freedreno/ir3: debug cleanup
[mesa.git] / src / gallium / drivers / r300 / r300_state_derived.c
index d5fc8ece252923bedc244619e00dc4df1d44c1d0..bae669e78cb16b26f87295b0610788f508832dd8 100644 (file)
 
 #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"
@@ -54,7 +52,6 @@ enum r300_rs_col_write_type {
 
 static void r300_draw_emit_attrib(struct r300_context* r300,
                                   enum attrib_emit emit,
-                                  enum interp_mode interp,
                                   int index)
 {
     struct r300_vertex_shader* vs = r300->vs_state.state;
@@ -64,7 +61,7 @@ static void r300_draw_emit_attrib(struct r300_context* r300,
     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);
+    draw_emit_vertex_attr(&r300->vertex_info, emit, output);
 }
 
 static void r300_draw_emit_all_attribs(struct r300_context* r300)
@@ -75,31 +72,27 @@ static void r300_draw_emit_all_attribs(struct r300_context* r300)
 
     /* Position. */
     if (vs_outputs->pos != ATTR_UNUSED) {
-        r300_draw_emit_attrib(r300, EMIT_4F, INTERP_PERSPECTIVE,
-                              vs_outputs->pos);
+        r300_draw_emit_attrib(r300, EMIT_4F, 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);
+        r300_draw_emit_attrib(r300, EMIT_1F_PSIZE, 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]);
+            r300_draw_emit_attrib(r300, EMIT_4F, vs_outputs->color[i]);
         }
     }
 
     /* Back-face colors. */
     for (i = 0; i < ATTR_COLOR_COUNT; i++) {
         if (vs_outputs->bcolor[i] != ATTR_UNUSED) {
-            r300_draw_emit_attrib(r300, EMIT_4F, INTERP_LINEAR,
-                                  vs_outputs->bcolor[i]);
+            r300_draw_emit_attrib(r300, EMIT_4F, vs_outputs->bcolor[i]);
         }
     }
 
@@ -110,16 +103,14 @@ static void r300_draw_emit_all_attribs(struct r300_context* r300)
     for (i = 0; i < ATTR_GENERIC_COUNT && gen_count < 8; i++) {
         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]);
+            r300_draw_emit_attrib(r300, EMIT_4F, vs_outputs->generic[i]);
             gen_count++;
         }
     }
 
     /* Fog coordinates. */
     if (gen_count < 8 && vs_outputs->fog != ATTR_UNUSED) {
-        r300_draw_emit_attrib(r300, EMIT_4F, INTERP_PERSPECTIVE,
-                              vs_outputs->fog);
+        r300_draw_emit_attrib(r300, EMIT_4F, vs_outputs->fog);
         gen_count++;
     }
 
@@ -127,8 +118,7 @@ static void r300_draw_emit_all_attribs(struct r300_context* r300)
     if (r300_fs(r300)->shader->inputs.wpos != ATTR_UNUSED && gen_count < 8) {
         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);
+        r300_draw_emit_attrib(r300, EMIT_4F, vs_outputs->wpos);
     }
 }
 
@@ -318,6 +308,7 @@ 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);
@@ -438,9 +429,61 @@ 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));
@@ -490,7 +533,8 @@ 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);
         }
     }
 
@@ -525,7 +569,8 @@ 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");
             }
         }
     }
@@ -552,7 +597,8 @@ 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");
         }
     }
 
@@ -591,13 +637,19 @@ 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)
 {
     const struct util_format_description *desc;
     float border_swizzled[4] = {0};
-    unsigned i;
     union util_color uc = {0};
 
     desc = util_format_description(format);
@@ -608,7 +660,7 @@ 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 {
@@ -621,35 +673,58 @@ 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[2] = border[i];
-            break;
-        case UTIL_FORMAT_SWIZZLE_Y:
-            border_swizzled[1] = border[i];
-            break;
-        case UTIL_FORMAT_SWIZZLE_Z:
-            border_swizzled[0] = 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)) {
-        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[0];
+        case PIPE_FORMAT_RGTC2_SNORM:
+        case PIPE_FORMAT_LATC2_SNORM:
+            util_pack_color(border_swizzled, PIPE_FORMAT_R8G8B8A8_SNORM, &uc);
+            return uc.ui[0];
+        case PIPE_FORMAT_RGTC2_UNORM:
+        case PIPE_FORMAT_LATC2_UNORM:
+            util_pack_color(border_swizzled, PIPE_FORMAT_R8G8B8A8_UNORM, &uc);
+            return uc.ui[0];
+        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[0];
+        default:
+            util_pack_color(border_swizzled, PIPE_FORMAT_B8G8R8A8_UNORM, &uc);
+            return uc.ui[0];
+        }
     }
 
     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) {
@@ -661,24 +736,52 @@ 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 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_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;
     }
 
-    return uc.ui;
+    return uc.ui[0];
 }
 
 static void r300_merge_textures_and_samplers(struct r300_context* r300)
@@ -688,10 +791,11 @@ 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)
@@ -706,7 +810,7 @@ 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];
@@ -717,44 +821,54 @@ 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->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.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 |=
-                        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);
             }
 
             /* 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(view->base.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(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;
+                        depth_swizzle[j] = PIPE_SWIZZLE_Y;
                 } else {
                     for (j = 0; j < 4; j++)
-                        depth_swizzle[j] = UTIL_FORMAT_SWIZZLE_X;
+                        depth_swizzle[j] = PIPE_SWIZZLE_X;
                 }
 
                 /* If compare mode is disabled, sampler view swizzles
@@ -772,17 +886,23 @@ 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(view->base.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.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.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;
@@ -807,13 +927,39 @@ 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(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
@@ -842,7 +988,7 @@ 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;
             }
         }
@@ -851,55 +997,75 @@ 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)) {
-            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;
     }
 }
 
-/* We can't use compressed zbuffers as samplers. */
-static void r300_flush_depth_textures(struct r300_context *r300)
+static void r300_decompress_depth_textures(struct r300_context *r300)
 {
     struct r300_textures_state *state =
         (struct r300_textures_state*)r300->textures_state.state;
-    unsigned i, level;
+    struct pipe_resource *tex;
     unsigned count = MIN2(state->sampler_view_count,
                           state->sampler_state_count);
+    unsigned i;
 
-    if (r300->z_decomp_rd)
+    if (!r300->locked_zbuffer) {
         return;
+    }
 
-    for (i = 0; i < count; i++)
+    for (i = 0; i < count; i++) {
         if (state->sampler_views[i] && state->sampler_states[i]) {
-            struct pipe_resource *tex = state->sampler_views[i]->base.texture;
-
-            if (tex->target == PIPE_TEXTURE_3D ||
-                tex->target == PIPE_TEXTURE_CUBE)
-                continue;
-
-            /* Ignore non-depth textures.
-             * Also ignore reinterpreted depth textures, e.g. resource_copy. */
-            if (!util_format_is_depth_or_stencil(tex->format))
-                continue;
-
-            for (level = 0; level <= tex->last_level; level++)
-                if (r300_texture(tex)->zmask_in_use[level]) {
-                    /* We don't handle 3D textures and cubemaps yet. */
-                    r300_flush_depth_stencil(&r300->context, tex, level, 0);
-                 }
+            tex = state->sampler_views[i]->base.texture;
+
+            if (tex == r300->locked_zbuffer->texture) {
+                r300_decompress_zmask_locked(r300);
+                return;
+            }
         }
+    }
 }
 
-void r300_update_derived_state(struct r300_context* r300)
+static void r300_validate_fragment_shader(struct r300_context *r300)
 {
-    r300_flush_depth_textures(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);