123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896 |
- /*
- * DSP utils : average functions are compiled twice for 3dnow/mmx2
- * Copyright (c) 2000, 2001 Fabrice Bellard
- * Copyright (c) 2002-2004 Michael Niedermayer
- *
- * MMX optimization by Nick Kurshev <nickols_k@mail.ru>
- * mostly rewritten by Michael Niedermayer <michaelni@gmx.at>
- * and improved by Zdenek Kabelac <kabi@users.sf.net>
- *
- * This file is part of FFmpeg.
- *
- * FFmpeg is free software; you can redistribute it and/or
- * modify it under the terms of the GNU Lesser General Public
- * License as published by the Free Software Foundation; either
- * version 2.1 of the License, or (at your option) any later version.
- *
- * FFmpeg is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
- * Lesser General Public License for more details.
- *
- * You should have received a copy of the GNU Lesser General Public
- * License along with FFmpeg; if not, write to the Free Software
- * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
- */
- /* XXX: we use explicit registers to avoid a gcc 2.95.2 register asm
- clobber bug - now it will work with 2.95.2 and also with -fPIC
- */
- static void DEF(put_pixels8_x2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
- {
- __asm__ volatile(
- "lea (%3, %3), %%"REG_a" \n\t"
- "1: \n\t"
- "movq (%1), %%mm0 \n\t"
- "movq (%1, %3), %%mm1 \n\t"
- PAVGB" 1(%1), %%mm0 \n\t"
- PAVGB" 1(%1, %3), %%mm1 \n\t"
- "movq %%mm0, (%2) \n\t"
- "movq %%mm1, (%2, %3) \n\t"
- "add %%"REG_a", %1 \n\t"
- "add %%"REG_a", %2 \n\t"
- "movq (%1), %%mm0 \n\t"
- "movq (%1, %3), %%mm1 \n\t"
- PAVGB" 1(%1), %%mm0 \n\t"
- PAVGB" 1(%1, %3), %%mm1 \n\t"
- "add %%"REG_a", %1 \n\t"
- "movq %%mm0, (%2) \n\t"
- "movq %%mm1, (%2, %3) \n\t"
- "add %%"REG_a", %2 \n\t"
- "subl $4, %0 \n\t"
- "jnz 1b \n\t"
- :"+g"(h), "+S"(pixels), "+D"(block)
- :"r" ((x86_reg)line_size)
- :"%"REG_a, "memory");
- }
- static void DEF(put_pixels4_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
- {
- __asm__ volatile(
- "testl $1, %0 \n\t"
- " jz 1f \n\t"
- "movd (%1), %%mm0 \n\t"
- "movd (%2), %%mm1 \n\t"
- "add %4, %1 \n\t"
- "add $4, %2 \n\t"
- PAVGB" %%mm1, %%mm0 \n\t"
- "movd %%mm0, (%3) \n\t"
- "add %5, %3 \n\t"
- "decl %0 \n\t"
- "1: \n\t"
- "movd (%1), %%mm0 \n\t"
- "add %4, %1 \n\t"
- "movd (%1), %%mm1 \n\t"
- "movd (%2), %%mm2 \n\t"
- "movd 4(%2), %%mm3 \n\t"
- "add %4, %1 \n\t"
- PAVGB" %%mm2, %%mm0 \n\t"
- PAVGB" %%mm3, %%mm1 \n\t"
- "movd %%mm0, (%3) \n\t"
- "add %5, %3 \n\t"
- "movd %%mm1, (%3) \n\t"
- "add %5, %3 \n\t"
- "movd (%1), %%mm0 \n\t"
- "add %4, %1 \n\t"
- "movd (%1), %%mm1 \n\t"
- "movd 8(%2), %%mm2 \n\t"
- "movd 12(%2), %%mm3 \n\t"
- "add %4, %1 \n\t"
- PAVGB" %%mm2, %%mm0 \n\t"
- PAVGB" %%mm3, %%mm1 \n\t"
- "movd %%mm0, (%3) \n\t"
- "add %5, %3 \n\t"
- "movd %%mm1, (%3) \n\t"
- "add %5, %3 \n\t"
- "add $16, %2 \n\t"
- "subl $4, %0 \n\t"
- "jnz 1b \n\t"
- #ifdef PIC //Note "+bm" and "+mb" are buggy too (with gcc 3.2.2 at least) and cannot be used
- :"+m"(h), "+a"(src1), "+c"(src2), "+d"(dst)
- #else
- :"+b"(h), "+a"(src1), "+c"(src2), "+d"(dst)
- #endif
- :"S"((x86_reg)src1Stride), "D"((x86_reg)dstStride)
- :"memory");
- }
- static void DEF(put_pixels8_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
- {
- __asm__ volatile(
- "testl $1, %0 \n\t"
- " jz 1f \n\t"
- "movq (%1), %%mm0 \n\t"
- "movq (%2), %%mm1 \n\t"
- "add %4, %1 \n\t"
- "add $8, %2 \n\t"
- PAVGB" %%mm1, %%mm0 \n\t"
- "movq %%mm0, (%3) \n\t"
- "add %5, %3 \n\t"
- "decl %0 \n\t"
- "1: \n\t"
- "movq (%1), %%mm0 \n\t"
- "add %4, %1 \n\t"
- "movq (%1), %%mm1 \n\t"
- "add %4, %1 \n\t"
- PAVGB" (%2), %%mm0 \n\t"
- PAVGB" 8(%2), %%mm1 \n\t"
- "movq %%mm0, (%3) \n\t"
- "add %5, %3 \n\t"
- "movq %%mm1, (%3) \n\t"
- "add %5, %3 \n\t"
- "movq (%1), %%mm0 \n\t"
- "add %4, %1 \n\t"
- "movq (%1), %%mm1 \n\t"
- "add %4, %1 \n\t"
- PAVGB" 16(%2), %%mm0 \n\t"
- PAVGB" 24(%2), %%mm1 \n\t"
- "movq %%mm0, (%3) \n\t"
- "add %5, %3 \n\t"
- "movq %%mm1, (%3) \n\t"
- "add %5, %3 \n\t"
- "add $32, %2 \n\t"
- "subl $4, %0 \n\t"
- "jnz 1b \n\t"
- #ifdef PIC //Note "+bm" and "+mb" are buggy too (with gcc 3.2.2 at least) and cannot be used
- :"+m"(h), "+a"(src1), "+c"(src2), "+d"(dst)
- #else
- :"+b"(h), "+a"(src1), "+c"(src2), "+d"(dst)
- #endif
- :"S"((x86_reg)src1Stride), "D"((x86_reg)dstStride)
- :"memory");
- //the following should be used, though better not with gcc ...
