dxGraphicsStateGuardian9.cxx

Go to the documentation of this file.

/**
 * PANDA 3D SOFTWARE
 * Copyright (c) Carnegie Mellon University.  All rights reserved.
 *
 * All use of this software is subject to the terms of the revised BSD
 * license.  You should have received a copy of this license along
 * with this source code in a file named "LICENSE."
 *
 * @file dxGraphicsStateGuardian9.cxx
 * @author mike
 * @date 1999-02-02
 * @author fperazzi, PandaSE
 * @date 2010-05-05
 */

#include "dxGraphicsStateGuardian9.h"
#include "config_dxgsg9.h"
#include "displayRegion.h"
#include "renderBuffer.h"
#include "geom.h"
#include "graphicsWindow.h"
#include "graphicsEngine.h"
#include "lens.h"
#include "ambientLight.h"
#include "directionalLight.h"
#include "pointLight.h"
#include "spotlight.h"
#include "textureAttrib.h"
#include "texGenAttrib.h"
#include "shadeModelAttrib.h"
#include "cullFaceAttrib.h"
#include "transparencyAttrib.h"
#include "alphaTestAttrib.h"
#include "depthTestAttrib.h"
#include "depthWriteAttrib.h"
#include "colorWriteAttrib.h"
#include "texMatrixAttrib.h"
#include "materialAttrib.h"
#include "renderModeAttrib.h"
#include "rescaleNormalAttrib.h"
#include "fogAttrib.h"
#include "depthOffsetAttrib.h"
#include "lightAttrib.h"
#include "stencilAttrib.h"
#include "scissorAttrib.h"
#include "clipPlaneAttrib.h"
#include "fog.h"
#include "throw_event.h"
#include "geomVertexFormat.h"
#include "geomVertexData.h"
#include "geomTriangles.h"
#include "geomTristrips.h"
#include "geomTrifans.h"
#include "geomLines.h"
#include "geomLinestrips.h"
#include "geomPoints.h"
#include "geomVertexReader.h"
#include "dxGeomMunger9.h"
#include "config_gobj.h"
#include "dxVertexBufferContext9.h"
#include "dxIndexBufferContext9.h"
#include "dxOcclusionQueryContext9.h"
#include "pStatTimer.h"
#include "pStatCollector.h"
#include "wdxGraphicsBuffer9.h"
#include "config_pgraph.h"
#include "shaderGenerator.h"
#ifdef HAVE_CG
#include <Cg/cgD3D9.h>
#endif

#include <mmsystem.h>

#include <dxsdkver.h>

#define tostring(x) #x
#define SDK_VERSION(major,minor) tostring(major) << "." << tostring(minor)
#define DIRECTX_SDK_VERSION  SDK_VERSION (_DXSDK_PRODUCT_MAJOR, _DXSDK_PRODUCT_MINOR) << "." << SDK_VERSION (_DXSDK_BUILD_MAJOR, _DXSDK_BUILD_MINOR)

using std::endl;
using std::max;
using std::min;

TypeHandle DXGraphicsStateGuardian9::_type_handle;

D3DMATRIX DXGraphicsStateGuardian9::_d3d_ident_mat;

unsigned char *DXGraphicsStateGuardian9::_temp_buffer = nullptr;
unsigned char *DXGraphicsStateGuardian9::_safe_buffer_start = nullptr;

#ifdef HAVE_CG
LPDIRECT3DDEVICE9 DXGraphicsStateGuardian9::_cg_device = nullptr;
#endif

#define __D3DLIGHT_RANGE_MAX ((PN_stdfloat)sqrt(FLT_MAX))  //for some reason this is missing in dx9 hdrs

#define MY_D3DRGBA(r, g, b, a) ((D3DCOLOR) D3DCOLOR_COLORVALUE(r, g, b, a))

/**
 *
 */
DXGraphicsStateGuardian9::
DXGraphicsStateGuardian9(GraphicsEngine *engine, GraphicsPipe *pipe) :
  GraphicsStateGuardian(CS_yup_left, engine, pipe)
{
  if (dxgsg9_cat.is_debug()) {
    dxgsg9_cat.debug()
      << "DXGraphicsStateGuardian9 " << this << " constructing\n";
  }

  // Assume that we will get a hardware-accelerated context, unless the window
  // tells us otherwise.
  _is_hardware = true;

  _screen = nullptr;
  _d3d_device = nullptr;

