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/x86util.asm"
34 ;------------------------------------------------------------------------------
35 ; void ff_int32_to_float_fmul_scalar(float *dst, const int32_t *src, float mul,
37 ;------------------------------------------------------------------------------
38 %macro INT32_TO_FLOAT_FMUL_SCALAR 1
40 cglobal int32_to_float_fmul_scalar, 3, 3, %1, dst, src, len
42 cglobal int32_to_float_fmul_scalar, 4, 4, %1, dst, src, mul, len
56 cvtdq2ps m1, [srcq+lenq ]
57 cvtdq2ps m2, [srcq+lenq+16]
59 cvtpi2ps m1, [srcq+lenq ]
60 cvtpi2ps m3, [srcq+lenq+ 8]
61 cvtpi2ps m2, [srcq+lenq+16]
62 cvtpi2ps m4, [srcq+lenq+24]
69 mova [dstq+lenq+16], m2
76 INT32_TO_FLOAT_FMUL_SCALAR 5
78 INT32_TO_FLOAT_FMUL_SCALAR 3
80 ;------------------------------------------------------------------------------
81 ; void ff_int32_to_float_fmul_array8(FmtConvertContext *c, float *dst, const int32_t *src,
82 ; const float *mul, int len);
83 ;------------------------------------------------------------------------------
84 %macro INT32_TO_FLOAT_FMUL_ARRAY8 0
85 cglobal int32_to_float_fmul_array8, 5, 5, 5, c, dst, src, mul, len
94 cvtdq2ps m1, [srcq+lenq ]
95 cvtdq2ps m2, [srcq+lenq+16]
97 cvtpi2ps m1, [srcq+lenq ]
98 cvtpi2ps m3, [srcq+lenq+ 8]
99 cvtpi2ps m2, [srcq+lenq+16]
100 cvtpi2ps m4, [srcq+lenq+24]
106 mova [dstq+lenq ], m1
107 mova [dstq+lenq+16], m2
115 INT32_TO_FLOAT_FMUL_ARRAY8
117 INT32_TO_FLOAT_FMUL_ARRAY8
119 ;------------------------------------------------------------------------------
120 ; void ff_float_to_int16(int16_t *dst, const float *src, long len);
121 ;------------------------------------------------------------------------------
122 %macro FLOAT_TO_INT16 1
123 cglobal float_to_int16, 3, 3, %1, dst, src, len
125 lea srcq, [srcq+2*lenq]
130 cvtps2dq m0, [srcq+2*lenq ]
131 cvtps2dq m1, [srcq+2*lenq+16]
135 CVTPS2PI m0, [srcq+2*lenq ]
136 CVTPS2PI m1, [srcq+2*lenq+ 8]
137 CVTPS2PI m2, [srcq+2*lenq+16]
138 CVTPS2PI m3, [srcq+2*lenq+24]
141 mova [dstq+lenq ], m0
142 mova [dstq+lenq+8], m2
159 ;------------------------------------------------------------------------------
160 ; void ff_float_to_int16_step(int16_t *dst, const float *src, long len, long step);
161 ;------------------------------------------------------------------------------
162 %macro FLOAT_TO_INT16_STEP 1
163 cglobal float_to_int16_step, 4, 7, %1, dst, src, len, step, step3, v1, v2
165 lea srcq, [srcq+2*lenq]
166 lea step3q, [stepq*3]
170 cvtps2dq m0, [srcq+2*lenq ]
171 cvtps2dq m1, [srcq+2*lenq+16]
178 mov [dstq+stepq*4], v2w
181 mov [dstq+stepq*2], v1w
182 mov [dstq+step3q*2], v2w
183 lea dstq, [dstq+stepq*8]
188 mov [dstq+stepq*4], v2w
191 mov [dstq+stepq*2], v1w
192 mov [dstq+step3q*2], v2w
193 lea dstq, [dstq+stepq*8]
195 CVTPS2PI m0, [srcq+2*lenq ]
196 CVTPS2PI m1, [srcq+2*lenq+ 8]
197 CVTPS2PI m2, [srcq+2*lenq+16]
198 CVTPS2PI m3, [srcq+2*lenq+24]
205 mov [dstq+stepq*4], v2w
208 mov [dstq+stepq*2], v1w
209 mov [dstq+step3q*2], v2w
210 lea dstq, [dstq+stepq*8]
215 mov [dstq+stepq*4], v2w
218 mov [dstq+stepq*2], v1w
219 mov [dstq+step3q*2], v2w
220 lea dstq, [dstq+stepq*8]
231 FLOAT_TO_INT16_STEP 2
233 FLOAT_TO_INT16_STEP 0
