Implement instanced indexed draw.

6 files changed, 26 insertions, 12 deletions

diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch.c b/src/gallium/auxiliary/draw/draw_pt_fetch.c
index e8174a29711..f88a839f61e 100644
--- a/src/gallium/auxiliary/draw/draw_pt_fetch.c
+++ b/src/gallium/auxiliary/draw/draw_pt_fetch.c
@@ -160,6 +160,7 @@ void draw_pt_fetch_run( struct pt_fetch *fetch,
    translate->run_elts( translate,
 			elts, 
 			count,
+                        draw->instance_id,
 			verts );
 
 }
diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c b/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c
index 40bfc0fbb22..771d94b973c 100644
--- a/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c
+++ b/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c
@@ -257,6 +257,7 @@ static void fetch_emit_run( struct draw_pt_middle_end *middle,
    feme->translate->run_elts( feme->translate, 
 			      fetch_elts,
 			      fetch_count,
+                              draw->instance_id,
 			      hw_verts );
 
    if (0) {
diff --git a/src/gallium/auxiliary/draw/draw_vs_varient.c b/src/gallium/auxiliary/draw/draw_vs_varient.c
index 4cc080f8039..8e14bdd6bde 100644
--- a/src/gallium/auxiliary/draw/draw_vs_varient.c
+++ b/src/gallium/auxiliary/draw/draw_vs_varient.c
@@ -142,6 +142,7 @@ static void PIPE_CDECL vsvg_run_elts( struct draw_vs_varient *varient,
    vsvg->fetch->run_elts( vsvg->fetch, 
                           elts,
                           count,
+                          vsvg->draw->instance_id,
                           temp_buffer );
 
    vsvg->base.vs->run_linear( vsvg->base.vs, 
diff --git a/src/gallium/auxiliary/translate/translate.h b/src/gallium/auxiliary/translate/translate.h
index fb298471b8b..9ae7a482a07 100644
--- a/src/gallium/auxiliary/translate/translate.h
+++ b/src/gallium/auxiliary/translate/translate.h
@@ -75,6 +75,7 @@ struct translate {
    void (PIPE_CDECL *run_elts)( struct translate *,
                                 const unsigned *elts,
                                 unsigned count,
+                                unsigned instance_id,
                                 void *output_buffer);
 
    void (PIPE_CDECL *run)( struct translate *,
diff --git a/src/gallium/auxiliary/translate/translate_generic.c b/src/gallium/auxiliary/translate/translate_generic.c
index 0fa99274099..742f03b5031 100644
--- a/src/gallium/auxiliary/translate/translate_generic.c
+++ b/src/gallium/auxiliary/translate/translate_generic.c
@@ -569,6 +569,7 @@ static emit_func get_emit_func( enum pipe_format format )
 static void PIPE_CDECL generic_run_elts( struct translate *translate,
                                          const unsigned *elts,
                                          unsigned count,
+                                         unsigned instance_id,
                                          void *output_buffer )
 {
    struct translate_generic *tg = translate_generic(translate);
@@ -584,13 +585,20 @@ static void PIPE_CDECL generic_run_elts( struct translate *translate,
 
       for (attr = 0; attr < nr_attrs; attr++) {
 	 float data[4];
-
-	 const char *src = (tg->attrib[attr].input_ptr + 
-			    tg->attrib[attr].input_stride * elt);
+         const char *src;
 
 	 char *dst = (vert + 
 		      tg->attrib[attr].output_offset);
 
+         if (tg->attrib[attr].instance_divisor) {
+            src = tg->attrib[attr].input_ptr +
+                  tg->attrib[attr].input_stride *
+                  (instance_id / tg->attrib[attr].instance_divisor);
+         } else {
+            src = tg->attrib[attr].input_ptr +
+                  tg->attrib[attr].input_stride * elt;
+         }
+
 	 tg->attrib[attr].fetch( src, data );
 
          if (0) debug_printf("vert %d/%d attr %d: %f %f %f %f\n",
diff --git a/src/gallium/auxiliary/translate/translate_sse.c b/src/gallium/auxiliary/translate/translate_sse.c
index ffc5fe9e818..ba4a246fdbf 100644
--- a/src/gallium/auxiliary/translate/translate_sse.c
+++ b/src/gallium/auxiliary/translate/translate_sse.c
@@ -376,13 +376,14 @@ static boolean init_inputs( struct translate_sse *p,
                             boolean linear )
 {
    unsigned i;
-   if (linear) {
-      struct x86_reg instance_id = x86_make_disp(p->machine_EDX,
-                                                 get_offset(p, &p->instance_id));
+   struct x86_reg instance_id = x86_make_disp(p->machine_EDX,
+                                              get_offset(p, &p->instance_id));
 
-      for (i = 0; i < p->nr_buffer_varients; i++) {
-         struct translate_buffer_varient *varient = &p->buffer_varient[i];
-         struct translate_buffer *buffer = &p->buffer[varient->buffer_index];
+   for (i = 0; i < p->nr_buffer_varients; i++) {
+      struct translate_buffer_varient *varient = &p->buffer_varient[i];
+      struct translate_buffer *buffer = &p->buffer[varient->buffer_index];
+
+      if (linear || varient->instance_divisor) {
          struct x86_reg buf_stride   = x86_make_disp(p->machine_EDX,
                                                      get_offset(p, &buffer->stride));
          struct x86_reg buf_ptr      = x86_make_disp(p->machine_EDX,
@@ -426,7 +427,7 @@ static boolean init_inputs( struct translate_sse *p,
          /* In the linear case, keep the buffer pointer instead of the
           * index number.
           */
-         if (p->nr_buffer_varients == 1) 
+         if (linear && p->nr_buffer_varients == 1)
             x86_mov(p->func, elt, tmp_EAX);
          else
             x86_mov(p->func, buf_ptr, tmp_EAX);
@@ -445,7 +446,7 @@ static struct x86_reg get_buffer_ptr( struct translate_sse *p,
    if (linear && p->nr_buffer_varients == 1) {
       return p->idx_EBX;
    }
-   else if (linear) {
+   else if (linear || p->buffer_varient[var_idx].instance_divisor) {
       struct x86_reg ptr = p->tmp_EAX;
       struct x86_reg buf_ptr = 
          x86_make_disp(p->machine_EDX, 
@@ -687,6 +688,7 @@ static void translate_sse_release( struct translate *translate )
 static void PIPE_CDECL translate_sse_run_elts( struct translate *translate,
 			      const unsigned *elts,
 			      unsigned count,
+                              unsigned instance_id,
 			      void *output_buffer )
 {
    struct translate_sse *p = (struct translate_sse *)translate;
@@ -694,7 +696,7 @@ static void PIPE_CDECL translate_sse_run_elts( struct translate *translate,
    p->gen_run_elts( translate,
 		    elts,
 		    count,
-                    0,
+                    instance_id,
 		    output_buffer );
 }