  _dx_is_ready = false;
  _vertex_blending_enabled = false;
  _overlay_windows_supported = false;
  _tex_stats_retrieval_impossible = false;
  _supports_render_texture = false;

  _active_ibuffer = nullptr;

  // This is a static member, but we initialize it here in the constructor
  // anyway.  It won't hurt if it gets repeatedly initalized.
  ZeroMemory(&_d3d_ident_mat, sizeof(D3DMATRIX));
  _d3d_ident_mat._11 = _d3d_ident_mat._22 = _d3d_ident_mat._33 = _d3d_ident_mat._44 = 1.0f;

  _cur_read_pixel_buffer = RenderBuffer::T_front;

  // DirectX drivers seem to consistently invert the texture when they copy
  // framebuffer-to-texture.  Ok.
  _copy_texture_inverted = true;

  _gsg_managed_textures = dx_management | dx_texture_management;
  _gsg_managed_vertex_buffers = dx_management;
  _gsg_managed_index_buffers = dx_management;

  _last_fvf = 0;
  _num_bound_streams = 0;
  _white_vbuffer = nullptr;

  _vertex_shader_version_major = 0;
  _vertex_shader_version_minor = 0;
  _pixel_shader_version_major = 0;
  _pixel_shader_version_minor = 0;

  _vertex_shader_profile = 0;
  _pixel_shader_profile = 0;

  _vertex_shader_maximum_constants = 0;

  _supports_stream_offset = false;

#ifdef HAVE_CG
  _cg_context = 0;
#endif

  get_gamma_table();
  atexit (atexit_function);
}

/**
 *
 */
DXGraphicsStateGuardian9::
~DXGraphicsStateGuardian9() {
  if (dxgsg9_cat.is_debug()) {
    dxgsg9_cat.debug()
      << "DXGraphicsStateGuardian9 " << this << " destructing\n";
  }

  if (IS_VALID_PTR(_d3d_device)) {
    _d3d_device->SetTexture(0, nullptr);  // this frees reference to the old texture
  }

  free_nondx_resources();
}

/**
 * Creates a new retained-mode representation of the given texture, and
 * returns a newly-allocated TextureContext pointer to reference it.  It is
 * the responsibility of the calling function to later call release_texture()
 * with this same pointer (which will also delete the pointer).
 *
 * This function should not be called directly to prepare a texture.  Instead,
 * call Texture::prepare().
 */
TextureContext *DXGraphicsStateGuardian9::
prepare_texture(Texture *tex, int view) {
  DXTextureContext9 *dtc = new DXTextureContext9(_prepared_objects, tex, view);

  if (!get_supports_compressed_texture_format(tex->get_ram_image_compression())) {
    dxgsg9_cat.error()
      << *dtc->get_texture() << " is stored in an unsupported compressed format.\n";
    return nullptr;
  }

  return dtc;
}

/**
 * Makes the texture the currently available texture for rendering on the ith
 * stage.
 */
void DXGraphicsStateGuardian9::
apply_texture(int i, TextureContext *tc, const SamplerState &sampler) {
  if (tc == nullptr) {
    // The texture wasn't bound properly or something, so ensure texturing is
    // disabled and just return.
    set_texture_stage_state(i, D3DTSS_COLOROP, D3DTOP_DISABLE);
    return;
  }
  if (!update_texture(tc, false)) {
    // Couldn't get the texture image or something.
    set_texture_stage_state(i, D3DTSS_COLOROP, D3DTOP_DISABLE);
    return;
  }

  tc->set_active(true);

  DXTextureContext9 *dtc = DCAST(DXTextureContext9, tc);
  Texture *tex = tc->get_texture();

  // if (tex->get_color_space() == CS_srgb) {
  if (Texture::is_srgb(tex->get_format())) {
    set_sampler_state(i, D3DSAMP_SRGBTEXTURE, TRUE);
  } else {
    set_sampler_state(i, D3DSAMP_SRGBTEXTURE, FALSE);
  }

  set_sampler_state(i, D3DSAMP_ADDRESSU,
    get_texture_wrap_mode(sampler.get_wrap_u()));

  set_sampler_state(i, D3DSAMP_ADDRESSV,
    get_texture_wrap_mode(sampler.get_wrap_v()));

  set_sampler_state(i, D3DSAMP_ADDRESSW,
    get_texture_wrap_mode(sampler.get_wrap_w()));

