Propellor Icc Measurements
tomcrawford
Posts: 1,127
Here are some Icc measurements for a single 40-pin DIP Propellor. These numbers are at room temperature. A micropower 3.3V regulator with quiescent current less than 0.1 mA is used. This is a sample of one Prop with a 32K EEPROM. All numbers are mA.
I took measurements at five PLL settings because I am doing battery-powered designs and wanted to know how much power I could save at reduced clock rates. In some cases there is an order of magnitude difference between 5 MHz and 80 MHz.
A spreadsheet is attached as a pdf file.
1. Repeat waitcnt(clkfreq*5 + cnt) This was as good a baseline as I could come up with. The PROP is almost completely idle; we still see a 3:1 difference in current.
2. Repeat symbol++. Symbol is defined as a long.
3. Parallax Serial Terminal is started (in cog 1) at 9600 baud. It is likely polling to see if anything is in the buffer, cog 0 is idling.
4. Cog 0 is repeating symbol++. Every 256 passes, it calls pst.
5. A red led connected to ground via a 1Kohm resistor is switched on and off very slowly with waitcnts between. It consistently takes 1.4 mA. This would be a 1.4V drop across the resistor, suggesting a Vforward of 1.9V. I guess that sounds about right.
6. Same LED, same resistor, returned to Vcc. Exact same results
7. Same LED, same resistor, switched on and off as fast as spin can.
8. Here we have all eight cogs running, one spin, one whatever pst is written in, and six PASM. The program is attached. Cog zero starts up pst (in cog one) and then starts up six PASM cogs. From then on, it gives each PASM cog a number to fiddle with and then calls pst to display the results returned from the six PASM cogs.
Each PASM cogs initializes itself by claiming two IO pins and calculating real addresses in hub memory. At loop it polls a flag in hub memory, executes a 16 x 6 unsigned multiply, and serializes the product. I thought this would be a reasonable simulation of a busy cog.
I took measurements at five PLL settings because I am doing battery-powered designs and wanted to know how much power I could save at reduced clock rates. In some cases there is an order of magnitude difference between 5 MHz and 80 MHz.
A spreadsheet is attached as a pdf file.
1. Repeat waitcnt(clkfreq*5 + cnt) This was as good a baseline as I could come up with. The PROP is almost completely idle; we still see a 3:1 difference in current.
2. Repeat symbol++. Symbol is defined as a long.
3. Parallax Serial Terminal is started (in cog 1) at 9600 baud. It is likely polling to see if anything is in the buffer, cog 0 is idling.
4. Cog 0 is repeating symbol++. Every 256 passes, it calls pst.
5. A red led connected to ground via a 1Kohm resistor is switched on and off very slowly with waitcnts between. It consistently takes 1.4 mA. This would be a 1.4V drop across the resistor, suggesting a Vforward of 1.9V. I guess that sounds about right.
6. Same LED, same resistor, returned to Vcc. Exact same results
7. Same LED, same resistor, switched on and off as fast as spin can.
8. Here we have all eight cogs running, one spin, one whatever pst is written in, and six PASM. The program is attached. Cog zero starts up pst (in cog one) and then starts up six PASM cogs. From then on, it gives each PASM cog a number to fiddle with and then calls pst to display the results returned from the six PASM cogs.
Each PASM cogs initializes itself by claiming two IO pins and calculating real addresses in hub memory. At loop it polls a flag in hub memory, executes a 16 x 6 unsigned multiply, and serializes the product. I thought this would be a reasonable simulation of a busy cog.
