[[Category:SPO600 Labs- Retired]]{{Admon/lab|Purpose of this Lab|In this lab, you will investigate the impact of different algorithms which produce the same effect.}}{{Admon/important|x86_64 and AArch64 Systems|This lab must be performed on both x86_64 and AArch64 systems. You will test may use the [[SPO600 Servers]] or you may use other system(s) -- it might make sense to use your own x86_64 system and select one of three algorithms [[SPO600_Servers#AArch64:_israel.cdot.systems|israel.cdot.systems]] for adjusting the volume of PCM audio samples based on benchmarkingAArch64.}}

== Lab 4 5 ==

* Digital sound is typically represented, uncompressed, as signed 16-bit integer signal samples. There is are two streams of samples, one each for the left and right stereo channels, at typical sample rates of 44.1 or 48 thousand samples (kHz)per second per channel, for a total of 88.2 or 96 thousand samples per second (kHz). Since there are 16 bits (2 bytes) per sample, the data rate is 88.2 * 1000 * 2 = 176,400 bytes/second (~172 KiB/sec) or 96 * 1000 * 2 = 192,000 bytes/second (~187.5 KiB/sec).* To change the volume of sound, each sample can be scaled (multiplied) by a volume factor, in the range of 0.00 (silence) to 1.00 (full volume).

=== Basic Sound Scale Program Multiple Approaches ===

Get Six programs are provided, each with a different approach to the files problem, named <code>vol0.c</code> through <code>vol5.c</code>. A header file, <code>vol.h</code>, defines how much data (in number of sample) will be processed by each program, as well as the volume level to be used for scaling (50%). These are the six programs: # vol0.c is the basic or naive algorithm. This approach multiplies each sound sample by the volume scaling factor, casting from signed 16-bit integer to floating point and back again. Casting between integer and floating point can be [[Expensive|expensive]] operations.# vol1.c does the math using fixed-point calculations. This avoids the overhead of casting between integer and floating point and back again.# vol2.c pre-calculates all 65536 different results, and then looks up the answer for each input value.# vol3.c is a dummy program - it doesn't scale the volume at all. It can be used to determine some of the overhead of the rest of the processing (besides scaling the volume) done by the other programs.# vol4.c uses Single Instruction, Multiple Data (SIMD) instructions accessed through inline assembley (assembly language code inserted into a C program). This program is specific to the AArch64 architecture and will not build for x86_64.# vol5.c uses SIMD instructions accessed through Complier Intrinsics. This program is also specific to AArch64. '''Note that vol4.c and vol5.c will build only on AArch64 systems because they use architecture-specific SIMD instructions.''' === Don't Compare Across Machines === In this lab , ''do not'' compare the relative performance across different machines, because various systems have different microarchitectures, memory configurations, peripheral implementations, and clock speeds, from mobile-class to server-class (e.g. Intel Atom vs. Xeon; AMD APU vs. Threadripper; ARM Cortex-A55 vs. Neoverse-V2). However, ''do'' compare the relative performance of the various algorithms on the ''same'' machine. === Important! === The hardest part of this lab, and the most critical component, is being able to separate the performance of the volume scaling code from the rest of the code (which only exists to set up the test of the scaling code). The volume scaling code runs ''very'' quickly, and is dwarfed by the rest of the code. You '''must''':* Control variables in your test environment.** What else is the machine doing while you are testing?** Who else is logged in to the machine?** What background operations are being performed?** How does your login on the machine affect performance (e.g., network activity)?* <span style="background: #ffff00">Isolate the performance of the volume scaling code.</span> This is one of the most important parts of this lab! There are two practical approaches:** Subtract the performance of the dummy version of the program from each of the other versions, or** Add code to the program to measure and report just the performance of the volume-scaling code* Repeat the tests multiple times to ensure that the results you are getting are consistent, valid, and accurately reflect the performance of the volume scaling code.** Make sure you are performing enough calculation to give a useful result -- adjust the SAMPLES value in <code>vol.h</code> to a sufficiently high value** Discard outliers (unusually high or low results)** Average the results.** Take some measure of the amount of variation of your results (e.g., tolerance limits or standard deviation). === Resources ===* ARM Aarch64 documentation** [http://developer.arm.com/ ARM Developer Information Centre]*** [SPO600 Servershttps://developer.arm.com/docs/den0024/latest ARM Cortex-A Series Programmer’s Guide for ARMv8-A]*** The ''short'' guide to the ARMv8 instruction set: [https://www.element14.com/community/servlet/JiveServlet/previewBody/41836-102-1-229511/ARM.Reference_Manual.pdf ARMv8 Instruction Set Overview] ("ARM ISA Overview")*** The ''long'' guide to the ARMv8 instruction set: [https://developer.arm.com/docs/ddi0487/latest/arm-architecture-reference-manual-armv8-for-armv8- but you can perform a-architecture-profile ARM Architecture Reference Manual ARMv8, for ARMv8-A architecture profile] ("ARM ARM")** [https://developer.arm.com/docs/ihi0055/latest/procedure-call-standard-for-the-arm-64-bit-architecture Procedure Call Standard for the ARM 64-bit Architecture (AArch64)] * x86_64 Documentation** AMD: https://developer.amd.com/resources/developer-guides-manuals/ (see the AMD64 Architecture section, particularly the ''AMD64 Architecture Programmer’s Manual Volume 3: General Purpose and System Instructions'')** Intel: http://www.intel.com/content/www/us/en/processors/architectures-software-developer-manuals.html** Web sites*** http://ref.x86asm.net/*** http://sandpile.org/ * Assembler and Inline Assembler** [[Assembler Basics]]** [[Inline Assembly Language]]** GAS Manual - Using as, The GNU Assembler: https://sourceware.org/binutils/docs/as/*** Specifically, the lab wherever you want[http://gcc.gnu.org/onlinedocs/gcc-4.8.2/gcc/Extended-Asm.html Assembler Instructions with C Expression Operands] section