  DWORD border_color;
  border_color = LColor_to_D3DCOLOR(sampler.get_border_color());

  set_sampler_state(i, D3DSAMP_BORDERCOLOR, border_color);

  uint aniso_degree = sampler.get_effective_anisotropic_degree();
  SamplerState::FilterType ft = sampler.get_effective_magfilter();

  if (aniso_degree >= 1) {
    set_sampler_state(i, D3DSAMP_MAXANISOTROPY, aniso_degree);
  }

  int supports_anisotropic_mag_filter;
  D3DTEXTUREFILTERTYPE new_mag_filter;

  supports_anisotropic_mag_filter = (_screen -> _d3dcaps.TextureFilterCaps & D3DPTFILTERCAPS_MAGFANISOTROPIC) != 0;
  if (aniso_degree <= 1 || supports_anisotropic_mag_filter == 0) {
    new_mag_filter = ((ft != SamplerState::FT_nearest) ? D3DTEXF_LINEAR : D3DTEXF_POINT);
  } else {
    new_mag_filter = D3DTEXF_ANISOTROPIC;
  }

  HRESULT hr;
  hr = set_sampler_state(i, D3DSAMP_MAGFILTER, new_mag_filter);
  if (hr != D3D_OK) {
    dxgsg9_cat.error()
      << "ERROR: set_sampler_state (D3DSAMP_MAGFILTER, "
      << new_mag_filter << ") failed for sampler: " << sampler << endl;
  }

  // map Panda composite min+mip filter types to d3d's separate min & mip
  // filter types
  D3DTEXTUREFILTERTYPE new_min_filter = get_d3d_min_type(sampler.get_effective_minfilter());
  D3DTEXTUREFILTERTYPE new_mip_filter = get_d3d_mip_type(sampler.get_effective_minfilter());

  if (!tex->might_have_ram_image()) {
    // If the texture is completely dynamic, don't try to issue mipmaps--
    // pandadx doesn't support auto-generated mipmaps at this point.
    new_mip_filter = D3DTEXF_NONE;
  }

  // sanity check
  if (!dtc->has_mipmaps()) {
    new_mip_filter = D3DTEXF_NONE;
  }

  if (aniso_degree >= 2) {
    new_min_filter = D3DTEXF_ANISOTROPIC;
  }

  set_sampler_state(i, D3DSAMP_MINFILTER, new_min_filter);
  set_sampler_state(i, D3DSAMP_MIPFILTER, new_mip_filter);

  float lod_bias = sampler.get_lod_bias();
  set_sampler_state(i, D3DSAMP_MIPMAPLODBIAS, *(DWORD*)&lod_bias);

  _d3d_device->SetTexture(i, dtc->get_d3d_texture());
}

/**
 * Ensures that the current Texture data is refreshed onto the GSG.  This
 * means updating the texture properties and/or re-uploading the texture
 * image, if necessary.  This should only be called within the draw thread.
 *
 * If force is true, this function will not return until the texture has been
 * fully uploaded.  If force is false, the function may choose to upload a
 * simple version of the texture instead, if the texture is not fully resident
 * (and if get_incomplete_render() is true).
 */
bool DXGraphicsStateGuardian9::
update_texture(TextureContext *tc, bool force) {
  DXTextureContext9 *dtc = DCAST(DXTextureContext9, tc);

  // If the texture image has changed, or if its use of mipmaps has changed,
  // we need to re-create the image.

  if (dtc->was_modified()) {
    if (!upload_texture(dtc, force)) {
      // Oops, we can't re-create the texture for some reason.
      Texture *tex = tc->get_texture();
      dxgsg9_cat.error()
        << "Unable to re-create texture " << *tex << endl;
      return false;
    }
  }
  dtc->enqueue_lru(&_prepared_objects->_graphics_memory_lru);

  return true;
}

/**
 * Creates a texture surface on the graphics card and fills it with its pixel
 * data.
 */
bool DXGraphicsStateGuardian9::
upload_texture(DXTextureContext9 *dtc, bool force) {
  Texture *tex = dtc->get_texture();
  if (!get_supports_compressed_texture_format(tex->get_ram_image_compression())) {
    dxgsg9_cat.error()
      << *tex << " is stored in an unsupported compressed format.\n";
    return false;
  }

  dtc->delete_texture();
  dtc->update_data_size_bytes(0);
  dtc->mark_unloaded();

