Commit fab9df63 authored by Diego Biurrun's avatar Diego Biurrun

dsputil: Split off global motion compensation bits into a separate context

parent f23d26a6
......@@ -63,8 +63,8 @@ OBJS-$(CONFIG_MPEGAUDIODSP) += mpegaudiodsp.o \
mpegaudiodsp_data.o \
mpegaudiodsp_fixed.o \
mpegaudiodsp_float.o
OBJS-$(CONFIG_MPEGVIDEO) += mpegvideo.o mpegvideo_motion.o \
mpegutils.o
OBJS-$(CONFIG_MPEGVIDEO) += mpegvideo.o mpegvideodsp.o \
mpegvideo_motion.o mpegutils.o
OBJS-$(CONFIG_MPEGVIDEOENC) += mpegvideo_enc.o mpeg12data.o \
motion_est.o ratecontrol.o
OBJS-$(CONFIG_QPELDSP) += qpeldsp.o
......
......@@ -3,8 +3,6 @@
* Copyright (c) 2000, 2001 Fabrice Bellard
* Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
*
* gmc & q-pel & 32/64 bit based MC by Michael Niedermayer <michaelni@gmx.at>
*
* This file is part of Libav.
*
* Libav is free software; you can redistribute it and/or
......@@ -352,92 +350,6 @@ static int sum_abs_dctelem_c(int16_t *block)
#define avg2(a, b) ((a + b + 1) >> 1)
#define avg4(a, b, c, d) ((a + b + c + d + 2) >> 2)
static void gmc1_c(uint8_t *dst, uint8_t *src, int stride, int h,
int x16, int y16, int rounder)
{
const int A = (16 - x16) * (16 - y16);
const int B = (x16) * (16 - y16);
const int C = (16 - x16) * (y16);
const int D = (x16) * (y16);
int i;
for (i = 0; i < h; i++) {
dst[0] = (A * src[0] + B * src[1] + C * src[stride + 0] + D * src[stride + 1] + rounder) >> 8;
dst[1] = (A * src[1] + B * src[2] + C * src[stride + 1] + D * src[stride + 2] + rounder) >> 8;
dst[2] = (A * src[2] + B * src[3] + C * src[stride + 2] + D * src[stride + 3] + rounder) >> 8;
dst[3] = (A * src[3] + B * src[4] + C * src[stride + 3] + D * src[stride + 4] + rounder) >> 8;
dst[4] = (A * src[4] + B * src[5] + C * src[stride + 4] + D * src[stride + 5] + rounder) >> 8;
dst[5] = (A * src[5] + B * src[6] + C * src[stride + 5] + D * src[stride + 6] + rounder) >> 8;
dst[6] = (A * src[6] + B * src[7] + C * src[stride + 6] + D * src[stride + 7] + rounder) >> 8;
dst[7] = (A * src[7] + B * src[8] + C * src[stride + 7] + D * src[stride + 8] + rounder) >> 8;
dst += stride;
src += stride;
}
}
void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
int dxx, int dxy, int dyx, int dyy, int shift, int r,
int width, int height)
{
int y, vx, vy;
const int s = 1 << shift;
width--;
height--;
for (y = 0; y < h; y++) {
int x;
vx = ox;
vy = oy;
for (x = 0; x < 8; x++) { // FIXME: optimize
int index;
int src_x = vx >> 16;
int src_y = vy >> 16;
int frac_x = src_x & (s - 1);
int frac_y = src_y & (s - 1);
src_x >>= shift;
src_y >>= shift;
if ((unsigned) src_x < width) {
if ((unsigned) src_y < height) {
index = src_x + src_y * stride;
dst[y * stride + x] =
((src[index] * (s - frac_x) +
src[index + 1] * frac_x) * (s - frac_y) +
(src[index + stride] * (s - frac_x) +
src[index + stride + 1] * frac_x) * frac_y +
r) >> (shift * 2);
} else {
index = src_x + av_clip(src_y, 0, height) * stride;
dst[y * stride + x] =
((src[index] * (s - frac_x) +
src[index + 1] * frac_x) * s +
r) >> (shift * 2);
}
} else {
if ((unsigned) src_y < height) {
index = av_clip(src_x, 0, width) + src_y * stride;
dst[y * stride + x] =
((src[index] * (s - frac_y) +
src[index + stride] * frac_y) * s +
r) >> (shift * 2);
} else {
index = av_clip(src_x, 0, width) +
av_clip(src_y, 0, height) * stride;
dst[y * stride + x] = src[index];
}
}
vx += dxx;
vy += dyx;
}
ox += dxy;
oy += dyy;
}
}
static inline int pix_abs16_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
int line_size, int h)
{
......@@ -1346,9 +1258,6 @@ av_cold void ff_dsputil_init(DSPContext *c, AVCodecContext *avctx)
c->sum_abs_dctelem = sum_abs_dctelem_c;
c->gmc1 = gmc1_c;
c->gmc = ff_gmc_c;
c->pix_sum = pix_sum_c;
c->pix_norm1 = pix_norm1_c;
......
