Thursday, April 18, 2019

for loop multiple conditions

what is the right?
1. for (i = 0, j = 0; i < loop, j < loop2; i++, j++)
2. for (i = 0, j = 0; (i < loop) && (j < loop2); i++, j++)

The answers are
Are multiple conditions allowed in a for loop?
GeeksforGeeks: Output of C Program | Set 22, question 2

for example
i = 1, 2; is same as i = 2;

Wednesday, April 17, 2019

TI DSP memory & memory map

L2SRAM: very very fast
MSMCSRAM (Multi-core Shared Memory Controller, SRAM): very fast
DDR3: fast

TI Linker Command File Primer

Thursday, March 14, 2019

TI DSP code optimization

Texas Instruments TMS320C6x DSP code optimization

1. Hand-Tuning Loops and Control Code on the TMS320C6000
1.1 loop optimization
 -nrestrict
 - #pragma MUST_ITERATE(lower_bound, upper_bound, factor)
 - _nasserts()
1.2 if statement optimization

2. Introduction to TMS320C6000 DSP Optimization
2.1 Cannot make "Pipelined Loop"
 - exceed 14 executed packets (1 packet is 8 instructions)
 - nested loops
 - conditional branches inside loops
 - function calls inside loops
2.2 "Pipelined Loop" consists of
 - Prolog: above Kernel
 - Kernel: pipeline is fully utilized
 - Epilog: below Kernel
2.3 ii
 - ii = iteration interval
 - software pipeline loop can be approximated with ii * number_of_iterations.
 - ii is bounded below by two factors: the loop carried dependency bound and the partitined resource bound.
 - the loop carried dependency bound: the distance of the largest loop path


* Reference
Hand-Tuning Loops and Control Code on the TMS320C6000
Introduction to TMS320C6000 DSP Optimization
TMS320C6000 DSP Optimization Workshop - Texas Instruments Wiki
TMS320C6000 Programmer's Guide (Rev. K) - Texas Instruments
http://processors.wiki.ti.com/index.php/Software_libraries
http://processors.wiki.ti.com/index.php/Profiler
DSP/BIOS timers and benchmarking Tips SPRA829: Profile



Tuesday, February 19, 2019

Performance / Optimization

1. Practical Performance
 - gprof
 - PAPI
 - Callgrind
 - Compiler Flags

1.1
 - IA-32 (32 bit, intel architecture 32 bit, i386): the 32 bit version of the x86 instruction set
 - AMD64 (64 bit, x64, x86_64, AMD64); the 64 bit version of the x86 instruction set
1.2 SIMD
 SISD (single instruction, single data) vs. SIMD (single instruction, multiple data)
 - SIMD 병렬 프로그래밍


Sunday, February 17, 2019

Duplicated code detection

1. PMD (https://pmd.github.io/)
CPD (copy paste detector)
If you use the Java 11 or more version, please use the PMD 6.11.0 or more

windows
> bin\cpdgui.bat

linux
$ ./run.sh cpd --minimum-tokens 100 -- files /usr/local/java/src/java
https://stackoverflow.com/questions/53260053/current-eclipse-plugin-for-duplicated-code



2. ConQAT
ConQAT end of life
You can download old versions from here.