From 5e00ae3fea51afd5de14d559f693645837b3903b Mon Sep 17 00:00:00 2001
From: Brian <brian.paul@tungstengraphics.com>
Date: Thu, 3 Jan 2008 09:56:48 -0700
Subject: [PATCH] Cell: initial work for getting/putting Z tiles

---
 src/mesa/pipe/cell/spu/main.c | 47 ++++++++++++++++++++++++++++---------------
 src/mesa/pipe/cell/spu/main.h |  7 ++++---
 src/mesa/pipe/cell/spu/tri.c  |  8 ++++----
 3 files changed, 39 insertions(+), 23 deletions(-)

diff --git a/src/mesa/pipe/cell/spu/main.c b/src/mesa/pipe/cell/spu/main.c
index bdb814e5d89..9580281971a 100644
--- a/src/mesa/pipe/cell/spu/main.c
+++ b/src/mesa/pipe/cell/spu/main.c
@@ -49,7 +49,8 @@ volatile struct cell_init_info init;
 
 struct framebuffer fb;
 
-uint tile[TILE_SIZE][TILE_SIZE] ALIGN16_ATTRIB;
+uint ctile[TILE_SIZE][TILE_SIZE] ALIGN16_ATTRIB;
+uint ztile[TILE_SIZE][TILE_SIZE] ALIGN16_ATTRIB;
 
 int DefaultTag;
 
@@ -66,11 +67,13 @@ wait_on_mask(unsigned tag)
 
 void
 get_tile(const struct framebuffer *fb, uint tx, uint ty, uint *tile,
-         int tag)
+         int tag, int zBuf)
 {
-   uint offset = ty * fb->width_tiles + tx;
-   uint bytesPerTile = TILE_SIZE * TILE_SIZE * 4;
-   ubyte *src = (ubyte *) fb->color_start + offset * bytesPerTile;
+   const uint offset = ty * fb->width_tiles + tx;
+   const uint bytesPerTile = TILE_SIZE * TILE_SIZE * 4;
+   const ubyte *src = zBuf ? fb->depth_start : fb->color_start;
+
+   src += offset * bytesPerTile;
 
    assert(tx < fb->width_tiles);
    assert(ty < fb->height_tiles);
@@ -90,11 +93,13 @@ get_tile(const struct framebuffer *fb, uint tx, uint ty, uint *tile,
 
 void
 put_tile(const struct framebuffer *fb, uint tx, uint ty, const uint *tile,
-         int tag)
+         int tag, int zBuf)
 {
-   uint offset = ty * fb->width_tiles + tx;
-   uint bytesPerTile = TILE_SIZE * TILE_SIZE * 4;
-   ubyte *dst = (ubyte *) fb->color_start + offset * bytesPerTile;
+   const uint offset = ty * fb->width_tiles + tx;
+   const uint bytesPerTile = TILE_SIZE * TILE_SIZE * 4;
+   ubyte *dst = zBuf ? fb->depth_start : fb->color_start;
+
+   dst += offset * bytesPerTile;
 
    assert(tx < fb->width_tiles);
    assert(ty < fb->height_tiles);
@@ -117,12 +122,12 @@ static void
 clear_tiles(const struct cell_command_clear_tiles *clear)
 {
    uint num_tiles = fb.width_tiles * fb.height_tiles;
-   uint i;
-   uint tile[TILE_SIZE * TILE_SIZE] ALIGN16_ATTRIB;
+   uint i, j;
    int tag = init.id;
 
-   for (i = 0; i < TILE_SIZE * TILE_SIZE; i++)
-      tile[i] = clear->value;
+   for (i = 0; i < TILE_SIZE; i++)
+      for (j = 0; j < TILE_SIZE; j++)
+         ctile[i][j] = clear->value;
 
    /*
    printf("SPU: %s num=%d w=%d h=%d\n",
@@ -132,7 +137,7 @@ clear_tiles(const struct cell_command_clear_tiles *clear)
    for (i = init.id; i < num_tiles; i += init.num_spus) {
       uint tx = i % fb.width_tiles;
       uint ty = i / fb.width_tiles;
-      put_tile(&fb, tx, ty, tile, tag);
+      put_tile(&fb, tx, ty, (uint *) ctile, tag, 0);
       /* XXX we don't want this here, but it fixes bad tile results */
       wait_on_mask(1 << tag);
    }
@@ -219,9 +224,14 @@ render(const struct cell_command_render *render)
       assert(tx < fb.width_tiles);
       assert(ty < fb.height_tiles);
 
