s/Tungsten Graphics/VMware/
[mesa.git] / src / gallium / auxiliary / draw / draw_context.c
index d99f94edc43a590059333e431bac94d9b32793a5..baebc16036c028d4c7028e4f40201e8caeace163 100644 (file)
@@ -1,6 +1,6 @@
 /**************************************************************************
  * 
- * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * Copyright 2007 VMware, Inc.
  * All Rights Reserved.
  * 
  * Permission is hereby granted, free of charge, to any person obtaining a
@@ -18,7 +18,7 @@
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
- * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
@@ -27,7 +27,7 @@
 
  /*
   * Authors:
-  *   Keith Whitwell <keith@tungstengraphics.com>
+  *   Keith Whitwell <keithw@vmware.com>
   */
 
 
 #include "util/u_math.h"
 #include "util/u_cpu_detect.h"
 #include "util/u_inlines.h"
+#include "util/u_helpers.h"
+#include "util/u_prim.h"
+#include "util/u_format.h"
 #include "draw_context.h"
+#include "draw_pipe.h"
+#include "draw_prim_assembler.h"
 #include "draw_vs.h"
 #include "draw_gs.h"
 
 #if HAVE_LLVM
 #include "gallivm/lp_bld_init.h"
+#include "gallivm/lp_bld_limits.h"
 #include "draw_llvm.h"
 
-static boolean
+boolean
 draw_get_option_use_llvm(void)
 {
    static boolean first = TRUE;
@@ -55,7 +61,7 @@ draw_get_option_use_llvm(void)
 
 #ifdef PIPE_ARCH_X86
       util_cpu_detect();
-      /* require SSE2 due to LLVM PR6960. */
+      /* require SSE2 due to LLVM PR6960. XXX Might be fixed by now? */
       if (!util_cpu_caps.has_sse2)
          value = FALSE;
 #endif
@@ -65,53 +71,64 @@ draw_get_option_use_llvm(void)
 #endif
 
 
-
-/**
- * Create new draw module context.
- */
-struct draw_context *
-draw_create(struct pipe_context *pipe)
-{
-   return draw_create_gallivm(pipe, NULL);
-}
-
-
-
 /**
  * Create new draw module context with gallivm state for LLVM JIT.
  */
-struct draw_context *
-draw_create_gallivm(struct pipe_context *pipe, struct gallivm_state *gallivm)
+static struct draw_context *
+draw_create_context(struct pipe_context *pipe, boolean try_llvm)
 {
    struct draw_context *draw = CALLOC_STRUCT( draw_context );
    if (draw == NULL)
-      goto fail;
+      goto err_out;
 
-#if HAVE_LLVM
-   if (draw_get_option_use_llvm()) {
-      if (!gallivm) {
-         gallivm = gallivm_create();
-         draw->own_gallivm = gallivm;
-      }
+   /* we need correct cpu caps for disabling denorms in draw_vbo() */
+   util_cpu_detect();
 
-      if (gallivm)
-         draw->llvm = draw_llvm_create(draw, gallivm);
+#if HAVE_LLVM
+   if (try_llvm && draw_get_option_use_llvm()) {
+      draw->llvm = draw_llvm_create(draw);
+      if (!draw->llvm)
+         goto err_destroy;
    }
 #endif
 
+   draw->pipe = pipe;
+
    if (!draw_init(draw))
-      goto fail;
+      goto err_destroy;
 
-   draw->pipe = pipe;
+   draw->ia = draw_prim_assembler_create(draw);
+   if (!draw->ia)
+      goto err_destroy;
 
    return draw;
 
-fail:
+err_destroy:
    draw_destroy( draw );
+err_out:
    return NULL;
 }
 
 
+/**
+ * Create new draw module context, with LLVM JIT.
+ */
+struct draw_context *
+draw_create(struct pipe_context *pipe)
+{
+   return draw_create_context(pipe, TRUE);
+}
+
+
+/**
+ * Create a new draw context, without LLVM JIT.
+ */
+struct draw_context *
+draw_create_no_llvm(struct pipe_context *pipe)
+{
+   return draw_create_context(pipe, FALSE);
+}
+
 
 boolean draw_init(struct draw_context *draw)
 {
@@ -127,13 +144,11 @@ boolean draw_init(struct draw_context *draw)
    ASSIGN_4V( draw->plane[3],  0,  1,  0, 1 );
    ASSIGN_4V( draw->plane[4],  0,  0,  1, 1 ); /* yes these are correct */
    ASSIGN_4V( draw->plane[5],  0,  0, -1, 1 ); /* mesa's a bit wonky */
-   draw->nr_planes = 6;
-   draw->clip_xy = 1;
-   draw->clip_z = 1;
-
-
-   draw->reduced_prim = ~0; /* != any of PIPE_PRIM_x */
+   draw->clip_xy = TRUE;
+   draw->clip_z = TRUE;
 
