static void offset_tri( struct draw_stage *stage,
struct prim_header *header )
{
struct prim_header tmp;
tmp.det = header->det;
tmp.flags = header->flags;
tmp.pad = header->pad;
tmp.v[0] = dup_vert(stage, header->v[0], 0);
tmp.v[1] = dup_vert(stage, header->v[1], 1);
tmp.v[2] = dup_vert(stage, header->v[2], 2);
do_offset_tri( stage, &tmp );
}
static void flatshade_tri_2( struct draw_stage *stage,
struct prim_header *header )
{
struct prim_header tmp;
tmp.det = header->det;
tmp.flags = header->flags;
tmp.pad = header->pad;
tmp.v[0] = dup_vert(stage, header->v[0], 0);
tmp.v[1] = dup_vert(stage, header->v[1], 1);
tmp.v[2] = header->v[2];
copy_colors2(stage, tmp.v[0], tmp.v[1], tmp.v[2]);
stage->next->tri( stage->next, &tmp );
}
static void
emit_segment(struct draw_stage *stage, struct prim_header *header,
float t0, float t1)
{
struct vertex_header *v0new = dup_vert(stage, header->v[0], 0);
struct vertex_header *v1new = dup_vert(stage, header->v[1], 1);
struct prim_header newprim = *header;
if (t0 > 0.0) {
screen_interp( stage->draw, v0new, t0, header->v[0], header->v[1] );
newprim.v[0] = v0new;
}
if (t1 < 1.0) {
screen_interp( stage->draw, v1new, t1, header->v[0], header->v[1] );
newprim.v[1] = v1new;
}
stage->next->line( stage->next, &newprim );
}
static void flatshade_line_1( struct draw_stage *stage,
struct prim_header *header )
{
struct prim_header tmp;
tmp.v[0] = dup_vert(stage, header->v[0], 0);
tmp.v[1] = header->v[1];
copy_colors(stage, tmp.v[0], tmp.v[1]);
stage->next->line( stage->next, &tmp );
}
/**
* Copy back color(s) to front color(s).
*/
static INLINE struct vertex_header *
copy_bfc( struct twoside_stage *twoside,
const struct vertex_header *v,
unsigned idx )
{
struct vertex_header *tmp = dup_vert( &twoside->stage, v, idx );
if (twoside->attrib_back0) {
COPY_4FV(tmp->data[twoside->attrib_front0],
tmp->data[twoside->attrib_back0]);
}
if (twoside->attrib_back1) {
COPY_4FV(tmp->data[twoside->attrib_front1],
tmp->data[twoside->attrib_back1]);
}
return tmp;
}
/* Clip a triangle against the viewport and user clip planes.
*/
static void
do_clip_tri( struct draw_stage *stage,
struct prim_header *header,
unsigned clipmask )
{
struct clip_stage *clipper = clip_stage( stage );
struct vertex_header *a[MAX_CLIPPED_VERTICES];
struct vertex_header *b[MAX_CLIPPED_VERTICES];
struct vertex_header **inlist = a;
struct vertex_header **outlist = b;
unsigned tmpnr = 0;
unsigned n = 3;
unsigned i;
boolean aEdges[MAX_CLIPPED_VERTICES];
boolean bEdges[MAX_CLIPPED_VERTICES];
boolean *inEdges = aEdges;
boolean *outEdges = bEdges;
inlist[0] = header->v[0];
inlist[1] = header->v[1];
inlist[2] = header->v[2];
/*
* Note: at this point we can't just use the per-vertex edge flags.
* We have to observe the edge flag bits set in header->flags which
* were set during primitive decomposition. Put those flags into
* an edge flags array which parallels the vertex array.
* Later, in the 'unfilled' pipeline stage we'll draw the edge if both
* the header.flags bit is set AND the per-vertex edgeflag field is set.
