[Gmp-commit] /var/hg/gmp: 5 new changesets
mercurial at gmplib.org
mercurial at gmplib.org
Thu Oct 27 15:48:50 CEST 2011
details: /var/hg/gmp/rev/e14bf9a101a2
changeset: 14385:e14bf9a101a2
user: Torbjorn Granlund <tege at gmplib.org>
date: Thu Oct 27 15:41:57 2011 +0200
description:
Add s390_32 sqr_basecase, remove separate integrated sqr_diag_addlsh1.asm.
details: /var/hg/gmp/rev/a6e7c37f9885
changeset: 14386:a6e7c37f9885
user: Torbjorn Granlund <tege at gmplib.org>
date: Thu Oct 27 15:42:12 2011 +0200
description:
Add s390_32/esame mul_basecase.
details: /var/hg/gmp/rev/1314035734c9
changeset: 14387:1314035734c9
user: Torbjorn Granlund <tege at gmplib.org>
date: Thu Oct 27 15:43:03 2011 +0200
description:
Fix rp=up code. Remove a leftover insn.
details: /var/hg/gmp/rev/d2ab59bfaaa5
changeset: 14388:d2ab59bfaaa5
user: Torbjorn Granlund <tege at gmplib.org>
date: Thu Oct 27 15:45:37 2011 +0200
description:
New file.
details: /var/hg/gmp/rev/cfe0a56e1f8c
changeset: 14389:cfe0a56e1f8c
user: Torbjorn Granlund <tege at gmplib.org>
date: Thu Oct 27 15:48:39 2011 +0200
description:
*** empty log message ***
diffstat:
ChangeLog | 10 +
mpn/s390_32/esame/mul_basecase.asm | 119 +++++++++++++
mpn/s390_32/esame/sqr_basecase.asm | 202 +++++++++++++++++++++++
mpn/s390_32/esame/sqr_diag_addlsh1.asm | 77 --------
mpn/s390_32/logops_n.asm | 284 +++++++++++++++++++++++++++++++++
mpn/s390_64/logops_n.asm | 7 +-
6 files changed, 618 insertions(+), 81 deletions(-)
diffs (truncated from 746 to 300 lines):
diff -r 6c6fc952ae46 -r cfe0a56e1f8c ChangeLog
--- a/ChangeLog Wed Oct 26 15:19:19 2011 +0200
+++ b/ChangeLog Thu Oct 27 15:48:39 2011 +0200
@@ -1,3 +1,13 @@
+2011-10-27 Torbjorn Granlund <tege at gmplib.org>
+
+ * mpn/s390_32/esame/mul_basecase.asm: New file.
+
+ * mpn/s390_32/esame/sqr_basecase.asm: New file.
+
+ * mpn/s390_32/logops_n.asm: New file.
+
+ * mpn/s390_64/logops_n.asm: Fix rp=up code. Remove a leftover insn.
+
2011-10-26 Niels Möller <nisse at lysator.liu.se>
* gmp-impl.h (mpn_hgcd_reduce, mpn_hgcd_reduce_itch): Added
diff -r 6c6fc952ae46 -r cfe0a56e1f8c mpn/s390_32/esame/mul_basecase.asm
--- /dev/null Thu Jan 01 00:00:00 1970 +0000
+++ b/mpn/s390_32/esame/mul_basecase.asm Thu Oct 27 15:48:39 2011 +0200
@@ -0,0 +1,119 @@
+dnl S/390-32/esame mpn_mul_basecase.
+
+dnl Copyright 2011 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 z900 ?
+C z990 ?
+C z9 ?
+C z10 ?
+C z196 ?
+
+C TODO
+C * Perhaps add special case for un <= 2.
+C * Replace loops by faster code. The mul_1 and addmul_1 loops could be sped
+C up by about 10%.
+
+C INPUT PARAMETERS
+define(`rp', `%r2')
+define(`up', `%r3')
+define(`un', `%r4')
+define(`vp', `%r5')
+define(`vn', `%r6')
+
+define(`zero', `%r8')
+
+ASM_START()
+PROLOGUE(mpn_mul_basecase)
+ chi un, 2
+ jhe L(ge2)
+
+C un = vn = 1
+ l %r1, 0(vp)
+ ml %r0, 0(up)
+ st %r1, 0(rp)
+ st %r0, 4(rp)
+ br %r14
+
+L(ge2): C jne L(gen)
+
+
+L(gen):
+C mul_1 =======================================================================
+
+ stm %r6, %r12, 24(%r15)
+ lhi zero, 0
+ ahi un, -1
+
+ l %r7, 0(vp)
+ l %r11, 0(up)
+ lhi %r12, 4 C init index register
+ mlr %r10, %r7
+ lr %r9, un
+ st %r11, 0(rp)
+ cr %r15, %r15 C clear carry flag
+
+L(tm): l %r1, 0(%r12,up)
+ mlr %r0, %r7
+ alcr %r1, %r10
+ lr %r10, %r0 C copy high part to carry limb
+ st %r1, 0(%r12,rp)
+ la %r12, 4(%r12)
+ brct %r9, L(tm)
+
+ alcr %r0, zero
+ st %r0, 0(%r12,rp)
+
+C addmul_1 loop ===============================================================
+
+ ahi vn, -1
+ je L(outer_end)
+L(outer_loop):
+
+ la rp, 4(rp) C rp += 1
+ la vp, 4(vp) C up += 1
+ l %r7, 0(vp)
+ l %r11, 0(up)
+ lhi %r12, 4 C init index register
+ mlr %r10, %r7
+ lr %r9, un
+ al %r11, 0(rp)
+ st %r11, 0(rp)
+
+L(tam): l %r1, 0(%r12,up)
+ l %r11, 0(%r12,rp)
+ mlr %r0, %r7
+ alcr %r1, %r11
+ alcr %r0, zero
+ alr %r1, %r10
+ lr %r10, %r0
+ st %r1, 0(%r12,rp)
+ la %r12, 4(%r12)
+ brct %r9, L(tam)
+
+ alcr %r0, zero
+ st %r0, 0(%r12,rp)
+
+ brct vn, L(outer_loop)
+L(outer_end):
+
+ lm %r6, %r12, 24(%r15)
+ br %r14
+EPILOGUE()
diff -r 6c6fc952ae46 -r cfe0a56e1f8c mpn/s390_32/esame/sqr_basecase.asm
--- /dev/null Thu Jan 01 00:00:00 1970 +0000
+++ b/mpn/s390_32/esame/sqr_basecase.asm Thu Oct 27 15:48:39 2011 +0200
@@ -0,0 +1,202 @@
+dnl S/390-32 mpn_sqr_basecase.