+   draw->pt.user.planes = (float (*) [DRAW_TOTAL_CLIP_PLANES][4]) &(draw->plane[0]);
+   draw->pt.user.eltMax = ~0;
 
    if (!draw_pipeline_init( draw ))
       return FALSE;
@@ -147,14 +162,32 @@ boolean draw_init(struct draw_context *draw)
    if (!draw_gs_init( draw ))
       return FALSE;
 
+   draw->quads_always_flatshade_last = !draw->pipe->screen->get_param(
+      draw->pipe->screen, PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION);
+
+   draw->floating_point_depth = false;
+
    return TRUE;
 }
 
+/*
+ * Called whenever we're starting to draw a new instance.
+ * Some internal structures don't want to have to reset internal
+ * members on each invocation (because their state might have to persist
+ * between multiple primitive restart rendering call) but might have to 
+ * for each new instance. 
+ * This is particularly the case for primitive id's in geometry shader.
+ */
+void draw_new_instance(struct draw_context *draw)
+{
+   draw_geometry_shader_new_instance(draw->gs.geometry_shader);
+}
+
 
 void draw_destroy( struct draw_context *draw )
 {
    struct pipe_context *pipe;
-   int i, j;
+   unsigned i, j;
 
    if (!draw)
       return;
@@ -181,6 +214,7 @@ void draw_destroy( struct draw_context *draw )
       draw->render->destroy( draw->render );
    */
 
+   draw_prim_assembler_destroy(draw->ia);
    draw_pipeline_destroy( draw );
    draw_pt_destroy( draw );
    draw_vs_destroy( draw );
@@ -188,9 +222,6 @@ void draw_destroy( struct draw_context *draw )
 #ifdef HAVE_LLVM
    if (draw->llvm)
       draw_llvm_destroy( draw->llvm );
-
-   if (draw->own_gallivm)
-      gallivm_destroy(draw->own_gallivm);
 #endif
 
    FREE( draw );
@@ -205,24 +236,36 @@ void draw_flush( struct draw_context *draw )
 
 
 /**
- * Specify the Minimum Resolvable Depth factor for polygon offset.
+ * Specify the depth stencil format for the draw pipeline. This function
+ * determines the Minimum Resolvable Depth factor for polygon offset.
  * This factor potentially depends on the number of Z buffer bits,
  * the rasterization algorithm and the arithmetic performed on Z
- * values between vertex shading and rasterization.  It will vary
- * from one driver to another.
+ * values between vertex shading and rasterization.
  */
-void draw_set_mrd(struct draw_context *draw, double mrd)
+void draw_set_zs_format(struct draw_context *draw, enum pipe_format format)
 {
-   draw->mrd = mrd;
+   const struct util_format_description *desc = util_format_description(format);
+
+   draw->floating_point_depth =
+      (util_get_depth_format_type(desc) == UTIL_FORMAT_TYPE_FLOAT);
+
+   draw->mrd = util_get_depth_format_mrd(desc);
 }
 
 
 static void update_clip_flags( struct draw_context *draw )
 {
    draw->clip_xy = !draw->driver.bypass_clip_xy;
+   draw->guard_band_xy = (!draw->driver.bypass_clip_xy &&
+                          draw->driver.guard_band_xy);
    draw->clip_z = (!draw->driver.bypass_clip_z &&
-                   !draw->depth_clamp);
-   draw->clip_user = (draw->nr_planes > 6);
+                   draw->rasterizer && draw->rasterizer->depth_clip);
+   draw->clip_user = draw->rasterizer &&
+                     draw->rasterizer->clip_plane_enable != 0;
+   draw->clip_points_xy = draw->clip_xy &&
+                          (!draw->driver.bypass_clip_points ||
+                          (draw->rasterizer &&
+                          !draw->rasterizer->point_tri_clip));
 }
 
 /**
@@ -238,8 +281,8 @@ void draw_set_rasterizer_state( struct draw_context *draw,
 
       draw->rasterizer = raster;
       draw->rast_handle = rast_handle;
-
-  }
+      update_clip_flags(draw);
+   }
 }
 
