Counter Strike : Global Offensive Source Code
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.
 
 
 
 
 
 

1420 lines
39 KiB

//============ Copyright (c) Valve Corporation, All rights reserved. ============
//
// cglmprogram.cpp
//
//===============================================================================
#include "glmgr/glmgr.h"
#include "glmgr/cglmprogram.h"
#include "filesystem.h"
#include "tier1/keyvalues.h"
// memdbgon -must- be the last include file in a .cpp file.
#include "tier0/memdbgon.h"
//===============================================================================
ConVar gl_shaderpair_cacherows_lg2( "gl_paircache_rows_lg2", "10"); // 10 is minimum
ConVar gl_shaderpair_cacheways_lg2( "gl_paircache_ways_lg2", "5"); // 5 is minimum
ConVar gl_shaderpair_cachelog( "gl_shaderpair_cachelog", "0" );
//===============================================================================
GLenum GLMProgTypeToARBEnum( EGLMProgramType type )
{
GLenum result = 0;
switch(type)
{
case kGLMVertexProgram: result = GL_VERTEX_PROGRAM_ARB; break;
case kGLMFragmentProgram: result = GL_FRAGMENT_PROGRAM_ARB; break;
default: Assert( !"bad program type"); result = 0; break;
}
return result;
}
GLenum GLMProgTypeToGLSLEnum( EGLMProgramType type )
{
GLenum result = 0;
switch(type)
{
case kGLMVertexProgram: result = GL_VERTEX_SHADER_ARB; break;
case kGLMFragmentProgram: result = GL_FRAGMENT_SHADER_ARB; break;
default: Assert( !"bad program type"); result = 0; break;
}
return result;
}
static uint g_shader_serial = 1000000;
CGLMProgram::CGLMProgram( GLMContext *ctx, EGLMProgramType type )
{
m_ctx = ctx;
m_ctx->CheckCurrent();
m_type = type;
m_serial = g_shader_serial++;
m_text = NULL; // no text yet
#if GLMDEBUG
m_editable = NULL;
#endif
memset( &m_descs, 0, sizeof( m_descs ) );
m_samplerMask = 0; // dxabstract sets this field later
// create an ARB vp/fp program object name. No need to bind it yet.
GLMShaderDesc *arbDesc = &m_descs[ kGLMARB ];
glGenProgramsARB( 1, &arbDesc->m_object.arb );
// create a GLSL shader object.
GLMShaderDesc *glslDesc = &m_descs[ kGLMGLSL ];
GLenum glslStage = GLMProgTypeToGLSLEnum( m_type );
glslDesc->m_object.glsl = glCreateShaderObjectARB( glslStage );;
// no text has arrived yet. That's done in SetProgramText.
}
CGLMProgram::~CGLMProgram( )
{
m_ctx->CheckCurrent();
// if there is an arb program, delete it
GLMShaderDesc *arbDesc = &m_descs[ kGLMARB ];
if (arbDesc->m_object.arb)
{
glDeleteProgramsARB( 1, &arbDesc->m_object.arb );
GLMCheckError();
arbDesc->m_object.arb = 0;
}
// if there is a GLSL shader, delete it
GLMShaderDesc *glslDesc = &m_descs[kGLMGLSL];
if (glslDesc->m_object.glsl)
{
glDeleteShader( (uint)glslDesc->m_object.glsl ); // why do I need a cast here again ?
GLMCheckError();
glslDesc->m_object.glsl = 0;
}
#if GLMDEBUG
if (m_editable)
{
delete m_editable;
m_editable = NULL;
}
#endif
if (m_text)
{
free( m_text );
m_text = NULL;
}
m_ctx = NULL;
}
enum EShaderSection
{
kGLMARBVertex, kGLMARBVertexDisabled,
kGLMARBFragment, kGLMARBFragmentDisabled,
kGLMGLSLVertex, kGLMGLSLVertexDisabled,
kGLMGLSLFragment, kGLMGLSLFragmentDisabled,
};
const char *g_shaderSectionMarkers[] = // match ordering of enum
{
"!!ARBvp", "-!!ARBvp", // enabled and disabled markers. so you can have multiple flavors in a blob and activate the one you want.
"!!ARBfp", "-!!ARBfp",
"//GLSLvp", "-//GLSLvp",
"//GLSLfp", "-//GLSLfp",
NULL
};
void CGLMProgram::SetProgramText( char *text )
{
// free old text if any
// clone new text
// scan newtext to find sections
// walk sections, and mark descs to indicate where text is at
if (m_text)
{
free( m_text );
m_text = NULL;
}
// scrub desc text references
for( int i=0; i<kGLMNumProgramTypes; i++)
{
GLMShaderDesc *desc = &m_descs[i];
desc->m_textPresent = false;
desc->m_textOffset = 0;
desc->m_textLength = 0;
}
m_text = strdup( text );
Assert( m_text != NULL );
#if GLMDEBUG
// create editable text item, if it does not already exist
if (!m_editable)
{
char *suffix = "";
switch(m_type)
{
case kGLMVertexProgram: suffix = ".vsh"; break;
case kGLMFragmentProgram: suffix = ".fsh"; break;
default: GLMDebugger();
}
m_editable = new CGLMEditableTextItem( m_text, strlen(m_text), false, "/debugshaders/", suffix );
// pull our string back from the editable (it has probably munged it)
if (m_editable->HasData())
{
ReloadStringFromEditable();
}
}
#endif
// scan the text and find sections
CGLMTextSectioner sections( m_text, strlen( m_text ), g_shaderSectionMarkers );
int sectionCount = sections.Count();
for( int i=0; i < sectionCount; i++ )
{
uint subtextOffset = 0;
uint subtextLength = 0;
int markerIndex = 0;
sections.GetSection( i, &subtextOffset, &subtextLength, &markerIndex );
// act on the section
GLMShaderDesc *desc = NULL;
switch( m_type )
{
case kGLMVertexProgram:
switch( markerIndex )
{
case kGLMARBVertex:
case kGLMGLSLVertex:
desc = &m_descs[ (markerIndex==kGLMARBVertex) ? kGLMARB : kGLMGLSL];
// these steps are generic across both langs
desc->m_textPresent = true;
desc->m_textOffset = subtextOffset;
desc->m_textLength = subtextLength;
desc->m_compiled = false;
desc->m_valid = false;
break;
case kGLMARBVertexDisabled:
case kGLMGLSLVertexDisabled:
// ignore quietly
break;
default: Assert(!"Mismatched section marker seen in SetProgramText (VP)"); break;
}
break;
case kGLMFragmentProgram:
switch( markerIndex )
{
case kGLMARBFragment:
case kGLMGLSLFragment:
desc = &m_descs[ (markerIndex==kGLMARBFragment) ? kGLMARB : kGLMGLSL];
// these steps are generic across both langs
desc->m_textPresent = true;
desc->m_textOffset = subtextOffset;
desc->m_textLength = subtextLength;
desc->m_compiled = false;
desc->m_valid = false;
break;
case kGLMARBFragmentDisabled:
case kGLMGLSLFragmentDisabled:
// ignore quietly
break;
default: Assert(!"Mismatched section marker seen in SetProgramText (VP)"); break;
}
break;
}
}
}
bool CGLMProgram::CompileActiveSources ( void )
{
bool result = true; // assume success
// compile everything we have text for
for( int i=0; i<kGLMNumProgramTypes; i++)
{
if (m_descs[i].m_textPresent)
{
if (!Compile( (EGLMProgramLang)i ))
{
result = false;
}
}
}
return result;
}
bool CGLMProgram::Compile( EGLMProgramLang lang )
{
bool result = true; // indicating validity..
bool noisy = false;
int loglevel = gl_shaderpair_cachelog.GetInt();
switch( lang )
{
case kGLMARB:
{
GLMShaderDesc *arbDesc;
arbDesc = &m_descs[ kGLMARB ];
// make sure no GLSL program is set up
glUseProgram(0);
GLMCheckError();
// bind our program container to context
GLenum arbTarget = GLMProgTypeToARBEnum( m_type );
glSetEnable( arbTarget, true ); // unclear if I need this to compile or just to draw...
GLMCheckError();
glBindProgramARB( arbTarget, arbDesc->m_object.arb ); // object created or just re-bound
GLMCheckError();
char *section = m_text + arbDesc->m_textOffset;
char *lastCharOfSection = section + arbDesc->m_textLength; // actually it's one past the last textual character
#if GLMDEBUG
if(noisy)
{
GLMPRINTF((">-D- CGLMProgram::Compile submitting following text for ARB %s program (name %d) ---------------------",
arbTarget == GL_FRAGMENT_PROGRAM_ARB ? "fragment" : "vertex",
arbDesc->m_object.arb ));
// we don't have a "print this many chars" call yet
// just temporarily null terminate the text we want to print
char saveChar = *lastCharOfSection;
*lastCharOfSection= 0;
GLMPRINTTEXT(( section, eDebugDump ));
*lastCharOfSection= saveChar;
GLMPRINTF(("<-D- CGLMProgram::Compile ARB EOT--" ));
}
#endif
glProgramStringARB( arbTarget, GL_PROGRAM_FORMAT_ASCII_ARB, arbDesc->m_textLength, section );
GLMCheckError( true, false );
arbDesc->m_compiled = true; // compiled but not necessarily valid
CheckValidity( lang );
GLMCheckError();
// leave it bound n enabled, don't care (draw will sort it all out)
result = arbDesc->m_valid;
}
break;
case kGLMGLSL:
{
GLMShaderDesc *glslDesc;
glslDesc = &m_descs[ kGLMGLSL ];
GLenum glslStage = GLMProgTypeToGLSLEnum( m_type );
// there's no binding to do for GLSL. but make sure no ARB stuff is bound for tidiness.
glSetEnable( GL_VERTEX_PROGRAM_ARB, false );
glSetEnable( GL_FRAGMENT_PROGRAM_ARB, false ); // add check errors on these
glBindProgramARB( GL_VERTEX_PROGRAM_ARB, 0 );
glBindProgramARB( GL_FRAGMENT_PROGRAM_ARB, 0 );
// no GLSL program either
glUseProgram(0);
// pump text into GLSL shader object
char *section = m_text + glslDesc->m_textOffset;
char *lastCharOfSection = section + glslDesc->m_textLength; // actually it's one past the last textual character
#if GLMDEBUG
if(noisy)
{
GLMPRINTF((">-D- CGLMProgram::Compile submitting following text for GLSL %s program (name %d) ---------------------",
glslStage == GL_FRAGMENT_SHADER_ARB ? "fragment" : "vertex",
glslDesc->m_object.glsl ));
// we don't have a "print this many chars" call yet
// just temporarily null terminate the text we want to print
char saveChar = *lastCharOfSection;
*lastCharOfSection= 0;
GLMPRINTTEXT(( section, eDebugDump ));
*lastCharOfSection= saveChar;
GLMPRINTF(("<-D- CGLMProgram::Compile GLSL EOT--" ));
}
#endif
glShaderSourceARB( glslDesc->m_object.glsl, 1, (const GLchar **)&section, &glslDesc->m_textLength);
GLMCheckError( true, false );
// compile
glCompileShaderARB( glslDesc->m_object.glsl );
glslDesc->m_compiled = true; // compiled but not necessarily valid
GLMCheckError( true, false );
CheckValidity( lang );
GLMCheckError();
if (loglevel>=2)
{
char tempname[128];
int tempindex = -1;
int tempcombo = -1;
//GetLabelIndexCombo( tempname, sizeof(tempname), &tempindex, &tempcombo );
//printf("\ncompile: - [ %s/%d/%d ] on GL name %d ", tempname, tempindex, tempcombo, glslDesc->m_object.glsl );
GetComboIndexNameString( tempname, sizeof(tempname) );
printf("\ncompile: %s on GL name %d ", tempname, glslDesc->m_object.glsl );
}
result = glslDesc->m_valid;
}
break;
}
return result;
}
#if GLMDEBUG
bool CGLMProgram::PollForChanges( void )
{
bool result = false;
if (m_editable)
{
result = m_editable->PollForChanges();
}
return result;
}
void CGLMProgram::ReloadStringFromEditable( void )
{
uint dataSize=0;
char *data=NULL;
m_editable->GetCurrentText( &data, &dataSize );
char *buf = (char *)malloc( dataSize+1 ); // we will NULL terminate it, since the mirror copy might not be
memcpy( buf, data, dataSize );
buf[dataSize] = 0;
SetProgramText( buf );
free( buf );
}
bool CGLMProgram::SyncWithEditable( void )
{
bool result = false;
if (m_editable->PollForChanges())
{
ReloadStringFromEditable();
CompileActiveSources();
// invalidate shader pair cache entries using this shader..
m_ctx->m_pairCache->PurgePairsWithShader( this );
result = true; // result true means "it changed"
}
return result;
}
#endif
// attributes which are general to both stages
// VP and FP:
//
// 0x88A0 PROGRAM_INSTRUCTIONS_ARB VP FP
// 0x88A1 MAX_PROGRAM_INSTRUCTIONS_ARB VP FP
// 0x88A2 PROGRAM_NATIVE_INSTRUCTIONS_ARB VP FP
// 0x88A3 MAX_PROGRAM_NATIVE_INSTRUCTIONS_ARB VP FP
//
// 0x88A4 PROGRAM_TEMPORARIES_ARB VP FP
// 0x88A5 MAX_PROGRAM_TEMPORARIES_ARB VP FP
// 0x88A6 PROGRAM_NATIVE_TEMPORARIES_ARB VP FP
// 0x88A7 MAX_PROGRAM_NATIVE_TEMPORARIES_ARB VP FP
//
// 0x88A8 PROGRAM_PARAMETERS_ARB VP FP
// 0x88A9 MAX_PROGRAM_PARAMETERS_ARB VP FP
// 0x88AA PROGRAM_NATIVE_PARAMETERS_ARB VP FP
// 0x88AB MAX_PROGRAM_NATIVE_PARAMETERS_ARB VP FP
//
// 0x88AC PROGRAM_ATTRIBS_ARB VP FP
// 0x88AD MAX_PROGRAM_ATTRIBS_ARB VP FP
// 0x88AE PROGRAM_NATIVE_ATTRIBS_ARB VP FP
// 0x88AF MAX_PROGRAM_NATIVE_ATTRIBS_ARB VP FP
//
// 0x88B4 MAX_PROGRAM_LOCAL_PARAMETERS_ARB VP FP
// 0x88B5 MAX_PROGRAM_ENV_PARAMETERS_ARB VP FP
// 0x88B6 PROGRAM_UNDER_NATIVE_LIMITS_ARB VP FP
//
// VP only:
//
// 0x88B0 PROGRAM_ADDRESS_REGISTERS_ARB VP
// 0x88B1 MAX_PROGRAM_ADDRESS_REGISTERS_ARB VP
// 0x88B2 PROGRAM_NATIVE_ADDRESS_REGISTERS_ARB VP
// 0x88B3 MAX_PROGRAM_NATIVE_ADDRESS_REGISTERS_ARB VP
//
// FP only:
//
// 0x8805 PROGRAM_ALU_INSTRUCTIONS_ARB FP
// 0x880B MAX_PROGRAM_ALU_INSTRUCTIONS_ARB FP
// 0x8808 PROGRAM_NATIVE_ALU_INSTRUCTIONS_ARB FP
// 0x880E MAX_PROGRAM_NATIVE_ALU_INSTRUCTIONS_ARB FP
// 0x8806 PROGRAM_TEX_INSTRUCTIONS_ARB FP
// 0x880C MAX_PROGRAM_TEX_INSTRUCTIONS_ARB FP
// 0x8809 PROGRAM_NATIVE_TEX_INSTRUCTIONS_ARB FP
// 0x880F MAX_PROGRAM_NATIVE_TEX_INSTRUCTIONS_ARB FP
// 0x8807 PROGRAM_TEX_INDIRECTIONS_ARB FP
// 0x880D MAX_PROGRAM_TEX_INDIRECTIONS_ARB FP
// 0x880A PROGRAM_NATIVE_TEX_INDIRECTIONS_ARB FP
// 0x8810 MAX_PROGRAM_NATIVE_TEX_INDIRECTIONS_ARB FP
struct GLMShaderLimitDesc
{
GLenum m_valueEnum;
GLenum m_limitEnum;
const char *m_debugName;
char m_flags;
// m_flags - 0x01 for VP, 0x02 for FP, or set both if applicable to both
};
// macro to help make the table of what to check
#ifndef LMD( val, limit, flags )
#define LMD( val, flags ) { GL_PROGRAM_##val##_ARB, GL_MAX_PROGRAM_##val##_ARB, #val, flags }
#else
#error you need to use a different name for this macro.
#endif
GLMShaderLimitDesc g_glmShaderLimitDescs[] =
{
// VP and FP..
LMD( INSTRUCTIONS, 3 ),
LMD( NATIVE_INSTRUCTIONS, 3 ),
LMD( NATIVE_TEMPORARIES, 3 ),
LMD( PARAMETERS, 3 ),
LMD( NATIVE_PARAMETERS, 3 ),
LMD( ATTRIBS, 3 ),
LMD( NATIVE_ATTRIBS, 3 ),
// VP only..
LMD( ADDRESS_REGISTERS, 1 ),
LMD( NATIVE_ADDRESS_REGISTERS, 1 ),
// FP only..
LMD( ALU_INSTRUCTIONS, 2 ),
LMD( NATIVE_ALU_INSTRUCTIONS, 2 ),
LMD( TEX_INSTRUCTIONS, 2 ),
LMD( NATIVE_TEX_INSTRUCTIONS, 2 ),
LMD( TEX_INDIRECTIONS, 2 ),
LMD( NATIVE_TEX_INDIRECTIONS, 2 ),
{ 0, 0, NULL, 0 }
};
#undef LMD
bool CGLMProgram::CheckValidity( EGLMProgramLang lang )
{
static char *targnames[] = { "vertex", "fragment" };
switch(lang)
{
case kGLMARB:
{
GLMShaderDesc *arbDesc;
arbDesc = &m_descs[ kGLMARB ];
GLenum arbTarget = GLMProgTypeToARBEnum( m_type );
Assert( arbDesc->m_compiled );
arbDesc->m_valid = true; // assume success til we see otherwise
// assume program is bound. is there anything wrong with it ?
GLint isNative=0;
glGetProgramivARB( arbTarget, GL_PROGRAM_UNDER_NATIVE_LIMITS_ARB, &isNative );
GLMCheckError();
// If the program is over the hardware's limits, print out some information
if (isNative!=1)
{
arbDesc->m_valid = false;
// check everything we can check
char checkmask = (1<<m_type); // 1 for VP, 2 for FP
for( GLMShaderLimitDesc *desc = g_glmShaderLimitDescs; desc->m_valueEnum !=0; desc++ )
{
if ( desc->m_flags & checkmask )
{
// test it
GLint value = 0;
GLint limit = 0;
glGetProgramivARB(arbTarget, desc->m_valueEnum, &value);
GLMCheckError();
glGetProgramivARB(arbTarget, desc->m_limitEnum, &limit);
GLMCheckError();
if (value > limit)
{
GLMPRINTF(("-D- Invalid %s program: program has %d %s; limit is %d", targnames[ m_type ], value, desc->m_debugName, limit ));
}
}
}
}
// syntax error check
GLint errorLine;
glGetIntegerv( GL_PROGRAM_ERROR_POSITION_ARB, &errorLine );
GLMCheckError();
if ( errorLine!=-1 )
{
const GLubyte* errorString = glGetString(GL_PROGRAM_ERROR_STRING_ARB);
GLMPRINTF(( "-D- Syntax error in ARB %s program: %s",targnames[ m_type ], errorString ));
arbDesc->m_valid = false;
}
if (!arbDesc->m_valid)
{
char *temp = strdup(m_text);
temp[ arbDesc->m_textOffset + arbDesc->m_textLength ] = 0;
GLMPRINTF(("-D- ----- ARB compile failed; bad source follows -----" ));
GLMPRINTTEXT(( temp + arbDesc->m_textOffset, eDebugDump, GLMPRINTTEXT_NUMBEREDLINES ));
GLMPRINTF(("-D- -----end-----" ));
free( temp );
}
return arbDesc->m_valid;
}
break;
case kGLMGLSL:
{
GLMShaderDesc *glslDesc;
glslDesc = &m_descs[ kGLMGLSL ];
GLenum glslStage = GLMProgTypeToGLSLEnum( m_type );
Assert( glslDesc->m_compiled );
glslDesc->m_valid = true; // assume success til we see otherwise
// GLSL error check
int compiled = 0, length = 0, laux = 0;
glGetObjectParameterivARB( (GLhandleARB)glslDesc->m_object.glsl, GL_OBJECT_COMPILE_STATUS_ARB, &compiled);
glGetObjectParameterivARB( (GLhandleARB)glslDesc->m_object.glsl, GL_OBJECT_INFO_LOG_LENGTH_ARB, &length);
GLcharARB *logString = (GLcharARB *)malloc(length * sizeof(GLcharARB));
glGetInfoLogARB((GLhandleARB)glslDesc->m_object.glsl, length, &laux, logString);
// we may not be able to check "native limits" stuff until link time. meh
if (!compiled)
{
glslDesc->m_valid = false;
}
if (!glslDesc->m_valid)
{
char *temp = strdup(m_text);
temp[ glslDesc->m_textOffset + glslDesc->m_textLength ] = 0;
GLMPRINTF(("-D- ----- GLSL compile failed: \n %s \n",logString ));
GLMPRINTTEXT(( temp + glslDesc->m_textOffset, eDebugDump, GLMPRINTTEXT_NUMBEREDLINES ));
GLMPRINTF(("-D- -----end-----" ));
free( temp );
}
free( logString );
return glslDesc->m_valid;
}
break;
}
}
void CGLMProgram::LogSlow( EGLMProgramLang lang )
{
// find the desc, see if it's marked
GLMShaderDesc *desc = &m_descs[ lang ];
if (!desc->m_slowMark)
{
// log it
printf( "\n-------------- Slow %s ( CGLMProgram @ %08x, lang %s, name %d ) : \n%s \n",
m_type==kGLMVertexProgram ? "VS" : "FS",
this,
lang==kGLMGLSL ? "GLSL" : "ARB",
(int)(lang==kGLMGLSL ? (int)desc->m_object.glsl : (int)desc->m_object.arb),
m_text
);
}
else // complain on a decreasing basis (powers of two)
{
if ( (desc->m_slowMark & (desc->m_slowMark-1)) == 0 )
{
// short blurb
printf( "\n Slow %s ( CGLMProgram @ %08x, lang %s, name %d ) (%d times)",
m_type==kGLMVertexProgram ? "VS" : "FS",
this,
lang==kGLMGLSL ? "GLSL" : "ARB",
(int)(lang==kGLMGLSL ? (int)desc->m_object.glsl : (int)desc->m_object.arb),
desc->m_slowMark+1
);
}
}
// mark it
desc->m_slowMark++;
}
void CGLMProgram::GetLabelIndexCombo ( char *labelOut, int labelOutMaxChars, int *indexOut, int *comboOut )
{
// find the label string
// example:
// trans#2871 label:vs-file vertexlit_and_unlit_generic_vs20 vs-index 294912 vs-combo 1234
*labelOut = 0;
*indexOut = -1;
char *lineStr = strstr( m_text, "// trans#" );
if (lineStr)
{
char temp1[1024];
int temp2,temp3;
int scratch=-1;
temp1[0] = 0;
temp2 = -1;
temp3 = -1;
if (this->m_type==kGLMVertexProgram)
{
sscanf( lineStr, "// trans#%d label:vs-file %s vs-index %d vs-combo %d", &scratch, temp1, &temp2, &temp3 );
}
else
{
sscanf( lineStr, "// trans#%d label:ps-file %s ps-index %d ps-combo %d", &scratch, temp1, &temp2, &temp3 );
}
if ( (strlen(temp1)!=0) )
{
strncpy( labelOut, temp1, labelOutMaxChars );
*indexOut = temp2;
*comboOut = temp3;
}
}
}
void CGLMProgram::GetComboIndexNameString ( char *stringOut, int stringOutMaxChars ) // mmmmmmmm-nnnnnnnn-filename
{
// find the label string
// example:
// trans#2871 label:vs-file vertexlit_and_unlit_generic_vs20 vs-index 294912 vs-combo 1234
*stringOut = 0;
char *lineStr = strstr( m_text, "// trans#" );
if (lineStr)
{
char temp1[1024];
int temp2,temp3;
int scratch=-1;
temp1[0] = 0;
temp2 = -1;
temp3 = -1;
if (this->m_type==kGLMVertexProgram)
{
sscanf( lineStr, "// trans#%d label:vs-file %s vs-index %d vs-combo %d", &scratch, temp1, &temp2, &temp3 );
}
else
{
sscanf( lineStr, "// trans#%d label:ps-file %s ps-index %d ps-combo %d", &scratch, temp1, &temp2, &temp3 );
}
int len = strlen(temp1);
if ( (len+20) < stringOutMaxChars )
{
// output formatted version
sprintf( stringOut, "%08X-%08X-%s", temp3, temp2, temp1 );
}
}
}
//===============================================================================
CGLMShaderPair::CGLMShaderPair( GLMContext *ctx )
{
m_ctx = ctx;
m_ctx->MakeCurrent();
m_vertexProg = m_fragmentProg = NULL;
m_program = glCreateProgramObjectARB();
GLMCheckError();
m_locVertexParams = -1;
m_locVertexInteger0 = -1; // "i0"
m_locVertexBool0 = -1; // "b0"
m_locVertexBool1 = -1; // "b1"
m_locVertexBool2 = -1; // "b2"
m_locVertexBool3 = -1; // "b3"
m_locFragmentParams = -1;
m_locFragmentFakeSRGBEnable = -1;
m_fakeSRGBEnableValue = -1.0f;
memset( m_locSamplers, 0xFF, sizeof( m_locSamplers ) );
m_valid = false;
m_samplersFixed = false; // fix them at draw time, and only do it once.
m_revision = 0; // bumps to 1 once linked
}
CGLMShaderPair::~CGLMShaderPair( )
{
if (m_program)
{
glDeleteObjectARB( (GLhandleARB)m_program );
m_program = 0;
}
}
bool CGLMShaderPair::SetProgramPair ( CGLMProgram *vp, CGLMProgram *fp )
{
m_valid = false; // assume failure
// true result means successful link and query
bool vpgood = (vp!=NULL) && (vp->m_descs[ kGLMGLSL ].m_valid);
bool fpgood = (fp!=NULL) && (fp->m_descs[ kGLMGLSL ].m_valid);
if (!fpgood)
{
// fragment side allowed to be "null".
fp = m_ctx->m_nullFragmentProgram;
}
if (vpgood && fpgood)
{
// attempt link. but first, detach any previously attached programs
if (m_vertexProg)
{
glDetachObjectARB(m_program, m_vertexProg->m_descs[kGLMGLSL].m_object.glsl);
GLMCheckError();
m_vertexProg = NULL;
}
if (m_fragmentProg)
{
glDetachObjectARB(m_program, m_fragmentProg->m_descs[kGLMGLSL].m_object.glsl);
GLMCheckError();
m_fragmentProg = NULL;
}
// now attach
glAttachObjectARB(m_program, vp->m_descs[kGLMGLSL].m_object.glsl);
m_vertexProg = vp;
GLMCheckError();
glAttachObjectARB(m_program, fp->m_descs[kGLMGLSL].m_object.glsl);
m_fragmentProg = fp;
GLMCheckError();
// force the locations for input attributes v0-vN to be at locations 0-N
// use the vertex attrib map to know which slots are live or not... oy! we don't have that map yet... but it's OK.
// fallback - just force v0-v15 to land in locations 0-15 as a standard.
if (vp->m_descs[kGLMGLSL].m_valid)
{
for( int i=0; i < 16; i++)
{
char tmp[16];
sprintf(tmp, "v%d", i); // v0 v1 v2 ... et al
glBindAttribLocationARB( m_program, i, tmp );
GLMCheckError();
}
}
// now link
glLinkProgramARB( m_program );
GLMCheckError();
// check for success
GLint result = 0;
glGetObjectParameterivARB(m_program,GL_OBJECT_LINK_STATUS_ARB,&result); // want GL_TRUE
if (result == GL_TRUE)
{
// success
m_valid = true;
m_revision++;
}
else
{
GLint length = 0;
GLint laux = 0;
// do some digging
glGetObjectParameterivARB(m_program,GL_OBJECT_INFO_LOG_LENGTH_ARB,&length);
GLcharARB *logString = (GLcharARB *)malloc(length * sizeof(GLcharARB));
glGetInfoLogARB(m_program, length, &laux, logString);
char *vtemp = strdup(vp->m_text);
vtemp[ vp->m_descs[kGLMGLSL].m_textOffset + vp->m_descs[kGLMGLSL].m_textLength ] = 0;
char *ftemp = strdup(fp->m_text);
ftemp[ fp->m_descs[kGLMGLSL].m_textOffset + fp->m_descs[kGLMGLSL].m_textLength ] = 0;
GLMPRINTF(("-D- ----- GLSL link failed: \n %s ",logString ));
GLMPRINTF(("-D- ----- GLSL vertex program selected: %08x (handle %08x)", vp, vp->m_descs[kGLMGLSL].m_object.glsl ));
GLMPRINTTEXT(( vtemp + vp->m_descs[kGLMGLSL].m_textOffset, eDebugDump, GLMPRINTTEXT_NUMBEREDLINES ));
GLMPRINTF(("-D- ----- GLSL fragment program selected: %08x (handle %08x)", fp, vp->m_descs[kGLMGLSL].m_object.glsl ));
GLMPRINTTEXT(( ftemp + fp->m_descs[kGLMGLSL].m_textOffset, eDebugDump, GLMPRINTTEXT_NUMBEREDLINES ));
GLMPRINTF(("-D- -----end-----" ));
free( ftemp );
free( vtemp );
free( logString );
}
}
else
{
// fail
Assert(!"Can't link these programs");
}
if (m_valid)
{
m_locVertexParams = glGetUniformLocationARB( m_program, "vc");
GLMCheckError();
m_locVertexInteger0 = glGetUniformLocationARB( m_program, "i0");
GLMCheckError();
m_locVertexBool0 = glGetUniformLocationARB( m_program, "b0");
GLMCheckError();
m_locVertexBool1 = glGetUniformLocationARB( m_program, "b1");
GLMCheckError();
m_locVertexBool2 = glGetUniformLocationARB( m_program, "b2");
GLMCheckError();
m_locVertexBool3 = glGetUniformLocationARB( m_program, "b3");
GLMCheckError();
m_locFragmentParams = glGetUniformLocationARB( m_program, "pc");
GLMCheckError();
m_locFragmentFakeSRGBEnable = glGetUniformLocationARB( m_program, "flSRGBWrite");
GLMCheckError();
m_fakeSRGBEnableValue = -1.0f;
for( int sampler=0; sampler<16; sampler++)
{
char tmp[16];
sprintf(tmp, "sampler%d", sampler); // sampler0 .. sampler1.. etc
m_locSamplers[sampler] = glGetUniformLocationARB( m_program, tmp );
GLMCheckError();
}
}
else
{
m_locVertexParams = -1;
m_locVertexInteger0 = -1;
m_locVertexBool0 = -1;
m_locVertexBool1 = -1;
m_locVertexBool2 = -1;
m_locVertexBool3 = -1;
m_locFragmentParams = -1;
m_locFragmentFakeSRGBEnable = -1;
m_fakeSRGBEnableValue = -999;
memset( m_locSamplers, 0xFF, sizeof( m_locSamplers ) );
m_revision = 0;
}
return m_valid;
}
bool CGLMShaderPair::RefreshProgramPair ( void )
{
// re-link and re-query the uniforms.
// since SetProgramPair knows how to detach previously attached shader objects, just pass the same ones in again.
CGLMProgram *vp = m_vertexProg;
CGLMProgram *fp = m_fragmentProg;
bool vpgood = (vp!=NULL) && (vp->m_descs[ kGLMGLSL ].m_valid);
bool fpgood = (fp!=NULL) && (fp->m_descs[ kGLMGLSL ].m_valid);
if (vpgood && fpgood)
{
SetProgramPair( vp, fp );
}
else
{
Debugger();
}
}
//===============================================================================
CGLMShaderPairCache::CGLMShaderPairCache( GLMContext *ctx )
{
m_ctx = ctx;
m_mark = 1;
m_rowsLg2 = gl_shaderpair_cacherows_lg2.GetInt();
if (m_rowsLg2 < 10)
m_rowsLg2 = 10;
m_rows = 1<<m_rowsLg2;
m_waysLg2 = gl_shaderpair_cacheways_lg2.GetInt();
if (m_waysLg2 < 5)
m_waysLg2 = 5;
m_ways = 1<<m_waysLg2;
m_entryCount = m_rows * m_ways;
uint entryTableSize = m_rows * m_ways * sizeof(CGLMPairCacheEntry);
m_entries = (CGLMPairCacheEntry*)malloc( entryTableSize ); // array[ m_rows ][ m_ways ]
memset( m_entries, 0, entryTableSize );
uint evictTableSize = m_rows * sizeof(uint);
m_evictions = (uint*)malloc( evictTableSize );
memset (m_evictions, 0, evictTableSize);
// hit counter table is same size
m_hits = (uint*)malloc( evictTableSize );
memset (m_hits, 0, evictTableSize);
}
CGLMShaderPairCache::~CGLMShaderPairCache( )
{
if (gl_shaderpair_cachelog.GetInt())
{
DumpStats();
}
// free all the built pairs
// free the entry table
bool purgeResult = this->Purge();
Assert( !purgeResult );
if (m_entries)
{
free( m_entries );
m_entries = NULL;
}
if (m_evictions)
{
free( m_evictions );
m_evictions = NULL;
}
if (m_hits)
{
free( m_hits );
m_hits = NULL;
}
}
// Set this convar internally to build or add to the shader pair cache file (link hints)
// We really only expect this to work on POSIX
static ConVar glm_cacheprograms( "glm_cacheprograms", "0", FCVAR_DEVELOPMENTONLY );
#define PROGRAM_CACHE_FILE "program_cache.cfg"
static void WriteToProgramCache( CGLMShaderPair *pair )
{
KeyValues *pProgramCache = new KeyValues( "programcache" );
pProgramCache->LoadFromFile( g_pFullFileSystem, PROGRAM_CACHE_FILE, "MOD" );
if ( !pProgramCache )
{
Warning( "Could not write to program cache file!\n" );
