Merge remote branch 'upstream/gallium-0.1' into nouveau-gallium-0.1

Conflicts:
	configs/default

39 files changed, 690 insertions, 414 deletions

diff --git a/src/gallium/drivers/cell/common.h b/src/gallium/drivers/cell/common.h
index f430e88b9c..c0ca201e1d 100644
--- a/src/gallium/drivers/cell/common.h
+++ b/src/gallium/drivers/cell/common.h
@@ -34,7 +34,6 @@
 #define CELL_COMMON_H
 
 #include "pipe/p_compiler.h"
-#include "pipe/p_util.h"
 #include "pipe/p_format.h"
 #include "pipe/p_state.h"
 
@@ -107,6 +106,9 @@
 #define CELL_BUFFER_STATUS_USED 20
 
 
+#define CELL_DEBUG_CHECKER  (1 << 0)
+#define CELL_DEBUG_SYNC     (1 << 1)
+
 
 /**
  */
@@ -264,6 +266,7 @@ struct cell_init_info
 {
    unsigned id;
    unsigned num_spus;
+   unsigned debug_flags;  /**< mask of CELL_DEBUG_x flags */
    struct cell_command *cmd;
 
    /** Buffers for command batches, vertex/index data */
diff --git a/src/gallium/drivers/cell/ppu/Makefile b/src/gallium/drivers/cell/ppu/Makefile
index 0389a9554c..25473e200c 100644
--- a/src/gallium/drivers/cell/ppu/Makefile
+++ b/src/gallium/drivers/cell/ppu/Makefile
@@ -5,7 +5,7 @@
 
 
 TOP = ../../../../..
-include $(TOP)/configs/linux-cell
+include $(TOP)/configs/current
 
 
 # This is the "top-level" cell PPU driver code, will get pulled into libGL.so
diff --git a/src/gallium/drivers/cell/ppu/cell_batch.c b/src/gallium/drivers/cell/ppu/cell_batch.c
index f45e5f25b6..16882c0129 100644
--- a/src/gallium/drivers/cell/ppu/cell_batch.c
+++ b/src/gallium/drivers/cell/ppu/cell_batch.c
@@ -32,6 +32,13 @@
 
 
 
+/**
+ * Search the buffer pool for an empty/free buffer and return its index.
+ * Buffers are used for storing vertex data, state and commands which
+ * will be sent to the SPUs.
+ * If no empty buffers are available, wait for one.
+ * \return buffer index in [0, CELL_NUM_BUFFERS-1]
+ */
 uint
 cell_get_empty_buffer(struct cell_context *cell)
 {
@@ -74,6 +81,11 @@ cell_get_empty_buffer(struct cell_context *cell)
 }
 
 
+/**
+ * Flush the current batch buffer to the SPUs.
+ * An empty buffer will be found and set as the new current batch buffer
+ * for subsequent commands/data.
+ */
 void
 cell_batch_flush(struct cell_context *cell)
 {
@@ -93,11 +105,11 @@ cell_batch_flush(struct cell_context *cell)
 
    /*
    printf("cell_batch_dispatch: buf %u at %p, size %u\n",
-          batch, &cell->batch_buffer[batch][0], size);
+          batch, &cell->buffer[batch][0], size);
    */
      
    /*
-    * Build "BATCH" command and sent to all SPUs.
+    * Build "BATCH" command and send to all SPUs.
     */
    cmd_word = CELL_CMD_BATCH | (batch << 8) | (size << 16);
 
@@ -120,6 +132,9 @@ cell_batch_flush(struct cell_context *cell)
 }
 
 
+/**
+ * Return the number of bytes free in the current batch buffer.
+ */
 uint
 cell_batch_free_space(const struct cell_context *cell)
 {
@@ -129,7 +144,9 @@ cell_batch_free_space(const struct cell_context *cell)
 
 
 /**
- * Append data to current batch.
+ * Append data to the current batch buffer.
+ * \param data  address of block of bytes to append
+ * \param bytes  size of block of bytes
  */
 void
 cell_batch_append(struct cell_context *cell, const void *data, uint bytes)
@@ -165,6 +182,10 @@ cell_batch_append(struct cell_context *cell, const void *data, uint bytes)
 }
 
 
+/**
+ * Allocate space in the current batch buffer for 'bytes' space.
+ * \return address in batch buffer to put data
+ */
 void *
 cell_batch_alloc(struct cell_context *cell, uint bytes)
 {
@@ -172,6 +193,10 @@ cell_batch_alloc(struct cell_context *cell, uint bytes)
 }
 
 
+/**
+ * Same as \sa cell_batch_alloc, but return an address at a particular
+ * alignment.
+ */
 void *
 cell_batch_alloc_aligned(struct cell_context *cell, uint bytes,
                          uint alignment)
@@ -215,3 +240,28 @@ cell_batch_alloc_aligned(struct cell_context *cell, uint bytes,
 
    return pos;
 }
+
+
+/**
+ * One-time init of batch buffers.
+ */
+void
+cell_init_batch_buffers(struct cell_context *cell)
+{
+   uint spu, buf;
+
+   /* init command, vertex/index buffer info */
+   for (buf = 0; buf < CELL_NUM_BUFFERS; buf++) {
+      cell->buffer_size[buf] = 0;
+
+      /* init batch buffer status values,
+       * mark 0th buffer as used, rest as free.
+       */
+      for (spu = 0; spu < cell->num_spus; spu++) {
+         if (buf == 0)
+            cell->buffer_status[spu][buf][0] = CELL_BUFFER_STATUS_USED;
+         else
+            cell->buffer_status[spu][buf][0] = CELL_BUFFER_STATUS_FREE;
+      }
+   }
+}
diff --git a/src/gallium/drivers/cell/ppu/cell_batch.h b/src/gallium/drivers/cell/ppu/cell_batch.h
index a6eee0a8b1..f74dd60079 100644
--- a/src/gallium/drivers/cell/ppu/cell_batch.h
+++ b/src/gallium/drivers/cell/ppu/cell_batch.h
@@ -54,5 +54,8 @@ extern void *
 cell_batch_alloc_aligned(struct cell_context *cell, uint bytes,
                          uint alignment);
 
+extern void
+cell_init_batch_buffers(struct cell_context *cell);
+
 
 #endif /* CELL_BATCH_H */
diff --git a/src/gallium/drivers/cell/ppu/cell_clear.c b/src/gallium/drivers/cell/ppu/cell_clear.c
index 3ffe09add6..c9c0c721bb 100644
--- a/src/gallium/drivers/cell/ppu/cell_clear.c
+++ b/src/gallium/drivers/cell/ppu/cell_clear.c
@@ -34,7 +34,8 @@
 #include <assert.h>
 #include <stdint.h>
 #include "pipe/p_inlines.h"
-#include "pipe/p_util.h"
+#include "util/u_memory.h"
+#include "util/u_pack_color.h"
 #include "cell/common.h"
 #include "cell_clear.h"
 #include "cell_context.h"
@@ -44,10 +45,32 @@
 #include "cell_state.h"
 
 
+/**
+ * Convert packed pixel from one format to another.
+ */
+static unsigned
+convert_color(enum pipe_format srcFormat, unsigned srcColor,
+              enum pipe_format dstFormat)
+{
+   ubyte r, g, b, a;
+   unsigned dstColor;
+
+   util_unpack_color_ub(srcFormat, &srcColor, &r, &g, &b, &a);
+   util_pack_color_ub(r, g, b, a, dstFormat, &dstColor);
+
+   return dstColor;
+}
+
+
+
+/**
+ * Called via pipe->clear()
+ */
 void
 cell_clear_surface(struct pipe_context *pipe, struct pipe_surface *ps,
                    unsigned clearValue)
 {
+   struct pipe_screen *screen = pipe->screen;
    struct cell_context *cell = cell_context(pipe);
    uint surfIndex;
 
@@ -56,16 +79,25 @@ cell_clear_surface(struct pipe_context *pipe, struct pipe_surface *ps,
 
 
    if (!cell->cbuf_map[0])
-      cell->cbuf_map[0] = pipe_surface_map(ps);
+      cell->cbuf_map[0] = screen->surface_map(screen, ps,
+                                              PIPE_BUFFER_USAGE_GPU_WRITE);
 
    if (ps == cell->framebuffer.zsbuf) {
+      /* clear z/stencil buffer */
       surfIndex = 1;
    }
    else {
+      /* clear color buffer */
       surfIndex = 0;
+
+      if (ps->format != PIPE_FORMAT_A8R8G8B8_UNORM) {
+         clearValue = convert_color(PIPE_FORMAT_A8R8G8B8_UNORM, clearValue,
+                                    ps->format);
+      }
    }
 
 
+   /* Build a CLEAR command and place it in the current batch buffer */
    {
       struct cell_command_clear_surface *clr
          = (struct cell_command_clear_surface *)
diff --git a/src/gallium/drivers/cell/ppu/cell_context.c b/src/gallium/drivers/cell/ppu/cell_context.c
index 12eb5aa254..71f1a3049d 100644
--- a/src/gallium/drivers/cell/ppu/cell_context.c
+++ b/src/gallium/drivers/cell/ppu/cell_context.c
@@ -35,7 +35,7 @@
 
 #include "pipe/p_defines.h"
 #include "pipe/p_format.h"
-#include "pipe/p_util.h"
+#include "util/u_memory.h"
 #include "pipe/p_winsys.h"
 #include "pipe/p_screen.h"
 
@@ -43,11 +43,11 @@
 #include "draw/draw_private.h"
 
 #include "cell/common.h"
+#include "cell_batch.h"
 #include "cell_clear.h"
 #include "cell_context.h"
 #include "cell_draw_arrays.h"
 #include "cell_flush.h"
-#include "cell_render.h"
 #include "cell_state.h"
 #include "cell_surface.h"
 #include "cell_spu.h"
@@ -73,22 +73,32 @@ cell_draw_create(struct cell_context *cell)
 {
    struct draw_context *draw = draw_create();
 
+#if 0 /* broken */
    if (getenv("GALLIUM_CELL_VS")) {
       /* plug in SPU-based vertex transformation code */
       draw->shader_queue_flush = cell_vertex_shader_queue_flush;
       draw->driver_private = cell;
    }
+#endif
 
    return draw;
 }
 
 
+#ifdef DEBUG
+static const struct debug_named_value cell_debug_flags[] = {
+   {"checker", CELL_DEBUG_CHECKER},/**< modulate tile clear color by SPU ID */
+   {"sync", CELL_DEBUG_SYNC},      /**< SPUs do synchronous DMA */
+   {NULL, 0}
+};
+#endif
+
+
 struct pipe_context *
 cell_create_context(struct pipe_screen *screen,
                     struct cell_winsys *cws)
 {
    struct cell_context *cell;
-   uint spu, buf;
 
    /* some fields need to be 16-byte aligned, so align the whole object */
    cell = (struct cell_context*) align_malloc(sizeof(struct cell_context), 16);
@@ -102,13 +112,6 @@ cell_create_context(struct pipe_screen *screen,
    cell->pipe.screen = screen;
    cell->pipe.destroy = cell_destroy_context;
 
-   /* state setters */
-   cell->pipe.set_vertex_buffers = cell_set_vertex_buffers;
-   cell->pipe.set_vertex_elements = cell_set_vertex_elements;
-
-   cell->pipe.draw_arrays = cell_draw_arrays;
-   cell->pipe.draw_elements = cell_draw_elements;
-
    cell->pipe.clear = cell_clear_surface;
    cell->pipe.flush = cell_flush;
 
@@ -118,20 +121,28 @@ cell_create_context(struct pipe_screen *screen,
    cell->pipe.wait_query = cell_wait_query;
 #endif
 
+   cell_init_draw_functions(cell);
    cell_init_state_functions(cell);
    cell_init_shader_functions(cell);
    cell_init_surface_functions(cell);
    cell_init_texture_functions(cell);
+   cell_init_vertex_functions(cell);
 
    cell->draw = cell_draw_create(cell);
 
    cell_init_vbuf(cell);
+
    draw_set_rasterize_stage(cell->draw, cell->vbuf);
 
    /* convert all points/lines to tris for the time being */
    draw_wide_point_threshold(cell->draw, 0.0);
    draw_wide_line_threshold(cell->draw, 0.0);
 
+   /* get env vars or read config file to get debug flags */
+   cell->debug_flags = debug_get_flags_option("CELL_DEBUG", 
+                                              cell_debug_flags, 
+                                              0 );
+
    /*
     * SPU stuff
     */
@@ -142,20 +153,7 @@ cell_create_context(struct pipe_screen *screen,
 
    cell_start_spus(cell);
 
