[patch][rfa] Inheriting from cache_component and friends

[Dave Brolley]

Following up on my recent posting regarding specializing on the existing 
cache_component. Here is the patch which allows use of cache_component 
and cache_line as base classes. The purpose of this is to allow me to 
implement a target-specific cache with a unique cache line 
implementation. The main changes are:

1) Make cache_line into a virtual base class. The original cache_line 
class is now called internal_cache_line (it keeps its data internally) 
and its functionality is unchanged.

2) Introduce a new cache_line_factory class which is used to allocate 
cache lines. Each cache_component is now passed a reference to a 
cache_line_factory which it uses to allocate its cache lines.  An 
instance called internal_line_factory is used to create lines of type 
internal_cache_line.

3) The find mechanism of cache and cache_set previously passed cache 
lines by assignment via reference arguments. This is no longer possible 
with a virtual class implementation. The mechanism now returns a pointer 
to the found line (if found) and NULL otherwise.

4) Similarly, the 'replace' method of cache_replacement_algorithm 
previously passed in a new line and returned the old one by assignment 
using reference arguments. Now, a pointer to the expelled line is 
returned (if any --- all lines might be locked) and NULL otherwise. The 
expelled line is updated using access methods. The 'replace' method has 
been renamed to 'expell' since the line is no longer replaced. This also 
eliminates construction and destruction overhead for each cache miss. 
Also, the copy constructor and assignment operator for cache_line are no 
longer necessary.

5) The 'insert' and 'extract' methods of cache_line are now declared 
using a macro for the cartesian product of possible datatypes as is 
common elsewhere. It was not possible to instantiate the necessary 
termplate methods automatically in the virtual base class.

6) The changes the sid/component/cfgroot were found necessary while 
debugging problems in my implementation.

7) The changes to main/dynamic/commonCfg.h were necessary in order to 
allow me to inherit from CacheCfg for my specialized cache configurator 
and to get a handle to the loader configuration from the SessionCfg so 
that I could connect the loader's "set-endian" pin to my cache line 
implementation.

I have tested this against my own port (with specialized cache and cache 
lines) as well as against xstormy16. All of the sid testsuite checks 
without regression (including the cache-specific tests). I'm seeking 
comments and approval to commit.

Regards,
Dave


-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: sid-cache.ChangeLog
URL: <http://sourceware.org/pipermail/sid/attachments/20040427/59e05370/attachment.ksh>
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: sid-cache.patch.txt
URL: <http://sourceware.org/pipermail/sid/attachments/20040427/59e05370/attachment.txt>