......@@ -34,10 +34,6 @@
extern uint32_t ff_square_tab[512];
void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
int dxx, int dxy, int dyx, int dyy, int shift, int r,
int width, int height);
struct MpegEncContext;
/* Motion estimation:
* h is limited to { width / 2, width, 2 * width },
......@@ -84,18 +80,6 @@ typedef struct DSPContext {
uint8_t *pixels /* align 8 */,
int line_size);
int (*sum_abs_dctelem)(int16_t *block /* align 16 */);
/**
* translational global motion compensation.
*/
void (*gmc1)(uint8_t *dst /* align 8 */, uint8_t *src /* align 1 */,
int srcStride, int h, int x16, int y16, int rounder);
/**
* global motion compensation.
*/
void (*gmc)(uint8_t *dst /* align 8 */, uint8_t *src /* align 1 */,
int stride, int h, int ox, int oy,
int dxx, int dxy, int dyx, int dyy,
int shift, int r, int width, int height);
int (*pix_sum)(uint8_t *pix, int line_size);
int (*pix_norm1)(uint8_t *pix, int line_size);
......
......@@ -380,6 +380,7 @@ av_cold int ff_dct_common_init(MpegEncContext *s)
ff_blockdsp_init(&s->bdsp, s->avctx);
ff_dsputil_init(&s->dsp, s->avctx);
ff_hpeldsp_init(&s->hdsp, s->avctx->flags);
ff_mpegvideodsp_init(&s->mdsp);
ff_videodsp_init(&s->vdsp, s->avctx->bits_per_raw_sample);
s->dct_unquantize_h263_intra = dct_unquantize_h263_intra_c;
......
......@@ -35,6 +35,7 @@
#include "get_bits.h"
#include "h263dsp.h"
#include "hpeldsp.h"
#include "mpegvideodsp.h"
#include "put_bits.h"
#include "ratecontrol.h"
#include "parser.h"
......@@ -351,6 +352,7 @@ typedef struct MpegEncContext {
BlockDSPContext bdsp;
DSPContext dsp; ///< pointers for accelerated dsp functions
HpelDSPContext hdsp;
MpegVideoDSPContext mdsp;
QpelDSPContext qdsp;
VideoDSPContext vdsp;
H263DSPContext h263dsp;
......
......@@ -25,7 +25,6 @@
#include "libavutil/internal.h"
#include "avcodec.h"
#include "dsputil.h"
#include "h261.h"
#include "mpegutils.h"
#include "mpegvideo.h"
......@@ -72,10 +71,10 @@ static void gmc1_motion(MpegEncContext *s,
}
if ((motion_x | motion_y) & 7) {
s->dsp.gmc1(dest_y, ptr, linesize, 16,
motion_x & 15, motion_y & 15, 128 - s->no_rounding);
s->dsp.gmc1(dest_y + 8, ptr + 8, linesize, 16,
motion_x & 15, motion_y & 15, 128 - s->no_rounding);
s->mdsp.gmc1(dest_y, ptr, linesize, 16,
motion_x & 15, motion_y & 15, 128 - s->no_rounding);
s->mdsp.gmc1(dest_y + 8, ptr + 8, linesize, 16,
motion_x & 15, motion_y & 15, 128 - s->no_rounding);
} else {
int dxy;
......@@ -115,8 +114,8 @@ static void gmc1_motion(MpegEncContext *s,
ptr = s->edge_emu_buffer;
emu = 1;
}
s->dsp.gmc1(dest_cb, ptr, uvlinesize, 8,
motion_x & 15, motion_y & 15, 128 - s->no_rounding);
s->mdsp.gmc1(dest_cb, ptr, uvlinesize, 8,
motion_x & 15, motion_y & 15, 128 - s->no_rounding);
ptr = ref_picture[2] + offset;
