2 # Copyright 2002,2005 Free Software Foundation, Inc.
4 # This file is part of GNU Radio
6 # GNU Radio is free software; you can redistribute it and/or modify
7 # it under the terms of the GNU General Public License as published by
8 # the Free Software Foundation; either version 2, or (at your option)
11 # GNU Radio is distributed in the hope that it will be useful,
12 # but WITHOUT ANY WARRANTY; without even the implied warranty of
13 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 # GNU General Public License for more details.
16 # You should have received a copy of the GNU General Public License
17 # along with GNU Radio; see the file COPYING. If not, write to
18 # the Free Software Foundation, Inc., 51 Franklin Street,
19 # Boston, MA 02110-1301, USA.
23 # input and taps are guarenteed to be 16 byte aligned.
24 # n_2_ccomplex_blocks is != 0
27 # ccomplex_dotprod_generic (const float *input,
28 # const float *taps, unsigned n_2_ccomplex_blocks, float *result)
37 # sum0 += input[0] * taps[0] - input[1] * taps[1];
38 # sum1 += input[0] * taps[1] + input[1] * taps[0];
39 # sum2 += input[2] * taps[2] - input[3] * taps[3];
40 # sum3 += input[2] * taps[3] + input[3] * taps[2];
45 # } while (--n_2_ccomplex_blocks != 0);
48 # result[0] = sum0 + sum2;
49 # result[1] = sum1 + sum3;
53 # TODO: prefetch and better scheduling
57 .file "ccomplex_dotprod_3dnowext64.S"
61 .globl GLOB_SYMB(ccomplex_dotprod_3dnowext)
62 DEF_FUNC_HEAD(ccomplex_dotprod_3dnowext)
63 GLOB_SYMB(ccomplex_dotprod_3dnowext):
65 # intput: rdi, taps: rsi, n_2_ccomplex_blocks: rdx, result: rcx
72 pxor %mm6, %mm6 # mm6 = 0 0
73 pxor %mm7, %mm7 # mm7 = 0 0
78 shr $1, %rax # rax = n_2_ccomplex_blocks / 2
88 # something like ?? cycles / loop
94 # complex prod: C += A * B, w/ temp Z
105 # A=mm0, B=mm2, Z=mm4
106 # A'=mm1, B'=mm3, Z'=mm5
149 # We've handled the bulk of multiplies up to here.
150 # Let's see if original n_2_ccomplex_blocks was odd.
151 # If so, we've got 2 more taps to do.
156 # The count was odd, do 2 more taps.
157 # Note that we've already got mm0/mm2 & mm1/mm3 preloaded
158 # from the main loop.
160 # A=mm0, B=mm2, Z=mm4
161 # A'=mm1, B'=mm3, Z'=mm5
176 # at this point mm6 and mm7 contain partial sums
180 movq %mm6, (%rcx) # result
186 FUNC_TAIL(ccomplex_dotprod_3dnowext)
187 .ident "Hand coded x86_64 3DNow!Ext assembly"