sigrid/ffmpeg

libavcodec/alpha/Makefile+0−10

@@ -1,10 +0,0 @@
1		−OBJS-$(CONFIG_BLOCKDSP) += alpha/blockdsp_alpha.o
2		−OBJS-$(CONFIG_ME_CMP) += alpha/me_cmp_alpha.o \
3		− alpha/me_cmp_mvi_asm.o
4		−OBJS-$(CONFIG_HPELDSP) += alpha/hpeldsp_alpha.o \
5		− alpha/hpeldsp_alpha_asm.o
6		−OBJS-$(CONFIG_IDCTDSP) += alpha/idctdsp_alpha.o \
7		− alpha/idctdsp_alpha_asm.o \
8		− alpha/simple_idct_alpha.o
9		−OBJS-$(CONFIG_MPEGVIDEO) += alpha/mpegvideo_alpha.o
10		−OBJS-$(CONFIG_PIXBLOCKDSP) += alpha/pixblockdsp_alpha.o

libavcodec/alpha/asm.h+0−153

@@ -1,153 +0,0 @@
1		−/*
2		− * Alpha optimized DSP utils
3		− * Copyright (c) 2002 Falk Hueffner <falk@debian.org>
4		− *
5		− * This file is part of FFmpeg.
6		− *
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.
11		− *
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.
16		− *
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		− */
21		−
22		−#ifndef AVCODEC_ALPHA_ASM_H
23		−#define AVCODEC_ALPHA_ASM_H
24		−
25		−#include <inttypes.h>
26		−
27		−#include "libavutil/common.h"
28		−
29		−#if AV_GCC_VERSION_AT_LEAST(2,96)
30		−# define likely(x) __builtin_expect((x) != 0, 1)
31		−# define unlikely(x) __builtin_expect((x) != 0, 0)
32		−#else
33		−# define likely(x) (x)
34		−# define unlikely(x) (x)
35		−#endif
36		−
37		−#define AMASK_BWX (1 << 0)
38		−#define AMASK_FIX (1 << 1)
39		−#define AMASK_CIX (1 << 2)
40		−#define AMASK_MVI (1 << 8)
41		−
42		−static inline uint64_t BYTE_VEC(uint64_t x)
43		−{
44		− x \|= x << 8;
45		− x \|= x << 16;
46		− x \|= x << 32;
47		− return x;
48		−}
49		−static inline uint64_t WORD_VEC(uint64_t x)
50		−{
51		− x \|= x << 16;
52		− x \|= x << 32;
53		− return x;
54		−}
55		−
56		−#define sextw(x) ((int16_t) (x))
57		−
58		−#ifdef __GNUC__
59		−#define ldq(p) \
60		− (((const union { \
61		− uint64_t __l; \
62		− __typeof__((p)) __s[sizeof (uint64_t) / sizeof (p)]; \
63		− } *) (p))->__l)
64		−#define ldl(p) \
65		− (((const union { \
66		− int32_t __l; \
67		− __typeof__((p)) __s[sizeof (int32_t) / sizeof (p)]; \
68		− } *) (p))->__l)
69		−#define stq(l, p) \
70		− do { \
71		− (((union { \
72		− uint64_t __l; \
73		− __typeof__((p)) __s[sizeof (uint64_t) / sizeof (p)]; \
74		− } *) (p))->__l) = l; \
75		− } while (0)
76		−#define stl(l, p) \
77		− do { \
78		− (((union { \
79		− int32_t __l; \
80		− __typeof__((p)) __s[sizeof (int32_t) / sizeof (p)]; \
81		− } *) (p))->__l) = l; \
82		− } while (0)
83		−struct unaligned_long { uint64_t l; } __attribute__((packed));
84		−#define ldq_u(p) ((const uint64_t ) (((uint64_t) (p)) & ~7ul))
85		−#define uldq(a) (((const struct unaligned_long *) (a))->l)
86		−
87		−#if AV_GCC_VERSION_AT_LEAST(3,3)
88		−#define prefetch(p) __builtin_prefetch((p), 0, 1)
89		−#define prefetch_en(p) __builtin_prefetch((p), 0, 0)
90		−#define prefetch_m(p) __builtin_prefetch((p), 1, 1)
91		−#define prefetch_men(p) __builtin_prefetch((p), 1, 0)
92		−#define cmpbge __builtin_alpha_cmpbge
93		−/* Avoid warnings. */
94		−#define extql(a, b) __builtin_alpha_extql(a, (uint64_t) (b))
95		−#define extwl(a, b) __builtin_alpha_extwl(a, (uint64_t) (b))
96		−#define extqh(a, b) __builtin_alpha_extqh(a, (uint64_t) (b))
97		−#define zap __builtin_alpha_zap
98		−#define zapnot __builtin_alpha_zapnot
99		−#define amask __builtin_alpha_amask
100		−#define implver __builtin_alpha_implver
101		−#define rpcc __builtin_alpha_rpcc
102		−#else
103		−#define prefetch(p) __asm__ volatile("ldl $31,%0" : : "m"((const char ) (p)) : "memory")
104		−#define prefetch_en(p) __asm__ volatile("ldq $31,%0" : : "m"((const char ) (p)) : "memory")
105		−#define prefetch_m(p) __asm__ volatile("lds $f31,%0" : : "m"((const char ) (p)) : "memory")
106		−#define prefetch_men(p) __asm__ volatile("ldt $f31,%0" : : "m"((const char ) (p)) : "memory")
107		−#define cmpbge(a, b) ({ uint64_t __r; __asm__ ("cmpbge %r1,%2,%0" : "=r" (__r) : "rJ" (a), "rI" (b)); __r; })
108		−#define extql(a, b) ({ uint64_t __r; __asm__ ("extql %r1,%2,%0" : "=r" (__r) : "rJ" (a), "rI" (b)); __r; })
109		−#define extwl(a, b) ({ uint64_t __r; __asm__ ("extwl %r1,%2,%0" : "=r" (__r) : "rJ" (a), "rI" (b)); __r; })
110		−#define extqh(a, b) ({ uint64_t __r; __asm__ ("extqh %r1,%2,%0" : "=r" (__r) : "rJ" (a), "rI" (b)); __r; })
111		−#define zap(a, b) ({ uint64_t __r; __asm__ ("zap %r1,%2,%0" : "=r" (__r) : "rJ" (a), "rI" (b)); __r; })
112		−#define zapnot(a, b) ({ uint64_t __r; __asm__ ("zapnot %r1,%2,%0" : "=r" (__r) : "rJ" (a), "rI" (b)); __r; })
113		−#define amask(a) ({ uint64_t __r; __asm__ ("amask %1,%0" : "=r" (__r) : "rI" (a)); __r; })
114		−#define implver() ({ uint64_t __r; __asm__ ("implver %0" : "=r" (__r)); __r; })
115		−#define rpcc() ({ uint64_t __r; __asm__ volatile ("rpcc %0" : "=r" (__r)); __r; })
116		−#endif
117		−#define wh64(p) __asm__ volatile("wh64 (%0)" : : "r"(p) : "memory")
118		−
119		−#if AV_GCC_VERSION_AT_LEAST(3,3) && defined(__alpha_max__)
120		−#define minub8 __builtin_alpha_minub8
121		−#define minsb8 __builtin_alpha_minsb8
122		−#define minuw4 __builtin_alpha_minuw4
123		−#define minsw4 __builtin_alpha_minsw4
124		−#define maxub8 __builtin_alpha_maxub8
125		−#define maxsb8 __builtin_alpha_maxsb8
126		−#define maxuw4 __builtin_alpha_maxuw4
127		−#define maxsw4 __builtin_alpha_maxsw4
128		−#define perr __builtin_alpha_perr
129		−#define pklb __builtin_alpha_pklb
130		−#define pkwb __builtin_alpha_pkwb
131		−#define unpkbl __builtin_alpha_unpkbl
132		−#define unpkbw __builtin_alpha_unpkbw
133		−#else
134		−#define minub8(a, b) ({ uint64_t __r; __asm__ (".arch ev6; minub8 %r1,%2,%0" : "=r" (__r) : "%rJ" (a), "rI" (b)); __r; })
135		−#define minsb8(a, b) ({ uint64_t __r; __asm__ (".arch ev6; minsb8 %r1,%2,%0" : "=r" (__r) : "%rJ" (a), "rI" (b)); __r; })
136		−#define minuw4(a, b) ({ uint64_t __r; __asm__ (".arch ev6; minuw4 %r1,%2,%0" : "=r" (__r) : "%rJ" (a), "rI" (b)); __r; })
137		−#define minsw4(a, b) ({ uint64_t __r; __asm__ (".arch ev6; minsw4 %r1,%2,%0" : "=r" (__r) : "%rJ" (a), "rI" (b)); __r; })
138		−#define maxub8(a, b) ({ uint64_t __r; __asm__ (".arch ev6; maxub8 %r1,%2,%0" : "=r" (__r) : "%rJ" (a), "rI" (b)); __r; })
139		−#define maxsb8(a, b) ({ uint64_t __r; __asm__ (".arch ev6; maxsb8 %r1,%2,%0" : "=r" (__r) : "%rJ" (a), "rI" (b)); __r; })
140		−#define maxuw4(a, b) ({ uint64_t __r; __asm__ (".arch ev6; maxuw4 %r1,%2,%0" : "=r" (__r) : "%rJ" (a), "rI" (b)); __r; })
141		−#define maxsw4(a, b) ({ uint64_t __r; __asm__ (".arch ev6; maxsw4 %r1,%2,%0" : "=r" (__r) : "%rJ" (a), "rI" (b)); __r; })
142		−#define perr(a, b) ({ uint64_t __r; __asm__ (".arch ev6; perr %r1,%r2,%0" : "=r" (__r) : "%rJ" (a), "rJ" (b)); __r; })
143		−#define pklb(a) ({ uint64_t __r; __asm__ (".arch ev6; pklb %r1,%0" : "=r" (__r) : "rJ" (a)); __r; })
144		−#define pkwb(a) ({ uint64_t __r; __asm__ (".arch ev6; pkwb %r1,%0" : "=r" (__r) : "rJ" (a)); __r; })
145		−#define unpkbl(a) ({ uint64_t __r; __asm__ (".arch ev6; unpkbl %r1,%0" : "=r" (__r) : "rJ" (a)); __r; })
146		−#define unpkbw(a) ({ uint64_t __r; __asm__ (".arch ev6; unpkbw %r1,%0" : "=r" (__r) : "rJ" (a)); __r; })
147		−#endif
148		−
149		−#else
150		−#error "Unknown compiler!"
151		−#endif
152		−
153		−#endif /* AVCODEC_ALPHA_ASM_H */

