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=3aa8deb63c88fb0f4e31bbd517679c63c8aa09ab;hpb=bcc13b74443137043e8a34f8cb64a5add0d8af93;p=mesa.git diff --git a/src/gallium/drivers/r300/r300_state_derived.c b/src/gallium/drivers/r300/r300_state_derived.c index 3aa8deb63c8..1013557bde9 100644 --- a/src/gallium/drivers/r300/r300_state_derived.c +++ b/src/gallium/drivers/r300/r300_state_derived.c @@ -25,13 +25,12 @@ #include "util/u_math.h" #include "util/u_memory.h" +#include "util/u_pack_color.h" #include "r300_context.h" #include "r300_fs.h" -#include "r300_hyperz.h" #include "r300_screen.h" #include "r300_shader_semantics.h" -#include "r300_state_derived.h" #include "r300_state_inlines.h" #include "r300_texture.h" #include "r300_vs.h" @@ -46,6 +45,11 @@ enum r300_rs_swizzle { SWIZ_0001, }; +enum r300_rs_col_write_type { + WRITE_COLOR = 0, + WRITE_FACE +}; + static void r300_draw_emit_attrib(struct r300_context* r300, enum attrib_emit emit, enum interp_mode interp, @@ -102,7 +106,8 @@ static void r300_draw_emit_all_attribs(struct r300_context* r300) * they won't be rasterized. */ gen_count = 0; for (i = 0; i < ATTR_GENERIC_COUNT && gen_count < 8; i++) { - if (vs_outputs->generic[i] != ATTR_UNUSED) { + if (vs_outputs->generic[i] != ATTR_UNUSED && + !(r300->sprite_coord_enable & (1 << i))) { r300_draw_emit_attrib(r300, EMIT_4F, INTERP_PERSPECTIVE, vs_outputs->generic[i]); gen_count++; @@ -118,7 +123,7 @@ static void r300_draw_emit_all_attribs(struct r300_context* r300) /* WPOS. */ if (r300_fs(r300)->shader->inputs.wpos != ATTR_UNUSED && gen_count < 8) { - DBG(r300, DBG_DRAW, "draw_emit_attrib: WPOS, index: %i\n", + DBG(r300, DBG_SWTCL, "draw_emit_attrib: WPOS, index: %i\n", vs_outputs->wpos); r300_draw_emit_attrib(r300, EMIT_4F, INTERP_PERSPECTIVE, vs_outputs->wpos); @@ -140,18 +145,19 @@ static void r300_swtcl_vertex_psc(struct r300_context *r300) /* 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); + DBG(r300, DBG_SWTCL, "r300: attrib count: %d\n", attrib_count); for (i = 0; i < attrib_count; i++) { - DBG(r300, DBG_DRAW, "r300: attrib: index %d, interp %d, emit %d," - " vs_output_tab %d\n", vinfo->attrib[i].src_index, - vinfo->attrib[i].interp_mode, vinfo->attrib[i].emit, - vs_output_tab[i]); - - /* Make sure we have a proper destination for our attribute. */ - assert(vs_output_tab[i] != -1); + if (vs_output_tab[i] == -1) { + assert(0); + abort(); + } format = draw_translate_vinfo_format(vinfo->attrib[i].emit); + DBG(r300, DBG_SWTCL, + "r300: swtcl_vertex_psc [%i] <- %s\n", + vs_output_tab[i], util_format_short_name(format)); + /* Obtain the type of data in this attribute. */ type = r300_translate_vertex_data_type(format); if (type == R300_INVALID_FORMAT) { @@ -185,7 +191,7 @@ static void r300_swtcl_vertex_psc(struct r300_context *r300) (R300_LAST_VEC << (i & 1 ? 16 : 0)); vstream->count = (i >> 1) + 1; - r300->vertex_stream_state.dirty = TRUE; + r300_mark_atom_dirty(r300, &r300->vertex_stream_state); r300->vertex_stream_state.size = (1 + vstream->count) * 2; } @@ -201,8 +207,10 @@ static void r300_rs_col(struct r300_rs_block* rs, int id, int ptr, rs->inst[id] |= R300_RS_INST_COL_ID(id); } -static void r300_rs_col_write(struct r300_rs_block* rs, int id, int fp_offset) +static void r300_rs_col_write(struct r300_rs_block* rs, int id, int fp_offset, + enum r300_rs_col_write_type type) { + assert(type == WRITE_COLOR); rs->inst[id] |= R300_RS_INST_COL_CN_WRITE | R300_RS_INST_COL_ADDR(fp_offset); } @@ -211,19 +219,19 @@ static void r300_rs_tex(struct r300_rs_block* rs, int id, int ptr, enum r300_rs_swizzle swiz) { if (swiz == SWIZ_X001) { - rs->ip[id] |= R300_RS_TEX_PTR(ptr*4) | + rs->ip[id] |= R300_RS_TEX_PTR(ptr) | R300_RS_SEL_S(R300_RS_SEL_C0) | R300_RS_SEL_T(R300_RS_SEL_K0) | R300_RS_SEL_R(R300_RS_SEL_K0) | R300_RS_SEL_Q(R300_RS_SEL_K1); } else if (swiz == SWIZ_XY01) { - rs->ip[id] |= R300_RS_TEX_PTR(ptr*4) | + rs->ip[id] |= R300_RS_TEX_PTR(ptr) | R300_RS_SEL_S(R300_RS_SEL_C0) | R300_RS_SEL_T(R300_RS_SEL_C1) | R300_RS_SEL_R(R300_RS_SEL_K0) | R300_RS_SEL_Q(R300_RS_SEL_K1); } else { - rs->ip[id] |= R300_RS_TEX_PTR(ptr*4) | + rs->ip[id] |= R300_RS_TEX_PTR(ptr) | R300_RS_SEL_S(R300_RS_SEL_C0) | R300_RS_SEL_T(R300_RS_SEL_C1) | R300_RS_SEL_R(R300_RS_SEL_C2) | @@ -250,32 +258,36 @@ static void r500_rs_col(struct r300_rs_block* rs, int id, int ptr, rs->inst[id] |= R500_RS_INST_COL_ID(id); } -static void r500_rs_col_write(struct r300_rs_block* rs, int id, int fp_offset) +static void r500_rs_col_write(struct r300_rs_block* rs, int id, int fp_offset, + enum r300_rs_col_write_type type) { - rs->inst[id] |= R500_RS_INST_COL_CN_WRITE | - R500_RS_INST_COL_ADDR(fp_offset); + if (type == WRITE_FACE) + rs->inst[id] |= R500_RS_INST_COL_CN_WRITE_BACKFACE | + R500_RS_INST_COL_ADDR(fp_offset); + else + rs->inst[id] |= R500_RS_INST_COL_CN_WRITE | + R500_RS_INST_COL_ADDR(fp_offset); + } static void r500_rs_tex(struct r300_rs_block* rs, int id, int ptr, enum r300_rs_swizzle swiz) { - int rs_tex_comp = ptr*4; - if (swiz == SWIZ_X001) { - rs->ip[id] |= R500_RS_SEL_S(rs_tex_comp) | + rs->ip[id] |= R500_RS_SEL_S(ptr) | R500_RS_SEL_T(R500_RS_IP_PTR_K0) | R500_RS_SEL_R(R500_RS_IP_PTR_K0) | R500_RS_SEL_Q(R500_RS_IP_PTR_K1); } else if (swiz == SWIZ_XY01) { - rs->ip[id] |= R500_RS_SEL_S(rs_tex_comp) | - R500_RS_SEL_T(rs_tex_comp + 1) | + rs->ip[id] |= R500_RS_SEL_S(ptr) | + R500_RS_SEL_T(ptr + 1) | R500_RS_SEL_R(R500_RS_IP_PTR_K0) | R500_RS_SEL_Q(R500_RS_IP_PTR_K1); } else { - rs->ip[id] |= R500_RS_SEL_S(rs_tex_comp) | - R500_RS_SEL_T(rs_tex_comp + 1) | - R500_RS_SEL_R(rs_tex_comp + 2) | - R500_RS_SEL_Q(rs_tex_comp + 3); + rs->ip[id] |= R500_RS_SEL_S(ptr) | + R500_RS_SEL_T(ptr + 1) | + R500_RS_SEL_R(ptr + 2) | + R500_RS_SEL_Q(ptr + 3); } rs->inst[id] |= R500_RS_INST_TEX_ID(id); } @@ -303,14 +315,16 @@ static void r300_update_rs_block(struct r300_context *r300) struct r300_shader_semantics *vs_outputs = &vs->outputs; 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; + 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); + 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); void (*rX00_rs_tex_write)(struct r300_rs_block*, int, int); boolean any_bcolor_used = vs_outputs->bcolor[0] != ATTR_UNUSED || vs_outputs->bcolor[1] != ATTR_UNUSED; int *stream_loc_notcl = r300->stream_loc_notcl; + uint32_t stuffing_enable = 0; if (r300->screen->caps.is_r500) { rX00_rs_col = r500_rs_col; @@ -324,6 +338,11 @@ static void r300_update_rs_block(struct r300_context *r300) rX00_rs_tex_write = r300_rs_tex_write; } + /* 0x5555 copied from classic, which means: + * Select user color 0 for COLOR0 up to COLOR7. + * What the hell does that mean? */ + rs.vap_vtx_state_cntl = 0x5555; + /* The position is always present in VAP. */ rs.vap_vsm_vtx_assm |= R300_INPUT_CNTL_POS; rs.vap_out_vtx_fmt[0] |= R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT; @@ -350,7 +369,7 @@ static void r300_update_rs_block(struct