jit-compilation

Name: jit-compilation
Author: mohitmishra786

63installs19forks

How to install jit-compilation?

mohitmishra786/low-level-dev-skills/jit-compilation

$npx -y skills add mohitmishra786/low-level-dev-skills --skill jit-compilation

Installs into the current project.

Summary

JIT compilation skill for runtime code generation. Use when building LLVM ORC JIT, LLJIT, Cranelift JIT, inline caches, trampolines, or Rust dynasm codegen. Activates on queries about ORC JIT, LLJIT, Cranelift, ExecutionSession, inline cache, W^X, or dynasm.

Files · 1

View on GitHub

SKILL.md

1	# JIT Compilation
2
3	## Purpose
4
5	Guide agents through just-in-time compilation: LLVM ORC JIT v2 (`ExecutionSession`, `IRLayer`, `ObjectLayer`), LLJIT for simpler use cases, Cranelift JIT, inline caches for dynamic dispatch, trampolines for lazy compilation, security considerations (W^X, code signing), and Rust `dynasm` for x86 codegen.
6
7	## When to Use
8
9	- Building an interpreter with a JIT hot-path
10	- Implementing lazy function compilation on first call
11	- Embedding dynamic code generation in a REPL or game engine
12	- Prototyping codegen without writing a full linker
13	- Creating inline caches for polymorphic call sites
14	- Generating x86 machine code from Rust with `dynasm`
15
16	## Workflow
17
18	### 1. JIT architecture overview
19
20	```
21	Source/AST/Bytecode
22	→ IR (LLVM IR or Cranelift CLIF)
23	→ Object file (in memory)
24	→ Runtime linker (RTDyldObjectLinkingLayer)
25	→ Executable code in R+X memory
26	→ Function pointer call
27	```
28
29	### 2. LLJIT — simplest LLVM JIT
30
31	```cpp
32	#include "llvm/ExecutionEngine/Orc/LLJIT.h"
33	#include "llvm/IR/LLVMContext.h"
34	#include "llvm/IR/Module.h"
35	#include "llvm/IR/IRBuilder.h"
36
37	using namespace llvm;
38	using namespace llvm::orc;
39
40	int main() {
41	auto JIT = cantFail(LLJITBuilder().create());
42
43	LLVMContext Context;
44	auto M = std::make_unique<Module>("jit", Context);
45	IRBuilder<> Builder(Context);
46
47	// int add(int a, int b) { return a + b; }
48	Function *AddFn = Function::Create(
49	FunctionType::get(Builder.getInt32Ty(),
50	{Builder.getInt32Ty(), Builder.getInt32Ty()}, false),
51	Function::ExternalLinkage, "add", M.get());
52
53	BasicBlock *BB = BasicBlock::Create(Context, "entry", AddFn);
54	Builder.SetInsertPoint(BB);
55	auto Args = AddFn->arg_begin();
56	Value *Sum = Builder.CreateAdd(Args, Args + 1);
57	Builder.CreateRet(Sum);
58
59	cantFail(JIT->addIRModule(ThreadSafeModule(std::move(M), std::make_unique<LLVMContext>())));
60
61	auto AddSym = JIT->lookup("add");
62	auto AddPtr = (int ()(int, int))AddSym->getValue();
63	int result = AddPtr(3, 4); // 7
64	return 0;
65	}
66	```
67
68	```bash
69	clang++ -std=c++17 jit.cpp $(llvm-config --cxxflags --ldflags --libs core orcjit native) -o jit