-   /* init command, vertex/index buffer info */
-   for (buf = 0; buf < CELL_NUM_BUFFERS; buf++) {
-      cell->buffer_size[buf] = 0;
-
-      /* init batch buffer status values,
-       * mark 0th buffer as used, rest as free.
-       */
-      for (spu = 0; spu < cell->num_spus; spu++) {
-         if (buf == 0)
-            cell->buffer_status[spu][buf][0] = CELL_BUFFER_STATUS_USED;
-         else
-            cell->buffer_status[spu][buf][0] = CELL_BUFFER_STATUS_FREE;
-      }
-   }
+   cell_init_batch_buffers(cell);
 
    return &cell->pipe;
 }
diff --git a/src/gallium/drivers/cell/ppu/cell_context.h b/src/gallium/drivers/cell/ppu/cell_context.h
index f1d1ca89a9..8cec9f45b2 100644
--- a/src/gallium/drivers/cell/ppu/cell_context.h
+++ b/src/gallium/drivers/cell/ppu/cell_context.h
@@ -39,8 +39,13 @@
 #include "rtasm/rtasm_ppc_spe.h"
 #include "tgsi/tgsi_scan.h"
 
+
 struct cell_vbuf_render;
 
+
+/**
+ * Cell vertex shader state, subclass of pipe_shader_state.
+ */
 struct cell_vertex_shader_state
 {
    struct pipe_shader_state shader;
@@ -49,6 +54,9 @@ struct cell_vertex_shader_state
 };
 
 
+/**
+ * Cell fragment shader state, subclass of pipe_shader_state.
+ */
 struct cell_fragment_shader_state
 {
    struct pipe_shader_state shader;
@@ -57,7 +65,11 @@ struct cell_fragment_shader_state
 };
 
 
-struct cell_blend_state {
+/**
+ * Cell blend state atom, subclass of pipe_blend_state.
+ */
+struct cell_blend_state
+{
    struct pipe_blend_state base;
 
    /**
@@ -67,17 +79,24 @@ struct cell_blend_state {
 };
 
 
-struct cell_depth_stencil_alpha_state {
-   struct pipe_depth_stencil_alpha_state   base;
+/**
+ * Cell depth/stencil/alpha state atom, subclass of
+ * pipe_depth_stencil_alpha_state.
+ */
+struct cell_depth_stencil_alpha_state
+{
+   struct pipe_depth_stencil_alpha_state base;
 
    /**
     * Generated code to perform alpha, stencil, and depth testing on the SPE
     */
    struct spe_function code;
-
 };
 
 
+/**
+ * Per-context state, subclass of pipe_context.
+ */
 struct cell_context
 {
    struct pipe_context pipe;
@@ -144,6 +163,8 @@ struct cell_context
 
    struct spe_function attrib_fetch;
    unsigned attrib_fetch_offsets[PIPE_MAX_ATTRIBS];
+
+   unsigned debug_flags;
 };
 
 
diff --git a/src/gallium/drivers/cell/ppu/cell_draw_arrays.c b/src/gallium/drivers/cell/ppu/cell_draw_arrays.c
index 6e08cf6fe8..880d535320 100644
--- a/src/gallium/drivers/cell/ppu/cell_draw_arrays.c
+++ b/src/gallium/drivers/cell/ppu/cell_draw_arrays.c
@@ -34,6 +34,7 @@
 #include "pipe/p_defines.h"
 #include "pipe/p_context.h"
 #include "pipe/p_winsys.h"
+#include "pipe/p_inlines.h"
 
 #include "cell_context.h"
 #include "cell_draw_arrays.h"
@@ -59,7 +60,8 @@ cell_map_constant_buffers(struct cell_context *sp)
    }
 
    draw_set_mapped_constant_buffer(sp->draw,
-                                   sp->mapped_constants[PIPE_SHADER_VERTEX]);
+                                   sp->mapped_constants[PIPE_SHADER_VERTEX],
+                                   sp->constants[PIPE_SHADER_VERTEX].size);
 }
 
 static void
@@ -75,14 +77,6 @@ cell_unmap_constant_buffers(struct cell_context *sp)
 }
 
 
-boolean
-cell_draw_arrays(struct pipe_context *pipe, unsigned mode,
-                     unsigned start, unsigned count)
-{
-   return cell_draw_elements(pipe, NULL, 0, mode, start, count);
-}
-
-
 
 /**
  * Draw vertex arrays, with optional indexing.
@@ -91,11 +85,13 @@ cell_draw_arrays(struct pipe_context *pipe, unsigned mode,
  *
  * XXX should the element buffer be specified/bound with a separate function?
  */
-boolean
-cell_draw_elements(struct pipe_context *pipe,
-                       struct pipe_buffer *indexBuffer,
-                       unsigned indexSize,
-                       unsigned mode, unsigned start, unsigned count)
+static boolean
+cell_draw_range_elements(struct pipe_context *pipe,
+                         struct pipe_buffer *indexBuffer,
+                         unsigned indexSize,
+                         unsigned min_index,
+                         unsigned max_index,
+                         unsigned mode, unsigned start, unsigned count)
 {
    struct cell_context *sp = cell_context(pipe);
    struct draw_context *draw = sp->draw;
@@ -113,7 +109,7 @@ cell_draw_elements(struct pipe_context *pipe,
     * Map vertex buffers
     */
    for (i = 0; i < sp->num_vertex_buffers; i++) {
-      void *buf = pipe->winsys->buffer_map(pipe->winsys,
+      void *buf = pipe_buffer_map(pipe->screen,
                                            sp->vertex_buffer[i].buffer,
                                            PIPE_BUFFER_USAGE_CPU_READ);
       cell_flush_buffer_range(sp, buf, sp->vertex_buffer[i].buffer->size);
@@ -121,7 +117,7 @@ cell_draw_elements(struct pipe_context *pipe,
    }
    /* Map index buffer, if present */
    if (indexBuffer) {
-      void *mapped_indexes = pipe->winsys->buffer_map(pipe->winsys,
+      void *mapped_indexes = pipe_buffer_map(pipe->screen,
                                                       indexBuffer,
                                                       PIPE_BUFFER_USAGE_CPU_READ);
       draw_set_mapped_element_buffer(draw, indexSize, mapped_indexes);
@@ -140,11 +136,11 @@ cell_draw_elements(struct pipe_context *pipe,
     */
    for (i = 0; i < sp->num_vertex_buffers; i++) {
       draw_set_mapped_vertex_buffer(draw, i, NULL);
-      pipe->winsys->buffer_unmap(pipe->winsys, sp->vertex_buffer[i].buffer);
+      pipe_buffer_unmap(pipe->screen, sp->vertex_buffer[i].buffer);
    }
    if (indexBuffer) {
       draw_set_mapped_element_buffer(draw, 0, NULL);
-      pipe->winsys->buffer_unmap(pipe->winsys, indexBuffer);
+      pipe_buffer_unmap(pipe->screen, indexBuffer);
    }
 
    /* Note: leave drawing surfaces mapped */
@@ -152,3 +148,44 @@ cell_draw_elements(struct pipe_context *pipe,
 
    return TRUE;
 }
+
+
+static boolean
+cell_draw_elements(struct pipe_context *pipe,
+                   struct pipe_buffer *indexBuffer,
+                   unsigned indexSize,
+                   unsigned mode, unsigned start, unsigned count)
+{
+   return cell_draw_range_elements( pipe, indexBuffer,
+                                    indexSize,
+                                    0, 0xffffffff,
+                                    mode, start, count );
+}
+
+
+static boolean
+cell_draw_arrays(struct pipe_context *pipe, unsigned mode,
+                     unsigned start, unsigned count)
+{
+   return cell_draw_elements(pipe, NULL, 0, mode, start, count);
+}
+
+
+static void
+cell_set_edgeflags(struct pipe_context *pipe, const unsigned *edgeflags)
+{
+   struct cell_context *cell = cell_context(pipe);
+   draw_set_edgeflags(cell->draw, edgeflags);
+}
+
+
+
+void
+cell_init_draw_functions(struct cell_context *cell)
+{
+   cell->pipe.draw_arrays = cell_draw_arrays;
+   cell->pipe.draw_elements = cell_draw_elements;
+   cell->pipe.draw_range_elements = cell_draw_range_elements;
+   cell->pipe.set_edgeflags = cell_set_edgeflags;
+}
+
diff --git a/src/gallium/drivers/cell/ppu/cell_draw_arrays.h b/src/gallium/drivers/cell/ppu/cell_draw_arrays.h
index d5df4aa05f..148873aa67 100644
--- a/src/gallium/drivers/cell/ppu/cell_draw_arrays.h
+++ b/src/gallium/drivers/cell/ppu/cell_draw_arrays.h
@@ -29,14 +29,8 @@
 #define CELL_DRAW_ARRAYS_H
 
 
-boolean cell_draw_arrays(struct pipe_context *pipe, unsigned mode,
-                         unsigned start, unsigned count);
-
-boolean cell_draw_elements(struct pipe_context *pipe,
-                           struct pipe_buffer *indexBuffer,
-                           unsigned indexSize,
-                           unsigned mode, unsigned start, unsigned count);
-
+extern void
+cell_init_draw_functions(struct cell_context *cell);
 
 
 #endif /* CELL_DRAW_ARRAYS_H */
diff --git a/src/gallium/drivers/cell/ppu/cell_flush.c b/src/gallium/drivers/cell/ppu/cell_flush.c
index 3aaf3de668..6596b72010 100644
--- a/src/gallium/drivers/cell/ppu/cell_flush.c
+++ b/src/gallium/drivers/cell/ppu/cell_flush.c
@@ -34,6 +34,9 @@
 #include "draw/draw_context.h"
 
 
+/**
+ * Called via pipe->flush()
+ */
 void
 cell_flush(struct pipe_context *pipe, unsigned flags,
            struct pipe_fence_handle **fence)
@@ -50,16 +53,19 @@ cell_flush(struct pipe_context *pipe, unsigned flags,
       flags |= CELL_FLUSH_WAIT;
 
    draw_flush( cell->draw );
-   cell_flush_int(pipe, flags);
+   cell_flush_int(cell, flags);
 }
 
 
-/** internal flush */
+/**
+ * Cell internal flush function.  Send the current batch buffer to all SPUs.
+ * If flags & CELL_FLUSH_WAIT, do not return until the SPUs are idle.
+ * \param flags  bitmask of flags CELL_FLUSH_WAIT, or zero
+ */
 void
-cell_flush_int(struct pipe_context *pipe, unsigned flags)
+cell_flush_int(struct cell_context *cell, unsigned flags)
 {
    static boolean flushing = FALSE;  /* recursion catcher */
-   struct cell_context *cell = cell_context(pipe);
    uint i;
 
    ASSERT(!flushing);
diff --git a/src/gallium/drivers/cell/ppu/cell_flush.h b/src/gallium/drivers/cell/ppu/cell_flush.h
index 8f0645c429..509ae6239a 100644
--- a/src/gallium/drivers/cell/ppu/cell_flush.h
+++ b/src/gallium/drivers/cell/ppu/cell_flush.h
@@ -36,7 +36,7 @@ cell_flush(struct pipe_context *pipe, unsigned flags,
            struct pipe_fence_handle **fence);
 
 extern void
-cell_flush_int(struct pipe_context *pipe, unsigned flags);
+cell_flush_int(struct cell_context *cell, unsigned flags);
 
 extern void
 cell_flush_buffer_range(struct cell_context *cell, void *ptr,
diff --git a/src/gallium/drivers/cell/ppu/cell_pipe_state.c b/src/gallium/drivers/cell/ppu/cell_pipe_state.c
index 67b87f16d7..e04cf5f274 100644
--- a/src/gallium/drivers/cell/ppu/cell_pipe_state.c
+++ b/src/gallium/drivers/cell/ppu/cell_pipe_state.c
@@ -30,10 +30,11 @@
  *  Brian Paul
  */
 
-#include "pipe/p_util.h"
+#include "util/u_memory.h"
 #include "pipe/p_inlines.h"
 #include "draw/draw_context.h"
 #include "cell_context.h"
+#include "cell_flush.h"
 #include "cell_state.h"
 #include "cell_texture.h"
 #include "cell_state_per_fragment.h"
@@ -130,8 +131,9 @@ cell_delete_depth_stencil_alpha_state(struct pipe_context *pipe, void *depth)
 }
 
