GAL: OpenGL draw calls pooling/splitting
Improves rendering performance, esp. for large designs by: - using a separate glDrawArrays calls() for large buffers of contiguous vertices (e.g. large zone fills) - pooling smaller items into a small-sized index buffer held in system RAM (has to be DMAed to the GPU by the driver anyway)
This commit is contained in:
Tomasz Wlostowski
parent
569c39ac37
commit
2c6e9778a1
|
|
@ -30,6 +30,9 @@
|
|||
#include <gal/opengl/noncached_container.h>
|
||||
#include <gal/opengl/shader.h>
|
||||
#include <gal/opengl/utils.h>
|
||||
#include <gal/opengl/vertex_item.h>
|
||||
|
||||
#include <profile.h>
|
||||
|
||||
#include <typeinfo>
|
||||
#include <confirm.h>
|
||||
|
|
@ -82,26 +85,13 @@ void GPU_MANAGER::SetShader( SHADER& aShader )
|
|||
GPU_CACHED_MANAGER::GPU_CACHED_MANAGER( VERTEX_CONTAINER* aContainer ) :
|
||||
GPU_MANAGER( aContainer ),
|
||||
m_buffersInitialized( false ),
|
||||
m_indicesPtr( nullptr ),
|
||||
m_indicesBuffer( 0 ),
|
||||
m_indicesSize( 0 ),
|
||||
m_indicesCapacity( 0 )
|
||||
{
|
||||
// Allocate the biggest possible buffer for indices
|
||||
resizeIndices( aContainer->GetSize() );
|
||||
}
|
||||
|
||||
|
||||
GPU_CACHED_MANAGER::~GPU_CACHED_MANAGER()
|
||||
{
|
||||
if( m_buffersInitialized )
|
||||
{
|
||||
if( glBindBuffer )
|
||||
glBindBuffer( GL_ARRAY_BUFFER, 0 );
|
||||
|
||||
if( glDeleteBuffers )
|
||||
glDeleteBuffers( 1, &m_indicesBuffer );
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
|
|
@ -109,54 +99,40 @@ void GPU_CACHED_MANAGER::BeginDrawing()
|
|||
{
|
||||
wxASSERT( !m_isDrawing );
|
||||
|
||||
if( !m_buffersInitialized )
|
||||
{
|
||||
glGenBuffers( 1, &m_indicesBuffer );
|
||||
checkGlError( "generating vertices buffer", __FILE__, __LINE__ );
|
||||
m_buffersInitialized = true;
|
||||
}
|
||||
|
||||
if( m_container->IsDirty() )
|
||||
resizeIndices( m_container->GetSize() );
|
||||
|
||||
// Number of vertices to be drawn in the EndDrawing()
|
||||
m_indicesSize = 0;
|
||||
// Set the indices pointer to the beginning of the indices-to-draw buffer
|
||||
m_indicesPtr = m_indices.get();
|
||||
m_curVrangeSize = 0;
|
||||
m_indexBufMaxSize = 0;
|
||||
m_indexBufSize = 0;
|
||||
m_vranges.clear();
|
||||
|
||||
m_isDrawing = true;
|
||||
}
|
||||
|
||||
|
||||
void GPU_CACHED_MANAGER::DrawIndices( unsigned int aOffset, unsigned int aSize )
|
||||
void GPU_CACHED_MANAGER::DrawIndices( const VERTEX_ITEM *aItem )
|
||||
{
|
||||
wxASSERT( m_isDrawing );
|
||||
|
||||
// Copy indices of items that should be drawn to GPU memory
|
||||
for( unsigned int i = aOffset; i < aOffset + aSize; *m_indicesPtr++ = i++ )
|
||||
;
|
||||
unsigned int offset = aItem->GetOffset();
|
||||
unsigned int size = aItem->GetSize();
