-#include "nv30/nv30_context.h"
-#include "nv40/nv40_context.h"
+#include "nvfx_context.h"
#include "nvfx_state.h"
+#include "nvfx_resource.h"
#include "draw/draw_context.h"
-#define RENDER_STATES(name, nvxx, vbo) \
-static struct nvfx_state_entry *name##_render_states[] = { \
- &nvxx##_state_framebuffer, \
- &nvfx_state_rasterizer, \
- &nvfx_state_scissor, \
- &nvxx##_state_stipple, \
- &nvxx##_state_fragprog, \
- &nvxx##_state_fragtex, \
- &nvxx##_state_vertprog, \
- &nvfx_state_blend, \
- &nvfx_state_blend_colour, \
- &nvfx_state_zsa, \
- &nvfx_state_sr, \
- &nvxx##_state_viewport, \
- &nvxx##_state_##vbo, \
- NULL \
-}
-
-RENDER_STATES(nv30, nv30, vbo);
-RENDER_STATES(nv30_swtnl, nv30, vbo); /* TODO: replace with vtxfmt once draw is unified */
-RENDER_STATES(nv40, nv40, vbo);
-RENDER_STATES(nv40_swtnl, nv40, vtxfmt);
-
-static void
-nvfx_state_do_validate(struct nvfx_context *nvfx,
- struct nvfx_state_entry **states)
+static boolean
+nvfx_state_validate_common(struct nvfx_context *nvfx)
{
- while (*states) {
- struct nvfx_state_entry *e = *states;
+ struct nouveau_channel* chan = nvfx->screen->base.channel;
+ unsigned dirty;
+ int all_swizzled = -1;
+ boolean flush_tex_cache = FALSE;
+
+ if(nvfx != nvfx->screen->cur_ctx)
+ {
+ nvfx->dirty = ~0;
+ nvfx->screen->cur_ctx = nvfx;
+ }
- if (nvfx->dirty & e->dirty.pipe) {
- if (e->validate(nvfx))
- nvfx->state.dirty |= (1ULL << e->dirty.hw);
+ /* These can trigger use the of 3D engine to copy temporaries.
+ * That will recurse here and thus dirty all 3D state, so we need to this before anything else, and in a loop..
+ * This converges to having clean temps, then binding both fragtexes and framebuffers.
+ */
+ while(nvfx->dirty & (NVFX_NEW_FB | NVFX_NEW_SAMPLER))
+ {
+ if(nvfx->dirty & NVFX_NEW_SAMPLER)
+ {
+ nvfx->dirty &=~ NVFX_NEW_SAMPLER;
+ nvfx_fragtex_validate(nvfx);
+
+ // TODO: only set this if really necessary
+ flush_tex_cache = TRUE;
}
- states++;
+ if(nvfx->dirty & NVFX_NEW_FB)
+ {
+ nvfx->dirty &=~ NVFX_NEW_FB;
+ all_swizzled = nvfx_framebuffer_prepare(nvfx);
+
+ // TODO: make sure this doesn't happen, i.e. fbs have matching formats
+ assert(all_swizzled >= 0);
+ }
}
- nvfx->dirty = 0;
-}
-void
-nvfx_state_emit(struct nvfx_context *nvfx)
-{
- struct nvfx_state *state = &nvfx->state;
- struct nvfx_screen *screen = nvfx->screen;
- struct nouveau_channel *chan = screen->base.channel;
- struct nouveau_grobj *eng3d = screen->eng3d;
- unsigned i;
- uint64_t states;
-
- /* XXX: race conditions
- */
- if (nvfx != screen->cur_ctx) {
- for (i = 0; i < NVFX_STATE_MAX; i++) {
- if (state->hw[i] && screen->state[i] != state->hw[i])
- state->dirty |= (1ULL << i);
+ dirty = nvfx->dirty;
+
+ if(nvfx->render_mode == HW)
+ {
+ if(dirty & (NVFX_NEW_VERTPROG | NVFX_NEW_VERTCONST | NVFX_NEW_UCP))
+ {
+ if(!nvfx_vertprog_validate(nvfx))
+ return FALSE;
}
- screen->cur_ctx = nvfx;
+ if(dirty & (NVFX_NEW_ARRAYS))
+ {
+ if(!nvfx_vbo_validate(nvfx))
+ return FALSE;
+ }
}
+ else
+ {
+ /* TODO: this looks a bit misdesigned */
+ if(dirty & (NVFX_NEW_VERTPROG | NVFX_NEW_UCP))
+ nvfx_vertprog_validate(nvfx);
- for (i = 0, states = state->dirty; states; i++) {
- if (!(states & (1ULL << i)))
- continue;
- so_ref (state->hw[i], &nvfx->screen->state[i]);
- if (state->hw[i])
- so_emit(chan, nvfx->screen->state[i]);
- states &= ~(1ULL << i);
+ if(dirty & (NVFX_NEW_ARRAYS | NVFX_NEW_FRAGPROG))
+ nvfx_vtxfmt_validate(nvfx);
}
- /* TODO: could nv30 need this or something similar too? */
- if(nvfx->is_nv4x) {
- if (state->dirty & ((1ULL << NVFX_STATE_FRAGPROG) |
- (1ULL << NVFX_STATE_FRAGTEX0))) {
- BEGIN_RING(chan, eng3d, NV40TCL_TEX_CACHE_CTL, 1);
- OUT_RING (chan, 2);
- BEGIN_RING(chan, eng3d, NV40TCL_TEX_CACHE_CTL, 1);
- OUT_RING (chan, 1);
+ if(dirty & NVFX_NEW_RAST)
+ sb_emit(chan, nvfx->rasterizer->sb, nvfx->rasterizer->sb_len);
+
+ if(dirty & NVFX_NEW_SCISSOR)
+ nvfx_state_scissor_validate(nvfx);
+
+ if(dirty & NVFX_NEW_STIPPLE)
+ nvfx_state_stipple_validate(nvfx);
+
+ if(dirty & (NVFX_NEW_FRAGPROG | NVFX_NEW_FRAGCONST))
+ {
+ nvfx_fragprog_validate(nvfx);
+ if(dirty & NVFX_NEW_FRAGPROG)
+ flush_tex_cache = TRUE; // TODO: do we need this?
+ }
+
+ if(all_swizzled >= 0)
+ nvfx_framebuffer_validate(nvfx, all_swizzled);
+
+ if(dirty & NVFX_NEW_BLEND)
+ sb_emit(chan, nvfx->blend->sb, nvfx->blend->sb_len);
+
+ if(dirty & NVFX_NEW_BCOL)
+ nvfx_state_blend_colour_validate(nvfx);
+
+ if(dirty & NVFX_NEW_ZSA)
+ sb_emit(chan, nvfx->zsa->sb, nvfx->zsa->sb_len);
+
+ if(dirty & NVFX_NEW_SR)
+ nvfx_state_sr_validate(nvfx);
+
+/* Having this depend on FB looks wrong, but it seems
+ necessary to make this work on nv3x
+ TODO: find the right fix
+*/
+ if(dirty & (NVFX_NEW_VIEWPORT | NVFX_NEW_FB))
+ nvfx_state_viewport_validate(nvfx);
+
+ if(flush_tex_cache)
+ {
+ // TODO: what about nv30?