 
-static void cell_set_clip_state( struct pipe_context *pipe,
-			     const struct pipe_clip_state *clip )
+static void
+cell_set_clip_state(struct pipe_context *pipe,
+                    const struct pipe_clip_state *clip)
 {
    struct cell_context *cell = cell_context(pipe);
 
@@ -294,6 +296,8 @@ cell_set_framebuffer_state(struct pipe_context *pipe,
       struct pipe_surface *csurf = fb->cbufs[0];
       struct pipe_surface *zsurf = fb->zsbuf;
       uint i;
+      uint flags = (PIPE_BUFFER_USAGE_GPU_WRITE |
+                    PIPE_BUFFER_USAGE_GPU_READ);
 
       /* unmap old surfaces */
       for (i = 0; i < PIPE_MAX_COLOR_BUFS; i++) {
@@ -308,15 +312,28 @@ cell_set_framebuffer_state(struct pipe_context *pipe,
          cell->zsbuf_map = NULL;
       }
 
-      /* update my state */
-      cell->framebuffer = *fb;
+      /* Finish any pending rendering to the current surface before
+       * installing a new surface!
+       */
+      cell_flush_int(cell, CELL_FLUSH_WAIT);
+
+      /* update my state
+       * (this is also where old surfaces will finally get freed)
+       */
+      cell->framebuffer.width = fb->width;
+      cell->framebuffer.height = fb->height;
+      cell->framebuffer.num_cbufs = fb->num_cbufs;
+      for (i = 0; i < PIPE_MAX_COLOR_BUFS; i++) {
+         pipe_surface_reference(&cell->framebuffer.cbufs[i], fb->cbufs[i]);
+      }
+      pipe_surface_reference(&cell->framebuffer.zsbuf, fb->zsbuf);
 
       /* map new surfaces */
       if (csurf)
-         cell->cbuf_map[0] = pipe_surface_map(csurf);
+         cell->cbuf_map[0] = pipe_surface_map(csurf, flags);
 
       if (zsurf)
-         cell->zsbuf_map = pipe_surface_map(zsurf);
+         cell->zsbuf_map = pipe_surface_map(zsurf, flags);
 
       cell->dirty |= CELL_NEW_FRAMEBUFFER;
    }
diff --git a/src/gallium/drivers/cell/ppu/cell_render.c b/src/gallium/drivers/cell/ppu/cell_render.c
index b663b37622..dd25ae880e 100644
--- a/src/gallium/drivers/cell/ppu/cell_render.c
+++ b/src/gallium/drivers/cell/ppu/cell_render.c
@@ -33,7 +33,7 @@
 #include "cell_context.h"
 #include "cell_render.h"
 #include "cell_spu.h"
-#include "pipe/p_util.h"
+#include "util/u_memory.h"
 #include "draw/draw_private.h"
 
 
diff --git a/src/gallium/drivers/cell/ppu/cell_screen.c b/src/gallium/drivers/cell/ppu/cell_screen.c
index 2bf441a0c5..139b3719b6 100644
--- a/src/gallium/drivers/cell/ppu/cell_screen.c
+++ b/src/gallium/drivers/cell/ppu/cell_screen.c
@@ -26,7 +26,7 @@
  **************************************************************************/
 
 
-#include "pipe/p_util.h"
+#include "util/u_memory.h"
 #include "pipe/p_winsys.h"
 #include "pipe/p_defines.h"
 #include "pipe/p_screen.h"
diff --git a/src/gallium/drivers/cell/ppu/cell_spu.c b/src/gallium/drivers/cell/ppu/cell_spu.c
index 973c0b1aa1..9508227e29 100644
--- a/src/gallium/drivers/cell/ppu/cell_spu.c
+++ b/src/gallium/drivers/cell/ppu/cell_spu.c
@@ -26,6 +26,11 @@
  **************************************************************************/
 
 
+/**
+ * Utility/wrappers for communicating with the SPUs.
+ */
+
+
 #include <pthread.h>
 
 #include "cell_spu.h"
@@ -40,6 +45,9 @@ helpful headers:
 */
 
 
+/**
+ * Cell/SPU info that's not per-context.
+ */
 struct cell_global_info cell_global;
 
 
@@ -74,7 +82,11 @@ wait_mbox_message(spe_context_ptr_t ctx)
 }
 
 
-static void *cell_thread_function(void *arg)
+/**
+ * Called by pthread_create() to spawn an SPU thread.
+ */
+static void *
+cell_thread_function(void *arg)
 {
    struct cell_init_info *init = (struct cell_init_info *) arg;
    unsigned entry = SPE_DEFAULT_ENTRY;
@@ -92,7 +104,10 @@ static void *cell_thread_function(void *arg)
 
 
 /**
- * Create the SPU threads
+ * Create the SPU threads.  This is done once during driver initialization.
+ * This involves setting the the "init" message which is sent to each SPU.
+ * The init message specifies an SPU id, total number of SPUs, location
+ * and number of batch buffers, etc.
  */
 void
 cell_start_spus(struct cell_context *cell)
@@ -100,7 +115,6 @@ cell_start_spus(struct cell_context *cell)
    static boolean one_time_init = FALSE;
    uint i, j;
 
-
    if (one_time_init) {
       fprintf(stderr, "PPU: Multiple rendering contexts not yet supported "
 	      "on Cell.\n");
@@ -120,6 +134,7 @@ cell_start_spus(struct cell_context *cell)
    for (i = 0; i < cell->num_spus; i++) {
       cell_global.inits[i].id = i;
       cell_global.inits[i].num_spus = cell->num_spus;
+      cell_global.inits[i].debug_flags = cell->debug_flags;
       cell_global.inits[i].cmd = &cell_global.command[i];
       for (j = 0; j < CELL_NUM_BUFFERS; j++) {
          cell_global.inits[i].buffers[j] = cell->buffer[j];
@@ -137,14 +152,17 @@ cell_start_spus(struct cell_context *cell)
          exit(1);
       }
       
-      pthread_create(&cell_global.spe_threads[i], NULL, &cell_thread_function,
-		     &cell_global.inits[i]);
+      pthread_create(&cell_global.spe_threads[i], /* returned thread handle */
+                     NULL,                        /* pthread attribs */
+                     &cell_thread_function,       /* start routine */
+		     &cell_global.inits[i]);      /* thread argument */
    }
 }
 
 
 /**
  * Tell all the SPUs to stop/exit.
+ * This is done when the driver's exiting / cleaning up.
  */
 void
 cell_spu_exit(struct cell_context *cell)
diff --git a/src/gallium/drivers/cell/ppu/cell_state.h b/src/gallium/drivers/cell/ppu/cell_state.h
index 82580ea35a..a7771a55a3 100644
--- a/src/gallium/drivers/cell/ppu/cell_state.h
+++ b/src/gallium/drivers/cell/ppu/cell_state.h
@@ -48,19 +48,17 @@
 #define CELL_NEW_VERTEX_INFO   0x8000
 
 
-void cell_set_vertex_elements(struct pipe_context *,
-                              unsigned count,
-                              const struct pipe_vertex_element *);
+extern void
+cell_update_derived( struct cell_context *softpipe );
 
-void cell_set_vertex_buffers(struct pipe_context *,
-                             unsigned count,
-                             const struct pipe_vertex_buffer *);
 
-void cell_update_derived( struct cell_context *softpipe );
+extern void
+cell_init_shader_functions(struct cell_context *cell);
 
 
-void
-cell_init_shader_functions(struct cell_context *cell);
+extern void
+cell_init_vertex_functions(struct cell_context *cell);
+
 
 #endif /* CELL_STATE_H */
 
diff --git a/src/gallium/drivers/cell/ppu/cell_state_derived.c b/src/gallium/drivers/cell/ppu/cell_state_derived.c
index 5480534ad9..efc4f78364 100644
--- a/src/gallium/drivers/cell/ppu/cell_state_derived.c
+++ b/src/gallium/drivers/cell/ppu/cell_state_derived.c
@@ -25,7 +25,7 @@
  * 
  **************************************************************************/
 
-#include "pipe/p_util.h"
+#include "util/u_memory.h"
 #include "pipe/p_shader_tokens.h"
 #include "draw/draw_context.h"
 #include "draw/draw_vertex.h"
@@ -35,21 +35,6 @@
 #include "cell_state_emit.h"
 
 
-static int
-find_vs_output(const struct cell_vertex_shader_state *vs,
-               uint semantic_name,
-               uint semantic_index)
-{
-   uint i;
-   for (i = 0; i < vs->info.num_outputs; i++) {
-      if (vs->info.output_semantic_name[i] == semantic_name &&
-          vs->info.output_semantic_index[i] == semantic_index)
-         return i;
-   }
-   return -1;
-}
-
-
 /**
  * Determine how to map vertex program outputs to fragment program inputs.
  * Basically, this will be used when computing the triangle interpolation
@@ -58,7 +43,6 @@ find_vs_output(const struct cell_vertex_shader_state *vs,
 static void
 calculate_vertex_layout( struct cell_context *cell )
 {
-   const struct cell_vertex_shader_state *vs = cell->vs;
    const struct cell_fragment_shader_state *fs = cell->fs;
    const enum interp_mode colorInterp
       = cell->rasterizer->flatshade ? INTERP_CONSTANT : INTERP_LINEAR;
@@ -82,7 +66,7 @@ calculate_vertex_layout( struct cell_context *cell )
    vinfo->num_attribs = 0;
 
    /* we always want to emit vertex pos */
-   src = find_vs_output(vs, TGSI_SEMANTIC_POSITION, 0);
+   src = draw_find_vs_output(cell->draw, TGSI_SEMANTIC_POSITION, 0);
    assert(src >= 0);
    draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_POS, src);
 
@@ -98,14 +82,14 @@ calculate_vertex_layout( struct cell_context *cell )
          break;
 
       case TGSI_SEMANTIC_COLOR:
-         src = find_vs_output(vs, TGSI_SEMANTIC_COLOR, 
-                              fs->info.input_semantic_index[i]);
+         src = draw_find_vs_output(cell->draw, TGSI_SEMANTIC_COLOR, 
+                                   fs->info.input_semantic_index[i]);
          assert(src >= 0);
          draw_emit_vertex_attr(vinfo, EMIT_4F, colorInterp, src);
          break;
 
       case TGSI_SEMANTIC_FOG:
-         src = find_vs_output(vs, TGSI_SEMANTIC_FOG, 0);
+         src = draw_find_vs_output(cell->draw, TGSI_SEMANTIC_FOG, 0);
 #if 1
          if (src < 0) /* XXX temp hack, try demos/fogcoord.c with this */
             src = 0;
@@ -116,7 +100,7 @@ calculate_vertex_layout( struct cell_context *cell )
 
       case TGSI_SEMANTIC_GENERIC:
          /* this includes texcoords and varying vars */
-         src = find_vs_output(vs, TGSI_SEMANTIC_GENERIC,
+         src = draw_find_vs_output(cell->draw, TGSI_SEMANTIC_GENERIC,
                               fs->info.input_semantic_index[i]);
          assert(src >= 0);
          draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE, src);
@@ -163,6 +147,9 @@ compute_cliprect(struct cell_context *sp)
 
 
 
+/**
+ * Update derived state, send current state to SPUs prior to rendering.
+ */
 void cell_update_derived( struct cell_context *cell )
 {
    if (cell->dirty & (CELL_NEW_RASTERIZER |
diff --git a/src/gallium/drivers/cell/ppu/cell_state_emit.c b/src/gallium/drivers/cell/ppu/cell_state_emit.c
index 9cae67f091..f2feaa329a 100644
--- a/src/gallium/drivers/cell/ppu/cell_state_emit.c
+++ b/src/gallium/drivers/cell/ppu/cell_state_emit.c
@@ -25,7 +25,7 @@
  * 
  **************************************************************************/
 
-#include "pipe/p_util.h"
+#include "util/u_memory.h"
 #include "cell_context.h"
 #include "cell_state.h"
 #include "cell_state_emit.h"
@@ -47,7 +47,10 @@ emit_state_cmd(struct cell_context *cell, uint cmd,
 }
 