|
||||
|
||||
m_indicesSize += aSize;
|
||||
}
|
||||
|
||||
|
||||
void GPU_CACHED_MANAGER::DrawAll()
|
||||
{
|
||||
wxASSERT( m_isDrawing );
|
||||
|
||||
for( unsigned int i = 0; i < m_indicesSize; *m_indicesPtr++ = i++ )
|
||||
;
|
||||
|
||||
m_indicesSize = m_container->GetSize();
|
||||
if( size > 1000 )
|
||||
{
|
||||
m_totalHuge += size;
|
||||
m_vranges.emplace_back( offset, offset + size - 1, true );
|
||||
m_indexBufSize = std::max( m_curVrangeSize, m_indexBufSize );
|
||||
m_curVrangeSize = 0;
|
||||
}
|
||||
else if ( size > 0 )
|
||||
{
|
||||
m_totalNormal += size;
|
||||
m_vranges.emplace_back( offset, offset + size - 1, false );
|
||||
m_curVrangeSize += size;
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
void GPU_CACHED_MANAGER::EndDrawing()
|
||||
{
|
||||
#ifdef KICAD_GAL_PROFILE
|
||||
PROF_COUNTER totalRealTime;
|
||||
#endif /* KICAD_GAL_PROFILE */
|
||||
|
||||
wxASSERT( m_isDrawing );
|
||||
|
||||
CACHED_CONTAINER* cached = static_cast<CACHED_CONTAINER*>( m_container );
|
||||
|
|
@ -164,11 +140,10 @@ void GPU_CACHED_MANAGER::EndDrawing()
|
|||
if( cached->IsMapped() )
|
||||
cached->Unmap();
|
||||
|
||||
if( m_indicesSize == 0 )
|
||||
{
|
||||
m_isDrawing = false;
|
||||
return;
|
||||
}
|
||||
m_indexBufSize = std::max( m_curVrangeSize, m_indexBufSize );
|
||||
m_indexBufMaxSize = std::max( 2*m_indexBufSize, m_indexBufMaxSize );
|
||||
|
||||
resizeIndices( m_indexBufMaxSize );
|
||||
|
||||
if( m_enableDepthTest )
|
||||
glEnable( GL_DEPTH_TEST );
|
||||
|
|
@ -192,18 +167,58 @@ void GPU_CACHED_MANAGER::EndDrawing()
|
|||
(GLvoid*) SHADER_OFFSET );
|
||||
}
|
||||
|
||||
glBindBuffer( GL_ELEMENT_ARRAY_BUFFER, m_indicesBuffer );
|
||||
glBufferData( GL_ELEMENT_ARRAY_BUFFER, m_indicesSize * sizeof( int ), (GLvoid*) m_indices.get(),
|
||||
GL_DYNAMIC_DRAW );
|
||||
PROF_COUNTER cntDraw( "gl-draw-elements" );
|
||||
|
||||
glDrawElements( GL_TRIANGLES, m_indicesSize, GL_UNSIGNED_INT, nullptr );
|
||||
int n_ranges = m_vranges.size();
|
||||
int n = 0;
|
||||
GLuint* iptr = m_indices.get();
|
||||
GLuint icnt = 0;
|
||||
|
||||
#ifdef KICAD_GAL_PROFILE
|
||||
wxLogTrace( traceGalProfile, wxT( "Cached manager size: %d" ), m_indicesSize );
|
||||
#endif /* KICAD_GAL_PROFILE */
|
||||
int drawCalls = 0;
|
||||
|
||||
while( n < n_ranges )
|
||||
{
|
||||
VRANGE* cur = &m_vranges[n];
|
||||
|
||||
if( cur->m_isContinuous )
|
||||
{
|
||||
if( icnt > 0 )
|
||||
{
|
||||
glDrawElements( GL_TRIANGLES, icnt, GL_UNSIGNED_INT, m_indices.get() );
|
||||
drawCalls++;
|
||||
}
|
||||
|
||||
icnt = 0;
|
||||
iptr = m_indices.get();
|
||||
|
||||
glDrawArrays( GL_TRIANGLES, cur->m_start, cur->m_end - cur->m_start + 1 );
|
||||
drawCalls++;
|
||||
}
|
