You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
630 lines
20 KiB
630 lines
20 KiB
//========= Copyright 1996-2009, Valve Corporation, All rights reserved. ============//
|
|
//
|
|
// Purpose: Defines a group of app systems that all have the same lifetime
|
|
// that need to be connected/initialized, etc. in a well-defined order
|
|
//
|
|
// $Revision: $
|
|
// $NoKeywords: $
|
|
//=============================================================================//
|
|
|
|
//===============================================================================
|
|
|
|
GLMRendererInfo::GLMRendererInfo( void )
|
|
{
|
|
m_display = NULL;
|
|
Q_memset( &m_info, 0, sizeof( m_info ) );
|
|
}
|
|
|
|
GLMRendererInfo::~GLMRendererInfo( void )
|
|
{
|
|
SDLAPP_FUNC;
|
|
|
|
if (m_display)
|
|
{
|
|
delete m_display;
|
|
m_display = NULL;
|
|
}
|
|
}
|
|
|
|
// !!! FIXME: sync this function with the Mac version in case anything important has changed.
|
|
void GLMRendererInfo::Init( GLMRendererInfoFields *info )
|
|
{
|
|
SDLAPP_FUNC;
|
|
|
|
m_info = *info;
|
|
m_display = NULL;
|
|
|
|
m_info.m_fullscreen = 0;
|
|
m_info.m_accelerated = 1;
|
|
m_info.m_windowed = 1;
|
|
|
|
m_info.m_ati = true;
|
|
m_info.m_atiNewer = true;
|
|
|
|
m_info.m_hasGammaWrites = true;
|
|
|
|
// If you haven't created a GL context by now (and initialized gGL), you're about to crash.
|
|
|
|
m_info.m_hasMixedAttachmentSizes = gGL->m_bHave_GL_ARB_framebuffer_object;
|
|
m_info.m_hasBGRA = gGL->m_bHave_GL_EXT_vertex_array_bgra;
|
|
|
|
// !!! FIXME: what do these do on the Mac?
|
|
m_info.m_hasNewFullscreenMode = false;
|
|
m_info.m_hasNativeClipVertexMode = true;
|
|
|
|
// if user disabled them
|
|
if (CommandLine()->FindParm("-glmdisableclipplanes"))
|
|
{
|
|
m_info.m_hasNativeClipVertexMode = false;
|
|
}
|
|
|
|
// or maybe enabled them..
|
|
if (CommandLine()->FindParm("-glmenableclipplanes"))
|
|
{
|
|
m_info.m_hasNativeClipVertexMode = true;
|
|
}
|
|
|
|
m_info.m_hasOcclusionQuery = gGL->m_bHave_GL_ARB_occlusion_query;
|
|
m_info.m_hasFramebufferBlit = gGL->m_bHave_GL_EXT_framebuffer_blit || gGL->m_bHave_GL_ARB_framebuffer_object;
|
|
|
|
GLint nMaxAniso = 0;
|
|
gGL->glGetIntegerv( GL_MAX_TEXTURE_MAX_ANISOTROPY_EXT, &nMaxAniso );
|
|
m_info.m_maxAniso = clamp<int>( nMaxAniso, 0, 16 );
|
|
|
|
// We don't currently used bindable uniforms, but I've been experimenting with them so I might as well check this in just in case they turn out to be useful.