if (emu) {
......@@ -127,8 +126,8 @@ static void gmc1_motion(MpegEncContext *s,
s->h_edge_pos >> 1, s->v_edge_pos >> 1);
ptr = s->edge_emu_buffer;
}
s->dsp.gmc1(dest_cr, ptr, uvlinesize, 8,
motion_x & 15, motion_y & 15, 128 - s->no_rounding);
s->mdsp.gmc1(dest_cr, ptr, uvlinesize, 8,
motion_x & 15, motion_y & 15, 128 - s->no_rounding);
}
static void gmc_motion(MpegEncContext *s,
......@@ -150,19 +149,19 @@ static void gmc_motion(MpegEncContext *s,
oy = s->sprite_offset[0][1] + s->sprite_delta[1][0] * s->mb_x * 16 +
s->sprite_delta[1][1] * s->mb_y * 16;
s->dsp.gmc(dest_y, ptr, linesize, 16,
ox, oy,
s->sprite_delta[0][0], s->sprite_delta[0][1],
s->sprite_delta[1][0], s->sprite_delta[1][1],
a + 1, (1 << (2 * a + 1)) - s->no_rounding,
s->h_edge_pos, s->v_edge_pos);
s->dsp.gmc(dest_y + 8, ptr, linesize, 16,
ox + s->sprite_delta[0][0] * 8,
oy + s->sprite_delta[1][0] * 8,
s->sprite_delta[0][0], s->sprite_delta[0][1],
s->sprite_delta[1][0], s->sprite_delta[1][1],
a + 1, (1 << (2 * a + 1)) - s->no_rounding,
s->h_edge_pos, s->v_edge_pos);
s->mdsp.gmc(dest_y, ptr, linesize, 16,
ox, oy,
s->sprite_delta[0][0], s->sprite_delta[0][1],
s->sprite_delta[1][0], s->sprite_delta[1][1],
a + 1, (1 << (2 * a + 1)) - s->no_rounding,
s->h_edge_pos, s->v_edge_pos);
s->mdsp.gmc(dest_y + 8, ptr, linesize, 16,
ox + s->sprite_delta[0][0] * 8,
oy + s->sprite_delta[1][0] * 8,
s->sprite_delta[0][0], s->sprite_delta[0][1],
s->sprite_delta[1][0], s->sprite_delta[1][1],
a + 1, (1 << (2 * a + 1)) - s->no_rounding,
s->h_edge_pos, s->v_edge_pos);
if (CONFIG_GRAY && s->flags & CODEC_FLAG_GRAY)
return;
......@@ -173,20 +172,20 @@ static void gmc_motion(MpegEncContext *s,
s->sprite_delta[1][1] * s->mb_y * 8;
ptr = ref_picture[1];
s->dsp.gmc(dest_cb, ptr, uvlinesize, 8,
ox, oy,
s->sprite_delta[0][0], s->sprite_delta[0][1],
s->sprite_delta[1][0], s->sprite_delta[1][1],
a + 1, (1 << (2 * a + 1)) - s->no_rounding,
s->h_edge_pos >> 1, s->v_edge_pos >> 1);
s->mdsp.gmc(dest_cb, ptr, uvlinesize, 8,
ox, oy,
s->sprite_delta[0][0], s->sprite_delta[0][1],
s->sprite_delta[1][0], s->sprite_delta[1][1],
a + 1, (1 << (2 * a + 1)) - s->no_rounding,
s->h_edge_pos >> 1, s->v_edge_pos >> 1);
ptr = ref_picture[2];
s->dsp.gmc(dest_cr, ptr, uvlinesize, 8,
ox, oy,
s->sprite_delta[0][0], s->sprite_delta[0][1],
s->sprite_delta[1][0], s->sprite_delta[1][1],
a + 1, (1 << (2 * a + 1)) - s->no_rounding,
s->h_edge_pos >> 1, s->v_edge_pos >> 1);
s->mdsp.gmc(dest_cr, ptr, uvlinesize, 8,
ox, oy,
s->sprite_delta[0][0], s->sprite_delta[0][1],
s->sprite_delta[1][0], s->sprite_delta[1][1],
a + 1, (1 << (2 * a + 1)) - s->no_rounding,
s->h_edge_pos >> 1, s->v_edge_pos >> 1);
}
static inline int hpel_motion(MpegEncContext *s,
......
