User Tools

Site Tools


floating_point_support

Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Both sides previous revision Previous revision
floating_point_support [2012/09/04 18:40]
xiryan
floating_point_support [2012/10/04 22:44] (current)
xiryan
Line 11: Line 11:
  
 ===Known Issues=== ===Known Issues===
-  *IR not passing ​- A few of the return values are very close to expected (One case was 1 off, and 4 in another) ​ For these cases, I'm suspecting that it's issue with the accuracy of FP representations. ​ FPToSI issue actually was found because ​of these cases.  ​Howeverthere are cases that are completely off, and one case even returned a 0+  *IR not passing. ​There'​s ​an issue with FPtoSI where when the input FP is out of range of integer, LLVM would return 0x80000000 but Altera core returns 0x79999999.  ​If the value is used in operation such as divisionthe result becomes ​completely off. 
-  *In some of the cores, using the constant 0 directly would fail+  *In some of the cores, using the constant 0 directly would fail (using variable containing 0 is fine)
   *Scheduler will not schedule FP instruction every cycle even when there is no dependency   *Scheduler will not schedule FP instruction every cycle even when there is no dependency
floating_point_support.txt · Last modified: 2012/10/04 22:44 by xiryan