- /* :"+g"(h), "+r"(src1), "+r"(src2), "+r"(dst)
- :"r"(src1Stride), "r"(dstStride)
- :"memory");*/
- }
- static void DEF(put_no_rnd_pixels8_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
- {
- __asm__ volatile(
- "pcmpeqb %%mm6, %%mm6 \n\t"
- "testl $1, %0 \n\t"
- " jz 1f \n\t"
- "movq (%1), %%mm0 \n\t"
- "movq (%2), %%mm1 \n\t"
- "add %4, %1 \n\t"
- "add $8, %2 \n\t"
- "pxor %%mm6, %%mm0 \n\t"
- "pxor %%mm6, %%mm1 \n\t"
- PAVGB" %%mm1, %%mm0 \n\t"
- "pxor %%mm6, %%mm0 \n\t"
- "movq %%mm0, (%3) \n\t"
- "add %5, %3 \n\t"
- "decl %0 \n\t"
- "1: \n\t"
- "movq (%1), %%mm0 \n\t"
- "add %4, %1 \n\t"
- "movq (%1), %%mm1 \n\t"
- "add %4, %1 \n\t"
- "movq (%2), %%mm2 \n\t"
- "movq 8(%2), %%mm3 \n\t"
- "pxor %%mm6, %%mm0 \n\t"
- "pxor %%mm6, %%mm1 \n\t"
- "pxor %%mm6, %%mm2 \n\t"
- "pxor %%mm6, %%mm3 \n\t"
- PAVGB" %%mm2, %%mm0 \n\t"
- PAVGB" %%mm3, %%mm1 \n\t"
- "pxor %%mm6, %%mm0 \n\t"
- "pxor %%mm6, %%mm1 \n\t"
- "movq %%mm0, (%3) \n\t"
- "add %5, %3 \n\t"
- "movq %%mm1, (%3) \n\t"
- "add %5, %3 \n\t"
- "movq (%1), %%mm0 \n\t"
- "add %4, %1 \n\t"
- "movq (%1), %%mm1 \n\t"
- "add %4, %1 \n\t"
- "movq 16(%2), %%mm2 \n\t"
- "movq 24(%2), %%mm3 \n\t"
- "pxor %%mm6, %%mm0 \n\t"
- "pxor %%mm6, %%mm1 \n\t"
- "pxor %%mm6, %%mm2 \n\t"
- "pxor %%mm6, %%mm3 \n\t"
- PAVGB" %%mm2, %%mm0 \n\t"
- PAVGB" %%mm3, %%mm1 \n\t"
- "pxor %%mm6, %%mm0 \n\t"
- "pxor %%mm6, %%mm1 \n\t"
- "movq %%mm0, (%3) \n\t"
- "add %5, %3 \n\t"
- "movq %%mm1, (%3) \n\t"
- "add %5, %3 \n\t"
- "add $32, %2 \n\t"
- "subl $4, %0 \n\t"
- "jnz 1b \n\t"
- #ifdef PIC //Note "+bm" and "+mb" are buggy too (with gcc 3.2.2 at least) and cannot be used
- :"+m"(h), "+a"(src1), "+c"(src2), "+d"(dst)
- #else
- :"+b"(h), "+a"(src1), "+c"(src2), "+d"(dst)
- #endif
- :"S"((x86_reg)src1Stride), "D"((x86_reg)dstStride)
- :"memory");
- //the following should be used, though better not with gcc ...
