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

6 files changed, 99 insertions, 8 deletions

diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.c b/src/gallium/auxiliary/tgsi/tgsi_exec.c
index 7892a67f04c..35b27423513 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_exec.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_exec.c
@@ -1038,7 +1038,6 @@ fetch_src_file_channel(const struct tgsi_exec_machine *mach,
       break;
 
    case TGSI_FILE_INPUT:
-   case TGSI_FILE_SYSTEM_VALUE:
       for (i = 0; i < QUAD_SIZE; i++) {
          /*
          if (TGSI_PROCESSOR_GEOMETRY == mach->Processor) {
@@ -1053,6 +1052,15 @@ fetch_src_file_channel(const struct tgsi_exec_machine *mach,
       }
       break;
 
+   case TGSI_FILE_SYSTEM_VALUE:
+      /* XXX no swizzling at this point.  Will be needed if we put
+       * gl_FragCoord, for example, in a sys value register.
+       */
+      for (i = 0; i < QUAD_SIZE; i++) {
+         chan->f[i] = mach->SystemValue[index->i[i]][0];
+      }
+      break;
+
    case TGSI_FILE_TEMPORARY:
       for (i = 0; i < QUAD_SIZE; i++) {
          assert(index->i[i] < TGSI_EXEC_NUM_TEMPS);
@@ -1907,8 +1915,7 @@ exec_declaration(struct tgsi_exec_machine *mach,
                  const struct tgsi_full_declaration *decl)
 {
    if (mach->Processor == TGSI_PROCESSOR_FRAGMENT) {
-      if (decl->Declaration.File == TGSI_FILE_INPUT ||
-          decl->Declaration.File == TGSI_FILE_SYSTEM_VALUE) {
+      if (decl->Declaration.File == TGSI_FILE_INPUT) {
          uint first, last, mask;
 
          first = decl->Range.First;
@@ -1921,6 +1928,7 @@ exec_declaration(struct tgsi_exec_machine *mach,
           * ureg code to emit the right UsageMask value (WRITEMASK_X).
           * Then, we could remove the tgsi_exec_machine::Face field.
           */
+         /* XXX make FACE a system value */
          if (decl->Semantic.Name == TGSI_SEMANTIC_FACE) {
             uint i;
 
@@ -1962,8 +1970,13 @@ exec_declaration(struct tgsi_exec_machine *mach,
          }
       }
    }
+
+   if (decl->Declaration.File == TGSI_FILE_SYSTEM_VALUE) {
+      mach->SysSemanticToIndex[decl->Declaration.Semantic] = decl->Range.First;
+   }
 }
 
+
 typedef void (* micro_op)(union tgsi_exec_channel *dst);
 
 static void
diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.h b/src/gallium/auxiliary/tgsi/tgsi_exec.h
index b5ebbfbfaab..6c204c73714 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_exec.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_exec.h
@@ -31,6 +31,7 @@
 
 #include "pipe/p_compiler.h"
 #include "pipe/p_state.h"
+#include "pipe/p_shader_tokens.h"
 
 #if defined __cplusplus
 extern "C" {
@@ -181,6 +182,8 @@ struct tgsi_sampler
 /* The maximum total number of vertices */
 #define TGSI_MAX_TOTAL_VERTICES (TGSI_MAX_PRIM_VERTICES * TGSI_MAX_PRIMITIVES * PIPE_MAX_ATTRIBS)
 
+#define TGSI_MAX_MISC_INPUTS 8
+
 /** function call/activation record */
 struct tgsi_call_record
 {
@@ -228,6 +231,10 @@ struct tgsi_exec_machine
    struct tgsi_exec_vector       Inputs[TGSI_MAX_PRIM_VERTICES * PIPE_MAX_ATTRIBS];
    struct tgsi_exec_vector       Outputs[TGSI_MAX_TOTAL_VERTICES];
 
+   /* System values */
+   unsigned                      SysSemanticToIndex[TGSI_SEMANTIC_COUNT];
+   float                         SystemValue[TGSI_MAX_MISC_INPUTS][4];
+
    struct tgsi_exec_vector       *Addrs;
    struct tgsi_exec_vector       *Predicates;
 
diff --git a/src/gallium/auxiliary/tgsi/tgsi_ppc.c b/src/gallium/auxiliary/tgsi/tgsi_ppc.c
index 3521847b619..537a0f6c5e4 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_ppc.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_ppc.c
@@ -294,7 +294,6 @@ emit_fetch(struct gen_context *gen,
    case TGSI_SWIZZLE_W:
       switch (reg->Register.File) {
       case TGSI_FILE_INPUT:
-      case TGSI_FILE_SYSTEM_VALUE:
          {
             int offset = (reg->Register.Index * 4 + swizzle) * 16;
             int offset_reg = emit_li_offset(gen, offset);
@@ -302,6 +301,9 @@ emit_fetch(struct gen_context *gen,
             ppc_lvx(gen->f, dst_vec, gen->inputs_reg, offset_reg);
          }
          break;
+      case TGSI_FILE_SYSTEM_VALUE:
+         assert(!"unhandled system value in tgsi_ppc.c");
+         break;
       case TGSI_FILE_TEMPORARY:
          if (is_ppc_vec_temporary(reg)) {
             /* use PPC vec register */
diff --git a/src/gallium/auxiliary/tgsi/tgsi_scan.c b/src/gallium/auxiliary/tgsi/tgsi_scan.c
index 6585da3e838..83c6ac75e54 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_scan.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_scan.c
@@ -143,7 +143,7 @@ tgsi_scan_shader(const struct tgsi_token *tokens,
                info->file_count[file]++;
                info->file_max[file] = MAX2(info->file_max[file], (int)reg);
 
