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

8 files changed, 683 insertions, 448 deletions

diff --git a/src/gallium/drivers/cell/spu/Makefile b/src/gallium/drivers/cell/spu/Makefile
index d49abb2e82..1ae0dfb8c1 100644
--- a/src/gallium/drivers/cell/spu/Makefile
+++ b/src/gallium/drivers/cell/spu/Makefile
@@ -22,12 +22,15 @@ SOURCES = \
 	spu_render.c \
 	spu_texture.c \
 	spu_tile.c \
-	spu_tri.c \
+	spu_tri.c
+
+OLD_SOURCES = \
 	spu_exec.c \
 	spu_util.c \
 	spu_vertex_fetch.c \
 	spu_vertex_shader.c
 
+
 SPU_OBJECTS = $(SOURCES:.c=.o) \
 
 SPU_ASM_OUT = $(SOURCES:.c=.s) \
@@ -43,7 +46,7 @@ INCLUDE_DIRS = \
 	$(SPU_CC) $(SPU_CFLAGS) -c $<
 
 .c.s:
-	$(SPU_CC) $(SPU_CFLAGS) -S $<
+	$(SPU_CC) $(SPU_CFLAGS) -O3 -S $<
 
 
 # The .a file will be linked into the main/PPU executable
diff --git a/src/gallium/drivers/cell/spu/spu_colorpack.h b/src/gallium/drivers/cell/spu/spu_colorpack.h
index e9fee8a3a6..fd8dc6ded3 100644
--- a/src/gallium/drivers/cell/spu/spu_colorpack.h
+++ b/src/gallium/drivers/cell/spu/spu_colorpack.h
@@ -79,14 +79,14 @@ spu_pack_color_shuffle(vector float rgba, vector unsigned char shuffle)
 
 
 static INLINE vector float
-spu_unpack_color(uint color)
+spu_unpack_B8G8R8A8(uint color)
 {
    vector unsigned int color_u4 = spu_splats(color);
    color_u4 = spu_shuffle(color_u4, color_u4,
                           ((vector unsigned char) {
-                             0, 0, 0, 0,
-                             5, 5, 5, 5,
                              10, 10, 10, 10,
+                             5, 5, 5, 5,
+                             0, 0, 0, 0,
                              15, 15, 15, 15}) );
    return spu_convtf(color_u4, 32);
 }
diff --git a/src/gallium/drivers/cell/spu/spu_main.c b/src/gallium/drivers/cell/spu/spu_main.c
index d223f32d94..78260c4259 100644
--- a/src/gallium/drivers/cell/spu/spu_main.c
+++ b/src/gallium/drivers/cell/spu/spu_main.c
@@ -34,6 +34,7 @@
 
 #include "spu_main.h"
 #include "spu_render.h"
+#include "spu_per_fragment_op.h"
 #include "spu_texture.h"
 #include "spu_tile.h"
 //#include "spu_test.h"
@@ -46,7 +47,7 @@
 /*
 helpful headers:
 /usr/lib/gcc/spu/4.1.1/include/spu_mfcio.h
-/opt/ibm/cell-sdk/prototype/sysroot/usr/include/libmisc.h
+/opt/cell/sdk/usr/include/libmisc.h
 */
 
 boolean Debug = FALSE;
@@ -62,14 +63,6 @@ struct spu_vs_context draw;
 static unsigned char attribute_fetch_code_buffer[136 * PIPE_MAX_ATTRIBS]
     ALIGN16_ATTRIB;
 
-static unsigned char depth_stencil_code_buffer[4 * 64]
-    ALIGN16_ATTRIB;
-
-static unsigned char fb_blend_code_buffer[4 * 64]
-    ALIGN16_ATTRIB;
-
-static unsigned char logicop_code_buffer[4 * 64]
-    ALIGN16_ATTRIB;
 
 
 /**
@@ -226,6 +219,46 @@ cmd_release_verts(const struct cell_command_release_verts *release)
 }
 
 
+/**
+ * Process a CELL_CMD_STATE_FRAGMENT_OPS command.
+ * This involves installing new fragment ops SPU code.
+ * If this function is never called, we'll use a regular C fallback function
+ * for fragment processing.
+ */
+static void
+cmd_state_fragment_ops(const struct cell_command_fragment_ops *fops)
+{
+   if (Debug)
+      printf("SPU %u: CMD_STATE_FRAGMENT_OPS\n", spu.init.id);
+   /* Copy SPU code from batch buffer to spu buffer */
+   memcpy(spu.fragment_ops_code, fops->code, SPU_MAX_FRAGMENT_OPS_INSTS * 4);
+   /* Copy state info (for fallback case only) */
+   memcpy(&spu.depth_stencil_alpha, &fops->dsa, sizeof(fops->dsa));
+   memcpy(&spu.blend, &fops->blend, sizeof(fops->blend));
+
+   /* Point function pointer at new code */
+   spu.fragment_ops = (spu_fragment_ops_func) spu.fragment_ops_code;
+
+   spu.read_depth = spu.depth_stencil_alpha.depth.enabled;
+   spu.read_stencil = spu.depth_stencil_alpha.stencil[0].enabled;
+}
+
+
+static void
+cmd_state_fragment_program(const struct cell_command_fragment_program *fp)
+{
+   if (Debug)
+      printf("SPU %u: CMD_STATE_FRAGMENT_PROGRAM\n", spu.init.id);
+   /* Copy SPU code from batch buffer to spu buffer */
+   memcpy(spu.fragment_program_code, fp->code,
+          SPU_MAX_FRAGMENT_PROGRAM_INSTS * 4);
+#if 01
+   /* Point function pointer at new code */
+   spu.fragment_program = (spu_fragment_program_func)spu.fragment_program_code;
+#endif
+}
+
+
 static void
 cmd_state_framebuffer(const struct cell_command_framebuffer *cmd)
 {
@@ -252,102 +285,24 @@ cmd_state_framebuffer(const struct cell_command_framebuffer *cmd)
 
