[Gmp-commit] /var/hg/gmp: 2 new changesets
mercurial at gmplib.org
mercurial at gmplib.org
Sun Apr 22 21:03:07 CEST 2012
details: /var/hg/gmp/rev/93c2ed5f26fe
changeset: 14879:93c2ed5f26fe
user: Torbjorn Granlund <tege at gmplib.org>
date: Sun Apr 22 20:46:16 2012 +0200
description:
*** empty log message ***
details: /var/hg/gmp/rev/4717833d2b43
changeset: 14880:4717833d2b43
user: Torbjorn Granlund <tege at gmplib.org>
date: Sun Apr 22 21:03:03 2012 +0200
description:
New ARM files.
diffstat:
ChangeLog | 7 ++
mpn/arm/README | 18 +-----
mpn/arm/bdiv_dbm1c.asm | 101 ++++++++++++++++++++++++++++++++++++++++
mpn/arm/v6/addmul_1.asm | 119 ++++++++++++++++++++++++++++++++++++++++++++++++
mpn/arm/v6/mul_1.asm | 108 +++++++++++++++++++++++++++++++++++++++++++
5 files changed, 339 insertions(+), 14 deletions(-)
diffs (truncated from 383 to 300 lines):
diff -r 7f5429959ef8 -r 4717833d2b43 ChangeLog
--- a/ChangeLog Sun Apr 22 12:23:10 2012 +0200
+++ b/ChangeLog Sun Apr 22 21:03:03 2012 +0200
@@ -1,3 +1,10 @@
+2012-04-22 Torbjorn Granlund <tege at gmplib.org>
+
+ * mpn/arm/bdiv_dbm1c.asm: New file.
+
+ * mpn/arm/v6/mul_1.asm: New file.
+ * mpn/arm/v6/addmul_1.asm: New file.
+
2012-04-22 Marco Bodrato <bodrato at mail.dm.unipi.it>
* gen-fac.c: Renamed, was gen-fac_ui.c .
diff -r 7f5429959ef8 -r 4717833d2b43 mpn/arm/README
--- a/mpn/arm/README Sun Apr 22 12:23:10 2012 +0200
+++ b/mpn/arm/README Sun Apr 22 21:03:03 2012 +0200
@@ -1,4 +1,4 @@
-Copyright 2002 Free Software Foundation, Inc.
+Copyright 2002, 2012 Free Software Foundation, Inc.
This file is part of the GNU MP Library.
@@ -19,16 +19,6 @@
-This directory contains mpn functions for ARM processors.
-It has been optimized for StrongARM.
-
-TODO
-
-Write mpn_addmul_2. The speed of mpn_addmul_1 is 9.75 c/l;
-mpn_addmul_2 could run at 8 c/l. mpn_addmul_N could
-approach 6 c/l, but register shortage will make this hard.
-
-Perhaps nails is the way to go even for an embedded processor like
-this, since the umlal accumulation could be used very effectively in
-that case. with just 2 nail bits, we should get close to 5 c/l for a
-mpn_addmul_N or mpn_mul_basecase.
+This directory contains mpn functions for ARM processors. It has been
+optimised for Cortex-A9, but the code in the top-level directory should run
+on all ARM processors at architecture level v4 or later.
diff -r 7f5429959ef8 -r 4717833d2b43 mpn/arm/bdiv_dbm1c.asm
--- /dev/null Thu Jan 01 00:00:00 1970 +0000
+++ b/mpn/arm/bdiv_dbm1c.asm Sun Apr 22 21:03:03 2012 +0200
@@ -0,0 +1,101 @@
+dnl ARM mpn_bdiv_dbm1c.
+
+dnl Copyright 2008, 2011, 2012 Free Software Foundation, Inc.
+
+dnl This file is part of the GNU MP Library.
+
+dnl The GNU MP Library is free software; you can redistribute it and/or modify
+dnl it under the terms of the GNU Lesser General Public License as published
+dnl by the Free Software Foundation; either version 3 of the License, or (at
+dnl your option) any later version.
