X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fgallium%2Fdrivers%2Fr300%2Fr300_state_derived.c;h=1013557bde9727c5295fe7a51d69fd4d397aa763;hb=5f7faff61bc3ebe80f262fac6f58225d8005631b;hp=50cde5528ef2f3ef41c0ff43fdb3dad6eca5a550;hpb=2e6a69939c26ab8cc27bcab29ed64cec5e9428ee;p=mesa.git diff --git a/src/gallium/drivers/r300/r300_state_derived.c b/src/gallium/drivers/r300/r300_state_derived.c index 50cde5528ef..1013557bde9 100644 --- a/src/gallium/drivers/r300/r300_state_derived.c +++ b/src/gallium/drivers/r300/r300_state_derived.c @@ -316,6 +316,7 @@ static void r300_update_rs_block(struct r300_context *r300) struct r300_shader_semantics *fs_inputs = &r300_fs(r300)->shader->inputs; struct r300_rs_block rs = {0}; int i, col_count = 0, tex_count = 0, fp_offset = 0, count, loc = 0, tex_ptr = 0; + int gen_offset = 0; void (*rX00_rs_col)(struct r300_rs_block*, int, int, enum r300_rs_swizzle); void (*rX00_rs_col_write)(struct r300_rs_block*, int, int, enum r300_rs_col_write_type); void (*rX00_rs_tex)(struct r300_rs_block*, int, int, enum r300_rs_swizzle); @@ -436,9 +437,61 @@ static void r300_update_rs_block(struct r300_context *r300) fprintf(stderr, "r300: ERROR: FS input FACE unassigned.\n"); } + /* Re-use color varyings for texcoords if possible. + * + * The colors are interpolated as 20-bit floats (reduced precision), + * Use this hack only if there are too many generic varyings. + * (number of generic varyings + fog + wpos > 8) */ + if (r300->screen->caps.is_r500 && !any_bcolor_used && !r300->flatshade && + fs_inputs->face == ATTR_UNUSED && + vs_outputs->num_generic + (vs_outputs->fog != ATTR_UNUSED) + + (fs_inputs->wpos != ATTR_UNUSED) > 8) { + for (i = 0; i < ATTR_GENERIC_COUNT && col_count < 2; i++) { + /* Cannot use color varyings for sprite coords. */ + if (fs_inputs->generic[i] != ATTR_UNUSED && + (r300->sprite_coord_enable & (1 << i))) { + break; + } + + if (vs_outputs->generic[i] != ATTR_UNUSED) { + /* Set up the color in VAP. */ + rs.vap_vsm_vtx_assm |= R300_INPUT_CNTL_COLOR; + rs.vap_out_vtx_fmt[0] |= + R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT << col_count; + stream_loc_notcl[loc++] = 2 + col_count; + + /* Rasterize it. */ + rX00_rs_col(&rs, col_count, col_count, SWIZ_XYZW); + + /* Write it to the FS input register if it's needed by the FS. */ + if (fs_inputs->generic[i] != ATTR_UNUSED) { + rX00_rs_col_write(&rs, col_count, fp_offset, WRITE_COLOR); + fp_offset++; + + DBG(r300, DBG_RS, + "r300: Rasterized generic %i redirected to color %i and written to FS.\n", + i, col_count); + } else { + DBG(r300, DBG_RS, "r300: Rasterized generic %i redirected to color %i unused.\n", + i, col_count); + } + col_count++; + } else { + /* Skip the FS input register, leave it uninitialized. */ + /* If we try to set it to (0,0,0,1), it will lock up. */ + if (fs_inputs->generic[i] != ATTR_UNUSED) { + fp_offset++; + + DBG(r300, DBG_RS, "r300: FS input generic %i unassigned%s.\n", i); + } + } + } + gen_offset = i; + } + /* Rasterize texture coordinates. */ - for (i = 0; i < ATTR_GENERIC_COUNT && tex_count < 8; i++) { - bool sprite_coord = false; + for (i = gen_offset; i < ATTR_GENERIC_COUNT && tex_count < 8; i++) { + boolean sprite_coord = false; if (fs_inputs->generic[i] != ATTR_UNUSED) { sprite_coord = !!