cell: support for cubemaps

Though, progs/demos/cubemap.c doesn't quite work right...

8 files changed, 214 insertions, 41 deletions

diff --git a/src/gallium/drivers/cell/common.h b/src/gallium/drivers/cell/common.h
index e4de9a551..c1e78f4db 100644
--- a/src/gallium/drivers/cell/common.h
+++ b/src/gallium/drivers/cell/common.h
@@ -251,6 +251,7 @@ struct cell_command_sampler
 struct cell_command_texture
 {
    uint64_t opcode;     /**< CELL_CMD_STATE_TEXTURE */
+   uint target;         /**< PIPE_TEXTURE_x */
    uint unit;
    void *start[CELL_MAX_TEXTURE_LEVELS];   /**< Address in main memory */
    ushort width[CELL_MAX_TEXTURE_LEVELS];
diff --git a/src/gallium/drivers/cell/ppu/cell_state_emit.c b/src/gallium/drivers/cell/ppu/cell_state_emit.c
index cae546b70..d4a867ffc 100644
--- a/src/gallium/drivers/cell/ppu/cell_state_emit.c
+++ b/src/gallium/drivers/cell/ppu/cell_state_emit.c
@@ -217,6 +217,7 @@ cell_emit_state(struct cell_context *cell)
                texture->width[level] = cell->texture[i]->base.width[level];
                texture->height[level] = cell->texture[i]->base.height[level];
             }
+            texture->target = cell->texture[i]->base.target;
          }
          else {
             uint level;
@@ -225,6 +226,7 @@ cell_emit_state(struct cell_context *cell)
                texture->width[level] = 0;
                texture->height[level] = 0;
             }
+            texture->target = 0;
          }
       }
    }
diff --git a/src/gallium/drivers/cell/ppu/cell_texture.c b/src/gallium/drivers/cell/ppu/cell_texture.c
index 4fd66bdea..4c92ef154 100644
--- a/src/gallium/drivers/cell/ppu/cell_texture.c
+++ b/src/gallium/drivers/cell/ppu/cell_texture.c
@@ -137,6 +137,7 @@ cell_texture_release(struct pipe_screen *screen,
    */
    if (--(*pt)->refcount <= 0) {
       struct cell_texture *ct = cell_texture(*pt);
+      uint i;
 
       /*
       DBG("%s deleting %p\n", __FUNCTION__, (void *) ct);
@@ -144,6 +145,12 @@ cell_texture_release(struct pipe_screen *screen,
 
       pipe_buffer_reference(screen, &ct->buffer, NULL);
 
+      for (i = 0; i < CELL_MAX_TEXTURE_LEVELS; i++) {
+         if (ct->tiled_data[i]) {
+            FREE(ct->tiled_data[i]);
+         }
+      }
+
       FREE(ct);
    }
    *pt = NULL;
@@ -204,27 +211,33 @@ static void
 cell_twiddle_texture(struct pipe_screen *screen,
                      struct pipe_surface *surface)
 {
-   struct cell_texture *texture = cell_texture(surface->texture);
+   struct cell_texture *ct = cell_texture(surface->texture);
    const uint level = surface->level;
-   const uint texWidth = texture->base.width[level];
-   const uint texHeight = texture->base.height[level];
+   const uint texWidth = ct->base.width[level];
+   const uint texHeight = ct->base.height[level];
    const uint bufWidth = align(texWidth, TILE_SIZE);
    const uint bufHeight = align(texHeight, TILE_SIZE);
    const void *map = pipe_buffer_map(screen, surface->buffer,
                                      PIPE_BUFFER_USAGE_CPU_READ);
    const uint *src = (const uint *) ((const ubyte *) map + surface->offset);
 