/*
* This file is part of Libav.
*
* Libav 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.
*
* Libav 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 Libav; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
#include "config.h"
#include "libavutil/attributes.h"
#include "libavutil/common.h"
#include "mpegvideodsp.h"
static void gmc1_c(uint8_t *dst, uint8_t *src, int stride, int h,
int x16, int y16, int rounder)
{
const int A = (16 - x16) * (16 - y16);
const int B = (x16) * (16 - y16);
const int C = (16 - x16) * (y16);
const int D = (x16) * (y16);
int i;
for (i = 0; i < h; i++) {
dst[0] = (A * src[0] + B * src[1] + C * src[stride + 0] + D * src[stride + 1] + rounder) >> 8;
dst[1] = (A * src[1] + B * src[2] + C * src[stride + 1] + D * src[stride + 2] + rounder) >> 8;
dst[2] = (A * src[2] + B * src[3] + C * src[stride + 2] + D * src[stride + 3] + rounder) >> 8;
dst[3] = (A * src[3] + B * src[4] + C * src[stride + 3] + D * src[stride + 4] + rounder) >> 8;
dst[4] = (A * src[4] + B * src[5] + C * src[stride + 4] + D * src[stride + 5] + rounder) >> 8;
dst[5] = (A * src[5] + B * src[6] + C * src[stride + 5] + D * src[stride + 6] + rounder) >> 8;
dst[6] = (A * src[6] + B * src[7] + C * src[stride + 6] + D * src[stride + 7] + rounder) >> 8;
dst[7] = (A * src[7] + B * src[8] + C * src[stride + 7] + D * src[stride + 8] + rounder) >> 8;
dst += stride;
src += stride;
}
}
void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
int dxx, int dxy, int dyx, int dyy, int shift, int r,
int width, int height)
{
int y, vx, vy;
const int s = 1 << shift;
width--;
height--;
for (y = 0; y < h; y++) {
int x;
vx = ox;
vy = oy;
for (x = 0; x < 8; x++) { // FIXME: optimize
int index;
int src_x = vx >> 16;
int src_y = vy >> 16;
int frac_x = src_x & (s - 1);
int frac_y = src_y & (s - 1);
src_x >>= shift;
src_y >>= shift;
if ((unsigned) src_x < width) {
if ((unsigned) src_y < height) {
index = src_x + src_y * stride;
dst[y * stride + x] =
((src[index] * (s - frac_x) +
src[index + 1] * frac_x) * (s - frac_y) +
(src[index + stride] * (s - frac_x) +
src[index + stride + 1] * frac_x) * frac_y +
r) >> (shift * 2);
} else {
index = src_x + av_clip(src_y, 0, height) * stride;
dst[y * stride + x] =
((src[index] * (s - frac_x) +
src[index + 1] * frac_x) * s +
r) >> (shift * 2);
}
} else {
if ((unsigned) src_y < height) {
index = av_clip(src_x, 0, width) + src_y * stride;
dst[y * stride + x] =
((src[index] * (s - frac_y) +
src[index + stride] * frac_y) * s +
r) >> (shift * 2);
} else {
index = av_clip(src_x, 0, width) +
av_clip(src_y, 0, height) * stride;
dst[y * stride + x] = src[index];
}
}
vx += dxx;
vy += dyx;
}
ox += dxy;
oy += dyy;
}
}
av_cold void ff_mpegvideodsp_init(MpegVideoDSPContext *c)
{
c->gmc1 = gmc1_c;
c->gmc = ff_gmc_c;
if (ARCH_PPC)
ff_mpegvideodsp_init_ppc(c);
if (ARCH_X86)
ff_mpegvideodsp_init_x86(c);
}
/*
* This file is part of Libav.
*
* Libav 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.
*
* Libav 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 Libav; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
#ifndef AVCODEC_MPEGVIDEODSP_H
#define AVCODEC_MPEGVIDEODSP_H
#include <stdint.h>
void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
int dxx, int dxy, int dyx, int dyy, int shift, int r,
int width, int height);
typedef struct MpegVideoDSPContext {
/**
* translational global motion compensation.
*/
void (*gmc1)(uint8_t *dst /* align 8 */, uint8_t *src /* align 1 */,
int srcStride, int h, int x16, int y16, int rounder);
/**
* global motion compensation.
