jsuya
/
efl
forked from enlightenment/efl
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

partial swaps++ to evas gl engine. 

SVN revision: 77069
This commit is contained in:
Carsten Haitzler 2012-09-26 03:49:33 +00:00
parent fd794e2a54
commit 53f2bc0638
4 changed files with 250 additions and 138 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
legacy/evas/ChangeLog
View File

@ -1052,8 +1052,15 @@
2012-09-24 Sung W. Park (sung_)
* Refactored Evas GL engine code so wayland_egl and gl_x11 and other
engines can share the same code. The common codes are in gl_common/
directory and evas_engine just has to implement a few engine functions.
* Refactored Evas GL engine code so wayland_egl and gl_x11 and other
engines can share the same code. The common codes are in gl_common/
directory and evas_engine just has to implement a few engine functions.
2012-09-26 Carsten Haitzler (The Rasterman)
* Add the ability via env vars to do partial swaps in the gl
engine. This requires that we also can get from gl info as to
if a swap was a copy, swap, discard and if a swap with 2 or 3
buffers. don't have that but env var will do for now for testing.

1
legacy/evas/NEWS
View File

@ -13,6 +13,7 @@ Improvements:
* Function to rotate an evas map with a quaternion: evas_map_util_quat_rotate().
* EVAS_GL_NO_BLACKLIST env var to turn off blacklisted drivers in gl
* Evas gl enigne can do partial swaps now.
Fixes:

23
legacy/evas/src/modules/engines/gl_common/evas_gl_context.c
View File

@ -2212,16 +2212,27 @@ evas_gl_common_context_image_map_push(Evas_Engine_GL_Context *gc,
w = w - x;
h = h - y;
if (!flat)
{
// FUZZZZ!
x -= 3;
y -= 3;
w += 6;
h += 6;
}
if (clip)
{
int nx = x, ny = y, nw = w, nh = h;
RECTS_CLIP_TO_RECT(nx, ny, nw, nh, cx, cy, cw, ch);
if ((nx == x) && (ny == y) && (nw == w) && (nh == h))
if (flat)
{
clip = 0; cx = 0; cy = 0; cw = 0; ch = 0;
int nx = x, ny = y, nw = w, nh = h;
RECTS_CLIP_TO_RECT(nx, ny, nw, nh, cx, cy, cw, ch);
if ((nx == x) && (ny == y) && (nw == w) && (nh == h))
{
clip = 0; cx = 0; cy = 0; cw = 0; ch = 0;
}
x = nx; y = ny; w = nw; h = nh;
}
x = nx; y = ny; w = nw; h = nh;
}
if (!flat)

