Merge branch 'draw-instanced'

Conflicts:
	src/gallium/auxiliary/draw/draw_llvm.c
	src/gallium/drivers/llvmpipe/lp_state_fs.c
	src/glsl/ir_set_program_inouts.cpp
	src/mesa/tnl/t_vb_program.c

11 files changed, 80 insertions, 9 deletions

diff --git a/src/mesa/main/extensions.c b/src/mesa/main/extensions.c
index c41ae4ad2d..24404993c6 100644
--- a/src/mesa/main/extensions.c
+++ b/src/mesa/main/extensions.c
@@ -380,6 +380,7 @@ _mesa_enable_sw_extensions(struct gl_context *ctx)
    ctx->Extensions.ARB_depth_texture = GL_TRUE;
    /*ctx->Extensions.ARB_draw_buffers = GL_TRUE;*/
    ctx->Extensions.ARB_draw_elements_base_vertex = GL_TRUE;
+   ctx->Extensions.ARB_draw_instanced = GL_TRUE;
    ctx->Extensions.ARB_explicit_attrib_location = GL_TRUE;
    ctx->Extensions.ARB_fragment_coord_conventions = GL_TRUE;
 #if FEATURE_ARB_fragment_program
diff --git a/src/mesa/main/mtypes.h b/src/mesa/main/mtypes.h
index f76e1eb2ee..1522332e8a 100644
--- a/src/mesa/main/mtypes.h
+++ b/src/mesa/main/mtypes.h
@@ -1748,11 +1748,24 @@ typedef enum
    PROGRAM_WRITE_ONLY,  /**< A dummy, write-only register */
    PROGRAM_ADDRESS,     /**< machine->AddressReg */
    PROGRAM_SAMPLER,     /**< for shader samplers, compile-time only */
+   PROGRAM_SYSTEM_VALUE,/**< InstanceId, PrimitiveID, etc. */
    PROGRAM_UNDEFINED,   /**< Invalid/TBD value */
    PROGRAM_FILE_MAX
 } gl_register_file;
 
 
+/**
+ * If the register file is PROGRAM_SYSTEM_VALUE, the register index will be
+ * one of these values.
+ */
+typedef enum
+{
+   SYSTEM_VALUE_FRONT_FACE,  /**< Fragment shader only (not done yet) */
+   SYSTEM_VALUE_INSTANCE_ID, /**< Vertex shader only */
+   SYSTEM_VALUE_MAX          /**< Number of values */
+} gl_system_value;
+
+
 /** Vertex and fragment instructions */
 struct prog_instruction;
 struct gl_program_parameter_list;
@@ -1775,6 +1788,7 @@ struct gl_program
 
    GLbitfield InputsRead;     /**< Bitmask of which input regs are read */
    GLbitfield64 OutputsWritten; /**< Bitmask of which output regs are written */
+   GLbitfield SystemValuesRead;   /**< Bitmask of SYSTEM_VALUE_x inputs used */
    GLbitfield InputFlags[MAX_PROGRAM_INPUTS];   /**< PROG_PARAM_BIT_x flags */
    GLbitfield OutputFlags[MAX_PROGRAM_OUTPUTS]; /**< PROG_PARAM_BIT_x flags */
    GLbitfield TexturesUsed[MAX_TEXTURE_UNITS];  /**< TEXTURE_x_BIT bitmask */
diff --git a/src/mesa/program/ir_to_mesa.cpp b/src/mesa/program/ir_to_mesa.cpp
index 0929b95950..786fdfbd23 100644
--- a/src/mesa/program/ir_to_mesa.cpp
+++ b/src/mesa/program/ir_to_mesa.cpp
@@ -1462,6 +1462,7 @@ ir_to_mesa_visitor::visit(ir_dereference_variable *ir)
       case ir_var_in:
       case ir_var_out:
       case ir_var_inout:
+      case ir_var_system_value:
 	 /* The linker assigns locations for varyings and attributes,
 	  * including deprecated builtins (like gl_Color), user-assign
 	  * generic attributes (glBindVertexLocation), and
@@ -1484,6 +1485,10 @@ ir_to_mesa_visitor::visit(ir_dereference_variable *ir)
 				   ir->var->type->gl_type,
 				   ir->var->location - VERT_ATTRIB_GENERIC0);
 	    }
+         } else if (ir->var->mode == ir_var_system_value) {
+	    entry = new(mem_ctx) variable_storage(ir->var,
+						  PROGRAM_SYSTEM_VALUE,
+						  ir->var->location);
 	 } else {
 	    entry = new(mem_ctx) variable_storage(ir->var,
 						  PROGRAM_OUTPUT,
diff --git a/src/mesa/program/prog_execute.c b/src/mesa/program/prog_execute.c
index dd15e9a1cc..e7553c69db 100644
--- a/src/mesa/program/prog_execute.c
+++ b/src/mesa/program/prog_execute.c
@@ -159,6 +159,10 @@ get_src_register_pointer(const struct prog_src_register *source,
          return ZeroVec;
       return prog->Parameters->ParameterValues[reg];
 
