…jl into gpucompiler-basic

* Basic rewrite using @jpsamaroo's generate_shlib_fptr.

* update, remove junk, fix tests, use `gcc` for linking

* remove windows from CI

* note that this doesn't work on windows

* fix `compile`, add `native_code_llvm`, `native_code_typed`

* clarify fib problem, add LoopVectorization test :)

* add tests for stack allocated mutable arrays

* test on version 1.7

* rename `compile` to `native_llvm_module`

Co-authored-by: Jorge Alberto Vieyra Salas <[email protected]>
Co-authored-by: Mason Protter <[email protected]>

reloactability problems

…cases

Use .bc extension for LLVM bitcode, some docs

New `compile` interface

* Allow changeing filename base in generate_shlib, though leave obj as default

* Add `generate_executable`

* Add `compile_executable`

* Update docstring for compile_executable

* Add tests for compile_executable

* llvmcall used in standalone test only works on Julia 1.8+

* Looks like clang_jll ain't gonna cut it for the standalone executables even on linux

* Try annotating compiled function with Base.@ccallable

* Let's try some inlining

* Ah, so `@ccallable` requires a return type

* Ok, so no `@ccallable`

* Potential workaround on non-apple systems: use minimal wrapper

* We can probably use clang_jll after all

* Probably about time for a version bump

* Add more minimal test for v1.7, fix specifier on advanced test

* Implement changes requested in code review

* Patch and relocate points

* comment out erroneous `@show`

* doc update and `filename` support in `compile`

* Update src/StaticCompiler.jl

Co-authored-by: Julian Samaroo <[email protected]>

* Update README.md

Co-authored-by: Julian Samaroo <[email protected]>

* Update src/StaticCompiler.jl

Co-authored-by: Julian Samaroo <[email protected]>

* Update src/StaticCompiler.jl

Co-authored-by: Julian Samaroo <[email protected]>

* Update src/StaticCompiler.jl

Co-authored-by: Julian Samaroo <[email protected]>

* Update src/StaticCompiler.jl

Co-authored-by: Julian Samaroo <[email protected]>

* Update src/pointer_patching.jl

Co-authored-by: Julian Samaroo <[email protected]>

* Apply suggestions from code review

Co-authored-by: Julian Samaroo <[email protected]>

* reenable blas test

* go back to serialization stdlib

* note on gvs

* try using JITTargetMachine for the dynamic linker

* try running CI on windows

* do some code reorganization

* refine warning

* allow windows failures

* add options to strip the llvm / asm code

* remove `@showln`

* add `pointer_patching_diff` function to quickly see the effect of `relocation_table!`

* improve pointer relocation

* consolidate logic and reduce segfaults during compilation

* remove unnecessary stuff

* don't duplicate global values, switch from Dict to IdDict for reloc

* avoid re-creation of global variables

* more tests

* add dummy test for error handing

* test with ErrorTypes.jl

* bump version

Co-authored-by: Julian Samaroo <[email protected]>

* tell llvm the correct types of objects; verify the IR after after modification

* Update src/code_loading.jl

Co-authored-by: Valentin Churavy <[email protected]>

* use julia's optimizer, enable tests for error handling

* make ManualMemory and StrideArraysCore available on other process

* run optimization twice so that reloc can happen on optimized code

* grab `jl_add_optimization_passes` from `libjulia-codegen` on newer versions

* don't test on windows for now, it's pointless

* remove unnecessary func

* add some comments

* reorganize

* catch an error pathway

Co-authored-by: Valentin Churavy <[email protected]>

* Use Enzyme's optimization and late lowering mechanisms

* BLAS calls work again

* update README

* add another test case

* re-enable the `@test_skip` on remotely calling BLAS

* Clean up `generate_executable`

* Delete unused line

* Rename generate_shlib to generate_obj

* Add generate_dylib

* Test generate_dylib

* Add and test `compile_dylib`

* s/_dylib/_shlib/g to keep original terminology

* Remove JLD2 from Project.toml

* Delete Manifest.toml

Did I miss anyone?

Update LICENSE

This should fix the breakages that happened on the MacOS master builds

Bump LLVM version

add v1.8 testing

Fix typo

… to modify these to avoid problems with argv pointers in the future

Add more integration tests, reorganize CI, update to GPUCompiler 0.16

to register new version

Bump version to 0.4.3

Multiple function shared lib

Add `cflags` kwarg to `compile_executable` and friends