X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fgallium%2Fdrivers%2Fr300%2Fr300_state_derived.c;h=f63114e7eb74a67cbd3227717aa5f6c2e34f603c;hb=636d01bd61cac83e13c3c64874e7e34e828ca93a;hp=95be7849f8f976380a366d5e105e5b99519a9288;hpb=3332229b3ba7a183a9f120ae4bbf9865e96df110;p=mesa.git diff --git a/src/gallium/drivers/r300/r300_state_derived.c b/src/gallium/drivers/r300/r300_state_derived.c index 95be7849f8f..f63114e7eb7 100644 --- a/src/gallium/drivers/r300/r300_state_derived.c +++ b/src/gallium/drivers/r300/r300_state_derived.c @@ -29,7 +29,6 @@ #include "r300_context.h" #include "r300_fs.h" -#include "r300_hyperz.h" #include "r300_screen.h" #include "r300_shader_semantics.h" #include "r300_state_inlines.h" @@ -439,7 +438,7 @@ static void r300_update_rs_block(struct r300_context *r300) /* Rasterize texture coordinates. */ for (i = 0; i < ATTR_GENERIC_COUNT && tex_count < 8; i++) { - bool sprite_coord = false; + boolean sprite_coord = false; if (fs_inputs->generic[i] != ATTR_UNUSED) { sprite_coord = !!(r300->sprite_coord_enable & (1 << i)); @@ -489,7 +488,8 @@ static void r300_update_rs_block(struct r300_context *r300) for (; i < ATTR_GENERIC_COUNT; i++) { if (fs_inputs->generic[i] != ATTR_UNUSED) { fprintf(stderr, "r300: ERROR: FS input generic %i unassigned, " - "not enough hardware slots.\n", i); + "not enough hardware slots (it's not a bug, do not " + "report it).\n", i); } } @@ -524,7 +524,8 @@ static void r300_update_rs_block(struct r300_context *r300) DBG(r300, DBG_RS, "r300: FS input fog unassigned.\n"); } else { fprintf(stderr, "r300: ERROR: FS input fog unassigned, " - "not enough hardware slots.\n"); + "not enough hardware slots. (it's not a bug, " + "do not report it)\n"); } } } @@ -551,7 +552,8 @@ static void r300_update_rs_block(struct r300_context *r300) } else { if (fs_inputs->wpos != ATTR_UNUSED && tex_count >= 8) { fprintf(stderr, "r300: ERROR: FS input WPOS unassigned, " - "not enough hardware slots.\n"); + "not enough hardware slots. (it's not a bug, do not " + "report it)\n"); } } @@ -590,6 +592,13 @@ static void r300_update_rs_block(struct r300_context *r300) } } +static void rgba_to_bgra(float color[4]) +{ + float x = color[0]; + color[0] = color[2]; + color[2] = x; +} + static uint32_t r300_get_border_color(enum pipe_format format, const float border[4], boolean is_r500) @@ -623,13 +632,13 @@ static uint32_t r300_get_border_color(enum pipe_format format, for (i = 0; i < 4; i++) { switch (desc->swizzle[i]) { case UTIL_FORMAT_SWIZZLE_X: - border_swizzled[2] = border[i]; + border_swizzled[0] = border[i]; break; case UTIL_FORMAT_SWIZZLE_Y: border_swizzled[1] = border[i]; break; case UTIL_FORMAT_SWIZZLE_Z: - border_swizzled[0] = border[i]; + border_swizzled[2] = border[i]; break; case UTIL_FORMAT_SWIZZLE_W: border_swizzled[3] = border[i]; @@ -639,20 +648,48 @@ static uint32_t r300_get_border_color(enum pipe_format format, /* Compressed formats. */ if (util_format_is_compressed(format)) { - util_pack_color(border_swizzled, PIPE_FORMAT_R8G8B8A8_UNORM, &uc); - return uc.ui; + switch (format) { + case PIPE_FORMAT_RGTC1_SNORM: + case PIPE_FORMAT_LATC1_SNORM: + border_swizzled[0] = border_swizzled[0] < 0 ? + border_swizzled[0]*0.5+1 : + border_swizzled[0]*0.5; + /* Pass through. */ + + case PIPE_FORMAT_RGTC1_UNORM: + case PIPE_FORMAT_LATC1_UNORM: + /* Add 1/32 to round the border color instead of truncating. */ + /* The Y component is used for the border color. */ + border_swizzled[1] = border_swizzled[0] + 1.0f/32; + util_pack_color(border_swizzled, PIPE_FORMAT_B4G4R4A4_UNORM, &uc); + return