Merge branch 'master' into pipe-video

Conflicts:
	src/gallium/include/pipe/p_format.h

8 files changed, 130 insertions, 29 deletions

diff --git a/src/gallium/drivers/softpipe/sp_fs_exec.c b/src/gallium/drivers/softpipe/sp_fs_exec.c
index 67e2c8f8bc4..346e1b402ba 100644
--- a/src/gallium/drivers/softpipe/sp_fs_exec.c
+++ b/src/gallium/drivers/softpipe/sp_fs_exec.c
@@ -158,9 +158,17 @@ exec_run( const struct sp_fragment_shader *base,
          case TGSI_SEMANTIC_POSITION:
             {
                uint j;
-               for (j = 0; j < 4; j++) {
+
+               for (j = 0; j < 4; j++)
                   quad->output.depth[j] = machine->Outputs[i].xyzw[2].f[j];
-               }
+            }
+            break;
+         case TGSI_SEMANTIC_STENCIL:
+            {
+               uint j;
+
+               for (j = 0; j < 4; j++)
+                  quad->output.stencil[j] = (unsigned)machine->Outputs[i].xyzw[1].f[j];
             }
             break;
          }
diff --git a/src/gallium/drivers/softpipe/sp_fs_sse.c b/src/gallium/drivers/softpipe/sp_fs_sse.c
index daa158df7c4..5b18cd035e3 100644
--- a/src/gallium/drivers/softpipe/sp_fs_sse.c
+++ b/src/gallium/drivers/softpipe/sp_fs_sse.c
@@ -169,9 +169,15 @@ fs_sse_run( const struct sp_fragment_shader *base,
          case TGSI_SEMANTIC_POSITION:
             {
                uint j;
-               for (j = 0; j < 4; j++) {
-                  quad->output.depth[j] = machine->Outputs[0].xyzw[2].f[j];
-               }
+               for (j = 0; j < 4; j++)
+                  quad->output.depth[j] = machine->Outputs[i].xyzw[2].f[j];
+            }
+            break;
+         case TGSI_SEMANTIC_STENCIL:
+            {
+               uint j;
+               for (j = 0; j < 4; j++)
+                  quad->output.stencil[j] = machine->Outputs[i].xyzw[1].f[j];
             }
             break;
          }
diff --git a/src/gallium/drivers/softpipe/sp_quad.h b/src/gallium/drivers/softpipe/sp_quad.h
index a3236bd1169..e745aa80619 100644
--- a/src/gallium/drivers/softpipe/sp_quad.h
+++ b/src/gallium/drivers/softpipe/sp_quad.h
@@ -85,6 +85,7 @@ struct quad_header_output
    /** colors in SOA format (rrrr, gggg, bbbb, aaaa) */
    float color[PIPE_MAX_COLOR_BUFS][NUM_CHANNELS][QUAD_SIZE];
    float depth[QUAD_SIZE];
+   uint8_t stencil[QUAD_SIZE];
 };
 
 
diff --git a/src/gallium/drivers/softpipe/sp_quad_depth_test.c b/src/gallium/drivers/softpipe/sp_quad_depth_test.c
index e9b92626176..c8f5f89568a 100644
--- a/src/gallium/drivers/softpipe/sp_quad_depth_test.c
+++ b/src/gallium/drivers/softpipe/sp_quad_depth_test.c
@@ -47,6 +47,8 @@ struct depth_data {
    unsigned bzzzz[QUAD_SIZE];  /**< Z values fetched from depth buffer */
    unsigned qzzzz[QUAD_SIZE];  /**< Z values from the quad */
    ubyte stencilVals[QUAD_SIZE];
+   boolean use_shader_stencil_refs;
+   ubyte shader_stencil_refs[QUAD_SIZE];
    struct softpipe_cached_tile *tile;
 };
 