*/
void (*gmc)(uint8_t *dst /* align 8 */, uint8_t *src /* align 1 */,
int stride, int h, int ox, int oy,
int dxx, int dxy, int dyx, int dyy,
int shift, int r, int width, int height);
} MpegVideoDSPContext;
void ff_mpegvideodsp_init(MpegVideoDSPContext *c);
void ff_mpegvideodsp_init_ppc(MpegVideoDSPContext *c);
void ff_mpegvideodsp_init_x86(MpegVideoDSPContext *c);
#endif /* AVCODEC_MPEGVIDEODSP_H */
......@@ -10,7 +10,8 @@ OBJS-$(CONFIG_H264QPEL) += ppc/h264qpel.o
OBJS-$(CONFIG_HPELDSP) += ppc/hpeldsp_altivec.o
OBJS-$(CONFIG_HUFFYUVDSP) += ppc/huffyuvdsp_altivec.o
OBJS-$(CONFIG_MPEGAUDIODSP) += ppc/mpegaudiodsp_altivec.o
OBJS-$(CONFIG_MPEGVIDEO) += ppc/mpegvideo_altivec.o
OBJS-$(CONFIG_MPEGVIDEO) += ppc/mpegvideo_altivec.o \
ppc/mpegvideodsp.o
OBJS-$(CONFIG_VIDEODSP) += ppc/videodsp_ppc.o
OBJS-$(CONFIG_VP3DSP) += ppc/vp3dsp_altivec.o
......@@ -23,7 +24,6 @@ OBJS-$(CONFIG_VP8_DECODER) += ppc/vp8dsp_altivec.o
ALTIVEC-OBJS-$(CONFIG_DSPUTIL) += ppc/dsputil_altivec.o \
ppc/fdct_altivec.o \
ppc/gmc_altivec.o \
ppc/idct_altivec.o \
FFT-OBJS-$(HAVE_GNU_AS) += ppc/fft_altivec_s.o
......
......@@ -28,8 +28,7 @@
#include "libavcodec/dsputil.h"
void ff_fdct_altivec(int16_t *block);
void ff_gmc1_altivec(uint8_t *dst, uint8_t *src, int stride, int h,
int x16, int y16, int rounder);
void ff_idct_put_altivec(uint8_t *dest, int line_size, int16_t *block);
void ff_idct_add_altivec(uint8_t *dest, int line_size, int16_t *block);
......
......@@ -35,8 +35,6 @@ av_cold void ff_dsputil_init_ppc(DSPContext *c, AVCodecContext *avctx,
if (PPC_ALTIVEC(av_get_cpu_flags())) {
ff_dsputil_init_altivec(c, avctx, high_bit_depth);
c->gmc1 = ff_gmc1_altivec;
if (!high_bit_depth) {
#if CONFIG_ENCODERS
if (avctx->dct_algo == FF_DCT_AUTO ||
......
......@@ -23,12 +23,13 @@
#include "libavutil/mem.h"
#include "libavutil/ppc/types_altivec.h"
#include "libavutil/ppc/util_altivec.h"
#include "dsputil_altivec.h"
#include "libavcodec/mpegvideodsp.h"
#if HAVE_ALTIVEC
/* AltiVec-enhanced gmc1. ATM this code assumes stride is a multiple of 8
* to preserve proper dst alignment. */
void ff_gmc1_altivec(uint8_t *dst /* align 8 */, uint8_t *src /* align1 */,
int stride, int h, int x16, int y16, int rounder)
static void gmc1_altivec(uint8_t *dst /* align 8 */, uint8_t *src /* align1 */,
int stride, int h, int x16, int y16, int rounder)
{
int i;
const DECLARE_ALIGNED(16, unsigned short, rounder_a) = rounder;
......@@ -122,3 +123,11 @@ void ff_gmc1_altivec(uint8_t *dst /* align 8 */, uint8_t *src /* align1 */,
src += stride;
}
}
#endif /* HAVE_ALTIVEC */
av_cold void ff_mpegvideodsp_init_ppc(MpegVideoDSPContext *c)
{
#if HAVE_ALTIVEC
c->gmc1 = gmc1_altivec;
#endif /* HAVE_ALTIVEC */
}
......@@ -20,7 +20,8 @@ OBJS-$(CONFIG_HUFFYUVDSP) += x86/huffyuvdsp_init.o
OBJS-$(CONFIG_HUFFYUVENCDSP) += x86/huffyuvencdsp_mmx.o
OBJS-$(CONFIG_LPC) += x86/lpc.o
OBJS-$(CONFIG_MPEGAUDIODSP) += x86/mpegaudiodsp.o
OBJS-$(CONFIG_MPEGVIDEO) += x86/mpegvideo.o
OBJS-$(CONFIG_MPEGVIDEO) += x86/mpegvideo.o \
x86/mpegvideodsp.o
OBJS-$(CONFIG_MPEGVIDEOENC) += x86/mpegvideoenc.o
OBJS-$(CONFIG_QPELDSP) += x86/qpeldsp_init.o
OBJS-$(CONFIG_VIDEODSP) += x86/videodsp_init.o
......