- /* :"+g"(h), "+r"(src1), "+r"(src2), "+r"(dst)
- :"r"(src1Stride), "r"(dstStride)
- :"memory");*/
- }
- static void DEF(avg_pixels4_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
- {
- __asm__ volatile(
- "testl $1, %0 \n\t"
- " jz 1f \n\t"
- "movd (%1), %%mm0 \n\t"
- "movd (%2), %%mm1 \n\t"
- "add %4, %1 \n\t"
- "add $4, %2 \n\t"
- PAVGB" %%mm1, %%mm0 \n\t"
- PAVGB" (%3), %%mm0 \n\t"
- "movd %%mm0, (%3) \n\t"
- "add %5, %3 \n\t"
- "decl %0 \n\t"
- "1: \n\t"
- "movd (%1), %%mm0 \n\t"
- "add %4, %1 \n\t"
- "movd (%1), %%mm1 \n\t"
- "add %4, %1 \n\t"
- PAVGB" (%2), %%mm0 \n\t"
- PAVGB" 4(%2), %%mm1 \n\t"
- PAVGB" (%3), %%mm0 \n\t"
- "movd %%mm0, (%3) \n\t"
- "add %5, %3 \n\t"
- PAVGB" (%3), %%mm1 \n\t"
- "movd %%mm1, (%3) \n\t"
- "add %5, %3 \n\t"
- "movd (%1), %%mm0 \n\t"
- "add %4, %1 \n\t"
- "movd (%1), %%mm1 \n\t"
- "add %4, %1 \n\t"
- PAVGB" 8(%2), %%mm0 \n\t"
- PAVGB" 12(%2), %%mm1 \n\t"
- PAVGB" (%3), %%mm0 \n\t"
- "movd %%mm0, (%3) \n\t"
- "add %5, %3 \n\t"
- PAVGB" (%3), %%mm1 \n\t"
- "movd %%mm1, (%3) \n\t"
- "add %5, %3 \n\t"
- "add $16, %2 \n\t"
- "subl $4, %0 \n\t"
- "jnz 1b \n\t"
- #ifdef PIC //Note "+bm" and "+mb" are buggy too (with gcc 3.2.2 at least) and cannot be used
- :"+m"(h), "+a"(src1), "+c"(src2), "+d"(dst)
- #else
- :"+b"(h), "+a"(src1), "+c"(src2), "+d"(dst)
- #endif
- :"S"((x86_reg)src1Stride), "D"((x86_reg)dstStride)
- :"memory");
- }
- static void DEF(avg_pixels8_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
- {
- __asm__ volatile(
- "testl $1, %0 \n\t"
- " jz 1f \n\t"
- "movq (%1), %%mm0 \n\t"
- "movq (%2), %%mm1 \n\t"
- "add %4, %1 \n\t"
- "add $8, %2 \n\t"
- PAVGB" %%mm1, %%mm0 \n\t"
- PAVGB" (%3), %%mm0 \n\t"
- "movq %%mm0, (%3) \n\t"
- "add %5, %3 \n\t"
- "decl %0 \n\t"
- "1: \n\t"
- "movq (%1), %%mm0 \n\t"
- "add %4, %1 \n\t"
- "movq (%1), %%mm1 \n\t"
- "add %4, %1 \n\t"
- PAVGB" (%2), %%mm0 \n\t"
- PAVGB" 8(%2), %%mm1 \n\t"
- PAVGB" (%3), %%mm0 \n\t"
- "movq %%mm0, (%3) \n\t"
- "add %5, %3 \n\t"
- PAVGB" (%3), %%mm1 \n\t"
- "movq %%mm1, (%3) \n\t"
- "add %5, %3 \n\t"
- "movq (%1), %%mm0 \n\t"
- "add %4, %1 \n\t"
- "movq (%1), %%mm1 \n\t"
- "add %4, %1 \n\t"
- PAVGB" 16(%2), %%mm0 \n\t"
- PAVGB" 24(%2), %%mm1 \n\t"
- PAVGB" (%3), %%mm0 \n\t"
- "movq %%mm0, (%3) \n\t"
- "add %5, %3 \n\t"
- PAVGB" (%3), %%mm1 \n\t"
- "movq %%mm1, (%3) \n\t"
- "add %5, %3 \n\t"
- "add $32, %2 \n\t"
- "subl $4, %0 \n\t"
- "jnz 1b \n\t"
- #ifdef PIC //Note "+bm" and "+mb" are buggy too (with gcc 3.2.2 at least) and cannot be used
- :"+m"(h), "+a"(src1), "+c"(src2), "+d"(dst)
- #else
- :"+b"(h), "+a"(src1), "+c"(src2), "+d"(dst)
- #endif
- :"S"((x86_reg)src1Stride), "D"((x86_reg)dstStride)
- :"memory");
- //the following should be used, though better not with gcc ...