libavcodec/alpha/blockdsp_alpha.c+0−49

@@ -1,49 +0,0 @@
1		−/*
2		− * Alpha optimised block operations
3		− * Copyright (c) 2002 Falk Hueffner <falk@debian.org>
4		− *
5		− * This file is part of FFmpeg.
6		− *
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.
11		− *
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.
16		− *
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		− */
21		−
22		−#include <stdint.h>
23		−
24		−#include "libavutil/attributes.h"
25		−#include "libavcodec/blockdsp.h"
26		−#include "asm.h"
27		−
28		−static void clear_blocks_axp(int16_t *blocks) {
29		− uint64_t p = (uint64_t ) blocks;
30		− int n = sizeof(int16_t) * 6 * 64;
31		−
32		− do {
33		− p[0] = 0;
34		− p[1] = 0;
35		− p[2] = 0;
36		− p[3] = 0;
37		− p[4] = 0;
38		− p[5] = 0;
39		− p[6] = 0;
40		− p[7] = 0;
41		− p += 8;
42		− n -= 8 * 8;
43		− } while (n);
44		−}
45		−
46		−av_cold void ff_blockdsp_init_alpha(BlockDSPContext *c)
47		−{
48		− c->clear_blocks = clear_blocks_axp;
49		−}

libavcodec/alpha/hpeldsp_alpha.c+0−213

@@ -1,213 +0,0 @@
1		−/*
2		− * Alpha optimized DSP utils
3		− * Copyright (c) 2002 Falk Hueffner <falk@debian.org>
4		− *
5		− * This file is part of FFmpeg.
6		− *
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.
11		− *
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.
16		− *
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		− */
21		−
22		−#include "libavutil/attributes.h"
23		−#include "libavcodec/hpeldsp.h"
24		−#include "hpeldsp_alpha.h"
25		−#include "asm.h"
26		−
27		−static inline uint64_t avg2_no_rnd(uint64_t a, uint64_t b)
28		−{
29		− return (a & b) + (((a ^ b) & BYTE_VEC(0xfe)) >> 1);
30		−}
31		−
32		−static inline uint64_t avg2(uint64_t a, uint64_t b)
33		−{
34		− return (a \| b) - (((a ^ b) & BYTE_VEC(0xfe)) >> 1);
35		−}
36		−
37		−#if 0
38		−/* The XY2 routines basically utilize this scheme, but reuse parts in
39		− each iteration. */
40		−static inline uint64_t avg4(uint64_t l1, uint64_t l2, uint64_t l3, uint64_t l4)
41		−{
42		− uint64_t r1 = ((l1 & ~BYTE_VEC(0x03)) >> 2)
43		− + ((l2 & ~BYTE_VEC(0x03)) >> 2)
44		− + ((l3 & ~BYTE_VEC(0x03)) >> 2)
45		− + ((l4 & ~BYTE_VEC(0x03)) >> 2);
46		− uint64_t r2 = (( (l1 & BYTE_VEC(0x03))
47		− + (l2 & BYTE_VEC(0x03))
48		− + (l3 & BYTE_VEC(0x03))
49		− + (l4 & BYTE_VEC(0x03))
50		− + BYTE_VEC(0x02)) >> 2) & BYTE_VEC(0x03);
51		− return r1 + r2;
52		−}
53		−#endif
54		−
55		−#define OP(LOAD, STORE) \
56		− do { \
57		− STORE(LOAD(pixels), block); \
58		− pixels += line_size; \
59		− block += line_size; \
60		− } while (--h)
61		−
62		−#define OP_X2(LOAD, STORE) \
63		− do { \
64		− uint64_t pix1, pix2; \
65		− \
66		− pix1 = LOAD(pixels); \
67		− pix2 = pix1 >> 8 \| ((uint64_t) pixels[8] << 56); \
68		− STORE(AVG2(pix1, pix2), block); \
69		− pixels += line_size; \
70		− block += line_size; \
71		− } while (--h)
72		−
73		−#define OP_Y2(LOAD, STORE) \
74		− do { \
75		− uint64_t pix = LOAD(pixels); \
76		− do { \
77		− uint64_t next_pix; \
78		− \
79		− pixels += line_size; \
80		− next_pix = LOAD(pixels); \
81		− STORE(AVG2(pix, next_pix), block); \
82		− block += line_size; \
83		− pix = next_pix; \
84		− } while (--h); \
85		− } while (0)
86		−
87		−#define OP_XY2(LOAD, STORE) \
88		− do { \
89		− uint64_t pix1 = LOAD(pixels); \
90		− uint64_t pix2 = pix1 >> 8 \| ((uint64_t) pixels[8] << 56); \
91		− uint64_t pix_l = (pix1 & BYTE_VEC(0x03)) \
92		− + (pix2 & BYTE_VEC(0x03)); \
93		− uint64_t pix_h = ((pix1 & ~BYTE_VEC(0x03)) >> 2) \
94		− + ((pix2 & ~BYTE_VEC(0x03)) >> 2); \
95		− \
96		− do { \
97		− uint64_t npix1, npix2; \
98		− uint64_t npix_l, npix_h; \
99		− uint64_t avg; \
100		− \
101		− pixels += line_size; \
102		− npix1 = LOAD(pixels); \
103		− npix2 = npix1 >> 8 \| ((uint64_t) pixels[8] << 56); \
104		− npix_l = (npix1 & BYTE_VEC(0x03)) \
105		− + (npix2 & BYTE_VEC(0x03)); \
106		− npix_h = ((npix1 & ~BYTE_VEC(0x03)) >> 2) \
107		− + ((npix2 & ~BYTE_VEC(0x03)) >> 2); \
108		− avg = (((pix_l + npix_l + AVG4_ROUNDER) >> 2) & BYTE_VEC(0x03)) \
109		− + pix_h + npix_h; \
110		− STORE(avg, block); \
111		− \
112		− block += line_size; \
113		− pix_l = npix_l; \
114		− pix_h = npix_h; \
115		− } while (--h); \
116		− } while (0)
117		−
118		−#define MAKE_OP(OPNAME, SUFF, OPKIND, STORE) \
119		−static void OPNAME ## _pixels ## SUFF ## _axp \
120		− (uint8_t restrict block, const uint8_t restrict pixels, \
121		− ptrdiff_t line_size, int h) \
122		−{ \
123		− if ((size_t) pixels & 0x7) { \
124		− OPKIND(uldq, STORE); \
125		− } else { \
126		− OPKIND(ldq, STORE); \
127		− } \
128		−} \
129		− \
130		−static void OPNAME ## _pixels16 ## SUFF ## _axp \
131		− (uint8_t restrict block, const uint8_t restrict pixels, \
132		− ptrdiff_t line_size, int h) \
133		−{ \
134		− OPNAME ## _pixels ## SUFF ## _axp(block, pixels, line_size, h); \
135		− OPNAME ## _pixels ## SUFF ## _axp(block + 8, pixels + 8, line_size, h); \
136		−}
137		−
138		−#define PIXOP(OPNAME, STORE) \
139		− MAKE_OP(OPNAME, , OP, STORE) \
140		− MAKE_OP(OPNAME, _x2, OP_X2, STORE) \
141		− MAKE_OP(OPNAME, _y2, OP_Y2, STORE) \
142		− MAKE_OP(OPNAME, _xy2, OP_XY2, STORE)
143		−
144		−/* Rounding primitives. */
145		−#define AVG2 avg2
146		−#define AVG4 avg4
147		−#define AVG4_ROUNDER BYTE_VEC(0x02)
148		−#define STORE(l, b) stq(l, b)
149		−PIXOP(put, STORE);
150		−
151		−#undef STORE
152		−#define STORE(l, b) stq(AVG2(l, ldq(b)), b);
153		−PIXOP(avg, STORE);
154		−
155		−/* Not rounding primitives. */
156		−#undef AVG2
157		−#undef AVG4
158		−#undef AVG4_ROUNDER
159		−#undef STORE
160		−#define AVG2 avg2_no_rnd
161		−#define AVG4 avg4_no_rnd
162		−#define AVG4_ROUNDER BYTE_VEC(0x01)
163		−#define STORE(l, b) stq(l, b)
164		−PIXOP(put_no_rnd, STORE);
165		−
166		−#undef STORE
167		−#define STORE(l, b) stq(AVG2(l, ldq(b)), b);
168		−PIXOP(avg_no_rnd, STORE);
169		−
170		−static void put_pixels16_axp_asm(uint8_t block, const uint8_t pixels,
171		− ptrdiff_t line_size, int h)
172		−{
173		− put_pixels_axp_asm(block, pixels, line_size, h);
174		− put_pixels_axp_asm(block + 8, pixels + 8, line_size, h);
175		−}
176		−
177		−av_cold void ff_hpeldsp_init_alpha(HpelDSPContext *c, int flags)
178		−{
179		− c->put_pixels_tab[0][0] = put_pixels16_axp_asm;
180		− c->put_pixels_tab[0][1] = put_pixels16_x2_axp;
181		− c->put_pixels_tab[0][2] = put_pixels16_y2_axp;
182		− c->put_pixels_tab[0][3] = put_pixels16_xy2_axp;
183		−
184		− c->put_no_rnd_pixels_tab[0][0] = put_pixels16_axp_asm;
185		− c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_axp;
186		− c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_axp;
187		− c->put_no_rnd_pixels_tab[0][3] = put_no_rnd_pixels16_xy2_axp;
188		−
189		− c->avg_pixels_tab[0][0] = avg_pixels16_axp;
190		− c->avg_pixels_tab[0][1] = avg_pixels16_x2_axp;
191		− c->avg_pixels_tab[0][2] = avg_pixels16_y2_axp;
192		− c->avg_pixels_tab[0][3] = avg_pixels16_xy2_axp;
193		−
194		− c->avg_no_rnd_pixels_tab[0] = avg_no_rnd_pixels16_axp;
195		− c->avg_no_rnd_pixels_tab[1] = avg_no_rnd_pixels16_x2_axp;
196		− c->avg_no_rnd_pixels_tab[2] = avg_no_rnd_pixels16_y2_axp;
197		− c->avg_no_rnd_pixels_tab[3] = avg_no_rnd_pixels16_xy2_axp;
198		−
199		− c->put_pixels_tab[1][0] = put_pixels_axp_asm;
200		− c->put_pixels_tab[1][1] = put_pixels_x2_axp;
201		− c->put_pixels_tab[1][2] = put_pixels_y2_axp;
202		− c->put_pixels_tab[1][3] = put_pixels_xy2_axp;
203		−
204		− c->put_no_rnd_pixels_tab[1][0] = put_pixels_axp_asm;
205		− c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels_x2_axp;
206		− c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels_y2_axp;
207		− c->put_no_rnd_pixels_tab[1][3] = put_no_rnd_pixels_xy2_axp;
208		−
209		− c->avg_pixels_tab[1][0] = avg_pixels_axp;
210		− c->avg_pixels_tab[1][1] = avg_pixels_x2_axp;
211		− c->avg_pixels_tab[1][2] = avg_pixels_y2_axp;
212		− c->avg_pixels_tab[1][3] = avg_pixels_xy2_axp;
213		−}

