This is an automated email from the git hooks/post-receive script.
git pushed a commit to branch evas_gl_buffer_optimization
in repository efl.
View the commit online.
commit a46f547aa5fdcb823ee5a7ae9fcfe24a420f860c
Author: Carsten Haitzler <ras...@rasterman.com>
AuthorDate: Tue Jan 10 22:23:46 2017 +0900
evas gl engine - add rolling per frame vbo buff pool for glmapbuffer
this adds a buffer pool for glmapbuffer over 6 frames with a pool per
frame of buffers to re-use for holding vertices, colors and more.
---
.../evas/engines/gl_common/evas_gl_common.h | 29 ++-
.../evas/engines/gl_common/evas_gl_context.c | 200 ++++++++++++++++++---
src/modules/evas/engines/gl_x11/evas_engine.c | 4 +
3 files changed, 207 insertions(+), 26 deletions(-)
diff --git a/src/modules/evas/engines/gl_common/evas_gl_common.h b/src/modules/evas/engines/gl_common/evas_gl_common.h
index 55f5ab95cc..3440a61bdb 100644
--- a/src/modules/evas/engines/gl_common/evas_gl_common.h
+++ b/src/modules/evas/engines/gl_common/evas_gl_common.h
@@ -246,12 +246,22 @@ enum _Shader_Type {
#define ARRAY_BUFFER_USE 500
#define ARRAY_BUFFER_USE_SHIFT 100
+typedef struct _Evas_GL_Vbo Evas_GL_Vbo;
+
+struct _Evas_GL_Vbo
+{
+ GLuint buffer;
+ unsigned int size;
+ unsigned int unused;
+};
+
struct _Evas_Engine_GL_Context
{
int references;
int w, h;
int rot;
int foc, z0, px, py;
+ unsigned int frame;
RGBA_Draw_Context *dc;
Evas_GL_Shared *shared;
@@ -271,7 +281,7 @@ struct _Evas_Engine_GL_Context
Eina_Bool anti_alias : 1;
} current;
} state;
-
+
struct {
int x, y, w, h;
Eina_Bool enabled : 1;
@@ -307,6 +317,8 @@ struct _Evas_Engine_GL_Context
GLfloat *texsam;
GLfloat *mask;
GLfloat *masksam;
+ Evas_GL_Image *im;
+ GLuint buffer;
Eina_Bool line: 1;
Eina_Bool use_vertex : 1; // always true
Eina_Bool use_color : 1;
@@ -318,17 +330,21 @@ struct _Evas_Engine_GL_Context
Eina_Bool use_mask : 1;
Eina_Bool use_masksam : 1;
Eina_Bool anti_alias : 1;
- Evas_GL_Image *im;
- GLuint buffer;
- int buffer_alloc;
- int buffer_use;
} array;
} pipe[MAX_PIPES];
struct {
Eina_Bool size : 1;
} change;
-
+
+#define VBO_POOL_SLOTS 6
+ struct {
+ struct {
+ Eina_List *spare;
+ Eina_List *used;
+ } pool[VBO_POOL_SLOTS];
+ } vbos;
+
Eina_List *font_glyph_textures;
Eina_Bool havestuff : 1;
@@ -519,6 +535,7 @@ EAPI Eina_Bool evas_gl_preload_enabled(void);
EAPI Evas_Engine_GL_Context *evas_gl_common_context_new(void);
EAPI void evas_gl_common_context_flush(Evas_Engine_GL_Context *gc);
