fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-05-06 02:58:05 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

nv40: construct vbo state the same way as the rest

Browse source
This commit is contained in:
Ben Skeggs 2008-02-25 13:29:16 +11:00
parent 14de997d5d
commit 4058a90127
5 changed files with 137 additions and 136 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

11
src/gallium/drivers/nv40/nv40_context.h
View file

@ -54,7 +54,9 @@ enum nv40_state_index {
NV40_STATE_VERTTEX1 = 28,
NV40_STATE_VERTTEX2 = 29,
NV40_STATE_VERTTEX3 = 30,
NV40_STATE_MAX = 31
NV40_STATE_VTXBUF = 31,
NV40_STATE_VTXFMT = 32,
NV40_STATE_MAX = 33
};
#define NV40_NEW_BLEND (1 << 0)
@ -117,7 +119,7 @@ struct nv40_state {
unsigned stipple_enabled;
unsigned fp_samplers;
unsigned dirty;
uint64_t dirty;
struct nouveau_stateobj *hw[NV40_STATE_MAX];
};
@ -149,13 +151,13 @@ struct nv40_context {
struct pipe_blend_color blend_colour;
struct pipe_viewport_state viewport;
struct pipe_framebuffer_state framebuffer;
struct pipe_buffer *idxbuf;
unsigned idxbuf_format;
} pipe_state;
struct nv40_state state;
unsigned fallback;
struct nouveau_stateobj *so_vtxbuf;
struct pipe_vertex_buffer vtxbuf[PIPE_ATTRIB_MAX];
struct pipe_vertex_element vtxelt[PIPE_ATTRIB_MAX];
};
@ -208,6 +210,7 @@ extern struct nv40_state_entry nv40_state_zsa;
extern struct nv40_state_entry nv40_state_viewport;
extern struct nv40_state_entry nv40_state_framebuffer;
extern struct nv40_state_entry nv40_state_fragtex;
extern struct nv40_state_entry nv40_state_vbo;
/* nv40_vbo.c */
extern boolean nv40_draw_arrays(struct pipe_context *, unsigned mode,

4
src/gallium/drivers/nv40/nv40_fragtex.c
View file

@ -140,7 +140,7 @@ nv40_fragtex_validate(struct nv40_context *nv40)
so_method(so, nv40->hw->curie, NV40TCL_TEX_ENABLE(unit), 1);
so_data (so, 0);
so_ref(so, &nv40->state.hw[NV40_STATE_FRAGTEX0 + unit]);
state->dirty |= (1 << (NV40_STATE_FRAGTEX0 + unit));
state->dirty |= (1ULL << (NV40_STATE_FRAGTEX0 + unit));
}
samplers = nv40->dirty_samplers & fp->samplers;
@ -150,7 +150,7 @@ nv40_fragtex_validate(struct nv40_context *nv40)
so = nv40_fragtex_build(nv40, unit);
so_ref(so, &nv40->state.hw[NV40_STATE_FRAGTEX0 + unit]);
state->dirty |= (1 << (NV40_STATE_FRAGTEX0 + unit));
state->dirty |= (1ULL << (NV40_STATE_FRAGTEX0 + unit));
}
nv40->state.fp_samplers = fp->samplers;

2
src/gallium/drivers/nv40/nv40_state.c
View file

@ -628,7 +628,6 @@ nv40_set_vertex_buffer(struct pipe_context *pipe, unsigned index,
struct nv40_context *nv40 = nv40_context(pipe);
nv40->vtxbuf[index] = *vb;
nv40->dirty |= NV40_NEW_ARRAYS;
}
@ -639,7 +638,6 @@ nv40_set_vertex_element(struct pipe_context *pipe, unsigned index,
struct nv40_context *nv40 = nv40_context(pipe);
nv40->vtxelt[index] = *ve;
nv40->dirty |= NV40_NEW_ARRAYS;
}

