Query on llvm-gcc v/s llc/as

Hi all,

I have started using LLVM recently, and I have a basic question. There are two different ways in which an executable can be generated from a .c file - by just using llvm-gcc, and using a combination of llvm-gcc (to emit bitcode .bc), llc (to emit assembly code), as (to assemble the code) and collect2 (linker). For example, consider compiling test.c :

1) llvm-gcc -O3 test.c -o testExecutable

2) llvm-gcc -O3 -emit-llvm -c test.c -o test.bc
   llc -O3 test.bc -o test.s
   as test.s -o test.o
   gcc test.o -o testExecutable2

My question is, are the two executables (testExecutable and testExecutable2) basically the same ? Is there any huge difference between the two processes which will impact the executable's running time ? I am asking this because I compared the executables by their wordcounts (wc -c), and there was a slight difference. However, when I timed the test runs, both performances were identical.

Kindly help me out on this.

Thanks.
Raghu.

Hi Raghu,

I think the difference lies in the fact that llc has 4 as its maximum
optimization level to allow link-time optimization. If this is what you're
noticing, then the second should be modified to "llc -O4 test.bc -o test.s". I
could be wrong though.

--Sam

The two control flows should be basically identical. In practice there may be a few differences due to differing defaults for things like relocation model, processor tuning, etc, between llc and llvm-gcc. When generating code directly, llvm-gcc uses the same back-end code generator that llc encapsulates, and then calls as and ld to assemble and link the code (pass "-v" to llvm-gcc to see the command lines it's using to do so).

-Jim

Hi Raghu,

1) llvm-gcc -O3 test.c -o testExecutable

2) llvm-gcc -O3 -emit-llvm -c test.c -o test.bc
    llc -O3 test.bc -o test.s
    as test.s -o test.o
    gcc test.o -o testExecutable2

My question is, are the two executables (testExecutable and testExecutable2) basically the same ? Is there any huge difference between the two processes which will impact the executable's running time ? I am asking this because I compared the executables by their wordcounts (wc -c), and there was a slight difference. However, when I timed the test runs, both performances were identical.

the default options are different, for example llc eliminates the frame pointer
by default while llvm-gcc doesn't (on the 32 bit x86 platform). Also, I think
llc automatically turns on features (eg: SSE) if your cpu supports it, while
with llvm-gcc this is not automatic.

Ciao,

Duncan.