Am i Mad?!: PowerPC Assembler

[lethe]

i m speculating, but perhaps would sun s C compiler be an example of a good compiler? it is written specifically for one OS, for an OS that exists mostly for only one processor family, and it is written by the same company that engineered the processor. so based on lavacdmn statements, i would bet that the solaris c compiler would be able to substantially outperform gcc on sparc/solaris.

i wonder if this is the case. it just so happens that i have a sparc/solaris box, but it only has gcc installed. i could get solaris cc though, since my university has a site license. maybe i could try some back of the envelope benchmarks....

 

[ladavacm]

Originally posted by lethe
i m speculating, but perhaps would sun s C compiler be an example of a good compiler? it is written specifically for one OS, for an OS that exists mostly for only one processor family, and it is written by the same company that engineered the processor. so based on lavacdmn statements, i would bet that the solaris c compiler would be able to substantially outperform gcc on sparc/solaris.

i wonder if this is the case. it just so happens that i have a sparc/solaris box, but it only has gcc installed. i could get solaris cc though, since my university has a site license. maybe i could try some back of the envelope benchmarks....

It does, according to my measurements about 2 years ago. In some cases, it is up to 4 times faster, especially on newest SPARC generations, some of which are not even correctly supported by gcc (i.e. incorrect code is generated)

 

[lethe]

if gcc killed off most other UNIX c compilers because it is free, then i would think that all those computer companies would have a vested interest in making it good at doing whatever architecture they use. for example, i ll bet apple has a whole little department working on improving gcc for OSX/PPC, since it is the compiler they use to compile their OS. if enough different people did this, then it could become a great compiler for every architecture.


with so much money behind it, don t you think it would eventually start becoming a good compiler?

on the other hand, perhaps its structure isn t very modular, and doesn t lend itself well to large rewrites or overhauls, and if it is already adequate enough for most people, then why bother.

 

[ladavacm]

Originally posted by lethe
if gcc killed off most other UNIX c compilers because it is free, then i would think that all those computer companies would have a vested interest in making it good at doing whatever architecture they use. for example, i ll bet apple has a whole little department working on improving gcc for OSX/PPC, since it is the compiler they use to compile their OS. if enough different people did this, then it could become a great compiler for every architecture.

I am sorry if I implied abundance of financial support for gcc; in fact, such support is apparently very limited.

with so much money behind it, don t you think it would eventually start becoming a good compiler?

Except that this money could never be recovered; this has to do with the particular license under which gcc is released. Therefore, not that much money is actually invested. If at all any serious money is spent, it is recovered through hardware sales.

on the other hand, perhaps its structure isn t very modular, and doesn t lend itself well to large rewrites or overhauls, and if it is already adequate enough for most people, then why bother. [/B]

It is quite modular, but not modular enough. It does not model data-flow architectures well, hence the lack of deep optimizations. It's design is relatively antiquated, dating to late 80ies. gcc's internals would have to be completely rewritten to bring it up to the modern CPU architectures; this is a lot of work with bleek chances for financial gains to the company undertaking such work. HW vendors still build their own compilers, but due to the market size, they charge an arm, a leg and both kidneys for them