=== Alternate Approaches =Benchmarking ==

The sample program uses Get the most basic, obvious algorithm files for the problem. Let's call this "Algorithm 0", or lab from one of the "Naive Algorithm". Note that it uses casting between integer and floating-point formats as well as multiplication -- both of which can be [[Expensive|expensiveSPO600 Servers]] operations-- but you can perform the lab wherever you want (feel free to use your laptop or home system). Test on both an x86_64 and an AArch64 system.

Try these alternate algorithms The files for scaling this lab are in the sound archive <code>/public/spo600-volume-examples.tgz</code> on each of the SPO600 servers. The archive contains:* <code>vol.h</code> controls the number of samples by modifying copies of to be processed and the volume level to be used* <code>vol1vol0.c</code>through <code>vol5. Edit c</code> implement the various algorithms* <code>vol_createsample.c</code> contains a function to create dummy samples* The <code>Makefile</code> can be used to build your modified the programs as well as the original. Test each approach to see the performance impact:

Perform these steps '''on both x86_64 and AArch64 systems''':# PreUnpack the archive <code>/public/spo600-volume-calculate examples.tgz</code># Study each of the source code files and make sure that you understand what the code is doing.# '''Make a lookup table (array) prediction''' of all possible sample values multiplied by the volume factor, relative performance of each scaling algorithm.# Build and look up test each sample in that table to get of the scaled valuesprograms. (You'll have #* Do all of the algorithms produce the same output?#** How can you verify this?#** If there is a difference, is it significant enough to handle matter?#* Change the fact number of samples so that each program takes a reasonable amount of time to execute (suggested minimum is 20 seconds).# Test the input values range from -32768 to +32767performance of each program (vol0 through vol3 on x86_64, while C arrays accept only a positive indexand vol0 through vol5 on AArch64).# Convert * Find a way to measure performance ''without'' the volume factor 0.75 time taken to a fixperform the test setup pre-point integer by multiplying by a binary number representing a fixedprocessing (generating the samples) and post-point value "1". For example, you could use 0b100000000 processing (= 256 in decimalsumming the results) so that you can measure ''only'' the time taken to represent 1scale the samples.00, and therefore use 0.75 '''This is the hard part!'''#* How much time is spent scaling the sound samples?#* 256 = 192 for your volume factor. Multiply Do multiple runs take the same time? How much variation do you observe? What is the likely cause of this fixed-point integer volume factor variation?#* Is there any difference in the results produced by each sample, then shift the result to various algorithms?#* Does the difference between the right algorithms vary depending on the required number architecture and implementation on which you test?#* What is the relative memory usage of bits after each program?# See if you can measurably increase performance by changing the multiplication compiler option (via the Makefile)# Was your prediction about performance accurate?# Find all of the questions, marked with <code>'''Q:'''</code>8 if you're using 256 as , in the multiplier)program comments, and answer those questions.

