I found that knowing simd programming is plus skill if you want to work in AAA companies like rockstars and insomniac games so I want to learn this and apply this inside my 3d renderer rather than glm so any resources can you recommend use simd for game engine math libs, don't shader git hub repos I want to learn it not to get some libraries implementing it
#Learn SIMD for math engine
vast breach
young zephyr
SIMD for individual math operations is not usually a great use for it
The real power of using SIMD is for vectorizing a hot loop that is bottlenecked on arithmetic
vast breach
SIMD for individual math operations is not usually a great use for it
I want it for matrix multiplication it gives me x12 time faster than glm
young zephyr
glm already has SIMD implementations
timber summit
read the simd intrinsic guide, look at simd implementations, use simd
vast breach
glm already has SIMD implementations
I tried it and My implementation still faster by 100 ms i guess
young zephyr
I'm not sure what you're asking for then
timber summit
You can have a look at the implementations inside rtm https://github.com/nfrechette/rtm which is more focused on speed
vast breach
I'm looking for more resources and applications for it in any book
timber summit
the reality is though, getting it working on one device, with one program layout, is a lot simpler than maintaining a cross-arch fast simd implementation
young zephyr
Yeah and like I said the real power is in vectorizing a specific problem, not just trying to turn a single 4-vector into a register
timber summit
I tried it and My implementation still faster by 100 ms i guess
what does "by 100ms" mean without context anyway? 
vast breach
Ah ok so using vectorized Algorithms make it faster in processing, so I want to learn how to be able to do this
young zephyr
Run N loop iterations at once where N is the number of lanes you have
It's basically the same as how a GPU CU executes a wavefront
vast breach
if it like you say so what these companies want from you when the want people understanding simd
young zephyr
They will probably have wanted you to do some rigorous performance analysis to make empirically motivated performance optimizations using SIMD
young zephyr
You'll need to find or create a problem that is bottlenecked by arithmetic
thorny idol
Think about cutting vegetables. https://deplinenoise.files.wordpress.com/2015/03/gdc2015_afredriksson_simd.pdf
white solar
grab either Vector class library (from agner) or xsimd. I think directly going with the low level intrinsics is annoying. Their syntax/usability is horrid. but with those libs you have a vec4 and vec8 type and can do arithmetic with it
some typical programs you can practice with are:
frustum culling
cpu rasterization (harder)
transforming a mesh vertices by a matrix on cpu (easy)