*/
inEdges[0] = !!(header->flags & DRAW_PIPE_EDGE_FLAG_0);
inEdges[1] = !!(header->flags & DRAW_PIPE_EDGE_FLAG_1);
inEdges[2] = !!(header->flags & DRAW_PIPE_EDGE_FLAG_2);
while (clipmask && n >= 3) {
const unsigned plane_idx = ffs(clipmask)-1;
const boolean is_user_clip_plane = plane_idx >= 6;
const float *plane = clipper->plane[plane_idx];
struct vertex_header *vert_prev = inlist[0];
boolean *edge_prev = &inEdges[0];
float dp_prev = dot4( vert_prev->clip, plane );
unsigned outcount = 0;
clipmask &= ~(1<<plane_idx);
assert(n < MAX_CLIPPED_VERTICES);
if (n >= MAX_CLIPPED_VERTICES)
return;
inlist[n] = inlist[0]; /* prevent rotation of vertices */
inEdges[n] = inEdges[0];
for (i = 1; i <= n; i++) {
struct vertex_header *vert = inlist[i];
boolean *edge = &inEdges[i];
float dp = dot4( vert->clip, plane );
if (!IS_NEGATIVE(dp_prev)) {
assert(outcount < MAX_CLIPPED_VERTICES);
if (outcount >= MAX_CLIPPED_VERTICES)
return;
outEdges[outcount] = *edge_prev;
outlist[outcount++] = vert_prev;
}
if (DIFFERENT_SIGNS(dp, dp_prev)) {
struct vertex_header *new_vert;
boolean *new_edge;
assert(tmpnr < MAX_CLIPPED_VERTICES + 1);
if (tmpnr >= MAX_CLIPPED_VERTICES + 1)
return;
new_vert = clipper->stage.tmp[tmpnr++];
assert(outcount < MAX_CLIPPED_VERTICES);
if (outcount >= MAX_CLIPPED_VERTICES)
return;
new_edge = &outEdges[outcount];
outlist[outcount++] = new_vert;
if (IS_NEGATIVE(dp)) {
/* Going out of bounds. Avoid division by zero as we
* know dp != dp_prev from DIFFERENT_SIGNS, above.
*/
float t = dp / (dp - dp_prev);
interp( clipper, new_vert, t, vert, vert_prev );
/* Whether or not to set edge flag for the new vert depends
* on whether it's a user-defined clipping plane. We're
* copying NVIDIA's behaviour here.
*/
if (is_user_clip_plane) {
/* we want to see an edge along the clip plane */
*new_edge = TRUE;
new_vert->edgeflag = TRUE;
}
else {
/* we don't want to see an edge along the frustum clip plane */
*new_edge = *edge_prev;
new_vert->edgeflag = FALSE;
}
}
else {
/* Coming back in.
*/
float t = dp_prev / (dp_prev - dp);
interp( clipper, new_vert, t, vert_prev, vert );
/* Copy starting vert's edgeflag:
*/
new_vert->edgeflag = vert_prev->edgeflag;
*new_edge = *edge_prev;
}
}
vert_prev = vert;
edge_prev = edge;
dp_prev = dp;
}
/* swap in/out lists */
{
struct vertex_header **tmp = inlist;
inlist = outlist;
outlist = tmp;
n = outcount;
}
{
boolean *tmp = inEdges;
inEdges = outEdges;
outEdges = tmp;
}
}
/* If flat-shading, copy provoking vertex color to polygon vertex[0]
*/
if (n >= 3) {
if (clipper->flat) {
if (stage->draw->rasterizer->flatshade_first) {
if (inlist[0] != header->v[0]) {
assert(tmpnr < MAX_CLIPPED_VERTICES + 1);
if (tmpnr >= MAX_CLIPPED_VERTICES + 1)
return;
inlist[0] = dup_vert(stage, inlist[0], tmpnr++);
copy_colors(stage, inlist[0], header->v[0]);
}
}
else {
if (inlist[0] != header->v[2]) {
assert(tmpnr < MAX_CLIPPED_VERTICES + 1);
if (tmpnr >= MAX_CLIPPED_VERTICES + 1)
return;
inlist[0] = dup_vert(stage, inlist[0], tmpnr++);
copy_colors(stage, inlist[0], header->v[2]);
}
}
}
/* Emit the polygon as triangles to the setup stage:
*/
emit_poly( stage, inlist, inEdges, n, header );
}
}
/**
* Draw an AA point by drawing a quad.
