entry sum_35() {
new_env(1, 0);
ildc(0);
load();
null();
oequal();
jz(label2);
ildc(0);
ret();
jmp(label3);
label2:
ildc(0);
load();
call(get_element_29);
ildc(0);
load();
ildc(1);
getfield();
call(sum_35);
iadd();
ret();
label3:
null();
ret();
}
entry length_34() {
new_env(1, 0);
ildc(0);
load();
null();
oequal();
jz(label0);
ildc(0);
ret();
jmp(label1);
label0:
ildc(1);
ildc(0);
load();
ildc(1);
getfield();
call(length_34);
iadd();
ret();
label1:
null();
ret();
}
entry print_36() {
new_env(1, 1);
ildc(1);
ildc(0);
load();
store();
pop();
label4:
ildc(1);
load();
null();
onotequal();
jz(label5);
ildc(1);
load();
call(get_element_29);
call(print_7);
pop();
ildc(4);
call(print_13);
pop();
ildc(1);
ildc(1);
load();
ildc(1);
getfield();
store();
pop();
jmp(label4);
label5:
ildc(5);
call(println_21);
pop();
null();
ret();
}
void divu1() {
// 整数オーバーフローは発生しないはず
enum { _1div, _end };
mov32_reg_mem(_ecx, _GPR_D(_rt, 0));
tst32_reg_reg(_ecx, _ecx);
jz(_end);
mov32_reg_mem(_eax, _GPR_D(_rs, 0));
xor32_reg_reg(_edx, _edx);
div_reg(_ecx);
mov32_reg_reg(_ecx, _edx);
// 商をloに格納
cdq();
mov32_mem_reg((u32)&g_cpu->m_lo.d[2], _eax);
mov32_mem_reg((u32)&g_cpu->m_lo.d[3], _edx);
mov32_reg_reg(_eax, _ecx);
// 余をhiに格納
cdq();
mov32_mem_reg((u32)&g_cpu->m_hi.d[2], _eax);
mov32_mem_reg((u32)&g_cpu->m_hi.d[3], _edx);
LABEL_DEF(_end);
}
void cnbint(
int i,
const double pos[][3],
const double vel[][3],
const double mass[],
int nnb,
int list[],
double f[3],
double fdot[3]){
const int NBMAX = 512;
assert(nnb <= NBMAX);
float xbuf[NBMAX] __attribute__ ((aligned(16)));
float ybuf[NBMAX] __attribute__ ((aligned(16)));
float zbuf[NBMAX] __attribute__ ((aligned(16)));
float vxbuf[NBMAX] __attribute__ ((aligned(16)));
float vybuf[NBMAX] __attribute__ ((aligned(16)));
float vzbuf[NBMAX] __attribute__ ((aligned(16)));
float mbuf[NBMAX] __attribute__ ((aligned(16)));
assert((unsigned long)xbuf % 16 == 0);
double xi = pos[i][0];
double yi = pos[i][1];
double zi = pos[i][2];
float vxi = vel[i][0];
float vyi = vel[i][1];
float vzi = vel[i][2];
for(int k=0; k<nnb; k++){
int j = list[k];
#if 1
int jj = list[k+4];
__builtin_prefetch(pos[jj]);
__builtin_prefetch(pos[jj]+2);
__builtin_prefetch(vel[jj]);
__builtin_prefetch(vel[jj]+2);
__builtin_prefetch(&mass[jj]);
#endif
#if 0
xbuf[k] = pos[j][0] - xi;
ybuf[k] = pos[j][1] - yi;
zbuf[k] = pos[j][2] - zi;
vxbuf[k] = vel[j][0] - vxi;
vybuf[k] = vel[j][1] - vyi;
vzbuf[k] = vel[j][2] - vzi;
mbuf[k] = mass[j];
#else
double xj = pos[j][0];
double yj = pos[j][1];
double zj = pos[j][2];
float vxj = vel[j][0];
float vyj = vel[j][1];
float vzj = vel[j][2];
float mj = mass[j];
xj -= xi;
yj -= yi;
zj -= zi;
vxj -= vxi;
vyj -= vyi;
vzj -= vzi;
xbuf[k] = xj;
ybuf[k] = yj;
zbuf[k] = zj;
vxbuf[k] = vxj;
vybuf[k] = vyj;
vzbuf[k] = vzj;
mbuf[k] = mj;
#endif
}
