#Nexirol&Renirol

1 messages · Page 2 of 1

deft kestrel
#

Updated initial message thing

deft kestrel
#

ah yes
₵Ʉ฿Ɇ

image.png
#

that's
a bit better

image.png
#

also a lot more epileptic

#

... why did the first two faces work and none of the other ones KEKW
and the second one is backwards because ofc, why wouldn't it be

terse fog
#

use the debugger

deft kestrel
#

you seem to be under the assumption I'm not already using a debugger

lapis turtle
#

Check your winding order

deft kestrel
#

there is
a lot more wrong here than just winding order KEKW

terse fog
#

just posting random renderdoc shots doesnt fix it

deft kestrel
#

I-
am using a debugger

terse fog
#

and i mean debugger debugger

#

add vertex colors

#

and shade each face in a different color, makes it easier to visually debug shit

deft kestrel
lapis turtle
#

you never specified you weren't KEKW

deft kestrel
#

this is my thread
I can post whatever issues I want iea

deft kestrel
# terse fog yes because its obvious 
            MemoryUtil.memCopy(
                    MemoryUtil.memAddress(faces[i].fb),
                    fb + i * (faceStride), faceStride
            );```also like
debugger kinda doesn't help with java when I'm dealing with native memory
#

which, I'mma not do that for right now KEKW

#

ok now next thing is colors

image.png
#

next problem: each face has a different vertex order

image.png
deft kestrel
#

a UV cube

#

actually don't think I've ever made one of these

#

actually the negatives were wrong

image.png
# 
        BoxMesh mesh = new BoxMeshBuilder()
                .withElement(3)
                .withElement(4)
                .withElement(2)
                .withSize(1, 1, 1)

                .allFaces()
                .colors(
                        0, 0, 0, 1,
                        1, 0, 0, 1,
                        1, 1, 0, 1,
                        0, 1, 0, 1
                )
                .uvs(
                        0, 0,
                        1, 0,
                        1, 1,
                        0, 1
                )

                .build();```
#

but hey at least I don't need like 5 constructors for box meshes!

#

this abomination can do all the box meshes I will need!

#

including this one which only has positive faces```java
BoxMesh mesh = new BoxMeshBuilder()
.withElement(3)
.withElement(4)
.withElement(2)
.withSize(1, 1, 1)

// .allFaces()
.emitFace(true, 0)
.emitFace(true, 1)
.emitFace(true, 2)
.colors(
0, 0, 0, 1,
1, 0, 0, 1,
1, 1, 0, 1,
0, 1, 0, 1
)
.uvs(
0, 0,
1, 0,
1, 1,
0, 1
)

            .build();```
image.png image.png
# 
        BoxMesh mesh1 = new BoxMeshBuilder()
                .withElement(3).withElement(4).withElement(2).withSize(10, 1, 10)

                .emitFace(true, 0).emitFace(false, 0).emitFace(true, 1).emitFace(false, 1)
                .colors(1, 1, 1, 1,
                        1, 1, 1, 1,
                        1, 1, 1, 1,
                        1, 1, 1, 1)
                .uvs(0, 0,
                        10, 0,
                        10, 1,
                        0, 1)

                .buildFace(true, 2)
                .colors(1, 1, 1, 1,
                        1, 1, 1, 1,
                        1, 1, 1, 1,
                        1, 1, 1, 1)
                .uvs(0, 0,
                        10, 0,
                        10, 10,
                        0, 10)
                .finishFace()

                .buildFace(false, 2)
                .colors(1, 1, 1, 1,
                        1, 1, 1, 1,
                        1, 1, 1, 1,
                        1, 1, 1, 1)
                .uvs(0, 0,
                        10, 0,
                        10, 10,
                        0, 10)
                .finishFace()

                .build();```I think it is obj, gltf, and scene loading time after I add normals
image.png
#

normals frOK

image.png
deft kestrel
# 
    public static ShaderProgram FLAT = new ShaderProgram(
            ShaderAttachment.of(GL20.GL_VERTEX_SHADER, FileUtils.readCL(Main.class, "flat_shader.vsh")),
            ShaderAttachment.of(GL20.GL_FRAGMENT_SHADER, FileUtils.readCL(Main.class, "flat_shader.fsh"))
    );```created a second constructor a bit ago which takes `ShaderAttachment`s which does  the stuff with the Shader class behind the scenes
🤷 basically the same thing, but it abstracts away having to deal with closing the `Shader` instances
#

KEKW how is the glb larger than the gltf

image.png
#

3dviewer cannot handle deccer's cubes agonyfrog
(reason: 3dviewer does not support nearest neighbor sampling)

image.png
#

3d viewer tooo

image.png image.png
#

frOK thanks microsoft

terse fog
deft kestrel
terse fog
#

you can ignore geometry and tessellation

#

the attachment thig is realy annoying when you dont even need to use it as a consumer of your api

#

but its just the first iteration, so its, fine, you can refactor stuff later, or rewrite the whole thing

lethal axle
#

It pains me to observe the amount of time spent worrying about code structure instead of making progress on drawing pixels.

#

Having said that you’re already drawing deccer cubes and I am nowhere close to that

#

So I should be quiet, have fun navel gazing your classification systems

#

Looking great though

deft kestrel
deft kestrel
#

It just tells the code hey I want you to load this shader as a fragment shader

lapis turtle
deft kestrel
deft kestrel
lethal axle
#

Close enough

#

Maybe another level of abstraction will do the trick?

deft kestrel
#

I think I need a gltf loader impl KEKW

#

(aka an impl of assimp)

#

Ok so I think

  • gltf
  • scene loading
  • data driven shader stuff (hey look something that will make developers not have to even know ShaderAttachment exists)
  • resume learnopengl
deft kestrel
#

So for my terrain object impl
… I kinda wanna support caves out of the box

#

Like actually have a feature for that

terse fog
#

what happened to loading gltfs

deft kestrel
#

I’m thinking ahead to features that will come after it

lapis turtle
#

You can always rewrite and refactor, sometimes thinking ahead takes longer than actually implementing the features

deft kestrel
#

Aware
But this feature is sorta one of the large reasons I want my own game engine; making caves in like…
UE4 or godot or unity is a massive pain

#

And caves are sorta a large plot point of the game I wanna make

terse fog
#

so either, voxels or just mesh and then embedded into terrain

deft kestrel
#

probably voxels smoothed out using surface nets, the somehow connected to be flush with the terrain

#

anyway, time to start looking at assimp

#

that is a bizzare name for a library imo

#

just data structures
which I can deal with on my own, designing them around what I need

#

bleaker_kekw Assimp::Importer does not seem to exist in the lwjgl bindings

terse fog
#

skill issue again

deft kestrel
# 
    AIScene scene = Assimp.aiImportFile(buffer, flags);

apparently lwjgl translated it to this

#

... first time I've seen lwjgl 3 stray away from the library that it was creating bindings for

#

that's
confusing and annoying

terse fog
#

its literally the same api 🙂

deft kestrel
#

Assimp::Importer importer;
const aiScene *scene = importer.ReadFile(path, aiProcess_Triangulate | aiProcess_FlipUVs);

    AIScene scene = Assimp.aiImportFile(buffer, flags);

ah yes because ReadFile is the same words as aiImportFile

#

lwjgl usually has a 1:1 match on the names

#

although
maybe it's not the same thing

# 
compile 'com.github.kotlin-graphics:assimp:1be72df603febc1b6b22dd5c7085af12aa1a832b'

h

#

GLTF: could not open referenced file "SM_Deccer_Cubes.bin"
wh

#

oh yeah that makes sense ig

#

... so I suppose I'm not gonna be able to load deccer cubes from RAM with assimp, huh? (Like without writing it to the disk)

deft kestrel
#

it's something

#

microsoft's bee mesh is like 700 units away

image.png
lapis turtle
deft kestrel
# lapis turtle what?

deccer cubes requires a few extra files that aren't embedded into the gltf
so assimp then tries to find those on the disk, but can't because the files aren't in the working dir, since I'm trying to load stuff from memory instead

lapis turtle
#

why not

#

move them into the current working directory

deft kestrel
#

and that's a gltf loaded

image.png
deft kestrel
#

I like stuff being able to be done entirely in memory

lapis turtle
#

so you're going to read a file from disk to memory to read it?

#

Im not quite sure I understand what you're saying here

deft kestrel
#

basically
my file can be in a zip file
Assimp assume it is either in RAM or on the disk (as far as I know, compressed in a zip file does not count)
if it is in RAM, the file must either have no dependencies or have those dependencies extracted to the disk ahead of time

I do not want to extract files at all if I don't have to, but deccer's cubes have a dependency file (SM_Deccer_Cubes.bin in the case of the deccer's cube file I was trying to load)
so Assimp would automatically look for it in the working directory, where it was not located
and the working directory cannot be set inside of a zip file

lapis turtle
#

if you're not constantly updating the mesh its not a problem

deft kestrel
#

KEKW I've seen multiple programs cripple themselves in this sorta assumption

#

(granted, most of them are zip manipulation utilities and such)

#

doing like a single funny thing with zip operations can make a program take like 5* as long to run

#

and like
most of the time it's not even difficult to not do said funny thing

#

deccer just vanished from the thread?

deft kestrel
#

... why is this an int buffer?

image.png
# 
/**
 * Specifies the number of components for a given UV channel.
 * 
 * <p>Up to three channels are supported (UVW, for accessing volume or cube maps). If the value is 2 for a given channel n, the component {@code p.z} of
 * {@code mTextureCoords[n][p]} is set to 0.0f. If the value is 1 for a given channel, {@code p.y} is set to 0.0f, too.</p>
 * 
 * <p>Note: 4D coordinates are not supported.</p>
 */

... okay?

#

then why is it a 2d array?

#

wait no

#

why is it 8 elements

deft kestrel
#

this is as far as I can currently go with GLTF

image.png
#

agonyfrog learn opengl doesn't do lighting in compute or post

deft kestrel
# 
    nMap = pow(abs(nMap), 0.5) * sign(nMap);

wh-

#

Warning, (version, profile) forced to be (110, compatibility), while in source code it is (330, core)
ERROR: #version: versions before 140 do not allow a profile token
ERROR: 0:17: '' : syntax error, unexpected IDENTIFIER
ERROR: 2 compilation errors. No code generated.
why does that cause this KEKW

#

regardless; normal maps approximately applies without using tangents and binormals

image.png
#

temporary workaround for the fact that I do not have tangents and binormals KEKW

#

also looking at this with specular on is making me think of voices of the void

#

oh hey, assimp just casually provides me with tangents and bitangents
guess that's a good next thing to hook up 🤷

# 
            SimpleRenderState state = new SimpleRenderState();
            state
                    .withTexture(textures.get(2), 0)
                    .withTexture(textures.get(1), 1)
                    .withTexture(textures.get(0), 2)
                    .withTexture(textures.get(3), 3)
                    .withSetup(() -> {
                        ShaderPrograms.PBR_SHADER.uniformInt(
                                "textureMaps", 1  | 2 | 4 | 8
                        );
                        ShaderPrograms.PBR_SHADER.uniformInt("ambient", 1);
                        ShaderPrograms.PBR_SHADER.uniformInt("normalMap", 2);
                        ShaderPrograms.PBR_SHADER.uniformInt("specularMetalic", 3);
                    });```this should also be abstracted away 'cuz it's a bit ugly ~~and also using hardcoded constants that aren't guaranteed to be the case for all models~~
#

KEKW ah yes the bee is like 700* as large as the dragon

image.png
#

the bee is actually only ~17.5 times the size of the dragon, apparently
which is a difference of 1,400 units vs 80 units

#

how exactly
is a colored ambient occlusion map supposed to be interpreted?

image.png
#

3d viewer seems to just be looking at .r

#

is-
this like a screenspace AO approximation system?

