水仙花数指的是一个 n 位数 ( n≥3 )，它的每个位上的数字的 n 次幂之和等于它本身。(例如:1^3 + 5^3 + 3^3 = 153)

下面代码，求19位，在E7200 2.53G的win xp 32b上耗时1秒左右。欢迎大家指出更多的优化方法。

#include <math.h> #include <string.h> #include <stdio.h> #include <stdlib.h> #include <windows.h> unsigned long long g_maxValue; unsigned long long g_minValue; unsigned long long POWXN[10]; void preCalculation(unsigned char n) { g_maxValue = pow((long double)10, n); g_minValue = pow((long double)10, n-1); POWXN[0] = 0ui64; POWXN[1] = 1ui64; for(int i = 2ui64; i < 10ui64; i++) { POWXN[i] = (unsigned long long)(pow((long double)i, n) + 0.5); } } bool search(unsigned long long sum, char *nums, unsigned char n) { char *p; int c = 0; int i = 0; int e = n; char tmp; char buf[256]; memcpy(buf,nums,n); /* // 19位时，交换花费的计算量仍然超过节约下来的搜索花费。 for(i = 0; i < n; ++i) { c = sum%10; sum /= 10; p = (char*)memchr(buf,c,e); if(NULL == p) { return false; } else { –e; if(p != buf + e) { tmp = buf[e]; buf[e] = *p; *p = tmp; } } } */ for(i = 0; i < n; ++i) { c = sum%10; sum /= 10; p = (char*)memchr(buf,c,n); if(NULL == p) { return false; } else { *p = ‘n’; } } return true; } int NarcissusNum(unsigned char n, unsigned long long *out) { preCalculation(n); char *nums = (char *)malloc(sizeof(char) * n); memset(nums, 0, sizeof(char) * n); unsigned long long sum = 0; int c = 0; // 遍历所有组合。不用排列，和即是所需排列 // 000001, 000002, 000003, … ,000009, 000011, 000012, …, 000019, 000022, 000023, …, 000111, …, 00122, …, // 001111, …, 011111, …, 111111, …, 899999, 999999 int xx = 0; for(;;) { while(nums[0] <= 9) { //xx++; sum = 0; for(unsigned char j = 0; j < n; ++j) { sum += POWXN[nums[j]]; } if(sum < g_minValue) { nums[0]++; continue; } else if(sum > g_maxValue) { //break; // 寻找下一个不超过最大值的组合。 // 与直接break出去比较，位数越大，优化效果越明显。19位时大约节约10~20ms unsigned char i = 1; unsigned char ii; unsigned long long tmp = sum; while(tmp > g_maxValue && i < n) { tmp = sum; for(ii = 0; ii < i; ++ii) { if(9 == nums[i]) { ++i; continue; } tmp -= POWXN[nums[ii]]; tmp += POWXN[nums[i]+1]; } ++i; } –i; ++nums[i]; for(unsigned char ii = 0; ii < i; ++ii) { nums[ii] = nums[i]; } } else { if(search(sum, nums, n)) { *out = sum; out++; } } nums[0]++; } // 组合进位。 // 00009 => 000011, 01299 => 01333, 39999 => 44444 unsigned char k = 1; for(; k < n; ++k) { nums[k]++; if(nums[k] < 10) { for(; k > 1; –k) { nums[k-1] = nums[k]; } break; } } if(nums[n-1] > 9) { free(nums); return xx; } nums[0] = nums[1]; } free(nums); return xx; } // 校验结果 bool Check(unsigned long long val, unsigned char n) { unsigned long long tmp = val; unsigned long long sum = 0; unsigned char c; while(0i64 != tmp) { c = tmp % 10; tmp /= 10; sum += (unsigned long long )pow((long double)c, n); } if(sum == val) return true; return false; } int _tmain(int argc, _TCHAR* argv[]) { FILETIME ct; FILETIME et; FILETIME kt; FILETIME ut; __int64 ktStart=0; __int64 utStart=0; __int64 ktStop=0; __int64 utStop=0; LARGE_INTEGER ss,ee,fq; HANDLE cur; int i,j,dump; cur = GetCurrentThread(); QueryPerformanceCounter(&ss); //QueryPerformanceCounter的单位应该是cpu的时间周期时间，1/cpu频率 GetThreadTimes(cur,&ct,&et,&kt,&ut); // 起始时的内核时间和用户时间。