- /* :"+g"(h), "+r"(src1), "+r"(src2), "+r"(dst)
- :"r"(src1Stride), "r"(dstStride)
- :"memory");*/
- }
- static void DEF(put_pixels16_x2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
- {
- __asm__ volatile(
- "lea (%3, %3), %%"REG_a" \n\t"
- "1: \n\t"
- "movq (%1), %%mm0 \n\t"
- "movq (%1, %3), %%mm1 \n\t"
- "movq 8(%1), %%mm2 \n\t"
- "movq 8(%1, %3), %%mm3 \n\t"
- PAVGB" 1(%1), %%mm0 \n\t"
- PAVGB" 1(%1, %3), %%mm1 \n\t"
- PAVGB" 9(%1), %%mm2 \n\t"
- PAVGB" 9(%1, %3), %%mm3 \n\t"
- "movq %%mm0, (%2) \n\t"
- "movq %%mm1, (%2, %3) \n\t"
- "movq %%mm2, 8(%2) \n\t"
- "movq %%mm3, 8(%2, %3) \n\t"
- "add %%"REG_a", %1 \n\t"
- "add %%"REG_a", %2 \n\t"
- "movq (%1), %%mm0 \n\t"
- "movq (%1, %3), %%mm1 \n\t"
- "movq 8(%1), %%mm2 \n\t"
- "movq 8(%1, %3), %%mm3 \n\t"
- PAVGB" 1(%1), %%mm0 \n\t"
- PAVGB" 1(%1, %3), %%mm1 \n\t"
- PAVGB" 9(%1), %%mm2 \n\t"
- PAVGB" 9(%1, %3), %%mm3 \n\t"
- "add %%"REG_a", %1 \n\t"
- "movq %%mm0, (%2) \n\t"
- "movq %%mm1, (%2, %3) \n\t"
- "movq %%mm2, 8(%2) \n\t"
- "movq %%mm3, 8(%2, %3) \n\t"
- "add %%"REG_a", %2 \n\t"
- "subl $4, %0 \n\t"
- "jnz 1b \n\t"
- :"+g"(h), "+S"(pixels), "+D"(block)
- :"r" ((x86_reg)line_size)
- :"%"REG_a, "memory");
- }
- static void DEF(put_pixels16_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
- {
- __asm__ volatile(
- "testl $1, %0 \n\t"
- " jz 1f \n\t"
- "movq (%1), %%mm0 \n\t"
- "movq 8(%1), %%mm1 \n\t"
- PAVGB" (%2), %%mm0 \n\t"
- PAVGB" 8(%2), %%mm1 \n\t"
- "add %4, %1 \n\t"
- "add $16, %2 \n\t"
- "movq %%mm0, (%3) \n\t"
- "movq %%mm1, 8(%3) \n\t"
- "add %5, %3 \n\t"
- "decl %0 \n\t"
- "1: \n\t"
- "movq (%1), %%mm0 \n\t"
- "movq 8(%1), %%mm1 \n\t"
- "add %4, %1 \n\t"
- PAVGB" (%2), %%mm0 \n\t"
- PAVGB" 8(%2), %%mm1 \n\t"
- "movq %%mm0, (%3) \n\t"
- "movq %%mm1, 8(%3) \n\t"
- "add %5, %3 \n\t"
- "movq (%1), %%mm0 \n\t"
- "movq 8(%1), %%mm1 \n\t"
- "add %4, %1 \n\t"
- PAVGB" 16(%2), %%mm0 \n\t"
- PAVGB" 24(%2), %%mm1 \n\t"
- "movq %%mm0, (%3) \n\t"
- "movq %%mm1, 8(%3) \n\t"
- "add %5, %3 \n\t"
- "add $32, %2 \n\t"
- "subl $2, %0 \n\t"
- "jnz 1b \n\t"
- #ifdef PIC //Note "+bm" and "+mb" are buggy too (with gcc 3.2.2 at least) and cannot be used
- :"+m"(h), "+a"(src1), "+c"(src2), "+d"(dst)
- #else
- :"+b"(h), "+a"(src1), "+c"(src2), "+d"(dst)
- #endif
- :"S"((x86_reg)src1Stride), "D"((x86_reg)dstStride)
- :"memory");
- //the following should be used, though better not with gcc ...
- /* :"+g"(h), "+r"(src1), "+r"(src2), "+r"(dst)
- :"r"(src1Stride), "r"(dstStride)
- :"memory");*/
- }
- static void DEF(avg_pixels16_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
- {
- __asm__ volatile(
- "testl $1, %0 \n\t"
- " jz 1f \n\t"
- "movq (%1), %%mm0 \n\t"
- "movq 8(%1), %%mm1 \n\t"
- PAVGB" (%2), %%mm0 \n\t"
- PAVGB" 8(%2), %%mm1 \n\t"
- "add %4, %1 \n\t"
- "add $16, %2 \n\t"
- PAVGB" (%3), %%mm0 \n\t"
- PAVGB" 8(%3), %%mm1 \n\t"
- "movq %%mm0, (%3) \n\t"
- "movq %%mm1, 8(%3) \n\t"
- "add %5, %3 \n\t"
- "decl %0 \n\t"
- "1: \n\t"
- "movq (%1), %%mm0 \n\t"
- "movq 8(%1), %%mm1 \n\t"
- "add %4, %1 \n\t"
- PAVGB" (%2), %%mm0 \n\t"
- PAVGB" 8(%2), %%mm1 \n\t"
- PAVGB" (%3), %%mm0 \n\t"
- PAVGB" 8(%3), %%mm1 \n\t"
- "movq %%mm0, (%3) \n\t"
- "movq %%mm1, 8(%3) \n\t"
- "add %5, %3 \n\t"
- "movq (%1), %%mm0 \n\t"
- "movq 8(%1), %%mm1 \n\t"
- "add %4, %1 \n\t"
- PAVGB" 16(%2), %%mm0 \n\t"
- PAVGB" 24(%2), %%mm1 \n\t"
- PAVGB" (%3), %%mm0 \n\t"
- PAVGB" 8(%3), %%mm1 \n\t"
- "movq %%mm0, (%3) \n\t"
- "movq %%mm1, 8(%3) \n\t"
- "add %5, %3 \n\t"
- "add $32, %2 \n\t"
- "subl $2, %0 \n\t"
- "jnz 1b \n\t"
- #ifdef PIC //Note "+bm" and "+mb" are buggy too (with gcc 3.2.2 at least) and cannot be used
- :"+m"(h), "+a"(src1), "+c"(src2), "+d"(dst)
- #else
- :"+b"(h), "+a"(src1), "+c"(src2), "+d"(dst)
- #endif
- :"S"((x86_reg)src1Stride), "D"((x86_reg)dstStride)
- :"memory");
- //the following should be used, though better not with gcc ...
