void
_jit_putargi(jit_state_t *_jit, jit_word_t u, jit_node_t *v)
{
jit_int32_t regno;
assert(v->code == jit_code_arg);
if (jit_arg_reg_p(v->u.w))
jit_movi(_I0 + v->u.w, u);
else {
regno = jit_get_reg(jit_class_gpr);
jit_movi(regno, u);
jit_stxi(v->u.w, JIT_FP, regno);
jit_unget_reg(regno);
}
}
int compile_pixel(int retreg, Cell* args) {
Cell* arg_x = car(args);
if (!arg_x) return argnum_error("(pixel x y color)");
args = cdr(args);
Cell* arg_y = car(args);
if (!arg_y) return argnum_error("(pixel x y color)");
args = cdr(args);
Cell* arg_c = car(args);
if (!arg_c) return argnum_error("(pixel x y color)");
compile_arg(JIT_R0, arg_c, TAG_PURE_INT);
stack_push(JIT_R0, &stack_ptr);
compile_arg(JIT_R0, arg_y, TAG_PURE_INT);
stack_push(JIT_R0, &stack_ptr);
compile_arg(JIT_R0, arg_x, TAG_PURE_INT);
stack_push(JIT_R0, &stack_ptr);
jit_prepare();
stack_pop(JIT_R0, &stack_ptr);
jit_pushargr(JIT_R0);
stack_pop(JIT_R0, &stack_ptr);
jit_pushargr(JIT_R0);
stack_pop(JIT_R0, &stack_ptr);
jit_pushargr(JIT_R0);
jit_finishi(machine_video_set_pixel);
jit_movi(retreg, 0);
return 1;
}
int compile_blit_mono_inv(int retreg, Cell* args) {
compile_arg(JIT_R0, car(args), TAG_BYTES);
jit_ldr(JIT_R0, JIT_R0); // load bytes addr
stack_push(JIT_R0, &stack_ptr);
compile_int_arg();
compile_int_arg();
compile_int_arg();
compile_int_arg();
compile_int_arg();
compile_int_arg();
compile_int_arg();
compile_int_arg();
jit_prepare();
push_stack_arg();
push_stack_arg();
push_stack_arg();
push_stack_arg();
push_stack_arg();
push_stack_arg();
push_stack_arg();
push_stack_arg();
push_stack_arg(); // pop bytes addr
jit_finishi(blit_vector1_invert);
jit_movi(retreg, 0);
return 1;
}
void
_jit_pushargi(jit_state_t *_jit, jit_word_t u)
{
jit_int32_t regno;
if (jit_arg_reg_p(_jitc->function->call.argi)) {
jit_movi(_O0 + _jitc->function->call.argi, u);
++_jitc->function->call.argi;
}
else {
regno = jit_get_reg(jit_class_gpr);
jit_movi(regno, u);
jit_stxi(_jitc->function->call.size + stack_framesize, JIT_SP, regno);
jit_unget_reg(regno);
_jitc->function->call.size += sizeof(jit_word_t);
}
}
int compile_flip(int retreg) {
jit_prepare();
jit_finishi(machine_video_flip);
jit_movi(retreg, 0);
return 1;
}
jit_value pop_int(Frame* frame) {
jit_value reg = pop_reg(frame);
struct jit* jit = frame->jit;
jit_ldxi(jit, frame->accum[1], reg, offsetof(YValue, type), sizeof(void*));
jit_movi(jit, frame->accum[2], 0);
jit_op* not_int_jump = jit_bner(jit, (intptr_t) JIT_FORWARD, frame->accum[1], frame->IntType);
jit_ldxi(jit, frame->accum[2], reg, offsetof(YInteger, value), sizeof(int64_t));
jit_patch(jit, not_int_jump);
jit_movr(jit, frame->accum[0], frame->accum[2]);
return frame->accum[0];
}
static void
load_exception(jit_int32_t regval, oword_t offset)
{
if (otype(current_record) == t_namespace) {
if (GPR[GLOBAL] != JIT_NOREG)