libavcodec/alpha/hpeldsp_alpha.h+0−28

@@ -1,28 +0,0 @@
1		−/*
2		− * This file is part of FFmpeg.
3		− *
4		− * FFmpeg is free software; you can redistribute it and/or
5		− * modify it under the terms of the GNU Lesser General Public
6		− * License as published by the Free Software Foundation; either
7		− * version 2.1 of the License, or (at your option) any later version.
8		− *
9		− * FFmpeg is distributed in the hope that it will be useful,
10		− * but WITHOUT ANY WARRANTY; without even the implied warranty of
11		− * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12		− * Lesser General Public License for more details.
13		− *
14		− * You should have received a copy of the GNU Lesser General Public
15		− * License along with FFmpeg; if not, write to the Free Software
16		− * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17		− */
18		−
19		−#ifndef AVCODEC_ALPHA_HPELDSP_ALPHA_H
20		−#define AVCODEC_ALPHA_HPELDSP_ALPHA_H
21		−
22		−#include <stdint.h>
23		−#include <stddef.h>
24		−
25		−void put_pixels_axp_asm(uint8_t block, const uint8_t pixels,
26		− ptrdiff_t line_size, int h);
27		−
28		−#endif /* AVCODEC_ALPHA_HPELDSP_ALPHA_H */

libavcodec/alpha/hpeldsp_alpha_asm.S+0−125

@@ -1,125 +0,0 @@
1		−/*
2		− * Alpha optimized DSP utils
3		− * Copyright (c) 2002 Falk Hueffner <falk@debian.org>
4		− *
5		− * This file is part of FFmpeg.
6		− *
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.
11		− *
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.
16		− *
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		− */
21		−
22		−/*
23		− * These functions are scheduled for pca56. They should work
24		− * reasonably on ev6, though.
25		− */
26		−
27		−#include "regdef.h"
28		−
29		−
30		− .set noat
31		− .set noreorder
32		− .arch pca56
33		− .text
34		−
35		−/************************************************************************
36		− * void put_pixels_axp_asm(uint8_t block, const uint8_t pixels,
37		− * int line_size, int h)
38		− */
39		− .align 6
40		− .globl put_pixels_axp_asm
41		− .ent put_pixels_axp_asm
42		−put_pixels_axp_asm:
43		− .frame sp, 0, ra
44		− .prologue 0
45		−
46		− and a1, 7, t0
47		− beq t0, $aligned
48		−
49		− .align 4
50		−$unaligned:
51		− ldq_u t0, 0(a1)
52		− ldq_u t1, 8(a1)
53		− addq a1, a2, a1
54		− nop
55		−
56		− ldq_u t2, 0(a1)
57		− ldq_u t3, 8(a1)
58		− addq a1, a2, a1
59		− nop
60		−
61		− ldq_u t4, 0(a1)
62		− ldq_u t5, 8(a1)
63		− addq a1, a2, a1
64		− nop
65		−
66		− ldq_u t6, 0(a1)
67		− ldq_u t7, 8(a1)
68		− extql t0, a1, t0
69		− addq a1, a2, a1
70		−
71		− extqh t1, a1, t1
72		− addq a0, a2, t8
73		− extql t2, a1, t2
74		− addq t8, a2, t9
75		−
76		− extqh t3, a1, t3
77		− addq t9, a2, ta
78		− extql t4, a1, t4
79		− or t0, t1, t0
80		−
81		− extqh t5, a1, t5
82		− or t2, t3, t2
83		− extql t6, a1, t6
84		− or t4, t5, t4
85		−
86		− extqh t7, a1, t7
87		− or t6, t7, t6
88		− stq t0, 0(a0)
89		− stq t2, 0(t8)
90		−
91		− stq t4, 0(t9)
92		− subq a3, 4, a3
93		− stq t6, 0(ta)
94		− addq ta, a2, a0
95		−
96		− bne a3, $unaligned
97		− ret
98		−
99		− .align 4
100		−$aligned:
101		− ldq t0, 0(a1)
102		− addq a1, a2, a1
103		− ldq t1, 0(a1)
104		− addq a1, a2, a1
105		−
106		− ldq t2, 0(a1)
107		− addq a1, a2, a1
108		− ldq t3, 0(a1)
109		−
110		− addq a0, a2, t4
111		− addq a1, a2, a1
112		− addq t4, a2, t5
113		− subq a3, 4, a3
114		−
115		− stq t0, 0(a0)
116		− addq t5, a2, t6
117		− stq t1, 0(t4)
118		− addq t6, a2, a0
119		−
120		− stq t2, 0(t5)
121		− stq t3, 0(t6)
122		−
123		− bne a3, $aligned
124		− ret
125		− .end put_pixels_axp_asm

libavcodec/alpha/idctdsp_alpha.c+0−127

@@ -1,127 +0,0 @@
1		−/*
2		− * Copyright (c) 2002 Falk Hueffner <falk@debian.org>
3		− *
4		− * This file is part of FFmpeg.
5		− *
6		− * FFmpeg is free software; you can redistribute it and/or
7		− * modify it under the terms of the GNU Lesser General Public
8		− * License as published by the Free Software Foundation; either
9		− * version 2.1 of the License, or (at your option) any later version.
10		− *
11		− * FFmpeg is distributed in the hope that it will be useful,
12		− * but WITHOUT ANY WARRANTY; without even the implied warranty of
13		− * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14		− * Lesser General Public License for more details.
15		− *
16		− * You should have received a copy of the GNU Lesser General Public
17		− * License along with FFmpeg; if not, write to the Free Software
18		− * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19		− */
20		−
21		−#include "libavutil/attributes.h"
22		−#include "libavcodec/avcodec.h"
23		−#include "libavcodec/idctdsp.h"
24		−#include "idctdsp_alpha.h"
25		−#include "asm.h"
26		−
27		−void put_pixels_clamped_mvi_asm(const int16_t block, uint8_t pixels,
28		− ptrdiff_t line_size);
29		−void add_pixels_clamped_mvi_asm(const int16_t block, uint8_t pixels,
30		− ptrdiff_t line_size);
31		−
32		−void (put_pixels_clamped_axp_p)(const int16_t block, uint8_t *pixels,
33		− ptrdiff_t line_size);
34		−void (add_pixels_clamped_axp_p)(const int16_t block, uint8_t *pixels,
35		− ptrdiff_t line_size);
36		−
37		−#if 0
38		−/* These functions were the base for the optimized assembler routines,
39		− and remain here for documentation purposes. */
40		−static void put_pixels_clamped_mvi(const int16_t block, uint8_t pixels,
41		− ptrdiff_t line_size)
42		−{
43		− int i = 8;
44		− uint64_t clampmask = zap(-1, 0xaa); /* 0x00ff00ff00ff00ff */
45		−
46		− do {
47		− uint64_t shorts0, shorts1;
48		−
49		− shorts0 = ldq(block);
50		− shorts0 = maxsw4(shorts0, 0);
51		− shorts0 = minsw4(shorts0, clampmask);
52		− stl(pkwb(shorts0), pixels);
53		−
54		− shorts1 = ldq(block + 4);
55		− shorts1 = maxsw4(shorts1, 0);
56		− shorts1 = minsw4(shorts1, clampmask);
57		− stl(pkwb(shorts1), pixels + 4);
58		−
59		− pixels += line_size;
60		− block += 8;
61		− } while (--i);
62		−}
63		−
64		−void add_pixels_clamped_mvi(const int16_t block, uint8_t pixels,
65		− ptrdiff_t line_size)
66		−{
67		− int h = 8;
68		− /* Keep this function a leaf function by generating the constants
69		− manually (mainly for the hack value ;-). */
70		− uint64_t clampmask = zap(-1, 0xaa); /* 0x00ff00ff00ff00ff */
71		− uint64_t signmask = zap(-1, 0x33);
72		− signmask ^= signmask >> 1; /* 0x8000800080008000 */
73		−
74		− do {
75		− uint64_t shorts0, pix0, signs0;
76		− uint64_t shorts1, pix1, signs1;
77		−
78		− shorts0 = ldq(block);
79		− shorts1 = ldq(block + 4);
80		−
81		− pix0 = unpkbw(ldl(pixels));
82		− /* Signed subword add (MMX paddw). */
83		− signs0 = shorts0 & signmask;
84		− shorts0 &= ~signmask;
85		− shorts0 += pix0;
86		− shorts0 ^= signs0;
87		− /* Clamp. */
88		− shorts0 = maxsw4(shorts0, 0);
89		− shorts0 = minsw4(shorts0, clampmask);
90		−
91		− /* Next 4. */
92		− pix1 = unpkbw(ldl(pixels + 4));
93		− signs1 = shorts1 & signmask;
94		− shorts1 &= ~signmask;
95		− shorts1 += pix1;
96		− shorts1 ^= signs1;
97		− shorts1 = maxsw4(shorts1, 0);
98		− shorts1 = minsw4(shorts1, clampmask);
99		−
100		− stl(pkwb(shorts0), pixels);
101		− stl(pkwb(shorts1), pixels + 4);
102		−
103		− pixels += line_size;
104		− block += 8;
105		− } while (--h);
106		−}
107		−#endif
108		−
109		−av_cold void ff_idctdsp_init_alpha(IDCTDSPContext c, AVCodecContext avctx,
110		− unsigned high_bit_depth)
111		−{
112		− /* amask clears all bits that correspond to present features. */
113		− if (amask(AMASK_MVI) == 0) {
114		− c->put_pixels_clamped = put_pixels_clamped_mvi_asm;
115		− c->add_pixels_clamped = add_pixels_clamped_mvi_asm;
116		− }
117		−
118		− put_pixels_clamped_axp_p = c->put_pixels_clamped;
119		− add_pixels_clamped_axp_p = c->add_pixels_clamped;
120		−
121		− if (!high_bit_depth && !avctx->lowres &&
122		− (avctx->idct_algo == FF_IDCT_AUTO)) {
123		− c->idct_put = ff_simple_idct_put_axp;
124		− c->idct_add = ff_simple_idct_add_axp;
125		− c->idct = ff_simple_idct_axp;
126		− }
127		−}

