inline static unsigned int
count_leading_zeros (UTItype x)
{
qword c = si_clz (*(qword *) & x);
qword cmp0 = si_cgti (c, 31);
qword cmp1 = si_and (cmp0, si_shlqbyi (cmp0, 4));
qword cmp2 = si_and (cmp1, si_shlqbyi (cmp0, 8));
qword s = si_a (c, si_and (cmp0, si_shlqbyi (c, 4)));
s = si_a (s, si_and (cmp1, si_shlqbyi (c, 8)));
s = si_a (s, si_and (cmp2, si_shlqbyi (c, 12)));
return si_to_uint (s);
}
UTItype
__udivmodti4 (UTItype num, UTItype den, UTItype * rp)
{
qword shift =
si_from_uint (count_leading_zeros (den) - count_leading_zeros (num));
qword n0 = si_from_UTItype (num);
qword d0 = si_from_UTItype (den);
qword bit = si_andi (si_fsmbi (1), 1);
qword r0 = si_il (0);
qword m1 = si_fsmbi (0x000f);
qword mask, r1, n1;
d0 = si_shlqbybi (si_shlqbi (d0, shift), shift);
bit = si_shlqbybi (si_shlqbi (bit, shift), shift);
do
{
r1 = si_or (r0, bit);
// n1 = n0 - d0 in TImode
n1 = si_bg (d0, n0);
n1 = si_shlqbyi (n1, 4);
n1 = si_sf (m1, n1);
n1 = si_bgx (d0, n0, n1);
n1 = si_shlqbyi (n1, 4);
n1 = si_sf (m1, n1);
n1 = si_bgx (d0, n0, n1);
n1 = si_shlqbyi (n1, 4);
n1 = si_sf (m1, n1);
n1 = si_sfx (d0, n0, n1);
mask = si_fsm (si_cgti (n1, -1));
r0 = si_selb (r0, r1, mask);
n0 = si_selb (n0, n1, mask);
bit = si_rotqmbii (bit, -1);
d0 = si_rotqmbii (d0, -1);
}
while (si_to_uint (si_orx (bit)));
if (rp)
*rp = si_to_UTItype (n0);
return si_to_UTItype (r0);
}
/**
* Sort vertices from top to bottom.
* Compute area and determine front vs. back facing.
* Do coarse clip test against tile bounds
* \return FALSE if tri is totally outside tile, TRUE otherwise
*/
static boolean
setup_sort_vertices(const qword vs)
{
float area, sign;
#if DEBUG_VERTS
if (spu.init.id==0) {
fprintf(stderr, "SPU %u: Triangle:\n", spu.init.id);
print_vertex(v0);
print_vertex(v1);
print_vertex(v2);
}
#endif
{
/* Load the float values for various processing... */
const qword f0 = (qword)(((const struct vertex_header*)si_to_ptr(vs))->data[0]);
const qword f1 = (qword)(((const struct vertex_header*)si_to_ptr(si_rotqbyi(vs, 4)))->data[0]);
const qword f2 = (qword)(((const struct vertex_header*)si_to_ptr(si_rotqbyi(vs, 8)))->data[0]);
/* Check if triangle is completely outside the tile bounds
* Find the min and max x and y positions of the three poits */
const qword minf = min3fq(f0, f1, f2);
const qword maxf = max3fq(f0, f1, f2);
/* Compare min and max against cliprect vals */
const qword maxsmins = si_shufb(maxf, minf, SHUFB4(A,B,a,b));
const qword outside = si_fcgt(maxsmins, si_csflt(setup.cliprect, 0));
/* Use a little magic to work out of the tri is visible or not */
if(si_to_uint(si_xori(si_gb(outside), 0xc))) return FALSE;
/* determine bottom to top order of vertices */
/* A table of shuffle patterns for putting vertex_header pointers into
correct order. Quite magical. */
const qword sort_order_patterns[] = {
SHUFB4(A,B,C,C),
SHUFB4(C,A,B,C),
SHUFB4(A,C,B,C),
SHUFB4(B,C,A,C),
SHUFB4(B,A,C,C),
SHUFB4(C,B,A,C) };
/* Collate y values into two vectors for comparison.
Using only one shuffle constant! ;) */
const qword y_02_ = si_shufb(f0, f2, SHUFB4(0,B,b,C));
const qword y_10_ = si_shufb(f1, f0, SHUFB4(0,B,b,C));
const qword y_012 = si_shufb(y_02_, f1, SHUFB4(0,B,b,C));
const qword y_120 = si_shufb(y_10_, f2, SHUFB4(0,B,b,C));
/* Perform comparison: {y0,y1,y2} > {y1,y2,y0} */
const qword compare = si_fcgt(y_012, y_120);
/* Compress the result of the comparison into 4 bits */
const qword gather = si_gb(compare);
/* Subtract one to attain the index into the LUT. Magical. */
const unsigned int index = si_to_uint(gather) - 1;
/* Load the appropriate pattern and construct the desired vector. */
setup.vertex_headers = si_shufb(vs, vs, sort_order_patterns[index]);
/* Using the result of the comparison, set sign.
Very magical. */
sign = ((si_to_uint(si_cntb(gather)) == 2) ? 1.0f : -1.0f);
}
setup.ebot.ds = spu_sub(setup.vmid->data[0], setup.vmin->data[0]);
setup.emaj.ds = spu_sub(setup.vmax->data[0], setup.vmin->data[0]);
setup.etop.ds = spu_sub(setup.vmax->data[0], setup.vmid->data[0]);
/*
* Compute triangle's area. Use 1/area to compute partial
* derivatives of attributes later.
*/
area = setup.emaj.dx * setup.ebot.dy - setup.ebot.dx * setup.emaj.dy;
setup.oneOverArea = 1.0f / area;
/* The product of area * sign indicates front/back orientation (0/1).
* Just in case someone gets the bright idea of switching the front
* and back constants without noticing that we're assuming their
* values in this operation, also assert that the values are
* what we think they are.
*/
ASSERT(CELL_FACING_FRONT == 0);
ASSERT(CELL_FACING_BACK == 1);
setup.facing = (area * sign > 0.0f)
^ (spu.rasterizer.front_winding == PIPE_WINDING_CW);
return TRUE;
}