r300g: derive user buffer sizes at draw time
[mesa.git] / src / gallium / drivers / r300 / r300_state.c
index d4e20635586076b616b490ef171f460556d76775..b55262a95229396bc83888d3b3d757877cbca786 100644 (file)
@@ -52,7 +52,7 @@
 #define UPDATE_STATE(cso, atom) \
     if (cso != atom.state) { \
         atom.state = cso;    \
-        atom.dirty = TRUE;   \
+        r300_mark_atom_dirty(r300, &(atom));   \
     }
 
 static boolean blend_discard_if_src_alpha_0(unsigned srcRGB, unsigned srcA,
@@ -417,7 +417,7 @@ static void r300_set_blend_color(struct pipe_context* pipe,
         END_CB;
     }
 
-    r300->blend_color_state.dirty = TRUE;
+    r300_mark_atom_dirty(r300, &r300->blend_color_state);
 }
 
 static void r300_set_clip_state(struct pipe_context* pipe,
@@ -446,7 +446,7 @@ static void r300_set_clip_state(struct pipe_context* pipe,
                 (state->depth_clamp ? R300_CLIP_DISABLE : 0));
         END_CB;
 
-        r300->clip_state.dirty = TRUE;
+        r300_mark_atom_dirty(r300, &r300->clip_state);
     } else {
         draw_set_clip_state(r300->draw, state);
     }
@@ -594,7 +594,7 @@ static void r300_bind_dsa_state(struct pipe_context* pipe,
 
     UPDATE_STATE(state, r300->dsa_state);
 
-    r300->hyperz_state.dirty = TRUE; /* Will be updated before the emission. */
+    r300_mark_atom_dirty(r300, &r300->hyperz_state); /* Will be updated before the emission. */
     r300_dsa_inject_stencilref(r300);
 }
 
@@ -613,7 +613,7 @@ static void r300_set_stencil_ref(struct pipe_context* pipe,
     r300->stencil_ref = *sr;
 
     r300_dsa_inject_stencilref(r300);
-    r300->dsa_state.dirty = TRUE;
+    r300_mark_atom_dirty(r300, &r300->dsa_state);
 }
 
 static void r300_tex_set_tiling_flags(struct r300_context *r300,
@@ -686,14 +686,23 @@ void r300_mark_fb_state_dirty(struct r300_context *r300,
     struct pipe_framebuffer_state *state = r300->fb_state.state;
     boolean can_hyperz = r300->rws->get_value(r300->rws, R300_CAN_HYPERZ);
 
-    /* What is marked as dirty depends on the enum r300_fb_state_change. */
-    r300->gpu_flush.dirty = TRUE;
-    r300->fb_state.dirty = TRUE;
-    r300->hyperz_state.dirty = TRUE;
+    r300_mark_atom_dirty(r300, &r300->gpu_flush);
+    r300_mark_atom_dirty(r300, &r300->fb_state);
 
+    /* What is marked as dirty depends on the enum r300_fb_state_change. */
     if (change == R300_CHANGED_FB_STATE) {
-        r300->aa_state.dirty = TRUE;
-        r300->fb_state_pipelined.dirty = TRUE;
+        r300_mark_atom_dirty(r300, &r300->aa_state);
+    }
+
+    if (change == R300_CHANGED_FB_STATE ||
+        change == R300_CHANGED_CBZB_FLAG ||
+        change == R300_CHANGED_ZCLEAR_FLAG) {
+        r300_mark_atom_dirty(r300, &r300->hyperz_state);
+    }
+
+    if (change == R300_CHANGED_FB_STATE ||
+        change == R300_CHANGED_MULTIWRITE) {
+        r300_mark_atom_dirty(r300, &r300->fb_state_pipelined);
     }
 
     /* Now compute the fb_state atom size. */
@@ -738,11 +747,11 @@ static void
 
     /* If nr_cbufs is changed from zero to non-zero or vice versa... */
     if (!!old_state->nr_cbufs != !!state->nr_cbufs) {
-        r300->blend_state.dirty = TRUE;
+        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->dsa_state.dirty = TRUE;
+        r300_mark_atom_dirty(r300, &r300->dsa_state);
     }
 