libavcodec/alpha/idctdsp_alpha.h+0−34

@@ -1,34 +0,0 @@
1		−/*
2		− * This file is part of FFmpeg.
3		− *
4		− * FFmpeg is free software; you can redistribute it and/or
5		− * modify it under the terms of the GNU Lesser General Public
6		− * License as published by the Free Software Foundation; either
7		− * version 2.1 of the License, or (at your option) any later version.
8		− *
9		− * FFmpeg is distributed in the hope that it will be useful,
10		− * but WITHOUT ANY WARRANTY; without even the implied warranty of
11		− * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12		− * Lesser General Public License for more details.
13		− *
14		− * You should have received a copy of the GNU Lesser General Public
15		− * License along with FFmpeg; if not, write to the Free Software
16		− * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17		− */
18		−
19		−#ifndef AVCODEC_ALPHA_IDCTDSP_ALPHA_H
20		−#define AVCODEC_ALPHA_IDCTDSP_ALPHA_H
21		−
22		−#include <stddef.h>
23		−#include <stdint.h>
24		−
25		−extern void (put_pixels_clamped_axp_p)(const int16_t block, uint8_t *pixels,
26		− ptrdiff_t line_size);
27		−extern void (add_pixels_clamped_axp_p)(const int16_t block, uint8_t *pixels,
28		− ptrdiff_t line_size);
29		−
30		−void ff_simple_idct_axp(int16_t *block);
31		−void ff_simple_idct_put_axp(uint8_t dest, ptrdiff_t line_size, int16_t block);
32		−void ff_simple_idct_add_axp(uint8_t dest, ptrdiff_t line_size, int16_t block);
33		−
34		−#endif /* AVCODEC_ALPHA_IDCTDSP_ALPHA_H */

libavcodec/alpha/idctdsp_alpha_asm.S+0−167

@@ -1,167 +0,0 @@
1		−/*
2		− * Alpha optimized IDCT-related routines
3		− * Copyright (c) 2002 Falk Hueffner <falk@debian.org>
4		− *
5		− * This file is part of FFmpeg.
6		− *
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.
11		− *
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.
16		− *
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		− */
21		−
22		−/*
23		− * These functions are scheduled for pca56. They should work
24		− * reasonably on ev6, though.
25		− */
26		−
27		−#include "regdef.h"
28		−
29		− .set noat
30		− .set noreorder
31		− .arch pca56
32		− .text
33		−
34		−/************************************************************************
35		− * void put_pixels_clamped_mvi_asm(const int16_t block, uint8_t pixels,
36		− * ptrdiff_t line_size)
37		− */
38		− .align 6
39		− .globl put_pixels_clamped_mvi_asm
40		− .ent put_pixels_clamped_mvi_asm
41		−put_pixels_clamped_mvi_asm:
42		− .frame sp, 0, ra
43		− .prologue 0
44		−
45		− lda t8, -1
46		− lda t9, 8 # loop counter
47		− zap t8, 0xaa, t8 # 00ff00ff00ff00ff
48		−
49		− .align 4
50		−1: ldq t0, 0(a0)
51		− ldq t1, 8(a0)
52		− ldq t2, 16(a0)
53		− ldq t3, 24(a0)
54		−
55		− maxsw4 t0, zero, t0
56		− subq t9, 2, t9
57		− maxsw4 t1, zero, t1
58		− lda a0, 32(a0)
59		−
60		− maxsw4 t2, zero, t2
61		− addq a1, a2, ta
62		− maxsw4 t3, zero, t3
63		− minsw4 t0, t8, t0
64		−
65		− minsw4 t1, t8, t1
66		− minsw4 t2, t8, t2
67		− minsw4 t3, t8, t3
68		− pkwb t0, t0
69		−
70		− pkwb t1, t1
71		− pkwb t2, t2
72		− pkwb t3, t3
73		− stl t0, 0(a1)
74		−
75		− stl t1, 4(a1)
76		− addq ta, a2, a1
77		− stl t2, 0(ta)
78		− stl t3, 4(ta)
79		−
80		− bne t9, 1b
81		− ret
82		− .end put_pixels_clamped_mvi_asm
83		−
84		−/************************************************************************
85		− * void add_pixels_clamped_mvi_asm(const int16_t block, uint8_t pixels,
86		− * ptrdiff_t line_size)
87		− */
88		− .align 6
89		− .globl add_pixels_clamped_mvi_asm
90		− .ent add_pixels_clamped_mvi_asm
91		−add_pixels_clamped_mvi_asm:
92		− .frame sp, 0, ra
93		− .prologue 0
94		−
95		− lda t1, -1
96		− lda th, 8
97		− zap t1, 0x33, tg
98		− nop
99		−
100		− srl tg, 1, t0
101		− xor tg, t0, tg # 0x8000800080008000
102		− zap t1, 0xaa, tf # 0x00ff00ff00ff00ff
103		−
104		− .align 4
105		−1: ldl t1, 0(a1) # pix0 (try to hit cache line soon)
106		− ldl t4, 4(a1) # pix1
107		− addq a1, a2, te # pixels += line_size
108		− ldq t0, 0(a0) # shorts0
109		−
110		− ldl t7, 0(te) # pix2 (try to hit cache line soon)
111		− ldl ta, 4(te) # pix3
112		− ldq t3, 8(a0) # shorts1
113		− ldq t6, 16(a0) # shorts2
114		−
115		− ldq t9, 24(a0) # shorts3
116		− unpkbw t1, t1 # 0 0 (quarter/op no.)
117		− and t0, tg, t2 # 0 1
118		− unpkbw t4, t4 # 1 0
119		−
120		− bic t0, tg, t0 # 0 2
121		− unpkbw t7, t7 # 2 0
122		− and t3, tg, t5 # 1 1
123		− addq t0, t1, t0 # 0 3
124		−
125		− xor t0, t2, t0 # 0 4
126		− unpkbw ta, ta # 3 0
127		− and t6, tg, t8 # 2 1
128		− maxsw4 t0, zero, t0 # 0 5
129		−
130		− bic t3, tg, t3 # 1 2
131		− bic t6, tg, t6 # 2 2
132		− minsw4 t0, tf, t0 # 0 6
133		− addq t3, t4, t3 # 1 3
134		−
135		− pkwb t0, t0 # 0 7
136		− xor t3, t5, t3 # 1 4
137		− maxsw4 t3, zero, t3 # 1 5
138		− addq t6, t7, t6 # 2 3
139		−
140		− xor t6, t8, t6 # 2 4
141		− and t9, tg, tb # 3 1
142		− minsw4 t3, tf, t3 # 1 6
143		− bic t9, tg, t9 # 3 2
144		−
145		− maxsw4 t6, zero, t6 # 2 5
146		− addq t9, ta, t9 # 3 3
147		− stl t0, 0(a1) # 0 8
148		− minsw4 t6, tf, t6 # 2 6
149		−
150		− xor t9, tb, t9 # 3 4
151		− maxsw4 t9, zero, t9 # 3 5
152		− lda a0, 32(a0) # block += 16;
153		− pkwb t3, t3 # 1 7
154		−
155		− minsw4 t9, tf, t9 # 3 6
156		− subq th, 2, th
157		− pkwb t6, t6 # 2 7
158		− pkwb t9, t9 # 3 7
159		−
160		− stl t3, 4(a1) # 1 8
161		− addq te, a2, a1 # pixels += line_size
162		− stl t6, 0(te) # 2 8
163		− stl t9, 4(te) # 3 8
164		−
165		− bne th, 1b
166		− ret
167		− .end add_pixels_clamped_mvi_asm