......@@ -52,8 +52,6 @@ static av_cold void dsputil_init_mmx(DSPContext *c, AVCodecContext *avctx,
break;
}
}
c->gmc = ff_gmc_mmx;
#endif /* HAVE_MMX_INLINE */
}
......
......@@ -260,118 +260,4 @@ void ff_draw_edges_mmx(uint8_t *buf, int wrap, int width, int height,
}
}
void ff_gmc_mmx(uint8_t *dst, uint8_t *src,
int stride, int h, int ox, int oy,
int dxx, int dxy, int dyx, int dyy,
int shift, int r, int width, int height)
{
const int w = 8;
const int ix = ox >> (16 + shift);
const int iy = oy >> (16 + shift);
const int oxs = ox >> 4;
const int oys = oy >> 4;
const int dxxs = dxx >> 4;
const int dxys = dxy >> 4;
const int dyxs = dyx >> 4;
const int dyys = dyy >> 4;
const uint16_t r4[4] = { r, r, r, r };
const uint16_t dxy4[4] = { dxys, dxys, dxys, dxys };
const uint16_t dyy4[4] = { dyys, dyys, dyys, dyys };
const uint64_t shift2 = 2 * shift;
int x, y;
const int dxw = (dxx - (1 << (16 + shift))) * (w - 1);
const int dyh = (dyy - (1 << (16 + shift))) * (h - 1);
const int dxh = dxy * (h - 1);
const int dyw = dyx * (w - 1);
if ( // non-constant fullpel offset (3% of blocks)
((ox ^ (ox + dxw)) | (ox ^ (ox + dxh)) | (ox ^ (ox + dxw + dxh)) |
(oy ^ (oy + dyw)) | (oy ^ (oy + dyh)) | (oy ^ (oy + dyw + dyh))) >> (16 + shift) ||
// uses more than 16 bits of subpel mv (only at huge resolution)
(dxx | dxy | dyx | dyy) & 15 ||
(unsigned) ix >= width - w ||
(unsigned) iy >= height - h) {
// FIXME could still use mmx for some of the rows
ff_gmc_c(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy,
shift, r, width, height);
return;
}
src += ix + iy * stride;
__asm__ volatile (
"movd %0, %%mm6 \n\t"
"pxor %%mm7, %%mm7 \n\t"
"punpcklwd %%mm6, %%mm6 \n\t"
"punpcklwd %%mm6, %%mm6 \n\t"
:: "r" (1 << shift));
for (x = 0; x < w; x += 4) {
uint16_t dx4[4] = { oxs - dxys + dxxs * (x + 0),
oxs - dxys + dxxs * (x + 1),
oxs - dxys + dxxs * (x + 2),
oxs - dxys + dxxs * (x + 3) };
uint16_t dy4[4] = { oys - dyys + dyxs * (x + 0),
oys - dyys + dyxs * (x + 1),
oys - dyys + dyxs * (x + 2),
oys - dyys + dyxs * (x + 3) };
for (y = 0; y < h; y++) {
__asm__ volatile (
"movq %0, %%mm4 \n\t"
"movq %1, %%mm5 \n\t"
"paddw %2, %%mm4 \n\t"
"paddw %3, %%mm5 \n\t"
"movq %%mm4, %0 \n\t"
"movq %%mm5, %1 \n\t"
"psrlw $12, %%mm4 \n\t"
"psrlw $12, %%mm5 \n\t"
: "+m" (*dx4), "+m" (*dy4)
: "m" (*dxy4), "m" (*dyy4));
__asm__ volatile (
"movq %%mm6, %%mm2 \n\t"
"movq %%mm6, %%mm1 \n\t"
"psubw %%mm4, %%mm2 \n\t"
"psubw %%mm5, %%mm1 \n\t"
"movq %%mm2, %%mm0 \n\t"
"movq %%mm4, %%mm3 \n\t"
"pmullw %%mm1, %%mm0 \n\t" // (s - dx) * (s - dy)
"pmullw %%mm5, %%mm3 \n\t" // dx * dy
