cell: initial work to support multi-texture

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];
 };
 
 

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));
    }

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);
+   }
 }
 
 

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);
 

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);
 

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);