-   switch (texture->base.format) {
+   switch (ct->base.format) {
    case PIPE_FORMAT_A8R8G8B8_UNORM:
-      /* free old tiled data */
-      if (texture->tiled_data[level]) {
-         align_free(texture->tiled_data[level]);
+      {
+         int numFaces = ct->base.target == PIPE_TEXTURE_CUBE ? 6 : 1;
+         int offset = bufWidth * bufHeight * 4 * surface->face;
+         uint *dst;
+
+         if (!ct->tiled_data[level]) {
+            ct->tiled_data[level] =
+               align_malloc(bufWidth * bufHeight * 4 * numFaces, 16);
+         }
+
+         dst = (uint *) ((ubyte *) ct->tiled_data[level] + offset);
+
+         twiddle_image_uint(texWidth, texHeight, TILE_SIZE, dst,
+                            surface->stride, src);
       }
-      /* alloc new tiled data */
-      texture->tiled_data[level] = align_malloc(bufWidth * bufHeight * 4, 16);
-      twiddle_image_uint(texWidth, texHeight, TILE_SIZE,
-                         texture->tiled_data[level],
-                         surface->stride, src);
       break;
    default:
       printf("Cell: twiddle unsupported texture format\n");
diff --git a/src/gallium/drivers/cell/spu/spu_command.c b/src/gallium/drivers/cell/spu/spu_command.c
index b1efe97e7..c951fa6f3 100644
--- a/src/gallium/drivers/cell/spu/spu_command.c
+++ b/src/gallium/drivers/cell/spu/spu_command.c
@@ -301,7 +301,8 @@ cmd_state_framebuffer(const struct cell_command_framebuffer *cmd)
  */
 static void
 update_tex_masks(struct spu_texture *texture,
-                 const struct pipe_sampler_state *sampler)
+                 const struct pipe_sampler_state *sampler,
+                 uint unit)
 {
    uint i;
 
@@ -328,6 +329,11 @@ update_tex_masks(struct spu_texture *texture,
          texture->level[i].scale_t = spu_splats(1.0f);
       }
    }
+
+   /* XXX temporary hack */
+   if (texture->target == PIPE_TEXTURE_CUBE) {
+      spu.sample_texture4[unit] = sample_texture4_cube;
+   }
 }
 
 
@@ -378,7 +384,7 @@ cmd_state_sampler(const struct cell_command_sampler *sampler)
       ASSERT(0);
    }
 
-   update_tex_masks(&spu.texture[unit], &spu.sampler[unit]);
+   update_tex_masks(&spu.texture[unit], &spu.sampler[unit], unit);
 }
 
 
@@ -393,6 +399,7 @@ cmd_state_texture(const struct cell_command_texture *texture)
    DEBUG_PRINTF("TEXTURE [%u]\n", texture->unit);
 
    spu.texture[unit].max_level = 0;
+   spu.texture[unit].target = texture->target;
 
    for (i = 0; i < CELL_MAX_TEXTURE_LEVELS; i++) {
       uint width = texture->width[i];
@@ -408,6 +415,10 @@ cmd_state_texture(const struct cell_command_texture *texture)
       spu.texture[unit].level[i].tiles_per_row =
          (width + TILE_SIZE - 1) / TILE_SIZE;
 
+      spu.texture[unit].level[i].bytes_per_image =
+         4 * ((width + TILE_SIZE - 1) & ~(TILE_SIZE-1))
+         * ((height + TILE_SIZE - 1) & ~(TILE_SIZE-1));
+
       spu.texture[unit].level[i].max_s = spu_splats((int) width - 1);
       spu.texture[unit].level[i].max_t = spu_splats((int) height - 1);
 
@@ -415,7 +426,7 @@ cmd_state_texture(const struct cell_command_texture *texture)
          spu.texture[unit].max_level = i;
    }
 
-   update_tex_masks(&spu.texture[unit], &spu.sampler[unit]);
+   update_tex_masks(&spu.texture[unit], &spu.sampler[unit], unit);
 