    switch (spu.fb.depth_format) {
    case PIPE_FORMAT_Z32_UNORM:
+      spu.fb.zsize = 4;
+      spu.fb.zscale = (float) 0xffffffffu;
+      break;
    case PIPE_FORMAT_Z24S8_UNORM:
    case PIPE_FORMAT_S8Z24_UNORM:
+   case PIPE_FORMAT_Z24X8_UNORM:
+   case PIPE_FORMAT_X8Z24_UNORM:
       spu.fb.zsize = 4;
+      spu.fb.zscale = (float) 0x00ffffffu;
       break;
    case PIPE_FORMAT_Z16_UNORM:
       spu.fb.zsize = 2;
+      spu.fb.zscale = (float) 0xffffu;
       break;
    default:
       spu.fb.zsize = 0;
       break;
    }
-
-   if (spu.fb.color_format == PIPE_FORMAT_A8R8G8B8_UNORM)
-      spu.color_shuffle = ((vector unsigned char) {
-                              12, 0, 4, 8, 0, 0, 0, 0, 
-                              0, 0, 0, 0, 0, 0, 0, 0});
-   else if (spu.fb.color_format == PIPE_FORMAT_B8G8R8A8_UNORM)
-      spu.color_shuffle = ((vector unsigned char) {
-                              8, 4, 0, 12, 0, 0, 0, 0, 
-                              0, 0, 0, 0, 0, 0, 0, 0});
-   else
-      ASSERT(0);
-}
-
-
-static void
-cmd_state_blend(const struct cell_command_blend *state)
-{
-   if (Debug)
-      printf("SPU %u: BLEND: enabled %d\n",
-             spu.init.id,
-             (state->size != 0));
-
-   ASSERT_ALIGN16(state->base);
-
-   if (state->size != 0) {
-      mfc_get(fb_blend_code_buffer,
-              (unsigned int) state->base,  /* src */
-              ROUNDUP16(state->size),
-              TAG_BATCH_BUFFER,
-              0, /* tid */
-              0  /* rid */);
-      wait_on_mask(1 << TAG_BATCH_BUFFER);
-      spu.blend = (blend_func) fb_blend_code_buffer;
-      spu.read_fb = state->read_fb;
-   } else {
-      spu.read_fb = FALSE;
-   }
-}
-
-
-static void
-cmd_state_depth_stencil(const struct cell_command_depth_stencil_alpha_test *state)
-{
-   if (Debug)
-      printf("SPU %u: DEPTH_STENCIL: ztest %d\n",
-             spu.init.id,
-             state->read_depth);
-
-   ASSERT_ALIGN16(state->base);
-
-   if (state->size != 0) {
-      mfc_get(depth_stencil_code_buffer,
-	      (unsigned int) state->base,  /* src */
-	      ROUNDUP16(state->size),
-	      TAG_BATCH_BUFFER,
-	      0, /* tid */
-	      0  /* rid */);
-      wait_on_mask(1 << TAG_BATCH_BUFFER);
-   } else {
-      /* If there is no code, emit a return instruction.
-       */
-      depth_stencil_code_buffer[0] = 0x35;
-      depth_stencil_code_buffer[1] = 0x00;
-      depth_stencil_code_buffer[2] = 0x00;
-      depth_stencil_code_buffer[3] = 0x00;
-   }
-
-   spu.frag_test = (frag_test_func) depth_stencil_code_buffer;
-   spu.read_depth = state->read_depth;
-   spu.read_stencil = state->read_stencil;
-}
-
-
-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;
 }
 
 
@@ -450,7 +405,9 @@ cmd_finish(void)
 
 
 /**
- * Execute a batch of commands
+ * Execute a batch of commands which was sent to us by the PPU.
+ * See the cell_emit_state.c code to see where the commands come from.
+ *
  * The opcode param encodes the location of the buffer and its size.
  */
 static void
@@ -487,16 +444,14 @@ cmd_batch(uint opcode)
       printf("SPU %u: release batch buf %u\n", spu.init.id, buf);
    release_buffer(buf);
 
+   /*
+    * Loop over commands in the batch buffer
+    */
    for (pos = 0; pos < usize; /* no incr */) {
       switch (buffer[pos]) {
-      case CELL_CMD_STATE_FRAMEBUFFER:
-         {
-            struct cell_command_framebuffer *fb
-               = (struct cell_command_framebuffer *) &buffer[pos];
-            cmd_state_framebuffer(fb);
-            pos += sizeof(*fb) / 8;
-         }
-         break;
+      /*
+       * rendering commands
+       */
       case CELL_CMD_CLEAR_SURFACE:
          {
             struct cell_command_clear_surface *clr
@@ -514,26 +469,32 @@ cmd_batch(uint opcode)
             pos += pos_incr;
          }
          break;
-      case CELL_CMD_RELEASE_VERTS:
+      /*
+       * state-update commands
+       */
+      case CELL_CMD_STATE_FRAMEBUFFER:
          {
-            struct cell_command_release_verts *release
-               = (struct cell_command_release_verts *) &buffer[pos];
-            cmd_release_verts(release);
-            pos += sizeof(*release) / 8;
+            struct cell_command_framebuffer *fb
+               = (struct cell_command_framebuffer *) &buffer[pos];
+            cmd_state_framebuffer(fb);
+            pos += sizeof(*fb) / 8;
          }
          break;
-      case CELL_CMD_FINISH:
-         cmd_finish();
-         pos += 1;
-         break;
-      case CELL_CMD_STATE_BLEND:
-         cmd_state_blend((struct cell_command_blend *) &buffer[pos+1]);
-         pos += (1 + ROUNDUP8(sizeof(struct cell_command_blend)) / 8);
+      case CELL_CMD_STATE_FRAGMENT_OPS:
+         {
+            struct cell_command_fragment_ops *fops
+               = (struct cell_command_fragment_ops *) &buffer[pos];
+            cmd_state_fragment_ops(fops);
+            pos += sizeof(*fops) / 8;
+         }
          break;
-      case CELL_CMD_STATE_DEPTH_STENCIL:
-         cmd_state_depth_stencil((struct cell_command_depth_stencil_alpha_test *)
-                                 &buffer[pos+1]);
-         pos += (1 + ROUNDUP8(sizeof(struct cell_command_depth_stencil_alpha_test)) / 8);
+      case CELL_CMD_STATE_FRAGMENT_PROGRAM:
+         {
+            struct cell_command_fragment_program *fp
+               = (struct cell_command_fragment_program *) &buffer[pos];
+            cmd_state_fragment_program(fp);
+            pos += sizeof(*fp) / 8;
+         }
          break;
       case CELL_CMD_STATE_SAMPLER:
          {
@@ -569,8 +530,10 @@ cmd_batch(uint opcode)
          pos += (1 + ROUNDUP8(sizeof(struct cell_array_info)) / 8);
          break;
       case CELL_CMD_STATE_BIND_VS:
+#if 0
          spu_bind_vertex_shader(&draw,
                                 (struct cell_shader_info *) &buffer[pos+1]);
+#endif
          pos += (1 + ROUNDUP8(sizeof(struct cell_shader_info)) / 8);
          break;
       case CELL_CMD_STATE_ATTRIB_FETCH:
@@ -578,9 +541,20 @@ cmd_batch(uint opcode)
                                 &buffer[pos+1]);
          pos += (1 + ROUNDUP8(sizeof(struct cell_attribute_fetch_code)) / 8);
          break;
-      case CELL_CMD_STATE_LOGICOP:
-         cmd_state_logicop((struct cell_command_logicop *) &buffer[pos+1]);
-         pos += (1 + ROUNDUP8(sizeof(struct cell_command_logicop)) / 8);
+      /*
+       * misc commands
+       */
+      case CELL_CMD_FINISH:
+         cmd_finish();
+         pos += 1;
+         break;
+      case CELL_CMD_RELEASE_VERTS:
+         {
+            struct cell_command_release_verts *release
+               = (struct cell_command_release_verts *) &buffer[pos];
+            cmd_release_verts(release);
+            pos += sizeof(*release) / 8;
+         }
          break;
       case CELL_CMD_FLUSH_BUFFER_RANGE: {
 	 struct cell_buffer_range *br = (struct cell_buffer_range *)
@@ -650,7 +624,9 @@ main_loop(void)
          exitFlag = 1;
          break;
       case CELL_CMD_VS_EXECUTE:
+#if 0
          spu_execute_vertex_shader(&draw, &cmd.vs);
+#endif
          break;
       case CELL_CMD_BATCH:
          cmd_batch(opcode);
@@ -675,6 +651,11 @@ one_time_init(void)
    memset(spu.ctile_status, TILE_STATUS_DEFINED, sizeof(spu.ctile_status));
    memset(spu.ztile_status, TILE_STATUS_DEFINED, sizeof(spu.ztile_status));
    invalidate_tex_cache();
+
+   /* Install default/fallback fragment processing function.
+    * This will normally be overriden by a code-gen'd function.
+    */
+   spu.fragment_ops = spu_fallback_fragment_ops;
 }
 
