X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fgallium%2Fdrivers%2Fnv50%2Fnv50_state_validate.c;h=956a70061541429f19b3ed4c3af155c210987a2c;hb=0ea575d721821262a862ceef010db9b1a8b4a6d9;hp=d313e9de4fefd50b9d65ff38e197183b12db22a9;hpb=862488075c5537b0613753b0d14c267527fc6199;p=mesa.git diff --git a/src/gallium/drivers/nv50/nv50_state_validate.c b/src/gallium/drivers/nv50/nv50_state_validate.c index d313e9de4fe..956a7006154 100644 --- a/src/gallium/drivers/nv50/nv50_state_validate.c +++ b/src/gallium/drivers/nv50/nv50_state_validate.c @@ -23,6 +23,12 @@ #include "nv50_context.h" #include "nouveau/nouveau_stateobj.h" +#define NV50_CBUF_FORMAT_CASE(n) \ + case PIPE_FORMAT_##n: so_data(so, NV50TCL_RT_FORMAT_##n); break + +#define NV50_ZETA_FORMAT_CASE(n) \ + case PIPE_FORMAT_##n: so_data(so, NV50TCL_ZETA_FORMAT_##n); break + static void nv50_state_validate_fb(struct nv50_context *nv50) { @@ -33,7 +39,7 @@ nv50_state_validate_fb(struct nv50_context *nv50) for (i = 0; i < fb->nr_cbufs; i++) { struct pipe_texture *pt = fb->cbufs[i]->texture; - struct nouveau_bo *bo = nv50_miptree(pt)->bo; + struct nouveau_bo *bo = nv50_miptree(pt)->base.bo; if (!gw) { w = fb->cbufs[i]->width; @@ -54,19 +60,22 @@ nv50_state_validate_fb(struct nv50_context *nv50) so_reloc (so, bo, fb->cbufs[i]->offset, NOUVEAU_BO_VRAM | NOUVEAU_BO_LOW | NOUVEAU_BO_RDWR, 0, 0); switch (fb->cbufs[i]->format) { - case PIPE_FORMAT_A8R8G8B8_UNORM: - so_data(so, 0xcf); - break; - case PIPE_FORMAT_R5G6B5_UNORM: - so_data(so, 0xe8); - break; + NV50_CBUF_FORMAT_CASE(A8R8G8B8_UNORM); + NV50_CBUF_FORMAT_CASE(X8R8G8B8_UNORM); + NV50_CBUF_FORMAT_CASE(R5G6B5_UNORM); + NV50_CBUF_FORMAT_CASE(R16G16B16A16_SNORM); + NV50_CBUF_FORMAT_CASE(R16G16B16A16_UNORM); + NV50_CBUF_FORMAT_CASE(R32G32B32A32_FLOAT); + NV50_CBUF_FORMAT_CASE(R16G16_SNORM); + NV50_CBUF_FORMAT_CASE(R16G16_UNORM); default: NOUVEAU_ERR("AIIII unknown format %s\n", pf_name(fb->cbufs[i]->format)); - so_data(so, 0xe6); + so_data(so, NV50TCL_RT_FORMAT_X8R8G8B8_UNORM); break; } - so_data(so, bo->tile_mode << 4); + so_data(so, nv50_miptree(pt)-> + level[fb->cbufs[i]->level].tile_mode << 4); so_data(so, 0x00000000); so_method(so, tesla, 0x1224, 1); @@ -75,7 +84,7 @@ nv50_state_validate_fb(struct nv50_context *nv50) if (fb->zsbuf) { struct pipe_texture *pt = fb->zsbuf->texture; - struct nouveau_bo *bo = nv50_miptree(pt)->bo; + struct nouveau_bo *bo = nv50_miptree(pt)->base.bo; if (!gw) { w = fb->zsbuf->width; @@ -92,25 +101,23 @@ nv50_state_validate_fb(struct nv50_context *nv50) so_reloc (so, bo, fb->zsbuf->offset, NOUVEAU_BO_VRAM | NOUVEAU_BO_LOW | NOUVEAU_BO_RDWR, 0, 0); switch (fb->zsbuf->format) { - case PIPE_FORMAT_Z24S8_UNORM: - case PIPE_FORMAT_Z24X8_UNORM: - so_data(so, 0x16); - break; - case PIPE_FORMAT_Z16_UNORM: - so_data(so, 0x15); - break; + NV50_ZETA_FORMAT_CASE(S8Z24_UNORM); + NV50_ZETA_FORMAT_CASE(X8Z24_UNORM); + NV50_ZETA_FORMAT_CASE(Z24S8_UNORM); + NV50_ZETA_FORMAT_CASE(Z32_FLOAT); default: NOUVEAU_ERR("AIIII unknown format %s\n", pf_name(fb->zsbuf->format)); - so_data(so, 0x16); + so_data(so, NV50TCL_ZETA_FORMAT_S8Z24_UNORM); break; } - so_data(so, bo->tile_mode << 4); + so_data(so, nv50_miptree(pt)-> + level[fb->zsbuf->level].tile_mode << 4); so_data(so, 0x00000000); so_method(so, tesla, 0x1538, 1); so_data (so, 1); - so_method(so, tesla, 0x1228, 3); + so_method(so, tesla, NV50TCL_ZETA_HORIZ, 3); so_data (so, fb->zsbuf->width); so_data (so, fb->zsbuf->height); so_data (so, 0x00010001); @@ -119,12 +126,18 @@ nv50_state_validate_fb(struct nv50_context *nv50) so_method(so, tesla, NV50TCL_VIEWPORT_HORIZ, 2); so_data (so, w << 16); so_data (so, h << 16); - so_method(so, tesla, 0x0e04, 2); + /* set window lower left corner */ + so_method(so, tesla, NV50TCL_WINDOW_LEFT, 2); + so_data (so, 0); + so_data (so, 0); + /* set screen scissor rectangle */ + so_method(so, tesla, NV50TCL_SCREEN_SCISSOR_HORIZ, 2); so_data (so, w << 16); so_data (so, h << 16); - so_method(so, tesla, 0xdf8, 2); - so_data (so, 0); - so_data (so, h); + + /* we set scissors to framebuffer size when they're 'turned off' */ + nv50->dirty |= NV50_NEW_SCISSOR; + so_ref(NULL, &nv50->state.scissor); so_ref(so, &nv50->state.fb); so_ref(NULL, &so); @@ -137,7 +150,32 @@ nv50_state_emit(struct nv50_context *nv50) struct nouveau_channel *chan = screen->base.channel; if (nv50->pctx_id != screen->cur_pctx) { - nv50->state.dirty |= 0xffffffff; + if (nv50->state.fb) + nv50->state.dirty |= NV50_NEW_FRAMEBUFFER; + if (nv50->state.blend) + nv50->state.dirty |= NV50_NEW_BLEND; + if (nv50->state.zsa) + nv50->state.dirty |= NV50_NEW_ZSA; + if (nv50->state.vertprog) + nv50->state.dirty |= NV50_NEW_VERTPROG; + if (nv50->state.fragprog) + nv50->state.dirty |= NV50_NEW_FRAGPROG; + if (nv50->state.rast) + nv50->state.dirty |= NV50_NEW_RASTERIZER; + if (nv50->state.blend_colour) + nv50->state.dirty |= NV50_NEW_BLEND_COLOUR; + if (nv50->state.stipple) + nv50->state.dirty |= NV50_NEW_STIPPLE; + if (nv50->state.scissor) + nv50->state.dirty |= NV50_NEW_SCISSOR; + if (nv50->state.viewport) + nv50->state.dirty |= NV50_NEW_VIEWPORT; + if (nv50->state.tsc_upload) + nv50->state.dirty |= NV50_NEW_SAMPLER; + if (nv50->state.tic_upload) + nv50->state.dirty |= NV50_NEW_TEXTURE; + if (nv50->state.vtxfmt && nv50->state.vtxbuf) + nv50->state.dirty |= NV50_NEW_ARRAYS; screen->cur_pctx = nv50->pctx_id; } @@ -151,6 +189,8 @@ nv50_state_emit(struct nv50_context *nv50) so_emit(chan, nv50->state.vertprog); if (nv50->state.dirty & NV50_NEW_FRAGPROG) so_emit(chan, nv50->state.fragprog); + if (nv50->state.dirty & (NV50_NEW_FRAGPROG | NV50_NEW_VERTPROG)) + so_emit(chan, nv50->state.programs); if (nv50->state.dirty & NV50_NEW_RASTERIZER) so_emit(chan, nv50->state.rast); if (nv50->state.dirty & NV50_NEW_BLEND_COLOUR) @@ -168,8 +208,19 @@ nv50_state_emit(struct nv50_context *nv50) if (nv50->state.dirty & NV50_NEW_ARRAYS) { so_emit(chan, nv50->state.vtxfmt); so_emit(chan, nv50->state.vtxbuf); + if (nv50->state.vtxattr) + so_emit(chan, nv50->state.vtxattr); } nv50->state.dirty = 0; +} + +void +nv50_state_flush_notify(struct nouveau_channel *chan) +{ + struct nv50_context *nv50 = chan->user_private; + + if (nv50->state.tic_upload && !(nv50->dirty & NV50_NEW_TEXTURE)) + so_emit(chan, nv50->state.tic_upload); so_emit_reloc_markers(chan, nv50->state.fb); so_emit_reloc_markers(chan, nv50->state.vertprog); @@ -182,6 +233,7 @@ boolean nv50_state_validate(struct nv50_context *nv50) { struct nouveau_grobj *tesla = nv50->screen->tesla; + struct nouveau_grobj *eng2d = nv50->screen->eng2d; struct nouveau_stateobj *so; unsigned i; @@ -200,6 +252,9 @@ nv50_state_validate(struct nv50_context *nv50) if (nv50->dirty & (NV50_NEW_FRAGPROG | NV50_NEW_FRAGPROG_CB)) nv50_fragprog_validate(nv50); + if (nv50->dirty & (NV50_NEW_FRAGPROG | NV50_NEW_VERTPROG)) + nv50_linkage_validate(nv50); + if (nv50->dirty & NV50_NEW_RASTERIZER) so_ref(nv50->rasterizer->so, &nv50->state.rast); @@ -233,13 +288,13 @@ nv50_state_validate(struct nv50_context *nv50) nv50->state.scissor_enabled = rast->scissor; so = so_new(3, 0); - so_method(so, tesla, 0x0ff4, 2); + so_method(so, tesla, NV50TCL_SCISSOR_HORIZ, 2); if (nv50->state.scissor_enabled) { - so_data(so, ((s->maxx - s->minx) << 16) | s->minx); - so_data(so, ((s->maxy - s->miny) << 16) | s->miny); + so_data(so, (s->maxx << 16) | s->minx); + so_data(so, (s->maxy << 16) | s->miny); } else { - so_data(so, (8192 << 16)); - so_data(so, (8192 << 16)); + so_data(so, (nv50->framebuffer.width << 16)); + so_data(so, (nv50->framebuffer.height << 16)); } so_ref(so, &nv50->state.scissor); so_ref(NULL, &so); @@ -261,23 +316,34 @@ scissor_uptodate: goto viewport_uptodate; nv50->state.viewport_bypass = bypass; - so = so_new(12, 0); + so = so_new(14, 0); if (!bypass) { - so_method(so, tesla, NV50TCL_VIEWPORT_UNK1(0), 3); + so_method(so, tesla, NV50TCL_VIEWPORT_TRANSLATE(0), 3); so_data (so, fui(nv50->viewport.translate[0])); so_data (so, fui(nv50->viewport.translate[1])); so_data (so, fui(nv50->viewport.translate[2])); - so_method(so, tesla, NV50TCL_VIEWPORT_UNK0(0), 3); + so_method(so, tesla, NV50TCL_VIEWPORT_SCALE(0), 3); so_data (so, fui(nv50->viewport.scale[0])); - so_data (so, fui(-nv50->viewport.scale[1])); + so_data (so, fui(nv50->viewport.scale[1])); so_data (so, fui(nv50->viewport.scale[2])); - so_method(so, tesla, 0x192c, 1); + + so_method(so, tesla, NV50TCL_VIEWPORT_TRANSFORM_EN, 1); so_data (so, 1); + /* 0x0000 = remove whole primitive only (xyz) + * 0x1018 = remove whole primitive only (xy), clamp z + * 0x1080 = clip primitive (xyz) + * 0x1098 = clip primitive (xy), clamp z + */ + so_method(so, tesla, NV50TCL_VIEW_VOLUME_CLIP_CTRL, 1); + so_data (so, 0x1080); + /* no idea what 0f90 does */ so_method(so, tesla, 0x0f90, 1); so_data (so, 0); } else { - so_method(so, tesla, 0x192c, 1); + so_method(so, tesla, NV50TCL_VIEWPORT_TRANSFORM_EN, 1); so_data (so, 0); + so_method(so, tesla, NV50TCL_VIEW_VOLUME_CLIP_CTRL, 1); + so_data (so, 0x0000); so_method(so, tesla, 0x0f90, 1); so_data (so, 1); } @@ -289,14 +355,25 @@ scissor_uptodate: viewport_uptodate: if (nv50->dirty & NV50_NEW_SAMPLER) { - int i; + unsigned i; + + so = so_new(nv50->sampler_nr * 9 + 23 + 4, 2); - so = so_new(nv50->sampler_nr * 8 + 3, 0); - so_method(so, tesla, 0x0f00, 1); - so_data (so, NV50_CB_TSC); - so_method(so, tesla, 0x40000f04, nv50->sampler_nr * 8); - for (i = 0; i < nv50->sampler_nr; i++) + nv50_so_init_sifc(nv50, so, nv50->screen->tsc, NOUVEAU_BO_VRAM, + nv50->sampler_nr * 8 * 4); + + for (i = 0; i < nv50->sampler_nr; i++) { + if (!nv50->sampler[i]) + continue; + so_method(so, eng2d, NV50_2D_SIFC_DATA | (2 << 29), 8); so_datap (so, nv50->sampler[i]->tsc, 8); + } + + so_method(so, tesla, 0x1440, 1); /* sync SIFC */ + so_data (so, 0); + so_method(so, tesla, 0x1334, 1); /* flush TSC */ + so_data (so, 0); + so_ref(so, &nv50->state.tsc_upload); so_ref(NULL, &so); } @@ -314,3 +391,33 @@ viewport_uptodate: return TRUE; } +void nv50_so_init_sifc(struct nv50_context *nv50, + struct nouveau_stateobj *so, + struct nouveau_bo *bo, unsigned reloc, unsigned size) +{ + struct nouveau_grobj *eng2d = nv50->screen->eng2d; + + so_method(so, eng2d, NV50_2D_DST_FORMAT, 2); + so_data (so, NV50_2D_DST_FORMAT_R8_UNORM); + so_data (so, 1); + so_method(so, eng2d, NV50_2D_DST_PITCH, 5); + so_data (so, 262144); + so_data (so, 65536); + so_data (so, 1); + so_reloc (so, bo, 0, reloc | NOUVEAU_BO_WR | NOUVEAU_BO_HIGH, 0, 0); + so_reloc (so, bo, 0, reloc | NOUVEAU_BO_WR | NOUVEAU_BO_LOW, 0, 0); + so_method(so, eng2d, NV50_2D_SIFC_UNK0800, 2); + so_data (so, 0); + so_data (so, NV50_2D_SIFC_FORMAT_R8_UNORM); + so_method(so, eng2d, NV50_2D_SIFC_WIDTH, 10); + so_data (so, size); + so_data (so, 1); + so_data (so, 0); + so_data (so, 1); + so_data (so, 0); + so_data (so, 1); + so_data (so, 0); + so_data (so, 0); + so_data (so, 0); + so_data (so, 0); +}