摘要:
A mechanism is described for facilitating improved copying of graphics data at computing devices according to one embodiment. A method of embodiments, as described herein, includes detecting a first data having a first set of primitives at a one-dimensional (“1D”) source buffer. The first data is detected to be copied to a 1D destination buffer. The method may further include re-describing the 1D source buffer and the 1D destination buffer into a two-dimensional (“2D”) source buffer and a 2D destination buffer, respectively, where re-describing may include re-describing the first data having the first set of primitives to a second data having a second set of primitives. The method may further include copying the second data having the second set of primitives from the 2D source buffer to the 2D destination buffer.
摘要:
Techniques to sort events of a graphics workload executed by a graphics processing unit to provide identification of events, that if addressed, may result in an improvement in performance are disclosed. The techniques can include: generating a signature and a weight for each event of a graphics workload; generating an event priority tree by organizing the events into parent and leaf nodes, where parent nodes comprise leaf nodes having a shared hash; and sorting frames based on a global weight of events corresponding to the frames.
摘要:
Various embodiments are presented herein that reallocate partitions of a shared physical memory between processing units. An apparatus and a computer-implemented method may determine an amount of memory space in the physical memory allocated to a first processing unit during system initialization. The determined amount of the memory space may be consolidated. The consolidated memory space may be allocated to the second processing unit during runtime. Other embodiments are described and claimed.