    //Debug=0;
 }
diff --git a/src/gallium/drivers/cell/spu/spu_funcs.c b/src/gallium/drivers/cell/spu/spu_funcs.c
index 66b82f673..5c3ee305d 100644
--- a/src/gallium/drivers/cell/spu/spu_funcs.c
+++ b/src/gallium/drivers/cell/spu/spu_funcs.c
@@ -106,7 +106,7 @@ spu_txp(vector float s, vector float t, vector float r, vector float q,
         unsigned unit)
 {
    struct vec_4x4 colors;
-   spu.sample_texture4[unit](s, t, r, q, unit, 0, colors.v);
+   spu.sample_texture4[unit](s, t, r, q, unit, 0, 0, colors.v);
    return colors;
 }
 
diff --git a/src/gallium/drivers/cell/spu/spu_main.h b/src/gallium/drivers/cell/spu/spu_main.h
index 45c6f4ced..8781041bf 100644
--- a/src/gallium/drivers/cell/spu/spu_main.h
+++ b/src/gallium/drivers/cell/spu/spu_main.h
@@ -68,7 +68,7 @@ typedef void (*spu_sample_texture4_func)(vector float s,
                                          vector float t,
                                          vector float r,
                                          vector float q,
-                                         uint unit, uint level,
+                                         uint unit, uint level, uint face,
                                          vector float colors[4]);
 
 
@@ -113,6 +113,7 @@ struct spu_texture_level
    void *start;
    ushort width, height;
    ushort tiles_per_row;
+   uint bytes_per_image;
    /** texcoord scale factors */
    vector float scale_s, scale_t;
    /** texcoord masks (if REPEAT then size-1, else ~0) */
@@ -126,6 +127,7 @@ struct spu_texture
 {
    struct spu_texture_level level[CELL_MAX_TEXTURE_LEVELS];
    uint max_level;
+   uint target;  /**< PIPE_TEXTURE_x */
 } ALIGN16_ATTRIB;
 
 
diff --git a/src/gallium/drivers/cell/spu/spu_texture.c b/src/gallium/drivers/cell/spu/spu_texture.c
index b21c43a46..2570f02c7 100644
--- a/src/gallium/drivers/cell/spu/spu_texture.c
+++ b/src/gallium/drivers/cell/spu/spu_texture.c
@@ -48,6 +48,9 @@ invalidate_tex_cache(void)
       uint bytes = 4 * spu.texture[unit].level[lvl].width
          * spu.texture[unit].level[lvl].height;
 
+      if (spu.texture[unit].target == PIPE_TEXTURE_CUBE)
+         bytes *= 6;
+
       spu_dcache_mark_dirty((unsigned) spu.texture[unit].level[lvl].start, bytes);
    }
 }
@@ -67,11 +70,11 @@ invalidate_tex_cache(void)
  * a time.
  */
 static void
-get_four_texels(uint unit, uint level, vec_int4 x, vec_int4 y,
+get_four_texels(uint unit, uint level, uint face, vec_int4 x, vec_int4 y,
                 vec_uint4 *texels)
 {
    const struct spu_texture_level *tlevel = &spu.texture[unit].level[level];
-   const unsigned texture_ea = (uintptr_t) tlevel->start;
+   unsigned texture_ea = (uintptr_t) tlevel->start;
    const vec_int4 tile_x = spu_rlmask(x, -5);  /* tile_x = x / 32 */
    const vec_int4 tile_y = spu_rlmask(y, -5);  /* tile_y = y / 32 */
    const qword offset_x = si_andi((qword) x, 0x1f); /* offset_x = x & 0x1f */
@@ -88,6 +91,8 @@ get_four_texels(uint unit, uint level, vec_int4 x, vec_int4 y,
    
    vec_uint4 offset = (vec_uint4) si_a(tile_offset, texel_offset);
    
