Skip to content

A
angle
Commit c828b149 by daniel@transgaming.com
Options

Improve TranslatedVertexBuffer::map performance

TRAC #11392 - For small draws, load stride 0 streams into regular streaming buffer Signed-off-by: Shannon Woods Signed-off-by: Daniel Koch Author: Andrew Lewycky git-svn-id: https://angleproject.googlecode.com/svn/trunk@265 736b8ea6-26fd-11df-bfd4-992fa37f6226
parent 9a0606cf
Showing with 77 additions and 47 deletions
src/libGLESv2/geometry/VertexDataManager.cpp
...@@ -20,6 +20,8 @@ ...@@ -20,6 +20,8 @@
namespace namespace
{ {
enum { INITIAL_STREAM_BUFFER_SIZE = 1024*1024 }; enum { INITIAL_STREAM_BUFFER_SIZE = 1024*1024 };
enum { MAX_CURRENT_VALUE_EXPANSION = 16 };
enum { CURRENT_VALUES_REQUIRED_SPACE = 4 * sizeof(float) * gl::MAX_VERTEX_ATTRIBS * MAX_CURRENT_VALUE_EXPANSION };
} }
namespace gl namespace gl
...@@ -31,7 +33,7 @@ VertexDataManager::VertexDataManager(Context *context, BufferBackEnd *backend) ...@@ -31,7 +33,7 @@ VertexDataManager::VertexDataManager(Context *context, BufferBackEnd *backend)
mStreamBuffer = mBackend->createVertexBuffer(INITIAL_STREAM_BUFFER_SIZE); mStreamBuffer = mBackend->createVertexBuffer(INITIAL_STREAM_BUFFER_SIZE);
try try
{ {
mCurrentValueBuffer = mBackend->createVertexBufferForStrideZero(4*sizeof(float)*MAX_VERTEX_ATTRIBS); mCurrentValueBuffer = mBackend->createVertexBufferForStrideZero(4 * sizeof(float) * MAX_VERTEX_ATTRIBS);
} }
catch (...) catch (...)
{ {
...@@ -77,8 +79,18 @@ GLenum VertexDataManager::preRenderValidate(GLint start, GLsizei count, ...@@ -77,8 +79,18 @@ GLenum VertexDataManager::preRenderValidate(GLint start, GLsizei count,
translated[i].enabled = activeAttribs[i]; translated[i].enabled = activeAttribs[i];
} }
processNonArrayAttributes(attribs, activeAttribs, translated); bool usesCurrentValues = false;
for (int i = 0; i < MAX_VERTEX_ATTRIBS; i++)
{
if (activeAttribs[i] && !attribs[i].mEnabled)
{
usesCurrentValues = true;
break;
}
}
// Handle the identity-mapped attributes.
// Process array attributes. // Process array attributes.
std::size_t requiredSpace = 0; std::size_t requiredSpace = 0;
...@@ -91,6 +103,11 @@ GLenum VertexDataManager::preRenderValidate(GLint start, GLsizei count, ...@@ -91,6 +103,11 @@ GLenum VertexDataManager::preRenderValidate(GLint start, GLsizei count,
} }
} }
if (usesCurrentValues)
{
requiredSpace += CURRENT_VALUES_REQUIRED_SPACE;
}
if (requiredSpace > mStreamBuffer->size()) if (requiredSpace > mStreamBuffer->size())
{ {
std::size_t newSize = std::max(requiredSpace, 3 * mStreamBuffer->size() / 2); // 1.5 x mStreamBuffer->size() is arbitrary and should be checked to see we don't have too many reallocations. std::size_t newSize = std::max(requiredSpace, 3 * mStreamBuffer->size() / 2); // 1.5 x mStreamBuffer->size() is arbitrary and should be checked to see we don't have too many reallocations.
...@@ -145,30 +162,12 @@ GLenum VertexDataManager::preRenderValidate(GLint start, GLsizei count, ...@@ -145,30 +162,12 @@ GLenum VertexDataManager::preRenderValidate(GLint start, GLsizei count,
} }
} }
return GL_NO_ERROR; if (usesCurrentValues)
}
void VertexDataManager::reloadCurrentValues(const AttributeState *attribs, std::size_t *offset)
{
if (mDirtyCurrentValues)
{ {
std::size_t totalSize = 4 * sizeof(float) * MAX_VERTEX_ATTRIBS; processNonArrayAttributes(attribs, activeAttribs, translated, count);
mCurrentValueBuffer->reserveSpace(totalSize);
float* p = static_cast<float*>(mCurrentValueBuffer->map(totalSize, &mCurrentValueOffset));
for (int i = 0; i < MAX_VERTEX_ATTRIBS; i++)
{
memcpy(&p[i*4], attribs[i].mCurrentValue, sizeof(attribs[i].mCurrentValue)); // FIXME: this should be doing a translation. This assumes that GL_FLOATx4 is supported.
}
mCurrentValueBuffer->unmap();
mDirtyCurrentValues = false;
} }
*offset = mCurrentValueOffset; return GL_NO_ERROR;
} }
std::size_t VertexDataManager::typeSize(GLenum type) const std::size_t VertexDataManager::typeSize(GLenum type) const
...@@ -214,37 +213,67 @@ std::size_t VertexDataManager::spaceRequired(const AttributeState &attrib, std:: ...@@ -214,37 +213,67 @@ std::size_t VertexDataManager::spaceRequired(const AttributeState &attrib, std::
return roundUp(size, 4 * sizeof(GLfloat)); return roundUp(size, 4 * sizeof(GLfloat));