 
diff --git a/src/gallium/drivers/cell/spu/spu_main.h b/src/gallium/drivers/cell/spu/spu_main.h
index 4879f8c9c8..2c7b625840 100644
--- a/src/gallium/drivers/cell/spu/spu_main.h
+++ b/src/gallium/drivers/cell/spu/spu_main.h
@@ -41,6 +41,10 @@
 #define MAX_HEIGHT 1024
 
 
+/**
+ * A tile is basically a TILE_SIZE x TILE_SIZE block of 4-byte pixels.
+ * The data may be addressed through several different types.
+ */
 typedef union {
    ushort us[TILE_SIZE][TILE_SIZE];
    uint   ui[TILE_SIZE][TILE_SIZE];
@@ -56,38 +60,29 @@ typedef union {
 #define TILE_STATUS_GETTING 5  /**< mfc_get() called but not yet arrived */
 
 
-struct spu_frag_test_results {
-   qword mask;
-   qword depth;
-   qword stencil;
-};
-
-typedef struct spu_frag_test_results (*frag_test_func)(qword frag_mask,
-    qword pixel_depth, qword pixel_stencil, qword frag_depth,
-    qword frag_alpha, qword facing);
-
-
-struct spu_blend_results {
-   qword r;
-   qword g;
-   qword b;
-   qword a;
-};
+/** Function for sampling textures */
+typedef vector float (*spu_sample_texture_func)(uint unit,
+                                                vector float texcoord);
 
-typedef struct spu_blend_results (*blend_func)(
-    qword frag_r, qword frag_g, qword frag_b, qword frag_a,
-    qword pixel_r, qword pixel_g, qword pixel_b, qword pixel_a,
-    qword const_r, qword const_g, qword const_b, qword const_a);
+/** Function for performing per-fragment ops */
+typedef void (*spu_fragment_ops_func)(uint x, uint y,
+                                      tile_t *colorTile,
+                                      tile_t *depthStencilTile,
+                                      vector float fragZ,
+                                      vector float fragRed,
+                                      vector float fragGreen,
+                                      vector float fragBlue,
+                                      vector float fragAlpha,
+                                      vector unsigned int mask);
 
-typedef struct spu_blend_results (*logicop_func)(
-    qword pixel_r, qword pixel_g, qword pixel_b, qword pixel_a,
-    qword frag_r, qword frag_g, qword frag_b, qword frag_a,
-    qword frag_mask);
+/** Function for running fragment program */
+typedef void (*spu_fragment_program_func)(vector float *inputs,
+                                          vector float *outputs,
+                                          vector float *constants);
 
 
-typedef vector float (*sample_texture_func)(uint unit, vector float texcoord);
-
-struct spu_framebuffer {
+struct spu_framebuffer
+{
    void *color_start;              /**< addr of color surface in main memory */
    void *depth_start;              /**< addr of depth surface in main memory */
    enum pipe_format color_format;
@@ -99,6 +94,7 @@ struct spu_framebuffer {
    uint depth_clear_value;
 
    uint zsize;                     /**< 0, 2 or 4 bytes per Z */
+   float zscale;                   /**< 65535.0, 2^24-1 or 2^32-1 */
 } ALIGN16_ATTRIB;
 
 
@@ -115,35 +111,31 @@ struct spu_texture
 
 
 /**
- * All SPU global/context state will be in singleton object of this type:
+ * All SPU global/context state will be in a singleton object of this type:
  */
 struct spu_global
 {
+   /** One-time init/constant info */
    struct cell_init_info init;
 
+   /*
+    * Current state
+    */
    struct spu_framebuffer fb;
-   boolean read_depth;
-   boolean read_stencil;
-   frag_test_func frag_test;  /**< Current depth/stencil test code */
-   
-   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;  /**< Current logicop code **/
-
+   struct pipe_depth_stencil_alpha_state depth_stencil_alpha;
+   struct pipe_blend_state blend;
    struct pipe_sampler_state sampler[PIPE_MAX_SAMPLERS];
    struct spu_texture texture[PIPE_MAX_SAMPLERS];
-
    struct vertex_info vertex_info;
 
-   /* XXX more state to come */
-
-
-   /** current color and Z tiles */
+   /** Current color and Z tiles */
    tile_t ctile ALIGN16_ATTRIB;
    tile_t ztile ALIGN16_ATTRIB;
 
+   /** Read depth/stencil tiles? */
+   boolean read_depth;
+   boolean read_stencil;
+
    /** Current tiles' status */
    ubyte cur_ctile_status, cur_ztile_status;
 
@@ -151,11 +143,22 @@ struct spu_global
    ubyte ctile_status[MAX_HEIGHT/TILE_SIZE][MAX_WIDTH/TILE_SIZE] ALIGN16_ATTRIB;
    ubyte ztile_status[MAX_HEIGHT/TILE_SIZE][MAX_WIDTH/TILE_SIZE] ALIGN16_ATTRIB;
 
+   /** Current fragment ops machine code */
+   uint fragment_ops_code[SPU_MAX_FRAGMENT_OPS_INSTS];
+   /** Current fragment ops function */
+   spu_fragment_ops_func fragment_ops;
+
+   /** Current fragment program machine code */
+   uint fragment_program_code[SPU_MAX_FRAGMENT_PROGRAM_INSTS];
+   /** Current fragment ops function */
+   spu_fragment_program_func fragment_program;
 