  if (_effective_incomplete_render && !force) {
    bool has_image = _supports_compressed_texture ? tex->has_ram_image() : tex->has_uncompressed_ram_image();
    if (!has_image && tex->might_have_ram_image() &&
        tex->has_simple_ram_image() &&
        !_loader.is_null()) {
      // If we don't have the texture data right now, go get it, but in the
      // meantime load a temporary simple image in its place.
      async_reload_texture(dtc);
      has_image = _supports_compressed_texture ? tex->has_ram_image() : tex->has_uncompressed_ram_image();
      if (!has_image) {
        if (dtc->was_simple_image_modified()) {
          return dtc->create_simple_texture(*_screen);
        }
        return true;
      }
    }
  }

  return dtc->create_texture(*_screen);
}

/**
 * Frees the GL resources previously allocated for the texture.
 */
void DXGraphicsStateGuardian9::
release_texture(TextureContext *tc) {
  DXTextureContext9 *dtc = DCAST(DXTextureContext9, tc);
  delete dtc;
}

/**
 * This method should only be called by the GraphicsEngine.  Do not call it
 * directly; call GraphicsEngine::extract_texture_data() instead.
 *
 * This method will be called in the draw thread to download the texture
 * memory's image into its ram_image value.  It returns true on success, false
 * otherwise.
 */
bool DXGraphicsStateGuardian9::
extract_texture_data(Texture *tex) {
  bool success = true;

  int num_views = tex->get_num_views();
  for (int view = 0; view < num_views; ++view) {
    TextureContext *tc = tex->prepare_now(view, get_prepared_objects(), this);
    nassertr(tc != nullptr, false);
    DXTextureContext9 *dtc = DCAST(DXTextureContext9, tc);

    if (!dtc->extract_texture_data(*_screen)) {
      success = false;
    }
  }

  return success;
}

/**
 *
 */
ShaderContext *DXGraphicsStateGuardian9::
prepare_shader(Shader *se) {
  PStatTimer timer(_prepare_shader_pcollector);

  switch (se->get_language()) {
  case Shader::SL_GLSL:
    dxgsg9_cat.error()
      << "Tried to load GLSL shader, but GLSL shaders not supported by Direct3D 9.\n";
    return nullptr;

  case Shader::SL_Cg:
#ifdef HAVE_CG
    if (_supports_basic_shaders) {
      return new DXShaderContext9(se, this);
    } else {
      dxgsg9_cat.error()
        << "Tried to load Cg shader, but basic shaders not supported.\n";
      return nullptr;
    }
#else
    dxgsg9_cat.error()
      << "Tried to load Cg shader, but Cg support not compiled in.\n";
    return nullptr;
#endif

  default:
    dxgsg9_cat.error()
      << "Tried to load shader with unsupported shader language!\n";
    return nullptr;
  }

  return nullptr;
}

/**
 *
 */
void DXGraphicsStateGuardian9::
release_shader(ShaderContext *sc) {
  DXShaderContext9 *gsc = DCAST(DXShaderContext9, sc);
  delete gsc;
}

/**
 * Creates a new retained-mode representation of the given data, and returns a
 * newly-allocated VertexBufferContext pointer to reference it.  It is the
 * responsibility of the calling function to later call
 * release_vertex_buffer() with this same pointer (which will also delete the
 * pointer).
 *
 * This function should not be called directly to prepare a buffer.  Instead,
 * call Geom::prepare().
 */
VertexBufferContext *DXGraphicsStateGuardian9::
prepare_vertex_buffer(GeomVertexArrayData *data) {
  DXVertexBufferContext9 *dvbc = new DXVertexBufferContext9(this, _prepared_objects, data);

  DWORD usage;
  D3DPOOL pool;
  if (_screen->_managed_vertex_buffers) {
    pool = D3DPOOL_MANAGED;
    usage = D3DUSAGE_WRITEONLY;
  } else {
    pool = D3DPOOL_DEFAULT;
    usage = D3DUSAGE_WRITEONLY | D3DUSAGE_DYNAMIC;
  }

  int num_bytes = data->get_data_size_bytes();

  PStatTimer timer(_create_vertex_buffer_pcollector, Thread::get_current_thread());