#

looking closer it seems like that's probably what it's supposed to be?

#

can't tell

#

my problem is that I like the java syntax too much and I can't get rid of it KEKW

# 
    if (str0 .= str1) {

also due to me not entirely liking operator overloads
custom operators may or may not be a thing at some point
for now, hardcoded special ones that redirect to functions

deft kestrel
#

Exception in thread "main" java.lang.UnsatisfiedLinkError: Failed to locate library: LLVM.dll
excuse me lwjgl
but how am I supposed to provide a file that I can't even find on the internet?

#

like I literally cannot find anything about LLVM.dll
closest I can find is LLVM-C.dll, which lwjgl is not happy with linking against

#

even though it seems like it's also supposed to link against LLVM-C in addition to LLVM

lethal axle
#

oh interesting

deft kestrel
#

I might go for a modified mockup of blender’s walk camera controls as the default for my game engine

#

"default" KEKW
I am only implementing one camera control scheme for my editor’s camera

#

So more like just "the camera controls"

deft kestrel
deft kestrel
#

Thinking of using abi as the keyword to indicate C/C++ calls

deft kestrel
#

Result: 5.0

local var0 = 5
return var0```
to learn LLVM
I'm making an LUA->LLVM JIT compiler
*and it's going pretty smoothly so far*
#

only caused like 5 segfaults or so so far

#

maybe 10

deft kestrel
lapis turtle
deft kestrel
#

oop
wrote something llvm wasn't able to properly optimize with my configuration of passes

#

running my passes twice makes LLVM fully optimize it

# 
local varE = 5 + 3
local varE = varE + 6

for i=5, 10, 1 do
    local varE = varE + 1
end

if varE <= 5
then
    local varE = 0
elseif varE >= 20
then
    local varE = 2.5
    
    for i=5, 10, 1 do
        local varE = varE + 1
        for i1=5, 10, 1 do
            local varE = varE + 1
        end
    end
end

return varE```the (partially inaccurate) LUA code that generated that IR
#

currently my compiler crashes if you don't write local due to me not having global variables yet, and instead local variables act as globals wheeze

#

ig actually only semi-globals

#

eh yeah globals
had to think about how LUA works for a bit there

lapis turtle
deft kestrel
#

I'm sorta just loading everything as longs and casting to the proper data type afterwards because

  1. LLVM will optimize it for me
  2. the LLVM bindings I'm using don't seem to support loading doubles and floating point numbers wheeze
  3. it's easier
deft kestrel
#

my compiler's code flow is made with it in mind
it's just not in yet

lapis turtle
#

Keep a global array of scope

#

When you go in a scope append

#

When you go out of scope pop

deft kestrel
#

yeah I'm aware of how scoping works iea

lapis turtle
#

Its super simple

#

You should implement it immediately

deft kestrel
# 
global = {'print': <external_function>}
scope_0 = {}

global['var0'] = 5
scope_0['var1'] = 2
global['var'] = global['var0'] * scope_0['var1'] * global['var0'] + scope_0['var1'] * global['var0']

global['func0'] = function
  global['func1']()
end

global['func1'] = function
  global['print']("test")
end

global['print']("a")
global['func0']()```there's uh
also this
# 
var0 = 5 var1 = 2
var = var0 * var1 * var0 + var1 * var0

function func0()  
    func1()
end

function func1()  
    print("test")
end

print("a")```lua equivalent of that
lapis turtle
deft kestrel
#

if you call func0 before the line where func1 is defined, LUA will throw an error du to func1 not being defined yet
but if you call func0 after the line of func1, it'll run fine

#

functions are defined as they come in the file, when they come in the file

#

my compiler is also single pass, as is the actual LUA compiler

#

no AST to be found here

lapis turtle
deft kestrel
#

just sorta a quirk of how the interpreter works

image.png
deft kestrel
deft kestrel
deft kestrel
#

I kinda fear trying to link this stuff to JNI functions though KEKW
(I'm gonna want this to be able to call java code at some point)

deft kestrel
deft kestrel
#

LLVM ERROR: MCJIT::runFunction does not support full-featured argument passing. Please use ExecutionEngine::getFunctionAddress and cast the result to the desired function pointer type.
yipee apparently I can't use LLVM to run functions that take 64 bit integers as parameters

#

incredible

#

welp I guess I need a metacall function thing... or something

#

long result = JNI.invokePP(Double.doubleToLongBits(20), LLVM.LLVMGetFunctionAddress(engine, functionEntry.getName()));
eh this works...
for now

#

won't work in the long run

#

there's actually a really funny thing I can do with JNI and LLVM
where I generate a class at runtime (synthetic class), have it have a native method that takes in the required arg types, and bind it to the function I generate using LLVM

#

and call it using either method handles or jni

deft kestrel
# 
local varE = !0 + 3
local varE = varE + 6

for i=5, 10, 1 do
    local varE = varE + 1
end

if varE <= 5
then
    local varE = 0
elseif varE >= 20
then
    local varE = 2.5
    
    for i=5, 10, 1 do
        local varE = varE + 1
        for i1=5, 10, 1 do
            local varE = varE + 1
            for i2=5, 10, 1 do
                local varE = varE + 1
            end
            for i2=5, 10, 1 do
                local varE = varE + 1
            end
        end
        for i1=5, 10, 1 do
            local varE = varE + 1
            for i2=5, 10, 1 do
                local varE = varE + 1
            end
            for i2=5, 10, 1 do
                local varE = varE + 1
            end
        end
    end
end

return varE```broke my compiiler ![animal](https://cdn.discordapp.com/emojis/1202796733965271080.webp?size=128 "animal")
actually I think this might have to do with scopes
so, scoping time
#

yeah, it's scopes

#

scopes are now in; next thing is to update how local behaves to make it accurate to LUA
which took, very little time to setup thumblob

deft kestrel
#

~63-73 ms vs ~65-75 ms
(data sorta inaccurate, since I'm kinda excluding outliers)
optimization pass collection seems to shave off ~2ms

#

... though I don't think this data is accurate at all, considering I don't think executing arg0+3+6 should take longer than optimizing asm code that is more complex than that

#

jni linkage moment

#

4.7, 4.9, 6.9, 4.5, 5.4, 3.9, 5.2, 4.5

5.5, 5.8, 6.5, 4.1, 5.7, 5.9, 5.0, 4.0

ok, around 1ms

#

I think there might still be some JNI overhead or something going on there but that's close enough of a metric I think

#

and my code outputs 17, just as it should

deft kestrel
#

actually it throws an error without
115 seconds is the bare minimum optimization passes for it to not throw an error

deft kestrel
# 
%struct.JNINativeInterface_ = type { ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr }```

I-
what the heck
#

feels like an AI went rogue to generate that

#

oh ofc it compiled for 32 bit

#

Exception in thread "main" java.lang.NoSuchMethodError: LMyJavaClass;.nativeMethod(Ljava/lang/Class;Ljava/lang/String;Ljava/lang/String;)J
lol what

#

ah right, static and non-static are different id get methods

#

2227764525016
okay

#

... why are there so many ptrs wheeze

lethal axle
#

what's happening

deft kestrel
# lethal axle what's happening

I’m making adventures into compiling LUA for sake of learning LLVM compiler infrastructure so that I don’t have to learn it in my ralux project

lethal axle
#

Fun!

deft kestrel
#

Of which I need 170 (release chars utf), 113 (get method id), and 169 (get chars utf)

deft kestrel
# 
            LLVMValueRef envRef = LLVM.LLVMGetParam(builder.function, 0);
            LLVMValueRef alloca = LLVM.LLVMBuildAlloca(root.builder, jniEnv.type(), "envptr");
            LLVMValueRef env = LLVM.LLVMBuildLoad2(root.builder, jniEnv.type(), envRef, "envro");
            LLVMValueRef why = LLVM.LLVMBuildStore(root.builder, env, alloca);

            PointerPointer<LLVMValueRef> params = new PointerPointer<>(2);
            params.put(0, root.loadLong(0));
            params.put(1, root.loadInt(169));
            LLVM.LLVMBuildInBoundsGEP(
                    root.builder, alloca,
                    params, 2,
                    "llvm_is_being_dumb"
            );```after literally like an hour trying to get this one call to work I randomly decided to stop using my abstraction
#

... for some reason doing that made it work

#

even though behind the scenes I'm pretty sure it's the same thing

#

ok well time to translate it back to my abstraction

#

... and now it just works

#

bleaker_kekw without even doing anything dumb (such as pointlessly boxing and unboxing the pointer)

#

bleaker_kekw no idea what changed

# 
        LLVMValueRef nameChr = root.call(freeUTF, envRef, nameJSTR, jniFalse);

now this is giving me problems

#

0
incredible (this is very much so NOT the right value at all)

deft kestrel
#

actually I think I might know what's going on here

#

maybe not

#

and hey
I actually have information now!
this older version of llvm generates much nicer IR (for readability, at least)!

# 
; Function Attrs: nounwind uwtable
define dso_local dllexport i64 @Java_MyJavaClass_nativeMethod(
    %struct.JNINativeInterface_** %env_ptr,
    %struct._jobject* nocapture readnone %1,
    %struct._jobject* %2,
    %struct._jobject* %3,
    %struct._jobject %4*
) local_unnamed_addr #0 {
  %env_loaded = load %struct.JNINativeInterface_*, %struct.JNINativeInterface_** %env_ptr, align 8, !tbaa !3
  %getUTF = getelementptr inbounds %struct.JNINativeInterface_, %struct.JNINativeInterface_* %env_loaded, i64 0, i32 169
  %getUTFLoaded = load i8* (%struct.JNINativeInterface_**, %struct._jobject*, i8*)*, i8* (%struct.JNINativeInterface_**, %struct._jobject*, i8*)** %getUTF, align 8, !tbaa !7
  %9 = tail call i8* %getUTFLoaded(%struct.JNINativeInterface_** %env_ptr, %struct._jobject* %3, i8* null) #1```ok so now I need to recreate this
#

and if I can do that
I should actually have a functional file

#

140709365796720
yes

#

I have done it
I have called a jni method without segfaulting
next step: use jni

deft kestrel
# 
        ProtoJNI.init();
        System.out.println(ProtoJNI.getJNIEnv());
        System.out.println("MID");
        long methodID = ProtoJNI.getStaticMethodID(Test.class, "print", "(Ljava/lang/String;)V");
        System.out.println(methodID);
        ProtoJNI.callStaticVoid(ProtoJNI.class, methodID, "Hello!");```

> 140708482422784
> MID
> 2138686952888
> Hello!
#

LLVM's optimization passes agree with my code, other than the lack of alignments
and lack of tail call

#

pretty sure this is also more optimal than O3

lethal axle
#

is this llvm lua thing going to render triangles

deft kestrel
#

well
if I get it hooked up to computer craft like I want to, I can then add on tom’s perhiperals to get a software implementation of like GL11 or something (no direct hardware access, unfortunately)
But this llvm lua thing is mostly just a learning project for a bigger project that will be used for drawing triangles

lethal axle
#

what are computer craft and tom's perhiperals

deft kestrel
#

It uh
Doesn’t do depth clipping with the near plane last I saw it, so it can’t really be used for first person stuff bleakekw