351
legacy/evas/src/modules/engines/gl_x11/evas_engine.c
View File

@ -11,19 +11,31 @@
#define EVAS_GL_NO_GL_H_CHECK 1
#include "Evas_GL.h"
enum {
MODE_FULL,
MODE_COPY,
MODE_DOUBLE,
MODE_TRIPLE
};
typedef struct _Render_Engine Render_Engine;
struct _Render_Engine
{
Tilebuf_Rect *rects;
Tilebuf_Rect *rects_prev[3];
Eina_Inlist *cur_rect;
Evas_GL_X11_Window *win;
Evas_Engine_Info_GL_X11 *info;
Evas *evas;
Tilebuf *tb;
int end;
int w, h;
int vsync;
int mode;
int w, h;
int vsync;
EVGL_Engine *evgl_engine;
EVGL_Engine *evgl_engine;
};
static int initted = 0;
@ -627,6 +639,7 @@ eng_setup(Evas *e, void *in)
{
Render_Engine *re;
Evas_Engine_Info_GL_X11 *info;
const char *s;
info = (Evas_Engine_Info_GL_X11 *)in;
if (!e->engine.data.output)
@ -740,6 +753,36 @@ eng_setup(Evas *e, void *in)
}
}
}
if ((s = getenv("EVAS_GL_SWAP_MODE")))
{
if ((!strcasecmp(s, "full")) ||
(!strcasecmp(s, "f")))
re->mode = MODE_FULL;
else if ((!strcasecmp(s, "copy")) ||
(!strcasecmp(s, "c")))
re->mode = MODE_COPY;
else if ((!strcasecmp(s, "double")) ||
(!strcasecmp(s, "d")) ||
(!strcasecmp(s, "2")))
re->mode = MODE_DOUBLE;
else if ((!strcasecmp(s, "triple")) ||
(!strcasecmp(s, "t")) ||
(!strcasecmp(s, "3")))
re->mode = MODE_TRIPLE;
}
else
{
// in most gl implementations - egl and glx here that we care about the TEND
// to either swap or copy backbuffer and front buffer, but strictly that is
// not true. technically backbuffer content is totally undefined after a swap
// and thus you MUST re-render all of it, thus MODE_FULL
// re->mode = MODE_FULL;
// BUT... reality is that lmost every implementation copies or swaps so
// triple buffer mode can be used as it is a superset of double buffer and
// copy (though using those explicitly is more efficient). so let's play with
// triple buffer mdoe as a default and see.
re->mode = MODE_TRIPLE;
}
if (!re->win)
{
free(re);
@ -815,6 +858,10 @@ eng_output_free(void *data)
evgl_engine_destroy(re->evgl_engine);
}
evas_common_tilebuf_free(re->tb);
if (re->rects) evas_common_tilebuf_free_render_rects(re->rects);
if (re->rects_prev[0]) evas_common_tilebuf_free_render_rects(re->rects_prev[0]);
if (re->rects_prev[1]) evas_common_tilebuf_free_render_rects(re->rects_prev[1]);
if (re->rects_prev[2]) evas_common_tilebuf_free_render_rects(re->rects_prev[2]);
free(re);
}
if ((initted == 1) && (gl_wins == 0))
@ -860,6 +907,7 @@ eng_output_redraws_rect_add(void *data, int x, int y, int w, int h)
evas_gl_common_context_resize(re->win->gl_context, re->win->w, re->win->h, re->win->rot);
evas_common_tilebuf_add_redraw(re->tb, x, y, w, h);
/* bounding box track */
RECTS_CLIP_TO_RECT(x, y, w, h, 0, 0, re->win->w, re->win->h);
if ((w <= 0) || (h <= 0)) return;
if (!re->win->draw.redraw)
@ -906,6 +954,64 @@ eng_output_redraws_clear(void *data)
// INF("GL: finish update cycle!");
}
static Tilebuf_Rect *
_merge_rects(Tilebuf *tb, Tilebuf_Rect *r1, Tilebuf_Rect *r2, Tilebuf_Rect *r3)
{
Tilebuf_Rect *r, *rects;
int x1, y1, x2, y2;
if (r1)
{
EINA_INLIST_FOREACH(EINA_INLIST_GET(r1), r)
{
evas_common_tilebuf_add_redraw(tb, r->x, r->y, r->w, r->h);