{Object_Title_and_Purpose}
CON
_clkmode = xtal1 + pll1x '5 MHz times pll multiplier
_xinfreq = 5_000_000
NPasm = 6 'number of PASM cogs to start
VAR
byte Cog
long MyPar[4]
byte debug, t1 'if not zero, start up terminal
long symbol
long IValue[NPasm] 'input values for each cog
long Flag[NPasm] '<> 0 says start, =0 says done
long Count[NPasm] 'return values from each cog
long MasterRVal '
OBJ
pst : "Parallax Serial Terminal"
PUB main
debug := 1
if Debug <> 0
pst.Start (9600) 'start up ser terminal: Note 9600 baud
pst.str(String("hello, world ")) 'runs in cog 1
pst.NewLine
MasterRVal? 'get the first random number
repeat t1 from 0 to NPasm - 1 'which cog it is
MyPar[0] := t1 'a value, not an address
MyPar[1] := @IValue 'address of list of input values
MyPar[2] := @Flag 'start/done flags
Flag[t1] := 0 'not ready to run
MyPar[3] := @Count 'Iterations
cog := cognew(@PasmCog, @MyPar) 'start up the PASM cog
pst.str(String("Start Cog: "))
pst.dec(t1)
pst.str(String(" Prop Cog Number: ")) '
pst.hex(Cog,1)
pst.NewLine
repeat
repeat t1 from 0 to NPasm - 1
IValue[t1] := MasterRVal? 'a random long
FLag[t1] := 1 'tell pasm to start
repeat while Flag[0] <> 0 'wait here until first one is done
pst.char(">")
repeat t1 from 0 to NPasm - 1 'and now print the results
pst.hex(Count[t1], 8)
pst.str(string(" "))
pst.newline
dat 'this is the PASM cog
PasmCog org 0
mov AdPar,Par 'get the address of input parameters
rdlong PasmNbr, AdPar 'this is my number
mov Work1, PasmNbr 'get a copy
shl Work1, #1 'times two
add Work1, #8 'the pins begin with IO8
mov DPMask, #1 'form the mask corresponding to my data pin
shl DPMask, Work1 'by shifting appropriately
andn outa, DPMask 'start out low for sure
or dira, DPMask 'make it an output
add Work1, #1 'to next pin number
mov CPMask, #1 'and the mask corresponding to clock pin
shl CPMask, Work1
andn outa, CPMask
or dira, CPMask
mov Work1, PasmNbr 'get my number back
shl Work1, #2 'multiply by four to get offset into tables
add AdPar, #4 'to MyPar[1] containing address of IValue list
rdlong AdIVal, AdPar 'get it
add AdIVal, Work1 'address of my very entry
add AdPar, #4 'to MyPar[2] containing address of flag list
rdlong AdFlag, AdPar 'address of flag list
add AdFlag, Work1 'my very one
add AdPar, #4 'to MyPar[3] containing address of return value list
rdlong AdCnt, AdPar 'get it
add AdCnt, Work1 'mine
loop rdlong Work1, AdFlag 'get the flag
test Work1, #$FF wz 'see if any bits set in the low order byte
if_Z jmp #loop 'just wait here if not
rdlong Work1, AdIVal 'get the input value
mov MPlyer, Work1 'the multiplier
andn MPlyer, FFFF 'extract, leaving in high order word
mov MCand, Work1 'multicand
and MCand, FFFF 'extract
mov Prod, #0 'start out zero
mov Work1, #16 'will look at 16 bits
mpyLp shl Prod, #1 'shift what we have
rol MPlyer, #1 wc 'high order bit into carry
if_c add Prod, MCand 'add multicand if needful
djnz Work1, #mpyLp '16 times
wrlong Prod, AdCnt 'store the result
mov work1, #32 'will shift out 32 bits
shlp rol Prod, #1 wc 'high order bit first
if_c or outa, DPMask 'set a one
if_nc andn outa, DPMask 'or a zero
nop 'simulate data setup time
or outa, CPMask 'clock high
nop 'clock width
andn outa,CPMask 'clock low
andn outa, DPMask 'leave data low
djnz work1, #Shlp '32 times
wrlong zero, AdFlag 'only useful for first pasm cog
jmp #loop 'and wait for another
Zero long 0 'zero value
FFFF long $FFFF 'Word-size constant
AdPar RES 'input parameters address
PasmNbr RES 'what pasm number I am
AdTab RES 'address of beginning of return table
DPMask RES 'data pin mask
CPMask RES 'clock pin mask
Work1 RES 'working register
AdIVal RES 'address of my particular input value
AdFlag RES 'address of my flag
AdCnt RES 'address of my count
MPlyer RES 'multiplier
MCand RES 'multiplicand
Prod RES 'product