"pmullw %%mm5, %%mm2 \n\t" // (s - dx) * dy
"pmullw %%mm4, %%mm1 \n\t" // dx * (s - dy)
"movd %4, %%mm5 \n\t"
"movd %3, %%mm4 \n\t"
"punpcklbw %%mm7, %%mm5 \n\t"
"punpcklbw %%mm7, %%mm4 \n\t"
"pmullw %%mm5, %%mm3 \n\t" // src[1, 1] * dx * dy
"pmullw %%mm4, %%mm2 \n\t" // src[0, 1] * (s - dx) * dy
"movd %2, %%mm5 \n\t"
"movd %1, %%mm4 \n\t"
"punpcklbw %%mm7, %%mm5 \n\t"
"punpcklbw %%mm7, %%mm4 \n\t"
"pmullw %%mm5, %%mm1 \n\t" // src[1, 0] * dx * (s - dy)
"pmullw %%mm4, %%mm0 \n\t" // src[0, 0] * (s - dx) * (s - dy)
"paddw %5, %%mm1 \n\t"
"paddw %%mm3, %%mm2 \n\t"
"paddw %%mm1, %%mm0 \n\t"
"paddw %%mm2, %%mm0 \n\t"
"psrlw %6, %%mm0 \n\t"
"packuswb %%mm0, %%mm0 \n\t"
"movd %%mm0, %0 \n\t"
: "=m" (dst[x + y * stride])
: "m" (src[0]), "m" (src[1]),
"m" (src[stride]), "m" (src[stride + 1]),
"m" (*r4), "m" (shift2));
src += stride;
}
src += 4 - h * stride;
}
}
#endif /* HAVE_INLINE_ASM */
......@@ -41,9 +41,4 @@ void ff_put_signed_pixels_clamped_mmx(const int16_t *block, uint8_t *pixels,
void ff_draw_edges_mmx(uint8_t *buf, int wrap, int width, int height,
int w, int h, int sides);
void ff_gmc_mmx(uint8_t *dst, uint8_t *src,
int stride, int h, int ox, int oy,
int dxx, int dxy, int dyx, int dyy,
int shift, int r, int width, int height);
#endif /* AVCODEC_X86_DSPUTIL_X86_H */
/*
* This file is part of Libav.
*
* Libav 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.
*
* Libav 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 Libav; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
#include "config.h"
#include "libavutil/attributes.h"
#include "libavutil/cpu.h"
#include "libavutil/x86/asm.h"
#include "libavutil/x86/cpu.h"
#include "libavcodec/mpegvideodsp.h"
#if HAVE_INLINE_ASM
static void gmc_mmx(uint8_t *dst, uint8_t *src,
int stride, int h, int ox, int oy,
int dxx, int dxy, int dyx, int dyy,
int shift, int r, int width, int height)
{
const int w = 8;
const int ix = ox >> (16 + shift);
const int iy = oy >> (16 + shift);
const int oxs = ox >> 4;
const int oys = oy >> 4;
const int dxxs = dxx >> 4;
const int dxys = dxy >> 4;
const int dyxs = dyx >> 4;
const int dyys = dyy >> 4;
const uint16_t r4[4] = { r, r, r, r };
const uint16_t dxy4[4] = { dxys, dxys, dxys, dxys };
const uint16_t dyy4[4] = { dyys, dyys, dyys, dyys };
const uint64_t shift2 = 2 * shift;
int x, y;
const int dxw = (dxx - (1 << (16 + shift))) * (w - 1);
const int dyh = (dyy - (1 << (16 + shift))) * (h - 1);
const int dxh = dxy * (h - 1);
const int dyw = dyx * (w - 1);
if ( // non-constant fullpel offset (3% of blocks)
((ox ^ (ox + dxw)) | (ox ^ (ox + dxh)) | (ox ^ (ox + dxw + dxh)) |
