mirrors/0ad
1
0
Fork 2
mirror of https://gitea.wildfiregames.com/0ad/0ad.git synced 2026-09-21 20:06:40 +00:00
Code Issues Packages Projects Releases Wiki Activity

Reduce memory allocations in the renderer

Browse Source 
Use an arena allocator in ShaderModelRenderer::Render, to reduce
the allocation cost in STL containers.

Avoid unnecessary copying of std::vectors.

This was SVN commit r13911.
This commit is contained in:
Ykkrosh
2013-09-29 14:48:11 +00:00
parent 6b0a0e83eb
commit c2583e42da
6 changed files with 61 additions and 22 deletions
Download Patch File Download Diff File Expand all files Collapse all files
source/graphics/ShaderDefines.cpp
+15 -1
View File
@@ -104,10 +104,21 @@ typename CShaderParams<value_t>::SItems* CShaderParams<value_t>::GetInterned(con
template<typename value_t>
CShaderParams<value_t>::CShaderParams()
{
*this = s_Empty;
}
template<typename value_t>
CShaderParams<value_t>::CShaderParams(SItems* items) : m_Items(items)
{
}
template<typename value_t>
CShaderParams<value_t> CShaderParams<value_t>::CreateEmpty()
{
SItems items;
items.RecalcHash();
m_Items = GetInterned(items);
return CShaderParams(GetInterned(items));
}
template<typename value_t>
@@ -267,5 +278,8 @@ void CShaderConditionalDefines::Add(const char* defname, const char* defvalue, i
template<> CShaderParams<CStrIntern>::InternedItems_t CShaderParams<CStrIntern>::s_InternedItems = CShaderParams<CStrIntern>::InternedItems_t();
template<> CShaderParams<CVector4D>::InternedItems_t CShaderParams<CVector4D>::s_InternedItems = CShaderParams<CVector4D>::InternedItems_t();
template<> CShaderParams<CStrIntern> CShaderParams<CStrIntern>::s_Empty = CShaderParams<CStrIntern>::CreateEmpty();
template<> CShaderParams<CVector4D> CShaderParams<CVector4D>::s_Empty = CShaderParams<CVector4D>::CreateEmpty();
template class CShaderParams<CStrIntern>;
template class CShaderParams<CVector4D>;
source/graphics/ShaderDefines.h
+4
View File
@@ -118,6 +118,10 @@ private:
* for any subsequent requests for an equal items list.
*/
static SItems* GetInterned(const SItems& items);
CShaderParams(SItems* items);
static CShaderParams CreateEmpty();
static CShaderParams s_Empty;
};
/**
source/lib/allocators/allocator_adapters.h
+3
View File
@@ -115,6 +115,9 @@ public:
typedef ProxyAllocator<U, Allocator> other;
};
// (required to be declared by boost::unordered_map, but should never be called)
explicit NOTHROW_DEFINE ProxyAllocator();
explicit NOTHROW_DEFINE ProxyAllocator(Allocator& allocator)
: allocator(&allocator)
{
source/renderer/DecalRData.cpp
+2 -2
View File
@@ -121,12 +121,12 @@ void CDecalRData::RenderDecals(std::vector<CDecalRData*>& decals, const CShaderD
if (material.GetSamplers().size() != 0)
{
CMaterial::SamplersVector samplers = material.GetSamplers();
const CMaterial::SamplersVector& samplers = material.GetSamplers();
size_t samplersNum = samplers.size();
for (size_t s = 0; s < samplersNum; ++s)
{
CMaterial::TextureSampler &samp = samplers[s];
const CMaterial::TextureSampler& samp = samplers[s];
shader->BindTexture(samp.Name, samp.Sampler);
}
source/renderer/ModelRenderer.cpp
+33 -15
View File
@@ -17,6 +17,8 @@
#include "precompiled.h"
#include "lib/allocators/allocator_adapters.h"
#include "lib/allocators/arena.h"
#include "lib/ogl.h"
#include "maths/Vector3D.h"
#include "maths/Vector4D.h"
@@ -417,8 +419,13 @@ void ShaderModelRenderer::Render(const RenderModifierPtr& modifier, const CShade
* list in each, rebinding the GL state whenever it changes.
