Merge remote-tracking branch 'upstream/master'

neo/d3xp/gamesys/Class.cpp

@@ -460,18 +460,18 @@ idClass::new
 #endif
 
 void * idClass::operator new( size_t s ) {
-	int *p;
+	intptr_t *p;
 
-	s += sizeof( int );
-	p = (int *)Mem_Alloc( s );
+	s += sizeof( intptr_t );
+	p = (intptr_t *)Mem_Alloc( s );
 	*p = s;
 	memused += s;
 	numobjects++;
 
 #ifdef ID_DEBUG_UNINITIALIZED_MEMORY
 	unsigned int *ptr = (unsigned int *)p;
 	int size = s;
-	assert( ( size & 3 ) == 0 );
+	assert( ( size & (sizeof(intptr_t) - 1) ) == 0 );
 	size >>= 3;
 	for ( int i = 1; i < size; i++ ) {
 		ptr[i] = 0xcdcdcdcd;
@@ -482,18 +482,18 @@ void * idClass::operator new( size_t s ) {
 }
 
 void * idClass::operator new( size_t s, int, int, char *, int ) {
-	int *p;
+	intptr_t *p;
 
-	s += sizeof( int );
-	p = (int *)Mem_Alloc( s );
+	s += sizeof( intptr_t );
+	p = (intptr_t *)Mem_Alloc( s );
 	*p = s;
 	memused += s;
 	numobjects++;
 
 #ifdef ID_DEBUG_UNINITIALIZED_MEMORY
 	unsigned int *ptr = (unsigned int *)p;
 	int size = s;
-	assert( ( size & 3 ) == 0 );
+	assert( ( size & (sizeof(intptr_t) - 1) ) == 0 );
 	size >>= 3;
 	for ( int i = 1; i < size; i++ ) {
 		ptr[i] = 0xcdcdcdcd;
@@ -513,21 +513,21 @@ idClass::delete
 ================
 */
 void idClass::operator delete( void *ptr ) {
-	int *p;
+	intptr_t *p;
 
 	if ( ptr ) {
-		p = ( ( int * )ptr ) - 1;
+		p = ( ( intptr_t * )ptr ) - 1;
 		memused -= *p;
 		numobjects--;
 		Mem_Free( p );
 	}
 }
 
 void idClass::operator delete( void *ptr, int, int, char *, int ) {
-	int *p;
+	intptr_t *p;
 
 	if ( ptr ) {
-		p = ( ( int * )ptr ) - 1;
+		p = ( ( intptr_t * )ptr ) - 1;
 		memused -= *p;
 		numobjects--;
 		Mem_Free( p );

neo/d3xp/script/Script_Interpreter.cpp

@@ -781,11 +781,22 @@ void idInterpreter::CallEvent( const function_t *func, int argsize ) {
 		switch( format[ i ] ) {
 		case D_EVENT_INTEGER :
 			var.intPtr = ( int * )&localstack[ start + pos ];
-			( *( int * )&data[ i ] ) = int( *var.floatPtr );
+			//( *( int * )&data[ i ] ) = int( *var.floatPtr );
+			// DG: for int/intrptr_t arguments, the callbacks from Callbacks.cpp pass
+			//     data[i] directly, not *(int*)&data[i] or some nonsense like that
+			//     so the integer must be assigned to the intptr_t for the value to be
+			//     passed correctly, esp. on 64bit Big Endian machines.
+			data[ i ] = int( *var.floatPtr );
 			break;
 
 		case D_EVENT_FLOAT :
 			var.intPtr = ( int * )&localstack[ start + pos ];
+			// NOTE: floats are the only type not passed as int or pointer in intptr_t,
+			//       but as float-data in the first 4 bytes of data[i].
+			//       So (unlike in the other cases), here this awkward code casting &data[i]
+			//       to another pointer type is actually necessary (same in CallSysEvent()).
+			//       In the other cases one could also use `data[i] = (intptr_t)var.blaPtr;`
+			//       (not doing those changes here now to minimize potential merge conflicts)
 			( *( float * )&data[ i ] ) = *var.floatPtr;
 			break;
 
