Heterogeneous Computing with OpenCL by Benedict R. Gaster & Lee Howes & David R. Kaeli & Perhaad Mistry & Dana Schaa

Author:Benedict R. Gaster & Lee Howes & David R. Kaeli & Perhaad Mistry & Dana Schaa
Language: eng
Format: epub, pdf
ISBN: 9780124055209
Publisher: Elsevier Inc.
Published: 2012-10-29T16:00:00+00:00

This is a pair of instruction packets, each containing up to four operations. The first one is only half full: this is where vector operations came in. In addition, the four banks of the register file that we see in Figure 6.7 were accessed from the four VLIW slots (with some flexibility) – such that vector variables were optimally stored. The architectural change to the four dynamically scheduled SIMD units we see in Figure 6.7 means that the extra arithmetic instructions may not be necessary, rather that this may be seen as purely an arithmetic intensity question. More significantly, use of an OpenCL short vector consumes multiple consecutive registers, and with no gain in terms of register packing efficiency this may lead to overuse of the register file. Note that four wavefronts are active in the space that would previously have been occupied by one: with extra intermediate registers to match.

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.