9 ms·

> LLVM is smaller at 1.6 million lines, but is newer, supports only C and C++ by default Unless I'm misunderstanding "by default", this isn't true. LLVM is lan

> LLVM is smaller at 1.6 million lines, but is newer, supports only C and C++ by default Unless I'm misunderstanding "by default", this isn't true. LLVM is language agnostic and supports many languages - even the primary C-like frontend Clang supports more than just C/C++.

webkike 9y ago
You are correct. The article is misinformed
pebers 9y ago
Perhaps they mean that within those 1.6 million lines, the only frontend is Clang? But the line count for gcc includes e.g. gfortran as well?
- valarauca1 9y ago
  The GCC addons (gcc-fortran, gcc-cobal, and gcc-go) just convert the existing language into C that is just ingested into the standard GCC. It was only around 2008 that gcc-c++ stopped doing the internal translation/name mangling dance and gained native first party support [1]. Much like how the older versions of GCC used to just produce ASM files to be ingested into GAS to create the binary objects. [1] https://lwn.net/Articles/542457/ https://lwn.net/Articles/542457/
  - lgg 9y ago
    The link you posted has nothing to do with how GCC does codegen. It is about GCC transitioning itself from being written in C to being written in C++.
  - mhh__ 9y ago
    GCC IR is not C
a_humean 9y ago
Yeah, I thought that part of the point of the LLVM project was that front end language compilers (entirely language agnostic) could target non-platform specific LLVM IR code, and then let the backend LLVM compiler compile to multiple backend targets. It simplifies front-end compiler development by having a single IR target over multiple native target.
wolfgke 9y ago
> LLVM is language agnostic There exists no such thing as "language agnostic". Every intermediate representation contains properties (very often implicitly) that are specific at least to a class of programming languages.
- whatshisface 9y ago
  If the intermediate representation is no more language-specific than x86, then it isn't really the compiler that's language-specific: it's modern computer architecture.
  - FeepingCreature 9y ago
    LLVM IR is fairly C specific. For instance, you better not be trying to write a language where null pointer access is defined...
    - nicwilson 9y ago
      We compile D with LLVM. Null pointer access is defined to crash, unlike C/C++ where it is undefined.
      - pcwalton 9y ago
        Are you adding explicit checks (possibly using fault maps to optimize them)? Because if not, you're depending on undefined behavior.
        nicwilson 9y ago
        That is only necessary for single objects that are bigger than one page which, we dont currently do (but should). Arrays are no problem since we check the bounds, only raw pointers cause problems.
        pcwalton 9y ago
pipio21 9y ago
We have worked with LLVM a lot for own own compilers and DSPs. LLVM Intermediate representation in practice is C++, AST is C++. LLVM do not support many languages,many languages support LLVM, which is different. That is many people, or languages' designers, have made the necessary work to make their languages compatible with LLVM IR. If the language is similar to C++, this is a simple task, if it is very different, it is very hard. LLVM does not support other languages, in fact, codebase changes a lot making maintenance painful.
- Arnt 9y ago
  Could you elaborate on the C++ resemblance? Specifically, in my limited experience LLVM IR resembles typical hardware rather than C++. Modeling the hardware is a design goal for C++, of course, so anything hardware-like must be C++-like. Can you describe some ways in which IR resembles C++ that are not plausible ways to resemble hardware? I'm really curious.

webkike 9y ago

You are correct. The article is misinformed

pebers 9y ago

Perhaps they mean that within those 1.6 million lines, the only frontend is Clang? But the line count for gcc includes e.g. gfortran as well?

valarauca1 9y ago
The GCC addons (gcc-fortran, gcc-cobal, and gcc-go) just convert the existing language into C that is just ingested into the standard GCC. It was only around 2008 that gcc-c++ stopped doing the internal translation/name mangling dance and gained native first party support [1]. Much like how the older versions of GCC used to just produce ASM files to be ingested into GAS to create the binary objects. [1] https://lwn.net/Articles/542457/ https://lwn.net/Articles/542457/
- lgg 9y ago
  The link you posted has nothing to do with how GCC does codegen. It is about GCC transitioning itself from being written in C to being written in C++.
- mhh__ 9y ago
  GCC IR is not C

a_humean 9y ago

Yeah, I thought that part of the point of the LLVM project was that front end language compilers (entirely language agnostic) could target non-platform specific LLVM IR code, and then let the backend LLVM compiler compile to multiple backend targets. It simplifies front-end compiler development by having a single IR target over multiple native target.

wolfgke 9y ago

> LLVM is language agnostic There exists no such thing as "language agnostic". Every intermediate representation contains properties (very often implicitly) that are specific at least to a class of programming languages.

whatshisface 9y ago
If the intermediate representation is no more language-specific than x86, then it isn't really the compiler that's language-specific: it's modern computer architecture.
- FeepingCreature 9y ago
  LLVM IR is fairly C specific. For instance, you better not be trying to write a language where null pointer access is defined...
  - nicwilson 9y ago
    We compile D with LLVM. Null pointer access is defined to crash, unlike C/C++ where it is undefined.
    - pcwalton 9y ago
      Are you adding explicit checks (possibly using fault maps to optimize them)? Because if not, you're depending on undefined behavior.
      - nicwilson 9y ago
        That is only necessary for single objects that are bigger than one page which, we dont currently do (but should). Arrays are no problem since we check the bounds, only raw pointers cause problems.
        pcwalton 9y ago

pipio21 9y ago

We have worked with LLVM a lot for own own compilers and DSPs. LLVM Intermediate representation in practice is C++, AST is C++. LLVM do not support many languages,many languages support LLVM, which is different. That is many people, or languages' designers, have made the necessary work to make their languages compatible with LLVM IR. If the language is similar to C++, this is a simple task, if it is very different, it is very hard. LLVM does not support other languages, in fact, codebase changes a lot making maintenance painful.

Arnt 9y ago
Could you elaborate on the C++ resemblance? Specifically, in my limited experience LLVM IR resembles typical hardware rather than C++. Modeling the hardware is a design goal for C++, of course, so anything hardware-like must be C++-like. Can you describe some ways in which IR resembles C++ that are not plausible ways to resemble hardware? I'm really curious.