70	```
71
72	### 3. ORC JIT v2 layers
73
74	```cpp
75	ExecutionSession ES;
76	auto &MainJD = ES.createBareJITDylib("main");
77
78	RTDyldObjectLinkingLayer ObjectLayer(
79	ES, []() { return std::make_unique<SectionMemoryManager>(); });
80
81	IRCompileLayer CompileLayer(
82	ES, ObjectLayer, std::make_unique<TargetMachineBuilder>());
83
84	// Add IR module to JITDylib
85	ThreadSafeModule TSM = ...;
86	CompileLayer.add(MainJD, std::move(TSM));
87
88	// Resolve symbol
89	auto Sym = ES.lookup({&MainJD}, "my_func");
90	```
91
92	Layers:
93	- IRLayer — compiles LLVM IR to object
94	- ObjectLayer — links relocatable objects
95	- ExecutionSession — symbol lookup and JITDylib management
96
97	### 4. Lazy compilation with trampolines
98
99	```
100	First call → trampoline → compile function → patch trampoline → direct call
101	```
102
103	```cpp
104	// Simplified lazy compile on first invocation
105	void *LazyCompile(const std::string &Name) {
106	if (!Compiled.count(Name)) {
107	auto Fn = CompileFromAST(Name);
108	Compiled[Name] = Fn;
109	// Patch call site or update function pointer table
110	}
111	return Compiled[Name];
112	}
113	```
114
115	ORC supports lazy reexports and lazy compilation via `LazyCallThroughManager`.
116
117	### 5. Inline caches
118
119	```javascript
120	// Concept: monomorphic call site caches resolved target
121	// Pseudocode for dynamic language
122	function call_site(obj, method, args) {
123	if (obj.class_id === cache.class_id) {
124	return cache.fn_ptr(args); // fast path
125	}
126	cache.class_id = obj.class_id;
127	cache.fn_ptr = resolve_method(obj, method);
128	return cache.fn_ptr(args);
129	}
130	```
131
132	JIT generates specialized code per cached type; deoptimize on cache miss.
133
134	### 6. Cranelift JIT
135
136	```rust
137	use cranelift::prelude::*;
138	use cranelift_jit::{JITBuilder, JITModule};
139	use cranelift_module::{Linkage, Module};
140
141	let isa = cranelift_native::builder().finish(settings::Flags::new(settings::builder()))?;
142	let jit_builder = JITBuilder::with_isa(isa, cranelift_module::default_libcall_names());
143	let mut module = JITModule::new(jit_builder);
144
145	let mut ctx = module.make_context();
146	ctx.func = Function::with_name_signature(
147	module.declare_function("add", Linkage::Export, &sig)?,
148	sig,
149	);
150	// ... build IR in ctx.func ...
151	module.define_function(func_id, &mut ctx)?;
152	module.finalize_definitions()?;
153	let code = module.get_finalized_function(func_id);
154	let add_fn: fn(i32, i32) -> i32 = unsafe { std::mem::transmute(code) };
155	```
156
157	Cranelift: faster compile times than LLVM, good for embeddable JITs.
158
159	### 7. Security — W^X policy
160
161	```
162	W^X (Write XOR Execute)
163	├── Memory page is writable OR exe

