代码

void sort(unsigned *a, int n){
    #define N 100000000
    #define D 256
    #define D1 255
    #define cal(w,w2,tw,op)\
    p=w2-1;\
    for(i=0;i<D;++i)rs[i]=p,p+=tw[i];\
    for(i=0;i<n;++i){\
    	p=w+i,\
    	*++rs[p[0]op]=p[0];\
    }
    unsigned b[N],*rs[D],cnt0[D],cnt1[D],cnt2[D],cnt3[D];
    register unsigned*p,i,x;
    for(i=0;i<D;++i)
    	cnt0[i]=cnt1[i]=cnt2[i]=cnt3[i]=0;
    for(i=0;i<n;++i)
        x=a[i],++cnt0[x&D1],++cnt1[x>>8&D1],++cnt2[x>>16&D1],++cnt3[x>>24];
    cal(a,b,cnt0,&D1);
    cal(b,a,cnt1,>>8&D1);
    cal(a,b,cnt2,>>16&D1);
    cal(b,a,cnt3,>>24);
}

评测结果