nv50: adapt to vertex elements cso

This commit is contained in:
Christoph Bumiller 2010-02-26 14:16:46 +01:00
parent 51d139f038
commit f2656c3e3c
3 changed files with 85 additions and 43 deletions

View File

@ -72,6 +72,12 @@ struct nv50_sampler_stateobj {
unsigned tsc[8];
};
struct nv50_vtxelt_stateobj {
struct pipe_vertex_element pipe[16];
unsigned num_elements;
uint32_t hw[16];
};
static INLINE unsigned
get_tile_height(uint32_t tile_mode)
{
@ -168,8 +174,7 @@ struct nv50_context {
struct pipe_buffer *constbuf[PIPE_SHADER_TYPES];
struct pipe_vertex_buffer vtxbuf[PIPE_MAX_ATTRIBS];
unsigned vtxbuf_nr;
struct pipe_vertex_element vtxelt[PIPE_MAX_ATTRIBS];
unsigned vtxelt_nr;
struct nv50_vtxelt_stateobj *vtxelt;
struct nv50_sampler_stateobj *sampler[PIPE_SHADER_TYPES][PIPE_MAX_SAMPLERS];
unsigned sampler_nr[PIPE_SHADER_TYPES];
struct nv50_miptree *miptree[PIPE_SHADER_TYPES][PIPE_MAX_SAMPLERS];
@ -217,6 +222,7 @@ extern void nv50_draw_elements_instanced(struct pipe_context *pipe,
unsigned count,
unsigned startInstance,
unsigned instanceCount);
extern void nv50_vtxelt_construct(struct nv50_vtxelt_stateobj *cso);
extern void nv50_vbo_validate(struct nv50_context *nv50);
/* nv50_clear.c */

View File

@ -720,15 +720,34 @@ nv50_set_vertex_buffers(struct pipe_context *pipe, unsigned count,
nv50->dirty |= NV50_NEW_ARRAYS;
}
static void *
nv50_vtxelts_state_create(struct pipe_context *pipe,
unsigned num_elements,
const struct pipe_vertex_element *elements)
{
struct nv50_vtxelt_stateobj *cso = CALLOC_STRUCT(nv50_vtxelt_stateobj);
assert(num_elements < 16); /* not doing fallbacks yet */
cso->num_elements = num_elements;
memcpy(cso->pipe, elements, num_elements * sizeof(*elements));
nv50_vtxelt_construct(cso);
return (void *)cso;
}
static void
nv50_set_vertex_elements(struct pipe_context *pipe, unsigned count,
const struct pipe_vertex_element *ve)
nv50_vtxelts_state_delete(struct pipe_context *pipe, void *hwcso)
{
FREE(hwcso);
}
static void
nv50_vtxelts_state_bind(struct pipe_context *pipe, void *hwcso)
{
struct nv50_context *nv50 = nv50_context(pipe);
memcpy(nv50->vtxelt, ve, sizeof(*ve) * count);
nv50->vtxelt_nr = count;
nv50->vtxelt = hwcso;
nv50->dirty |= NV50_NEW_ARRAYS;
}
@ -778,7 +797,10 @@ nv50_init_state_functions(struct nv50_context *nv50)
nv50->pipe.set_scissor_state = nv50_set_scissor_state;
nv50->pipe.set_viewport_state = nv50_set_viewport_state;
nv50->pipe.create_vertex_elements_state = nv50_vtxelts_state_create;
nv50->pipe.delete_vertex_elements_state = nv50_vtxelts_state_delete;
nv50->pipe.bind_vertex_elements_state = nv50_vtxelts_state_bind;
nv50->pipe.set_vertex_buffers = nv50_set_vertex_buffers;
nv50->pipe.set_vertex_elements = nv50_set_vertex_elements;
}

