Files
test2/source/blender/draw/intern/draw_instance_data.cc

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

205 lines
5.6 KiB
C++
Raw Normal View History

/* SPDX-FileCopyrightText: 2016 Blender Authors
*
* SPDX-License-Identifier: GPL-2.0-or-later */
/** \file
* \ingroup draw
*/
/**
* DRW Instance Data Manager
* This is a special memory manager that keeps memory blocks ready to send as VBO data in one
* continuous allocation. This way we avoid feeding #gpu::Batch each instance data one by one and
2024-12-09 09:14:44 +11:00
* unnecessary `memcpy`. Since we lose which memory block was used each #DRWShadingGroup we need to
2019-05-01 10:35:46 +10:00
* redistribute them in the same order/size to avoid to realloc each frame. This is why
* #DRWInstanceDatas are sorted in a list for each different data size.
2019-03-19 15:17:46 +11:00
*/
#include "draw_instance_data.hh"
2024-01-05 11:16:57 -05:00
#include "DRW_engine.hh"
#include "DRW_render.hh" /* For DRW_shgroup_get_instance_count() */
#include "BLI_listbase.h"
#include "BLI_mempool.h"
#include "MEM_guardedalloc.h"
struct DRWInstanceData {
DRWInstanceData *next;
bool used; /* If this data is used or not. */
size_t data_size; /* Size of one instance data. */
BLI_mempool *mempool;
};
struct DRWInstanceDataList {
DRWInstanceDataList *next, *prev;
/* Linked lists for all possible data pool size */
DRWInstanceData *idata_head[MAX_INSTANCE_DATA_SIZE];
DRWInstanceData *idata_tail[MAX_INSTANCE_DATA_SIZE];
};
struct DRWTempBufferHandle {
blender::gpu::VertBuf *buf;
/** Format pointer for reuse. */
GPUVertFormat *format;
/** Touched vertex length for resize. */
int *vert_len;
};
struct DRWTempInstancingHandle {
/** Copy of geom but with the per-instance attributes. */
blender::gpu::Batch *batch;
/** Batch containing instancing attributes. */
blender::gpu::Batch *instancer;
/** Call-buffer to be used instead of instancer. */
blender::gpu::VertBuf *buf;
/** Original non-instanced batch pointer. */
blender::gpu::Batch *geom;
};
Gawain: Refactor: VAOs caching AND use new VAOs manager. A major bottleneck of current implementation is the call to create_bindings() for basically every drawcalls. This is due to the VAO being tagged dirty when assigning a new shader to the Batch, defeating the purpose of the Batch (reuse it for drawing). Since managing hundreds of batches in DrawManager and DrawCache seems not fun enough to me, I prefered rewritting the batches itself. --- Batch changes --- For this to happen I needed to change the Instancing to be part of the Batch rather than being another batch supplied at drawtime. The Gwn_VertBuffers are copied from the batch to be instanciated and a new Gwn_VertBuffer is supplied for instancing attribs. This mean a VAO can be generated and cached for this instancing case. A Batch can be rendered with instancing, without instancing attribs and without the need for a new VAO using the GWN_batch_draw_range_ex with the force_instance parameter set to true. --- Draw manager changes --- The downside with this approach is that we must track the validity of the instanced batch (the original one). For this the only way (I could think of) is to set a callback for when the batch is getting free. This means a bit of refactor in the DrawManager with the separation of batching and instancing Batches. --- VAO cache --- Each VAO is generated for a given ShaderInterface. This means we can keep it alive as long as the shader interface lives. If a ShaderInterface is discarded, it needs to destroy every VAO associated to it. Otherwise, a new ShaderInterface with the same adress could be generated and reuse the same VAO with incorrect bindings. The VAO cache itself is using a mix between a static array of VAO and a dynamic array if the is not enough space in the static. Using this hybrid approach is a bit more performant than the dynamic array alone. The array will not resize down but empty entries will be filled up again. It's unlikely we get a buffer overflow from this. Resizing could be done on next allocation if needed. --- Results --- Using Cached VAOs means that we are not querying each vertex attrib for each vbo for each drawcall, every redraw! In a CPU limited test scene (10000 cubes in Clay engine) I get a reduction of CPU drawing time from ~20ms to 13ms. The only area that is not caching VAOs is the instancing from particles (see comment DRW_shgroup_instance_batch).
