1
- ADDSUBC x8, x4, x6
- str x8, [rp],#8
- cbz n, L(rt)
+L(ent): lsr x18, n, #2
+ tbz n, #0, L(bx0)
-L(b0): ldp x4, x5, [up],#16
- ldp x6, x7, [vp],#16
- sub n, n, #2
- ADDSUBC x8, x4, x6
- ADDSUBC x9, x5, x7
- cbz n, L(end)
+L(bx1): ldr x7, [up]
+ ldr x11, [vp]
+ ADDSUBC x15, x7, x11
+ str x15, [rp],#8
+ tbnz n, #1, L(b11)
-L(top): ldp x4, x5, [up],#16
- ldp x6, x7, [vp],#16
- sub n, n, #2
- stp x8, x9, [rp],#16
- ADDSUBC x8, x4, x6
- ADDSUBC x9, x5, x7
- cbnz n, L(top)
+L(b01): cbz x18, L(ret)
+ ldp x4, x5, [up,#8]
+ ldp x8, x9, [vp,#8]
+ sub up, up, #8
+ sub vp, vp, #8
+ b L(mid)
-L(end): stp x8, x9, [rp]
-L(rt): RETVAL
+L(b11): ldp x6, x7, [up,#8]
+ ldp x10, x11, [vp,#8]
+ add up, up, #8
+ add vp, vp, #8
+ cbz x18, L(end)
+ b L(top)
+
+L(bx0): tbnz n, #1, L(b10)
+
+L(b00): ldp x4, x5, [up]
+ ldp x8, x9, [vp]
+ sub up, up, #16
+ sub vp, vp, #16
+ b L(mid)
+
+L(b10): ldp x6, x7, [up]
+ ldp x10, x11, [vp]
+ cbz x18, L(end)
+
+ ALIGN(16)
+L(top): ldp x4, x5, [up,#16]
+ ldp x8, x9, [vp,#16]
+ ADDSUBC x14, x6, x10
+ ADDSUBC x15, x7, x11
+ stp x14, x15, [rp],#16
+L(mid): ldp x6, x7, [up,#32]!
+ ldp x10, x11, [vp,#32]!
+ ADDSUBC x12, x4, x8
+ ADDSUBC x13, x5, x9
+ stp x12, x13, [rp],#16
+ sub x18, x18, #1
+ cbnz x18, L(top)
+
+L(end): ADDSUBC x14, x6, x10
+ ADDSUBC x15, x7, x11
+ stp x14, x15, [rp]
+L(ret): RETVAL
ret
EPILOGUE()
diff -r 65c543170956 -r f1416254f510 mpn/arm64/aorsorrlsh1_n.asm
--- /dev/null Thu Jan 01 00:00:00 1970 +0000
+++ b/mpn/arm64/aorsorrlsh1_n.asm Thu Feb 16 21:09:42 2017 +0100
@@ -0,0 +1,43 @@
+dnl ARM64 mpn_addlsh1_n, mpn_sublsh1_n, mpn_rsblsh1_n.
+
+dnl Copyright 2017 Free Software Foundation, Inc.
+
+dnl This file is part of the GNU MP Library.
+dnl
+dnl The GNU MP Library is free software; you can redistribute it and/or modify
+dnl it under the terms of either:
+dnl
+dnl * the GNU Lesser General Public License as published by the Free
+dnl Software Foundation; either version 3 of the License, or (at your
+dnl option) any later version.
+dnl
+dnl or
+dnl
+dnl * the GNU General Public License as published by the Free Software
+dnl Foundation; either version 2 of the License, or (at your option) any
+dnl later version.
+dnl
+dnl or both in parallel, as here.
+dnl
+dnl The GNU MP Library is distributed in the hope that it will be useful, but
+dnl WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+dnl or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
+dnl for more details.
+dnl
+dnl You should have received copies of the GNU General Public License and the
+dnl GNU Lesser General Public License along with the GNU MP Library. If not,
+dnl see https://www.gnu.org/licenses/.
+
+include(`../config.m4')
+
+
+define(LSH, 1)
+define(RSH, 63)
+
+ifdef(`OPERATION_addlsh1_n',`define(`DO_add')')
+ifdef(`OPERATION_sublsh1_n',`define(`DO_sub')')
+ifdef(`OPERATION_rsblsh1_n',`define(`DO_rsb')')
+
+MULFUNC_PROLOGUE(mpn_addlsh1_n mpn_sublsh1_n mpn_rsblsh1_n)
+
+include_mpn(`arm64/aorsorrlshC_n.asm')
diff -r 65c543170956 -r f1416254f510 mpn/arm64/aorsorrlsh2_n.asm
--- /dev/null Thu Jan 01 00:00:00 1970 +0000
+++ b/mpn/arm64/aorsorrlsh2_n.asm Thu Feb 16 21:09:42 2017 +0100
@@ -0,0 +1,43 @@
+dnl ARM64 mpn_addlsh2_n, mpn_sublsh2_n, mpn_rsblsh2_n.
+
+dnl Copyright 2017 Free Software Foundation, Inc.
