nv30: Emit framebuffer state using state objects

This commit is contained in:
Patrice Mandin 2008-07-03 20:58:31 +02:00
parent 9833aec6cb
commit 36488ed052
5 changed files with 92 additions and 164 deletions

View file

@ -99,6 +99,7 @@ struct nv30_state {
unsigned stipple_enabled;
unsigned viewport_bypass;
uint64_t dirty;
struct nouveau_stateobj *hw[NV30_STATE_MAX];
};
@ -208,8 +209,10 @@ extern void nv30_fragprog_destroy(struct nv30_context *,
extern void nv30_fragtex_bind(struct nv30_context *);
/* nv30_state.c and friends */
extern boolean nv30_state_validate(struct nv30_context *nv30);
extern void nv30_emit_hw_state(struct nv30_context *nv30);
extern void nv30_state_tex_update(struct nv30_context *nv30);
extern struct nv30_state_entry nv30_state_framebuffer;
/* nv30_vbo.c */
extern boolean nv30_draw_arrays(struct pipe_context *, unsigned mode,

View file

@ -565,110 +565,9 @@ nv30_set_framebuffer_state(struct pipe_context *pipe,
const struct pipe_framebuffer_state *fb)
{
struct nv30_context *nv30 = nv30_context(pipe);
struct pipe_surface *rt[2], *zeta = NULL;
uint32_t rt_enable, rt_format, w = 0, h = 0;
int i, colour_format = 0, zeta_format = 0;
rt_enable = 0;
for (i = 0; i < 2; i++) {
if (!fb->cbufs[i])
continue;
if (colour_format) {
assert(w == fb->cbufs[i]->width);
assert(h == fb->cbufs[i]->height);
assert(colour_format == fb->cbufs[i]->format);
} else {
w = fb->cbufs[i]->width;
h = fb->cbufs[i]->height;
colour_format = fb->cbufs[i]->format;
rt_enable |= (NV34TCL_RT_ENABLE_COLOR0 << i);
rt[i] = fb->cbufs[i];
}
}
if (rt_enable & (NV34TCL_RT_ENABLE_COLOR1 | NV34TCL_RT_ENABLE_COLOR2 |
NV34TCL_RT_ENABLE_COLOR3))
rt_enable |= NV34TCL_RT_ENABLE_MRT;
if (fb->zsbuf) {
if (colour_format) {
assert(w == fb->zsbuf->width);
assert(h == fb->zsbuf->height);
} else {
w = fb->zsbuf->width;
h = fb->zsbuf->height;
}
zeta_format = fb->zsbuf->format;
zeta = fb->zsbuf;
}
rt_format = NV34TCL_RT_FORMAT_TYPE_LINEAR;
switch (colour_format) {
case PIPE_FORMAT_A8R8G8B8_UNORM:
case 0:
rt_format |= NV34TCL_RT_FORMAT_COLOR_A8R8G8B8;
break;
case PIPE_FORMAT_R5G6B5_UNORM:
rt_format |= NV34TCL_RT_FORMAT_COLOR_R5G6B5;
break;
default:
assert(0);
}
switch (zeta_format) {
case PIPE_FORMAT_Z16_UNORM:
rt_format |= NV34TCL_RT_FORMAT_ZETA_Z16;
break;
case PIPE_FORMAT_Z24S8_UNORM:
case 0:
rt_format |= NV34TCL_RT_FORMAT_ZETA_Z24S8;
break;
default:
assert(0);
}
if (rt_enable & NV34TCL_RT_ENABLE_COLOR0) {
uint32_t pitch = rt[0]->stride;
if (zeta) {
pitch |= (zeta->stride << 16);
} else {
pitch |= (pitch << 16);
}
BEGIN_RING(rankine, NV34TCL_COLOR0_PITCH, 1);
OUT_RING (pitch);
nv30->rt[0] = rt[0]->buffer;
}
if (rt_enable & NV34TCL_RT_ENABLE_COLOR1) {
BEGIN_RING(rankine, NV34TCL_COLOR1_PITCH, 1);
OUT_RING (rt[1]->stride);
nv30->rt[1] = rt[1]->buffer;
}
if (zeta_format)
{
nv30->zeta = zeta->buffer;
}
nv30->rt_enable = rt_enable;
BEGIN_RING(rankine, NV34TCL_RT_ENABLE, 1);
OUT_RING (rt_enable);
BEGIN_RING(rankine, NV34TCL_RT_HORIZ, 3);
OUT_RING ((w << 16) | 0);
OUT_RING ((h << 16) | 0);
OUT_RING (rt_format);
BEGIN_RING(rankine, NV34TCL_VIEWPORT_HORIZ, 2);
OUT_RING ((w << 16) | 0);
OUT_RING ((h << 16) | 0);
BEGIN_RING(rankine, NV34TCL_VIEWPORT_CLIP_HORIZ(0), 2);
OUT_RING (((w - 1) << 16) | 0);
OUT_RING (((h - 1) << 16) | 0);
BEGIN_RING(rankine, NV34TCL_VIEWPORT_TX_ORIGIN, 1);
OUT_RING (0);
nv30->framebuffer = *fb;
nv30->dirty |= NV30_NEW_FB;
}
static void

