1 ;******************************************************************************
2 ;* SIMD optimized non-power-of-two MDCT functions
4 ;* Copyright (C) 2017 Rostislav Pehlivanov <atomnuker@gmail.com>
6 ;* This file is part of FFmpeg.
8 ;* FFmpeg is free software; you can redistribute it and/or
9 ;* modify it under the terms of the GNU Lesser General Public
10 ;* License as published by the Free Software Foundation; either
11 ;* version 2.1 of the License, or (at your option) any later version.
13 ;* FFmpeg is distributed in the hope that it will be useful,
14 ;* but WITHOUT ANY WARRANTY; without even the implied warranty of
15 ;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 ;* Lesser General Public License for more details.
18 ;* You should have received a copy of the GNU Lesser General Public
19 ;* License along with FFmpeg; if not, write to the Free Software
20 ;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21 ;******************************************************************************
23 %include "libavutil/x86/x86util.asm"
27 perm_neg: dd 2, 5, 3, 4, 6, 1, 7, 0
28 perm_pos: dd 0, 7, 1, 6, 4, 3, 5, 2
29 sign_adjust_r: times 4 dd 0x80000000, 0x00000000
31 sign_adjust_5: dd 0x00000000, 0x80000000, 0x80000000, 0x00000000
37 ;*****************************************************************************************
38 ;void ff_fft15_avx(FFTComplex *out, FFTComplex *in, FFTComplex *exptab, ptrdiff_t stride);
39 ;*****************************************************************************************
40 %macro FFT5 3 ; %1 - in_offset, %2 - dst1 (64bit used), %3 - dst2
41 VBROADCASTSD m0, [inq + %1] ; in[ 0].re, in[ 0].im, in[ 0].re, in[ 0].im
42 movsd xm1, [inq + 1*16 + 8 + %1] ; in[ 3].re, in[ 3].im, 0, 0
43 movsd xm4, [inq + 6*16 + 0 + %1] ; in[12].re, in[12].im, 0, 0
44 movhps xm1, [inq + 3*16 + 0 + %1] ; in[ 3].re, in[ 3].im, in[ 6].re, in[ 6].im
45 movhps xm4, [inq + 4*16 + 8 + %1] ; in[12].re, in[12].im, in[ 9].re, in[ 9].im
47 subps xm2, xm1, xm4 ; t[2].im, t[2].re, t[3].im, t[3].re
48 addps xm1, xm4 ; t[0].re, t[0].im, t[1].re, t[1].im
50 movhlps %2, xm1 ; t[0].re, t[1].re, t[0].im, t[1].im
52 addps %2, xm0 ; DC[0].re, DC[0].im, junk...
53 movlhps %2, %2 ; DC[0].re, DC[0].im, DC[0].re, DC[0].im
55 shufps xm3, xm1, xm2, q0110 ; t[0].re, t[0].im, t[2].re, t[2].im
56 shufps xm1, xm2, q2332 ; t[1].re, t[1].im, t[3].re, t[3].im
64 addsubps xm3, xm1 ; t[0].re, t[0].im, t[2].re, t[2].im
65 subps xm%3, xm4 ; t[4].re, t[4].im, t[5].re, t[5].im
67 movhlps xm2, xm%3, xm3 ; t[2].re, t[2].im, t[5].re, t[5].im
68 movlhps xm3, xm%3 ; t[0].re, t[0].im, t[4].re, t[4].im
74 shufps xm3, xm3, q1032
75 vinsertf128 m%3, m%3, xm3, 1 ; All ACs (tmp[1] through to tmp[4])
76 addps m%3, m%3, m0 ; Finally offset with DCs
79 %macro BUTTERFLIES_DC 2 ; %1 - exptab_offset, %2 - out
80 mulps xm0, xm9, [exptabq + %1 + 16*0]
81 mulps xm1, xm10, [exptabq + %1 + 16*1]
84 movhlps xm1, xm0 ; t[0].re, t[1].re, t[0].im, t[1].im
92 %macro BUTTERFLIES_AC 2 ; exptab, exptab_offset, src1, src2, src3, out (uses m0-m3)
93 mulps m0, m12, [exptabq + 64*0 + 0*mmsize + %1]
94 mulps m1, m12, [exptabq + 64*0 + 1*mmsize + %1]
95 mulps m2, m13, [exptabq + 64*1 + 0*mmsize + %1]
96 mulps m3, m13, [exptabq + 64*1 + 1*mmsize + %1]