+EAPI void evas_gl_common_context_idle_flush(Evas_Engine_GL_Context *gc);
EAPI void evas_gl_common_context_free(Evas_Engine_GL_Context *gc);
EAPI void evas_gl_common_context_use(Evas_Engine_GL_Context *gc);
EAPI void evas_gl_common_context_newframe(Evas_Engine_GL_Context *gc);
diff --git a/src/modules/evas/engines/gl_common/evas_gl_context.c b/src/modules/evas/engines/gl_common/evas_gl_context.c
index 470b4f7e81..4a446abbcb 100644
--- a/src/modules/evas/engines/gl_common/evas_gl_context.c
+++ b/src/modules/evas/engines/gl_common/evas_gl_context.c
@@ -273,7 +273,7 @@ evas_gl_symbols(void *(*GetProcAddress)(const char *name))
FINDSYM(glsym_glStartTiling, "glActivateTileQCOM", NULL, glsym_func_void);
FINDSYM(glsym_glEndTiling, "glEndTilingQCOM", "GL_QCOM_tiled_rendering", glsym_func_void);
FINDSYM(glsym_glEndTiling, "glEndTiling", NULL, glsym_func_void);
-
+
if (!getenv("EVAS_GL_MAPBUFFER_DISABLE"))
{
// Not sure there's an EXT variant. (probably no KHR variant)
@@ -773,6 +773,141 @@ _evas_gl_common_viewport_set(Evas_Engine_GL_Context *gc)
}
}
+///////////////////////////////////////////////////////////////////////////
+
+// debugging just for the fbo pool/cache
+#if 0
+# define VD(...) printf(__VA_ARGS__)
+#else
+# define VD(...)
+#endif
+
+// a vbo bigger than MAX_VBO_SIZE is stupid and likely an error so dont
+// even try
+#define MAX_VBO_SIZE (0x7fffffff)
+// if the final vbo we found is more than MAX_VBO_DIFF bigger than what
+// we need then let's not use it and alloc a new vbo anyway
+#define MAX_VBO_DIFF (128 * 1024)
+// if the buffer is up to MIN_VBO_DIFF buffer than what we want just
+// take it and don't look for a better match as this just costs us
+// time hunting rfor a better one and since this is O(n) why bother
+// with that effort when near enough is good enough
+#define MIN_VBO_DIFF (512)
+
+static void
+_vbo_pool_vbo_del(Evas_GL_Vbo *vbo)
+{
+ glDeleteBuffers(1, &(vbo->buffer));
+ free(vbo);
+}
+
+static void
+_vbo_pool_clear(Evas_Engine_GL_Context *gc)
+{
+ int i;
+ Evas_GL_Vbo *vbo;
+
+ if (!(glsym_glMapBuffer && glsym_glUnmapBuffer)) return;
+ for (i = 0; i < VBO_POOL_SLOTS; i++)
+ {
+ EINA_LIST_FREE(gc->vbos.pool[i].spare, vbo)
+ {
+ VD("FREE UP SPARE buf=%i size=%i\n", vbo->buffer, vbo->size);
+ _vbo_pool_vbo_del(vbo);
+ }
+ EINA_LIST_FREE(gc->vbos.pool[i].used, vbo)
+ {
+ VD("FREE UP USED buf=%i size=%i\n", vbo->buffer, vbo->size);
+ _vbo_pool_vbo_del(vbo);
+ }
+ }
+}
+
+static void
+_vbo_pool_newframe(Evas_Engine_GL_Context *gc)
+{
+ int slot;
