1 ;******************************************************************************
2 ;* x86 optimized Format Conversion Utils
3 ;* Copyright (c) 2008 Loren Merritt
5 ;* This file is part of Libav.
7 ;* Libav 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 ;* Libav 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 Libav; 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 int32_to_float_fmul_scalar(float *dst, const int *src, float mul, int len);
36 ;---------------------------------------------------------------------------------
37 %macro INT32_TO_FLOAT_FMUL_SCALAR 1
39 cglobal int32_to_float_fmul_scalar, 3, 3, %1, dst, src, len
41 cglobal int32_to_float_fmul_scalar, 4, 4, %1, dst, src, mul, len
55 cvtdq2ps m1, [srcq+lenq ]
56 cvtdq2ps m2, [srcq+lenq+16]
58 cvtpi2ps m1, [srcq+lenq ]
59 cvtpi2ps m3, [srcq+lenq+ 8]
60 cvtpi2ps m2, [srcq+lenq+16]
61 cvtpi2ps m4, [srcq+lenq+24]
68 mova [dstq+lenq+16], m2
75 INT32_TO_FLOAT_FMUL_SCALAR 5
77 INT32_TO_FLOAT_FMUL_SCALAR 3
80 ;------------------------------------------------------------------------------
81 ; void ff_float_to_int16(int16_t *dst, const float *src, long len);
82 ;------------------------------------------------------------------------------
83 %macro FLOAT_TO_INT16 1
84 cglobal float_to_int16, 3, 3, %1, dst, src, len
86 lea srcq, [srcq+2*lenq]
91 cvtps2dq m0, [srcq+2*lenq ]
92 cvtps2dq m1, [srcq+2*lenq+16]
96 CVTPS2PI m0, [srcq+2*lenq ]
97 CVTPS2PI m1, [srcq+2*lenq+ 8]
98 CVTPS2PI m2, [srcq+2*lenq+16]
99 CVTPS2PI m3, [srcq+2*lenq+24]
102 mova [dstq+lenq ], m0
103 mova [dstq+lenq+8], m2
120 ;------------------------------------------------------------------------------
121 ; void ff_float_to_int16_step(int16_t *dst, const float *src, long len, long step);
122 ;------------------------------------------------------------------------------
123 %macro FLOAT_TO_INT16_STEP 1
124 cglobal float_to_int16_step, 4, 7, %1, dst, src, len, step, step3, v1, v2
126 lea srcq, [srcq+2*lenq]
127 lea step3q, [stepq*3]
131 cvtps2dq m0, [srcq+2*lenq ]
132 cvtps2dq m1, [srcq+2*lenq+16]
139 mov [dstq+stepq*4], v2w
142 mov [dstq+stepq*2], v1w
143 mov [dstq+step3q*2], v2w
144 lea dstq, [dstq+stepq*8]
149 mov [dstq+stepq*4], v2w
152 mov [dstq+stepq*2], v1w
153 mov [dstq+step3q*2], v2w
154 lea dstq, [dstq+stepq*8]
156 CVTPS2PI m0, [srcq+2*lenq ]
157 CVTPS2PI m1, [srcq+2*lenq+ 8]
158 CVTPS2PI m2, [srcq+2*lenq+16]
159 CVTPS2PI m3, [srcq+2*lenq+24]
166 mov [dstq+stepq*4], v2w
169 mov [dstq+stepq*2], v1w
170 mov [dstq+step3q*2], v2w
171 lea dstq, [dstq+stepq*8]
176 mov [dstq+stepq*4], v2w
179 mov [dstq+stepq*2], v1w
180 mov [dstq+step3q*2], v2w
181 lea dstq, [dstq+stepq*8]
192 FLOAT_TO_INT16_STEP 2
194 FLOAT_TO_INT16_STEP 0
196 FLOAT_TO_INT16_STEP 0
198 ;-------------------------------------------------------------------------------
199 ; void ff_float_to_int16_interleave2(int16_t *dst, const float **src, long len);
