Llorenç Martí García

3D Architect at IMVU Inc. I like to make games and benchmarking tools (like BlueSteel) on my free time :D. I'm from Barcelona. Opinions my own.

[Llorenç Martí García @LlorensMarti]

@idoccor How long was a normal build+executions of those tests ?

[Llorenç Martí García @LlorensMarti]

@gallifantes Enveja sana! 🤗🥘

[Llorenç Martí García @LlorensMarti]

@Wunkolo Totally agree!

[Llorenç Martí García @LlorensMarti]

@DavidGR1714 @Rosell_Advocat I no només això sinó que la FED té eines brutals per a protegir el USD

[Llorenç Martí García @LlorensMarti]

@DavidGR1714 @Rosell_Advocat Jo coincideixo en que l’opció intelligent és diversificar. Que no dubto que el dollar pugui perdre influència, però, si l’USD baixés molt, altres monedes com l’EUR també ho farien crec

[Llorenç Martí García @LlorensMarti]

@Jonathan_Blow What I found, specifically for opaque objects is that it is better to sort by minimal state change than for distance.
Transparent objects are another beast and, specially on UGC, the algorithm can be tricky very soon.

[Llorenç Martí García @LlorensMarti]

@SebAaltonen It’s funny because when I changed from an Intel Macbook to the M1 the C++ compilation times halved :D

[Llorenç Martí García @LlorensMarti]

@SebAaltonen Something that I experienced is that, as you reach maximum scale on a CPU, the performance you got with few cores degrades when you use all of them. I think it needs to do with how caches are designed/used, but it will be interesting. (In my case it was Intel CPUs)

[Llorenç Martí García @LlorensMarti]

@ocornut I am currently integrating Dawn in a side project repo, and it needs to be integrated as a manual managed repo (no git submodules), otherwise the git submodule update requires a google dev account to update some Dawn submodules. Its very tiring altogether :(

[Llorenç Martí García @LlorensMarti]

@SebAaltonen Also, the cache line size is important if the elements are going to be accessed in a multithreaded environment.

I learned some time ago about “false sharing” where 2 threads access 2 different elements both living in the same cache line.

^ Performance killer

[Llorenç Martí García @LlorensMarti]

@JLarky I much prefer trunk base development with release branches.
Almost no feature-branches, no long lasting dev branches, etc. Linear history, and cherry-pick for hotfixes.

[Llorenç Martí García @LlorensMarti]

@Hasen_Judi Not only build and run locally but run fast!

[Llorenç Martí García @LlorensMarti]

@SebAaltonen I still have an M1 Max, and super happy with it :)

[Llorenç Martí García @LlorensMarti]

@simplex_fx Yep, 250 users is too low, but, I must ask, “serverS” ? How many servers ?

[Llorenç Martí García @LlorensMarti]

@SebAaltonen “- Bloated unit test suites taking several hours to execute. Slows down devs and causes merge conflicts as pushes are delayed.”

I am interested, how much time does it take to execute the suit of tests?, here we have about 5600 tests for our 3D engine and it takes ~0.8s total.

[Llorenç Martí García @LlorensMarti]

@SebAaltonen If embedding SQL was an error, I wonder what we think about HTTP, where we need to parse headers to know the length of the message ;)

[Llorenç Martí García @LlorensMarti]

@SheriefFYI Actually in IMVU having alpha as separate texture is very common.

It allows derivation of products that are opaque or with certain alpha to have another design of alpha and be a different product.

When the assets are served for runtime, then, alpha comes inside RGBA.

[Llorenç Martí García @LlorensMarti]

@SheriefFYI Sounds complicated!! :D

[Llorenç Martí García @LlorensMarti]

@iquilezles Sublime Text for everything

[Llorenç Martí García @LlorensMarti]

@Hasen_Judi @FlohOfWoe @Dae8alus Actually, green tests, formatted code, atomic commits, are characteristics that makes medium to big teams more efficient.

You can read/understand code faster, bisect bugs faster, etc.