+   case PROGRAM_SYSTEM_VALUE:
+      assert(reg < Elements(machine->SystemValues));
+      return machine->SystemValues[reg];
+
    default:
       _mesa_problem(NULL,
          "Invalid src register file %d in get_src_register_pointer()",
diff --git a/src/mesa/program/prog_execute.h b/src/mesa/program/prog_execute.h
index cefd468c36..cdf37082a0 100644
--- a/src/mesa/program/prog_execute.h
+++ b/src/mesa/program/prog_execute.h
@@ -61,6 +61,7 @@ struct gl_program_machine
    GLfloat (*EnvParams)[4]; /**< Vertex or Fragment env parameters */
    GLuint CondCodes[4];  /**< COND_* value for x/y/z/w */
    GLint AddressReg[MAX_PROGRAM_ADDRESS_REGS][4];
+   GLfloat SystemValues[SYSTEM_VALUE_MAX][4];
 
    const GLubyte *Samplers;  /** Array mapping sampler var to tex unit */
 
diff --git a/src/mesa/program/prog_print.c b/src/mesa/program/prog_print.c
index abebf392c0..484596af76 100644
--- a/src/mesa/program/prog_print.c
+++ b/src/mesa/program/prog_print.c
@@ -72,6 +72,8 @@ _mesa_register_file_name(gl_register_file f)
       return "ADDR";
    case PROGRAM_SAMPLER:
       return "SAMPLER";
+   case PROGRAM_SYSTEM_VALUE:
+      return "SYSVAL";
    case PROGRAM_UNDEFINED:
       return "UNDEFINED";
    default:
@@ -310,6 +312,9 @@ reg_string(gl_register_file f, GLint index, gl_prog_print_mode mode,
       case PROGRAM_UNIFORM: /* extension */
          sprintf(str, "uniform[%s%d]", addr, index);
          break;
+      case PROGRAM_SYSTEM_VALUE:
+         sprintf(str, "sysvalue[%s%d]", addr, index);
+         break;
       case PROGRAM_STATE_VAR:
          {
             struct gl_program_parameter *param
diff --git a/src/mesa/state_tracker/st_extensions.c b/src/mesa/state_tracker/st_extensions.c
index d240cab1a7..df4f5cf8b5 100644
--- a/src/mesa/state_tracker/st_extensions.c
+++ b/src/mesa/state_tracker/st_extensions.c
@@ -461,4 +461,8 @@ void st_init_extensions(struct st_context *st)
    if (screen->get_param(screen, PIPE_CAP_SHADER_STENCIL_EXPORT)) {
       ctx->Extensions.ARB_shader_stencil_export = GL_TRUE;
    }
+
+   if (screen->get_param(screen, PIPE_CAP_INSTANCED_DRAWING)) {
+      ctx->Extensions.ARB_draw_instanced = GL_TRUE;
+   }
 }
diff --git a/src/mesa/state_tracker/st_mesa_to_tgsi.c b/src/mesa/state_tracker/st_mesa_to_tgsi.c
index f848462310..42f1c2017f 100644
--- a/src/mesa/state_tracker/st_mesa_to_tgsi.c
+++ b/src/mesa/state_tracker/st_mesa_to_tgsi.c
@@ -72,6 +72,7 @@ struct st_translate {
    struct ureg_src inputs[PIPE_MAX_SHADER_INPUTS];
    struct ureg_dst address[1];
    struct ureg_src samplers[PIPE_MAX_SAMPLERS];
+   struct ureg_src systemValues[SYSTEM_VALUE_MAX];
 
    /* Extra info for handling point size clamping in vertex shader */
    struct ureg_dst pointSizeResult; /**< Actual point size output register */
@@ -104,6 +105,13 @@ struct st_translate {
 };
 
 
+/** Map Mesa's SYSTEM_VALUE_x to TGSI_SEMANTIC_x */
+static unsigned mesa_sysval_to_semantic[SYSTEM_VALUE_MAX] = {
+   TGSI_SEMANTIC_FACE,
+   TGSI_SEMANTIC_INSTANCEID
+};
+
+
 /**
  * Make note of a branch to a label in the TGSI code.
  * After we've emitted all instructions, we'll go over the list
@@ -245,6 +253,10 @@ src_register( struct st_translate *t,
    case PROGRAM_ADDRESS:
       return ureg_src(t->address[index]);
 
+   case PROGRAM_SYSTEM_VALUE:
+      assert(index < Elements(t->systemValues));
+      return t->systemValues[index];
+
    default:
       debug_assert( 0 );
       return ureg_src_undef();
@@ -1089,6 +1101,21 @@ st_translate_mesa_program(
       t->address[0] = ureg_DECL_address( ureg );
    }
 