View File

@ -223,11 +223,10 @@ nv50_set_static_vtxattr(struct nv50_context *nv50, unsigned i, void *data)
struct nouveau_grobj *tesla = nv50->screen->tesla;
struct nouveau_channel *chan = tesla->channel;
float v[4];
unsigned nr_components = util_format_get_nr_components(nv50->vtxelt[i].src_format);
enum pipe_format pf = nv50->vtxelt->pipe[i].src_format;
unsigned nr_components = util_format_get_nr_components(pf);
util_format_read_4f(nv50->vtxelt[i].src_format,
v, 0, data, 0, 0, 0, 1, 1);
util_format_read_4f(pf, v, 0, data, 0, 0, 0, 1, 1);
switch (nr_components) {
case 4:
@ -266,16 +265,17 @@ init_per_instance_arrays_immd(struct nv50_context *nv50,
struct nouveau_bo *bo;
unsigned i, b, count = 0;
for (i = 0; i < nv50->vtxelt_nr; ++i) {
if (!nv50->vtxelt[i].instance_divisor)
for (i = 0; i < nv50->vtxelt->num_elements; ++i) {
if (!nv50->vtxelt->pipe[i].instance_divisor)
continue;
++count;
b = nv50->vtxelt[i].vertex_buffer_index;
b = nv50->vtxelt->pipe[i].vertex_buffer_index;
pos[i] = nv50->vtxelt[i].src_offset +
pos[i] = nv50->vtxelt->pipe[i].src_offset +
nv50->vtxbuf[b].buffer_offset +
startInstance * nv50->vtxbuf[b].stride;
step[i] = startInstance % nv50->vtxelt[i].instance_divisor;
step[i] = startInstance %
nv50->vtxelt->pipe[i].instance_divisor;
bo = nouveau_bo(nv50->vtxbuf[b].buffer);
if (!bo->map)
@ -296,22 +296,22 @@ init_per_instance_arrays(struct nv50_context *nv50,
struct nouveau_channel *chan = tesla->channel;
struct nouveau_bo *bo;
struct nouveau_stateobj *so;
unsigned i, b, count = 0;
unsigned i, b, count = 0, num_elements = nv50->vtxelt->num_elements;
const uint32_t rl = NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD;
if (nv50->vbo_fifo)
return init_per_instance_arrays_immd(nv50, startInstance,
pos, step);
so = so_new(nv50->vtxelt_nr, nv50->vtxelt_nr * 2, nv50->vtxelt_nr * 2);
so = so_new(num_elements, num_elements * 2, num_elements * 2);
for (i = 0; i < nv50->vtxelt_nr; ++i) {
if (!nv50->vtxelt[i].instance_divisor)
for (i = 0; i < nv50->vtxelt->num_elements; ++i) {
if (!nv50->vtxelt->pipe[i].instance_divisor)
continue;
++count;
b = nv50->vtxelt[i].vertex_buffer_index;
b = nv50->vtxelt->pipe[i].vertex_buffer_index;
pos[i] = nv50->vtxelt[i].src_offset +
pos[i] = nv50->vtxelt->pipe[i].src_offset +
nv50->vtxbuf[b].buffer_offset +
startInstance * nv50->vtxbuf[b].stride;
@ -319,7 +319,8 @@ init_per_instance_arrays(struct nv50_context *nv50,
step[i] = 0;
continue;
}
step[i] = startInstance % nv50->vtxelt[i].instance_divisor;
step[i] = startInstance %
nv50->vtxelt->pipe[i].instance_divisor;
bo = nouveau_bo(nv50->vtxbuf[b].buffer);
@ -344,12 +345,12 @@ step_per_instance_arrays_immd(struct nv50_context *nv50,
struct nouveau_bo *bo;
unsigned i, b;
for (i = 0; i < nv50->vtxelt_nr; ++i) {
if (!nv50->vtxelt[i].instance_divisor)
for (i = 0; i < nv50->vtxelt->num_elements; ++i) {
if (!nv50->vtxelt->pipe[i].instance_divisor)
continue;
if (++step[i] != nv50->vtxelt[i].instance_divisor)
if (++step[i] != nv50->vtxelt->pipe[i].instance_divisor)
continue;
b = nv50->vtxelt[i].vertex_buffer_index;
b = nv50->vtxelt->pipe[i].vertex_buffer_index;
bo = nouveau_bo(nv50->vtxbuf[b].buffer);
step[i] = 0;
@ -367,7 +368,7 @@ step_per_instance_arrays(struct nv50_context *nv50,