@@ -186,6 +188,33 @@ convert_quad_depth( struct depth_data *data,
 }
 
 
+/**
+ * Compute the depth_data::shader_stencil_refs[] values from the float fragment stencil values.
+ */
+static void
+convert_quad_stencil( struct depth_data *data, 
+                      const struct quad_header *quad )
+{
+   unsigned j;
+
+   data->use_shader_stencil_refs = TRUE;
+   /* Copy quads stencil values
+    */
+   switch (data->format) {
+   case PIPE_FORMAT_Z24X8_UNORM:
+   case PIPE_FORMAT_Z24_UNORM_S8_USCALED:
+   case PIPE_FORMAT_X8Z24_UNORM:
+   case PIPE_FORMAT_S8_USCALED_Z24_UNORM:
+      {
+         for (j = 0; j < QUAD_SIZE; j++) {
+            data->shader_stencil_refs[j] = ((unsigned)(quad->output.stencil[j]));
+         }
+      }
+      break;
+   default:
+      assert(0);
+   }
+}
 
 /**
  * Write data->bzzzz[] values and data->stencilVals into the Z/stencil buffer.
@@ -272,8 +301,14 @@ do_stencil_test(struct depth_data *data,
 {
    unsigned passMask = 0x0;
    unsigned j;
+   ubyte refs[QUAD_SIZE];
 
-   ref &= valMask;
+   for (j = 0; j < QUAD_SIZE; j++) {
+      if (data->use_shader_stencil_refs)
+         refs[j] = data->shader_stencil_refs[j] & valMask;
+      else 
+         refs[j] = ref & valMask;
+   }
 
    switch (func) {
    case PIPE_FUNC_NEVER:
@@ -281,42 +316,42 @@ do_stencil_test(struct depth_data *data,
       break;
    case PIPE_FUNC_LESS:
       for (j = 0; j < QUAD_SIZE; j++) {
-         if (ref < (data->stencilVals[j] & valMask)) {
+         if (refs[j] < (data->stencilVals[j] & valMask)) {
             passMask |= (1 << j);
          }
       }
       break;
    case PIPE_FUNC_EQUAL:
       for (j = 0; j < QUAD_SIZE; j++) {
-         if (ref == (data->stencilVals[j] & valMask)) {
+         if (refs[j] == (data->stencilVals[j] & valMask)) {
             passMask |= (1 << j);
          }
       }
       break;
    case PIPE_FUNC_LEQUAL:
       for (j = 0; j < QUAD_SIZE; j++) {
-         if (ref <= (data->stencilVals[j] & valMask)) {
+         if (refs[j] <= (data->stencilVals[j] & valMask)) {
             passMask |= (1 << j);
          }
       }
       break;
    case PIPE_FUNC_GREATER:
       for (j = 0; j < QUAD_SIZE; j++) {
-         if (ref > (data->stencilVals[j] & valMask)) {
+         if (refs[j] > (data->stencilVals[j] & valMask)) {
             passMask |= (1 << j);
          }
       }
       break;
    case PIPE_FUNC_NOTEQUAL:
       for (j = 0; j < QUAD_SIZE; j++) {
-         if (ref != (data->stencilVals[j] & valMask)) {
+         if (refs[j] != (data->stencilVals[j] & valMask)) {
             passMask |= (1 << j);
          }
       }
       break;
    case PIPE_FUNC_GEQUAL:
       for (j = 0; j < QUAD_SIZE; j++) {
-         if (ref >= (data->stencilVals[j] & valMask)) {
+         if (refs[j] >= (data->stencilVals[j] & valMask)) {
             passMask |= (1 << j);
          }
       }
@@ -348,9 +383,14 @@ apply_stencil_op(struct depth_data *data,
 {
    unsigned j;
    ubyte newstencil[QUAD_SIZE];
+   ubyte refs[QUAD_SIZE];
 
    for (j = 0; j < QUAD_SIZE; j++) {
       newstencil[j] = data->stencilVals[j];
+      if (data->use_shader_stencil_refs)
+         refs[j] = data->shader_stencil_refs[j];
+      else
+         refs[j] = ref;
    }
 
    switch (op) {
@@ -367,7 +407,7 @@ apply_stencil_op(struct depth_data *data,
    case PIPE_STENCIL_OP_REPLACE:
       for (j = 0; j < QUAD_SIZE; j++) {
          if (mask & (1 << j)) {
-            newstencil[j] = ref;
+            newstencil[j] = refs[j];
          }
       }
       break;
@@ -688,8 +728,10 @@ depth_test_quads_fallback(struct quad_stage *qs,
    unsigned i, pass = 0;
    const struct sp_fragment_shader *fs = qs->softpipe->fs;
    boolean interp_depth = !fs->info.writes_z;
+   boolean shader_stencil_ref = fs->info.writes_stencil;
    struct depth_data data;
 