(r300->sprite_coord_enable & (1 << i)); @@ -605,7 +658,6 @@ static uint32_t r300_get_border_color(enum pipe_format format, { const struct util_format_description *desc; float border_swizzled[4] = {0}; - unsigned i; union util_color uc = {0}; desc = util_format_description(format); @@ -616,7 +668,7 @@ static uint32_t r300_get_border_color(enum pipe_format format, case PIPE_FORMAT_Z16_UNORM: return util_pack_z(PIPE_FORMAT_Z16_UNORM, border[0]); case PIPE_FORMAT_X8Z24_UNORM: - case PIPE_FORMAT_S8_USCALED_Z24_UNORM: + case PIPE_FORMAT_S8_UINT_Z24_UNORM: if (is_r500) { return util_pack_z(PIPE_FORMAT_X8Z24_UNORM, border[0]); } else { @@ -629,29 +681,19 @@ static uint32_t r300_get_border_color(enum pipe_format format, } /* Apply inverse swizzle of the format. */ - for (i = 0; i < 4; i++) { - switch (desc->swizzle[i]) { - case UTIL_FORMAT_SWIZZLE_X: - border_swizzled[0] = border[i]; - break; - case UTIL_FORMAT_SWIZZLE_Y: - border_swizzled[1] = border[i]; - break; - case UTIL_FORMAT_SWIZZLE_Z: - border_swizzled[2] = border[i]; - break; - case UTIL_FORMAT_SWIZZLE_W: - border_swizzled[3] = border[i]; - break; - } - } + util_format_unswizzle_4f(border_swizzled, border, desc->swizzle); /* Compressed formats. */ if (util_format_is_compressed(format)) { switch (format) { case PIPE_FORMAT_RGTC1_SNORM: - case PIPE_FORMAT_RGTC1_UNORM: 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. */ @@ -666,6 +708,12 @@ static uint32_t r300_get_border_color(enum pipe_format format, case PIPE_FORMAT_LATC2_UNORM: util_pack_color(border_swizzled, PIPE_FORMAT_R8G8B8A8_UNORM, &uc); return uc.ui; + case PIPE_FORMAT_DXT1_SRGB: + case PIPE_FORMAT_DXT1_SRGBA: + case PIPE_FORMAT_DXT3_SRGBA: + case PIPE_FORMAT_DXT5_SRGBA: + util_pack_color(border_swizzled, PIPE_FORMAT_B8G8R8A8_SRGB, &uc); + return uc.ui; default: util_pack_color(border_swizzled, PIPE_FORMAT_B8G8R8A8_UNORM, &uc); return uc.ui; @@ -696,10 +744,18 @@ static uint32_t r300_get_border_color(enum pipe_format format, default: case 8: - 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); + if (desc->channel[0].type == UTIL_FORMAT_TYPE_SIGNED) { + util_pack_color(border_swizzled, PIPE_FORMAT_R8G8B8A8_SNORM, &uc); + } else if (desc->colorspace == UTIL_FORMAT_COLORSPACE_SRGB) { + if (desc->nr_channels == 2) { + border_swizzled[3] = border_swizzled[1]; + util_pack_color(border_swizzled, PIPE_FORMAT_L8A8_SRGB, &uc); + } else { + util_pack_color(border_swizzled, PIPE_FORMAT_R8G8B8A8_SRGB, &uc); + } + } else { + util_pack_color(border_swizzled, PIPE_FORMAT_R8G8B8A8_UNORM, &uc); + } break; case 10: @@ -736,25 +792,6 @@ static uint32_t r300_get_border_color(enum pipe_format format, return uc.ui; } -static boolean util_format_is_float(enum pipe_format format) -{ - const struct util_format_description *desc = util_format_description(format); - unsigned i; - - if (!format) - return FALSE; - - /* Find the first non-void channel. */ - for (i = 0; i < 4; i++) - if (desc->channel[i].type != UTIL_FORMAT_TYPE_VOID) - break; - - if (i == 4) - return FALSE; - - return desc->channel[i].type == UTIL_FORMAT_TYPE_FLOAT ? TRUE : FALSE; -} - static void r300_merge_textures_and_samplers(struct r300_context* r300) { struct r300_textures_state *state = @@ -763,9 +800,10 @@ static void r300_merge_textures_and_samplers(struct r300_context* r300) struct r300_sampler_state *sampler; struct r300_sampler_view *view; struct r300_resource *tex; - unsigned min_level, max_level, i, j, size; + 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) @@ -791,42 +829,47 @@ static void r300_merge_textures_and_samplers(struct r300_context* r300) /* Set the border color. */ texstate->border_color = r300_get_border_color(view->base.format, - sampler->state.border_color, + sampler->state.border_color.f, r300->screen->caps.is_r500); /* determine min/max levels */ - max_level = MIN3(sampler->max_lod + view->base.u.tex.first_level, - tex->b.