r300_context *r300) /* Write it to the FS input register if it's needed by the FS. */ if (fs_inputs->color[i] != ATTR_UNUSED) { - rX00_rs_col_write(&rs, col_count, fp_offset); + rX00_rs_col_write(&rs, col_count, fp_offset, WRITE_COLOR); fp_offset++; DBG(r300, DBG_RS, @@ -391,15 +410,92 @@ static void r300_update_rs_block(struct r300_context *r300) stream_loc_notcl[loc++] = 6 + tex_count; /* Rasterize it. */ - rX00_rs_tex(&rs, tex_count, tex_count, SWIZ_XYZW); + rX00_rs_tex(&rs, tex_count, tex_ptr, SWIZ_XYZW); tex_count++; + tex_ptr += 4; } } } + /* gl_FrontFacing. + * Note that we can use either the two-sided color selection based on + * the front and back vertex shader colors, or gl_FrontFacing, + * but not both! It locks up otherwise. + * + * In Direct3D 9, the two-sided color selection can be used + * with shaders 2.0 only, while gl_FrontFacing can be used + * with shaders 3.0 only. The hardware apparently hasn't been designed + * to support both at the same time. */ + if (r300->screen->caps.is_r500 && fs_inputs->face != ATTR_UNUSED && + !(any_bcolor_used && r300->two_sided_color)) { + rX00_rs_col(&rs, col_count, col_count, SWIZ_XYZW); + rX00_rs_col_write(&rs, col_count, fp_offset, WRITE_FACE); + fp_offset++; + col_count++; + DBG(r300, DBG_RS, "r300: Rasterized FACE written to FS.\n"); + } else if (fs_inputs->face != ATTR_UNUSED) { + 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 = !!(r300->sprite_coord_enable & (1 << i)); + 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)); + } if (vs_outputs->generic[i] != ATTR_UNUSED || sprite_coord) { if (!sprite_coord) { @@ -407,10 +503,12 @@ static void r300_update_rs_block(struct r300_context *r300) rs.vap_vsm_vtx_assm |= (R300_INPUT_CNTL_TC0 << tex_count); rs.vap_out_vtx_fmt[1] |= (4 << (3 * tex_count)); stream_loc_notcl[loc++] = 6 + tex_count; - } + } else + stuffing_enable |= + R300_GB_TEX_ST << (R300_GB_TEX0_SOURCE_SHIFT + (tex_count*2)); /* Rasterize it. */ - rX00_rs_tex(&rs, tex_count, tex_count, + rX00_rs_tex(&rs, tex_count, tex_ptr, sprite_coord ? SWIZ_XY01 : SWIZ_XYZW); /* Write it to the FS input register if it's needed by the FS. */ @@ -419,14 +517,15 @@ static void r300_update_rs_block(struct r300_context *r300) fp_offset++; DBG(r300, DBG_RS, - "r300: Rasterized generic %i written to FS%s.\n", - i, sprite_coord ? " (sprite coord)" : ""); + "r300: Rasterized generic %i written to FS%s in texcoord %d.\n", + i, sprite_coord ? " (sprite coord)" : "", tex_count); } else { DBG(r300, DBG_RS, "r300: Rasterized generic %i unused%s.\n", i, sprite_coord ? " (sprite coord)" : ""); } tex_count++; + tex_ptr += sprite_coord ? 2 : 4; } else { /* Skip the FS input register, leave it uninitialized. */ /* If we try to set it to (0,0,0,1), it will lock up. */ @@ -439,6 +538,14 @@ 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 (it's not a bug, do not " + "report it).\n", i); + } + } + /* Rasterize fog coordinates. */ if (vs_outputs->fog != ATTR_UNUSED && tex_count < 8) { /* Set up the fog coordinates in VAP. */ @@ -447,7 +554,7 @@ static void r300_update_rs_block(struct r300_context *r300) stream_loc_notcl[loc++] = 6 + tex_count; /* Rasterize it. */ - rX00_rs_tex(&rs, tex_count, tex_count, SWIZ_X001); + rX00_rs_tex(&rs, tex_count, tex_ptr, SWIZ_X001); /* Write it to the FS input register if it's needed by the FS. */ if (fs_inputs->fog != ATTR_UNUSED) { @@ -459,13 +566,20 @@ static void r300_update_rs_block(struct r300_context *r300) DBG(r300, DBG_RS, "r300: Rasterized fog unused.