+   data.use_shader_stencil_refs = FALSE;
 
    if (qs->softpipe->depth_stencil->alpha.enabled) {
       nr = alpha_test_quads(qs, quads, nr);
@@ -716,6 +758,9 @@ depth_test_quads_fallback(struct quad_stage *qs,
          }
 
          if (qs->softpipe->depth_stencil->stencil[0].enabled) {
+            if (shader_stencil_ref)
+               convert_quad_stencil(&data, quads[i]);
+            
             depth_stencil_test_quad(qs, &data, quads[i]);
             write_depth_stencil_values(&data, quads[i]);
          }
diff --git a/src/gallium/drivers/softpipe/sp_quad_pipe.c b/src/gallium/drivers/softpipe/sp_quad_pipe.c
index 43b8e88e334..2cfd02a22c6 100644
--- a/src/gallium/drivers/softpipe/sp_quad_pipe.c
+++ b/src/gallium/drivers/softpipe/sp_quad_pipe.c
@@ -47,7 +47,8 @@ sp_build_quad_pipeline(struct softpipe_context *sp)
       sp->framebuffer.zsbuf &&
       !sp->depth_stencil->alpha.enabled &&
       !sp->fs->info.uses_kill &&
-      !sp->fs->info.writes_z;
+      !sp->fs->info.writes_z &&
+      !sp->fs->info.writes_stencil;
 
    sp->quad.first = sp->quad.blend;
 
diff --git a/src/gallium/drivers/softpipe/sp_screen.c b/src/gallium/drivers/softpipe/sp_screen.c
index 25a0a622179..edc2a6dacf2 100644
--- a/src/gallium/drivers/softpipe/sp_screen.c
+++ b/src/gallium/drivers/softpipe/sp_screen.c
@@ -113,8 +113,12 @@ softpipe_get_param(struct pipe_screen *screen, enum pipe_cap param)
       return 1;
    case PIPE_CAP_STREAM_OUTPUT:
       return 1;
+   case PIPE_CAP_PRIMITIVE_RESTART:
+      return 1;
    case PIPE_CAP_DEPTHSTENCIL_CLEAR_SEPARATE:
       return 0;
+   case PIPE_CAP_SHADER_STENCIL_EXPORT:
+      return 1;
    default:
       return 0;
    }
@@ -206,13 +210,6 @@ softpipe_is_format_supported( struct pipe_screen *screen,
       if (format_desc->block.width != 1 ||
           format_desc->block.height != 1)
          return FALSE;
-
-      /*
-       * TODO: Unfortunately we cannot render into anything more than 32 bits
-       * because we encode color clear values into a 32bit word.
-       */
-      if (format_desc->block.bits > 32)
-         return FALSE;
    }
 
    if (bind & PIPE_BIND_DEPTH_STENCIL) {
diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.c b/src/gallium/drivers/softpipe/sp_tex_sample.c
index 088e48f81fe..2eac4c7a82b 100644
--- a/src/gallium/drivers/softpipe/sp_tex_sample.c
+++ b/src/gallium/drivers/softpipe/sp_tex_sample.c
@@ -44,6 +44,9 @@
 #include "sp_tex_tile_cache.h"
 
 
+/** Set to one to help debug texture sampling */
+#define DEBUG_TEX 0
+
 
 /*
  * Return fractional part of 'f'.  Used for computing interpolation weights.
@@ -774,6 +777,18 @@ pot_level_size(unsigned base_pot, unsigned level)
 }
 
 
+static void
+print_sample(const char *function, float rgba[NUM_CHANNELS][QUAD_SIZE])
+{
+   debug_printf("%s %g %g %g %g, %g %g %g %g, %g %g %g %g, %g %g %g %g\n",
+                function,
+                rgba[0][0], rgba[1][0], rgba[2][0], rgba[3][0],
+                rgba[0][1], rgba[1][1], rgba[2][1], rgba[3][1],
+                rgba[0][2], rgba[1][2], rgba[2][2], rgba[3][2],
+                rgba[0][3], rgba[1][3], rgba[2][3], rgba[3][3]);
+}
+
+
 /* Some image-filter fastpaths:
  */
 static INLINE void
@@ -832,6 +847,10 @@ img_filter_2d_linear_repeat_POT(struct tgsi_sampler *tgsi_sampler,
                               tx[2][c], tx[3][c]);
       }
    }
+
+   if (DEBUG_TEX) {
+      print_sample(__FUNCTION__, rgba);
+   }
 }
 
