T

T

H E

H E

W

W

O R L D

O R L D

L

L

E A D E R I N

E A D E R I N

D

D

S

S

P

P

S

S

O L U T I O N S

O L U T I O N S

Agenda

Agenda



Overview

Overview

Hello, TI DSP,

Hello, TI DSP,

‘C6000

‘C6000



DSK

DSK

Tour of ‘C6711 DSK

Tour of ‘C6711 DSK



C Code

C Code

Code Composer

Code Composer



System

System

1. McBSP,

1. McBSP,

Interrupts

Interrupts

2. EDMA, Cache

2. EDMA, Cache



DSP/BIOS

DSP/BIOS

Real-Time Tools

Real-Time Tools



VAB

VAB

Visual Design

Visual Design



Epilogue

Epilogue

What Next?

What Next?

DSP/BIOS

DSP/BIOS



Real-time Scheduling

Real-time Scheduling



Simple example of real-time

Simple example of real-time

problem

problem



HWI, SWI, and TSK

HWI, SWI, and TSK



Scheduler

Scheduler



DSP/BIOS II adds multi-tasking

DSP/BIOS II adds multi-tasking



Real-time analysis (RTA)

Real-time analysis (RTA)

To:

To:

Engineering

Engineering

From:

From:

Marketing

Marketing

Re:

Re:

Design Spec

Design Spec

Change

Change

Due to customer

Due to customer

request, we need to

request, we need to

add DTMF functions to

add DTMF functions to

the Audio Player

the Audio Player

ASAP!!!

ASAP!!!

Cannot increase

Cannot increase

system cost !!!

system cost !!!

TI DSP

TI DSP

New System Requirement -

New System Requirement -

Abstract

Abstract

DTMF

DTMF

Filter

Filter

Previous Requirement

Previous Requirement



DSP filters audio signal

DSP filters audio signal

New Requirement

New Requirement



Add DTMF function

Add DTMF function



DTMF is independent of filter

DTMF is independent of filter



Issues:

Issues:



Do we have enough bandwidth (MIPS)?

Do we have enough bandwidth (MIPS)?



Will one routine conflict with the other?

Will one routine conflict with the other?



How do we create the compound system?

How do we create the compound system?

Interrupt Driven System

Interrupt Driven System

TI DSP

TI DSP

main

main

{

{

while(1);

while(1);

}

}

Timer1_ISR

Timer1_ISR

{

{

}

}

Timer2_ISR

Timer2_ISR

{

{

}

}

B

B

A

A

The choice of most designers:

The choice of most designers:

Put each routine in it’s own ISR

Put each routine in it’s own ISR

Interrupt Driven System

Interrupt Driven System

A

A

running

running

idle

idle

Time

Time

1

1

2

2

3

3

5

5

4

4

6

6

7

7

0

0

B

B

Only one can run at a time...

Only one can run at a time...

TI DSP

TI DSP

main

main

{

{

while(1);

while(1);

}

}

Timer1_ISR

Timer1_ISR

{

{

}

}

Timer2_ISR

Timer2_ISR

{

{

}

}

B

B

A

A

Period

Period

Compute

Compute

CPU Usage

CPU Usage

Routine

Routine

A

A

:

:

22

22





s

s

11

11





s

s

(50%)

(50%)

Routine

Routine

B

B

:

:

125

125





s

s

33

33





s

s

(26%)

(26%)

76%

76%

Interrupt Driven System - Problem

Interrupt Driven System - Problem

Period

Period

Compute

Compute

CPU Usage

CPU Usage

Routine

Routine

A

A

:

:

22

22





s

s

11

11





s

s

(50%)

(50%)

Routine

Routine

B

B

:

:

125

125





s

s

33

33





s

s

(26%)

(26%)

76%

76%

Time

Time

1

1

2

2

3

3

5

5

4

4

6

6

7

7

0

0

B

B

A

A

running

running

idle

idle

y

y

1

1

y

y

2

2

y

y

3

3

y

y

4

4

Missed !

Missed !