102 shufps m1, m1, m1, q2301
105 vextractf128 xm1, m0, 1
107 movlps [%2q + strideq*1], xm0
108 movhps [%2q + strideq*2], xm0
109 movlps [%2q + stride3q], xm1
110 movhps [%2q + strideq*4], xm1
114 cglobal fft15, 4, 6, 14, out, in, exptab, stride, stride3, stride5
118 movaps xm5, [exptabq + 480 + 16*0]
119 movaps xm6, [exptabq + 480 + 16*1]
120 movaps xm7, [sign_adjust_5]
126 lea stride3q, [strideq + strideq*2]
127 lea stride5q, [strideq + strideq*4]
131 BUTTERFLIES_DC (8*6 + 4*0)*2*4, out0
132 lea outq, [out0q + stride5q*1]
133 BUTTERFLIES_DC (8*6 + 4*1)*2*4, out
134 lea outq, [out0q + stride5q*2]
135 BUTTERFLIES_DC (8*6 + 4*2)*2*4, out
137 BUTTERFLIES_AC (8*0)*2*4, out0
138 lea outq, [out0q + stride5q*1]
139 BUTTERFLIES_AC (8*2)*2*4, out
140 lea outq, [out0q + stride5q*2]
141 BUTTERFLIES_AC (8*4)*2*4, out
147 ;*******************************************************************************************************
148 ;void ff_mdct15_postreindex(FFTComplex *out, FFTComplex *in, FFTComplex *exp, int *lut, ptrdiff_t len8);
149 ;*******************************************************************************************************
151 mov r4d, [lutq + %3q*4 + 0]
152 movsd xmm%1, [inq + r4q*8]
153 mov r4d, [lutq + %3q*4 + 4]
154 movhps xmm%1, [inq + r4q*8]
156 mov r4d, [lutq + %3q*4 + 8]
157 movsd %2, [inq + r4q*8]
158 mov r4d, [lutq + %3q*4 + 12]
159 movhps %2, [inq + r4q*8]
160 vinsertf128 %1, %1, %2, 1
164 %macro POSTROTATE_FN 1
165 cglobal mdct15_postreindex, 5, 7, 8 + cpuflag(avx2)*2, out, in, exp, lut, len8, offset_p, offset_n
167 xor offset_nq, offset_nq
168 lea offset_pq, [len8q*2 - %1]
170 movaps m7, [sign_adjust_r]
173 movaps m8, [perm_pos]
174 movaps m9, [perm_neg]
178 movups m0, [expq + offset_pq*8] ; exp[p0].re, exp[p0].im, exp[p1].re, exp[p1].im, exp[p2].re, exp[p2].im, exp[p3].re, exp[p3].im
179 movups m1, [expq + offset_nq*8] ; exp[n3].re, exp[n3].im, exp[n2].re, exp[n2].im, exp[n1].re, exp[n1].im, exp[n0].re, exp[n0].im
181 LUT_LOAD_4D m3, xm4, offset_p ; in[p0].re, in[p0].im, in[p1].re, in[p1].im, in[p2].re, in[p2].im, in[p3].re, in[p3].im
182 LUT_LOAD_4D m4, xm5, offset_n ; in[n3].re, in[n3].im, in[n2].re, in[n2].im, in[n1].re, in[n1].im, in[n0].re, in[n0].im
184 mulps m5, m3, m0 ; in[p].reim * exp[p].reim
185 mulps m6, m4, m1 ; in[n].reim * exp[n].reim
187 xorps m5, m7 ; in[p].re *= -1, in[p].im *= 1
188 xorps m6, m7 ; in[n].re *= -1, in[n].im *= 1
190 shufps m3, m3, m3, q2301 ; in[p].imre
191 shufps m4, m4, m4, q2301 ; in[n].imre
193 mulps m3, m0 ; in[p].imre * exp[p].reim
194 mulps m4, m1 ; in[n].imre * exp[n].reim
196 haddps m3, m6 ; out[n0].im, out[n1].im, out[n3].re, out[n2].re, out[n2].im, out[n3].im, out[n1].re, out[n0].re
197 haddps m5, m4 ; out[p0].re, out[p1].re, out[p3].im, out[p2].im, out[p2].re, out[p3].re, out[p1].im, out[p0].im
200 vpermps m3, m9, m3 ; out[n3].im, out[n3].re, out[n2].im, out[n2].re, out[n1].im, out[n1].re, out[n0].im, out[n0].re
201 vpermps m5, m8, m5 ; out[p0].re, out[p0].im, out[p1].re, out[p1].im, out[p2].re, out[p2].im, out[p3].re, out[p3].im
203 shufps m3, m3, m3, q0312
204 shufps m5, m5, m5, q2130
207 movups [outq + offset_nq*8], m3
208 movups [outq + offset_pq*8], m5
212 cmp offset_nq, offset_pq
221 %if ARCH_X86_64 && HAVE_AVX2_EXTERNAL