|
|
m_info.m_hasBindableUniforms = gGL->m_bHave_GL_EXT_bindable_uniform;
|
|
m_info.m_hasBindableUniforms = false; // !!! FIXME hardwiring this path to false until we see how to accelerate it properly
|
|
m_info.m_maxVertexBindableUniforms = 0;
|
|
m_info.m_maxFragmentBindableUniforms = 0;
|
|
m_info.m_maxBindableUniformSize = 0;
|
|
|
|
if (m_info.m_hasBindableUniforms)
|
|
{
|
|
gGL->glGetIntegerv(GL_MAX_VERTEX_BINDABLE_UNIFORMS_EXT, &m_info.m_maxVertexBindableUniforms);
|
|
gGL->glGetIntegerv(GL_MAX_FRAGMENT_BINDABLE_UNIFORMS_EXT, &m_info.m_maxFragmentBindableUniforms);
|
|
gGL->glGetIntegerv(GL_MAX_BINDABLE_UNIFORM_SIZE_EXT, &m_info.m_maxBindableUniformSize);
|
|
if ( ( m_info.m_maxVertexBindableUniforms < 1 ) || ( m_info.m_maxFragmentBindableUniforms < 1 ) || ( m_info.m_maxBindableUniformSize < ( sizeof( float ) * 4 * 256 ) ) )
|
|
{
|
|
m_info.m_hasBindableUniforms = false;
|
|
}
|
|
}
|
|
|
|
m_info.m_hasUniformBuffers = gGL->m_bHave_GL_ARB_uniform_buffer;
|
|
m_info.m_hasPerfPackage1 = true; // this flag is Mac-specific. We do slower things if you don't have Mac OS X 10.x.y or later. Linux always does the fast path!
|
|
|
|
//-------------------------------------------------------------------
|
|
// runtime options that aren't negotiable once set
|
|
|
|
m_info.m_hasDualShaders = CommandLine()->FindParm("-glmdualshaders") != 0;
|
|
|
|
//-------------------------------------------------------------------
|
|
// "can'ts "
|
|
|
|
#if defined( OSX )
|
|
m_info.m_cantBlitReliably = m_info.m_intel; //FIXME X3100&10.6.3 has problems blitting.. adjust this if bug fixed in 10.6.4
|
|
#else
|
|
// m_cantBlitReliably path doesn't work right now, and the Intel path is different for us on Linux/Win7 anyway
|
|
m_info.m_cantBlitReliably = false;
|
|
#endif
|
|
|
|
if (CommandLine()->FindParm("-glmenabletrustblit"))
|
|
{
|
|
m_info.m_cantBlitReliably = false; // we trust the blit, so set the cant-blit cap to false
|
|
}
|
|
if (CommandLine()->FindParm("-glmdisabletrustblit"))
|
|
{
|
|
m_info.m_cantBlitReliably = true; // we do not trust the blit, so set the cant-blit cap to true
|
|
}
|
|
|
|
// MSAA resolve issues
|
|
m_info.m_cantResolveFlipped = false;
|
|
|
|
|
|
#if defined( OSX )
|
|
m_info.m_cantResolveScaled = true; // generally true until new extension ships
|
|
#else
|
|
// DON'T just slam this to false and run without first testing with -gl_debug enabled on NVidia/AMD/etc.
|
|
// This path needs the m_bHave_GL_EXT_framebuffer_multisample_blit_scaled extension.
|
|
m_info.m_cantResolveScaled = true;
|
|
|
|
if ( gGL->m_bHave_GL_EXT_framebuffer_multisample_blit_scaled )
|
|
{
|
|
m_info.m_cantResolveScaled = false;
|
|
}
|
|
#endif
|
|
|
|
// gamma decode impacting shader codegen
|
|
m_info.m_costlyGammaFlips = false;
|
|
}
|
|
|
|
void GLMRendererInfo::PopulateDisplays()
|
|
{
|
|
SDLAPP_FUNC;
|
|
|
|
Assert( !m_display );
|
|
m_display = new GLMDisplayInfo;
|
|
|
|
// Populate display mode table.