+
+dnl This file is part of the GNU MP Library.
+dnl
+dnl The GNU MP Library is free software; you can redistribute it and/or modify
+dnl it under the terms of either:
+dnl
+dnl * the GNU Lesser General Public License as published by the Free
+dnl Software Foundation; either version 3 of the License, or (at your
+dnl option) any later version.
+dnl
+dnl or
+dnl
+dnl * the GNU General Public License as published by the Free Software
+dnl Foundation; either version 2 of the License, or (at your option) any
+dnl later version.
+dnl
+dnl or both in parallel, as here.
+dnl
+dnl The GNU MP Library is distributed in the hope that it will be useful, but
+dnl WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+dnl or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
+dnl for more details.
+dnl
+dnl You should have received copies of the GNU General Public License and the
+dnl GNU Lesser General Public License along with the GNU MP Library. If not,
+dnl see https://www.gnu.org/licenses/.
+
+include(`../config.m4')
+
+
+define(LSH, 2)
+define(RSH, 62)
+
+ifdef(`OPERATION_addlsh2_n',`define(`DO_add')')
+ifdef(`OPERATION_sublsh2_n',`define(`DO_sub')')
+ifdef(`OPERATION_rsblsh2_n',`define(`DO_rsb')')
+
+MULFUNC_PROLOGUE(mpn_addlsh2_n mpn_sublsh2_n mpn_rsblsh2_n)
+
+include_mpn(`arm64/aorsorrlshC_n.asm')
diff -r 65c543170956 -r f1416254f510 mpn/arm64/aorsorrlshC_n.asm
--- /dev/null Thu Jan 01 00:00:00 1970 +0000
+++ b/mpn/arm64/aorsorrlshC_n.asm Thu Feb 16 21:09:42 2017 +0100
@@ -0,0 +1,139 @@
+dnl ARM64 mpn_addlshC_n, mpn_sublshC_n, mpn_rsblshC_n.
+
+dnl Contributed to the GNU project by Torbjörn Granlund.
+
+dnl Copyright 2017 Free Software Foundation, Inc.
+
+dnl This file is part of the GNU MP Library.
+dnl
+dnl The GNU MP Library is free software; you can redistribute it and/or modify
+dnl it under the terms of either:
+dnl
+dnl * the GNU Lesser General Public License as published by the Free
+dnl Software Foundation; either version 3 of the License, or (at your
+dnl option) any later version.
+dnl
+dnl or
+dnl
+dnl * the GNU General Public License as published by the Free Software
+dnl Foundation; either version 2 of the License, or (at your option) any
+dnl later version.
+dnl
+dnl or both in parallel, as here.
+dnl
+dnl The GNU MP Library is distributed in the hope that it will be useful, but
+dnl WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+dnl or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
+dnl for more details.
+dnl
+dnl You should have received copies of the GNU General Public License and the
+dnl GNU Lesser General Public License along with the GNU MP Library. If not,
+dnl see https://www.gnu.org/licenses/.
+
+include(`../config.m4')
+
+C cycles/limb
+C Cortex-A53 3.25-3.75
+C Cortex-A57 2.18
+C X-Gene 2.5
+
+changecom(blah)
+
+define(`rp', `x0')
+define(`up', `x1')
+define(`vp', `x2')
+define(`n', `x3')
+
+ifdef(`DO_add', `
+ define(`ADDSUB', `adds $1, $2, $3')
+ define(`ADDSUBC', `adcs $1, $2, $3')
+ define(`CLRRCY', `adds $1, xzr, xzr')
+ define(`RETVAL', `adc x0, $1, xzr')
+ define(`func_n', mpn_addlsh`'LSH`'_n)')
+ifdef(`DO_sub', `
+ define(`ADDSUB', `subs $1, $3, $2')
+ define(`ADDSUBC', `sbcs $1, $3, $2')
+ define(`CLRRCY', `subs $1, xzr, xzr')
+ define(`RETVAL', `cinc x0, $1, cc')
+ define(`func_n', mpn_sublsh`'LSH`'_n)')
+ifdef(`DO_rsb', `
+ define(`ADDSUB', `subs $1, $2, $3')
+ define(`ADDSUBC', `sbcs $1, $2, $3')
+ define(`CLRRCY', `subs $1, xzr, xzr')
+ define(`RETVAL', `sbc x0, $1, xzr')
+ define(`func_n', mpn_rsblsh`'LSH`'_n)')
+
+ASM_START()
+PROLOGUE(func_n)
+ lsr x18, n, #2
+ tbz n, #0, L(bx0)
+
+L(bx1): ldr x5, [up]
+ tbnz n, #1, L(b11)
+
+L(b01): ldr x11, [vp]
+ cbz x18, L(1)
+ ldp x8, x9, [vp,#8]
+ lsl x13, x11, #LSH
+ ADDSUB( x15, x13, x5)
+ str x15, [rp],#8
+ sub up, up, #24
+ sub vp, vp, #8
+ b L(mid)
+
+L(1): lsl x13, x11, #LSH
More information about the gmp-commit
mailing list