@@ -911,7 +922,12 @@ void idInterpreter::CallSysEvent( const function_t *func, int argsize ) {
 		switch( format[ i ] ) {
 		case D_EVENT_INTEGER :
 			source.intPtr = ( int * )&localstack[ start + pos ];
-			*( int * )&data[ i ] = int( *source.floatPtr );
+			//*( int * )&data[ i ] = int( *source.floatPtr );
+			// DG: for int/intrptr_t arguments, the callbacks from Callbacks.cpp pass
+			//     data[i] directly, not *(int*)&data[i] or some nonsense like that
+			//     so the integer must be assigned to the intptr_t for the value to be
+			//     passed correctly, esp. on 64bit Big Endian machines.
+			data[ i ] = int( *source.floatPtr );
 			break;
 
 		case D_EVENT_FLOAT :

neo/d3xp/script/Script_Interpreter.h

@@ -140,7 +140,26 @@ ID_INLINE void idInterpreter::Push( intptr_t value ) {
 	if ( localstackUsed + sizeof( intptr_t ) > LOCALSTACK_SIZE ) {
 		Error( "Push: locals stack overflow\n" );
 	}
-	*( intptr_t * )&localstack[ localstackUsed ]	= value;
+	// DG: fix for 64bit Big Endian machines
+	//*( intptr_t * )&localstack[ localstackUsed ]	= value;
+	int val = value;
+	assert(value == val && "I really assumed that value would always fit into 32bit");
+	// dhewm3 used to store these values on localstack[] as intptr_t, even though
+	// they always seem to be int32. Unfortunately, all the code reading localstack[]
+	// gets the pointer to &localstack[ localstackUsed ] and then interprets that
+	// as int* or float* or whatever.
+	// So with 64bit machines it's stored as the wrong size (intptr_t is int64, int is int32),
+	// and on Big Endian the values are just 0 (or UINT_MAX if value < 0) - on Little Endian
+	// the first 4 bytes are the ones with actual data so the bug was hidden).
+	// To fix this, now a regular itn is stored at `&localstack[ localstackUsed ]`,
+	// as expected by the code using localstack[].
+	// TODO: once this has been tested more (and the assertion above has never triggered),
+	//       change idInterpreter::Push() to take a regular int argument instead of intptr_t.
+	int *stackVar = ( int * )&localstack[ localstackUsed ];
+	*stackVar = val;
+
+	// even though a 32bit int is put onto the stack, increase it by sizeof(intptr_t)
+	// so it remains aligned to multiples of the native pointer size
 	localstackUsed += sizeof( intptr_t );
 }
 

neo/framework/Common.cpp

@@ -2725,13 +2725,27 @@ void idCommonLocal::LoadGameDLL( void ) {
 	// there was no gamelib for this mod, use default one from base game
 	if (!gameDLL) {
 		common->Printf( "\n" );
-		common->Warning( "couldn't load mod-specific %s, defaulting to base game's library!\n", dll );
-		sys->DLL_GetFileName(fs_base, dll, sizeof(dll)); // DEFUNKT sys->DLL_GetFileName(BASE_GAMEDIR, dll, sizeof(dll));
-		LoadGameDLLbyName(dll, s);
+
+		const char *fs_base = cvarSystem->GetCVarString("fs_game_base");
+		if (fs_base && fs_base[0]) {
+			common->Warning( "couldn't load mod-specific %s, defaulting to library of fs_game_base (%s)!\n", dll, fs_base);
+			sys->DLL_GetFileName(fs_base, dll, sizeof(dll));
+			LoadGameDLLbyName(dll, s);
+			if ( !gameDLL ) {
+				common->Warning( "couldn't load fs_game_base lib %s either, defaulting to base game's library!\n", dll);
+			}
+		} else {
+			common->Warning( "couldn't load mod-specific %s, defaulting to base game's library!\n", dll );
+		}
+
+		if ( !gameDLL ) {
+			sys->DLL_GetFileName(BASE_GAMEDIR, dll, sizeof(dll));
+			LoadGameDLLbyName(dll, s);
+		}
 	}
 
 	if ( !gameDLL ) {
-		common->FatalError( "couldn't load game dynamic library" );
+		common->FatalError( "couldn't load game dynamic library '%s'", dll );
 		return;
 	}
 

neo/game/gamesys/Class.cpp

@@ -460,18 +460,18 @@ idClass::new
 #endif
 
 void * idClass::operator new( size_t s ) {
-	int *p;
+	intptr_t *p;
 
-	s += sizeof( int );
-	p = (int *)Mem_Alloc( s );
+	s += sizeof( intptr_t );
+	p = (intptr_t *)Mem_Alloc( s );
 	*p = s;
 	memused += s;
 	numobjects++;
 
 #ifdef ID_DEBUG_UNINITIALIZED_MEMORY
 	unsigned int *ptr = (unsigned int *)p;
 	int size = s;
-	assert( ( size & 3 ) == 0 );
+	assert( ( size & (sizeof(intptr_t) - 1) ) == 0 );
 	size >>= 3;
 	for ( int i = 1; i < size; i++ ) {
 		ptr[i] = 0xcdcdcdcd;
@@ -482,18 +482,18 @@ void * idClass::operator new( size_t s ) {
 }
 
 void * idClass::operator new( size_t s, int, int, char *, int ) {
-	int *p;
+	intptr_t *p;
 
