/**
* Render a horizontal span of quads
*/
static void flush_spans( struct setup_context *setup )
{
const int step = 16;
const int xleft0 = setup->span.left[0];
const int xleft1 = setup->span.left[1];
const int xright0 = setup->span.right[0];
const int xright1 = setup->span.right[1];
struct quad_stage *pipe = setup->softpipe->quad.first;
int minleft = block_x(MIN2(xleft0, xleft1));
int maxright = MAX2(xright0, xright1);
int x;
for (x = minleft; x < maxright; x += step) {
unsigned skip_left0 = CLAMP(xleft0 - x, 0, step);
unsigned skip_left1 = CLAMP(xleft1 - x, 0, step);
unsigned skip_right0 = CLAMP(x + step - xright0, 0, step);
unsigned skip_right1 = CLAMP(x + step - xright1, 0, step);
unsigned lx = x;
unsigned q = 0;
unsigned skipmask_left0 = (1U << skip_left0) - 1U;
unsigned skipmask_left1 = (1U << skip_left1) - 1U;
/* These calculations fail when step == 32 and skip_right == 0.
*/
unsigned skipmask_right0 = ~0U << (unsigned)(step - skip_right0);
unsigned skipmask_right1 = ~0U << (unsigned)(step - skip_right1);
unsigned mask0 = ~skipmask_left0 & ~skipmask_right0;
unsigned mask1 = ~skipmask_left1 & ~skipmask_right1;
if (mask0 | mask1) {
do {
unsigned quadmask = (mask0 & 3) | ((mask1 & 3) << 2);
if (quadmask) {
setup->quad[q].input.x0 = lx;
setup->quad[q].input.y0 = setup->span.y;
setup->quad[q].input.facing = setup->facing;
setup->quad[q].inout.mask = quadmask;
setup->quad_ptrs[q] = &setup->quad[q];
q++;
}
mask0 >>= 2;
mask1 >>= 2;
lx += 2;
} while (mask0 | mask1);
pipe->run( pipe, setup->quad_ptrs, q );
}
}
setup->span.y = 0;
setup->span.right[0] = 0;
setup->span.right[1] = 0;
setup->span.left[0] = 1000000; /* greater than right[0] */
setup->span.left[1] = 1000000; /* greater than right[1] */
}
/**
* Render a horizontal span of quads
*/
static void flush_spans( struct setup_context *setup )
{
const int step = TILE_VECTOR_WIDTH;
const int xleft0 = setup->span.left[0];
const int xleft1 = setup->span.left[1];
const int xright0 = setup->span.right[0];
const int xright1 = setup->span.right[1];
int minleft = block_x(MIN2(xleft0, xleft1));
int maxright = MAX2(xright0, xright1);
int x;
for (x = minleft; x < maxright; x += step) {
unsigned skip_left0 = CLAMP(xleft0 - x, 0, step);
unsigned skip_left1 = CLAMP(xleft1 - x, 0, step);
unsigned skip_right0 = CLAMP(x + step - xright0, 0, step);
unsigned skip_right1 = CLAMP(x + step - xright1, 0, step);
unsigned lx = x;
const unsigned nr_quads = TILE_VECTOR_HEIGHT*TILE_VECTOR_WIDTH/QUAD_SIZE;
unsigned q = 0;
unsigned skipmask_left0 = (1U << skip_left0) - 1U;
unsigned skipmask_left1 = (1U << skip_left1) - 1U;
/* These calculations fail when step == 32 and skip_right == 0.
*/
unsigned skipmask_right0 = ~0U << (unsigned)(step - skip_right0);
unsigned skipmask_right1 = ~0U << (unsigned)(step - skip_right1);
unsigned mask0 = ~skipmask_left0 & ~skipmask_right0;
unsigned mask1 = ~skipmask_left1 & ~skipmask_right1;
if (mask0 | mask1) {
for(q = 0; q < nr_quads; ++q) {
unsigned quadmask = (mask0 & 3) | ((mask1 & 3) << 2);
setup->quad[q].input.x0 = lx;
setup->quad[q].input.y0 = setup->span.y;
setup->quad[q].inout.mask = quadmask;
setup->quad_ptrs[q] = &setup->quad[q];
mask0 >>= 2;
mask1 >>= 2;
lx += 2;
}
assert(!(mask0 | mask1));
shade_quads(setup->llvmpipe, setup->quad_ptrs, nr_quads );
}
}
setup->span.y = 0;
setup->span.right[0] = 0;
setup->span.right[1] = 0;
setup->span.left[0] = 1000000; /* greater than right[0] */
setup->span.left[1] = 1000000; /* greater than right[1] */
}
/**
* Emit a quad (pass to next stage) with clipping.
*/
static INLINE void
clip_emit_quad( struct setup_context *setup, struct quad_header *quad )
{
quad_clip( setup, quad );
if (quad->inout.mask) {
struct llvmpipe_context *lp = setup->llvmpipe;
#if 1
/* XXX: The blender expects 4 quads. This is far from efficient, but
* until we codegenerate single-quad variants of the fragment pipeline
* we need this hack. */
const unsigned nr_quads = TILE_VECTOR_HEIGHT*TILE_VECTOR_WIDTH/QUAD_SIZE;
struct quad_header quads[4];
struct quad_header *quad_ptrs[4];
int x0 = block_x(quad->input.x0);
unsigned i;
assert(nr_quads == 4);
for(i = 0; i < nr_quads; ++i) {
int x = x0 + 2*i;
if(x == quad->input.x0)
memcpy(&quads[i], quad, sizeof quads[i]);
else {
memset(&quads[i], 0, sizeof quads[i]);
quads[i].input.x0 = x;
quads[i].input.y0 = quad->input.y0;
quads[i].coef = quad->coef;
}
quad_ptrs[i] = &quads[i];
}
shade_quads( lp, quad_ptrs, nr_quads );
#else
shade_quads( lp, &quad, 1 );
#endif
}
}