jit_addi(regval, GPR[GLOBAL], offset);
else
jit_movi(regval, (oword_t)gd + offset);
}
else {
if (GPR[FRAME] != JIT_NOREG)
jit_addi(regval, GPR[FRAME], offset);
else {
jit_ldxi(regval, JIT_V0, offsetof(othread_t, fp));
jit_addi(regval, regval, offset);
}
}
}
jit_node_t *compile_rpn(char *expr)
{
jit_node_t *in, *fn;
int stack_base, stack_ptr;
fn = jit_note(NULL, 0);
jit_prolog();
in = jit_arg();
stack_ptr = stack_base = jit_allocai (32 * sizeof (int));
jit_getarg_i(JIT_R2, in);
while (*expr) {
char buf[32];
int n;
if (sscanf(expr, "%[0-9]%n", buf, &n)) {
expr += n - 1;
stack_push(JIT_R0, &stack_ptr);
jit_movi(JIT_R0, atoi(buf));
} else if (*expr == 'x') {
stack_push(JIT_R0, &stack_ptr);
jit_movr(JIT_R0, JIT_R2);
} else if (*expr == '+') {
stack_pop(JIT_R1, &stack_ptr);
jit_addr(JIT_R0, JIT_R1, JIT_R0);
} else if (*expr == '-') {
stack_pop(JIT_R1, &stack_ptr);
jit_subr(JIT_R0, JIT_R1, JIT_R0);
} else if (*expr == '*') {
stack_pop(JIT_R1, &stack_ptr);
jit_mulr(JIT_R0, JIT_R1, JIT_R0);
} else if (*expr == '/') {
stack_pop(JIT_R1, &stack_ptr);
jit_divr(JIT_R0, JIT_R1, JIT_R0);
} else {
fprintf(stderr, "cannot compile: %s\n", expr);
abort();
}
++expr;
}
jit_retr(JIT_R0);
jit_epilog();
return fn;
}
void jit_addi(int dreg, int imm) {
jit_movi(9,imm);
jit_addr(dreg,9);
}
void jit_lea(int reg, void* addr) {
jit_movi(reg, (uint32_t)addr);
}
jit_pointer_t compile(const std::vector<Oper>& ops, jit_word_t *memory, const bool flush = true)
{
jit_prolog();
jit_movi(JIT_V0, reinterpret_cast<jit_word_t>(memory));
jit_movi(JIT_V1, 0);
std::stack<Loop> loops;
jit_node_t* start = jit_note(__FILE__, __LINE__);
for ( size_t n=0; n<ops.size(); ++n ) {
switch ( ops[n].code ) {
case '<':
jit_str(JIT_V0, JIT_V1);
jit_subi(JIT_V0, JIT_V0, ops[n].count * sizeof(jit_word_t));
jit_ldr(JIT_V1, JIT_V0);
break;
case '>':
jit_str(JIT_V0, JIT_V1);
jit_addi(JIT_V0, JIT_V0, ops[n].count * sizeof(jit_word_t));
jit_ldr(JIT_V1, JIT_V0);
break;
case 'z':
jit_movi(JIT_V1, 0);
break;
case '+':
jit_addi(JIT_V1, JIT_V1, ops[n].count);
break;
case '-':
jit_subi(JIT_V1, JIT_V1, ops[n].count);
break;
case '.':
jit_prepare();
jit_pushargr(JIT_V1);
jit_finishi(reinterpret_cast<jit_pointer_t>(putchar));
if ( flush ) {
jit_prepare();
jit_pushargi(reinterpret_cast<jit_word_t>(stdout));
jit_finishi(reinterpret_cast<jit_pointer_t>(fflush));
}
break;
case ',':
jit_prepare();
jit_finishi(reinterpret_cast<jit_pointer_t>(getchar));
jit_retval(JIT_V1);
break;
case '[': {
Loop loop;
loop.end = jit_forward();
jit_node_t *j = jit_beqi(JIT_V1, 0);
jit_patch_at(j, loop.end);
loop.body = jit_label();
loops.push(loop);
} break;
case ']': {
Loop loop = loops.top();
jit_node_t *j = jit_bnei(JIT_V1, 0);
jit_patch_at(j, loop.body);
jit_link(loop.end);
loops.pop();
break;
}
default:
break;
}
}
jit_node_t* stop = jit_note(__FILE__, __LINE__);
