Skip to main content

SIMD - Single Instruction Multiple Data


Hi! Today’s lecture, we learned SIMD - Single Instruction Multiple Data. This is a great tool to process data in a bulk fashion. So, instead of doing one by one, based on the variable size, we can do 16, 8, 4 or 2 at the time. This technique is called auto-vectorization resources, and it falls into the category of machine instruction optimization that I mentioned in my last post.

If the machine is SIMD enabled, the compiler can use it when translating a sum loop, for example. If we are summing 8 bits numbers, using SIMD, it will be 16 times faster. However, the compiler can figure that it is not safe to use SIMD due to overlapping or non-aligned data. In fact, the compiler will not apply SIMD in most cases, so we need to get our hands dirty and inject some assembly. I’ll show you how to do it in a second.

Here are the lanes of the 128-bit AArch64 Advanced SIMD:
16 x 8 bits
8 x 16 bits
4 x 32 bits
2 x 64 bits
1 x 128 bits

Reading the ARM manual, we can find a lot of SIMD functions. Bringing back the volume example, we can process 8 values each time, and not worry about the overflow. The magic instruction is SQDMULH – Signed Integer Saturating Doubling Multiply returning High Half. With that name, it must make coffee too! Well, no. It multiplies the first parameter with the second. It puts the result into the third, discarding the fraction portion and not overflowing – on overflow, it will keep the minimum or maximum value. It is precisely what we need to deal with the volume in one instruction.

Now let’s mix some C and Assembly, shall we?

The syntax is:
__asm__ ("assembly code" : outputs : inputs : clobbers);

Warning: this will break the portability. It is a good idea to have compiler flags to “pick” the right portion of the code based on the architecture being compiled. Here we are not doing that.

This is the code provided by our instructor. Do you see the loop in C and the ASM instruction inside? The line 52 is doing 8 values per iteration using the magic single instruction SQDMULH. It is fast! The code, as it is, will only work on Arch64, though.


If you don’t like assembly like me, intrinsics will help. The GCC compiler has some sort of functions representations of the assembly instructions. I think that it helps, but it also has its limitations. Here is the same example, but using intrinsics. Take a look at line 42.


This is it for today. I’m working on profiling my awk build. Stay tuned!

Comments

Popular posts from this blog

Two-digit Numeric Display - Final

Photo by  Nick Hillier  on  Unsplash In this post, I’ll continue the two-digit numeric display. If you miss it, click here and check it out . To finish this project, we just need to show the numbers in the matrix-pixel (the black-box in the 6502 emulator ). To kickstart, our instructor gave us one example of how to display graphs, which was a lot helpful. The first thing that I’ve noticed was the bitmap table at the bottom. So, I mimic it and made ten tables like that to represent each number (zero to nine). So far, so good! Then I grabbed the logic to display one digit, and then my nightmares just started. How to place two graphs (one for each digit)? How to switch from one number to another? How to reuse code? Where is my coffee?! To emulate some if-elseif-else statements, I used jmp (jump). They are all over the place! However, the 6502 limits the jump range from -127 to 128. That means moving the code-blocks to satisfy all jumps limit. For e...

Project Stage 1

Photo by  SpaceX  on  Unsplash Hello! This is my SPO 600 blog, and this post will be long – sorry. The goal is to pick one project that is CPU intensive, written in C or C++, and experiment different compiler options and present the results. That’s why it will be long – lots of data to show. I choose the AWK project ( https://github.com/onetrueawk/awk ). It is a handy tool to process files. Parse, sort, and filter are some trivial operations that are CPU intensive. To make it harder, I created a huge XML file to parse it and count the tags. I've described the machines in my last post, if you miss it, here it is . I also created a script to run and collect the data. I planned to run each candidate 10 times, but a few attempts didn’t receive any data. So, I decided to nest the loop in a way that even if someone kills my process, the data could be used. Guess what? It happened! To produce the candidates, I just changed the CFLAGS inside the makefile and ran the...

Assembly?

Photo by  Jonas Svidras  on  Unsplash Last week on my SPO course, I had my first experience writing Assembly code. I won’t lie; it was struggling. For me, Assembly is like the Latin of the codding languages and “carpe diem” wasn’t my first lesson. Hexadecimal, binary and a list of instructions is a must know to guarantee survival. Our instructor introduced us to the 6502 processor: it is an old school chip that was used in many home solutions such as PCs and video games. Internally, it has three general-purpose registers, three special-purpose registers, memory and input and output ports. Fortunately, there are emulators on the internet that helps us to focus on the development, hiding the electronic part from us. http://6502.cdot.systems/ Using the emulator, our first task was to copy, paste and execute a piece of code to change the colour of every pixel in the display matrix. That was easy! The result was a yellow screen. Then we were asked t...