10
src/gallium/drivers/nv40/nv40_state_emit.c
View file

@ -14,6 +14,7 @@ static struct nv40_state_entry *render_states[] = {
&nv40_state_blend_colour,
&nv40_state_zsa,
&nv40_state_viewport,
&nv40_state_vbo,
NULL
};
@ -31,7 +32,7 @@ nv40_state_validate(struct nv40_context *nv40)
if (nv40->dirty & e->dirty.pipe) {
if (e->validate(nv40))
nv40->state.dirty |= (1 << e->dirty.hw);
nv40->state.dirty |= (1ULL << e->dirty.hw);
}
states++;
@ -64,20 +65,21 @@ nv40_state_emit(struct nv40_context *nv40)
unsigned i, samplers;
while (state->dirty) {
unsigned idx = ffs(state->dirty) - 1;
unsigned idx = ffsll(state->dirty) - 1;
so_ref (state->hw[idx], &nv40->hw->state[idx]);
so_emit(nv40->nvws, nv40->hw->state[idx]);
state->dirty &= ~(1 << idx);
state->dirty &= ~(1ULL << idx);
}
so_emit_reloc_markers(nv40->nvws, state->hw[NV40_STATE_FB]);
for (i = 0, samplers = state->fp_samplers; i < 16 && samplers; i++) {
so_emit_reloc_markers(nv40->nvws,
state->hw[NV40_STATE_FRAGTEX0+i]);
samplers &= ~(1 << i);
samplers &= ~(1ULL << i);
}
so_emit_reloc_markers(nv40->nvws, state->hw[NV40_STATE_FRAGPROG]);
so_emit_reloc_markers(nv40->nvws, state->hw[NV40_STATE_VTXBUF]);
}
void

