/* driver */
static void scale(int n, data_t **a, data_t **b, data_t c) {
struct v_reg dst,src, src_end, v0, v1, al[10];
struct v_label loop1;
static unsigned insn[1024];
v_vptr vp;
/* simple unroll */
/* assert(n >= 4 && (n % 4) == 0); */
v_lambda("foo", "%p%p", al, V_LEAF, insn);
/* row & c come in as parameters */
dst = al[0];
src = al[1];
if(!v_getreg(&src_end, V_P, V_TEMP))
v_fatal("scale: out of registers\n");
if(!v_getreg(&v0, V_U, V_TEMP))
v_fatal("scale: out of registers\n");
if(!v_getreg(&v1, V_U, V_TEMP))
v_fatal("scale: out of registers\n");
loop1 = v_genlabel();
/* relies on contigous memory */
v_raw_load(v_ldpi(src, src, (0)), 1); /* perform loads without interlocks */
v_raw_load(v_ldpi(dst, dst, (0)), 1);
v_addpi(src_end, src, (n * n) * sizeof **a);
v_label(loop1);
/* load 2 to get rid of delay slots */
v_raw_load(v_ldui(v0, src, (0 * sizeof **a)), 1);
v_raw_load(v_ldui(v1, src, (1 * sizeof **a)), 1);
/* multiplies will be strength reduced */
if(strength_reduce) cmuli(v0, v0, c);
else v_mului(v0, v0, c);
v_addpi(dst, dst, (2 * sizeof **a));
if(strength_reduce) cmuli(v1, v1, c);
else v_mului(v1, v1, c);
v_stui(v0, dst, -(2 * sizeof **a));
v_addpi(src, src, (2 * sizeof **a));
/* schedule delay slot instructions */
v_schedule_delay(
v_bltp(src, src_end, loop1),
v_stui(v1, dst, -(1 * sizeof **a))
);
vp = v_end().v;
if(disass)
v_dump((void *)vp);
if(!pixie)
vp(a,b); /* perform multiplication */
}
/* A predicate is represented as a sum-of-products, that is
(A1 A2 ... ) OR (B1 B2 ...) OR ...
where each element in a product (the A?'s and B?'s) are simple
predicates like v > 10.
Predicates are represented in memory as an array of wk_term's,
one term for each immediate, variable, operator, conjunction or
disjunction. A single product is considered to be a group of
contiguous wk_term's that are not WK_ORs. The whole mess is
terminated by a WK_END. */
#include <vcode/vcode.h>
#include <xok/wk.h>
#include <xok/mmu.h>
#include <xok/sys_proto.h>
#include <xok/kerrno.h>
#include <xok/malloc.h>
#include <xok_include/assert.h>
#include <xok/printf.h>
#ifndef __CAP__
#include <xok/pmapP.h>
#else
#include <xok/pmap.h>
#endif
#define WK_MAX_CODE_BYTES 4096
#define OVERRUN_SAFETY 20
#define OVERRUN_CHECK \
{ \
if (v_ip > code + WK_MAX_CODE_BYTES - OVERRUN_SAFETY) { \
warn ("wk_compile: out of code space\n"); \
ret = -E_INVAL; \
goto error; \
} \
}
static int next_pp; /* outside function so can be used by cleanup code */
static int wk_compile (struct wk_term *t, int sz, char *code,
u_int *pred_pages) {
int i;
v_reg_t r1, r2, z, tag;
v_label_t end_of_term;
int start_term = 1;
int op1 = 1;
cap c;
struct Ppage *pp;
u_int ppn;
int ret = 0;
next_pp = 0;
v_lambda ("", "", NULL, 1, code, WK_MAX_CODE_BYTES);
if (!v_getreg (&r1, V_U, V_TEMP) ||
!v_getreg (&r2, V_U, V_TEMP) ||
!v_getreg (&z, V_U, V_TEMP) ||
!v_getreg (&tag, V_U, V_TEMP))
panic ("wk_compile: architecture doesn't have enough registers.");
v_setu (tag, -1);
v_setu (z, 0);
for (i = 0; i < sz; i++) {
if (start_term) {
end_of_term = v_genlabel ();
start_term = 0;
}
OVERRUN_CHECK;
switch (t[i].wk_type) {
case WK_VAR:
if (next_pp >= WK_MAX_PP-1) {
warn ("wk_compile: too many pages in predicate\n");
ret = -E_INVAL;
goto error;
}
if ((ret = env_getcap (curenv, t[i].wk_cap, &c)) < 0) {
goto error;
}
ppn = PGNO((u_int)t[i].wk_var);
if (!ppn || ppn >= nppage) {
printf ("at index %d\n", i);
warn ("wk_compile: invalid physical page\n");
ret = -E_INVAL;
goto error;
}
pp = ppages_get(ppn);
switch (Ppage_pp_status_get(pp)) {
case PP_USER:
if ((ret = ppage_acl_check(pp,&c,PP_ACL_LEN,0)) < 0) {
goto error;
}
ppage_pin (pp);
pred_pages[next_pp++] = ppn;
break;
case PP_KERNRO:
/* user can access pages that each env get's mapped r/o */
break;
default:
printf ("at index %d\n", i);
warn ("wk_compile: attempt to reference non PP_KERNRO or PP_USER page\n");
ret = -E_INVAL;
goto error;
}
if (op1) {
v_ldui (r1, z, (int )ptov (t[i].wk_var));
op1 = 0;
} else {
v_ldui (r2, z, (int )ptov (t[i].wk_var));
op1 = 1;
}
break;
case WK_IMM:
if (op1) {
v_setu (r1, t[i].wk_imm);
op1 = 0;
} else {
v_setu (r2, t[i].wk_imm);
op1 = 1;
}
break;
case WK_TAG: {
v_setu (tag, t[i].wk_tag);
break;
}
case WK_OP: {
switch (t[i].wk_op) {
case WK_GT: {
v_bleu (r1, r2, end_of_term);
break;
}
case WK_GTE: {
v_bltu (r1, r2, end_of_term);
break;
}
case WK_LT: {
v_bgeu (r1, r2, end_of_term);
break;
}
case WK_LTE: {
v_bgtu (r1, r2, end_of_term);
break;
}
case WK_EQ: {
v_bneu (r1, r2, end_of_term);
break;
}
case WK_NEQ: {
v_bequ (r1, r2, end_of_term);
break;
}
case WK_OR: {
v_retu (tag);
v_label (end_of_term);
start_term = 1;
break;
}
default: {
printf ("at index %d\n", i);
warn ("wk_compile: invalid wk-pred instruction\n");
ret = -E_INVAL;
goto error;
}
}
break;
}
default:
printf ("at index %d\n", i);
warn ("wk_compile: invalid wk-pred type\n");
ret = -E_INVAL;
goto error;
}
}
/* end the last term */
OVERRUN_CHECK;
v_retu (tag);
v_label (end_of_term);
v_retui (0);
v_end (NULL);
error:
/* have to do this even on error so that our caller can just call
wk_free to clean memory/ref counts up */
pred_pages[next_pp] = 0;
curenv->env_pred_pgs = pred_pages;
curenv->env_pred = (Spred)code;
return ret;
}