*/
static void
aapoint_point(struct draw_stage *stage, struct prim_header *header)
{
const struct aapoint_stage *aapoint = aapoint_stage(stage);
struct prim_header tri;
struct vertex_header *v[4];
const uint tex_slot = aapoint->tex_slot;
const uint pos_slot = aapoint->pos_slot;
float radius, *pos, *tex;
uint i;
float k;
if (aapoint->psize_slot >= 0) {
radius = 0.5f * header->v[0]->data[aapoint->psize_slot][0];
}
else {
radius = aapoint->radius;
}
/*
* Note: the texcoords (generic attrib, really) we use are special:
* The S and T components simply vary from -1 to +1.
* The R component is k, below.
* The Q component is 1.0 and will used as a handy constant in the
* fragment shader.
*/
/*
* k is the threshold distance from the point's center at which
* we begin alpha attenuation (the coverage value).
* Operating within a unit circle, we'll compute the fragment's
* distance 'd' from the center point using the texcoords.
* IF d > 1.0 THEN
* KILL fragment
* ELSE IF d > k THEN
* compute coverage in [0,1] proportional to d in [k, 1].
* ELSE
* coverage = 1.0; // full coverage
* ENDIF
*
* Note: the ELSEIF and ELSE clauses are actually implemented with CMP to
* avoid using IF/ELSE/ENDIF TGSI opcodes.
*/
#if !NORMALIZE
k = 1.0f / radius;
k = 1.0f - 2.0f * k + k * k;
#else
k = 1.0f - 1.0f / radius;
#endif
/* allocate/dup new verts */
for (i = 0; i < 4; i++) {
v[i] = dup_vert(stage, header->v[0], i);
}
/* new verts */
pos = v[0]->data[pos_slot];
pos[0] -= radius;
pos[1] -= radius;
pos = v[1]->data[pos_slot];
pos[0] += radius;
pos[1] -= radius;
pos = v[2]->data[pos_slot];
pos[0] += radius;
pos[1] += radius;
pos = v[3]->data[pos_slot];
pos[0] -= radius;
pos[1] += radius;
/* new texcoords */
tex = v[0]->data[tex_slot];
ASSIGN_4V(tex, -1, -1, k, 1);
tex = v[1]->data[tex_slot];
ASSIGN_4V(tex, 1, -1, k, 1);
tex = v[2]->data[tex_slot];
ASSIGN_4V(tex, 1, 1, k, 1);
tex = v[3]->data[tex_slot];
ASSIGN_4V(tex, -1, 1, k, 1);
/* emit 2 tris for the quad strip */
tri.v[0] = v[0];
tri.v[1] = v[1];
tri.v[2] = v[2];
stage->next->tri( stage->next, &tri );
tri.v[0] = v[0];
tri.v[1] = v[2];
tri.v[2] = v[3];
stage->next->tri( stage->next, &tri );
}
/**
* Draw a wide line by drawing a quad (two triangles).
*/
static void wideline_line( struct draw_stage *stage,
struct prim_header *header )
{
/*const struct wideline_stage *wide = wideline_stage(stage);*/
const unsigned pos = draw_current_shader_position_output(stage->draw);
const float half_width = 0.5f * stage->draw->rasterizer->line_width;
struct prim_header tri;
struct vertex_header *v0 = dup_vert(stage, header->v[0], 0);
struct vertex_header *v1 = dup_vert(stage, header->v[0], 1);
struct vertex_header *v2 = dup_vert(stage, header->v[1], 2);
struct vertex_header *v3 = dup_vert(stage, header->v[1], 3);
float *pos0 = v0->data[pos];
float *pos1 = v1->data[pos];
float *pos2 = v2->data[pos];
float *pos3 = v3->data[pos];
const float dx = fabsf(pos0[0] - pos2[0]);
const float dy = fabsf(pos0[1] - pos2[1]);
const boolean half_pixel_center =
stage->draw->rasterizer->half_pixel_center;
/* small tweak to meet GL specification */
const float bias = half_pixel_center ? 0.125f : 0.0f;
/*
* Draw wide line as a quad (two tris) by "stretching" the line along
* X or Y.