for(int k=nnb; k%4; k++){
xbuf[k] = 16.0f;
ybuf[k] = 16.0f;
zbuf[k] = 16.0f;
vxbuf[k] = 0.0f;
vybuf[k] = 0.0f;
vzbuf[k] = 0.0f;
mbuf[k] = 0.0f;
}
v4df ax(0.0), ay(0.0), az(0.0);
v4sf jx(0.0), jy(0.0), jz(0.0);
for(int k=0; k<nnb; k+=4){
v4sf dx(xbuf + k);
v4sf dy(ybuf + k);
v4sf dz(zbuf + k);
v4sf dvx(vxbuf + k);
v4sf dvy(vybuf + k);
v4sf dvz(vzbuf + k);
v4sf mj(mbuf + k);
v4sf r2 = dx*dx + dy*dy + dz*dz;
v4sf rv = dx*dvx + dy*dvy + dz*dvz;
rv *= v4sf(-3.0f);
// v4sf rinv1 = r2.rsqrt() & v4sf(mask);
#if 1
v4sf rinv1 = r2.rsqrt();
#else
v4sf rinv1 = v4sf(v4df(1.0) / v4df(r2).sqrt());
#endif
v4sf rinv2 = rinv1 * rinv1;
rv *= rinv2;
v4sf rinv3 = mj * rinv1 * rinv2;
dx *= rinv3; ax += v4df(dx);
dy *= rinv3; ay += v4df(dy);
dz *= rinv3; az += v4df(dz);
dvx *= rinv3; jx += dvx;
dvy *= rinv3; jy += dvy;
dvz *= rinv3; jz += dvz;
dx *= rv; jx += dx;
dy *= rv; jy += dy;
dz *= rv; jz += dz;
}
f[0] = ax.sum();
f[1] = ay.sum();
f[2] = az.sum();
fdot[0] = (v4df(jx)).sum();
fdot[1] = (v4df(jy)).sum();
fdot[2] = (v4df(jz)).sum();
assert(f[0] == f[0]);
assert(f[1] == f[1]);
assert(f[2] == f[2]);
assert(fdot[0] == fdot[0]);
assert(fdot[1] == fdot[1]);
assert(fdot[2] == fdot[2]);
}
void cnbint(
int i,
const double pos[][3],
const double vel[][3],
const double mass[],
int nnb,
int list[],
double f[3],
double fdot[3]){
const int NBMAX = 512;
assert(nnb <= NBMAX);
double xbuf[NBMAX] __attribute__ ((aligned(16)));
double ybuf[NBMAX] __attribute__ ((aligned(16)));
double zbuf[NBMAX] __attribute__ ((aligned(16)));
float vxbuf[NBMAX] __attribute__ ((aligned(16)));
float vybuf[NBMAX] __attribute__ ((aligned(16)));
float vzbuf[NBMAX] __attribute__ ((aligned(16)));
float mbuf[NBMAX] __attribute__ ((aligned(16)));
for(int k=0; k<nnb; k++){
int j = list[k];
xbuf[k] = pos[j][0];
ybuf[k] = pos[j][1];
zbuf[k] = pos[j][2];
vxbuf[k] = vel[j][0];
vybuf[k] = vel[j][1];
vzbuf[k] = vel[j][2];
mbuf[k] = mass[j];
}
for(int k=nnb; k%4; k++){
xbuf[k] = 16.0;
ybuf[k] = 16.0;
zbuf[k] = 16.0;
vxbuf[k] = 0.0f;
vybuf[k] = 0.0f;
vzbuf[k] = 0.0f;
mbuf[k] = 0.0f;
}
v4df xi(pos[i][0]);
v4df yi(pos[i][1]);
v4df zi(pos[i][2]);
v4sf vxi(vel[i][0]);
v4sf vyi(vel[i][1]);
v4sf vzi(vel[i][2]);
v4df ax(0.0), ay(0.0), az(0.0);
v4sf jx(0.0), jy(0.0), jz(0.0);
for(int k=0; k<nnb; k+=4){
v4df xj(xbuf + k);
v4df yj(ybuf + k);
v4df zj(zbuf + k);
v4sf vxj(vxbuf + k);
v4sf vyj(vybuf + k);
v4sf vzj(vzbuf + k);
v4sf mj(mbuf + k);
v4sf dx = v4sf::_v4sf(xj - xi);
v4sf dy = v4sf::_v4sf(yj - yi);
v4sf dz = v4sf::_v4sf(zj - zi);
v4sf dvx = vxj - vxi;
v4sf dvy = vyj - vyi;
v4sf dvz = vzj - vzi;
v4sf r2 = dx*dx + dy*dy + dz*dz;
v4sf rv = dx*dvx + dy*dvy + dz*dvz;
rv *= v4sf(-3.0f);
// v4sf rinv1 = r2.rsqrt() & v4sf(mask);
v4sf rinv1 = r2.rsqrt();
v4sf rinv2 = rinv1 * rinv1;
rv *= rinv2;
v4sf rinv3 = mj * rinv1 * rinv2;
dx *= rinv3; ax += v4df(dx);