Blog about your experiments with a detailed analysis of your results, including memory usage, time performance, accuracy, and trade-offs. Include answers to all of the questions marked with Q: in the source code.

Important'''Make sure you convincingly <u>prove</u> your results to your reader! ''' Re-- read the [[#Important.21|section marked ''Important'' above]] and make sure you address the issues explained there. Also be sure to explain what you're doing so that a reader coming across your blog post understands the context (in other words, don't just jump into a discussion of optimization results -- give your post some context).

'''Optional - Recommended:''' Compare results across several '''implementations ''' of AArch64 and x86_64 systems. Note that on different CPU implementations, the relative performance of different algorithms will vary; for example, table lookup may outperform other algorithms on a system with a fast memory system (cache), but not on a system with a slower memory system.* For AArch64, you could compare the performance on AArchie against another a Raspberry Pi 4 (in 64-bit ARM system such as a Raspberry Pi 3 mode) or an ARM Chromebook.* For x86_64, you could compare the performance of different processors, such as xerxesportugal.cdot.systems, your own laptop or desktop, and Seneca systems such as Matrix, Zenit, or lab desktops.

* Most solutions for a problem of this type involve generating a large amount of data in an array, processing that array using the function being evaluated, and then storing that data back into an array. The test setup can take more time than the actual test! Make sure that you measure the time taken for the code in question (the part that scales the test function only sound samples) ONLY -- you need to be able to remove the rest of the processing time from your evaluation.* You may need to run a very massive large amount of sample data through the function to be able to detect its performance. Feel free to edit the sample count in <code>vol.h</code> as necessary.

==== Analyzing Results =Tips ===* What is the impact {{Admon/tip|Analysis|Do a thorough analysis of various optimization levels on the software performance? results. Be certain (For example, compiling with -O0 / -O1 / -O2 / -O3and prove!)* Does that your performance measurement ''does not'' include the distribution generation or summarization of the test data matter? (e.gDo multiple runs and discard the outliers.Decide whether to use mean, is there any difference if there are no absolute large numbersminimum, or no negative numbers?)* If samples are fed at CD rate (44100 samples per second x 2 channels x 2 bytes per sample), can each of maximum time values from the algorithms keep up?* What is the memory footprint of each approach?* What is the performance of each approach?* What is the energy consumption of each approach? (What information do you need to calculate this?)* Various machines within an architecture have very different performance profiles, energy consumptionmultiple runs, and hardware costs -- so it's not reasonable to compare explain why you made that decision. Control your variables well. Show relative performance between machinesas percentage change, but it is reasonable to compare the relative performance of the algorithms in each contexte.g. Does the ratio of performance of the various approaches remain constant across the machines? Why or why not?* What other optimizations can be applied to , "this problem?approach was NN% faster than that approach".}}

Another The version of the <code>time</code> command, located in <code>/bin/time</code>, gives slightly different information, than the version built in to bash -- including maximum resident memory usage: <code>/bin/time ''./programName''</code>}}

{{Admon/tip|Stack Limitstdint.h|Fixed-size, non-static arrays will be placed in the stack space. The size of the stack space is controlled by per-process limits, inherited from the shell, and adjustable with the <code>ulimitstdint.h</code> command. Allocating an array larger than the stack header provides definitions for many specialized integer size limit will cause a segmentation fault, usually on the first write. To see the current stack limit, use <code>ulimit -s</code> (displayed value is in KB; default is usually 8192 KB or 8 MB)types. To set the current stack limit, place a new size in KB or the keyword Use <code>unlimitedint16_t</code>after the <code>for 16-s</code> argument.<br /><br />Alternate (bit signed integers and preferred) approach, as used in the provided sample code: allocate the array space with <code>malloc()</code> or <code>calloc()uint16_t</code>for 16-bit unsigned integers.}}

{{Admon/tip|stdint.hScripting|The <code>stdint.h</code> header provides definitions for many specialized integer size types. Use <code>int16_t</code> for 16-bit signed integers.bash scripting capabilities to reduce tedious manual steps!}}