From b642730be93149baa7556e5791393168ab396175 Mon Sep 17 00:00:00 2001
From: José Fonseca <jrfonseca@tungstengraphics.com>
Date: Fri, 15 Feb 2008 17:35:24 +0900
Subject: Code reorganization: move files into their places.

This is in a separate commit to ensure renames are properly preserved.
---
 src/gallium/drivers/cell/spu/spu_texture.c | 217 +++++++++++++++++++++++++++++
 1 file changed, 217 insertions(+)
 create mode 100644 src/gallium/drivers/cell/spu/spu_texture.c

(limited to 'src/gallium/drivers/cell/spu/spu_texture.c')

diff --git a/src/gallium/drivers/cell/spu/spu_texture.c b/src/gallium/drivers/cell/spu/spu_texture.c
new file mode 100644
index 0000000000..3962aaa4a9
--- /dev/null
+++ b/src/gallium/drivers/cell/spu/spu_texture.c
@@ -0,0 +1,217 @@
+/**************************************************************************
+ * 
+ * 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 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.
+ * 
+ **************************************************************************/
+
+
+#include "pipe/p_compiler.h"
+#include "spu_main.h"
+#include "spu_texture.h"
+#include "spu_tile.h"
+#include "spu_colorpack.h"
+
+
+/**
+ * Number of texture tiles to cache.
+ * Note that this will probably be the largest consumer of SPU local store/
+ * memory for this driver!
+ */
+#define CACHE_SIZE 16
+
+static tile_t tex_tiles[CACHE_SIZE]  ALIGN16_ATTRIB;
+
+static vector unsigned int tex_tile_xy[CACHE_SIZE];
+
+
+
+/**
+ * Mark all tex cache entries as invalid.
+ */
+void
+invalidate_tex_cache(void)
+{
+   /* XXX memset? */
+   uint i;
+   for (i = 0; i < CACHE_SIZE; i++) {
+      tex_tile_xy[i] = ((vector unsigned int) { ~0U, ~0U, ~0U, ~0U });
+   }
+}
+
+
+/**
+ * Return the cache pos/index which corresponds to tile (tx,ty)
+ */
+static INLINE uint
+cache_pos(vector unsigned int txty)
+{
+   uint pos = (spu_extract(txty,0) + spu_extract(txty,1) * 4) % CACHE_SIZE;
+   return pos;
+}
+
+
+/**
+ * Make sure the tile for texel (i,j) is present, return its position/index
+ * in the cache.
+ */
+static uint
+get_tex_tile(vector unsigned int ij)
+{
+   /* tile address: tx,ty */
+   const vector unsigned int txty = spu_rlmask(ij, -5);  /* divide by 32 */
+   const uint pos = cache_pos(txty);
+
+   if ((spu_extract(tex_tile_xy[pos], 0) != spu_extract(txty, 0)) ||
+       (spu_extract(tex_tile_xy[pos], 1) != spu_extract(txty, 1))) {
+
+      /* texture cache miss, fetch tile from main memory */
+      const uint tiles_per_row = spu.texture.width / TILE_SIZE;
+      const uint bytes_per_tile = sizeof(tile_t);
+      const void *src = (const ubyte *) spu.texture.start
+         + (spu_extract(txty,1) * tiles_per_row + spu_extract(txty,0)) * bytes_per_tile;
+
+      printf("SPU %u: tex cache miss at %d, %d  pos=%u  old=%d,%d\n",
+             spu.init.id,
+             spu_extract(txty,0),
+             spu_extract(txty,1),
+             pos,
+             spu_extract(tex_tile_xy[pos],0),
+             spu_extract(tex_tile_xy[pos],1));
+
+      ASSERT_ALIGN16(tex_tiles[pos].ui);
+      ASSERT_ALIGN16(src);
+
+      mfc_get(tex_tiles[pos].ui,  /* dest */
+              (unsigned int) src,
+              bytes_per_tile,      /* size */
+              TAG_TEXTURE_TILE,
+              0, /* tid */
+              0  /* rid */);
+
+      wait_on_mask(1 << TAG_TEXTURE_TILE);
+
+      tex_tile_xy[pos] = txty;
+   }
+   else {
+#if 0
+      printf("SPU %u: tex cache HIT at %d, %d\n",
+             spu.init.id, tx, ty);
+#endif
+   }
+
+   return pos;
+}
+
+
+/**
+ * Get texture sample at texcoord.
+ * XXX this is extremely primitive for now.
+ */
+vector float
+sample_texture_nearest(vector float texcoord)
+{
+   vector float tc = spu_mul(texcoord, spu.tex_size);
+   vector unsigned int itc = spu_convtu(tc, 0);  /* convert to int */
+   itc = spu_and(itc, spu.tex_size_mask);        /* mask (GL_REPEAT) */
+   vector unsigned int ij = spu_and(itc, TILE_SIZE-1); /* intra tile addr */
+   uint pos = get_tex_tile(itc);
+   uint texel = tex_tiles[pos].ui[spu_extract(ij, 1)][spu_extract(ij, 0)];
+   return spu_unpack_A8R8G8B8(texel);
+}
+
+
+vector float
+sample_texture_bilinear(vector float texcoord)
+{
+   static const vector unsigned int offset10 = {1, 0, 0, 0};
+   static const vector unsigned int offset01 = {0, 1, 0, 0};
+
+   vector float tc = spu_mul(texcoord, spu.tex_size);
+   tc = spu_add(tc, spu_splats(-0.5f));  /* half texel bias */
+
+   /* integer texcoords S,T: */
+   vector unsigned int itc00 = spu_convtu(tc, 0);  /* convert to int */
+   vector unsigned int itc01 = spu_add(itc00, offset01);
+   vector unsigned int itc10 = spu_add(itc00, offset10);
+   vector unsigned int itc11 = spu_add(itc10, offset01);
+
+   /* mask (GL_REPEAT) */
+   itc00 = spu_and(itc00, spu.tex_size_mask);
+   itc01 = spu_and(itc01, spu.tex_size_mask);
+   itc10 = spu_and(itc10, spu.tex_size_mask);
+   itc11 = spu_and(itc11, spu.tex_size_mask);
+
+   /* intra tile addr */
+   vector unsigned int ij00 = spu_and(itc00, TILE_SIZE-1);
+   vector unsigned int ij01 = spu_and(itc01, TILE_SIZE-1);
+   vector unsigned int ij10 = spu_and(itc10, TILE_SIZE-1);
+   vector unsigned int ij11 = spu_and(itc11, TILE_SIZE-1);
+
+   /* get tile cache positions */
+   uint pos00 = get_tex_tile(itc00);
+   uint pos01, pos10, pos11;
+   if ((spu_extract(ij00, 0) < TILE_SIZE-1) &&
+       (spu_extract(ij00, 1) < TILE_SIZE-1)) {
+      /* all texels are in the same tile */
+      pos01 = pos10 = pos11 = pos00;
+   }
+   else {
+      pos01 = get_tex_tile(itc01);
+      pos10 = get_tex_tile(itc10);
+      pos11 = get_tex_tile(itc11);
+   }
+
+   /* get texels from tiles and convert to float[4] */
+   vector float texel00 = spu_unpack_A8R8G8B8(tex_tiles[pos00].ui[spu_extract(ij00, 1)][spu_extract(ij00, 0)]);
+   vector float texel01 = spu_unpack_A8R8G8B8(tex_tiles[pos01].ui[spu_extract(ij01, 1)][spu_extract(ij01, 0)]);
+   vector float texel10 = spu_unpack_A8R8G8B8(tex_tiles[pos10].ui[spu_extract(ij10, 1)][spu_extract(ij10, 0)]);
+   vector float texel11 = spu_unpack_A8R8G8B8(tex_tiles[pos11].ui[spu_extract(ij11, 1)][spu_extract(ij11, 0)]);
+
+   /* Compute weighting factors in [0,1]
+    * Multiply texcoord by 1024, AND with 1023, convert back to float.
+    */
+   vector float tc1024 = spu_mul(tc, spu_splats(1024.0f));
+   vector signed int itc1024 = spu_convts(tc1024, 0);
+   itc1024 = spu_and(itc1024, spu_splats((1 << 10) - 1));
+   vector float weight = spu_convtf(itc1024, 10);
+
+   /* smeared frac and 1-frac */
+   vector float sfrac = spu_splats(spu_extract(weight, 0));
+   vector float tfrac = spu_splats(spu_extract(weight, 1));
+   vector float sfrac1 = spu_sub(spu_splats(1.0f), sfrac);
+   vector float tfrac1 = spu_sub(spu_splats(1.0f), tfrac);
+
+   /* multiply the samples (colors) by the S/T weights */
+   texel00 = spu_mul(spu_mul(texel00, sfrac1), tfrac1);
+   texel10 = spu_mul(spu_mul(texel10, sfrac ), tfrac1);
+   texel01 = spu_mul(spu_mul(texel01, sfrac1), tfrac );
+   texel11 = spu_mul(spu_mul(texel11, sfrac ), tfrac );
+
+   /* compute sum of weighted samples */
+   vector float texel_sum = spu_add(texel00, texel01);
+   texel_sum = spu_add(texel_sum, texel10);
+   texel_sum = spu_add(texel_sum, texel11);
+
+   return texel_sum;
+}
-- 
cgit v1.2.3


From 1c50ea2cd9ab8752793c99b4a7a2a6656bdde1ac Mon Sep 17 00:00:00 2001
From: Ian Romanick <idr@us.ibm.com>
Date: Wed, 27 Feb 2008 13:40:23 -0800
Subject: cell: Use unified data cache for textures too

