X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fgallium%2Fdrivers%2Fnv30%2Fnv30_state_emit.c;h=9480695d6e57b1a147286cf3c84a16d306b81d5c;hb=e3734593aea202e99e77febea7b86c904080939f;hp=eda2fd45f569e24ff3112f4cef2a3e84188ac95e;hpb=92120851947ad4a47d2150a7cf3f8dc5fdde396c;p=mesa.git diff --git a/src/gallium/drivers/nv30/nv30_state_emit.c b/src/gallium/drivers/nv30/nv30_state_emit.c index eda2fd45f56..9480695d6e5 100644 --- a/src/gallium/drivers/nv30/nv30_state_emit.c +++ b/src/gallium/drivers/nv30/nv30_state_emit.c @@ -1,85 +1,118 @@ #include "nv30_context.h" #include "nv30_state.h" -void -nv30_emit_hw_state(struct nv30_context *nv30) +static struct nv30_state_entry *render_states[] = { + &nv30_state_framebuffer, + &nv30_state_rasterizer, + &nv30_state_scissor, + &nv30_state_stipple, + &nv30_state_fragprog, + &nv30_state_fragtex, + &nv30_state_vertprog, + &nv30_state_blend, + &nv30_state_blend_colour, + &nv30_state_zsa, + &nv30_state_viewport, + &nv30_state_vbo, + NULL +}; + +static void +nv30_state_do_validate(struct nv30_context *nv30, + struct nv30_state_entry **states) { - int i; + const struct pipe_framebuffer_state *fb = &nv30->framebuffer; + unsigned i; - if (nv30->dirty & NV30_NEW_FRAGPROG) { - nv30_fragprog_bind(nv30, nv30->fragprog.current); - /*XXX: clear NV30_NEW_FRAGPROG if no new program uploaded */ - } + 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; - if (nv30->dirty_samplers || (nv30->dirty & NV30_NEW_FRAGPROG)) { - nv30_fragtex_bind(nv30); -/* - BEGIN_RING(rankine, NV34TCL_TX_CACHE_CTL, 1); - OUT_RING (2); - BEGIN_RING(rankine, NV34TCL_TX_CACHE_CTL, 1); - OUT_RING (1);*/ - nv30->dirty &= ~NV30_NEW_FRAGPROG; - } + 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); + } + } - if (nv30->dirty & NV30_NEW_VERTPROG) { - nv30_vertprog_bind(nv30, nv30->vertprog.current); - nv30->dirty &= ~NV30_NEW_VERTPROG; + states++; } + nv30->dirty = 0; +} - nv30->dirty_samplers = 0; +void +nv30_state_emit(struct nv30_context *nv30) +{ + struct nv30_state *state = &nv30->state; + struct nv30_screen *screen = nv30->screen; + unsigned i, samplers; + uint64_t states; - /* 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 - */ + if (nv30->pctx_id != screen->cur_pctx) { + for (i = 0; i < NV30_STATE_MAX; i++) { + if (state->hw[i] && screen->state[i] != state->hw[i]) + state->dirty |= (1ULL << i); + } - /* 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); + screen->cur_pctx = nv30->pctx_id; } - 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); + 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->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);*/ + state->dirty = 0; + + so_emit_reloc_markers(nv30->nvws, state->hw[NV30_STATE_FB]); + for (i = 0, samplers = state->fp_samplers; i < 16 && samplers; i++) { + if (!(samplers & (1 << i))) + continue; + so_emit_reloc_markers(nv30->nvws, + state->hw[NV30_STATE_FRAGTEX0+i]); + samplers &= ~(1ULL << i); } + so_emit_reloc_markers(nv30->nvws, state->hw[NV30_STATE_FRAGPROG]); + if (state->hw[NV30_STATE_VTXBUF] /*&& nv30->render_mode == HW*/) + so_emit_reloc_markers(nv30->nvws, state->hw[NV30_STATE_VTXBUF]); +} - /* Texture images, emitted in nv30_fragtex_build */ +boolean +nv30_state_validate(struct nv30_context *nv30) +{ #if 0 - for (i = 0; i < 16; i++) { - if (!(nv30->fp_samplers & (1 << i))) - continue; - BEGIN_RING(rankine, NV34TCL_TX_OFFSET(i), 2); - OUT_RELOCl(nv30->tex[i].buffer, 0, NOUVEAU_BO_VRAM | - NOUVEAU_BO_GART | NOUVEAU_BO_RD); - OUT_RELOCd(nv30->tex[i].buffer, nv30->tex[i].format, - NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD | - NOUVEAU_BO_OR, NV34TCL_TX_FORMAT_DMA0, - NV34TCL_TX_FORMAT_DMA1); + 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 - - /* Fragment program */ - BEGIN_RING(rankine, NV34TCL_FP_ACTIVE_PROGRAM, 1); - OUT_RELOC (nv30->fragprog.active->buffer, 0, NOUVEAU_BO_VRAM | - NOUVEAU_BO_GART | NOUVEAU_BO_RD | NOUVEAU_BO_LOW | - NOUVEAU_BO_OR, NV34TCL_FP_ACTIVE_PROGRAM_DMA0, - NV34TCL_FP_ACTIVE_PROGRAM_DMA1); + 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; } -