LLVM vs. GCC MIPS code generation, any tests?

Question

I am interested to know what is the “best” free / OSS compiler for MIPS, GCC or LLVM code, or is there something even better than those?

I am interested in learning more about the fast and limited memory generated assembly code, and not about the size of the code.

In other words, llvm-opt does a better job than gcc -O3?

Answer 1

http://www.phoronix.com/scan.php?page=news_item&px=OTI1MA "LLVM 2.9 Release Candidate 2 Is Here

Posted by Michael Larabel on March 25, 2011 No ARM LLVM tests due to lack of hardware ... "

maybe someone with fast oblique arc / quadrant and ARM LLVM etc. can run the http://openbenchmarking.org/ scanner before Monday, and Michael can add these to his other results

Answer 2

I don't know about mips, I tried ARM, and the llvm code was 10-20% slower than the current gcc. These tests were based on zlib. decompression in itself and compression, and then decompression. both clang and llvm-gcc are used. I preferred clang because -m32 actually runs on a 64-bit host. For the test in question, I found that NOT using -O2 (or -O3) produced the fastest code. tied the bytecode modules into one large module and performed one option with standard optimizations to get the fastest code. llc was -O2 by default, and this helped improve performance.

EDIT:

An interesting test between gcc and llvm / clang for mips.

void dummy ( unsigned int ); void dowait ( void ) { unsigned int ra; for(ra=0x80000;ra;ra--) dummy(ra); } 

gcc:

 9d006034 <dowait>: 9d006034: 27bdffe8 addiu sp,sp,-24 9d006038: afb00010 sw s0,16(sp) 9d00603c: afbf0014 sw ra,20(sp) 9d006040: 3c100008 lui s0,0x8 9d006044: 02002021 move a0,s0 9d006048: 0f40180a jal 9d006028 <dummy> 9d00604c: 2610ffff addiu s0,s0,-1 9d006050: 1600fffd bnez s0,9d006048 <dowait+0x14> 9d006054: 02002021 move a0,s0 9d006058: 8fbf0014 lw ra,20(sp) 9d00605c: 8fb00010 lw s0,16(sp) 9d006060: 03e00008 jr ra 9d006064: 27bd0018 addiu sp,sp,24 

And llvm after assembly

 9d006034 <dowait>: 9d006034: 27bdffe8 addiu sp,sp,-24 9d006038: afbf0014 sw ra,20(sp) 9d00603c: afb00010 sw s0,16(sp) 9d006040: 3c020008 lui v0,0x8 9d006044: 34440000 ori a0,v0,0x0 9d006048: 2490ffff addiu s0,a0,-1 9d00604c: 0f40180a jal 9d006028 <dummy> 9d006050: 00000000 nop 9d006054: 00102021 addu a0,zero,s0 9d006058: 1600fffb bnez s0,9d006048 <dowait+0x14> 9d00605c: 00000000 nop 9d006060: 8fb00010 lw s0,16(sp) 9d006064: 8fbf0014 lw ra,20(sp) 9d006068: 27bd0018 addiu sp,sp,24 9d00606c: 03e00008 jr ra 9d006070: 00000000 nop 

I speak after assembly because I saw gnu, like all that is.

 .globl PUT32 PUT32: sw $a1,0($a0) jr $ra nop 

and reinstall the assembly for me:

 9d00601c <PUT32>: 9d00601c: 03e00008 jr ra 9d006020: ac850000 sw a1,0(a0) 9d006024: 00000000 nop 

The difference between the code generated by llvm and gcc is the instructions that are placed in the branch delay slot. I used clang and llc to build the assembly, and then used binutils, gnu as, to create the binary. So the curiosity is that for my hand the code was compiled:

 ori $sp,$sp,0x2000 jal notmain nop 

It is optimized for me:

 9d006004: 0f401820 jal 9d006080 <notmain> 9d006008: 37bd2000 ori sp,sp,0x2000 9d00600c: 00000000 nop 

but generated llc code

 addiu $16, $4, -1 jal dummy nop 

did not have

 9d006048: 2490ffff addiu s0,a0,-1 9d00604c: 0f40180a jal 9d006028 <dummy> 9d006050: 00000000 nop 

Answer 3

LLVM is usually better than GCC on x86, but I have not found tests for MIPS. Since you are asking this question, I assume that you have access to the MIPS machine, so why don’t you compile the Benchmarks Game C code Computer language with GCC and LLVM, and see which is faster. I assume that GCC will be, since the MIPS backend is relatively new, but the code is much cleaner, and I expect LLVM to ultimately win.