dnl Contributed to the GNU project by Torbjörn Granlund.
dnl Copyright 2013 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 cycles/limb good
C aligned unaligned best seen forcpu?
C StrongARM - -
C XScale - -
C Cortex-A7 ? ?
C Cortex-A8 ? ?
C Cortex-A9 3.5 3.5 Y
C Cortex-A15 1.75 1.75 Y
C We read 64 bits at a time at 32-bit aligned addresses, and except for the
C first and last store, we write using 64-bit aligned addresses. All shifting
C is done on 64-bit words in 'extension' registers.
C
C It should be possible to read also using 64-bit alignment, by manipulating
C the shift count for unaligned operands. Not done, since it does not seem to
C matter for A9 or A15.
C
C This will not work in big-endian mode.
C TODO
C * Try using 128-bit operations. Note that Neon lacks pure 128-bit shifts,
C which might make it tricky.
C * Clean up and simplify.
C * Consider sharing most of the codefor lshift and rshift, since the feed-in
C code, the loop, and most of the wind-down code are identical.
C * Replace the basecase code with code using 'extension' registers.
C * Optimise. It is not clear that this loop insn permutation is optimal for
C either A9 or A15.
ASM_START(neon)
TEXT ALIGN(64)
PROLOGUE(mpn_lshiftc) mov r12, n, lsl #2 add rp, rp, r12 add ap, ap, r12
cmp n, #4 C SIMD code n limit
ble L(base)
vdup.32 d6, r3 C left shift count is positive sub r3, r3, #64 C right shift count is negative
vdup.32 d7, r3 mov r12, #-8 C lshift pointer update offset
sub ap, ap, #8
vld1.32 {d19}, [ap], r12 C load initial 2 limbs
vshl.u64 d18, d19, d7 C retval
tst rp, #4 C is rp 64-bit aligned already?
beq L(rp_aligned) C yes, skip
vmvn d19, d19 add ap, ap, #4 C move back ap pointer
vshl.u64 d4, d19, d6 sub n, n, #1 C first limb handled sub rp, rp, #4
vst1.32 {d4[1]}, [rp] C store first limb, rp gets aligned
vld1.32 {d19}, [ap], r12 C load ap[1] and ap[2]
L(rp_aligned): sub rp, rp, #8
subs n, n, #6
vmvn d19, d19
blt L(two_or_three_more)
tst n, #2
beq L(2)
L(1): vld1.32 {d17}, [ap], r12
vshl.u64 d5, d19, d6
vmvn d17, d17
vld1.32 {d16}, [ap], r12
vshl.u64 d0, d17, d7
vshl.u64 d4, d17, d6 sub n, n, #2
b L(mid)
Die Informationen auf dieser Webseite wurden
nach bestem Wissen sorgfältig zusammengestellt. Es wird jedoch weder Vollständigkeit, noch Richtigkeit,
noch Qualität der bereit gestellten Informationen zugesichert.
Bemerkung:
Die farbliche Syntaxdarstellung und die Messung sind noch experimentell.