[Openmcl-devel] Short-float in 1.1_060826...

[Gary Byers]

Thanks.  The bug (which has been there for months and only affects PPC32)
is in the kernel.


cvs diff -u -r1.93 ppc-spentry.s
--- lisp-kernel/ppc-spentry.s	21 Jun 2006 21:15:19 -0000	1.93
+++ lisp-kernel/ppc-spentry.s	28 Aug 2006 23:17:51 -0000
@@ -4333,7 +4333,7 @@
  	 __(blr)
  local_label(misc_set_single_float_vector):
           __(extract_lisptag(imm2,arg_z))
-         __(cmpwi cr7,imm0,tag_misc)
+         __(cmpwi cr7,imm2,tag_misc)
           __(la imm0,misc_data_offset(arg_y))
  	 __(bne- cr7,local_label(set_bad))
  	 __(extract_header(imm2,arg_z))


What that's saying is that in the checked-in version, the code typechecks
to ensure that the value being stored in the single-float vector is in
fact a single-float by first getting its tag to the imm2 register and
then comparing the (random) value in the imm0 register to "tag_misc";
it'd work better if that comparison ("cmpwi") involved imm2 instead of
imm0 ...

On Tue, 29 Aug 2006, Alberto Santini wrote:

> Hello,
>  I report you a little feedback about 1.1_060826.
>
> Compiling the code (dated 1991) you can find at this url:
>
> <http://www.cs.cmu.edu/afs/cs/project/ai-repository/ai/areas/neural/systems/cascor/lisp/cascor1.cl>
>
> I see some difference between 1.0 and 1.1_060826.
>
> With 1.0 this example works fine, but not with 1.1:
>
>  (make-array 10 :element-type 'short-float :initial-element 0.0)
>
> With 1.1 it doesn't work using nor (coerce 'short-float 0.0)
> nor 0.0S0 in :initial-element.
>
> But with 1.1 the above code is compiled without problem: ok we have
> a problem with short-float. Compiling with 1.0 the above code gives
> the following error:
>
> While compiling an anonymous function :
> unsupported vector type: :SINGLE-FLOAT-VECTOR
>   [Condition of type CCL::COMPILE-TIME-PROGRAM-ERROR]
>
>
> Sorry... I haven't the knowledge to investigate further.
>
> Thanks in advance,
> Alberto Santini
> _______________________________________________
> Openmcl-devel mailing list
> Openmcl-devel at clozure.com
> http://clozure.com/mailman/listinfo/openmcl-devel
>
>