 /* With a little more work, llvmpipe will be able to turn this off and
@@ -248,15 +291,23 @@ void draw_set_rasterizer_state( struct draw_context *draw,
  * Some hardware can turn off clipping altogether - in particular any
  * hardware with a TNL unit can do its own clipping, even if it is
  * relying on the draw module for some other reason.
+ * Setting bypass_clip_points to achieve d3d-style point clipping (the driver
+ * will need to do the "vp scissoring") _requires_ the driver to implement
+ * wide points / point sprites itself (points will still be clipped if rasterizer
+ * point_tri_clip isn't set). Only relevant if bypass_clip_xy isn't set.
  */
 void draw_set_driver_clipping( struct draw_context *draw,
                                boolean bypass_clip_xy,
-                               boolean bypass_clip_z )
+                               boolean bypass_clip_z,
+                               boolean guard_band_xy,
+                               boolean bypass_clip_points)
 {
    draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE );
 
    draw->driver.bypass_clip_xy = bypass_clip_xy;
    draw->driver.bypass_clip_z = bypass_clip_z;
+   draw->driver.guard_band_xy = guard_band_xy;
+   draw->driver.bypass_clip_points = bypass_clip_points;
    update_clip_flags(draw);
 }
 
@@ -281,49 +332,52 @@ void draw_set_rasterize_stage( struct draw_context *draw,
 void draw_set_clip_state( struct draw_context *draw,
                           const struct pipe_clip_state *clip )
 {
-   draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE );
-
-   assert(clip->nr <= PIPE_MAX_CLIP_PLANES);
-   memcpy(&draw->plane[6], clip->ucp, clip->nr * sizeof(clip->ucp[0]));
-   draw->nr_planes = 6 + clip->nr;
-   draw->depth_clamp = clip->depth_clamp;
+   draw_do_flush(draw, DRAW_FLUSH_PARAMETER_CHANGE);
 
-   update_clip_flags(draw);
+   memcpy(&draw->plane[6], clip->ucp, sizeof(clip->ucp));
 }
 
 
 /**
  * Set the draw module's viewport state.
  */
-void draw_set_viewport_state( struct draw_context *draw,
-                              const struct pipe_viewport_state *viewport )
+void draw_set_viewport_states( struct draw_context *draw,
+                               unsigned start_slot,
+                               unsigned num_viewports,
+                               const struct pipe_viewport_state *vps )
 {
-   draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE );
-   draw->viewport = *viewport; /* struct copy */
-   draw->identity_viewport = (viewport->scale[0] == 1.0f &&
-                              viewport->scale[1] == 1.0f &&
-                              viewport->scale[2] == 1.0f &&
-                              viewport->scale[3] == 1.0f &&
-                              viewport->translate[0] == 0.0f &&
-                              viewport->translate[1] == 0.0f &&
-                              viewport->translate[2] == 0.0f &&
-                              viewport->translate[3] == 0.0f);
-
-   draw_vs_set_viewport( draw, viewport );
+   const struct pipe_viewport_state *viewport = vps;
+   draw_do_flush(draw, DRAW_FLUSH_PARAMETER_CHANGE);
+
+   debug_assert(start_slot < PIPE_MAX_VIEWPORTS);
+   debug_assert((start_slot + num_viewports) <= PIPE_MAX_VIEWPORTS);
+
+   memcpy(draw->viewports + start_slot, vps,
+          sizeof(struct pipe_viewport_state) * num_viewports);
+
+   draw->identity_viewport = (num_viewports == 1) &&
+      (viewport->scale[0] == 1.0f &&
+       viewport->scale[1] == 1.0f &&
+       viewport->scale[2] == 1.0f &&
+       viewport->scale[3] == 1.0f &&
+       viewport->translate[0] == 0.0f &&
+       viewport->translate[1] == 0.0f &&
+       viewport->translate[2] == 0.0f &&
+       viewport->translate[3] == 0.0f);
 }
 
 
 
 void
 draw_set_vertex_buffers(struct draw_context *draw,
-                        unsigned count,
+                        unsigned start_slot, unsigned count,
                         const struct pipe_vertex_buffer *buffers)
 {
-   assert(count <= PIPE_MAX_ATTRIBS);
+   assert(start_slot + count <= PIPE_MAX_ATTRIBS);
 