+   /* Declare misc input registers
+    */
+   {
+      GLbitfield sysInputs = program->SystemValuesRead;
+      unsigned numSys = 0;
+      for (i = 0; sysInputs; i++) {
+         if (sysInputs & (1 << i)) {
+            unsigned semName = mesa_sysval_to_semantic[i];
+            t->systemValues[i] = ureg_DECL_system_value(ureg, numSys, semName, 0);
+            numSys++;
+            sysInputs &= ~(1 << i);
+         }
+      }
+   }
+
    if (program->IndirectRegisterFiles & (1 << PROGRAM_TEMPORARY)) {
       /* If temps are accessed with indirect addressing, declare temporaries
        * in sequential order.  Else, we declare them on demand elsewhere.
diff --git a/src/mesa/tnl/t_context.h b/src/mesa/tnl/t_context.h
index bc01646247..6a9444216c 100644
--- a/src/mesa/tnl/t_context.h
+++ b/src/mesa/tnl/t_context.h
@@ -527,6 +527,8 @@ typedef struct
    GLubyte *block[VERT_ATTRIB_MAX];
    GLuint nr_blocks;
 
+   GLuint CurInstance;
+
 } TNLcontext;
 
 
diff --git a/src/mesa/tnl/t_draw.c b/src/mesa/tnl/t_draw.c
index 30f1bf323c..bdb893eba2 100644
--- a/src/mesa/tnl/t_draw.c
+++ b/src/mesa/tnl/t_draw.c
@@ -453,6 +453,7 @@ void _tnl_draw_prims( struct gl_context *ctx,
        */
       struct gl_buffer_object *bo[VERT_ATTRIB_MAX + 1];
       GLuint nr_bo = 0;
+      GLuint inst;
 
       for (i = 0; i < nr_prims;) {
 	 GLuint this_nr_prims;
@@ -470,15 +471,19 @@ void _tnl_draw_prims( struct gl_context *ctx,
 	 /* Binding inputs may imply mapping some vertex buffer objects.
 	  * They will need to be unmapped below.
 	  */
-	 bind_prims(ctx, &prim[i], this_nr_prims);
-	 bind_inputs(ctx, arrays, max_index + prim[i].basevertex + 1,
-		     bo, &nr_bo);
-	 bind_indices(ctx, ib, bo, &nr_bo);
+         for (inst = 0; inst < prim[i].num_instances; inst++) {
 
-	 TNL_CONTEXT(ctx)->Driver.RunPipeline(ctx);
+            bind_prims(ctx, &prim[i], this_nr_prims);
+            bind_inputs(ctx, arrays, max_index + prim[i].basevertex + 1,
+                        bo, &nr_bo);
+            bind_indices(ctx, ib, bo, &nr_bo);
 
-	 unmap_vbos(ctx, bo, nr_bo);
-	 free_space(ctx);
+            tnl->CurInstance = inst;
+            TNL_CONTEXT(ctx)->Driver.RunPipeline(ctx);
+
+            unmap_vbos(ctx, bo, nr_bo);
+            free_space(ctx);
+         }
 
 	 i += this_nr_prims;
       }
diff --git a/src/mesa/tnl/t_vb_program.c b/src/mesa/tnl/t_vb_program.c
index 94372bbafb..f5d8f7477a 100644
--- a/src/mesa/tnl/t_vb_program.c
+++ b/src/mesa/tnl/t_vb_program.c
@@ -220,7 +220,8 @@ _tnl_program_string(struct gl_context *ctx, GLenum target, struct gl_program *pr
  * Initialize virtual machine state prior to executing vertex program.
  */
 static void
-init_machine(struct gl_context *ctx, struct gl_program_machine *machine)
+init_machine(struct gl_context *ctx, struct gl_program_machine *machine,
+             GLuint instID)
 {
    /* Input registers get initialized from the current vertex attribs */
    memcpy(machine->VertAttribs, ctx->Current.Attrib,
@@ -256,6 +257,8 @@ init_machine(struct gl_context *ctx, struct gl_program_machine *machine)
    machine->FetchTexelDeriv = NULL; /* not used by vertex programs */
 
    machine->Samplers = ctx->VertexProgram._Current->Base.SamplerUnits;
+
+   machine->SystemValues[SYSTEM_VALUE_INSTANCE_ID][0] = (GLfloat) instID;
 }
 
 
@@ -341,7 +344,7 @@ run_vp( struct gl_context *ctx, struct tnl_pipeline_stage *stage )
    for (i = 0; i < VB->Count; i++) {
       GLuint attr;
 
-      init_machine(ctx, machine);
+      init_machine(ctx, machine, tnl->CurInstance);
 
 #if 0
       printf("Input  %d: %f, %f, %f, %f\n", i,