Hi all,
I am new to cortex m3...
For my application I need to calculate the number of cycles for a particular function....
I set the debug options as high optimization level and optimize for time....
How can I calculate the MIPS/cycles manually for a particular function.
Please suggest me a solution...
Thanks in advance.
Krish.
»
Quick Links
Page 1 of 1
Cycle count in cortex m3
Rate Topic:
#2
isogen74 
- Super Contributor
-
- Group: Members
- Posts: 1097
- Joined: 20-March 07
Posted 10 December 2009 - 09:50 AM
The low tech solution:
Run it in a loop a _lot_ of times.
Use a stop watch.
((Seconds * MHz) / Loops) = cycles per loop.
Run it in a loop a _lot_ of times.
Use a stop watch.
((Seconds * MHz) / Loops) = cycles per loop.
When optimizing software, consider that the quickest code to run is the bit you removed from the call path.
#3
Krish
- Member
-
- Group: Members
- Posts: 8
- Joined: 10-December 09
Posted 10 December 2009 - 11:58 AM
Thanks for the reply....
does armulator serve my purpose?
If so how to start with realview debugger?(any settings).
does armulator serve my purpose?
If so how to start with realview debugger?(any settings).
isogen74, on Dec 10 2009, 10:50 AM, said:
The low tech solution:
Run it in a loop a _lot_ of times.
Use a stop watch.
((Seconds * MHz) / Loops) = cycles per loop.
Run it in a loop a _lot_ of times.
Use a stop watch.
((Seconds * MHz) / Loops) = cycles per loop.
#4
mharnisch
- Contributor
-
- Group: Members
- Posts: 80
- Joined: 11-April 08
Posted 10 December 2009 - 03:08 PM
Krish, on Dec 10 2009, 01:58 PM, said:
If so how to start with realview debugger?(any settings).
The ISSM for Cortex-M3 is semi-accurate. It models an ideal memory sub-system. Not sure about the pipeline accuracy, but I guess CM3 is simple enough.
If you are really concerned about accuracy, get a cheap eval board and read the cycle count register in the DWT unit.
Regards
Marcus
#5
Krish
- Member
-
- Group: Members
- Posts: 8
- Joined: 10-December 09
Posted 11 December 2009 - 08:36 AM
Thanks Marcus,
so we can access the DWT only by connecting the board?
I tried like this but did not work....
int count = 0;
int *DWT_CPICNT = (int *)0xE0001008; //address of the register
*DWT_CPICNT = 0; // reset the counter
.......
......
count = *DWT_CPICNT;
can you please suggest me how to access the DWT unit without the board?
Thanks & regards,
Krish.
so we can access the DWT only by connecting the board?
I tried like this but did not work....
int count = 0;
int *DWT_CPICNT = (int *)0xE0001008; //address of the register
*DWT_CPICNT = 0; // reset the counter
.......
......
count = *DWT_CPICNT;
can you please suggest me how to access the DWT unit without the board?
Thanks & regards,
Krish.
#6
mharnisch
- Contributor
-
- Group: Members
- Posts: 80
- Joined: 11-April 08
Posted 11 December 2009 - 01:19 PM
Krish, on Dec 11 2009, 10:36 AM, said:
so we can access the DWT only by connecting the board?
I tried like this but did not work....
I tried like this but did not work....
Not sure if this part of DWT is modelled in ISSM. Did you follow the initialisation procedure in the Cortex-M3 TRM?
Regards
Marcus
#7 Guest_Joseph Yiu_*
Posted 11 December 2009 - 06:29 PM
Hi Krish,
I few things you need to do:
1. Define the registers as "volatile", and "unsigned"
2. Enable TRCENA bit in DEMCR (0xE000EDFC, bit 24) before using DWT
3. You should use DWT's Cycle Count Register (0xE0001004) for timing measurement.
The CPI counter is only 8-bit, and need to have trace support to detect the number of time it overflow.
While the DWT Cycle Count Register is 32-bit.
4. You need to enable the Cycle Count counter.
So based on the code you have prepared, it can be modified to:
int count = 0;
volatile unsigned int *DWT_CYCCNT = (int *)0xE0001004; //address of the register
volatile unsigned int *DWT_CONTROL = (int *)0xE0001000; //address of the register
volatile unsigned int *SCB_DEMCR = (int *)0xE000EDFC; //address of the register
*SCB_DEMCR = *SCB_DEMCR | 0x01000000;
*DWT_CYCCNT = 0; // reset the counter
*DWT_CONTROL = *DWT_CONTROL | 1 ; // enable the counter
.......
......
count = *DWT_CYCCNT;
I haven't test this code in hardware but it should work.... :-)
The ISSM might not have the detail model of debug components, so it is best to do the timing measurement on actual hardware, either on actual microcontroller or on our FPGA prototyping system.
Note that if the debugger you are using is trying to use this counter at the same time (e.g. if you have got some sort of profiling feature turn on in your debug environment), you might find incorrect profiling result by doing this.
regards,
Joseph
I few things you need to do:
1. Define the registers as "volatile", and "unsigned"
2. Enable TRCENA bit in DEMCR (0xE000EDFC, bit 24) before using DWT
3. You should use DWT's Cycle Count Register (0xE0001004) for timing measurement.
The CPI counter is only 8-bit, and need to have trace support to detect the number of time it overflow.
While the DWT Cycle Count Register is 32-bit.
4. You need to enable the Cycle Count counter.
So based on the code you have prepared, it can be modified to:
int count = 0;
volatile unsigned int *DWT_CYCCNT = (int *)0xE0001004; //address of the register
volatile unsigned int *DWT_CONTROL = (int *)0xE0001000; //address of the register
volatile unsigned int *SCB_DEMCR = (int *)0xE000EDFC; //address of the register
*SCB_DEMCR = *SCB_DEMCR | 0x01000000;
*DWT_CYCCNT = 0; // reset the counter
*DWT_CONTROL = *DWT_CONTROL | 1 ; // enable the counter
.......
......
count = *DWT_CYCCNT;
I haven't test this code in hardware but it should work.... :-)
The ISSM might not have the detail model of debug components, so it is best to do the timing measurement on actual hardware, either on actual microcontroller or on our FPGA prototyping system.
Note that if the debugger you are using is trying to use this counter at the same time (e.g. if you have got some sort of profiling feature turn on in your debug environment), you might find incorrect profiling result by doing this.
regards,
Joseph
#9
BillNeifert
- Member
-
- Group: Members
- Posts: 2
- Joined: 26-January 10
Posted 26 January 2010 - 07:36 PM
The best way to get accurate results for this is to use accurate models. Carbon Design Systems has models of the M3 (as well as all the other processors and peripherals) which are compiled directly from ARM's RTL and are 100% implementation accurate. These models can work in pretty much any virtual or RTL environment and also feature debugger integrations to enable you to debug your software as well.
<disclaimer> I work for Carbon
<disclaimer> I work for Carbon
Share this topic:
Page 1 of 1