+   texture_ea = texture_ea + face * tlevel->bytes_per_image;
+
    spu_dcache_fetch_unaligned((qword *) & texels[0],
                               texture_ea + spu_extract(offset, 0), 4);
    spu_dcache_fetch_unaligned((qword *) & texels[1],
@@ -121,7 +126,8 @@ spu_clamp(vector signed int vec, vector signed int max)
 void
 sample_texture4_nearest(vector float s, vector float t,
                         vector float r, vector float q,
-                        uint unit, uint level, vector float colors[4])
+                        uint unit, uint level, uint face,
+                        vector float colors[4])
 {
    const struct spu_texture_level *tlevel = &spu.texture[unit].level[level];
    vector float ss = spu_mul(s, tlevel->scale_s);
@@ -138,7 +144,7 @@ sample_texture4_nearest(vector float s, vector float t,
    is = spu_clamp(is, tlevel->max_s);
    it = spu_clamp(it, tlevel->max_t);
 
-   get_four_texels(unit, level, is, it, texels);
+   get_four_texels(unit, level, face, is, it, texels);
 
    /* convert four packed ARGBA pixels to float RRRR,GGGG,BBBB,AAAA */
    spu_unpack_A8R8G8B8_transpose4(texels, colors);
@@ -152,11 +158,14 @@ sample_texture4_nearest(vector float s, vector float t,
 void
 sample_texture4_bilinear(vector float s, vector float t,
                          vector float r, vector float q,
-                         uint unit, uint level, vector float colors[4])
+                         uint unit, uint level, uint face,
+                         vector float colors[4])
 {
    const struct spu_texture_level *tlevel = &spu.texture[unit].level[level];
-   vector float ss = spu_madd(s, tlevel->scale_s,  spu_splats(-0.5f));
-   vector float tt = spu_madd(t, tlevel->scale_t, spu_splats(-0.5f));
+   static const vector float half = {-0.5f, -0.5f, -0.5f, -0.5f};
+
+   vector float ss = spu_madd(s, tlevel->scale_s, half);
+   vector float tt = spu_madd(t, tlevel->scale_t, half);
 
    vector signed int is0 = spu_convts(ss, 0);
    vector signed int it0 = spu_convts(tt, 0);
@@ -179,10 +188,10 @@ sample_texture4_bilinear(vector float s, vector float t,
 
    /* get packed int texels */
    vector unsigned int texels[16];
-   get_four_texels(unit, level, is0, it0, texels + 0);  /* upper-left */
-   get_four_texels(unit, level, is1, it0, texels + 4);  /* upper-right */
-   get_four_texels(unit, level, is0, it1, texels + 8);  /* lower-left */
-   get_four_texels(unit, level, is1, it1, texels + 12); /* lower-right */
+   get_four_texels(unit, level, face, is0, it0, texels + 0);  /* upper-left */
+   get_four_texels(unit, level, face, is1, it0, texels + 4);  /* upper-right */
+   get_four_texels(unit, level, face, is0, it1, texels + 8);  /* lower-left */
+   get_four_texels(unit, level, face, is1, it1, texels + 12); /* lower-right */
 
    /* XXX possibly rework following code to compute the weighted sample
     * colors with integer arithmetic for fewer int->float conversions.
@@ -299,10 +308,12 @@ transpose(vector unsigned int *mOut0,
 void
 sample_texture4_bilinear_2(vector float s, vector float t,
                            vector float r, vector float q,
-                           uint unit, uint level, vector float colors[4])
+                           uint unit, uint level, uint face,
+                           vector float colors[4])
 {
    const struct spu_texture_level *tlevel = &spu.texture[unit].level[level];
    static const vector float half = {-0.5f, -0.5f, -0.5f, -0.5f};
+
    /* Scale texcoords by size of texture, and add half pixel bias */
    vector float ss = spu_madd(s, tlevel->scale_s, half);
    vector float tt = spu_madd(t, tlevel->scale_t, half);
@@ -339,10 +350,10 @@ sample_texture4_bilinear_2(vector float s, vector float t,
 