不过精度也就10ms级 ktStart = (((__int64)kt.dwHighDateTime)<<32) + (__int64)kt.dwLowDateTime; utStart = (((__int64)ut.dwHighDateTime)<<32) + (__int64)ut.dwLowDateTime; unsigned long long nums[10] = {0}; unsigned char n = 19; for(int c = 0; c < 10; c++) { printf(“%d/n”,NarcissusNum(n, nums)); } GetThreadTimes(cur,&ct,&et,&kt,&ut); // 结束时的内核时间和用户时间 ktStop = (((__int64)kt.dwHighDateTime)<<32) + (__int64)kt.dwLowDateTime; utStop = (((__int64)ut.dwHighDateTime)<<32) + (__int64)ut.dwLowDateTime; printf(“Kernel Time: %llu ms/n”, (ktStop – ktStart)/10000); //FILETIME的单位是100ns，除10000变ms printf(“User Time: %llu ms/n”, (utStop – utStart)/10000); QueryPerformanceCounter(&ee); QueryPerformanceFrequency(&fq); // QueryPerformanceCounter高精度的，但记得并不是代码实际消耗的时间。 printf(“Time: %llu us/n”, (ee.QuadPart-ss.QuadPart)/(fq.QuadPart/1000000)); for(int i = 0; i < 10 && 0 != nums[i]; i++) { if(Check(nums[i], n)) { printf(“%llu/n”, nums[i]); } else { printf(“%llu worng/n”, nums[i]); } } return 0; }

优化了下，800ms

#include <math.h> #include <string.h> #include <stdio.h> #include <stdlib.h> #include <windows.h> unsigned long long g_maxValue; unsigned long long g_minValue; unsigned long long POWXN[10]; unsigned long long INC2N[10]; void preCalculation(unsigned char n) { g_maxValue = pow((long double)10, n); g_minValue = pow((long double)10, n-1); POWXN[0] = 0ui64; POWXN[1] = 1ui64; for(int i = 2; i < 10; i++) { POWXN[i] = (unsigned long long)(pow((long double)i, n) + 0.5); } INC2N[0] = 0; INC2N[1] = 1; for(int i = 2; i < 10; i++) { INC2N[i] = POWXN[i] – POWXN[i-1]; } } bool search(unsigned long long sum, char *nums, unsigned char n) { char c; unsigned char nn; unsigned char i; unsigned char j; char buf[256]; char *p=buf; memcpy(buf,nums,n); /* for(i = 0; i < n; ++i) { c = sum%10; sum /= 10; p = (char*)memchr(buf,c,n); if(NULL == p) { return false; } else { *p = ‘n’; } }*/ if(n < 4) { nn = 0; } else { nn = n – 4; } for(i = 0; i < n; i++) { c = sum%10; sum /= 10; for(j=0; j < nn; j+=4) { if(c == p[j]) { p[j] = ‘n’; c = ‘f’; break; } else if(c == p[j+1]) { p[j+1] = ‘n’; c = ‘f’; break; } else if(c == p[j+2]) { p[j+2] = ‘n’; c = ‘f’; break; } else if(c == p[j+3]) { p[j+3] = ‘n’; c = ‘f’; break; } } if(‘f’ != c) { for(; j < n; j++) { if(c == p[j]) { p[j] = ‘n’; c = ‘f’; break; } } if(‘f’!= c) return false; } } return true; } int NarcissusNum(unsigned char n, unsigned long long *out) { char c; unsigned char i,j,k,nn; char buf[256]; unsigned long long tmp; char *nums; int xx = 0; unsigned long long sum = 0; nums = (char *)malloc(sizeof(char) * n + 1); memset(nums, 0, sizeof(char) * n + 1); memset(buf,0,256); if(n < 4) { nn = 0; } else { nn = n – 4; } preCalculation(n); // 遍历所有组合。不用排列，和即是所需排列 // 000001, 000002, 000003, … ,000009, 000011, 000012, …, 000019, 000022, 000023, …, 000111, …, 00122, …, // 001111, …, 011111, …, 111111, …, 899999, 999999 for(;;) { while(nums[0] <= 9) { //++xx; sum += INC2N[nums[0]]; if(sum < g_minValue) { ++nums[0]; continue; } else if(sum > g_maxValue) {//++xx; //break; // 寻找下一个不超过最大值的组合。 // 与直接break出去比较，位数越大，优化效果越明显。 i = 1; tmp = sum; // 假设1 3 8 9 9过大，1 4 4 4 4在范围内 while(tmp > g_maxValue && i < n) { tmp = sum; // 第一次1 3 9 9 9 仍然过大 // 第二次1 4 4 4 4 范围内 for(j = 0; j <= i; ++j) { if(9 == nums[i]) { ++i; –j; continue; } tmp -= POWXN[nums[j]]; tmp += POWXN[nums[i]+1]; } ++i; } // tmp 为 1 4 4 4 4的和 –i; if(9 == nums[n – 1]) { free(nums); return xx; //break; } ++nums[i]; for(j = 0; j < i; ++j) { nums[j] = nums[i]; } // sum 降为 1 4 4 4 3的和，上面循环处再加回去。 