+
+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 Lesser General Public
+dnl License for more details.
+
+dnl You should have received a copy of the GNU Lesser General Public License
+dnl along with the GNU MP Library. If not, see http://www.gnu.org/licenses/.
+
+include(`../config.m4')
+
+C cycles/limb
+C StrongARM ?
+C XScale ?
+C Cortex-A8 ?
+C Cortex-A9 4.25
+C Cortex-A15 ?
+
+C TODO
+C * Try using umlal or umaal.
+C * Try using ldm/stm.
+
+define(`qp', `r0')
+define(`up', `r1')
+define(`n', `r2')
+define(`bd', `r3')
+define(`cy', `sp,#0')
+
+ASM_START()
+ TEXT
+ ALIGN(16)
+PROLOGUE(mpn_bdiv_dbm1c)
+ push {r4, r5, r6, r7, r8, r9}
+ ldr r4, [up], #4
+ ldr r5, [sp, #24]
+ ands r9, n, #3
+ beq L(fi0)
+ cmp r9, #2
+ bcc L(fi1)
+ beq L(fi2)
+
+L(fi3): umull r8, r9, r4, bd
+ ldr r4, [up], #4
+ b L(lo3)
+
+L(fi0): umull r6, r7, r4, bd
+ ldr r4, [up], #4
+ b L(lo0)
+
+L(fi1): subs n, n, #1
+ umull r8, r9, r4, bd
+ bls L(wd1)
+ ldr r4, [up], #4
+ b L(lo1)
+
+L(fi2): umull r6, r7, r4, bd
+ ldr r4, [up], #4
+ b L(lo2)
+
+L(top): ldr r4, [up], #4
+ subs r5, r5, r6
+ str r5, [qp], #4
+ sbc r5, r5, r7
+L(lo1): umull r6, r7, r4, bd
+ ldr r4, [up], #4
+ subs r5, r5, r8
+ str r5, [qp], #4
+ sbc r5, r5, r9
+L(lo0): umull r8, r9, r4, bd
+ ldr r4, [up], #4
+ subs r5, r5, r6
+ str r5, [qp], #4
+ sbc r5, r5, r7
+L(lo3): umull r6, r7, r4, bd
+ ldr r4, [up], #4
+ subs r5, r5, r8
+ str r5, [qp], #4
+ sbc r5, r5, r9
+L(lo2): subs n, n, #4
+ umull r8, r9, r4, bd
+ bhi L(top)
+
+L(wd2): subs r5, r5, r6
+ str r5, [qp], #4
+ sbc r5, r5, r7
+L(wd1): subs r5, r5, r8
+ str r5, [qp]
+ sbc r0, r5, r9
+ pop {r4, r5, r6, r7, r8, r9}
+ bx lr
+EPILOGUE()
diff -r 7f5429959ef8 -r 4717833d2b43 mpn/arm/v6/addmul_1.asm
--- /dev/null Thu Jan 01 00:00:00 1970 +0000
+++ b/mpn/arm/v6/addmul_1.asm Sun Apr 22 21:03:03 2012 +0200
@@ -0,0 +1,119 @@
+dnl ARM mpn_addmul_1.
+
+dnl Copyright 2012 Free Software Foundation, Inc.
+
+dnl This file is part of the GNU MP Library.
+
+dnl The GNU MP Library is free software; you can redistribute it and/or modify
+dnl it under the terms of the GNU Lesser General Public License as published
+dnl by the Free Software Foundation; either version 3 of the License, or (at
+dnl your option) any later version.
+
+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 Lesser General Public
+dnl License for more details.
+
+dnl You should have received a copy of the GNU Lesser General Public License
+dnl along with the GNU MP Library. If not, see http://www.gnu.org/licenses/.
+
+include(`../config.m4')
+
+C cycles/limb
+C StrongARM: -
+C XScale -
+C Cortex-A8 ?
+C Cortex-A9 3.25
+C Cortex-A15 ?
+
+C TODO
+C * Micro-optimise feed-in code.
+C * Optimise for n=1,2 by delaying register saving.