\n"); } tex_count++; + tex_ptr += 4; } 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->fog != ATTR_UNUSED) { fp_offset++; - DBG(r300, DBG_RS, "r300: FS input fog unassigned.\n"); + if (tex_count < 8) { + DBG(r300, DBG_RS, "r300: FS input fog unassigned.\n"); + } else { + fprintf(stderr, "r300: ERROR: FS input fog unassigned, " + "not enough hardware slots. (it's not a bug, " + "do not report it)\n"); + } } } @@ -478,7 +592,7 @@ static void r300_update_rs_block(struct r300_context *r300) stream_loc_notcl[loc++] = 6 + tex_count; /* Rasterize it. */ - rX00_rs_tex(&rs, tex_count, tex_count, SWIZ_XYZW); + rX00_rs_tex(&rs, tex_count, tex_ptr, SWIZ_XYZW); /* Write it to the FS input register. */ rX00_rs_tex_write(&rs, tex_count, fp_offset); @@ -487,6 +601,13 @@ static void r300_update_rs_block(struct r300_context *r300) fp_offset++; tex_count++; + tex_ptr += 4; + } else { + if (fs_inputs->wpos != ATTR_UNUSED && tex_count >= 8) { + fprintf(stderr, "r300: ERROR: FS input WPOS unassigned, " + "not enough hardware slots. (it's not a bug, do not " + "report it)\n"); + } } /* Invalidate the rest of the no-TCL (GA) stream locations. */ @@ -505,19 +626,172 @@ static void r300_update_rs_block(struct r300_context *r300) DBG(r300, DBG_RS, "r300: --- Rasterizer status ---: colors: %i, " "generics: %i.\n", col_count, tex_count); - rs.count = (tex_count*4) | (col_count << R300_IC_COUNT_SHIFT) | + rs.count = MIN2(tex_ptr, 32) | (col_count << R300_IC_COUNT_SHIFT) | R300_HIRES_EN; count = MAX3(col_count, tex_count, 1); rs.inst_count = count - 1; + /* set the GB enable flags */ + if (r300->sprite_coord_enable) + stuffing_enable |= R300_GB_POINT_STUFF_ENABLE; + + rs.gb_enable = stuffing_enable; + /* Now, after all that, see if we actually need to update the state. */ if (memcmp(r300->rs_block_state.state, &rs, sizeof(struct r300_rs_block))) { memcpy(r300->rs_block_state.state, &rs, sizeof(struct r300_rs_block)); - r300->rs_block_state.size = 11 + count*2; + r300->rs_block_state.size = 13 + count*2; } } +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) +{ + const struct util_format_description *desc; + float border_swizzled[4] = {0}; + union util_color uc = {0}; + + desc = util_format_description(format); + + /* Do depth formats first. */ + if (util_format_is_depth_or_stencil(format)) { + switch (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_UINT_Z24_UNORM: + if (is_r500) { + return util_pack_z(PIPE_FORMAT_X8Z24_UNORM, border[0]); + } else { + return util_pack_z(PIPE_FORMAT_Z16_UNORM, border[0]) << 16; + } + default: + assert(0); + return 0; + } + } + + /* Apply inverse swizzle of the format. */ + 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_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; + 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; + } + } + + 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) { + util_pack_color(border_swizzled, PIPE_FORMAT_B5G6R5_UNORM, &uc); + } else { + assert(0); + } + break; + + default: + case 8: + 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: + util_pack_color(border_swizzled, PIPE_FORMAT_R10G10B10A2_UNORM, &uc); + break; + + case 16: + if (desc->nr_channels <= 2) { + 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 { + 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; + } + + return uc.ui; +} + static void r300_merge_textures_and_samplers(struct