-   /** for converting RGBA to PIPE_FORMAT_x colors */
-   vector unsigned char color_shuffle;
+   /** Current texture sampler function */
+   spu_sample_texture_func sample_texture[CELL_MAX_SAMPLERS];
 
-   sample_texture_func sample_texture[CELL_MAX_SAMPLERS];
+   /** Fragment program constants (XXX preliminary/used) */
+#define MAX_CONSTANTS 32
+   vector float constants[MAX_CONSTANTS];
 
 } ALIGN16_ATTRIB;
 
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 c0a729b3d2..03dd547845 100644
--- a/src/gallium/drivers/cell/spu/spu_per_fragment_op.c
+++ b/src/gallium/drivers/cell/spu/spu_per_fragment_op.c
@@ -1,211 +1,475 @@
-/*
- * (C) Copyright IBM Corporation 2008
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
  * All Rights Reserved.
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the "Software"),
- * to deal in the Software without restriction, including without limitation
- * on the rights to use, copy, modify, merge, publish, distribute, sub
- * license, and/or sell copies of the Software, and to permit persons to whom
- * the Software is furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice (including the next
- * paragraph) shall be included in all copies or substantial portions of the
- * Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.  IN NO EVENT SHALL
- * AUTHORS, COPYRIGHT HOLDERS, AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
- * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
- * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
- * USE OR OTHER DEALINGS IN THE SOFTWARE.
- */
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
 
 /**
- * \file spu_per_fragment_op.c
- * SPU implementation various per-fragment operations.
- *
- * \author Ian Romanick <idr@us.ibm.com>
+ * \author Brian Paul
  */
 
+
+#include <transpose_matrix4x4.h>
 #include "pipe/p_format.h"
 #include "spu_main.h"
+#include "spu_colorpack.h"
 #include "spu_per_fragment_op.h"
 
-#define ZERO 0x80
 
-static void
-read_ds_quad(tile_t *buffer, unsigned x, unsigned y,
-             enum pipe_format depth_format, qword *depth,
-             qword *stencil)
-{
-   const int ix = x / 2;
-   const int iy = y / 2;
+#define LINEAR_QUAD_LAYOUT 1
 
-   switch (depth_format) {
-   case PIPE_FORMAT_Z16_UNORM: {
-      qword *ptr = (qword *) &buffer->us8[iy][ix / 2];
 
-      const qword shuf_vec = (qword) {
-         ZERO, ZERO, 0, 1, ZERO, ZERO, 2, 3,
-         ZERO, ZERO, 4, 5, ZERO, ZERO, 6, 7
-      };
+/**
+ * Called by rasterizer for each quad after the shader has run.  Do
+ * all the per-fragment operations including alpha test, z test,
+ * stencil test, blend, colormask and logicops.  This is a
+ * fallback/debug function.  In reality we'll use a generated function
+ * produced by the PPU.  But this function is useful for
+ * debug/validation.
+ */
+void
+spu_fallback_fragment_ops(uint x, uint y,
+                          tile_t *colorTile,
+                          tile_t *depthStencilTile,
+                          vector float fragZ,
+                          vector float fragR,
+                          vector float fragG,
+                          vector float fragB,
+                          vector float fragA,
+                          vector unsigned int mask)
+{
+   vector float frag_aos[4];
+   unsigned int c0, c1, c2, c3;
+
+   /* do alpha test */
+   if (spu.depth_stencil_alpha.alpha.enabled) {
+      vector float ref = spu_splats(spu.depth_stencil_alpha.alpha.ref);
+      vector unsigned int amask;
+
+      switch (spu.depth_stencil_alpha.alpha.func) {
+      case PIPE_FUNC_LESS:
+         amask = spu_cmpgt(ref, fragA);  /* mask = (fragA < ref) */
+         break;
+      case PIPE_FUNC_GREATER:
+         amask = spu_cmpgt(fragA, ref);  /* mask = (fragA > ref) */
+         break;
+      case PIPE_FUNC_GEQUAL:
+         amask = spu_cmpgt(ref, fragA);
+         amask = spu_nor(amask, amask);
+         break;
+      case PIPE_FUNC_LEQUAL:
+         amask = spu_cmpgt(fragA, ref);
+         amask = spu_nor(amask, amask);
+         break;
+      case PIPE_FUNC_EQUAL:
+         amask = spu_cmpeq(ref, fragA);
+         break;
+      case PIPE_FUNC_NOTEQUAL:
+         amask = spu_cmpeq(ref, fragA);
+         amask = spu_nor(amask, amask);
+         break;
+      case PIPE_FUNC_ALWAYS:
+         amask = spu_splats(0xffffffffU);
+         break;
+      case PIPE_FUNC_NEVER:
+         amask = spu_splats( 0x0U);
+         break;
+      default:
+         ;
+      }
+
+      mask = spu_and(mask, amask);
+   }
 
+   /* Z and/or stencil testing... */
+   if (spu.depth_stencil_alpha.depth.enabled ||
+       spu.depth_stencil_alpha.stencil[0].enabled) {
+
+      /* get four Z/Stencil values from tile */
+      vector unsigned int mask24 = spu_splats((unsigned int)0x00ffffffU);
+      vector unsigned int ifbZS = depthStencilTile->ui4[y/2][x/2];
+      vector unsigned int ifbZ = spu_and(ifbZS, mask24);
+      vector unsigned int ifbS = spu_andc(ifbZS, mask24);
+
+      if (spu.depth_stencil_alpha.stencil[0].enabled) {
+         /* do stencil test */
+         ASSERT(spu.fb.depth_format == PIPE_FORMAT_S8Z24_UNORM);
+
+      }
+      else if (spu.depth_stencil_alpha.depth.enabled) {
+         /* do depth test */
+
+         ASSERT(spu.fb.depth_format == PIPE_FORMAT_S8Z24_UNORM ||
+                spu.fb.depth_format == PIPE_FORMAT_X8Z24_UNORM);
+
+         vector unsigned int ifragZ;
+         vector unsigned int zmask;
+
+         /* convert four fragZ from float to uint */
+         fragZ = spu_mul(fragZ, spu_splats((float) 0xffffff));
+         ifragZ = spu_convtu(fragZ, 0);
+
+         /* do depth comparison, setting zmask with results */
+         switch (spu.depth_stencil_alpha.depth.func) {
+         case PIPE_FUNC_LESS:
+            zmask = spu_cmpgt(ifbZ, ifragZ);  /* mask = (ifragZ < ifbZ) */
+            break;
+         case PIPE_FUNC_GREATER:
+            zmask = spu_cmpgt(ifragZ, ifbZ);  /* mask = (ifbZ > ifragZ) */
+            break;
+         case PIPE_FUNC_GEQUAL:
+            zmask = spu_cmpgt(ifbZ, ifragZ);
+            zmask = spu_nor(zmask, zmask);
+            break;
+         case PIPE_FUNC_LEQUAL:
+            zmask = spu_cmpgt(ifragZ, ifbZ);
+            zmask = spu_nor(zmask, zmask);
+            break;
+         case PIPE_FUNC_EQUAL:
+            zmask = spu_cmpeq(ifbZ, ifragZ);
+            break;
+         case PIPE_FUNC_NOTEQUAL:
+            zmask = spu_cmpeq(ifbZ, ifragZ);
+            zmask = spu_nor(zmask, zmask);
+            break;
+         case PIPE_FUNC_ALWAYS:
+            zmask = spu_splats(0xffffffffU);
+            break;
+         case PIPE_FUNC_NEVER:
+            zmask = spu_splats( 0x0U);
+            break;
+         default:
+            ;
+         }
+
+         mask = spu_and(mask, zmask);
+
+         /* merge framebuffer Z and fragment Z according to the mask */
+         ifbZ = spu_or(spu_and(ifragZ, mask),
+                       spu_andc(ifbZ, mask));
+      }
+
+      if (spu_extract(spu_orx(mask), 0)) {
+         /* put new fragment Z/Stencil values back into Z/Stencil tile */
+         depthStencilTile->ui4[y/2][x/2] = spu_or(ifbZ, ifbS);
+
+         spu.cur_ztile_status = TILE_STATUS_DIRTY;
+      }
+   }
 