---
 src/gallium/drivers/cell/spu/spu_main.c    |   2 +
 src/gallium/drivers/cell/spu/spu_main.h    |   3 +-
 src/gallium/drivers/cell/spu/spu_texture.c | 184 +++++++++++------------------
 3 files changed, 72 insertions(+), 117 deletions(-)

(limited to 'src/gallium/drivers/cell/spu/spu_texture.c')

diff --git a/src/gallium/drivers/cell/spu/spu_main.c b/src/gallium/drivers/cell/spu/spu_main.c
index cc4bafdb3a..59300028d4 100644
--- a/src/gallium/drivers/cell/spu/spu_main.c
+++ b/src/gallium/drivers/cell/spu/spu_main.c
@@ -286,6 +286,8 @@ cmd_state_texture(const struct cell_command_texture *texture)
       { spu.texture.width, spu.texture.height, 0.0, 0.0};
    spu.tex_size_mask = (vector unsigned int)
       { spu.texture.width - 1, spu.texture.height - 1, 0, 0 };
+   spu.tex_size_x_mask = spu_splats(spu.texture.width - 1);
+   spu.tex_size_y_mask = spu_splats(spu.texture.height - 1);
 }
 
 
diff --git a/src/gallium/drivers/cell/spu/spu_main.h b/src/gallium/drivers/cell/spu/spu_main.h
index d14f1abbe7..a13edd1702 100644
--- a/src/gallium/drivers/cell/spu/spu_main.h
+++ b/src/gallium/drivers/cell/spu/spu_main.h
@@ -107,6 +107,8 @@ struct spu_global
 
    vector float tex_size;
    vector unsigned int tex_size_mask; /**< == int(size - 1) */
+   vector unsigned int tex_size_x_mask; /**< == int(size - 1) */
+   vector unsigned int tex_size_y_mask; /**< == int(size - 1) */
 
    vector float (*sample_texture)(vector float texcoord);
 
@@ -130,7 +132,6 @@ extern boolean Debug;
 #define TAG_INDEX_BUFFER      16
 #define TAG_BATCH_BUFFER      17
 #define TAG_MISC              18
-#define TAG_TEXTURE_TILE      19
 #define TAG_DCACHE0           20
 #define TAG_DCACHE1           21
 #define TAG_DCACHE2           22
diff --git a/src/gallium/drivers/cell/spu/spu_texture.c b/src/gallium/drivers/cell/spu/spu_texture.c
index 3962aaa4a9..67eb08196a 100644
--- a/src/gallium/drivers/cell/spu/spu_texture.c
+++ b/src/gallium/drivers/cell/spu/spu_texture.c
@@ -31,19 +31,7 @@
 #include "spu_texture.h"
 #include "spu_tile.h"
 #include "spu_colorpack.h"
-
-
-/**
- * Number of texture tiles to cache.
- * Note that this will probably be the largest consumer of SPU local store/
- * memory for this driver!
- */
-#define CACHE_SIZE 16
-
-static tile_t tex_tiles[CACHE_SIZE]  ALIGN16_ATTRIB;
-
-static vector unsigned int tex_tile_xy[CACHE_SIZE];
-
+#include "spu_dcache.h"
 
 
 /**
@@ -52,78 +40,60 @@ static vector unsigned int tex_tile_xy[CACHE_SIZE];
 void
 invalidate_tex_cache(void)
 {
-   /* XXX memset? */
-   uint i;
-   for (i = 0; i < CACHE_SIZE; i++) {
-      tex_tile_xy[i] = ((vector unsigned int) { ~0U, ~0U, ~0U, ~0U });
-   }
+   spu_dcache_mark_dirty((unsigned) spu.texture.start,
+                         4 * spu.texture.width * spu.texture.height);
 }
 
 
-/**
- * Return the cache pos/index which corresponds to tile (tx,ty)
- */
-static INLINE uint
-cache_pos(vector unsigned int txty)
+static uint
+get_texel(vec_uint4 coordinate)
 {
-   uint pos = (spu_extract(txty,0) + spu_extract(txty,1) * 4) % CACHE_SIZE;
-   return pos;
+   vec_uint4 tmp;
+   unsigned x = spu_extract(coordinate, 0);
+   unsigned y = spu_extract(coordinate, 1);
+   const unsigned tiles_per_row = spu.texture.width / TILE_SIZE;
+   unsigned tile_offset = sizeof(tile_t) * ((y / TILE_SIZE * tiles_per_row) 
+                                            + (x / TILE_SIZE));
+   unsigned texel_offset = 4 * (((y % TILE_SIZE) * TILE_SIZE)
+                                + (x % TILE_SIZE));
+
+   spu_dcache_fetch_unaligned((qword *) & tmp,
+                              spu.texture.start + tile_offset + texel_offset,
+                              4);
+   return spu_extract(tmp, 0);
 }
 
 
-/**
- * Make sure the tile for texel (i,j) is present, return its position/index
- * in the cache.
- */
-static uint
-get_tex_tile(vector unsigned int ij)
+static void
+get_four_texels(vec_uint4 x, vec_uint4 y, vec_uint4 *texels)
 {
-   /* tile address: tx,ty */
-   const vector unsigned int txty = spu_rlmask(ij, -5);  /* divide by 32 */
-   const uint pos = cache_pos(txty);
-
-   if ((spu_extract(tex_tile_xy[pos], 0) != spu_extract(txty, 0)) ||
-       (spu_extract(tex_tile_xy[pos], 1) != spu_extract(txty, 1))) {
-
-      /* texture cache miss, fetch tile from main memory */
-      const uint tiles_per_row = spu.texture.width / TILE_SIZE;
-      const uint bytes_per_tile = sizeof(tile_t);
-      const void *src = (const ubyte *) spu.texture.start
-         + (spu_extract(txty,1) * tiles_per_row + spu_extract(txty,0)) * bytes_per_tile;
-
-      printf("SPU %u: tex cache miss at %d, %d  pos=%u  old=%d,%d\n",
-             spu.init.id,
-             spu_extract(txty,0),
-             spu_extract(txty,1),
-             pos,
-             spu_extract(tex_tile_xy[pos],0),
-             spu_extract(tex_tile_xy[pos],1));
-
-      ASSERT_ALIGN16(tex_tiles[pos].ui);
-      ASSERT_ALIGN16(src);
-
-      mfc_get(tex_tiles[pos].ui,  /* dest */
-              (unsigned int) src,
-              bytes_per_tile,      /* size */
-              TAG_TEXTURE_TILE,
-              0, /* tid */
-              0  /* rid */);
-
-      wait_on_mask(1 << TAG_TEXTURE_TILE);
-
-      tex_tile_xy[pos] = txty;
-   }
-   else {
-#if 0
-      printf("SPU %u: tex cache HIT at %d, %d\n",
-             spu.init.id, tx, ty);
-#endif
-   }
-
-   return pos;
+   const unsigned texture_ea = (uintptr_t) spu.texture.start;
+   vec_uint4 tile_x = spu_rlmask(x, -5);
+   vec_uint4 tile_y = spu_rlmask(y, -5);
+   const qword offset_x = si_andi((qword) x, 0x1f);
+   const qword offset_y = si_andi((qword) y, 0x1f);
+
+   const qword tiles_per_row = (qword) spu_splats(spu.texture.width / TILE_SIZE);
+   const qword tile_size = (qword) spu_splats(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);
+
+   qword texel_offset = si_a(si_mpyui(offset_y, 32), offset_x);
+   texel_offset = si_mpyui(texel_offset, 4);
+   
+   vec_uint4 offset = (vec_uint4) si_a(tile_offset, texel_offset);
+   
+   spu_dcache_fetch_unaligned((qword *) & texels[0],
+                              texture_ea + spu_extract(offset, 0), 4);
+   spu_dcache_fetch_unaligned((qword *) & texels[1],
+                              texture_ea + spu_extract(offset, 1), 4);
+   spu_dcache_fetch_unaligned((qword *) & texels[2],
+                              texture_ea + spu_extract(offset, 2), 4);
+   spu_dcache_fetch_unaligned((qword *) & texels[3],
+                              texture_ea + spu_extract(offset, 3), 4);
 }
 
-
 /**
  * Get texture sample at texcoord.
  * XXX this is extremely primitive for now.
@@ -134,9 +104,7 @@ sample_texture_nearest(vector float texcoord)
    vector float tc = spu_mul(texcoord, spu.tex_size);
    vector unsigned int itc = spu_convtu(tc, 0);  /* convert to int */
    itc = spu_and(itc, spu.tex_size_mask);        /* mask (GL_REPEAT) */
-   vector unsigned int ij = spu_and(itc, TILE_SIZE-1); /* intra tile addr */
-   uint pos = get_tex_tile(itc);
-   uint texel = tex_tiles[pos].ui[spu_extract(ij, 1)][spu_extract(ij, 0)];
+   uint texel = get_texel(itc);
    return spu_unpack_A8R8G8B8(texel);
 }
 
@@ -144,49 +112,33 @@ sample_texture_nearest(vector float texcoord)
 vector float
 sample_texture_bilinear(vector float texcoord)
 {
-   static const vector unsigned int offset10 = {1, 0, 0, 0};
-   static const vector unsigned int offset01 = {0, 1, 0, 0};
+   static const vec_uint4 offset_x = {0, 0, 1, 1};
+   static const vec_uint4 offset_y = {0, 1, 0, 1};
 
    vector float tc = spu_mul(texcoord, spu.tex_size);
    tc = spu_add(tc, spu_splats(-0.5f));  /* half texel bias */
 
