On 8/02/11 12:01 PM, Tristan Gingold wrote:
> David,
>
> feel free to post a patch. You seem to be able to track down this issue
> completely.
>
> Tristan.
>
With a ghdl-0.29 source distribution
cd ghdl/vhdl (or 'cd ghdl' for svn distribution)
patch -c < PATH_TO_EVALUATION_DIFF/evaluation.diff
The diff/patch file was produced by:
diff -c evaluation.adb ~/Desktop/ghdl/evaluation.adb
Note the patch target file name is evaluation.adb in the immediate context.
Three lines plus a comment.
A shorter description:
The patch basically short circuits further processing when rol and ror
rotation is zero after modulo reduction. At this point in the code Dir_Left
is the actual rotation direction, True for left. For actual right rotations
where Cnt is 0, Cnt will be assigned Len - Cnt which is Len (array length)
and Cnt becomes an offset index. Cnt won't point to a valid element of
Array_List in the later Get_Nth_Element function calls if it is equal to Len.
I don't currently have a build environment handy to try it in.
*** evaluation.adb 2010-02-28 10:37:11.000000000 +1300
--- /Users/david_koontz/Desktop/ghdl/evaluation.adb 2011-02-08
14:02:31.000000000 +1300
***************
*** 677,682 ****
--- 677,687 ----
when Iir_Predefined_Array_Rol
| Iir_Predefined_Array_Ror =>
Cnt := Cnt mod Len;
+ -- if Cnt modulo Len = 0, return original Left array
+ if Cnt = 0 then
+ return Build_Simple_Aggregate
+ (Arr_List, Origin, Get_Type (Left));
+ end if;
if not Dir_Left then
Cnt := Len - Cnt;
end if;
_______________________________________________
Ghdl-discuss mailing list
[email protected]
https://mail.gna.org/listinfo/ghdl-discuss