235 FLOAT_TO_INT16_STEP 0
237 ;-------------------------------------------------------------------------------
238 ; void ff_float_to_int16_interleave2(int16_t *dst, const float **src, long len);
239 ;-------------------------------------------------------------------------------
240 %macro FLOAT_TO_INT16_INTERLEAVE2 0
241 cglobal float_to_int16_interleave2, 3, 4, 2, dst, src0, src1, len
243 mov src1q, [src0q+gprsize]
251 cvtps2dq m0, [src0q+lenq]
252 cvtps2dq m1, [src1q+lenq]
258 CVTPS2PI m0, [src0q+lenq ]
259 CVTPS2PI m1, [src0q+lenq+8]
260 CVTPS2PI m2, [src1q+lenq ]
261 CVTPS2PI m3, [src1q+lenq+8]
267 mova [dstq+lenq ], m0
268 mova [dstq+lenq+8], m1
279 FLOAT_TO_INT16_INTERLEAVE2
281 FLOAT_TO_INT16_INTERLEAVE2
283 FLOAT_TO_INT16_INTERLEAVE2
285 ;-----------------------------------------------------------------------------
286 ; void ff_float_to_int16_interleave6(int16_t *dst, const float **src, int len)
287 ;-----------------------------------------------------------------------------
288 %macro FLOAT_TO_INT16_INTERLEAVE6 0
289 cglobal float_to_int16_interleave6, 2, 8, 0, dst, src, src1, src2, src3, src4, src5, len
293 %define lend dword r2m
295 mov src1q, [srcq+1*gprsize]
296 mov src2q, [srcq+2*gprsize]
297 mov src3q, [srcq+3*gprsize]
298 mov src4q, [srcq+4*gprsize]
299 mov src5q, [srcq+5*gprsize]
308 CVTPS2PI mm1, [srcq+src1q]
309 CVTPS2PI mm2, [srcq+src2q]
310 CVTPS2PI mm3, [srcq+src3q]
311 CVTPS2PI mm4, [srcq+src4q]
312 CVTPS2PI mm5, [srcq+src5q]
333 %endmacro ; FLOAT_TO_INT16_INTERLEAVE6
336 FLOAT_TO_INT16_INTERLEAVE6
338 FLOAT_TO_INT16_INTERLEAVE6
340 FLOAT_TO_INT16_INTERLEAVE6
342 ;-----------------------------------------------------------------------------
343 ; void ff_float_interleave6(float *dst, const float **src, unsigned int len);
344 ;-----------------------------------------------------------------------------
346 %macro FLOAT_INTERLEAVE6 1
347 cglobal float_interleave6, 2, 8, %1, dst, src, src1, src2, src3, src4, src5, len
351 %define lend dword r2m
353 mov src1q, [srcq+1*gprsize]
354 mov src2q, [srcq+2*gprsize]
355 mov src3q, [srcq+3*gprsize]
356 mov src4q, [srcq+4*gprsize]
357 mov src5q, [srcq+5*gprsize]
367 movaps m1, [srcq+src1q]
368 movaps m2, [srcq+src2q]
369 movaps m3, [srcq+src3q]
370 movaps m4, [srcq+src4q]
371 movaps m5, [srcq+src5q]
394 movq m1, [srcq+src1q]
395 movq m2, [srcq+src2q]
396 movq m3, [srcq+src3q]
397 movq m4, [srcq+src4q]
398 movq m5, [srcq+src5q]
400 SBUTTERFLY dq, 0, 1, 6
401 SBUTTERFLY dq, 2, 3, 6
402 SBUTTERFLY dq, 4, 5, 6
425 ;-----------------------------------------------------------------------------
426 ; void ff_float_interleave2(float *dst, const float **src, unsigned int len);
427 ;-----------------------------------------------------------------------------
429 %macro FLOAT_INTERLEAVE2 1
430 cglobal float_interleave2, 3, 4, %1, dst, src, len, src1
431 mov src1q, [srcq+gprsize]
436 mova m1, [srcq+src1q ]
437 mova m3, [srcq +mmsize]
438 mova m4, [srcq+src1q+mmsize]
449 mova [dstq+1*mmsize], m2
450 mova [dstq+2*mmsize], m3
451 mova [dstq+3*mmsize], m1
464 %define PUNPCKLDQ punpckldq
465 %define PUNPCKHDQ punpckhdq
468 %define PUNPCKLDQ unpcklps
469 %define PUNPCKHDQ unpckhps