r300_context* r300) { struct r300_textures_state *state = @@ -525,16 +799,11 @@ 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, 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); - unsigned char depth_swizzle[4] = { - UTIL_FORMAT_SWIZZLE_X, - UTIL_FORMAT_SWIZZLE_X, - UTIL_FORMAT_SWIZZLE_X, - UTIL_FORMAT_SWIZZLE_X - }; + boolean has_us_format = r300->screen->caps.has_us_format; /* The KIL opcode fix, see below. */ if (!count && !r300->screen->caps.is_r500) @@ -549,39 +818,99 @@ 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]; texstate->format = view->format; texstate->filter0 = sampler->filter0; texstate->filter1 = sampler->filter1; - texstate->border_color = sampler->border_color; + + /* Set the border color. */ + texstate->border_color = + r300_get_border_color(view->base.format, + sampler->state.border_color.f, + r300->screen->caps.is_r500); + + /* determine min/max levels */ + 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, + view->base.format, + base_level, + view->width0_override, + view->height0_override, + &texstate->format); + texstate->format.tile_config |= offset & 0xffffffe0; + assert((offset & 0x1f) == 0); + } /* Assign a texture cache region. */ texstate->format.format1 |= view->texcache_region; - /* If compare mode is disabled, the sampler view swizzles - * are stored in the format. - * Otherwise, swizzles must be applied after the compare mode - * in the fragment shader. */ - if (util_format_is_depth_or_stencil(tex->b.b.format)) { + /* Depth textures are kinda special. */ + 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(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++) + depth_swizzle[j] = UTIL_FORMAT_SWIZZLE_Y; + } else { + for (j = 0; j < 4; j++) + depth_swizzle[j] = UTIL_FORMAT_SWIZZLE_X; + } + + /* If compare mode is disabled, sampler view swizzles + * are stored in the format. + * Otherwise, the swizzles must be applied after the compare + * mode in the fragment shader. */ if (sampler->state.compare_mode == PIPE_TEX_COMPARE_NONE) { texstate->format.format1 |= - r300_get_swizzle_combined(depth_swizzle, view->swizzle); + r300_get_swizzle_combined(depth_swizzle, + view->swizzle, FALSE); } else { texstate->format.format1 |= - r300_get_swizzle_combined(depth_swizzle, 0); + r300_get_swizzle_combined(depth_swizzle, 0, FALSE); } } + if (r300->screen->caps.dxtc_swizzle && + 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.target == PIPE_TEXTURE_1D) { texstate->filter0 &= ~R300_TX_WRAP_T_MASK; texstate->filter0 |= R300_TX_WRAP_T(R300_TX_CLAMP_TO_EDGE); } - if (tex->uses_pitch) { + /* 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. */ texstate->filter0 &= ~R300_TX_MIN_FILTER_MIP_MASK; @@ -605,19 +934,40 @@ static void r300_merge_textures_and_samplers(struct r300_context* r300) texstate->filter0 |= R300_TX_WRAP_T(R300_TX_CLAMP_TO_EDGE); } } else { - /* determine min/max levels */ /* the MAX_MIP level is the largest (finest) one */ - max_level = MIN3(sampler->max_lod + view->base.first_level, - tex->b.b.last_level, view->base.last_level); - min_level = MIN2(sampler->min_lod + view->base.first_level, - max_level); - 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(view->base.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 @@ -646,7 +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; } } @@ -655,20 +1005,75 @@ 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) { 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);