-   util_copy_vertex_buffers(draw->pt.vertex_buffer,
-                            &draw->pt.nr_vertex_buffers,
-                            buffers, count);
+   util_set_vertex_buffers_count(draw->pt.vertex_buffer,
+                                 &draw->pt.nr_vertex_buffers,
+                                 buffers, start_slot, count);
 }
 
 
@@ -334,6 +388,10 @@ draw_set_vertex_elements(struct draw_context *draw,
 {
    assert(count <= PIPE_MAX_ATTRIBS);
 
+   /* We could improve this by only flushing the frontend and the fetch part
+    * of the middle. This would avoid recalculating the emit keys.*/
+   draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE );
+
    memcpy(draw->pt.vertex_element, elements, count * sizeof(elements[0]));
    draw->pt.nr_vertex_elements = count;
 }
@@ -344,9 +402,11 @@ draw_set_vertex_elements(struct draw_context *draw,
  */
 void
 draw_set_mapped_vertex_buffer(struct draw_context *draw,
-                              unsigned attr, const void *buffer)
+                              unsigned attr, const void *buffer,
+                              size_t size)
 {
-   draw->pt.user.vbuffer[attr] = buffer;
+   draw->pt.user.vbuffer[attr].map  = buffer;
+   draw->pt.user.vbuffer[attr].size = size;
 }
 
 
@@ -361,17 +421,16 @@ draw_set_mapped_constant_buffer(struct draw_context *draw,
                 shader_type == PIPE_SHADER_GEOMETRY);
    debug_assert(slot < PIPE_MAX_CONSTANT_BUFFERS);
 
+   draw_do_flush(draw, DRAW_FLUSH_PARAMETER_CHANGE);
+
    switch (shader_type) {
    case PIPE_SHADER_VERTEX:
       draw->pt.user.vs_constants[slot] = buffer;
       draw->pt.user.vs_constants_size[slot] = size;
-      draw->pt.user.planes = (float (*) [12][4]) &(draw->plane[0]);
-      draw_vs_set_constants(draw, slot, buffer, size);
       break;
    case PIPE_SHADER_GEOMETRY:
       draw->pt.user.gs_constants[slot] = buffer;
       draw->pt.user.gs_constants_size[slot] = size;
-      draw_gs_set_constants(draw, slot, buffer, size);
       break;
    default:
       assert(0 && "invalid shader type in draw_set_mapped_constant_buffer");
@@ -446,7 +505,9 @@ draw_set_force_passthrough( struct draw_context *draw, boolean enable )
 
 
 /**
- * Allocate an extra vertex/geometry shader vertex attribute.
+ * Allocate an extra vertex/geometry shader vertex attribute, if it doesn't
+ * exist already.
+ *
  * This is used by some of the optional draw module stages such
  * as wide_point which may need to allocate additional generic/texcoord
  * attributes.
@@ -455,8 +516,17 @@ int
 draw_alloc_extra_vertex_attrib(struct draw_context *draw,
                                uint semantic_name, uint semantic_index)
 {
-   const int num_outputs = draw_current_shader_outputs(draw);
-   const int n = draw->extra_shader_outputs.num;
+   int slot;
+   uint num_outputs;
+   uint n;
+
+   slot = draw_find_shader_output(draw, semantic_name, semantic_index);
+   if (slot >= 0) {
+      return slot;
+   }
+
+   num_outputs = draw_current_shader_outputs(draw);
+   n = draw->extra_shader_outputs.num;
 
    assert(n < Elements(draw->extra_shader_outputs.semantic_name));
 
@@ -480,6 +550,44 @@ draw_remove_extra_vertex_attribs(struct draw_context *draw)
 }
 
