X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fgallium%2Fdrivers%2Fr300%2Fr300_state.c;h=b55262a95229396bc83888d3b3d757877cbca786;hb=be1af4394e060677b7db6bbb8e3301e38a3363da;hp=93cc0dbe985269931e4492abbb88173d3f907877;hpb=0b50fcbd556ead8d35c2b543f13de433996a5822;p=mesa.git diff --git a/src/gallium/drivers/r300/r300_state.c b/src/gallium/drivers/r300/r300_state.c index 93cc0dbe985..b55262a9522 100644 --- a/src/gallium/drivers/r300/r300_state.c +++ b/src/gallium/drivers/r300/r300_state.c @@ -23,8 +23,9 @@ #include "draw/draw_context.h" -#include "util/u_blitter.h" +#include "util/u_framebuffer.h" #include "util/u_math.h" +#include "util/u_mm.h" #include "util/u_memory.h" #include "util/u_pack_color.h" @@ -43,6 +44,7 @@ #include "r300_texture.h" #include "r300_vs.h" #include "r300_winsys.h" +#include "r300_hyperz.h" /* r300_state: Functions used to intialize state context by translating * Gallium state objects into semi-native r300 state objects. */ @@ -50,7 +52,7 @@ #define UPDATE_STATE(cso, atom) \ if (cso != atom.state) { \ atom.state = cso; \ - atom.dirty = TRUE; \ + r300_mark_atom_dirty(r300, &(atom)); \ } static boolean blend_discard_if_src_alpha_0(unsigned srcRGB, unsigned srcA, @@ -415,7 +417,7 @@ static void r300_set_blend_color(struct pipe_context* pipe, END_CB; } - r300->blend_color_state.dirty = TRUE; + r300_mark_atom_dirty(r300, &r300->blend_color_state); } static void r300_set_clip_state(struct pipe_context* pipe, @@ -429,19 +431,23 @@ static void r300_set_clip_state(struct pipe_context* pipe, clip->clip = *state; if (r300->screen->caps.has_tcl) { - BEGIN_CB(clip->cb, 29); - OUT_CB_REG(R300_VAP_PVS_VECTOR_INDX_REG, - (r300->screen->caps.is_r500 ? - R500_PVS_UCP_START : R300_PVS_UCP_START)); - OUT_CB_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, 6 * 4); - OUT_CB_TABLE(state->ucp, 6 * 4); + r300->clip_state.size = 2 + !!state->nr * 3 + state->nr * 4; + + BEGIN_CB(clip->cb, r300->clip_state.size); + if (state->nr) { + OUT_CB_REG(R300_VAP_PVS_VECTOR_INDX_REG, + (r300->screen->caps.is_r500 ? + R500_PVS_UCP_START : R300_PVS_UCP_START)); + OUT_CB_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, state->nr * 4); + OUT_CB_TABLE(state->ucp, state->nr * 4); + } OUT_CB_REG(R300_VAP_CLIP_CNTL, ((1 << state->nr) - 1) | - R300_PS_UCP_MODE_CLIP_AS_TRIFAN); + R300_PS_UCP_MODE_CLIP_AS_TRIFAN | + (state->depth_clamp ? R300_CLIP_DISABLE : 0)); END_CB; - r300->clip_state.dirty = TRUE; + r300_mark_atom_dirty(r300, &r300->clip_state); } else { - draw_flush(r300->draw); draw_set_clip_state(r300->draw, state); } } @@ -468,14 +474,14 @@ static void* dsa->dsa = *state; - /* Depth test setup. */ + /* Depth test setup. - separate write mask depth for decomp flush */ + if (state->depth.writemask) { + dsa->z_buffer_control |= R300_Z_WRITE_ENABLE; + } + if (state->depth.enabled) { dsa->z_buffer_control |= R300_Z_ENABLE; - if (state->depth.writemask) { - dsa->z_buffer_control |= R300_Z_WRITE_ENABLE; - } - dsa->z_stencil_control |= (r300_translate_depth_stencil_function(state->depth.func) << R300_Z_FUNC_SHIFT); @@ -588,6 +594,7 @@ static void r300_bind_dsa_state(struct pipe_context* pipe, UPDATE_STATE(state, r300->dsa_state); + r300_mark_atom_dirty(r300, &r300->hyperz_state); /* Will be updated before the emission. */ r300_dsa_inject_stencilref(r300); } @@ -606,35 +613,46 @@ static void r300_set_stencil_ref(struct pipe_context* pipe, r300->stencil_ref = *sr; r300_dsa_inject_stencilref(r300); - r300->dsa_state.dirty = TRUE; + r300_mark_atom_dirty(r300, &r300->dsa_state); +} + +static void r300_tex_set_tiling_flags(struct r300_context *r300, + struct r300_texture *tex, unsigned level) +{ + /* Check if the macrotile flag needs to be changed. + * Skip changing the flags otherwise. */ + if (tex->desc.macrotile[tex->surface_level] != + tex->desc.macrotile[level]) { + /* Tiling determines how DRM treats the buffer data. + * We must flush CS when changing it if the buffer is referenced. */ + if (r300->rws->cs_is_buffer_referenced(r300->cs, + tex->cs_buffer, R300_REF_CS)) + r300->context.flush(&r300->context, 0, NULL); + + r300->rws->buffer_set_tiling(r300->rws, tex->buffer, + tex->desc.microtile, tex->desc.macrotile[level], + tex->desc.stride_in_bytes[0]); + + tex->surface_level = level; + } } /* This switcheroo is needed just because of goddamned MACRO_SWITCH. */ static void r300_fb_set_tiling_flags(struct r300_context *r300, - const struct pipe_framebuffer_state *old_state, - const struct pipe_framebuffer_state *new_state) + const struct pipe_framebuffer_state *state) { - struct r300_texture *tex; - unsigned i, level; + unsigned i; /* Set tiling flags for new surfaces. */ - for (i = 0; i < new_state->nr_cbufs; i++) { - tex = r300_texture(new_state->cbufs[i]->texture); - level = new_state->cbufs[i]->level; - - r300->rws->buffer_set_tiling(r300->rws, tex->buffer, - tex->pitch[0] * util_format_get_blocksize(tex->b.b.format), - tex->microtile, - tex->mip_macrotile[level]); + for (i = 0; i < state->nr_cbufs; i++) { + r300_tex_set_tiling_flags(r300, + r300_texture(state->cbufs[i]->texture), + state->cbufs[i]->u.tex.level); } - if (new_state->zsbuf) { - tex = r300_texture(new_state->zsbuf->texture); - level = new_state->zsbuf->level; - - r300->rws->buffer_set_tiling(r300->rws, tex->buffer, - tex->pitch[0] * util_format_get_blocksize(tex->b.b.format), - tex->microtile, - tex->mip_macrotile[level]); + if (state->zsbuf) { + r300_tex_set_tiling_flags(r300, + r300_texture(state->zsbuf->texture), + state->zsbuf->u.tex.level); } } @@ -645,21 +663,62 @@ static void r300_print_fb_surf_info(struct pipe_surface *surf, unsigned index, struct r300_texture *rtex = r300_texture(tex); fprintf(stderr, - "r300: %s[%i] Dim: %ix%i, Offset: %i, ZSlice: %i, " - "Face: %i, Level: %i, Format: %s\n" + "r300: %s[%i] Dim: %ix%i, Firstlayer: %i, " + "Lastlayer: %i, Level: %i, Format: %s\n" "r300: TEX: Macro: %s, Micro: %s, Pitch: %i, " "Dim: %ix%ix%i, LastLevel: %i, Format: %s\n", - binding, index, surf->width, surf->height, surf->offset, - surf->zslice, surf->face, surf->level, + binding, index, surf->width, surf->height, + surf->u.tex.first_layer, surf->u.tex.last_layer, surf->u.tex.level, util_format_short_name(surf->format), - rtex->macrotile ? "YES" : " NO", rtex->microtile ? "YES" : " NO", - rtex->hwpitch[0], tex->width0, tex->height0, tex->depth0, + rtex->desc.macrotile[0] ? "YES" : " NO", + rtex->desc.microtile ? "YES" : " NO", + rtex->desc.stride_in_pixels[0], + tex->width0, tex->height0, tex->depth0, tex->last_level, util_format_short_name(tex->format)); } +void r300_mark_fb_state_dirty(struct r300_context *r300, + enum r300_fb_state_change change) +{ + struct pipe_framebuffer_state *state = r300->fb_state.state; + boolean can_hyperz = r300->rws->get_value(r300->rws, R300_CAN_HYPERZ); + + r300_mark_atom_dirty(r300, &r300->gpu_flush); + r300_mark_atom_dirty(r300, &r300->fb_state); + + /* What is marked as dirty depends on the enum r300_fb_state_change. */ + if (change == R300_CHANGED_FB_STATE) { + r300_mark_atom_dirty(r300, &r300->aa_state); + } + + if (change == R300_CHANGED_FB_STATE || + change == R300_CHANGED_CBZB_FLAG || + change == R300_CHANGED_ZCLEAR_FLAG) { + r300_mark_atom_dirty(r300, &r300->hyperz_state); + } + + if (change == R300_CHANGED_FB_STATE || + change == R300_CHANGED_MULTIWRITE) { + r300_mark_atom_dirty(r300, &r300->fb_state_pipelined); + } + + /* Now compute the fb_state atom size. */ + r300->fb_state.size = 2 + (8 * state->nr_cbufs); + + if (r300->cbzb_clear) + r300->fb_state.size += 10; + else if (state->zsbuf) { + r300->fb_state.size += 10; + if (can_hyperz) + r300->fb_state.size += r300->screen->caps.hiz_ram ? 8 : 4; + } + + /* The size of the rest of atoms stays the same. */ +} + static void r300_set_framebuffer_state(struct pipe_context* pipe, const struct pipe_framebuffer_state* state) @@ -667,8 +726,10 @@ static void struct r300_context* r300 = r300_context(pipe); struct r300_aa_state *aa = (struct r300_aa_state*)r300->aa_state.state; struct pipe_framebuffer_state *old_state = r300->fb_state.state; + boolean can_hyperz = r300->rws->get_value(r300->rws, R300_CAN_HYPERZ); unsigned max_width, max_height, i; uint32_t zbuffer_bpp = 0; + int blocksize; if (r300->screen->caps.is_r500) { max_width = max_height = 4096; @@ -684,49 +745,67 @@ static void return; } - if (r300->draw) { - draw_flush(r300->draw); - } - - r300->gpu_flush.dirty = TRUE; - r300->aa_state.dirty = TRUE; - r300->fb_state.dirty = TRUE; - r300->hyperz_state.dirty = TRUE; - r300->fb_state_pipelined.dirty = TRUE; - /* If nr_cbufs is changed from zero to non-zero or vice versa... */ if (!!old_state->nr_cbufs != !!state->nr_cbufs) { - r300->blend_state.dirty = TRUE; + r300_mark_atom_dirty(r300, &r300->blend_state); } /* If zsbuf is set from NULL to non-NULL or vice versa.. */ if (!!old_state->zsbuf != !!state->zsbuf) { - r300->dsa_state.dirty = TRUE; + r300_mark_atom_dirty(r300, &r300->dsa_state); } /* The tiling flags are dependent on the surface miplevel, unfortunately. */ - r300_fb_set_tiling_flags(r300, r300->fb_state.state, state); + r300_fb_set_tiling_flags(r300, state); - util_assign_framebuffer_state(r300->fb_state.state, state); + util_copy_framebuffer_state(r300->fb_state.state, state); - r300->fb_state.size = - 2 + - (8 * state->nr_cbufs) + - (state->zsbuf ? (r300->screen->caps.has_hiz ? 18 : 14) : 0); + r300_mark_fb_state_dirty(r300, R300_CHANGED_FB_STATE); + r300->validate_buffers = TRUE; - /* Polygon offset depends on the zbuffer bit depth. */ - if (state->zsbuf && r300->polygon_offset_enabled) { - switch (util_format_get_blocksize(state->zsbuf->texture->format)) { - case 2: - zbuffer_bpp = 16; - break; - case 4: - zbuffer_bpp = 24; - break; + r300->z_compression = false; + + if (state->zsbuf) { + blocksize = util_format_get_blocksize(state->zsbuf->texture->format); + switch (blocksize) { + case 2: + zbuffer_bpp = 16; + break; + case 4: + zbuffer_bpp = 24; + break; + } + if (can_hyperz) { + struct r300_surface *zs_surf = r300_surface(state->zsbuf); + struct r300_texture *tex; + int compress = r300->screen->caps.is_rv350 ? RV350_Z_COMPRESS_88 : R300_Z_COMPRESS_44; + int level = zs_surf->base.u.tex.level; + + tex = r300_texture(zs_surf->base.texture); + + /* work out whether we can support hiz on this buffer */ + r300_hiz_alloc_block(r300, zs_surf); + + /* work out whether we can support zmask features on this buffer */ + r300_zmask_alloc_block(r300, zs_surf, compress); + + if (tex->zmask_mem[level]) { + /* compression causes hangs on 16-bit */ + if (zbuffer_bpp == 24) + r300->z_compression = compress; + } + DBG(r300, DBG_HYPERZ, + "hyper-z features: hiz: %d @ %08x z-compression: %d z-fastfill: %d @ %08x\n", tex->hiz_mem[level] ? 1 : 0, + tex->hiz_mem[level] ? tex->hiz_mem[level]->ofs : 0xdeadbeef, + r300->z_compression, tex->zmask_mem[level] ? 1 : 0, + tex->zmask_mem[level] ? tex->zmask_mem[level]->ofs : 0xdeadbeef); } + /* Polygon offset depends on the zbuffer bit depth. */ if (r300->zbuffer_bpp != zbuffer_bpp) { r300->zbuffer_bpp = zbuffer_bpp; - r300->rs_state.dirty = TRUE; + + if (r300->polygon_offset_enabled) + r300_mark_atom_dirty(r300, &r300->rs_state); } } @@ -784,9 +863,9 @@ void r300_mark_fs_code_dirty(struct r300_context *r300) { struct r300_fragment_shader* fs = r300_fs(r300); - r300->fs.dirty = TRUE; - r300->fs_rc_constant_state.dirty = TRUE; - r300->fs_constants.dirty = TRUE; + r300_mark_atom_dirty(r300, &r300->fs); + r300_mark_atom_dirty(r300, &r300->fs_rc_constant_state); + r300_mark_atom_dirty(r300, &r300->fs_constants); r300->fs.size = fs->shader->cb_code_size; if (r300->screen->caps.is_r500) { @@ -796,6 +875,9 @@ void r300_mark_fs_code_dirty(struct r300_context *r300) r300->fs_rc_constant_state.size = fs->shader->rc_state_count * 5; r300->fs_constants.size = fs->shader->externals_count * 4 + 1; } + + ((struct r300_constant_buffer*)r300->fs_constants.state)->remap_table = + fs->shader->code.constants_remap_table; } /* Bind fragment shader state. */ @@ -803,17 +885,26 @@ static void r300_bind_fs_state(struct pipe_context* pipe, void* shader) { struct r300_context* r300 = r300_context(pipe); struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader; + struct pipe_framebuffer_state *fb = r300->fb_state.state; + boolean last_multi_write; if (fs == NULL) { r300->fs.state = NULL; return; } + last_multi_write = r300_fragment_shader_writes_all(r300_fs(r300)); + r300->fs.state = fs; r300_pick_fragment_shader(r300); r300_mark_fs_code_dirty(r300); - r300->rs_block_state.dirty = TRUE; /* Will be updated before the emission. */ + if (fb->nr_cbufs > 1 && + last_multi_write != r300_fragment_shader_writes_all(fs)) { + r300_mark_fb_state_dirty(r300, R300_CHANGED_MULTIWRITE); + } + + r300_mark_atom_dirty(r300, &r300->rs_block_state); /* Will be updated before the emission. */ } /* Delete fragment shader state. */ @@ -850,7 +941,6 @@ static void* r300_create_rs_state(struct pipe_context* pipe, const struct pipe_rasterizer_state* state) { struct r300_rs_state* rs = CALLOC_STRUCT(r300_rs_state); - int i; float psiz; uint32_t vap_control_status; /* R300_VAP_CNTL_STATUS: 0x2140 */ uint32_t point_size; /* R300_GA_POINT_SIZE: 0x421c */ @@ -863,21 +953,20 @@ static void* r300_create_rs_state(struct pipe_context* pipe, uint32_t polygon_mode; /* R300_GA_POLY_MODE: 0x4288 */ uint32_t clip_rule; /* R300_SC_CLIP_RULE: 0x43D0 */ - /* Specifies top of Raster pipe specific enable controls, - * i.e. texture coordinates stuffing for points, lines, triangles */ - uint32_t stuffing_enable; /* R300_GB_ENABLE: 0x4008 */ - /* Point sprites texture coordinates, 0: lower left, 1: upper right */ - float point_texcoord_left; /* R300_GA_POINT_S0: 0x4200 */ - float point_texcoord_bottom; /* R300_GA_POINT_T0: 0x4204 */ - float point_texcoord_right; /* R300_GA_POINT_S1: 0x4208 */ - float point_texcoord_top; /* R300_GA_POINT_T1: 0x420c */ + float point_texcoord_left = 0; /* R300_GA_POINT_S0: 0x4200 */ + float point_texcoord_bottom = 0;/* R300_GA_POINT_T0: 0x4204 */ + float point_texcoord_right = 1; /* R300_GA_POINT_S1: 0x4208 */ + float point_texcoord_top = 0; /* R300_GA_POINT_T1: 0x420c */ CB_LOCALS; /* Copy rasterizer state. */ rs->rs = *state; rs->rs_draw = *state; + rs->rs.sprite_coord_enable = state->point_quad_rasterization * + state->sprite_coord_enable; + /* Override some states for Draw. */ rs->rs_draw.sprite_coord_enable = 0; /* We can do this in HW. */ @@ -964,6 +1053,9 @@ static void* r300_create_rs_state(struct pipe_context* pipe, R300_GA_LINE_STIPPLE_CONFIG_STIPPLE_SCALE_MASK); /* XXX this might need to be scaled up */ line_stipple_value = state->line_stipple_pattern; + } else { + line_stipple_config = 0; + line_stipple_value = 0; } if (state->flatshade) { @@ -974,19 +1066,8 @@ static void* r300_create_rs_state(struct pipe_context* pipe, clip_rule = state->scissor ? 0xAAAA : 0xFFFF; - /* Point sprites */ - stuffing_enable = 0; - if (state->sprite_coord_enable) { - stuffing_enable = R300_GB_POINT_STUFF_ENABLE; - for (i = 0; i < 8; i++) { - if (state->sprite_coord_enable & (1 << i)) - stuffing_enable |= - R300_GB_TEX_STR << (R300_GB_TEX0_SOURCE_SHIFT + (i*2)); - } - - point_texcoord_left = 0.0f; - point_texcoord_right = 1.0f; - + /* Point sprites coord mode */ + if (rs->rs.sprite_coord_enable) { switch (state->sprite_coord_mode) { case PIPE_SPRITE_COORD_UPPER_LEFT: point_texcoord_top = 0.0f; @@ -1000,7 +1081,7 @@ static void* r300_create_rs_state(struct pipe_context* pipe, } /* Build the main command buffer. */ - BEGIN_CB(rs->cb_main, 25); + BEGIN_CB(rs->cb_main, RS_STATE_MAIN_SIZE); OUT_CB_REG(R300_VAP_CNTL_STATUS, vap_control_status); OUT_CB_REG(R300_GA_POINT_SIZE, point_size); OUT_CB_REG_SEQ(R300_GA_POINT_MINMAX, 2); @@ -1014,7 +1095,6 @@ static void* r300_create_rs_state(struct pipe_context* pipe, OUT_CB_REG(R300_GA_LINE_STIPPLE_VALUE, line_stipple_value); OUT_CB_REG(R300_GA_POLY_MODE, polygon_mode); OUT_CB_REG(R300_SC_CLIP_RULE, clip_rule); - OUT_CB_REG(R300_GB_ENABLE, stuffing_enable); OUT_CB_REG_SEQ(R300_GA_POINT_S0, 4); OUT_CB_32F(point_texcoord_left); OUT_CB_32F(point_texcoord_bottom); @@ -1058,14 +1138,11 @@ static void r300_bind_rs_state(struct pipe_context* pipe, void* state) boolean last_two_sided_color = r300->two_sided_color; if (r300->draw && rs) { - draw_flush(r300->draw); draw_set_rasterizer_state(r300->draw, &rs->rs_draw, state); } if (rs) { - r300->polygon_offset_enabled = (rs->rs.offset_point || - rs->rs.offset_line || - rs->rs.offset_tri); + r300->polygon_offset_enabled = rs->polygon_offset_enable; r300->sprite_coord_enable = rs->rs.sprite_coord_enable; r300->two_sided_color = rs->rs.light_twoside; } else { @@ -1075,11 +1152,11 @@ static void r300_bind_rs_state(struct pipe_context* pipe, void* state) } UPDATE_STATE(state, r300->rs_state); - r300->rs_state.size = 25 + (r300->polygon_offset_enabled ? 5 : 0); + r300->rs_state.size = RS_STATE_MAIN_SIZE + (r300->polygon_offset_enabled ? 5 : 0); if (last_sprite_coord_enable != r300->sprite_coord_enable || last_two_sided_color != r300->two_sided_color) { - r300->rs_block_state.dirty = TRUE; + r300_mark_atom_dirty(r300, &r300->rs_block_state); } } @@ -1097,7 +1174,6 @@ static void* struct r300_sampler_state* sampler = CALLOC_STRUCT(r300_sampler_state); boolean is_r500 = r300->screen->caps.is_r500; int lod_bias; - union util_color uc; sampler->state = *state; @@ -1139,12 +1215,12 @@ static void* /* Unfortunately, r300-r500 don't support floating-point mipmap lods. */ /* We must pass these to the merge function to clamp them properly. */ - sampler->min_lod = MAX2((unsigned)state->min_lod, 0); - sampler->max_lod = MAX2((unsigned)ceilf(state->max_lod), 0); + sampler->min_lod = (unsigned)MAX2(state->min_lod, 0); + sampler->max_lod = (unsigned)MAX2(ceilf(state->max_lod), 0); lod_bias = CLAMP((int)(state->lod_bias * 32 + 1), -(1 << 9), (1 << 9) - 1); - sampler->filter1 |= lod_bias << R300_LOD_BIAS_SHIFT; + sampler->filter1 |= (lod_bias << R300_LOD_BIAS_SHIFT) & R300_LOD_BIAS_MASK; /* This is very high quality anisotropic filtering for R5xx. * It's good for benchmarking the performance of texturing but @@ -1154,9 +1230,6 @@ static void* sampler->filter1 |= r500_anisotropy(state->max_anisotropy); } - util_pack_color(state->border_color, PIPE_FORMAT_B8G8R8A8_UNORM, &uc); - sampler->border_color = uc.ui; - /* R500-specific fixups and optimizations */ if (r300->screen->caps.is_r500) { sampler->filter1 |= R500_BORDER_FIX; @@ -1181,7 +1254,7 @@ static void r300_bind_sampler_states(struct pipe_context* pipe, memcpy(state->sampler_states, states, sizeof(void*) * count); state->sampler_state_count = count; - r300->textures_state.dirty = TRUE; + r300_mark_atom_dirty(r300, &r300->textures_state); } static void r300_lacks_vertex_textures(struct pipe_context* pipe, @@ -1243,29 +1316,27 @@ static void r300_set_fragment_sampler_views(struct pipe_context* pipe, } for (i = 0; i < count; i++) { - if (&state->sampler_views[i]->base != views[i]) { - pipe_sampler_view_reference( - (struct pipe_sampler_view**)&state->sampler_views[i], - views[i]); + pipe_sampler_view_reference( + (struct pipe_sampler_view**)&state->sampler_views[i], + views[i]); - if (!views[i]) { - continue; - } + if (!views[i]) { + continue; + } - /* A new sampler view (= texture)... */ - dirty_tex = TRUE; + /* A new sampler view (= texture)... */ + dirty_tex = TRUE; - /* Set the texrect factor in the fragment shader. + /* Set the texrect factor in the fragment shader. * Needed for RECT and NPOT fallback. */ - texture = r300_texture(views[i]->texture); - if (texture->uses_pitch) { - r300->fs_rc_constant_state.dirty = TRUE; - } + texture = r300_texture(views[i]->texture); + if (texture->desc.is_npot) { + r300_mark_atom_dirty(r300, &r300->fs_rc_constant_state); + } - state->sampler_views[i]->texcache_region = + state->sampler_views[i]->texcache_region = r300_assign_texture_cache_region(view_index, real_num_views); - view_index++; - } + view_index++; } for (i = count; i < tex_units; i++) { @@ -1278,10 +1349,11 @@ static void r300_set_fragment_sampler_views(struct pipe_context* pipe, state->sampler_view_count = count; - r300->textures_state.dirty = TRUE; + r300_mark_atom_dirty(r300, &r300->textures_state); + r300->validate_buffers = TRUE; if (dirty_tex) { - r300->texture_cache_inval.dirty = TRUE; + r300_mark_atom_dirty(r300, &r300->texture_cache_inval); } } @@ -1292,6 +1364,8 @@ r300_create_sampler_view(struct pipe_context *pipe, { struct r300_sampler_view *view = CALLOC_STRUCT(r300_sampler_view); struct r300_texture *tex = r300_texture(texture); + boolean is_r500 = r300_screen(pipe->screen)->caps.is_r500; + boolean dxtc_swizzle = r300_screen(pipe->screen)->caps.dxtc_swizzle; if (view) { view->base = *templ; @@ -1307,8 +1381,10 @@ r300_create_sampler_view(struct pipe_context *pipe, view->format = tex->tx_format; view->format.format1 |= r300_translate_texformat(templ->format, - view->swizzle); - if (r300_screen(pipe->screen)->caps.is_r500) { + view->swizzle, + is_r500, + dxtc_swizzle); + if (is_r500) { view->format.format2 |= r500_tx_format_msb_bit(templ->format); } } @@ -1332,7 +1408,7 @@ static void r300_set_scissor_state(struct pipe_context* pipe, memcpy(r300->scissor_state.state, state, sizeof(struct pipe_scissor_state)); - r300->scissor_state.dirty = TRUE; + r300_mark_atom_dirty(r300, &r300->scissor_state); } static void r300_set_viewport_state(struct pipe_context* pipe, @@ -1345,7 +1421,6 @@ static void r300_set_viewport_state(struct pipe_context* pipe, r300->viewport = *state; if (r300->draw) { - draw_flush(r300->draw); draw_set_viewport_state(r300->draw, state); viewport->vte_control = R300_VTX_XY_FMT | R300_VTX_Z_FMT; return; @@ -1379,9 +1454,9 @@ static void r300_set_viewport_state(struct pipe_context* pipe, viewport->vte_control |= R300_VPORT_Z_OFFSET_ENA; } - r300->viewport_state.dirty = TRUE; + r300_mark_atom_dirty(r300, &r300->viewport_state); if (r300->fs.state && r300_fs(r300)->shader->inputs.wpos != ATTR_UNUSED) { - r300->fs_rc_constant_state.dirty = TRUE; + r300_mark_atom_dirty(r300, &r300->fs_rc_constant_state); } } @@ -1390,9 +1465,18 @@ static void r300_set_vertex_buffers(struct pipe_context* pipe, const struct pipe_vertex_buffer* buffers) { struct r300_context* r300 = r300_context(pipe); - struct pipe_vertex_buffer *vbo; + const struct pipe_vertex_buffer *vbo; unsigned i, max_index = (1 << 24) - 1; boolean any_user_buffer = FALSE; + boolean any_nonuser_buffer = FALSE; + struct pipe_vertex_buffer dummy_vb = {0}; + + /* There must be at least one vertex buffer set, otherwise it locks up. */ + if (!count) { + dummy_vb.buffer = r300->dummy_vb; + buffers = &dummy_vb; + count = 1; + } if (count == r300->vertex_buffer_count && memcmp(r300->vertex_buffer, buffers, @@ -1416,55 +1500,95 @@ static void r300_set_vertex_buffers(struct pipe_context* pipe, } for (i = 0; i < count; i++) { - /* Why, yes, I AM casting away constness. How did you know? */ - vbo = (struct pipe_vertex_buffer*)&buffers[i]; + vbo = &buffers[i]; /* Skip NULL buffers */ - if (!buffers[i].buffer) { + if (!vbo->buffer) { continue; } - if (r300_buffer_is_user_buffer(vbo->buffer)) { + /* User buffers have no info about maximum index, + * we will have to compute it in draw_vbo. */ + if (r300_is_user_buffer(vbo->buffer)) { any_user_buffer = TRUE; + continue; } + any_nonuser_buffer = TRUE; - if (vbo->max_index == ~0) { - /* if no VBO stride then only one vertex value so max index is 1 */ - /* should think about converting to VS constants like svga does */ - if (!vbo->stride) - vbo->max_index = 1; - else - vbo->max_index = - (vbo->buffer->width0 - vbo->buffer_offset) / vbo->stride; - } + /* The stride of zero means we will be fetching only the first + * vertex, so don't care about max_index. */ + if (!vbo->stride) + continue; - max_index = MIN2(vbo->max_index, max_index); + /* Update the maximum index. */ + { + unsigned vbo_max_index = + (vbo->buffer->width0 - vbo->buffer_offset) / vbo->stride; + max_index = MIN2(max_index, vbo_max_index); + } } r300->any_user_vbs = any_user_buffer; r300->vertex_buffer_max_index = max_index; - + r300->vertex_arrays_dirty = TRUE; + if (any_nonuser_buffer) + r300->validate_buffers = TRUE; + if (!any_user_buffer) + r300->upload_vb_validated = FALSE; } else { /* SW TCL. */ - draw_flush(r300->draw); draw_set_vertex_buffers(r300->draw, count, buffers); } /* Common code. */ for (i = 0; i < count; i++) { + vbo = &buffers[i]; + /* Reference our buffer. */ - pipe_resource_reference(&r300->vertex_buffer[i].buffer, buffers[i].buffer); + pipe_resource_reference(&r300->vertex_buffer[i].buffer, vbo->buffer); + if (vbo->buffer && r300_is_user_buffer(vbo->buffer)) { + pipe_resource_reference(&r300->valid_vertex_buffer[i], NULL); + } else { + pipe_resource_reference(&r300->valid_vertex_buffer[i], vbo->buffer); + } } for (; i < r300->vertex_buffer_count; i++) { /* Dereference any old buffers. */ pipe_resource_reference(&r300->vertex_buffer[i].buffer, NULL); + pipe_resource_reference(&r300->valid_vertex_buffer[i], NULL); } memcpy(r300->vertex_buffer, buffers, - sizeof(struct pipe_vertex_buffer) * count); + sizeof(struct pipe_vertex_buffer) * count); + r300->vertex_buffer_count = count; } +static void r300_set_index_buffer(struct pipe_context* pipe, + const struct pipe_index_buffer *ib) +{ + struct r300_context* r300 = r300_context(pipe); + + if (ib && ib->buffer) { + pipe_resource_reference(&r300->index_buffer.buffer, ib->buffer); + memcpy(&r300->index_buffer, ib, sizeof(r300->index_buffer)); + + if (r300->screen->caps.has_tcl && + !r300_is_user_buffer(ib->buffer)) { + r300->validate_buffers = TRUE; + r300->upload_ib_validated = FALSE; + } + } + else { + pipe_resource_reference(&r300->index_buffer.buffer, NULL); + memset(&r300->index_buffer, 0, sizeof(r300->index_buffer)); + } + + if (!r300->screen->caps.has_tcl) { + draw_set_index_buffer(r300->draw, ib); + } +} + /* Initialize the PSC tables. */ static void r300_vertex_psc(struct r300_vertex_element_state *velems) { @@ -1525,6 +1649,14 @@ static void* r300_create_vertex_elements_state(struct pipe_context* pipe, struct r300_vertex_element_state *velems; unsigned i; enum pipe_format *format; + struct pipe_vertex_element dummy_attrib = {0}; + + /* R300 Programmable Stream Control (PSC) doesn't support 0 vertex elements. */ + if (!count) { + dummy_attrib.src_format = PIPE_FORMAT_R8G8B8A8_UNORM; + attribs = &dummy_attrib; + count = 1; + } assert(count <= PIPE_MAX_ATTRIBS); velems = CALLOC_STRUCT(r300_vertex_element_state); @@ -1591,7 +1723,8 @@ static void* r300_create_vertex_elements_state(struct pipe_context* pipe, * swizzles are already set up. * Also compute the vertex size. */ for (i = 0; i < count; i++) { - /* This is OK because we check for aligned strides too. */ + /* This is OK because we check for aligned strides too + * elsewhere. */ velems->hw_format_size[i] = align(util_format_get_blocksize(velems->hw_format[i]), 4); velems->vertex_size_dwords += velems->hw_format_size[i] / 4; @@ -1614,13 +1747,13 @@ static void r300_bind_vertex_elements_state(struct pipe_context *pipe, r300->velems = velems; if (r300->draw) { - draw_flush(r300->draw); draw_set_vertex_elements(r300->draw, velems->count, velems->velem); return; } UPDATE_STATE(&velems->vertex_stream, r300->vertex_stream_state); r300->vertex_stream_state.size = (1 + velems->vertex_stream.count) * 2; + r300->vertex_arrays_dirty = TRUE; } static void r300_delete_vertex_elements_state(struct pipe_context *pipe, void *state) @@ -1663,24 +1796,26 @@ static void r300_bind_vs_state(struct pipe_context* pipe, void* shader) r300->vs_state.state = vs; /* The majority of the RS block bits is dependent on the vertex shader. */ - r300->rs_block_state.dirty = TRUE; /* Will be updated before the emission. */ + r300_mark_atom_dirty(r300, &r300->rs_block_state); /* Will be updated before the emission. */ if (r300->screen->caps.has_tcl) { - r300->vs_state.dirty = TRUE; + unsigned fc_op_dwords = r300->screen->caps.is_r500 ? 3 : 2; + r300_mark_atom_dirty(r300, &r300->vs_state); r300->vs_state.size = vs->code.length + 9 + + (vs->code.num_fc_ops ? vs->code.num_fc_ops * fc_op_dwords + 4 : 0); + + r300_mark_atom_dirty(r300, &r300->vs_constants); + r300->vs_constants.size = + 2 + + (vs->externals_count ? vs->externals_count * 4 + 3 : 0) + (vs->immediates_count ? vs->immediates_count * 4 + 3 : 0); - if (vs->externals_count) { - r300->vs_constants.dirty = TRUE; - r300->vs_constants.size = vs->externals_count * 4 + 3; - } else { - r300->vs_constants.size = 0; - } + ((struct r300_constant_buffer*)r300->vs_constants.state)->remap_table = + vs->code.constants_remap_table; - r300->pvs_flush.dirty = TRUE; + r300_mark_atom_dirty(r300, &r300->pvs_flush); } else { - draw_flush(r300->draw); draw_bind_vertex_shader(r300->draw, (struct draw_vertex_shader*)vs->draw_vs); } @@ -1693,6 +1828,8 @@ static void r300_delete_vs_state(struct pipe_context* pipe, void* shader) if (r300->screen->caps.has_tcl) { rc_constants_destroy(&vs->code.constants); + if (vs->code.constants_remap_table) + FREE(vs->code.constants_remap_table); } else { draw_delete_vertex_shader(r300->draw, (struct draw_vertex_shader*)vs->draw_vs); @@ -1708,77 +1845,61 @@ static void r300_set_constant_buffer(struct pipe_context *pipe, { struct r300_context* r300 = r300_context(pipe); struct r300_constant_buffer *cbuf; - struct pipe_transfer *tr; - float *mapped; - int max_size = 0, max_size_bytes = 0, clamped_size = 0; + struct r300_buffer *rbuf = r300_buffer(buf); + uint32_t *mapped; switch (shader) { case PIPE_SHADER_VERTEX: cbuf = (struct r300_constant_buffer*)r300->vs_constants.state; - max_size = 256; break; case PIPE_SHADER_FRAGMENT: cbuf = (struct r300_constant_buffer*)r300->fs_constants.state; - if (r300->screen->caps.is_r500) { - max_size = 256; - } else { - max_size = 32; - } break; default: - assert(0); return; } - max_size_bytes = max_size * 4 * sizeof(float); - if (buf == NULL || buf->width0 == 0 || - (mapped = pipe_buffer_map(pipe, buf, PIPE_TRANSFER_READ, &tr)) == NULL) - { - cbuf->count = 0; + if (buf == NULL || buf->width0 == 0) + return; + + if (rbuf->user_buffer) + mapped = (uint32_t*)rbuf->user_buffer; + else if (rbuf->constant_buffer) + mapped = (uint32_t*)rbuf->constant_buffer; + else return; - } if (shader == PIPE_SHADER_FRAGMENT || (shader == PIPE_SHADER_VERTEX && r300->screen->caps.has_tcl)) { assert((buf->width0 % (4 * sizeof(float))) == 0); - - /* Check the size of the constant buffer. */ - /* XXX Subtract immediates and RC_STATE_* variables. */ - if (buf->width0 > max_size_bytes) { - fprintf(stderr, "r300: Max size of the constant buffer is " - "%i*4 floats.\n", max_size); - } - - clamped_size = MIN2(buf->width0, max_size_bytes); - cbuf->count = clamped_size / (4 * sizeof(float)); - - if (shader == PIPE_SHADER_FRAGMENT && !r300->screen->caps.is_r500) { - unsigned i,j; - - /* Convert constants to float24. */ - for (i = 0; i < cbuf->count; i++) - for (j = 0; j < 4; j++) - cbuf->constants[i][j] = pack_float24(mapped[i*4+j]); - } else { - memcpy(cbuf->constants, mapped, clamped_size); - } + cbuf->ptr = mapped; } if (shader == PIPE_SHADER_VERTEX) { if (r300->screen->caps.has_tcl) { - if (r300->vs_constants.size) { - r300->vs_constants.dirty = TRUE; + struct r300_vertex_shader *vs = + (struct r300_vertex_shader*)r300->vs_state.state; + + if (!vs) { + cbuf->buffer_base = 0; + return; } - r300->pvs_flush.dirty = TRUE; + + cbuf->buffer_base = r300->vs_const_base; + r300->vs_const_base += vs->code.constants.Count; + if (r300->vs_const_base > R500_MAX_PVS_CONST_VECS) { + r300->vs_const_base = vs->code.constants.Count; + cbuf->buffer_base = 0; + r300_mark_atom_dirty(r300, &r300->pvs_flush); + } + r300_mark_atom_dirty(r300, &r300->vs_constants); } else if (r300->draw) { draw_set_mapped_constant_buffer(r300->draw, PIPE_SHADER_VERTEX, 0, mapped, buf->width0); } } else if (shader == PIPE_SHADER_FRAGMENT) { - r300->fs_constants.dirty = TRUE; + r300_mark_atom_dirty(r300, &r300->fs_constants); } - - pipe_buffer_unmap(pipe, buf, tr); } void r300_init_state_functions(struct r300_context* r300) @@ -1826,6 +1947,7 @@ void r300_init_state_functions(struct r300_context* r300) r300->context.set_viewport_state = r300_set_viewport_state; r300->context.set_vertex_buffers = r300_set_vertex_buffers; + r300->context.set_index_buffer = r300_set_index_buffer; r300->context.create_vertex_elements_state = r300_create_vertex_elements_state; r300->context.bind_vertex_elements_state = r300_bind_vertex_elements_state;