-               if (file == TGSI_FILE_INPUT || file == TGSI_FILE_SYSTEM_VALUE) {
+               if (file == TGSI_FILE_INPUT) {
                   info->input_semantic_name[reg] = (ubyte)fulldecl->Semantic.Name;
                   info->input_semantic_index[reg] = (ubyte)fulldecl->Semantic.Index;
                   info->input_interpolate[reg] = (ubyte)fulldecl->Declaration.Interpolate;
@@ -151,6 +151,23 @@ tgsi_scan_shader(const struct tgsi_token *tokens,
                   info->input_cylindrical_wrap[reg] = (ubyte)fulldecl->Declaration.CylindricalWrap;
                   info->num_inputs++;
                }
+               else if (file == TGSI_FILE_SYSTEM_VALUE) {
+                  unsigned index = fulldecl->Range.First;
+                  unsigned semName = fulldecl->Semantic.Name;
+
+                  info->system_value_semantic_name[index] = semName;
+                  info->num_system_values = MAX2(info->num_system_values,
+                                                 index + 1);
+
+                  /*
+                  info->system_value_semantic_name[info->num_system_values++] = 
+                     fulldecl->Semantic.Name;
+                  */
+
+                  if (fulldecl->Semantic.Name == TGSI_SEMANTIC_INSTANCEID) {
+                     info->uses_instanceid = TRUE;
+                  }
+               }
                else if (file == TGSI_FILE_OUTPUT) {
                   info->output_semantic_name[reg] = (ubyte)fulldecl->Semantic.Name;
                   info->output_semantic_index[reg] = (ubyte)fulldecl->Semantic.Index;
diff --git a/src/gallium/auxiliary/tgsi/tgsi_scan.h b/src/gallium/auxiliary/tgsi/tgsi_scan.h
index 104097fbc03..53ab3d509dd 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_scan.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_scan.h
@@ -51,6 +51,9 @@ struct tgsi_shader_info
    ubyte output_semantic_name[PIPE_MAX_SHADER_OUTPUTS]; /**< TGSI_SEMANTIC_x */
    ubyte output_semantic_index[PIPE_MAX_SHADER_OUTPUTS];
 
+   ubyte num_system_values;
+   ubyte system_value_semantic_name[PIPE_MAX_SHADER_INPUTS];
+
    uint file_mask[TGSI_FILE_COUNT];  /**< bitmask of declared registers */
    uint file_count[TGSI_FILE_COUNT];  /**< number of declared registers */
    int file_max[TGSI_FILE_COUNT];  /**< highest index of declared registers */
@@ -64,6 +67,7 @@ struct tgsi_shader_info
    boolean writes_stencil; /**< does fragment shader write stencil value? */
    boolean writes_edgeflag; /**< vertex shader outputs edgeflag */
    boolean uses_kill;  /**< KIL or KILP instruction used? */
+   boolean uses_instanceid;
 
    /**
     * Bitmask indicating which register files are accessed with
diff --git a/src/gallium/auxiliary/tgsi/tgsi_sse2.c b/src/gallium/auxiliary/tgsi/tgsi_sse2.c
index 086d983a73a..3f2cda860e0 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_sse2.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_sse2.c
@@ -163,6 +163,14 @@ get_immediate_base( void )
       reg_DX );
 }
 
+static struct x86_reg
+get_system_value_base( void )
+{
+   return x86_make_disp(
+      get_machine_base(),
+      Offset(struct tgsi_exec_machine, SystemValue) );
+}
+
 
 /**
  * Data access helpers.
@@ -229,6 +237,16 @@ get_temp(
 }
 
 static struct x86_reg
+get_system_value(
+   unsigned vec,
+   unsigned chan )
+{
+   return x86_make_disp(
+      get_system_value_base(), /* base */
+      (vec * 4 + chan) * 4 );  /* byte offset from base */
+}
+
+static struct x86_reg
 get_coef(
    unsigned vec,
    unsigned chan,
@@ -423,6 +441,30 @@ emit_tempf(
 }
 
 /**
+ * Copy a system value to xmm register
+ * \param xmm  the destination xmm register
+ * \param vec  the source system value register
+ * \param chan  src channel to fetch (X, Y, Z or W)
+ */
+static void
+emit_system_value(
+   struct x86_function *func,
+   unsigned xmm,
+   unsigned vec,
+   unsigned chan )
+{
+   sse_movss(
+      func,
+      make_xmm( xmm ),
+      get_system_value( vec, chan ) );
+   sse_shufps(
+      func,
+      make_xmm( xmm ),
+      make_xmm( xmm ),
+      SHUF( 0, 0, 0, 0 ) );
+}
+
+/**
  * Load an xmm register with an input attrib coefficient (a0, dadx or dady)
  * \param xmm  the destination xmm register
  * \param vec  the src input/attribute coefficient index
@@ -1281,8 +1323,15 @@ emit_fetch(
             swizzle );
          break;
 
-      case TGSI_FILE_INPUT:
       case TGSI_FILE_SYSTEM_VALUE:
+         emit_system_value(
+            func,
+            xmm,
+            reg->Register.Index,
+            swizzle );
+         break;
+
+      case TGSI_FILE_INPUT:
          emit_inputf(
             func,
             xmm,
@@ -2636,8 +2685,7 @@ emit_declaration(
    struct x86_function *func,
    struct tgsi_full_declaration *decl )
 {
-   if( decl->Declaration.File == TGSI_FILE_INPUT ||
-       decl->Declaration.File == TGSI_FILE_SYSTEM_VALUE ) {
+   if( decl->Declaration.File == TGSI_FILE_INPUT ) {
       unsigned first, last, mask;
       unsigned i, j;