     /* The tiling flags are dependent on the surface miplevel, unfortunately. */
@@ -751,6 +760,7 @@ static void
     util_copy_framebuffer_state(r300->fb_state.state, state);
 
     r300_mark_fb_state_dirty(r300, R300_CHANGED_FB_STATE);
+    r300->validate_buffers = TRUE;
 
     r300->z_compression = false;
     
@@ -795,7 +805,7 @@ static void
             r300->zbuffer_bpp = zbuffer_bpp;
 
             if (r300->polygon_offset_enabled)
-                r300->rs_state.dirty = TRUE;
+                r300_mark_atom_dirty(r300, &r300->rs_state);
         }
     }
 
@@ -853,9 +863,9 @@ void r300_mark_fs_code_dirty(struct r300_context *r300)
 {
     struct r300_fragment_shader* fs = r300_fs(r300);
 
-    r300->fs.dirty = TRUE;
-    r300->fs_rc_constant_state.dirty = TRUE;
-    r300->fs_constants.dirty = TRUE;
+    r300_mark_atom_dirty(r300, &r300->fs);
+    r300_mark_atom_dirty(r300, &r300->fs_rc_constant_state);
+    r300_mark_atom_dirty(r300, &r300->fs_constants);
     r300->fs.size = fs->shader->cb_code_size;
 
     if (r300->screen->caps.is_r500) {
@@ -875,17 +885,26 @@ 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);
 
-    r300->rs_block_state.dirty = TRUE; /* Will be updated before the emission. */
+    if (fb->nr_cbufs > 1 &&
+        last_multi_write != r300_fragment_shader_writes_all(fs)) {
+        r300_mark_fb_state_dirty(r300, R300_CHANGED_MULTIWRITE);
+    }
+
+    r300_mark_atom_dirty(r300, &r300->rs_block_state); /* Will be updated before the emission. */
 }
 
 /* Delete fragment shader state. */
@@ -1137,7 +1156,7 @@ static void r300_bind_rs_state(struct pipe_context* pipe, void* state)
 
     if (last_sprite_coord_enable != r300->sprite_coord_enable ||
         last_two_sided_color != r300->two_sided_color) {
-        r300->rs_block_state.dirty = TRUE;
+        r300_mark_atom_dirty(r300, &r300->rs_block_state);
     }
 }
 
@@ -1235,7 +1254,7 @@ static void r300_bind_sampler_states(struct pipe_context* pipe,
     memcpy(state->sampler_states, states, sizeof(void*) * count);
     state->sampler_state_count = count;
 
-    r300->textures_state.dirty = TRUE;
+    r300_mark_atom_dirty(r300, &r300->textures_state);
 }
 
 static void r300_lacks_vertex_textures(struct pipe_context* pipe,
@@ -1297,29 +1316,27 @@ static void r300_set_fragment_sampler_views(struct pipe_context* pipe,
     }
 
     for (i = 0; i < count; i++) {
-        if (&state->sampler_views[i]->base != views[i]) {
-            pipe_sampler_view_reference(
-                    (struct pipe_sampler_view**)&state->sampler_views[i],
-                    views[i]);
+        pipe_sampler_view_reference(
+                (struct pipe_sampler_view**)&state->sampler_views[i],
+                views[i]);
 
-            if (!views[i]) {
-                continue;
-            }
+        if (!views[i]) {
+            continue;
+        }
 
-            /* A new sampler view (= texture)... */
-            dirty_tex = TRUE;
+        /* A new sampler view (= texture)... */
+        dirty_tex = TRUE;
 
-            /* Set the texrect factor in the fragment shader.
+        /* Set the texrect factor in the fragment shader.
              * Needed for RECT and NPOT fallback. */
-            texture = r300_texture(views[i]->texture);
-            if (texture->desc.is_npot) {
-                r300->fs_rc_constant_state.dirty = TRUE;
-            }
+        texture = r300_texture(views[i]->texture);
+        if (texture->desc.is_npot) {
+            r300_mark_atom_dirty(r300, &r300->fs_rc_constant_state);
+        }
 
