Rockbox

Tasklist

FS#11716 - mpa: move dct32 to IRAM

Attached to Project: Rockbox
Opened by Andree Buschmann (Buschel) - Wednesday, 03 November 2010, 21:32 GMT
Last edited by Andree Buschmann (Buschel) - Sunday, 07 November 2010, 22:31 GMT
Task Type Patches
Category Codecs
Status Closed
Assigned To No-one
Operating System All players
Severity Low
Priority Normal
Reported Version Release 3.6
Due in Version Undecided
Due Date Undecided
Percent Complete 100%
Votes 0
Private No

Details

Speed-up on S5L870x (nano 2g) is ~4-5%. No impact on PP502x / PP5002.

Should be measured on other CPUs as well.
   mpa_v01.patch (0.5 KiB)
This task depends upon

Closed by  Andree Buschmann (Buschel)
Sunday, 07 November 2010, 22:31 GMT
Reason for closing:  Invalid
Comment by Andree Buschmann (Buschel) - Thursday, 04 November 2010, 20:34 GMT
Tested with mp3 local testfile, 220 kbps.

nano 2g (S5L8701): svn 56.2 MHz, v01 53.7 MHz
Comment by Andree Buschmann (Buschel) - Sunday, 07 November 2010, 09:41 GMT
Minor refacturing of the iram-defines. During refacturing I came by the fact that ICODE was used for Coldfire CPUs MCF5250 and MCF5249. Normally those CPUs suffer from moving code to IRAM, therefor the v02-patch do not use ICODE for those both CPUs anymore. Patch needs to be benchmarked on those CPUs.
   mpa_v02.patch (1.7 KiB)
Comment by Andree Buschmann (Buschel) - Sunday, 07 November 2010, 11:53 GMT
v02 tests by n1s:
h300 (mcf5249): svn = 26.50MHz, patch = 26.47MHz for a 128k sample
fuzev1 (AS3525): patch 1 MHz slower than svn
Comment by Andree Buschmann (Buschel) - Sunday, 07 November 2010, 13:24 GMT
Sorry to say, but the speed-up on S5L seems to be a mis-measurement by myself. I cannot reproduce the result...

Loading...