#

(I do not really use computer craft ever myself; someone mentioned compiling computer craft scripts and I decided hey that’d probably be a good way to learn LLVM IR)

lethal axle
#

are you also active on the minecraft mod discord?

deft kestrel
#

Which one?
There are many minecraft mod discords

#

There are a large number of mod authors who have their own discord server (or servers)
There are probably several general modding servers
There’s probably several general mod pack dev servers
There’s also probably several general mod dev servers

deft kestrel
lethal axle
#

I don't know anything about them, just I imagined there must be one

deft kestrel
#

I’m active on a few
mostly just in the communities on them though, and they’re the servers owned by certain devs; not for just general modding

#

Sometimes in life
You try to say something meaningful
Then realize you broke the functionality of loops as a syntax feature while doing so

IMG_4020.png
#

git logs are fun when they tell a story

deft kestrel
#

he did infact not break loops

image.png
lethal axle
#

you have your own minecraft mod discord

deft kestrel
#

mhm

deft kestrel
#

it works and I don’t know why and this concerns me

deft kestrel
#

So at this rate
It’ll be like a week till I get stuff polished enough for me to begin implementing functions
since I kinda wanna get all the type coceriencio coercion weird word properly implemented first

#

And that’s talking calling JNI functions, not defining functions using LUA’s syntax

#

Actually an LUA based dependency management system sounds fun

lethal axle
#

that's awesome

#

I would be curious about your lua deps management system

#

I like lua

#

I used it in blockens

deft kestrel
#

Basically
In java
We have this tool called gradle
It’s what a lot of java devs use for dependency management and as a build system
But a lot of people sorta treat it like a black box and don’t like to work with it

Personally, I’m a fan of it because I can actually script it to automate tasks
But it’s slow as heck and not easy to figure out how to do complex stuff with it

#

My way of learning LUA
was programming a graphing calculator
Which used TI BASIC, not LUA

#

but it had the if then end thing going on like LUA has so close enough tbh

deft kestrel
lethal axle
#

nice

deft kestrel
#

LLVM ERROR: Cannot select: 0x18d2bd1a798: i64 = bitcast Constant:i8<116>
0x18d2bd1b0f0: i8 = Constant<116>
In function: $$module_lua_jit_$root_entry$_tfc.jlluavm.parse.LUASyntaxConsumer5ae9a829
I broke LLVM

elder pondBOT
#

Lyzantra
Compile Error! Click the errors reaction for more information.
(You may edit your message to recompile.)

930203537776996412.png
deft kestrel
deft kestrel
#

So is this like
Going to just be a thing from now on when using block quotes?

#

test

#

test
test

#

heuh?

deft kestrel
#

Bjorn, thanks for reminding me of wuw

lethal axle
#

np np

#

yours is slightly different

deft kestrel
#

(Clip from me working on optimizing display logic)

deft kestrel
#

As far as I know
I have two options for optimizing this further:

  • go modify CEF itself to keep its buffers in consistent locations because it does not seem to do that, which is… absolutely bizarre to me
  • go modify JCEF to make it support OnAcceleratedPaint and figure out context sharing (more viable option)
deft kestrel
#

… I am now under the impression I would have to modify chromium to do that first one and I’m finding people saying that onacceleratedpaint has been removed and can find barely any info on it

lethal axle
#

that's your app?

#

is this another project?

deft kestrel
#

CEF and JCEF are dependencies of MCEF
MCEF is not a project I own, but it is one I work on

lethal axle
#

this is a browser in a minecraft plugin?

#

I didn't realize this was minecraft related

#

is minecraft just like an operating system now

deft kestrel
#

mod, not plugin
plugins exist only on the server, mods exist on both client and server

lethal axle
#

I just play vanilla minecraft, I have never once installed a minecraft mod

#

it's like this alternative universe of minecraft that just is not all interesting to me, minecraft doesn't need any mods imo

#

and it's just all this unecessary stuff that doesn't make the game better or more interesting

#

I mean some of the early popular mods had a ton of automation

#

but now...a browser mod

#

cool

#

I already have a browser, I didn't need minecraft to have one

#

it's just people showing off I guess

#

idk your fun isn't wrong, it's just not making minecraft a more fun game though

#

imo anyway

#

people should use all that pretty amazing talent and time to build something that's theirs and exist external to minecraft that people will actually use because it doesn't exist in minecraft and people don't want to use a browser in minecraft

#

like it's almost like minecraft mods are blackhole, a virus, that captures talented people and consumes all the value that they might have improved people's lives with

#

instead it just goes into a fucking meme that might get some attention on minecraft forum

#

anyway

#

+1 cool, very impressive stuff

deft kestrel
deft kestrel
lethal axle
#

hrmm

#

why

deft kestrel
#

because people want to be able to display content in game and watch youtube with eachother and stuff

#

from within a game

#

now what is a meme
is the fact that at some point I was working on a smart watch mod

lethal axle
#

I asked my daughter if she would do that, she plays a lot of roblox and minecraft and she didn't know it existed but she said it sounded cool

#

just now

#

idk ok I guess it has value, there are lot of kids playing minecraft

#

sorry lol

deft kestrel
#

^ there is genuinely no real purpose to this KEKW

#

also this item can add actual gameplay value

2024-10-20_00.54.16.png
lethal axle
#

you know, alt tab is a thing

deft kestrel
#

yeah I use alt tab like every 5 seconds

lethal axle
#

add a chrome browser to counter strike

#

I bet garry's mod has one

deft kestrel
#

also, I'm quite literally having to work on a modified version of JCEF in order to make this mod

deft kestrel
lethal axle
#

studio execs should consider if modders haven't added a chrome browser to their "successful" game, was it really successful?

deft kestrel
#

lol

#

like I don't think I've ever googled so much for a single thing in my entire time that I've been writing code

#

jcef doesn't even support audio handlers, I had to go implement that in C++

deft kestrel
# lethal axle it's just people showing off I guess

mc modding is sorta just
you're given this massive toolbox that is really easy to use
you don't have to use it, but using it makes your life far easier and reduces the odds of incompatibilities
so if you wanna go learn something, you can just spin up a minecraft dev env and ignore whatever parts of the toolbox you don't want to use and use whatever ones you want for convenience sake
or if you wanna make something that you plan to release, you can choose as much as the toolbox as you want

lethal axle
#

sure, that's true

deft kestrel
#

ok so on my LUA compiler
I think I've actually reached a point where it would be easier for me to start making a compiler for my own language, to force myself to get up a good wrapper for LLVM and force myself to figure out GEP and stuff

deft kestrel
#

JNI decided to be a road block when dealing with strings it seems

#

I'm not sure

lapis turtle
#

simple solution

#

use c

deft kestrel
#

That’s such a boring solution though

lapis turtle
#

patch jni yourself

deft kestrel
#

Oh no you missunderstand
The program is segfaulting when I try to call JNI
And I think this is because of strings but I’m not actually sure
So I kinda wanna get an actual thing going that will compile stuff properly and such

lapis turtle
#

write it in c

deft kestrel
#

That’s a boring solution and does not teach me how to write a compiler

lapis turtle
deft kestrel
#

KEKW writing the compiler in C also wouldn’t solve my JNI issue

#

The code I’m compiling does JNI stuff

lapis turtle
#

Hostage style

#

Not boring solution

deft kestrel
#

It’s probably my fault that there’s problems anyway

lapis turtle
#

Stab the problems in the heart

deft kestrel
#

I am the problem

lethal axle
#

write the compiler in zig

deft kestrel
# 
        root.toTargetMachine(
                new Target(
                        Architecture.X86_64,
                        Vendor.APPLE,
                        OperatingSystem.WINDOWS,
                        Environment.NEWLIB
                ),
                CPU.GENERIC
        );```this seems like a good target config
#

(according to ChatGPT, setting APPLE as the vendor is good for debugger support)

#

I'll probably play around with environment a bit more when I actually have a compiler up and going for ralux

deft kestrel
#

for simplicity reasons I'm making my own int to string function for now

deft kestrel
#

it will have one small problem for now: it builds the string backwards KEKW

#

so if you give it 1234, it'll output 4321

#

that is, if it works

#

which I would be shocked if this works first try

#

it segfaults the ir builder frOK

#

ech

#

GEP really likes to cause me pain

#

fortunately, adding a little bit to my C test file and telling LLVM to compile it to IR tells me exactly why it's causing me pain in this case

#

Call parameter type does not match function signature!
[6 x i8]* @Hello
i8* %callPuts = call i32 @puts([6 x i8]* @Hello)
Failed to verify function main
Both operands to a binary operator are not of the same type!
%offset_to_utf = add i32 %get_last_digit, i8 48
Invalid bitcast
%as_int8 = bitcast i32 %offset_to_utf to i8
Invalid bitcast
%as_int64 = bitcast i32 %index1 to i64
Failed to verify function intToString
Call parameter type does not match function signature!
[6 x i8]* @Hello
i8* %callPuts = call i32 @puts([6 x i8]* @Hello)
Both operands to a binary operator are not of the same type!
%offset_to_utf = add i32 %get_last_digit, i8 48
Invalid bitcast
%as_int8 = bitcast i32 %offset_to_utf to i8
Invalid bitcast
%as_int64 = bitcast i32 %index1 to i64
makes sense

#

i8* %callPuts = call i32 @puts([6 x i8]* @Hello)
wait that's not valid?

#

... but it works

deft kestrel
#

%callPuts = call i32 @puts(i8* getelementptr inbounds ([6 x i8], [6 x i8]* @Hello, i64 0, i64 0))

well this is what is valid, so 🤷 ig

#

now to debug why I get empty string from my int to str

deft kestrel
#

reason for my code not working: I used EQ instead of NE

#

👏

#

that was the only bug after fixing the stuff that was flat out invalid and prevented it from building

#

LLVM ERROR: Unsupported library call operation!
ah yes
unsupported library call operation
clearly the correct error message for someone using a 32768 bit integer

#

it performed the analysis and then told me it couldn't perform analysis

image.png
#

(blue icon is labeled "View Analysis" and shows the same code as the previous requests)

#

either that or it still generates the analysis and then doesn't run it

lethal axle
#

claude recently got updated

#

it's supposed to be better at code now

#

I can't tell the difference

#

they didn't version it though, just "new"

#

also computer use

deft kestrel
#

ok bug located

image.png
#

this happens if you use negatives

#

that's not a fun one to fix

#

I know exactly what's happening
rem returns a signed integer
so if I give it a negative number, I get a negative number back

#

which... results in that

#

I... think?

#

yeah

#

this website just flat out lied to me 👏

image.png
#

result there should be -3

#

I'mma just use a branch for now I guess
or maybe forever, since I can just setup the ralux compiler to always go agressive mode when optimizing this function and LLVM is incredible at optimizing stuff and should actually get optimal code for this assuming the actual logic is optimal

#

also I kinda wanna implement as much stuff in ralux as I can to maximize portability
'cuz I want ralux to feel like a high level lang

lethal axle
#

is that ui you made?

#

oh it's a website

deft kestrel
#

... and suddenly my code depends on msvc it seems..?

image.png
#

... llvm is inserting stack protection stuff automatically, what

lethal axle
#

I just always alt tab to my browser and open the dev console and do modulo or whatever in there frogegreenexcited

#

I have a powershell helm thing I should really use it

deft kestrel
#

I'm starting to just go to one compiler lua when I want to run stuff :iea:

#

slowly

#

turns out not using branching is really easyjava LLVMValueRef conditionPN = root.compareInt(ECompOp.GE, arg1, zero, "comp_val_pos"); dig = root.select( conditionPN, dig, root.negate(dig) );

#

that's a nice thing to have added to my toolkit; will really come in handy for null init and mentally reboots ternaries

#

close enough
why is my - sign a ]? KEKW

image.png
#

is this ascii?

