r300g: rework vertex format fallback

1) Only translate the [min_index, max_index] range.
2) Upload translated vertices via the uploader.
3) Rename valid_vertex_buffer[] to real_vertex_buffer[]

7 files changed, 86 insertions, 78 deletions

diff --git a/src/gallium/drivers/r300/r300_context.c b/src/gallium/drivers/r300/r300_context.c
index c8966ee59a..114fb316c0 100644
--- a/src/gallium/drivers/r300/r300_context.c
+++ b/src/gallium/drivers/r300/r300_context.c
@@ -84,9 +84,9 @@ static void r300_release_referenced_objects(struct r300_context *r300)
     pipe_resource_reference(&r300->vbo, NULL);
 
     /* Vertex buffers. */
-    for (i = 0; i < r300->vertex_buffer_count; i++) {
+    for (i = 0; i < r300->real_vertex_buffer_count; i++) {
         pipe_resource_reference(&r300->vertex_buffer[i].buffer, NULL);
-        pipe_resource_reference(&r300->valid_vertex_buffer[i], NULL);
+        pipe_resource_reference(&r300->real_vertex_buffer[i], NULL);
     }
 
     /* If there are any queries pending or not destroyed, remove them now. */
diff --git a/src/gallium/drivers/r300/r300_context.h b/src/gallium/drivers/r300/r300_context.h
index 57ecfb168f..430a0ddbb5 100644
--- a/src/gallium/drivers/r300/r300_context.h
+++ b/src/gallium/drivers/r300/r300_context.h
@@ -445,6 +445,7 @@ struct r300_translate_context {
 
     /* Saved and new vertex element state. */
     void *saved_velems, *new_velems;
+    unsigned vb_slot;
 };
 
 struct r300_context {
@@ -560,8 +561,9 @@ struct r300_context {
     /* May contain user buffers. */
     struct pipe_vertex_buffer vertex_buffer[PIPE_MAX_ATTRIBS];
     /* Contains only non-user buffers. */
-    struct pipe_resource *valid_vertex_buffer[PIPE_MAX_ATTRIBS];
+    struct pipe_resource *real_vertex_buffer[PIPE_MAX_ATTRIBS];
     int vertex_buffer_count;
+    int real_vertex_buffer_count; /* with the translated buffer. */
     int vertex_buffer_max_index;
     boolean any_user_vbs;
     /* Vertex elements for Gallium. */
diff --git a/src/gallium/drivers/r300/r300_emit.c b/src/gallium/drivers/r300/r300_emit.c
index 2157cb3ede..34f87f74d3 100644
--- a/src/gallium/drivers/r300/r300_emit.c
+++ b/src/gallium/drivers/r300/r300_emit.c
@@ -847,7 +847,7 @@ static void r300_update_vertex_arrays_cb(struct r300_context *r300, unsigned pac
 void r300_emit_vertex_arrays(struct r300_context* r300, int offset, boolean indexed)
 {
     struct pipe_vertex_buffer *vbuf = r300->vertex_buffer;
-    struct pipe_resource **valid_vbuf = r300->valid_vertex_buffer;
+    struct pipe_resource **valid_vbuf = r300->real_vertex_buffer;
     struct pipe_vertex_element *velem = r300->velems->velem;
     struct r300_buffer *buf;
     int i;
@@ -1227,9 +1227,9 @@ validate:
                                 r300_buffer(r300->vbo)->domain, 0);
     /* ...vertex buffers for HWTCL path... */
     if (do_validate_vertex_buffers) {
-        struct pipe_resource **buf = r300->valid_vertex_buffer;
-        struct pipe_resource **last = r300->valid_vertex_buffer +
-                                      r300->vertex_buffer_count;
+        struct pipe_resource **buf = r300->real_vertex_buffer;
+        struct pipe_resource **last = r300->real_vertex_buffer +
+                                      r300->real_vertex_buffer_count;
         for (; buf != last; buf++) {
             if (!*buf)
                 continue;
diff --git a/src/gallium/drivers/r300/r300_render.c b/src/gallium/drivers/r300/r300_render.c
index b35822c82f..41ddd748bb 100644
--- a/src/gallium/drivers/r300/r300_render.c
+++ b/src/gallium/drivers/r300/r300_render.c
@@ -325,7 +325,7 @@ static boolean immd_is_good_idea(struct r300_context *r300,
         vbi = velem->vertex_buffer_index;
 
         if (!checked[vbi]) {
-            buf = r300->valid_vertex_buffer[vbi];
+            buf = r300->real_vertex_buffer[vbi];
 