*/
typedef boost::unordered_map<SMRMaterialBucketKey, std::vector<CModel*>, SMRMaterialBucketKeyHash> MaterialBuckets_t;
MaterialBuckets_t materialBuckets;
Allocators::Arena<> arena(1*MiB);
typedef ProxyAllocator<void*, Allocators::Arena<> > ArenaProxyAllocator;
typedef std::vector<CModel*, ArenaProxyAllocator> ModelList_t;
typedef boost::unordered_map<SMRMaterialBucketKey, ModelList_t, SMRMaterialBucketKeyHash,
std::equal_to<SMRMaterialBucketKey>, ProxyAllocator<void*, Allocators::Arena<> >
> MaterialBuckets_t;
MaterialBuckets_t materialBuckets((MaterialBuckets_t::allocator_type(arena)));
{
PROFILE3("bucketing by material");
@@ -432,7 +439,7 @@ void ShaderModelRenderer::Render(const RenderModifierPtr& modifier, const CShade
const CShaderConditionalDefines& condefs = model->GetMaterial().GetConditionalDefines();
for (size_t j = 0; j < condefs.GetSize(); ++j)
{
const CShaderConditionalDefines::CondDefine &item = condefs.GetItem(j);
const CShaderConditionalDefines::CondDefine& item = condefs.GetItem(j);
int type = item.m_CondType;
switch (type)
{
@@ -454,21 +461,32 @@ void ShaderModelRenderer::Render(const RenderModifierPtr& modifier, const CShade
CShaderDefines defs = model->GetMaterial().GetShaderDefines(condFlags);
SMRMaterialBucketKey key(model->GetMaterial().GetShaderEffect(), defs);
std::vector<CModel*>& bucketItems = materialBuckets[key];
bucketItems.push_back(model);
MaterialBuckets_t::iterator it = materialBuckets.find(key);
if (it == materialBuckets.end())
{
std::pair<MaterialBuckets_t::iterator, bool> inserted = materialBuckets.insert(
std::make_pair(key, ModelList_t(ModelList_t::allocator_type(arena))));
inserted.first->second.reserve(32);
inserted.first->second.push_back(model);
}
else
{
it->second.push_back(model);
}
}
}
std::vector<SMRSortByDistItem> sortByDistItems;
std::vector<SMRSortByDistItem, ArenaProxyAllocator> sortByDistItems((ArenaProxyAllocator(arena)));
std::vector<CShaderTechniquePtr> sortByDistTechs;
std::vector<CShaderTechniquePtr, ArenaProxyAllocator> sortByDistTechs((ArenaProxyAllocator(arena)));
// indexed by sortByDistItems[i].techIdx
// (which stores indexes instead of CShaderTechniquePtr directly
// to avoid the shared_ptr copy cost when sorting; maybe it'd be better
// if we just stored raw CShaderTechnique* and assumed the shader manager
// will keep it alive long enough)
std::vector<SMRTechBucket> techBuckets;
std::vector<SMRTechBucket, ArenaProxyAllocator> techBuckets((ArenaProxyAllocator(arena)));
{
PROFILE3("processing material buckets");
@@ -528,7 +546,7 @@ void ShaderModelRenderer::Render(const RenderModifierPtr& modifier, const CShade
// (This exists primarily because techBuckets wants a CModel**;
// we could avoid the cost of copying into this list by adding
// a stride length into techBuckets and not requiring contiguous CModel*s)
std::vector<CModel*> sortByDistModels;
std::vector<CModel*, ArenaProxyAllocator> sortByDistModels((ArenaProxyAllocator(arena)));
if (!sortByDistItems.empty())
{
@@ -577,15 +595,15 @@ void ShaderModelRenderer::Render(const RenderModifierPtr& modifier, const CShade
// This vector keeps track of texture changes during rendering. It is kept outside the
// loops to avoid excessive reallocations. The token allocation of 64 elements
// should be plenty, though it is reallocated below (at a cost) if necessary.