uc.ui; + case PIPE_FORMAT_RGTC2_SNORM: + case PIPE_FORMAT_LATC2_SNORM: + util_pack_color(border_swizzled, PIPE_FORMAT_R8G8B8A8_SNORM, &uc); + return uc.ui; + case PIPE_FORMAT_RGTC2_UNORM: + case PIPE_FORMAT_LATC2_UNORM: + util_pack_color(border_swizzled, PIPE_FORMAT_R8G8B8A8_UNORM, &uc); + return uc.ui; + default: + util_pack_color(border_swizzled, PIPE_FORMAT_B8G8R8A8_UNORM, &uc); + return uc.ui; + } } switch (desc->channel[0].size) { case 2: + rgba_to_bgra(border_swizzled); util_pack_color(border_swizzled, PIPE_FORMAT_B2G3R3_UNORM, &uc); break; case 4: + rgba_to_bgra(border_swizzled); util_pack_color(border_swizzled, PIPE_FORMAT_B4G4R4A4_UNORM, &uc); break; case 5: + rgba_to_bgra(border_swizzled); if (desc->channel[1].size == 5) { util_pack_color(border_swizzled, PIPE_FORMAT_B5G5R5A1_UNORM, &uc); } else if (desc->channel[1].size == 6) { @@ -664,19 +701,39 @@ static uint32_t r300_get_border_color(enum pipe_format format, default: case 8: - util_pack_color(border_swizzled, PIPE_FORMAT_B8G8R8A8_UNORM, &uc); + if (desc->channel[0].type == UTIL_FORMAT_TYPE_SIGNED) + util_pack_color(border_swizzled, PIPE_FORMAT_R8G8B8A8_SNORM, &uc); + else + util_pack_color(border_swizzled, PIPE_FORMAT_R8G8B8A8_UNORM, &uc); break; case 10: - util_pack_color(border_swizzled, PIPE_FORMAT_B10G10R10A2_UNORM, &uc); + util_pack_color(border_swizzled, PIPE_FORMAT_R10G10B10A2_UNORM, &uc); break; case 16: if (desc->nr_channels <= 2) { - border_swizzled[0] = border_swizzled[2]; - util_pack_color(border_swizzled, PIPE_FORMAT_R16G16_UNORM, &uc); + if (desc->channel[0].type == UTIL_FORMAT_TYPE_FLOAT) { + util_pack_color(border_swizzled, PIPE_FORMAT_R16G16_FLOAT, &uc); + } else if (desc->channel[0].type == UTIL_FORMAT_TYPE_SIGNED) { + util_pack_color(border_swizzled, PIPE_FORMAT_R16G16_SNORM, &uc); + } else { + util_pack_color(border_swizzled, PIPE_FORMAT_R16G16_UNORM, &uc); + } } else { - util_pack_color(border_swizzled, PIPE_FORMAT_B8G8R8A8_UNORM, &uc); + if (desc->channel[0].type == UTIL_FORMAT_TYPE_SIGNED) { + util_pack_color(border_swizzled, PIPE_FORMAT_R8G8B8A8_SNORM, &uc); + } else { + util_pack_color(border_swizzled, PIPE_FORMAT_R8G8B8A8_UNORM, &uc); + } + } + break; + + case 32: + if (desc->nr_channels == 1) { + util_pack_color(border_swizzled, PIPE_FORMAT_R32_FLOAT, &uc); + } else { + util_pack_color(border_swizzled, PIPE_FORMAT_R8G8B8A8_UNORM, &uc); } break; } @@ -691,10 +748,11 @@ static void r300_merge_textures_and_samplers(struct r300_context* r300) struct r300_texture_sampler_state *texstate; struct r300_sampler_state *sampler; struct r300_sampler_view *view; - struct r300_texture *tex; - unsigned min_level, max_level, i, j, size; + struct r300_resource *tex; + unsigned base_level, min_level, level_count, i, j, size; unsigned count = MIN2(state->sampler_view_count, state->sampler_state_count); + boolean has_us_format = r300->screen->caps.has_us_format; /* The KIL opcode fix, see below. */ if (!count && !r300->screen->caps.is_r500) @@ -709,7 +767,7 @@ static void r300_merge_textures_and_samplers(struct r300_context* r300) state->tx_enable |= 1 << i; view = state->sampler_views[i]; - tex = r300_texture(view->base.texture); + tex = r300_resource(view->base.texture); sampler = state->sampler_states[i]; texstate = &state->regs[i]; @@ -724,33 +782,44 @@ static void r300_merge_textures_and_samplers(struct r300_context* r300) r300->screen->caps.is_r500); /* determine min/max levels */ - max_level = MIN3(sampler->max_lod + view->base.u.tex.first_level, - tex->desc.b.b.last_level, view->base.u.tex.last_level); - min_level = MIN2(sampler->min_lod + view->base.u.tex.first_level, - max_level); - - if (tex->desc.is_npot && min_level > 0) { - /* Even though we do not implement mipmapping for NPOT - * textures, we should at least honor the minimum level - * which is allowed to be displayed. We do this by setting up - * an i-th mipmap level as the zero level. */ - r300_texture_setup_format_state(r300->screen, &tex->desc, - min_level, + base_level = view->base.u.tex.first_level; + min_level = sampler->min_lod; + level_count = MIN3(sampler->max_lod, + tex->b.b.b.last_level - base_level, + view->base.u.tex.last_level - base_level); + + if (base_level + min_level) { + unsigned offset; + + if (tex->tex.is_npot) { + /* Even though we do not implement mipmapping for NPOT + * textures, we should at least honor the minimum level + * which is allowed to be displayed. We do this by setting up + * an i-th mipmap level as the zero level. */ + base_level += min_level; + } + offset = tex->tex_offset + + tex->tex.offset_in_bytes[base_level]; + + r300_texture_setup_format_state(r300->screen, tex, + base_level, &texstate->format); - texstate->format.tile_config |= - tex->desc.offset_in_bytes[min_level] & 0xffffffe0; - assert((tex->desc.offset_in_bytes[min_level] & 0x1f) == 0); + texstate->format.tile_config |= offset & 0xffffffe0; + assert((offset & 0x1f) == 0); + } else { + texstate->format.tile_config |= tex->tex_offset & 0xffffffe0; + assert((tex->tex_offset & 0x1f) == 0); } /* Assign a texture cache region. */ texstate->format.format1 |= view->texcache_region; /* Depth textures are kinda special. */ - if (util_format_is_depth_or_stencil(tex->desc.b.b.format)) { + if (util_format_is_depth_or_stencil(tex->b.b.b.format)) { unsigned char depth_swizzle[4]; if (!r300->screen->caps.is_r500 && - util_format_get_blocksizebits(tex->desc.b.b.format) == 32) { + util_format_get_blocksizebits(tex->b.b.b.format) == 32) { /* X24x8 is sampled as Y16X16 on r3xx-r4xx. * The depth here is at the Y component. */ for (j = 0; j < 4; j++) @@ -775,17 +844,23 @@ static void r300_merge_textures_and_samplers(struct r300_context* r300) } if (r300->screen->caps.dxtc_swizzle && - util_format_is_compressed(tex->desc.b.b.format)) { + util_format_is_compressed(tex->b.b.b.format)) { texstate->filter1 |= R400_DXTC_SWIZZLE_ENABLE; } /* to emulate 1D textures through 2D ones correctly */ - if (tex->desc.b.b.target == PIPE_TEXTURE_1D) { + if (tex->b.b.b.target == PIPE_TEXTURE_1D) { texstate->filter0 &= ~R300_TX_WRAP_T_MASK; texstate->filter0 |= R300_TX_WRAP_T(R300_TX_CLAMP_TO_EDGE); } - if (tex->desc.is_npot) { + /* The hardware doesn't like CLAMP and CLAMP_TO_BORDER + * for the 3rd coordinate if the texture isn't 3D. */ + if (tex->b.b.b.target != PIPE_TEXTURE_3D) { + texstate->filter0 &= ~R300_TX_WRAP_R_MASK; + } + + if (tex->tex.is_npot) { /* NPOT textures don't support mip filter, unfortunately. * This prevents incorrect rendering. */ texstate->filter0 &= ~R300_TX_MIN_FILTER_MIP_MASK; @@ -810,13 +885,39 @@ static void r300_merge_textures_and_samplers(struct r300_context* r300) } } else { /* the MAX_MIP level is the largest (finest) one */ - texstate->format.format0 |= R300_TX_NUM_LEVELS(max_level); + texstate->format.format0 |= R300_TX_NUM_LEVELS(level_count); texstate->filter0 |= R300_TX_MAX_MIP_LEVEL(min_level); } + /* Float textures only support nearest and mip-nearest filtering. */ + if (util_format_is_float(tex->b.b.b.format)) { + /* No MAG linear filtering. */ + if ((texstate->filter0 & R300_TX_MAG_FILTER_MASK) == + R300_TX_MAG_FILTER_LINEAR) { + texstate->filter0 &= ~R300_TX_MAG_FILTER_MASK; + texstate->filter0 |= R300_TX_MAG_FILTER_NEAREST; + } + /* No MIN linear filtering. */ + if ((texstate->filter0 & R300_TX_MIN_FILTER_MASK) == + R300_TX_MIN_FILTER_LINEAR) { + texstate->filter0 &= ~R300_TX_MIN_FILTER_MASK; + texstate->filter0 |= R300_TX_MIN_FILTER_NEAREST; + } + /* No mipmap linear filtering. */ + if ((texstate->filter0 & R300_TX_MIN_FILTER_MIP_MASK) == + R300_TX_MIN_FILTER_MIP_LINEAR) { + texstate->filter0 &= ~R300_TX_MIN_FILTER_MIP_MASK; + texstate->filter0 |= R300_TX_MIN_FILTER_MIP_NEAREST; + } + /* No anisotropic filtering. */ + texstate->filter0 &= ~R300_TX_MAX_ANISO_MASK; + texstate->filter1 &= ~R500_TX_MAX_ANISO_MASK; + texstate->filter1 &= ~R500_TX_ANISO_HIGH_QUALITY; + } + texstate->filter0 |= i << 28; - size += 16; + size += 16 + (has_us_format ? 2 : 0); state->count = i+1; } else { /* For the KIL opcode to work on r3xx-r4xx, the texture unit @@ -845,7 +946,7 @@ static void r300_merge_textures_and_samplers(struct r300_context* r300) texstate->border_color = 0; texstate->filter0 |= i << 28; - size += 16; + size += 16 + (has_us_format ? 2 : 0); state->count = i+1; } } @@ -854,22 +955,75 @@ static void r300_merge_textures_and_samplers(struct r300_context* r300) r300->textures_state.size = size; /* Pick a fragment shader based on either the texture compare state - * or the uses_pitch flag. */ - if (r300->fs.state && count) { - if (r300_pick_fragment_shader(r300)) { + * or the uses_pitch flag or some other external state. */ + if (count && + r300->fs_status == FRAGMENT_SHADER_VALID) { + r300->fs_status = FRAGMENT_SHADER_MAYBE_DIRTY; + } +} + +static void r300_decompress_depth_textures(struct r300_context *r300) +{ + struct r300_textures_state *state = + (struct r300_textures_state*)r300->textures_state.state; + struct pipe_resource *tex; + unsigned count = MIN2(state->sampler_view_count, + state->sampler_state_count); + unsigned i; + + if (!r300->locked_zbuffer) { + return; + } + + for (i = 0; i < count; i++) { + if (state->sampler_views[i] && state->sampler_states[i]) { + tex = state->sampler_views[i]->base.texture; + + if (tex == r300->locked_zbuffer->texture) { + r300_decompress_zmask_locked(r300); + return; + } + } + } +} + +static void r300_validate_fragment_shader(struct r300_context *r300) +{ + struct pipe_framebuffer_state *fb = r300->fb_state.state; + + if (r300->fs.state && r300->fs_status != FRAGMENT_SHADER_VALID) { + /* Pick the fragment shader based on external states. + * Then mark the state dirty if the fragment shader is either dirty + * or the function r300_pick_fragment_shader changed the shader. */ + if (r300_pick_fragment_shader(r300) || + r300->fs_status == FRAGMENT_SHADER_DIRTY) { + /* Mark the state atom as dirty. */ r300_mark_fs_code_dirty(r300); + + /* Does Multiwrite need to be changed? */ + if (fb->nr_cbufs > 1) { + boolean new_multiwrite = + r300_fragment_shader_writes_all(r300_fs(r300)); + + if (r300->fb_multiwrite != new_multiwrite) { + r300->fb_multiwrite = new_multiwrite; + r300_mark_fb_state_dirty(r300, R300_CHANGED_MULTIWRITE); + } + } } + r300->fs_status = FRAGMENT_SHADER_VALID; } } void r300_update_derived_state(struct r300_context* r300) { - r300_flush_depth_textures(r300); - if (r300->textures_state.dirty) { + r300_decompress_depth_textures(r300); r300_merge_textures_and_samplers(r300); } + r300_validate_fragment_shader(r300); + if (r300->rs_block_state.dirty) { r300_update_rs_block(r300);