#

no

#

-3 gets a -, but I'm not sure if it's the correct -

#

it is

#

... I'mma guess this is a locale thing, because chatgpt agrees that that should not be the case

#

in like every encoding that exists, 45 is -
not -3

#

now the only thing left for my to string function is to reverse it

#

oh also I need to fix the case of number==0

#

because 0 currently yields an empty string but should yield 0 as a string
but I think I'll fix that after I have a compiler, so I don't have to do it in raw LLVM IR

#

reverse can also come after a compiler ig
I can deal with the backwards strings while debugging stuff 🤷

# 
    public static abi int puts([+ AbiType: byte* +] String text);
# 
    [+ NumericPrims->T int +]
    [+ T->SizeOf->size +]
    public static String intToString(T integer) {
        char[] buffer = new char[size];
        // TODO: implement
        return null;
    }```ralux compiler hints are gonna be so fun to mess with
deft kestrel
#

I love IDEA plugins

lapis turtle
# deft kestrel

average java programmer trying not to have CompilerInvokerFactoryPatternMaker

deft kestrel
lapis turtle
#

its just funny

deft kestrel
#

It just runs the compiler with a simple test file

#

I hate the dumb enterprise object factory builder blah blah blah nonsense

lapis turtle
#

based java programmer

deft kestrel
#

bleaker_kekw java enterprise codebases are a nightmare

#

Builders are nice
But you don’t need builders for bloody everything
You got like two parameters? Just use a freaking constructor like a sane person
Got 20 parameters, or like 3+ optional parameters? Now THAT’S a topic for a builder

lapis turtle
#

ye

deft kestrel
#

Builder style setters are also just nice to use instead of regular setters
‘Cuz then you can chain .setThing()
But that doesn’t require a dedicated builder class bleakekw

lapis turtle
#

I dont like getters and setters personally, unless its used for like some internal thing

deft kestrel
#

Agreed

lapis turtle
#

but even then

#

if its internal, and private you can just access it anyways

deft kestrel
#

There are cases where setters on public things are nice
But usually I’d prefer to just access the field directly

lapis turtle
#

ye

deft kestrel
#

getters on public things are good for mod support though

#

or
ig just getters and setters in general

#

‘Cuz it creates a convenient place to hijack stuff

#

BuilderRoot extends ModuleRoot extends ContextRoot
That’s a trick I learned from mojang; BuilderRoot is a large file so I separated it into three files to break the code apart a bit
Probably gonna split it up even further later on

#

The Level (World) class in minecraft is a tree of like… I think at least 10 interfaces and probably like 3-5 abstract classes
Because it’s just such a large thing that’d be a mess elsewise

deft kestrel
# 
    // TODO: scope nonsense (and I really mean nonsense, 'cuz it is gonna be some really real nonsense)
    Value currentValue;
#

for a second I just got really scared about ternaries
then realized I have nothing to worry about

#

I'm also probably gonna implement arrays as a pointer and an integer tbh

#

antlr->llvm for expressions is gonna not be fun

image.png
deft kestrel
# 
; ModuleID = 'module'
source_filename = "module"

define i32 @main() {
entry:
  %i = alloca i32
  store i32 0, i32* %i
}```I can now compile```java
int i = 0;
#

store i32 10, i32* %i

and various other integer constants

#

because it would be silly if I couldn't do other int constants

#

next comes
i += 5;

and then
return i;

#

public static int main() {
being a native lang, main returns an int

lapis turtle
#

are you using this for like minecraft mods?

#

the compiler

deft kestrel
#

this compiler is not going to be used for mc mods

#

antlr can compile to java, so java ended up being convenient in the short term
... probably would've been easier to use C# or C++ or like anything that is not java for this compiler though

#

but at this point I have enough of a system setup that I don't really care to switch 🤷

deft kestrel
#

Might also give vulkan another try with ralux
Not sure
Probably gonna design the wrapper before I hook up vulkan this time if I do, to force myself to create an API I’m happy with instead of making the API around vulkan

lapis turtle
#

I wouldnt recommend fitting an opengl api to vulkan

#

From what I’ve seen from other game engines/renderers is vulkan focused rhis apply well to other gApis, while other gApi focused rhi’s do not apply well

deft kestrel
#

Ok but also I don’t want creating a framebuffer to require me to through a 5 step process and track like 20 objects or whatever

lapis turtle
#

Do you not use dynamic rendering

#

Also in most abstractions you shouldn’t even need to think about framebuffers

deft kestrel
#

My framebuffers were a CPU data structure, ‘cuz yes, I was using dynamic rendering
But I was keeping track of far too many objects throughout my program with my vulkan based renderer

#

I’m just used to having framebuffers due to working with opengl and minecraft KEKW

lapis turtle
#

Yeah you’re doing something fuckedd then

lapis turtle
deft kestrel
#

My FrameBuffer class just sorta held the list of things to use for the dynamic rendering pass

lapis turtle
#

Commandbuffer != framebuffer

#

Two separate concepts

#

Command buffer is the actual object we submit gpu commands to, framebuffer describes attachements, draw extent and other state

deft kestrel
deft kestrel
lapis turtle
#

Why do you have an explicit framebuffer object when you already have enough info to setup a framebuffer on the fly 😭

deft kestrel
#

smth smth render passes

lapis turtle
#

A renderpass object should contain the info to setup a framebuffer

deft kestrel
lapis turtle
deft kestrel
#

And again, used to opengl KEKW

#

Opengl framebuffer setup is a nightmare for memorization

#

I tend to like a mixture of high and low level, personally

lapis turtle
#

Abstraction over gl is a nightmare in general

deft kestrel
#

Mojang did it pretty well

deft kestrel
# 
define i32 @main() {
entry:
  %i = alloca i32
  store i32 15, i32* %i
  %compute_sum = add i32* %i, i32 5
  store i32* %compute_sum, i32* %i
  %get_value = load i32, i32* %i
  ret i32 %get_value
}```nice
#

next is expression trees

deft kestrel
#

there's a plugin to represent classes as 3d graphs of methods and variables and params and constructors
what. the heck.

image.png
#

(Graph Buddy)

lethal axle
#

are these s-expressions?

deft kestrel
#

s-expressions?

lethal axle
#

polish notation

#
  • 1 2
#

= 3

#

the operator followed by the arguments

#

instead of a LH operator RH

#

1 + 2

deft kestrel
#

oh
previously it was onlyjava int i = 15; i += 5; return i;

lethal axle
#

= 3

deft kestrel
#

now it'sjava int i = 15; i += 5 + i; return i;

lethal axle
deft kestrel
lethal axle
#

no I mean the app

#

what's that a screenshot of

deft kestrel
#

ah
it's an intellij plugin called Graph Buddy
so the app is intellij :iea:

#

looks a lot nicer in 2d with small classes

image.png
#

same class with vars and params shown

image.png
#

define i32 @main() {
entry:
%i = alloca i32
store i32 15, i32* %i
%get_value = load i32, i32* %i
%compute_quotient = sdiv i32 5, %get_value
%compute_sum = add i32 %compute_quotient, 3
%compute_diff = sub i32* %i, i32 %compute_sum
store i32* %compute_diff, i32* %i
%get_value1 = load i32, i32* %i
ret i32 %get_value1
}

int i = 15;
i -= 5 / i + 3;
return i;

#

compiling multiple functions (of course) also works

#

granted they can't
interact with eachother at all yet

deft kestrel
#

probably gonna make my own IR format at some point
'cuz performing analysis of my own is kinda difficult with LLVM IR

lapis turtle
#

It can be fun

deft kestrel
#

Both operands to a binary operator are not of the same type!
%compute_diff = sub i32* %i, i32 %compute_sum6
Stored value type does not match pointer operand type!
store i32* %compute_diff, i32* %i
i32Both operands to a binary operator are not of the same type!
%compute_sum10 = add i32* %i, i32 %get_value9
Stored value type does not match pointer operand type!
store i32* %compute_sum10, i32* %i
i32Failed to verify function main
Both operands to a binary operator are not of the same type!
%compute_diff = sub i32* %i, i32 %compute_sum6
Stored value type does not match pointer operand type!
store i32* %compute_diff, i32* %i
i32Both operands to a binary operator are not of the same type!
%compute_sum10 = add i32* %i, i32 %get_value9
Stored value type does not match pointer operand type!
store i32* %compute_sum10, i32* %i
this is why you make sure you validate your llvm ir before you try to implement new features

#

(I forgot a .getValue())

#

er

#

yeah

#

-13492
-13492```java
pkg tfc.test;

public class TestClass {
public static int main() {
int i = 15;
i -= 5 / (i + 3) + i * i * i + (5 + 8);

    int 3a = 3 * i;
    i += 3a;
    
    return i;
}

}```sweet

image.png
#

next up: the rest of the integer types

#

or ig coercion
whatever breaks the compiler first

#

coercion it is!

