X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fgallium%2Fdrivers%2Fr300%2Fr300_state.c;h=71ac89715cc47ff6e2b091208bb4b0231d12b2b3;hb=42910ebe7b9748c0ecb6a597bae3e7d43c7e170f;hp=281ff6844950a77d72e16f661231eacc8bd19d2e;hpb=5184b02b0793136a15b7edf309d79a1efe1bc73d;p=mesa.git diff --git a/src/gallium/drivers/r300/r300_state.c b/src/gallium/drivers/r300/r300_state.c index 281ff684495..71ac89715cc 100644 --- a/src/gallium/drivers/r300/r300_state.c +++ b/src/gallium/drivers/r300/r300_state.c @@ -30,18 +30,26 @@ #include "tgsi/tgsi_parse.h" #include "pipe/p_config.h" -#include "pipe/internal/p_winsys_screen.h" #include "r300_context.h" #include "r300_reg.h" #include "r300_screen.h" +#include "r300_screen_buffer.h" #include "r300_state_inlines.h" #include "r300_fs.h" #include "r300_vs.h" +#include "radeon_winsys.h" + /* r300_state: Functions used to intialize state context by translating * Gallium state objects into semi-native r300 state objects. */ +#define UPDATE_STATE(cso, atom) \ + if (cso != atom.state) { \ + atom.state = cso; \ + atom.dirty = TRUE; \ + } + static boolean blend_discard_if_src_alpha_0(unsigned srcRGB, unsigned srcA, unsigned dstRGB, unsigned dstA) { @@ -156,23 +164,33 @@ static boolean blend_discard_if_src_alpha_color_1(unsigned srcRGB, unsigned srcA dstA == PIPE_BLENDFACTOR_ONE); } +static unsigned bgra_cmask(unsigned mask) +{ + /* Gallium uses RGBA color ordering while R300 expects BGRA. */ + + return ((mask & PIPE_MASK_R) << 2) | + ((mask & PIPE_MASK_B) >> 2) | + (mask & (PIPE_MASK_G | PIPE_MASK_A)); +} + /* Create a new blend state based on the CSO blend state. * * This encompasses alpha blending, logic/raster ops, and blend dithering. */ static void* r300_create_blend_state(struct pipe_context* pipe, const struct pipe_blend_state* state) { + struct r300_screen* r300screen = r300_screen(pipe->screen); struct r300_blend_state* blend = CALLOC_STRUCT(r300_blend_state); - if (state->blend_enable) + if (state->rt[0].blend_enable) { - unsigned eqRGB = state->rgb_func; - unsigned srcRGB = state->rgb_src_factor; - unsigned dstRGB = state->rgb_dst_factor; + unsigned eqRGB = state->rt[0].rgb_func; + unsigned srcRGB = state->rt[0].rgb_src_factor; + unsigned dstRGB = state->rt[0].rgb_dst_factor; - unsigned eqA = state->alpha_func; - unsigned srcA = state->alpha_src_factor; - unsigned dstA = state->alpha_dst_factor; + unsigned eqA = state->rt[0].alpha_func; + unsigned srcA = state->rt[0].alpha_src_factor; + unsigned dstA = state->rt[0].alpha_dst_factor; /* despite the name, ALPHA_BLEND_ENABLE has nothing to do with alpha, * this is just the crappy D3D naming */ @@ -289,18 +307,18 @@ static void* r300_create_blend_state(struct pipe_context* pipe, (state->logicop_func) << R300_RB3D_ROPCNTL_ROP_SHIFT; } - /* Color Channel Mask */ - if (state->colormask & PIPE_MASK_R) { - blend->color_channel_mask |= RB3D_COLOR_CHANNEL_MASK_RED_MASK0; - } - if (state->colormask & PIPE_MASK_G) { - blend->color_channel_mask |= RB3D_COLOR_CHANNEL_MASK_GREEN_MASK0; - } - if (state->colormask & PIPE_MASK_B) { - blend->color_channel_mask |= RB3D_COLOR_CHANNEL_MASK_BLUE_MASK0; - } - if (state->colormask & PIPE_MASK_A) { - blend->color_channel_mask |= RB3D_COLOR_CHANNEL_MASK_ALPHA_MASK0; + /* Color channel masks for all MRTs. */ + blend->color_channel_mask = bgra_cmask(state->rt[0].colormask); + if (r300screen->caps->is_r500 && state->independent_blend_enable) { + if (state->rt[1].blend_enable) { + blend->color_channel_mask |= bgra_cmask(state->rt[1].colormask) << 4; + } + if (state->rt[2].blend_enable) { + blend->color_channel_mask |= bgra_cmask(state->rt[2].colormask) << 8; + } + if (state->rt[3].blend_enable) { + blend->color_channel_mask |= bgra_cmask(state->rt[3].colormask) << 12; + } } if (state->dither) { @@ -317,8 +335,7 @@ static void r300_bind_blend_state(struct pipe_context* pipe, { struct r300_context* r300 = r300_context(pipe); - r300->blend_state.state = state; - r300->blend_state.dirty = TRUE; + UPDATE_STATE(state, r300->blend_state); } /* Free blend state. */ @@ -345,7 +362,7 @@ static void r300_set_blend_color(struct pipe_context* pipe, (struct r300_blend_color_state*)r300->blend_color_state.state; union util_color uc; - util_pack_color(color->color, PIPE_FORMAT_A8R8G8B8_UNORM, &uc); + util_pack_color(color->color, PIPE_FORMAT_B8G8R8A8_UNORM, &uc); state->blend_color = uc.ui; /* XXX if FP16 blending is enabled, we should use the FP16 format */ @@ -365,6 +382,8 @@ static void r300_set_clip_state(struct pipe_context* pipe, { struct r300_context* r300 = r300_context(pipe); + r300->clip = *state; + if (r300_screen(pipe->screen)->caps->has_tcl) { memcpy(r300->clip_state.state, state, sizeof(struct pipe_clip_state)); r300->clip_state.size = 29; @@ -416,7 +435,7 @@ static void* (r300_translate_stencil_op(state->stencil[0].zfail_op) << R300_S_FRONT_ZFAIL_OP_SHIFT); - dsa->stencil_ref_mask = (state->stencil[0].ref_value) | + dsa->stencil_ref_mask = (state->stencil[0].valuemask << R300_STENCILMASK_SHIFT) | (state->stencil[0].writemask << R300_STENCILWRITEMASK_SHIFT); @@ -432,11 +451,10 @@ static void* (r300_translate_stencil_op(state->stencil[1].zfail_op) << R300_S_BACK_ZFAIL_OP_SHIFT); - /* XXX it seems r3xx doesn't support STENCILREFMASK_BF */ if (caps->is_r500) { dsa->z_buffer_control |= R500_STENCIL_REFMASK_FRONT_BACK; - dsa->stencil_ref_bf = (state->stencil[1].ref_value) | + dsa->stencil_ref_bf = (state->stencil[1].valuemask << R300_STENCILMASK_SHIFT) | (state->stencil[1].writemask << @@ -451,8 +469,7 @@ static void* r300_translate_alpha_function(state->alpha.func) | R300_FG_ALPHA_FUNC_ENABLE; - /* XXX figure out why emitting 10bit alpha ref causes CS to dump */ - /* always use 8bit alpha ref */ + /* We could use 10bit alpha ref but who needs that? */ dsa->alpha_function |= float_to_ubyte(state->alpha.ref_value); if (caps->is_r500) @@ -467,11 +484,8 @@ static void r300_bind_dsa_state(struct pipe_context* pipe, void* state) { struct r300_context* r300 = r300_context(pipe); - struct r300_screen* r300screen = r300_screen(pipe->screen); - r300->dsa_state.state = state; - r300->dsa_state.size = r300screen->caps->is_r500 ? 8 : 6; - r300->dsa_state.dirty = TRUE; + UPDATE_STATE(state, r300->dsa_state); } /* Free DSA state. */ @@ -481,24 +495,149 @@ static void r300_delete_dsa_state(struct pipe_context* pipe, FREE(state); } +static void r300_set_stencil_ref(struct pipe_context* pipe, + const struct pipe_stencil_ref* sr) +{ + struct r300_context* r300 = r300_context(pipe); + r300->stencil_ref = *sr; + r300->dsa_state.dirty = TRUE; +} + +/* This switcheroo is needed just because of goddamned MACRO_SWITCH. */ +static void r300_fb_update_tiling_flags(struct r300_context *r300, + const struct pipe_framebuffer_state *old_state, + const struct pipe_framebuffer_state *new_state) +{ + struct r300_texture *tex; + unsigned i, j, level; + + /* Reset tiling flags for old surfaces to default values. */ + for (i = 0; i < old_state->nr_cbufs; i++) { + for (j = 0; j < new_state->nr_cbufs; j++) { + if (old_state->cbufs[i]->texture == new_state->cbufs[j]->texture) { + break; + } + } + /* If not binding the surface again... */ + if (j != new_state->nr_cbufs) { + continue; + } + + tex = (struct r300_texture*)old_state->cbufs[i]->texture; + + if (tex) { + r300->rws->buffer_set_tiling(r300->rws, tex->buffer, + tex->pitch[0], + tex->microtile != 0, + tex->macrotile != 0); + } + } + if (old_state->zsbuf && + (!new_state->zsbuf || + old_state->zsbuf->texture != new_state->zsbuf->texture)) { + tex = (struct r300_texture*)old_state->zsbuf->texture; + + if (tex) { + r300->rws->buffer_set_tiling(r300->rws, tex->buffer, + tex->pitch[0], + tex->microtile != 0, + tex->macrotile != 0); + } + } + + /* Set tiling flags for new surfaces. */ + for (i = 0; i < new_state->nr_cbufs; i++) { + tex = (struct r300_texture*)new_state->cbufs[i]->texture; + level = new_state->cbufs[i]->level; + + r300->rws->buffer_set_tiling(r300->rws, tex->buffer, + tex->pitch[level], + tex->microtile != 0, + tex->mip_macrotile[level] != 0); + } + if (new_state->zsbuf) { + tex = (struct r300_texture*)new_state->zsbuf->texture; + level = new_state->zsbuf->level; + + r300->rws->buffer_set_tiling(r300->rws, tex->buffer, + tex->pitch[level], + tex->microtile != 0, + tex->mip_macrotile[level] != 0); + } +} + static void r300_set_framebuffer_state(struct pipe_context* pipe, const struct pipe_framebuffer_state* state) { struct r300_context* r300 = r300_context(pipe); + struct r300_screen* r300screen = r300_screen(pipe->screen); + struct pipe_framebuffer_state *old_state = r300->fb_state.state; + unsigned max_width, max_height; + uint32_t zbuffer_bpp = 0; + + + if (state->nr_cbufs > 4) { + debug_printf("r300: Implementation error: Too many MRTs in %s, " + "refusing to bind framebuffer state!\n", __FUNCTION__); + return; + } + + if (r300screen->caps->is_r500) { + max_width = max_height = 4096; + } else if (r300screen->caps->is_r400) { + max_width = max_height = 4021; + } else { + max_width = max_height = 2560; + } + + if (state->width > max_width || state->height > max_height) { + debug_printf("r300: Implementation error: Render targets are too " + "big in %s, refusing to bind framebuffer state!\n", __FUNCTION__); + return; + } if (r300->draw) { draw_flush(r300->draw); } - r300->framebuffer_state = *state; + r300->fb_state.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; + } + /* If zsbuf is set from NULL to non-NULL or vice versa.. */ + if (!!old_state->zsbuf != !!state->zsbuf) { + r300->dsa_state.dirty = TRUE; + } + if (!r300->scissor_enabled) { + r300->scissor_state.dirty = TRUE; + } - /* Don't rely on the order of states being set for the first time. */ - r300->dirty_state |= R300_NEW_FRAMEBUFFERS; + r300_fb_update_tiling_flags(r300, r300->fb_state.state, state); - r300->blend_state.dirty = TRUE; - r300->dsa_state.dirty = TRUE; - r300->scissor_state.dirty = TRUE; + memcpy(r300->fb_state.state, state, sizeof(struct pipe_framebuffer_state)); + + r300->fb_state.size = (10 * state->nr_cbufs) + (2 * (4 - state->nr_cbufs)) + + (state->zsbuf ? 10 : 0) + 8; + + /* 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; + } + + if (r300->zbuffer_bpp != zbuffer_bpp) { + r300->zbuffer_bpp = zbuffer_bpp; + r300->rs_state.dirty = TRUE; + } + } } /* Create fragment shader state. */ @@ -533,8 +672,10 @@ static void r300_bind_fs_state(struct pipe_context* pipe, void* shader) r300->fs = fs; r300_pick_fragment_shader(r300); - if (r300->vs && r300_vertex_shader_setup_wpos(r300)) { - r300->dirty_state |= R300_NEW_VERTEX_FORMAT; + r300->rs_block_state.dirty = TRUE; /* Will be updated before the emission. */ + + if (r300->vs_state.state && r300_vertex_shader_setup_wpos(r300)) { + r300->vap_output_state.dirty = TRUE; } r300->dirty_state |= R300_NEW_FRAGMENT_SHADER | R300_NEW_FRAGMENT_SHADER_CONSTANTS; @@ -572,6 +713,7 @@ static void r300_set_polygon_stipple(struct pipe_context* pipe, static void* r300_create_rs_state(struct pipe_context* pipe, const struct pipe_rasterizer_state* state) { + struct r300_screen* r300screen = r300_screen(pipe->screen); struct r300_rs_state* rs = CALLOC_STRUCT(r300_rs_state); /* Copy rasterizer state for Draw. */ @@ -583,28 +725,17 @@ static void* r300_create_rs_state(struct pipe_context* pipe, rs->vap_control_status = R300_VC_32BIT_SWAP; #endif - /* If bypassing TCL, or if no TCL engine is present, turn off the HW TCL. - * Else, enable HW TCL and force Draw's TCL off. */ - if (state->bypass_vs_clip_and_viewport || - !r300_screen(pipe->screen)->caps->has_tcl) { + /* If no TCL engine is present, turn off the HW TCL. */ + if (!r300screen->caps->has_tcl) { rs->vap_control_status |= R300_VAP_TCL_BYPASS; } rs->point_size = pack_float_16_6x(state->point_size) | (pack_float_16_6x(state->point_size) << R300_POINTSIZE_X_SHIFT); - rs->point_minmax = - ((int)(state->point_size_min * 6.0) << - R300_GA_POINT_MINMAX_MIN_SHIFT) | - ((int)(state->point_size_max * 6.0) << - R300_GA_POINT_MINMAX_MAX_SHIFT); - rs->line_control = pack_float_16_6x(state->line_width) | R300_GA_LINE_CNTL_END_TYPE_COMP; - /* XXX I think there is something wrong with the polygon mode, - * XXX re-test when r300g is in a better shape */ - /* Enable polygon mode */ if (state->fill_cw != PIPE_POLYGON_MODE_FILL || state->fill_ccw != PIPE_POLYGON_MODE_FILL) { @@ -657,10 +788,8 @@ static void* r300_create_rs_state(struct pipe_context* pipe, } if (rs->polygon_offset_enable) { - rs->depth_offset_front = rs->depth_offset_back = - fui(state->offset_units); - rs->depth_scale_front = rs->depth_scale_back = - fui(state->offset_scale); + rs->depth_offset = state->offset_units; + rs->depth_scale = state->offset_scale; } if (state->line_stipple_enable) { @@ -686,6 +815,7 @@ static void r300_bind_rs_state(struct pipe_context* pipe, void* state) { struct r300_context* r300 = r300_context(pipe); struct r300_rs_state* rs = (struct r300_rs_state*)state; + boolean scissor_was_enabled = r300->scissor_enabled; if (r300->draw) { draw_flush(r300->draw); @@ -693,21 +823,18 @@ static void r300_bind_rs_state(struct pipe_context* pipe, void* state) } if (rs) { - r300->tcl_bypass = rs->rs.bypass_vs_clip_and_viewport; + r300->polygon_offset_enabled = rs->rs.offset_cw || rs->rs.offset_ccw; + r300->scissor_enabled = rs->rs.scissor; } else { - r300->tcl_bypass = FALSE; + r300->polygon_offset_enabled = FALSE; + r300->scissor_enabled = FALSE; } - r300->rs_state.state = rs; - r300->rs_state.dirty = TRUE; - /* XXX Why is this still needed, dammit!? */ - r300->scissor_state.dirty = TRUE; - r300->viewport_state.dirty = TRUE; + UPDATE_STATE(state, r300->rs_state); + r300->rs_state.size = 17 + (r300->polygon_offset_enabled ? 5 : 0); - /* XXX Clean these up when we move to atom emits */ - r300->dirty_state |= R300_NEW_RS_BLOCK; - if (r300->fs && r300->fs->inputs.wpos != ATTR_UNUSED) { - r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS; + if (scissor_was_enabled != r300->scissor_enabled) { + r300->scissor_state.dirty = TRUE; } } @@ -723,6 +850,7 @@ static void* { struct r300_context* r300 = r300_context(pipe); struct r300_sampler_state* sampler = CALLOC_STRUCT(r300_sampler_state); + boolean is_r500 = r300_screen(pipe->screen)->caps->is_r500; int lod_bias; union util_color uc; @@ -736,10 +864,12 @@ static void* sampler->filter0 |= r300_translate_tex_filters(state->min_img_filter, state->mag_img_filter, state->min_mip_filter, - state->max_anisotropy > 1.0); + state->max_anisotropy > 0); + + sampler->filter0 |= r300_anisotropy(state->max_anisotropy); /* Unfortunately, r300-r500 don't support floating-point mipmap lods. */ - /* We must pass these to the emit function to clamp them properly. */ + /* 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); @@ -747,9 +877,15 @@ static void* sampler->filter1 |= lod_bias << R300_LOD_BIAS_SHIFT; - sampler->filter1 |= r300_anisotropy(state->max_anisotropy); + /* This is very high quality anisotropic filtering for R5xx. + * It's good for benchmarking the performance of texturing but + * in practice we don't want to slow down the driver because it's + * a pretty good performance killer. Feel free to play with it. */ + if (DBG_ON(r300, DBG_ANISOHQ) && is_r500) { + sampler->filter1 |= r500_anisotropy(state->max_anisotropy); + } - util_pack_color(state->border_color, PIPE_FORMAT_A8R8G8B8_UNORM, &uc); + util_pack_color(state->border_color, PIPE_FORMAT_B8G8R8A8_UNORM, &uc); sampler->border_color = uc.ui; /* R500-specific fixups and optimizations */ @@ -765,23 +901,20 @@ static void r300_bind_sampler_states(struct pipe_context* pipe, void** states) { struct r300_context* r300 = r300_context(pipe); - int i; + struct r300_textures_state* state = + (struct r300_textures_state*)r300->textures_state.state; if (count > 8) { return; } - for (i = 0; i < count; i++) { - if (r300->sampler_states[i] != states[i]) { - r300->sampler_states[i] = (struct r300_sampler_state*)states[i]; - r300->dirty_state |= (R300_NEW_SAMPLER << i); - } - } + memcpy(state->sampler_states, states, sizeof(void*) * count); + state->sampler_count = count; - r300->sampler_count = count; + r300->textures_state.dirty = TRUE; /* Pick a fragment shader based on the texture compare state. */ - if (r300->fs && (r300->dirty_state & R300_ANY_NEW_SAMPLERS)) { + if (r300->fs && count) { if (r300_pick_fragment_shader(r300)) { r300->dirty_state |= R300_NEW_FRAGMENT_SHADER | R300_NEW_FRAGMENT_SHADER_CONSTANTS; @@ -800,27 +933,32 @@ static void r300_delete_sampler_state(struct pipe_context* pipe, void* state) FREE(state); } -static void r300_set_sampler_textures(struct pipe_context* pipe, - unsigned count, - struct pipe_texture** texture) +static void r300_set_fragment_sampler_views(struct pipe_context* pipe, + unsigned count, + struct pipe_sampler_view** views) { struct r300_context* r300 = r300_context(pipe); + unsigned i; boolean is_r500 = r300_screen(r300->context.screen)->caps->is_r500; - int i; + boolean dirty_tex = FALSE; /* XXX magic num */ if (count > 8) { return; } - + for (i = 0; i < count; i++) { - if (r300->textures[i] != (struct r300_texture*)texture[i]) { - pipe_texture_reference((struct pipe_texture**)&r300->textures[i], - texture[i]); - r300->dirty_state |= (R300_NEW_TEXTURE << i); + if (r300->fragment_sampler_views[i] != views[i]) { + struct r300_texture *texture; + + pipe_sampler_view_reference(&r300->fragment_sampler_views[i], + views[i]); + dirty_tex = TRUE; + + texture = (struct r300_texture *)views[i]->texture; /* R300-specific - set the texrect factor in a fragment shader */ - if (!is_r500 && r300->textures[i]->is_npot) { + if (!is_r500 && texture->is_npot) { /* XXX It would be nice to re-emit just 1 constant, * XXX not all of them */ r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS; @@ -829,14 +967,47 @@ static void r300_set_sampler_textures(struct pipe_context* pipe, } for (i = count; i < 8; i++) { - if (r300->textures[i]) { - pipe_texture_reference((struct pipe_texture**)&r300->textures[i], + if (r300->fragment_sampler_views[i]) { + pipe_sampler_view_reference(&r300->fragment_sampler_views[i], NULL); - r300->dirty_state |= (R300_NEW_TEXTURE << i); } } - r300->texture_count = count; + r300->fragment_sampler_view_count = count; + + r300->textures_state.dirty = TRUE; + + if (dirty_tex) { + r300->texture_cache_inval.dirty = TRUE; + } +} + +static struct pipe_sampler_view * +r300_create_sampler_view(struct pipe_context *pipe, + struct pipe_texture *texture, + const struct pipe_sampler_view *templ) +{ + struct r300_context *r300 = r300_context(pipe); + struct pipe_sampler_view *view = CALLOC_STRUCT(pipe_sampler_view); + + if (view) { + *view = *templ; + view->reference.count = 1; + view->texture = NULL; + pipe_texture_reference(&view->texture, texture); + view->context = pipe; + } + + return view; +} + + +static void +r300_sampler_view_destroy(struct pipe_context *pipe, + struct pipe_sampler_view *view) +{ + pipe_texture_reference(&view->texture, NULL); + FREE(view); } static void r300_set_scissor_state(struct pipe_context* pipe, @@ -847,7 +1018,9 @@ 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; + if (r300->scissor_enabled) { + r300->scissor_state.dirty = TRUE; + } } static void r300_set_viewport_state(struct pipe_context* pipe, @@ -857,6 +1030,8 @@ static void r300_set_viewport_state(struct pipe_context* pipe, struct r300_viewport_state* viewport = (struct r300_viewport_state*)r300->viewport_state.state; + r300->viewport = *state; + /* Do the transform in HW. */ viewport->vte_control = R300_VTX_W0_FMT; @@ -896,34 +1071,271 @@ static void r300_set_vertex_buffers(struct pipe_context* pipe, const struct pipe_vertex_buffer* buffers) { struct r300_context* r300 = r300_context(pipe); + int i; + unsigned max_index = (1 << 24) - 1; + boolean any_user_buffer = false; + + if (count == r300->vertex_buffer_count && + memcmp(r300->vertex_buffer, buffers, count * sizeof(buffers[0])) == 0) + return; + + for (i = 0; i < count; i++) { + pipe_buffer_reference(&r300->vertex_buffer[i].buffer, buffers[i].buffer); + if (r300_buffer_is_user_buffer(buffers[i].buffer)) + any_user_buffer = true; + max_index = MIN2(buffers[i].max_index, max_index); + } + + for ( ; i < r300->vertex_buffer_count; i++) + pipe_buffer_reference(&r300->vertex_buffer[i].buffer, NULL); memcpy(r300->vertex_buffer, buffers, - sizeof(struct pipe_vertex_buffer) * count); + sizeof(struct pipe_vertex_buffer) * count); + r300->vertex_buffer_count = count; + r300->vertex_buffer_max_index = max_index; + r300->any_user_vbs = any_user_buffer; if (r300->draw) { draw_flush(r300->draw); draw_set_vertex_buffers(r300->draw, count, buffers); } +} + +static boolean r300_validate_aos(struct r300_context *r300) +{ + struct pipe_vertex_buffer *vbuf = r300->vertex_buffer; + struct pipe_vertex_element *velem = r300->velems->velem; + int i; - r300->dirty_state |= R300_NEW_VERTEX_FORMAT; + /* Check if formats and strides are aligned to the size of DWORD. */ + for (i = 0; i < r300->velems->count; i++) { + if (vbuf[velem[i].vertex_buffer_index].stride % 4 != 0 || + util_format_get_blocksize(velem[i].src_format) % 4 != 0) { + return FALSE; + } + } + return TRUE; } -static void r300_set_vertex_elements(struct pipe_context* pipe, - unsigned count, - const struct pipe_vertex_element* elements) +static void r300_draw_emit_attrib(struct r300_context* r300, + enum attrib_emit emit, + enum interp_mode interp, + int index) { - struct r300_context* r300 = r300_context(pipe); + struct r300_vertex_shader* vs = r300->vs_state.state; + struct tgsi_shader_info* info = &vs->info; + int output; + + output = draw_find_shader_output(r300->draw, + info->output_semantic_name[index], + info->output_semantic_index[index]); + draw_emit_vertex_attr(&r300->vertex_info, emit, interp, output); +} + +static void r300_draw_emit_all_attribs(struct r300_context* r300) +{ + struct r300_vertex_shader* vs = r300->vs_state.state; + struct r300_shader_semantics* vs_outputs = &vs->outputs; + int i, gen_count; + + /* Position. */ + if (vs_outputs->pos != ATTR_UNUSED) { + r300_draw_emit_attrib(r300, EMIT_4F, INTERP_PERSPECTIVE, + vs_outputs->pos); + } else { + assert(0); + } + + /* Point size. */ + if (vs_outputs->psize != ATTR_UNUSED) { + r300_draw_emit_attrib(r300, EMIT_1F_PSIZE, INTERP_POS, + vs_outputs->psize); + } + + /* Colors. */ + for (i = 0; i < ATTR_COLOR_COUNT; i++) { + if (vs_outputs->color[i] != ATTR_UNUSED) { + r300_draw_emit_attrib(r300, EMIT_4F, INTERP_LINEAR, + vs_outputs->color[i]); + } + } + + /* XXX Back-face colors. */ + + /* Texture coordinates. */ + gen_count = 0; + for (i = 0; i < ATTR_GENERIC_COUNT; i++) { + if (vs_outputs->generic[i] != ATTR_UNUSED) { + r300_draw_emit_attrib(r300, EMIT_4F, INTERP_PERSPECTIVE, + vs_outputs->generic[i]); + gen_count++; + } + } + + /* Fog coordinates. */ + if (vs_outputs->fog != ATTR_UNUSED) { + r300_draw_emit_attrib(r300, EMIT_4F, INTERP_PERSPECTIVE, + vs_outputs->fog); + gen_count++; + } + + /* XXX magic */ + assert(gen_count <= 8); +} + +/* Update the PSC tables. */ +static void r300_vertex_psc(struct r300_vertex_element_state *velems) +{ + struct r300_vertex_stream_state *vstream = &velems->vertex_stream; + uint16_t type, swizzle; + enum pipe_format format; + unsigned i; + + assert(velems->count <= 16); + + /* Vertex shaders have no semantics on their inputs, + * so PSC should just route stuff based on the vertex elements, + * and not on attrib information. */ + for (i = 0; i < velems->count; i++) { + format = velems->velem[i].src_format; + + type = r300_translate_vertex_data_type(format) | + (i << R300_DST_VEC_LOC_SHIFT); + swizzle = r300_translate_vertex_data_swizzle(format); + + if (i & 1) { + vstream->vap_prog_stream_cntl[i >> 1] |= type << 16; + vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle << 16; + } else { + vstream->vap_prog_stream_cntl[i >> 1] |= type; + vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle; + } + } + + /* Set the last vector in the PSC. */ + if (i) { + i -= 1; + } + vstream->vap_prog_stream_cntl[i >> 1] |= + (R300_LAST_VEC << (i & 1 ? 16 : 0)); + + vstream->count = (i >> 1) + 1; +} + +/* Update the PSC tables for SW TCL, using Draw. */ +static void r300_swtcl_vertex_psc(struct r300_context *r300, + struct r300_vertex_element_state *velems) +{ + struct r300_vertex_stream_state *vstream = &velems->vertex_stream; + struct r300_vertex_shader* vs = r300->vs_state.state; + struct vertex_info* vinfo = &r300->vertex_info; + uint16_t type, swizzle; + enum pipe_format format; + unsigned i, attrib_count; + int* vs_output_tab = vs->stream_loc_notcl; + + /* For each Draw attribute, route it to the fragment shader according + * to the vs_output_tab. */ + attrib_count = vinfo->num_attribs; + DBG(r300, DBG_DRAW, "r300: attrib count: %d\n", attrib_count); + for (i = 0; i < attrib_count; i++) { + DBG(r300, DBG_DRAW, "r300: attrib: offset %d, interp %d, size %d," + " vs_output_tab %d\n", vinfo->attrib[i].src_index, + vinfo->attrib[i].interp_mode, vinfo->attrib[i].emit, + vs_output_tab[i]); + } + + for (i = 0; i < attrib_count; i++) { + /* Make sure we have a proper destination for our attribute. */ + assert(vs_output_tab[i] != -1); + + format = draw_translate_vinfo_format(vinfo->attrib[i].emit); + + /* Obtain the type of data in this attribute. */ + type = r300_translate_vertex_data_type(format) | + vs_output_tab[i] << R300_DST_VEC_LOC_SHIFT; + + /* Obtain the swizzle for this attribute. Note that the default + * swizzle in the hardware is not XYZW! */ + swizzle = r300_translate_vertex_data_swizzle(format); + + /* Add the attribute to the PSC table. */ + if (i & 1) { + vstream->vap_prog_stream_cntl[i >> 1] |= type << 16; + vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle << 16; + } else { + vstream->vap_prog_stream_cntl[i >> 1] |= type; + vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle; + } + } + + /* Set the last vector in the PSC. */ + if (i) { + i -= 1; + } + vstream->vap_prog_stream_cntl[i >> 1] |= + (R300_LAST_VEC << (i & 1 ? 16 : 0)); - memcpy(r300->vertex_element, - elements, - sizeof(struct pipe_vertex_element) * count); - r300->vertex_element_count = count; + vstream->count = (i >> 1) + 1; +} + +static void* r300_create_vertex_elements_state(struct pipe_context* pipe, + unsigned count, + const struct pipe_vertex_element* attribs) +{ + struct r300_context *r300 = r300_context(pipe); + struct r300_screen* r300screen = r300_screen(pipe->screen); + struct r300_vertex_element_state *velems; + + assert(count <= PIPE_MAX_ATTRIBS); + velems = CALLOC_STRUCT(r300_vertex_element_state); + if (velems != NULL) { + velems->count = count; + memcpy(velems->velem, attribs, sizeof(struct pipe_vertex_element) * count); + + if (r300screen->caps->has_tcl) { + r300_vertex_psc(velems); + } else { + memset(&r300->vertex_info, 0, sizeof(struct vertex_info)); + r300_draw_emit_all_attribs(r300); + draw_compute_vertex_size(&r300->vertex_info); + r300_swtcl_vertex_psc(r300, velems); + } + } + return velems; +} + +static void r300_bind_vertex_elements_state(struct pipe_context *pipe, + void *state) +{ + struct r300_context *r300 = r300_context(pipe); + struct r300_vertex_element_state *velems = state; + + if (velems == NULL) { + return; + } + + r300->velems = velems; if (r300->draw) { draw_flush(r300->draw); - draw_set_vertex_elements(r300->draw, count, elements); + draw_set_vertex_elements(r300->draw, velems->count, velems->velem); + } + + if (!r300_validate_aos(r300)) { + /* XXX We should fallback using draw. */ + assert(0); + abort(); } + + UPDATE_STATE(&velems->vertex_stream, r300->vertex_stream_state); + r300->vertex_stream_state.size = (1 + velems->vertex_stream.count) * 2; +} + +static void r300_delete_vertex_elements_state(struct pipe_context *pipe, void *state) +{ + FREE(state); } static void* r300_create_vs_state(struct pipe_context* pipe, @@ -931,86 +1343,127 @@ static void* r300_create_vs_state(struct pipe_context* pipe, { struct r300_context* r300 = r300_context(pipe); - if (r300_screen(pipe->screen)->caps->has_tcl) { - struct r300_vertex_shader* vs = CALLOC_STRUCT(r300_vertex_shader); - /* Copy state directly into shader. */ - vs->state = *shader; - vs->state.tokens = tgsi_dup_tokens(shader->tokens); - - tgsi_scan_shader(shader->tokens, &vs->info); + struct r300_vertex_shader* vs = CALLOC_STRUCT(r300_vertex_shader); + r300_vertex_shader_common_init(vs, shader); - return (void*)vs; + if (r300_screen(pipe->screen)->caps->has_tcl) { + r300_translate_vertex_shader(r300, vs); } else { - return draw_create_vertex_shader(r300->draw, shader); + vs->draw_vs = draw_create_vertex_shader(r300->draw, shader); } + + return vs; } static void r300_bind_vs_state(struct pipe_context* pipe, void* shader) { struct r300_context* r300 = r300_context(pipe); + struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader; - if (r300_screen(pipe->screen)->caps->has_tcl) { - struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader; + if (vs == NULL) { + r300->vs_state.state = NULL; + return; + } + if (vs == r300->vs_state.state) { + return; + } + r300->vs_state.state = vs; - if (vs == NULL) { - r300->vs = NULL; - return; - } else if (!vs->translated) { - r300_translate_vertex_shader(r300, vs); - } + // VS output mapping for HWTCL or stream mapping for SWTCL to the RS block + if (r300->fs) { + r300_vertex_shader_setup_wpos(r300); + } + memcpy(r300->vap_output_state.state, &vs->vap_out, + sizeof(struct r300_vap_output_state)); + r300->vap_output_state.dirty = TRUE; - r300->vs = vs; - if (r300->fs) { - r300_vertex_shader_setup_wpos(r300); - } + /* 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. */ + + if (r300_screen(pipe->screen)->caps->has_tcl) { + r300->vs_state.dirty = TRUE; + r300->vs_state.size = vs->code.length + 9; - r300->dirty_state |= - R300_NEW_VERTEX_SHADER | R300_NEW_VERTEX_SHADER_CONSTANTS | - R300_NEW_VERTEX_FORMAT; + r300->pvs_flush.dirty = TRUE; + + r300->dirty_state |= R300_NEW_VERTEX_SHADER_CONSTANTS; } else { draw_flush(r300->draw); draw_bind_vertex_shader(r300->draw, - (struct draw_vertex_shader*)shader); + (struct draw_vertex_shader*)vs->draw_vs); } } static void r300_delete_vs_state(struct pipe_context* pipe, void* shader) { struct r300_context* r300 = r300_context(pipe); + struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader; if (r300_screen(pipe->screen)->caps->has_tcl) { - struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader; - rc_constants_destroy(&vs->code.constants); - FREE((void*)vs->state.tokens); - FREE(shader); } else { draw_delete_vertex_shader(r300->draw, - (struct draw_vertex_shader*)shader); + (struct draw_vertex_shader*)vs->draw_vs); } + + FREE((void*)vs->state.tokens); + FREE(shader); } static void r300_set_constant_buffer(struct pipe_context *pipe, uint shader, uint index, - const struct pipe_constant_buffer *buf) + struct pipe_buffer *buf) { struct r300_context* r300 = r300_context(pipe); + struct r300_screen *r300screen = r300_screen(pipe->screen); void *mapped; + int max_size = 0; - if (buf == NULL || buf->buffer->size == 0 || - (mapped = pipe_buffer_map(pipe->screen, buf->buffer, PIPE_BUFFER_USAGE_CPU_READ)) == NULL) + if (buf == NULL || buf->size == 0 || + (mapped = pipe_buffer_map(pipe->screen, buf, PIPE_BUFFER_USAGE_CPU_READ)) == NULL) { r300->shader_constants[shader].count = 0; return; } - assert((buf->buffer->size % 4 * sizeof(float)) == 0); - memcpy(r300->shader_constants[shader].constants, mapped, buf->buffer->size); - r300->shader_constants[shader].count = buf->buffer->size / (4 * sizeof(float)); - pipe_buffer_unmap(pipe->screen, buf->buffer); + assert((buf->size % 4 * sizeof(float)) == 0); + + /* Check the size of the constant buffer. */ + switch (shader) { + case PIPE_SHADER_VERTEX: + max_size = 256; + break; + case PIPE_SHADER_FRAGMENT: + if (r300screen->caps->is_r500) { + max_size = 256; + /* XXX Implement emission of r400's extended constant buffer. */ + /*} else if (r300screen->caps->is_r400) { + max_size = 64;*/ + } else { + max_size = 32; + } + break; + default: + assert(0); + } - if (shader == PIPE_SHADER_VERTEX) - r300->dirty_state |= R300_NEW_VERTEX_SHADER_CONSTANTS; + /* XXX Subtract immediates and RC_STATE_* variables. */ + if (buf->size > (sizeof(float) * 4 * max_size)) { + debug_printf("r300: Max size of the constant buffer is " + "%i*4 floats.\n", max_size); + abort(); + } + + memcpy(r300->shader_constants[shader].constants, mapped, buf->size); + r300->shader_constants[shader].count = buf->size / (4 * sizeof(float)); + pipe_buffer_unmap(pipe->screen, buf); + + if (shader == PIPE_SHADER_VERTEX) { + if (r300screen->caps->has_tcl) { + r300->dirty_state |= R300_NEW_VERTEX_SHADER_CONSTANTS; + r300->pvs_flush.dirty = TRUE; + } + } else if (shader == PIPE_SHADER_FRAGMENT) r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS; } @@ -1031,6 +1484,8 @@ void r300_init_state_functions(struct r300_context* r300) r300->context.bind_depth_stencil_alpha_state = r300_bind_dsa_state; r300->context.delete_depth_stencil_alpha_state = r300_delete_dsa_state; + r300->context.set_stencil_ref = r300_set_stencil_ref; + r300->context.set_framebuffer_state = r300_set_framebuffer_state; r300->context.create_fs_state = r300_create_fs_state; @@ -1048,14 +1503,19 @@ void r300_init_state_functions(struct r300_context* r300) r300->context.bind_vertex_sampler_states = r300_lacks_vertex_textures; r300->context.delete_sampler_state = r300_delete_sampler_state; - r300->context.set_fragment_sampler_textures = r300_set_sampler_textures; + r300->context.set_fragment_sampler_views = r300_set_fragment_sampler_views; + r300->context.create_sampler_view = r300_create_sampler_view; + r300->context.sampler_view_destroy = r300_sampler_view_destroy; r300->context.set_scissor_state = r300_set_scissor_state; r300->context.set_viewport_state = r300_set_viewport_state; r300->context.set_vertex_buffers = r300_set_vertex_buffers; - r300->context.set_vertex_elements = r300_set_vertex_elements; + + r300->context.create_vertex_elements_state = r300_create_vertex_elements_state; + r300->context.bind_vertex_elements_state = r300_bind_vertex_elements_state; + r300->context.delete_vertex_elements_state = r300_delete_vertex_elements_state; r300->context.create_vs_state = r300_create_vs_state; r300->context.bind_vs_state = r300_bind_vs_state;