2018-02-20 01:55:19 +01:00
static ListBase g_idatalists = {nullptr, nullptr};
/* -------------------------------------------------------------------- */
/** \name Instance Data (DRWInstanceData)
* \{ */
static DRWInstanceData *drw_instance_data_create(DRWInstanceDataList *idatalist, uint attr_size)
{
DRWInstanceData *idata = static_cast<DRWInstanceData *>(
MEM_callocN(sizeof(DRWInstanceData), "DRWInstanceData"));
idata->next = nullptr;
idata->used = true;
idata->data_size = attr_size;
idata->mempool = BLI_mempool_create(sizeof(float) * idata->data_size, 0, 16, 0);
BLI_assert(attr_size > 0);
/* Push to linked list. */
if (idatalist->idata_head[attr_size - 1] == nullptr) {
idatalist->idata_head[attr_size - 1] = idata;
}
else {
idatalist->idata_tail[attr_size - 1]->next = idata;
}
idatalist->idata_tail[attr_size - 1] = idata;
return idata;
}
static void DRW_instance_data_free(DRWInstanceData *idata)
{
BLI_mempool_destroy(idata->mempool);
}
void *DRW_instance_data_next(DRWInstanceData *idata)
{
return BLI_mempool_alloc(idata->mempool);
}
DRWInstanceData *DRW_instance_data_request(DRWInstanceDataList *idatalist, uint attr_size)
{
BLI_assert(attr_size > 0 && attr_size <= MAX_INSTANCE_DATA_SIZE);
DRWInstanceData *idata = idatalist->idata_head[attr_size - 1];
/* Search for an unused data chunk. */
for (; idata; idata = idata->next) {
if (idata->used == false) {
idata->used = true;
return idata;
}
}
return drw_instance_data_create(idatalist, attr_size);
}
/** \} */
/* -------------------------------------------------------------------- */
/** \name Instance Data List (DRWInstanceDataList)
* \{ */
DRWInstanceDataList *DRW_instance_data_list_create()
{
DRWInstanceDataList *idatalist = static_cast<DRWInstanceDataList *>(
MEM_callocN(sizeof(DRWInstanceDataList), "DRWInstanceDataList"));
Gawain: Refactor: VAOs caching AND use new VAOs manager. A major bottleneck of current implementation is the call to create_bindings() for basically every drawcalls. This is due to the VAO being tagged dirty when assigning a new shader to the Batch, defeating the purpose of the Batch (reuse it for drawing). Since managing hundreds of batches in DrawManager and DrawCache seems not fun enough to me, I prefered rewritting the batches itself. --- Batch changes --- For this to happen I needed to change the Instancing to be part of the Batch rather than being another batch supplied at drawtime. The Gwn_VertBuffers are copied from the batch to be instanciated and a new Gwn_VertBuffer is supplied for instancing attribs. This mean a VAO can be generated and cached for this instancing case. A Batch can be rendered with instancing, without instancing attribs and without the need for a new VAO using the GWN_batch_draw_range_ex with the force_instance parameter set to true. --- Draw manager changes --- The downside with this approach is that we must track the validity of the instanced batch (the original one). For this the only way (I could think of) is to set a callback for when the batch is getting free. This means a bit of refactor in the DrawManager with the separation of batching and instancing Batches. --- VAO cache --- Each VAO is generated for a given ShaderInterface. This means we can keep it alive as long as the shader interface lives. If a ShaderInterface is discarded, it needs to destroy every VAO associated to it. Otherwise, a new ShaderInterface with the same adress could be generated and reuse the same VAO with incorrect bindings. The VAO cache itself is using a mix between a static array of VAO and a dynamic array if the is not enough space in the static. Using this hybrid approach is a bit more performant than the dynamic array alone. The array will not resize down but empty entries will be filled up again. It's unlikely we get a buffer overflow from this. Resizing could be done on next allocation if needed. --- Results --- Using Cached VAOs means that we are not querying each vertex attrib for each vbo for each drawcall, every redraw! In a CPU limited test scene (10000 cubes in Clay engine) I get a reduction of CPU drawing time from ~20ms to 13ms. The only area that is not caching VAOs is the instancing from particles (see comment DRW_shgroup_instance_batch).