TI DSP

TI DSP

main

main

{

{

while(1);

while(1);

}

}

Timer1_ISR

Timer1_ISR

{

{

}

}

Timer2_ISR

Timer2_ISR

{

{

}

}

B

B

A

A

There are

There are

two

two

elements of CPU loading:

elements of CPU loading:

average & instantaneous

average & instantaneous

Nested Interrupts

Nested Interrupts

A common solution is to allow

A common solution is to allow

hardware interrupts to preempt each

hardware interrupts to preempt each

other - called

other - called

'nesting'

'nesting'

interrupts

interrupts



Problem is, the user must handle all

Problem is, the user must handle all

context save and restore

context save and restore



While reasonable for 1-2 interrupts, it

While reasonable for 1-2 interrupts, it

becomes tedious and trouble-prone for

becomes tedious and trouble-prone for

more

more



Can become near impossible when

Can become near impossible when

using object-coded algorithms or

using object-coded algorithms or

libraries

libraries

(thus, making it difficult to purchase

(thus, making it difficult to purchase

algo's)

algo's)



DSP/BIOS handles this for you with

DSP/BIOS handles this for you with

simple drag -n- drop editing

simple drag -n- drop editing

TI DSP

TI DSP

main

main

{

{

while(1);

while(1);

}

}

Timer1_ISR

Timer1_ISR

{

{

}

}

Timer2_ISR

Timer2_ISR

{

{

}

}

B

B

A

A

A

A

running

running

idle

idle

Time

Time

1

1

2

2

3

3

5

5

4

4

6

6

7

7

0

0

B

B

The DSP/BIOS Solution

The DSP/BIOS Solution

B

B

A

A

main

main

{

{

return;

return;

}

}





DSP/BIOS

DSP/BIOS

A

A

running

running

idle

idle

Time

Time

1

1

2

2

3

3

5

5

4

4

6

6

7

7

0

0

B

B

DSP/BIOS provides scheduling:

DSP/BIOS provides scheduling:



You needn’t build a custom (inflexible)

You needn’t build a custom (inflexible)

state-machine for each DSP design

state-machine for each DSP design



Instead, you can leverage 1000’s of

Instead, you can leverage 1000’s of

hours of development, proven in 100’s

hours of development, proven in 100’s

of systems!

of systems!

Easy to write

Easy to write

- Modules written

- Modules written

independently

independently

Easy to maintain

Easy to maintain

- Module interaction

- Module interaction

minimized

minimized

Built-in Scheduling

Built-in Scheduling

- Managed by

- Managed by

DSP/BIOS

DSP/BIOS

DSP/BIOS

DSP/BIOS



Real-time Scheduling

Real-time Scheduling



Simple example of real-time

Simple example of real-time

problem

problem



HWI, SWI, and TSK

HWI, SWI, and TSK



Scheduler

Scheduler



DSP/BIOS II adds multi-tasking

DSP/BIOS II adds multi-tasking



Real-time analysis (RTA)

Real-time analysis (RTA)

DSP/BIOS Thread Types

DSP/BIOS Thread Types

P

ri

o

ri

ty

P

ri

o

ri

ty

HWI

HWI

Hardware Interrupts

Hardware Interrupts



Used to implement 'urgent' part

Used to implement 'urgent' part

of real-time event

of real-time event



Triggered by hardware interrupt

Triggered by hardware interrupt



HWI priorities set by hardware

HWI priorities set by hardware

SWI

SWI

Software Interrupts

Software Interrupts



Use SWI to perform HWI '

Use SWI to perform HWI '

follow-

follow-

up

up

' activity

' activity



SWI's are '

SWI's are '

posted

posted

' by HWI's or

' by HWI's or

SWI's

SWI's



Multiple SWIs at each of 14

Multiple SWIs at each of 14

priority levels

priority levels

IDL

IDL

Background

Background



Multiple IDL functions

Multiple IDL functions



Run round-robin

Run round-robin

DSP/BIOS Scheduler

DSP/BIOS Scheduler

h/w INT

h/w INT

HWI:

HWI:

urgent code

urgent code

post SWI

post SWI

SWI (or TSK)

SWI (or TSK)

ints disabled

ints disabled

rather than all this time

rather than all this time

read

read

serial port

serial port

run filter with new data

run filter with new data

HWI

HWI



Fast response to

Fast response to

interrupts

interrupts



Minimal context

Minimal context

switching

switching



High priority only

High priority only



Can post SWI or TSK

Can post SWI or TSK



Danger of missing an

Danger of missing an

interrupt while

interrupt while

executing ISR

executing ISR

SWI or TSK

SWI or TSK



Latency in response

Latency in response

time

time



Context switch

Context switch

performed

performed



Selectable priority

Selectable priority

levels

levels



Can post another SWI

Can post another SWI



Execution managed by

Execution managed by

scheduler

scheduler

Let’s look at a scheduling example...

Let’s look at a scheduling example...