    /* integer texcoords S,T: */
-   vector unsigned int itc00 = spu_convtu(tc, 0);  /* convert to int */
-   vector unsigned int itc01 = spu_add(itc00, offset01);
-   vector unsigned int itc10 = spu_add(itc00, offset10);
-   vector unsigned int itc11 = spu_add(itc10, offset01);
-
-   /* mask (GL_REPEAT) */
-   itc00 = spu_and(itc00, spu.tex_size_mask);
-   itc01 = spu_and(itc01, spu.tex_size_mask);
-   itc10 = spu_and(itc10, spu.tex_size_mask);
-   itc11 = spu_and(itc11, spu.tex_size_mask);
-
-   /* intra tile addr */
-   vector unsigned int ij00 = spu_and(itc00, TILE_SIZE-1);
-   vector unsigned int ij01 = spu_and(itc01, TILE_SIZE-1);
-   vector unsigned int ij10 = spu_and(itc10, TILE_SIZE-1);
-   vector unsigned int ij11 = spu_and(itc11, TILE_SIZE-1);
-
-   /* get tile cache positions */
-   uint pos00 = get_tex_tile(itc00);
-   uint pos01, pos10, pos11;
-   if ((spu_extract(ij00, 0) < TILE_SIZE-1) &&
-       (spu_extract(ij00, 1) < TILE_SIZE-1)) {
-      /* all texels are in the same tile */
-      pos01 = pos10 = pos11 = pos00;
-   }
-   else {
-      pos01 = get_tex_tile(itc01);
-      pos10 = get_tex_tile(itc10);
-      pos11 = get_tex_tile(itc11);
-   }
-
-   /* get texels from tiles and convert to float[4] */
-   vector float texel00 = spu_unpack_A8R8G8B8(tex_tiles[pos00].ui[spu_extract(ij00, 1)][spu_extract(ij00, 0)]);
-   vector float texel01 = spu_unpack_A8R8G8B8(tex_tiles[pos01].ui[spu_extract(ij01, 1)][spu_extract(ij01, 0)]);
-   vector float texel10 = spu_unpack_A8R8G8B8(tex_tiles[pos10].ui[spu_extract(ij10, 1)][spu_extract(ij10, 0)]);
-   vector float texel11 = spu_unpack_A8R8G8B8(tex_tiles[pos11].ui[spu_extract(ij11, 1)][spu_extract(ij11, 0)]);
+   vec_uint4 itc = spu_convtu(tc, 0);  /* convert to int */
+
+   vec_uint4 texels[4];
+   
+   vec_uint4 x = spu_splats(spu_extract(itc, 0));
+   vec_uint4 y = spu_splats(spu_extract(itc, 1));
+
+   x = spu_add(x, offset_x);
+   y = spu_add(y, offset_y);
+
+   x = spu_and(x, spu.tex_size_x_mask);
+   y = spu_and(y, spu.tex_size_y_mask);
+
+   get_four_texels(x, y, texels);
+
+   vector float texel00 = spu_unpack_A8R8G8B8(spu_extract(texels[0], 0));
+   vector float texel01 = spu_unpack_A8R8G8B8(spu_extract(texels[1], 0));
+   vector float texel10 = spu_unpack_A8R8G8B8(spu_extract(texels[2], 0));
+   vector float texel11 = spu_unpack_A8R8G8B8(spu_extract(texels[3], 0));
+
 
    /* Compute weighting factors in [0,1]
     * Multiply texcoord by 1024, AND with 1023, convert back to float.
-- 
cgit v1.2.3


From 14452aee73e16f2ede075cf894e69d62cc539f5e Mon Sep 17 00:00:00 2001
From: Brian <brian.paul@tungstengraphics.com>
Date: Mon, 31 Mar 2008 17:38:21 -0600
Subject: cell: initial work to support multi-texture

---
 src/gallium/drivers/cell/common.h              |  8 +++++--
 src/gallium/drivers/cell/ppu/cell_state_emit.c | 18 +++++++---------
 src/gallium/drivers/cell/spu/spu_main.c        | 27 ++++++++++++++++--------
 src/gallium/drivers/cell/spu/spu_main.h        |  8 +++----
 src/gallium/drivers/cell/spu/spu_texture.c     | 29 ++++++++++++++++----------
 src/gallium/drivers/cell/spu/spu_tri.c         |  2 +-
 6 files changed, 55 insertions(+), 37 deletions(-)

(limited to 'src/gallium/drivers/cell/spu/spu_texture.c')

diff --git a/src/gallium/drivers/cell/common.h b/src/gallium/drivers/cell/common.h
index c9e873b35c..298812fc20 100644
--- a/src/gallium/drivers/cell/common.h
+++ b/src/gallium/drivers/cell/common.h
@@ -66,6 +66,8 @@
 
 #define CELL_MAX_SPUS 6
 
+#define CELL_MAX_SAMPLERS 4
+
 #define TILE_SIZE 32
 
 
@@ -228,8 +230,10 @@ struct cell_command_release_verts
 
 struct cell_command_texture
 {
-   void *start;         /**< Address in main memory */
-   uint width, height;
+   struct {
+      void *start;         /**< Address in main memory */
+      ushort width, height;
+   } texture[CELL_MAX_SAMPLERS];
 };
 
 
diff --git a/src/gallium/drivers/cell/ppu/cell_state_emit.c b/src/gallium/drivers/cell/ppu/cell_state_emit.c
index 4c75caa025..4fbe1a21b8 100644
--- a/src/gallium/drivers/cell/ppu/cell_state_emit.c
+++ b/src/gallium/drivers/cell/ppu/cell_state_emit.c
@@ -129,17 +129,15 @@ cell_emit_state(struct cell_context *cell)
 
    if (cell->dirty & CELL_NEW_TEXTURE) {
       struct cell_command_texture texture;
-      if (cell->texture[0]) {
-         texture.start = cell->texture[0]->tiled_data;
-         texture.width = cell->texture[0]->base.width[0];
-         texture.height = cell->texture[0]->base.height[0];
+      uint i;
+      memset(&texture, 0, sizeof(texture));
+      for (i = 0;i < CELL_MAX_SAMPLERS; i++) {
+         if (cell->texture[i]) {
+            texture.texture[i].start = cell->texture[i]->tiled_data;
+            texture.texture[i].width = cell->texture[i]->base.width[0];
+            texture.texture[i].height = cell->texture[i]->base.height[0];
+         }
       }
-      else {
-         texture.start = NULL;
-         texture.width = 0;
-         texture.height = 0;
-      }
-
       emit_state_cmd(cell, CELL_CMD_STATE_TEXTURE,
                      &texture, sizeof(struct cell_command_texture));
    }
diff --git a/src/gallium/drivers/cell/spu/spu_main.c b/src/gallium/drivers/cell/spu/spu_main.c
index d7f46f8024..80fa5f7859 100644
--- a/src/gallium/drivers/cell/spu/spu_main.c
+++ b/src/gallium/drivers/cell/spu/spu_main.c
@@ -329,17 +329,26 @@ cmd_state_sampler(const struct pipe_sampler_state *state)
 static void
 cmd_state_texture(const struct cell_command_texture *texture)
 {
-   if (Debug)
-      printf("SPU %u: TEXTURE at %p  size %u x %u\n",
-             spu.init.id, texture->start, texture->width, texture->height);
+   uint i;
+
+   if (1||Debug) {
+      printf("SPU %u: TEXTURE\n", spu.init.id);
+      for (i = 0; i < CELL_MAX_SAMPLERS; i++) {
+         printf("  %d: at %p  size %u x %u\n", i, texture->texture[i].start,
+                texture->texture[i].width, texture->texture[i].height);
+      }
+   }
 
    memcpy(&spu.texture, texture, sizeof(*texture));
-   spu.tex_size = (vector float)
-      { spu.texture.width, spu.texture.height, 0.0, 0.0};
-   spu.tex_size_mask = (vector unsigned int)
-      { spu.texture.width - 1, spu.texture.height - 1, 0, 0 };
-   spu.tex_size_x_mask = spu_splats(spu.texture.width - 1);
-   spu.tex_size_y_mask = spu_splats(spu.texture.height - 1);
+   for (i = 0; i < CELL_MAX_SAMPLERS; i++) {
+      const uint width = texture->texture[i].width;
+      const uint height = texture->texture[i].height;
+      spu.tex_size[i] = (vector float) { width, height, 0.0, 0.0};
+      spu.tex_size_mask[i] = (vector unsigned int)
+         { width - 1, height - 1, 0, 0 };
+      spu.tex_size_x_mask[i] = spu_splats(width - 1);
+      spu.tex_size_y_mask[i] = spu_splats(height - 1);
+   }
 }
 
 
diff --git a/src/gallium/drivers/cell/spu/spu_main.h b/src/gallium/drivers/cell/spu/spu_main.h
index c20452931a..8a87787537 100644
--- a/src/gallium/drivers/cell/spu/spu_main.h
+++ b/src/gallium/drivers/cell/spu/spu_main.h
@@ -141,10 +141,10 @@ struct spu_global
    /** for converting RGBA to PIPE_FORMAT_x colors */
    vector unsigned char color_shuffle;
 