sum = tmp – INC2N[nums[0]]; } else {//++xx; /* 19位下，掉用函数大概慢了40ms if(search(sum, nums, n)) { *out = sum; ++out; } ++nums[0]; */ memcpy(buf,nums,n); tmp = sum; for(i = 0; i < n; i++) { c = tmp%10; tmp /= 10; // 展开循环 for(j=0; j < nn; j+=4) { if(c == buf[j]) { buf[j] = ‘n’; c = ‘f’; break; } else if(c == buf[j+1]) { buf[j+1] = ‘n’; c = ‘f’; break; } else if(c == buf[j+2]) { buf[j+2] = ‘n’; c = ‘f’; break; } else if(c == buf[j+3]) { buf[j+3] = ‘n’; c = ‘f’; break; } } if(‘f’ != c) { for(; j < n; j++) { if(c == buf[j]) { buf[j] = ‘n’; c = ‘f’; break; } } if(‘f’!= c) { goto NEXT; } } } *out = sum; ++out; NEXT: ++nums[0]; } } // 组合进位。 // 00009 => 0000a => 000011, 01299 => 012aa => 01333, 39999 => 3aaaa => 44444 //++xx; for(k = 1; k < n; ++k) { nums[k]++; if(nums[k] < 10) { for(; k > 0; –k) { nums[k-1] = nums[k]; } break; } } if(nums[n-1] > 9) { free(nums); return xx; } sum = POWXN[nums[0]-1]; // 进位后，最低位（nums[0]）不可能是0 for(j = 1; j < n; ++j) { sum += POWXN[nums[j]]; } } free(nums); return xx; } // 校验结果 bool Check(unsigned long long val, unsigned char n) { unsigned long long tmp = val; unsigned long long sum = 0; unsigned char c; while(0i64 != tmp) { c = tmp % 10; tmp /= 10; sum += (unsigned long long )pow((long double)c, n); } if(sum == val) return true; return false; } int _tmain(int argc, _TCHAR* argv[]) { FILETIME ct; FILETIME et; FILETIME kt; FILETIME ut; __int64 ktStart=0; __int64 utStart=0; __int64 ktStop=0; __int64 utStop=0; LARGE_INTEGER ss,ee,fq; HANDLE cur; cur = GetCurrentThread(); QueryPerformanceCounter(&ss); //QueryPerformanceCounter的单位应该是cpu的时间周期时间，1/cpu频率 GetThreadTimes(cur,&ct,&et,&kt,&ut); // 起始时的内核时间和用户时间。不过精度也就10ms级 ktStart = (((__int64)kt.dwHighDateTime)<<32) + (__int64)kt.dwLowDateTime; utStart = (((__int64)ut.dwHighDateTime)<<32) + (__int64)ut.dwLowDateTime; unsigned long long nums[10] = {0}; unsigned char n = 19; for(int c = 0; c < 10; c++) { NarcissusNum(n, nums); //printf(“%d/n”,NarcissusNum(n, nums)); /*NarcissusNum(c, nums); printf(“%d:/n”,c); for(int i = 0; i < 10 && 0 != nums[i]; i++) { if(Check(nums[i], c)) { printf(“%llu/n”, nums[i]); } else { printf(“%llu worng/n”, nums[i]); } } memset(nums,0,sizeof(unsigned long long) * 10);*/ } GetThreadTimes(cur,&ct,&et,&kt,&ut); // 结束时的内核时间和用户时间 ktStop = (((__int64)kt.dwHighDateTime)<<32) + (__int64)kt.dwLowDateTime; utStop = (((__int64)ut.dwHighDateTime)<<32) + (__int64)ut.dwLowDateTime; printf(“Kernel Time: %llu ms/n”, (ktStop – ktStart)/10000); //FILETIME的单位是100ns，除10000变ms printf(“User Time: %llu ms/n”, (utStop – utStart)/10000); QueryPerformanceCounter(&ee); QueryPerformanceFrequency(&fq); // QueryPerformanceCounter高精度的，但记得并不是代码实际消耗的时间。 printf(“Time: %llu us/n”, (ee.QuadPart-ss.QuadPart)/(fq.QuadPart/1000000)); //printf(“Time: %llu ns/n”, (ee.QuadPart-ss.QuadPart)/(fq.QuadPart/1000000000)); for(int i = 0; i < 10 && 0 != nums[i]; i++) { if(Check(nums[i], n)) { printf(“%llu/n”, nums[i]); } else { printf(“%llu worng/n”, nums[i]); } } return 0; }