Return to mul_1.s CVS log

Up to [local] / OpenXM_contrib / gmp / mpn / hppa / hppa1_1

Annotation of OpenXM_contrib/gmp/mpn/hppa/hppa1_1/mul_1.s, Revision 1.1.1.2

1.1.1.2 ! maekawa     1: ; HP-PA-1.1 __gmpn_mul_1 -- Multiply a limb vector with a limb and store
1.1       maekawa     2: ; the result in a second limb vector.
                      3:
1.1.1.2 ! maekawa     4: ; Copyright (C) 1992, 1993, 1994, 2000 Free Software Foundation, Inc.
1.1       maekawa     5:
                      6: ; This file is part of the GNU MP Library.
                      7:
                      8: ; The GNU MP Library is free software; you can redistribute it and/or modify
1.1.1.2 ! maekawa     9: ; it under the terms of the GNU Lesser General Public License as published by
        !            10: ; the Free Software Foundation; either version 2.1 of the License, or (at your
1.1       maekawa    11: ; option) any later version.
                     12:
                     13: ; The GNU MP Library is distributed in the hope that it will be useful, but
                     14: ; WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
1.1.1.2 ! maekawa    15: ; or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
1.1       maekawa    16: ; License for more details.
                     17:
1.1.1.2 ! maekawa    18: ; You should have received a copy of the GNU Lesser General Public License
1.1       maekawa    19: ; along with the GNU MP Library; see the file COPYING.LIB.  If not, write to
                     20: ; the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
                     21: ; MA 02111-1307, USA.
                     22:
                     23:
                     24: ; INPUT PARAMETERS
                     25: ; res_ptr      r26
                     26: ; s1_ptr       r25
                     27: ; size         r24
                     28: ; s2_limb      r23
                     29:
                     30: ; This runs at 9 cycles/limb on a PA7000.  With the used instructions, it can
                     31: ; not become faster due to data cache contention after a store.  On the
                     32: ; PA7100 it runs at 7 cycles/limb, and that can not be improved either, since
                     33: ; only the xmpyu does not need the integer pipeline, so the only dual-issue
                     34: ; we will get are addc+xmpyu.  Unrolling would not help either CPU.
                     35:
                     36: ; We could use fldds to read two limbs at a time from the S1 array, and that
                     37: ; could bring down the times to 8.5 and 6.5 cycles/limb for the PA7000 and
                     38: ; PA7100, respectively.  We don't do that since it does not seem worth the
                     39: ; (alignment) troubles...
                     40:
                     41: ; At least the PA7100 is rumored to be able to deal with cache-misses
                     42: ; without stalling instruction issue.  If this is true, and the cache is
                     43: ; actually also lockup-free, we should use a deeper software pipeline, and
                     44: ; load from S1 very early!  (The loads and stores to -12(sp) will surely be
                     45: ; in the cache.)
                     46:
                     47:        .code
1.1.1.2 ! maekawa    48:        .export         __gmpn_mul_1
        !            49: __gmpn_mul_1
1.1       maekawa    50:        .proc
                     51:        .callinfo       frame=64,no_calls
                     52:        .entry
                     53:
                     54:        ldo             64(%r30),%r30
                     55:        fldws,ma        4(%r25),%fr5
                     56:        stw             %r23,-16(%r30)          ; move s2_limb ...
                     57:        addib,=         -1,%r24,L$just_one_limb
                     58:         fldws          -16(%r30),%fr4          ; ... into fr4
                     59:        add             %r0,%r0,%r0             ; clear carry
                     60:        xmpyu           %fr4,%fr5,%fr6
                     61:        fldws,ma        4(%r25),%fr7
                     62:        fstds           %fr6,-16(%r30)
                     63:        xmpyu           %fr4,%fr7,%fr8
                     64:        ldw             -12(%r30),%r19          ; least significant limb in product
                     65:        ldw             -16(%r30),%r28
                     66:
                     67:        fstds           %fr8,-16(%r30)
                     68:        addib,=         -1,%r24,L$end
                     69:         ldw            -12(%r30),%r1
                     70:
                     71: ; Main loop
                     72: L$loop fldws,ma        4(%r25),%fr5
                     73:        stws,ma         %r19,4(%r26)
                     74:        addc            %r28,%r1,%r19
                     75:        xmpyu           %fr4,%fr5,%fr6
                     76:        ldw             -16(%r30),%r28
                     77:        fstds           %fr6,-16(%r30)
                     78:        addib,<>        -1,%r24,L$loop
                     79:         ldw            -12(%r30),%r1
                     80:
                     81: L$end  stws,ma         %r19,4(%r26)
                     82:        addc            %r28,%r1,%r19
                     83:        ldw             -16(%r30),%r28
                     84:        stws,ma         %r19,4(%r26)
                     85:        addc            %r0,%r28,%r28
                     86:        bv              0(%r2)
                     87:         ldo            -64(%r30),%r30
                     88:
                     89: L$just_one_limb
                     90:        xmpyu           %fr4,%fr5,%fr6
                     91:        fstds           %fr6,-16(%r30)
                     92:        ldw             -16(%r30),%r28
                     93:        ldo             -64(%r30),%r30
                     94:        bv              0(%r2)
                     95:         fstws          %fr6R,0(%r26)
                     96:
                     97:        .exit
                     98:        .procend