* We need to tweak coords in several ways to be conformant here.
*/
if (dx > dy) {
/* x-major line */
pos0[1] = pos0[1] - half_width - bias;
pos1[1] = pos1[1] + half_width - bias;
pos2[1] = pos2[1] - half_width - bias;
pos3[1] = pos3[1] + half_width - bias;
if (half_pixel_center) {
if (pos0[0] < pos2[0]) {
/* left to right line */
pos0[0] -= 0.5f;
pos1[0] -= 0.5f;
pos2[0] -= 0.5f;
pos3[0] -= 0.5f;
}
else {
/* right to left line */
pos0[0] += 0.5f;
pos1[0] += 0.5f;
pos2[0] += 0.5f;
pos3[0] += 0.5f;
}
}
}
else {
/* y-major line */
pos0[0] = pos0[0] - half_width + bias;
pos1[0] = pos1[0] + half_width + bias;
pos2[0] = pos2[0] - half_width + bias;
pos3[0] = pos3[0] + half_width + bias;
if (half_pixel_center) {
if (pos0[1] < pos2[1]) {
/* top to bottom line */
pos0[1] -= 0.5f;
pos1[1] -= 0.5f;
pos2[1] -= 0.5f;
pos3[1] -= 0.5f;
}
else {
/* bottom to top line */
pos0[1] += 0.5f;
pos1[1] += 0.5f;
pos2[1] += 0.5f;
pos3[1] += 0.5f;
}
}
}
tri.det = header->det; /* only the sign matters */
tri.v[0] = v0;
tri.v[1] = v2;
tri.v[2] = v3;
stage->next->tri( stage->next, &tri );
tri.v[0] = v0;
tri.v[1] = v3;
tri.v[2] = v1;
stage->next->tri( stage->next, &tri );
}
/**
* Draw a wide line by drawing a quad, using geometry which will
* fullfill GL's antialiased line requirements.
*/
static void
aaline_line(struct draw_stage *stage, struct prim_header *header)
{
const struct aaline_stage *aaline = aaline_stage(stage);
const float half_width = aaline->half_line_width;
struct prim_header tri;
struct vertex_header *v[8];
uint texPos = aaline->tex_slot;
uint posPos = aaline->pos_slot;
float *pos, *tex;
float dx = header->v[1]->data[posPos][0] - header->v[0]->data[posPos][0];
float dy = header->v[1]->data[posPos][1] - header->v[0]->data[posPos][1];
double a = atan2(dy, dx);
float c_a = (float) cos(a), s_a = (float) sin(a);
uint i;
/* XXX the ends of lines aren't quite perfect yet, but probably passable */
dx = 0.5F * half_width;
dy = half_width;
/* allocate/dup new verts */
for (i = 0; i < 8; i++) {
v[i] = dup_vert(stage, header->v[i/4], i);
}
/*
* Quad strip for line from v0 to v1 (*=endpoints):
*
* 1 3 5 7
* +---+---------------------+---+
* | |
* | *v0 v1* |
* | |
* +---+---------------------+---+
* 0 2 4 6
*/
/* new verts */
pos = v[0]->data[posPos];
pos[0] += (-dx * c_a - dy * s_a);
pos[1] += (-dx * s_a + dy * c_a);
pos = v[1]->data[posPos];
pos[0] += (-dx * c_a - -dy * s_a);
pos[1] += (-dx * s_a + -dy * c_a);
pos = v[2]->data[posPos];
pos[0] += ( dx * c_a - dy * s_a);
pos[1] += ( dx * s_a + dy * c_a);
pos = v[3]->data[posPos];
pos[0] += ( dx * c_a - -dy * s_a);
pos[1] += ( dx * s_a + -dy * c_a);
pos = v[4]->data[posPos];
pos[0] += (-dx * c_a - dy * s_a);
pos[1] += (-dx * s_a + dy * c_a);
pos = v[5]->data[posPos];
pos[0] += (-dx * c_a - -dy * s_a);
pos[1] += (-dx * s_a + -dy * c_a);
pos = v[6]->data[posPos];
pos[0] += ( dx * c_a - dy * s_a);
pos[1] += ( dx * s_a + dy * c_a);
pos = v[7]->data[posPos];
pos[0] += ( dx * c_a - -dy * s_a);
pos[1] += ( dx * s_a + -dy * c_a);
/* new texcoords */
tex = v[0]->data[texPos];
ASSIGN_4V(tex, 0, 0, 0, 1);
tex = v[1]->data[texPos];
ASSIGN_4V(tex, 0, 1, 0, 1);
tex = v[2]->data[texPos];
ASSIGN_4V(tex, .5, 0, 0, 1);
tex = v[3]->data[texPos];
ASSIGN_4V(tex, .5, 1, 0, 1);
tex = v[4]->data[texPos];
ASSIGN_4V(tex, .5, 0, 0, 1);
tex = v[5]->data[texPos];
ASSIGN_4V(tex, .5, 1, 0, 1);
tex = v[6]->data[texPos];
ASSIGN_4V(tex, 1, 0, 0, 1);
tex = v[7]->data[texPos];
ASSIGN_4V(tex, 1, 1, 0, 1);
/* emit 6 tris for the quad strip */
tri.v[0] = v[2]; tri.v[1] = v[1]; tri.v[2] = v[0];
stage->next->tri( stage->next, &tri );
tri.v[0] = v[3]; tri.v[1] = v[1]; tri.v[2] = v[2];
stage->next->tri( stage->next, &tri );
tri.v[0] = v[4]; tri.v[1] = v[3]; tri.v[2] = v[2];
stage->next->tri( stage->next, &tri );
tri.v[0] = v[5]; tri.v[1] = v[3]; tri.v[2] = v[4];
stage->next->tri( stage->next, &tri );
tri.v[0] = v[6]; tri.v[1] = v[5]; tri.v[2] = v[4];
stage->next->tri( stage->next, &tri );
tri.v[0] = v[7]; tri.v[1] = v[5]; tri.v[2] = v[6];
stage->next->tri( stage->next, &tri );
}
/* Clip a triangle against the viewport and user clip planes.
*/
static void
do_clip_tri( struct draw_stage *stage,
struct prim_header *header,
unsigned clipmask )
{
struct clipper *clipper = clipper_stage( stage );
struct vertex_header *a[MAX_CLIPPED_VERTICES];
struct vertex_header *b[MAX_CLIPPED_VERTICES];
struct vertex_header **inlist = a;
struct vertex_header **outlist = b;
unsigned tmpnr = 0;
unsigned n = 3;
unsigned i;
inlist[0] = header->v[0];
inlist[1] = header->v[1];
inlist[2] = header->v[2];
while (clipmask && n >= 3) {
const unsigned plane_idx = ffs(clipmask)-1;
const float *plane = clipper->plane[plane_idx];
struct vertex_header *vert_prev = inlist[0];
float dp_prev = dot4( vert_prev->clip, plane );
unsigned outcount = 0;
clipmask &= ~(1<<plane_idx);
inlist[n] = inlist[0]; /* prevent rotation of vertices */
for (i = 1; i <= n; i++) {
struct vertex_header *vert = inlist[i];
float dp = dot4( vert->clip, plane );
if (!IS_NEGATIVE(dp_prev)) {
outlist[outcount++] = vert_prev;
}
if (DIFFERENT_SIGNS(dp, dp_prev)) {
struct vertex_header *new_vert = clipper->stage.tmp[tmpnr++];
outlist[outcount++] = new_vert;
if (IS_NEGATIVE(dp)) {
/* Going out of bounds. Avoid division by zero as we
* know dp != dp_prev from DIFFERENT_SIGNS, above.