# 
            if (impl.symbol.getText().equals("boolean")) isInt = !(isBool = true);
```java is fun
#

... what

image.png
#

80 bit floating point number

#

that's not even a power of 2, what

#

why

#

regardless; fp types come next though

# 
    public Type coerce(BuilderRoot root, Type type) {
//        if (!this.equals(type)) throw new RuntimeException("NYI");
        if (intInt(type)) {
            if (signed && type.signed) {
                int selfSize = root.getIntSize(llvm);
                int otherSize = root.getIntSize(type.llvm);

                if (selfSize < otherSize) return type;
                return this;
            } else if (!signed && !type.signed) {
//                int selfSize = root.getIntSize(llvm);
//                int otherSize = root.getIntSize(type.llvm);
//
//                if (selfSize < otherSize) return type;
//                return this;
                throw new RuntimeException("NYI");
            } else throw new RuntimeException("NYI");
        } else {
            if (isInt || type.isInt) {
                if (isFloat() && type.isInt) {
                    throw new RuntimeException("NYI");
                } else if (isInt && type.isFloat()) {
                    throw new RuntimeException("NYI");
                } else {
                    throw new RuntimeException("NYI");
                }
            }
            throw new RuntimeException("NYI");
        }
    }```type coercion functions are
quite something ![KEKW](https://cdn.discordapp.com/emojis/666849321462792234.webp?size=128 "KEKW")
deft kestrel
#

%14 = call double @pow(double %13, double 4.000000e+00) #5
funfact: LLVM does not optimize pow

#

%14 = %13 * %13
%15 = %14 * %13

#

(or at least, I'm pretty sure that'd be faster, gonna wanna profile that at some point)

#

probably faster for ints, considering C pow function is only applicable for doubles

# 
quadruple fv = 2;
fv += 2;
half i = fv * 10;
fv /= 2;
fv /= 4;
fv *= 100;
i += fv;

int asInt = i;
return asInt;```
quadruples, stretches, and halves
#

… things I will probably never use

#

Actually I might end up using halves

#

Next things:

  • return types
  • params
  • calls
deft kestrel
#

... I just realized
I'm gonna need to make LLVM not always optimize out casts...

#

'cuz I do sometimes use casts to manipulate data

#

(i << 24) >> 24 is annoying way to do what (byte) i does

#

I used nasm like once and suddenly I can figure out bitshifts in my head

#

well not like once but like

# 
                    public static stretch main() {
                        quadruple fv = 2;
                        fv += 2;
                        half i = fv * 10;
                        fv /= 2;
                        fv /= 4;
                        fv *= 100;
                        i += fv;
                        
                        return i;
                    }```anyway, return types can now be specified 🎉
# 

    public static abi int puts([+ AbiType: byte* +] String text);
    public static abi [+ AbiType: void* +] void[] memcpy(
        [+ AbiType: void* +] void[] dst,
        [+ AbiType: void* +] void[] src,
        unsigned long count
    );
    public static abi [+ AbiType: void* +] void[] realloc(
        [+ AbiType: void* +] void[] ptr,
        unsigned long new_size
    );```abi is gonna be
interesting to implement
to say the least
#

(void[] will probably only be valid for abi; outside of abi I'm probably gonna enforce generic types)

deft kestrel
#

OH
RIGHT
KEKW

#

I should probably parse all method headers before I start parsing methods

#

yeah that would probably save me a truckload of problems, lol

# 
define i32 @main() {
entry:
}

define i32 @test(i32, i32) {
entry:
  %lh = alloca i32
  store i32 %0, i32* %lh
  %rh = alloca i32
  store i32 %1, i32* %rh
  %fv = alloca fp128
  store fp128 0xL00000000000000004000000000000000, fp128* %fv
  %get_value = load fp128, fp128* %fv
  %compute_sum = fadd fp128 %get_value, 0xL00000000000000004000000000000000
  store fp128 %compute_sum, fp128* %fv
  %i = alloca half
  %get_value1 = load fp128, fp128* %fv
  %compute_prod = fmul fp128 %get_value1, 0xL00000000000000004002400000000000
  %cast_fp_to_fp = fptrunc fp128 %compute_prod to half
  store half %cast_fp_to_fp, half* %i
  %get_value2 = load fp128, fp128* %fv
  %compute_quotient = fdiv fp128 %get_value2, 0xL00000000000000004000000000000000
  store fp128 %compute_quotient, fp128* %fv
  %get_value3 = load fp128, fp128* %fv
  %compute_quotient4 = fdiv fp128 %get_value3, 0xL00000000000000004001000000000000
  store fp128 %compute_quotient4, fp128* %fv
  %get_value5 = load fp128, fp128* %fv
  %compute_prod6 = fmul fp128 %get_value5, 0xL00000000000000004005900000000000
  store fp128 %compute_prod6, fp128* %fv
  %get_value7 = load fp128, fp128* %fv
  %get_value8 = load half, half* %i
  %cast_fp_to_fp9 = fptrunc fp128 %get_value7 to half
  %compute_sum10 = fadd half %get_value8, %cast_fp_to_fp9
  store half %compute_sum10, half* %i
}```uh
#

that ain- oh

# 
define i32 @main() {
entry:
  %fv = alloca fp128
  store fp128 0xL00000000000000004000000000000000, fp128* %fv
  %get_value = load fp128, fp128* %fv
  %compute_sum = fadd fp128 %get_value, 0xL00000000000000004000000000000000
  store fp128 %compute_sum, fp128* %fv
  %i = alloca half
  %get_value1 = load fp128, fp128* %fv
  %compute_prod = fmul fp128 %get_value1, 0xL00000000000000004002400000000000
  %cast_fp_to_fp = fptrunc fp128 %compute_prod to half
  store half %cast_fp_to_fp, half* %i
  %get_value2 = load fp128, fp128* %fv
  %compute_quotient = fdiv fp128 %get_value2, 0xL00000000000000004000000000000000
  store fp128 %compute_quotient, fp128* %fv
  %get_value3 = load fp128, fp128* %fv
  %compute_quotient4 = fdiv fp128 %get_value3, 0xL00000000000000004001000000000000
  store fp128 %compute_quotient4, fp128* %fv
  %get_value5 = load fp128, fp128* %fv
  %compute_prod6 = fmul fp128 %get_value5, 0xL00000000000000004005900000000000
  store fp128 %compute_prod6, fp128* %fv
  %get_value7 = load fp128, fp128* %fv
  %get_value8 = load half, half* %i
  %cast_fp_to_fp9 = fptrunc fp128 %get_value7 to half
  %compute_sum10 = fadd half %get_value8, %cast_fp_to_fp9
  store half %compute_sum10, half* %i
}

define i32 @test(i32, i32) {
entry:
  %lh = alloca i32
  store i32 %0, i32* %lh
  %rh = alloca i32
  store i32 %1, i32* %rh
}```🎉
#

the number of switches I have in this compiler is unholy

#

I have never written so many switches in a single project

lethal axle
#

zig's core language repo features a ton of nested switches

#

it's also probably why zig's switches are so good, because the devs use them a lot

deft kestrel
# 
Object o = getObject();
o ?-> { // if o!=null, do block
  System.out.println(o.toString());
}```
#

as someone who likes nullability
nullability related features are something I'm inclined to think about

#

still don't know what exactly I want to do console IO stuff as

# 
List<Pair<Pair<String, List<Type>>, Pair<FunctionBuilder, Type>>> funcsByParams = new ArrayList<>();

java devs when being lazy

#

lld-link: error: undefined symbol: __gnu_f2h_ieee
ugh

#

I'm gonna have to figure out these link errors at some point

#

this would be a good thing to generate as a 3d graph imo

#

... maybe I look into having a graph based file browsing system for ralux dev

# 
define void @recursive() {
entry:
  br label %tailrecurse

tailrecurse:                                      ; preds = %tailrecurse, %entry
  br label %tailrecurse
}```ah yes, llvm optimized my recursive function
# 
define void @recursive() {
entry:
  call void @recursive()
  ret void
}```for reference, this is the unoptimized version
#

due to this optimization, it does not work properly; the program does not crash

#

it also is not perfectly optimized for the logic it was optimized to, could be optimized to define void @recursive() { entry: br label %entry }

deft kestrel
#

solution: link against clang_rt.builtins-x86_64.lib

#

unfortunately, still need to link against ucrt.lib for puts though

#

TIL
in javajava byte b = 0; b = -b;is invalid
why?
because -b casts b to an int

#

what.
the heck.

#

0
1
10
-10
-1
2147483647
-2147483648

    public static String toString(int integer) {
        char[] buf = new char[11];
        boolean condPN = integer < 0;
        if (integer == 0) {
            buf[0] = '0';
            return new String(buf, 0, 1);
        } else {
            int index = 0;
            while (integer != 0) {
                byte rem = (byte) (integer % 10);
                if (condPN) rem = (byte) -rem;
                integer /= 10;
                buf[index] = (char) (rem + '0');
                index++;
            }

            if (condPN)
                buf[index] = '-';
            else index -= 1;

            for (int i = 0; i < (index + 1) / 2; i++) {
                char cl = buf[i];
                char cr = buf[index - i];
                buf[index - i] = cl;
                buf[i] = cr;
            }
            return new String(buf, 0, index + 1);
        }
    }```welp that's int to string done (java)
#

not perfectly optimized even for java, but meh

# 
            buf[i] <-> buf[index - i];

value trading

deft kestrel
# deft kestrel 0 1 10 -10 -1 2147483647 -2147483648 ```java public static String toString(... 
    [+ NumericPrims->T int +]
    [+ T->SizeOf->size +]
    public static String intToString(T integer) {
        if (integer == 0) {
            return "0";
        } else {
            char[] buf = new char[size];
            boolean condPN = integer < 0;

            int index = 0;
            while (integer != 0) {
                T quotient = integer / 10;
                byte rem = (byte) (integer - (quotient * 10));
                if (condPN) rem = -rem;
                integer = quotient;
                buf[index] = (char) (rem + '0');
                index++;
            }

            if (condPN)
                buf[index] = '-';
            else index -= 1;

            for (int i = 0; i < (index + 1) / 2; i++) {
                buf[i] <-> buf[index - i];
            }
            buf = realloc(buf, index + 1);
            return (String) buf;
        }
    }```ralux
deft kestrel
#

line 20:8 mismatched input 'return' expecting {PRIMITIVE, 'new', 'if', 'for', 'while', 'do', 'break', 'continue', 'return', 'unsigned', '+', '-', '*', '**', '/', '^', '|', '&', '%', '||', '&&', '<', '>', '<=', '>=', '==', '!=', '.=', '.!=', '.#', '{', '}', ';', '%infer_semicolon%', '%enforce_semicolon%', WORD}

unexpected return
one of the things I was expecting is return

#

👏

#

(it's actually expecting specifically else because my parser is bugged)

deft kestrel
# 
        if (child.getChildCount() > 5) {
            if (child.getChild(5) instanceof RuleContext ctx) {
                switch (ctx.getRuleIndex()) {
                    case RaluxParser.RULE_semi_truck -> {
                        if (child.getChildCount() > 6)
                            throw new RuntimeException("NYI");
                    }
                    default -> throw new RuntimeException("NYI");
                }
            } else throw new RuntimeException("NYI");
        }```![KEKW](https://cdn.discordapp.com/emojis/666849321462792234.webp?size=128 "KEKW")
#

the amount of throw statements in this compiler is also absurd

#

ya know what
i don't care

#

it doesn't matter for this

#

what does matter
is that this is printing 94 instead of 0

image.png
deft kestrel
#

ah right, debug output is backwards
that's 49, not 94

#

frOK yeah if statements work properly

#

flawless string reversal logic

image.png
deft kestrel
#

outputs are no longer backwards frOK

image.png
#

yeah I got annoyed with this that I fixed the ir builder which is a temporary solution until I can compile the standard library

deft kestrel
#

it's really nice
if I do ANYTHING the compiler doesn't support, it just throws an erorr
which immediately tells me where that unimplemented thing should be implemented

deft kestrel
#

next:

  • the various types of loops
  • &&, ||
  • ??
lethal axle
#

is this in a public repo?

deft kestrel
#

absolutely not usable at all yet

#

when it is usable, I'm probably gonna go for opengl to get a sense of linkage and stuff
then tackle either vulkan or directx12
might go for directx12, since that's built in to the windows sdk

lapis turtle
lapis turtle
deft kestrel
lapis turtle
deft kestrel
#

Basically KEKW

