http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53081
Bug #: 53081 Summary: memcpy/memset loop recognition Classification: Unclassified Product: gcc Version: unknown Status: UNCONFIRMED Severity: normal Priority: P3 Component: middle-end AssignedTo: unassig...@gcc.gnu.org ReportedBy: xinlian...@gmail.com Both LLVM and icc recognize initialization and copy loop and synthesize calls to memcpy and memset. memmove call can also be synthesized when src/target may overlap. Option needs to provided to disable such optimization in signal handlers. I consider this as optimization for benchmarking ;) For instance, the prime number finder program sieve.c is one of the benchmarks in LLVM. Both LLVM and icc beats gcc in this one because of the missing optimization. #ifndef T #define T int #endif T arr[1000]; void foo(int n) { int i; for (i = 0; i < n; i++) { arr[i] = 0; } } void foo2(int n, T* p) { int i; for (i = 0; i < n; i++) { *p++ = 0; } } #ifndef T #define T int #endif T arr[1000]; T arr2[1000]; void foo(int n) { int i; for (i = 0; i < n; i++) { arr[i] = arr2[i]; } } // sieve.c /* -*- mode: c -*- * $Id: sieve.c 36673 2007-05-03 16:55:46Z laurov $ * http://www.bagley.org/~doug/shootout/ */ #include <stdio.h> #include <stdlib.h> int main(int argc, char *argv[]) { #ifdef SMALL_PROBLEM_SIZE #define LENGTH 17000 #else #define LENGTH 170000 #endif int NUM = ((argc == 2) ? atoi(argv[1]) : LENGTH); static char flags[8192 + 1]; long i, k; int count = 0; while (NUM--) { count = 0; for (i=2; i <= 8192; i++) { flags[i] = 1; } for (i=2; i <= 8192; i++) { if (flags[i]) { /* remove all multiples of prime: i */ for (k=i+i; k <= 8192; k+=i) { flags[k] = 0; } count++; } } } printf("Count: %d\n", count); return(0); }