-	s += sizeof( int );
-	p = (int *)Mem_Alloc( s );
+	s += sizeof( intptr_t );
+	p = (intptr_t *)Mem_Alloc( s );
 	*p = s;
 	memused += s;
 	numobjects++;
 
 #ifdef ID_DEBUG_UNINITIALIZED_MEMORY
 	unsigned int *ptr = (unsigned int *)p;
 	int size = s;
-	assert( ( size & 3 ) == 0 );
+	assert( ( size & (sizeof(intptr_t) - 1) ) == 0 );
 	size >>= 3;
 	for ( int i = 1; i < size; i++ ) {
 		ptr[i] = 0xcdcdcdcd;
@@ -513,21 +513,21 @@ idClass::delete
 ================
 */
 void idClass::operator delete( void *ptr ) {
-	int *p;
+	intptr_t *p;
 
 	if ( ptr ) {
-		p = ( ( int * )ptr ) - 1;
+		p = ( ( intptr_t * )ptr ) - 1;
 		memused -= *p;
 		numobjects--;
 		Mem_Free( p );
 	}
 }
 
 void idClass::operator delete( void *ptr, int, int, char *, int ) {
-	int *p;
+	intptr_t *p;
 
 	if ( ptr ) {
-		p = ( ( int * )ptr ) - 1;
+		p = ( ( intptr_t * )ptr ) - 1;
 		memused -= *p;
 		numobjects--;
 		Mem_Free( p );

neo/game/script/Script_Interpreter.cpp

@@ -781,11 +781,22 @@ void idInterpreter::CallEvent( const function_t *func, int argsize ) {
 		switch( format[ i ] ) {
 		case D_EVENT_INTEGER :
 			var.intPtr = ( int * )&localstack[ start + pos ];
-			( *( int * )&data[ i ] ) = int( *var.floatPtr );
+			//( *( int * )&data[ i ] ) = int( *var.floatPtr );
+			// DG: for int/intrptr_t arguments, the callbacks from Callbacks.cpp pass
+			//     data[i] directly, not *(int*)&data[i] or some nonsense like that
+			//     so the integer must be assigned to the intptr_t for the value to be
+			//     passed correctly, esp. on 64bit Big Endian machines.
+			data[ i ] = int( *var.floatPtr );
 			break;
 
 		case D_EVENT_FLOAT :
 			var.intPtr = ( int * )&localstack[ start + pos ];
+			// NOTE: floats are the only type not passed as int or pointer in intptr_t,
+			//       but as float-data in the first 4 bytes of data[i].
+			//       So (unlike in the other cases), here this awkward code casting &data[i]
+			//       to another pointer type is actually necessary (same in CallSysEvent()).
+			//       In the other cases one could also use `data[i] = (intptr_t)var.blaPtr;`
+			//       (not doing those changes here now to minimize potential merge conflicts)
 			( *( float * )&data[ i ] ) = *var.floatPtr;
 			break;
 
@@ -911,7 +922,12 @@ void idInterpreter::CallSysEvent( const function_t *func, int argsize ) {
 		switch( format[ i ] ) {
 		case D_EVENT_INTEGER :
 			source.intPtr = ( int * )&localstack[ start + pos ];
-			*( int * )&data[ i ] = int( *source.floatPtr );
+			//*( int * )&data[ i ] = int( *source.floatPtr );
+			// DG: for int/intrptr_t arguments, the callbacks from Callbacks.cpp pass
+			//     data[i] directly, not *(int*)&data[i] or some nonsense like that
+			//     so the integer must be assigned to the intptr_t for the value to be
+			//     passed correctly, esp. on 64bit Big Endian machines.
+			data[ i ] = int( *source.floatPtr );
 			break;
 
 		case D_EVENT_FLOAT :

