static void
nv50_draw_arrays_instanced(struct pipe_context *pipe,
unsigned mode, unsigned start, unsigned count,
unsigned startInstance, unsigned instanceCount)
{
struct nv50_context *nv50 = nv50_context(pipe);
struct nouveau_channel *chan = nv50->screen->tesla->channel;
struct nouveau_grobj *tesla = nv50->screen->tesla;
struct instance a[16];
unsigned prim = nv50_prim(mode);
instance_init(nv50, a, startInstance);
if (!nv50_state_validate(nv50, 10 + 16*3))
return;
if (nv50->vbo_fifo) {
nv50_push_elements_instanced(pipe, NULL, 0, 0, mode, start,
count, startInstance,
instanceCount);
return;
}
BEGIN_RING(chan, tesla, NV50TCL_CB_ADDR, 2);
OUT_RING (chan, NV50_CB_AUX | (24 << 8));
OUT_RING (chan, startInstance);
while (instanceCount--) {
if (AVAIL_RING(chan) < (7 + 16*3)) {
FIRE_RING(chan);
if (!nv50_state_validate(nv50, 7 + 16*3)) {
assert(0);
return;
}
}
instance_step(nv50, a);
BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
OUT_RING (chan, prim);
BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BUFFER_FIRST, 2);
OUT_RING (chan, start);
OUT_RING (chan, count);
BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
OUT_RING (chan, 0);
prim |= (1 << 28);
}
}
void
nv50_push_elements_instanced(struct pipe_context *pipe,
struct pipe_resource *idxbuf,
unsigned idxsize, int idxbias,
unsigned mode, unsigned start, unsigned count,
unsigned i_start, unsigned i_count)
{
struct nv50_context *nv50 = nv50_context(pipe);
struct nouveau_grobj *tesla = nv50->screen->tesla;
struct nouveau_channel *chan = tesla->channel;
struct push_context ctx;
const unsigned p_overhead = 4 + /* begin/end */
4; /* potential edgeflag enable/disable */
const unsigned v_overhead = 1 + /* VERTEX_DATA packet header */
2; /* potential edgeflag modification */
struct util_split_prim s;
unsigned vtx_size;
boolean nzi = FALSE;
int i;
ctx.nv50 = nv50;
ctx.attr_nr = 0;
ctx.idxbuf = NULL;
ctx.vtx_size = 0;
ctx.edgeflag = 0.5f;
ctx.edgeflag_attr = nv50->vertprog->vp.edgeflag;
/* map vertex buffers, determine vertex size */
for (i = 0; i < nv50->vtxelt->num_elements; i++) {
struct pipe_vertex_element *ve = &nv50->vtxelt->pipe[i];
struct pipe_vertex_buffer *vb = &nv50->vtxbuf[ve->vertex_buffer_index];
struct nouveau_bo *bo = nv50_resource(vb->buffer)->bo;
unsigned size, nr_components, n;
if (!(nv50->vbo_fifo & (1 << i)))
continue;
n = ctx.attr_nr++;
if (nouveau_bo_map(bo, NOUVEAU_BO_RD)) {
assert(bo->map);
return;
}
ctx.attr[n].map = (uint8_t *)bo->map + vb->buffer_offset + ve->src_offset;
nouveau_bo_unmap(bo);
ctx.attr[n].stride = vb->stride;
ctx.attr[n].divisor = ve->instance_divisor;
if (ctx.attr[n].divisor) {
ctx.attr[n].step = i_start % ve->instance_divisor;
ctx.attr[n].map = (uint8_t *)ctx.attr[n].map + i_start * vb->stride;
}
size = util_format_get_component_bits(ve->src_format,
UTIL_FORMAT_COLORSPACE_RGB, 0);
nr_components = util_format_get_nr_components(ve->src_format);
switch (size) {
case 8:
switch (nr_components) {
case 1: ctx.attr[n].push = emit_b08_1; break;
case 2: ctx.attr[n].push = emit_b16_1; break;
case 3: ctx.attr[n].push = emit_b08_3; break;
case 4: ctx.attr[n].push = emit_b32_1; break;
}
ctx.vtx_size++;
break;
case 16:
switch (nr_components) {
case 1: ctx.attr[n].push = emit_b16_1; break;
case 2: ctx.attr[n].push = emit_b32_1; break;
case 3: ctx.attr[n].push = emit_b16_3; break;
case 4: ctx.attr[n].push = emit_b32_2; break;
}
ctx.vtx_size += (nr_components + 1) >> 1;
break;
case 32:
switch (nr_components) {
case 1: ctx.attr[n].push = emit_b32_1; break;
case 2: ctx.attr[n].push = emit_b32_2; break;
case 3: ctx.attr[n].push = emit_b32_3; break;
case 4: ctx.attr[n].push = emit_b32_4; break;
}
ctx.vtx_size += nr_components;
break;
default:
assert(0);
return;
}
}
vtx_size = ctx.vtx_size + v_overhead;
/* map index buffer, if present */
if (idxbuf) {
struct nouveau_bo *bo = nv50_resource(idxbuf)->bo;
if (nouveau_bo_map(bo, NOUVEAU_BO_RD)) {
assert(bo->map);
return;
}
ctx.idxbuf = bo->map;
ctx.idxbias = idxbias;
ctx.idxsize = idxsize;
nouveau_bo_unmap(bo);
}
s.priv = &ctx;
s.edge = emit_edgeflag;
if (idxbuf) {
if (idxsize == 1)
s.emit = idxbias ? emit_elt08_biased : emit_elt08;
else
if (idxsize == 2)
s.emit = idxbias ? emit_elt16_biased : emit_elt16;
else
s.emit = idxbias ? emit_elt32_biased : emit_elt32;
} else
s.emit = emit_verts;
/* per-instance loop */
BEGIN_RING(chan, tesla, NV50TCL_CB_ADDR, 2);
OUT_RING (chan, NV50_CB_AUX | (24 << 8));
OUT_RING (chan, i_start);
while (i_count--) {
unsigned max_verts;
boolean done;
for (i = 0; i < ctx.attr_nr; i++) {
if (!ctx.attr[i].divisor ||
ctx.attr[i].divisor != ++ctx.attr[i].step)
continue;
ctx.attr[i].step = 0;
ctx.attr[i].map = (uint8_t *)ctx.attr[i].map + ctx.attr[i].stride;
}
util_split_prim_init(&s, mode, start, count);
do {
if (AVAIL_RING(chan) < p_overhead + (6 * vtx_size)) {
FIRE_RING(chan);
if (!nv50_state_validate(nv50, p_overhead + (6 * vtx_size))) {
assert(0);
return;
}
}
max_verts = AVAIL_RING(chan);
max_verts -= p_overhead;
max_verts /= vtx_size;
BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
OUT_RING (chan, nv50_prim(s.mode) | (nzi ? (1 << 28) : 0));
done = util_split_prim_next(&s, max_verts);
BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
OUT_RING (chan, 0);
} while (!done);
nzi = TRUE;
}
}