+ if(nvfx->is_nv4x)
+ {
+ WAIT_RING(chan, 4);
+ OUT_RING(chan, RING_3D(NV40TCL_TEX_CACHE_CTL, 1));
+ OUT_RING(chan, 2);
+ OUT_RING(chan, RING_3D(NV40TCL_TEX_CACHE_CTL, 1));
+ OUT_RING(chan, 1);
}
}
- state->dirty = 0;
+ nvfx->dirty = 0;
+ return TRUE;
}
void
-nvfx_state_flush_notify(struct nouveau_channel *chan)
+nvfx_state_emit(struct nvfx_context *nvfx)
{
- struct nvfx_context *nvfx = chan->user_private;
- struct nvfx_state *state = &nvfx->state;
- unsigned i, samplers;
-
- so_emit_reloc_markers(chan, state->hw[NVFX_STATE_FB]);
- for (i = 0, samplers = state->fp_samplers; i < 16 && samplers; i++) {
- if (!(samplers & (1 << i)))
- continue;
- so_emit_reloc_markers(chan,
- state->hw[NVFX_STATE_FRAGTEX0+i]);
- samplers &= ~(1ULL << i);
+ struct nouveau_channel* chan = nvfx->screen->base.channel;
+ /* we need to ensure there is enough space to output relocations in one go */
+ unsigned max_relocs = 0
+ + 16 /* vertex buffers, incl. dma flag */
+ + 2 /* index buffer plus format+dma flag */
+ + 2 * 5 /* 4 cbufs + zsbuf, plus dma objects */
+ + 2 * 16 /* fragment textures plus format+dma flag */
+ + 2 * 4 /* vertex textures plus format+dma flag */
+ + 1 /* fragprog incl dma flag */
+ ;
+ MARK_RING(chan, max_relocs * 2, max_relocs * 2);
+ nvfx_state_relocate(nvfx);
+
+ unsigned render_temps = nvfx->state.render_temps;
+ if(render_temps)
+ {
+ for(int i = 0; i < nvfx->framebuffer.nr_cbufs; ++i)
+ {
+ if(render_temps & (1 << i))
+ util_dirty_surface_set_dirty(nvfx_surface_get_dirty_surfaces(nvfx->framebuffer.cbufs[i]),
+ (struct util_dirty_surface*)nvfx->framebuffer.cbufs[i]);
+ }
+
+ if(render_temps & 0x80)
+ util_dirty_surface_set_dirty(nvfx_surface_get_dirty_surfaces(nvfx->framebuffer.zsbuf),
+ (struct util_dirty_surface*)nvfx->framebuffer.zsbuf);
}
- so_emit_reloc_markers(chan, state->hw[NVFX_STATE_FRAGPROG]);
- if (state->hw[NVFX_STATE_VTXBUF] && nvfx->render_mode == HW)
- so_emit_reloc_markers(chan, state->hw[NVFX_STATE_VTXBUF]);
+}
+
+void
+nvfx_state_relocate(struct nvfx_context *nvfx)
+{
+ nvfx_framebuffer_relocate(nvfx);
+ nvfx_fragtex_relocate(nvfx);
+ nvfx_fragprog_relocate(nvfx);
+ if (nvfx->render_mode == HW)
+ nvfx_vbo_relocate(nvfx);
}
boolean
return FALSE;
/* Attempt to go to hwtnl again */
- nvfx->pipe.flush(&nvfx->pipe, 0, NULL);
nvfx->dirty |= (NVFX_NEW_VIEWPORT |
NVFX_NEW_VERTPROG |
NVFX_NEW_ARRAYS);
nvfx->render_mode = HW;
}
- if(!nvfx->is_nv4x)
- nvfx_state_do_validate(nvfx, nv30_render_states);
- else
- nvfx_state_do_validate(nvfx, nv40_render_states);
-
- if (nvfx->fallback_swtnl || nvfx->fallback_swrast)
+ if(!nvfx_state_validate_common(nvfx))
return FALSE;
if (was_sw)
draw_bind_vertex_shader(draw, nvfx->vertprog->draw);
if (nvfx->draw_dirty & NVFX_NEW_RAST)
- draw_set_rasterizer_state(draw, &nvfx->rasterizer->pipe);
+ draw_set_rasterizer_state(draw, &nvfx->rasterizer->pipe,
+ nvfx->rasterizer);
if (nvfx->draw_dirty & NVFX_NEW_UCP)
draw_set_clip_state(draw, &nvfx->clip);
draw_set_vertex_elements(draw, nvfx->vtxelt->num_elements, nvfx->vtxelt->pipe);
}
- if(!nvfx->is_nv4x)
- nvfx_state_do_validate(nvfx, nv30_swtnl_render_states);
- else
- nvfx_state_do_validate(nvfx, nv40_swtnl_render_states);
-
- if (nvfx->fallback_swrast) {
- NOUVEAU_ERR("swtnl->swrast 0x%08x\n", nvfx->fallback_swrast);
- return FALSE;
- }
+ nvfx_state_validate_common(nvfx);
nvfx->draw_dirty = 0;
return TRUE;