neo/game/script/Script_Interpreter.h

@@ -140,7 +140,26 @@ ID_INLINE void idInterpreter::Push( intptr_t value ) {
 	if ( localstackUsed + sizeof( intptr_t ) > LOCALSTACK_SIZE ) {
 		Error( "Push: locals stack overflow\n" );
 	}
-	*( intptr_t * )&localstack[ localstackUsed ]	= value;
+	// DG: fix for 64bit Big Endian machines
+	//*( intptr_t * )&localstack[ localstackUsed ]	= value;
+	int val = value;
+	assert(value == val && "I really assumed that value would always fit into 32bit");
+	// dhewm3 used to store these values on localstack[] as intptr_t, even though
+	// they always seem to be int32. Unfortunately, all the code reading localstack[]
+	// gets the pointer to &localstack[ localstackUsed ] and then interprets that
+	// as int* or float* or whatever.
+	// So with 64bit machines it's stored as the wrong size (intptr_t is int64, int is int32),
+	// and on Big Endian the values are just 0 (or UINT_MAX if value < 0) - on Little Endian
+	// the first 4 bytes are the ones with actual data so the bug was hidden).
+	// To fix this, now a regular itn is stored at `&localstack[ localstackUsed ]`,
+	// as expected by the code using localstack[].
+	// TODO: once this has been tested more (and the assertion above has never triggered),
+	//       change idInterpreter::Push() to take a regular int argument instead of intptr_t.
+	int *stackVar = ( int * )&localstack[ localstackUsed ];
+	*stackVar = val;
+
+	// even though a 32bit int is put onto the stack, increase it by sizeof(intptr_t)
+	// so it remains aligned to multiples of the native pointer size
 	localstackUsed += sizeof( intptr_t );
 }
 

neo/idlib/Heap.h

@@ -139,6 +139,24 @@ __inline void operator delete[]( void *p ) {
 #endif /* ID_DEBUG_MEMORY */
 
 
+// allocate SIZE bytes, aligned to 16 bytes - possibly on the stack (like _alloca16())
+// if it's too big (> ID_MAX_ALLOCA_SIZE, 1MB), it gets allocated on the Heap instead.
+// ON_STACK should be a bool and will be set to true if it was allocated on the stack
+// and false if it was allocated on the heap.
+// if ON_STACK is false, you must free this with Mem_FreeA() or Mem_Free16()!
+// (just pass your ON_STACK bool to Mem_FreeA() and it will do the right thing)
+#define Mem_MallocA( SIZE, ON_STACK ) \
+	( (SIZE) < ID_MAX_ALLOCA_SIZE ? ( ON_STACK=true, _alloca16(SIZE) ) : ( ON_STACK=false, Mem_Alloc16(SIZE) ) )
+
+// free memory allocated with Mem_MallocA()
+ID_INLINE void Mem_FreeA( void* ptr, bool onStack )
+{
+	if( !onStack ) {
+		Mem_Free16( ptr );
+	}
+}
+
+
 /*
 ===============================================================================
 

neo/libs/imgui/imconfig.h

@@ -137,3 +137,13 @@ namespace ImGui
     void MyFunction(const char* name, MyMatrix44* mtx);
 }
 */
+
+// DG: on Big Endian systems the order of color channels in an uint32 color is inverted.
+//     this seems to be the official way to support Big Endian platforms in ImGUI:
+#if D3_IS_BIG_ENDIAN
+  #define IM_COL32_R_SHIFT    24
+  #define IM_COL32_G_SHIFT    16
+  #define IM_COL32_B_SHIFT    8
+  #define IM_COL32_A_SHIFT    0
+  #define IM_COL32_A_MASK     0x000000FF
+#endif

neo/renderer/ModelOverlay.cpp

@@ -118,8 +118,16 @@ void idRenderModelOverlay::CreateOverlay( const idRenderModel *model, const idPl
 	}
 
 	// make temporary buffers for the building process
-	overlayVertex_t	*overlayVerts = (overlayVertex_t *)_alloca( maxVerts * sizeof( *overlayVerts ) );
-	glIndex_t *overlayIndexes = (glIndex_t *)_alloca16( maxIndexes * sizeof( *overlayIndexes ) );
+	// DG: using Mem_MallocA() instead of alloca() to avoid stack overflows with large models
+	size_t vertSize = maxVerts * sizeof( overlayVertex_t );
+	bool vertsOnStack;
+	overlayVertex_t	*overlayVerts = (overlayVertex_t *)Mem_MallocA( vertSize, vertsOnStack );
+
+	// Note: here we have two Mem_MallocA() calls, this relies on the stack being
+	//       big enough for two alloca(ID_MAX_ALLOCA_SIZE) calls!
+	size_t idxSize = maxIndexes * sizeof( glIndex_t );
+	bool idxOnStack;
+	glIndex_t *overlayIndexes = (glIndex_t *)Mem_MallocA( idxSize, idxOnStack );
 
 	// pull out the triangles we need from the base surfaces
 	for ( surfNum = 0; surfNum < model->NumBaseSurfaces(); surfNum++ ) {
@@ -225,6 +233,9 @@ void idRenderModelOverlay::CreateOverlay( const idRenderModel *model, const idPl
 			materials[i]->surfaces.RemoveIndex( 0 );
 		}
 	}
+
+	Mem_FreeA(overlayVerts, vertsOnStack);
+	Mem_FreeA(overlayIndexes, idxOnStack);
 }
 
 /*