Checkpoint: rework shader input/output register mapping.

This is a step toward removing TGSI_ATTRIB_ tokens.
Basically, when translating Mesa programs to TGSI programs, pass in input and
output register re-maps, plus interpolation info.
There's some known breakage (cubemap.c) so more to be done...

2 files changed, 32 insertions, 21 deletions

diff --git a/src/mesa/pipe/draw/draw_vertex_fetch.c b/src/mesa/pipe/draw/draw_vertex_fetch.c
index 0dbbdf17f2..ce402d681f 100644
--- a/src/mesa/pipe/draw/draw_vertex_fetch.c
+++ b/src/mesa/pipe/draw/draw_vertex_fetch.c
@@ -67,6 +67,10 @@ fetch_attrib4(const void *ptr, unsigned format, float attrib[4])
    }
 }
 
+
+/**
+ * Fetch vertex attributes for 'count' vertices.
+ */
 void draw_vertex_fetch( struct draw_context *draw,
 			struct tgsi_exec_machine *machine,
 			const unsigned *elts,
@@ -74,27 +78,26 @@ void draw_vertex_fetch( struct draw_context *draw,
 {
    unsigned j;
 
-
-   /* load machine inputs */
+   /* loop over vertices */
    for (j = 0; j < count; j++) {
-      unsigned attr;
-      for (attr = 0; attr < 16; attr++) {
-         if (draw->vertex_shader.inputs_read & (1 << attr)) {
-            unsigned buf = draw->vertex_element[attr].vertex_buffer_index;
-            const void *src
-               = (const void *) ((const ubyte *) draw->mapped_vbuffer[buf]
-                                 + draw->vertex_buffer[buf].buffer_offset
-                                 + draw->vertex_element[attr].src_offset
-                                 + elts[j] * draw->vertex_buffer[buf].pitch);
-            float p[4];
+      uint attr;
+      /* loop over vertex attributes (vertex shader inputs) */
+      for (attr = 0; attr < draw->vertex_shader.num_inputs; attr++) {
+
+         unsigned buf = draw->vertex_element[attr].vertex_buffer_index;
+         const void *src
+            = (const void *) ((const ubyte *) draw->mapped_vbuffer[buf]
+                              + draw->vertex_buffer[buf].buffer_offset
+                              + draw->vertex_element[attr].src_offset
+                              + elts[j] * draw->vertex_buffer[buf].pitch);
+         float p[4];
 
-            fetch_attrib4(src, draw->vertex_element[attr].src_format, p);
+         fetch_attrib4(src, draw->vertex_element[attr].src_format, p);
 
-            machine->Inputs[attr].xyzw[0].f[j] = p[0]; /*X*/
-            machine->Inputs[attr].xyzw[1].f[j] = p[1]; /*Y*/
-            machine->Inputs[attr].xyzw[2].f[j] = p[2]; /*Z*/
-            machine->Inputs[attr].xyzw[3].f[j] = p[3]; /*W*/
-         }
+         machine->Inputs[attr].xyzw[0].f[j] = p[0]; /*X*/
+         machine->Inputs[attr].xyzw[1].f[j] = p[1]; /*Y*/
+         machine->Inputs[attr].xyzw[2].f[j] = p[2]; /*Z*/
+         machine->Inputs[attr].xyzw[3].f[j] = p[3]; /*W*/
       }
    }
 }
diff --git a/src/mesa/pipe/draw/draw_vertex_shader.c b/src/mesa/pipe/draw/draw_vertex_shader.c
index 8effc74cbe..cb6c605b8d 100644
--- a/src/mesa/pipe/draw/draw_vertex_shader.c
+++ b/src/mesa/pipe/draw/draw_vertex_shader.c
@@ -114,7 +114,6 @@ run_vertex_program(struct draw_context *draw,
 
    draw_vertex_fetch( draw, &machine, elts, count );
 
-
    /* run shader */
    if( draw->vertex_shader.executable != NULL ) {
 #if defined(USE_X86_ASM) || defined(SLANG_X86)
@@ -159,14 +158,23 @@ run_vertex_program(struct draw_context *draw,
       vOut[j]->data[0][2] = z * scale[2] + trans[2];
       vOut[j]->data[0][3] = w;
 
-      /* remaining attributes are packed into sequential post-transform
+      /* Remaining attributes are packed into sequential post-transform
        * vertex attrib slots.
+       * Skip 0 since we just did it above.
+       * Subtract two because of the VERTEX_HEADER, CLIP_POS attribs.
        */
-      for (slot = 1; slot < draw->vertex_info.num_attribs; slot++) {
+      for (slot = 1; slot < draw->vertex_info.num_attribs - 2; slot++) {
          vOut[j]->data[slot][0] = machine.Outputs[slot].xyzw[0].f[j];
          vOut[j]->data[slot][1] = machine.Outputs[slot].xyzw[1].f[j];
          vOut[j]->data[slot][2] = machine.Outputs[slot].xyzw[2].f[j];
          vOut[j]->data[slot][3] = machine.Outputs[slot].xyzw[3].f[j];
+         /*
+         printf("output %d: %f %f %f %f\n", slot,
+                vOut[j]->data[slot][0],
+                vOut[j]->data[slot][1],
+                vOut[j]->data[slot][2],
+                vOut[j]->data[slot][3]);
+         */
       }
    } /* loop over vertices */
 }