 
+/**
+ * If a geometry shader is present, return its info, else the vertex shader's
+ * info.
+ */
+struct tgsi_shader_info *
+draw_get_shader_info(const struct draw_context *draw)
+{
+
+   if (draw->gs.geometry_shader) {
+      return &draw->gs.geometry_shader->info;
+   } else {
+      return &draw->vs.vertex_shader->info;
+   }
+}
+
+/**
+ * Prepare outputs slots from the draw module
+ *
+ * Certain parts of the draw module can emit additional
+ * outputs that can be quite useful to the backends, a good
+ * example of it is the process of decomposing primitives
+ * into wireframes (aka. lines) which normally would lose
+ * the face-side information, but using this method we can
+ * inject another shader output which passes the original
+ * face side information to the backend.
+ */
+void
+draw_prepare_shader_outputs(struct draw_context *draw)
+{
+   draw_remove_extra_vertex_attribs(draw);
+   draw_prim_assembler_prepare_outputs(draw->ia);
+   draw_unfilled_prepare_outputs(draw, draw->pipeline.unfilled);
+   if (draw->pipeline.aapoint)
+      draw_aapoint_prepare_outputs(draw, draw->pipeline.aapoint);
+   if (draw->pipeline.aaline)
+      draw_aaline_prepare_outputs(draw, draw->pipeline.aaline);
+}
+
 /**
  * Ask the draw module for the location/slot of the given vertex attribute in
  * a post-transformed vertex.
@@ -491,21 +599,17 @@ draw_remove_extra_vertex_attribs(struct draw_context *draw)
  * attributes (such as texcoords for AA lines).  The driver can call this
  * function to find those attributes.
  *
- * Zero is returned if the attribute is not found since this is
- * a don't care / undefined situtation.  Returning -1 would be a bit more
- * work for the drivers.
+ * -1 is returned if the attribute is not found since this is
+ * an undefined situation. Note, that zero is valid and can
+ * be used by any of the attributes, because position is not
+ * required to be attribute 0 or even at all present.
  */
 int
 draw_find_shader_output(const struct draw_context *draw,
                         uint semantic_name, uint semantic_index)
 {
-   const struct draw_vertex_shader *vs = draw->vs.vertex_shader;
-   const struct draw_geometry_shader *gs = draw->gs.geometry_shader;
+   const struct tgsi_shader_info *info = draw_get_shader_info(draw);
    uint i;
-   const struct tgsi_shader_info *info = &vs->info;
-
-   if (gs)
-      info = &gs->info;
 
    for (i = 0; i < info->num_outputs; i++) {
       if (info->output_semantic_name[i] == semantic_name &&
@@ -521,7 +625,7 @@ draw_find_shader_output(const struct draw_context *draw,
       }
    }
 
-   return 0;
+   return -1;
 }
 
 
@@ -537,40 +641,65 @@ draw_find_shader_output(const struct draw_context *draw,
 uint
 draw_num_shader_outputs(const struct draw_context *draw)
 {
+   const struct tgsi_shader_info *info = draw_get_shader_info(draw);
    uint count;
 
-   /* If a geometry shader is present, its outputs go to the
-    * driver, else the vertex shader's outputs.
-    */
-   if (draw->gs.geometry_shader)
-      count = draw->gs.geometry_shader->info.num_outputs;
-   else
-      count = draw->vs.vertex_shader->info.num_outputs;
-
+   count = info->num_outputs;
    count += draw->extra_shader_outputs.num;
 
    return count;
 }
 
 
+/**
+ * Return total number of the vertex shader outputs.  This function
+ * also counts any extra vertex output attributes that may
+ * be filled in by some draw stages (such as AA point, AA line,
+ * front face).
+ */
+uint
+draw_total_vs_outputs(const struct draw_context *draw)
+{
+   const struct tgsi_shader_info *info = &draw->vs.vertex_shader->info;
+
+   return info->num_outputs + draw->extra_shader_outputs.num;;
+}
+
+/**
+ * Return total number of the geometry shader outputs. This function
+ * also counts any extra geometry output attributes that may
+ * be filled in by some draw stages (such as AA point, AA line, front
+ * face).
+ */
+uint
+draw_total_gs_outputs(const struct draw_context *draw)
+{   
+   const struct tgsi_shader_info *info;
+
+   if (!draw->gs.geometry_shader)
+      return 0;
+
+   info = &draw->gs.geometry_shader->info;
+
+   return info->num_outputs + draw->extra_shader_outputs.num;
+}
+
+
 /**
  * Provide TGSI sampler objects for vertex/geometry shaders that use
- * texture fetches.
+ * texture fetches.  This state only needs to be set once per context.
  * This might only be used by software drivers for the time being.
  */
 void
-draw_texture_samplers(struct draw_context *draw,
-                      uint shader,
-                      uint num_samplers,
-                      struct tgsi_sampler **samplers)
+draw_texture_sampler(struct draw_context *draw,
+                     uint shader,
+                     struct tgsi_sampler *sampler)
 {
    if (shader == PIPE_SHADER_VERTEX) {
-      draw->vs.num_samplers = num_samplers;
-      draw->vs.samplers = samplers;
+      draw->vs.tgsi.sampler = sampler;
    } else {
       debug_assert(shader == PIPE_SHADER_GEOMETRY);
-      draw->gs.num_samplers = num_samplers;
-      draw->gs.samplers = samplers;
+      draw->gs.tgsi.sampler = sampler;
    }
 }
 