2018-02-20 01:55:19 +01:00
BLI_addtail(&g_idatalists, idatalist);
return idatalist;
}
void DRW_instance_data_list_free(DRWInstanceDataList *idatalist)
{
DRWInstanceData *idata, *next_idata;
for (int i = 0; i < MAX_INSTANCE_DATA_SIZE; i++) {
for (idata = idatalist->idata_head[i]; idata; idata = next_idata) {
next_idata = idata->next;
DRW_instance_data_free(idata);
MEM_freeN(idata);
}
idatalist->idata_head[i] = nullptr;
idatalist->idata_tail[i] = nullptr;
}
Gawain: Refactor: VAOs caching AND use new VAOs manager. A major bottleneck of current implementation is the call to create_bindings() for basically every drawcalls. This is due to the VAO being tagged dirty when assigning a new shader to the Batch, defeating the purpose of the Batch (reuse it for drawing). Since managing hundreds of batches in DrawManager and DrawCache seems not fun enough to me, I prefered rewritting the batches itself. --- Batch changes --- For this to happen I needed to change the Instancing to be part of the Batch rather than being another batch supplied at drawtime. The Gwn_VertBuffers are copied from the batch to be instanciated and a new Gwn_VertBuffer is supplied for instancing attribs. This mean a VAO can be generated and cached for this instancing case. A Batch can be rendered with instancing, without instancing attribs and without the need for a new VAO using the GWN_batch_draw_range_ex with the force_instance parameter set to true. --- Draw manager changes --- The downside with this approach is that we must track the validity of the instanced batch (the original one). For this the only way (I could think of) is to set a callback for when the batch is getting free. This means a bit of refactor in the DrawManager with the separation of batching and instancing Batches. --- VAO cache --- Each VAO is generated for a given ShaderInterface. This means we can keep it alive as long as the shader interface lives. If a ShaderInterface is discarded, it needs to destroy every VAO associated to it. Otherwise, a new ShaderInterface with the same adress could be generated and reuse the same VAO with incorrect bindings. The VAO cache itself is using a mix between a static array of VAO and a dynamic array if the is not enough space in the static. Using this hybrid approach is a bit more performant than the dynamic array alone. The array will not resize down but empty entries will be filled up again. It's unlikely we get a buffer overflow from this. Resizing could be done on next allocation if needed. --- Results --- Using Cached VAOs means that we are not querying each vertex attrib for each vbo for each drawcall, every redraw! In a CPU limited test scene (10000 cubes in Clay engine) I get a reduction of CPU drawing time from ~20ms to 13ms. The only area that is not caching VAOs is the instancing from particles (see comment DRW_shgroup_instance_batch).
2018-02-20 01:55:19 +01:00
BLI_remlink(&g_idatalists, idatalist);
MEM_freeN(idatalist);
}
void DRW_instance_data_list_reset(DRWInstanceDataList *idatalist)
{
DRWInstanceData *idata;
for (int i = 0; i < MAX_INSTANCE_DATA_SIZE; i++) {
for (idata = idatalist->idata_head[i]; idata; idata = idata->next) {
idata->used = false;
}
}
}
void DRW_instance_data_list_free_unused(DRWInstanceDataList *idatalist)
{
DRWInstanceData *idata, *next_idata;
/* Remove unused data blocks and sanitize each list. */
for (int i = 0; i < MAX_INSTANCE_DATA_SIZE; i++) {
idatalist->idata_tail[i] = nullptr;
for (idata = idatalist->idata_head[i]; idata; idata = next_idata) {
next_idata = idata->next;
if (idata->used == false) {
if (idatalist->idata_head[i] == idata) {
idatalist->idata_head[i] = next_idata;
}
else {
2018-09-19 18:19:49 +02:00
/* idatalist->idata_tail[i] is guaranteed not to be null in this case. */
idatalist->idata_tail[i]->next = next_idata;
}
DRW_instance_data_free(idata);
MEM_freeN(idata);
}
else {
if (idatalist->idata_tail[i] != nullptr) {
idatalist->idata_tail[i]->next = idata;
}
idatalist->idata_tail[i] = idata;
}
}
}
}
void DRW_instance_data_list_resize(DRWInstanceDataList *idatalist)
{
DRWInstanceData *idata;
for (int i = 0; i < MAX_INSTANCE_DATA_SIZE; i++) {
for (idata = idatalist->idata_head[i]; idata; idata = idata->next) {
BLI_mempool_clear_ex(idata->mempool, BLI_mempool_len(idata->mempool));
}
}
}
/** \} */