 
-
+/**
+ * For state marked as 'dirty', construct a state-update command block
+ * and insert it into the current batch buffer.
+ */
 void
 cell_emit_state(struct cell_context *cell)
 {
@@ -90,7 +93,8 @@ cell_emit_state(struct cell_context *cell)
          blend.size = (char *) cell->blend->code.csr
              - (char *) cell->blend->code.store;
          blend.read_fb = TRUE;
-      } else {
+      }
+      else {
          blend.base = 0;
          blend.size = 0;
          blend.read_fb = FALSE;
@@ -101,7 +105,6 @@ cell_emit_state(struct cell_context *cell)
 
    if (cell->dirty & CELL_NEW_DEPTH_STENCIL) {
       struct cell_command_depth_stencil_alpha_test dsat;
-      
 
       if (cell->depth_stencil != NULL) {
 	 dsat.base = (intptr_t) cell->depth_stencil->code.store;
@@ -109,15 +112,15 @@ cell_emit_state(struct cell_context *cell)
 	     - (char *) cell->depth_stencil->code.store;
 	 dsat.read_depth = TRUE;
 	 dsat.read_stencil = FALSE;
-      } else {
+      }
+      else {
 	 dsat.base = 0;
 	 dsat.size = 0;
 	 dsat.read_depth = FALSE;
 	 dsat.read_stencil = FALSE;
       }
 
-      emit_state_cmd(cell, CELL_CMD_STATE_DEPTH_STENCIL, &dsat,
-		     sizeof(dsat));
+      emit_state_cmd(cell, CELL_CMD_STATE_DEPTH_STENCIL, &dsat, sizeof(dsat));
    }
 
    if (cell->dirty & CELL_NEW_SAMPLER) {
@@ -162,15 +165,14 @@ cell_emit_state(struct cell_context *cell)
       const struct draw_context *const draw = cell->draw;
       struct cell_shader_info info;
 
-      info.num_outputs = draw->num_vs_outputs;
-      info.declarations = (uintptr_t) draw->machine.Declarations;
-      info.num_declarations = draw->machine.NumDeclarations;
-      info.instructions = (uintptr_t) draw->machine.Instructions;
-      info.num_instructions = draw->machine.NumInstructions;
-      info.immediates = (uintptr_t) draw->machine.Imms;
-      info.num_immediates = draw->machine.ImmLimit / 4;
+      info.num_outputs = draw_num_vs_outputs(draw);
+      info.declarations = (uintptr_t) draw->vs.machine.Declarations;
+      info.num_declarations = draw->vs.machine.NumDeclarations;
+      info.instructions = (uintptr_t) draw->vs.machine.Instructions;
+      info.num_instructions = draw->vs.machine.NumInstructions;
+      info.immediates = (uintptr_t) draw->vs.machine.Imms;
+      info.num_immediates = draw->vs.machine.ImmLimit / 4;
 
-      emit_state_cmd(cell, CELL_CMD_STATE_BIND_VS,
-		     & info, sizeof(info));
+      emit_state_cmd(cell, CELL_CMD_STATE_BIND_VS, &info, sizeof(info));
    }
 }
diff --git a/src/gallium/drivers/cell/ppu/cell_state_shader.c b/src/gallium/drivers/cell/ppu/cell_state_shader.c
index f5707f2bb8..97e44eeb1a 100644
--- a/src/gallium/drivers/cell/ppu/cell_state_shader.c
+++ b/src/gallium/drivers/cell/ppu/cell_state_shader.c
@@ -26,7 +26,7 @@
  **************************************************************************/
 
 #include "pipe/p_defines.h"
-#include "pipe/p_util.h"
+#include "util/u_memory.h"
 #include "pipe/p_inlines.h"
 #include "pipe/p_winsys.h"
 #include "draw/draw_context.h"
@@ -41,7 +41,7 @@
 static INLINE struct cell_fragment_shader_state *
 cell_fragment_shader_state(void *shader)
 {
-   return (struct pipe_shader_state *) shader;
+   return (struct cell_fragment_shader_state *) shader;
 }
 
 
@@ -49,11 +49,14 @@ cell_fragment_shader_state(void *shader)
 static INLINE struct cell_vertex_shader_state *
 cell_vertex_shader_state(void *shader)
 {
-   return (struct pipe_shader_state *) shader;
+   return (struct cell_vertex_shader_state *) shader;
 }
 
 
-
+/**
+ * Create fragment shader state.
+ * Called via pipe->create_fs_state()
+ */
 static void *
 cell_create_fs_state(struct pipe_context *pipe,
                      const struct pipe_shader_state *templ)
@@ -77,6 +80,9 @@ cell_create_fs_state(struct pipe_context *pipe,
 }
 
 
+/**
+ * Called via pipe->bind_fs_state()
+ */
 static void
 cell_bind_fs_state(struct pipe_context *pipe, void *fs)
 {
@@ -88,6 +94,9 @@ cell_bind_fs_state(struct pipe_context *pipe, void *fs)
 }
 
 
+/**
+ * Called via pipe->delete_fs_state()
+ */
 static void
 cell_delete_fs_state(struct pipe_context *pipe, void *fs)
 {
@@ -98,6 +107,10 @@ cell_delete_fs_state(struct pipe_context *pipe, void *fs)
 }
 
 
+/**
+ * Create vertex shader state.
+ * Called via pipe->create_vs_state()
+ */
 static void *
 cell_create_vs_state(struct pipe_context *pipe,
                      const struct pipe_shader_state *templ)
@@ -128,6 +141,9 @@ cell_create_vs_state(struct pipe_context *pipe,
 }
 
 
+/**
+ * Called via pipe->bind_vs_state()
+ */
 static void
 cell_bind_vs_state(struct pipe_context *pipe, void *vs)
 {
@@ -142,6 +158,9 @@ cell_bind_vs_state(struct pipe_context *pipe, void *vs)
 }
 
 
+/**
+ * Called via pipe->delete_vs_state()
+ */
 static void
 cell_delete_vs_state(struct pipe_context *pipe, void *vs)
 {
@@ -154,6 +173,9 @@ cell_delete_vs_state(struct pipe_context *pipe, void *vs)
 }
 
 
+/**
+ * Called via pipe->set_constant_buffer()
+ */
 static void
 cell_set_constant_buffer(struct pipe_context *pipe,
                          uint shader, uint index,
@@ -166,7 +188,7 @@ cell_set_constant_buffer(struct pipe_context *pipe,
    assert(index == 0);
 
    /* note: reference counting */
-   pipe_buffer_reference(ws,
+   winsys_buffer_reference(ws,
                         &cell->constants[shader].buffer,
                         buf->buffer);
    cell->constants[shader].size = buf->size;
diff --git a/src/gallium/drivers/cell/ppu/cell_state_vertex.c b/src/gallium/drivers/cell/ppu/cell_state_vertex.c
index 114684c2a3..fbe55c8472 100644
--- a/src/gallium/drivers/cell/ppu/cell_state_vertex.c
+++ b/src/gallium/drivers/cell/ppu/cell_state_vertex.c
@@ -35,7 +35,7 @@
 #include "draw/draw_context.h"
 
 
-void
+static void
 cell_set_vertex_elements(struct pipe_context *pipe,
                          unsigned count,
                          const struct pipe_vertex_element *elements)
@@ -53,7 +53,7 @@ cell_set_vertex_elements(struct pipe_context *pipe,
 }
 
 
-void
+static void
 cell_set_vertex_buffers(struct pipe_context *pipe,
                         unsigned count,
                         const struct pipe_vertex_buffer *buffers)
@@ -69,3 +69,11 @@ cell_set_vertex_buffers(struct pipe_context *pipe,
 
    draw_set_vertex_buffers(cell->draw, count, buffers);
 }
+
+
+void
+cell_init_vertex_functions(struct cell_context *cell)
+{
+   cell->pipe.set_vertex_buffers = cell_set_vertex_buffers;
+   cell->pipe.set_vertex_elements = cell_set_vertex_elements;
+}
diff --git a/src/gallium/drivers/cell/ppu/cell_surface.c b/src/gallium/drivers/cell/ppu/cell_surface.c
index 5549eb496d..732c64082e 100644
--- a/src/gallium/drivers/cell/ppu/cell_surface.c
+++ b/src/gallium/drivers/cell/ppu/cell_surface.c
@@ -25,106 +25,13 @@
  * 
  **************************************************************************/
 
-#include "pipe/p_defines.h"
-#include "pipe/p_util.h"
-#include "pipe/p_inlines.h"
-#include "pipe/p_winsys.h"
-#include "util/p_tile.h"
+#include "util/u_rect.h"
 #include "cell_context.h"
-#include "cell_surface.h"
-
-
-static void
-cell_surface_copy(struct pipe_context *pipe,
-                  boolean do_flip,
-                  struct pipe_surface *dst,
-                  unsigned dstx, unsigned dsty,
-                  struct pipe_surface *src,
-                  unsigned srcx, unsigned srcy,
-                  unsigned width, unsigned height)
-{
-   assert( dst->cpp == src->cpp );
-
-   pipe_copy_rect(pipe_surface_map(dst),
-                  &dst->block,
-                  dst->stride,
-                  dstx, dsty,
-                  width, height,
-                  pipe_surface_map(src),
-                  do_flip ? -src->stride : src->stride,
-                  srcx, do_flip ? height - 1 - srcy : srcy);
-
-   pipe_surface_unmap(src);
-   pipe_surface_unmap(dst);
-}
-
-
-static void *
-get_pointer(struct pipe_surface *dst, void *dst_map, unsigned x, unsigned y)
-{
-   return (char *)dst_map + y / dst->block.height * dst->stride + x / dst->block.width * dst->block.size;
-}
-
-
-#define UBYTE_TO_USHORT(B) ((B) | ((B) << 8))
-
-
-/**
- * Fill a rectangular sub-region.  Need better logic about when to
- * push buffers into AGP - will currently do so whenever possible.
- */
-static void
-cell_surface_fill(struct pipe_context *pipe,
-                  struct pipe_surface *dst,
-                  unsigned dstx, unsigned dsty,
-                  unsigned width, unsigned height, unsigned value)
-{
-   unsigned i, j;
-   void *dst_map = pipe_surface_map(dst);
-
-   assert(dst->stride > 0);
-
-   switch (dst->block.size) {
-   case 1:
-   case 2:
-   case 4:
-      pipe_fill_rect(dst_map, &dst->block, dst->stride, dstx, dsty, width, height, value);
-      break;
-   case 8:
-      {
-         /* expand the 4-byte clear value to an 8-byte value */
-         ushort *row = (ushort *) get_pointer(dst, dst_map, dstx, dsty);
-         ushort val0 = UBYTE_TO_USHORT((value >>  0) & 0xff);
-         ushort val1 = UBYTE_TO_USHORT((value >>  8) & 0xff);
-         ushort val2 = UBYTE_TO_USHORT((value >> 16) & 0xff);
-         ushort val3 = UBYTE_TO_USHORT((value >> 24) & 0xff);
-         val0 = (val0 << 8) | val0;
-         val1 = (val1 << 8) | val1;
-         val2 = (val2 << 8) | val2;
-         val3 = (val3 << 8) | val3;
-         for (i = 0; i < height; i++) {
-            for (j = 0; j < width; j++) {
-               row[j*4+0] = val0;
-               row[j*4+1] = val1;
-               row[j*4+2] = val2;
-               row[j*4+3] = val3;
-            }
-            row += dst->stride/2;
-         }
-      }
-      break;
-   default:
-      assert(0);
-      break;
-   }
-
-   pipe_surface_unmap( dst );
-}
 
 
 void
 cell_init_surface_functions(struct cell_context *cell)
 {
-   cell->pipe.surface_copy = cell_surface_copy;
-   cell->pipe.surface_fill = cell_surface_fill;
+   cell->pipe.surface_copy = util_surface_copy;
+   cell->pipe.surface_fill = util_surface_fill;
 }
diff --git a/src/gallium/drivers/cell/ppu/cell_texture.c b/src/gallium/drivers/cell/ppu/cell_texture.c
index 533b64227d..b6590dfb86 100644
--- a/src/gallium/drivers/cell/ppu/cell_texture.c
+++ b/src/gallium/drivers/cell/ppu/cell_texture.c
@@ -33,8 +33,9 @@
 #include "pipe/p_context.h"
 #include "pipe/p_defines.h"
 #include "pipe/p_inlines.h"
-#include "pipe/p_util.h"
 #include "pipe/p_winsys.h"
+#include "util/u_math.h"
+#include "util/u_memory.h"
 
