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 "x86util.asm"
26 ;---------------------------------------------------------------------------------
27 ; void int32_to_float_fmul_scalar(float *dst, const int *src, float mul, int len);
28 ;---------------------------------------------------------------------------------
29 %macro INT32_TO_FLOAT_FMUL_SCALAR 2
31 cglobal int32_to_float_fmul_scalar_%1, 3,3,%2, dst, src, len
33 cglobal int32_to_float_fmul_scalar_%1, 4,4,%2, dst, src, mul, len
47 cvtdq2ps m1, [srcq+lenq ]
48 cvtdq2ps m2, [srcq+lenq+16]
50 cvtpi2ps m1, [srcq+lenq ]
51 cvtpi2ps m3, [srcq+lenq+ 8]
52 cvtpi2ps m2, [srcq+lenq+16]
53 cvtpi2ps m4, [srcq+lenq+24]
60 mova [dstq+lenq+16], m2
67 %define SPLATD SPLATD_SSE
69 INT32_TO_FLOAT_FMUL_SCALAR sse, 5
71 %define SPLATD SPLATD_SSE2
72 INT32_TO_FLOAT_FMUL_SCALAR sse2, 3
76 ;------------------------------------------------------------------------------
77 ; void ff_float_to_int16(int16_t *dst, const float *src, long len);
78 ;------------------------------------------------------------------------------
79 %macro FLOAT_TO_INT16 2
80 cglobal float_to_int16_%1, 3,3,%2, dst, src, len
82 lea srcq, [srcq+2*lenq]
87 cvtps2dq m0, [srcq+2*lenq ]
88 cvtps2dq m1, [srcq+2*lenq+16]
92 cvtps2pi m0, [srcq+2*lenq ]
93 cvtps2pi m1, [srcq+2*lenq+ 8]
94 cvtps2pi m2, [srcq+2*lenq+16]
95 cvtps2pi m3, [srcq+2*lenq+24]
99 mova [dstq+lenq+8], m2
110 FLOAT_TO_INT16 sse2, 2
112 FLOAT_TO_INT16 sse, 0
113 %define cvtps2pi pf2id
114 FLOAT_TO_INT16 3dnow, 0
117 ;------------------------------------------------------------------------------
118 ; void ff_float_to_int16_step(int16_t *dst, const float *src, long len, long step);
119 ;------------------------------------------------------------------------------
120 %macro FLOAT_TO_INT16_STEP 2
121 cglobal float_to_int16_step_%1, 4,7,%2, dst, src, len, step, step3, v1, v2
123 lea srcq, [srcq+2*lenq]
124 lea step3q, [stepq*3]
128 cvtps2dq m0, [srcq+2*lenq ]
129 cvtps2dq m1, [srcq+2*lenq+16]
136 mov [dstq+stepq*4], v2w
139 mov [dstq+stepq*2], v1w
140 mov [dstq+step3q*2], v2w
141 lea dstq, [dstq+stepq*8]
146 mov [dstq+stepq*4], v2w
149 mov [dstq+stepq*2], v1w
150 mov [dstq+step3q*2], v2w
151 lea dstq, [dstq+stepq*8]
153 cvtps2pi m0, [srcq+2*lenq ]
154 cvtps2pi m1, [srcq+2*lenq+ 8]
155 cvtps2pi m2, [srcq+2*lenq+16]
156 cvtps2pi m3, [srcq+2*lenq+24]
163 mov [dstq+stepq*4], v2w
166 mov [dstq+stepq*2], v1w
167 mov [dstq+step3q*2], v2w
168 lea dstq, [dstq+stepq*8]
173 mov [dstq+stepq*4], v2w
176 mov [dstq+stepq*2], v1w
177 mov [dstq+step3q*2], v2w
178 lea dstq, [dstq+stepq*8]
189 FLOAT_TO_INT16_STEP sse2, 2
191 FLOAT_TO_INT16_STEP sse, 0
192 %define cvtps2pi pf2id
193 FLOAT_TO_INT16_STEP 3dnow, 0
196 ;-------------------------------------------------------------------------------
197 ; void ff_float_to_int16_interleave2(int16_t *dst, const float **src, long len);
198 ;-------------------------------------------------------------------------------
199 %macro FLOAT_TO_INT16_INTERLEAVE2 1