    /* get packed int texels */
    vector unsigned int texels[16];
-   get_four_texels(unit, level, is0, it0, texels + 0);  /* upper-left */
-   get_four_texels(unit, level, is1, it0, texels + 4);  /* upper-right */
-   get_four_texels(unit, level, is0, it1, texels + 8);  /* lower-left */
-   get_four_texels(unit, level, is1, it1, texels + 12); /* lower-right */
+   get_four_texels(unit, level, face, is0, it0, texels + 0);  /* upper-left */
+   get_four_texels(unit, level, face, is1, it0, texels + 4);  /* upper-right */
+   get_four_texels(unit, level, face, is0, it1, texels + 8);  /* lower-left */
+   get_four_texels(unit, level, face, is1, it1, texels + 12); /* lower-right */
 
    /* twiddle packed 32-bit BGRA pixels into RGBA as four unsigned ints */
    {
@@ -433,7 +444,8 @@ compute_lambda(uint unit, vector float s, vector float t)
 void
 sample_texture4_lod(vector float s, vector float t,
                     vector float r, vector float q,
-                    uint unit, uint level_ignored, vector float colors[4])
+                    uint unit, uint level_ignored, uint face,
+                    vector float colors[4])
 {
    /*
     * Note that we're computing a lambda/lod here that's used for all
@@ -452,15 +464,136 @@ sample_texture4_lod(vector float s, vector float t,
 
    if (lambda <= 0.0f) {
       /* magnify */
-      spu.mag_sample_texture4[unit](s, t, r, q, unit, 0, colors);
+      spu.mag_sample_texture4[unit](s, t, r, q, unit, 0, 0, colors);
    }
    else {
       /* minify */
       int level = (int) (lambda + 0.5f);
       if (level > (int) spu.texture[unit].max_level)
          level = spu.texture[unit].max_level;
-      spu.min_sample_texture4[unit](s, t, r, q, unit, level, colors);
+      spu.min_sample_texture4[unit](s, t, r, q, unit, level, 0, colors);
       /* XXX to do: mipmap level interpolation */
    }
 }
 