deft kestrel
#

I really wish vr was less annoying to work with
and I don't mean code wise
I mean the actually taking on and off the headset a million times while developing and testing stuff

#

I still love the name semi truck KEKW

image.png
#

hm I just thought of something really silly and now I need to test it in java

#

aw it's not valid

# 
            for (int i1 = -1; ((i1 == -1) ? (i1 = 0) : i1) < 5; i1++) {
            }```but this is valid
# 
        for (System.out.print("Honestly quite"); ; System.out.println(" incredible")) {
        }```and so is this
# 
                        for (int a = 0; a < 100; a+=1) {
                            print(a);
                        }```anyway, this now works
#

gonna go implement break and continue nextjava for (int a = 0; a < 100; continue;) { print(a); a += 1; }KEKW

#

^ this is the silly idea I had for java, which ended up not being valid

#

do I expect anyone to be crazy enough to do this?
definitely freaking not
so am I going to make it explicitly illegal to do it?
no

# 
                        for (int a = 0; a < 100; if (a == 50) a += 1; else continue) {
                            print(a);
                            a += 1;
                        }```should also be able to do stuff like this I think ![bleaker_kekw](https://cdn.discordapp.com/emojis/1271044788086509619.webp?size=128 "bleaker_kekw")
#

which is really silly

#

huh
apparently I wrote this a bit smarter than that

#

ngl I kinda wanna see control flow in a loop header used in an actual codebase though

# 
                        for (int a = 0; a < 100; a += 1) {
                            if (a > 50) continue;
                            print(a);
                        }```this now works
now I want to support the meme ![KEKW](https://cdn.discordapp.com/emojis/666849321462792234.webp?size=128 "KEKW")
deft kestrel
#

... I think I'll make header control flow act upon parent loopjava for (int b = 0; b < 100; b += 1) { for (int a = 0; a < 100; if (a == 50) continue; else a += 1;) { print(a); a += 1; } }so continuing loop with var a in header would continue loop with var b here

#

I have no reference point for how this should work and it's kinda a meme feature anyway ¯_(ツ)_/¯

#

but ofc, the show must go on, so I must test it

# 
line 20:61 mismatched input 'a' expecting 'if'
line 20:68 mismatched input ')' expecting {PRIMITIVE, 'new', 'if', 'for', 'while', 'do', 'break', 'continue', 'return', 'unsigned', '+', '-', '*', '**', '/', '^', '|', '&', '%', '||', '&&', '<', '>', '<=', '>=', '==', '!=', '.=', '.!=', '.#', '{', '}', ';', '%infer_semicolon%', '%enforce_semicolon%', WORD}
line 27:17 missing WORD at '('
line 27:18 mismatched input '-' expecting {'<', '[', WORD}
line 29:8 extraneous input 'return' expecting {PRIMITIVE, C_TYPE, MODIFIER, 'static', 'unsigned', '<', '[', '}', ';', WORD}
line 29:16 mismatched input ';' expecting {'<', '[', WORD}
line 32:18 extraneous input 'int' expecting {C_TYPE, MODIFIER, 'static'}```o uh what
#

oh I don't have else implemented, right KEKW

#

yeah, works as I described

deft kestrel
# 

tree_exit22:                                      ; preds = %condition_false25
  br label %head19

condition_false25:                                ; preds = %tail21
  br label %tree_exit22

condition_true26:                                 ; preds = %tail21
  br label %tail

conclusion27:                                     ; preds = %head19
  br label %tail
}```there's a lot of just jump branches that I emit
good thing CFGSimplicifcation pass exists ![KEKW](https://cdn.discordapp.com/emojis/666849321462792234.webp?size=128 "KEKW")
#

I'mma go through llvm versions to try to find the newest that has bindings for passes

#

looks like 13.0.1 is the latest I can use

#

after that, some of the things I actually need in order for my compiler to even work disappear

#

🤷 I can look into porting later when I'm not limited by premade bindings

#

KEKW I think it somehow got stuck in a loop

image.png
#

feel like it's cycling between the same five files or something

#

though that may be it downoading multiple files at once
hard to tell

deft kestrel
#

oooo
13.0.1 supports coroutines

#

ok so change of plans

  • &&, ||
  • support more loop variants
  • if/else chains
#

then maybe... switches?
then improve method lookup
then I suppose garbage collector

#

how do switches actually work behind the scenes?
I've never understood that

#

I suppose the best way to find out would be to write a switch in C and compile it to llvm ir 🤷

deft kestrel
#

And whatever else is missing that I don’t immediately care about KEKW

#

Just found out
Forums posts can be organized in the channel list by unfollowing and refollowing the post

deft kestrel
lethal axle
#

Just wanted to look at the code

deft kestrel
#

Think you’ll dabble in ralux at all?

lethal axle
#

Never heard of it

deft kestrel
#

ralux is the name of the lang I’m working on

lethal axle
#

Oh

#

What is its use case?

#

I can test it out for you if you want

deft kestrel
#

Likely gonna try for general purpose
But I’m primarily planning on using it for game dev stuff

#

Low level, but meant to feel high level

lethal axle
#

That’s awesome, sure I will try it out

deft kestrel
lethal axle
#

Will your compiler be an exe or a jar?

deft kestrel
#

jar for a while, will eventually convert to exe

#

Wanna make sure lang is stable enough before I port the compiler to the lang so I don’t shoot myself in the foot KEKW

#

Because I have shot myself in the foot at least once by doing stuff like that KEKW

deft kestrel
lethal axle
#

I probably haven't had a java runtime installed in a decade now :P

deft kestrel
#

I have no java runtime installed on my desktop
Only development kits
If I wanna run a jar, I need to do it from command prompt KEKW

#

Also, intelliJ idea comes with a jdk, and I only know plugin dev for ij, so naturally I’m only making a ralux plugin for ij for now

lethal axle
#

I don't have intellij

#

I probably am not your target user, but I will be happy to give it a try, seems cool

deft kestrel
#

oh!
java ships with a command to convert a jar to an exe KEKW

image.png
#

wonder when that was added

#

I'mma see if I can get that working

#

KEKW this takes a long time to jar task due to including natives for windows, linux, and mac

#

which I should look into not doing

#

ok now I should only be including the natives for the architecture and os I'm building it for

#

and it still takes really long KEKW

# 
probably the largest jar I've ever made
#

incredible

image.png
#

I think I may need to make my own JIT if I want to make a VM
'cuz 600mb is ridiculous for embedding bleaker_kekw

image.png
lethal axle
#

you don't need to go out of your way for me on this project, I am unlikely user of it, I do like seeing what you're up to. I imagine you must have an idea of what problem you wanted to solve though since you are doing all this work

#

I am in the business of selecting pixel colors to put into frame buffers

deft kestrel
#

I never knew about jpackage before so I wanted to see if it's viable for stuff

#

... yeah I'm not downloading something for it

deft kestrel
#

KEKW found a bug

#

for loops don't define scopes properly

#

2 line fix though
ez

deft kestrel
#

entry:
%is_it_zero = icmp eq i32 %1, 0
br i1 %is_it_zero, label %zero, label %not_zero

common.ret: ; preds = %loop_body, %reverse, %zero
%common.ret.op = phi i32 [ 1, %zero ], [ 0, %reverse ], [ %div_by_2, %loop_body ]
ret i32 %common.ret.op

zero: ; preds = %entry
store i8 48, i8* %0, align 1
br label %common.ret

not_zero: ; preds = %entry
%comp_val_pos = icmp sgt i32 %1, -1
br i1 %comp_val_pos, label %body.us, label %body
not a fan of having 4 2 instruction branches in a single function though

#

but I don't think I can resolve that one

deft kestrel
# 
        LLVMPassManagerBuilderRef builderRef = LLVM.LLVMPassManagerBuilderCreate();
        LLVM.LLVMPassManagerBuilderSetOptLevel(builderRef, llvm);

        LLVMPassManagerRef pass = LLVM.LLVMCreatePassManager();
        LLVM.LLVMPassManagerBuilderPopulateModulePassManager(builderRef, pass);```just found out about this ![KEKW](https://cdn.discordapp.com/emojis/666849321462792234.webp?size=128 "KEKW")
deft kestrel
#

KEKW my breakpoints are bugged, what

#

that's a new one to me

#

oh my order of ops is hecked KEKW

image.png
#

bleaker_kekw just found out this library throws like a million errors before initializing

image.png
#

(java errors are slow btw!)

#

br <null operand!>, label %condition_true, label %condition_false
what :car:

#

why did my compiler not explode

#

I'm concerned
I forgot a throw 👏

# 
  store i32 %19, i32* %4, align 4, !dbg !146
  %20 = call zeroext i1 @boolFunc(), !dbg !147
  br i1 %20, label %21, label %25, !dbg !147

; <label>:21:                                     ; preds = %0
  %22 = call zeroext i1 @otherFunc(), !dbg !147
  br i1 %22, label %23, label %25, !dbg !147

; <label>:23:                                     ; preds = %21
  %24 = call i32 @puts(i8* getelementptr inbounds ([4 x i8], [4 x i8]* @"??_C@_03CFIIMAOO@lol?$AA@", i32 0, i32 0)), !dbg !148
  br label %25, !dbg !151```ok so short circuit eval really is just branch
makes sense
deft kestrel
# 
            LLVM.LLVMPassManagerBuilderPopulateModulePassManager(builderRef, pass);
            LLVM.LLVMAddDeadArgEliminationPass(pass);
            LLVM.LLVMAddInternalizePass(pass, 1);
            LLVM.LLVMAddMergeFunctionsPass(pass);
            LLVM.LLVMAddMergeFunctionsPass(pass);
            LLVM.LLVMAddFunctionInliningPass(pass);
            LLVM.LLVMAddAggressiveDCEPass(pass);
            LLVM.LLVMAddGlobalDCEPass(pass);
            LLVM.LLVMAddArgumentPromotionPass(pass);
            LLVM.LLVMAddAlwaysInlinerPass(pass);
            LLVM.LLVMAddPruneEHPass(pass);
            LLVM.LLVMPassManagerBuilderPopulateModulePassManager(builderRef, pass);
            root.hyperAggressiveOptimizer(false, pass);
            LLVM.LLVMAddStripSymbolsPass(pass);
            LLVM.LLVMAddLowerExpectIntrinsicPass(pass);
            LLVM.LLVMAddLowerConstantIntrinsicsPass(pass);```when your one and only purpose is to optimize the program
#

Exception in thread "main" java.lang.ClassCastException: class tfc.ralux.compiler.parse.RaluxParser$DOperandContext cannot be cast to class tfc.ralux.compiler.parse.RaluxParser$OperandContext (tfc.ralux.compiler.parse.RaluxParser$DOperandContext and tfc.ralux.compiler.parse.RaluxParser$OperandContext are in unnamed module of loader 'app')

th is a DOperandContext

elder pondBOT
#

Lyzantra

930203537776996412.png
deft kestrel
#

ohh
++

#

well one of the two is faster

image.png
#

KEKW not sure which

image.png
#

module 2 looks to be flickering faster actually

#

... I can't tell KEKW

#

actually I think module is the faster one

deft kestrel
#

🤷 it's too early for me to be caring this much about optimization

deft kestrel
#

ever wondered what a codebase looks like as a graph?

image.png
deft kestrel
# lethal axle you don't need to go out of your way for me on this project, I am unlikely user ...

The problem I’m trying to solve is
one that not many people care about, since in most cases just use a VM lang or just use a high level lang is an option
and it, honestly is an option for me to, but I want to be free to mess around with bindings without writing this monolithic C or C++ codebase to go along with my java project (kinda, defeating the purpose of java) and I want to be able to Just Use IntelliJ IDEA because that’s what I’m used to
So I’m basically just trying to get this to work for me
if it becomes something people use, then it becomes something people use 🤷
I’d love to see that happen, but I don’t expect it
also I don’t, entirely want to learn native interop with C#, it irrationally scares me for some reason KEKW

also I wanna learn compiler infrastructure stuff, and making a compiler is the best way to do that frogapprove

#

C# is a great language
unfortunately I’m an intellij idealist bleaker_kekw

#

also learned something about java today that now bothers me
protected is package OR subclasses, not just subclasses

#

Also I’m kinda curious how far LLVM would optimize a software raster that is entirely inlined to be a single function

lethal axle
#

you can write C++ and C in intellij

#

all the features of C Lion work in Intellij ultimate I would expect

deft kestrel