- /* :"+g"(h), "+r"(src1), "+r"(src2), "+r"(dst)
- :"r"(src1Stride), "r"(dstStride)
- :"memory");*/
- }
- static void DEF(put_no_rnd_pixels16_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
- {
- __asm__ volatile(
- "pcmpeqb %%mm6, %%mm6 \n\t"
- "testl $1, %0 \n\t"
- " jz 1f \n\t"
- "movq (%1), %%mm0 \n\t"
- "movq 8(%1), %%mm1 \n\t"
- "movq (%2), %%mm2 \n\t"
- "movq 8(%2), %%mm3 \n\t"
- "pxor %%mm6, %%mm0 \n\t"
- "pxor %%mm6, %%mm1 \n\t"
- "pxor %%mm6, %%mm2 \n\t"
- "pxor %%mm6, %%mm3 \n\t"
- PAVGB" %%mm2, %%mm0 \n\t"
- PAVGB" %%mm3, %%mm1 \n\t"
- "pxor %%mm6, %%mm0 \n\t"
- "pxor %%mm6, %%mm1 \n\t"
- "add %4, %1 \n\t"
- "add $16, %2 \n\t"
- "movq %%mm0, (%3) \n\t"
- "movq %%mm1, 8(%3) \n\t"
- "add %5, %3 \n\t"
- "decl %0 \n\t"
- "1: \n\t"
- "movq (%1), %%mm0 \n\t"
- "movq 8(%1), %%mm1 \n\t"
- "add %4, %1 \n\t"
- "movq (%2), %%mm2 \n\t"
- "movq 8(%2), %%mm3 \n\t"
- "pxor %%mm6, %%mm0 \n\t"
- "pxor %%mm6, %%mm1 \n\t"
- "pxor %%mm6, %%mm2 \n\t"
- "pxor %%mm6, %%mm3 \n\t"
- PAVGB" %%mm2, %%mm0 \n\t"
- PAVGB" %%mm3, %%mm1 \n\t"
- "pxor %%mm6, %%mm0 \n\t"
- "pxor %%mm6, %%mm1 \n\t"
- "movq %%mm0, (%3) \n\t"
- "movq %%mm1, 8(%3) \n\t"
- "add %5, %3 \n\t"
- "movq (%1), %%mm0 \n\t"
- "movq 8(%1), %%mm1 \n\t"
- "add %4, %1 \n\t"
- "movq 16(%2), %%mm2 \n\t"
- "movq 24(%2), %%mm3 \n\t"
- "pxor %%mm6, %%mm0 \n\t"
- "pxor %%mm6, %%mm1 \n\t"
- "pxor %%mm6, %%mm2 \n\t"
- "pxor %%mm6, %%mm3 \n\t"
- PAVGB" %%mm2, %%mm0 \n\t"
- PAVGB" %%mm3, %%mm1 \n\t"
- "pxor %%mm6, %%mm0 \n\t"
- "pxor %%mm6, %%mm1 \n\t"
- "movq %%mm0, (%3) \n\t"
- "movq %%mm1, 8(%3) \n\t"
- "add %5, %3 \n\t"
- "add $32, %2 \n\t"
- "subl $2, %0 \n\t"
- "jnz 1b \n\t"
- #ifdef PIC //Note "+bm" and "+mb" are buggy too (with gcc 3.2.2 at least) and cannot be used
- :"+m"(h), "+a"(src1), "+c"(src2), "+d"(dst)
- #else
- :"+b"(h), "+a"(src1), "+c"(src2), "+d"(dst)
- #endif
- :"S"((x86_reg)src1Stride), "D"((x86_reg)dstStride)
- :"memory");
- //the following should be used, though better not with gcc ...