return;
}
// extract values of interest which represent a pair of shaders
char vprogramName[128];
int vprogramStaticIndex = -1;
int vprogramDynamicIndex = -1;
pair->m_vertexProg->GetLabelIndexCombo( vprogramName, sizeof(vprogramName), &vprogramStaticIndex, &vprogramDynamicIndex );
char pprogramName[128];
int pprogramStaticIndex = -1;
int pprogramDynamicIndex = -1;
pair->m_fragmentProg->GetLabelIndexCombo( pprogramName, sizeof(pprogramName), &pprogramStaticIndex, &pprogramDynamicIndex );
// make up a key - this thing is really a list of tuples, so need not be keyed by anything particular
KeyValues *pProgramKey = pProgramCache->CreateNewKey();
Assert( pProgramKey );
pProgramKey->SetString ( "vs", vprogramName );
pProgramKey->SetString ( "ps", pprogramName );
pProgramKey->SetInt ( "vs_static", vprogramStaticIndex );
pProgramKey->SetInt ( "ps_static", pprogramStaticIndex );
pProgramKey->SetInt ( "vs_dynamic", vprogramDynamicIndex );
pProgramKey->SetInt ( "ps_dynamic", pprogramDynamicIndex );
pProgramCache->SaveToFile( g_pFullFileSystem, PROGRAM_CACHE_FILE, "MOD" );
pProgramCache->deleteThis();
}
CGLMShaderPair *CGLMShaderPairCache::SelectShaderPair( CGLMProgram *vp, CGLMProgram *fp, uint extraKeyBits )
{
CGLMShaderPair *result = NULL;
int loglevel = gl_shaderpair_cachelog.GetInt();
char vtempname[128];
int vtempindex = -1;
int vtempcombo = -1;
char ptempname[128];
int ptempindex = -1;
int ptempcombo = -1;
// select row where pair would be found if it exists
uint rowIndex = HashRowIndex( vp, fp, extraKeyBits );
CGLMPairCacheEntry *row = HashRowPtr( rowIndex );
// probe row and see if we get a hit
int hitway = -1;int emptyway = -1; int oldestway = -1;
HashRowProbe( row, vp, fp, extraKeyBits, &hitway, &emptyway, &oldestway );
if (hitway >=0)
{
// found it. mark it and return
CGLMPairCacheEntry *hit = row + hitway;
hit->m_lastMark = m_mark;
m_mark = m_mark+1;
if (!m_mark) // somewhat unlikely this will ever be reached.. but we need to avoid zero as a mark value
{
m_mark = 1;
}
// count the hit
m_hits[ rowIndex ] ++;
if (loglevel >= 3) // hits logged at level 3 and higher
{
printf("\nSSP: hit - row %05d - pair $%08x (%d'th hit on row)",rowIndex, hit->m_pair, m_hits[ rowIndex ] );
}
result = hit->m_pair;
}
else
{
// we missed. if there is no empty way, then somebody's getting evicted.
int destway = -1;
if (emptyway>=0)
{
destway = emptyway;
if (loglevel >= 2) // misses logged at level 3 and higher
{
printf("\nSSP: miss - row %05d - ", rowIndex );
}
}
else
{
// evict the oldest way
Assert( oldestway >= 0); // better not come back negative
CGLMPairCacheEntry *evict = row + oldestway;
Assert( evict->m_pair != NULL );
Assert( evict->m_pair != m_ctx->m_boundPair ); // just check
///////////////////////FIXME may need to do a shoot-down if the pair being evicted is currently active in the context
m_evictions[ rowIndex ]++;
// log eviction if desired
if (loglevel >= 2) // misses logged at level 3 and higher
{
//evict->m_vertexProg->GetLabelIndexCombo( vtempname, sizeof(vtempname), &vtempindex, &vtempcombo );
//evict->m_fragmentProg->GetLabelIndexCombo( ptempname, sizeof(ptempname), &ptempindex, &ptempcombo );
//printf("\nSSP: miss - row %05d - [ %s/%d/%d %s/%d/%d ]'s %d'th eviction - ", rowIndex, vtempname, vtempindex, vtempcombo, ptempname, ptempindex, ptempcombo, m_evictions[ rowIndex ] );
evict->m_vertexProg->GetComboIndexNameString( vtempname, sizeof(vtempname) );
evict->m_fragmentProg->GetComboIndexNameString( ptempname, sizeof(ptempname) );
printf("\nSSP: miss - row %05d - [ %s + %s ]'s %d'th eviction - ", rowIndex, vtempname, ptempname, m_evictions[ rowIndex ] );
}
delete evict->m_pair; evict->m_pair = NULL;
memset( evict, 0, sizeof(*evict) );
destway = oldestway;
}
// make the new entry
CGLMPairCacheEntry *newentry = row + destway;
newentry->m_lastMark = m_mark;
newentry->m_vertexProg = vp;
newentry->m_fragmentProg = fp;
newentry->m_extraKeyBits = extraKeyBits;
newentry->m_pair = new CGLMShaderPair( m_ctx );
newentry->m_pair->SetProgramPair( vp, fp );
if (loglevel >= 2) // say a little bit more
{
//newentry->m_vertexProg->GetLabelIndexCombo( vtempname, sizeof(vtempname), &vtempindex, &vtempcombo );
//newentry->m_fragmentProg->GetLabelIndexCombo( ptempname, sizeof(ptempname), &ptempindex, &ptempcombo );
//printf("new [ %s/%d/%d %s/%d/%d ]", vtempname, vtempindex, vtempcombo, ptempname, ptempindex, ptempcombo );
newentry->m_vertexProg->GetComboIndexNameString( vtempname, sizeof(vtempname) );
newentry->m_fragmentProg->GetComboIndexNameString( ptempname, sizeof(ptempname) );
printf("new [ %s + %s ]", vtempname, ptempname );
}
m_mark = m_mark+1;
if (!m_mark) // somewhat unlikely this will ever be reached.. but we need to avoid zero as a mark value
{
m_mark = 1;
}
result = newentry->m_pair;
if (glm_cacheprograms.GetInt())
{
WriteToProgramCache( newentry->m_pair );
}
}
return result;
}
void CGLMShaderPairCache::QueryShaderPair( int index, GLMShaderPairInfo *infoOut )
{
if ( (index<0) || ( index >= (m_rows*m_ways) ) )
{
// no such location
memset( infoOut, sizeof(*infoOut), 0 );
infoOut->m_status = -1;
}
else
{
// locate the entry, and see if an active pair is present.