-      /* At even X values we want the first 4 shorts, and at odd X values we
-       * want the second 4 shorts.
+   if (spu.blend.blend_enable) {
+      /* blending terms, misc regs */
+      vector float term1r, term1g, term1b, term1a;
+      vector float term2r, term2g, term2b, term2a;
+      vector float one, tmp;
+
+      vector float fbRGBA[4];  /* current framebuffer colors */
+
+      /* get colors from framebuffer/tile */
+      {
+         vector float fc[4];
+         uint c0, c1, c2, c3;
+
+#if LINEAR_QUAD_LAYOUT /* See comments/diagram below */
+         c0 = colorTile->ui[y][x*2+0];
+         c1 = colorTile->ui[y][x*2+1];
+         c2 = colorTile->ui[y][x*2+2];
+         c3 = colorTile->ui[y][x*2+3];
+#else
+         c0 = colorTile->ui[y+0][x+0];
+         c1 = colorTile->ui[y+0][x+1];
+         c2 = colorTile->ui[y+1][x+0];
+         c3 = colorTile->ui[y+1][x+1];
+#endif
+         switch (spu.fb.color_format) {
+         case PIPE_FORMAT_B8G8R8A8_UNORM:
+            fc[0] = spu_unpack_B8G8R8A8(c0);
+            fc[1] = spu_unpack_B8G8R8A8(c1);
+            fc[2] = spu_unpack_B8G8R8A8(c2);
+            fc[3] = spu_unpack_B8G8R8A8(c3);
+            break;
+         case PIPE_FORMAT_A8R8G8B8_UNORM:
+            fc[0] = spu_unpack_A8R8G8B8(c0);
+            fc[1] = spu_unpack_A8R8G8B8(c1);
+            fc[2] = spu_unpack_A8R8G8B8(c2);
+            fc[3] = spu_unpack_A8R8G8B8(c3);
+            break;
+         default:
+            ASSERT(0);
+         }
+         _transpose_matrix4x4(fbRGBA, fc);
+      }
+
+      /*
+       * Compute Src RGB terms
        */
-      qword bias = (qword) spu_splats((unsigned char) ((ix & 0x01) << 3));
-      qword bias_mask = si_fsmbi(0x3333);
-      qword sv = si_a(shuf_vec, si_and(bias_mask, bias));
-
-      *depth = si_shufb(*ptr, *ptr, sv);
-      *stencil = si_il(0);
-      break;
+      switch (spu.blend.rgb_src_factor) {
+      case PIPE_BLENDFACTOR_ONE:
+         term1r = fragR;
+         term1g = fragG;
+         term1b = fragB;
+         break;
+      case PIPE_BLENDFACTOR_ZERO:
+         term1r =
+         term1g =
+         term1b = spu_splats(0.0f);
+         break;
+      case PIPE_BLENDFACTOR_SRC_COLOR:
+         term1r = spu_mul(fragR, fragR);
+         term1g = spu_mul(fragG, fragG);
+         term1b = spu_mul(fragB, fragB);
+         break;
+      case PIPE_BLENDFACTOR_SRC_ALPHA:
+         term1r = spu_mul(fragR, fragA);
+         term1g = spu_mul(fragG, fragA);
+         term1b = spu_mul(fragB, fragA);
+         break;
+      /* XXX more cases */
+      default:
+         ASSERT(0);
+      }
+
+      /*
+       * Compute Src Alpha term
+       */
+      switch (spu.blend.alpha_src_factor) {
+      case PIPE_BLENDFACTOR_ONE:
+         term1a = fragA;
+         break;
+      case PIPE_BLENDFACTOR_SRC_COLOR:
+         term1a = spu_splats(0.0f);
+         break;
+      case PIPE_BLENDFACTOR_SRC_ALPHA:
+         term1a = spu_mul(fragA, fragA);
+         break;
+      /* XXX more cases */
+      default:
+         ASSERT(0);
+      }
+
+      /*
+       * Compute Dest RGB terms
+       */
+      switch (spu.blend.rgb_dst_factor) {
+      case PIPE_BLENDFACTOR_ONE:
+         term2r = fragR;
+         term2g = fragG;
+         term2b = fragB;
+         break;
+      case PIPE_BLENDFACTOR_ZERO:
+         term2r =
+         term2g =
+         term2b = spu_splats(0.0f);
+         break;
+      case PIPE_BLENDFACTOR_SRC_COLOR:
+         term2r = spu_mul(fbRGBA[0], fragR);
+         term2g = spu_mul(fbRGBA[1], fragG);
+         term2b = spu_mul(fbRGBA[2], fragB);
+         break;
+      case PIPE_BLENDFACTOR_SRC_ALPHA:
+         term2r = spu_mul(fbRGBA[0], fragA);
+         term2g = spu_mul(fbRGBA[1], fragA);
+         term2b = spu_mul(fbRGBA[2], fragA);
+         break;
+      case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
+         one = spu_splats(1.0f);
+         tmp = spu_sub(one, fragA);
+         term2r = spu_mul(fbRGBA[0], tmp);
+         term2g = spu_mul(fbRGBA[1], tmp);
+         term2b = spu_mul(fbRGBA[2], tmp);
+         break;
+      /* XXX more cases */
+      default:
+         ASSERT(0);
+      }
+
+      /*
+       * Compute Dest Alpha term
+       */
+      switch (spu.blend.alpha_dst_factor) {
+      case PIPE_BLENDFACTOR_ONE:
+         term2a = fragA;
+         break;
+      case PIPE_BLENDFACTOR_SRC_COLOR:
+         term2a = spu_splats(0.0f);
+         break;
+      case PIPE_BLENDFACTOR_SRC_ALPHA:
+         term2a = spu_mul(fbRGBA[3], fragA);
+         break;
+      case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
+         one = spu_splats(1.0f);
+         tmp = spu_sub(one, fragA);
+         term2a = spu_mul(fbRGBA[3], tmp);
+         break;
+      /* XXX more cases */
+      default:
+         ASSERT(0);
+      }
+
+      /*
+       * Combine Src/Dest RGB terms
+       */
+      switch (spu.blend.rgb_func) {
+      case PIPE_BLEND_ADD:
+         fragR = spu_add(term1r, term2r);
+         fragG = spu_add(term1g, term2g);
+         fragB = spu_add(term1b, term2b);
+         break;
+      case PIPE_BLEND_SUBTRACT:
+         fragR = spu_sub(term1r, term2r);
+         fragG = spu_sub(term1g, term2g);
+         fragB = spu_sub(term1b, term2b);
+         break;
+      /* XXX more cases */
+      default:
+         ASSERT(0);
+      }
+
+      /*
+       * Combine Src/Dest A term
+       */
+      switch (spu.blend.alpha_func) {
+      case PIPE_BLEND_ADD:
+         fragA = spu_add(term1a, term2a);
+         break;
+      case PIPE_BLEND_SUBTRACT:
+         fragA = spu_sub(term1a, term2a);
+         break;
+      /* XXX more cases */
+      default:
+         ASSERT(0);
+      }
    }
 