246
src/gallium/drivers/nv40/nv40_vbo.c
View file

@ -39,6 +39,43 @@ nv40_vbo_type(uint format)
}
}
static boolean
nv40_vbo_set_idxbuf(struct nv40_context *nv40, struct pipe_buffer *ib,
unsigned ib_size)
{
unsigned type;
if (!ib) {
nv40->pipe_state.idxbuf = NULL;
nv40->pipe_state.idxbuf_format = 0xdeadbeef;
return FALSE;
}
/* No support for 8bit indices, no support at all on 0x4497 chips */
if (nv40->hw->curie->grclass == NV44TCL || ib_size == 1)
return FALSE;
switch (ib_size) {
case 2:
type = NV40TCL_IDXBUF_FORMAT_TYPE_U16;
break;
case 4:
type = NV40TCL_IDXBUF_FORMAT_TYPE_U32;
break;
default:
return FALSE;
}
if (ib != nv40->pipe_state.idxbuf ||
type != nv40->pipe_state.idxbuf_format) {
nv40->dirty |= NV40_NEW_ARRAYS;
nv40->pipe_state.idxbuf = ib;
nv40->pipe_state.idxbuf_format = type;
}
return TRUE;
}
static boolean
nv40_vbo_static_attrib(struct nv40_context *nv40, int attrib,
struct pipe_vertex_element *ve,
@ -101,104 +138,15 @@ nv40_vbo_static_attrib(struct nv40_context *nv40, int attrib,
return TRUE;
}
static void
nv40_vbo_arrays_update(struct nv40_context *nv40, struct pipe_buffer *ib,
unsigned ib_format)
{
struct nv40_vertex_program *vp = nv40->pipe_state.vertprog;
struct nouveau_stateobj *vtxbuf, *vtxfmt;
unsigned inputs, hw, num_hw;
unsigned vb_flags = NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD;
inputs = vp->ir;
for (hw = 0; hw < 16 && inputs; hw++) {
if (inputs & (1 << hw)) {
num_hw = hw;
inputs &= ~(1 << hw);
}
}
num_hw++;
vtxbuf = so_new(20, 18);
so_method(vtxbuf, nv40->hw->curie, NV40TCL_VTXBUF_ADDRESS(0), num_hw);
vtxfmt = so_new(17, 0);
so_method(vtxfmt, nv40->hw->curie, NV40TCL_VTXFMT(0), num_hw);
inputs = vp->ir;
for (hw = 0; hw < num_hw; hw++) {
struct pipe_vertex_element *ve;
struct pipe_vertex_buffer *vb;
if (!(inputs & (1 << hw))) {
so_data(vtxbuf, 0);
so_data(vtxfmt, NV40TCL_VTXFMT_TYPE_FLOAT);
continue;
}
ve = &nv40->vtxelt[hw];
vb = &nv40->vtxbuf[ve->vertex_buffer_index];
if (!vb->pitch && nv40_vbo_static_attrib(nv40, hw, ve, vb)) {
so_data(vtxbuf, 0);
so_data(vtxfmt, NV40TCL_VTXFMT_TYPE_FLOAT);
continue;
}
so_reloc(vtxbuf, vb->buffer, vb->buffer_offset + ve->src_offset,
vb_flags | NOUVEAU_BO_LOW | NOUVEAU_BO_OR,
0, NV40TCL_VTXBUF_ADDRESS_DMA1);
so_data (vtxfmt, ((vb->pitch << NV40TCL_VTXFMT_STRIDE_SHIFT) |
(nv40_vbo_ncomp(ve->src_format) <<
NV40TCL_VTXFMT_SIZE_SHIFT) |
nv40_vbo_type(ve->src_format)));
}
if (ib) {
so_method(vtxbuf, nv40->hw->curie, NV40TCL_IDXBUF_ADDRESS, 2);
so_reloc (vtxbuf, ib, 0, vb_flags | NOUVEAU_BO_LOW, 0, 0);
so_reloc (vtxbuf, ib, ib_format, vb_flags | NOUVEAU_BO_OR,
0, NV40TCL_IDXBUF_FORMAT_DMA1);
}
so_emit(nv40->nvws, vtxfmt);
so_emit(nv40->nvws, vtxbuf);
so_ref (vtxbuf, &nv40->so_vtxbuf);
so_ref (NULL, &vtxfmt);
}
static boolean
nv40_vbo_validate_state(struct nv40_context *nv40,
struct pipe_buffer *ib, unsigned ib_format)
{
unsigned vdn = nv40->dirty & NV40_NEW_ARRAYS;
nv40_emit_hw_state(nv40);
if (vdn || ib) {
nv40_vbo_arrays_update(nv40, ib, ib_format);
nv40->dirty &= ~NV40_NEW_ARRAYS;
}
so_emit_reloc_markers(nv40->nvws, nv40->so_vtxbuf);
BEGIN_RING(curie, 0x1710, 1);
OUT_RING (0); /* vtx cache flush */
return TRUE;
}
boolean
nv40_draw_arrays(struct pipe_context *pipe, unsigned mode, unsigned start,
unsigned count)
{
struct nv40_context *nv40 = nv40_context(pipe);
unsigned nr;
boolean ret;
ret = nv40_vbo_validate_state(nv40, NULL, 0);
if (!ret) {
NOUVEAU_ERR("state validate failed\n");
return FALSE;
}
nv40_vbo_set_idxbuf(nv40, NULL, 0);
nv40_emit_hw_state(nv40);
BEGIN_RING(curie, NV40TCL_BEGIN_END, 1);
OUT_RING (nvgl_primitive(mode));
@ -305,14 +253,9 @@ nv40_draw_elements_inline(struct pipe_context *pipe,
{
struct nv40_context *nv40 = nv40_context(pipe);