-   vector float tex_size;
-   vector unsigned int tex_size_mask; /**< == int(size - 1) */
-   vector unsigned int tex_size_x_mask; /**< == int(size - 1) */
-   vector unsigned int tex_size_y_mask; /**< == int(size - 1) */
+   vector float tex_size[CELL_MAX_SAMPLERS];
+   vector unsigned int tex_size_mask[CELL_MAX_SAMPLERS]; /**< == int(size - 1) */
+   vector unsigned int tex_size_x_mask[CELL_MAX_SAMPLERS]; /**< == int(size - 1) */
+   vector unsigned int tex_size_y_mask[CELL_MAX_SAMPLERS]; /**< == int(size - 1) */
 
    vector float (*sample_texture)(vector float texcoord);
 
diff --git a/src/gallium/drivers/cell/spu/spu_texture.c b/src/gallium/drivers/cell/spu/spu_texture.c
index 67eb08196a..91a6aec5ec 100644
--- a/src/gallium/drivers/cell/spu/spu_texture.c
+++ b/src/gallium/drivers/cell/spu/spu_texture.c
@@ -40,25 +40,29 @@
 void
 invalidate_tex_cache(void)
 {
-   spu_dcache_mark_dirty((unsigned) spu.texture.start,
-                         4 * spu.texture.width * spu.texture.height);
+   uint unit = 0;
+   uint bytes = 4 * spu.texture.texture[unit].width
+      * spu.texture.texture[unit].height;
+
+   spu_dcache_mark_dirty((unsigned) spu.texture.texture[unit].start, bytes);
 }
 
 
 static uint
 get_texel(vec_uint4 coordinate)
 {
+   const uint unit = 0;
    vec_uint4 tmp;
    unsigned x = spu_extract(coordinate, 0);
    unsigned y = spu_extract(coordinate, 1);
-   const unsigned tiles_per_row = spu.texture.width / TILE_SIZE;
+   const unsigned tiles_per_row = spu.texture.texture[unit].width / TILE_SIZE;
    unsigned tile_offset = sizeof(tile_t) * ((y / TILE_SIZE * tiles_per_row) 
                                             + (x / TILE_SIZE));
    unsigned texel_offset = 4 * (((y % TILE_SIZE) * TILE_SIZE)
                                 + (x % TILE_SIZE));
 
    spu_dcache_fetch_unaligned((qword *) & tmp,
-                              spu.texture.start + tile_offset + texel_offset,
+                              spu.texture.texture[unit].start + tile_offset + texel_offset,
                               4);
    return spu_extract(tmp, 0);
 }
@@ -67,13 +71,14 @@ get_texel(vec_uint4 coordinate)
 static void
 get_four_texels(vec_uint4 x, vec_uint4 y, vec_uint4 *texels)
 {
-   const unsigned texture_ea = (uintptr_t) spu.texture.start;
+   const uint unit = 0;
+   const unsigned texture_ea = (uintptr_t) spu.texture.texture[unit].start;
    vec_uint4 tile_x = spu_rlmask(x, -5);
    vec_uint4 tile_y = spu_rlmask(y, -5);
    const qword offset_x = si_andi((qword) x, 0x1f);
    const qword offset_y = si_andi((qword) y, 0x1f);
 
-   const qword tiles_per_row = (qword) spu_splats(spu.texture.width / TILE_SIZE);
+   const qword tiles_per_row = (qword) spu_splats(spu.texture.texture[unit].width / TILE_SIZE);
    const qword tile_size = (qword) spu_splats(sizeof(tile_t));
 
    qword tile_offset = si_mpya((qword) tile_y, tiles_per_row, (qword) tile_x);
@@ -101,9 +106,10 @@ get_four_texels(vec_uint4 x, vec_uint4 y, vec_uint4 *texels)
 vector float
 sample_texture_nearest(vector float texcoord)
 {
-   vector float tc = spu_mul(texcoord, spu.tex_size);
+   const uint unit = 0;
+   vector float tc = spu_mul(texcoord, spu.tex_size[unit]);
    vector unsigned int itc = spu_convtu(tc, 0);  /* convert to int */
-   itc = spu_and(itc, spu.tex_size_mask);        /* mask (GL_REPEAT) */
+   itc = spu_and(itc, spu.tex_size_mask[unit]);        /* mask (GL_REPEAT) */
    uint texel = get_texel(itc);
    return spu_unpack_A8R8G8B8(texel);
 }
@@ -112,10 +118,11 @@ sample_texture_nearest(vector float texcoord)
 vector float
 sample_texture_bilinear(vector float texcoord)
 {
+   const uint unit = 0;
    static const vec_uint4 offset_x = {0, 0, 1, 1};
    static const vec_uint4 offset_y = {0, 1, 0, 1};
 
-   vector float tc = spu_mul(texcoord, spu.tex_size);
+   vector float tc = spu_mul(texcoord, spu.tex_size[unit]);
    tc = spu_add(tc, spu_splats(-0.5f));  /* half texel bias */
 
    /* integer texcoords S,T: */
@@ -129,8 +136,8 @@ sample_texture_bilinear(vector float texcoord)
    x = spu_add(x, offset_x);
    y = spu_add(y, offset_y);
 
-   x = spu_and(x, spu.tex_size_x_mask);
-   y = spu_and(y, spu.tex_size_y_mask);
+   x = spu_and(x, spu.tex_size_x_mask[unit]);
+   y = spu_and(y, spu.tex_size_y_mask[unit]);
 
    get_four_texels(x, y, texels);
 
diff --git a/src/gallium/drivers/cell/spu/spu_tri.c b/src/gallium/drivers/cell/spu/spu_tri.c
index 95c629a8aa..9f63317b1f 100644
--- a/src/gallium/drivers/cell/spu/spu_tri.c
+++ b/src/gallium/drivers/cell/spu/spu_tri.c
@@ -309,7 +309,7 @@ emit_quad( int x, int y, mask_t mask )
 
       spu.cur_ctile_status = TILE_STATUS_DIRTY;
 
-      if (spu.texture.start) {
+      if (spu.texture.texture[0].start) {
          /* texture mapping */
          vector float texcoords[4];
          eval_coeff(2, (float) x, (float) y, texcoords);
-- 
cgit v1.2.3


From e6c981f22c0b6469ef44e9d7a34113db34647fef Mon Sep 17 00:00:00 2001
From: Brian <brian.paul@tungstengraphics.com>
Date: Mon, 31 Mar 2008 21:09:02 -0600
Subject: cell: more work for multi-texture support

---
 src/gallium/drivers/cell/common.h              | 17 ++++++--
 src/gallium/drivers/cell/ppu/cell_state_emit.c | 31 ++++++++++-----
 src/gallium/drivers/cell/spu/spu_main.c        | 55 +++++++++++++++-----------
 src/gallium/drivers/cell/spu/spu_main.h        | 18 ++++++---
 src/gallium/drivers/cell/spu/spu_texture.c     | 24 +++++------
 src/gallium/drivers/cell/spu/spu_tri.c         |  2 +-
 6 files changed, 90 insertions(+), 57 deletions(-)

(limited to 'src/gallium/drivers/cell/spu/spu_texture.c')

diff --git a/src/gallium/drivers/cell/common.h b/src/gallium/drivers/cell/common.h
index 298812fc20..f430e88b9c 100644
--- a/src/gallium/drivers/cell/common.h
+++ b/src/gallium/drivers/cell/common.h
@@ -36,6 +36,7 @@
 #include "pipe/p_compiler.h"
 #include "pipe/p_util.h"
 #include "pipe/p_format.h"
+#include "pipe/p_state.h"
 
 
 /** The standard assert macro doesn't seem to work reliably */
@@ -228,12 +229,20 @@ struct cell_command_release_verts
 };
 
 
+struct cell_command_sampler
+{
+   uint64_t opcode;         /**< CELL_CMD_STATE_SAMPLER */
+   uint unit;
+   struct pipe_sampler_state state;
+};
+
+
 struct cell_command_texture
 {
-   struct {
-      void *start;         /**< Address in main memory */
-      ushort width, height;
-   } texture[CELL_MAX_SAMPLERS];
+   uint64_t opcode;     /**< CELL_CMD_STATE_TEXTURE */
+   uint unit;
+   void *start;         /**< Address in main memory */
+   ushort width, height;
 };
 
 
diff --git a/src/gallium/drivers/cell/ppu/cell_state_emit.c b/src/gallium/drivers/cell/ppu/cell_state_emit.c
index 4fbe1a21b8..9cae67f091 100644
--- a/src/gallium/drivers/cell/ppu/cell_state_emit.c
+++ b/src/gallium/drivers/cell/ppu/cell_state_emit.c
@@ -121,25 +121,36 @@ cell_emit_state(struct cell_context *cell)
    }
 
    if (cell->dirty & CELL_NEW_SAMPLER) {
-      if (cell->sampler[0]) {
-         emit_state_cmd(cell, CELL_CMD_STATE_SAMPLER,
-                        cell->sampler[0], sizeof(struct pipe_sampler_state));
+      uint i;
+      for (i = 0; i < CELL_MAX_SAMPLERS; i++) {
+         if (cell->sampler[i]) {
+            struct cell_command_sampler *sampler
+               = cell_batch_alloc(cell, sizeof(*sampler));
+            sampler->opcode = CELL_CMD_STATE_SAMPLER;
+            sampler->unit = i;
+            sampler->state = *cell->sampler[i];
+         }
       }
    }
 