             if (!(r300_buffer(buf)->domain & R300_DOMAIN_GTT)) {
                 return FALSE;
@@ -390,7 +390,7 @@ static void r300_emit_draw_arrays_immediate(struct r300_context *r300,
         /* Map the buffer. */
         if (!transfer[vbi]) {
             map[vbi] = (uint32_t*)pipe_buffer_map(&r300->context,
-                                                  r300->valid_vertex_buffer[vbi],
+                                                  r300->real_vertex_buffer[vbi],
                                                   PIPE_TRANSFER_READ,
 						  &transfer[vbi]);
             map[vbi] += (vbuf->buffer_offset / 4) + stride[i] * start;
@@ -787,7 +787,7 @@ static void r300_swtcl_draw_vbo(struct pipe_context* pipe,
             (indexed ? PREP_INDEXED : 0),
             indexed ? 256 : 6);
 
-    for (i = 0; i < r300->vertex_buffer_count; i++) {
+    for (i = 0; i < r300->real_vertex_buffer_count; i++) {
         if (r300->vertex_buffer[i].buffer) {
             void *buf = pipe_buffer_map(pipe,
                                   r300->vertex_buffer[i].buffer,
@@ -810,7 +810,7 @@ static void r300_swtcl_draw_vbo(struct pipe_context* pipe,
     draw_flush(r300->draw);
     r300->draw_vbo_locked = FALSE;
 
-    for (i = 0; i < r300->vertex_buffer_count; i++) {
+    for (i = 0; i < r300->real_vertex_buffer_count; i++) {
         if (r300->vertex_buffer[i].buffer) {
             pipe_buffer_unmap(pipe, vb_transfer[i]);
             draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
diff --git a/src/gallium/drivers/r300/r300_render_translate.c b/src/gallium/drivers/r300/r300_render_translate.c
index c48062c808..3b544ae63b 100644
--- a/src/gallium/drivers/r300/r300_render_translate.c
+++ b/src/gallium/drivers/r300/r300_render_translate.c
@@ -30,9 +30,8 @@
 #include "r300_context.h"
 #include "translate/translate.h"
 #include "util/u_index_modify.h"
+#include "util/u_upload_mgr.h"
 
-/* XXX Optimization: use min_index and translate only that range. */
-/* XXX Use the uploader. */
 void r300_begin_vertex_translate(struct r300_context *r300,
                                  int min_index, int max_index)
 {
@@ -43,11 +42,12 @@ void r300_begin_vertex_translate(struct r300_context *r300,
     struct translate *tr;
     struct r300_vertex_element_state *ve = r300->velems;
     boolean vb_translated[PIPE_MAX_ATTRIBS] = {0};
-    void *vb_map[PIPE_MAX_ATTRIBS] = {0}, *out_map;
-    struct pipe_transfer *vb_transfer[PIPE_MAX_ATTRIBS] = {0}, *out_transfer;
-    struct pipe_resource *out_buffer;
-    unsigned i, num_verts;
-    unsigned slot;
+    uint8_t *vb_map[PIPE_MAX_ATTRIBS] = {0}, *out_map;
+    struct pipe_transfer *vb_transfer[PIPE_MAX_ATTRIBS] = {0};
+    struct pipe_resource *out_buffer = NULL;
+    unsigned i, num_verts, out_offset;
+    struct pipe_vertex_element new_velems[PIPE_MAX_ATTRIBS];
+    boolean flushed;
 
     /* Initialize the translate key, i.e. the recipe how vertices should be
      * translated. */
@@ -59,6 +59,7 @@ void r300_begin_vertex_translate(struct r300_context *r300,
 
         /* Check for support. */
         if (ve->velem[i].src_format == ve->hw_format[i] &&
+            /* These two are r300-specific.  */
             (vb->buffer_offset + ve->velem[i].src_offset) % 4 == 0 &&
             vb->stride % 4 == 0) {
             continue;
@@ -66,23 +67,23 @@ void r300_begin_vertex_translate(struct r300_context *r300,
 
         /* Workaround for translate: output floats instead of halfs. */
         switch (output_format) {
-            case PIPE_FORMAT_R16_FLOAT:
-                output_format = PIPE_FORMAT_R32_FLOAT;
-                output_format_size = 4;
-                break;
-            case PIPE_FORMAT_R16G16_FLOAT:
-                output_format = PIPE_FORMAT_R32G32_FLOAT;
-                output_format_size = 8;
-                break;
-            case PIPE_FORMAT_R16G16B16_FLOAT:
-                output_format = PIPE_FORMAT_R32G32B32_FLOAT;
-                output_format_size = 12;
-                break;
-            case PIPE_FORMAT_R16G16B16A16_FLOAT:
-                output_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
-                output_format_size = 16;
-                break;
-            default:;
+        case PIPE_FORMAT_R16_FLOAT:
+            output_format = PIPE_FORMAT_R32_FLOAT;
+            output_format_size = 4;
+            break;
+        case PIPE_FORMAT_R16G16_FLOAT:
+            output_format = PIPE_FORMAT_R32G32_FLOAT;
+            output_format_size = 8;
+            break;
+        case PIPE_FORMAT_R16G16B16_FLOAT:
+            output_format = PIPE_FORMAT_R32G32B32_FLOAT;
+            output_format_size = 12;
+            break;
+        case PIPE_FORMAT_R16G16B16A16_FLOAT:
+            output_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
+            output_format_size = 16;
+            break;
+        default:;
         }
 
