Galaxy S3 ARMv7 benchmark: Rust, Ocaml, Haskell, Go, Racket, Lua, C# and Java

[logicchains]

http://togototo.wordpress.com/2014/08/05/fibonacci-numbers-on-the-galaxy-s3-arm-benchmarks-of-rust-ocaml-haskell-go-racket-lua-c-and-java/

[higherpurpose]

It seems the Rust team is really optimizing the language for ARM chips.

[logicchains → higherpurpose]

They use the LLVM as the backend, which is quite well optimised. The relatively poorer performance of that program on x86 may just be a bug (as the x86 version is compiled with a slightly different version of the Rust compiler); the program uses recursion, which may not be a Rust idiom and hence may not be well tested for performance regressions.

[dbaupp → logicchains]

It doesn't actually use recursion; it is tail recursion and LLVM optimises it to a loop:

  .LBB0_77:
  	movq	%rax, %rbx
  	movq	%rcx, %rax
  	addq	%rbx, %rax
  	incq	%rdx
  	movq	%rbx, %rcx
  	jne	.LBB0_77

The difference on x64-64 is the Rust is using 64-bit integers (`int` is pointer sized), while the C is only using 32-bit ones; changing them to the 32-bit `i32` to match the C makes Rust super-fast:

  75699
  LANGUAGE Rust 2167
  75699
  LANGUAGE C 2220 # gcc
  75699
  LANGUAGE C 2271 # clang

I opened https://github.com/logicchains/armfib/pull/1 .

[logicchains → dbaupp]

Thanks. The tables have been updated to use the new values; I'm really surprised how much of a difference changing from int to int32 made.