    if (cell->dirty & CELL_NEW_TEXTURE) {
-      struct cell_command_texture texture;
       uint i;
-      memset(&texture, 0, sizeof(texture));
       for (i = 0;i < CELL_MAX_SAMPLERS; i++) {
+         struct cell_command_texture *texture
+            =  cell_batch_alloc(cell, sizeof(*texture));
+         texture->opcode = CELL_CMD_STATE_TEXTURE;
+         texture->unit = i;
          if (cell->texture[i]) {
-            texture.texture[i].start = cell->texture[i]->tiled_data;
-            texture.texture[i].width = cell->texture[i]->base.width[0];
-            texture.texture[i].height = cell->texture[i]->base.height[0];
+            texture->start = cell->texture[i]->tiled_data;
+            texture->width = cell->texture[i]->base.width[0];
+            texture->height = cell->texture[i]->base.height[0];
+         }
+         else {
+            texture->start = NULL;
+            texture->width = 1;
+            texture->height = 1;
          }
       }
-      emit_state_cmd(cell, CELL_CMD_STATE_TEXTURE,
-                     &texture, sizeof(struct cell_command_texture));
    }
 
    if (cell->dirty & CELL_NEW_VERTEX_INFO) {
diff --git a/src/gallium/drivers/cell/spu/spu_main.c b/src/gallium/drivers/cell/spu/spu_main.c
index 80fa5f7859..7f0473d198 100644
--- a/src/gallium/drivers/cell/spu/spu_main.c
+++ b/src/gallium/drivers/cell/spu/spu_main.c
@@ -312,13 +312,13 @@ cmd_state_depth_stencil(const struct cell_command_depth_stencil_alpha_test *stat
 
 
 static void
-cmd_state_sampler(const struct pipe_sampler_state *state)
+cmd_state_sampler(const struct cell_command_sampler *sampler)
 {
    if (Debug)
-      printf("SPU %u: SAMPLER\n",
-             spu.init.id);
+      printf("SPU %u: SAMPLER [%u]\n",
+             spu.init.id, sampler->unit);
 
-   memcpy(&spu.sampler[0], state, sizeof(*state));
+   spu.sampler[sampler->unit] = sampler->state;
    if (spu.sampler[0].min_img_filter == PIPE_TEX_FILTER_LINEAR)
       spu.sample_texture = sample_texture_bilinear;
    else
@@ -329,26 +329,25 @@ cmd_state_sampler(const struct pipe_sampler_state *state)
 static void
 cmd_state_texture(const struct cell_command_texture *texture)
 {
-   uint i;
+   const uint unit = texture->unit;
+   const uint width = texture->width;
+   const uint height = texture->height;
 
-   if (1||Debug) {
-      printf("SPU %u: TEXTURE\n", spu.init.id);
-      for (i = 0; i < CELL_MAX_SAMPLERS; i++) {
-         printf("  %d: at %p  size %u x %u\n", i, texture->texture[i].start,
-                texture->texture[i].width, texture->texture[i].height);
-      }
+   if (Debug) {
+      printf("SPU %u: TEXTURE [%u] at %p  size %u x %u\n", spu.init.id,
+             texture->unit, texture->start,
+             texture->width, texture->height);
    }
 
-   memcpy(&spu.texture, texture, sizeof(*texture));
-   for (i = 0; i < CELL_MAX_SAMPLERS; i++) {
-      const uint width = texture->texture[i].width;
-      const uint height = texture->texture[i].height;
-      spu.tex_size[i] = (vector float) { width, height, 0.0, 0.0};
-      spu.tex_size_mask[i] = (vector unsigned int)
+   spu.texture[unit].start = texture->start;
+   spu.texture[unit].width = width;
+   spu.texture[unit].height = height;
+
+   spu.texture[unit].tex_size = (vector float) { width, height, 0.0, 0.0};
+   spu.texture[unit].tex_size_mask = (vector unsigned int)
          { width - 1, height - 1, 0, 0 };
-      spu.tex_size_x_mask[i] = spu_splats(width - 1);
-      spu.tex_size_y_mask[i] = spu_splats(height - 1);
-   }
+   spu.texture[unit].tex_size_x_mask = spu_splats(width - 1);
+   spu.texture[unit].tex_size_y_mask = spu_splats(height - 1);
 }
 
 
@@ -480,12 +479,20 @@ cmd_batch(uint opcode)
          pos += (1 + ROUNDUP8(sizeof(struct cell_command_depth_stencil_alpha_test)) / 8);
          break;
       case CELL_CMD_STATE_SAMPLER:
-         cmd_state_sampler((struct pipe_sampler_state *) &buffer[pos+1]);
-         pos += (1 + ROUNDUP8(sizeof(struct pipe_sampler_state)) / 8);
+         {
+            struct cell_command_sampler *sampler
+               = (struct cell_command_sampler *) &buffer[pos];
+            cmd_state_sampler(sampler);
+            pos += sizeof(*sampler) / 8;
+         }
          break;
       case CELL_CMD_STATE_TEXTURE:
-         cmd_state_texture((struct cell_command_texture *) &buffer[pos+1]);
-         pos += (1 + ROUNDUP8(sizeof(struct cell_command_texture)) / 8);
+         {
+            struct cell_command_texture *texture
+               = (struct cell_command_texture *) &buffer[pos];
+            cmd_state_texture(texture);
+            pos += sizeof(*texture) / 8;
+         }
          break;
       case CELL_CMD_STATE_VERTEX_INFO:
          cmd_state_vertex_info((struct vertex_info *) &buffer[pos+1]);
diff --git a/src/gallium/drivers/cell/spu/spu_main.h b/src/gallium/drivers/cell/spu/spu_main.h
index 8a87787537..2bfad3535a 100644
--- a/src/gallium/drivers/cell/spu/spu_main.h
+++ b/src/gallium/drivers/cell/spu/spu_main.h
@@ -100,6 +100,17 @@ struct spu_framebuffer {
 } ALIGN16_ATTRIB;
 
 
+struct spu_texture
+{
+   void *start;
+   uint width, height;
+   vector float tex_size;
+   vector unsigned int tex_size_mask; /**< == int(size - 1) */
+   vector unsigned int tex_size_x_mask; /**< == int(size - 1) */
+   vector unsigned int tex_size_y_mask; /**< == int(size - 1) */
+} ALIGN16_ATTRIB;
+
+
 /**
  * All SPU global/context state will be in singleton object of this type:
  */
@@ -119,7 +130,7 @@ struct spu_global
    logicop_func logicop;
 
    struct pipe_sampler_state sampler[PIPE_MAX_SAMPLERS];
-   struct cell_command_texture texture;
+   struct spu_texture texture[PIPE_MAX_SAMPLERS];
 
    struct vertex_info vertex_info;
 
@@ -141,11 +152,6 @@ struct spu_global
    /** for converting RGBA to PIPE_FORMAT_x colors */
    vector unsigned char color_shuffle;
 
-   vector float tex_size[CELL_MAX_SAMPLERS];
-   vector unsigned int tex_size_mask[CELL_MAX_SAMPLERS]; /**< == int(size - 1) */
-   vector unsigned int tex_size_x_mask[CELL_MAX_SAMPLERS]; /**< == int(size - 1) */
-   vector unsigned int tex_size_y_mask[CELL_MAX_SAMPLERS]; /**< == int(size - 1) */
-
    vector float (*sample_texture)(vector float texcoord);
 
 } ALIGN16_ATTRIB;
diff --git a/src/gallium/drivers/cell/spu/spu_texture.c b/src/gallium/drivers/cell/spu/spu_texture.c
index 91a6aec5ec..4612501eb3 100644
--- a/src/gallium/drivers/cell/spu/spu_texture.c
+++ b/src/gallium/drivers/cell/spu/spu_texture.c
@@ -41,10 +41,10 @@ void
 invalidate_tex_cache(void)
 {
    uint unit = 0;
-   uint bytes = 4 * spu.texture.texture[unit].width
-      * spu.texture.texture[unit].height;
+   uint bytes = 4 * spu.texture[unit].width
+      * spu.texture[unit].height;
 
-   spu_dcache_mark_dirty((unsigned) spu.texture.texture[unit].start, bytes);
+   spu_dcache_mark_dirty((unsigned) spu.texture[unit].start, bytes);
 }
 
 
@@ -55,14 +55,14 @@ get_texel(vec_uint4 coordinate)
    vec_uint4 tmp;
    unsigned x = spu_extract(coordinate, 0);
    unsigned y = spu_extract(coordinate, 1);
-   const unsigned tiles_per_row = spu.texture.texture[unit].width / TILE_SIZE;
+   const unsigned tiles_per_row = spu.texture[unit].width / TILE_SIZE;
    unsigned tile_offset = sizeof(tile_t) * ((y / TILE_SIZE * tiles_per_row) 
                                             + (x / TILE_SIZE));
    unsigned texel_offset = 4 * (((y % TILE_SIZE) * TILE_SIZE)
                                 + (x % TILE_SIZE));
 
    spu_dcache_fetch_unaligned((qword *) & tmp,
-                              spu.texture.texture[unit].start + tile_offset + texel_offset,
+                              spu.texture[unit].start + tile_offset + texel_offset,
                               4);
    return spu_extract(tmp, 0);
 }
@@ -72,13 +72,13 @@ static void
 get_four_texels(vec_uint4 x, vec_uint4 y, vec_uint4 *texels)
 {
    const uint unit = 0;
-   const unsigned texture_ea = (uintptr_t) spu.texture.texture[unit].start;
+   const unsigned texture_ea = (uintptr_t) spu.texture[unit].start;
    vec_uint4 tile_x = spu_rlmask(x, -5);
    vec_uint4 tile_y = spu_rlmask(y, -5);
    const qword offset_x = si_andi((qword) x, 0x1f);
    const qword offset_y = si_andi((qword) y, 0x1f);
 
