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 %define SPLATD SPLATD_SSE
76 INT32_TO_FLOAT_FMUL_SCALAR 5
78 %define SPLATD SPLATD_SSE2
79 INT32_TO_FLOAT_FMUL_SCALAR 3
83 ;------------------------------------------------------------------------------
84 ; void ff_float_to_int16(int16_t *dst, const float *src, long len);
85 ;------------------------------------------------------------------------------
86 %macro FLOAT_TO_INT16 1
87 cglobal float_to_int16, 3, 3, %1, dst, src, len
89 lea srcq, [srcq+2*lenq]
94 cvtps2dq m0, [srcq+2*lenq ]
95 cvtps2dq m1, [srcq+2*lenq+16]
99 CVTPS2PI m0, [srcq+2*lenq ]
100 CVTPS2PI m1, [srcq+2*lenq+ 8]
101 CVTPS2PI m2, [srcq+2*lenq+16]
102 CVTPS2PI m3, [srcq+2*lenq+24]
105 mova [dstq+lenq ], m0
106 mova [dstq+lenq+8], m2
123 ;------------------------------------------------------------------------------
124 ; void ff_float_to_int16_step(int16_t *dst, const float *src, long len, long step);
125 ;------------------------------------------------------------------------------
126 %macro FLOAT_TO_INT16_STEP 1
127 cglobal float_to_int16_step, 4, 7, %1, dst, src, len, step, step3, v1, v2
129 lea srcq, [srcq+2*lenq]
130 lea step3q, [stepq*3]
134 cvtps2dq m0, [srcq+2*lenq ]
135 cvtps2dq m1, [srcq+2*lenq+16]
142 mov [dstq+stepq*4], v2w
145 mov [dstq+stepq*2], v1w
146 mov [dstq+step3q*2], v2w
147 lea dstq, [dstq+stepq*8]
152 mov [dstq+stepq*4], v2w
155 mov [dstq+stepq*2], v1w
156 mov [dstq+step3q*2], v2w
157 lea dstq, [dstq+stepq*8]
159 CVTPS2PI m0, [srcq+2*lenq ]
160 CVTPS2PI m1, [srcq+2*lenq+ 8]
161 CVTPS2PI m2, [srcq+2*lenq+16]
162 CVTPS2PI m3, [srcq+2*lenq+24]
169 mov [dstq+stepq*4], v2w
172 mov [dstq+stepq*2], v1w
173 mov [dstq+step3q*2], v2w
174 lea dstq, [dstq+stepq*8]
179 mov [dstq+stepq*4], v2w
182 mov [dstq+stepq*2], v1w
183 mov [dstq+step3q*2], v2w
184 lea dstq, [dstq+stepq*8]
195 FLOAT_TO_INT16_STEP 2
197 FLOAT_TO_INT16_STEP 0
199 FLOAT_TO_INT16_STEP 0
201 ;-------------------------------------------------------------------------------
202 ; void ff_float_to_int16_interleave2(int16_t *dst, const float **src, long len);
203 ;-------------------------------------------------------------------------------
204 %macro FLOAT_TO_INT16_INTERLEAVE2 0
205 cglobal float_to_int16_interleave2, 3, 4, 2, dst, src0, src1, len
207 mov src1q, [src0q+gprsize]
215 cvtps2dq m0, [src0q+lenq]
216 cvtps2dq m1, [src1q+lenq]
222 CVTPS2PI m0, [src0q+lenq ]
223 CVTPS2PI m1, [src0q+lenq+8]
224 CVTPS2PI m2, [src1q+lenq ]
225 CVTPS2PI m3, [src1q+lenq+8]
231 mova [dstq+lenq ], m0
232 mova [dstq+lenq+8], m1
243 FLOAT_TO_INT16_INTERLEAVE2
245 FLOAT_TO_INT16_INTERLEAVE2
247 FLOAT_TO_INT16_INTERLEAVE2
249 %macro FLOAT_TO_INT16_INTERLEAVE6 0
250 ; void float_to_int16_interleave6_sse(int16_t *dst, const float **src, int len)
251 cglobal float_to_int16_interleave6, 2, 8, 0, dst, src, src1, src2, src3, src4, src5, len
255 %define lend dword r2m
257 mov src1q, [srcq+1*gprsize]
258 mov src2q, [srcq+2*gprsize]
259 mov src3q, [srcq+3*gprsize]
260 mov src4q, [srcq+4*gprsize]
261 mov src5q, [srcq+5*gprsize]
270 CVTPS2PI mm1, [srcq+src1q]
271 CVTPS2PI mm2, [srcq+src2q]
272 CVTPS2PI mm3, [srcq+src3q]
273 CVTPS2PI mm4, [srcq+src4q]
274 CVTPS2PI mm5, [srcq+src5q]
295 %endmacro ; FLOAT_TO_INT16_INTERLEAVE6
298 FLOAT_TO_INT16_INTERLEAVE6
300 FLOAT_TO_INT16_INTERLEAVE6
302 FLOAT_TO_INT16_INTERLEAVE6
304 ;-----------------------------------------------------------------------------
305 ; void ff_float_interleave6(float *dst, const float **src, unsigned int len);
306 ;-----------------------------------------------------------------------------
308 %macro FLOAT_INTERLEAVE6 1
309 cglobal float_interleave6, 2, 8, %1, dst, src, src1, src2, src3, src4, src5, len
313 %define lend dword r2m
315 mov src1q, [srcq+1*gprsize]
316 mov src2q, [srcq+2*gprsize]
317 mov src3q, [srcq+3*gprsize]
318 mov src4q, [srcq+4*gprsize]
319 mov src5q, [srcq+5*gprsize]
329 movaps m1, [srcq+src1q]
330 movaps m2, [srcq+src2q]
331 movaps m3, [srcq+src3q]
332 movaps m4, [srcq+src4q]
333 movaps m5, [srcq+src5q]
356 movq m1, [srcq+src1q]
357 movq m2, [srcq+src2q]
358 movq m3, [srcq+src3q]
359 movq m4, [srcq+src4q]
360 movq m5, [srcq+src5q]
362 SBUTTERFLY dq, 0, 1, 6
363 SBUTTERFLY dq, 2, 3, 6
364 SBUTTERFLY dq, 4, 5, 6
387 ;-----------------------------------------------------------------------------
388 ; void ff_float_interleave2(float *dst, const float **src, unsigned int len);
389 ;-----------------------------------------------------------------------------
391 %macro FLOAT_INTERLEAVE2 1
392 cglobal float_interleave2, 3, 4, %1, dst, src, len, src1
393 mov src1q, [srcq+gprsize]
398 mova m1, [srcq+src1q ]
399 mova m3, [srcq +mmsize]
400 mova m4, [srcq+src1q+mmsize]
411 mova [dstq+1*mmsize], m2
412 mova [dstq+2*mmsize], m3
413 mova [dstq+3*mmsize], m1
426 %define PUNPCKLDQ punpckldq
427 %define PUNPCKHDQ punpckhdq
430 %define PUNPCKLDQ unpcklps
431 %define PUNPCKHDQ unpckhps