On Fri, Jul 7, 2017 at 6:39 AM, Luc Van Oostenryck <luc.vanoostenryck@xxxxxxxxx> wrote: > Here is some patches reworking the whole ptrlist API > which move all the logic currently present in the macros > in a small set of primitive functions using an iterator > structure. I am not considering it for this release for sure. Actually, NACK on the function based iterator API. There is no much of a real function enhancement. And it is a big slow down. I actually like Linus' macro based implementation. I want to keep it that way. It looks very clean on the caller side. It perform extremely fast. All the internal variable will easy promote to register. I am all about using code to squeezing very last bit of performance out of the hardware. The Linus' ptrlist does an amazing job. Now the function based iterator: The loop variable will have to keep in the struct member. It does not inline well. On each function call, the compiler have to sync loop variable into the memory based struct. Take a look at the machine code it generated. It is no where near as good as the original one. The benchmark confirms that as well. Here is the program just iterate over a long list for many times with some condition testing inside the list item. With Linus' ptrlist: $ time ./test-ptrlist sum 4992000000 real 0m9.071s user 0m9.028s sys 0m0.002s Exactly same program with your iterator: $ time ./test-ptrlist sum 4992000000 real 0m31.601s user 0m31.320s sys 0m0.003s That is 344% slow down. I have another test, for just the loop iteration part with nop inside the loop. That is 10x speed difference. Just make it out there, I am not willing to take a pure cosmetic change with like 0.1% performance penalty. So NACK on this. I feel sorry you have spent some many time on this. If I know earlier I would have tell you that is not going to work. I actually try that myself long time ago and find out Linus' ptrlist is very hard to beat in terms of raw performance. The git branh for this test prgram is at: https://git.kernel.org/pub/scm/devel/sparse/chrisl/sparse.git/log/?h=test-ptrlist Here is the test program. Sorry gmail cause some white space damage to the code. Chris /* * sparse ptrlist abuse */ #include <stdio.h> #include <stdlib.h> #include <stdarg.h> #include <string.h> #include <assert.h> #include "lib.h" #include "symbol.h" #include "expression.h" #include "linearize.h" #include "flow.h" #include "storage.h" #include "target.h" int main(int argc, char **argv) { struct string_list *filelist = NULL; struct symbol_list *syms = NULL; struct symbol *sym = (struct symbol *)0x10; unsigned long sum = 0; sparse_initialize(argc, argv, &filelist); for (int i= 0; i < 10000; i++, sym++) add_symbol(&syms, sym); for (int i= 0; i < 100; i++) { for (int j= 0; j < 10000; j++) { FOR_EACH_PTR(syms, sym) { if ((unsigned long) sym & 0x1000) sum ++; } END_FOR_EACH_PTR(sym); } } printf("sum %ld\n", sum); return 0; } -- To unsubscribe from this list: send the line "unsubscribe linux-sparse" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