// if so, extract info and return with m_status=1.
// if not, exit with m_status = 0.
CGLMPairCacheEntry *entry = &m_entries[index];
if (entry->m_pair)
{
// live
// extract values of interest for caller
entry->m_pair->m_vertexProg->GetLabelIndexCombo ( infoOut->m_vsName, sizeof(infoOut->m_vsName), &infoOut->m_vsStaticIndex, &infoOut->m_vsDynamicIndex );
entry->m_pair->m_fragmentProg->GetLabelIndexCombo ( infoOut->m_psName, sizeof(infoOut->m_psName), &infoOut->m_psStaticIndex, &infoOut->m_psDynamicIndex );
infoOut->m_status = 1;
}
else
{
// not
memset( infoOut, sizeof(*infoOut), 0 );
infoOut->m_status = 0;
}
}
}
bool CGLMShaderPairCache::PurgePairsWithShader( CGLMProgram *prog )
{
bool result = false;
// walk all rows*ways
int limit = m_rows * m_ways;
for( int i=0; i < limit; i++)
{
CGLMPairCacheEntry *entry = &m_entries[i];
if (entry->m_pair)
{
//scrub it, if not currently bound, and if the supplied shader matches either stage
if ( (entry->m_vertexProg==prog) || (entry->m_fragmentProg==prog) )
{
// found it, but does it conflict with bound pair ?
if (entry->m_pair == m_ctx->m_boundPair)
{
m_ctx->m_boundPair = NULL;
}
delete entry->m_pair;
memset( entry, 0, sizeof(*entry) );
}
}
}
return result;
}
bool CGLMShaderPairCache::Purge( void )
{
bool result = false;
// walk all rows*ways
int limit = m_rows * m_ways;
for( int i=0; i < limit; i++)
{
CGLMPairCacheEntry *entry = &m_entries[i];
if (entry->m_pair)
{
//scrub it, unless the pair is the currently bound pair in our parent glm context
if (entry->m_pair != m_ctx->m_boundPair)
{
delete entry->m_pair;
memset( entry, 0, sizeof(*entry) );
}
else
{
result = true;
}
}
}
return result;
}
void CGLMShaderPairCache::DumpStats ( void )
{
printf("\n------------------\npair cache stats");
int total = 0;
for( int row=0; row < m_rows; row++ )
{
if ( (m_evictions[row] != 0) || (m_hits[row] != 0) )
{
printf("\n row %d : %d evictions, %d hits",row,m_evictions[row], m_hits[row]);
total += m_evictions[row];
}
}
printf("\n\npair cache evictions: %d\n-----------------------\n",total );
}
//===============================
uint CGLMShaderPairCache::HashRowIndex ( CGLMProgram *vp, CGLMProgram *fp, uint extraKeyBits )
{
// calculate row index for this pair
uint vp_hash = vp->m_serial * 47;
uint fp_hash = fp->m_serial;
uint hash_row_index = ((vp_hash * fp_hash) + (vp_hash ^ fp_hash) + (extraKeyBits * 7) ) & (m_rows-1);
return hash_row_index;
}
CGLMPairCacheEntry* CGLMShaderPairCache::HashRowPtr ( uint hashRowIndex )
{
return &m_entries[ hashRowIndex * m_ways ];
}
void CGLMShaderPairCache::HashRowProbe ( CGLMPairCacheEntry *row, CGLMProgram *vp, CGLMProgram *fp, uint extraKeyBits, int *hitwayOut, int *emptywayOut, int *oldestwayOut )
{
// scan this row to see if the desired pair is present
CGLMPairCacheEntry *cursor = row;
int hitway = -1;
int emptyway = -1;
int oldestway = -1;
long long oldestmark = 0xFFFFFFFFFFFFFFFFLL;
for( int way = 0; (way < m_ways) && (hitway<0); way++)
{
if (cursor->m_lastMark != 0) // occupied slot
{
if ( (cursor->m_vertexProg == vp) && (cursor->m_fragmentProg == fp) && (cursor->m_extraKeyBits == extraKeyBits) ) // match?
{
// found it
hitway = way;
}
// check if this is the oldest one on the row - only occupied slots are checked
if (cursor->m_lastMark < oldestmark)
{
oldestway = way;
oldestmark = cursor->m_lastMark;
}
}
else
{
// empty way, log it if first one seen
if (emptyway<0)
{
emptyway = way;
}
}
cursor++;
}
if (hitwayOut) *hitwayOut = hitway;
if (emptywayOut) *emptywayOut = emptyway;
if (oldestwayOut) *oldestwayOut = oldestway;
}