dy *= rinv3; ay += v4df(dy);
dz *= rinv3; az += v4df(dz);
dvx *= rinv3; jx += dvx;
dvy *= rinv3; jy += dvy;
dvz *= rinv3; jz += dvz;
dx *= rv; jx += dx;
dy *= rv; jy += dy;
dz *= rv; jz += dz;
}
f[0] = ax.sum();
f[1] = ay.sum();
f[2] = az.sum();
fdot[0] = (v4df(jx)).sum();
fdot[1] = (v4df(jy)).sum();
fdot[2] = (v4df(jz)).sum();
assert(f[0] == f[0]);
assert(f[1] == f[1]);
assert(f[2] == f[2]);
assert(fdot[0] == fdot[0]);
assert(fdot[1] == fdot[1]);
assert(fdot[2] == fdot[2]);
}
void
codegen(xJIT *xjit)
{
xOperand *pputc, *pgetc, *stack;
char c;
const void *code;
int lstack[100], *lcur = lstack, lno = 0;
pputc = rbx;
pgetc = rbp;
stack = r12;
push(rbx);
push(rbp);
push(r12);
mov(pputc, rdi); /* putchar */
mov(pgetc, rsi); /* getchar */
mov(stack, rdx); /* stack */
while (~(c = getchar())) {
switch (c) {
case '+':
case '-':
{
int cnt;
cnt = countc(c);
if (cnt == 1) {
c == '+'
? inc(dword(stack))
: dec(dword(stack));
}
else {
addi(dword(stack), c == '+' ? cnt : -cnt);
}
}
break;
case '>':
case '<':
{
int cnt;
cnt = countc(c);
addi(stack, 4 * (c == '>' ? cnt : -cnt));
}
break;
case '.':
{
mov(rdi, dword(stack));
call(pputc,XJIT_CALL_OPERAND);
}
break;
case ',':
{
call(pgetc,XJIT_CALL_OPERAND);
mov(dword(stack), eax);
}
break;
case '[':
{
L(numl(lno, 'B')); /* backward label */
mov(eax, dword(stack));
test(eax, eax);
jz(numl(lno, 'F'), XJIT_LABEL_NEAR);
*lcur++ = lno++;
}
break;
case ']':
{
int no;
no = *--lcur;
jmp(numl(no, 'B'), XJIT_JMP_LABEL);
L(numl(no, 'F')); /* forward label */
}
break;
default:
break;
}
}
pop(r12);
pop(rbp);
pop(rbx);
ret();
}
entry main_40() {
new_env(0, 3);
ildc(1);
newobj(2);
ildc(0);
call(List_28);
store();
pop();
ildc(0);
ildc(1);
store();
pop();
label6:
ildc(0);
load();
ildc(6);
ilt();
jz(label7);
ildc(1);
ildc(1);
load();
ildc(0);
load();
call(push_front_30);
store();
pop();
label8:
ildc(0);
dup();
load();
swap();
dup();
load();
ildc(1);
iadd();
store();
pop();
pop();
jmp(label6);
label7:
ildc(1);
load();
call(print_36);
pop();
ildc(6);
call(print_13);
pop();
ildc(1);
load();
call(length_34);
call(println_15);
pop();
ildc(9);
call(print_13);
pop();
ildc(1);
load();
call(sum_35);
call(println_15);
pop();
ildc(2);
newobj(2);
ildc(1);
load();
call(get_element_29);
call(List_28);
store();
pop();
ildc(1);
ildc(1);
load();
call(pop_front_33);
store();
pop();
label9:
ildc(1);
load();
null();
onotequal();
jz(label10);
ildc(2);
ildc(2);
load();
ildc(1);
load();
call(get_element_29);
call(push_front_30);
store();
pop();
ildc(1);
ildc(1);
load();
call(pop_front_33);
store();
pop();
jmp(label9);
label10:
ildc(2);
load();
call(print_36);
pop();
ildc(11);
call(print_13);
pop();
ildc(2);
load();
call(length_34);
call(println_15);
pop();
ildc(14);
call(print_13);
pop();
ildc(2);
load();
call(sum_35);
call(println_15);
pop();
null();
ret();
}