- /* :"+g"(h), "+r"(src1), "+r"(src2), "+r"(dst)
- :"r"(src1Stride), "r"(dstStride)
- :"memory");*/
- }
- /* GL: this function does incorrect rounding if overflow */
- static void DEF(put_no_rnd_pixels8_x2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
- {
- MOVQ_BONE(mm6);
- __asm__ volatile(
- "lea (%3, %3), %%"REG_a" \n\t"
- "1: \n\t"
- "movq (%1), %%mm0 \n\t"
- "movq (%1, %3), %%mm2 \n\t"
- "movq 1(%1), %%mm1 \n\t"
- "movq 1(%1, %3), %%mm3 \n\t"
- "add %%"REG_a", %1 \n\t"
- "psubusb %%mm6, %%mm0 \n\t"
- "psubusb %%mm6, %%mm2 \n\t"
- PAVGB" %%mm1, %%mm0 \n\t"
- PAVGB" %%mm3, %%mm2 \n\t"
- "movq %%mm0, (%2) \n\t"
- "movq %%mm2, (%2, %3) \n\t"
- "movq (%1), %%mm0 \n\t"
- "movq 1(%1), %%mm1 \n\t"
- "movq (%1, %3), %%mm2 \n\t"
- "movq 1(%1, %3), %%mm3 \n\t"
- "add %%"REG_a", %2 \n\t"
- "add %%"REG_a", %1 \n\t"
- "psubusb %%mm6, %%mm0 \n\t"
- "psubusb %%mm6, %%mm2 \n\t"
- PAVGB" %%mm1, %%mm0 \n\t"
- PAVGB" %%mm3, %%mm2 \n\t"
- "movq %%mm0, (%2) \n\t"
- "movq %%mm2, (%2, %3) \n\t"
- "add %%"REG_a", %2 \n\t"
- "subl $4, %0 \n\t"
- "jnz 1b \n\t"
- :"+g"(h), "+S"(pixels), "+D"(block)
- :"r" ((x86_reg)line_size)
- :"%"REG_a, "memory");
- }
- static void DEF(put_pixels8_y2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
- {
- __asm__ volatile(
- "lea (%3, %3), %%"REG_a" \n\t"
- "movq (%1), %%mm0 \n\t"
- "sub %3, %2 \n\t"
- "1: \n\t"
- "movq (%1, %3), %%mm1 \n\t"
- "movq (%1, %%"REG_a"), %%mm2 \n\t"
- "add %%"REG_a", %1 \n\t"
- PAVGB" %%mm1, %%mm0 \n\t"
- PAVGB" %%mm2, %%mm1 \n\t"
- "movq %%mm0, (%2, %3) \n\t"
- "movq %%mm1, (%2, %%"REG_a") \n\t"
- "movq (%1, %3), %%mm1 \n\t"
- "movq (%1, %%"REG_a"), %%mm0 \n\t"
- "add %%"REG_a", %2 \n\t"
- "add %%"REG_a", %1 \n\t"
- PAVGB" %%mm1, %%mm2 \n\t"
- PAVGB" %%mm0, %%mm1 \n\t"
- "movq %%mm2, (%2, %3) \n\t"
- "movq %%mm1, (%2, %%"REG_a") \n\t"
- "add %%"REG_a", %2 \n\t"
- "subl $4, %0 \n\t"
- "jnz 1b \n\t"
- :"+g"(h), "+S"(pixels), "+D" (block)
- :"r" ((x86_reg)line_size)
- :"%"REG_a, "memory");
- }
- /* GL: this function does incorrect rounding if overflow */
- static void DEF(put_no_rnd_pixels8_y2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
- {
- MOVQ_BONE(mm6);
- __asm__ volatile(
- "lea (%3, %3), %%"REG_a" \n\t"
- "movq (%1), %%mm0 \n\t"
- "sub %3, %2 \n\t"
- "1: \n\t"
- "movq (%1, %3), %%mm1 \n\t"
- "movq (%1, %%"REG_a"), %%mm2 \n\t"
- "add %%"REG_a", %1 \n\t"
- "psubusb %%mm6, %%mm1 \n\t"
- PAVGB" %%mm1, %%mm0 \n\t"
- PAVGB" %%mm2, %%mm1 \n\t"
- "movq %%mm0, (%2, %3) \n\t"
- "movq %%mm1, (%2, %%"REG_a") \n\t"
- "movq (%1, %3), %%mm1 \n\t"
- "movq (%1, %%"REG_a"), %%mm0 \n\t"
- "add %%"REG_a", %2 \n\t"
- "add %%"REG_a", %1 \n\t"
- "psubusb %%mm6, %%mm1 \n\t"
- PAVGB" %%mm1, %%mm2 \n\t"
- PAVGB" %%mm0, %%mm1 \n\t"
- "movq %%mm2, (%2, %3) \n\t"
- "movq %%mm1, (%2, %%"REG_a") \n\t"
- "add %%"REG_a", %2 \n\t"
- "subl $4, %0 \n\t"
- "jnz 1b \n\t"
- :"+g"(h), "+S"(pixels), "+D" (block)
- :"r" ((x86_reg)line_size)
- :"%"REG_a, "memory");
- }
- static void DEF(avg_pixels8)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
- {
- __asm__ volatile(
- "lea (%3, %3), %%"REG_a" \n\t"
- "1: \n\t"
- "movq (%2), %%mm0 \n\t"
- "movq (%2, %3), %%mm1 \n\t"
- PAVGB" (%1), %%mm0 \n\t"
- PAVGB" (%1, %3), %%mm1 \n\t"
- "movq %%mm0, (%2) \n\t"
- "movq %%mm1, (%2, %3) \n\t"
- "add %%"REG_a", %1 \n\t"
- "add %%"REG_a", %2 \n\t"
- "movq (%2), %%mm0 \n\t"
- "movq (%2, %3), %%mm1 \n\t"
- PAVGB" (%1), %%mm0 \n\t"
- PAVGB" (%1, %3), %%mm1 \n\t"
- "add %%"REG_a", %1 \n\t"
- "movq %%mm0, (%2) \n\t"
- "movq %%mm1, (%2, %3) \n\t"
- "add %%"REG_a", %2 \n\t"
- "subl $4, %0 \n\t"
- "jnz 1b \n\t"
- :"+g"(h), "+S"(pixels), "+D"(block)
- :"r" ((x86_reg)line_size)
- :"%"REG_a, "memory");
- }
- static void DEF(avg_pixels8_x2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
- {