|
|
m_display->PopulateModes();
|
|
}
|
|
|
|
|
|
void GLMRendererInfo::Dump( int which )
|
|
{
|
|
SDLAPP_FUNC;
|
|
|
|
GLMPRINTF(("\n #%d: GLMRendererInfo @ %p, renderer-id=(%08x) display-mask=%08x vram=%dMB",
|
|
which, this,
|
|
m_info.m_rendererID,
|
|
m_info.m_displayMask,
|
|
m_info.m_vidMemory >> 20
|
|
));
|
|
GLMPRINTF(("\n VendorID=%04x DeviceID=%04x Model=%s",
|
|
m_info.m_pciVendorID,
|
|
m_info.m_pciDeviceID,
|
|
m_info.m_pciModelString
|
|
));
|
|
|
|
m_display->Dump( which );
|
|
}
|
|
|
|
|
|
|
|
|
|
GLMDisplayDB::GLMDisplayDB ()
|
|
{
|
|
SDLAPP_FUNC;
|
|
|
|
m_renderer.m_display = NULL;
|
|
}
|
|
|
|
GLMDisplayDB::~GLMDisplayDB ( void )
|
|
{
|
|
SDLAPP_FUNC;
|
|
|
|
if ( m_renderer.m_display )
|
|
{
|
|
delete m_renderer.m_display;
|
|
m_renderer.m_display = NULL;
|
|
}
|
|
}
|
|
|
|
#ifndef GL_GPU_MEMORY_INFO_DEDICATED_VIDMEM_NVX
|
|
#define GL_GPU_MEMORY_INFO_DEDICATED_VIDMEM_NVX 0x9047
|
|
#endif
|
|
|
|
#ifndef GL_GPU_MEMORY_INFO_TOTAL_AVAILABLE_MEMORY_NVX
|
|
#define GL_GPU_MEMORY_INFO_TOTAL_AVAILABLE_MEMORY_NVX 0x9048
|
|
#endif
|
|
|
|
#ifndef GL_GPU_MEMORY_INFO_CURRENT_AVAILABLE_VIDMEM_NVX
|
|
#define GL_GPU_MEMORY_INFO_CURRENT_AVAILABLE_VIDMEM_NVX 0x9049
|
|
#endif
|
|
|
|
#ifndef GL_VBO_FREE_MEMORY_ATI
|
|
#define GL_VBO_FREE_MEMORY_ATI 0x87FB
|
|
#endif
|
|
|
|
#ifndef GL_TEXTURE_FREE_MEMORY_ATI
|
|
#define GL_TEXTURE_FREE_MEMORY_ATI 0x87FC
|
|
#endif
|
|
|
|
#ifndef GL_RENDERBUFFER_FREE_MEMORY_ATI
|
|
#define GL_RENDERBUFFER_FREE_MEMORY_ATI 0x87FD
|
|
#endif
|
|
|
|
void GLMDisplayDB::PopulateRenderers( void )
|
|
{
|
|
SDLAPP_FUNC;
|
|
|
|
Assert( !m_renderer.m_display );
|
|
|
|
GLMRendererInfoFields fields;
|
|
memset( &fields, 0, sizeof(fields) );
|
|
|
|
// Assume 512MB of available video memory
|
|
fields.m_vidMemory = 512 * 1024 * 1024;
|
|
|
|
DebugPrintf( "GL_NVX_gpu_memory_info: %s\n", gGL->m_bHave_GL_NVX_gpu_memory_info ? "AVAILABLE" : "UNAVAILABLE" );
|
|
DebugPrintf( "GL_ATI_meminfo: %s\n", gGL->m_bHave_GL_ATI_meminfo ? "AVAILABLE" : "UNAVAILABLE" );
|
|
|
|
if ( gGL->m_bHave_GL_NVX_gpu_memory_info )
|
|
{
|
|
gGL->glGetError();
|
|
|
|
GLint nTotalDedicated = 0, nTotalAvail = 0, nCurrentAvail = 0;
|
|
gGL->glGetIntegerv( GL_GPU_MEMORY_INFO_DEDICATED_VIDMEM_NVX, &nTotalDedicated );
|
|
gGL->glGetIntegerv( GL_GPU_MEMORY_INFO_TOTAL_AVAILABLE_MEMORY_NVX, &nTotalAvail );
|
|
gGL->glGetIntegerv( GL_GPU_MEMORY_INFO_CURRENT_AVAILABLE_VIDMEM_NVX, &nCurrentAvail );
|
|
|
|
if ( gGL->glGetError() )
|
|
{
|
|
DebugPrintf( "GL_NVX_gpu_memory_info: Failed retrieving available GPU memory\n" );
|
|
}
|
|
else
|
|
{
|
|
DebugPrintf( "GL_NVX_gpu_memory_info: Total Dedicated: %u, Total Avail: %u, Current Avail: %u\n", nTotalDedicated, nTotalAvail, nCurrentAvail );
|
|
|
|
// Try to do something reasonable. Should we report dedicated or total available to the engine here?
