Cell: basic texture mapping

Texture images are tiled in PPU code.  SPUs use a texture cache for getting
texels from textures.
This is very rough code, but demos/texcyl.c works.

4 files changed, 59 insertions, 22 deletions

diff --git a/src/mesa/pipe/cell/spu/Makefile b/src/mesa/pipe/cell/spu/Makefile
index 417ae1b072..011fdcefe3 100644
--- a/src/mesa/pipe/cell/spu/Makefile
+++ b/src/mesa/pipe/cell/spu/Makefile
@@ -17,6 +17,7 @@ PROG_SPU_EMBED_O = $(PROG)_spu-embed.o
 
 SOURCES = \
 	spu_main.c \
+	spu_texture.c \
 	spu_tile.c \
 	spu_tri.c
 
diff --git a/src/mesa/pipe/cell/spu/spu_main.c b/src/mesa/pipe/cell/spu/spu_main.c
index c2b05ed5a2..5a5b17dd89 100644
--- a/src/mesa/pipe/cell/spu/spu_main.c
+++ b/src/mesa/pipe/cell/spu/spu_main.c
@@ -34,6 +34,7 @@
 #include <spu_mfcio.h>
 
 #include "spu_main.h"
+#include "spu_texture.h"
 #include "spu_tri.h"
 #include "spu_tile.h"
 #include "pipe/cell/common.h"
@@ -447,6 +448,17 @@ 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);
+
+   memcpy(&spu.texture, texture, sizeof(*texture));
+}
+
+
+static void
 cmd_state_vertex_info(const struct vertex_info *vinfo)
 {
    if (Debug) {
@@ -561,6 +573,10 @@ cmd_batch(uint opcode)
          cmd_state_sampler((struct pipe_sampler_state *) &buffer[pos+1]);
          pos += (1 + sizeof(struct pipe_sampler_state) / 4);
          break;
+      case CELL_CMD_STATE_TEXTURE:
+         cmd_state_texture((struct cell_command_texture *) &buffer[pos+1]);
+         pos += (1 + sizeof(struct cell_command_texture) / 4);
+         break;
       case CELL_CMD_STATE_VERTEX_INFO:
          cmd_state_vertex_info((struct vertex_info *) &buffer[pos+1]);
          pos += (1 + sizeof(struct vertex_info) / 4);
@@ -656,6 +672,7 @@ one_time_init(void)
 {
    memset(tile_status, TILE_STATUS_DEFINED, sizeof(tile_status));
    memset(tile_status_z, TILE_STATUS_DEFINED, sizeof(tile_status_z));
+   invalidate_tex_cache();
 }
 
 
diff --git a/src/mesa/pipe/cell/spu/spu_main.h b/src/mesa/pipe/cell/spu/spu_main.h
index 5bc5d9fa99..480c54ebd0 100644
--- a/src/mesa/pipe/cell/spu/spu_main.h
+++ b/src/mesa/pipe/cell/spu/spu_main.h
@@ -60,6 +60,7 @@ struct spu_global
    struct pipe_depth_stencil_alpha_state depth_stencil;
    struct pipe_blend_state blend;
    struct pipe_sampler_state sampler[PIPE_MAX_SAMPLERS];
+   struct cell_command_texture texture;
 
    struct vertex_info vertex_info;
 
@@ -84,6 +85,8 @@ extern struct spu_global spu;
 #define TAG_INDEX_BUFFER      16
 #define TAG_BATCH_BUFFER      17
 #define TAG_MISC              18
+#define TAG_TEXTURE_TILE      19
+
 
 
 extern void
diff --git a/src/mesa/pipe/cell/spu/spu_tri.c b/src/mesa/pipe/cell/spu/spu_tri.c
index 3d0d106c10..aad28f1036 100644
--- a/src/mesa/pipe/cell/spu/spu_tri.c
+++ b/src/mesa/pipe/cell/spu/spu_tri.c
@@ -33,6 +33,7 @@
 #include "pipe/p_format.h"
 #include "pipe/p_util.h"
 #include "spu_main.h"
+#include "spu_texture.h"
 #include "spu_tile.h"
 #include "spu_tri.h"
 
