1 Date: Fri, 1 Jun 2001 16:38:17 -0500 (CDT) 2 From: Chris Lattner <sabre (a] nondot.org> 3 To: Vikram S. Adve <vadve (a] cs.uiuc.edu> 4 Subject: Interesting: GCC passes 5 6 7 Take a look at this document (which describes the order of optimizations 8 that GCC performs): 9 10 http://gcc.gnu.org/onlinedocs/gcc_17.html 11 12 The rundown is that after RTL generation, the following happens: 13 14 1 . [t] jump optimization (jumps to jumps, etc) 15 2 . [t] Delete unreachable code 16 3 . Compute live ranges for CSE 17 4 . [t] Jump threading (jumps to jumps with identical or inverse conditions) 18 5 . [t] CSE 19 6 . *** Conversion to SSA 20 7 . [t] SSA Based DCE 21 8 . *** Conversion to LLVM 22 9 . UnSSA 23 10. GCSE 24 11. LICM 25 12. Strength Reduction 26 13. Loop unrolling 27 14. [t] CSE 28 15. [t] DCE 29 16. Instruction combination, register movement, scheduling... etc. 30 31 I've marked optimizations with a [t] to indicate things that I believe to 32 be relatively trivial to implement in LLVM itself. The time consuming 33 things to reimplement would be SSA based PRE, Strength reduction & loop 34 unrolling... these would be the major things we would miss out on if we 35 did LLVM creation from tree code [inlining and other high level 36 optimizations are done on the tree representation]. 37 38 Given the lack of "strong" optimizations that would take a long time to 39 reimplement, I am leaning a bit more towards creating LLVM from the tree 40 code. Especially given that SGI has GPL'd their compiler, including many 41 SSA based optimizations that could be adapted (besides the fact that their 42 code looks MUCH nicer than GCC :) 43 44 Even if we choose to do LLVM code emission from RTL, we will almost 45 certainly want to move LLVM emission from step 8 down until at least CSE 46 has been rerun... which causes me to wonder if the SSA generation code 47 will still work (due to global variable dependencies and stuff). I assume 48 that it can be made to work, but might be a little more involved than we 49 would like. 50 51 I'm continuing to look at the Tree -> RTL code. It is pretty gross 52 because they do some of the translation a statement at a time, and some 53 of it a function at a time... I'm not quite clear why and how the 54 distinction is drawn, but it does not appear that there is a wonderful 55 place to attach extra info. 56 57 Anyways, I'm proceeding with the RTL -> LLVM conversion phase for now. We 58 can talk about this more on Monday. 59 60 Wouldn't it be nice if there were a obvious decision to be made? :) 61 62 -Chris 63 64