@@ -584,25 +713,28 @@ void draw_set_render( struct draw_context *draw,
 }
 
 
-void
-draw_set_index_buffer(struct draw_context *draw,
-                      const struct pipe_index_buffer *ib)
-{
-   if (ib)
-      memcpy(&draw->pt.index_buffer, ib, sizeof(draw->pt.index_buffer));
-   else
-      memset(&draw->pt.index_buffer, 0, sizeof(draw->pt.index_buffer));
-}
-
-
 /**
- * Tell drawing context where to find mapped index/element buffer.
+ * Tell the draw module where vertex indexes/elements are located, and
+ * their size (in bytes).
+ *
+ * Note: the caller must apply the pipe_index_buffer::offset value to
+ * the address.  The draw module doesn't do that.
  */
 void
-draw_set_mapped_index_buffer(struct draw_context *draw,
-                             const void *elements)
+draw_set_indexes(struct draw_context *draw,
+                 const void *elements, unsigned elem_size,
+                 unsigned elem_buffer_space)
 {
-    draw->pt.user.elts = elements;
+   assert(elem_size == 0 ||
+          elem_size == 1 ||
+          elem_size == 2 ||
+          elem_size == 4);
+   draw->pt.user.elts = elements;
+   draw->pt.user.eltSizeIB = elem_size;
+   if (elem_size)
+      draw->pt.user.eltMax = elem_buffer_space / elem_size;
+   else
+      draw->pt.user.eltMax = 0;
 }
 
 
@@ -618,8 +750,8 @@ void draw_do_flush( struct draw_context *draw, unsigned flags )
 
       draw_pipeline_flush( draw, flags );
 
-      draw->reduced_prim = ~0; /* is reduced_prim needed any more? */
-      
+      draw_pt_flush( draw, flags );
+
       draw->flushing = FALSE;
    }
 }
@@ -653,6 +785,77 @@ draw_current_shader_position_output(const struct draw_context *draw)
 }
 
 
+/**
+ * Return the index of the shader output which will contain the
+ * viewport index.
+ */
+uint
+draw_current_shader_viewport_index_output(const struct draw_context *draw)
+{
+   if (draw->gs.geometry_shader)
+      return draw->gs.geometry_shader->viewport_index_output;
+   return 0;
+}
+
+/**
+ * Returns true if there's a geometry shader bound and the geometry
+ * shader writes out a viewport index.
+ */
+boolean
+draw_current_shader_uses_viewport_index(const struct draw_context *draw)
+{
+   if (draw->gs.geometry_shader)
+      return draw->gs.geometry_shader->info.writes_viewport_index;
+   return FALSE;
+}
+
+
+/**
+ * Return the index of the shader output which will contain the
+ * vertex position.
+ */
+uint
+draw_current_shader_clipvertex_output(const struct draw_context *draw)
+{
+   return draw->vs.clipvertex_output;
+}
+
+uint
+draw_current_shader_clipdistance_output(const struct draw_context *draw, int index)
+{
+   debug_assert(index < PIPE_MAX_CLIP_OR_CULL_DISTANCE_ELEMENT_COUNT);
+   if (draw->gs.geometry_shader)
+      return draw->gs.geometry_shader->clipdistance_output[index];
+   return draw->vs.clipdistance_output[index];
+}
+
+
+uint
+draw_current_shader_num_written_clipdistances(const struct draw_context *draw)
+{
+   if (draw->gs.geometry_shader)
+      return draw->gs.geometry_shader->info.num_written_clipdistance;
+   return draw->vs.vertex_shader->info.num_written_clipdistance;
+}
+
+
+uint
+draw_current_shader_culldistance_output(const struct draw_context *draw, int index)
+{
+   debug_assert(index < PIPE_MAX_CLIP_OR_CULL_DISTANCE_ELEMENT_COUNT);
+   if (draw->gs.geometry_shader)
+      return draw->gs.geometry_shader->culldistance_output[index];
+   return draw->vs.vertex_shader->culldistance_output[index];
+}
+
+uint
+draw_current_shader_num_written_culldistances(const struct draw_context *draw)
+{
+   if (draw->gs.geometry_shader)
+      return draw->gs.geometry_shader->info.num_written_culldistance;
+   return draw->vs.vertex_shader->info.num_written_culldistance;
+}
+
 /**
  * Return a pointer/handle for a driver/CSO rasterizer object which
  * disabled culling, stippling, unfilled tris, etc.
@@ -678,7 +881,9 @@ draw_get_rasterizer_no_cull( struct draw_context *draw,
       rast.scissor = scissor;
       rast.flatshade = flatshade;
       rast.front_ccw = 1;
-      rast.gl_rasterization_rules = draw->rasterizer->gl_rasterization_rules;
+      rast.half_pixel_center = draw->rasterizer->half_pixel_center;
+      rast.bottom_edge_rule = draw->rasterizer->bottom_edge_rule;
+      rast.clip_halfz = draw->rasterizer->clip_halfz;
 