- __asm__ volatile(
- "lea (%3, %3), %%"REG_a" \n\t"
- "1: \n\t"
- "movq (%1), %%mm0 \n\t"
- "movq (%1, %3), %%mm2 \n\t"
- PAVGB" 1(%1), %%mm0 \n\t"
- PAVGB" 1(%1, %3), %%mm2 \n\t"
- PAVGB" (%2), %%mm0 \n\t"
- PAVGB" (%2, %3), %%mm2 \n\t"
- "add %%"REG_a", %1 \n\t"
- "movq %%mm0, (%2) \n\t"
- "movq %%mm2, (%2, %3) \n\t"
- "movq (%1), %%mm0 \n\t"
- "movq (%1, %3), %%mm2 \n\t"
- PAVGB" 1(%1), %%mm0 \n\t"
- PAVGB" 1(%1, %3), %%mm2 \n\t"
- "add %%"REG_a", %2 \n\t"
- "add %%"REG_a", %1 \n\t"
- PAVGB" (%2), %%mm0 \n\t"
- PAVGB" (%2, %3), %%mm2 \n\t"
- "movq %%mm0, (%2) \n\t"
- "movq %%mm2, (%2, %3) \n\t"
- "add %%"REG_a", %2 \n\t"
- "subl $4, %0 \n\t"
- "jnz 1b \n\t"
- :"+g"(h), "+S"(pixels), "+D"(block)
- :"r" ((x86_reg)line_size)
- :"%"REG_a, "memory");
- }
- static void DEF(avg_pixels8_y2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
- {
- __asm__ volatile(
- "lea (%3, %3), %%"REG_a" \n\t"
- "movq (%1), %%mm0 \n\t"
- "sub %3, %2 \n\t"
- "1: \n\t"
- "movq (%1, %3), %%mm1 \n\t"
- "movq (%1, %%"REG_a"), %%mm2 \n\t"
- "add %%"REG_a", %1 \n\t"
- PAVGB" %%mm1, %%mm0 \n\t"
- PAVGB" %%mm2, %%mm1 \n\t"
- "movq (%2, %3), %%mm3 \n\t"
- "movq (%2, %%"REG_a"), %%mm4 \n\t"
- PAVGB" %%mm3, %%mm0 \n\t"
- PAVGB" %%mm4, %%mm1 \n\t"
- "movq %%mm0, (%2, %3) \n\t"
- "movq %%mm1, (%2, %%"REG_a") \n\t"
- "movq (%1, %3), %%mm1 \n\t"
- "movq (%1, %%"REG_a"), %%mm0 \n\t"
- PAVGB" %%mm1, %%mm2 \n\t"
- PAVGB" %%mm0, %%mm1 \n\t"
- "add %%"REG_a", %2 \n\t"
- "add %%"REG_a", %1 \n\t"
- "movq (%2, %3), %%mm3 \n\t"
- "movq (%2, %%"REG_a"), %%mm4 \n\t"
- PAVGB" %%mm3, %%mm2 \n\t"
- PAVGB" %%mm4, %%mm1 \n\t"
- "movq %%mm2, (%2, %3) \n\t"
- "movq %%mm1, (%2, %%"REG_a") \n\t"
- "add %%"REG_a", %2 \n\t"
- "subl $4, %0 \n\t"
- "jnz 1b \n\t"
- :"+g"(h), "+S"(pixels), "+D"(block)
- :"r" ((x86_reg)line_size)
- :"%"REG_a, "memory");
- }
- /* Note this is not correctly rounded, but this function is only
- * used for B-frames so it does not matter. */
- static void DEF(avg_pixels8_xy2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
- {
- MOVQ_BONE(mm6);
- __asm__ volatile(
- "lea (%3, %3), %%"REG_a" \n\t"
- "movq (%1), %%mm0 \n\t"
- PAVGB" 1(%1), %%mm0 \n\t"
- ASMALIGN(3)
- "1: \n\t"
- "movq (%1, %%"REG_a"), %%mm2 \n\t"
- "movq (%1, %3), %%mm1 \n\t"
- "psubusb %%mm6, %%mm2 \n\t"
- PAVGB" 1(%1, %3), %%mm1 \n\t"
- PAVGB" 1(%1, %%"REG_a"), %%mm2 \n\t"
- "add %%"REG_a", %1 \n\t"
- PAVGB" %%mm1, %%mm0 \n\t"
- PAVGB" %%mm2, %%mm1 \n\t"
- PAVGB" (%2), %%mm0 \n\t"
- PAVGB" (%2, %3), %%mm1 \n\t"
- "movq %%mm0, (%2) \n\t"
- "movq %%mm1, (%2, %3) \n\t"
- "movq (%1, %3), %%mm1 \n\t"
- "movq (%1, %%"REG_a"), %%mm0 \n\t"
- PAVGB" 1(%1, %3), %%mm1 \n\t"
- PAVGB" 1(%1, %%"REG_a"), %%mm0 \n\t"
- "add %%"REG_a", %2 \n\t"
- "add %%"REG_a", %1 \n\t"
- PAVGB" %%mm1, %%mm2 \n\t"
- PAVGB" %%mm0, %%mm1 \n\t"
- PAVGB" (%2), %%mm2 \n\t"
- PAVGB" (%2, %3), %%mm1 \n\t"
- "movq %%mm2, (%2) \n\t"
- "movq %%mm1, (%2, %3) \n\t"
- "add %%"REG_a", %2 \n\t"
- "subl $4, %0 \n\t"
- "jnz 1b \n\t"
- :"+g"(h), "+S"(pixels), "+D"(block)
- :"r" ((x86_reg)line_size)
- :"%"REG_a, "memory");
- }
- static void DEF(avg_pixels4)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
- {
- do {
- __asm__ volatile(
- "movd (%1), %%mm0 \n\t"
- "movd (%1, %2), %%mm1 \n\t"
- "movd (%1, %2, 2), %%mm2 \n\t"
- "movd (%1, %3), %%mm3 \n\t"
- PAVGB" (%0), %%mm0 \n\t"
- PAVGB" (%0, %2), %%mm1 \n\t"
- PAVGB" (%0, %2, 2), %%mm2 \n\t"
- PAVGB" (%0, %3), %%mm3 \n\t"
- "movd %%mm0, (%1) \n\t"
- "movd %%mm1, (%1, %2) \n\t"
- "movd %%mm2, (%1, %2, 2) \n\t"
- "movd %%mm3, (%1, %3) \n\t"
- ::"S"(pixels), "D"(block),
- "r" ((x86_reg)line_size), "r"((x86_reg)3L*line_size)
- :"memory");
- block += 4*line_size;
- pixels += 4*line_size;
- h -= 4;
- } while(h > 0);
- }
- //FIXME the following could be optimized too ...