 
-   case PIPE_FORMAT_Z32_UNORM: {
-      qword *ptr = (qword *) &buffer->ui4[iy][ix];
-
-      *depth = *ptr;
-      *stencil = si_il(0);
-      break;
+   /*
+    * Convert RRRR,GGGG,BBBB,AAAA to RGBA,RGBA,RGBA,RGBA.
+    */
+#if 0
+   /* original code */
+   {
+      vector float frag_soa[4];
+      frag_soa[0] = fragR;
+      frag_soa[1] = fragG;
+      frag_soa[2] = fragB;
+      frag_soa[3] = fragA;
+      _transpose_matrix4x4(frag_aos, frag_soa);
    }
-      
-
-   case PIPE_FORMAT_Z24S8_UNORM: {
-      qword *ptr = (qword *) &buffer->ui4[iy][ix];
-      qword mask = si_fsmbi(0xEEEE);
-
-      *depth = si_rotmai(si_and(*ptr, mask), -8);
-      *stencil = si_andc(*ptr, mask);
+#else
+   /* short-cut relying on function parameter layout: */
+   _transpose_matrix4x4(frag_aos, &fragR);
+   (void) fragG;
+   (void) fragB;
+#endif
+
+   /*
+    * Pack float colors into 32-bit RGBA words.
+    */
+   switch (spu.fb.color_format) {
+   case PIPE_FORMAT_A8R8G8B8_UNORM:
+      c0 = spu_pack_A8R8G8B8(frag_aos[0]);
+      c1 = spu_pack_A8R8G8B8(frag_aos[1]);
+      c2 = spu_pack_A8R8G8B8(frag_aos[2]);
+      c3 = spu_pack_A8R8G8B8(frag_aos[3]);
       break;
-   }
-
-
-   case PIPE_FORMAT_S8Z24_UNORM: {
-      qword *ptr = (qword *) &buffer->ui4[iy][ix];
 
-      *depth = si_and(*ptr, si_fsmbi(0x7777));
-      *stencil = si_andi(si_roti(*ptr, 8), 0x0ff);
+   case PIPE_FORMAT_B8G8R8A8_UNORM:
+      c0 = spu_pack_B8G8R8A8(frag_aos[0]);
+      c1 = spu_pack_B8G8R8A8(frag_aos[1]);
+      c2 = spu_pack_B8G8R8A8(frag_aos[2]);
+      c3 = spu_pack_B8G8R8A8(frag_aos[3]);
       break;
-   }
-
-
    default:
+      fprintf(stderr, "SPU: Bad pixel format in spu_default_fragment_ops\n");
       ASSERT(0);
-      break;
-   }
-}
-
-
-static void
-write_ds_quad(tile_t *buffer, unsigned x, unsigned y,
-              enum pipe_format depth_format,
-              qword depth, qword stencil)
-{
-   const int ix = x / 2;
-   const int iy = y / 2;
-
-   (void) stencil;
-
-   switch (depth_format) {
-   case PIPE_FORMAT_Z16_UNORM: {
-      qword *ptr = (qword *) &buffer->us8[iy][ix / 2];
-
-      qword sv = ((ix & 0x01) == 0) 
-          ? (qword) { 2, 3, 6, 7, 10, 11, 14, 15,
-                      24, 25, 26, 27, 28, 29, 30, 31 }
-          : (qword) { 16, 17, 18, 19, 20 , 21, 22, 23,
-                      2, 3, 6, 7, 10, 11, 14, 15 };
-      *ptr = si_shufb(depth, *ptr, sv);
-      break;
-   }
-
-
-   case PIPE_FORMAT_Z32_UNORM: {
-      qword *ptr = (qword *) &buffer->ui4[iy][ix];
-      *ptr = depth;
-      break;
    }
 
 
-   case PIPE_FORMAT_Z24S8_UNORM: {
-      qword *ptr = (qword *) &buffer->ui4[iy][ix];
-      qword mask = si_fsmbi(0xEEEE);
-
-      depth = si_shli(depth, 8);
-      *ptr = si_selb(stencil, depth, mask);
-      break;
-   }
-
-
-   case PIPE_FORMAT_S8Z24_UNORM: {
-      qword *ptr = (qword *) &buffer->ui4[iy][ix];
-      qword mask = si_fsmbi(0x7777);
-
-      stencil = si_shli(stencil, 24);
-      *ptr = si_selb(stencil, depth, mask);
-      break;
+   /*
+    * Color masking
+    */
+   if (spu.blend.colormask != 0xf) {
+      /* XXX to do */
+      /* apply color mask to 32-bit packed colors */
    }
 
 
-   default:
-      ASSERT(0);
-      break;
+   /*
+    * Logic Ops
+    */
+   if (spu.blend.logicop_enable) {
+      /* XXX to do */
+      /* apply logicop to 32-bit packed colors */
    }
-}
 