+ Eina_List *l, *ll;
+ Evas_GL_Vbo *vbo;
+
+ if (!(glsym_glMapBuffer && glsym_glUnmapBuffer)) return;
+ VD("-------------------------------- FRAME %i\n", gc->frame + 1);
+ slot = gc->frame % VBO_POOL_SLOTS;
+ EINA_LIST_FOREACH_SAFE(gc->vbos.pool[slot].spare, l, ll, vbo)
+ {
+ vbo->unused++;
+ if (vbo->unused >= 2)
+ {
+ VD("DEL UNUSED buf=%i size=%i\n", vbo->buffer, vbo->size);
+ _vbo_pool_vbo_del(vbo);
+ gc->vbos.pool[slot].spare =
+ eina_list_remove_list(gc->vbos.pool[slot].spare, l);
+ }
+ }
+ EINA_LIST_FREE(gc->vbos.pool[slot].used, vbo)
+ {
+ gc->vbos.pool[slot].spare =
+ eina_list_prepend(gc->vbos.pool[slot].spare, vbo);
+ }
+}
+
+static GLuint
+_vbo_pool_buffer_find(Evas_Engine_GL_Context *gc, size_t size)
+{
+ Eina_List *l, *l_sel = NULL;
+ Evas_GL_Vbo *vbo, *vbo_sel = NULL;
+ int slot;
+ unsigned int diff, diff_sel = MAX_VBO_SIZE;
+ int hunt = 0;
+
+ if (size > MAX_VBO_SIZE) return 0; // stupidly too big buffer
+ // look up in a specifc buffer pool that is a ring buffer of the last
+ // N frames (one slot with a pool per frame) where N is VBO_POOL_SLOTS
+ slot = gc->frame % VBO_POOL_SLOTS;
+ EINA_LIST_FOREACH(gc->vbos.pool[slot].spare, l, vbo)
+ {
+ hunt++;
+ if (vbo->size < size) continue;
+ diff = size - vbo->size;
+ if (diff < diff_sel)
+ {
+ vbo_sel = vbo;
+ diff_sel = diff;
+ l_sel = l;
+ if (diff < MIN_VBO_DIFF) break;
+ }
+ }
+ VD("FIND SIZE=%i from #=%i/%i - ", (int)size, hunt, eina_list_count(gc->vbos.pool[slot].spare));
+ if ((vbo_sel) && (diff_sel > MAX_VBO_DIFF))
+ {
+ VD("[BIG %i] ", diff_sel);
+ vbo_sel = NULL;
+ }
+ if (vbo_sel)
+ {
+ vbo = vbo_sel;
+ VD("FOUND buf=%i size=%i unused=%i\n", vbo->buffer, vbo->size, vbo->unused);
+ vbo->unused = 0;;
+ gc->vbos.pool[slot].spare =
+ eina_list_remove_list(gc->vbos.pool[slot].spare, l_sel);
+ gc->vbos.pool[slot].used =
+ eina_list_append(gc->vbos.pool[slot].used, vbo);
+ return vbo->buffer;
+ }
+ vbo = calloc(1, sizeof(Evas_GL_Vbo));
+ if (!vbo) return 0;
+ glGenBuffers(1, &(vbo->buffer));
+ glBindBuffer(GL_ARRAY_BUFFER, vbo->buffer);
+ glBufferData(GL_ARRAY_BUFFER, (long)size, NULL, GL_DYNAMIC_DRAW);
+ vbo->size = size;
+ gc->vbos.pool[slot].used =
+ eina_list_append(gc->vbos.pool[slot].used, vbo);
+ VD("** ALLOC ** buf=%i size=%i\n", vbo->buffer, vbo->size);
+ return vbo->buffer;
+}
+
+///////////////////////////////////////////////////////////////////////////
+
EAPI Evas_Engine_GL_Context *
evas_gl_common_context_new(void)
{
@@ -806,12 +941,12 @@ evas_gl_common_context_new(void)