 #include "cell_context.h"
 #include "cell_state.h"
@@ -62,17 +63,30 @@ cell_texture_layout(struct cell_texture * spt)
    spt->buffer_size = 0;
 
    for ( level = 0 ; level <= pt->last_level ; level++ ) {
+      unsigned size;
+      unsigned w_tile, h_tile;
+
+      /* width, height, rounded up to tile size */
+      w_tile = align(width, TILE_SIZE);
+      h_tile = align(height, TILE_SIZE);
+
       pt->width[level] = width;
       pt->height[level] = height;
       pt->depth[level] = depth;
-      pt->nblocksx[level] = pf_get_nblocksx(&pt->block, width);  
-      pt->nblocksy[level] = pf_get_nblocksy(&pt->block, height);  
+      pt->nblocksx[level] = pf_get_nblocksx(&pt->block, w_tile);  
+      pt->nblocksy[level] = pf_get_nblocksy(&pt->block, h_tile);  
+
+      spt->stride[level] = pt->nblocksx[level] * pt->block.size;
 
       spt->level_offset[level] = spt->buffer_size;
 
-      spt->buffer_size += (pt->nblocksy[level] *
-			  ((pt->target == PIPE_TEXTURE_CUBE) ? 6 : depth) *
-			  pt->nblocksx[level] * pt->block.size;
+      size = pt->nblocksx[level] * pt->nblocksy[level] * pt->block.size;
+      if (pt->target == PIPE_TEXTURE_CUBE)
+         size *= 6;
+      else
+         size *= depth;
+
+      spt->buffer_size += size;
 
       width  = minify(width);
       height = minify(height);
@@ -82,8 +96,8 @@ cell_texture_layout(struct cell_texture * spt)
 
 
 static struct pipe_texture *
-cell_texture_create_screen(struct pipe_screen *screen,
-                           const struct pipe_texture *templat)
+cell_texture_create(struct pipe_screen *screen,
+                    const struct pipe_texture *templat)
 {
    struct pipe_winsys *ws = screen->winsys;
    struct cell_texture *spt = CALLOC_STRUCT(cell_texture);
@@ -110,8 +124,8 @@ cell_texture_create_screen(struct pipe_screen *screen,
 
 
 static void
-cell_texture_release_screen(struct pipe_screen *screen,
-                            struct pipe_texture **pt)
+cell_texture_release(struct pipe_screen *screen,
+                     struct pipe_texture **pt)
 {
    if (!*pt)
       return;
@@ -127,7 +141,7 @@ cell_texture_release_screen(struct pipe_screen *screen,
       DBG("%s deleting %p\n", __FUNCTION__, (void *) spt);
       */
 
-      pipe_buffer_reference(screen->winsys, &spt->buffer, NULL);
+      pipe_buffer_reference(screen, &spt->buffer, NULL);
 
       FREE(spt);
    }
@@ -135,6 +149,7 @@ cell_texture_release_screen(struct pipe_screen *screen,
 }
 
 
+#if 0
 static void
 cell_texture_update(struct pipe_context *pipe, struct pipe_texture *texture,
                     uint face, uint levelsMask)
@@ -142,12 +157,14 @@ cell_texture_update(struct pipe_context *pipe, struct pipe_texture *texture,
    /* XXX TO DO:  re-tile the texture data ... */
 
 }
+#endif
 
 
 static struct pipe_surface *
-cell_get_tex_surface_screen(struct pipe_screen *screen,
-                            struct pipe_texture *pt,
-                            unsigned face, unsigned level, unsigned zslice)
+cell_get_tex_surface(struct pipe_screen *screen,
+                     struct pipe_texture *pt,
+                     unsigned face, unsigned level, unsigned zslice,
+                     unsigned usage)
 {
    struct pipe_winsys *ws = screen->winsys;
    struct cell_texture *spt = cell_texture(pt);
@@ -157,7 +174,7 @@ cell_get_tex_surface_screen(struct pipe_screen *screen,
    if (ps) {
       assert(ps->refcount);
       assert(ps->winsys);
-      pipe_buffer_reference(ws, &ps->buffer, spt->buffer);
+      winsys_buffer_reference(ws, &ps->buffer, spt->buffer);
       ps->format = pt->format;
       ps->block = pt->block;
       ps->width = pt->width[level];
@@ -166,12 +183,21 @@ cell_get_tex_surface_screen(struct pipe_screen *screen,
       ps->nblocksy = pt->nblocksy[level];
       ps->stride = spt->stride[level];
       ps->offset = spt->level_offset[level];
+      ps->usage = usage;
+
+      /* XXX may need to override usage flags (see sp_texture.c) */
+
+      pipe_texture_reference(&ps->texture, pt); 
+      ps->face = face;
+      ps->level = level;
+      ps->zslice = zslice;
 
       if (pt->target == PIPE_TEXTURE_CUBE || pt->target == PIPE_TEXTURE_3D) {
 	 ps->offset += ((pt->target == PIPE_TEXTURE_CUBE) ? face : zslice) *
 		       ps->nblocksy *
 		       ps->stride;
-      } else {
+      }
+      else {
 	 assert(face == 0);
 	 assert(zslice == 0);
       }
@@ -181,6 +207,11 @@ cell_get_tex_surface_screen(struct pipe_screen *screen,
 
 
 
+/**
+ * Copy tile data from linear layout to tiled layout.
+ * XXX this should be rolled into the future surface-creation code.
+ * XXX also need "untile" code...
+ */
 static void
 tile_copy_data(uint w, uint h, uint tile_size, uint *dst, const uint *src)
 {
@@ -211,6 +242,7 @@ tile_copy_data(uint w, uint h, uint tile_size, uint *dst, const uint *src)
 
 /**
  * Convert linear texture image data to tiled format for SPU usage.
+ * XXX recast this in terms of pipe_surfaces (aka texture views).
  */
 static void
 cell_tile_texture(struct cell_context *cell,
@@ -228,10 +260,11 @@ cell_tile_texture(struct cell_context *cell,
    assert(w % TILE_SIZE == 0);
    assert(h % TILE_SIZE == 0);
 
-   surf = screen->get_tex_surface(screen, &texture->base, face, level, zslice);
+   surf = screen->get_tex_surface(screen, &texture->base, face, level, zslice,
+                                  PIPE_BUFFER_USAGE_CPU_WRITE);
    ASSERT(surf);
 
-   src = (const uint *) pipe_surface_map(surf);
+   src = (const uint *) pipe_surface_map(surf, PIPE_BUFFER_USAGE_CPU_WRITE);
 
    if (texture->tiled_data) {
       align_free(texture->tiled_data);
@@ -246,11 +279,12 @@ cell_tile_texture(struct cell_context *cell,
 }
 
 
-
 void
 cell_update_texture_mapping(struct cell_context *cell)
 {
+#if 0
    uint face = 0, level = 0, zslice = 0;
+#endif
    uint i;
 
    for (i = 0; i < CELL_MAX_SAMPLERS; i++) {
@@ -275,16 +309,79 @@ cell_update_texture_mapping(struct cell_context *cell)
 }
 
 
+static void 
+cell_tex_surface_release(struct pipe_screen *screen, 
+                         struct pipe_surface **s)
+{
+   /* Effectively do the texture_update work here - if texture images
+    * needed post-processing to put them into hardware layout, this is
+    * where it would happen.  For softpipe, nothing to do.
+    */
+   assert ((*s)->texture);
+   pipe_texture_reference(&(*s)->texture, NULL); 
+
+   screen->winsys->surface_release(screen->winsys, s);
+}
+
+
+static void *
+cell_surface_map( struct pipe_screen *screen,
+                  struct pipe_surface *surface,
+                  unsigned flags )
+{
+   ubyte *map;
+
+   if (flags & ~surface->usage) {
+      assert(0);
+      return NULL;
+   }
+
+   map = pipe_buffer_map( screen, surface->buffer, flags );
+   if (map == NULL)
+      return NULL;
+
+   /* May want to different things here depending on read/write nature
+    * of the map:
+    */
+   if (surface->texture &&
+       (flags & PIPE_BUFFER_USAGE_CPU_WRITE)) 
+   {
+      /* Do something to notify sharing contexts of a texture change.
+       * In softpipe, that would mean flushing the texture cache.
+       */
+#if 0
+      cell_screen(screen)->timestamp++;
+#endif
+   }
+   
+   return map + surface->offset;
+}
+
+
+static void
+cell_surface_unmap(struct pipe_screen *screen,
+                   struct pipe_surface *surface)
+{
+   pipe_buffer_unmap( screen, surface->buffer );
+}
+
+
 void
 cell_init_texture_functions(struct cell_context *cell)
 {
-   cell->pipe.texture_update = cell_texture_update;
+   /*cell->pipe.texture_update = cell_texture_update;*/
 }
 
+
 void
 cell_init_screen_texture_funcs(struct pipe_screen *screen)
 {
-   screen->texture_create = cell_texture_create_screen;
-   screen->texture_release = cell_texture_release_screen;
-   screen->get_tex_surface = cell_get_tex_surface_screen;
+   screen->texture_create = cell_texture_create;
+   screen->texture_release = cell_texture_release;
+
+   screen->get_tex_surface = cell_get_tex_surface;
+   screen->tex_surface_release = cell_tex_surface_release;
+
+   screen->surface_map = cell_surface_map;
+   screen->surface_unmap = cell_surface_unmap;
 }
diff --git a/src/gallium/drivers/cell/ppu/cell_texture.h b/src/gallium/drivers/cell/ppu/cell_texture.h
index fcee069d05..6d37e95ebc 100644
--- a/src/gallium/drivers/cell/ppu/cell_texture.h
+++ b/src/gallium/drivers/cell/ppu/cell_texture.h
@@ -41,6 +41,7 @@ struct cell_texture
    struct pipe_texture base;
 
    unsigned long level_offset[PIPE_MAX_TEXTURE_LEVELS];
+   unsigned long stride[PIPE_MAX_TEXTURE_LEVELS];
 
    /* The data is held here:
     */
diff --git a/src/gallium/drivers/cell/ppu/cell_vbuf.c b/src/gallium/drivers/cell/ppu/cell_vbuf.c
index 3a181b585c..aa63435b93 100644
--- a/src/gallium/drivers/cell/ppu/cell_vbuf.c
+++ b/src/gallium/drivers/cell/ppu/cell_vbuf.c
@@ -26,6 +26,11 @@
  **************************************************************************/
 
 /**
+ * Vertex buffer code.  The draw module transforms vertices to window
+ * coords, etc. and emits the vertices into buffer supplied by this module.
+ * When a vertex buffer is full, or we flush, we'll send the vertex data
+ * to the SPUs.
+ *
  * Authors
  *  Brian Paul
  */
@@ -37,6 +42,7 @@
 #include "cell_spu.h"
 #include "cell_vbuf.h"
 #include "draw/draw_vbuf.h"
+#include "util/u_memory.h"
 
 
 /** Allow vertex data to be inlined after RENDER command */
@@ -112,7 +118,7 @@ cell_vbuf_release_vertices(struct vbuf_render *vbr, void *vertices,
    }
 
    cvbr->vertex_buf = ~0;
-   cell_flush_int(&cell->pipe, 0x0);
+   cell_flush_int(cell, 0x0);
 
    assert(vertices == cvbr->vertex_buffer);
    cvbr->vertex_buffer = NULL;
@@ -120,12 +126,13 @@ cell_vbuf_release_vertices(struct vbuf_render *vbr, void *vertices,
 
 
 
-static void
+static boolean
 cell_vbuf_set_primitive(struct vbuf_render *vbr, unsigned prim)
 {
    struct cell_vbuf_render *cvbr = cell_vbuf_render(vbr);
    cvbr->prim = prim;
    /*printf("cell_set_prim %u\n", prim);*/
+   return TRUE;
 }
 