-            state->sampler_views[i]->texcache_region =
+        state->sampler_views[i]->texcache_region =
                 r300_assign_texture_cache_region(view_index, real_num_views);
-            view_index++;
-        }
+        view_index++;
     }
 
     for (i = count; i < tex_units; i++) {
@@ -1332,10 +1349,11 @@ static void r300_set_fragment_sampler_views(struct pipe_context* pipe,
 
     state->sampler_view_count = count;
 
-    r300->textures_state.dirty = TRUE;
+    r300_mark_atom_dirty(r300, &r300->textures_state);
+    r300->validate_buffers = TRUE;
 
     if (dirty_tex) {
-        r300->texture_cache_inval.dirty = TRUE;
+        r300_mark_atom_dirty(r300, &r300->texture_cache_inval);
     }
 }
 
@@ -1390,7 +1408,7 @@ static void r300_set_scissor_state(struct pipe_context* pipe,
     memcpy(r300->scissor_state.state, state,
         sizeof(struct pipe_scissor_state));
 
-    r300->scissor_state.dirty = TRUE;
+    r300_mark_atom_dirty(r300, &r300->scissor_state);
 }
 
 static void r300_set_viewport_state(struct pipe_context* pipe,
@@ -1436,9 +1454,9 @@ static void r300_set_viewport_state(struct pipe_context* pipe,
         viewport->vte_control |= R300_VPORT_Z_OFFSET_ENA;
     }
 
-    r300->viewport_state.dirty = TRUE;
+    r300_mark_atom_dirty(r300, &r300->viewport_state);
     if (r300->fs.state && r300_fs(r300)->shader->inputs.wpos != ATTR_UNUSED) {
-        r300->fs_rc_constant_state.dirty = TRUE;
+        r300_mark_atom_dirty(r300, &r300->fs_rc_constant_state);
     }
 }
 
@@ -1447,15 +1465,15 @@ static void r300_set_vertex_buffers(struct pipe_context* pipe,
                                     const struct pipe_vertex_buffer* buffers)
 {
     struct r300_context* r300 = r300_context(pipe);
-    struct pipe_vertex_buffer *vbo;
+    const struct pipe_vertex_buffer *vbo;
     unsigned i, max_index = (1 << 24) - 1;
     boolean any_user_buffer = FALSE;
+    boolean any_nonuser_buffer = FALSE;
     struct pipe_vertex_buffer dummy_vb = {0};
 
     /* There must be at least one vertex buffer set, otherwise it locks up. */
     if (!count) {
         dummy_vb.buffer = r300->dummy_vb;
-        dummy_vb.max_index = r300->dummy_vb->width0 / 4;
         buffers = &dummy_vb;
         count = 1;
     }
@@ -1482,34 +1500,41 @@ static void r300_set_vertex_buffers(struct pipe_context* pipe,
         }
 
         for (i = 0; i < count; i++) {
-            /* Why, yes, I AM casting away constness. How did you know? */
-            vbo = (struct pipe_vertex_buffer*)&buffers[i];
+            vbo = &buffers[i];
 
             /* Skip NULL buffers */
-            if (!buffers[i].buffer) {
+            if (!vbo->buffer) {
                 continue;
             }
 
-            if (r300_buffer_is_user_buffer(vbo->buffer)) {
+            /* User buffers have no info about maximum index,
+             * we will have to compute it in draw_vbo. */
+            if (r300_is_user_buffer(vbo->buffer)) {
                 any_user_buffer = TRUE;
+                continue;
             }
+            any_nonuser_buffer = TRUE;
 
-            if (vbo->max_index == ~0) {
-                /* if no VBO stride then only one vertex value so max index is 1 */
-                /* should think about converting to VS constants like svga does */
-                if (!vbo->stride)
-                    vbo->max_index = 1;
-                else
-                    vbo->max_index =
-                             (vbo->buffer->width0 - vbo->buffer_offset) / vbo->stride;
-            }
+            /* The stride of zero means we will be fetching only the first
+             * vertex, so don't care about max_index. */
+            if (!vbo->stride)
+                continue;
 
