AGA Fetch mode testing

earok · 28 September 2017, 23:32

I've checked in a small Blitz app (rapidly coded so the source code is a mess) into The Zone which does some tests with different fetch modes. The results surprised me a little:

- It doesn't appear to have a noticeable effect on CPU operations, but:
- Fetch Modes 1 and 2 allow about 35% more blits to be done in the same time, Fetch Mode 3 allows about 50% more blits.

Daedalus · 29 September 2017, 01:25

The code will be waiting on the blits to finish, so CPU usage differences will be hard to detect from that alone - your code will always be throttled by the blitter. Fetch modes 1 and 2 have about the same bandwidth (one is double width, so 32 bit transfers instead of 16 bit, the other is double clocked, so 16 bit transfers at double speed, but I can't remember which is which), and fetch mode 3 is both combined (double speed and double width for 4x bandwidth). So that's where your extra blitter time is coming from since less time is spent fetching the display.

28 September 2017, 23:32	#1
earok Registered User Join Date: Dec 2013 Location: Auckland Posts: 3,539	AGA Fetch mode testing I've checked in a small Blitz app (rapidly coded so the source code is a mess) into The Zone which does some tests with different fetch modes. The results surprised me a little: - It doesn't appear to have a noticeable effect on CPU operations, but: - Fetch Modes 1 and 2 allow about 35% more blits to be done in the same time, Fetch Mode 3 allows about 50% more blits.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
[Solved] FS-UAE doesn't fetch my WHDLoad.key file.	browniesrgut	support.FS-UAE	6	28 February 2014 23:44
indivision aga and ham mode	pbareges	support.Hardware	3	03 February 2009 12:25
What's the best screen mode for an AGA shoot-um-up	NovaCoder	Coders. General	5	09 January 2009 13:19
Stop Interlace in AGA mode	andy_blah	support.WinUAE	135	21 August 2008 19:43
Details of the AGA double CAS fetch mode?	Leffmann	Coders. General	3	29 July 2008 11:26

29 September 2017, 01:25	#2
Daedalus Registered User Join Date: Jun 2009 Location: Dublin, then Glasgow Posts: 6,334	The code will be waiting on the blits to finish, so CPU usage differences will be hard to detect from that alone - your code will always be throttled by the blitter. Fetch modes 1 and 2 have about the same bandwidth (one is double width, so 32 bit transfers instead of 16 bit, the other is double clocked, so 16 bit transfers at double speed, but I can't remember which is which), and fetch mode 3 is both combined (double speed and double width for 4x bandwidth). So that's where your extra blitter time is coming from since less time is spent fetching the display.

Currently Active Users Viewing This Thread: 1 (0 members and 1 guests)