200 ;-------------------------------------------------------------------------------
201 %macro FLOAT_TO_INT16_INTERLEAVE2 0
202 cglobal float_to_int16_interleave2, 3, 4, 2, dst, src0, src1, len
204 mov src1q, [src0q+gprsize]
212 cvtps2dq m0, [src0q+lenq]
213 cvtps2dq m1, [src1q+lenq]
219 CVTPS2PI m0, [src0q+lenq ]
220 CVTPS2PI m1, [src0q+lenq+8]
221 CVTPS2PI m2, [src1q+lenq ]
222 CVTPS2PI m3, [src1q+lenq+8]
228 mova [dstq+lenq ], m0
229 mova [dstq+lenq+8], m1
240 FLOAT_TO_INT16_INTERLEAVE2
242 FLOAT_TO_INT16_INTERLEAVE2
244 FLOAT_TO_INT16_INTERLEAVE2
246 %macro FLOAT_TO_INT16_INTERLEAVE6 0
247 ; void float_to_int16_interleave6_sse(int16_t *dst, const float **src, int len)
248 cglobal float_to_int16_interleave6, 2, 8, 0, dst, src, src1, src2, src3, src4, src5, len
252 %define lend dword r2m
254 mov src1q, [srcq+1*gprsize]
255 mov src2q, [srcq+2*gprsize]
256 mov src3q, [srcq+3*gprsize]
257 mov src4q, [srcq+4*gprsize]
258 mov src5q, [srcq+5*gprsize]
267 CVTPS2PI mm1, [srcq+src1q]
268 CVTPS2PI mm2, [srcq+src2q]
269 CVTPS2PI mm3, [srcq+src3q]
270 CVTPS2PI mm4, [srcq+src4q]
271 CVTPS2PI mm5, [srcq+src5q]
292 %endmacro ; FLOAT_TO_INT16_INTERLEAVE6
295 FLOAT_TO_INT16_INTERLEAVE6
297 FLOAT_TO_INT16_INTERLEAVE6
299 FLOAT_TO_INT16_INTERLEAVE6
301 ;-----------------------------------------------------------------------------
302 ; void ff_float_interleave6(float *dst, const float **src, unsigned int len);
303 ;-----------------------------------------------------------------------------
305 %macro FLOAT_INTERLEAVE6 1
306 cglobal float_interleave6, 2, 8, %1, dst, src, src1, src2, src3, src4, src5, len
310 %define lend dword r2m
312 mov src1q, [srcq+1*gprsize]
313 mov src2q, [srcq+2*gprsize]
314 mov src3q, [srcq+3*gprsize]
315 mov src4q, [srcq+4*gprsize]
316 mov src5q, [srcq+5*gprsize]
326 movaps m1, [srcq+src1q]
327 movaps m2, [srcq+src2q]
328 movaps m3, [srcq+src3q]
329 movaps m4, [srcq+src4q]
330 movaps m5, [srcq+src5q]
353 movq m1, [srcq+src1q]
354 movq m2, [srcq+src2q]
355 movq m3, [srcq+src3q]
356 movq m4, [srcq+src4q]
357 movq m5, [srcq+src5q]
359 SBUTTERFLY dq, 0, 1, 6
360 SBUTTERFLY dq, 2, 3, 6
361 SBUTTERFLY dq, 4, 5, 6
384 ;-----------------------------------------------------------------------------
385 ; void ff_float_interleave2(float *dst, const float **src, unsigned int len);
386 ;-----------------------------------------------------------------------------
388 %macro FLOAT_INTERLEAVE2 1
389 cglobal float_interleave2, 3, 4, %1, dst, src, len, src1
390 mov src1q, [srcq+gprsize]
395 mova m1, [srcq+src1q ]
396 mova m3, [srcq +mmsize]
397 mova m4, [srcq+src1q+mmsize]
408 mova [dstq+1*mmsize], m2
409 mova [dstq+2*mmsize], m3
410 mova [dstq+3*mmsize], m1
423 %define PUNPCKLDQ punpckldq
424 %define PUNPCKHDQ punpckhdq
427 %define PUNPCKLDQ unpcklps
428 %define PUNPCKHDQ unpckhps