#

… well, I don’t have ultimate

lethal axle
#

you can probably get it as a student

#

Get free access to all developer tools from JetBrains!

deft kestrel
#

oh I think they’re also doing a thing recently where more of their IDEs are free, but you have to have data collection on if you use it for free

#

I’ll check that out when on my laptop

deft kestrel
#

ok, rider is downloading

#

long_circuit: ; preds = %entry
br label %tree_exit15
why is it so hard to get llvm to optimize out these branches that literally do nothing other than just jump to another branch

#

... regardless; next thing is to implement the ability to compile multiple files and import stuff

#

I don't think I even remember ever having this process occur with IDEA

image.png
#

I mean it must've but like
I have no memory of it

#

rider installed

#

rider installed

#

I will need to register an account KEKW

#

lld-link: error: undefined symbol: __fixtfsi
I love having no idea what library provides what

#

according to chat gpt it comes from rt builtins
I'm already linking against rt builtins

#

wait what

image.png
deft kestrel
lethal axle
#

Weird

deft kestrel
# 
    private T[] ts;
#

I sure do love when the ide automatically inserts the private qualilfier

#

wat

#

so far the weirdest thing to me with C# is this

image.png
#

because this is valid in java, but in C# it has to be this

image.png
#

but apparently this is valid, that's neat

image.png
#

lld-link: error: could not open 'libcmt.lib': no such file or directory
bleaker_kekw updating lld did not help

#

bleaker_kekw it made it worse

#

maybe I should just make an issue report on llvm
'cuz I can't find anything at all on this and no matter what random files I link against it doesn't help and AI keeps telling me to go depend on MSVC (visual studios compiler, which would mean that people would have to install visual studios to use my language, which seems dumb)

#

and AI also tells me to use mingw, which like
I'm using llvm because mingw is kinda eh at existing on windows?
so like I don't really want to use it because it's kinda more effort to install than its worth?

#

and literally nothing I do brings me any closoer to having it work

# 
    private T[] ts = Array.Empty<T>();

... why does java not have this

#

(generic typed arrays are pain to initialize on java)

#

ok this is awesome
but I just can't stand having stuff abruptly get typed

image.png
#

my complaint with this is exactly the same as my complaint with github copiloit
I end up spending a bunch of time undoing what the ML assistant does

... turns out that's not what's doing it, ech

#

well my quickly written list class works 🤷

image.png image.png
deft kestrel
#

maybe that's worth the effort at this point, idk
probably after I have ralux up and running though, so I can write the lib in ralux, since ON-5 doesn't trigger any of these so far, and I don't think ON-4 does either

#

oh nope, O4 can still cause it :yipee:

lapis turtle
deft kestrel
#

the compiler's written in java, there's no chance I run into this problem building the compiler

lapis turtle
#

I believe llvm does that

deft kestrel
# 
           LLVM.LLVMAddInternalizePass(pass, 1);
            LLVM.LLVMPassManagerBuilderPopulateModulePassManager(builderRef, pass);```this seems to be the minimum pass set for eliminating them
#

and I've tried linking against like every single lib I could find from both llvm and windows kit

deft kestrel
lapis turtle
#

also libcmt is the c runtime, so prolly important

deft kestrel
deft kestrel
lapis turtle
lapis turtle
deft kestrel
#

I have no idea how to dynamic link

deft kestrel
lapis turtle
deft kestrel
#

(I've gone over stuff like that like a million times with chatgpt already)

lapis turtle
deft kestrel
#

using clang does stuff like this

image.png
lapis turtle
deft kestrel
lapis turtle
deft kestrel
#

if anything it should be using uclib intrinsics, because I'm setting my environment as uclib

#

but it doesn't seem to care about that

lapis turtle
#

I would just *.obj

deft kestrel
#

I only have one obj file

image.png
lapis turtle
# 
lld-link test.obj /OUT:test.exe \
    pcre2-8.lib gc.lib iconv.lib \
    libcmt.lib libvcruntime.lib libucrt.lib kernel32.lib dbghelp.lib shell32.lib ole32.lib advapi32.lib \
    /ENTRY:wmainCRTStartup /LIBPATH:`crystal env CRYSTAL_LIBRARY_PATH`
deft kestrel
lapis turtle
#

not a copy paste

deft kestrel
# deft kestrel ```java LLVM.LLVMAddInternalizePass(pass, 1); LLVM.LLVMPa...

"lld-link.exe module.obj -entry:main /libpath:"C:/Program Files (x86)/Windows Kits/10/Lib/10.0.22621.0/ucrt/x64" ucrt.lib /libpath:"C:/Program Files/LLVM/lib/clang/8.0.1/lib/windows" clang_rt.builtins-x86_64.lib
yeah I uh
already have a command that works
problem is it only works if I use optimization level ON-5 or add these two passes because elsewise it fails to link against stuff that my code doesn't reference

lapis turtle
#

whats wrong with the two passes

# 
lld-link module.obj /OUT:module.exe \
    libcmt.lib    
/ENTRY:wmainCRTStartup /LIBPATH:`path to your libs`
deft kestrel
#

well considering I'm pretty sure llvm only inserts these calls to missing functions for sake of removing undefined behavior
I'm pretty sure those passes introduce undefined behavior
and even if not, it's still not O0, and can introduce behavior that doesn't necessarily line up considering this also requires LLVM level 1

also yes I know how to link against libcmt.lib, problem is I have no idea where it's located

lapis turtle
#

also libpath is necessary

deft kestrel
lapis turtle
#

sadly this isn't linux and nothings under one unfied directory

deft kestrel
#

wait no that's because I typed stuff wrong

lapis turtle
deft kestrel
#

nope lib path is entirely necessary

image.png
lapis turtle
#

you never provided a libpath for libcmt either

deft kestrel
#

yeah because I don't know what said libpath is, so how am I supposed to provide something that I don't know

#

only thing I have found about libcmt is that it's from msvc

lapis turtle
deft kestrel
#

C:\Program Files\Microsoft Visual Studio\2022\Community\VC\Tools\MSVC\14.37.32822\lib\x86\libcmt.lib
yeah
msvc
which I don't have a guarantee is on the user's computer

#

and is also not located where the internet told me it is

lapis turtle
#

again, if someone is installing and using your compiler, since msvc is a required dependency they must install msvc

deft kestrel
#

nor does it solve anything KEKW

image.png
deft kestrel
lapis turtle
deft kestrel
#

so I have no idea where the msvc dependency is coming from
but I don't want it

lapis turtle
deft kestrel
#

regardless; msvc libcmt doesn't solve anything either, so uh
🤷

lapis turtle
deft kestrel
#

seems I'm just gonna have to make my own runtime thingy

lapis turtle
deft kestrel
#

🤷 heck with msvc
I don't need it KEKW

lapis turtle
lapis turtle
deft kestrel
# lapis turtle https://compiler-rt.llvm.org/

builtins - a simple library that provides an implementation of the low-level target-specific hooks required by code generation and other runtime components. For example, when compiling for a 32-bit target, converting a double to a 64-bit unsigned integer is compiling into a runtime call to the "__fixunsdfdi" function. The builtins library provides optimized implementations of this and other low-level routines, either in target-independent C form, or as a heavily-optimized assembly.
... so, according to that, the thing that I'm already linking against should have already solved my issue but it hasn't

deft kestrel
# lapis turtle you do though

my stuff works without it exactly as well as with it
which is to say, it doesn't unless I do a dumb work around for the exact same issue with as I get without

#

so uh, no I don't think I do KEKW

lapis turtle
deft kestrel
#

well then what is the point

#

this ran exactly as intended here, though it's not using libcmt
so like, it sure doesn't seem to be explicitly required, and it also doesn't provide the function that's missing

image.png
#

I can set it up to auto link against msvc if it's present, but for the moment it serves no purpose if it's not gonna fix the issue I'm having when compiling without optimization passes, which is currently the only thing I care about

#

I mean I guess it's required for lld for 13.0.1, actually
but I'm not currently using that version of lld

#

and using that version doesn't solve anything either

deft kestrel
lapis turtle
deft kestrel
lapis turtle
lapis turtle
#

seems to have been implemented

deft kestrel
lapis turtle
#

strange

lapis turtle
deft kestrel
#

I'm also currently unable to use anything higher than llvm 13.0.1, since newer versions remove some things I currently need
but even then, what the heck
some of if not all of these things are generated by even 8.x

lapis turtle
deft kestrel
#

based on .lib, I'm inclined to assume static lib

lapis turtle
deft kestrel
#

I have no idea how I would check that then

lapis turtle
#

it doesn't seem to have an associated dll and builtins would make me think its a static lib

#

is this relevant to you at all?

image.png
#

https://discourse.llvm.org/t/improve-autolinking-of-compiler-rt-and-libc-on-windows-with-lld-link/71392/3

LLVM Discussion Forums
Improve autolinking of compiler-rt and libc++ on windows with lld-link

I think you wanted to say that you can ship a lld-link.bat and not cannot. We did consider this, but it would be very complicated bat script or changes to the clang driver in that case, considering how different the link.exe command line interface is from what clang is today. It’s possible, but it’s not a small effort.

deft kestrel
#

adding higher level directories doesn't help either

lapis turtle
#

whats wrong with the extra passes

deft kestrel
#

I want O0-0 to actually be no passes at all
just the raw compiler output, verbatim, exactly as is

lapis turtle
#

I would just post on the llvm forumns at this point

deft kestrel
#

also singleton datatypes seem like a neat idea```java
public singleton SingletonType {
private static Object someDataThatIsStoredHereButNeededElsewhere;

public SingletonType() {
}

public Object get() {
    return someDataThatIsStoredHereButNeededElsewhere;
}

}```

then in some file you do

private static final SingletonType accessor = new SingletonType();```

and if some other code snippet else tries that, well the program crashes
#

so just a data type that you can only ever initialize once

#

(this is the most common way I personally use singletons)

lapis turtle
deft kestrel
#

Sometimes internals need to access internals
and sometimes it’s nice to have those internals be in separate packages
this is an easy way to keep the internals protected while allowing them to be in separate packages

lapis turtle
deft kestrel
#

For those cases, having it just be one class file would be
… messy, at best

#

I also usually only do this with data loaders

lapis turtle
#

If I need to share values between two instances of a class

  1. make them the same class
    or
  2. just get a pointer to the value
deft kestrel
#

apparently, returning an integer from a void function segfaults the compiler

#

which, makes sense KEKW

#

so...

  • while
  • do while
  • else/else if
  • fields
  • assignability checking
  • fix short circuit eval
  • objects
  • gc
  • instance fields&methods
  • ralux specific syntax features
deft kestrel
#

shoot, llvm does not optimize my short circuit eval bleaker_kekw

#

Or at least not well

deft kestrel
#

I would imagine it would do better if I just kept reusing a single "dirt" variable instead of allocating one per short circuit branch

lethal axle
#

render the dirt as a dirt texture so we can see some pixels

lethal axle
#

btw #lang-dev might be interested in your stuff here

deft kestrel
#

@lethal axle tried claude today and
holy heck does claude know what it’s talking about

#

Based on LLMs from scratch, if asked, claude can actually generate something that looks relatively right for not just a general overview of an LLM architecture
Unlike chatgpt and llama, which struggle to even just do an overview of an llm model

lethal axle
#

claude is a better model in my experience

#

It doesn’t generate images though and doesn’t have other features chat-gpt has

deft kestrel
#

Also knows a good bit about apache mxnet it seems, including that it’s no longer mainted

lethal axle
#

I stopped paying for other models

deft kestrel