         /* Add this vertex element. */
@@ -91,7 +92,7 @@ void r300_begin_vertex_translate(struct r300_context *r300,
         te->instance_divisor;*/
         te->input_buffer = ve->velem[i].vertex_buffer_index;
         te->input_format = ve->velem[i].src_format;
-        te->input_offset = vb->buffer_offset + ve->velem[i].src_offset;
+        te->input_offset = ve->velem[i].src_offset;
         te->output_format = output_format;
         te->output_offset = key.output_stride;
 
@@ -112,19 +113,22 @@ void r300_begin_vertex_translate(struct r300_context *r300,
             vb_map[i] = pipe_buffer_map(pipe, vb->buffer,
                                         PIPE_TRANSFER_READ, &vb_transfer[i]);
 
-            tr->set_buffer(tr, i, vb_map[i], vb->stride, max_index);
+            tr->set_buffer(tr, i,
+                           vb_map[i] + vb->buffer_offset + vb->stride * min_index,
+                           vb->stride, ~0);
         }
     }
 
     /* Create and map the output buffer. */
-    num_verts = max_index + 1;
+    num_verts = max_index + 1 - min_index;
 
-    out_buffer = pipe_buffer_create(&r300->screen->screen,
-                                    PIPE_BIND_VERTEX_BUFFER,
-                                    key.output_stride * num_verts);
+    u_upload_alloc(r300->upload_vb,
+                   key.output_stride * min_index,
+                   key.output_stride * num_verts,
+                   &out_offset, &out_buffer, &flushed,
+                   (void**)&out_map);
 
-    out_map = pipe_buffer_map(pipe, out_buffer, PIPE_TRANSFER_WRITE,
-                              &out_transfer);
+    out_offset -= key.output_stride * min_index;
 
     /* Translate. */
     tr->run(tr, 0, num_verts, 0, out_map);
@@ -136,48 +140,46 @@ void r300_begin_vertex_translate(struct r300_context *r300,
         }
     }
 
-    pipe_buffer_unmap(pipe, out_transfer);
-
     /* Setup the new vertex buffer in the first free slot. */
-    slot = ~0;
+    r300->tran.vb_slot = ~0;
     for (i = 0; i < PIPE_MAX_ATTRIBS; i++) {
-        struct pipe_vertex_buffer *vb = &r300->vertex_buffer[i];
-
-        if (!vb->buffer) {
-            pipe_resource_reference(&r300->valid_vertex_buffer[i], out_buffer);
-            vb->buffer_offset = 0;
-            vb->stride = key.output_stride;
-            slot = i;
-            /* XXX probably need to preserve the real count for u_blitter_save_*. */
-            r300->vertex_buffer_count = MAX2(r300->vertex_buffer_count, i+1);
+        if (!r300->vertex_buffer[i].buffer) {
+            r300->tran.vb_slot = i;
+
+            if (i >= r300->vertex_buffer_count) {
+                r300->real_vertex_buffer_count = i+1;
+            }
+
+            /* r300-specific: */
             r300->validate_buffers = TRUE;
+            r300->vertex_arrays_dirty = TRUE;
             break;
         }
     }
-    /* XXX This may fail. */
-    assert(slot != ~0);
-
-    /* Save and replace vertex elements. */
-    {
-        struct pipe_vertex_element new_velems[PIPE_MAX_ATTRIBS];
 
-        r300->tran.saved_velems = r300->velems;
+    if (r300->tran.vb_slot != ~0) {
+		/* Setup the new vertex buffer. */
+		pipe_resource_reference(&r300->real_vertex_buffer[r300->tran.vb_slot], out_buffer);
+		r300->vertex_buffer[r300->tran.vb_slot].buffer_offset = out_offset;
+		r300->vertex_buffer[r300->tran.vb_slot].stride = key.output_stride;
 