+C * Try using ldm/stm.
+
+define(`rp',`r0')
+define(`up',`r1')
+define(`n', `r2')
+define(`v0',`r3')
+
+ASM_START()
+PROLOGUE(mpn_addmul_1)
+ stmfd sp!, { r4, r5, r6, r7, r8 }
+
+ ands r12, n, #3
+ beq L(fi0)
+ cmp r12, #2
+ bcc L(fi1)
+ beq L(fi2)
+
+L(fi3): ldr r4, [up], #4
+ ldr r12, [rp, #0]
+ ldr r5, [up], #4
+ ldr r6, [rp, #4]
+ mov r8, #0
+ umaal r8, r12, r4, v0
+ b L(lo3)
+
+L(fi0): ldr r5, [up], #4
+ ldr r8, [rp, #0]
+ ldr r4, [up], #4
+ ldr r12, [rp, #4]
+ mov r7, #0
+ umaal r7, r8, r5, v0
+ add rp, rp, #4
+ b L(lo0)
+
+L(fi1): ldr r4, [up], #4
+ ldr r7, [rp, #0]
+ subs n, n, #1
+ mov r6, #0
+ beq L(1)
+ ldr r5, [up], #4
+ ldr r8, [rp, #4]
+ umaal r6, r7, r4, v0
+ add rp, rp, #8
+ b L(lo1)
+
+L(fi2): ldr r5, [up], #4
+ ldr r6, [rp, #0]
+ ldr r4, [up], #4
+ ldr r7, [rp, #4]
+ mov r12, #0
+ umaal r12, r6, r5, v0
+ subs n, n, #2
+ add rp, rp, #12
+ beq L(end)
+
+ ALIGN(16)
+L(top): ldr r5, [up], #4
+ ldr r8, [rp, #-4]
+ umaal r6, r7, r4, v0
+ str r12, [rp, #-12]
+L(lo1): ldr r4, [up], #4
+ ldr r12, [rp, #0]
+ umaal r7, r8, r5, v0
+ str r6, [rp, #-8]
+L(lo0): ldr r5, [up], #4
+ ldr r6, [rp, #4]
+ umaal r8, r12, r4, v0
+ str r7, [rp, #-4]
+L(lo3): ldr r4, [up], #4
+ ldr r7, [rp, #8]
+ umaal r12, r6, r5, v0
+ str r8, [rp], #16
+ subs n, n, #4
+ bhi L(top)
+
+L(end): umaal r6, r7, r4, v0
+ str r12, [rp, #-12]
+ str r6, [rp, #-8]
+ mov r0, r7
+ ldmfd sp!, { r4, r5, r6, r7, r8 }
+ bx lr
+
+L(1): umaal r6, r7, r4, v0
+ str r6, [rp, #0]
+ mov r0, r7
+ ldmfd sp!, { r4, r5, r6, r7, r8 }
+ bx lr
+EPILOGUE()
diff -r 7f5429959ef8 -r 4717833d2b43 mpn/arm/v6/mul_1.asm
--- /dev/null Thu Jan 01 00:00:00 1970 +0000
+++ b/mpn/arm/v6/mul_1.asm Sun Apr 22 21:03:03 2012 +0200
@@ -0,0 +1,108 @@
+dnl ARM mpn_mul_1.
+
+dnl Copyright 2012 Free Software Foundation, Inc.
+
+dnl This file is part of the GNU MP Library.
+
+dnl The GNU MP Library is free software; you can redistribute it and/or modify
+dnl it under the terms of the GNU Lesser General Public License as published
+dnl by the Free Software Foundation; either version 3 of the License, or (at
+dnl your option) any later version.
+
+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 Lesser General Public
+dnl License for more details.
+
+dnl You should have received a copy of the GNU Lesser General Public License
+dnl along with the GNU MP Library. If not, see http://www.gnu.org/licenses/.
+
+include(`../config.m4')
+
+C cycles/limb
+C StrongARM: -
+C XScale -
+C Cortex-A8 ?
More information about the gmp-commit
mailing list