People who liked this also viewed

6 picks

Type

jit-compilation

63installs19forks

View on GitHub

How to install jit-compilation?

mohitmishra786/low-level-dev-skills/jit-compilation

$npx -y skills add mohitmishra786/low-level-dev-skills --skill jit-compilation

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# JIT Compilation
2
3	## Purpose
4
5	Guide agents through just-in-time compilation: LLVM ORC JIT v2 (`ExecutionSession`, `IRLayer`, `ObjectLayer`), LLJIT for simpler use cases, Cranelift JIT, inline caches for dynamic dispatch, trampolines for lazy compilation, security considerations (W^X, code signing), and Rust `dynasm` for x86 codegen.
6
7	## When to Use
8
9	- Building an interpreter with a JIT hot-path
10	- Implementing lazy function compilation on first call
11	- Embedding dynamic code generation in a REPL or game engine
12	- Prototyping codegen without writing a full linker
13	- Creating inline caches for polymorphic call sites
14	- Generating x86 machine code from Rust with `dynasm`
15
16	## Workflow
17
18	### 1. JIT architecture overview
19
20	```
21	Source/AST/Bytecode
22	→ IR (LLVM IR or Cranelift CLIF)
23	→ Object file (in memory)
24	→ Runtime linker (RTDyldObjectLinkingLayer)
25	→ Executable code in R+X memory
26	→ Function pointer call
27	```
28
29	### 2. LLJIT — simplest LLVM JIT
30
31	```cpp
32	#include "llvm/ExecutionEngine/Orc/LLJIT.h"
33	#include "llvm/IR/LLVMContext.h"
34	#include "llvm/IR/Module.h"
35	#include "llvm/IR/IRBuilder.h"
36
37	using namespace llvm;
38	using namespace llvm::orc;
39
40	int main() {
41	auto JIT = cantFail(LLJITBuilder().create());
42
43	LLVMContext Context;
44	auto M = std::make_unique<Module>("jit", Context);
45	IRBuilder<> Builder(Context);
46
47	// int add(int a, int b) { return a + b; }
48	Function *AddFn = Function::Create(
49	FunctionType::get(Builder.getInt32Ty(),
50	{Builder.getInt32Ty(), Builder.getInt32Ty()}, false),
51	Function::ExternalLinkage, "add", M.get());
52
53	BasicBlock *BB = BasicBlock::Create(Context, "entry", AddFn);
54	Builder.SetInsertPoint(BB);
55	auto Args = AddFn->arg_begin();
56	Value *Sum = Builder.CreateAdd(Args, Args + 1);
57	Builder.CreateRet(Sum);
58
59	cantFail(JIT->addIRModule(ThreadSafeModule(std::move(M), std::make_unique<LLVMContext>())));
60
61	auto AddSym = JIT->lookup("add");
62	auto AddPtr = (int ()(int, int))AddSym->getValue();
63	int result = AddPtr(3, 4); // 7
64	return 0;
65	}
66	```
67
68	```bash
69	clang++ -std=c++17 jit.cpp $(llvm-config --cxxflags --ldflags --libs core orcjit native) -o jit
70	```
71
72	### 3. ORC JIT v2 layers
73
74	```cpp
75	ExecutionSession ES;
76	auto &MainJD = ES.createBareJITDylib("main");
77
78	RTDyldObjectLinkingLayer ObjectLayer(
79	ES, []() { return std::make_unique<SectionMemoryManager>(); });
80
81	IRCompileLayer CompileLayer(
82	ES, ObjectLayer, std::make_unique<TargetMachineBuilder>());
83
84	// Add IR module to JITDylib
85	ThreadSafeModule TSM = ...;
86	CompileLayer.add(MainJD, std::move(TSM));
87
88	// Resolve symbol
89	auto Sym = ES.lookup({&MainJD}, "my_func");
90	```
91
92	Layers:
93	- IRLayer — compiles LLVM IR to object
94	- ObjectLayer — links relocatable objects
95	- ExecutionSession — symbol lookup and JITDylib management
96
97	### 4. Lazy compilation with trampolines
98
99	```
100	First call → trampoline → compile function → patch trampoline → direct call
101	```
102
103	```cpp
104	// Simplified lazy compile on first invocation
105	void *LazyCompile(const std::string &Name) {
106	if (!Compiled.count(Name)) {
107	auto Fn = CompileFromAST(Name);
108	Compiled[Name] = Fn;
109	// Patch call site or update function pointer table
110	}
111	return Compiled[Name];
112	}
113	```
114
115	ORC supports lazy reexports and lazy compilation via `LazyCallThroughManager`.
116
117	### 5. Inline caches
118
119	```javascript
120	// Concept: monomorphic call site caches resolved target
121	// Pseudocode for dynamic language
122	function call_site(obj, method, args) {
123	if (obj.class_id === cache.class_id) {
124	return cache.fn_ptr(args); // fast path
125	}
126	cache.class_id = obj.class_id;
127	cache.fn_ptr = resolve_method(obj, method);
128	return cache.fn_ptr(args);
129	}
130	```
131
132	JIT generates specialized code per cached type; deoptimize on cache miss.
133
134	### 6. Cranelift JIT
135
136	```rust
137	use cranelift::prelude::*;
138	use cranelift_jit::{JITBuilder, JITModule};
139	use cranelift_module::{Linkage, Module};
140
141	let isa = cranelift_native::builder().finish(settings::Flags::new(settings::builder()))?;
142	let jit_builder = JITBuilder::with_isa(isa, cranelift_module::default_libcall_names());
143	let mut module = JITModule::new(jit_builder);
144
145	let mut ctx = module.make_context();
146	ctx.func = Function::with_name_signature(
147	module.declare_function("add", Linkage::Export, &sig)?,
148	sig,
149	);
150	// ... build IR in ctx.func ...
151	module.define_function(func_id, &mut ctx)?;
152	module.finalize_definitions()?;
153	let code = module.get_finalized_function(func_id);
154	let add_fn: fn(i32, i32) -> i32 = unsafe { std::mem::transmute(code) };
155	```
156
157	Cranelift: faster compile times than LLVM, good for embeddable JITs.
158
159	### 7. Security — W^X policy
160
161	```
162	W^X (Write XOR Execute)
163	├── Memory page is writable OR exe

People who liked this also viewed

6 picks

Type