jit_ret();
jit_epilog();
jit_pointer_t r = jit_emit();
fprintf(stderr, "compiled to %zu bytes\n", (char*)jit_address(stop) -
(char*)jit_address(start));
return r;
}
void
_jit_reti(jit_state_t *_jit, jit_word_t u)
{
jit_movi(JIT_RET, u);
jit_ret();
}
void
run_gnulightning(void)
{
struct bfi * n = bfprog;
int maxstack = 0, stackptr = 0;
char *strbuf = 0;
size_t maxstrlen = 0;
#ifdef GNULIGHTv1
jit_insn** loopstack = 0;
jit_insn *codeBuffer;
void * startptr;
int argp;
#endif
#ifdef GNULIGHTv2
jit_node_t *start, *end; /* For size of code */
jit_node_t** loopstack = 0;
jit_node_t *argp;
#endif
if (cell_size == 8) tape_step = 1; else
tape_step = sizeof(int);
#ifdef GNULIGHTv1
/* TODO: Use mmap for allocating memory, the x86 execute protection
* bit is on the segment so Linux has to say thay everything below
* a specific address is executable. If you ask mmap for executable
* memory it can put it below the current value. The mprotect()
* function can't do this.
*/
if (total_nodes < 4096)
codeBuffer = malloc(65536);
else
codeBuffer = malloc(16 * total_nodes);
save_ptr_for_free(codeBuffer);
codeptr = (codeptr_t) jit_set_ip(codeBuffer).vptr;
startptr = jit_get_ip().ptr;
/* Function call prolog */
jit_prolog(1);
/* Get the data area pointer */
argp = jit_arg_p();
jit_getarg_p(REG_P, argp);
#endif
#ifdef GNULIGHTv2
init_jit(NULL); // argv[0]);
_jit = jit_new_state();
start = jit_note(__FILE__, __LINE__);
jit_prolog();
/* Get the data area pointer */
argp = jit_arg();
jit_getarg(REG_P, argp);
#endif
while(n)
{
switch(n->type)
{
case T_MOV:
if (acc_loaded)
acc_offset -= n->count;
jit_addi(REG_P, REG_P, n->count * tape_step);
break;
case T_ADD:
load_acc_offset(n->offset);
set_acc_offset(n->offset);
jit_addi(REG_ACC, REG_ACC, n->count);
break;
case T_SET:
set_acc_offset(n->offset);
if (acc_const && acc_const_val == n->count) {
;
} else if (acc_const && acc_const_val+1 == n->count) {
jit_addi(REG_ACC, REG_ACC, 1);
} else if (acc_const && acc_const_val-1 == n->count) {
jit_addi(REG_ACC, REG_ACC, -1);
} else {
jit_movi(REG_ACC, n->count);
}
acc_const = 1;
acc_const_val = n->count;
break;
case T_CALC:
if (n->offset == n->offset2 && n->count2 == 1) {
load_acc_offset(n->offset);
set_acc_offset(n->offset);
if (n->count)
jit_addi(REG_ACC, REG_ACC, n->count);
} else if (n->count2 != 0) {
load_acc_offset(n->offset2);
set_acc_offset(n->offset);
if (n->count2 == -1)
jit_negr(REG_ACC, REG_ACC);
else if (n->count2 != 1)
jit_muli(REG_ACC, REG_ACC, n->count2);
if (n->count)
jit_addi(REG_ACC, REG_ACC, n->count);
} else {
clean_acc();
set_acc_offset(n->offset);
jit_movi(REG_ACC, n->count);
if (n->count2 != 0) {
if (tape_step > 1)
jit_ldxi_i(REG_A1, REG_P, n->offset2 * tape_step);
else
jit_ldxi_uc(REG_A1, REG_P, n->offset2);
if (n->count2 == -1)
jit_negr(REG_A1, REG_A1);
else if (n->count2 != 1)