View file

@ -1,10 +1,55 @@
#include "nv30_context.h"
#include "nv30_state.h"
static struct nv30_state_entry *render_states[] = {
&nv30_state_framebuffer,
NULL
};
static void
nv30_state_do_validate(struct nv30_context *nv30,
struct nv30_state_entry **states)
{
const struct pipe_framebuffer_state *fb = &nv30->framebuffer;
unsigned i;
for (i = 0; i < fb->num_cbufs; i++)
fb->cbufs[i]->status = PIPE_SURFACE_STATUS_DEFINED;
if (fb->zsbuf)
fb->zsbuf->status = PIPE_SURFACE_STATUS_DEFINED;
while (*states) {
struct nv30_state_entry *e = *states;
if (nv30->dirty & e->dirty.pipe) {
if (e->validate(nv30)) {
nv30->state.dirty |= (1ULL << e->dirty.hw);
}
}
states++;
}
/* TODO: uncomment when finished converting
nv30->dirty = 0;
*/
}
void
nv30_emit_hw_state(struct nv30_context *nv30)
{
int i;
struct nv30_state *state = &nv30->state;
unsigned i;
uint64 states;
for (i = 0, states = state->dirty; states; i++) {
if (!(states & (1ULL << i)))
continue;
so_ref (state->hw[i], &nv30->screen->state[i]);
if (state->hw[i])
so_emit(nv30->nvws, nv30->screen->state[i]);
states &= ~(1ULL << i);
}
if (nv30->dirty & NV30_NEW_FRAGPROG) {
nv30_fragprog_bind(nv30, nv30->fragprog.current);
@ -28,37 +73,7 @@ nv30_emit_hw_state(struct nv30_context *nv30)
nv30->dirty_samplers = 0;
/* Emit relocs for every referenced buffer.
* This is to ensure the bufmgr has an accurate idea of how
* the buffer is used. This isn't very efficient, but we don't
* seem to take a significant performance hit. Will be improved
* at some point. Vertex arrays are emitted by nv30_vbo.c
*/
/* Render targets */
if (nv30->rt_enable & NV34TCL_RT_ENABLE_COLOR0) {
BEGIN_RING(rankine, NV34TCL_DMA_COLOR0, 1);
OUT_RELOCo(nv30->rt[0], NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
BEGIN_RING(rankine, NV34TCL_COLOR0_OFFSET, 1);
OUT_RELOCl(nv30->rt[0], 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
}
if (nv30->rt_enable & NV34TCL_RT_ENABLE_COLOR1) {
BEGIN_RING(rankine, NV34TCL_DMA_COLOR1, 1);
OUT_RELOCo(nv30->rt[1], NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
BEGIN_RING(rankine, NV34TCL_COLOR1_OFFSET, 1);
OUT_RELOCl(nv30->rt[1], 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
}
if (nv30->zeta) {
BEGIN_RING(rankine, NV34TCL_DMA_ZETA, 1);
OUT_RELOCo(nv30->zeta, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
BEGIN_RING(rankine, NV34TCL_ZETA_OFFSET, 1);
OUT_RELOCl(nv30->zeta, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
/* XXX allocate LMA */
/* BEGIN_RING(rankine, NV34TCL_LMA_DEPTH_OFFSET, 1);
OUT_RING(0);*/
}
so_emit_reloc_markers(nv30->nvws, state->hw[NV30_STATE_FB]);
/* Texture images, emitted in nv30_fragtex_build */
#if 0
@ -83,3 +98,35 @@ nv30_emit_hw_state(struct nv30_context *nv30)
NV34TCL_FP_ACTIVE_PROGRAM_DMA1);
}
boolean
nv30_state_validate(struct nv30_context *nv30)
{
#if 0
boolean was_sw = nv30->fallback_swtnl ? TRUE : FALSE;
if (nv30->render_mode != HW) {
/* Don't even bother trying to go back to hw if none
* of the states that caused swtnl previously have changed.
*/
if ((nv30->fallback_swtnl & nv30->dirty)
!= nv30->fallback_swtnl)
return FALSE;
/* Attempt to go to hwtnl again */
nv30->pipe.flush(&nv30->pipe, 0, NULL);
nv30->dirty |= (NV30_NEW_VIEWPORT |
NV30_NEW_VERTPROG |
NV30_NEW_ARRAYS);
nv30->render_mode = HW;
}
#endif
nv30_state_do_validate(nv30, render_states);
#if 0
if (nv30->fallback_swtnl || nv30->fallback_swrast)
return FALSE;
if (was_sw)
NOUVEAU_ERR("swtnl->hw\n");
#endif
return TRUE;
}

