<html> <head> <base href="https://bugs.freedesktop.org/" /> </head> <body> <div> <a class="bz_bug_link bz_status_NEW " title="NEW - [HSW, regression, bisected] i965/pixel_read: Use meta_pbo_GetTexSubImage for PBO ReadPixels (ef0499af255ecd)" href="https://bugs.freedesktop.org/show_bug.cgi?id=89314#c19">Comment # 19</a> on <a class="bz_bug_link bz_status_NEW " title="NEW - [HSW, regression, bisected] i965/pixel_read: Use meta_pbo_GetTexSubImage for PBO ReadPixels (ef0499af255ecd)" href="https://bugs.freedesktop.org/show_bug.cgi?id=89314">bug 89314</a> from <a class="email" href="mailto:kenneth@whitecape.org" title="Kenneth Graunke <kenneth@whitecape.org>"> Kenneth Graunke</a> <pre>It seems pretty clear to me. The specification allows rounding in either direction. Reading from GPU memory into GPU memory (a PBO) vs. reading into CPU memory (a malloc'd buffer) are different operations, and already have differing properties. For example, reading into CPU memory could stall due to cross-device synchronization, while reading into GPU memory might be fully pipelined and only stall if/when the buffer is eventually mapped. Those are substantially different behaviors; applications don't expect them to behave identically. Our implementation of GetTexSubImage/ReadPixels into CPU memory is valid. Our implementation of GetTexSubImage/ReadPixels into GPU memory is also valid. The GL implementation is expected to perform these operations as quickly as possible, so it needs to be free to choose the most performant valid implementation as it sees fit. The fact that they differ, ever so slightly, is not a problem. Both results are allowed by the specification.</pre> </div> <hr> You are receiving this mail because: <ul> <li>You are the QA Contact for the bug.</li> </ul> </body> </html>