Merge branch 'master' into asm-shader-rework-1

Conflicts:
	src/mesa/shader/arbprogparse.c

5 files changed, 105 insertions, 45 deletions

diff --git a/src/gallium/auxiliary/draw/draw_pipe.c b/src/gallium/auxiliary/draw/draw_pipe.c
index 2e3f5b2fc0..1c6d657297 100644
--- a/src/gallium/auxiliary/draw/draw_pipe.c
+++ b/src/gallium/auxiliary/draw/draw_pipe.c
@@ -158,6 +158,60 @@ static void do_triangle( struct draw_context *draw,
 
 
 
+#define QUAD(i0,i1,i2,i3)                       \
+   do_triangle( draw,                           \
+                ( DRAW_PIPE_RESET_STIPPLE |     \
+                  DRAW_PIPE_EDGE_FLAG_0 |       \
+                  DRAW_PIPE_EDGE_FLAG_2 ),      \
+                verts + stride * elts[i0],      \
+                verts + stride * elts[i1],      \
+                verts + stride * elts[i3]);     \
+   do_triangle( draw,                           \
+                ( DRAW_PIPE_EDGE_FLAG_0 |       \
+                  DRAW_PIPE_EDGE_FLAG_1 ),      \
+                verts + stride * elts[i1],      \
+                verts + stride * elts[i2],      \
+                verts + stride * elts[i3])
+
+#define TRIANGLE(flags,i0,i1,i2)                                        \
+   do_triangle( draw,                                                   \
+                elts[i0],  /* flags */                          \
+                verts + stride * (elts[i0] & ~DRAW_PIPE_FLAG_MASK),     \
+                verts + stride * elts[i1],                              \
+                verts + stride * elts[i2])
+
+#define LINE(flags,i0,i1)                                       \
+   do_line( draw,                                               \
+            elts[i0],                                   \
+            verts + stride * (elts[i0] & ~DRAW_PIPE_FLAG_MASK), \
+            verts + stride * elts[i1])
+
+#define POINT(i0)                               \
+   do_point( draw,                              \
+             verts + stride * elts[i0] )
+
+#define FUNC pipe_run
+#define ARGS                                    \
+    struct draw_context *draw,                  \
+    unsigned prim,                              \
+    struct vertex_header *vertices,             \
+    unsigned stride,                            \
+    const ushort *elts
+
+#define LOCAL_VARS                                           \
+   char *verts = (char *)vertices;                           \
+   boolean flatfirst = (draw->rasterizer->flatshade &&       \
+                        draw->rasterizer->flatshade_first);  \
+   unsigned i;                                               \
+   ushort flags
+
+#define FLUSH
+
+#include "draw_pt_decompose.h"
+#undef ARGS
+#undef LOCAL_VARS
+
+
 
 /* Code to run the pipeline on a fairly arbitary collection of vertices.
  *
@@ -178,34 +232,12 @@ void draw_pipeline_run( struct draw_context *draw,
                         unsigned count )
 {
    char *verts = (char *)vertices;
-   unsigned i;
 
    draw->pipeline.verts = verts;
    draw->pipeline.vertex_stride = stride;
    draw->pipeline.vertex_count = vertex_count;
    
-   switch (prim) {
-   case PIPE_PRIM_POINTS:
-      for (i = 0; i < count; i++) 
-         do_point( draw, 
-                   verts + stride * elts[i] );
-      break;
-   case PIPE_PRIM_LINES:
-      for (i = 0; i+1 < count; i += 2) 
-         do_line( draw, 
-                  elts[i+0],  /* flags */
-                  verts + stride * (elts[i+0] & ~DRAW_PIPE_FLAG_MASK),
-                  verts + stride * elts[i+1]);
-      break;
-   case PIPE_PRIM_TRIANGLES:
-      for (i = 0; i+2 < count; i += 3)
-         do_triangle( draw, 
-                      elts[i+0],  /* flags */
-                      verts + stride * (elts[i+0] & ~DRAW_PIPE_FLAG_MASK),
-                      verts + stride * elts[i+1],
-                      verts + stride * elts[i+2]);
-      break;
-   }
+   pipe_run(draw, prim, vertices, stride, elts, count);
    