  HRESULT hr;
  int attempts = 0;
  do
  {
    hr = _screen->_d3d_device->CreateVertexBuffer(num_bytes, usage, dvbc->_fvf, pool, &dvbc->_vbuffer, nullptr);
    attempts++;
  }
  while (check_dx_allocation(hr, num_bytes, attempts));

  if (!FAILED(hr)) {
    #if 0
    if (dxgsg9_cat.is_debug() && DXdebug_buffers9) {
      dxgsg9_cat.debug()
        << "creating vertex buffer " << dvbc->_vbuffer << ": "
        << data->get_num_rows() << " vertices "
        << *data->get_array_format() << "\n";
    }
    #endif

    return dvbc;
  } else {
    dxgsg9_cat.error()
      << "CreateVertexBuffer failed" << D3DERRORSTRING(hr);

    dvbc->_vbuffer = nullptr;
  }

  return nullptr;
}

/**
 * Updates the vertex buffer with the current data, and makes it the current
 * vertex buffer for rendering.
 */
bool DXGraphicsStateGuardian9::
apply_vertex_buffer(VertexBufferContext *vbc,
                    const GeomVertexArrayDataHandle *reader, bool force ) {

  DXVertexBufferContext9 *dvbc = DCAST(DXVertexBufferContext9, vbc);

  if (dvbc->was_modified(reader)) {
    int num_bytes = reader->get_data_size_bytes();
    #if 0
    if (dxgsg9_cat.is_debug() && DXdebug_buffers9) {
      dxgsg9_cat.debug()
        << "copying " << num_bytes
        << " bytes into vertex buffer " << dvbc->_vbuffer << "\n";
    }
    #endif

    if ( num_bytes != 0 ) {
      const unsigned char *client_pointer = reader->get_read_pointer(force);
      if (client_pointer == nullptr) {
        return false;
      }

      PStatTimer timer(_load_vertex_buffer_pcollector, reader->get_current_thread());

      #if 0
      if (dvbc->changed_size(reader)) {
        // We have to destroy the old vertex buffer and create a new one.
        dvbc->create_vbuffer(*_screen, reader);
      }
      #endif

      HRESULT hr;
      BYTE *local_pointer;
      if (_screen->_managed_vertex_buffers) {
        hr = dvbc->_vbuffer->Lock(0, num_bytes, (void **) &local_pointer, 0);
      } else {
        hr = dvbc->_vbuffer->Lock(0, num_bytes, (void **) &local_pointer, D3DLOCK_DISCARD);
      }
      if (FAILED(hr)) {
        dxgsg9_cat.error()
          << "VertexBuffer::Lock failed" << D3DERRORSTRING(hr);
        return false;
      }

      memcpy(local_pointer, client_pointer, num_bytes);

      dvbc->_vbuffer->Unlock();

      _data_transferred_pcollector.add_level(num_bytes);
    }

    dvbc->mark_loaded(reader);
  }
  dvbc->enqueue_lru(&_prepared_objects->_graphics_memory_lru);

  return true;
}

/**
 * Frees the GL resources previously allocated for the data.  This function
 * should never be called directly; instead, call Data::release() (or simply
 * let the Data destruct).
 */
void DXGraphicsStateGuardian9::
release_vertex_buffer(VertexBufferContext *vbc) {

  DXVertexBufferContext9 *dvbc = DCAST(DXVertexBufferContext9, vbc);

  #if 0
  if (dxgsg9_cat.is_debug() && DXdebug_buffers9) {
    dxgsg9_cat.debug()
      << "deleting vertex buffer " << dvbc->_vbuffer << "\n";
  }
  #endif

  dvbc->_vbuffer->Release();
  dvbc->_vbuffer = nullptr;

  delete dvbc;
}

/**
 * Internal function to bind a buffer object for the indicated data array, if
 * appropriate, or to unbind a buffer object if it should be rendered from
 * client memory.
 *
 * If the buffer object is bound, this function sets client_pointer to NULL
 * (representing the start of the buffer object in server memory); if the
 * buffer object is not bound, this function sets client_pointer the pointer
 * to the data array in client memory, that is, the data array passed in.
 *
 * If force is not true, the function may return false indicating the data is
 * not currently available.
 */
bool DXGraphicsStateGuardian9::
setup_array_data(DXVertexBufferContext9*& dvbc,
                 const GeomVertexArrayDataHandle* array_reader,
                 bool force) {