struct pipe_winsys *ws = pipe->winsys;
boolean ret;
void *map;
ret = nv40_vbo_validate_state(nv40, NULL, 0);
if (!ret) {
NOUVEAU_ERR("state validate failed\n");
return FALSE;
}
nv40_emit_hw_state(nv40);
map = ws->buffer_map(ws, ib, PIPE_BUFFER_USAGE_CPU_READ);
if (!ib) {
@ -348,30 +291,12 @@ nv40_draw_elements_inline(struct pipe_context *pipe,
static boolean
nv40_draw_elements_vbo(struct pipe_context *pipe,
struct pipe_buffer *ib, unsigned ib_size,
unsigned mode, unsigned start, unsigned count)
{
struct nv40_context *nv40 = nv40_context(pipe);
unsigned nr, type;
boolean ret;
unsigned nr;
switch (ib_size) {
case 2:
type = NV40TCL_IDXBUF_FORMAT_TYPE_U16;
break;
case 4:
type = NV40TCL_IDXBUF_FORMAT_TYPE_U32;
break;
default:
NOUVEAU_ERR("invalid idxbuf fmt %d\n", ib_size);
return FALSE;
}
ret = nv40_vbo_validate_state(nv40, ib, type);
if (!ret) {
NOUVEAU_ERR("failed state validation\n");
return FALSE;
}
nv40_emit_hw_state(nv40);
BEGIN_RING(curie, NV40TCL_BEGIN_END, 1);
OUT_RING (nvgl_primitive(mode));
@ -409,19 +334,92 @@ nv40_draw_elements(struct pipe_context *pipe,
{
struct nv40_context *nv40 = nv40_context(pipe);
/* 0x4497 doesn't support real index buffers, and there doesn't appear
* to be support on any chipset for 8-bit indices.
*/
if (nv40->hw->curie->grclass == NV44TCL || indexSize == 1) {
if (nv40_vbo_set_idxbuf(nv40, indexBuffer, indexSize)) {
nv40_draw_elements_vbo(pipe, mode, start, count);
} else {
nv40_draw_elements_inline(pipe, indexBuffer, indexSize,
mode, start, count);
} else {
nv40_draw_elements_vbo(pipe, indexBuffer, indexSize,
mode, start, count);
}
pipe->flush(pipe, 0);
return TRUE;
}
static boolean
nv40_vbo_validate(struct nv40_context *nv40)
{
struct nv40_vertex_program *vp = nv40->pipe_state.vertprog;
struct nouveau_stateobj *vtxbuf, *vtxfmt;
struct pipe_buffer *ib = nv40->pipe_state.idxbuf;
unsigned ib_format = nv40->pipe_state.idxbuf_format;
unsigned inputs, hw, num_hw;
unsigned vb_flags = NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD;
inputs = vp->ir;
for (hw = 0; hw < 16 && inputs; hw++) {
if (inputs & (1 << hw)) {
num_hw = hw;
inputs &= ~(1 << hw);
}
}
num_hw++;
vtxbuf = so_new(20, 18);
so_method(vtxbuf, nv40->hw->curie, NV40TCL_VTXBUF_ADDRESS(0), num_hw);
vtxfmt = so_new(17, 0);
so_method(vtxfmt, nv40->hw->curie, NV40TCL_VTXFMT(0), num_hw);
inputs = vp->ir;
for (hw = 0; hw < num_hw; hw++) {
struct pipe_vertex_element *ve;
struct pipe_vertex_buffer *vb;
if (!(inputs & (1 << hw))) {
so_data(vtxbuf, 0);
so_data(vtxfmt, NV40TCL_VTXFMT_TYPE_FLOAT);
continue;
}
ve = &nv40->vtxelt[hw];
vb = &nv40->vtxbuf[ve->vertex_buffer_index];
if (!vb->pitch && nv40_vbo_static_attrib(nv40, hw, ve, vb)) {
so_data(vtxbuf, 0);
so_data(vtxfmt, NV40TCL_VTXFMT_TYPE_FLOAT);
continue;
}
so_reloc(vtxbuf, vb->buffer, vb->buffer_offset + ve->src_offset,
vb_flags | NOUVEAU_BO_LOW | NOUVEAU_BO_OR,
0, NV40TCL_VTXBUF_ADDRESS_DMA1);
so_data (vtxfmt, ((vb->pitch << NV40TCL_VTXFMT_STRIDE_SHIFT) |
(nv40_vbo_ncomp(ve->src_format) <<
NV40TCL_VTXFMT_SIZE_SHIFT) |
nv40_vbo_type(ve->src_format)));
}
if (ib) {
so_method(vtxbuf, nv40->hw->curie, NV40TCL_IDXBUF_ADDRESS, 2);
so_reloc (vtxbuf, ib, 0, vb_flags | NOUVEAU_BO_LOW, 0, 0);
so_reloc (vtxbuf, ib, ib_format, vb_flags | NOUVEAU_BO_OR,
0, NV40TCL_IDXBUF_FORMAT_DMA1);
}
so_method(vtxbuf, nv40->hw->curie, 0x1710, 1);
so_data (vtxbuf, 0);
so_ref(vtxbuf, &nv40->state.hw[NV40_STATE_VTXBUF]);
nv40->state.dirty |= (1ULL << NV40_STATE_VTXBUF);
so_ref(vtxfmt, &nv40->state.hw[NV40_STATE_VTXFMT]);
nv40->state.dirty |= (1ULL << NV40_STATE_VTXFMT);
return FALSE;
}
struct nv40_state_entry nv40_state_vbo = {
.validate = nv40_vbo_validate,
.dirty = {
.pipe = NV40_NEW_ARRAYS,
.hw = 0,
}
};