jit_muli(REG_A1, REG_A1, n->count2);
jit_addr(REG_ACC, REG_ACC, REG_A1);
}
}
if (n->count3 != 0) {
if (tape_step > 1)
jit_ldxi_i(REG_A1, REG_P, n->offset3 * tape_step);
else
jit_ldxi_uc(REG_A1, REG_P, n->offset3);
if (n->count3 == -1)
jit_negr(REG_A1, REG_A1);
else if (n->count3 != 1)
jit_muli(REG_A1, REG_A1, n->count3);
jit_addr(REG_ACC, REG_ACC, REG_A1);
}
break;
case T_IF: case T_MULT: case T_CMULT:
case T_WHL:
load_acc_offset(n->offset);
clean_acc();
acc_const = acc_loaded = 0;
if (stackptr >= maxstack) {
loopstack = realloc(loopstack,
((maxstack+=32)+2)*sizeof(*loopstack));
if (loopstack == 0) {
perror("loop stack realloc failure");
exit(1);
}
}
if (cell_mask > 0 && acc_hi_dirty) {
if (cell_mask == 0xFF)
jit_extr_uc(REG_ACC,REG_ACC);
else
jit_andi(REG_ACC, REG_ACC, cell_mask);
}
#ifdef GNULIGHTv1
loopstack[stackptr] = jit_beqi_i(jit_forward(), REG_ACC, 0);
loopstack[stackptr+1] = jit_get_label();
#endif
#ifdef GNULIGHTv2
loopstack[stackptr] = jit_beqi(REG_ACC, 0);
loopstack[stackptr+1] = jit_label();
#endif
stackptr += 2;
break;
case T_END:
load_acc_offset(n->offset);
clean_acc();
stackptr -= 2;
if (stackptr < 0) {
fprintf(stderr, "Code gen failure: Stack pointer negative.\n");
exit(1);
}
if (cell_mask > 0 && acc_hi_dirty) {
if (cell_mask == 0xFF)
jit_extr_uc(REG_ACC,REG_ACC);
else
jit_andi(REG_ACC, REG_ACC, cell_mask);
}
#ifdef GNULIGHTv1
jit_bnei_i(loopstack[stackptr+1], REG_ACC, 0);
jit_patch(loopstack[stackptr]);
#endif
#ifdef GNULIGHTv2
{
jit_node_t *ref;
ref = jit_bnei(REG_ACC, 0);
jit_patch_at(ref, loopstack[stackptr+1]);
jit_patch(loopstack[stackptr]);
}
#endif
break;
case T_ENDIF:
clean_acc();
acc_const = acc_loaded = 0;
stackptr -= 2;
if (stackptr < 0) {
fprintf(stderr, "Code gen failure: Stack pointer negative.\n");
exit(1);
}
jit_patch(loopstack[stackptr]);
break;
case T_PRT:
clean_acc();
load_acc_offset(n->offset);
acc_loaded = 0;
#ifdef GNULIGHTv1
jit_prepare_i(1);
jit_pusharg_i(REG_ACC);
jit_finish(putch);
#endif
#ifdef GNULIGHTv2
jit_prepare();
jit_pushargr(REG_ACC);
jit_finishi(putch);
#endif
break;
case T_CHR:
clean_acc();
acc_const = acc_loaded = 0;
if (n->count <= 0 || (n->count >= 127 && iostyle == 1) ||
!n->next || n->next->type != T_CHR) {
jit_movi(REG_ACC, n->count);
#ifdef GNULIGHTv1
jit_prepare_i(1);
jit_pusharg_i(REG_ACC);
jit_finish(putch);
#endif
#ifdef GNULIGHTv2
jit_prepare();
jit_pushargr(REG_ACC);
jit_finishi(putch);
#endif
} else {
unsigned i = 0;
struct bfi * v = n;
char *s;
while(v->next && v->next->type == T_CHR &&
v->next->count > 0 &&
(v->next->count < 127 || iostyle != 1)) {
if (i+2 > maxstrlen) {
if (maxstrlen) maxstrlen *= 2; else maxstrlen = 4096;
strbuf = realloc(strbuf, maxstrlen);
if (!strbuf) {
fprintf(stderr, "Reallocate of string buffer failed\n");
exit(42);
}
}
strbuf[i++] = (char) /*GCC -Wconversion*/ v->count;