 
@@ -243,7 +250,7 @@ cell_vbuf_draw(struct vbuf_render *vbr,
 
 #if 0
    /* helpful for debug */
-   cell_flush_int(&cell->pipe, CELL_FLUSH_WAIT);
+   cell_flush_int(cell, CELL_FLUSH_WAIT);
 #endif
 }
 
diff --git a/src/gallium/drivers/cell/ppu/cell_vertex_fetch.c b/src/gallium/drivers/cell/ppu/cell_vertex_fetch.c
index 49d5443cde..2ece0250f6 100644
--- a/src/gallium/drivers/cell/ppu/cell_vertex_fetch.c
+++ b/src/gallium/drivers/cell/ppu/cell_vertex_fetch.c
@@ -260,6 +260,7 @@ emit_fetch(struct spe_function *p,
 
 void cell_update_vertex_fetch(struct draw_context *draw)
 {
+#if 0
    struct cell_context *const cell =
        (struct cell_context *) draw->driver_private;
    struct spe_function *p = &cell->attrib_fetch;
@@ -337,4 +338,7 @@ void cell_update_vertex_fetch(struct draw_context *draw)
 	     cell->attrib_fetch_offsets[function_index[i]];
       }
    }
+#else
+   assert(0);
+#endif
 }
diff --git a/src/gallium/drivers/cell/ppu/cell_vertex_shader.c b/src/gallium/drivers/cell/ppu/cell_vertex_shader.c
index f753960a0f..2b10c116fa 100644
--- a/src/gallium/drivers/cell/ppu/cell_vertex_shader.c
+++ b/src/gallium/drivers/cell/ppu/cell_vertex_shader.c
@@ -32,6 +32,7 @@
 #include "pipe/p_defines.h"
 #include "pipe/p_context.h"
 #include "pipe/p_winsys.h"
+#include "util/u_math.h"
 
 #include "cell_context.h"
 #include "cell_draw_arrays.h"
@@ -50,6 +51,7 @@
 void
 cell_vertex_shader_queue_flush(struct draw_context *draw)
 {
+#if 0
    struct cell_context *const cell =
        (struct cell_context *) draw->driver_private;
    struct cell_command_vs *const vs = &cell_global.command[0].vs;
@@ -133,9 +135,12 @@ cell_vertex_shader_queue_flush(struct draw_context *draw)
       vs->num_elts = n;
       send_mbox_message(cell_global.spe_contexts[0], CELL_CMD_VS_EXECUTE);
 
-      cell_flush_int(& cell->pipe, CELL_FLUSH_WAIT);
+      cell_flush_int(cell, CELL_FLUSH_WAIT);
    }
 
    draw->vs.post_nr = draw->vs.queue_nr;
    draw->vs.queue_nr = 0;
+#else
+   assert(0);
+#endif
 }
diff --git a/src/gallium/drivers/cell/ppu/cell_winsys.c b/src/gallium/drivers/cell/ppu/cell_winsys.c
index ebabce3c8f..d570bbd2f9 100644
--- a/src/gallium/drivers/cell/ppu/cell_winsys.c
+++ b/src/gallium/drivers/cell/ppu/cell_winsys.c
@@ -26,7 +26,7 @@
  **************************************************************************/
 
 
-#include "pipe/p_util.h"
+#include "util/u_memory.h"
 #include "cell_winsys.h"
 
 
diff --git a/src/gallium/drivers/cell/spu/Makefile b/src/gallium/drivers/cell/spu/Makefile
index 8e83610790..d49abb2e82 100644
--- a/src/gallium/drivers/cell/spu/Makefile
+++ b/src/gallium/drivers/cell/spu/Makefile
@@ -5,7 +5,7 @@
 
 
 TOP = ../../../../..
-include $(TOP)/configs/linux-cell
+include $(TOP)/configs/current
 
 
 PROG = g3d
diff --git a/src/gallium/drivers/cell/spu/spu_exec.c b/src/gallium/drivers/cell/spu/spu_exec.c
index 42e5022f30..e27df2dfb3 100644
--- a/src/gallium/drivers/cell/spu/spu_exec.c
+++ b/src/gallium/drivers/cell/spu/spu_exec.c
@@ -63,7 +63,6 @@
 
 #include "pipe/p_compiler.h"
 #include "pipe/p_state.h"
-#include "pipe/p_util.h"
 #include "pipe/p_shader_tokens.h"
 #include "tgsi/tgsi_parse.h"
 #include "tgsi/tgsi_util.h"
@@ -383,10 +382,10 @@ fetch_src_file_channel(
          break;
 
       case TGSI_FILE_IMMEDIATE:
-         assert( index->i[0] < (int) mach->ImmLimit );
-         assert( index->i[1] < (int) mach->ImmLimit );
-         assert( index->i[2] < (int) mach->ImmLimit );
-         assert( index->i[3] < (int) mach->ImmLimit );
+         ASSERT( index->i[0] < (int) mach->ImmLimit );
+         ASSERT( index->i[1] < (int) mach->ImmLimit );
+         ASSERT( index->i[2] < (int) mach->ImmLimit );
+         ASSERT( index->i[3] < (int) mach->ImmLimit );
 
          chan->f[0] = mach->Imms[index->i[0]][swizzle];
          chan->f[1] = mach->Imms[index->i[1]][swizzle];
@@ -410,7 +409,7 @@ fetch_src_file_channel(
          break;
 
       default:
-         assert( 0 );
+         ASSERT( 0 );
       }
       break;
 
@@ -423,7 +422,7 @@ fetch_src_file_channel(
       break;
 
    default:
-      assert( 0 );
+      ASSERT( 0 );
    }
 }
 
@@ -472,7 +471,7 @@ fetch_source(
          index.q = si_shli(index.q, 12);
          break;
       default:
-         assert( 0 );
+         ASSERT( 0 );
       }
 
       index.i[0] += reg->SrcRegisterDim.Index;
@@ -559,7 +558,7 @@ store_dest(
       break;
 
    default:
-      assert( 0 );
+      ASSERT( 0 );
       return;
    }
 
@@ -583,11 +582,11 @@ store_dest(
       break;
 
    case TGSI_SAT_MINUS_PLUS_ONE:
-      assert( 0 );
+      ASSERT( 0 );
       break;
 
    default:
-      assert( 0 );
+      ASSERT( 0 );
    }
 }
 
@@ -770,7 +769,7 @@ exec_tex(struct spu_exec_machine *mach,
       break;
 
    default:
-      assert (0);
+      ASSERT (0);
    }
 
    FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
@@ -862,7 +861,7 @@ exec_declaration(struct spu_exec_machine *mach,
             break;
 
          default:
-            assert( 0 );
+            ASSERT( 0 );
          }
 
          if( mask == TGSI_WRITEMASK_XYZW ) {
@@ -972,11 +971,11 @@ exec_instruction(
       break;
 
    case TGSI_OPCODE_EXP:
-      assert (0);
+      ASSERT (0);
       break;
 
    case TGSI_OPCODE_LOG:
-      assert (0);
+      ASSERT (0);
       break;
 
    case TGSI_OPCODE_MUL:
@@ -1152,24 +1151,24 @@ exec_instruction(
       break;
 
    case TGSI_OPCODE_CND:
-      assert (0);
+      ASSERT (0);
       break;
 
    case TGSI_OPCODE_CND0:
-      assert (0);
+      ASSERT (0);
       break;
 
    case TGSI_OPCODE_DOT2ADD:
       /* TGSI_OPCODE_DP2A */
-      assert (0);
+      ASSERT (0);
       break;
 
    case TGSI_OPCODE_INDEX:
-      assert (0);
+      ASSERT (0);
       break;
 
    case TGSI_OPCODE_NEGATE:
-      assert (0);
+      ASSERT (0);
       break;
 
    case TGSI_OPCODE_FRAC:
@@ -1182,7 +1181,7 @@ exec_instruction(
       break;
 
    case TGSI_OPCODE_CLAMP:
-      assert (0);
+      ASSERT (0);
       break;
 
    case TGSI_OPCODE_FLOOR:
@@ -1277,7 +1276,7 @@ exec_instruction(
       break;
 
     case TGSI_OPCODE_MULTIPLYMATRIX:
-       assert (0);
+       ASSERT (0);
        break;
 
     case TGSI_OPCODE_ABS:
@@ -1291,7 +1290,7 @@ exec_instruction(
        break;
 
    case TGSI_OPCODE_RCC:
-      assert (0);
+      ASSERT (0);
       break;
 
    case TGSI_OPCODE_DPH:
@@ -1354,23 +1353,23 @@ exec_instruction(
       break;
 
    case TGSI_OPCODE_PK2H:
-      assert (0);
+      ASSERT (0);
       break;
 
    case TGSI_OPCODE_PK2US:
-      assert (0);
+      ASSERT (0);
       break;
 
    case TGSI_OPCODE_PK4B:
-      assert (0);
+      ASSERT (0);
       break;
 
    case TGSI_OPCODE_PK4UB:
-      assert (0);
+      ASSERT (0);
       break;
 
    case TGSI_OPCODE_RFL:
-      assert (0);
+      ASSERT (0);
       break;
 
    case TGSI_OPCODE_SEQ:
@@ -1385,7 +1384,7 @@ exec_instruction(
       break;
 
    case TGSI_OPCODE_SFL:
-      assert (0);
+      ASSERT (0);
       break;
 
    case TGSI_OPCODE_SGT:
@@ -1430,7 +1429,7 @@ exec_instruction(
       break;
 
    case TGSI_OPCODE_STR:
-      assert (0);
+      ASSERT (0);
       break;
 
    case TGSI_OPCODE_TEX:
@@ -1453,7 +1452,7 @@ exec_instruction(
       /* src[1] = d[strq]/dx */
       /* src[2] = d[strq]/dy */
       /* src[3] = sampler unit */
-      assert (0);
+      ASSERT (0);
       break;
 
    case TGSI_OPCODE_TXL:
@@ -1471,35 +1470,35 @@ exec_instruction(
       break;
 
    case TGSI_OPCODE_UP2H:
-      assert (0);
+      ASSERT (0);
       break;
 
    case TGSI_OPCODE_UP2US:
-      assert (0);
+      ASSERT (0);
       break;
 
    case TGSI_OPCODE_UP4B:
-      assert (0);
+      ASSERT (0);
       break;
 
    case TGSI_OPCODE_UP4UB:
-      assert (0);
+      ASSERT (0);
       break;
 
    case TGSI_OPCODE_X2D:
-      assert (0);
+      ASSERT (0);
       break;
 
    case TGSI_OPCODE_ARA:
-      assert (0);
+      ASSERT (0);
       break;
 
    case TGSI_OPCODE_ARR:
-      assert (0);
+      ASSERT (0);
       break;
 
    case TGSI_OPCODE_BRA:
-      assert (0);
+      ASSERT (0);
       break;
 
    case TGSI_OPCODE_CAL:
@@ -1508,14 +1507,14 @@ exec_instruction(
          /* do the call */
 
          /* push the Cond, Loop, Cont stacks */
-         assert(mach->CondStackTop < TGSI_EXEC_MAX_COND_NESTING);
+         ASSERT(mach->CondStackTop < TGSI_EXEC_MAX_COND_NESTING);
          mach->CondStack[mach->CondStackTop++] = mach->CondMask;
-         assert(mach->LoopStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
+         ASSERT(mach->LoopStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
          mach->LoopStack[mach->LoopStackTop++] = mach->LoopMask;
-         assert(mach->ContStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
+         ASSERT(mach->ContStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
          mach->ContStack[mach->ContStackTop++] = mach->ContMask;
 
-         assert(mach->FuncStackTop < TGSI_EXEC_MAX_CALL_NESTING);
+         ASSERT(mach->FuncStackTop < TGSI_EXEC_MAX_CALL_NESTING);
          mach->FuncStack[mach->FuncStackTop++] = mach->FuncMask;
 
          /* note that PC was already incremented above */
@@ -1539,13 +1538,13 @@ exec_instruction(
          *pc = mach->CallStack[--mach->CallStackTop];
 
          /* pop the Cond, Loop, Cont stacks */
-         assert(mach->CondStackTop > 0);
+         ASSERT(mach->CondStackTop > 0);
          mach->CondMask = mach->CondStack[--mach->CondStackTop];
-         assert(mach->LoopStackTop > 0);
+         ASSERT(mach->LoopStackTop > 0);
          mach->LoopMask = mach->LoopStack[--mach->LoopStackTop];
-         assert(mach->ContStackTop > 0);
+         ASSERT(mach->ContStackTop > 0);
          mach->ContMask = mach->ContStack[--mach->ContStackTop];
-         assert(mach->FuncStackTop > 0);
+         ASSERT(mach->FuncStackTop > 0);
          mach->FuncMask = mach->FuncStack[--mach->FuncStackTop];
 
          UPDATE_EXEC_MASK(mach);
@@ -1553,7 +1552,7 @@ exec_instruction(
       break;
 
    case TGSI_OPCODE_SSG:
-      assert (0);
+      ASSERT (0);
       break;
 
    case TGSI_OPCODE_CMP:
@@ -1593,11 +1592,11 @@ exec_instruction(
       break;
 
    case TGSI_OPCODE_NRM:
-      assert (0);
+      ASSERT (0);
       break;
 
    case TGSI_OPCODE_DIV:
-      assert( 0 );
+      ASSERT( 0 );
       break;
 
    case TGSI_OPCODE_DP2:
@@ -1616,7 +1615,7 @@ exec_instruction(
 
    case TGSI_OPCODE_IF:
       /* push CondMask */
-      assert(mach->CondStackTop < TGSI_EXEC_MAX_COND_NESTING);
+      ASSERT(mach->CondStackTop < TGSI_EXEC_MAX_COND_NESTING);
       mach->CondStack[mach->CondStackTop++] = mach->CondMask;
       FETCH( &r[0], 0, CHAN_X );
       /* update CondMask */
@@ -1640,7 +1639,7 @@ exec_instruction(
       /* invert CondMask wrt previous mask */
       {
          uint prevMask;
-         assert(mach->CondStackTop > 0);
+         ASSERT(mach->CondStackTop > 0);
          prevMask = mach->CondStack[mach->CondStackTop - 1];
          mach->CondMask = ~mach->CondMask & prevMask;
          UPDATE_EXEC_MASK(mach);
@@ -1650,7 +1649,7 @@ exec_instruction(
 
    case TGSI_OPCODE_ENDIF:
       /* pop CondMask */
-      assert(mach->CondStackTop > 0);
+      ASSERT(mach->CondStackTop > 0);
       mach->CondMask = mach->CondStack[--mach->CondStackTop];
       UPDATE_EXEC_MASK(mach);
       break;
@@ -1661,19 +1660,19 @@ exec_instruction(
       break;
 
    case TGSI_OPCODE_REP:
-      assert (0);
+      ASSERT (0);
       break;
 
    case TGSI_OPCODE_ENDREP:
-       assert (0);
+       ASSERT (0);
        break;
 
    case TGSI_OPCODE_PUSHA:
-      assert (0);
+      ASSERT (0);
       break;
 
    case TGSI_OPCODE_POPA:
-      assert (0);
+      ASSERT (0);
       break;
 
    case TGSI_OPCODE_CEIL:
@@ -1747,7 +1746,7 @@ exec_instruction(
       break;
 
    case TGSI_OPCODE_MOD:
-      assert (0);
+      ASSERT (0);
       break;
 
    case TGSI_OPCODE_XOR:
@@ -1760,15 +1759,15 @@ exec_instruction(
       break;
 
    case TGSI_OPCODE_SAD:
-      assert (0);
+      ASSERT (0);
       break;
 
    case TGSI_OPCODE_TXF:
-      assert (0);
+      ASSERT (0);
       break;
 
    case TGSI_OPCODE_TXQ:
-      assert (0);
+      ASSERT (0);
       break;
 
    case TGSI_OPCODE_EMIT:
@@ -1785,9 +1784,9 @@ exec_instruction(
       /* fall-through (for now) */
    case TGSI_OPCODE_BGNLOOP2:
       /* push LoopMask and ContMasks */
-      assert(mach->LoopStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
+      ASSERT(mach->LoopStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
       mach->LoopStack[mach->LoopStackTop++] = mach->LoopMask;
-      assert(mach->ContStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
+      ASSERT(mach->ContStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
       mach->ContStack[mach->ContStackTop++] = mach->ContMask;
       break;
 
@@ -1795,7 +1794,7 @@ exec_instruction(
       /* fall-through (for now at least) */
    case TGSI_OPCODE_ENDLOOP2:
       /* Restore ContMask, but don't pop */
-      assert(mach->ContStackTop > 0);
+      ASSERT(mach->ContStackTop > 0);
       mach->ContMask = mach->ContStack[mach->ContStackTop - 1];
       if (mach->LoopMask) {
          /* repeat loop: jump to instruction just past BGNLOOP */
@@ -1803,10 +1802,10 @@ exec_instruction(
       }
       else {
          /* exit loop: pop LoopMask */
-         assert(mach->LoopStackTop > 0);
+         ASSERT(mach->LoopStackTop > 0);
          mach->LoopMask = mach->LoopStack[--mach->LoopStackTop];
          /* pop ContMask */
-         assert(mach->ContStackTop > 0);
+         ASSERT(mach->ContStackTop > 0);
          mach->ContMask = mach->ContStack[--mach->ContStackTop];
       }
       UPDATE_EXEC_MASK(mach);
@@ -1835,26 +1834,26 @@ exec_instruction(
       break;
 
    case TGSI_OPCODE_NOISE1:
-      assert( 0 );
+      ASSERT( 0 );
       break;
 
    case TGSI_OPCODE_NOISE2:
-      assert( 0 );
+      ASSERT( 0 );
       break;
 
    case TGSI_OPCODE_NOISE3:
-      assert( 0 );
+      ASSERT( 0 );
       break;
 
    case TGSI_OPCODE_NOISE4:
-      assert( 0 );
+      ASSERT( 0 );
       break;
 
    case TGSI_OPCODE_NOP:
       break;
 
    default:
-      assert( 0 );
+      ASSERT( 0 );
    }
 }
 
@@ -1875,11 +1874,11 @@ spu_exec_machine_run( struct spu_exec_machine *mach )
    mach->FuncMask = 0xf;
    mach->ExecMask = 0xf;
 
-   mach->CondStackTop = 0; /* temporarily subvert this assertion */
-   assert(mach->CondStackTop == 0);
-   assert(mach->LoopStackTop == 0);
-   assert(mach->ContStackTop == 0);
-   assert(mach->CallStackTop == 0);
+   mach->CondStackTop = 0; /* temporarily subvert this ASSERTion */
+   ASSERT(mach->CondStackTop == 0);
+   ASSERT(mach->LoopStackTop == 0);
+   ASSERT(mach->ContStackTop == 0);
+   ASSERT(mach->CallStackTop == 0);
 
    mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0] = 0;
    mach->Temps[TEMP_OUTPUT_I].xyzw[TEMP_OUTPUT_C].u[0] = 0;
diff --git a/src/gallium/drivers/cell/spu/spu_exec.h b/src/gallium/drivers/cell/spu/spu_exec.h
index c68f78f59b..8605679940 100644
--- a/src/gallium/drivers/cell/spu/spu_exec.h
+++ b/src/gallium/drivers/cell/spu/spu_exec.h
@@ -99,7 +99,7 @@ struct spu_exec_machine
     * 1  address
     */
    struct spu_exec_vector       Temps[TGSI_EXEC_NUM_TEMPS 
-				      + TGSI_EXEC_NUM_ADDRS + 1]
+                                      + TGSI_EXEC_NUM_TEMP_EXTRAS + 1]
        ALIGN16_ATTRIB;
 
    struct spu_exec_vector       *Addrs;
diff --git a/src/gallium/drivers/cell/spu/spu_main.c b/src/gallium/drivers/cell/spu/spu_main.c
index e04ffeb9b1..d223f32d94 100644
--- a/src/gallium/drivers/cell/spu/spu_main.c
+++ b/src/gallium/drivers/cell/spu/spu_main.c
@@ -55,6 +55,10 @@ struct spu_global spu;
 
 struct spu_vs_context draw;
 
+
+/**
+ * Buffers containing dynamically generated SPU code:
+ */
 static unsigned char attribute_fetch_code_buffer[136 * PIPE_MAX_ATTRIBS]
     ALIGN16_ATTRIB;
 
@@ -136,54 +140,75 @@ really_clear_tiles(uint surfaceIndex)
 static void
 cmd_clear_surface(const struct cell_command_clear_surface *clear)
 {
-   const uint num_tiles = spu.fb.width_tiles * spu.fb.height_tiles;
-   uint i;
-
    if (Debug)
       printf("SPU %u: CLEAR SURF %u to 0x%08x\n", spu.init.id,
              clear->surface, clear->value);
 
-#define CLEAR_OPT 1
-#if CLEAR_OPT
-   /* set all tile's status to CLEAR */
    if (clear->surface == 0) {
-      memset(spu.ctile_status, TILE_STATUS_CLEAR, sizeof(spu.ctile_status));
       spu.fb.color_clear_value = clear->value;
+      if (spu.init.debug_flags & CELL_DEBUG_CHECKER) {
+         uint x = (spu.init.id << 4) | (spu.init.id << 12) |
+            (spu.init.id << 20) | (spu.init.id << 28);
+         spu.fb.color_clear_value ^= x;
+      }
    }
    else {
-      memset(spu.ztile_status, TILE_STATUS_CLEAR, sizeof(spu.ztile_status));
       spu.fb.depth_clear_value = clear->value;
    }
-   return;
-#endif
 
+#define CLEAR_OPT 1
+#if CLEAR_OPT
+
+   /* Simply set all tiles' status to CLEAR.
+    * When we actually begin rendering into a tile, we'll initialize it to
+    * the clear value.  If any tiles go untouched during the frame,
+    * really_clear_tiles() will set them to the clear value.
+    */
    if (clear->surface == 0) {
-      spu.fb.color_clear_value = clear->value;
-      clear_c_tile(&spu.ctile);
+      memset(spu.ctile_status, TILE_STATUS_CLEAR, sizeof(spu.ctile_status));
    }
    else {
-      spu.fb.depth_clear_value = clear->value;
-      clear_z_tile(&spu.ztile);
+      memset(spu.ztile_status, TILE_STATUS_CLEAR, sizeof(spu.ztile_status));
    }
 
+#else
+
+   /*
+    * This path clears the whole framebuffer to the clear color right now.
+    */
+
    /*
    printf("SPU: %s num=%d w=%d h=%d\n",
           __FUNCTION__, num_tiles, spu.fb.width_tiles, spu.fb.height_tiles);
    */
 
-   for (i = spu.init.id; i < num_tiles; i += spu.init.num_spus) {
-      uint tx = i % spu.fb.width_tiles;
-      uint ty = i / spu.fb.width_tiles;
-      if (clear->surface == 0)
-         put_tile(tx, ty, &spu.ctile, TAG_SURFACE_CLEAR, 0);
-      else
-         put_tile(tx, ty, &spu.ztile, TAG_SURFACE_CLEAR, 1);
-      /* XXX we don't want this here, but it fixes bad tile results */
+   /* init a single tile to the clear value */
+   if (clear->surface == 0) {
+      clear_c_tile(&spu.ctile);
+   }
+   else {
+      clear_z_tile(&spu.ztile);
    }
 
-#if 0
-   wait_on_mask(1 << TAG_SURFACE_CLEAR);
-#endif
+   /* walk over my tiles, writing the 'clear' tile's data */
+   {
+      const uint num_tiles = spu.fb.width_tiles * spu.fb.height_tiles;
+      uint i;
+      for (i = spu.init.id; i < num_tiles; i += spu.init.num_spus) {
+         uint tx = i % spu.fb.width_tiles;
+         uint ty = i / spu.fb.width_tiles;
+         if (clear->surface == 0)
+            put_tile(tx, ty, &spu.ctile, TAG_SURFACE_CLEAR, 0);
+         else
+            put_tile(tx, ty, &spu.ztile, TAG_SURFACE_CLEAR, 1);
+      }
+   }
+
+   if (spu.init.debug_flags & CELL_DEBUG_SYNC) {
+      wait_on_mask(1 << TAG_SURFACE_CLEAR);
+   }
+
+#endif /* CLEAR_OPT */
 
    if (Debug)
       printf("SPU %u: CLEAR SURF done\n", spu.init.id);
@@ -312,6 +337,21 @@ cmd_state_depth_stencil(const struct cell_command_depth_stencil_alpha_test *stat
 
 
 static void
+cmd_state_logicop(const struct cell_command_logicop * code)
+{
+   mfc_get(logicop_code_buffer,
+           (unsigned int) code->base,  /* src */
+           code->size,
+           TAG_BATCH_BUFFER,
+           0, /* tid */
+           0  /* rid */);
+   wait_on_mask(1 << TAG_BATCH_BUFFER);
+
+   spu.logicop = (logicop_func) logicop_code_buffer;
+}
+
+
+static void
 cmd_state_sampler(const struct cell_command_sampler *sampler)
 {
    if (Debug)
@@ -381,6 +421,21 @@ cmd_state_vs_array_info(const struct cell_array_info *vs_info)
 