@@ -362,9 +363,24 @@ emit_quad( struct setup_stage *setup, int x, int y, unsigned mask )
    /* Cell: "write" quad fragments to the tile by setting prim color */
    const int ix = x - setup->cliprect_minx;
    const int iy = y - setup->cliprect_miny;
-   float colors[4][4];
-
-   eval_coeff(setup, 1, (float) x, (float) y, colors);
+   uint colors[4];  /* indexed by QUAD_x */
+
+   if (spu.texture.start) {
+      float texcoords[4][4];
+      uint i;
+      eval_coeff(setup, 2, (float) x, (float) y, texcoords);
+      for (i = 0; i < 4; i++) {
+         colors[i] = sample_texture(texcoords[i]);
+      }
+   }
+   else {
+      float fcolors[4][4];
+      eval_coeff(setup, 1, (float) x, (float) y, fcolors);
+      colors[QUAD_TOP_LEFT] = pack_color(fcolors[QUAD_TOP_LEFT]);
+      colors[QUAD_TOP_RIGHT] = pack_color(fcolors[QUAD_TOP_RIGHT]);
+      colors[QUAD_BOTTOM_LEFT] = pack_color(fcolors[QUAD_BOTTOM_LEFT]);
+      colors[QUAD_BOTTOM_RIGHT] = pack_color(fcolors[QUAD_BOTTOM_RIGHT]);
+   }
 
    if (spu.depth_stencil.depth.enabled) {
       mask &= do_depth_test(setup, x, y, mask);
@@ -382,13 +398,13 @@ emit_quad( struct setup_stage *setup, int x, int y, unsigned mask )
       tile_status[setup->ty][setup->tx] = TILE_STATUS_DIRTY;
 
       if (mask & MASK_TOP_LEFT)
-         ctile.t32[iy][ix] = pack_color(colors[QUAD_TOP_LEFT]);
+         ctile.t32[iy][ix] = colors[QUAD_TOP_LEFT];
       if (mask & MASK_TOP_RIGHT)
-         ctile.t32[iy][ix+1] = pack_color(colors[QUAD_TOP_RIGHT]);
+         ctile.t32[iy][ix+1] = colors[QUAD_TOP_RIGHT];
       if (mask & MASK_BOTTOM_LEFT)
-         ctile.t32[iy+1][ix] = pack_color(colors[QUAD_BOTTOM_LEFT]);
+         ctile.t32[iy+1][ix] = colors[QUAD_BOTTOM_LEFT];
       if (mask & MASK_BOTTOM_RIGHT)
-         ctile.t32[iy+1][ix+1] = pack_color(colors[QUAD_BOTTOM_RIGHT]);
+         ctile.t32[iy+1][ix+1] = colors[QUAD_BOTTOM_RIGHT];
    }
 #endif
 }
@@ -606,7 +622,6 @@ static boolean setup_sort_vertices( struct setup_stage *setup,
 }
 
 
-#if 0
 /**
  * Compute a0 for a constant-valued coefficient (GL_FLAT shading).
  * The value value comes from vertex->data[slot][i].
@@ -614,21 +629,20 @@ static boolean setup_sort_vertices( struct setup_stage *setup,
  * \param slot  which attribute slot 
  * \param i  which component of the slot (0..3)
  */
-static void const_coeff( struct setup_stage *setup,
-			 unsigned slot,
-			 unsigned i )
+static void const_coeff(struct setup_stage *setup, uint slot)
 {
-   assert(slot < PIPE_MAX_SHADER_INPUTS);
-   assert(i <= 3);
+   uint i;
+   ASSERT(slot < PIPE_MAX_SHADER_INPUTS);
 
-   setup->coef[slot].dadx[i] = 0;
-   setup->coef[slot].dady[i] = 0;
+   for (i = 0; i < 4; i++) {
+      setup->coef[slot].dadx[i] = 0;
+      setup->coef[slot].dady[i] = 0;
 
-   /* need provoking vertex info!
-    */
-   setup->coef[slot].a0[i] = setup->vprovoke->data[slot][i];
+      /* need provoking vertex info!
+       */
+      setup->coef[slot].a0[i] = setup->vprovoke->data[slot][i];
+   }
 }
-#endif
 
 
 /**
@@ -735,15 +749,17 @@ static void setup_tri_coefficients( struct setup_stage *setup )
       case INTERP_NONE:
          break;
       case INTERP_POS:
-         tri_linear_coeff(setup, i, 2, 3);  /* slot 0, z */
+         tri_linear_coeff(setup, i, 2, 3);
          /* XXX interp W if PERSPECTIVE... */
          break;
       case INTERP_CONSTANT:
-         /* fall-through */
+         const_coeff(setup, i);
+         break;
       case INTERP_LINEAR:
-         tri_linear_coeff(setup, i, 0, 4);  /* slot 1, color */
+         tri_linear_coeff(setup, i, 0, 4);
          break;
       case INTERP_PERSPECTIVE:
+         tri_linear_coeff(setup, i, 0, 4); /* XXX temporary */
          break;
       default:
          ASSERT(0);