-   const qword tiles_per_row = (qword) spu_splats(spu.texture.texture[unit].width / TILE_SIZE);
+   const qword tiles_per_row = (qword) spu_splats(spu.texture[unit].width / TILE_SIZE);
    const qword tile_size = (qword) spu_splats(sizeof(tile_t));
 
    qword tile_offset = si_mpya((qword) tile_y, tiles_per_row, (qword) tile_x);
@@ -107,9 +107,9 @@ vector float
 sample_texture_nearest(vector float texcoord)
 {
    const uint unit = 0;
-   vector float tc = spu_mul(texcoord, spu.tex_size[unit]);
+   vector float tc = spu_mul(texcoord, spu.texture[unit].tex_size);
    vector unsigned int itc = spu_convtu(tc, 0);  /* convert to int */
-   itc = spu_and(itc, spu.tex_size_mask[unit]);        /* mask (GL_REPEAT) */
+   itc = spu_and(itc, spu.texture[unit].tex_size_mask); /* mask (GL_REPEAT) */
    uint texel = get_texel(itc);
    return spu_unpack_A8R8G8B8(texel);
 }
@@ -122,7 +122,7 @@ sample_texture_bilinear(vector float texcoord)
    static const vec_uint4 offset_x = {0, 0, 1, 1};
    static const vec_uint4 offset_y = {0, 1, 0, 1};
 
-   vector float tc = spu_mul(texcoord, spu.tex_size[unit]);
+   vector float tc = spu_mul(texcoord, spu.texture[unit].tex_size);
    tc = spu_add(tc, spu_splats(-0.5f));  /* half texel bias */
 
    /* integer texcoords S,T: */
@@ -136,8 +136,8 @@ sample_texture_bilinear(vector float texcoord)
    x = spu_add(x, offset_x);
    y = spu_add(y, offset_y);
 
-   x = spu_and(x, spu.tex_size_x_mask[unit]);
-   y = spu_and(y, spu.tex_size_y_mask[unit]);
+   x = spu_and(x, spu.texture[unit].tex_size_x_mask);
+   y = spu_and(y, spu.texture[unit].tex_size_y_mask);
 
    get_four_texels(x, y, texels);
 
diff --git a/src/gallium/drivers/cell/spu/spu_tri.c b/src/gallium/drivers/cell/spu/spu_tri.c
index 9f63317b1f..17e337bbdf 100644
--- a/src/gallium/drivers/cell/spu/spu_tri.c
+++ b/src/gallium/drivers/cell/spu/spu_tri.c
@@ -309,7 +309,7 @@ emit_quad( int x, int y, mask_t mask )
 
       spu.cur_ctile_status = TILE_STATUS_DIRTY;
 
-      if (spu.texture.texture[0].start) {
+      if (spu.texture[0].start) {
          /* texture mapping */
          vector float texcoords[4];
          eval_coeff(2, (float) x, (float) y, texcoords);
-- 
cgit v1.2.3


From e7b23d36df1ab3ac5b54ef8e4e56c4fd46db8257 Mon Sep 17 00:00:00 2001
From: Brian <brian.paul@tungstengraphics.com>
Date: Tue, 1 Apr 2008 11:35:53 -0600
Subject: cell: checkpoint: more multi-texture work

---
 src/gallium/drivers/cell/ppu/cell_texture.c |  8 +++++--
 src/gallium/drivers/cell/spu/spu_main.c     |  2 +-
 src/gallium/drivers/cell/spu/spu_main.h     |  2 +-
 src/gallium/drivers/cell/spu/spu_texture.c  |  6 ++---
 src/gallium/drivers/cell/spu/spu_texture.h  |  4 ++--
 src/gallium/drivers/cell/spu/spu_tri.c      | 34 +++++++++++++++++++++++++----
 6 files changed, 42 insertions(+), 14 deletions(-)

(limited to 'src/gallium/drivers/cell/spu/spu_texture.c')

diff --git a/src/gallium/drivers/cell/ppu/cell_texture.c b/src/gallium/drivers/cell/ppu/cell_texture.c
index 9c694e136d..c59d1f7f23 100644
--- a/src/gallium/drivers/cell/ppu/cell_texture.c
+++ b/src/gallium/drivers/cell/ppu/cell_texture.c
@@ -245,9 +245,13 @@ void
 cell_update_texture_mapping(struct cell_context *cell)
 {
    uint face = 0, level = 0, zslice = 0;
+   uint i;
+
+   for (i = 0; i < CELL_MAX_SAMPLERS; i++) {
+      if (cell->texture[i])
+         cell_tile_texture(cell, cell->texture[i]);
+   }
 
-   if (cell->texture[0])
-      cell_tile_texture(cell, cell->texture[0]);
 #if 0
    if (cell->tex_surf && cell->tex_map) {
       pipe_surface_unmap(cell->tex_surf);
diff --git a/src/gallium/drivers/cell/spu/spu_main.c b/src/gallium/drivers/cell/spu/spu_main.c
index 5b5a570a3c..a840d01596 100644
--- a/src/gallium/drivers/cell/spu/spu_main.c
+++ b/src/gallium/drivers/cell/spu/spu_main.c
@@ -333,7 +333,7 @@ cmd_state_texture(const struct cell_command_texture *texture)
    const uint width = texture->width;
    const uint height = texture->height;
 
-   if (Debug) {
+   if (1||Debug) {
       printf("SPU %u: TEXTURE [%u] at %p  size %u x %u\n", spu.init.id,
              texture->unit, texture->start,
              texture->width, texture->height);
diff --git a/src/gallium/drivers/cell/spu/spu_main.h b/src/gallium/drivers/cell/spu/spu_main.h
index 2bfad3535a..26e050cfc3 100644
--- a/src/gallium/drivers/cell/spu/spu_main.h
+++ b/src/gallium/drivers/cell/spu/spu_main.h
@@ -152,7 +152,7 @@ struct spu_global
    /** for converting RGBA to PIPE_FORMAT_x colors */
    vector unsigned char color_shuffle;
 
-   vector float (*sample_texture)(vector float texcoord);
+   vector float (*sample_texture)(uint unit, vector float texcoord);
 
 } ALIGN16_ATTRIB;
 
diff --git a/src/gallium/drivers/cell/spu/spu_texture.c b/src/gallium/drivers/cell/spu/spu_texture.c
index 4612501eb3..58a426cc40 100644
--- a/src/gallium/drivers/cell/spu/spu_texture.c
+++ b/src/gallium/drivers/cell/spu/spu_texture.c
@@ -104,9 +104,8 @@ get_four_texels(vec_uint4 x, vec_uint4 y, vec_uint4 *texels)
  * XXX this is extremely primitive for now.
  */
 vector float
-sample_texture_nearest(vector float texcoord)
+sample_texture_nearest(uint unit, vector float texcoord)
 {
-   const uint unit = 0;
    vector float tc = spu_mul(texcoord, spu.texture[unit].tex_size);
    vector unsigned int itc = spu_convtu(tc, 0);  /* convert to int */
    itc = spu_and(itc, spu.texture[unit].tex_size_mask); /* mask (GL_REPEAT) */
@@ -116,9 +115,8 @@ sample_texture_nearest(vector float texcoord)
 
 
 vector float
-sample_texture_bilinear(vector float texcoord)
+sample_texture_bilinear(uint unit, vector float texcoord)
 {
-   const uint unit = 0;
    static const vec_uint4 offset_x = {0, 0, 1, 1};
    static const vec_uint4 offset_y = {0, 1, 0, 1};
 
diff --git a/src/gallium/drivers/cell/spu/spu_texture.h b/src/gallium/drivers/cell/spu/spu_texture.h
index 95eb87080f..f7c9738be8 100644
--- a/src/gallium/drivers/cell/spu/spu_texture.h
+++ b/src/gallium/drivers/cell/spu/spu_texture.h
@@ -37,11 +37,11 @@ invalidate_tex_cache(void);
 
 
 extern vector float
-sample_texture_nearest(vector float texcoord);
+sample_texture_nearest(uint unit, vector float texcoord);
 
 
 extern vector float
-sample_texture_bilinear(vector float texcoord);
+sample_texture_bilinear(uint unit, vector float texcoord);
 