 
@@ -872,6 +891,10 @@ img_filter_2d_nearest_repeat_POT(struct tgsi_sampler *tgsi_sampler,
          rgba[c][j] = out[c];
       }
    }
+
+   if (DEBUG_TEX) {
+      print_sample(__FUNCTION__, rgba);
+   }
 }
 
 
@@ -921,6 +944,10 @@ img_filter_2d_nearest_clamp_POT(struct tgsi_sampler *tgsi_sampler,
          rgba[c][j] = out[c];
       }
    }
+
+   if (DEBUG_TEX) {
+      print_sample(__FUNCTION__, rgba);
+   }
 }
 
 
@@ -957,6 +984,10 @@ img_filter_1d_nearest(struct tgsi_sampler *tgsi_sampler,
          rgba[c][j] = out[c];
       }
    }
+
+   if (DEBUG_TEX) {
+      print_sample(__FUNCTION__, rgba);
+   }
 }
 
 
@@ -997,6 +1028,10 @@ img_filter_2d_nearest(struct tgsi_sampler *tgsi_sampler,
          rgba[c][j] = out[c];
       }
    }
+
+   if (DEBUG_TEX) {
+      print_sample(__FUNCTION__, rgba);
+   }
 }
 
 
@@ -1045,6 +1080,10 @@ img_filter_cube_nearest(struct tgsi_sampler *tgsi_sampler,
          rgba[c][j] = out[c];
       }
    }
+
+   if (DEBUG_TEX) {
+      print_sample(__FUNCTION__, rgba);
+   }
 }
 
 
@@ -1357,6 +1396,10 @@ mip_filter_linear(struct tgsi_sampler *tgsi_sampler,
          }
       }
    }
+
+   if (DEBUG_TEX) {
+      print_sample(__FUNCTION__, rgba);
+   }
 }
 
 
@@ -1402,13 +1445,9 @@ mip_filter_nearest(struct tgsi_sampler *tgsi_sampler,
       samp->min_img_filter(tgsi_sampler, s, t, p, NULL, tgsi_sampler_lod_bias, rgba);
    }
 
-#if 0
-   printf("RGBA %g %g %g %g, %g %g %g %g, %g %g %g %g, %g %g %g %g\n",
-          rgba[0][0], rgba[1][0], rgba[2][0], rgba[3][0],
-          rgba[0][1], rgba[1][1], rgba[2][1], rgba[3][1],
-          rgba[0][2], rgba[1][2], rgba[2][2], rgba[3][2],
-          rgba[0][3], rgba[1][3], rgba[2][3], rgba[3][3]);
-#endif
+   if (DEBUG_TEX) {
+      print_sample(__FUNCTION__, rgba);
+   }
 }
 
 
@@ -1510,6 +1549,10 @@ mip_filter_linear_2d_linear_repeat_POT(
          }
       }
    }
+
+   if (DEBUG_TEX) {
+      print_sample(__FUNCTION__, rgba);
+   }
 }
 
 
diff --git a/src/gallium/drivers/softpipe/sp_tile_cache.h b/src/gallium/drivers/softpipe/sp_tile_cache.h
index 031c7c1ea5c..4151a47c323 100644
--- a/src/gallium/drivers/softpipe/sp_tile_cache.h
+++ b/src/gallium/drivers/softpipe/sp_tile_cache.h
@@ -86,7 +86,7 @@ struct softpipe_tile_cache
    struct softpipe_cached_tile *entries[NUM_ENTRIES];
    uint clear_flags[(MAX_WIDTH / TILE_SIZE) * (MAX_HEIGHT / TILE_SIZE) / 32];
    float clear_color[4];  /**< for color bufs */
-   uint clear_val;        /**< for z+stencil, or packed color clear value */
+   uint clear_val;        /**< for z+stencil */
    boolean depth_stencil; /**< Is the surface a depth/stencil format? */
 
    struct softpipe_cached_tile *tile;  /**< scratch tile for clears */