  // Prepare the buffer object and bind it.
  VertexBufferContext* vbc = ((GeomVertexArrayData *)array_reader->get_object())->prepare_now(get_prepared_objects(), this);
  nassertr(vbc != nullptr, false);
  if (!apply_vertex_buffer(vbc, array_reader, force)) {
    return false;
  }

  dvbc = (DXVertexBufferContext9*)vbc;
  return true;
}

/**
 * Creates a new retained-mode representation of the given data, and returns a
 * newly-allocated IndexBufferContext pointer to reference it.  It is the
 * responsibility of the calling function to later call release_index_buffer()
 * with this same pointer (which will also delete the pointer).
 *
 * This function should not be called directly to prepare a buffer.  Instead,
 * call Geom::prepare().
 */
IndexBufferContext *DXGraphicsStateGuardian9::
prepare_index_buffer(GeomPrimitive *data) {
  DXIndexBufferContext9 *dibc = new DXIndexBufferContext9(_prepared_objects, data);
  return dibc;
}

/**
 * Updates the index buffer with the current data, and makes it the current
 * index buffer for rendering.
 */
bool DXGraphicsStateGuardian9::
apply_index_buffer(IndexBufferContext *ibc,
                   const GeomPrimitivePipelineReader *reader, bool force) {
  DXIndexBufferContext9 *dibc = DCAST(DXIndexBufferContext9, ibc);

  if (dibc->_ibuffer == nullptr) {
    // Attempt to create a new index buffer.
    dibc->create_ibuffer(*_screen, reader);

    if (dibc->_ibuffer != nullptr) {
      if (!dibc->upload_data(reader, force)) {
        return false;
      }
      dibc->mark_loaded(reader);

      _d3d_device->SetIndices(dibc->_ibuffer);
      _active_ibuffer = dibc;
      dibc->set_active(true);

    } else {
      _d3d_device->SetIndices(nullptr);
      _active_ibuffer = nullptr;
    }

  } else {
    if (dibc->was_modified(reader)) {
      if (dibc->changed_size(reader)) {
        // We have to destroy the old index buffer and create a new one.
        dibc->create_ibuffer(*_screen, reader);
      }

      if (!dibc->upload_data(reader, force)) {
        return false;
      }

      dibc->mark_loaded(reader);
      _active_ibuffer = nullptr;
    }

    if (_active_ibuffer != dibc) {
      _d3d_device->SetIndices(dibc->_ibuffer);
      _active_ibuffer = dibc;
      dibc->set_active(true);
    }
  }
  dibc->enqueue_lru(&_prepared_objects->_graphics_memory_lru);

  return true;
}

/**
 * Frees the GL resources previously allocated for the data.  This function
 * should never be called directly; instead, call Data::release() (or simply
 * let the Data destruct).
 */
void DXGraphicsStateGuardian9::
release_index_buffer(IndexBufferContext *ibc) {
  DXIndexBufferContext9 *dibc = DCAST(DXIndexBufferContext9, ibc);
  delete dibc;
}

/**
 * Begins a new occlusion query.  After this call, you may call
 * begin_draw_primitives() and draw_triangles()/draw_whatever() repeatedly.
 * Eventually, you should call end_occlusion_query() before the end of the
 * frame; that will return a new OcclusionQueryContext object that will tell
 * you how many pixels represented by the bracketed geometry passed the depth
 * test.
 *
 * It is not valid to call begin_occlusion_query() between another
 * begin_occlusion_query() .. end_occlusion_query() sequence.
 */
void DXGraphicsStateGuardian9::
begin_occlusion_query() {
  nassertv(_supports_occlusion_query);
  nassertv(_current_occlusion_query == nullptr);

  IDirect3DQuery9 *query;
  HRESULT hr = _d3d_device->CreateQuery(D3DQUERYTYPE_OCCLUSION, &query);
  if (FAILED(hr)) {
    dxgsg9_cat.warning()
      << "Occlusion query failed.\n";
    return;
  }

  PT(DXOcclusionQueryContext9) queryobj = new DXOcclusionQueryContext9(query);

  if (dxgsg9_cat.is_debug()) {
    dxgsg9_cat.debug()
      << "beginning occlusion query " << query << "\n";
  }

  query->Issue(D3DISSUE_BEGIN);
  _current_occlusion_query = queryobj;
}

/**
 * Ends a previous call to begin_occlusion_query(). This call returns the
 * OcclusionQueryContext object that will (eventually) report the number of
 * pixels that passed the depth test between the call to
 * begin_occlusion_query() and end_occlusion_query().
 */
PT(OcclusionQueryContext) DXGraphicsStateGuardian9::
end_occlusion_query() {
  if (_current_occlusion_query == nullptr) {
    return nullptr;
  }