for (i = 0; i < MAX_PIPES; i++)
{
gc->pipe[i].shader.render_op = EVAS_RENDER_BLEND;
- if (glsym_glMapBuffer && glsym_glUnmapBuffer)
- {
- glGenBuffers(1, &gc->pipe[i].array.buffer);
- gc->pipe[i].array.buffer_alloc = 0;
- gc->pipe[i].array.buffer_use = 0;
- }
+// if (glsym_glMapBuffer && glsym_glUnmapBuffer)
+// {
+// glGenBuffers(1, &gc->pipe[i].array.buffer);
+// gc->pipe[i].array.buffer_alloc = 0;
+// gc->pipe[i].array.buffer_use = 0;
+// }
}
gc->state.current.tex_target = GL_TEXTURE_2D;
@@ -930,11 +1065,13 @@ evas_gl_common_context_new(void)
else if (strstr(s, "NVIDIA Tegra"))
shared->info.tune.pipes.max = DEF_PIPES_TEGRA_2;
}
+#if 1
if (!getenv("EVAS_GL_MAPBUFFER"))
{
glsym_glMapBuffer = NULL;
glsym_glUnmapBuffer= NULL;
}
+#endif
#define GETENVOPT(name, tune_param, min, max) \
do { \
@@ -1094,6 +1231,12 @@ error:
return NULL;
}
+EAPI void
+evas_gl_common_context_idle_flush(Evas_Engine_GL_Context *gc)
+{
+ _vbo_pool_clear(gc);
+}
+
EAPI void
evas_gl_common_context_free(Evas_Engine_GL_Context *gc)
{
@@ -1109,11 +1252,12 @@ evas_gl_common_context_free(Evas_Engine_GL_Context *gc)
if (gc->font_surface)
evas_cache_image_drop(&gc->font_surface->cache_entry);
- if (glsym_glMapBuffer && glsym_glUnmapBuffer)
- {
- for (i = 0; i < MAX_PIPES; i++)
- glDeleteBuffers(1, &gc->pipe[i].array.buffer);
- }
+ _vbo_pool_clear(gc);
+// if (glsym_glMapBuffer && glsym_glUnmapBuffer)
+// {
+// for (i = 0; i < MAX_PIPES; i++)
+// glDeleteBuffers(1, &gc->pipe[i].array.buffer);
+// }
if (gc->shared)
{
@@ -1186,6 +1330,8 @@ evas_gl_common_context_newframe(Evas_Engine_GL_Context *gc)
{
int i;
+ _vbo_pool_newframe(gc);
+ gc->frame++;
if (_evas_gl_common_cutout_rects)
{
evas_common_draw_context_apply_clear_cutouts(_evas_gl_common_cutout_rects);
@@ -3313,6 +3459,7 @@ shader_array_flush(Evas_Engine_GL_Context *gc)
if (glsym_glMapBuffer && glsym_glUnmapBuffer)
{
unsigned char *x;
+ unsigned int pad = 0;
# define VERTEX_SIZE (gc->pipe[i].array.num * sizeof(GLshort) * VERTEX_CNT)
# define COLOR_SIZE (gc->pipe[i].array.num * sizeof(GLubyte) * COLOR_CNT)
@@ -3320,7 +3467,11 @@ shader_array_flush(Evas_Engine_GL_Context *gc)
# define SAM_SIZE (gc->pipe[i].array.num * sizeof(GLfloat) * SAM_CNT)
# define MASK_SIZE (gc->pipe[i].array.num * sizeof(GLfloat) * MASK_CNT)
vertex_ptr = NULL;
- color_ptr = vertex_ptr + VERTEX_SIZE;
+ // purely for correctness handle alignment. since we do
+ // quads which is 6 points this ends up not needed, but just
+ // in case in future...