|
|
// For now, just take the MAX of both.
|
|
uint64 nActualAvail = static_cast<uint64>( MAX( nTotalAvail, nTotalDedicated ) ) * 1024;
|
|
fields.m_vidMemory = static_cast< GLint >( MIN( nActualAvail, 0x7FFFFFFF ) );
|
|
}
|
|
}
|
|
else if ( gGL->m_bHave_GL_ATI_meminfo )
|
|
{
|
|
// As of 10/8/12 this extension is only available under Linux and Windows FireGL parts.
|
|
gGL->glGetError();
|
|
|
|
GLint nAvail[4] = { 0, 0, 0, 0 };
|
|
gGL->glGetIntegerv( GL_TEXTURE_FREE_MEMORY_ATI, nAvail );
|
|
|
|
if ( gGL->glGetError() )
|
|
{
|
|
DebugPrintf( "GL_ATI_meminfo: Failed retrieving available GPU memory\n" );
|
|
}
|
|
else
|
|
{
|
|
// param[0] - total memory free in the pool
|
|
// param[1] - largest available free block in the pool
|
|
// param[2] - total auxiliary memory free
|
|
// param[3] - largest auxiliary free block
|
|
|
|
DebugPrintf( "GL_ATI_meminfo: GL_TEXTURE_FREE_MEMORY_ATI: Total Free: %i, Largest Avail: %i, Total Aux: %i, Largest Aux Avail: %i\n",
|
|
nAvail[0], nAvail[1], nAvail[2], nAvail[3] );
|
|
|
|
uint64 nActualAvail = static_cast<uint64>( nAvail[0] ) * 1024;
|
|
fields.m_vidMemory = static_cast< GLint >( MIN( nActualAvail, 0x7FFFFFFF ) );
|
|
}
|
|
}
|
|
|
|
// Clamp the min amount of video memory to 256MB in case a query returned something bogus, or we interpreted it badly.
|
|
fields.m_vidMemory = MAX( fields.m_vidMemory, 128 * 1024 * 1024 );
|
|
fields.m_texMemory = fields.m_vidMemory;
|
|
|
|
fields.m_pciVendorID = GLM_OPENGL_VENDOR_ID;
|
|
fields.m_pciDeviceID = GLM_OPENGL_DEFAULT_DEVICE_ID;
|
|
if ( ( gGL->m_nDriverProvider == cGLDriverProviderIntel ) || ( gGL->m_nDriverProvider == cGLDriverProviderIntelOpenSource ) )
|
|
{
|
|
fields.m_pciDeviceID = GLM_OPENGL_LOW_PERF_DEVICE_ID;
|
|
}
|
|
|
|
/* fields.m_colorModes = (uint)-1;
|
|
fields.m_bufferModes = (uint)-1;
|
|
fields.m_depthModes = (uint)-1;
|
|
fields.m_stencilModes = (uint)-1;
|
|
fields.m_maxAuxBuffers = (uint)128;
|
|
fields.m_maxSampleBuffers = (uint)128;
|
|
fields.m_maxSamples = (uint)2048;
|
|
fields.m_sampleModes = (uint)128;
|
|
fields.m_sampleAlpha = (uint)32;
|
|
*/
|
|
|
|
GLint nMaxMultiSamples = 0;
|
|
gGL->glGetIntegerv( GL_MAX_SAMPLES_EXT, &nMaxMultiSamples );
|
|
fields.m_maxSamples = clamp<int>( nMaxMultiSamples, 0, 8 );
|
|
DebugPrintf( "GL_MAX_SAMPLES_EXT: %i\n", nMaxMultiSamples );
|
|
|
|
// We only have one GLMRendererInfo on Linux, unlike Mac OS X. Whatever libGL.so wants to do, we go with it.