||||
else
|
||||
{
|
||||
for( GLuint i = cur->m_start; i <= cur->m_end; i++ )
|
||||
{
|
||||
*iptr++ = i;
|
||||
icnt++;
|
||||
}
|
||||
}
|
||||
n++;
|
||||
}
|
||||
|
||||
if( icnt > 0 )
|
||||
{
|
||||
glDrawElements( GL_TRIANGLES, icnt, GL_UNSIGNED_INT, m_indices.get() );
|
||||
drawCalls++;
|
||||
}
|
||||
|
||||
cntDraw.Stop();
|
||||
|
||||
KI_TRACE( traceGalProfile,
|
||||
"Cached manager size: VBO size %u iranges %llu max elt size %u drawcalls %u\n",
|
||||
cached->AllItemsSize(), m_vranges.size(), m_indexBufMaxSize, drawCalls );
|
||||
KI_TRACE( traceGalProfile, "Timing: %s\n", cntDraw.to_string() );
|
||||
|
||||
glBindBuffer( GL_ARRAY_BUFFER, 0 );
|
||||
glBindBuffer( GL_ELEMENT_ARRAY_BUFFER, 0 );
|
||||
cached->ClearDirty();
|
||||
|
||||
// Deactivate vertex array
|
||||
|
|
@ -217,12 +232,6 @@ void GPU_CACHED_MANAGER::EndDrawing()
|
|||
}
|
||||
|
||||
m_isDrawing = false;
|
||||
|
||||
#ifdef KICAD_GAL_PROFILE
|
||||
totalRealTime.Stop();
|
||||
wxLogTrace( traceGalProfile, wxT( "GPU_CACHED_MANAGER::EndDrawing(): %.1f ms" ),
|
||||
totalRealTime.msecs() );
|
||||
#endif /* KICAD_GAL_PROFILE */
|
||||
}
|
||||
|
||||
|
||||
|
|
@ -249,19 +258,12 @@ void GPU_NONCACHED_MANAGER::BeginDrawing()
|
|||
}
|
||||
|
||||
|
||||
void GPU_NONCACHED_MANAGER::DrawIndices( unsigned int aOffset, unsigned int aSize )
|
||||
void GPU_NONCACHED_MANAGER::DrawIndices( const VERTEX_ITEM* aItem )
|
||||
{
|
||||
wxASSERT_MSG( false, wxT( "Not implemented yet" ) );
|
||||
}
|
||||
|
||||
|
||||
void GPU_NONCACHED_MANAGER::DrawAll()
|
||||
{
|
||||
// This is the default use case, nothing has to be done
|
||||
// The real rendering takes place in the EndDrawing() function
|
||||
}
|
||||
|
||||
|
||||
void GPU_NONCACHED_MANAGER::EndDrawing()
|
||||
{
|
||||
#ifdef KICAD_GAL_PROFILE
|
||||
|
|
|
|||
|
|
@ -243,10 +243,7 @@ void VERTEX_MANAGER::BeginDrawing() const
|
|||
|
||||
void VERTEX_MANAGER::DrawItem( const VERTEX_ITEM& aItem ) const
|
||||
{
|
||||
int size = aItem.GetSize();
|
||||
int offset = aItem.GetOffset();
|
||||
|
||||
m_gpu->DrawIndices( offset, size );
|
||||
m_gpu->DrawIndices( &aItem );
|
||||
}
|
||||
|
||||
|
||||
|
|
|
|||
|
|
@ -27,6 +27,7 @@
|
|||
#ifndef GPU_MANAGER_H_
|
||||
#define GPU_MANAGER_H_
|
||||
|
||||
#include <vector>
|
||||
#include <gal/opengl/vertex_common.h>
|
||||
#include <boost/scoped_array.hpp>
|
||||
|
||||
|
|
@ -34,6 +35,7 @@ namespace KIGFX
|
|||
{
|
||||
class SHADER;
|
||||
class VERTEX_CONTAINER;
|
||||
class VERTEX_ITEM;
|
||||
class CACHED_CONTAINER;
|
||||
class NONCACHED_CONTAINER;
|
||||
|
||||
|
|
@ -58,12 +60,7 @@ public:
|
|||
* @param aOffset is the beginning of the range.
|
||||
* @param aSize is the number of vertices to be drawn.