View file

@ -4,7 +4,7 @@ static boolean
nv30_state_framebuffer_validate(struct nv30_context *nv30)
{
struct pipe_framebuffer_state *fb = &nv30->framebuffer;
struct pipe_surface *rt[4], *zeta = NULL;
struct pipe_surface *rt[2], *zeta = NULL;
uint32_t rt_enable, rt_format;
int i, colour_format = 0, zeta_format = 0;
struct nouveau_stateobj *so = so_new(64, 10);
@ -23,8 +23,7 @@ nv30_state_framebuffer_validate(struct nv30_context *nv30)
}
}
if (rt_enable & (NV34TCL_RT_ENABLE_COLOR1 | NV34TCL_RT_ENABLE_COLOR2 |
NV34TCL_RT_ENABLE_COLOR3))
if (rt_enable & NV34TCL_RT_ENABLE_COLOR1)
rt_enable |= NV34TCL_RT_ENABLE_MRT;
if (fb->zsbuf) {
@ -86,31 +85,7 @@ nv30_state_framebuffer_validate(struct nv30_context *nv30)
NOUVEAU_BO_LOW, 0, 0);
so_data (so, rt[1]->stride);
}
/*
if (rt_enable & NV34TCL_RT_ENABLE_COLOR2) {
so_method(so, nv30->screen->rankine, NV34TCL_DMA_COLOR2, 1);
so_reloc (so, rt[2]->buffer, 0, rt_flags | NOUVEAU_BO_OR,
nv30->nvws->channel->vram->handle,
nv30->nvws->channel->gart->handle);
so_method(so, nv30->screen->rankine, NV34TCL_COLOR2_OFFSET, 1);
so_reloc (so, rt[2]->buffer, rt[2]->offset, rt_flags |
NOUVEAU_BO_LOW, 0, 0);
so_method(so, nv30->screen->rankine, NV34TCL_COLOR2_PITCH, 1);
so_data (so, rt[2]->pitch * rt[2]->cpp);
}
if (rt_enable & NV34TCL_RT_ENABLE_COLOR3) {
so_method(so, nv30->screen->rankine, NV34TCL_DMA_COLOR3, 1);
so_reloc (so, rt[3]->buffer, 0, rt_flags | NOUVEAU_BO_OR,
nv30->nvws->channel->vram->handle,
nv30->nvws->channel->gart->handle);
so_method(so, nv30->screen->rankine, NV34TCL_COLOR3_OFFSET, 1);
so_reloc (so, rt[3]->buffer, rt[3]->offset, rt_flags |
NOUVEAU_BO_LOW, 0, 0);
so_method(so, nv30->screen->rankine, NV34TCL_COLOR3_PITCH, 1);
so_data (so, rt[3]->pitch * rt[3]->cpp);
}
*/
if (zeta_format) {
so_method(so, nv30->screen->rankine, NV34TCL_DMA_ZETA, 1);
so_reloc (so, zeta->buffer, 0, rt_flags | NOUVEAU_BO_OR,
@ -119,8 +94,7 @@ nv30_state_framebuffer_validate(struct nv30_context *nv30)
so_method(so, nv30->screen->rankine, NV34TCL_ZETA_OFFSET, 1);
so_reloc (so, zeta->buffer, zeta->offset, rt_flags |
NOUVEAU_BO_LOW, 0, 0);
/*so_method(so, nv30->screen->rankine, NV34TCL_ZETA_PITCH, 1);
so_data (so, zeta->pitch * zeta->cpp);*/
/* TODO: allocate LMA depth buffer */
}
so_method(so, nv30->screen->rankine, NV34TCL_RT_ENABLE, 1);
@ -137,6 +111,9 @@ nv30_state_framebuffer_validate(struct nv30_context *nv30)
so_data (so, ((h - 1) << 16) | 0);
so_method(so, nv30->screen->rankine, 0x1d88, 1);
so_data (so, (1 << 12) | h);
/* Wonder why this is needed, context should all be set to zero on init */
so_method(so, nv30->screen->rankine, NV34TCL_VIEWPORT_TX_ORIGIN, 1);
so_data (so, 0);
so_ref(so, &nv30->state.hw[NV30_STATE_FB]);
return TRUE;

View file

@ -154,6 +154,8 @@ nv30_vbo_validate_state(struct nv30_context *nv30,
{
unsigned inputs;
nv30_state_validate(nv30);
nv30_emit_hw_state(nv30);
if (nv30->dirty & NV30_NEW_ARRAYS) {