    draw->pipeline.verts = NULL;
    draw->pipeline.vertex_count = 0;
diff --git a/src/gallium/auxiliary/draw/draw_pipe_pstipple.c b/src/gallium/auxiliary/draw/draw_pipe_pstipple.c
index 30a6d2919d..283502cdf3 100644
--- a/src/gallium/auxiliary/draw/draw_pipe_pstipple.c
+++ b/src/gallium/auxiliary/draw/draw_pipe_pstipple.c
@@ -256,7 +256,10 @@ pstip_transform_inst(struct tgsi_transform_context *ctx,
          uint size = 4;
          immed = tgsi_default_full_immediate();
          immed.Immediate.NrTokens = 1 + size; /* one for the token itself */
-         immed.u.Pointer = (void *) value;
+         immed.u[0].Float = value[0];
+         immed.u[1].Float = value[1];
+         immed.u[2].Float = value[2];
+         immed.u[3].Float = value[3];
          ctx->emit_immediate(ctx, &immed);
       }
 
diff --git a/src/gallium/auxiliary/draw/draw_pipe_wide_point.c b/src/gallium/auxiliary/draw/draw_pipe_wide_point.c
index 014d8c7346..7d76a7dbf3 100644
--- a/src/gallium/auxiliary/draw/draw_pipe_wide_point.c
+++ b/src/gallium/auxiliary/draw/draw_pipe_wide_point.c
@@ -28,6 +28,30 @@
 /* Authors:  Keith Whitwell <keith@tungstengraphics.com>
  */
 
+/**
+ * Notes on wide points and sprite mode:
+ *
+ * In wide point/sprite mode we effectively need to convert each incoming
+ * vertex into four outgoing vertices specifying the corners of a quad.
+ * Since we don't (yet) have geometry shaders, we have to handle this here
+ * in the draw module.
+ *
+ * For sprites, it also means that this is where we have to handle texcoords
+ * for the vertices of the quad.  OpenGL's GL_COORD_REPLACE state specifies
+ * if/how enabled texcoords are automatically generated for sprites.  We pass
+ * that info through gallium in the pipe_rasterizer_state::sprite_coord_mode
+ * array.
+ *
+ * Additionally, GLSL's gl_PointCoord fragment attribute has to be handled
+ * here as well.  This is basically an additional texture/generic attribute
+ * that varies .x from 0 to 1 horizontally across the point and varies .y
+ * vertically from 0 to 1 down the sprite.
+ *
+ * With geometry shaders, the state tracker could create a GS to do
+ * most/all of this.
+ */
+
+
 #include "util/u_math.h"
 #include "util/u_memory.h"
 #include "pipe/p_defines.h"
@@ -52,7 +76,7 @@ struct widepoint_stage {
 
    int psize_slot;
 
-   int point_coord_fs_input;  /**< input for pointcoord (and fog) */
+   int point_coord_fs_input;  /**< input for pointcoord */
 };
 
 
@@ -64,8 +88,6 @@ widepoint_stage( struct draw_stage *stage )
 }
 
 
-
-
 /**
  * Set the vertex texcoords for sprite mode.
  * Coords may be left untouched or set to a right-side-up or upside-down
@@ -89,10 +111,12 @@ static void set_texcoords(const struct widepoint_stage *wide,
    }
 
    if (wide->point_coord_fs_input >= 0) {
-      /* put gl_PointCoord into extra vertex output's zw components */
-      uint k = wide->stage.draw->extra_vp_outputs.slot;
-      v->data[k][2] = tc[0];
-      v->data[k][3] = tc[1];
+      /* put gl_PointCoord into the extra vertex slot */
+      uint slot = wide->stage.draw->extra_vp_outputs.slot;
+      v->data[slot][0] = tc[0];
+      v->data[slot][1] = tc[1];
+      v->data[slot][2] = 0.0F;
+      v->data[slot][3] = 1.0F;
    }
 }
 
@@ -182,10 +206,10 @@ static void widepoint_point( struct draw_stage *stage,
 
 
 static int
-find_fog_input_attrib(struct draw_context *draw)
+find_pntc_input_attrib(struct draw_context *draw)
 {
-   /* Scan the fragment program's input decls to find the fogcoord
-    * attribute.  The z/w components will store the point coord.
+   /* Scan the fragment program's input decls to find the pointcoord
+    * attribute.  The xy components will store the point coord.
     */
    return 0; /* XXX fix this */
 }
@@ -229,8 +253,8 @@ static void widepoint_first_point( struct draw_stage *stage,
       }
       wide->num_texcoords = j;
 
-      /* find fragment shader PointCoord/Fog input */
-      wide->point_coord_fs_input = find_fog_input_attrib(draw);
+      /* find fragment shader PointCoord input */
+      wide->point_coord_fs_input = find_pntc_input_attrib(draw);
 
       /* setup extra vp output (point coord implemented as a texcoord) */
       draw->extra_vp_outputs.semantic_name = TGSI_SEMANTIC_GENERIC;
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c
index 9e37a26c1e..62e04a65f3 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos.c
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.c
@@ -1758,24 +1758,24 @@ emit_instruction( struct aos_compilation *cp,
    case TGSI_OPCODE_SUB:
       return emit_SUB(cp, inst);
  
-   case TGSI_OPCODE_LERP:
+   case TGSI_OPCODE_LRP:
 //      return emit_LERP(cp, inst);
       return FALSE;
 
-   case TGSI_OPCODE_FRAC:
+   case TGSI_OPCODE_FRC:
       return emit_FRC(cp, inst);
 
    case TGSI_OPCODE_CLAMP:
 //      return emit_CLAMP(cp, inst);
       return FALSE;
 
-   case TGSI_OPCODE_FLOOR:
+   case TGSI_OPCODE_FLR:
       return emit_FLR(cp, inst);
 
    case TGSI_OPCODE_ROUND:
       return emit_RND(cp, inst);
 
-   case TGSI_OPCODE_EXPBASE2:
+   case TGSI_OPCODE_EX2:
 #if FAST_MATH
       return emit_EXPBASE2(cp, inst);
 #elif 0
@@ -1787,13 +1787,13 @@ emit_instruction( struct aos_compilation *cp,
       return FALSE;
 #endif
 
-   case TGSI_OPCODE_LOGBASE2:
+   case TGSI_OPCODE_LG2:
       return emit_LG2(cp, inst);
 
-   case TGSI_OPCODE_POWER:
+   case TGSI_OPCODE_POW:
       return emit_POW(cp, inst);
 
-   case TGSI_OPCODE_CROSSPRODUCT:
+   case TGSI_OPCODE_XPD:
       return emit_XPD(cp, inst);
 
    case TGSI_OPCODE_ABS:
@@ -1891,8 +1891,9 @@ static boolean note_immediate( struct aos_compilation *cp,
    unsigned pos = cp->num_immediates++;
    unsigned j;
 
+   assert( imm->Immediate.NrTokens <= 4 + 1 );
    for (j = 0; j < imm->Immediate.NrTokens - 1; j++) {
-      cp->vaos->machine->immediate[pos][j] = imm->u.ImmediateFloat32[j].Float;
+      cp->vaos->machine->immediate[pos][j] = imm->u[j].Float;
    }
 
    return TRUE;
diff --git a/src/gallium/auxiliary/draw/draw_vs_llvm.c b/src/gallium/auxiliary/draw/draw_vs_llvm.c
index 727977bc3a..b3535c0e48 100644
--- a/src/gallium/auxiliary/draw/draw_vs_llvm.c
+++ b/src/gallium/auxiliary/draw/draw_vs_llvm.c
@@ -119,7 +119,7 @@ draw_create_vs_llvm(struct draw_context *draw,
    vs->base.create_varient = draw_vs_varient_generic;
    vs->base.run_linear = vs_llvm_run_linear;
    vs->base.delete = vs_llvm_delete;
-   vs->machine = &draw->vs.machine;
+   vs->machine = draw->vs.machine;
 
    {
       struct gallivm_ir *ir = gallivm_ir_new(GALLIVM_VS);