|
||||
*/
|
||||
virtual void DrawIndices( unsigned int aOffset, unsigned int aSize ) = 0;
|
||||
|
||||
/**
|
||||
* Make the GPU draw all the vertices stored in the container.
|
||||
*/
|
||||
virtual void DrawAll() = 0;
|
||||
virtual void DrawIndices( const VERTEX_ITEM* aItem ) = 0;
|
||||
|
||||
/**
|
||||
* Clear the container after drawing routines.
|
||||
|
|
@ -105,6 +102,19 @@ protected:
|
|||
class GPU_CACHED_MANAGER : public GPU_MANAGER
|
||||
{
|
||||
public:
|
||||
|
||||
struct VRANGE
|
||||
{
|
||||
VRANGE( int aStart, int aEnd, bool aContinuous ) :
|
||||
m_start( aStart ),
|
||||
m_end( aEnd ),
|
||||
m_isContinuous ( aContinuous )
|
||||
{}
|
||||
unsigned int m_start, m_end;
|
||||
bool m_isContinuous;
|
||||
};
|
||||
|
||||
|
||||
GPU_CACHED_MANAGER( VERTEX_CONTAINER* aContainer );
|
||||
~GPU_CACHED_MANAGER();
|
||||
|
||||
|
|
@ -112,10 +122,7 @@ public:
|
|||
virtual void BeginDrawing() override;
|
||||
|
||||
///< @copydoc GPU_MANAGER::DrawIndices()
|
||||
virtual void DrawIndices( unsigned int aOffset, unsigned int aSize ) override;
|
||||
|
||||
///< @copydoc GPU_MANAGER::DrawAll()
|
||||
virtual void DrawAll() override;
|
||||
virtual void DrawIndices( const VERTEX_ITEM* aItem ) override;
|
||||
|
||||
///< @copydoc GPU_MANAGER::EndDrawing()
|
||||
virtual void EndDrawing() override;
|
||||
|
|
@ -136,17 +143,26 @@ protected:
|
|||
///< Pointer to the current indices buffer
|
||||
boost::scoped_array<GLuint> m_indices;
|
||||
|
||||
///< Pointer to the first free cell in the indices buffer
|
||||
GLuint* m_indicesPtr;
|
||||
|
||||
///< Handle to indices buffer
|
||||
GLuint m_indicesBuffer;
|
||||
|
||||
///< Number of indices stored in the indices buffer
|
||||
unsigned int m_indicesSize;
|
||||
|
||||
///< Current indices buffer size
|
||||
unsigned int m_indicesCapacity;
|
||||
|
||||
///< Ranges of visible vertex indices to render
|
||||
std::vector<VRANGE> m_vranges;
|
||||
|
||||
///< Number of huge VRANGEs (i.e. large zones) with separate draw calls
|
||||
int m_totalHuge;
|
||||
|
||||
///< Number of regular VRANGEs (small items) pooled into single draw call
|
||||
int m_totalNormal;
|
||||
|
||||
///< Current size of index buffer
|
||||
unsigned int m_indexBufSize;
|
||||
|
||||
///< Maximum size taken by the index buffer for all frames rendered so far
|
||||
unsigned int m_indexBufMaxSize;
|
||||
|
||||
///< Size of the current VRANGE
|
||||
unsigned int m_curVrangeSize;
|
||||
};
|
||||
|
||||
|
||||
|
|
@ -159,10 +175,7 @@ public:
|
|||
virtual void BeginDrawing() override;
|
||||
|
||||
///< @copydoc GPU_MANAGER::DrawIndices()
|
||||
virtual void DrawIndices( unsigned int aOffset, unsigned int aSize ) override;
|
||||
|
||||
///< @copydoc GPU_MANAGER::DrawAll()
|
||||
virtual void DrawAll() override;
|
||||
virtual void DrawIndices( const VERTEX_ITEM* aItem ) override;
|
||||
|
||||
///< @copydoc GPU_MANAGER::EndDrawing()
|
||||
virtual void EndDrawing() override;
|
||||
|
|
|
|||
Loading…
Reference in New Issue