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
118 ;------------------------------------------------------------------------------
119 ; void ff_float_to_int16_step(int16_t *dst, const float *src, long len, long step);
120 ;------------------------------------------------------------------------------
121 %macro FLOAT_TO_INT16_STEP 2
122 cglobal float_to_int16_step_%1, 4,7,%2, dst, src, len, step, step3, v1, v2
124 lea srcq, [srcq+2*lenq]
125 lea step3q, [stepq*3]
129 cvtps2dq m0, [srcq+2*lenq ]
130 cvtps2dq m1, [srcq+2*lenq+16]
137 mov [dstq+stepq*4], v2w
140 mov [dstq+stepq*2], v1w
141 mov [dstq+step3q*2], v2w
142 lea dstq, [dstq+stepq*8]
147 mov [dstq+stepq*4], v2w
150 mov [dstq+stepq*2], v1w
151 mov [dstq+step3q*2], v2w
152 lea dstq, [dstq+stepq*8]
154 cvtps2pi m0, [srcq+2*lenq ]
155 cvtps2pi m1, [srcq+2*lenq+ 8]
156 cvtps2pi m2, [srcq+2*lenq+16]
157 cvtps2pi m3, [srcq+2*lenq+24]
164 mov [dstq+stepq*4], v2w
167 mov [dstq+stepq*2], v1w
168 mov [dstq+step3q*2], v2w
169 lea dstq, [dstq+stepq*8]
174 mov [dstq+stepq*4], v2w
177 mov [dstq+stepq*2], v1w
178 mov [dstq+step3q*2], v2w
179 lea dstq, [dstq+stepq*8]
190 FLOAT_TO_INT16_STEP sse2, 2
192 FLOAT_TO_INT16_STEP sse, 0
193 %define cvtps2pi pf2id
194 FLOAT_TO_INT16_STEP 3dnow, 0
197 ;-------------------------------------------------------------------------------
198 ; void ff_float_to_int16_interleave2(int16_t *dst, const float **src, long len);
199 ;-------------------------------------------------------------------------------
200 %macro FLOAT_TO_INT16_INTERLEAVE2 1
201 cglobal float_to_int16_interleave2_%1, 3,4,2, dst, src0, src1, len
203 mov src1q, [src0q+gprsize]
211 cvtps2dq m0, [src0q+lenq]
212 cvtps2dq m1, [src1q+lenq]
218 cvtps2pi m0, [src0q+lenq ]
219 cvtps2pi m1, [src0q+lenq+8]
220 cvtps2pi m2, [src1q+lenq ]
221 cvtps2pi m3, [src1q+lenq+8]
227 mova [dstq+lenq ], m0
228 mova [dstq+lenq+8], m1
239 %define cvtps2pi pf2id
240 FLOAT_TO_INT16_INTERLEAVE2 3dnow
242 %define movdqa movaps
243 FLOAT_TO_INT16_INTERLEAVE2 sse
246 FLOAT_TO_INT16_INTERLEAVE2 sse2
258 %macro FLOAT_TO_INT16_INTERLEAVE6 1
259 ; void float_to_int16_interleave6_sse(int16_t *dst, const float **src, int len)
260 cglobal float_to_int16_interleave6_%1, 2,8,0, dst, src, src1, src2, src3, src4, src5, len
264 %define lend dword r2m
266 mov src1q, [srcq+1*gprsize]
267 mov src2q, [srcq+2*gprsize]
268 mov src3q, [srcq+3*gprsize]
269 mov src4q, [srcq+4*gprsize]
270 mov src5q, [srcq+5*gprsize]
279 cvtps2pi mm1, [srcq+src1q]
280 cvtps2pi mm2, [srcq+src2q]
281 cvtps2pi mm3, [srcq+src3q]
282 cvtps2pi mm4, [srcq+src4q]
283 cvtps2pi mm5, [srcq+src5q]
304 %endmacro ; FLOAT_TO_INT16_INTERLEAVE6
306 %define pswapd PSWAPD_SSE
307 FLOAT_TO_INT16_INTERLEAVE6 sse
308 %define cvtps2pi pf2id
309 %define pswapd PSWAPD_3DN1
310 FLOAT_TO_INT16_INTERLEAVE6 3dnow
312 FLOAT_TO_INT16_INTERLEAVE6 3dn2
315 ;-----------------------------------------------------------------------------
316 ; void ff_float_interleave6(float *dst, const float **src, unsigned int len);
317 ;-----------------------------------------------------------------------------
319 %macro FLOAT_INTERLEAVE6 2
320 cglobal float_interleave6_%1, 2,8,%2, dst, src, src1, src2, src3, src4, src5, len
324 %define lend dword r2m
326 mov src1q, [srcq+1*gprsize]
327 mov src2q, [srcq+2*gprsize]
328 mov src3q, [srcq+3*gprsize]
329 mov src4q, [srcq+4*gprsize]
330 mov src5q, [srcq+5*gprsize]
340 movaps m1, [srcq+src1q]
341 movaps m2, [srcq+src2q]
342 movaps m3, [srcq+src3q]
343 movaps m4, [srcq+src4q]
344 movaps m5, [srcq+src5q]
367 movq m1, [srcq+src1q]
368 movq m2, [srcq+src2q]
369 movq m3, [srcq+src3q]
370 movq m4, [srcq+src4q]
371 movq m5, [srcq+src5q]
373 SBUTTERFLY dq, 0, 1, 6
374 SBUTTERFLY dq, 2, 3, 6
375 SBUTTERFLY dq, 4, 5, 6
394 FLOAT_INTERLEAVE6 mmx, 0
396 FLOAT_INTERLEAVE6 sse, 7
398 ;-----------------------------------------------------------------------------
399 ; void ff_float_interleave2(float *dst, const float **src, unsigned int len);
400 ;-----------------------------------------------------------------------------
402 %macro FLOAT_INTERLEAVE2 2
403 cglobal float_interleave2_%1, 3,4,%2, dst, src, len, src1
404 mov src1q, [srcq+gprsize]
409 MOVPS m1, [srcq+src1q ]
410 MOVPS m3, [srcq +mmsize]
411 MOVPS m4, [srcq+src1q+mmsize]
422 MOVPS [dstq+1*mmsize], m2
423 MOVPS [dstq+2*mmsize], m3
424 MOVPS [dstq+3*mmsize], m1
438 %define PUNPCKLDQ punpckldq
439 %define PUNPCKHDQ punpckhdq
440 FLOAT_INTERLEAVE2 mmx, 0
443 %define PUNPCKLDQ unpcklps
444 %define PUNPCKHDQ unpckhps
445 FLOAT_INTERLEAVE2 sse, 5