[Gmp-commit] /var/hg/gmp: 3 new changesets
mercurial at gmplib.org
mercurial at gmplib.org
Sat Feb 11 02:40:48 UTC 2017
details: /var/hg/gmp/rev/67ec2317e891
changeset: 17244:67ec2317e891
user: Torbjorn Granlund <tg at gmplib.org>
date: Fri Feb 10 18:49:37 2017 +0100
description:
(EPILOGUE_cpu): Zap lea_list to avoid repetition.
details: /var/hg/gmp/rev/62969fff192e
changeset: 17245:62969fff192e
user: Torbjorn Granlund <tg at gmplib.org>
date: Fri Feb 10 18:50:32 2017 +0100
description:
New file, based on dive_1.asm.
details: /var/hg/gmp/rev/4ab74d8ba4d5
changeset: 17246:4ab74d8ba4d5
user: Torbjorn Granlund <tg at gmplib.org>
date: Sat Feb 11 03:40:32 2017 +0100
description:
New file.
diffstat:
mpn/arm/arm-defs.m4 | 3 +-
mpn/arm/v6/bdiv_q_1.asm | 164 ++++++++++++++++++++++++++++++++++++++++++++++++
mpn/arm64/bdiv_q_1.asm | 128 +++++++++++++++++++++++++++++++++++++
3 files changed, 294 insertions(+), 1 deletions(-)
diffs (truncated from 312 to 300 lines):
diff -r ca8b7ecb35ee -r 4ab74d8ba4d5 mpn/arm/arm-defs.m4
--- a/mpn/arm/arm-defs.m4 Fri Feb 10 02:06:27 2017 +0100
+++ b/mpn/arm/arm-defs.m4 Sat Feb 11 03:40:32 2017 +0100
@@ -89,6 +89,7 @@
define(`EPILOGUE_cpu',
`lea_list
- SIZE(`$1',.-`$1')')
+ SIZE(`$1',.-`$1')'
+`define(`lea_list', `')')
divert
diff -r ca8b7ecb35ee -r 4ab74d8ba4d5 mpn/arm/v6/bdiv_q_1.asm
--- /dev/null Thu Jan 01 00:00:00 1970 +0000
+++ b/mpn/arm/v6/bdiv_q_1.asm Sat Feb 11 03:40:32 2017 +0100
@@ -0,0 +1,164 @@
+dnl ARM v6 mpn_bdiv_q_1
+
+dnl Contributed to the GNU project by Torbjörn Granlund.
+
+dnl Copyright 2012, 2013, 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 cycles/limb
+C norm unorm modexact_1c_odd
+C StrongARM - -
+C XScale - -
+C Cortex-A7 ? ?
+C Cortex-A8 ? ?
+C Cortex-A9 9 10 9
+C Cortex-A15 7 7 7
+
+C Architecture requirements:
+C v5 -
+C v5t clz
+C v5te -
+C v6 umaal
+C v6t2 -
+C v7a -
+
+define(`rp', `r0')
+define(`up', `r1')
+define(`n', `r2')
+define(`d', `r3')
+define(`di_arg', `sp[0]')
+define(`cnt_arg', `sp[4]')
+
+define(`cy', `r7')
+define(`cnt', `r6')
+define(`tnc', `r4')
+
+ASM_START()
+PROLOGUE(mpn_bdiv_q_1)
+ push {r6,r7,r8,r9,r10,r11}
+
+ tst d, #1
+
+ rsb r10, d, #0
+ and r10, r10, d
+ clz r10, r10
+ rsb cnt, r10, #31 C count_trailing_zeros
+ mov d, d, lsr cnt
+
+C binvert limb
+ LEA( r10, binvert_limb_table)
+ and r12, d, #254
+ ldrb r10, [r10, r12, lsr #1]
+ mul r12, r10, r10
+ mul r12, d, r12
+ rsb r12, r12, r10, lsl #1
+ mul r10, r12, r12
+ mul r10, d, r10
+ rsb r10, r10, r12, lsl #1 C r10 = inverse
+
+ ldr r11, [up], #4 C up[0]
+ mov cy, #0
+ rsb r8, r10, #0 C r8 = -inverse
+ bne L(norm)
+ b L(unnorm)
+EPILOGUE()
+
+PROLOGUE(mpn_pi1_bdiv_q_1)
+ push {r6,r7,r8,r9,r10,r11}
+
+ ldr cnt, [sp, #28]
+ ldr r10, [sp, #24]
+ cmp cnt, #0
+
+ ldr r11, [up], #4 C up[0]
+ mov cy, #0
+ rsb r8, r10, #0 C r8 = -inverse
+
+ bne L(unnorm)
+
+L(norm):
+ subs n, n, #1
+ mul r11, r11, r10