 
-qword
-spu_do_depth_stencil(int x, int y,
-                     qword frag_mask, qword frag_depth, qword frag_alpha,
-                     qword facing)
-{
-   struct spu_frag_test_results  result;
-   qword pixel_depth;
-   qword pixel_stencil;
-
-   /* All of this preable code (everthing before the call to frag_test) should
-    * be generated on the PPU and upload to the SPU.
+   /*
+    * If mask is non-zero, mark tile as dirty.
     */
-   if (spu.read_depth || spu.read_stencil) {
-      read_ds_quad(&spu.ztile, x, y, spu.fb.depth_format,
-                   &pixel_depth, &pixel_stencil);
+   if (spu_extract(spu_orx(mask), 0)) {
+      spu.cur_ctile_status = TILE_STATUS_DIRTY;
    }
-   
-   switch (spu.fb.depth_format) {
-   case PIPE_FORMAT_Z16_UNORM:
-      frag_depth = si_fm(frag_depth, (qword)spu_splats((float)(0x0000ffffu)));
-      frag_depth = si_cfltu(frag_depth, 0);
-      break;
-   case PIPE_FORMAT_Z32_UNORM:
-      frag_depth = si_fm(frag_depth, (qword)spu_splats((float)(0xffffffffu)));
-      frag_depth = si_cfltu(frag_depth, 0);
-      break;
-   case PIPE_FORMAT_Z24S8_UNORM:
-   case PIPE_FORMAT_S8Z24_UNORM:
-      frag_depth = si_fm(frag_depth, (qword)spu_splats((float)(0x00ffffffu)));
-      frag_depth = si_cfltu(frag_depth, 0);
-      break;
-   default:
-      ASSERT(0);
-      break;
+   else {
+      return;
    }
 
-   result = (*spu.frag_test)(frag_mask, pixel_depth, pixel_stencil,
-                             frag_depth, frag_alpha, facing);
-
 
-   /* This code (everthing after the call to frag_test) should
-    * be generated on the PPU and upload to the SPU.
+   /*
+    * Write new quad colors to the framebuffer/tile.
+    * Only write pixels where the corresponding mask word is set.
     */
-   if (spu.read_depth || spu.read_stencil) {
-      write_ds_quad(&spu.ztile, x, y, spu.fb.depth_format,
-                    result.depth, result.stencil);
-   }
-
-   return result.mask;
+#if LINEAR_QUAD_LAYOUT
+   /*
+    * Quad layout:
+    *  +--+--+--+--+
+    *  |p0|p1|p2|p3|
+    *  +--+--+--+--+
+    */
+   if (spu_extract(mask, 0))
+      colorTile->ui[y][x*2] = c0;
+   if (spu_extract(mask, 1))
+      colorTile->ui[y][x*2+1] = c1;
+   if (spu_extract(mask, 2))
+      colorTile->ui[y][x*2+2] = c2;
+   if (spu_extract(mask, 3))
+      colorTile->ui[y][x*2+3] = c3;
+#else
+   /*
+    * Quad layout:
+    *  +--+--+
+    *  |p0|p1|
+    *  +--+--+
+    *  |p2|p3|
+    *  +--+--+
+    */
+   if (spu_extract(mask, 0))
+      colorTile->ui[y+0][x+0] = c0;
+   if (spu_extract(mask, 1))
+      colorTile->ui[y+0][x+1] = c1;
+   if (spu_extract(mask, 2))
+      colorTile->ui[y+1][x+0] = c2;
+   if (spu_extract(mask, 3))
+      colorTile->ui[y+1][x+1] = c3;
+#endif
 }
diff --git a/src/gallium/drivers/cell/spu/spu_per_fragment_op.h b/src/gallium/drivers/cell/spu/spu_per_fragment_op.h
index 6571258699..f817abf046 100644
--- a/src/gallium/drivers/cell/spu/spu_per_fragment_op.h
+++ b/src/gallium/drivers/cell/spu/spu_per_fragment_op.h
@@ -1,32 +1,44 @@
-/*
- * (C) Copyright IBM Corporation 2008
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
  * All Rights Reserved.
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the "Software"),
- * to deal in the Software without restriction, including without limitation
- * on the rights to use, copy, modify, merge, publish, distribute, sub
- * license, and/or sell copies of the Software, and to permit persons to whom
- * the Software is furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice (including the next
- * paragraph) shall be included in all copies or substantial portions of the
- * Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.  IN NO EVENT SHALL
- * AUTHORS, COPYRIGHT HOLDERS, AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
- * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
- * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
- * USE OR OTHER DEALINGS IN THE SOFTWARE.
- */
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
 
 #ifndef SPU_PER_FRAGMENT_OP
 #define SPU_PER_FRAGMENT_OP
 
-extern qword
-spu_do_depth_stencil(int x, int y, qword frag_mask, qword frag_depth,
-		     qword frag_alpha, qword facing);
+
+extern void
+spu_fallback_fragment_ops(uint x, uint y,
+                          tile_t *colorTile,
+                          tile_t *depthStencilTile,
+                          vector float fragZ,
+                          vector float fragRed,
+                          vector float fragGreen,
+                          vector float fragBlue,
+                          vector float fragAlpha,
+                          vector unsigned int mask);
+
 
 #endif /* SPU_PER_FRAGMENT_OP */
diff --git a/src/gallium/drivers/cell/spu/spu_texture.c b/src/gallium/drivers/cell/spu/spu_texture.c
index 5051774f00..117b8a36f8 100644
--- a/src/gallium/drivers/cell/spu/spu_texture.c
+++ b/src/gallium/drivers/cell/spu/spu_texture.c
@@ -97,7 +97,7 @@ get_four_texels(uint unit, vec_uint4 x, vec_uint4 y, vec_uint4 *texels)
    const qword offset_y = si_andi((qword) y, 0x1f);
 
    const qword tiles_per_row = (qword) spu_splats(spu.texture[unit].tiles_per_row);
-   const qword tile_size = (qword) spu_splats(sizeof(tile_t));
+   const qword tile_size = (qword) spu_splats((unsigned) sizeof(tile_t));
 
    qword tile_offset = si_mpya((qword) tile_y, tiles_per_row, (qword) tile_x);
    tile_offset = si_mpy((qword) tile_offset, tile_size);
diff --git a/src/gallium/drivers/cell/spu/spu_tri.c b/src/gallium/drivers/cell/spu/spu_tri.c
index 2a4e0b423c..8b93878192 100644
--- a/src/gallium/drivers/cell/spu/spu_tri.c
+++ b/src/gallium/drivers/cell/spu/spu_tri.c
@@ -38,7 +38,6 @@
 #include "spu_texture.h"
 #include "spu_tile.h"
 #include "spu_tri.h"
-#include "spu_per_fragment_op.h"
 
 
 /** Masks are uint[4] vectors with each element being 0 or 0xffffffff */
@@ -209,7 +208,7 @@ clip_emit_quad(struct setup_stage *setup)
 /**
  * Evaluate attribute coefficients (plane equations) to compute
  * attribute values for the four fragments in a quad.
- * Eg: four colors will be compute.
+ * Eg: four colors will be computed (in AoS format).
  */
 static INLINE void
 eval_coeff(uint slot, float x, float y, vector float result[4])
@@ -255,31 +254,6 @@ eval_z(float x, float y)
 }
 
