From 62ab89785b55e60b978dc2b32995676859299c80 Mon Sep 17 00:00:00 2001 From: Ben Skeggs Date: Wed, 3 Mar 2010 14:26:25 +1000 Subject: nv50: rework state emission --- src/gallium/drivers/nv50/nv50_state_validate.c | 430 ++++++++++++------------- 1 file changed, 209 insertions(+), 221 deletions(-) (limited to 'src/gallium/drivers/nv50/nv50_state_validate.c') diff --git a/src/gallium/drivers/nv50/nv50_state_validate.c b/src/gallium/drivers/nv50/nv50_state_validate.c index 0d30957ee3..abdb9a55c8 100644 --- a/src/gallium/drivers/nv50/nv50_state_validate.c +++ b/src/gallium/drivers/nv50/nv50_state_validate.c @@ -25,8 +25,8 @@ #include "nv50_context.h" #include "nouveau/nouveau_stateobj.h" -static void -nv50_state_validate_fb(struct nv50_context *nv50) +static struct nouveau_stateobj * +validate_fb(struct nv50_context *nv50) { struct nouveau_grobj *tesla = nv50->screen->tesla; struct nouveau_stateobj *so = so_new(32, 79, 18); @@ -169,10 +169,8 @@ nv50_state_validate_fb(struct nv50_context *nv50) /* 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); + so_ref(NULL, &nv50->state.hw[12]); + return so; } static void @@ -199,261 +197,251 @@ nv50_validate_samplers(struct nv50_context *nv50, struct nouveau_stateobj *so, } } -static void -nv50_state_emit(struct nv50_context *nv50) +static struct nouveau_stateobj * +validate_blend(struct nv50_context *nv50) { - struct nv50_screen *screen = nv50->screen; - struct nouveau_channel *chan = screen->base.channel; + struct nouveau_stateobj *so = NULL; + so_ref(nv50->blend->so, &so); + return so; +} - /* XXX: this is racy for multiple contexts active on separate - * threads. - */ - if (screen->cur_ctx != nv50) { - 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.geomprog) - nv50->state.dirty |= NV50_NEW_GEOMPROG; - 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.stencil_ref) - nv50->state.dirty |= NV50_NEW_STENCIL_REF; - 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_ctx = nv50; - } +static struct nouveau_stateobj * +validate_zsa(struct nv50_context *nv50) +{ + struct nouveau_stateobj *so = NULL; + so_ref(nv50->zsa->so, &so); + return so; +} - if (nv50->state.dirty & NV50_NEW_FRAMEBUFFER) - so_emit(chan, nv50->state.fb); - if (nv50->state.dirty & NV50_NEW_BLEND) - so_emit(chan, nv50->state.blend); - if (nv50->state.dirty & NV50_NEW_ZSA) - so_emit(chan, nv50->state.zsa); - if (nv50->state.dirty & NV50_NEW_VERTPROG) - 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_GEOMPROG && nv50->state.geomprog) - so_emit(chan, nv50->state.geomprog); - if (nv50->state.dirty & (NV50_NEW_FRAGPROG | NV50_NEW_VERTPROG | - NV50_NEW_GEOMPROG | NV50_NEW_RASTERIZER)) - so_emit(chan, nv50->state.fp_linkage); - if ((nv50->state.dirty & (NV50_NEW_VERTPROG | NV50_NEW_GEOMPROG)) - && nv50->state.gp_linkage) - so_emit(chan, nv50->state.gp_linkage); - if (nv50->state.dirty & NV50_NEW_RASTERIZER) - so_emit(chan, nv50->state.rast); - if (nv50->state.dirty & NV50_NEW_BLEND_COLOUR) - so_emit(chan, nv50->state.blend_colour); - if (nv50->state.dirty & NV50_NEW_STENCIL_REF) - so_emit(chan, nv50->state.stencil_ref); - if (nv50->state.dirty & NV50_NEW_STIPPLE) - so_emit(chan, nv50->state.stipple); - if (nv50->state.dirty & NV50_NEW_SCISSOR) - so_emit(chan, nv50->state.scissor); - if (nv50->state.dirty & NV50_NEW_VIEWPORT) - so_emit(chan, nv50->state.viewport); - if (nv50->state.dirty & NV50_NEW_SAMPLER) - so_emit(chan, nv50->state.tsc_upload); - if (nv50->state.dirty & NV50_NEW_TEXTURE) - so_emit(chan, nv50->state.tic_upload); - 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; +static struct nouveau_stateobj * +validate_rast(struct nv50_context *nv50) +{ + struct nouveau_stateobj *so = NULL; + so_ref(nv50->rasterizer->so, &so); + return so; } -void -nv50_state_flush_notify(struct nouveau_channel *chan) +static struct nouveau_stateobj * +validate_blend_colour(struct nv50_context *nv50) { - struct nv50_context *nv50 = chan->user_private; + struct nouveau_grobj *tesla = nv50->screen->tesla; + struct nouveau_stateobj *so = so_new(1, 4, 0); + + so_method(so, tesla, NV50TCL_BLEND_COLOR(0), 4); + so_data (so, fui(nv50->blend_colour.color[0])); + so_data (so, fui(nv50->blend_colour.color[1])); + so_data (so, fui(nv50->blend_colour.color[2])); + so_data (so, fui(nv50->blend_colour.color[3])); + return so; +} - nv50_tex_relocs(nv50); +static struct nouveau_stateobj * +validate_stencil_ref(struct nv50_context *nv50) +{ + struct nouveau_grobj *tesla = nv50->screen->tesla; + struct nouveau_stateobj *so = so = so_new(2, 2, 0); - so_emit_reloc_markers(chan, nv50->state.fb); - so_emit_reloc_markers(chan, nv50->state.vertprog); - so_emit_reloc_markers(chan, nv50->state.fragprog); - so_emit_reloc_markers(chan, nv50->state.vtxbuf); - so_emit_reloc_markers(chan, nv50->screen->static_init); + so_method(so, tesla, NV50TCL_STENCIL_FRONT_FUNC_REF, 1); + so_data (so, nv50->stencil_ref.ref_value[0]); + so_method(so, tesla, NV50TCL_STENCIL_BACK_FUNC_REF, 1); + so_data (so, nv50->stencil_ref.ref_value[1]); + return so; +} - if (nv50->state.instbuf) - so_emit_reloc_markers(chan, nv50->state.instbuf); +static struct nouveau_stateobj * +validate_stipple(struct nv50_context *nv50) +{ + struct nouveau_grobj *tesla = nv50->screen->tesla; + struct nouveau_stateobj *so = so_new(1, 32, 0); + int i; + + so_method(so, tesla, NV50TCL_POLYGON_STIPPLE_PATTERN(0), 32); + for (i = 0; i < 32; i++) + so_data(so, util_bswap32(nv50->stipple.stipple[i])); + return so; } -boolean -nv50_state_validate(struct nv50_context *nv50) +static struct nouveau_stateobj * +validate_scissor(struct nv50_context *nv50) { + struct pipe_rasterizer_state *rast = &nv50->rasterizer->pipe; + struct pipe_scissor_state *s = &nv50->scissor; struct nouveau_grobj *tesla = nv50->screen->tesla; struct nouveau_stateobj *so; - unsigned i; - if (nv50->dirty & NV50_NEW_FRAMEBUFFER) - nv50_state_validate_fb(nv50); + if (nv50->state.hw[12] && + (rast->scissor == 0 && nv50->state.scissor_enabled == 0)) + return NULL; + nv50->state.scissor_enabled = rast->scissor; - if (nv50->dirty & NV50_NEW_BLEND) - so_ref(nv50->blend->so, &nv50->state.blend); + so = so_new(1, 2, 0); + so_method(so, tesla, NV50TCL_SCISSOR_HORIZ(0), 2); + if (nv50->state.scissor_enabled) { + so_data(so, (s->maxx << 16) | s->minx); + so_data(so, (s->maxy << 16) | s->miny); + } else { + so_data(so, (nv50->framebuffer.width << 16)); + so_data(so, (nv50->framebuffer.height << 16)); + } - if (nv50->dirty & NV50_NEW_ZSA) - so_ref(nv50->zsa->so, &nv50->state.zsa); + return so; +} - if (nv50->dirty & (NV50_NEW_VERTPROG | NV50_NEW_VERTPROG_CB)) - nv50_vertprog_validate(nv50); +static struct nouveau_stateobj * +validate_viewport(struct nv50_context *nv50) +{ + struct nouveau_grobj *tesla = nv50->screen->tesla; + struct nouveau_stateobj *so = so_new(5, 9, 0); + + so_method(so, tesla, NV50TCL_VIEWPORT_TRANSLATE_X(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_SCALE_X(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[2])); + + 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); - if (nv50->dirty & (NV50_NEW_FRAGPROG | NV50_NEW_FRAGPROG_CB)) - nv50_fragprog_validate(nv50); + return so; +} - if (nv50->dirty & (NV50_NEW_GEOMPROG | NV50_NEW_GEOMPROG_CB)) - nv50_geomprog_validate(nv50); +static struct nouveau_stateobj * +validate_sampler(struct nv50_context *nv50) +{ + struct nouveau_grobj *tesla = nv50->screen->tesla; + struct nouveau_stateobj *so; + unsigned nr = 0, i; - if (nv50->dirty & (NV50_NEW_FRAGPROG | NV50_NEW_VERTPROG | - NV50_NEW_GEOMPROG | NV50_NEW_RASTERIZER)) - nv50_fp_linkage_validate(nv50); + for (i = 0; i < PIPE_SHADER_TYPES; ++i) + nr += nv50->sampler_nr[i]; - if (nv50->dirty & (NV50_NEW_GEOMPROG | NV50_NEW_VERTPROG)) - nv50_gp_linkage_validate(nv50); + so = so_new(1 + 5 * PIPE_SHADER_TYPES, + 1 + 19 * PIPE_SHADER_TYPES + nr * 8, + PIPE_SHADER_TYPES * 2); - if (nv50->dirty & NV50_NEW_RASTERIZER) - so_ref(nv50->rasterizer->so, &nv50->state.rast); + nv50_validate_samplers(nv50, so, PIPE_SHADER_VERTEX); + nv50_validate_samplers(nv50, so, PIPE_SHADER_FRAGMENT); - if (nv50->dirty & NV50_NEW_BLEND_COLOUR) { - so = so_new(1, 4, 0); - so_method(so, tesla, NV50TCL_BLEND_COLOR(0), 4); - so_data (so, fui(nv50->blend_colour.color[0])); - so_data (so, fui(nv50->blend_colour.color[1])); - so_data (so, fui(nv50->blend_colour.color[2])); - so_data (so, fui(nv50->blend_colour.color[3])); - so_ref(so, &nv50->state.blend_colour); - so_ref(NULL, &so); - } - - if (nv50->dirty & NV50_NEW_STENCIL_REF) { - so = so_new(2, 2, 0); - so_method(so, tesla, NV50TCL_STENCIL_FRONT_FUNC_REF, 1); - so_data (so, nv50->stencil_ref.ref_value[0]); - so_method(so, tesla, NV50TCL_STENCIL_BACK_FUNC_REF, 1); - so_data (so, nv50->stencil_ref.ref_value[1]); - so_ref(so, &nv50->state.stencil_ref); - so_ref(NULL, &so); - } + so_method(so, tesla, 0x1334, 1); /* flush TSC */ + so_data (so, 0); - if (nv50->dirty & NV50_NEW_STIPPLE) { - so = so_new(1, 32, 0); - so_method(so, tesla, NV50TCL_POLYGON_STIPPLE_PATTERN(0), 32); - for (i = 0; i < 32; i++) - so_data(so, util_bswap32(nv50->stipple.stipple[i])); - so_ref(so, &nv50->state.stipple); - so_ref(NULL, &so); - } + return so; +} - if (nv50->dirty & (NV50_NEW_SCISSOR | NV50_NEW_RASTERIZER)) { - struct pipe_rasterizer_state *rast = &nv50->rasterizer->pipe; - struct pipe_scissor_state *s = &nv50->scissor; +static struct nouveau_stateobj * +validate_vtxbuf(struct nv50_context *nv50) +{ + struct nouveau_stateobj *so = NULL; + so_ref(nv50->state.vtxbuf, &so); + return so; +} - if (nv50->state.scissor && - (rast->scissor == 0 && nv50->state.scissor_enabled == 0)) - goto scissor_uptodate; - nv50->state.scissor_enabled = rast->scissor; +static struct nouveau_stateobj * +validate_vtxattr(struct nv50_context *nv50) +{ + struct nouveau_stateobj *so = NULL; + so_ref(nv50->state.vtxattr, &so); + return so; +} - so = so_new(1, 2, 0); - so_method(so, tesla, NV50TCL_SCISSOR_HORIZ(0), 2); - if (nv50->state.scissor_enabled) { - so_data(so, (s->maxx << 16) | s->minx); - so_data(so, (s->maxy << 16) | s->miny); - } else { - so_data(so, (nv50->framebuffer.width << 16)); - so_data(so, (nv50->framebuffer.height << 16)); - } - so_ref(so, &nv50->state.scissor); - so_ref(NULL, &so); - nv50->state.dirty |= NV50_NEW_SCISSOR; - } -scissor_uptodate: - - if (nv50->dirty & (NV50_NEW_VIEWPORT | NV50_NEW_RASTERIZER)) { - if (nv50->state.viewport && - !(nv50->dirty & NV50_NEW_VIEWPORT)) - goto viewport_uptodate; - - so = so_new(5, 9, 0); - so_method(so, tesla, NV50TCL_VIEWPORT_TRANSLATE_X(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_SCALE_X(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[2])); - - 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); +struct state_validate { + struct nouveau_stateobj *(*func)(struct nv50_context *nv50); + unsigned states; +} validate_list[] = { + { validate_fb , NV50_NEW_FRAMEBUFFER }, + { validate_blend , NV50_NEW_BLEND }, + { validate_zsa , NV50_NEW_ZSA }, + { nv50_vertprog_validate , NV50_NEW_VERTPROG | NV50_NEW_VERTPROG_CB }, + { nv50_fragprog_validate , NV50_NEW_FRAGPROG | NV50_NEW_FRAGPROG_CB }, + { nv50_geomprog_validate , NV50_NEW_GEOMPROG | NV50_NEW_GEOMPROG_CB }, + { nv50_fp_linkage_validate, NV50_NEW_VERTPROG | NV50_NEW_GEOMPROG | + NV50_NEW_FRAGPROG | NV50_NEW_RASTERIZER }, + { nv50_gp_linkage_validate, NV50_NEW_VERTPROG | NV50_NEW_GEOMPROG }, + { validate_rast , NV50_NEW_RASTERIZER }, + { validate_blend_colour , NV50_NEW_BLEND_COLOUR }, + { validate_stencil_ref , NV50_NEW_STENCIL_REF }, + { validate_stipple , NV50_NEW_STIPPLE }, + { validate_scissor , NV50_NEW_SCISSOR | NV50_NEW_RASTERIZER }, + { validate_viewport , NV50_NEW_VIEWPORT }, + { validate_sampler , NV50_NEW_SAMPLER }, + { nv50_tex_validate , NV50_NEW_TEXTURE | NV50_NEW_SAMPLER }, + { nv50_vbo_validate , NV50_NEW_ARRAYS }, + { validate_vtxbuf , NV50_NEW_ARRAYS }, + { validate_vtxattr , NV50_NEW_ARRAYS }, + {} +}; +#define validate_list_len (sizeof(validate_list) / sizeof(validate_list[0])) - so_ref(so, &nv50->state.viewport); - so_ref(NULL, &so); - nv50->state.dirty |= NV50_NEW_VIEWPORT; - } -viewport_uptodate: +void +nv50_state_flush_notify(struct nouveau_channel *chan) +{ + struct nv50_context *nv50 = chan->user_private; - if (nv50->dirty & NV50_NEW_SAMPLER) { - unsigned nr = 0; + nv50_tex_relocs(nv50); - for (i = 0; i < PIPE_SHADER_TYPES; ++i) - nr += nv50->sampler_nr[i]; + so_emit_reloc_markers(chan, nv50->state.hw[0]); /* fb */ + so_emit_reloc_markers(chan, nv50->state.hw[3]); /* vp */ + so_emit_reloc_markers(chan, nv50->state.hw[4]); /* fp */ + so_emit_reloc_markers(chan, nv50->state.hw[17]); /* vb */ + so_emit_reloc_markers(chan, nv50->screen->static_init); - so = so_new(1 + 5 * PIPE_SHADER_TYPES, - 1 + 19 * PIPE_SHADER_TYPES + nr * 8, - PIPE_SHADER_TYPES * 2); +#if 0 + if (nv50->state.instbuf) + so_emit_reloc_markers(chan, nv50->state.instbuf); +#endif +} - nv50_validate_samplers(nv50, so, PIPE_SHADER_VERTEX); - nv50_validate_samplers(nv50, so, PIPE_SHADER_FRAGMENT); +boolean +nv50_state_validate(struct nv50_context *nv50) +{ + struct nouveau_channel *chan = nv50->screen->base.channel; + int i; - so_method(so, tesla, 0x1334, 1); /* flush TSC */ - so_data (so, 0); + if (nv50->screen->cur_ctx != nv50) { + for (i = 0; i < validate_list_len; i++) { + if (nv50->state.hw[i]) + nv50->state.hw_dirty |= (1 << i); + } - so_ref(so, &nv50->state.tsc_upload); - so_ref(NULL, &so); + nv50->screen->cur_ctx = nv50; } - if (nv50->dirty & (NV50_NEW_TEXTURE | NV50_NEW_SAMPLER)) - nv50_tex_validate(nv50); + for (i = 0; i < validate_list_len; i++) { + struct state_validate *validate = &validate_list[i]; + struct nouveau_stateobj *so; + + if (!(nv50->dirty & validate->states)) + continue; - if (nv50->dirty & NV50_NEW_ARRAYS) - nv50_vbo_validate(nv50); + so = validate->func(nv50); + if (!so) + continue; - nv50->state.dirty |= nv50->dirty; + so_ref(so, &nv50->state.hw[i]); + so_ref(NULL, &so); + nv50->state.hw_dirty |= (1 << i); + } nv50->dirty = 0; - nv50_state_emit(nv50); + + while (nv50->state.hw_dirty) { + i = ffs(nv50->state.hw_dirty) - 1; + nv50->state.hw_dirty &= ~(1 << i); + + so_emit(chan, nv50->state.hw[i]); + } return TRUE; } -- cgit v1.2.3