(oy ^ (oy + dyw)) | (oy ^ (oy + dyh)) | (oy ^ (oy + dyw + dyh))) >> (16 + shift) ||
// uses more than 16 bits of subpel mv (only at huge resolution)
(dxx | dxy | dyx | dyy) & 15 ||
(unsigned) ix >= width - w ||
(unsigned) iy >= height - h) {
// FIXME could still use mmx for some of the rows
ff_gmc_c(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy,
shift, r, width, height);
return;
}
src += ix + iy * stride;
__asm__ volatile (
"movd %0, %%mm6 \n\t"
"pxor %%mm7, %%mm7 \n\t"
"punpcklwd %%mm6, %%mm6 \n\t"
"punpcklwd %%mm6, %%mm6 \n\t"
:: "r" (1 << shift));
for (x = 0; x < w; x += 4) {
uint16_t dx4[4] = { oxs - dxys + dxxs * (x + 0),
oxs - dxys + dxxs * (x + 1),
oxs - dxys + dxxs * (x + 2),
oxs - dxys + dxxs * (x + 3) };
uint16_t dy4[4] = { oys - dyys + dyxs * (x + 0),
oys - dyys + dyxs * (x + 1),
oys - dyys + dyxs * (x + 2),
oys - dyys + dyxs * (x + 3) };
for (y = 0; y < h; y++) {
__asm__ volatile (
"movq %0, %%mm4 \n\t"
"movq %1, %%mm5 \n\t"
"paddw %2, %%mm4 \n\t"
"paddw %3, %%mm5 \n\t"
"movq %%mm4, %0 \n\t"
"movq %%mm5, %1 \n\t"
"psrlw $12, %%mm4 \n\t"
"psrlw $12, %%mm5 \n\t"
: "+m" (*dx4), "+m" (*dy4)
: "m" (*dxy4), "m" (*dyy4));
__asm__ volatile (
"movq %%mm6, %%mm2 \n\t"
"movq %%mm6, %%mm1 \n\t"
"psubw %%mm4, %%mm2 \n\t"
"psubw %%mm5, %%mm1 \n\t"
"movq %%mm2, %%mm0 \n\t"
"movq %%mm4, %%mm3 \n\t"
"pmullw %%mm1, %%mm0 \n\t" // (s - dx) * (s - dy)
"pmullw %%mm5, %%mm3 \n\t" // dx * dy
"pmullw %%mm5, %%mm2 \n\t" // (s - dx) * dy
"pmullw %%mm4, %%mm1 \n\t" // dx * (s - dy)
"movd %4, %%mm5 \n\t"
"movd %3, %%mm4 \n\t"
"punpcklbw %%mm7, %%mm5 \n\t"
"punpcklbw %%mm7, %%mm4 \n\t"
"pmullw %%mm5, %%mm3 \n\t" // src[1, 1] * dx * dy
"pmullw %%mm4, %%mm2 \n\t" // src[0, 1] * (s - dx) * dy
"movd %2, %%mm5 \n\t"
"movd %1, %%mm4 \n\t"
"punpcklbw %%mm7, %%mm5 \n\t"
"punpcklbw %%mm7, %%mm4 \n\t"
"pmullw %%mm5, %%mm1 \n\t" // src[1, 0] * dx * (s - dy)
"pmullw %%mm4, %%mm0 \n\t" // src[0, 0] * (s - dx) * (s - dy)
"paddw %5, %%mm1 \n\t"
"paddw %%mm3, %%mm2 \n\t"
"paddw %%mm1, %%mm0 \n\t"
"paddw %%mm2, %%mm0 \n\t"
"psrlw %6, %%mm0 \n\t"
"packuswb %%mm0, %%mm0 \n\t"
"movd %%mm0, %0 \n\t"
: "=m" (dst[x + y * stride])
: "m" (src[0]), "m" (src[1]),
"m" (src[stride]), "m" (src[stride + 1]),
"m" (*r4), "m" (shift2));
src += stride;
}
src += 4 - h * stride;
}
}
#endif /* HAVE_INLINE_ASM */
av_cold void ff_mpegvideodsp_init_x86(MpegVideoDSPContext *c)
{
#if HAVE_INLINE_ASM
int cpu_flags = av_get_cpu_flags();
if (INLINE_MMX(cpu_flags))
c->gmc = gmc_mmx;
#endif /* HAVE_INLINE_ASM */
}
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment