X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fgallium%2Fdrivers%2Fr300%2Fr300_state.c;h=7127ea1ac16b2641ecb50b89c08630fe24e50f3a;hb=37a85b1830188503feb286ebdee0cdb44fa298af;hp=b810f4081c8a9d61a4c53c52df65df4fad8fb17f;hpb=cf143c1f4d7c3636ddd5c767518b1b00ff46b16c;p=mesa.git diff --git a/src/gallium/drivers/r300/r300_state.c b/src/gallium/drivers/r300/r300_state.c index b810f4081c8..7127ea1ac16 100644 --- a/src/gallium/drivers/r300/r300_state.c +++ b/src/gallium/drivers/r300/r300_state.c @@ -24,6 +24,7 @@ #include "draw/draw_context.h" #include "util/u_framebuffer.h" +#include "util/u_half.h" #include "util/u_math.h" #include "util/u_mm.h" #include "util/u_memory.h" @@ -44,7 +45,6 @@ #include "r300_fs.h" #include "r300_texture.h" #include "r300_vs.h" -#include "r300_winsys.h" /* r300_state: Functions used to intialize state context by translating * Gallium state objects into semi-native r300 state objects. */ @@ -187,12 +187,16 @@ static void* r300_create_blend_state(struct pipe_context* pipe, struct r300_screen* r300screen = r300_screen(pipe->screen); struct r300_blend_state* blend = CALLOC_STRUCT(r300_blend_state); uint32_t blend_control = 0; /* R300_RB3D_CBLEND: 0x4e04 */ + uint32_t blend_control_noclamp = 0; /* R300_RB3D_CBLEND: 0x4e04 */ uint32_t alpha_blend_control = 0; /* R300_RB3D_ABLEND: 0x4e08 */ + uint32_t alpha_blend_control_noclamp = 0; /* R300_RB3D_ABLEND: 0x4e08 */ uint32_t color_channel_mask = 0; /* R300_RB3D_COLOR_CHANNEL_MASK: 0x4e0c */ uint32_t rop = 0; /* R300_RB3D_ROPCNTL: 0x4e18 */ uint32_t dither = 0; /* R300_RB3D_DITHER_CTL: 0x4e50 */ CB_LOCALS; + blend->state = *state; + if (state->rt[0].blend_enable) { unsigned eqRGB = state->rt[0].rgb_func; @@ -205,10 +209,14 @@ static void* r300_create_blend_state(struct pipe_context* pipe, /* despite the name, ALPHA_BLEND_ENABLE has nothing to do with alpha, * this is just the crappy D3D naming */ - blend_control = R300_ALPHA_BLEND_ENABLE | - r300_translate_blend_function(eqRGB) | + blend_control = blend_control_noclamp = + R300_ALPHA_BLEND_ENABLE | ( r300_translate_blend_factor(srcRGB) << R300_SRC_BLEND_SHIFT) | ( r300_translate_blend_factor(dstRGB) << R300_DST_BLEND_SHIFT); + blend_control |= + r300_translate_blend_function(eqRGB, TRUE); + blend_control_noclamp |= + r300_translate_blend_function(eqRGB, FALSE); /* Optimization: some operations do not require the destination color. * @@ -230,6 +238,7 @@ static void* r300_create_blend_state(struct pipe_context* pipe, srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE) { /* Enable reading from the colorbuffer. */ blend_control |= R300_READ_ENABLE; + blend_control_noclamp |= R300_READ_ENABLE; if (r300screen->caps.is_r500) { /* Optimization: Depending on incoming pixels, we can @@ -305,10 +314,14 @@ static void* r300_create_blend_state(struct pipe_context* pipe, /* separate alpha */ if (srcA != srcRGB || dstA != dstRGB || eqA != eqRGB) { blend_control |= R300_SEPARATE_ALPHA_ENABLE; - alpha_blend_control = - r300_translate_blend_function(eqA) | + blend_control_noclamp |= R300_SEPARATE_ALPHA_ENABLE; + alpha_blend_control = alpha_blend_control_noclamp = (r300_translate_blend_factor(srcA) << R300_SRC_BLEND_SHIFT) | (r300_translate_blend_factor(dstA) << R300_DST_BLEND_SHIFT); + alpha_blend_control |= + r300_translate_blend_function(eqA, TRUE); + alpha_blend_control_noclamp |= + r300_translate_blend_function(eqA, FALSE); } } @@ -345,7 +358,7 @@ static void* r300_create_blend_state(struct