|
|
m_renderer.Init( &fields );
|
|
|
|
// then go back and ask each renderer to populate its display info table.
|
|
m_renderer.PopulateDisplays();
|
|
}
|
|
|
|
|
|
|
|
void GLMDisplayDB::PopulateFakeAdapters( uint realRendererIndex ) // fake adapters = one real adapter times however many displays are on it
|
|
{
|
|
SDLAPP_FUNC;
|
|
|
|
Assert( realRendererIndex == 0 );
|
|
}
|
|
|
|
void GLMDisplayDB::Populate(void)
|
|
{
|
|
SDLAPP_FUNC;
|
|
|
|
this->PopulateRenderers();
|
|
|
|
this->PopulateFakeAdapters( 0 );
|
|
|
|
#if GLMDEBUG
|
|
this->Dump();
|
|
#endif
|
|
}
|
|
|
|
|
|
|
|
int GLMDisplayDB::GetFakeAdapterCount( void )
|
|
{
|
|
SDLAPP_FUNC;
|
|
|
|
return 1;
|
|
}
|
|
|
|
bool GLMDisplayDB::GetFakeAdapterInfo( int fakeAdapterIndex, int *rendererOut, int *displayOut, GLMRendererInfoFields *rendererInfoOut, GLMDisplayInfoFields *displayInfoOut )
|
|
{
|
|
SDLAPP_FUNC;
|
|
|
|
if (fakeAdapterIndex >= GetFakeAdapterCount() )
|
|
{
|
|
*rendererOut = 0;
|
|
*displayOut = 0;
|
|
return true; // fail
|
|
}
|
|
|
|
*rendererOut = 0;
|
|
*displayOut = 0;
|
|
|
|
bool rendResult = GetRendererInfo( *rendererOut, rendererInfoOut );
|
|
bool dispResult = GetDisplayInfo( *rendererOut, *displayOut, displayInfoOut );
|
|
|
|
return rendResult || dispResult;
|
|
}
|
|
|
|
|
|
int GLMDisplayDB::GetRendererCount( void )
|
|
{
|
|
SDLAPP_FUNC;
|
|
|
|
return 1;
|
|
}
|
|
|
|
bool GLMDisplayDB::GetRendererInfo( int rendererIndex, GLMRendererInfoFields *infoOut )
|
|
{
|
|
SDLAPP_FUNC;
|
|
|
|
memset( infoOut, 0, sizeof( GLMRendererInfoFields ) );
|
|
|
|
if (rendererIndex >= GetRendererCount())
|
|
return true; // fail
|
|
|
|
*infoOut = m_renderer.m_info;
|
|
|
|
return false;
|
|
}
|
|
|
|
int GLMDisplayDB::GetDisplayCount( int rendererIndex )
|
|
{
|
|
SDLAPP_FUNC;
|
|
|
|
if (rendererIndex >= GetRendererCount())
|
|
{
|
|
Assert( 0 );
|
|
return 0; // fail
|
|
}
|
|
|
|
return 1;
|
|
}
|
|
|
|
bool GLMDisplayDB::GetDisplayInfo( int rendererIndex, int displayIndex, GLMDisplayInfoFields *infoOut )
|
|
{
|
|
SDLAPP_FUNC;
|
|
|
|
memset( infoOut, 0, sizeof( GLMDisplayInfoFields ) );
|
|
|
|
if (rendererIndex >= GetRendererCount())
|
|
return true; // fail
|
|
|
|
if (displayIndex >= GetDisplayCount(rendererIndex))
|
|
return true; // fail
|
|
|
|
*infoOut = m_renderer.m_display->m_info;
|
|
|
|
return false;
|
|
}
|
|
|
|
int GLMDisplayDB::GetModeCount( int rendererIndex, int displayIndex )
|
|
{
|
|
SDLAPP_FUNC;
|
|
|
|
if (rendererIndex >= GetRendererCount())
|
|
return 0; // fail
|
|
|
|
if (displayIndex >= GetDisplayCount(rendererIndex))
|
|
return 0; // fail
|
|
|
|
return m_renderer.m_display->m_modes->Count();
|
|
}
|
|
|
|
bool GLMDisplayDB::GetModeInfo( int rendererIndex, int displayIndex, int modeIndex, GLMDisplayModeInfoFields *infoOut )
