Subject: Re: (No Subject)
From: Craig Chambers (chambers@cs.washington.edu)
Date: Wed Apr 19 2000 - 13:37:55 PDT
Mayur Naik wrote:
>
> Hello
>
> The experimental assessment in [CC99] considers the space cost of a pointer to be 1 byte. Wouldn't 4 bytes be more realistic?
Why do you say it's only a byte? I intended it to be 4. Actual SPARC data.
>
> I think there are a few minor errors in Table 3. For instance, the minimum number of =, <, and [] tests generated using the (=,<,[]) technique in both statically and dynamically typed versions is 0 but the size is 2 bytes and the avg. dispatch time is 2.5 cycles. The number of = tests must be changed to 1.
The reason for the apparent discrepancy is that (I just discovered) you can't
compare down columns, but only across rows. The minimum number of = tests in a
dispatch tree may be zero, and similarly for < and [] tests, but these minima
aren't all from the same tree. In each dispatch tree there was at least one =,
<, or [] test, leading to a minimum size > 0. I admit that this way of
reporting data can be confusing. I wish I knew of a more meaningful was of
presenting these results.
-- Craig
This archive was generated by hypermail 2b25 : Tue Oct 03 2000 - 15:21:30 PDT