void thread_entry(int cid, int nc)
{
const int R = 8;
int m, n, p;
uint64_t s = 0xdeadbeefU;
if (have_vec) {
m = HCBM;
n = HCBN;
p = HCBK;
} else {
m = CBM;
n = CBN;
p = CBK;
}
t a[m*p];
t b[p*n];
t c[m*n];
for (size_t i = 0; i < m; i++)
for (size_t j = 0; j < p; j++)
a[i*p+j] = (t)(s = lfsr(s));
for (size_t i = 0; i < p; i++)
for (size_t j = 0; j < n; j++)
b[i*n+j] = (t)(s = lfsr(s));
memset(c, 0, m*n*sizeof(c[0]));
size_t instret, cycles;
if (have_vec) {
for (int i = 0; i < R; i++)
{
instret = -rdinstret();
cycles = -rdcycle();
mm_rb_hwacha(m, n, p, a, p, b, n, c, n);
instret += rdinstret();
cycles += rdcycle();
}
} else {
for (int i = 0; i < R; i++)
{
instret = -rdinstret();
cycles = -rdcycle();
mm(m, n, p, a, p, b, n, c, n);
instret += rdinstret();
cycles += rdcycle();
}
}
asm volatile("fence");
printf("C%d: reg block %dx%dx%d, cache block %dx%dx%d\n",
cid, RBM, RBN, RBK, CBM, CBN, CBK);
printf("C%d: %d instructions\n", cid, (int)(instret));
printf("C%d: %d cycles\n", cid, (int)(cycles));
printf("C%d: %d flops\n", cid, 2*m*n*p);
printf("C%d: %d Mflops @ 1 GHz\n", cid, 2000*m*n*p/(cycles));
#if 1
for (size_t i = 0; i < m; i++)
{
for (size_t j = 0; j < n; j++)
{
t s = 0;
for (size_t k = 0; k < p; k++)
s += a[i*p+k] * b[k*n+j];
s *= R;
if (fabs(c[i*n+j]-s) > fabs(1e-6*s))
{
printf("C%d: c[%lu][%lu] %f != %f\n", cid, i, j, c[i*n+j], s);
exit(1);
}
}
}
#endif
barrier(nc);
exit(0);
}
static void run_loaded_program(size_t argc, char** argv, uintptr_t kstack_top)
{
// copy phdrs to user stack
size_t stack_top = current.stack_top - current.phdr_size;
memcpy((void*)stack_top, (void*)current.phdr, current.phdr_size);
current.phdr = stack_top;
// copy argv to user stack
for (size_t i = 0; i < argc; i++) {
size_t len = strlen((char*)(uintptr_t)argv[i])+1;
stack_top -= len;
memcpy((void*)stack_top, (void*)(uintptr_t)argv[i], len);
argv[i] = (void*)stack_top;
}
// copy envp to user stack
const char* envp[] = {
// environment goes here
};
size_t envc = sizeof(envp) / sizeof(envp[0]);
for (size_t i = 0; i < envc; i++) {
size_t len = strlen(envp[i]) + 1;
stack_top -= len;
memcpy((void*)stack_top, envp[i], len);
envp[i] = (void*)stack_top;
}
// align stack
stack_top &= -sizeof(void*);
struct {
long key;
long value;
} aux[] = {
{AT_ENTRY, current.entry},
{AT_PHNUM, current.phnum},
{AT_PHENT, current.phent},
{AT_PHDR, current.phdr},
{AT_PAGESZ, RISCV_PGSIZE},
{AT_SECURE, 0},
{AT_RANDOM, stack_top},
{AT_NULL, 0}
};
// place argc, argv, envp, auxp on stack
#define PUSH_ARG(type, value) do { \
*((type*)sp) = (type)value; \
sp += sizeof(type); \
} while (0)
#define STACK_INIT(type) do { \
unsigned naux = sizeof(aux)/sizeof(aux[0]); \
stack_top -= (1 + argc + 1 + envc + 1 + 2*naux) * sizeof(type); \
stack_top &= -16; \
long sp = stack_top; \
PUSH_ARG(type, argc); \
for (unsigned i = 0; i < argc; i++) \
PUSH_ARG(type, argv[i]); \
PUSH_ARG(type, 0); /* argv[argc] = NULL */ \
for (unsigned i = 0; i < envc; i++) \
PUSH_ARG(type, envp[i]); \
PUSH_ARG(type, 0); /* envp[envc] = NULL */ \
for (unsigned i = 0; i < naux; i++) { \
PUSH_ARG(type, aux[i].key); \
PUSH_ARG(type, aux[i].value); \
} \
} while (0)
STACK_INIT(uintptr_t);
if (current.cycle0) { // start timer if so requested
current.time0 = rdtime();
current.cycle0 = rdcycle();
current.instret0 = rdinstret();
}
trapframe_t tf;
init_tf(&tf, current.entry, stack_top);
__clear_cache(0, 0);
write_csr(sscratch, kstack_top);
start_user(&tf);
}