- static void DEF(put_no_rnd_pixels16_x2)(uint8_t *block, const uint8_t *pixels, int line_size, int h){
- DEF(put_no_rnd_pixels8_x2)(block , pixels , line_size, h);
- DEF(put_no_rnd_pixels8_x2)(block+8, pixels+8, line_size, h);
- }
- static void DEF(put_pixels16_y2)(uint8_t *block, const uint8_t *pixels, int line_size, int h){
- DEF(put_pixels8_y2)(block , pixels , line_size, h);
- DEF(put_pixels8_y2)(block+8, pixels+8, line_size, h);
- }
- static void DEF(put_no_rnd_pixels16_y2)(uint8_t *block, const uint8_t *pixels, int line_size, int h){
- DEF(put_no_rnd_pixels8_y2)(block , pixels , line_size, h);
- DEF(put_no_rnd_pixels8_y2)(block+8, pixels+8, line_size, h);
- }
- static void DEF(avg_pixels16)(uint8_t *block, const uint8_t *pixels, int line_size, int h){
- DEF(avg_pixels8)(block , pixels , line_size, h);
- DEF(avg_pixels8)(block+8, pixels+8, line_size, h);
- }
- static void DEF(avg_pixels16_x2)(uint8_t *block, const uint8_t *pixels, int line_size, int h){
- DEF(avg_pixels8_x2)(block , pixels , line_size, h);
- DEF(avg_pixels8_x2)(block+8, pixels+8, line_size, h);
- }
- static void DEF(avg_pixels16_y2)(uint8_t *block, const uint8_t *pixels, int line_size, int h){
- DEF(avg_pixels8_y2)(block , pixels , line_size, h);
- DEF(avg_pixels8_y2)(block+8, pixels+8, line_size, h);
- }
- static void DEF(avg_pixels16_xy2)(uint8_t *block, const uint8_t *pixels, int line_size, int h){
- DEF(avg_pixels8_xy2)(block , pixels , line_size, h);
- DEF(avg_pixels8_xy2)(block+8, pixels+8, line_size, h);
- }
- #define QPEL_2TAP_L3(OPNAME) \
- static void DEF(OPNAME ## 2tap_qpel16_l3)(uint8_t *dst, uint8_t *src, int stride, int h, int off1, int off2){\
- __asm__ volatile(\
- "1: \n\t"\
- "movq (%1,%2), %%mm0 \n\t"\
- "movq 8(%1,%2), %%mm1 \n\t"\
- PAVGB" (%1,%3), %%mm0 \n\t"\
- PAVGB" 8(%1,%3), %%mm1 \n\t"\
- PAVGB" (%1), %%mm0 \n\t"\
- PAVGB" 8(%1), %%mm1 \n\t"\
- STORE_OP( (%1,%4),%%mm0)\
- STORE_OP(8(%1,%4),%%mm1)\
- "movq %%mm0, (%1,%4) \n\t"\
- "movq %%mm1, 8(%1,%4) \n\t"\
- "add %5, %1 \n\t"\
- "decl %0 \n\t"\
- "jnz 1b \n\t"\
- :"+g"(h), "+r"(src)\
- :"r"((x86_reg)off1), "r"((x86_reg)off2),\
- "r"((x86_reg)(dst-src)), "r"((x86_reg)stride)\
- :"memory"\
- );\
- }\
- static void DEF(OPNAME ## 2tap_qpel8_l3)(uint8_t *dst, uint8_t *src, int stride, int h, int off1, int off2){\
- __asm__ volatile(\
- "1: \n\t"\
- "movq (%1,%2), %%mm0 \n\t"\
- PAVGB" (%1,%3), %%mm0 \n\t"\
- PAVGB" (%1), %%mm0 \n\t"\
- STORE_OP((%1,%4),%%mm0)\
- "movq %%mm0, (%1,%4) \n\t"\
- "add %5, %1 \n\t"\
- "decl %0 \n\t"\
- "jnz 1b \n\t"\
- :"+g"(h), "+r"(src)\
- :"r"((x86_reg)off1), "r"((x86_reg)off2),\
- "r"((x86_reg)(dst-src)), "r"((x86_reg)stride)\
- :"memory"\
- );\
- }
- #define STORE_OP(a,b) PAVGB" "#a","#b" \n\t"
- QPEL_2TAP_L3(avg_)
- #undef STORE_OP
- #define STORE_OP(a,b)
- QPEL_2TAP_L3(put_)
- #undef STORE_OP
- #undef QPEL_2TAP_L3
|