[Bf-blender-cvs] [ed555750eb8] blender2.8: DRW: Use pseudo persistent memory pool for the rendering data structure.
Clément Foucault
noreply at git.blender.org
Mon Nov 6 17:43:18 CET 2017
Commit: ed555750eb8294005f54425094cb692f29830631
Author: Clément Foucault
Date: Mon Nov 6 16:47:23 2017 +0100
Branches: blender2.8
https://developer.blender.org/rBed555750eb8294005f54425094cb692f29830631
DRW: Use pseudo persistent memory pool for the rendering data structure.
This gets rid of the bottleneck of allocation / free of thousands of elements every frame.
Cache time (Eevee) (test scene is default file with cube duplicated 3241 times)
pre-patch: 23ms
post-patch: 14ms
===================================================================
M source/blender/draw/intern/draw_manager.c
M source/blender/gpu/GPU_viewport.h
M source/blender/gpu/intern/gpu_viewport.c
===================================================================
diff --git a/source/blender/draw/intern/draw_manager.c b/source/blender/draw/intern/draw_manager.c
index 57489153ba0..f18bc9969b1 100644
--- a/source/blender/draw/intern/draw_manager.c
+++ b/source/blender/draw/intern/draw_manager.c
@@ -27,6 +27,7 @@
#include "BLI_dynstr.h"
#include "BLI_listbase.h"
+#include "BLI_mempool.h"
#include "BLI_rect.h"
#include "BLI_string.h"
@@ -130,13 +131,6 @@
/* Use draw manager to call GPU_select, see: DRW_draw_select_loop */
#define USE_GPU_SELECT
-/* Use BLI_memiter */
-#define USE_MEM_ITER
-
-#ifdef USE_MEM_ITER
-#include "BLI_memiter.h"
-#endif
-
#ifdef USE_GPU_SELECT
# include "ED_view3d.h"
# include "ED_armature.h"
@@ -147,6 +141,7 @@
#define MAX_ATTRIB_NAME 32
+#define MAX_ATTRIB_COUNT 6 /* Can be adjusted for more */
#define MAX_PASS_NAME 32
#define MAX_CLIP_PLANES 6 /* GL_MAX_CLIP_PLANES is at least 6 */
@@ -177,7 +172,7 @@ typedef enum {
} DRWAttribType;
struct DRWUniform {
- struct DRWUniform *next, *prev;
+ struct DRWUniform *next;
DRWUniformType type;
int location;
int length;
@@ -186,7 +181,7 @@ struct DRWUniform {
};
typedef struct DRWAttrib {
- struct DRWAttrib *next, *prev;
+ struct DRWAttrib *prev;
char name[MAX_ATTRIB_NAME];
int location;
int format_id;
@@ -195,8 +190,9 @@ typedef struct DRWAttrib {
} DRWAttrib;
struct DRWInterface {
- ListBase uniforms; /* DRWUniform */
- ListBase attribs; /* DRWAttrib */
+ DRWUniform *uniforms; /* DRWUniform, single-linked list */
+ DRWAttrib *attribs; /* DRWAttrib, single-linked list */
+ DRWAttrib *attribs_first; /* First added attrib to traverse in the right order */
int attribs_count;
int attribs_stride;
int attribs_size[16];
@@ -227,15 +223,16 @@ struct DRWInterface {
};
struct DRWPass {
- ListBase shgroups; /* DRWShadingGroup */
+ /* Single linked list with last member to append */
+ DRWShadingGroup *shgroups;
+ DRWShadingGroup *shgroups_last;
+
DRWState state;
char name[MAX_PASS_NAME];
};
typedef struct DRWCallHeader {
-#ifndef USE_MEM_ITER
- void *next, *prev;
-#endif
+ void *prev;
#ifdef USE_GPU_SELECT
int select_id;
@@ -265,21 +262,18 @@ typedef struct DRWCallGenerate {
typedef struct DRWCallDynamic {
DRWCallHeader head;
- const void *data[];
+ const void *data[MAX_ATTRIB_COUNT];
} DRWCallDynamic;
struct DRWShadingGroup {
- struct DRWShadingGroup *next, *prev;
+ struct DRWShadingGroup *next;
GPUShader *shader; /* Shader to bind */
- DRWInterface *interface; /* Uniforms pointers */
+ DRWInterface interface; /* Uniforms pointers */
/* DRWCall or DRWCallDynamic depending of type */
-#ifdef USE_MEM_ITER
- BLI_memiter *calls;
-#else
- ListBase calls;
-#endif
+ void *calls;
+ void *calls_first; /* To be able to traverse the list in the order of addition */