-      get_tile(&fb, tx, ty, (uint *) tile, tag);
+      get_tile(&fb, tx, ty, (uint *) ctile, tag, 0);
       wait_on_mask(1 << tag);  /* XXX temporary */
 
+      if (fb.depth_format == PIPE_FORMAT_Z32_UNORM) {
+         get_tile(&fb, tx, ty, (uint *) ztile, tag+1, 1);
+         wait_on_mask(1 << (tag+1));  /* XXX temporary */
+      }
+
       assert(render->prim_type == PIPE_PRIM_TRIANGLES);
 
       /* loop over tris */
@@ -252,8 +262,13 @@ render(const struct cell_command_render *render)
          tri_draw(&prim, tx, ty);
       }
 
-      put_tile(&fb, tx, ty, (uint *) tile, tag);
+      put_tile(&fb, tx, ty, (uint *) ctile, tag, 0);
       wait_on_mask(1 << tag); /* XXX temp */
+
+      if (fb.depth_format == PIPE_FORMAT_Z32_UNORM) {
+         put_tile(&fb, tx, ty, (uint *) ztile, tag+1, 1);
+         wait_on_mask(1 << (tag+1));  /* XXX temporary */
+      }
    }
 }
 
diff --git a/src/mesa/pipe/cell/spu/main.h b/src/mesa/pipe/cell/spu/main.h
index 60a565bfa77..c22679c5d99 100644
--- a/src/mesa/pipe/cell/spu/main.h
+++ b/src/mesa/pipe/cell/spu/main.h
@@ -49,7 +49,8 @@ struct framebuffer {
 
 extern struct framebuffer fb;
 
-extern uint tile[TILE_SIZE][TILE_SIZE] ALIGN16_ATTRIB;
+extern uint ctile[TILE_SIZE][TILE_SIZE] ALIGN16_ATTRIB;
+extern uint ztile[TILE_SIZE][TILE_SIZE] ALIGN16_ATTRIB;
 
 extern int DefaultTag;
 
@@ -59,11 +60,11 @@ wait_on_mask(unsigned tag);
 
 void
 get_tile(const struct framebuffer *fb, uint tx, uint ty, uint *tile,
-         int tag);
+          int tag, int zBuf);
 
 void
 put_tile(const struct framebuffer *fb, uint tx, uint ty, const uint *tile,
-         int tag);
+         int tag, int zBuf);
 
 
 #endif /* MAIN_H */
diff --git a/src/mesa/pipe/cell/spu/tri.c b/src/mesa/pipe/cell/spu/tri.c
index f58cc4b0249..36599297b98 100644
--- a/src/mesa/pipe/cell/spu/tri.c
+++ b/src/mesa/pipe/cell/spu/tri.c
@@ -267,13 +267,13 @@ emit_quad( struct setup_stage *setup, int x, int y, unsigned mask )
    eval_coeff(setup, 1, (float) x, (float) y, colors);
 
    if (mask & MASK_TOP_LEFT)
-      tile[iy][ix] = pack_color(colors[QUAD_TOP_LEFT]);
+      ctile[iy][ix] = pack_color(colors[QUAD_TOP_LEFT]);
    if (mask & MASK_TOP_RIGHT)
-      tile[iy][ix+1] = pack_color(colors[QUAD_TOP_RIGHT]);
+      ctile[iy][ix+1] = pack_color(colors[QUAD_TOP_RIGHT]);
    if (mask & MASK_BOTTOM_LEFT)
-      tile[iy+1][ix] = pack_color(colors[QUAD_BOTTOM_LEFT]);
+      ctile[iy+1][ix] = pack_color(colors[QUAD_BOTTOM_LEFT]);
    if (mask & MASK_BOTTOM_RIGHT)
-      tile[iy+1][ix+1] = pack_color(colors[QUAD_BOTTOM_RIGHT]);
+      ctile[iy+1][ix+1] = pack_color(colors[QUAD_BOTTOM_RIGHT]);
 #endif
 }
 
-- 
2.11.0