200 cglobal float_to_int16_interleave2_%1, 3,4,2, dst, src0, src1, len
202 mov src1q, [src0q+gprsize]
210 cvtps2dq m0, [src0q+lenq]
211 cvtps2dq m1, [src1q+lenq]
217 cvtps2pi m0, [src0q+lenq ]
218 cvtps2pi m1, [src0q+lenq+8]
219 cvtps2pi m2, [src1q+lenq ]
220 cvtps2pi m3, [src1q+lenq+8]
226 mova [dstq+lenq ], m0
227 mova [dstq+lenq+8], m1
238 %define cvtps2pi pf2id
239 FLOAT_TO_INT16_INTERLEAVE2 3dnow
241 %define movdqa movaps
242 FLOAT_TO_INT16_INTERLEAVE2 sse
245 FLOAT_TO_INT16_INTERLEAVE2 sse2
251 %macro PSWAPD_3DNOW 2
257 %macro FLOAT_TO_INT16_INTERLEAVE6 1
258 ; void float_to_int16_interleave6_sse(int16_t *dst, const float **src, int len)
259 cglobal float_to_int16_interleave6_%1, 2,8,0, dst, src, src1, src2, src3, src4, src5, len
263 %define lend dword r2m
265 mov src1q, [srcq+1*gprsize]
266 mov src2q, [srcq+2*gprsize]
267 mov src3q, [srcq+3*gprsize]
268 mov src4q, [srcq+4*gprsize]
269 mov src5q, [srcq+5*gprsize]
278 cvtps2pi mm1, [srcq+src1q]
279 cvtps2pi mm2, [srcq+src2q]
280 cvtps2pi mm3, [srcq+src3q]
281 cvtps2pi mm4, [srcq+src4q]
282 cvtps2pi mm5, [srcq+src5q]
303 %endmacro ; FLOAT_TO_INT16_INTERLEAVE6
305 %define pswapd PSWAPD_SSE
306 FLOAT_TO_INT16_INTERLEAVE6 sse
307 %define cvtps2pi pf2id
308 %define pswapd PSWAPD_3DNOW
309 FLOAT_TO_INT16_INTERLEAVE6 3dnow
311 FLOAT_TO_INT16_INTERLEAVE6 3dnowext
314 ;-----------------------------------------------------------------------------
315 ; void ff_float_interleave6(float *dst, const float **src, unsigned int len);
316 ;-----------------------------------------------------------------------------
318 %macro FLOAT_INTERLEAVE6 2
319 cglobal float_interleave6_%1, 2,8,%2, dst, src, src1, src2, src3, src4, src5, len
323 %define lend dword r2m
325 mov src1q, [srcq+1*gprsize]
326 mov src2q, [srcq+2*gprsize]
327 mov src3q, [srcq+3*gprsize]
328 mov src4q, [srcq+4*gprsize]
329 mov src5q, [srcq+5*gprsize]
339 movaps m1, [srcq+src1q]
340 movaps m2, [srcq+src2q]
341 movaps m3, [srcq+src3q]
342 movaps m4, [srcq+src4q]
343 movaps m5, [srcq+src5q]
366 movq m1, [srcq+src1q]
367 movq m2, [srcq+src2q]
368 movq m3, [srcq+src3q]
369 movq m4, [srcq+src4q]
370 movq m5, [srcq+src5q]
372 SBUTTERFLY dq, 0, 1, 6
373 SBUTTERFLY dq, 2, 3, 6
374 SBUTTERFLY dq, 4, 5, 6
393 FLOAT_INTERLEAVE6 mmx, 0
395 FLOAT_INTERLEAVE6 sse, 7
397 ;-----------------------------------------------------------------------------
398 ; void ff_float_interleave2(float *dst, const float **src, unsigned int len);
399 ;-----------------------------------------------------------------------------
401 %macro FLOAT_INTERLEAVE2 2
402 cglobal float_interleave2_%1, 3,4,%2, dst, src, len, src1
403 mov src1q, [srcq+gprsize]
408 MOVPS m1, [srcq+src1q ]
409 MOVPS m3, [srcq +mmsize]
410 MOVPS m4, [srcq+src1q+mmsize]
421 MOVPS [dstq+1*mmsize], m2
422 MOVPS [dstq+2*mmsize], m3
423 MOVPS [dstq+3*mmsize], m1
437 %define PUNPCKLDQ punpckldq
438 %define PUNPCKHDQ punpckhdq
439 FLOAT_INTERLEAVE2 mmx, 0
442 %define PUNPCKLDQ unpcklps
443 %define PUNPCKHDQ unpckhps
444 FLOAT_INTERLEAVE2 sse, 5