+
+dnl Copyright 2011 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 z900 ?
+C z990 23
+C z9 ?
+C z10 ?
+C z196 ?
+
+C TODO
+C * Clean up.
+C * Stop iterating addmul_1 loop at latest for n = 2, implement longer tail.
+C This will ask for basecase handling of n = 3.
+C * Update counters and pointers more straightforwardly, possibly lowering
+C register usage.
+C * Should we use this allocation-free style for more sqr_basecase asm
+C implementations? The only disadvantage is that it requires R != U.
+C * Replace loops by faster code. The mul_1 and addmul_1 loops could be sped
+C up by about 10%. The sqr_diag_addlsh1 loop could probably be sped up even
+C more.
+
+C INPUT PARAMETERS
+define(`rp', `%r2')
+define(`up', `%r3')
+define(`n', `%r4')
+
+define(`zero', `%r8')
+define(`rp_saved', `%r9')
+define(`up_saved', `%r13')
+define(`n_saved', `%r14')
+
+ASM_START()
+PROLOGUE(mpn_sqr_basecase)
+ ahi n, -2
+ jhe L(ge2)
+
+C n = 1
+ l %r5, 0(up)
+ mlr %r4, %r5
+ st %r5, 0(rp)
+ st %r4, 4(rp)
+ br %r14
+
+L(ge2): jne L(gen)
+
+C n = 2
+ stm %r6, %r8, 24(%r15)
+ lhi zero, 0
+
+ l %r5, 0(up)
+ mlr %r4, %r5 C u0 * u0
+ l %r1, 4(up)
+ mlr %r0, %r1 C u1 * u1
+ st %r5, 0(rp)
+
+ l %r7, 0(up)
+ ml %r6, 4(up) C u0 * u1
+ alr %r7, %r7
+ alcr %r6, %r6
+ alcr %r0, zero
+
+ alr %r4, %r7
+ alcr %r1, %r6
+ alcr %r0, zero
+ st %r4, 4(rp)
+ st %r1, 8(rp)
+ st %r0, 12(rp)
+
+ lm %r6, %r8, 24(%r15)
+ br %r14
+
+L(gen):
+C mul_1 =======================================================================
+
+ stm %r6, %r14, 24(%r15)
+ lhi zero, 0
+ lr up_saved, up
+ lr rp_saved, rp
+ lr n_saved, n
+
+ l %r6, 0(up)
+ l %r11, 4(up)
+ lhi %r12, 8 C init index register
+ mlr %r10, %r6
+ lr %r5, n
+ st %r11, 4(rp)
+ cr %r15, %r15 C clear carry flag
+
+L(tm): l %r1, 0(%r12,up)
+ mlr %r0, %r6
+ alcr %r1, %r10
+ lr %r10, %r0 C copy high part to carry limb
+ st %r1, 0(%r12,rp)
+ la %r12, 4(%r12)
+ brct %r5, L(tm)
+
+ alcr %r0, zero
+ st %r0, 0(%r12,rp)
+
+C addmul_1 loop ===============================================================
+
+ ahi n, -1
+ je L(outer_end)
+L(outer_loop):
+
+ la rp, 8(rp) C rp += 2
+ la up, 4(up) C up += 1
+ l %r6, 0(up)
+ l %r11, 4(up)
+ lhi %r12, 8 C init index register
+ mlr %r10, %r6
+ lr %r5, n
+ al %r11, 4(rp)
+ st %r11, 4(rp)
+
+L(tam): l %r1, 0(%r12,up)
+ l %r7, 0(%r12,rp)
+ mlr %r0, %r6
+ alcr %r1, %r7
+ alcr %r0, zero
+ alr %r1, %r10
+ lr %r10, %r0
+ st %r1, 0(%r12,rp)
+ la %r12, 4(%r12)
+ brct %r5, L(tam)
+
+ alcr %r0, zero
+ st %r0, 0(%r12,rp)
+
+ brct n, L(outer_loop)
+L(outer_end):
+
+ l %r6, 4(up)
+ l %r1, 8(up)
+ lr %r7, %r0 C Same as: l %r7, 12(,rp)
+ mlr %r0, %r6
+ alr %r1, %r7
More information about the gmp-commit
mailing list