|
|
{
|
|
SDLAPP_FUNC;
|
|
|
|
memset( infoOut, 0, sizeof( GLMDisplayModeInfoFields ) );
|
|
|
|
if ( rendererIndex >= GetRendererCount())
|
|
return true; // fail
|
|
|
|
if (displayIndex >= GetDisplayCount( rendererIndex ) )
|
|
return true; // fail
|
|
|
|
if ( modeIndex >= GetModeCount( rendererIndex, displayIndex ) )
|
|
return true; // fail
|
|
|
|
if ( modeIndex >= 0 )
|
|
{
|
|
GLMDisplayMode *displayModeInfo = m_renderer.m_display->m_modes->Element( modeIndex );
|
|
|
|
*infoOut = displayModeInfo->m_info;
|
|
}
|
|
else
|
|
{
|
|
const GLMDisplayInfoFields &info = m_renderer.m_display->m_info;
|
|
|
|
infoOut->m_modePixelWidth = info.m_displayPixelWidth;
|
|
infoOut->m_modePixelHeight = info.m_displayPixelHeight;
|
|
infoOut->m_modeRefreshHz = 0;
|
|
|
|
//return true; // fail
|
|
}
|
|
|
|
return false;
|
|
}
|
|
|
|
|
|
void GLMDisplayDB::Dump( void )
|
|
{
|
|
SDLAPP_FUNC;
|
|
|
|
GLMPRINTF(("\n GLMDisplayDB @ %p ",this ));
|
|
|
|
m_renderer.Dump( 0 );
|
|
}
|
|
|
|
//===============================================================================
|
|
|
|
GLMDisplayInfo::GLMDisplayInfo()
|
|
{
|
|
SDLAPP_FUNC;
|
|
|
|
m_modes = NULL;
|
|
|
|
int Width, Height;
|
|
GetLargestDisplaySize( Width, Height );
|
|
|
|
m_info.m_displayPixelWidth = ( uint )Width;
|
|
m_info.m_displayPixelHeight = ( uint )Height;
|
|
}
|
|
|
|
GLMDisplayInfo::~GLMDisplayInfo( void )
|
|
{
|
|
SDLAPP_FUNC;
|
|
}
|
|
|
|
extern "C" int DisplayModeSortFunction( GLMDisplayMode * const *A, GLMDisplayMode * const *B )
|
|
{
|
|
SDLAPP_FUNC;
|
|
|
|
int bigger = -1;
|
|
int smaller = 1; // adjust these for desired ordering
|
|
|
|
// check refreshrate - higher should win
|
|
if ( (*A)->m_info.m_modeRefreshHz > (*B)->m_info.m_modeRefreshHz )
|
|
{
|
|
return bigger;
|
|
}
|
|
else if ( (*A)->m_info.m_modeRefreshHz < (*B)->m_info.m_modeRefreshHz )
|
|
{
|
|
return smaller;
|
|
}
|
|
|
|
// check area - larger mode should win
|
|
int areaa = (*A)->m_info.m_modePixelWidth * (*A)->m_info.m_modePixelHeight;
|
|
int areab = (*B)->m_info.m_modePixelWidth * (*B)->m_info.m_modePixelHeight;
|
|
|
|
if ( areaa > areab )
|
|
{
|
|
return bigger;
|
|
}
|
|
else if ( areaa < areab )
|
|
{
|
|
return smaller;
|
|
}
|
|
|
|
return 0; // equal rank
|
|
}
|
|
|
|
|
|
void GLMDisplayInfo::PopulateModes( void )
|
|
{
|
|
SDLAPP_FUNC;
|
|
|
|
Assert( !m_modes );
|
|
m_modes = new CUtlVector< GLMDisplayMode* >;
|
|
|
|
int nummodes = SDL_GetNumVideoDisplays();
|
|
|
|
for ( int i = 0; i < nummodes; i++ )
|
|
{
|
|
SDL_Rect rect = { 0, 0, 0, 0 };
|
|
|
|
if ( !SDL_GetDisplayBounds( i, &rect ) && rect.w && rect.h )
|
|
{
|
|
m_modes->AddToTail( new GLMDisplayMode( rect.w, rect.h, 0 ) );
|
|
}
|
|
}
|
|
|
|
// Add a big pile of window resolutions.