}
}
if (r2)
{
EINA_INLIST_FOREACH(EINA_INLIST_GET(r2), r)
{
evas_common_tilebuf_add_redraw(tb, r->x, r->y, r->w, r->h);
}
}
if (r2)
{
EINA_INLIST_FOREACH(EINA_INLIST_GET(r3), r)
{
evas_common_tilebuf_add_redraw(tb, r->x, r->y, r->w, r->h);
}
}
rects = evas_common_tilebuf_get_render_rects(tb);
// bounding box -> make a bounding box single region update of all regions.
// yes we could try and be smart and figure out size of regions, how far
// apart etc. etc. to try and figure out an optimal "set". this is a tradeoff
// between multiple update regions to render and total pixels to render.
if (rects)
{
x1 = rects->x; y1 = rects->y;
x2 = rects->x + rects->w; y2 = rects->y + rects->h;
EINA_INLIST_FOREACH(EINA_INLIST_GET(rects), r)
{
if (r->x < x1) x1 = r->x;
if (r->y < y1) y1 = r->y;
if ((r->x + r->w) > x2) x2 = r->x + r->w;
if ((r->y + r->h) > y2) y2 = r->y + r->h;
}
evas_common_tilebuf_free_render_rects(rects);
rects = calloc(1, sizeof(Tilebuf_Rect));
if (rects)
{
rects->x = x1;
rects->y = y1;
rects->w = x2 - x1;
rects->h = y2 - y1;
}
}
evas_common_tilebuf_clear(tb);
return rects;
}
/* vsync games - not for now though */
#define VSYNC_TO_SCREEN 1
@ -913,90 +1019,126 @@ static void *
eng_output_redraws_next_update_get(void *data, int *x, int *y, int *w, int *h, int *cx, int *cy, int *cw, int *ch)
{
Render_Engine *re;
Tilebuf_Rect *rects;
Tilebuf_Rect *rect;
int i;
Eina_Bool first_rect = EINA_FALSE;
#define CLEAR_PREV_RECTS(x) \
do { \
if (re->rects_prev[i]) \
evas_common_tilebuf_free_render_rects(re->rects_prev[i]); \
re->rects_prev[i] = NULL; \
} while (0)
re = (Render_Engine *)data;
/* get the upate rect surface - return engine data as dummy */
rects = evas_common_tilebuf_get_render_rects(re->tb);
if (rects)
if (re->end)
{
/*
Tilebuf_Rect *r;
printf("REAAAAACCTS\n");
EINA_INLIST_FOREACH(EINA_INLIST_GET(rects), r)
{
printf(" %i %i %ix%i\n", r->x, r->y, r->w, r->h);
}
*/
evas_common_tilebuf_free_render_rects(rects);
re->end = 0;
return NULL;
}
if (!re->rects)
{
re->rects = evas_common_tilebuf_get_render_rects(re->tb);
evas_common_tilebuf_clear(re->tb);
if (re->rects)
{
/* ensure we get rid of previous rect lists we dont need if mode
* changed/is appropriate */
switch (re->mode)
{
case MODE_FULL:
case MODE_COPY: // no prev rects needed
for (i = 0; i < 3; i++) CLEAR_PREV_RECTS(i);
break;
case MODE_DOUBLE: // double mode - only 1 level of prev rect
for (i = 1; i < 3; i++) CLEAR_PREV_RECTS(i);
re->rects_prev[1] = re->rects_prev[0];
re->rects_prev[0] = re->rects;
// merge prev[1] + prev[0] -> rects
re->rects = _merge_rects(re->tb, re->rects_prev[0], re->rects_prev[1], NULL);
break;
case MODE_TRIPLE: // keep all
for (i = 2; i < 3; i++) CLEAR_PREV_RECTS(i);
re->rects_prev[2] = re->rects_prev[1];
re->rects_prev[1] = re->rects_prev[0];
re->rects_prev[0] = re->rects;
re->rects = NULL;
// merge prev[2] + prev[1] + prev[0] -> rects
re->rects = _merge_rects(re->tb, re->rects_prev[0], re->rects_prev[1], re->rects_prev[2]);
break;
default:
break;
}
re->cur_rect = EINA_INLIST_GET(re->rects);
first_rect = EINA_TRUE;
}
}