HWI

HWI

SWI 2

SWI 2

SWI 1

SWI 1

main()

main()

IDL

IDL

HWI with SWI & IDL

HWI with SWI & IDL

interrupt

interrupt

return

return

return

return

return

return

post swi2

post swi2

Skip Slide Animation

Skip Slide Animation

return

return

return

return

interrupt

interrupt

post swi1

post swi1

SWI Properties

SWI Properties

Managing SWI Priority

Managing SWI Priority



Drag and Drop SWIs to

Drag and Drop SWIs to

change priority

change priority



Equal priority SWIs run

Equal priority SWIs run

round-robin

round-robin



Drag and Drop SWIs to

Drag and Drop SWIs to

change priority

change priority



Equal priority SWIs run

Equal priority SWIs run

round-robin

round-robin

DSP/BIOS Thread Types

DSP/BIOS Thread Types

P

ri

o

ri

ty

P

ri

o

ri

ty

HWI

HWI

Hardware Interrupts

Hardware Interrupts



Used to implement 'urgent' part

Used to implement 'urgent' part

of real-time event

of real-time event



Triggered by hardware interrupt

Triggered by hardware interrupt



HWI priorities set by hardware

HWI priorities set by hardware

SWI

SWI

Software Interrupts

Software Interrupts



Use SWI to perform HWI '

Use SWI to perform HWI '

follow-

follow-

up

up

' activity

' activity



SWI's are '

SWI's are '

posted

posted

' by HWI's or

' by HWI's or

SWI's

SWI's



Multiple SWIs at each of 14

Multiple SWIs at each of 14

priority levels

priority levels

TSK

TSK

Tasks

Tasks



Use TSK to run different programs

Use TSK to run different programs

concurrently under separate

concurrently under separate

contexts

contexts



TSK's are usually enabled to run

TSK's are usually enabled to run

by setting a flag, called a

by setting a flag, called a

'

'

semaphore

semaphore

'

'

IDL

IDL

Background

Background



Multiple IDL functions

Multiple IDL functions



Run round-robin

Run round-robin

SWI vs. TSK

SWI vs. TSK



Similar to hardware

Similar to hardware

interrupt, but

interrupt, but

triggered by

triggered by

SWI_post() function

SWI_post() function

call

call



All SWI's share system

All SWI's share system

software stack (along

software stack (along

with HWI's)

with HWI's)

SWI

SWI

SWI_post

SWI_post

start

start

end

end

“

“

run to

run to

completion”

completion”



Each TSK has its own

Each TSK has its own

stack, which allows

stack, which allows

them to pause

them to pause



Usually implemented as

Usually implemented as

loop

loop



Executed conditionally

Executed conditionally

based on a semaphore

based on a semaphore

(condition, flag)

(condition, flag)



SEM_post function call

SEM_post function call

set’s flag to trigger

set’s flag to trigger

execution

execution

TSK

TSK

start

start

end

end

Pause

Pause

SEM_post

SEM_post

(blocked

(blocked

state)

state)

SEM_pend

SEM_pend

TSK Preemption Example

TSK Preemption Example

HWI

HWI

SWI 2

SWI 2

SWI 1

SWI 1

IDL

IDL

main()

main()

TSK 2

TSK 2

TSK 1

TSK 1

interrupt

interrupt

pend

pend

sem2

sem2

return

return

interrupt

interrupt

interrupt

interrupt

pend

pend

sem2

sem2

pend

pend

sem1

sem1

return

return

return

return

post

post

swi1

swi1

return

return

return

return

post

post

sem2

sem2

return

return

post

swi2

How do you set priorities ...

How do you set priorities ...

DSP/BIOS

DSP/BIOS



Real-time Scheduling

Real-time Scheduling



Simple example problem

Simple example problem



HWI, SWI, and TSK

HWI, SWI, and TSK



Scheduler

Scheduler



Real-time analysis (RTA)

Real-time analysis (RTA)



printf() and LOG_printf

printf() and LOG_printf



Statistics

Statistics



Visual Instrumentation

Visual Instrumentation

Why is printf() used?

Why is printf() used?



printf is widely used for logical debug

printf is widely used for logical debug

(checking

(checking

your answer)

your answer)



What’s wrong with printf?

What’s wrong with printf?



30,000+ cycles to perform printf

30,000+ cycles to perform printf



Requires extensive prog & data memory

Requires extensive prog & data memory



DSP has to format the text string

DSP has to format the text string



Worse yet, it's Non-Deterministic

Worse yet, it's Non-Deterministic



The DSP must stop and wait for MS Windows

The DSP must stop and wait for MS Windows

to send string back to debugger (is Windows

to send string back to debugger (is Windows

real-time?)

real-time?)