 
 static void
+cmd_state_attrib_fetch(const struct cell_attribute_fetch_code *code)
+{
+   mfc_get(attribute_fetch_code_buffer,
+           (unsigned int) code->base,  /* src */
+           code->size,
+           TAG_BATCH_BUFFER,
+           0, /* tid */
+           0  /* rid */);
+   wait_on_mask(1 << TAG_BATCH_BUFFER);
+
+   draw.vertex_fetch.code = attribute_fetch_code_buffer;
+}
+
+
+static void
 cmd_finish(void)
 {
    if (Debug)
@@ -518,38 +573,15 @@ cmd_batch(uint opcode)
                                 (struct cell_shader_info *) &buffer[pos+1]);
          pos += (1 + ROUNDUP8(sizeof(struct cell_shader_info)) / 8);
          break;
-      case CELL_CMD_STATE_ATTRIB_FETCH: {
-         struct cell_attribute_fetch_code *code =
-             (struct cell_attribute_fetch_code *) &buffer[pos+1];
-
-              mfc_get(attribute_fetch_code_buffer,
-                      (unsigned int) code->base,  /* src */
-                      code->size,
-                      TAG_BATCH_BUFFER,
-                      0, /* tid */
-                      0  /* rid */);
-         wait_on_mask(1 << TAG_BATCH_BUFFER);
-
-         draw.vertex_fetch.code = attribute_fetch_code_buffer;
+      case CELL_CMD_STATE_ATTRIB_FETCH:
+         cmd_state_attrib_fetch((struct cell_attribute_fetch_code *)
+                                &buffer[pos+1]);
          pos += (1 + ROUNDUP8(sizeof(struct cell_attribute_fetch_code)) / 8);
          break;
-      }
-      case CELL_CMD_STATE_LOGICOP: {
-         struct cell_command_logicop *code =
-             (struct cell_command_logicop *) &buffer[pos+1];
-
-              mfc_get(logicop_code_buffer,
-                      (unsigned int) code->base,  /* src */
-                      code->size,
-                      TAG_BATCH_BUFFER,
-                      0, /* tid */
-                      0  /* rid */);
-         wait_on_mask(1 << TAG_BATCH_BUFFER);
-
-	 spu.logicop = (logicop_func) logicop_code_buffer;
+      case CELL_CMD_STATE_LOGICOP:
+         cmd_state_logicop((struct cell_command_logicop *) &buffer[pos+1]);
          pos += (1 + ROUNDUP8(sizeof(struct cell_command_logicop)) / 8);
          break;
-      }
       case CELL_CMD_FLUSH_BUFFER_RANGE: {
 	 struct cell_buffer_range *br = (struct cell_buffer_range *)
 	     &buffer[pos+1];
diff --git a/src/gallium/drivers/cell/spu/spu_main.h b/src/gallium/drivers/cell/spu/spu_main.h
index e962e1426c..4879f8c9c8 100644
--- a/src/gallium/drivers/cell/spu/spu_main.h
+++ b/src/gallium/drivers/cell/spu/spu_main.h
@@ -124,13 +124,13 @@ struct spu_global
    struct spu_framebuffer fb;
    boolean read_depth;
    boolean read_stencil;
-   frag_test_func frag_test;
+   frag_test_func frag_test;  /**< Current depth/stencil test code */
    
-   boolean read_fb;
-   blend_func blend;
+   boolean read_fb;   /**< Does current blend mode require framebuffer read? */
+   blend_func blend;  /**< Current blend code */
    qword const_blend_color[4] ALIGN16_ATTRIB;
 
-   logicop_func logicop;
+   logicop_func logicop;  /**< Current logicop code **/
 
    struct pipe_sampler_state sampler[PIPE_MAX_SAMPLERS];
    struct spu_texture texture[PIPE_MAX_SAMPLERS];
diff --git a/src/gallium/drivers/cell/spu/spu_per_fragment_op.c b/src/gallium/drivers/cell/spu/spu_per_fragment_op.c
index b4cffeeb32..c0a729b3d2 100644
--- a/src/gallium/drivers/cell/spu/spu_per_fragment_op.c
+++ b/src/gallium/drivers/cell/spu/spu_per_fragment_op.c
@@ -95,7 +95,7 @@ read_ds_quad(tile_t *buffer, unsigned x, unsigned y,
 
 
    default:
-      assert(0);
+      ASSERT(0);
       break;
    }
 }
@@ -153,7 +153,7 @@ write_ds_quad(tile_t *buffer, unsigned x, unsigned y,
 
 
    default:
-      assert(0);
+      ASSERT(0);
       break;
    }
 }
diff --git a/src/gallium/drivers/cell/spu/spu_render.c b/src/gallium/drivers/cell/spu/spu_render.c
index 6df59abd36..305dc98881 100644
--- a/src/gallium/drivers/cell/spu/spu_render.c
+++ b/src/gallium/drivers/cell/spu/spu_render.c
@@ -35,7 +35,7 @@
 #include "spu_tri.h"
 #include "spu_tile.h"
 #include "cell/common.h"
-
+#include "util/u_memory.h"
 
 
 /**
diff --git a/src/gallium/drivers/cell/spu/spu_tile.c b/src/gallium/drivers/cell/spu/spu_tile.c
index 12dc246328..216a33126b 100644
--- a/src/gallium/drivers/cell/spu/spu_tile.c
+++ b/src/gallium/drivers/cell/spu/spu_tile.c
@@ -31,6 +31,9 @@
 #include "spu_main.h"
 
 
+/**
+ * Get tile of color or Z values from main memory, put into SPU memory.
+ */
 void
 get_tile(uint tx, uint ty, tile_t *tile, int tag, int zBuf)
 {
@@ -56,6 +59,9 @@ get_tile(uint tx, uint ty, tile_t *tile, int tag, int zBuf)
 }
 
 
+/**
+ * Move tile of color or Z values from SPU memory to main memory.
+ */
 void
 put_tile(uint tx, uint ty, const tile_t *tile, int tag, int zBuf)
 {
diff --git a/src/gallium/drivers/cell/spu/spu_tri.c b/src/gallium/drivers/cell/spu/spu_tri.c
index ab4ff8160a..2a4e0b423c 100644
--- a/src/gallium/drivers/cell/spu/spu_tri.c
+++ b/src/gallium/drivers/cell/spu/spu_tri.c
@@ -32,7 +32,7 @@
 #include <transpose_matrix4x4.h>
 #include "pipe/p_compiler.h"
 #include "pipe/p_format.h"
-#include "pipe/p_util.h"
+#include "util/u_math.h"
 #include "spu_colorpack.h"
 #include "spu_main.h"
 #include "spu_texture.h"
diff --git a/src/gallium/drivers/cell/spu/spu_util.c b/src/gallium/drivers/cell/spu/spu_util.c
index 74ab2bbd1f..b8a0d4a265 100644
--- a/src/gallium/drivers/cell/spu/spu_util.c
+++ b/src/gallium/drivers/cell/spu/spu_util.c
@@ -1,5 +1,7 @@
-#include "pipe/p_util.h"
+
+#include "cell/common.h"
 #include "pipe/p_shader_tokens.h"
+#include "pipe/p_debug.h"
 #include "tgsi/tgsi_parse.h"
 //#include "tgsi_build.h"
 #include "tgsi/tgsi_util.h"
@@ -19,7 +21,7 @@ tgsi_util_get_src_register_swizzle(
    case 3:
       return reg->SwizzleW;
    default:
-      assert( 0 );
+      ASSERT( 0 );
    }
    return 0;
 }
@@ -39,7 +41,7 @@ tgsi_util_get_src_register_extswizzle(
    case 3:
       return reg->ExtSwizzleW;
    default:
-      assert( 0 );
+      ASSERT( 0 );
    }
    return 0;
 }
@@ -59,12 +61,12 @@ tgsi_util_get_full_src_register_extswizzle(
       &reg->SrcRegisterExtSwz,
       component );
 
-   assert (TGSI_SWIZZLE_X == TGSI_EXTSWIZZLE_X);
-   assert (TGSI_SWIZZLE_Y == TGSI_EXTSWIZZLE_Y);
-   assert (TGSI_SWIZZLE_Z == TGSI_EXTSWIZZLE_Z);
-   assert (TGSI_SWIZZLE_W == TGSI_EXTSWIZZLE_W);
-   assert (TGSI_EXTSWIZZLE_ZERO > TGSI_SWIZZLE_W);
-   assert (TGSI_EXTSWIZZLE_ONE > TGSI_SWIZZLE_W);
+   ASSERT (TGSI_SWIZZLE_X == TGSI_EXTSWIZZLE_X);
+   ASSERT (TGSI_SWIZZLE_Y == TGSI_EXTSWIZZLE_Y);
+   ASSERT (TGSI_SWIZZLE_Z == TGSI_EXTSWIZZLE_Z);
+   ASSERT (TGSI_SWIZZLE_W == TGSI_EXTSWIZZLE_W);
+   ASSERT (TGSI_EXTSWIZZLE_ZERO > TGSI_SWIZZLE_W);
+   ASSERT (TGSI_EXTSWIZZLE_ONE > TGSI_SWIZZLE_W);
 
    /*
     * Second, calculate the simple  swizzle  for   the   unswizzled channel index.
@@ -94,7 +96,7 @@ tgsi_util_get_src_register_extnegate(
    case 3:
       return reg->NegateW;
    default:
-      assert( 0 );
+      ASSERT( 0 );
    }
    return 0;
 }
@@ -119,7 +121,7 @@ tgsi_util_set_src_register_extnegate(
       reg->NegateW = negate;
       break;
    default:
-      assert( 0 );
+      ASSERT( 0 );
    }
 }
 
diff --git a/src/gallium/drivers/cell/spu/spu_vertex_fetch.c b/src/gallium/drivers/cell/spu/spu_vertex_fetch.c
index 219fd90cc0..03375d84a5 100644
--- a/src/gallium/drivers/cell/spu/spu_vertex_fetch.c
+++ b/src/gallium/drivers/cell/spu/spu_vertex_fetch.c
@@ -32,7 +32,6 @@
   *   Ian Romanick <idr@us.ibm.com>
   */
 
-#include "pipe/p_util.h"
 #include "pipe/p_state.h"
 #include "pipe/p_shader_tokens.h"
 #include "spu_exec.h"
@@ -93,7 +92,7 @@ static void generic_vertex_fetch(struct spu_vs_context *draw,
    unsigned nr_attrs = draw->vertex_fetch.nr_attrs;
    unsigned attr;
 
-   assert(count <= 4);
+   ASSERT(count <= 4);
 
 #if DRAW_DBG
    printf("SPU: %s count = %u, nr_attrs = %u\n", 
diff --git a/src/gallium/drivers/cell/spu/spu_vertex_shader.c b/src/gallium/drivers/cell/spu/spu_vertex_shader.c
index 3119a78c06..fbe5b34d39 100644
--- a/src/gallium/drivers/cell/spu/spu_vertex_shader.c
+++ b/src/gallium/drivers/cell/spu/spu_vertex_shader.c
@@ -34,16 +34,37 @@
 
 #include <spu_mfcio.h>
 
-#include "pipe/p_util.h"
 #include "pipe/p_state.h"
 #include "pipe/p_shader_tokens.h"
-#include "spu_vertex_shader.h"
-#include "spu_exec.h"
+#include "util/u_math.h"
 #include "draw/draw_private.h"
 #include "draw/draw_context.h"
 #include "cell/common.h"
+#include "spu_vertex_shader.h"
+#include "spu_exec.h"
 #include "spu_main.h"
 
+
+#define MAX_VERTEX_SIZE ((2 + PIPE_MAX_SHADER_OUTPUTS) * 4 * sizeof(float))
+
+
+#define CLIP_RIGHT_BIT 0x01
+#define CLIP_LEFT_BIT 0x02
+#define CLIP_TOP_BIT 0x04
+#define CLIP_BOTTOM_BIT 0x08
+#define CLIP_FAR_BIT 0x10
+#define CLIP_NEAR_BIT 0x20
+
+
+static INLINE float
+dot4(const float *a, const float *b)
+{
+   return (a[0]*b[0] +
+           a[1]*b[1] +
+           a[2]*b[2] +
+           a[3]*b[3]);
+}
+
 static INLINE unsigned
 compute_clipmask(const float *clip, /*const*/ float plane[][4], unsigned nr)
 {
@@ -91,7 +112,7 @@ run_vertex_program(struct spu_vs_context *draw,
    const float *scale = draw->viewport.scale;
    const float *trans = draw->viewport.translate;
 
-   assert(count <= 4);
+   ASSERT(count <= 4);
 
    machine->Processor = TGSI_PROCESSOR_VERTEX;