 
-static INLINE mask_t
-do_depth_test(int x, int y, mask_t quadmask)
-{
-   float4 zvals;
-   mask_t mask;
-
-   if (spu.fb.depth_format == PIPE_FORMAT_NONE)
-      return quadmask;
-
-   zvals.v = eval_z((float) x, (float) y);
-
-   mask = (mask_t) spu_do_depth_stencil(x - setup.cliprect_minx,
-					y - setup.cliprect_miny,
-					(qword) quadmask, 
-					(qword) zvals.v,
-					(qword) spu_splats((unsigned char) 0x0ffu),
-					(qword) spu_splats((unsigned int) 0x01u));
-
-   if (spu_extract(spu_orx(mask), 0))
-      spu.cur_ztile_status = TILE_STATUS_DIRTY;
-
-   return mask;
-}
-
-
 /**
  * Emit a quad (pass to next stage).  No clipping is done.
  * Note: about 1/5 to 1/7 of the time, mask is zero and this function
@@ -289,18 +263,6 @@ do_depth_test(int x, int y, mask_t quadmask)
 static INLINE void
 emit_quad( int x, int y, mask_t mask )
 {
-#if 0
-   struct softpipe_context *sp = setup.softpipe;
-   setup.quad.x0 = x;
-   setup.quad.y0 = y;
-   setup.quad.mask = mask;
-   sp->quad.first->run(sp->quad.first, &setup.quad);
-#else
-
-   if (spu.read_depth) {
-      mask = do_depth_test(x, y, mask);
-   }
-
    /* If any bits in mask are set... */
    if (spu_extract(spu_orx(mask), 0)) {
       const int ix = x - setup.cliprect_minx;
@@ -308,6 +270,7 @@ emit_quad( int x, int y, mask_t mask )
       vector float colors[4];
 
       spu.cur_ctile_status = TILE_STATUS_DIRTY;
+      spu.cur_ztile_status = TILE_STATUS_DIRTY;
 
       if (spu.texture[0].start) {
          /* texture mapping */
@@ -351,59 +314,68 @@ emit_quad( int x, int y, mask_t mask )
       }
       else {
          /* simple shading */
+#if 0
          eval_coeff(1, (float) x, (float) y, colors);
-      }
-
 
-      /* Convert fragment data from AoS to SoA format.
-       */
-      qword soa_frag[4];
-      _transpose_matrix4x4((vec_float4 *) soa_frag, colors);
+#else
+         /* XXX new fragment program code */
+
+         if (spu.fragment_program) {
+            vector float inputs[4*4], outputs[2*4];
+
+            /* setup inputs */
+            eval_coeff(1, (float) x, (float) y, inputs);
+
+            /* Execute the current fragment program */
+            spu.fragment_program(inputs, outputs, spu.constants);
+
+            /* Copy outputs */
+            colors[0] = outputs[0*4+0];
+            colors[1] = outputs[0*4+1];
+            colors[2] = outputs[0*4+2];
+            colors[3] = outputs[0*4+3];
+
+            if (0 && spu.init.id==0 && y == 48) {
+               printf("colors[0] = %f %f %f %f\n",
+                      spu_extract(colors[0], 0),
+                      spu_extract(colors[0], 1),
+                      spu_extract(colors[0], 2),
+                      spu_extract(colors[0], 3));
+               printf("colors[1] = %f %f %f %f\n",
+                      spu_extract(colors[1], 0),
+                      spu_extract(colors[1], 1),
+                      spu_extract(colors[1], 2),
+                      spu_extract(colors[1], 3));
+            }
 
-      /* Read the current framebuffer values.
-       */
-      const qword pix[4] = {
-         (qword) spu_splats(spu.ctile.ui[iy+0][ix+0]),
-         (qword) spu_splats(spu.ctile.ui[iy+0][ix+1]),
-         (qword) spu_splats(spu.ctile.ui[iy+1][ix+0]),
-         (qword) spu_splats(spu.ctile.ui[iy+1][ix+1]),
-      };
+         }
+#endif
+      }
 
-      qword soa_pix[4];
 
-      if (spu.read_fb) {
-         /* Convert pixel data from AoS to SoA format.
+      {
+         /* Convert fragment data from AoS to SoA format.
+          * I.e. (RGBA,RGBA,RGBA,RGBA) -> (RRRR,GGGG,BBBB,AAAA)
+          * This is temporary!
           */
-         vec_float4 aos_pix[4] = {
-            spu_unpack_A8R8G8B8(spu.ctile.ui[iy+0][ix+0]),
-            spu_unpack_A8R8G8B8(spu.ctile.ui[iy+0][ix+1]),
-            spu_unpack_A8R8G8B8(spu.ctile.ui[iy+1][ix+0]),
-            spu_unpack_A8R8G8B8(spu.ctile.ui[iy+1][ix+1]),
-         };
-
-         _transpose_matrix4x4((vec_float4 *) soa_pix, aos_pix);
-      }
+         vector float soa_frag[4];
+         _transpose_matrix4x4(soa_frag, colors);
 
+         float4 fragZ;
 
-      struct spu_blend_results result =
-          (*spu.blend)(soa_frag[0], soa_frag[1], soa_frag[2], soa_frag[3],
-                       soa_pix[0], soa_pix[1], soa_pix[2], soa_pix[3],
-                       spu.const_blend_color[0], spu.const_blend_color[1],
-                       spu.const_blend_color[2], spu.const_blend_color[3]);
+         fragZ.v = eval_z((float) x, (float) y);
 
+         /* Do all per-fragment/quad operations here, including:
+          *  alpha test, z test, stencil test, blend and framebuffer writing.
+          */
+         spu.fragment_ops(ix, iy, &spu.ctile, &spu.ztile,
+                          fragZ.v,
+                          soa_frag[0], soa_frag[1],
+                          soa_frag[2], soa_frag[3],
+                          mask);
+      }
 
-      /* Convert final pixel data from SoA to AoS format.
-       */
-      result = (*spu.logicop)(pix[0], pix[1], pix[2], pix[3],
-                              result.r, result.g, result.b, result.a,
-                              (qword) mask);
-
-      spu.ctile.ui[iy+0][ix+0] = spu_extract((vec_uint4) result.r, 0);
-      spu.ctile.ui[iy+0][ix+1] = spu_extract((vec_uint4) result.g, 0);
-      spu.ctile.ui[iy+1][ix+0] = spu_extract((vec_uint4) result.b, 0);
-      spu.ctile.ui[iy+1][ix+1] = spu_extract((vec_uint4) result.a, 0);
    }
-#endif
 }