       draw->rasterizer_no_cull[scissor][flatshade] =
          pipe->create_rasterizer_state(pipe, &rast);
@@ -687,80 +892,191 @@ draw_get_rasterizer_no_cull( struct draw_context *draw,
 }
 
 void
-draw_set_mapped_so_buffers(struct draw_context *draw,
-                           void *buffers[PIPE_MAX_SO_BUFFERS],
-                           unsigned num_buffers)
+draw_set_mapped_so_targets(struct draw_context *draw,
+                           int num_targets,
+                           struct draw_so_target *targets[PIPE_MAX_SO_BUFFERS])
 {
    int i;
 
-   for (i = 0; i < num_buffers; ++i) {
-      draw->so.buffers[i] = buffers[i];
-   }
-   draw->so.num_buffers = num_buffers;
-}
+   for (i = 0; i < num_targets; i++)
+      draw->so.targets[i] = targets[i];
+   for (i = num_targets; i < PIPE_MAX_SO_BUFFERS; i++)
+      draw->so.targets[i] = NULL;
 
-void
-draw_set_so_state(struct draw_context *draw,
-                  struct pipe_stream_output_state *state)
-{
-   memcpy(&draw->so.state,
-          state,
-          sizeof(struct pipe_stream_output_state));
+   draw->so.num_targets = num_targets;
 }
 
 void
 draw_set_sampler_views(struct draw_context *draw,
+                       unsigned shader_stage,
                        struct pipe_sampler_view **views,
                        unsigned num)
 {
    unsigned i;
 
-   debug_assert(num <= PIPE_MAX_VERTEX_SAMPLERS);
+   debug_assert(shader_stage < PIPE_SHADER_TYPES);
+   debug_assert(num <= PIPE_MAX_SHADER_SAMPLER_VIEWS);
+
+   draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE );
 
    for (i = 0; i < num; ++i)
-      draw->sampler_views[i] = views[i];
-   for (i = num; i < PIPE_MAX_VERTEX_SAMPLERS; ++i)
-      draw->sampler_views[i] = NULL;
+      draw->sampler_views[shader_stage][i] = views[i];
+   for (i = num; i < PIPE_MAX_SHADER_SAMPLER_VIEWS; ++i)
+      draw->sampler_views[shader_stage][i] = NULL;
 
-   draw->num_sampler_views = num;
+   draw->num_sampler_views[shader_stage] = num;
 }
 
 void
 draw_set_samplers(struct draw_context *draw,
+                  unsigned shader_stage,
                   struct pipe_sampler_state **samplers,
                   unsigned num)
 {
    unsigned i;
 
-   debug_assert(num <= PIPE_MAX_VERTEX_SAMPLERS);
+   debug_assert(shader_stage < PIPE_SHADER_TYPES);
+   debug_assert(num <= PIPE_MAX_SAMPLERS);
+
+   draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE );
 
    for (i = 0; i < num; ++i)
-      draw->samplers[i] = samplers[i];
-   for (i = num; i < PIPE_MAX_VERTEX_SAMPLERS; ++i)
-      draw->samplers[i] = NULL;
+      draw->samplers[shader_stage][i] = samplers[i];
+   for (i = num; i < PIPE_MAX_SAMPLERS; ++i)
+      draw->samplers[shader_stage][i] = NULL;
 
-   draw->num_samplers = num;
+   draw->num_samplers[shader_stage] = num;
 