-            max_index = MIN2(vbo->max_index, max_index);
+            /* Update the maximum index. */
+            {
+                unsigned vbo_max_index =
+                      (vbo->buffer->width0 - vbo->buffer_offset) / vbo->stride;
+                max_index = MIN2(max_index, vbo_max_index);
+            }
         }
 
         r300->any_user_vbs = any_user_buffer;
         r300->vertex_buffer_max_index = max_index;
-
+        r300->vertex_arrays_dirty = TRUE;
+        if (any_nonuser_buffer)
+            r300->validate_buffers = TRUE;
+        if (!any_user_buffer)
+            r300->upload_vb_validated = FALSE;
     } else {
         /* SW TCL. */
         draw_set_vertex_buffers(r300->draw, count, buffers);
@@ -1517,16 +1542,25 @@ static void r300_set_vertex_buffers(struct pipe_context* pipe,
 
     /* Common code. */
     for (i = 0; i < count; i++) {
+        vbo = &buffers[i];
+
         /* Reference our buffer. */
-        pipe_resource_reference(&r300->vertex_buffer[i].buffer, buffers[i].buffer);
+        pipe_resource_reference(&r300->vertex_buffer[i].buffer, vbo->buffer);
+        if (vbo->buffer && r300_is_user_buffer(vbo->buffer)) {
+            pipe_resource_reference(&r300->valid_vertex_buffer[i], NULL);
+        } else {
+            pipe_resource_reference(&r300->valid_vertex_buffer[i], vbo->buffer);
+        }
     }
     for (; i < r300->vertex_buffer_count; i++) {
         /* Dereference any old buffers. */
         pipe_resource_reference(&r300->vertex_buffer[i].buffer, NULL);
+        pipe_resource_reference(&r300->valid_vertex_buffer[i], NULL);
     }
 
     memcpy(r300->vertex_buffer, buffers,
-        sizeof(struct pipe_vertex_buffer) * count);
+           sizeof(struct pipe_vertex_buffer) * count);
+
     r300->vertex_buffer_count = count;
 }
 
@@ -1535,20 +1569,23 @@ static void r300_set_index_buffer(struct pipe_context* pipe,
 {
     struct r300_context* r300 = r300_context(pipe);
 
-    if (ib) {
+    if (ib && ib->buffer) {
         pipe_resource_reference(&r300->index_buffer.buffer, ib->buffer);
         memcpy(&r300->index_buffer, ib, sizeof(r300->index_buffer));
+
+        if (r300->screen->caps.has_tcl &&
+            !r300_is_user_buffer(ib->buffer)) {
+            r300->validate_buffers = TRUE;
+            r300->upload_ib_validated = FALSE;
+        }
     }
     else {
         pipe_resource_reference(&r300->index_buffer.buffer, NULL);
         memset(&r300->index_buffer, 0, sizeof(r300->index_buffer));
     }
 
-    if (r300->screen->caps.has_tcl) {
-       /* TODO make this more like a state */
-    }
-    else {
-       draw_set_index_buffer(r300->draw, ib);
+    if (!r300->screen->caps.has_tcl) {
+        draw_set_index_buffer(r300->draw, ib);
     }
 }
 
@@ -1716,6 +1753,7 @@ static void r300_bind_vertex_elements_state(struct pipe_context *pipe,
 
     UPDATE_STATE(&velems->vertex_stream, r300->vertex_stream_state);
     r300->vertex_stream_state.size = (1 + velems->vertex_stream.count) * 2;
+    r300->vertex_arrays_dirty = TRUE;
 }
 
 static void r300_delete_vertex_elements_state(struct pipe_context *pipe, void *state)
@@ -1758,27 +1796,25 @@ static void r300_bind_vs_state(struct pipe_context* pipe, void* shader)
     r300->vs_state.state = vs;
 
     /* The majority of the RS block bits is dependent on the vertex shader. */
-    r300->rs_block_state.dirty = TRUE; /* Will be updated before the emission. */
+    r300_mark_atom_dirty(r300, &r300->rs_block_state); /* Will be updated before the emission. */
 
     if (r300->screen->caps.has_tcl) {
         unsigned fc_op_dwords = r300->screen->caps.is_r500 ? 3 : 2;
-        r300->vs_state.dirty = TRUE;
+        r300_mark_atom_dirty(r300, &r300->vs_state);
         r300->vs_state.size =
                 vs->code.length + 9 +
-                (vs->immediates_count ? vs->immediates_count * 4 + 3 : 0) +
         (vs->code.num_fc_ops ? vs->code.num_fc_ops * fc_op_dwords + 4 : 0);
 
-        if (vs->externals_count) {
-            r300->vs_constants.dirty = TRUE;
-            r300->vs_constants.size = vs->externals_count * 4 + 3;
-        } else {
-            r300->vs_constants.size = 0;
-        }
+        r300_mark_atom_dirty(r300, &r300->vs_constants);
+        r300->vs_constants.size =
+                2 +
+                (vs->externals_count ? vs->externals_count * 4 + 3 : 0) +
+                (vs->immediates_count ? vs->immediates_count * 4 + 3 : 0);
 
         ((struct r300_constant_buffer*)r300->vs_constants.state)->remap_table =
                 vs->code.constants_remap_table;
 
-        r300->pvs_flush.dirty = TRUE;
+        r300_mark_atom_dirty(r300, &r300->pvs_flush);
     } else {
         draw_bind_vertex_shader(r300->draw,
                 (struct draw_vertex_shader*)vs->draw_vs);
@@ -1809,6 +1845,7 @@ static void r300_set_constant_buffer(struct pipe_context *pipe,
 {
     struct r300_context* r300 = r300_context(pipe);
     struct r300_constant_buffer *cbuf;
+    struct r300_buffer *rbuf = r300_buffer(buf);
     uint32_t *mapped;
 
     switch (shader) {
@@ -1819,33 +1856,49 @@ static void r300_set_constant_buffer(struct pipe_context *pipe,
             cbuf = (struct r300_constant_buffer*)r300->fs_constants.state;
             break;
         default:
-            assert(0);
             return;
     }
 
-    if (buf == NULL || buf->width0 == 0 ||
-        (mapped = r300_buffer(buf)->constant_buffer) == NULL) {
+    if (buf == NULL || buf->width0 == 0)
+        return;
+
+    if (rbuf->user_buffer)
+        mapped = (uint32_t*)rbuf->user_buffer;
+    else if (rbuf->constant_buffer)
+        mapped = (uint32_t*)rbuf->constant_buffer;
+    else
         return;
-    }
 
     if (shader == PIPE_SHADER_FRAGMENT ||
         (shader == PIPE_SHADER_VERTEX && r300->screen->caps.has_tcl)) {
         assert((buf->width0 % (4 * sizeof(float))) == 0);
-        cbuf->ptr = mapped + index*4;
+        cbuf->ptr = mapped;
     }
 
     if (shader == PIPE_SHADER_VERTEX) {
         if (r300->screen->caps.has_tcl) {
-            if (r300->vs_constants.size) {
-                r300->vs_constants.dirty = TRUE;
+            struct r300_vertex_shader *vs =
+                    (struct r300_vertex_shader*)r300->vs_state.state;
+
+            if (!vs) {
+                cbuf->buffer_base = 0;
+                return;
+            }
+
+            cbuf->buffer_base = r300->vs_const_base;
+            r300->vs_const_base += vs->code.constants.Count;
+            if (r300->vs_const_base > R500_MAX_PVS_CONST_VECS) {
+                r300->vs_const_base = vs->code.constants.Count;
+                cbuf->buffer_base = 0;
+                r300_mark_atom_dirty(r300, &r300->pvs_flush);
             }
-            r300->pvs_flush.dirty = TRUE;
+            r300_mark_atom_dirty(r300, &r300->vs_constants);
         } else if (r300->draw) {
             draw_set_mapped_constant_buffer(r300->draw, PIPE_SHADER_VERTEX,
                 0, mapped, buf->width0);
         }
     } else if (shader == PIPE_SHADER_FRAGMENT) {
-        r300->fs_constants.dirty = TRUE;
+        r300_mark_atom_dirty(r300, &r300->fs_constants);
     }
 }