+ if (gc->pipe[i].array.num & 1) pad = 2;
+ color_ptr = vertex_ptr + VERTEX_SIZE + pad;
texuv_ptr = color_ptr + COLOR_SIZE;
texuv2_ptr = texuv_ptr + TEX_SIZE;
texuv3_ptr = texuv2_ptr + TEX_SIZE;
@@ -3330,15 +3481,23 @@ shader_array_flush(Evas_Engine_GL_Context *gc)
masksam_ptr = mask_ptr + MASK_SIZE;
# define END_POINTER (masksam_ptr + SAM_SIZE)
- glBindBuffer(GL_ARRAY_BUFFER, gc->pipe[i].array.buffer);
- if ((gc->pipe[i].array.buffer_alloc < (long)END_POINTER) ||
- (gc->pipe[i].array.buffer_use >= (ARRAY_BUFFER_USE + ARRAY_BUFFER_USE_SHIFT * i)))
+ if (gc->pipe[i].array.buffer == 0)
{
- glBufferData(GL_ARRAY_BUFFER, (long)END_POINTER, NULL, GL_STATIC_DRAW);
- gc->pipe[i].array.buffer_alloc = (long)END_POINTER;
- gc->pipe[i].array.buffer_use = 0;
+ gc->pipe[i].array.buffer =
+ _vbo_pool_buffer_find(gc, (size_t)END_POINTER);
}
- gc->pipe[i].array.buffer_use++;
+ glBindBuffer(GL_ARRAY_BUFFER, gc->pipe[i].array.buffer);
+// glBufferData(GL_ARRAY_BUFFER, (long)0, NULL, GL_DYNAMIC_DRAW);
+// glBufferData(GL_ARRAY_BUFFER, (long)END_POINTER, NULL, GL_DYNAMIC_DRAW);
+// if ((gc->pipe[i].array.buffer_alloc < (long)END_POINTER) ||
+// (gc->pipe[i].array.buffer_use >= (ARRAY_BUFFER_USE + ARRAY_BUFFER_USE_SHIFT * i)))
+// {
+// glBufferData(GL_ARRAY_BUFFER, (long)0, NULL, GL_DYNAMIC_DRAW);
+// glBufferData(GL_ARRAY_BUFFER, (long)END_POINTER, NULL, GL_DYNAMIC_DRAW);
+// gc->pipe[i].array.buffer_alloc = (long)END_POINTER;
+// gc->pipe[i].array.buffer_use = 0;
+// }
+// gc->pipe[i].array.buffer_use++;
x = glsym_glMapBuffer(GL_ARRAY_BUFFER, GL_WRITE_ONLY);
if (x)
@@ -3650,6 +3809,7 @@ shader_array_flush(Evas_Engine_GL_Context *gc)
if (glsym_glMapBuffer && glsym_glUnmapBuffer)
{
glBindBuffer(GL_ARRAY_BUFFER, 0);
+ gc->pipe[i].array.buffer = 0;
}
gc->pipe[i].region.x = 0;
diff --git a/src/modules/evas/engines/gl_x11/evas_engine.c b/src/modules/evas/engines/gl_x11/evas_engine.c
index 94bfdedbb9..f4c58025ff 100644
--- a/src/modules/evas/engines/gl_x11/evas_engine.c
+++ b/src/modules/evas/engines/gl_x11/evas_engine.c
@@ -61,6 +61,7 @@ EVGL_Current_Native_Context_Get_Call glsym_evgl_current_native_context_get = NUL
Evas_Gl_Symbols glsym_evas_gl_symbols = NULL;
Evas_GL_Common_Context_New glsym_evas_gl_common_context_new = NULL;
+Evas_GL_Common_Context_Call glsym_evas_gl_common_context_idle_flush = NULL;
Evas_GL_Common_Context_Call glsym_evas_gl_common_context_flush = NULL;
Evas_GL_Common_Context_Call glsym_evas_gl_common_context_free = NULL;
Evas_GL_Common_Context_Call glsym_evas_gl_common_context_use = NULL;
@@ -1283,6 +1284,7 @@ gl_symbols(void)
LINK2GENERIC(evas_gl_common_image_free);
LINK2GENERIC(evas_gl_common_image_native_enable);
LINK2GENERIC(evas_gl_common_context_new);
+ LINK2GENERIC(evas_gl_common_context_idle_flush);
LINK2GENERIC(evas_gl_common_context_flush);
LINK2GENERIC(evas_gl_common_context_free);
LINK2GENERIC(evas_gl_common_context_use);
@@ -1572,6 +1574,8 @@ eng_outbuf_idle_flush(Outbuf *ob)
{
if (glsym_evas_gl_common_shaders_flush)
glsym_evas_gl_common_shaders_flush(ob->gl_context->shared);
+ if (glsym_evas_gl_common_context_idle_flush)
+ glsym_evas_gl_common_context_idle_flush(ob->gl_context);
}
static void
--
To stop receiving notification emails like this one, please contact
the administrator of this repository.
