codon/docs
A. R. Shajii ebd344f894
GPU and other updates (#52)
* Add nvptx pass

* Fix spaces

* Don't change name

* Add runtime support

* Add init call

* Add more runtime functions

* Add launch function

* Add intrinsics

* Fix codegen

* Run GPU pass between general opt passes

* Set data layout

* Create context

* Link libdevice

* Add function remapping

* Fix linkage

* Fix libdevice link

* Fix linking

* Fix personality

* Fix linking

* Fix linking

* Fix linking

* Add internalize pass

* Add more math conversions

* Add more re-mappings

* Fix conversions

* Fix __str__

* Add decorator attribute for any decorator

* Update kernel decorator

* Fix kernel decorator

* Fix kernel decorator

* Fix kernel decorator

* Fix kernel decorator

* Remove old decorator

* Fix pointer calc

* Fix fill-in codegen

* Fix linkage

* Add comment

* Update list conversion

* Add more conversions

* Add dict and set conversions

* Add float32 type to IR/LLVM

* Add float32

* Add float32 stdlib

* Keep required global values in PTX module

* Fix PTX module pruning

* Fix malloc

* Set will-return

* Fix name cleanup

* Fix access

* Fix name cleanup

* Fix function renaming

* Update dimension API

* Fix args

* Clean up API

* Move GPU transformations to end of opt pipeline

* Fix alloc replacements

* Fix naming

* Target PTX 4.2

* Fix global renaming

* Fix early return in static blocks; Add __realized__ function

* Format

* Add __llvm_name__ for functions

* Add vector type to IR

* SIMD support [wip]

* Update kernel naming

* Fix early returns; Fix SIMD calls

* Fix kernel naming

* Fix IR matcher

* Remove module print

* Update realloc

* Add overloads for 32-bit float math ops

* Add gpu.Pointer type for working with raw pointers

* Add float32 conversion

* Add to_gpu and from_gpu

* clang-format

* Add f32 reduction support to OpenMP

* Fix automatic GPU class conversions

* Fix conversion functions

* Fix conversions

* Rename self

* Fix tuple conversion

* Fix conversions

* Fix conversions

* Update PTX filename

* Fix filename

* Add raw function

* Add GPU docs

* Allow nested object conversions

* Add tests (WIP)

* Update SIMD

* Add staticrange and statictuple loop support

* SIMD updates

* Add new Vec constructors

* Fix UInt conversion

* Fix size-0 allocs

* Add more tests

* Add matmul test

* Rename gpu test file

* Add more tests

* Add alloc cache

* Fix object_to_gpu

* Fix frees

* Fix str conversion

* Fix set conversion

* Fix conversions

* Fix class conversion

* Fix str conversion

* Fix byte conversion

* Fix list conversion

* Fix pointer conversions

* Fix conversions

* Fix conversions

* Update tests

* Fix conversions

* Fix tuple conversion

* Fix tuple conversion

* Fix auto conversions

* Fix conversion

* Fix magics

* Update tests

* Support GPU in JIT mode

* Fix GPU+JIT

* Fix kernel filename in JIT mode

* Add __static_print__; Add earlyDefines; Various domination bugfixes; SimplifyContext RAII base handling

* Fix global static handling

* Fix float32 tests

* FIx gpu module

* Support OpenMP "collapse" option

* Add more collapse tests

* Capture generics and statics

* TraitVar handling

* Python exceptions / isinstance [wip; no_ci]

* clang-format

* Add list comparison operators

* Support empty raise in IR

* Add dict 'or' operator

* Fix repr

* Add copy module

* Fix spacing

* Use sm_30

* Python exceptions

* TypeTrait support; Fix defaultDict

* Fix earlyDefines

* Add defaultdict

* clang-format

* Fix invalid canonicalizations

* Fix empty raise

* Fix copyright

* Add Python numerics option

* Support py-numerics in math module

* Update docs

* Add static Python division / modulus

* Add static py numerics tests

* Fix staticrange/tuple; Add KwTuple.__getitem__

* clang-format

* Add gpu parameter to par

* Fix globals

* Don't init loop vars on loop collapse

* Add par-gpu tests

* Update gpu docs

* Fix isinstance check

* Remove invalid test

* Add -libdevice to set custom path [skip ci]

* Add release notes; bump version [skip ci]

* Add libdevice docs [skip ci]

Co-authored-by: Ibrahim Numanagić <ibrahimpasa@gmail.com>
2022-09-15 15:40:00 -04:00
..
advanced GPU and other updates (#52) 2022-09-15 15:40:00 -04:00
img Update docs (#28) 2022-07-26 16:08:42 -04:00
interop Add Cython fixes (decorator arguments, pyvars support) (#42) 2022-08-03 16:36:03 -04:00
intro GPU and other updates (#52) 2022-09-15 15:40:00 -04:00
language GPU and other updates (#52) 2022-09-15 15:40:00 -04:00
README.md Update docs (#28) 2022-07-26 16:08:42 -04:00
SUMMARY.md GPU and other updates (#52) 2022-09-15 15:40:00 -04:00
docgen.py Update docs (#28) 2022-07-26 16:08:42 -04:00

README.md

Codon Pipeline

Codon is a high-performance Python compiler that compiles Python code to native machine code without any runtime overhead. Typical speedups over Python are on the order of 100x or more, on a single thread. Codon supports native multithreading which can lead to speedups many times higher still.

The Codon framework is fully modular and extensible, allowing for the seamless integration of new modules, compiler optimizations, domain-specific languages and so on. We actively develop Codon extensions for a number of domains such as bioinformatics and quantitative finance.

Codon at a glance

A simple Python program fib.py...

from time import time

def fib(n):
    return n if n < 2 else fib(n - 1) + fib(n - 2)

t0 = time()
ans = fib(40)
t1 = time()
print(f'Computed fib(40) = {ans} in {t1 - t0} seconds.')

... run through Python and Codon:

$ python3 fib.py
Computed fib(40) = 102334155 in 17.979357957839966 seconds.
$ codon run -release fib.py
Computed fib(40) = 102334155 in 0.275645 seconds.