AMCC suggested to set the PMU bit to 0 for best performace on the
PPC440 DDR controller. The 440er common DDR setup files (sdram.c &
spd_sdram.c) are changed accordingly. So all 440er boards using
these setup routines will automatically receive this performance
increase.

Please see below some benchmarks done by AMCC to demonstrate this
performance changes:

----------------------------------------
SDRAM0_CFG0[PMU] = 1 (U-boot default for Bamboo, Yosemite and Yellowstone)
----------------------------------------
Stream benchmark results
-------------------------------------------------------------
This system uses 8 bytes per DOUBLE PRECISION word.
-------------------------------------------------------------
Array size = 2000000, Offset = 0
Total memory required = 45.8 MB.
Each test is run 10 times, but only
the *best* time for each is used.
-------------------------------------------------------------
Your clock granularity/precision appears to be 1 microseconds.
Each test below will take on the order of 112345 microseconds.
(= 112345 clock ticks)
Increase the size of the arrays if this shows that you are not getting
at least 20 clock ticks per test.
-------------------------------------------------------------
WARNING -- The above is only a rough guideline.
For best results, please be sure you know the precision of your system
timer.
-------------------------------------------------------------
Function Rate (MB/s) RMS time Min time Max time
Copy: 256.7683 0.1248 0.1246 0.1250
Scale: 246.0157 0.1302 0.1301 0.1302
Add: 255.0316 0.1883 0.1882 0.1885
Triad: 253.1245 0.1897 0.1896 0.1899

TTCP Benchmark Results
ttcp-t: socket
ttcp-t: connect
ttcp-t: buflen=8192, nbuf=2048, align=16384/0, port=5000 tcp ->
localhost
ttcp-t: 16777216 bytes in 0.28 real seconds = 454.29 Mbit/sec +++
ttcp-t: 2048 I/O calls, msec/call = 0.14, calls/sec = 7268.57
ttcp-t: 0.0user 0.1sys 0:00real 60% 0i+0d 0maxrss 0+2pf 3+1506csw

----------------------------------------
SDRAM0_CFG0[PMU] = 0 (Suggested modification)
Setting PMU = 0 provides a noticeable performance improvement *2% to
5% improvement in memory performance.
*Improves the Mbit/sec for TTCP benchmark by almost 76%.
----------------------------------------
Stream benchmark results
-------------------------------------------------------------
This system uses 8 bytes per DOUBLE PRECISION word.
-------------------------------------------------------------
Array size = 2000000, Offset = 0
Total memory required = 45.8 MB.
Each test is run 10 times, but only
the *best* time for each is used.
-------------------------------------------------------------
Your clock granularity/precision appears to be 1 microseconds.
Each test below will take on the order of 120066 microseconds.
(= 120066 clock ticks)
Increase the size of the arrays if this shows that you are not getting
at least 20 clock ticks per test.
-------------------------------------------------------------
WARNING -- The above is only a rough guideline.
For best results, please be sure you know the precision of your system
timer.
-------------------------------------------------------------
Function Rate (MB/s) RMS time Min time Max time
Copy: 262.5167 0.1221 0.1219 0.1223
Scale: 258.4856 0.1238 0.1238 0.1240
Add: 262.5404 0.1829 0.1828 0.1831
Triad: 266.8594 0.1800 0.1799 0.1802

TTCP Benchmark Results
ttcp-t: socket
ttcp-t: connect
ttcp-t: buflen=8192, nbuf=2048, align=16384/0, port=5000 tcp ->
localhost
ttcp-t: 16777216 bytes in 0.16 real seconds = 804.06 Mbit/sec +++
ttcp-t: 2048 I/O calls, msec/call = 0.08, calls/sec = 12864.89
ttcp-t: 0.0user 0.0sys 0:00real 46% 0i+0d 0maxrss 0+2pf 120+1csw

2006-07-28, Stefan Roese