|
|
static const struct
|
|
{
|
|
uint w;
|
|
uint h;
|
|
} s_Resolutions[] =
|
|
{
|
|
{ 640, 480 }, // 4x3
|
|
{ 800, 600 },
|
|
{ 1024, 768 },
|
|
{ 1152, 864 },
|
|
{ 1280, 960 },
|
|
{ 1600, 1200 },
|
|
{ 1920, 1440 },
|
|
{ 2048, 1536 },
|
|
|
|
{ 1280, 720 }, // 16x9
|
|
{ 1366, 768 },
|
|
{ 1600, 900 },
|
|
{ 1920, 1080 },
|
|
|
|
{ 720, 480 }, // 16x10
|
|
{ 1280, 800 },
|
|
{ 1680, 1050 },
|
|
{ 1920, 1200 },
|
|
{ 2560, 1600 },
|
|
};
|
|
|
|
for ( int i = 0; i < ARRAYSIZE( s_Resolutions ); i++ )
|
|
{
|
|
uint w = s_Resolutions[ i ].w;
|
|
uint h = s_Resolutions[ i ].h;
|
|
|
|
if ( ( w <= m_info.m_displayPixelWidth ) && ( h <= m_info.m_displayPixelHeight ) )
|
|
{
|
|
m_modes->AddToTail( new GLMDisplayMode( w, h, 0 ) );
|
|
|
|
if ( ( w * 2 <= m_info.m_displayPixelWidth ) && ( h * 2 < m_info.m_displayPixelHeight ) )
|
|
{
|
|
// Add double of everything also - Retina proofing hopefully.
|
|
m_modes->AddToTail( new GLMDisplayMode( w * 2, h * 2, 0 ) );
|
|
}
|
|
}
|
|
}
|
|
|
|
m_modes->Sort( DisplayModeSortFunction );
|
|
|
|
// remove dupes.
|
|
nummodes = m_modes->Count();
|
|
int i = 1; // not zero!
|
|
while (i < nummodes)
|
|
{
|
|
GLMDisplayModeInfoFields& info0 = m_modes->Element( i - 1 )->m_info;
|
|
GLMDisplayModeInfoFields& info1 = m_modes->Element( i )->m_info;
|
|
|
|
if ( ( info0.m_modePixelWidth == info1.m_modePixelWidth ) &&
|
|
( info0.m_modePixelHeight == info1.m_modePixelHeight ) &&
|
|
( info0.m_modeRefreshHz == info1.m_modeRefreshHz ) )
|
|
{
|
|
m_modes->Remove(i);
|
|
nummodes--;
|
|
}
|
|
else
|
|
{
|
|
i++;
|
|
}
|
|
}
|
|
}
|
|
|
|
|
|
void GLMDisplayInfo::Dump( int which )
|
|
{
|
|
SDLAPP_FUNC;
|
|
|
|
GLMPRINTF(("\n #%d: GLMDisplayInfo @ %08x, pixwidth=%d pixheight=%d",
|
|
which, (int)this, m_info.m_displayPixelWidth, m_info.m_displayPixelHeight ));
|
|
|
|
FOR_EACH_VEC( *m_modes, i )
|
|
{
|
|
( *m_modes )[i]->Dump(i);
|
|
}
|
|
}
|