libavcodec/alpha/me_cmp_alpha.c+0−279

@@ -1,279 +0,0 @@
1		−/*
2		− * Alpha optimized DSP utils
3		− * Copyright (c) 2002 Falk Hueffner <falk@debian.org>
4		− *
5		− * This file is part of FFmpeg.
6		− *
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.
11		− *
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.
16		− *
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		− */
21		−
22		−#include "libavutil/attributes.h"
23		−#include "libavcodec/me_cmp.h"
24		−#include "asm.h"
25		−
26		−int pix_abs16x16_mvi_asm(struct MpegEncContext v, const uint8_t pix1, const uint8_t *pix2,
27		− ptrdiff_t line_size, int h);
28		−
29		−static inline uint64_t avg2(uint64_t a, uint64_t b)
30		−{
31		− return (a \| b) - (((a ^ b) & BYTE_VEC(0xfe)) >> 1);
32		−}
33		−
34		−static inline uint64_t avg4(uint64_t l1, uint64_t l2, uint64_t l3, uint64_t l4)
35		−{
36		− uint64_t r1 = ((l1 & ~BYTE_VEC(0x03)) >> 2)
37		− + ((l2 & ~BYTE_VEC(0x03)) >> 2)
38		− + ((l3 & ~BYTE_VEC(0x03)) >> 2)
39		− + ((l4 & ~BYTE_VEC(0x03)) >> 2);
40		− uint64_t r2 = (( (l1 & BYTE_VEC(0x03))
41		− + (l2 & BYTE_VEC(0x03))
42		− + (l3 & BYTE_VEC(0x03))
43		− + (l4 & BYTE_VEC(0x03))
44		− + BYTE_VEC(0x02)) >> 2) & BYTE_VEC(0x03);
45		− return r1 + r2;
46		−}
47		−
48		−static int pix_abs8x8_mvi(struct MpegEncContext v, const uint8_t pix1, const uint8_t *pix2,
49		− ptrdiff_t line_size, int h)
50		−{
51		− int result = 0;
52		−
53		− if ((size_t) pix2 & 0x7) {
54		− /* works only when pix2 is actually unaligned */
55		− do { /* do 8 pixel a time */
56		− uint64_t p1, p2;
57		−
58		− p1 = ldq(pix1);
59		− p2 = uldq(pix2);
60		− result += perr(p1, p2);
61		−
62		− pix1 += line_size;
63		− pix2 += line_size;
64		− } while (--h);
65		− } else {
66		− do {
67		− uint64_t p1, p2;
68		−
69		− p1 = ldq(pix1);
70		− p2 = ldq(pix2);
71		− result += perr(p1, p2);
72		−
73		− pix1 += line_size;
74		− pix2 += line_size;
75		− } while (--h);
76		− }
77		−
78		− return result;
79		−}
80		−
81		−static int pix_abs16x16_x2_mvi(struct MpegEncContext v, const uint8_t pix1, const uint8_t *pix2,
82		− ptrdiff_t line_size, int h)
83		−{
84		− int result = 0;
85		− uint64_t disalign = (size_t) pix2 & 0x7;
86		−
87		− switch (disalign) {
88		− case 0:
89		− do {
90		− uint64_t p1_l, p1_r, p2_l, p2_r;
91		− uint64_t l, r;
92		−
93		− p1_l = ldq(pix1);
94		− p1_r = ldq(pix1 + 8);
95		− l = ldq(pix2);
96		− r = ldq(pix2 + 8);
97		− p2_l = avg2(l, (l >> 8) \| ((uint64_t) r << 56));
98		− p2_r = avg2(r, (r >> 8) \| ((uint64_t) pix2[16] << 56));
99		− pix1 += line_size;
100		− pix2 += line_size;
101		−
102		− result += perr(p1_l, p2_l)
103		− + perr(p1_r, p2_r);
104		− } while (--h);
105		− break;
106		− case 7:
107		− /* \|.......l\|lllllllr\|rrrrrrr*\|
108		− This case is special because disalign1 would be 8, which
109		− gets treated as 0 by extqh. At least it is a bit faster
110		− that way :) */
111		− do {
112		− uint64_t p1_l, p1_r, p2_l, p2_r;
113		− uint64_t l, m, r;
114		−
115		− p1_l = ldq(pix1);
116		− p1_r = ldq(pix1 + 8);
117		− l = ldq_u(pix2);
118		− m = ldq_u(pix2 + 8);
119		− r = ldq_u(pix2 + 16);
120		− p2_l = avg2(extql(l, disalign) \| extqh(m, disalign), m);
121		− p2_r = avg2(extql(m, disalign) \| extqh(r, disalign), r);
122		− pix1 += line_size;
123		− pix2 += line_size;
124		−
125		− result += perr(p1_l, p2_l)
126		− + perr(p1_r, p2_r);
127		− } while (--h);
128		− break;
129		− default:
130		− do {
131		− uint64_t disalign1 = disalign + 1;
132		− uint64_t p1_l, p1_r, p2_l, p2_r;
133		− uint64_t l, m, r;
134		−
135		− p1_l = ldq(pix1);
136		− p1_r = ldq(pix1 + 8);
137		− l = ldq_u(pix2);
138		− m = ldq_u(pix2 + 8);
139		− r = ldq_u(pix2 + 16);
140		− p2_l = avg2(extql(l, disalign) \| extqh(m, disalign),
141		− extql(l, disalign1) \| extqh(m, disalign1));
142		− p2_r = avg2(extql(m, disalign) \| extqh(r, disalign),
143		− extql(m, disalign1) \| extqh(r, disalign1));
144		− pix1 += line_size;
145		− pix2 += line_size;
146		−
147		− result += perr(p1_l, p2_l)
148		− + perr(p1_r, p2_r);
149		− } while (--h);
150		− break;
151		− }
152		− return result;
153		−}
154		−
155		−static int pix_abs16x16_y2_mvi(struct MpegEncContext v, const uint8_t pix1, const uint8_t *pix2,
156		− ptrdiff_t line_size, int h)
157		−{
158		− int result = 0;
159		−
160		− if ((size_t) pix2 & 0x7) {
161		− uint64_t t, p2_l, p2_r;
162		− t = ldq_u(pix2 + 8);
163		− p2_l = extql(ldq_u(pix2), pix2) \| extqh(t, pix2);
164		− p2_r = extql(t, pix2) \| extqh(ldq_u(pix2 + 16), pix2);
165		−
166		− do {
167		− uint64_t p1_l, p1_r, np2_l, np2_r;
168		− uint64_t t;
169		−
170		− p1_l = ldq(pix1);
171		− p1_r = ldq(pix1 + 8);
172		− pix2 += line_size;
173		− t = ldq_u(pix2 + 8);
174		− np2_l = extql(ldq_u(pix2), pix2) \| extqh(t, pix2);
175		− np2_r = extql(t, pix2) \| extqh(ldq_u(pix2 + 16), pix2);
176		−
177		− result += perr(p1_l, avg2(p2_l, np2_l))
178		− + perr(p1_r, avg2(p2_r, np2_r));
179		−
180		− pix1 += line_size;
181		− p2_l = np2_l;
182		− p2_r = np2_r;
183		−
184		− } while (--h);
185		− } else {
186		− uint64_t p2_l, p2_r;
187		− p2_l = ldq(pix2);
188		− p2_r = ldq(pix2 + 8);
189		− do {
190		− uint64_t p1_l, p1_r, np2_l, np2_r;
191		−
192		− p1_l = ldq(pix1);
193		− p1_r = ldq(pix1 + 8);
194		− pix2 += line_size;
195		− np2_l = ldq(pix2);
196		− np2_r = ldq(pix2 + 8);
197		−
198		− result += perr(p1_l, avg2(p2_l, np2_l))
199		− + perr(p1_r, avg2(p2_r, np2_r));
200		−
201		− pix1 += line_size;
202		− p2_l = np2_l;
203		− p2_r = np2_r;
204		− } while (--h);
205		− }
206		− return result;
207		−}
208		−
209		−static int pix_abs16x16_xy2_mvi(struct MpegEncContext v, const uint8_t pix1, const uint8_t *pix2,
210		− ptrdiff_t line_size, int h)
211		−{
212		− int result = 0;
213		−
214		− uint64_t p1_l, p1_r;
215		− uint64_t p2_l, p2_r, p2_x;
216		−
217		− p1_l = ldq(pix1);
218		− p1_r = ldq(pix1 + 8);
219		−
220		− if ((size_t) pix2 & 0x7) { /* could be optimized a lot */
221		− p2_l = uldq(pix2);
222		− p2_r = uldq(pix2 + 8);
223		− p2_x = (uint64_t) pix2[16] << 56;
224		− } else {
225		− p2_l = ldq(pix2);
226		− p2_r = ldq(pix2 + 8);
227		− p2_x = ldq(pix2 + 16) << 56;
228		− }
229		−
230		− do {
231		− uint64_t np1_l, np1_r;
232		− uint64_t np2_l, np2_r, np2_x;
233		−
234		− pix1 += line_size;
235		− pix2 += line_size;
236		−
237		− np1_l = ldq(pix1);
238		− np1_r = ldq(pix1 + 8);
239		−
240		− if ((size_t) pix2 & 0x7) { /* could be optimized a lot */
241		− np2_l = uldq(pix2);
242		− np2_r = uldq(pix2 + 8);
243		− np2_x = (uint64_t) pix2[16] << 56;
244		− } else {
245		− np2_l = ldq(pix2);
246		− np2_r = ldq(pix2 + 8);
247		− np2_x = ldq(pix2 + 16) << 56;
248		− }
249		−
250		− result += perr(p1_l,
251		− avg4( p2_l, ( p2_l >> 8) \| ((uint64_t) p2_r << 56),
252		− np2_l, (np2_l >> 8) \| ((uint64_t) np2_r << 56)))
253		− + perr(p1_r,
254		− avg4( p2_r, ( p2_r >> 8) \| ((uint64_t) p2_x),
255		− np2_r, (np2_r >> 8) \| ((uint64_t) np2_x)));
256		−
257		− p1_l = np1_l;
258		− p1_r = np1_r;
259		− p2_l = np2_l;
260		− p2_r = np2_r;
261		− p2_x = np2_x;
262		− } while (--h);
263		−
264		− return result;
265		−}
266		−
267		−av_cold void ff_me_cmp_init_alpha(MECmpContext c, AVCodecContext avctx)
268		−{
269		− /* amask clears all bits that correspond to present features. */
270		− if (amask(AMASK_MVI) == 0) {
271		− c->sad[0] = pix_abs16x16_mvi_asm;
272		− c->sad[1] = pix_abs8x8_mvi;
273		− c->pix_abs[0][0] = pix_abs16x16_mvi_asm;
274		− c->pix_abs[1][0] = pix_abs8x8_mvi;
275		− c->pix_abs[0][1] = pix_abs16x16_x2_mvi;
276		− c->pix_abs[0][2] = pix_abs16x16_y2_mvi;
277		− c->pix_abs[0][3] = pix_abs16x16_xy2_mvi;
278		− }
279		−}