struct nouveau_channel *chan = tesla->channel;
struct nouveau_bo *bo;
struct nouveau_stateobj *so;
unsigned i, b;
unsigned i, b, num_elements = nv50->vtxelt->num_elements;
const uint32_t rl = NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD;
if (nv50->vbo_fifo) {
@ -375,14 +376,14 @@ step_per_instance_arrays(struct nv50_context *nv50,
return;
}
so = so_new(nv50->vtxelt_nr, nv50->vtxelt_nr * 2, nv50->vtxelt_nr * 2);
so = so_new(num_elements, num_elements * 2, num_elements * 2);
for (i = 0; i < nv50->vtxelt_nr; ++i) {
if (!nv50->vtxelt[i].instance_divisor)
for (i = 0; i < nv50->vtxelt->num_elements; ++i) {
if (!nv50->vtxelt->pipe[i].instance_divisor)
continue;
b = nv50->vtxelt[i].vertex_buffer_index;
b = nv50->vtxelt->pipe[i].vertex_buffer_index;
if (++step[i] == nv50->vtxelt[i].instance_divisor) {
if (++step[i] == nv50->vtxelt->pipe[i].instance_divisor) {
step[i] = 0;
pos[i] += nv50->vtxbuf[b].stride;
}
@ -740,7 +741,8 @@ nv50_vbo_static_attrib(struct nv50_context *nv50, unsigned attrib,
0, 0, 1, 1);
so = *pso;
if (!so)
*pso = so = so_new(nv50->vtxelt_nr, nv50->vtxelt_nr * 4, 0);
*pso = so = so_new(nv50->vtxelt->num_elements,
nv50->vtxelt->num_elements * 4, 0);
switch (nr_components) {
case 4:
@ -778,6 +780,18 @@ nv50_vbo_static_attrib(struct nv50_context *nv50, unsigned attrib,
return TRUE;
}
void
nv50_vtxelt_construct(struct nv50_vtxelt_stateobj *cso)
{
unsigned i;
for (i = 0; i < cso->num_elements; ++i) {
struct pipe_vertex_element *ve = &cso->pipe[i];
cso->hw[i] = nv50_vbo_vtxelt_to_hw(ve);
}
}
void
nv50_vbo_validate(struct nv50_context *nv50)
{
@ -798,19 +812,19 @@ nv50_vbo_validate(struct nv50_context *nv50)
if (NV50_USING_LOATHED_EDGEFLAG(nv50))
nv50->vbo_fifo = 0xffff; /* vertprog can't set edgeflag */
n_ve = MAX2(nv50->vtxelt_nr, nv50->state.vtxelt_nr);
n_ve = MAX2(nv50->vtxelt->num_elements, nv50->state.vtxelt_nr);
vtxattr = NULL;
vtxbuf = so_new(n_ve * 2, n_ve * 5, nv50->vtxelt_nr * 4);
vtxbuf = so_new(n_ve * 2, n_ve * 5, nv50->vtxelt->num_elements * 4);
vtxfmt = so_new(1, n_ve, 0);
so_method(vtxfmt, tesla, NV50TCL_VERTEX_ARRAY_ATTRIB(0), n_ve);
for (i = 0; i < nv50->vtxelt_nr; i++) {
struct pipe_vertex_element *ve = &nv50->vtxelt[i];
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 = nouveau_bo(vb->buffer);
uint32_t hw = nv50_vbo_vtxelt_to_hw(ve);
uint32_t hw = nv50->vtxelt->hw[i];
if (!vb->stride &&
nv50_vbo_static_attrib(nv50, i, &vtxattr, ve, vb)) {
@ -859,7 +873,7 @@ nv50_vbo_validate(struct nv50_context *nv50)
so_method(vtxbuf, tesla, NV50TCL_VERTEX_ARRAY_FORMAT(i), 1);
so_data (vtxbuf, 0);
}
nv50->state.vtxelt_nr = nv50->vtxelt_nr;
nv50->state.vtxelt_nr = nv50->vtxelt->num_elements;
so_ref (vtxfmt, &nv50->state.vtxfmt);
so_ref (vtxbuf, &nv50->state.vtxbuf);
@ -1020,13 +1034,13 @@ emit_prepare(struct nv50_context *nv50, struct nv50_vbo_emitctx *emit,
emit->nr_ve = 0;
emit->vtx_dwords = 0;
for (i = 0; i < nv50->vtxelt_nr; ++i) {
for (i = 0; i < nv50->vtxelt->num_elements; ++i) {
struct pipe_vertex_element *ve;
struct pipe_vertex_buffer *vb;
unsigned n, size, nr_components;
const struct util_format_description *desc;
ve = &nv50->vtxelt[i];
ve = &nv50->vtxelt->pipe[i];
vb = &nv50->vtxbuf[ve->vertex_buffer_index];
if (!(nv50->vbo_fifo & (1 << i)) || ve->instance_divisor)
continue;