* * Instead we now try to track the sizes of allocated buffer objects. * We watch calls to glBindBuffer to see which buffer is bound to * which target and to glBufferData to see how large the buffer's * allocated size is. When glMapBuffer is called, we consult our table * of buffer sizes to see if we can return an answer without a glGet * call.
* * We share the GLBufferSizeTracker objects among all GLContexts for * which sharing is enabled, because the namespace for buffer objects * is the same for these contexts.
* * Tracking the state of which buffer objects are bound is done in the * GLBufferStateTracker and is not completely trivial. In the face of * calls to glPushClientAttrib / glPopClientAttrib we currently punt * and re-fetch the bound buffer object for the state in question; * see, for example, glVertexPointer and the calls down to * GLBufferStateTracker.getBoundBufferObject(). Note that we currently * ignore new binding targets such as GL_TRANSFORM_FEEDBACK_BUFFER_NV; * the fact that new binding targets may be added in the future makes * it impossible to cache state for these new targets.
* * Ignoring new binding targets, the primary situation in which we may * not be able to return a cached answer is in the case of an error, * where glBindBuffer may not have been called before trying to call * glBufferData. Also, if external native code modifies a buffer * object, we may return an incorrect answer. (FIXME: this case * requires more thought, and perhaps stochastic and * exponential-fallback checking. However, note that it can only occur * in the face of external native code which requires that the * application be signed anyway, so there is no security risk in this * area.) */ public class GLBufferSizeTracker { // Map from buffer names to sizes. // Note: should probably have some way of shrinking this map, but // can't just make it a WeakHashMap because nobody holds on to the // keys; would have to always track creation and deletion of buffer // objects, which is probably sub-optimal. The expected usage // pattern of buffer objects indicates that the fact that this map // never shrinks is probably not that bad. private IntLongHashMap bufferSizeMap; protected static final boolean DEBUG = Debug.isPropertyDefined("jogl.debug.GLBufferSizeTracker", true); public GLBufferSizeTracker() { bufferSizeMap = new IntLongHashMap(); bufferSizeMap.setKeyNotFoundValue(0xFFFFFFFFFFFFFFFFL); } public final void setBufferSize(GLBufferStateTracker bufferStateTracker, int target, GL caller, long size) { // Need to do some similar queries to getBufferSize below int buffer = bufferStateTracker.getBoundBufferObject(target, caller); if (buffer != 0) { setDirectStateBufferSize(buffer, caller, size); } // We don't know the current buffer state. Note that the buffer // state tracker will have made the appropriate OpenGL query if it // didn't know what was going on, so at this point we have nothing // left to do except drop this piece of information on the floor. } public final void setDirectStateBufferSize(int buffer, GL caller, long size) { bufferSizeMap.put(buffer, size); } public final long getBufferSize(GLBufferStateTracker bufferStateTracker, int target, GL caller) { // See whether we know what buffer is currently bound to the given // state final int buffer = bufferStateTracker.getBoundBufferObject(target, caller); if (0 != buffer) { return getBufferSizeImpl(target, buffer, caller); } // We don't know what's going on in this case; query the GL for an answer // FIXME: both functions return 'int' types, which is not suitable, // since buffer lenght is 64bit ? int[] tmp = new int[1]; caller.glGetBufferParameteriv(target, GL.GL_BUFFER_SIZE, tmp, 0); if (DEBUG) { System.err.println("GLBufferSizeTracker.getBufferSize(): no cached buffer information"); } return (long) tmp[0]; } public final long getDirectStateBufferSize(int buffer, GL caller) { return getBufferSizeImpl(0, buffer, caller); } private final long getBufferSizeImpl(int target, int buffer, GL caller) { // See whether we know the size of this buffer object; at this // point we almost certainly should if the application is // written correctly long sz = bufferSizeMap.get(buffer); if (0xFFFFFFFFFFFFFFFFL == sz) { // For robustness, try to query this value from the GL as we used to // FIXME: both functions return 'int' types, which is not suitable, // since buffer lenght is 64bit ? int[] tmp = new int[1]; if(0==target) { // DirectState .. if(caller.isFunctionAvailable("glGetNamedBufferParameterivEXT")) { caller.getGL2().glGetNamedBufferParameterivEXT(buffer, GL.GL_BUFFER_SIZE, tmp, 0); } else { throw new GLException("Error: getDirectStateBufferSize called with unknown state and GL function 'glGetNamedBufferParameterivEXT' n/a to query size"); } } else { caller.glGetBufferParameteriv(target, GL.GL_BUFFER_SIZE, tmp, 0); } if (tmp[0] == 0) { // Assume something is wrong rather than silently going along throw new GLException("Error: buffer size returned by "+ ((0==target)?"glGetNamedBufferParameterivEXT":"glGetBufferParameteriv")+ " was zero; probably application error"); } // Assume we just don't know what's happening sz = (long) tmp[0]; bufferSizeMap.put(buffer, sz); if (DEBUG) { System.err.println("GLBufferSizeTracker.getBufferSize(): made slow query to cache size " + sz + " for buffer " + buffer); } } return sz; } // This should be called on any major event where we might start // producing wrong answers, such as OpenGL context creation and // destruction if we don't know whether there are other currently- // created contexts that might be keeping the buffer objects alive // that we're dealing with public final void clearCachedBufferSizes() { bufferSizeMap.clear(); } }