pipe_context* pipe, */ /* Build a command buffer. */ - BEGIN_CB(blend->cb, 8); + BEGIN_CB(blend->cb_clamp, 8); OUT_CB_REG(R300_RB3D_ROPCNTL, rop); OUT_CB_REG_SEQ(R300_RB3D_CBLEND, 3); OUT_CB(blend_control); @@ -354,6 +367,16 @@ static void* r300_create_blend_state(struct pipe_context* pipe, OUT_CB_REG(R300_RB3D_DITHER_CTL, dither); END_CB; + /* Build a command buffer. */ + BEGIN_CB(blend->cb_noclamp, 8); + OUT_CB_REG(R300_RB3D_ROPCNTL, rop); + OUT_CB_REG_SEQ(R300_RB3D_CBLEND, 3); + OUT_CB(blend_control_noclamp); + OUT_CB(alpha_blend_control_noclamp); + OUT_CB(color_channel_mask); + OUT_CB_REG(R300_RB3D_DITHER_CTL, dither); + END_CB; + /* The same as above, but with no colorbuffer reads and writes. */ BEGIN_CB(blend->cb_no_readwrite, 8); OUT_CB_REG(R300_RB3D_ROPCNTL, rop); @@ -395,22 +418,64 @@ static void r300_set_blend_color(struct pipe_context* pipe, const struct pipe_blend_color* color) { struct r300_context* r300 = r300_context(pipe); - struct r300_blend_color_state* state = + struct pipe_framebuffer_state *fb = r300->fb_state.state; + struct r300_blend_color_state *state = (struct r300_blend_color_state*)r300->blend_color_state.state; + struct pipe_blend_color c; + enum pipe_format format = fb->nr_cbufs ? fb->cbufs[0]->format : 0; CB_LOCALS; + state->state = *color; /* Save it, so that we can reuse it in set_fb_state */ + c = *color; + + /* The blend color is dependent on the colorbuffer format. */ + if (fb->nr_cbufs) { + switch (format) { + case PIPE_FORMAT_R8_UNORM: + case PIPE_FORMAT_L8_UNORM: + case PIPE_FORMAT_I8_UNORM: + c.color[1] = c.color[0]; + break; + + case PIPE_FORMAT_A8_UNORM: + c.color[1] = c.color[3]; + break; + + case PIPE_FORMAT_R8G8_UNORM: + c.color[2] = c.color[1]; + break; + + case PIPE_FORMAT_L8A8_UNORM: + c.color[2] = c.color[3]; + break; + + default:; + } + } + if (r300->screen->caps.is_r500) { - /* XXX if FP16 blending is enabled, we should use the FP16 format */ BEGIN_CB(state->cb, 3); OUT_CB_REG_SEQ(R500_RB3D_CONSTANT_COLOR_AR, 2); - OUT_CB(float_to_fixed10(color->color[0]) | - (float_to_fixed10(color->color[3]) << 16)); - OUT_CB(float_to_fixed10(color->color[2]) | - (float_to_fixed10(color->color[1]) << 16)); + + switch (format) { + case PIPE_FORMAT_R16G16B16A16_FLOAT: + OUT_CB(util_float_to_half(c.color[2]) | + (util_float_to_half(c.color[3]) << 16)); + OUT_CB(util_float_to_half(c.color[0]) | + (util_float_to_half(c.color[1]) << 16)); + break; + + default: + OUT_CB(float_to_fixed10(c.color[0]) | + (float_to_fixed10(c.color[3]) << 16)); + OUT_CB(float_to_fixed10(c.color[2]) | + (float_to_fixed10(c.color[1]) << 16)); + } + END_CB; } else { union util_color uc; - util_pack_color(color->color, PIPE_FORMAT_B8G8R8A8_UNORM, &uc); + util_pack_color(c.color, PIPE_FORMAT_B8G8R8A8_UNORM, &uc); BEGIN_CB(state->cb, 2); OUT_CB_REG(R300_RB3D_BLEND_COLOR, uc.ui); @@ -484,6 +549,10 @@ static void* dsa->z_stencil_control |= (r300_translate_depth_stencil_function(state->depth.func) << R300_Z_FUNC_SHIFT); + } else { + /* We must enable depth test, otherwise occlusion queries won't work. */ + dsa->z_buffer_control |= R300_Z_ENABLE; + dsa->z_stencil_control |= R300_ZS_ALWAYS; } /* Stencil buffer setup. */ @@ -537,29 +606,56 @@ static void* r300_translate_alpha_function(state->alpha.func) | R300_FG_ALPHA_FUNC_ENABLE; - /* We could use 10bit alpha ref but who needs that? */ dsa->alpha_function |= float_to_ubyte(state->alpha.ref_value); + dsa->alpha_value = util_float_to_half(state->alpha.ref_value); - if (caps->is_r500) + if (caps->is_r500) { + dsa->alpha_function_fp16 = dsa->alpha_function | + R500_FG_ALPHA_FUNC_FP16_ENABLE; dsa->alpha_function |= R500_FG_ALPHA_FUNC_8BIT; + } } - BEGIN_CB(&dsa->cb_begin, 8); + BEGIN_CB(&dsa->cb_begin, 10); OUT_CB_REG(R300_FG_ALPHA_FUNC, dsa->alpha_function); OUT_CB_REG_SEQ(R300_ZB_CNTL, 3); OUT_CB(dsa->z_buffer_control); OUT_CB(dsa->z_stencil_control); OUT_CB(dsa->stencil_ref_mask); OUT_CB_REG(R500_ZB_STENCILREFMASK_BF, dsa->stencil_ref_bf); + OUT_CB_REG(R500_FG_ALPHA_VALUE, dsa->alpha_value); END_CB; - BEGIN_CB(dsa->cb_no_readwrite, 8); + BEGIN_CB(&dsa->cb_begin_fp16, 10); + OUT_CB_REG(R300_FG_ALPHA_FUNC, dsa->alpha_function_fp16); + OUT_CB_REG_SEQ(R300_ZB_CNTL, 3); + OUT_CB(dsa->z_buffer_control); + OUT_CB(dsa->z_stencil_control); + OUT_CB(dsa->stencil_ref_mask); + OUT_CB_REG(R500_ZB_STENCILREFMASK_BF, dsa->stencil_ref_bf); + OUT_CB_REG(R500_FG_ALPHA_VALUE, dsa->alpha_value); + END_CB; + + /* We must enable depth test, otherwise occlusion queries won't work. + * We setup a dummy zbuffer to silent the CS checker, see emit_fb_state. */ + BEGIN_CB(dsa->cb_zb_no_readwrite, 10); OUT_CB_REG(R300_FG_ALPHA_FUNC, dsa->alpha_function); OUT_CB_REG_SEQ(R300_ZB_CNTL, 3); + OUT_CB(R300_Z_ENABLE); + OUT_CB(R300_ZS_ALWAYS); OUT_CB(0); - OUT_CB(0); + OUT_CB_REG(R500_ZB_STENCILREFMASK_BF, 0); + OUT_CB_REG(R500_FG_ALPHA_VALUE, dsa->alpha_value); + END_CB; + + BEGIN_CB(dsa->cb_fp16_zb_no_readwrite, 10); + OUT_CB_REG(R300_FG_ALPHA_FUNC, dsa->alpha_function_fp16); + OUT_CB_REG_SEQ(R300_ZB_CNTL, 3); + OUT_CB(R300_Z_ENABLE); + OUT_CB(R300_ZS_ALWAYS); OUT_CB(0); OUT_CB_REG(R500_ZB_STENCILREFMASK_BF, 0); + OUT_CB_REG(R500_FG_ALPHA_VALUE, dsa->alpha_value); END_CB; return (void*)dsa; @@ -678,7 +774,6 @@ 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); @@ -686,6 +781,8 @@ void r300_mark_fb_state_dirty(struct r300_context *r300, /* 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); + r300_mark_atom_dirty(r300, &r300->dsa_state); /* for AlphaRef */ + r300_set_blend_color(&r300->context, r300->blend_color_state.state); } if (change == R300_CHANGED_FB_STATE || @@ -701,12 +798,14 @@ void r300_mark_fb_state_dirty(struct r300_context *r300, /* Now compute the fb_state atom size. */ r300->fb_state.size = 2 + (8 * state->nr_cbufs); - if (r300->cbzb_clear) + if (r300->cbzb_clear) { r300->fb_state.size += 10; - else if (state->zsbuf) { + } else if (state->zsbuf) { r300->fb_state.size += 10; - if (can_hyperz) + if (r300->hyperz_enabled) r300->fb_state.size += 8; + } else if (state->nr_cbufs) { + r300->fb_state.size += 10; } /* The size of the rest of atoms stays the same. */ @@ -721,6 +820,7 @@ r300_set_framebuffer_state(struct pipe_context* pipe, struct pipe_framebuffer_state *old_state = r300->fb_state.state; unsigned max_width, max_height, i; uint32_t zbuffer_bpp = 0; + boolean unlock_zbuffer = FALSE; if (r300->screen->caps.is_r500) { max_width = max_height = 4096; @@ -736,7 +836,7 @@ r300_set_framebuffer_state(struct pipe_context* pipe, return; } - if (old_state->zsbuf && r300->zmask_in_use && !r300->hyperz_locked) { + if (old_state->zsbuf && r300->zmask_in_use && !r300->locked_zbuffer) { /* There is a zmask in use, what are we gonna do? */ if (state->zsbuf) { if (!pipe_surface_equal(old_state->zsbuf, state->zsbuf)) { @@ -746,10 +846,9 @@ r300_set_framebuffer_state(struct pipe_context* pipe, } } else { /* We don't bind another zbuffer, so lock the current one. */ - r300->hyperz_locked = TRUE; pipe_surface_reference(&r300->locked_zbuffer, old_state->zsbuf); } - } else if (r300->hyperz_locked && r300->locked_zbuffer) { + } else if (r300->locked_zbuffer) { /* We have a locked zbuffer now, what are we gonna do? */ if (state->zsbuf) { if (!pipe_surface_equal(r300->locked_zbuffer, state->zsbuf)) { @@ -759,15 +858,15 @@ r300_set_framebuffer_state(struct pipe_context* pipe, r300->hiz_in_use = FALSE; } else { /* We are binding the locked zbuffer again, so unlock it. */ - r300->hyperz_locked = FALSE; + unlock_zbuffer = TRUE; } } } + assert(state->zsbuf || (r300->locked_zbuffer && !unlock_zbuffer) || !r300->zmask_in_use); + + /* Need to reset clamping or colormask. */ + r300_mark_atom_dirty(r300, &r300->blend_state); - /* If nr_cbufs is changed from zero to non-zero or vice versa... */ - if (!!old_state->nr_cbufs != !!state->nr_cbufs) { - 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_mark_atom_dirty(r300, &r300->dsa_state); @@ -778,7 +877,7 @@ r300_set_framebuffer_state(struct pipe_context* pipe, util_copy_framebuffer_state(r300->fb_state.state, state); - if (!r300->hyperz_locked) { + if (unlock_zbuffer) { pipe_surface_reference(&r300->locked_zbuffer, NULL); } @@ -877,24 +976,14 @@ 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); - - if (fb->nr_cbufs > 1 && - last_multi_write != r300_fragment_shader_writes_all(fs)) { - r300_mark_fb_state_dirty(r300, R300_CHANGED_MULTIWRITE); - } + r300->fs_status = FRAGMENT_SHADER_DIRTY; r300_mark_atom_dirty(r300, &r300->rs_block_state); /* Will be updated before the emission. */ } @@ -944,12 +1033,14 @@ static void* r300_create_rs_state(struct pipe_context* pipe, uint32_t line_stipple_value; /* R300_GA_LINE_STIPPLE_VALUE: 0x4260 */ uint32_t polygon_mode; /* R300_GA_POLY_MODE: 0x4288 */ uint32_t clip_rule; /* R300_SC_CLIP_RULE: 0x43D0 */ + uint32_t round_mode; /* R300_GA_ROUND_MODE: 0x428c */ /* Point sprites texture coordinates, 0: lower left, 1: upper right */ 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 */ + boolean vclamp = state->clamp_vertex_color; CB_LOCALS; /* Copy rasterizer state. */ @@ -1072,6 +1163,12 @@ static void* r300_create_rs_state(struct pipe_context* pipe, } } + /* Vertex color clamping. FP20 means no clamping. */ + round_mode = + R300_GA_ROUND_MODE_GEOMETRY_ROUND_NEAREST | + (!vclamp ? (R300_GA_ROUND_MODE_RGB_CLAMP_FP20 | + R300_GA_ROUND_MODE_ALPHA_CLAMP_FP20) : 0); + /* Build the main command buffer. */ BEGIN_CB(rs->cb_main, RS_STATE_MAIN_SIZE); OUT_CB_REG(R300_VAP_CNTL_STATUS, vap_control_status); @@ -1086,6 +1183,7 @@ static void* r300_create_rs_state(struct pipe_context* pipe, OUT_CB_REG(R300_GA_LINE_STIPPLE_CONFIG, line_stipple_config); OUT_CB_REG(R300_GA_LINE_STIPPLE_VALUE, line_stipple_value); OUT_CB_REG(R300_GA_POLY_MODE, polygon_mode); + OUT_CB_REG(R300_GA_ROUND_MODE, round_mode); OUT_CB_REG(R300_SC_CLIP_RULE, clip_rule); OUT_CB_REG_SEQ(R300_GA_POINT_S0, 4); OUT_CB_32F(point_texcoord_left); @@ -1128,6 +1226,7 @@ static void r300_bind_rs_state(struct pipe_context* pipe, void* state) struct r300_rs_state* rs = (struct r300_rs_state*)state; int last_sprite_coord_enable = r300->sprite_coord_enable; boolean last_two_sided_color = r300->two_sided_color; + boolean last_frag_clamp = r300->frag_clamp; if (r300->draw && rs) { draw_set_rasterizer_state(r300->draw, &rs->rs_draw, state); @@ -1137,10 +1236,12 @@ static void r300_bind_rs_state(struct pipe_context* pipe, void* state) 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; + r300->frag_clamp = rs->rs.clamp_fragment_color; } else { r300->polygon_offset_enabled = FALSE; r300->sprite_coord_enable = 0; r300->two_sided_color = FALSE; + r300->frag_clamp = FALSE; } UPDATE_STATE(state, r300->rs_state); @@ -1150,6 +1251,11 @@ static void r300_bind_rs_state(struct pipe_context* pipe, void* state) last_two_sided_color != r300->two_sided_color) { r300_mark_atom_dirty(r300, &r300->rs_block_state); } + + if (last_frag_clamp != r300->frag_clamp && + r300->fs_status == FRAGMENT_SHADER_VALID) { + r300->fs_status = FRAGMENT_SHADER_MAYBE_DIRTY; + } } /* Free rasterizer state. */ @@ -1359,6 +1465,8 @@ r300_create_sampler_view(struct pipe_context *pipe, boolean dxtc_swizzle = r300_screen(pipe->screen)->caps.dxtc_swizzle; if (view) { + unsigned hwformat; + view->base = *templ; view->base.reference.count = 1; view->base.context = pipe; @@ -1370,11 +1478,19 @@ r300_create_sampler_view(struct pipe_context *pipe, view->swizzle[2] = templ->swizzle_b; view->swizzle[3] = templ->swizzle_a; + hwformat = r300_translate_texformat(templ->format, + view->swizzle, + is_r500, + dxtc_swizzle); + + if (hwformat == ~0) { + fprintf(stderr, "r300: Ooops. Got unsupported format %s in %s.\n", + util_format_short_name(templ->format), __func__); + } + assert(hwformat != ~0); + view->format = tex->tx_format; - view->format.format1 |= r300_translate_texformat(templ->format, - view->swizzle, - is_r500, - dxtc_swizzle); + view->format.format1 |= hwformat; if (is_r500) { view->format.format2 |= r500_tx_format_msb_bit(templ->format); } @@ -1446,7 +1562,8 @@ static void r300_set_viewport_state(struct pipe_context* pipe, } r300_mark_atom_dirty(r300, &r300->viewport_state); - if (r300->fs.state && r300_fs(r300)->shader->inputs.wpos != ATTR_UNUSED) { + if (r300->fs.state && r300_fs(r300)->shader && + r300_fs(r300)->shader->inputs.wpos != ATTR_UNUSED) { r300_mark_atom_dirty(r300, &r300->fs_rc_constant_state); } } @@ -1512,12 +1629,6 @@ static void r300_vertex_psc(struct r300_vertex_element_state *velems) enum pipe_format format; unsigned i; - if (velems->count > 16) { - fprintf(stderr, "r300: More than 16 vertex elements are not supported," - " requested %i, using 16.\n", velems->count); - 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. */ @@ -1568,9 +1679,12 @@ static void* r300_create_vertex_elements_state(struct pipe_context* pipe, dummy_attrib.src_format = PIPE_FORMAT_R8G8B8A8_UNORM; attribs = &dummy_attrib; count = 1; + } else if (count > 16) { + fprintf(stderr, "r300: More than 16 vertex elements are not supported," + " requested %i, using 16.\n", count); + count = 16; } - assert(count <= PIPE_MAX_ATTRIBS); velems = CALLOC_STRUCT(r300_vertex_element_state); if (!velems) return NULL; @@ -1768,6 +1882,14 @@ static void r300_set_constant_buffer(struct pipe_context *pipe, } } +static void r300_texture_barrier(struct pipe_context *pipe) +{ + struct r300_context *r300 = r300_context(pipe); + + r300_mark_atom_dirty(r300, &r300->gpu_flush); + r300_mark_atom_dirty(r300, &r300->texture_cache_inval); +} + void r300_init_state_functions(struct r300_context* r300) { r300->context.create_blend_state = r300_create_blend_state; @@ -1823,4 +1945,6 @@ void r300_init_state_functions(struct r300_context* r300) r300->context.create_vs_state = r300_create_vs_state; r300->context.bind_vs_state = r300_bind_vs_state; r300->context.delete_vs_state = r300_delete_vs_state; + + r300->context.texture_barrier = r300_texture_barrier; }