*/
float t = dp / (dp - dp_prev);
interp( clipper, new_vert, t, vert, vert_prev );
/* Force edgeflag true in this case:
*/
new_vert->edgeflag = 1;
} else {
/* Coming back in.
*/
float t = dp_prev / (dp_prev - dp);
interp( clipper, new_vert, t, vert_prev, vert );
/* Copy starting vert's edgeflag:
*/
new_vert->edgeflag = vert_prev->edgeflag;
}
}
vert_prev = vert;
dp_prev = dp;
}
{
struct vertex_header **tmp = inlist;
inlist = outlist;
outlist = tmp;
n = outcount;
}
}
/* If flat-shading, copy color to new provoking vertex.
*/
if (clipper->flat && inlist[0] != header->v[2]) {
if (1) {
inlist[0] = dup_vert(stage, inlist[0], tmpnr++);
}
copy_colors(stage, inlist[0], header->v[2]);
}
/* Emit the polygon as triangles to the setup stage:
*/
if (n >= 3)
emit_poly( stage, inlist, n, header );
}
/* If there are lots of sprite points (and why wouldn't there be?) it
* would probably be more sensible to change hardware setup to
* optimize this rather than doing the whole thing in software like
* this.
*/
static void widepoint_point( struct draw_stage *stage,
struct prim_header *header )
{
const struct widepoint_stage *wide = widepoint_stage(stage);
const unsigned pos = draw_current_shader_position_output(stage->draw);
const boolean sprite = (boolean) stage->draw->rasterizer->point_quad_rasterization;
float half_size;
float left_adj, right_adj, bot_adj, top_adj;
struct prim_header tri;
/* four dups of original vertex */
struct vertex_header *v0 = dup_vert(stage, header->v[0], 0);
struct vertex_header *v1 = dup_vert(stage, header->v[0], 1);
struct vertex_header *v2 = dup_vert(stage, header->v[0], 2);
struct vertex_header *v3 = dup_vert(stage, header->v[0], 3);
float *pos0 = v0->data[pos];
float *pos1 = v1->data[pos];
float *pos2 = v2->data[pos];
float *pos3 = v3->data[pos];
/* point size is either per-vertex or fixed size */
if (wide->psize_slot >= 0) {
half_size = header->v[0]->data[wide->psize_slot][0];
half_size *= 0.5f;
}
else {
half_size = wide->half_point_size;
}
left_adj = -half_size + wide->xbias;
right_adj = half_size + wide->xbias;
bot_adj = half_size + wide->ybias;
top_adj = -half_size + wide->ybias;
pos0[0] += left_adj;
pos0[1] += top_adj;
pos1[0] += left_adj;
pos1[1] += bot_adj;
pos2[0] += right_adj;
pos2[1] += top_adj;
pos3[0] += right_adj;
pos3[1] += bot_adj;
if (sprite) {
static const float tex00[4] = { 0, 0, 0, 1 };
static const float tex01[4] = { 0, 1, 0, 1 };
static const float tex11[4] = { 1, 1, 0, 1 };
static const float tex10[4] = { 1, 0, 0, 1 };
set_texcoords( wide, v0, tex00 );
set_texcoords( wide, v1, tex01 );
set_texcoords( wide, v2, tex10 );
set_texcoords( wide, v3, tex11 );
}
tri.det = header->det; /* only the sign matters */
tri.v[0] = v0;
tri.v[1] = v2;
tri.v[2] = v3;
stage->next->tri( stage->next, &tri );
tri.v[0] = v0;
tri.v[1] = v3;
tri.v[2] = v1;
stage->next->tri( stage->next, &tri );
}