DRWState state_extra; /* State changes for this batch only (or'd with the pass's state) */
DRWState state_extra_disable; /* State changes for this batch only (and'd with the pass's state) */
@@ -322,6 +316,15 @@ enum {
/** Render State: No persistent data between draw calls. */
static struct DRWGlobalState {
+ /* Cache generation */
+ ViewportMemoryPool *vmempool;
+ DRWUniform *last_uniform;
+ DRWAttrib *last_attrib;
+ DRWCall *last_call;
+ DRWCallGenerate *last_callgenerate;
+ DRWCallDynamic *last_calldynamic;
+ DRWShadingGroup *last_shgroup;
+
/* Rendering state */
GPUShader *shader;
@@ -652,10 +655,8 @@ void DRW_shader_free(GPUShader *shader)
/** \name Interface (DRW_interface)
* \{ */
-static DRWInterface *DRW_interface_create(GPUShader *shader)
+static void DRW_interface_create(DRWInterface *interface, GPUShader *shader)
{
- DRWInterface *interface = MEM_mallocN(sizeof(DRWInterface), "DRWInterface");
-
interface->model = GPU_shader_get_builtin_uniform(shader, GWN_UNIFORM_MODEL);
interface->modelinverse = GPU_shader_get_builtin_uniform(shader, GWN_UNIFORM_MODEL_INV);
interface->modelview = GPU_shader_get_builtin_uniform(shader, GWN_UNIFORM_MODELVIEW);
@@ -681,21 +682,11 @@ static DRWInterface *DRW_interface_create(GPUShader *shader)
memset(&interface->vbo_format, 0, sizeof(Gwn_VertFormat));
- BLI_listbase_clear(&interface->uniforms);
- BLI_listbase_clear(&interface->attribs);
-
- return interface;
+ interface->uniforms = NULL;
+ interface->attribs = NULL;
+ interface->attribs_first = NULL;
}
-#ifdef USE_GPU_SELECT
-static DRWInterface *DRW_interface_duplicate(DRWInterface *interface_src)
-{
- DRWInterface *interface_dst = MEM_dupallocN(interface_src);
- BLI_duplicatelist(&interface_dst->uniforms, &interface_src->uniforms);
- BLI_duplicatelist(&interface_dst->attribs, &interface_src->attribs);
- return interface_dst;
-}
-#endif
static void DRW_interface_uniform(DRWShadingGroup *shgroup, const char *name,
DRWUniformType type, const void *value, int length, int arraysize)
@@ -716,7 +707,7 @@ static void DRW_interface_uniform(DRWShadingGroup *shgroup, const char *name,
return;
}
- DRWUniform *uni = MEM_mallocN(sizeof(DRWUniform), "DRWUniform");
+ DRWUniform *uni = BLI_mempool_alloc(DST.vmempool->uniforms);
BLI_assert(arraysize > 0);
@@ -726,12 +717,14 @@ static void DRW_interface_uniform(DRWShadingGroup *shgroup, const char *name,
uni->length = length;
uni->arraysize = arraysize;
- BLI_addtail(&shgroup->interface->uniforms, uni);
+ /* Prepend */
+ uni->next = shgroup->interface.uniforms;
+ shgroup->interface.uniforms = uni;
}
static void DRW_interface_attrib(DRWShadingGroup *shgroup, const char *name, DRWAttribType type, int size, bool dummy)
{
- DRWAttrib *attrib = MEM_mallocN(sizeof(DRWAttrib), "DRWAttrib");
+ DRWAttrib *attrib = BLI_mempool_alloc(DST.vmempool->attribs);
GLuint program = GPU_shader_get_program(shgroup->shader);
attrib->location = glGetAttribLocation(program, name);
@@ -755,9 +748,19 @@ static void DRW_interface_attrib(DRWShadingGroup *shgroup, const char *name, DRW
BLI_assert(BLI_strnlen(name, 32) < 32);
BLI_strncpy(attrib->name, name, 32);
- shgroup->interface->attribs_count += 1;
+ shgroup->interface.attribs_count += 1;
+ BLI_assert(shgroup->interface.attribs_count < MAX_ATTRIB_COUNT);
- BLI_addtail(&shgroup->interface->attribs, attrib);
+ /* Prepend */
+ if (shgroup->interface.attribs == NULL) {
+ shgroup->interface.attribs = attrib;
+ shgroup->interface.attribs_first = attrib;
+ }
+ else {
+ shgroup->interface.attribs->prev = attrib;
+ shgroup->interface.attribs = attrib;
+ }
+ attrib->prev = NULL;
}
/** \} */