 
 #endif /* SPU_TEXTURE_H */
diff --git a/src/gallium/drivers/cell/spu/spu_tri.c b/src/gallium/drivers/cell/spu/spu_tri.c
index 17e337bbdf..51abcb1757 100644
--- a/src/gallium/drivers/cell/spu/spu_tri.c
+++ b/src/gallium/drivers/cell/spu/spu_tri.c
@@ -311,17 +311,43 @@ emit_quad( int x, int y, mask_t mask )
 
       if (spu.texture[0].start) {
          /* texture mapping */
+         const uint unit = 0;
          vector float texcoords[4];
          eval_coeff(2, (float) x, (float) y, texcoords);
 
          if (spu_extract(mask, 0))
-            colors[0] = spu.sample_texture(texcoords[0]);
+            colors[0] = spu.sample_texture(unit, texcoords[0]);
          if (spu_extract(mask, 1))
-            colors[1] = spu.sample_texture(texcoords[1]);
+            colors[1] = spu.sample_texture(unit, texcoords[1]);
          if (spu_extract(mask, 2))
-            colors[2] = spu.sample_texture(texcoords[2]);
+            colors[2] = spu.sample_texture(unit, texcoords[2]);
          if (spu_extract(mask, 3))
-            colors[3] = spu.sample_texture(texcoords[3]);
+            colors[3] = spu.sample_texture(unit, texcoords[3]);
+
+
+         if (spu.texture[1].start) {
+            /* multi-texture mapping */
+            const uint unit = 1;
+            vector float colors1[4];
+
+            eval_coeff(3, (float) x, (float) y, texcoords);
+
+            if (spu_extract(mask, 0))
+               colors1[0] = spu.sample_texture(unit, texcoords[0]);
+            if (spu_extract(mask, 1))
+               colors1[1] = spu.sample_texture(unit, texcoords[1]);
+            if (spu_extract(mask, 2))
+               colors1[2] = spu.sample_texture(unit, texcoords[2]);
+            if (spu_extract(mask, 3))
+               colors1[3] = spu.sample_texture(unit, texcoords[3]);
+
+            /* hack: modulate first texture by second */
+            colors[0] = spu_mul(colors[0], colors1[0]);
+            colors[1] = spu_mul(colors[1], colors1[1]);
+            colors[2] = spu_mul(colors[2], colors1[2]);
+            colors[3] = spu_mul(colors[3], colors1[3]);
+         }
+
       }
       else {
          /* simple shading */
-- 
cgit v1.2.3


From 9e7b730eb25f08065d718dee4a67c67d1d133b6e Mon Sep 17 00:00:00 2001
From: Brian <brian.paul@tungstengraphics.com>
Date: Tue, 1 Apr 2008 14:52:25 -0600
Subject: cell: pass tex unit to get_texel()

---
 src/gallium/drivers/cell/spu/spu_texture.c | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

(limited to 'src/gallium/drivers/cell/spu/spu_texture.c')

diff --git a/src/gallium/drivers/cell/spu/spu_texture.c b/src/gallium/drivers/cell/spu/spu_texture.c
index 58a426cc40..ba0d57b959 100644
--- a/src/gallium/drivers/cell/spu/spu_texture.c
+++ b/src/gallium/drivers/cell/spu/spu_texture.c
@@ -49,9 +49,8 @@ invalidate_tex_cache(void)
 
 
 static uint
-get_texel(vec_uint4 coordinate)
+get_texel(uint unit, vec_uint4 coordinate)
 {
-   const uint unit = 0;
    vec_uint4 tmp;
    unsigned x = spu_extract(coordinate, 0);
    unsigned y = spu_extract(coordinate, 1);
@@ -109,7 +108,7 @@ sample_texture_nearest(uint unit, vector float texcoord)
    vector float tc = spu_mul(texcoord, spu.texture[unit].tex_size);
    vector unsigned int itc = spu_convtu(tc, 0);  /* convert to int */
    itc = spu_and(itc, spu.texture[unit].tex_size_mask); /* mask (GL_REPEAT) */
-   uint texel = get_texel(itc);
+   uint texel = get_texel(unit, itc);
    return spu_unpack_A8R8G8B8(texel);
 }
 
-- 
cgit v1.2.3


From bccd3f138ccc717fad9073110d15e3321b590476 Mon Sep 17 00:00:00 2001
From: Brian <brian.paul@tungstengraphics.com>
Date: Tue, 1 Apr 2008 15:42:42 -0600
Subject: cell: more multi-texture fixes (mostly working now)

---
 src/gallium/drivers/cell/spu/spu_main.c    |  6 +++---
 src/gallium/drivers/cell/spu/spu_main.h    |  4 +++-
 src/gallium/drivers/cell/spu/spu_texture.c |  5 ++---
 src/gallium/drivers/cell/spu/spu_tri.c     | 18 +++++++++---------
 4 files changed, 17 insertions(+), 16 deletions(-)

(limited to 'src/gallium/drivers/cell/spu/spu_texture.c')

diff --git a/src/gallium/drivers/cell/spu/spu_main.c b/src/gallium/drivers/cell/spu/spu_main.c
index 5b5a570a3c..1ab1c40379 100644
--- a/src/gallium/drivers/cell/spu/spu_main.c
+++ b/src/gallium/drivers/cell/spu/spu_main.c
@@ -319,10 +319,10 @@ cmd_state_sampler(const struct cell_command_sampler *sampler)
              spu.init.id, sampler->unit);
 
    spu.sampler[sampler->unit] = sampler->state;
-   if (spu.sampler[0].min_img_filter == PIPE_TEX_FILTER_LINEAR)
-      spu.sample_texture = sample_texture_bilinear;
+   if (spu.sampler[sampler->unit].min_img_filter == PIPE_TEX_FILTER_LINEAR)
+      spu.sample_texture[sampler->unit] = sample_texture_bilinear;
    else
-      spu.sample_texture = sample_texture_nearest;
+      spu.sample_texture[sampler->unit] = sample_texture_nearest;
 }
 
 
diff --git a/src/gallium/drivers/cell/spu/spu_main.h b/src/gallium/drivers/cell/spu/spu_main.h
index 26e050cfc3..e9e39cbeab 100644
--- a/src/gallium/drivers/cell/spu/spu_main.h
+++ b/src/gallium/drivers/cell/spu/spu_main.h
@@ -85,6 +85,8 @@ typedef struct spu_blend_results (*logicop_func)(
     qword frag_mask);
 
 
+typedef vector float (*sample_texture_func)(uint unit, vector float texcoord);
+
 struct spu_framebuffer {
    void *color_start;              /**< addr of color surface in main memory */
    void *depth_start;              /**< addr of depth surface in main memory */
@@ -152,7 +154,7 @@ struct spu_global
    /** for converting RGBA to PIPE_FORMAT_x colors */
    vector unsigned char color_shuffle;
 
-   vector float (*sample_texture)(uint unit, vector float texcoord);
+   sample_texture_func sample_texture[CELL_MAX_SAMPLERS];
 
 } ALIGN16_ATTRIB;
 
diff --git a/src/gallium/drivers/cell/spu/spu_texture.c b/src/gallium/drivers/cell/spu/spu_texture.c
index ba0d57b959..ceab246980 100644
--- a/src/gallium/drivers/cell/spu/spu_texture.c
+++ b/src/gallium/drivers/cell/spu/spu_texture.c
@@ -68,9 +68,8 @@ get_texel(uint unit, vec_uint4 coordinate)
 
 
 static void
-get_four_texels(vec_uint4 x, vec_uint4 y, vec_uint4 *texels)
+get_four_texels(uint unit, vec_uint4 x, vec_uint4 y, vec_uint4 *texels)
 {
-   const uint unit = 0;
    const unsigned texture_ea = (uintptr_t) spu.texture[unit].start;
    vec_uint4 tile_x = spu_rlmask(x, -5);
    vec_uint4 tile_y = spu_rlmask(y, -5);
@@ -136,7 +135,7 @@ sample_texture_bilinear(uint unit, vector float texcoord)
    x = spu_and(x, spu.texture[unit].tex_size_x_mask);
    y = spu_and(y, spu.texture[unit].tex_size_y_mask);
 
-   get_four_texels(x, y, texels);
+   get_four_texels(unit, x, y, texels);
 
    vector float texel00 = spu_unpack_A8R8G8B8(spu_extract(texels[0], 0));
    vector float texel01 = spu_unpack_A8R8G8B8(spu_extract(texels[1], 0));
diff --git a/src/gallium/drivers/cell/spu/spu_tri.c b/src/gallium/drivers/cell/spu/spu_tri.c
index 51abcb1757..ab4ff8160a 100644
--- a/src/gallium/drivers/cell/spu/spu_tri.c
+++ b/src/gallium/drivers/cell/spu/spu_tri.c
@@ -316,13 +316,13 @@ emit_quad( int x, int y, mask_t mask )
          eval_coeff(2, (float) x, (float) y, texcoords);
 
          if (spu_extract(mask, 0))
-            colors[0] = spu.sample_texture(unit, texcoords[0]);
+            colors[0] = spu.sample_texture[unit](unit, texcoords[0]);
          if (spu_extract(mask, 1))
-            colors[1] = spu.sample_texture(unit, texcoords[1]);
+            colors[1] = spu.sample_texture[unit](unit, texcoords[1]);
          if (spu_extract(mask, 2))
-            colors[2] = spu.sample_texture(unit, texcoords[2]);
+            colors[2] = spu.sample_texture[unit](unit, texcoords[2]);
          if (spu_extract(mask, 3))
-            colors[3] = spu.sample_texture(unit, texcoords[3]);
+            colors[3] = spu.sample_texture[unit](unit, texcoords[3]);
 
 
          if (spu.texture[1].start) {
@@ -330,16 +330,16 @@ emit_quad( int x, int y, mask_t mask )
             const uint unit = 1;
             vector float colors1[4];
 
-            eval_coeff(3, (float) x, (float) y, texcoords);
+            eval_coeff(2, (float) x, (float) y, texcoords);
 
             if (spu_extract(mask, 0))
-               colors1[0] = spu.sample_texture(unit, texcoords[0]);
+               colors1[0] = spu.sample_texture[unit](unit, texcoords[0]);
             if (spu_extract(mask, 1))
-               colors1[1] = spu.sample_texture(unit, texcoords[1]);
+               colors1[1] = spu.sample_texture[unit](unit, texcoords[1]);
             if (spu_extract(mask, 2))
-               colors1[2] = spu.sample_texture(unit, texcoords[2]);
+               colors1[2] = spu.sample_texture[unit](unit, texcoords[2]);
             if (spu_extract(mask, 3))
-               colors1[3] = spu.sample_texture(unit, texcoords[3]);
+               colors1[3] = spu.sample_texture[unit](unit, texcoords[3]);
 