Bottom Line

Bottom Line



Not Deterministic

Not Deterministic



Why waste DSP's MIPS and Mbytes

Why waste DSP's MIPS and Mbytes



Real-time code can fail due to printf

Real-time code can fail due to printf

printf (“I'm a wasteful function = %d\n”,i++);

printf (“I'm a wasteful function = %d\n”,i++);

printf (“I'm a wasteful function = %d\n”,i++);

printf (“I'm a wasteful function = %d\n”,i++);

DSP/BIOS: Real-Time

DSP/BIOS: Real-Time

Instrumentation

Instrumentation





DSP

DSP

Real-Time Capture

Real-Time Capture

Host (CCS)

Host (CCS)

Analysis & Display

Analysis & Display

Two main printf real-time issues are …

Two main printf real-time issues are …

1.

1.

Not Deterministic

Not Deterministic

2.

2.

Why waste DSP's MIPS and Mbytes

Why waste DSP's MIPS and Mbytes

If captured in real-time, how is debug data transferred?

If captured in real-time, how is debug data transferred?

LOG_printf

LOG_printf



LOG_printf is answer to real-time issues

LOG_printf is answer to real-time issues



Data is captured to a buffer on the DSP

Data is captured to a buffer on the DSP



Data is transferred to buffer in background (IDL)

Data is transferred to buffer in background (IDL)

How is LOG_printf used in code?

How is LOG_printf used in code?

External

External

reference to

reference to

LOG object

LOG object

defined using

defined using

the

the

configuration

configuration

tool

tool

Call LOG_printf

Call LOG_printf

#include <std.h>

#include <std.h>

#include <log.h>

#include <log.h>

extern far LOG_Obj myLog;

extern far LOG_Obj myLog;

#include <std.h>

#include <std.h>

#include <log.h>

#include <log.h>

extern far LOG_Obj myLog;

extern far LOG_Obj myLog;

LOG_printf(&myLog, “New load = %d000 instructions”, loadVal);

LOG_printf(&myLog, “New load = %d000 instructions”, loadVal);

LOG_printf(&myLog, “New load = %d000 instructions”, loadVal);

LOG_printf(&myLog, “New load = %d000 instructions”, loadVal);

Call the

Call the

function

function

Run-time

Run-time

display of

display of

user-defined

user-defined

LOG

LOG

DSP/BIOS

DSP/BIOS



Real-time Scheduling

Real-time Scheduling



Real-time analysis (RTA)

Real-time analysis (RTA)



printf() and LOG_printf

printf() and LOG_printf



Statistics

Statistics



Visual Instrumentation

Visual Instrumentation

Built-in Statistics Tools

Built-in Statistics Tools



Execution Graph

Execution Graph

is useful for visualization of thread scheduling

is useful for visualization of thread scheduling



Statistics

Statistics

give you finer detail to profile thread execution

give you finer detail to profile thread execution

Is the thread meeting its real-time deadline?

Is the thread meeting its real-time deadline?

The number of

The number of

times the SWI has

times the SWI has

run

run

Max and Average

Max and Average

number of instructions

number of instructions

from SWI_post to

from SWI_post to

completion (can also

completion (can also

display as milliseconds

display as milliseconds

or microseconds)

or microseconds)

User-Defined STS example

User-Defined STS example

Reference the

Reference the

STS object

STS object

#include <std.h>

#include <std.h>

#include <sts.h>

#include <sts.h>

extern far STS_Obj myStat;

extern far STS_Obj myStat;

#include <std.h>

#include <std.h>

#include <sts.h>

#include <sts.h>

extern far STS_Obj myStat;

extern far STS_Obj myStat;

STS_add(&myStat, your_variable);

STS_add(&myStat, your_variable);

STS_add(&myStat, your_variable);

STS_add(&myStat, your_variable);

Monitor a

Monitor a

variable

variable

Insert STS object

Insert STS object

in configuration

in configuration

tool,

tool,

rename to

rename to

“myStat”

“myStat”



What does deterministic, low-overhead RTA mean to

What does deterministic, low-overhead RTA mean to

you?

you?



LOG, STS, and TRC module operations are very fast

LOG, STS, and TRC module operations are very fast

and execute in a consistent time, as shown in the

and execute in a consistent time, as shown in the

following list:

following list:



LOG_printf:

LOG_printf:

approximately 32 cycles

approximately 32 cycles



STS_add:

STS_add:

approximately 18 cycles

approximately 18 cycles



STS_delta:

STS_delta:

approximately 21 cycles

approximately 21 cycles



TRC_enable & _disable:

TRC_enable & _disable:

approximately 6 cycles

approximately 6 cycles

* (exact timing depends upon processor type)

* (exact timing depends upon processor type)



STS and LOG - low overhead

STS and LOG - low overhead



Each STS object uses only four words of data memory.