std::vector<CTexture*> currentTexs;
std::vector<CTexture*, ArenaProxyAllocator> currentTexs((ArenaProxyAllocator(arena)));
currentTexs.reserve(64);
// texBindings holds the identifier bindings in the shader, which can no longer be defined
// statically in the ShaderRenderModifier class. texBindingNames uses interned strings to
// keep track of when bindings need to be reevaluated.
std::vector<CShaderProgram::Binding> texBindings;
std::vector<CShaderProgram::Binding, ArenaProxyAllocator> texBindings((ArenaProxyAllocator(arena)));
texBindings.reserve(64);
std::vector<CStrIntern> texBindingNames;
std::vector<CStrIntern, ArenaProxyAllocator> texBindingNames((ArenaProxyAllocator(arena)));
texBindingNames.reserve(64);
while (idxTechStart < techBuckets.size())
@@ -633,7 +651,7 @@ void ShaderModelRenderer::Render(const RenderModifierPtr& modifier, const CShade
if (flags && !(model->GetFlags() & flags))
continue;
CMaterial::SamplersVector samplers = model->GetMaterial().GetSamplers();
const CMaterial::SamplersVector& samplers = model->GetMaterial().GetSamplers();
size_t samplersNum = samplers.size();
// make sure the vectors are the right virtual sizes, and also
@@ -653,7 +671,7 @@ void ShaderModelRenderer::Render(const RenderModifierPtr& modifier, const CShade
// bind the samplers to the shader
for (size_t s = 0; s < samplersNum; ++s)
{
CMaterial::TextureSampler &samp = samplers[s];
const CMaterial::TextureSampler& samp = samplers[s];
CShaderProgram::Binding bind = texBindings[s];
// check that the handles are current
@@ -694,7 +712,7 @@ void ShaderModelRenderer::Render(const RenderModifierPtr& modifier, const CShade
currentStaticUniforms.BindUniforms(shader);
}
CShaderRenderQueries renderQueries = model->GetMaterial().GetRenderQueries();
const CShaderRenderQueries& renderQueries = model->GetMaterial().GetRenderQueries();
for (size_t q = 0; q < renderQueries.GetSize(); q++)
{
source/renderer/PatchRData.cpp
+4 -4
View File
@@ -787,12 +787,12 @@ void CPatchRData::RenderBases(const std::vector<CPatchRData*>& patches, const CS
if (itt->first->GetMaterial().GetSamplers().size() != 0)
{
CMaterial::SamplersVector samplers = itt->first->GetMaterial().GetSamplers();
const CMaterial::SamplersVector& samplers = itt->first->GetMaterial().GetSamplers();
size_t samplersNum = samplers.size();
for (size_t s = 0; s < samplersNum; ++s)
{
CMaterial::TextureSampler &samp = samplers[s];
const CMaterial::TextureSampler& samp = samplers[s];
shader->BindTexture(samp.Name, samp.Sampler);
}
@@ -1017,12 +1017,12 @@ void CPatchRData::RenderBlends(const std::vector<CPatchRData*>& patches, const C
if (itt->m_Texture)
{
CMaterial::SamplersVector samplers = itt->m_Texture->GetMaterial().GetSamplers();
const CMaterial::SamplersVector& samplers = itt->m_Texture->GetMaterial().GetSamplers();
size_t samplersNum = samplers.size();
for (size_t s = 0; s < samplersNum; ++s)
{
CMaterial::TextureSampler &samp = samplers[s];
const CMaterial::TextureSampler& samp = samplers[s];
shader->BindTexture(samp.Name, samp.Sampler);
}