+ beq L(end)
+
+ ALIGN(16)
+L(top): ldr r9, [up], #4
+ mov r12, #0
+ str r11, [rp], #4
+ umaal r12, cy, r11, d
+ mul r11, r9, r10
+ mla r11, cy, r8, r11
+ subs n, n, #1
+ bne L(top)
+
+L(end): str r11, [rp]
+ pop {r10,r11,r6,r7,r8,r9}
+ bx r14
+
+L(unnorm):
+ push {r4,r5}
+ rsb tnc, cnt, #32
+ mov r5, r11, lsr cnt
+ subs n, n, #1
+ beq L(edx)
+
+ ldr r12, [up], #4
+ orr r9, r5, r12, lsl tnc
+ mov r5, r12, lsr cnt
+ mul r11, r9, r10
+ subs n, n, #1
+ beq L(edu)
+
+ ALIGN(16)
+L(tpu): ldr r12, [up], #4
+ orr r9, r5, r12, lsl tnc
+ mov r5, r12, lsr cnt
+ mov r12, #0
+ str r11, [rp], #4
+ umaal r12, cy, r11, d
+ mul r11, r9, r10
+ mla r11, cy, r8, r11
+ subs n, n, #1
+ bne L(tpu)
+
+L(edu): str r11, [rp], #4
+ mov r12, #0
+ umaal r12, cy, r11, d
+ mul r11, r5, r10
+ mla r11, cy, r8, r11
+ str r11, [rp]
+ pop {r4,r5,r6,r7,r8,r9,r10,r11}
+ bx r14
+
+L(edx): mul r11, r5, r10
+ str r11, [rp]
+ pop {r4,r5,r6,r7,r8,r9,r10,r11}
+ bx r14
+EPILOGUE()
diff -r ca8b7ecb35ee -r 4ab74d8ba4d5 mpn/arm64/bdiv_q_1.asm
--- /dev/null Thu Jan 01 00:00:00 1970 +0000
+++ b/mpn/arm64/bdiv_q_1.asm Sat Feb 11 03:40:32 2017 +0100
@@ -0,0 +1,128 @@
+dnl ARM64 mpn_bdiv_q_1, mpn_pi1_bdiv_q_1 -- Hensel division by 1-limb divisor.
+
+dnl Contributed to the GNU project by Torbjörn Granlund.
+
+dnl Copyright 2012, 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 norm unorm
+C Cortex-A53 12 15
+C Cortex-A57 12 12
+C Cortex-A72
+C Cortex-A73
+C X-Gene 11 11
+
+C TODO
+C * Scheduling of umulh later in the unorm loop brings A53 time to 12 c/l.
+C Unfortunately, that requires software pipelining.
+
+define(`rp', `x0')
+define(`up', `x1')
+define(`n', `x2')
+define(`d', `x3')
+define(`di', `x4') C just mpn_pi1_bdiv_q_1
+define(`cnt', `x5') C just mpn_pi1_bdiv_q_1
+
+define(`cy', `r7')
+define(`tnc', `x8')
+
+ASM_START()
+PROLOGUE(mpn_bdiv_q_1)
+
+ rbit x6, d
+ clz cnt, x6
+ lsr d, d, cnt
+
+ifdef(`PIC',`
+ adrp x7, :got:__gmp_binvert_limb_table
+ ubfx x6, d, 1, 7
+ ldr x7, [x7, #:got_lo12:__gmp_binvert_limb_table]
+',`
+ adrp x7, __gmp_binvert_limb_table
+ ubfx x6, d, 1, 7
+ add x7, x7, :lo12:__gmp_binvert_limb_table
+')
+ ldrb w6, [x7, x6]
+ ubfiz x7, x6, 1, 8
+ umull x6, w6, w6
+ msub x6, x6, d, x7
+ lsl x7, x6, 1
+ mul x6, x6, x6
+ msub x6, x6, d, x7
+ lsl x7, x6, 1
+ mul x6, x6, x6
+ msub di, x6, d, x7
+
+ b mpn_pi1_bdiv_q_1
+EPILOGUE()
+
+PROLOGUE(mpn_pi1_bdiv_q_1)
+ add n, n, #-1
+ subs x6, x6, x6 C clear r6 and C flag
+ ldr x9, [up],#8
+ cbz cnt, L(norm)
+
+L(unorm):
+ lsr x12, x9, cnt
+ cbz n, L(eu1)
+ sub tnc, xzr, cnt
+
+L(tpu): ldr x9, [up],#8
+ lsl x7, x9, tnc
+ orr x7, x7, x12
+ sbcs x6, x7, x6
+ mul x7, x6, di
+ str x7, [rp],#8
+ lsr x12, x9, cnt
+ umulh x6, x7, d
+ add n, n, #-1
+ cbnz n, L(tpu)
+
+L(eu1): sbcs x6, x12, x6
+ mul x6, x6, di
+ str x6, [rp]
+ ret
+
+L(norm):
+ mul x5, x9, di
+ str x5, [rp],#8
More information about the gmp-commit
mailing list