Each STS object uses only four words of data memory.

Therefore, only four words need to be uploaded from a

Therefore, only four words need to be uploaded from a

statistics object

statistics object



LOG functions only need to capture 4 words per

LOG functions only need to capture 4 words per

invocation

invocation

Low Instrumentation Overhead

Low Instrumentation Overhead

You can leave RTA in your code and get:

You can leave RTA in your code and get:

1.

1.

What you test is "exactly" what you ship

What you test is "exactly" what you ship

2.

2.

When (if) field problems arise, debug

When (if) field problems arise, debug

instrumentation is already in place!

instrumentation is already in place!

DSP/BIOS

DSP/BIOS



Real-time Scheduling

Real-time Scheduling



Real-time analysis (RTA)

Real-time analysis (RTA)



printf() and LOG_printf

printf() and LOG_printf



Statistics

Statistics



Visual Instrumentation

Visual Instrumentation



Execution Graph

Execution Graph



CPU Load Graph

CPU Load Graph

Viewing System Events

Viewing System Events



System Log is an

System Log is an

Event “Logic Analyzer”

Event “Logic Analyzer”



The HWI and SWI routines are on the left axis,

The HWI and SWI routines are on the left axis,

and the bottom axis are event ticks

and the bottom axis are event ticks

…

…

This makes it easy to follow the action

This makes it easy to follow the action

CPU Load Meter

CPU Load Meter



CPU load meter acts like a MIPs calculator

CPU load meter acts like a MIPs calculator



Basically, it keeps track of the IDLE time

Basically, it keeps track of the IDLE time

DSP/BIOS - API Modules

DSP/BIOS - API Modules

Instrumentation/Real-Time Analysis

Instrumentation/Real-Time Analysis

LOG

LOG

Message Log manger

Message Log manger

STS

STS

Statistics accumulator manager

Statistics accumulator manager

TRC

TRC

Trace manager

Trace manager

RTDX

RTDX

Real-Time Data Exchange manager

Real-Time Data Exchange manager

Thread Types

Thread Types

HWI

HWI

Hardware interrupt manager

Hardware interrupt manager

SWI

SWI

Software interrupt manager

Software interrupt manager

TSK

TSK

Multitasking manager

Multitasking manager

IDL

IDL

Idle function & processing loop manager

Idle function & processing loop manager

Clock and Periodic Functions

Clock and Periodic Functions

CLK

CLK

System clock manager

System clock manager

PRD

PRD

Periodic function manger

Periodic function manger

Comm/Synch between threads

Comm/Synch between threads

SEM

SEM

Semaphores manager

Semaphores manager

MBX

MBX

Mailboxes manager

Mailboxes manager

LCK

LCK

Resource lock manager

Resource lock manager

Input/Output

Input/Output

PIP

PIP

Data pipe manager

Data pipe manager

HST

HST

Host input/output manager

Host input/output manager

SIO

SIO

Stream I/O manager

Stream I/O manager

DEV

DEV

Device driver interface

Device driver interface

Memory and Low-level Primitives

Memory and Low-level Primitives

MEM

MEM

Memory manager

Memory manager

SYS

SYS

System services manager

System services manager

QUE

QUE

Queue manager

Queue manager

ATM

ATM

Atomic functions

Atomic functions

GBL

GBL

Global setting manager

Global setting manager

#include’s

#include’s

extern far SWI_Obj

extern far SWI_Obj

global variables

global variables

main( ) {

main( ) {

CSL_Init();

CSL_Init();

BSL_init();

BSL_init();

codec_init();

codec_init();

init_HWI();

init_HWI();

AD535_write(hAD53

AD535_write(hAD53

5, 0);

5, 0);

return;

return;

}

}

// HWI routines

// HWI routines

init_HWI();

init_HWI();

XINT0_HWI;

XINT0_HWI;

// Codec Routines

// Codec Routines

codec_init();

codec_init();

codec_out();

codec_out();

lab6.c

lab6.c

HWI

HWI

9: _XINT0_HWI

9: _XINT0_HWI

SWI

SWI

codec_swi:

codec_swi:

_codec_out

_codec_out

lab6.cdb

lab6.cdb

sineGen( )

sineGen( )

sine_float.c

sine_float.c

Lab 6

Lab 6

You get to

You get to

complete these

complete these