Dactyloidae Browser (Basilisk fork)

[athenian200]

The one thing that jumps out at me that I am not sure I'd want to upstream (at least on Windows) is the override of the system memcpy. I'm pretty sure the memcpy included in MSVC's CRT is probably better than a handwritten version using SSE2...

I checked, and MSVC actually implements memcpy as raw ASM that can take advantage of SSE and AVX where available, meaning it's probably faster than any C or C++ implementation could be.

Granted, I am not an expert on code review here, but that's my naive intuition as a former CompSci student who didn't graduate due to being bad at Calculus, but otherwise got a lot of the other lessons.

[ownedbywuigi]

The one thing that jumps out at me that I am not sure I'd want to upstream (at least on Windows) is the override of the system memcpy. I'm pretty sure the memcpy included in MSVC's CRT is probably better than a handwritten version using SSE2...

I checked, and MSVC actually implements memcpy as raw ASM that can take advantage of SSE and AVX where available, meaning it's probably faster than any C or C++ implementation could be.

Granted, I am not an expert on code review here, but that's my naive intuition as a former CompSci student who didn't graduate due to being bad at Calculus, but otherwise got a lot of the other lessons.

I’ll do a test build with the MSVC memcpy tonight and see if performance increases

[ownedbywuigi]

UPDATE: Turns out like 2 of my modifications HURT performance, so I'm reverting those. Working on trying to fix occasional browser "hiccups" and high RAM usage (800MB RAM on YT and this forum)

[jarsealer]

Name really doesn't go off the tongue. xD

Couldn't think of a better name.

Basiliscus (https://www.merriam-webster.com/dictionary/Basiliscus), Basilides (doesn't make sense meaning wise tho, https://en.wikipedia.org/wiki/Basilides)

[Gemmaugr]

Couldn't think of a better name.

At least it's not Quetzalcoatlus.. How about Kirin?

[ownedbywuigi]

Couldn't think of a better name.

At least it's not Quetzalcoatlus.. How about Kirin?

I already bought the domain so I guess it’s too late lol

[ownedbywuigi]

13.0 prerelease 2 out now
https://repo.dactyloidae.xyz/Dactyloida ... /13.0-pre2

I want to branch out to a Linux build by the time 13.0 becomes stable, but mach (at least on my fork) seems to hate me on Linux

[ownedbywuigi]

UPDATE: Turns out like 2 of my modifications HURT performance, so I'm reverting those. Working on trying to fix occasional browser "hiccups" and high RAM usage (800MB RAM on YT and this forum)

Can confirm pre2 now FULLY FIXES the "browser hiccups" except for loading in a site for the first time.

This is fucking big.

STABLE 13.0 TOMORROW.

[UCyborg]

How do you wizards figure these things out?

[ownedbywuigi]

fuck it we ball
13.0 "Warp" out now, now the fastest XUL browser (until it gets upstreamed)
https://repo.dactyloidae.xyz/Dactyloida ... s/tag/13.0

[UCyborg]

Wait, is there supposed to be much difference from before? Though my combination of extensions might be problematic.

The problem I outlined here is still present.

[ownedbywuigi]

Wait, is there supposed to be much difference from before? Though my combination of extensions might be problematic.

The problem I outlined here is still present.

Not much, it's just pre2 with it's internal version changed to 52.9 again for better extension support

Will look into the issue though, will hopefully be fixed by 13.1

[ownedbywuigi]

Update on 13.1, I got (mostly) stable e10s, making that only one page freezes instead of the whole browser. Only regression is the backspace keeps on making the page go back for some reason, which I will try to fix.

EDIT: REGRESSION FIXED. E10S IS FULLY STABLE!!

[q160765803]

Update on 13.1, I got (mostly) stable e10s, making that only one page freezes instead of the whole browser. Only regression is the backspace keeps on making the page go back for some reason, which I will try to fix.

EDIT: REGRESSION FIXED. E10S IS FULLY STABLE!!

Wow. but I wonder if your newly added SSE2 codes will affect on SSE(not SSE2) or even IA32(no any of SSE) builds?

[ownedbywuigi]

Update on 13.1, I got (mostly) stable e10s, making that only one page freezes instead of the whole browser. Only regression is the backspace keeps on making the page go back for some reason, which I will try to fix.

EDIT: REGRESSION FIXED. E10S IS FULLY STABLE!!

Wow. but I wonder if your newly added SSE2 codes will affect on SSE(not SSE2) or even IA32(no any of SSE) builds?

For some of the code, there is a CPU check for SSE/IA32 and for some it's just SSE2 only. Considering that Dactyloidae is 64 bit only, I do not think there will be a issue with IA-32 and/or SSE1

[q160765803]

Update on 13.1, I got (mostly) stable e10s, making that only one page freezes instead of the whole browser. Only regression is the backspace keeps on making the page go back for some reason, which I will try to fix.

EDIT: REGRESSION FIXED. E10S IS FULLY STABLE!!

Wow. but I wonder if your newly added SSE2 codes will affect on SSE(not SSE2) or even IA32(no any of SSE) builds?

For some of the code, there is a CPU check for SSE/IA32 and for some it's just SSE2 only. Considering that Dactyloidae is 64 bit only, I do not think there will be a issue with IA-32 and/or SSE1

Does it? Looked at this commit: https://repo.dactyloidae.xyz/Dactyloida ... 871b8be379
Using `vmovhlps` with 3 parameters, which means AVX, but you just check it with SSE2. Ref: https://www.felixcloutier.com/x86/movhlps

If these code go upstream, these are needed to be clarified.

[athenian200]

Does it? Looked at this commit: https://repo.dactyloidae.xyz/Dactyloida ... 871b8be379
Using `vmovhlps` with 3 parameters, which means AVX, but you just check it with SSE2. Ref: https://www.felixcloutier.com/x86/movhlps

If these code go upstream, these are needed to be clarified.

This kind of thing is actually a part of why we made AVX the baseline instead of SSE2. AVX instructions are useful for improving performance, and honestly it's easy to get mixed up and pull in AVX instructions as part of an optimization when you think you're staying well within SSE/SSE2 territory. Just making AVX the baseline means you don't have to worry as much and can just focus on using optimized instructions.

[q160765803]

Does it? Looked at this commit: https://repo.dactyloidae.xyz/Dactyloida ... 871b8be379
Using `vmovhlps` with 3 parameters, which means AVX, but you just check it with SSE2. Ref: https://www.felixcloutier.com/x86/movhlps

If these code go upstream, these are needed to be clarified.

This kind of thing is actually a part of why we made AVX the baseline instead of SSE2. AVX instructions are useful for improving performance, and honestly it's easy to get mixed up and pull in AVX instructions as part of an optimization when you think you're staying well within SSE/SSE2 territory. Just making AVX the baseline means you don't have to worry as much and can just focus on using optimized instructions.

It could be fine if it is guarding correctly, at least it won't crash when using it on pre-AVX processors. (while old code benefits older processors that has SSE3 but not AVX)

And to me, commits like this are mixing up in various purposes and it should be better to separated if these changes are going to upstream.

[ownedbywuigi]

with some major modifications to WebExtensions, I got latest uBlock Origin installed on 13.1.

[Mæstro]

Are you trying to add WebExtensions support to your browser? If not, what, precisely, are you doing?