+
+/** XXX need a SIMD version of this */
+static unsigned
+choose_cube_face(float rx, float ry, float rz, float *newS, float *newT)
+{
+   /*
+      major axis
+      direction     target                             sc     tc    ma
+      ----------    -------------------------------    ---    ---   ---
+       +rx          TEXTURE_CUBE_MAP_POSITIVE_X_EXT    -rz    -ry   rx
+       -rx          TEXTURE_CUBE_MAP_NEGATIVE_X_EXT    +rz    -ry   rx
+       +ry          TEXTURE_CUBE_MAP_POSITIVE_Y_EXT    +rx    +rz   ry
+       -ry          TEXTURE_CUBE_MAP_NEGATIVE_Y_EXT    +rx    -rz   ry
+       +rz          TEXTURE_CUBE_MAP_POSITIVE_Z_EXT    +rx    -ry   rz
+       -rz          TEXTURE_CUBE_MAP_NEGATIVE_Z_EXT    -rx    -ry   rz
+   */
+   const float arx = fabsf(rx);
+   const float ary = fabsf(ry);
+   const float arz = fabsf(rz);
+   unsigned face;
+   float sc, tc, ma;
+
+   if (arx > ary && arx > arz) {
+      if (rx >= 0.0F) {
+         face = PIPE_TEX_FACE_POS_X;
+         sc = -rz;
+         tc = -ry;
+         ma = arx;
+      }
+      else {
+         face = PIPE_TEX_FACE_NEG_X;
+         sc = rz;
+         tc = -ry;
+         ma = arx;
+      }
+   }
+   else if (ary > arx && ary > arz) {
+      if (ry >= 0.0F) {
+         face = PIPE_TEX_FACE_POS_Y;
+         sc = rx;
+         tc = rz;
+         ma = ary;
+      }
+      else {
+         face = PIPE_TEX_FACE_NEG_Y;
+         sc = rx;
+         tc = -rz;
+         ma = ary;
+      }
+   }
+   else {
+      if (rz > 0.0F) {
+         face = PIPE_TEX_FACE_POS_Z;
+         sc = rx;
+         tc = -ry;
+         ma = arz;
+      }
+      else {
+         face = PIPE_TEX_FACE_NEG_Z;
+         sc = -rx;
+         tc = -ry;
+         ma = arz;
+      }
+   }
+
+   *newS = (sc / ma + 1.0F) * 0.5F;
+   *newT = (tc / ma + 1.0F) * 0.5F;
+
+   return face;
+}
+
+
+
+void
+sample_texture4_cube(vector float s, vector float t,
+                     vector float r, vector float q,
+                     uint unit, uint level, int face_ignored,
+                     vector float colors[4])
+{
+   static const vector float zero = {0.0f, 0.0f, 0.0f, 0.0f};
+   uint p, faces[4];
+   float newS[4], newT[4];
+
+   /* Compute cube face referenced by the four sets of texcoords.
+    * XXX we should SIMD-ize this.
+    */
+   for (p = 0; p < 4; p++) {      
+      float rx = spu_extract(s, p);
+      float ry = spu_extract(t, p);
+      float rz = spu_extract(r, p);
+      faces[p] = choose_cube_face(rx, ry, rz, &newS[p], &newT[p]);
+   }
+
+   if (faces[0] == faces[1] &&
+       faces[0] == faces[2] &&
+       faces[0] == faces[3]) {
+      /* GOOD!  All four texcoords refer to the same cube face */
+      s = (vector float) {newS[0], newS[1], newS[2], newS[3]};
+      t = (vector float) {newT[0], newT[1], newT[2], newT[3]};
+      sample_texture4_nearest(s, t, zero, zero, unit, level, faces[0], colors);
+   }
+   else {
+      /* BAD!  The four texcoords refer to different faces */
+      for (p = 0; p < 4; p++) {      
+         vector float c[4];
+
+         sample_texture4_nearest(spu_splats(newS[p]), spu_splats(newT[p]),
+                                 zero, zero, unit, level, faces[p], c);
+
+         float red = spu_extract(c[0], p);
+         float green = spu_extract(c[1], p);
+         float blue = spu_extract(c[2], p);
+         float alpha = spu_extract(c[3], p);
+
+         colors[0] = spu_insert(red,   colors[0], p);
+         colors[1] = spu_insert(green, colors[1], p);
+         colors[2] = spu_insert(blue,  colors[2], p);
+         colors[3] = spu_insert(alpha, colors[3], p);
+      }
+   }
+}
diff --git a/src/gallium/drivers/cell/spu/spu_texture.h b/src/gallium/drivers/cell/spu/spu_texture.h
index ec06a50b4..08b891a4a 100644
--- a/src/gallium/drivers/cell/spu/spu_texture.h
+++ b/src/gallium/drivers/cell/spu/spu_texture.h
@@ -39,24 +39,35 @@ invalidate_tex_cache(void);
 extern void
 sample_texture4_nearest(vector float s, vector float t,
                         vector float r, vector float q,
-                        uint unit, uint level, vector float colors[4]);
+                        uint unit, uint level, uint face,
+                        vector float colors[4]);
 
 
 extern void
 sample_texture4_bilinear(vector float s, vector float t,
                          vector float r, vector float q,
-                         uint unit, uint level, vector float colors[4]);
+                         uint unit, uint level, uint face,
+                         vector float colors[4]);
 
 extern void
 sample_texture4_bilinear_2(vector float s, vector float t,
-                         vector float r, vector float q,
-                           uint unit, uint level, vector float colors[4]);
+                           vector float r, vector float q,
+                           uint unit, uint level, uint face,
+                           vector float colors[4]);
 
 
 extern void
 sample_texture4_lod(vector float s, vector float t,
                     vector float r, vector float q,
-                    uint unit, uint level, vector float colors[4]);
+                    uint unit, uint level, uint face,
+                    vector float colors[4]);
+
+
+extern void
+sample_texture4_cube(vector float s, vector float t,
+                     vector float r, vector float q,
+                     uint unit, uint level_ignored, int face_ignored,
+                     vector float colors[4]);
 
 
 #endif /* SPU_TEXTURE_H */