} }
void VertexDataManager::processNonArrayAttributes(const AttributeState *attribs, const std::bitset<MAX_VERTEX_ATTRIBS> &activeAttribs, TranslatedAttribute *translated) void VertexDataManager::processNonArrayAttributes(const AttributeState *attribs, const std::bitset<MAX_VERTEX_ATTRIBS> &activeAttribs, TranslatedAttribute *translated, std::size_t count)
{ {
bool usesCurrentValues = false; if (count <= MAX_CURRENT_VALUE_EXPANSION)
for (int i = 0; i < MAX_VERTEX_ATTRIBS; i++)
{ {
if (activeAttribs[i] && !attribs[i].mEnabled) if (mDirtyCurrentValues || mCurrentValueLoadBuffer != mStreamBuffer)
{ {
usesCurrentValues = true; float *p = static_cast<float*>(mStreamBuffer->map(CURRENT_VALUES_REQUIRED_SPACE, &mCurrentValueOffset));
break;
for (int i = 0; i < MAX_VERTEX_ATTRIBS; i++)
{
float *out = p + MAX_CURRENT_VALUE_EXPANSION * 4 * i;
for (unsigned int j = 0; j < MAX_CURRENT_VALUE_EXPANSION; j++)
{
*out++ = attribs[i].mCurrentValue[0];
*out++ = attribs[i].mCurrentValue[1];
*out++ = attribs[i].mCurrentValue[2];
*out++ = attribs[i].mCurrentValue[3];
}
}
mStreamBuffer->unmap();
mCurrentValueLoadBuffer = mStreamBuffer;
mCurrentValueSize = MAX_CURRENT_VALUE_EXPANSION;
mCurrentValueStride = 4 * sizeof(float);
} }
} }
else
if (usesCurrentValues)
{ {
std::size_t currentValueOffset; if (mDirtyCurrentValues || mCurrentValueLoadBuffer != mCurrentValueBuffer)
{
std::size_t totalSize = 4 * sizeof(float) * MAX_VERTEX_ATTRIBS;
reloadCurrentValues(attribs, &currentValueOffset); mCurrentValueBuffer->reserveSpace(totalSize);
for (std::size_t i = 0; i < MAX_VERTEX_ATTRIBS; i++) float* p = static_cast<float*>(mCurrentValueBuffer->map(totalSize, &mCurrentValueOffset));
{
if (activeAttribs[i] && !attribs[i].mEnabled)
{
translated[i].buffer = mCurrentValueBuffer;
translated[i].type = GL_FLOAT; for (int i = 0; i < MAX_VERTEX_ATTRIBS; i++)
translated[i].size = 4; {
translated[i].normalized = false; memcpy(&p[i*4], attribs[i].mCurrentValue, sizeof(attribs[i].mCurrentValue)); // FIXME: this should be doing a translation. This assumes that GL_FLOATx4 is supported.
translated[i].stride = 0;
translated[i].offset = currentValueOffset + 4 * sizeof(float) * i;
} }
mCurrentValueBuffer->unmap();
mCurrentValueLoadBuffer = mCurrentValueBuffer;
mCurrentValueSize = 1;
mCurrentValueStride = 0;
}
}
for (std::size_t i = 0; i < MAX_VERTEX_ATTRIBS; i++)
{
if (activeAttribs[i] && !attribs[i].mEnabled)
{
translated[i].buffer = mCurrentValueLoadBuffer;
translated[i].type = GL_FLOAT;
translated[i].size = 4;
translated[i].normalized = false;
translated[i].stride = mCurrentValueStride;
translated[i].offset = mCurrentValueOffset + 4 * sizeof(float) * i * mCurrentValueSize;
} }
} }
} }
......
src/libGLESv2/geometry/VertexDataManager.h
...@@ -43,9 +43,7 @@ class VertexDataManager ...@@ -43,9 +43,7 @@ class VertexDataManager
private: private:
std::bitset<MAX_VERTEX_ATTRIBS> getActiveAttribs() const; std::bitset<MAX_VERTEX_ATTRIBS> getActiveAttribs() const;
void reloadCurrentValues(const AttributeState *attribs, std::size_t *offset); void processNonArrayAttributes(const AttributeState *attribs, const std::bitset<MAX_VERTEX_ATTRIBS> &activeAttribs, TranslatedAttribute *translated, std::size_t count);
void processNonArrayAttributes(const AttributeState *attribs, const std::bitset<MAX_VERTEX_ATTRIBS> &activeAttribs, TranslatedAttribute *translated);
std::size_t typeSize(GLenum type) const; std::size_t typeSize(GLenum type) const;
std::size_t interpretGlStride(const AttributeState &attrib) const; std::size_t interpretGlStride(const AttributeState &attrib) const;
...@@ -61,6 +59,9 @@ class VertexDataManager ...@@ -61,6 +59,9 @@ class VertexDataManager
bool mDirtyCurrentValues; bool mDirtyCurrentValues;
std::size_t mCurrentValueOffset; // Offset within mCurrentValueBuffer that the current attribute values were last loaded at. std::size_t mCurrentValueOffset; // Offset within mCurrentValueBuffer that the current attribute values were last loaded at.
TranslatedVertexBuffer *mCurrentValueBuffer; TranslatedVertexBuffer *mCurrentValueBuffer;
TranslatedVertexBuffer *mCurrentValueLoadBuffer;
std::size_t mCurrentValueStride;
std::size_t mCurrentValueSize;
}; };
} }
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment