draw: fix vertex id computation

vertex id has to be unaffected by the start index (i.e. when calling
draw arrays with start_index = 5, the first vertex_id has to still
be 0, not 5) and it has to be equal to the index when performing
indexed rendering (in which case it has to be unaffected by the
index bias). This fixes our behavior.

Signed-off-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>

5 files changed, 37 insertions, 13 deletions

diff --git a/src/gallium/auxiliary/draw/draw_llvm.c b/src/gallium/auxiliary/draw/draw_llvm.c
index a3174b404c1..c195a2b4def 100644
--- a/src/gallium/auxiliary/draw/draw_llvm.c
+++ b/src/gallium/auxiliary/draw/draw_llvm.c
@@ -1486,7 +1486,7 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant,
    struct gallivm_state *gallivm = variant->gallivm;
    LLVMContextRef context = gallivm->context;
    LLVMTypeRef int32_type = LLVMInt32TypeInContext(context);
-   LLVMTypeRef arg_types[9];
+   LLVMTypeRef arg_types[10];
    unsigned num_arg_types =
       elts ? Elements(arg_types) : Elements(arg_types) - 1;
    LLVMTypeRef func_type;
@@ -1496,6 +1496,7 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant,
    struct lp_type vs_type;
    LLVMValueRef end, start;
    LLVMValueRef count, fetch_elts, fetch_elt_max, fetch_count;
+   LLVMValueRef vertex_id_offset;
    LLVMValueRef stride, step, io_itr;
    LLVMValueRef io_ptr, vbuffers_ptr, vb_ptr;
    LLVMValueRef zero = lp_build_const_int32(gallivm, 0);
@@ -1541,6 +1542,7 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant,
    arg_types[i++] = int32_type;                     /* stride */
    arg_types[i++] = get_vb_ptr_type(variant);       /* pipe_vertex_buffer's */
    arg_types[i++] = int32_type;                     /* instance_id */
+   arg_types[i++] = int32_type;                     /* vertex_id_offset */
 
    func_type = LLVMFunctionType(int32_type, arg_types, num_arg_types, 0);
 
@@ -1565,6 +1567,7 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant,
    stride                    = LLVMGetParam(variant_func, 5 + (elts ? 1 : 0));
    vb_ptr                    = LLVMGetParam(variant_func, 6 + (elts ? 1 : 0));
    system_values.instance_id = LLVMGetParam(variant_func, 7 + (elts ? 1 : 0));
+   vertex_id_offset          = LLVMGetParam(variant_func, 8 + (elts ? 1 : 0));
 
    lp_build_name(context_ptr, "context");
    lp_build_name(io_ptr, "io");
@@ -1572,6 +1575,7 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant,
    lp_build_name(stride, "stride");
    lp_build_name(vb_ptr, "vb");
    lp_build_name(system_values.instance_id, "instance_id");
+   lp_build_name(vertex_id_offset, "vertex_id_offset");
 
    if (elts) {
       fetch_elts    = LLVMGetParam(variant_func, 3);
@@ -1646,22 +1650,19 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant,
 #endif
       system_values.vertex_id = lp_build_zero(gallivm, lp_type_uint_vec(32, 32*vector_length));
       for (i = 0; i < vector_length; ++i) {
-         LLVMValueRef true_index =
+         LLVMValueRef vert_index =
             LLVMBuildAdd(builder,
                          lp_loop.counter,
                          lp_build_const_int32(gallivm, i), "");
-         true_index = LLVMBuildAdd(builder, start, true_index, "");
+         LLVMValueRef true_index =
+            LLVMBuildAdd(builder, start, vert_index, "");
+         LLVMValueRef vertex_id;
 
          /* make sure we're not out of bounds which can happen
           * if fetch_count % 4 != 0, because on the last iteration
           * a few of the 4 vertex fetches will be out of bounds */
          true_index = lp_build_min(&bld, true_index, fetch_max);
 
-         system_values.vertex_id = LLVMBuildInsertElement(
-            gallivm->builder,
-            system_values.vertex_id, true_index,
-            lp_build_const_int32(gallivm, i), "");
-
          if (elts) {
             LLVMValueRef fetch_ptr;
             LLVMValueRef index_overflowed;
@@ -1673,7 +1674,7 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant,
             index_overflowed = LLVMBuildICmp(builder, LLVMIntUGT,
                                              true_index, fetch_elt_max,
                                              "index_overflowed");
-            
+
             lp_build_if(&if_ctx, gallivm, index_overflowed);
             {
                /* Generate maximum possible index so that
@@ -1698,6 +1699,23 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant,
             lp_build_endif(&if_ctx);
             true_index = LLVMBuildLoad(builder, index_ptr, "true_index");
          }
+         /* in the paths with elts vertex id has to be unaffected by the
+          * index bias and because indices inside our elements array have
+          * already had index bias applied we need to subtract it here to
+          * get back to the original index.
+          * in the linear paths vertex id has to be unaffected by the
+          * original start index and because we abuse the 'start' variable
+          * to either represent the actual start index or the index at which
+          * the primitive was split (we split rendering into chunks of at
+          * most 4095-vertices) we need to back out the original start
+          * index out of our vertex id here.
+          */
+         vertex_id = LLVMBuildSub(builder, true_index, vertex_id_offset, "");
+
+         system_values.vertex_id = LLVMBuildInsertElement(
+            gallivm->builder,
+            system_values.vertex_id, vertex_id,
+            lp_build_const_int32(gallivm, i), "");
 
          for (j = 0; j < draw->pt.nr_vertex_elements; ++j) {
             struct pipe_vertex_element *velem = &draw->pt.vertex_element[j];
diff --git a/src/gallium/auxiliary/draw/draw_llvm.h b/src/gallium/auxiliary/draw/draw_llvm.h
index 347fde2b55f..0675e3bf17b 100644
--- a/src/gallium/auxiliary/draw/draw_llvm.h
+++ b/src/gallium/auxiliary/draw/draw_llvm.h
@@ -263,7 +263,8 @@ typedef int
                       unsigned count,
                       unsigned stride,
                       struct pipe_vertex_buffer *vertex_buffers,
-                      unsigned instance_id);
+                      unsigned instance_id,
+                      unsigned vertex_id_offset);
 
 
 typedef int
@@ -275,7 +276,8 @@ typedef int
                            unsigned fetch_count,
                            unsigned stride,
                            struct pipe_vertex_buffer *vertex_buffers,
-                           unsigned instance_id);
+                           unsigned instance_id,
+                           unsigned vertex_id_offset);
 
 
 typedef int
diff --git a/src/gallium/auxiliary/draw/draw_private.h b/src/gallium/auxiliary/draw/draw_private.h
index d8cd8ebdb64..868b6c78954 100644
--- a/src/gallium/auxiliary/draw/draw_private.h
+++ b/src/gallium/auxiliary/draw/draw_private.h
@@ -311,6 +311,7 @@ struct draw_context
 
    unsigned instance_id;
    unsigned start_instance;
+   unsigned start_index;
 
 #ifdef HAVE_LLVM
    struct draw_llvm *llvm;
diff --git a/src/gallium/auxiliary/draw/draw_pt.c b/src/gallium/auxiliary/draw/draw_pt.c
index fcc24057db0..5b16bc72ef4 100644
--- a/src/gallium/auxiliary/draw/draw_pt.c
+++ b/src/gallium/auxiliary/draw/draw_pt.c
@@ -535,6 +535,7 @@ draw_vbo(struct draw_context *draw,
    }
 
    draw->pt.max_index = index_limit - 1;
+   draw->start_index = info->start;
 
    /*
     * TODO: We could use draw->pt.max_index to further narrow
diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline_llvm.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline_llvm.c
index 2e47fad551a..17b948a7652 100644
--- a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline_llvm.c
+++ b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline_llvm.c
@@ -353,7 +353,8 @@ llvm_pipeline_generic( struct draw_pt_middle_end *middle,
                                        fetch_info->count,
                                        fpme->vertex_size,
                                        draw->pt.vertex_buffer,
-                                       draw->instance_id);
+                                       draw->instance_id,
+                                       draw->start_index);
    else
       clipped = fpme->current_variant->jit_func_elts( &fpme->llvm->jit_context,
                                             llvm_vert_info.verts,
@@ -363,7 +364,8 @@ llvm_pipeline_generic( struct draw_pt_middle_end *middle,
                                             fetch_info->count,
                                             fpme->vertex_size,
                                             draw->pt.vertex_buffer,
-                                            draw->instance_id);
+                                            draw->instance_id,
+                                            draw->pt.user.eltBias);
 
    /* Finished with fetch and vs:
     */