if (!re->cur_rect) return NULL;
rect = (Tilebuf_Rect *)re->cur_rect;
if (re->rects)
{
switch (re->mode)
{
case MODE_COPY:
case MODE_DOUBLE:
case MODE_TRIPLE:
rect = (Tilebuf_Rect *)re->cur_rect;
*x = rect->x;
*y = rect->y;
*w = rect->w;
*h = rect->h;
*cx = rect->x;
*cy = rect->y;
*cw = rect->w;
*ch = rect->h;
re->cur_rect = re->cur_rect->next;
break;
case MODE_FULL:
re->cur_rect = NULL;
if (x) *x = 0;
if (y) *y = 0;
if (w) *w = re->win->w;
if (h) *h = re->win->h;
if (cx) *cx = 0;
if (cy) *cy = 0;
if (cw) *cw = re->win->w;
if (ch) *ch = re->win->h;
break;
default:
break;
}
if (first_rect)
{
#ifdef GL_GLES
// dont need to for egl - eng_window_use() can check for other ctxt's
// dont need to for egl - eng_window_use() can check for other ctxt's
#else
eng_window_use(NULL);
eng_window_use(NULL);
#endif
eng_window_use(re->win);
if (!_re_wincheck(re)) return NULL;
evas_gl_common_context_flush(re->win->gl_context);
evas_gl_common_context_newframe(re->win->gl_context);
if (x) *x = 0;
if (y) *y = 0;
if (w) *w = re->win->w;
if (h) *h = re->win->h;
if (cx) *cx = 0;
if (cy) *cy = 0;
if (cw) *cw = re->win->w;
if (ch) *ch = re->win->h;
eng_window_use(re->win);
if (!_re_wincheck(re)) return NULL;
evas_gl_common_context_flush(re->win->gl_context);
evas_gl_common_context_newframe(re->win->gl_context);
//// debug partial updates :)
// glClearColor(1.0, 0.5, 0.2, 1.0);
// glClear(GL_COLOR_BUFFER_BIT);
}
if (!re->cur_rect)
{
re->end = 1;
}
return re->win->gl_context->def_surface;
}
return NULL;
/*
if (!re->win->draw.redraw) return NULL;
#ifdef GL_GLES
// dont need to for egl - eng_window_use() can check for other ctxt's
#else
eng_window_use(NULL);
#endif
eng_window_use(re->win);
if (!_re_wincheck(re)) return NULL;
evas_gl_common_context_flush(re->win->gl_context);
evas_gl_common_context_newframe(re->win->gl_context);
if (x) *x = re->win->draw.x1;
if (y) *y = re->win->draw.y1;
if (w) *w = re->win->draw.x2 - re->win->draw.x1 + 1;
if (h) *h = re->win->draw.y2 - re->win->draw.y1 + 1;
if (cx) *cx = re->win->draw.x1;
if (cy) *cy = re->win->draw.y1;
if (cw) *cw = re->win->draw.x2 - re->win->draw.x1 + 1;
if (ch) *ch = re->win->draw.y2 - re->win->draw.y1 + 1;
return re->win->gl_context->def_surface;
*/
}
//#define FRAMECOUNT 1
#ifdef FRAMECOUNT
static double
get_time(void)
{
struct timeval timev;
gettimeofday(&timev, NULL);
return (double)timev.tv_sec + (((double)timev.tv_usec) / 1000000);
}
#endif
static int safe_native = -1;
static void
eng_output_redraws_next_update_push(void *data, void *surface __UNUSED__, int x __UNUSED__, int y __UNUSED__, int w __UNUSED__, int h __UNUSED__)
{
Render_Engine *re;
#ifdef FRAMECOUNT
static double pt = 0.0;
double ta, tb;
#endif
re = (Render_Engine *)data;
/* put back update surface.. in this case just unflag redraw */
@ -1023,18 +1165,8 @@ eng_output_redraws_next_update_push(void *data, void *surface __UNUSED__, int x
#ifdef GL_GLES
// this is needed to make sure all previous rendering is flushed to
// buffers/surfaces
#ifdef FRAMECOUNT
double t0 = get_time();
ta = t0 - pt;
pt = t0;
#endif