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->tex.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 - * the i-th mipmap level as the zero level. */ - unsigned offset = tex->tex_offset + - tex->tex.offset_in_bytes[min_level]; + base_level = view->base.u.tex.first_level; + min_level = sampler->min_lod; + level_count = MIN3(sampler->max_lod, + tex->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_in_bytes[base_level]; r300_texture_setup_format_state(r300->screen, tex, - min_level, + view->base.format, + base_level, + view->width0_override, + view->height0_override, &texstate->format); 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->b.b.b.format)) { + if (util_format_is_depth_or_stencil(view->base.format)) { unsigned char depth_swizzle[4]; if (!r300->screen->caps.is_r500 && - util_format_get_blocksizebits(tex->b.b.b.format) == 32) { + util_format_get_blocksizebits(view->base.format) == 32) { /* X24x8 is sampled as Y16X16 on r3xx-r4xx. * The depth here is at the Y component. */ for (j = 0; j < 4; j++) @@ -851,16 +894,22 @@ static void r300_merge_textures_and_samplers(struct r300_context* r300) } if (r300->screen->caps.dxtc_swizzle && - util_format_is_compressed(tex->b.b.b.format)) { + util_format_is_compressed(view->base.format)) { texstate->filter1 |= R400_DXTC_SWIZZLE_ENABLE; } /* to emulate 1D textures through 2D ones correctly */ - if (tex->b.b.b.target == PIPE_TEXTURE_1D) { + if (tex->b.b.target == PIPE_TEXTURE_1D) { texstate->filter0 &= ~R300_TX_WRAP_T_MASK; texstate->filter0 |= R300_TX_WRAP_T(R300_TX_CLAMP_TO_EDGE); } + /* The hardware doesn't like CLAMP and CLAMP_TO_BORDER + * for the 3rd coordinate if the texture isn't 3D. */ + if (tex->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. */ @@ -886,12 +935,12 @@ 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)) { + if (util_format_is_float(view->base.format)) { /* No MAG linear filtering. */ if ((texstate->filter0 & R300_TX_MAG_FILTER_MASK) == R300_TX_MAG_FILTER_LINEAR) { @@ -918,7 +967,7 @@ static void r300_merge_textures_and_samplers(struct r300_context* r300) 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 @@ -947,7 +996,7 @@ 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; } } @@ -956,11 +1005,10 @@ 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)) { - r300_mark_fs_code_dirty(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; } } @@ -973,7 +1021,7 @@ static void r300_decompress_depth_textures(struct r300_context *r300) state->sampler_state_count); unsigned i; - if (!r300->hyperz_locked || !r300->locked_zbuffer) { + if (!r300->locked_zbuffer) { return; } @@ -989,6 +1037,34 @@ static void r300_decompress_depth_textures(struct r300_context *r300) } } +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) { if (r300->textures_state.dirty) { @@ -996,6 +1072,8 @@ void r300_update_derived_state(struct r300_context* r300) r300_merge_textures_and_samplers(r300); } + r300_validate_fragment_shader(r300); + if (r300->rs_block_state.dirty) { r300_update_rs_block(r300);