Mirror of forum.nim-lang.org

480 :: LLVM Backend

[2014-06-22T05:38:04+02:00]

sirinath (orginal) [2014-06-22T05:38:04+02:00] view original

Hi,

Any plans to by pass C++ and generate native code?

Suminda

Araq (orginal) [2014-06-23T00:26:17+02:00] view original
We have a native backend in the works but it's not based on LLVM. An LLVM codegen is currently not on our agenda.

Since many of you asked for a list of advantages/disadvantages I found my old notes about this issue. Here they are:

Advantages of C:
- easier bootstrap; especially because the library depends on C's structures and constants which may change
- the compiler will support embedded targets easily
- interfacing with C is easier
- code generator for it already written
- works better on Windows than LLVM
Advantages of LLVM:
- accurate debug information
- atomic instructions; transactional memory project is available
- instructions for overflow detection
- hopefully compilation times will be better
- only one assembler format to support
- we get code generators for C and CLI for free
- in the long run better exception support
- generated code is checked deeply; assurance that the generated code does not depend on implicit type conversions
- support for merging of generic procedures
Disadvantages of C:
- chars are flawed horribly
- cannot return arrays or sets; arrays are fucked up
- generating correct C types is hard, but the algorithm works now
- exceptions are inefficient to support (move to C++?)
Disadvantages of LLVM:
- code generation is harder than C
- generated code is even harder to read than C
- code generation for exceptions is hard
- bootstrapping is MUCH harder
- assembler not integrated
- code generation for coroutines is not any easier than in C

brianrogoff (orginal) [2014-06-24T16:37:08+02:00] view original

An LLVM backend might make a really good GSOC 2015 project.

It would be especially interesting if Nimrod on LLVM performed better than, say Swift or Rust on a few silly microbenchmarks. Not really meaningful, mind you, just good reddit-bait.

LLVM receives a lot of attention and people keep trying to make it a better back end, so it's certainly interesting, but I'd like to see Nimrod reach a 1.0 release sooner and a new backend would likely not help that.