@@ -770,23 +773,30 @@ static void DRW_interface_attrib(DRWShadingGroup *shgroup, const char *name, DRW
DRWShadingGroup *DRW_shgroup_create(struct GPUShader *shader, DRWPass *pass)
{
- DRWShadingGroup *shgroup = MEM_mallocN(sizeof(DRWShadingGroup), "DRWShadingGroup");
- BLI_addtail(&pass->shgroups, shgroup);
+ DRWShadingGroup *shgroup = BLI_mempool_alloc(DST.vmempool->shgroups);
+
+ /* Append */
+ if (pass->shgroups != NULL) {
+ pass->shgroups_last->next = shgroup;
+ }
+ else {
+ pass->shgroups = shgroup;
+ }
+ pass->shgroups_last = shgroup;
+ shgroup->next = NULL;
+
+ DRW_interface_create(&shgroup->interface, shader);
shgroup->type = DRW_SHG_NORMAL;
shgroup->shader = shader;
- shgroup->interface = DRW_interface_create(shader);
shgroup->state_extra = 0;
shgroup->state_extra_disable = ~0x0;
shgroup->batch_geom = NULL;
shgroup->instance_geom = NULL;
shgroup->instance_data = NULL;
-#ifdef USE_MEM_ITER
- shgroup->calls = BLI_memiter_create(BLI_MEMITER_DEFAULT_SIZE);
-#else
- BLI_listbase_clear(&shgroup->calls);
-#endif
+ shgroup->calls = NULL;
+ shgroup->calls_first = NULL;
#ifdef USE_GPU_SELECT
shgroup->pass_parent = pass;
@@ -885,7 +895,7 @@ DRWShadingGroup *DRW_shgroup_material_empty_tri_batch_create(
if (shgroup) {
shgroup->type = DRW_SHG_TRIANGLE_BATCH;
- shgroup->interface->instance_count = size * 3;
+ shgroup->interface.instance_count = size * 3;
DRW_interface_attrib(shgroup, "dummy", DRW_ATTRIB_FLOAT, 1, true);
}
@@ -930,7 +940,7 @@ DRWShadingGroup *DRW_shgroup_empty_tri_batch_create(struct GPUShader *shader, DR
DRWShadingGroup *shgroup = DRW_shgroup_create(shader, pass);
shgroup->type = DRW_SHG_TRIANGLE_BATCH;
- shgroup->interface->instance_count = size * 3;
+ shgroup->interface.instance_count = size * 3;
DRW_interface_attrib(shgroup, "dummy", DRW_ATTRIB_FLOAT, 1, true);
return shgroup;
@@ -938,46 +948,43 @@ DRWShadingGroup *DRW_shgroup_empty_tri_batch_create(struct GPUShader *shader, DR
void DRW_shgroup_free(struct DRWShadingGroup *shgroup)
{
-#ifdef USE_MEM_ITER
- BLI_memiter_destroy(shgroup->calls);
-#else
- BLI_freelistN(&shgroup->calls);
-#endif
-
- BLI_freelistN(&shgroup->interface->uniforms);
- BLI_freelistN(&shgroup->interface->attribs);
-
- if (shgroup->interface->instance_vbo &&
- (shgroup->interface->instance_batch == 0))
+ if (shgroup->interface.instance_vbo &&
+ (shgroup->interface.instance_batch == 0))
{
- glDeleteBuffers(1, &shgroup->interface->instance_vbo);
+ glDeleteBuffers(1, &shgroup->interface.instance_vbo);
}
- MEM_freeN(shgroup->interface);
-
GWN_BATCH_DISCARD_SAFE(shgroup->batch_geom);
}
void DRW_shgroup_instance_batch(DRWShadingGroup *shgroup, struct Gwn_Batch *instances)
{
BLI_assert(shgroup->type == DRW_SHG_INSTANCE);
- BLI_assert(shgroup->interface->instance_batch == NULL);
+ BLI_assert(shgroup->interface.instance_batch == NULL);
- shgroup->interface->instance_batch = instances;
+ shgroup->interface.instance_batch = instances;
}
+#define CALL_PREPEND(shgroup, call) { \
+ if (shgroup->calls == NULL) { \
+ shgroup->calls = call; \
+ shgroup->calls_first = call; \
+ } \
+ else { \
+ ((typeof(call))shgroup->calls)->head.prev = call; \
+ shgroup->calls = call; \
+ } \
+ call->head.prev = NULL; \
+} ((void)0)
+
+
void DRW_shgroup_call_add(DRWShadingGroup *shgroup, Gwn_Batch *geom, float (*obmat)[4])
{
BLI_assert(geom != NULL);
- DRWCall *call;
+ DRWCall *call = BLI_mempool_alloc(DST.vmempool->calls);
-#ifdef USE_MEM_ITER
- call = BLI_memiter_calloc(shgroup->calls, sizeof(DRWCall));
-#else
- call = MEM_callocN(sizeof(DRWCall), "DRWCall");
- BLI_addtail(&shgroup->calls, call);
-#endif
+
@@ Diff output truncated at 10240 characters. @@
More information about the Bf-blender-cvs
mailing list