Memory Benchmark with Smart Placement

Table of Contents

• Supported Combinations
• Introduction
• Building benchmark application
• Running benchmark application

Supported Combinations

Parameter	Value
CPU + OS	r5fss0-0 freertos
Toolchain	ti-arm-clang
Boards	am263x-cc, am263x-lp
Example folder	examples/benchmarks/ocmc_benchmarking

Introduction

• This demo provides a means of measuring the performance of a realistic application where the text of the application is sitting in various memory locations and the data is sitting in On-Chip-Memory RAM (referred to as OCM, OCMC or OCMRAM).
• The application executes 10 different configurations of the same text varying by data vs. instruction cache intensity. Each test calls 16 separate functions 500 total times in random order.
• The most instruction intensive example achieves a instruction cache miss rate (ICM/sec) of ~3-4 million per second when run entirely from OCMRAM. This is a rate that we have similarly seen in real-world customer examples.
• More data intensive tests have more repetitive code, achieving much lower ICM rates.

Application Output	Description
Mem Cpy size => 100	Size of the memcpy in bytes executed by each task
Exec Time in usec => 2567	Amount of time in microseconds
Iter => 1	Number of times the test was run
Task calls => 500	Number of randomly ordered calls to the 16 tasks
Inst Cache miss => 11421	Total instruction cache misses
Inst Cache acc => 650207	Total instruction cache accesses
num switches => 1469	Number of total context switches
num instr exec => 1029260	Total number of executed instructions
ICM/sec => 4449162	Instruction cache misses per second
INST/sec => 400958317	Instructions executed per second

Please read more on smart placement at Smart Placement, before moving further ahead.

annotations.S has a macro named ENABLE_SMART_PLACEMENT_ANNOTATION which can be used to enable and disable annotations which will effectively enable and disable smart placement.

Here it will be shown how code performance (or CPI) is improved with smart placement.

Building benchmark application

To build this application, compiler ti-cgt-armllvm >= 3.2.0 LTS or later is required. Application can be compiled using make command.

Running benchmark application

This particular application is used to show the effect of smart placement. To do this, we first remove ENABLE_SMART_PLACEMENT_ANNOTATION macro (which is in annotations.S file.) and compile it. To know how annotations.S has been generated then please refer to Software Instrument Smart Placement

Flash this application using uart_uniflash. For this please go through UART Uniflash

After running the application, console logs prints out in the above table format with different configuration like Mem Cpy Size i.e. for different values of Mem Cpy Size, exec time and ICM/sec (among with others) is printed. For example,

OCMC benchmarking:: Board_init success
Countdown...
1/5
2/5
3/5
4/5
5/5
Filling up the buffers
Inst Cache Miss: 2
Inst Cache Access: 22
Data Cache Miss: 39


master_task


master_task -- start sending

Mem Cpy Size    => 0
Start Time in Usec => 244351
Exec Time in Usec => 11867
Iter            => 1
Task Calls      => 500
Inst Cache Miss => 87023
Inst Cache Acc  => 856803
Num Instr Exec  => 1893125
ICM/sec         => 9
ICM/sec         => 9
INST/sec        => 197
INST/sec        => 197

Mem Cpy Size    => 100
Start Time in Usec => 269072
Exec Time in Usec => 12428
Iter            => 1
Task Calls      => 500
Inst Cache Miss => 85720
Inst Cache Acc  => 995992
Num Instr Exec  => 2087348
ICM/sec         => 8
ICM/sec         => 8
INST/sec        => 207
INST/sec        => 207

Mem Cpy Size    => 200
Start Time in Usec => 298477
Exec Time in Usec => 13130
Iter            => 1
Task Calls      => 500
Inst Cache Miss => 86529
Inst Cache Acc  => 1141634
Num Instr Exec  => 2287056
ICM/sec         => 8
ICM/sec         => 8
INST/sec        => 215
INST/sec        => 215

Mem Cpy Size    => 400
Start Time in Usec => 325456
Exec Time in Usec => 14606
Iter            => 1
Task Calls      => 500
Inst Cache Miss => 88361
Inst Cache Acc  => 1446200
Num Instr Exec  => 2688209
ICM/sec         => 7
ICM/sec         => 7
INST/sec        => 227
INST/sec        => 227

Mem Cpy Size    => 1000
Start Time in Usec => 354810
Exec Time in Usec => 18367
Iter            => 1
Task Calls      => 500
Inst Cache Miss => 84032
Inst Cache Acc  => 2342630
Num Instr Exec  => 3879466
ICM/sec         => 5
ICM/sec         => 5
INST/sec        => 260
INST/sec        => 260

Mem Cpy Size    => 1500
Start Time in Usec => 388418
Exec Time in Usec => 23678
Iter            => 1
Task Calls      => 500
Inst Cache Miss => 86357
Inst Cache Acc  => 3106798
Num Instr Exec  => 4892362
ICM/sec         => 4
ICM/sec         => 4
INST/sec        => 255
INST/sec        => 255

Mem Cpy Size    => 2000
Start Time in Usec => 426769
Exec Time in Usec => 28432
Iter            => 1
Task Calls      => 500
Inst Cache Miss => 86051
Inst Cache Acc  => 3855309
Num Instr Exec  => 5887390
ICM/sec         => 3
ICM/sec         => 3
INST/sec        => 255
INST/sec        => 255

Mem Cpy Size    => 2250
Start Time in Usec => 471196
Exec Time in Usec => 31236
Iter            => 1
Task Calls      => 500
Inst Cache Miss => 88306
Inst Cache Acc  => 4239124
Num Instr Exec  => 6397039
ICM/sec         => 3
ICM/sec         => 3
INST/sec        => 252
INST/sec        => 252

So from the above, mem Cpy Size is changing, ICM is almost same but Exec Time in Usec is increasing which is thereby changing ICM/sec. Here to find the base line, Exec Time in Usec should be selected for the case where ICM/sec is 3 (or ~3Millions Instruction Cache Miss/Sec) and from the above data, it can be determined that when mem Cpy Size is 2000, ICM/sec was 3 and for this, it took 28432 us to execute the application. Therefore, 28432 us is our baseline number.

Now after this, ENABLE_SMART_PLACEMENT_ANNOTATION macro is defined again and example is compiled back.

running that would give similar logs and execution time against the Mem Cpy Size of 2000 is to be noted and has been measured out to be 25758 us. Following table shows the same:

Mem Cpy Size	INST/sec	Exec Time in Usec without Smart Placement	Exec Time in Usec with Smart Placement
0000	7333192	11867	9424
0100	7002172	12428	10057
0200	6627799	13130	10653
0400	5958031	14606	12105
1000	4509122	18367	15924
1500	3647140	23678	21222
2000	3026765	28432	25758
2250	2827058	31236	28583

Therefore, with smart placement code was executed in 25758us and without smart placement same code executed in 28432us which corresponds to performance gain of 10% without changing of any code.