             /* hack: modulate first texture by second */
             colors[0] = spu_mul(colors[0], colors1[0]);
-- 
cgit v1.2.3


From 217d37940771dd02ff1aa365105eca2c7a09d623 Mon Sep 17 00:00:00 2001
From: Brian <brian.paul@tungstengraphics.com>
Date: Wed, 2 Apr 2008 14:01:42 -0600
Subject: cell: minor texture improvements

Precompute tiles_per_row.  Use ushort multiplies in a few places.  New comments.
---
 src/gallium/drivers/cell/spu/spu_main.c    |  2 ++
 src/gallium/drivers/cell/spu/spu_main.h    |  3 ++-
 src/gallium/drivers/cell/spu/spu_texture.c | 32 ++++++++++++++++++++----------
 3 files changed, 26 insertions(+), 11 deletions(-)

(limited to 'src/gallium/drivers/cell/spu/spu_texture.c')

diff --git a/src/gallium/drivers/cell/spu/spu_main.c b/src/gallium/drivers/cell/spu/spu_main.c
index 1ab1c40379..e04ffeb9b1 100644
--- a/src/gallium/drivers/cell/spu/spu_main.c
+++ b/src/gallium/drivers/cell/spu/spu_main.c
@@ -343,6 +343,8 @@ cmd_state_texture(const struct cell_command_texture *texture)
    spu.texture[unit].width = width;
    spu.texture[unit].height = height;
 
+   spu.texture[unit].tiles_per_row = width / TILE_SIZE;
+
    spu.texture[unit].tex_size = (vector float) { width, height, 0.0, 0.0};
    spu.texture[unit].tex_size_mask = (vector unsigned int)
          { width - 1, height - 1, 0, 0 };
diff --git a/src/gallium/drivers/cell/spu/spu_main.h b/src/gallium/drivers/cell/spu/spu_main.h
index e9e39cbeab..e962e1426c 100644
--- a/src/gallium/drivers/cell/spu/spu_main.h
+++ b/src/gallium/drivers/cell/spu/spu_main.h
@@ -105,7 +105,8 @@ struct spu_framebuffer {
 struct spu_texture
 {
    void *start;
-   uint width, height;
+   ushort width, height;
+   ushort tiles_per_row;
    vector float tex_size;
    vector unsigned int tex_size_mask; /**< == int(size - 1) */
    vector unsigned int tex_size_x_mask; /**< == int(size - 1) */
diff --git a/src/gallium/drivers/cell/spu/spu_texture.c b/src/gallium/drivers/cell/spu/spu_texture.c
index ceab246980..e9a2754e57 100644
--- a/src/gallium/drivers/cell/spu/spu_texture.c
+++ b/src/gallium/drivers/cell/spu/spu_texture.c
@@ -51,22 +51,25 @@ invalidate_tex_cache(void)
 static uint
 get_texel(uint unit, vec_uint4 coordinate)
 {
+   const unsigned texture_ea = (uintptr_t) spu.texture[unit].start;
+   ushort x = spu_extract(coordinate, 0);
+   ushort y = spu_extract(coordinate, 1);
+   unsigned tile_offset = sizeof(tile_t)
+      * ((y / TILE_SIZE * spu.texture[unit].tiles_per_row) + (x / TILE_SIZE));
+   ushort texel_offset = (ushort) 4
+      * (ushort) (((ushort) (y % TILE_SIZE) * (ushort) TILE_SIZE) + (x % TILE_SIZE));
    vec_uint4 tmp;
-   unsigned x = spu_extract(coordinate, 0);
-   unsigned y = spu_extract(coordinate, 1);
-   const unsigned tiles_per_row = spu.texture[unit].width / TILE_SIZE;
-   unsigned tile_offset = sizeof(tile_t) * ((y / TILE_SIZE * tiles_per_row) 
-                                            + (x / TILE_SIZE));
-   unsigned texel_offset = 4 * (((y % TILE_SIZE) * TILE_SIZE)
-                                + (x % TILE_SIZE));
 
    spu_dcache_fetch_unaligned((qword *) & tmp,
-                              spu.texture[unit].start + tile_offset + texel_offset,
+                              texture_ea + tile_offset + texel_offset,
                               4);
    return spu_extract(tmp, 0);
 }
 
 
+/**
+ * Get four texels from locations (x[0], y[0]), (x[1], y[1]) ...
+ */
 static void
 get_four_texels(uint unit, vec_uint4 x, vec_uint4 y, vec_uint4 *texels)
 {
@@ -76,7 +79,7 @@ get_four_texels(uint unit, vec_uint4 x, vec_uint4 y, vec_uint4 *texels)
    const qword offset_x = si_andi((qword) x, 0x1f);
    const qword offset_y = si_andi((qword) y, 0x1f);
 
-   const qword tiles_per_row = (qword) spu_splats(spu.texture[unit].width / TILE_SIZE);
+   const qword tiles_per_row = (qword) spu_splats(spu.texture[unit].tiles_per_row);
    const qword tile_size = (qword) spu_splats(sizeof(tile_t));
 
    qword tile_offset = si_mpya((qword) tile_y, tiles_per_row, (qword) tile_x);
@@ -97,6 +100,7 @@ get_four_texels(uint unit, vec_uint4 x, vec_uint4 y, vec_uint4 *texels)
                               texture_ea + spu_extract(offset, 3), 4);
 }
 
+
 /**
  * Get texture sample at texcoord.
  * XXX this is extremely primitive for now.
@@ -126,17 +130,25 @@ sample_texture_bilinear(uint unit, vector float texcoord)
 
    vec_uint4 texels[4];
    
+   /* setup texcoords for quad:
+    *  +-----+-----+
+    *  |x0,y0|x1,y1|
+    *  +-----+-----+
+    *  |x2,y2|x3,y3|
+    *  +-----+-----+
+    */
    vec_uint4 x = spu_splats(spu_extract(itc, 0));
    vec_uint4 y = spu_splats(spu_extract(itc, 1));
-
    x = spu_add(x, offset_x);
    y = spu_add(y, offset_y);
 
+   /* GL_REPEAT wrap mode: */
    x = spu_and(x, spu.texture[unit].tex_size_x_mask);
    y = spu_and(y, spu.texture[unit].tex_size_y_mask);
 
    get_four_texels(unit, x, y, texels);
 
+   /* integer A8R8G8B8 to float texel conversion */
    vector float texel00 = spu_unpack_A8R8G8B8(spu_extract(texels[0], 0));
    vector float texel01 = spu_unpack_A8R8G8B8(spu_extract(texels[1], 0));
    vector float texel10 = spu_unpack_A8R8G8B8(spu_extract(texels[2], 0));
-- 
cgit v1.2.3


From a7504ad587ee8cbfa9958ad23321a691ce0823d3 Mon Sep 17 00:00:00 2001
From: Brian <brian.paul@tungstengraphics.com>
Date: Wed, 2 Apr 2008 14:30:28 -0600
Subject: cell: added some comments/ideas about better texture sampling

---
 src/gallium/drivers/cell/spu/spu_texture.c | 18 +++++++++++++++++-
 1 file changed, 17 insertions(+), 1 deletion(-)

(limited to 'src/gallium/drivers/cell/spu/spu_texture.c')

diff --git a/src/gallium/drivers/cell/spu/spu_texture.c b/src/gallium/drivers/cell/spu/spu_texture.c
index e9a2754e57..5051774f00 100644
--- a/src/gallium/drivers/cell/spu/spu_texture.c
+++ b/src/gallium/drivers/cell/spu/spu_texture.c
@@ -48,9 +48,16 @@ invalidate_tex_cache(void)
 }
 
 
+/**
+ * XXX look into getting texels for all four pixels in a quad at once.
+ */
 static uint
 get_texel(uint unit, vec_uint4 coordinate)
 {
+   /*
+    * XXX we could do the "/ TILE_SIZE" and "% TILE_SIZE" operations as
+    * SIMD since X and Y are already in a SIMD register.
+    */
    const unsigned texture_ea = (uintptr_t) spu.texture[unit].start;
    ushort x = spu_extract(coordinate, 0);
    ushort y = spu_extract(coordinate, 1);
@@ -69,6 +76,16 @@ get_texel(uint unit, vec_uint4 coordinate)
 
 /**
  * Get four texels from locations (x[0], y[0]), (x[1], y[1]) ...
+ *
+ * NOTE: in the typical case of bilinear filtering, the four texels
+ * are in a 2x2 group so we could get by with just two dcache fetches
+ * (two side-by-side texels per fetch).  But when bilinear filtering
+ * wraps around a texture edge, we'll probably need code like we have
+ * now.
+ * FURTHERMORE: since we're rasterizing a quad of 2x2 pixels at a time,
+ * it's quite likely that the four pixels in a quad will need some of the
+ * same texels.  So look into doing texture fetches for four pixels at
+ * a time.
  */
 static void
 get_four_texels(uint unit, vec_uint4 x, vec_uint4 y, vec_uint4 *texels)
@@ -103,7 +120,6 @@ get_four_texels(uint unit, vec_uint4 x, vec_uint4 y, vec_uint4 *texels)
 
 /**
  * Get texture sample at texcoord.
- * XXX this is extremely primitive for now.
  */
 vector float
 sample_texture_nearest(uint unit, vector float texcoord)
-- 
cgit v1.2.3