 #ifdef HAVE_LLVM
    if (draw->llvm)
-      draw_llvm_set_sampler_state(draw);
+      draw_llvm_set_sampler_state(draw, shader_stage);
 #endif
 }
 
 void
 draw_set_mapped_texture(struct draw_context *draw,
-                        unsigned sampler_idx,
+                        unsigned shader_stage,
+                        unsigned sview_idx,
                         uint32_t width, uint32_t height, uint32_t depth,
                         uint32_t first_level, uint32_t last_level,
+                        const void *base_ptr,
                         uint32_t row_stride[PIPE_MAX_TEXTURE_LEVELS],
                         uint32_t img_stride[PIPE_MAX_TEXTURE_LEVELS],
-                        const void *data[PIPE_MAX_TEXTURE_LEVELS])
+                        uint32_t mip_offsets[PIPE_MAX_TEXTURE_LEVELS])
 {
 #ifdef HAVE_LLVM
-   if(draw->llvm)
+   if (draw->llvm)
       draw_llvm_set_mapped_texture(draw,
-                                sampler_idx,
-                                width, height, depth, first_level, last_level,
-                                row_stride, img_stride, data);
+                                   shader_stage,
+                                   sview_idx,
+                                   width, height, depth, first_level,
+                                   last_level, base_ptr,
+                                   row_stride, img_stride, mip_offsets);
+#endif
+}
+
+/**
+ * XXX: Results for PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS because there are two
+ * different ways of setting textures, and drivers typically only support one.
+ */
+int
+draw_get_shader_param_no_llvm(unsigned shader, enum pipe_shader_cap param)
+{
+   switch(shader) {
+   case PIPE_SHADER_VERTEX:
+   case PIPE_SHADER_GEOMETRY:
+      return tgsi_exec_get_shader_param(param);
+   default:
+      return 0;
+   }
+}
+
+/**
+ * XXX: Results for PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS because there are two
+ * different ways of setting textures, and drivers typically only support one.
+ */
+int
+draw_get_shader_param(unsigned shader, enum pipe_shader_cap param)
+{
+
+#ifdef HAVE_LLVM
+   if (draw_get_option_use_llvm()) {
+      switch(shader) {
+      case PIPE_SHADER_VERTEX:
+      case PIPE_SHADER_GEOMETRY:
+         return gallivm_get_shader_param(param);
+      default:
+         return 0;
+      }
+   }
 #endif
+
+   return draw_get_shader_param_no_llvm(shader, param);
+}
+
+/**
+ * Enables or disables collection of statistics.
+ *
+ * Draw module is capable of generating statistics for the vertex
+ * processing pipeline. Collection of that data isn't free and so
+ * it's disabled by default. The users of the module can enable
+ * (or disable) this functionality through this function.
+ * The actual data will be emitted through the VBUF interface,
+ * the 'pipeline_statistics' callback to be exact.
+ */
+void
+draw_collect_pipeline_statistics(struct draw_context *draw,
+                                 boolean enable)
+{
+   draw->collect_statistics = enable;
+}
+
+/**
+ * Computes clipper invocation statistics.
+ *
+ * Figures out how many primitives would have been
+ * sent to the clipper given the specified
+ * prim info data.
+ */
+void
+draw_stats_clipper_primitives(struct draw_context *draw,
+                              const struct draw_prim_info *prim_info)
+{
+   if (draw->collect_statistics) {
+      unsigned start, i;
+      for (start = i = 0;
+           i < prim_info->primitive_count;
+           start += prim_info->primitive_lengths[i], i++)
+      {
+         draw->statistics.c_invocations +=
+            u_decomposed_prims_for_vertices(prim_info->prim,
+                                            prim_info->primitive_lengths[i]);
+      }
+   }
+}
+
+
+/**
+ * Returns true if the draw module will inject the frontface
+ * info into the outputs.
+ *
+ * Given the specified primitive and rasterizer state
+ * the function will figure out if the draw module
+ * will inject the front-face information into shader
+ * outputs. This is done to preserve the front-facing
+ * info when decomposing primitives into wireframes.
+ */
+boolean
+draw_will_inject_frontface(const struct draw_context *draw)
+{
+   unsigned reduced_prim = u_reduced_prim(draw->pt.prim);
+   const struct pipe_rasterizer_state *rast = draw->rasterizer;
+
+   if (reduced_prim != PIPE_PRIM_TRIANGLES) {
+      return FALSE;
+   }
+
+   return (rast &&
+           (rast->fill_front != PIPE_POLYGON_MODE_FILL ||
+            rast->fill_back != PIPE_POLYGON_MODE_FILL));
 }