lethal axle
#

I agree

#

It can still be wildly inaccurate

#

It tries to agree to a fault

#

Anthropic calls this sycophantic behavior and they study it

deft kestrel
# 
                BlockBuilder builder = root.getBlockBuilding();
                LLVMValueRef valueRef = consumer.getDirtRef();
                BlockBuilder branchLong = consumer.getDirect().block("long_circuit");
                BlockBuilder branchJump = consumer.getDirect().block("jump_circuit");
                BlockBuilder branchShort = consumer.getDirect().block("short_circuit");
                builder.conditionalJump(left.llvm, branchLong, branchJump);

                branchJump.enableBuilding();
                root.setValue(valueRef, root.integer(0, 1));
                branchJump.jump(branchShort);

                branchLong.enableBuilding();
                Value right = compileExpr((RaluxParser.ExprContext) context.getChild(2));
                root.setValue(valueRef, right.llvm);
                root.getBlockBuilding().jump(branchShort);

                branchShort.enableBuilding();
                val = root.getValue(valueRef, "get_dirt");```well my short circuit eval was also broken anyway, so that's fixed now
deft kestrel
# 
        try((
            fis
        ) -> {
            // error prone code
        }).with(new FileInputStream("test.txt")).catch((IOException ioe) -> {
            ioe.printStackTrace();
        });```kinda like this for error handling syntax
#

System.out.println(new String(proc.getInputStream().readAllBytes().readAllBytes()));
good job intellij

#

(context: .readAllBytes() returns a byte[], which has no methods, so calling .readAllBytes() again is a compiler error)

#

(this is a sarcastic good job)

deft kestrel
#

as a human being
I get bored sometimes
lowLevelVirtualMachineParameterValueReferenceWrappedUsingAJavaObjectWhichHoldsALongWhichRepresentsTheMemoryAddressOfTheObject

image.png
lethal axle
#

average java variable name

deft kestrel
#

I think I'm gonna get started on RLXIR (intermediate between ralux and llvmir/other backends)

#

(I want to be able to have a system where people can just sorta drop a DLL into a folder and add a new compilation target or a new pass set or whatever)

# 
List<RlxCls> classes =  new ArrayList<>();

and because ralux is an object oriented lang, yeah I'm gonna make classes a proper thing in rlxir

deft kestrel
#

first try figuring out gradle subprojects!

image.png
deft kestrel
# 
    @Override
    public String toString() {
        return switch (typ) {
            case 'i' -> "int" + bits;
            case 'f' -> "float" + bits;
            case 'c' -> "char" + bits;
            case 'v' -> "void";
            case 'p' -> "pointer";
            default -> throw new RuntimeException("qhat");
        };
    }```I remember when switch statements were the most ugly things in the world with java
deft kestrel
# 
        ConstInstr<Integer> CONST_2 = new ConstInstr<>(2, RlxTypes.INT);
        ConstInstr<Integer> CONST_4 = new ConstInstr<>(4, RlxTypes.INT);

        MathInstr sum = function.sum(CONST_2, CONST_4);
        VarInstr var = function.makeVar(sum.valueType(), sum);```that's a lot simpler than LLVM IR
# 
        ConstInstr<Integer> CONST_2 = new ConstInstr<>(2, RlxTypes.INT);
        ConstInstr<Integer> CONST_4 = new ConstInstr<>(4, RlxTypes.INT);

        MathInstr sum = function.sum(CONST_2, CONST_4);
        VarInstr var = function.makeVar(sum.valueType(), sum);

        ValueInstr val = var.get();
        MathInstr sum1 = function.sum(val, CONST_4);
        var.set(sum1);```
deft kestrel
# 
    public void compile() {
        BlockBuilder block = builder.block("entry");
        block.enableBuilding();
        for (RlxInstr instr : instrs) {
            switch (instr.type()) {
                case CONST -> throw new RuntimeException("Consts shouldn't show up in functions");
                case MATH -> {
                    compileMath((MathInstr) instr);
                }
                case RETURN_VOID -> root.getBlockBuilding().ret();
                case RETURN_VALUE -> {
                    ReturnInstr ret = (ReturnInstr) instr;
                    ensureData(ret.valueInstr);
                    root.getBlockBuilding().ret(ret.valueInstr.getCompilerData());
                }
            }
        }
        if (!block.isTerminated()) block.ret();
    }```nice
#

already have support for:

  • addition
  • subtraction
  • division
  • multiplication
  • return
    from rlx ir->llvm ir compilation
    next is vars, then fp and type coercions
#

gonna also have to rewrite the whole ast -> ir compilation process to use rlx ir instead of llvm ir directly, but that should be... a good amount simplier than compiling to llvm ir directly

deft kestrel
#

ight, vars and rets are in

lapis turtle
#

why need intermediates for intermediates

deft kestrel
# lapis turtle I see a lot of beginner compiler developers do this, I dont see a point in the s...

I was sorta left unable to really properly do nearly as much analysis as I wanted with LLVM
my intermediate is just sorta a data structure with no file format (currently)

though I also want to be able to export libraries as a file that is not a machine binary but also not just the source code
🤷 I can do without the intermediate for the intermediate, but it just kinda simplifies stuff for me

#

I mean I probably could do the analysis with llvm but like
eh, it's just easier to have the extra step, lol

lapis turtle
deft kestrel
#

I kinda wanna do different backends at some point but like
eh I probably won't
but I am decently likely to do multiple frontends (largely due to build systems)

#

I wanna use a variant of LUA for my build system 🤷

lapis turtle
deft kestrel
lapis turtle
lethal axle
#

maybe it's just one

#

there's AIR and ZIR so I guess that's two

deft kestrel
# 
; ModuleID = 'module'
source_filename = "module"

define i32 @"TestClass#testFunction"() {
entry:
  %var = alloca i32, align 4
  %var1 = alloca i32, align 4
  store i32 8, i32* %var, align 4
  %get_var = load i32, i32* %var, align 4
  %si_sum = add i32 %get_var, 3
  store i32 %si_sum, i32* %var, align 4
  store i32 6, i32* %var1, align 4
  %get_var2 = load i32, i32* %var, align 4
  %si_sum3 = add i32 8, %get_var2
  store i32 %si_sum3, i32* %var1, align 4
  %get_var4 = load i32, i32* %var1, align 4
  %get_var5 = load i32, i32* %var, align 4
  %si_cmp = icmp sgt i32 %get_var4, %get_var5
  br i1 %si_cmp, label %test0, label %test1

test0:                                            ; preds = %entry
  %get_var6 = load i32, i32* %var1, align 4
  ret i32 %get_var6

test1:                                            ; preds = %entry
  %get_var7 = load i32, i32* %var, align 4
  ret i32 %get_var7
}```yipee!
deft kestrel
deft kestrel
lapis turtle
deft kestrel
#

I'm gearing rlxir to this whole object orientation centered design that languages like java and C# uses (where basically everything is an object) but in a way that it's easier to translate to LLVM than just the source code alone 🤷

lethal axle
deft kestrel
#

from a quick glance at a google image of zig code
zig looks to be a blend of a good amount of my favorite parts of a few languages I don't like
as well as some stuff that I would definitely have to learn to understand KEKW

image.png
#

also what is this?
jvm bytecode, where the type comes after the arguments?

image.png
lethal axle
#

Idk I don’t have any problems building anything I want to with zig gpAkkoShrug

deft kestrel
deft kestrel
#

does that tell zig to infer the size? or?

#

@lethal axle?

lapis turtle
lethal axle
#

Ok I guess it’s not for everyone, it’s working for me though idk gpAkkoShrug

lapis turtle
deft kestrel
lethal axle
#

Sorry I was afk

#

It’s pretty convenient

#

It’s a comptime syntax that determines the size of an array by the number of values in the literal

deft kestrel
#

alternatively; []u32{0};

lethal axle
#

That’s the slice syntax

#

[] is the zig version of **

#

Well sorta

#

It’s a pointer to a collection

#

You can get a slice from an array

azure zinc
azure zinc
#

ZIR and AIR serve different purposes. ZIR is untyped because a fully typed zig program requires comptime evaluation and AIR is a per-function IR used by codegen

deft kestrel
azure zinc
#

mhm iirc there's a proposal for that

#

although im not in favor of it as that might cause confusion between []u8 (the slice type) and []u8 (the array literal)

deft kestrel
#

ohh

lethal axle
#

I was trying to say that but I guess not as well

#

:P

deft kestrel
#

okay what the heck

image.png
#

there's like 2 elements that are responsive and the rest of the UI is frozen KEKW

#

how does this even occur KEKW

deft kestrel
#

type coercion time 🤷

lapis turtle
#

so now

#

make arguments about things you dont understand

#

You did that without any research into one link on my profile, I would say you should also consider this about yourself :^)

deft kestrel
#

Exception in thread "main" java.lang.RuntimeException: float32 cannot be set as a int32. Are you missing a cast?
at tfc.rlxir.instr.value.vars.VarInstr.set(VarInstr.java:23)
at ir_to_ir.Test.main(Test.java:35)
I-
:wha:

#

the math instruction should be the one crashing the compiler here thonk

#

if anything

#

I stand corrected

deft kestrel
# 
        ConstInstr<Integer> CONST_5 = new ConstInstr<>(5, RlxTypes.INT);
        ConstInstr<Integer> CONST_3 = new ConstInstr<>(3, RlxTypes.INT);
        ConstInstr<Float> CONST_6 = new ConstInstr<>(6.0f, RlxTypes.FLOAT);
        MathInstr sum = function.sum(CONST_5, CONST_3);

        VarInstr var = function.makeVar(sum);
        var.set(function.sum(var.get(), CONST_3));
        VarInstr var1 = function.makeVar(CONST_6);
        var1.set(function.autoCast(
                function.sum(sum, var1.get()),
                var1.type
        ));

        CompareInstr condition = function.compare(CompareOp.GT, var1.get(), var.get());

        RlxBlock blockA = function.makeBlock("test0");
        RlxBlock blockB = function.makeBlock("test1");
        function.jumpIf(condition, blockA, blockB);
        function.ret(var);
        function.buildBlock(blockA);
        function.ret(var1.get());```
#

%cast_i_fp_float16_to_float32 = sitofp half %get_var2 to float
incredible

#

KEKW something broke there

#

yeah that should not be picked up on as an INT_FLOAT cast

image.png
#

cause: I was assuming the left hand coercion cast would always be the same as the right hand coercion cast
which is almost if not actually exclusively false KEKW

#

I think exclusively false for how my logic was

deft kestrel
#

I'm not gonna bother making my IR readable to a human when dumped to a file
I only need it to be readable to the compiler 🤷
I can even just do a binary data representation and not bother beyond that truthfully, but I like being able to see it as a debug string

it's the code api for working with the ir that needs to be comprehensible

#

TestClass {
func 3s int32 main() {
entry: # 19 instructions
# TODO: list instr args (as indices)
cd int32 math
define_var # var
d set_var # var
d int32 get_var # var
d int32 math
d set_var # var
define_var # var1
d set_var # var1
d float16 get_var # var1
cd float32 cast
d float32 cast
d float32 math
d float16 cast
d set_var # var1
d float16 get_var # var1
d int32 get_var # var
d int32 cast
d int1 comparison
d cond_jump
test0: # 3 instructions
# TODO: list instr args (as indices)
d float16 get_var # var1
d int32 cast
d return_value
test1: # 2 instructions
# TODO: list instr args (as indices)
d int32 get_var # var
d return_value
}
}
like I'mma just refer to things with the instruction index for instr args, 'cuz while that's really annoying for a human to read, it's really easy to parse that way

#

(3s stands for package-protected, static)

azure zinc
lapis turtle
lapis turtle
#

my main gripe is the typing system it just feels like things are in the wrong order

#

const* makes it feel like the const is what we're taking a pointer of, which doesn't make sense

#

lemme get the whole rant I had

#

*const [N]<TYPE>

#

oh

#

god

#

pointer to constant N sized type array

#

?

azure zinc
#

correct

#

i'll brb in 1h

azure zinc
#

At the end of the day, just like every other language it's all about the creator's taste. There are people who like it and people who don't.

deft kestrel
#

(of note: I specified JVM and not any language for a reason)