<br><br><div class="gmail_quote">On Fri, May 18, 2012 at 11:28 AM, Brian Paul <span dir="ltr"><<a href="mailto:brianp@vmware.com" target="_blank">brianp@vmware.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<div class="im">On 05/18/2012 10:11 AM, Jose Fonseca wrote:<br>
</div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<br>
<br>
----- Original Message -----<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div class="im">
<br>
A while back I noticed that the piglit roundmode-pixelstore and<br>
roundmode-getinteger tests pass on my 64-bit Fedora system but fail<br>
on<br>
a 32-bit Ubuntu system. Both glGetIntegerv() and glPixelStoref()<br>
use<br>
the IROUND() function to convert floats to ints.<br>
<br>
The implementation if IROUND() that uses the x86 fistp instruction is<br>
protected with:<br>
<br></div>
#if defined(USE_X86_ASM)&& defined(__GNUC__)&& defined(__i386__)<div class="im"><br>
<br>
but that evaluates to 0 on x86-64 (neither USE_X86_ASM nor __i386__<br>
are defined) so we use the C fallback:<br>
<br>
#define IROUND(f) ((int) (((f)>= 0.0F) ? ((f) + 0.5F) : ((f) -<br>
0.5F)))<br>
<br>
The C version of IROUND() does what we want for the piglit tests but<br>
not the x86 version. I think the default x86 rounding mode is<br>
FE_UPWARD so that explains the failures.<br>
<br>
<br>
So I think I'd like to do the following:<br>
<br>
1. Enable the x86 fistp-based functions in imports.h for x86-64.<br>
</div></blockquote><div class="im">
<br>
It's illegal/inneficient to use x87 on x86-64. We should use the appropriate SSE intrisinsic instead.<br></div></blockquote></blockquote><div><br></div><div>The instruction is "cvtss2si". Even if you use SSE here, you depend on the rounding mode in the MXCSR register, which means you'll have to set that, because some applications change this mode to use a faster or more precise rounding mode. It's the parallel problem that you have with "fistp".</div>
<div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div class="im">
<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
2. Rename IROUND() to IROUND_FAST() and define it as float->int<br>
conversion by whatever method is fastest.<br>
<br>
3. Define IROUND() as round to nearest int. For the x86 fistp<br>
implementation this would involve setting/restoring the rounding<br>
mode.<br></blockquote></div></blockquote></blockquote><div><br></div><div>If I recall, it is generally run with some other rounding mode other than "truncate" by default, so usually float -> int conversions that involve truncation (C cast) require changing the rounding mode <i>to truncation</i>. This was such a problem that in SSE3 there is "fisttp" which is "FP integer store with truncation". I guess though if the default rounding mode causes problems, there isn't much that can be done but change it each time.</div>
<div><br></div><div>Patrick</div></div>