n = v;
v = v->next;
}
strbuf[i] = 0;
s = strdup(strbuf);
if (!s) {
fprintf(stderr, "Save of string failed\n");
exit(43);
}
save_ptr_for_free(s);
#ifdef GNULIGHTv1
jit_movi_p(REG_ACC, s);
jit_prepare_i(1);
jit_pusharg_i(REG_ACC);
jit_finish(puts_without_nl);
#endif
#ifdef GNULIGHTv2
jit_prepare();
jit_pushargi((jit_word_t) s);
jit_finishi(puts_without_nl);
#endif
}
break;
case T_INP:
load_acc_offset(n->offset);
set_acc_offset(n->offset);
#ifdef GNULIGHTv1
jit_prepare_i(1);
jit_pusharg_i(REG_ACC);
jit_finish(getch);
jit_retval_i(REG_ACC);
#endif
#ifdef GNULIGHTv2
jit_prepare();
jit_pushargr(REG_ACC);
jit_finishi(getch);
jit_retval(REG_ACC);
#endif
break;
case T_STOP:
#ifdef GNULIGHTv1
jit_prepare_i(0);
jit_finish(failout);
#endif
#ifdef GNULIGHTv2
jit_prepare();
jit_finishi(failout);
#endif
break;
case T_NOP:
case T_DUMP:
fprintf(stderr, "Warning on code generation: "
"%s node: ptr+%d, cnt=%d, @(%d,%d).\n",
tokennames[n->type],
n->offset, n->count, n->line, n->col);
break;
default:
fprintf(stderr, "Code gen error: "
"%s\t"
"%d:%d, %d:%d, %d:%d\n",
tokennames[n->type],
n->offset, n->count,
n->offset2, n->count2,
n->offset3, n->count3);
exit(1);
}
n=n->next;
#if 0 /*def GNULIGHTv2 */
if(n && enable_trace) {
char *p, buf[250];
clean_acc();
acc_loaded = 0;
sprintf(buf, "@(%d,%d)\n", n->line, n->col);
p = strdup(buf);
save_ptr_for_free(p);
jit_prepare();
jit_pushargi((jit_word_t) p);
jit_finishi(puts_without_nl);
}
#endif
#ifdef GNULIGHTv1
/* TODO -- Check for codeBuffer overflow (add jmp to new) */
#endif
}
jit_ret();
if (strbuf) { maxstrlen = 0; free(strbuf); strbuf = 0; }
delete_tree();
#ifdef GNULIGHTv1
jit_flush_code(startptr, jit_get_ip().ptr);
if (verbose)
fprintf(stderr, "Generated %d bytes of V1 GNU Lightning code, running\n",
(int)(jit_get_ip().ptr - (char*)startptr));
start_runclock();
codeptr(map_hugeram());
finish_runclock(&run_time, &io_time);
#endif
#ifdef GNULIGHTv2
jit_epilog();
end = jit_note(__FILE__, __LINE__);
codeptr = jit_emit();
if (verbose)
fprintf(stderr, "Generated %d bytes of V2 GNU Lightning code, running\n",
(int)((char*)jit_address(end) - (char*)jit_address(start)));
jit_clear_state();
// jit_disassemble();
start_runclock();
codeptr(map_hugeram());
finish_runclock(&run_time, &io_time);
#endif
#if 0
/* This code writes the generated instructions to a file
* so we can disassemble it using: ndisasm -b 32/64 code.bin */
{
#ifdef GNULIGHTv1
char *p = startptr;
int s = jit_get_ip().ptr - p;
#endif
#ifdef GNULIGHTv2
char *p = (char*)jit_address(start);
int s = (char*)jit_address(end) - p;
#endif
FILE *fp = fopen("code.bin", "w");
int i;
for (i = 0; i < s; ++i) {
fputc(p[i], fp);
}
fclose(fp);
}
#endif
#ifdef GNULIGHTv2
jit_destroy_state();
finish_jit();
#endif
codeptr = 0;
if (loopstack) { free(loopstack); loopstack = 0; }
free_saved_memory();
}
