1 ;******************************************************************************
2 ;* x86 optimized Format Conversion Utils
3 ;* Copyright (c) 2008 Loren Merritt
5 ;* This file is part of FFmpeg.
7 ;* FFmpeg is free software; you can redistribute it and/or
8 ;* modify it under the terms of the GNU Lesser General Public
9 ;* License as published by the Free Software Foundation; either
10 ;* version 2.1 of the License, or (at your option) any later version.
12 ;* FFmpeg is distributed in the hope that it will be useful,
13 ;* but WITHOUT ANY WARRANTY; without even the implied warranty of
14 ;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 ;* Lesser General Public License for more details.
17 ;* You should have received a copy of the GNU Lesser General Public
18 ;* License along with FFmpeg; if not, write to the Free Software
19 ;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 ;******************************************************************************
22 %include "libavutil/x86/x86inc.asm"
23 %include "libavutil/x86/x86util.asm"
27 ;---------------------------------------------------------------------------------
28 ; void int32_to_float_fmul_scalar(float *dst, const int *src, float mul, int len);
29 ;---------------------------------------------------------------------------------
30 %macro INT32_TO_FLOAT_FMUL_SCALAR 2
32 cglobal int32_to_float_fmul_scalar_%1, 3,3,%2, dst, src, len
34 cglobal int32_to_float_fmul_scalar_%1, 4,4,%2, dst, src, mul, len
48 cvtdq2ps m1, [srcq+lenq ]
49 cvtdq2ps m2, [srcq+lenq+16]
51 cvtpi2ps m1, [srcq+lenq ]
52 cvtpi2ps m3, [srcq+lenq+ 8]
53 cvtpi2ps m2, [srcq+lenq+16]
54 cvtpi2ps m4, [srcq+lenq+24]
61 mova [dstq+lenq+16], m2
68 %define SPLATD SPLATD_SSE
70 INT32_TO_FLOAT_FMUL_SCALAR sse, 5
72 %define SPLATD SPLATD_SSE2
73 INT32_TO_FLOAT_FMUL_SCALAR sse2, 3
77 ;------------------------------------------------------------------------------
78 ; void ff_float_to_int16(int16_t *dst, const float *src, long len);
79 ;------------------------------------------------------------------------------
80 %macro FLOAT_TO_INT16 2
81 cglobal float_to_int16_%1, 3,3,%2, dst, src, len
83 lea srcq, [srcq+2*lenq]
88 cvtps2dq m0, [srcq+2*lenq ]
89 cvtps2dq m1, [srcq+2*lenq+16]
93 cvtps2pi m0, [srcq+2*lenq ]
94 cvtps2pi m1, [srcq+2*lenq+ 8]
95 cvtps2pi m2, [srcq+2*lenq+16]
96 cvtps2pi m3, [srcq+2*lenq+24]
100 mova [dstq+lenq+8], m2
111 FLOAT_TO_INT16 sse2, 2
113 FLOAT_TO_INT16 sse, 0
114 %define cvtps2pi pf2id
115 FLOAT_TO_INT16 3dnow, 0
119 ;-------------------------------------------------------------------------------
120 ; void ff_float_to_int16_interleave2(int16_t *dst, const float **src, long len);
121 ;-------------------------------------------------------------------------------
122 %macro FLOAT_TO_INT16_INTERLEAVE2 1
123 cglobal float_to_int16_interleave2_%1, 3,4,2, dst, src0, src1, len
125 mov src1q, [src0q+gprsize]
133 cvtps2dq m0, [src0q+lenq]
134 cvtps2dq m1, [src1q+lenq]
140 cvtps2pi m0, [src0q+lenq ]
141 cvtps2pi m1, [src0q+lenq+8]
142 cvtps2pi m2, [src1q+lenq ]
143 cvtps2pi m3, [src1q+lenq+8]
149 mova [dstq+lenq ], m0
150 mova [dstq+lenq+8], m1
161 %define cvtps2pi pf2id
162 FLOAT_TO_INT16_INTERLEAVE2 3dnow
164 %define movdqa movaps
165 FLOAT_TO_INT16_INTERLEAVE2 sse
168 FLOAT_TO_INT16_INTERLEAVE2 sse2
180 %macro FLOAT_TO_INT16_INTERLEAVE6 1
181 ; void float_to_int16_interleave6_sse(int16_t *dst, const float **src, int len)
182 cglobal float_to_int16_interleave6_%1, 2,7,0, dst, src, src1, src2, src3, src4, src5
187 %define lend dword r2m
189 mov src1q, [srcq+1*gprsize]
190 mov src2q, [srcq+2*gprsize]
191 mov src3q, [srcq+3*gprsize]
192 mov src4q, [srcq+4*gprsize]
193 mov src5q, [srcq+5*gprsize]
202 cvtps2pi mm1, [srcq+src1q]
203 cvtps2pi mm2, [srcq+src2q]
204 cvtps2pi mm3, [srcq+src3q]
205 cvtps2pi mm4, [srcq+src4q]
206 cvtps2pi mm5, [srcq+src5q]
227 %endmacro ; FLOAT_TO_INT16_INTERLEAVE6
229 %define pswapd PSWAPD_SSE
230 FLOAT_TO_INT16_INTERLEAVE6 sse
231 %define cvtps2pi pf2id
232 %define pswapd PSWAPD_3DN1
233 FLOAT_TO_INT16_INTERLEAVE6 3dnow
235 FLOAT_TO_INT16_INTERLEAVE6 3dn2
238 ;-----------------------------------------------------------------------------
239 ; void ff_float_interleave6(float *dst, const float **src, unsigned int len);
240 ;-----------------------------------------------------------------------------
242 %macro FLOAT_INTERLEAVE6 2
243 cglobal float_interleave6_%1, 2,7,%2, dst, src, src1, src2, src3, src4, src5
248 %define lend dword r2m
250 mov src1q, [srcq+1*gprsize]
251 mov src2q, [srcq+2*gprsize]
252 mov src3q, [srcq+3*gprsize]
253 mov src4q, [srcq+4*gprsize]
254 mov src5q, [srcq+5*gprsize]
264 movaps m1, [srcq+src1q]
265 movaps m2, [srcq+src2q]
266 movaps m3, [srcq+src3q]
267 movaps m4, [srcq+src4q]
268 movaps m5, [srcq+src5q]
291 movq m1, [srcq+src1q]
292 movq m2, [srcq+src2q]
293 movq m3, [srcq+src3q]
294 movq m4, [srcq+src4q]
295 movq m5, [srcq+src5q]
297 SBUTTERFLY dq, 0, 1, 6
298 SBUTTERFLY dq, 2, 3, 6
299 SBUTTERFLY dq, 4, 5, 6
318 FLOAT_INTERLEAVE6 mmx, 0
320 FLOAT_INTERLEAVE6 sse, 7
322 ;-----------------------------------------------------------------------------
323 ; void ff_float_interleave2(float *dst, const float **src, unsigned int len);
324 ;-----------------------------------------------------------------------------
326 %macro FLOAT_INTERLEAVE2 2
327 cglobal float_interleave2_%1, 3,4,%2, dst, src, len, src1
328 mov src1q, [srcq+gprsize]
333 MOVPS m1, [srcq+src1q ]
334 MOVPS m3, [srcq +mmsize]
335 MOVPS m4, [srcq+src1q+mmsize]
346 MOVPS [dstq+1*mmsize], m2
347 MOVPS [dstq+2*mmsize], m3
348 MOVPS [dstq+3*mmsize], m1
362 %define PUNPCKLDQ punpckldq
363 %define PUNPCKHDQ punpckhdq
364 FLOAT_INTERLEAVE2 mmx, 0
367 %define PUNPCKLDQ unpcklps
368 %define PUNPCKHDQ unpckhps
369 FLOAT_INTERLEAVE2 sse, 5