+        /* Setup new vertex elements. */
         for (i = 0; i < ve->count; i++) {
             if (vb_translated[ve->velem[i].vertex_buffer_index]) {
                 te = &key.element[tr_elem_index[i]];
                 new_velems[i].instance_divisor = ve->velem[i].instance_divisor;
                 new_velems[i].src_format = te->output_format;
                 new_velems[i].src_offset = te->output_offset;
-                new_velems[i].vertex_buffer_index = slot;
+                new_velems[i].vertex_buffer_index = r300->tran.vb_slot;
             } else {
                 memcpy(&new_velems[i], &ve->velem[i],
                        sizeof(struct pipe_vertex_element));
             }
         }
 
+        r300->tran.saved_velems = r300->velems;
         r300->tran.new_velems =
-            pipe->create_vertex_elements_state(pipe, ve->count, new_velems);
+                pipe->create_vertex_elements_state(pipe, ve->count, new_velems);
         pipe->bind_vertex_elements_state(pipe, r300->tran.new_velems);
     }
 
@@ -188,9 +190,19 @@ void r300_end_vertex_translate(struct r300_context *r300)
 {
     struct pipe_context *pipe = &r300->context;
 
+    if (r300->tran.new_velems == NULL) {
+        return;
+    }
+
     /* Restore vertex elements. */
     pipe->bind_vertex_elements_state(pipe, r300->tran.saved_velems);
+    r300->tran.saved_velems = NULL;
     pipe->delete_vertex_elements_state(pipe, r300->tran.new_velems);
+    r300->tran.new_velems = NULL;
+
+    /* Delete the now-unused VBO. */
+    pipe_resource_reference(&r300->real_vertex_buffer[r300->tran.vb_slot], NULL);
+    r300->real_vertex_buffer_count = r300->vertex_buffer_count;
 }
 
 /* XXX Use the uploader. */
diff --git a/src/gallium/drivers/r300/r300_screen_buffer.c b/src/gallium/drivers/r300/r300_screen_buffer.c
index cc3c1d7687..af6fa1048c 100644
--- a/src/gallium/drivers/r300/r300_screen_buffer.c
+++ b/src/gallium/drivers/r300/r300_screen_buffer.c
@@ -87,7 +87,7 @@ void r300_upload_user_buffers(struct r300_context *r300,
     int i, nr = r300->velems->count;
     unsigned count = max_index + 1 - min_index;
     boolean flushed;
-    boolean uploaded[16] = {0};
+    boolean uploaded[32] = {0};
 
     for (i = 0; i < nr; i++) {
         unsigned index = r300->velems->velem[i].vertex_buffer_index;
@@ -105,16 +105,10 @@ void r300_upload_user_buffers(struct r300_context *r300,
                 size = r300->velems->hw_format_size[i];
             }
 
-            DBG(r300, DBG_UPLOAD,
-                "Uploading %i bytes, index: %i, buffer: %p, userptr: %p "
-                "offset: %i, stride: %i.\n",
-                size, index, userbuf, userbuf->user_buffer,
-                vb->buffer_offset, vb->stride);
-
             u_upload_data(r300->upload_vb, first, size,
                           userbuf->user_buffer + first,
                           &vb->buffer_offset,
-                          &r300->valid_vertex_buffer[index],
+                          &r300->real_vertex_buffer[index],
                           &flushed);
 
             vb->buffer_offset -= first;
@@ -127,10 +121,9 @@ void r300_upload_user_buffers(struct r300_context *r300,
             }
             uploaded[index] = TRUE;
         } else {
-            assert(r300->valid_vertex_buffer[index]);
+            assert(r300->real_vertex_buffer[index]);
         }
     }
-    DBG(r300, DBG_UPLOAD, "-------\n");
 }
 
 static void r300_buffer_destroy(struct pipe_screen *screen,
diff --git a/src/gallium/drivers/r300/r300_state.c b/src/gallium/drivers/r300/r300_state.c
index ba456d413f..dad41ab91e 100644
--- a/src/gallium/drivers/r300/r300_state.c
+++ b/src/gallium/drivers/r300/r300_state.c
@@ -1560,21 +1560,22 @@ static void r300_set_vertex_buffers(struct pipe_context* pipe,
         /* Reference our 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);
+            pipe_resource_reference(&r300->real_vertex_buffer[i], NULL);
         } else {
-            pipe_resource_reference(&r300->valid_vertex_buffer[i], vbo->buffer);
+            pipe_resource_reference(&r300->real_vertex_buffer[i], vbo->buffer);
         }
     }
-    for (; i < r300->vertex_buffer_count; i++) {
+    for (; i < r300->real_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);
+        pipe_resource_reference(&r300->real_vertex_buffer[i], NULL);
     }
 
     memcpy(r300->vertex_buffer, buffers,
            sizeof(struct pipe_vertex_buffer) * count);
 
     r300->vertex_buffer_count = count;
+    r300->real_vertex_buffer_count = count;
 }
 
 static void r300_set_index_buffer(struct pipe_context* pipe,