libavcodec/alpha/me_cmp_mvi_asm.S+0−179

@@ -1,179 +0,0 @@
1		−/*
2		− * Alpha optimized DSP utils
3		− * Copyright (c) 2002 Falk Hueffner <falk@debian.org>
4		− *
5		− * This file is part of FFmpeg.
6		− *
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.
11		− *
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.
16		− *
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		− */
21		−
22		−#include "regdef.h"
23		−
24		−/* Some nicer register names. */
25		−#define ta t10
26		−#define tb t11
27		−#define tc t12
28		−#define td AT
29		−/* Danger: these overlap with the argument list and the return value */
30		−#define te a5
31		−#define tf a4
32		−#define tg a3
33		−#define th v0
34		−
35		− .set noat
36		− .set noreorder
37		− .arch pca56
38		− .text
39		−
40		−/*****************************************************************************
41		− * int pix_abs16x16_mvi_asm(const uint8_t pix1, const uint8_t pix2, int line_size)
42		− *
43		− * This code is written with a pca56 in mind. For ev6, one should
44		− * really take the increased latency of 3 cycles for MVI instructions
45		− * into account.
46		− *
47		− * It is important to keep the loading and first use of a register as
48		− * far apart as possible, because if a register is accessed before it
49		− * has been fetched from memory, the CPU will stall.
50		− */
51		− .align 4
52		− .globl pix_abs16x16_mvi_asm
53		− .ent pix_abs16x16_mvi_asm
54		−pix_abs16x16_mvi_asm:
55		− .frame sp, 0, ra, 0
56		− .prologue 0
57		−
58		− and a2, 7, t0
59		− clr v0
60		− beq t0, $aligned
61		− .align 4
62		−$unaligned:
63		− /* Registers:
64		− line 0:
65		− t0: left_u -> left lo -> left
66		− t1: mid
67		− t2: right_u -> right hi -> right
68		− t3: ref left
69		− t4: ref right
70		− line 1:
71		− t5: left_u -> left lo -> left
72		− t6: mid
73		− t7: right_u -> right hi -> right
74		− t8: ref left
75		− t9: ref right
76		− temp:
77		− ta: left hi
78		− tb: right lo
79		− tc: error left
80		− td: error right */
81		−
82		− /* load line 0 */
83		− ldq_u t0, 0(a2) # left_u
84		− ldq_u t1, 8(a2) # mid
85		− ldq_u t2, 16(a2) # right_u
86		− ldq t3, 0(a1) # ref left
87		− ldq t4, 8(a1) # ref right
88		− addq a1, a3, a1 # pix1
89		− addq a2, a3, a2 # pix2
90		− /* load line 1 */
91		− ldq_u t5, 0(a2) # left_u
92		− ldq_u t6, 8(a2) # mid
93		− ldq_u t7, 16(a2) # right_u
94		− ldq t8, 0(a1) # ref left
95		− ldq t9, 8(a1) # ref right
96		− addq a1, a3, a1 # pix1
97		− addq a2, a3, a2 # pix2
98		− /* calc line 0 */
99		− extql t0, a2, t0 # left lo
100		− extqh t1, a2, ta # left hi
101		− extql t1, a2, tb # right lo
102		− or t0, ta, t0 # left
103		− extqh t2, a2, t2 # right hi
104		− perr t3, t0, tc # error left
105		− or t2, tb, t2 # right
106		− perr t4, t2, td # error right
107		− addq v0, tc, v0 # add error left
108		− addq v0, td, v0 # add error left
109		− /* calc line 1 */
110		− extql t5, a2, t5 # left lo
111		− extqh t6, a2, ta # left hi
112		− extql t6, a2, tb # right lo
113		− or t5, ta, t5 # left
114		− extqh t7, a2, t7 # right hi
115		− perr t8, t5, tc # error left
116		− or t7, tb, t7 # right
117		− perr t9, t7, td # error right
118		− addq v0, tc, v0 # add error left
119		− addq v0, td, v0 # add error left
120		− /* loop */
121		− subq a4, 2, a4 # h -= 2
122		− bne a4, $unaligned
123		− ret
124		−
125		− .align 4
126		−$aligned:
127		− /* load line 0 */
128		− ldq t0, 0(a2) # left
129		− ldq t1, 8(a2) # right
130		− addq a2, a3, a2 # pix2
131		− ldq t2, 0(a1) # ref left
132		− ldq t3, 8(a1) # ref right
133		− addq a1, a3, a1 # pix1
134		− /* load line 1 */
135		− ldq t4, 0(a2) # left
136		− ldq t5, 8(a2) # right
137		− addq a2, a3, a2 # pix2
138		− ldq t6, 0(a1) # ref left
139		− ldq t7, 8(a1) # ref right
140		− addq a1, a3, a1 # pix1
141		− /* load line 2 */
142		− ldq t8, 0(a2) # left
143		− ldq t9, 8(a2) # right
144		− addq a2, a3, a2 # pix2
145		− ldq ta, 0(a1) # ref left
146		− ldq tb, 8(a1) # ref right
147		− addq a1, a3, a1 # pix1
148		− /* load line 3 */
149		− ldq tc, 0(a2) # left
150		− ldq td, 8(a2) # right
151		− addq a2, a3, a2 # pix2
152		− ldq te, 0(a1) # ref left
153		− ldq a0, 8(a1) # ref right
154		− /* calc line 0 */
155		− perr t0, t2, t0 # error left
156		− addq a1, a3, a1 # pix1
157		− perr t1, t3, t1 # error right
158		− addq v0, t0, v0 # add error left
159		− /* calc line 1 */
160		− perr t4, t6, t0 # error left
161		− addq v0, t1, v0 # add error right
162		− perr t5, t7, t1 # error right
163		− addq v0, t0, v0 # add error left
164		− /* calc line 2 */
165		− perr t8, ta, t0 # error left
166		− addq v0, t1, v0 # add error right
167		− perr t9, tb, t1 # error right
168		− addq v0, t0, v0 # add error left
169		− /* calc line 3 */
170		− perr tc, te, t0 # error left
171		− addq v0, t1, v0 # add error right
172		− perr td, a0, t1 # error right
173		− addq v0, t0, v0 # add error left
174		− addq v0, t1, v0 # add error right
175		− /* loop */
176		− subq a4, 4, a4 # h -= 4
177		− bne a4, $aligned
178		− ret
179		− .end pix_abs16x16_mvi_asm

libavcodec/alpha/mpegvideo_alpha.c+0−110

@@ -1,110 +0,0 @@
1		−/*
2		− * Alpha optimized DSP utils
3		− * Copyright (c) 2002 Falk Hueffner <falk@debian.org>
4		− *
5		− * This file is part of FFmpeg.
6		− *
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.
11		− *
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.
16		− *
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		− */
21		−
22		−#include "libavutil/attributes.h"
23		−#include "libavcodec/mpegvideo.h"
24		−#include "asm.h"
25		−
26		−static void dct_unquantize_h263_axp(int16_t *block, int n_coeffs,
27		− uint64_t qscale, uint64_t qadd)
28		−{
29		− uint64_t qmul = qscale << 1;
30		− uint64_t correction = WORD_VEC(qmul * 255 >> 8);
31		− int i;
32		−
33		− qadd = WORD_VEC(qadd);
34		−
35		− for(i = 0; i <= n_coeffs; block += 4, i += 4) {
36		− uint64_t levels, negmask, zeros, add, sub;
37		−
38		− levels = ldq(block);
39		− if (levels == 0)
40		− continue;
41		−
42		−#ifdef __alpha_max__
43		− /* I don't think the speed difference justifies runtime
44		− detection. */
45		− negmask = maxsw4(levels, -1); /* negative -> ffff (-1) */
46		− negmask = minsw4(negmask, 0); /* positive -> 0000 (0) */
47		−#else
48		− negmask = cmpbge(WORD_VEC(0x7fff), levels);
49		− negmask &= (negmask >> 1) \| (1 << 7);
50		− negmask = zap(-1, negmask);
51		−#endif
52		−
53		− zeros = cmpbge(0, levels);
54		− zeros &= zeros >> 1;
55		− /* zeros \|= zeros << 1 is not needed since qadd <= 255, so
56		− zapping the lower byte suffices. */
57		−
58		− levels *= qmul;
59		− levels -= correction & (negmask << 16);
60		−
61		− add = qadd & ~negmask;
62		− sub = qadd & negmask;
63		− /* Set qadd to 0 for levels == 0. */
64		− add = zap(add, zeros);
65		− levels += add;
66		− levels -= sub;
67		−
68		− stq(levels, block);
69		− }
70		−}
71		−
72		−static void dct_unquantize_h263_intra_axp(MpegEncContext s, int16_t block,
73		− int n, int qscale)
74		−{
75		− int n_coeffs;
76		− uint64_t qadd;
77		− int16_t block0 = block[0];
78		−
79		− if (!s->h263_aic) {
80		− if (n < 4)
81		− block0 *= s->y_dc_scale;
82		− else
83		− block0 *= s->c_dc_scale;
84		− qadd = (qscale - 1) \| 1;
85		− } else {
86		− qadd = 0;
87		− }
88		−
89		− if(s->ac_pred)
90		− n_coeffs = 63;
91		− else
92		− n_coeffs = s->inter_scantable.raster_end[s->block_last_index[n]];
93		−
94		− dct_unquantize_h263_axp(block, n_coeffs, qscale, qadd);
95		−
96		− block[0] = block0;
97		−}
98		−
99		−static void dct_unquantize_h263_inter_axp(MpegEncContext s, int16_t block,
100		− int n, int qscale)
101		−{
102		− int n_coeffs = s->inter_scantable.raster_end[s->block_last_index[n]];
103		− dct_unquantize_h263_axp(block, n_coeffs, qscale, (qscale - 1) \| 1);
104		−}
105		−
106		−av_cold void ff_mpv_common_init_axp(MpegEncContext *s)
107		−{
108		− s->dct_unquantize_h263_intra = dct_unquantize_h263_intra_axp;
109		− s->dct_unquantize_h263_inter = dct_unquantize_h263_inter_axp;
110		−}

