------- Comment #8 from pinskia at gcc dot gnu dot org 2007-06-17 06:54 ------- > Size of the compilers, efficiency of both compiler and generated code are > clearly secondary Do you even know why I added POINTER_PLUS_EXPR? Did you even read my reply to Jeff Law on why I started working on this project? My reply to Jeff explains that generated code was my reason. On SPU, you have only loads that are only 128 bit long and the alignment is 128 bits only. So we need better alignment information prograted down from the tree level to the RTL level.
http://gcc.gnu.org/ml/gcc/2007-05/msg00691.html I guess you did not look at the other bug reports or commented on due to pointer_plus either? Where the code gen has been improved. I guess you are making comments to just make comments. And what does pointer_plus, SSE, Altivec have in common. Besides me caring about 2 of those. Nothing. In fact the later two have less in common than you think. Yes they are both vector extensions. I will mention about Altivec (VMX for me), I work on the PS3 toolchain and I really really do care about code generation and performance of the compiler. I work mostly on the PPU (PowerPC) side of things but every once in a while I work on the SPU side. If you want to compile only standards conforming code, then use -pedantic-errors as documented in the GCC manual which it seems you don't read. This is not a bug as --disable-decimal-float disables the C/C++ interfaces to the underlaying GCC support. Guess you don't understand that GCC is more than a C/C++ compiler. GCC stands for GNU Compiler Collection and NOT GNU C Compiler. One more thing about pointer plus, have you seen how much time I actually spent on working on this project? It took more around 6.5 months (including time working on other stuff). I guess you don't care that I needed approval from my manager to work on this project and then get approval to apply it and also review time also? If you don't see the last two, then you need to read the GCC lists better. I was very open about what currently works and what I could test. I don't think you have a right to tell me that my work is shit when you obviously did not read anything I have done about being straight forward about this project. Yes pointer plus was a big project, it was needed to allow GCC to complete with other compilers, even to get code generation back to where it was in 3.4 in many many cases. Take a look at PR 18412, PR 29708, PR 32273, PR 16913 (which still has issues but the main issue is solved by pointer_plus), PR 20643 (which is improved by pointer_plus but I don't know if it is all the way fixed), PR 21485 (where the original testcase fixed by pointer_plus), PR 28690 is helped by pointer plus (got SPEC results for Power6 with pointer plus vs without). So what more do you want? that is 7 code generation issues, some of them are regressions fixed/helped by pointer_plus. -- pinskia at gcc dot gnu dot org changed: What |Removed |Added ---------------------------------------------------------------------------- Status|UNCONFIRMED |RESOLVED Resolution| |INVALID http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32314