// previous rendering should be done and swapped
if (!safe_native) eglWaitNative(EGL_CORE_NATIVE_ENGINE);
#ifdef FRAMECOUNT
double t1 = get_time();
tb = t1 - t0;
printf("... %1.5f -> %1.5f | ", ta, tb);
#endif
// if (eglGetError() != EGL_SUCCESS)
// {
// printf("Error: eglWaitNative(EGL_CORE_NATIVE_ENGINE) fail.\n");
@ -1043,7 +1175,6 @@ eng_output_redraws_next_update_push(void *data, void *surface __UNUSED__, int x
// previous rendering should be done and swapped
if (!safe_native) glXWaitX();
#endif
//x// printf("frame -> push\n");
}
static void
@ -1054,14 +1185,11 @@ eng_output_flush(void *data)
re = (Render_Engine *)data;
if (!_re_wincheck(re)) return;
if (!re->win->draw.drew) return;
//x// printf("frame -> flush\n");
re->win->draw.drew = 0;
eng_window_use(re->win);
#ifdef GL_GLES
#ifdef FRAMECOUNT
double t0 = get_time();
#endif
if (!re->vsync)
{
if (re->info->vsync) eglSwapInterval(re->win->egl_disp, 1);
@ -1072,23 +1200,20 @@ eng_output_flush(void *data)
{
re->info->callback.pre_swap(re->info->callback.data, re->evas);
}
// XXX: if partial swaps can be done use re->rects
eglSwapBuffers(re->win->egl_disp, re->win->egl_surface[0]);
if (!safe_native) eglWaitGL();
if (re->info->callback.post_swap)
{
re->info->callback.post_swap(re->info->callback.data, re->evas);
}
#ifdef FRAMECOUNT
double t1 = get_time();
printf("%1.5f\n", t1 - t0);
#endif
// if (eglGetError() != EGL_SUCCESS)
// {
// printf("Error: eglSwapBuffers() fail.\n");
// }
#else
#ifdef VSYNC_TO_SCREEN
if ((re->info->vsync)/* || (1)*/)
if (re->info->vsync)
{
if (glsym_glXSwapIntervalEXT)
{
@ -1119,56 +1244,24 @@ eng_output_flush(void *data)
}
}
}
# endif
#endif
if (re->info->callback.pre_swap)
{
re->info->callback.pre_swap(re->info->callback.data, re->evas);
}
#if 1
if (1)
#else
if ((re->win->draw.x1 == 0) && (re->win->draw.y1 == 0) && (re->win->draw.x2 == (re->win->w - 1)) && (re->win->draw.y2 == (re->win->h - 1)))
#endif
{
// double t, t2 = 0.0;
// t = get_time();
glXSwapBuffers(re->win->disp, re->win->win);
// t = get_time() - t;
// if (!safe_native)
// {
// t2 = get_time();
// glXWaitGL();
// t2 = get_time() - t2;
// }
// printf("swap: %3.5f (%3.5fms), x wait gl: %3.5f (%3.5fms)\n",
// t, t * 1000.0, t2, t2 * 1000.0);
}
else
{
// FIXME: this doesn't work.. why oh why?
int sx, sy, sw, sh;
sx = re->win->draw.x1;
sy = re->win->draw.y1;
sw = (re->win->draw.x2 - re->win->draw.x1) + 1;
sh = (re->win->draw.y2 - re->win->draw.y1) + 1;
sy = re->win->h - sy - sh;
glBitmap(0, 0, 0, 0, sx, re->win->h - sy, NULL);
glEnable(GL_SCISSOR_TEST);
glScissor(sx, sy, sw, sh);
glDrawBuffer(GL_FRONT);
glCopyPixels(sx, sy, sw, sh, GL_COLOR);
glDrawBuffer(GL_BACK);
glDisable(GL_SCISSOR_TEST);
glBitmap(0, 0, 0, 0, 0, 0, NULL);
glFlush();
}
// XXX: if partial swaps can be done use re->rects
glXSwapBuffers(re->win->disp, re->win->win);
if (re->info->callback.post_swap)
{
re->info->callback.post_swap(re->info->callback.data, re->evas);
}
#endif
// clear out rects after swap as we may use them during swap
if (re->rects)
{
evas_common_tilebuf_free_render_rects(re->rects);
re->rects = NULL;
}
}
static void