libavcodec/alpha/pixblockdsp_alpha.c+0−79

@@ -1,79 +0,0 @@
1		−/*
2		− * SIMD-optimized pixel operations
3		− *
4		− * This file is part of FFmpeg.
5		− *
6		− * FFmpeg is free software; you can redistribute it and/or
7		− * modify it under the terms of the GNU Lesser General Public
8		− * License as published by the Free Software Foundation; either
9		− * version 2.1 of the License, or (at your option) any later version.
10		− *
11		− * FFmpeg is distributed in the hope that it will be useful,
12		− * but WITHOUT ANY WARRANTY; without even the implied warranty of
13		− * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14		− * Lesser General Public License for more details.
15		− *
16		− * You should have received a copy of the GNU Lesser General Public
17		− * License along with FFmpeg; if not, write to the Free Software
18		− * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19		− */
20		−
21		−#include "libavutil/attributes.h"
22		−#include "libavcodec/pixblockdsp.h"
23		−#include "asm.h"
24		−
25		−static void get_pixels_mvi(int16_t *restrict block,
26		− const uint8_t *restrict pixels, ptrdiff_t stride)
27		−{
28		− int h = 8;
29		−
30		− do {
31		− uint64_t p;
32		−
33		− p = ldq(pixels);
34		− stq(unpkbw(p), block);
35		− stq(unpkbw(p >> 32), block + 4);
36		−
37		− pixels += stride;
38		− block += 8;
39		− } while (--h);
40		−}
41		−
42		−static void diff_pixels_mvi(int16_t block, const uint8_t s1, const uint8_t *s2,
43		− ptrdiff_t stride)
44		−{
45		− int h = 8;
46		− uint64_t mask = 0x4040;
47		−
48		− mask \|= mask << 16;
49		− mask \|= mask << 32;
50		− do {
51		− uint64_t x, y, c, d, a;
52		− uint64_t signs;
53		−
54		− x = ldq(s1);
55		− y = ldq(s2);
56		− c = cmpbge(x, y);
57		− d = x - y;
58		− a = zap(mask, c); /* We use 0x4040404040404040 here... */
59		− d += 4 * a; /* ...so we can use s4addq here. */
60		− signs = zap(-1, c);
61		−
62		− stq(unpkbw(d) \| (unpkbw(signs) << 8), block);
63		− stq(unpkbw(d >> 32) \| (unpkbw(signs >> 32) << 8), block + 4);
64		−
65		− s1 += stride;
66		− s2 += stride;
67		− block += 8;
68		− } while (--h);
69		−}
70		−
71		−av_cold void ff_pixblockdsp_init_alpha(PixblockDSPContext c, AVCodecContext avctx,
72		− unsigned high_bit_depth)
73		−{
74		− if (amask(AMASK_MVI) == 0) {
75		− if (!high_bit_depth)
76		− c->get_pixels = get_pixels_mvi;
77		− c->diff_pixels = diff_pixels_mvi;
78		− }
79		−}

libavcodec/alpha/regdef.h+0−77

@@ -1,77 +0,0 @@
1		−/*
2		− * Alpha optimized DSP utils
3		− * copyright (c) 2002 Falk Hueffner <falk@debian.org>
4		− *
5		− * This file is part of FFmpeg.
6		− *
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.
11		− *
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.
16		− *
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		− */
21		−
22		−/* Some BSDs don't seem to have regdef.h... sigh */
23		−#ifndef AVCODEC_ALPHA_REGDEF_H
24		−#define AVCODEC_ALPHA_REGDEF_H
25		−
26		−#define v0 $0 /* function return value */
27		−
28		−#define t0 $1 /* temporary registers (caller-saved) */
29		−#define t1 $2
30		−#define t2 $3
31		−#define t3 $4
32		−#define t4 $5
33		−#define t5 $6
34		−#define t6 $7
35		−#define t7 $8
36		−
37		−#define s0 $9 /* saved-registers (callee-saved registers) */
38		−#define s1 $10
39		−#define s2 $11
40		−#define s3 $12
41		−#define s4 $13
42		−#define s5 $14
43		−#define s6 $15
44		−#define fp s6 /* frame-pointer (s6 in frame-less procedures) */
45		−
46		−#define a0 $16 /* argument registers (caller-saved) */
47		−#define a1 $17
48		−#define a2 $18
49		−#define a3 $19
50		−#define a4 $20
51		−#define a5 $21
52		−
53		−#define t8 $22 /* more temps (caller-saved) */
54		−#define t9 $23
55		−#define t10 $24
56		−#define t11 $25
57		−#define ra $26 /* return address register */
58		−#define t12 $27
59		−
60		−#define pv t12 /* procedure-variable register */
61		−#define AT $at /* assembler temporary */
62		−#define gp $29 /* global pointer */
63		−#define sp $30 /* stack pointer */
64		−#define zero $31 /* reads as zero, writes are noops */
65		−
66		−/* Some nicer register names. */
67		−#define ta t10
68		−#define tb t11
69		−#define tc t12
70		−#define td AT
71		−/* Danger: these overlap with the argument list and the return value */
72		−#define te a5
73		−#define tf a4
74		−#define tg a3
75		−#define th v0
76		−
77		−#endif /* AVCODEC_ALPHA_REGDEF_H */

libavcodec/alpha/simple_idct_alpha.c+0−303

@@ -1,303 +0,0 @@
1		−/*
2		− * Simple IDCT (Alpha optimized)
3		− *
4		− * Copyright (c) 2001 Michael Niedermayer <michaelni@gmx.at>
5		− *
6		− * based upon some outcommented C code from mpeg2dec (idct_mmx.c
7		− * written by Aaron Holtzman <aholtzma@ess.engr.uvic.ca>)
8		− *
9		− * Alpha optimizations by Måns Rullgård <mans@mansr.com>
10		− * and Falk Hueffner <falk@debian.org>
11		− *
12		− * This file is part of FFmpeg.
13		− *
14		− * FFmpeg is free software; you can redistribute it and/or
15		− * modify it under the terms of the GNU Lesser General Public
16		− * License as published by the Free Software Foundation; either
17		− * version 2.1 of the License, or (at your option) any later version.
18		− *
19		− * FFmpeg is distributed in the hope that it will be useful,
20		− * but WITHOUT ANY WARRANTY; without even the implied warranty of
21		− * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
22		− * Lesser General Public License for more details.
23		− *
24		− * You should have received a copy of the GNU Lesser General Public
25		− * License along with FFmpeg; if not, write to the Free Software
26		− * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
27		− */
28		−
29		−#include "idctdsp_alpha.h"
30		−#include "asm.h"
31		−
32		−// cos(i * M_PI / 16) * sqrt(2) * (1 << 14)
33		−// W4 is actually exactly 16384, but using 16383 works around
34		−// accumulating rounding errors for some encoders
35		−#define W1 22725
36		−#define W2 21407
37		−#define W3 19266
38		−#define W4 16383
39		−#define W5 12873
40		−#define W6 8867
41		−#define W7 4520
42		−#define ROW_SHIFT 11
43		−#define COL_SHIFT 20
44		−
45		−/* 0: all entries 0, 1: only first entry nonzero, 2: otherwise */
46		−static inline int idct_row(int16_t *row)
47		−{
48		− int a0, a1, a2, a3, b0, b1, b2, b3, t;
49		− uint64_t l, r, t2;
50		− l = ldq(row);
51		− r = ldq(row + 4);
52		−
53		− if (l == 0 && r == 0)
54		− return 0;
55		−
56		− a0 = W4 * sextw(l) + (1 << (ROW_SHIFT - 1));
57		−
58		− if (((l & ~0xffffUL) \| r) == 0) {
59		− a0 >>= ROW_SHIFT;
60		− t2 = (uint16_t) a0;
61		− t2 \|= t2 << 16;
62		− t2 \|= t2 << 32;
63		−
64		− stq(t2, row);
65		− stq(t2, row + 4);
66		− return 1;
67		− }
68		−
69		− a1 = a0;
70		− a2 = a0;
71		− a3 = a0;
72		−
73		− t = extwl(l, 4); /* row[2] */
74		− if (t != 0) {
75		− t = sextw(t);
76		− a0 += W2 * t;
77		− a1 += W6 * t;
78		− a2 -= W6 * t;
79		− a3 -= W2 * t;
80		− }
81		−
82		− t = extwl(r, 0); /* row[4] */
83		− if (t != 0) {
84		− t = sextw(t);
85		− a0 += W4 * t;
86		− a1 -= W4 * t;
87		− a2 -= W4 * t;
88		− a3 += W4 * t;
89		− }
90		−
91		− t = extwl(r, 4); /* row[6] */
92		− if (t != 0) {
93		− t = sextw(t);
94		− a0 += W6 * t;
95		− a1 -= W2 * t;
96		− a2 += W2 * t;
97		− a3 -= W6 * t;
98		− }
99		−
100		− t = extwl(l, 2); /* row[1] */
101		− if (t != 0) {
102		− t = sextw(t);
103		− b0 = W1 * t;
104		− b1 = W3 * t;
105		− b2 = W5 * t;
106		− b3 = W7 * t;
107		− } else {
108		− b0 = 0;
109		− b1 = 0;
110		− b2 = 0;
111		− b3 = 0;
112		− }
113		−
114		− t = extwl(l, 6); /* row[3] */
115		− if (t) {
116		− t = sextw(t);
117		− b0 += W3 * t;
118		− b1 -= W7 * t;
119		− b2 -= W1 * t;
120		− b3 -= W5 * t;
121		− }
122		−
123		−
124		− t = extwl(r, 2); /* row[5] */
125		− if (t) {
126		− t = sextw(t);
127		− b0 += W5 * t;
128		− b1 -= W1 * t;
129		− b2 += W7 * t;
130		− b3 += W3 * t;
131		− }
132		−
133		− t = extwl(r, 6); /* row[7] */
134		− if (t) {
135		− t = sextw(t);
136		− b0 += W7 * t;
137		− b1 -= W5 * t;
138		− b2 += W3 * t;
139		− b3 -= W1 * t;
140		− }
141		−
142		− row[0] = (a0 + b0) >> ROW_SHIFT;
143		− row[1] = (a1 + b1) >> ROW_SHIFT;
144		− row[2] = (a2 + b2) >> ROW_SHIFT;
145		− row[3] = (a3 + b3) >> ROW_SHIFT;
146		− row[4] = (a3 - b3) >> ROW_SHIFT;
147		− row[5] = (a2 - b2) >> ROW_SHIFT;
148		− row[6] = (a1 - b1) >> ROW_SHIFT;
149		− row[7] = (a0 - b0) >> ROW_SHIFT;
150		−
151		− return 2;
152		−}
153		−
154		−static inline void idct_col(int16_t *col)
155		−{
156		− int a0, a1, a2, a3, b0, b1, b2, b3;
157		−
158		− col[0] += (1 << (COL_SHIFT - 1)) / W4;
159		−
160		− a0 = W4 * col[8 * 0];
161		− a1 = W4 * col[8 * 0];
162		− a2 = W4 * col[8 * 0];
163		− a3 = W4 * col[8 * 0];
164		−
165		− if (col[8 * 2]) {
166		− a0 += W2 * col[8 * 2];
167		− a1 += W6 * col[8 * 2];
168		− a2 -= W6 * col[8 * 2];
169		− a3 -= W2 * col[8 * 2];
170		− }
171		−
172		− if (col[8 * 4]) {
173		− a0 += W4 * col[8 * 4];
174		− a1 -= W4 * col[8 * 4];
175		− a2 -= W4 * col[8 * 4];
176		− a3 += W4 * col[8 * 4];
177		− }
178		−
179		− if (col[8 * 6]) {
180		− a0 += W6 * col[8 * 6];
181		− a1 -= W2 * col[8 * 6];
182		− a2 += W2 * col[8 * 6];
183		− a3 -= W6 * col[8 * 6];
184		− }
185		−
186		− if (col[8 * 1]) {
187		− b0 = W1 * col[8 * 1];
188		− b1 = W3 * col[8 * 1];
189		− b2 = W5 * col[8 * 1];
190		− b3 = W7 * col[8 * 1];
191		− } else {
192		− b0 = 0;
193		− b1 = 0;
194		− b2 = 0;
195		− b3 = 0;
196		− }
197		−
198		− if (col[8 * 3]) {
199		− b0 += W3 * col[8 * 3];
200		− b1 -= W7 * col[8 * 3];
201		− b2 -= W1 * col[8 * 3];
202		− b3 -= W5 * col[8 * 3];
203		− }
204		−
205		− if (col[8 * 5]) {
206		− b0 += W5 * col[8 * 5];
207		− b1 -= W1 * col[8 * 5];
208		− b2 += W7 * col[8 * 5];
209		− b3 += W3 * col[8 * 5];
210		− }
211		−
212		− if (col[8 * 7]) {
213		− b0 += W7 * col[8 * 7];
214		− b1 -= W5 * col[8 * 7];
215		− b2 += W3 * col[8 * 7];
216		− b3 -= W1 * col[8 * 7];
217		− }
218		−
219		− col[8 * 0] = (a0 + b0) >> COL_SHIFT;
220		− col[8 * 7] = (a0 - b0) >> COL_SHIFT;
221		− col[8 * 1] = (a1 + b1) >> COL_SHIFT;
222		− col[8 * 6] = (a1 - b1) >> COL_SHIFT;
223		− col[8 * 2] = (a2 + b2) >> COL_SHIFT;
224		− col[8 * 5] = (a2 - b2) >> COL_SHIFT;
225		− col[8 * 3] = (a3 + b3) >> COL_SHIFT;
226		− col[8 * 4] = (a3 - b3) >> COL_SHIFT;
227		−}
228		−
229		−/* If all rows but the first one are zero after row transformation,
230		− all rows will be identical after column transformation. */
231		−static inline void idct_col2(int16_t *col)
232		−{
233		− int i;
234		− uint64_t l, r;
235		−
236		− for (i = 0; i < 8; ++i) {
237		− int a0 = col[i] + (1 << (COL_SHIFT - 1)) / W4;
238		−
239		− a0 *= W4;
240		− col[i] = a0 >> COL_SHIFT;
241		− }
242		−
243		− l = ldq(col + 0 * 4); r = ldq(col + 1 * 4);
244		− stq(l, col + 2 * 4); stq(r, col + 3 * 4);
245		− stq(l, col + 4 * 4); stq(r, col + 5 * 4);
246		− stq(l, col + 6 * 4); stq(r, col + 7 * 4);
247		− stq(l, col + 8 * 4); stq(r, col + 9 * 4);
248		− stq(l, col + 10 * 4); stq(r, col + 11 * 4);
249		− stq(l, col + 12 * 4); stq(r, col + 13 * 4);
250		− stq(l, col + 14 * 4); stq(r, col + 15 * 4);
251		−}
252		−
253		−void ff_simple_idct_axp(int16_t *block)
254		−{
255		−
256		− int i;
257		− int rowsZero = 1; /* all rows except row 0 zero */
258		− int rowsConstant = 1; /* all rows consist of a constant value */
259		−
260		− for (i = 0; i < 8; i++) {
261		− int sparseness = idct_row(block + 8 * i);
262		−
263		− if (i > 0 && sparseness > 0)
264		− rowsZero = 0;
265		− if (sparseness == 2)
266		− rowsConstant = 0;
267		− }
268		−
269		− if (rowsZero) {
270		− idct_col2(block);
271		− } else if (rowsConstant) {
272		− idct_col(block);
273		− for (i = 0; i < 8; i += 2) {
274		− uint64_t v = (uint16_t) block[0];
275		− uint64_t w = (uint16_t) block[8];
276		−
277		− v \|= v << 16;
278		− w \|= w << 16;
279		− v \|= v << 32;
280		− w \|= w << 32;
281		− stq(v, block + 0 * 4);
282		− stq(v, block + 1 * 4);
283		− stq(w, block + 2 * 4);
284		− stq(w, block + 3 * 4);
285		− block += 4 * 4;
286		− }
287		− } else {
288		− for (i = 0; i < 8; i++)
289		− idct_col(block + i);
290		− }
291		−}
292		−
293		−void ff_simple_idct_put_axp(uint8_t dest, ptrdiff_t line_size, int16_t block)
294		−{
295		− ff_simple_idct_axp(block);
296		− put_pixels_clamped_axp_p(block, dest, line_size);
297		−}
298		−
299		−void ff_simple_idct_add_axp(uint8_t dest, ptrdiff_t line_size, int16_t block)
300		−{
301		− ff_simple_idct_axp(block);
302		− add_pixels_clamped_axp_p(block, dest, line_size);
303		−}

libavcodec/blockdsp.c+1−3

@@ -63,9 +63,7 @@ av_cold void ff_blockdsp_init(BlockDSPContext *c) · 62 unmodified lines
63	63	c->fill_block_tab[0] = fill_block16_c;
64	64	c->fill_block_tab[1] = fill_block8_c;
65	65
66		−#if ARCH_ALPHA
67		− ff_blockdsp_init_alpha(c);
68		−#elif ARCH_ARM
	66	+#if ARCH_ARM
69	67	ff_blockdsp_init_arm(c);
70	68	#elif ARCH_PPC
71	69	ff_blockdsp_init_ppc(c);

libavcodec/hpeldsp.c+0−2

@@ -357,8 +357,6 @@ av_cold void ff_hpeldsp_init(HpelDSPContext *c, int flags) · 356 unmodified lines
357	357
358	358	#if ARCH_AARCH64
359	359	ff_hpeldsp_init_aarch64(c, flags);
360		−#elif ARCH_ALPHA
361		− ff_hpeldsp_init_alpha(c, flags);
362	360	#elif ARCH_ARM
363	361	ff_hpeldsp_init_arm(c, flags);
364	362	#elif ARCH_PPC

libavcodec/idctdsp.c+0−2

@@ -294,8 +294,6 @@ av_cold void ff_idctdsp_init(IDCTDSPContext c, AVCodecContext avctx) · 293 unmodified lines
294	294
295	295	#if ARCH_AARCH64
296	296	ff_idctdsp_init_aarch64(c, avctx, high_bit_depth);
297		−#elif ARCH_ALPHA
298		− ff_idctdsp_init_alpha(c, avctx, high_bit_depth);
299	297	#elif ARCH_ARM
300	298	ff_idctdsp_init_arm(c, avctx, high_bit_depth);
301	299	#elif ARCH_PPC

libavcodec/me_cmp.c+0−2

@@ -1058,8 +1058,6 @@ av_cold void ff_me_cmp_init(MECmpContext c, AVCodecContext avctx) · 1057 unmodified lines
1058	1058
1059	1059	#if ARCH_AARCH64
1060	1060	ff_me_cmp_init_aarch64(c, avctx);
1061		−#elif ARCH_ALPHA
1062		− ff_me_cmp_init_alpha(c, avctx);
1063	1061	#elif ARCH_ARM
1064	1062	ff_me_cmp_init_arm(c, avctx);
1065	1063	#elif ARCH_PPC

libavcodec/mpegvideo.c+1−3

@@ -305,9 +305,7 @@ static av_cold int dct_init(MpegEncContext *s) · 304 unmodified lines
305	305	ff_mpv_common_init_neon(s);
306	306	#endif
307	307
308		−#if ARCH_ALPHA
309		− ff_mpv_common_init_axp(s);
310		−#elif ARCH_ARM
	308	+#if ARCH_ARM
311	309	ff_mpv_common_init_arm(s);
312	310	#elif ARCH_PPC
313	311	ff_mpv_common_init_ppc(s);

libavcodec/pixblockdsp.c+0−2

@@ -103,8 +103,6 @@ av_cold void ff_pixblockdsp_init(PixblockDSPContext c, AVCodecContext avctx) · 102 unmodified lines
103	103
104	104	#if ARCH_AARCH64
105	105	ff_pixblockdsp_init_aarch64(c, avctx, high_bit_depth);
106		−#elif ARCH_ALPHA
107		− ff_pixblockdsp_init_alpha(c, avctx, high_bit_depth);
108	106	#elif ARCH_ARM
109	107	ff_pixblockdsp_init_arm(c, avctx, high_bit_depth);
110	108	#elif ARCH_PPC

libavcodec/alpha: remove DSP & support code

Keyboard shortcuts

General

Issues and pull requests

Issues, pull requests and Inbox

Repository code