  PT(OcclusionQueryContext) result = _current_occlusion_query;

  IDirect3DQuery9 *query = DCAST(DXOcclusionQueryContext9, result)->_query;

  if (dxgsg9_cat.is_debug()) {
    dxgsg9_cat.debug()
      << "ending occlusion query " << query << "\n";
  }

  _current_occlusion_query = nullptr;
  query->Issue(D3DISSUE_END);

  return result;
}

/**
 * Creates a new GeomMunger object to munge vertices appropriate to this GSG
 * for the indicated state.
 */
PT(GeomMunger) DXGraphicsStateGuardian9::
make_geom_munger(const RenderState *state, Thread *current_thread) {
  PT(DXGeomMunger9) munger = new DXGeomMunger9(this, state);
  return GeomMunger::register_munger(munger, current_thread);
}

/**
 * Clears all of the indicated buffers to their assigned colors.
 */
void DXGraphicsStateGuardian9::
clear(DrawableRegion *clearable) {

  DWORD main_flags = 0;
  DWORD aux_flags = 0;

  if ((!clearable->get_clear_color_active())&&
      (!clearable->get_clear_depth_active())&&
      (!clearable->get_clear_stencil_active())) {
    return;
  }

  set_state_and_transform(RenderState::make_empty(), _internal_transform);

  D3DCOLOR color_clear_value = LColor_to_D3DCOLOR(clearable->get_clear_color());
  PN_stdfloat depth_clear_value = clearable->get_clear_depth();
  DWORD stencil_clear_value = (DWORD)(clearable->get_clear_stencil());

  // set appropriate flags
  if (clearable->get_clear_color_active()) {
    main_flags |=  D3DCLEAR_TARGET;
  }

  if (clearable->get_clear_depth_active() &&
      _screen->_presentation_params.EnableAutoDepthStencil) {
    aux_flags |=  D3DCLEAR_ZBUFFER;
  }

  if (clearable->get_clear_stencil_active()) {
    // clear only if there is a stencil buffer
    if (_screen->_presentation_params.EnableAutoDepthStencil &&
      IS_STENCIL_FORMAT(_screen->_presentation_params.AutoDepthStencilFormat)) {
      aux_flags |=  D3DCLEAR_STENCIL;
    }
  }

  if ((main_flags | aux_flags) != 0) {
    HRESULT hr = _d3d_device->Clear(0, nullptr, main_flags | aux_flags, color_clear_value,
                                    depth_clear_value, stencil_clear_value);
    if (FAILED(hr) && main_flags == D3DCLEAR_TARGET && aux_flags != 0) {
      // Maybe there's a problem with the one or more of the auxiliary
      // buffers.
      hr = _d3d_device->Clear(0, nullptr, D3DCLEAR_TARGET, color_clear_value,
                              depth_clear_value, stencil_clear_value);
      if (!FAILED(hr)) {
        // Yep, it worked without them.  That's a problem.  Which buffer poses
        // the problem?
        if (clearable->get_clear_depth_active()) {
          aux_flags |=  D3DCLEAR_ZBUFFER;
          HRESULT hr2 = _d3d_device->Clear(0, nullptr, D3DCLEAR_ZBUFFER, color_clear_value,
                                           depth_clear_value, stencil_clear_value);
          if (FAILED(hr2)) {
            dxgsg9_cat.error()
              << "Unable to clear depth buffer; removing.\n";
            // This is really hacky code.
            ((FrameBufferProperties *)_current_properties)->set_depth_bits(0);
          }
        }
        if (clearable->get_clear_stencil_active()) {
          aux_flags |=  D3DCLEAR_STENCIL;
          HRESULT hr2 = _d3d_device->Clear(0, nullptr, D3DCLEAR_STENCIL, color_clear_value,
                                           stencil_clear_value, stencil_clear_value);
          if (FAILED(hr2)) {
            dxgsg9_cat.error()
              << "Unable to clear stencil buffer; removing.\n";
            // This is really hacky code.
            ((FrameBufferProperties *)_current_properties)->set_stencil_bits(0);
            _supports_stencil = false;
          }
        }
      }
    }

    if (FAILED(hr)) {
      dxgsg9_cat.error()
        << "clear_buffer failed:  Clear returned " << D3DERRORSTRING(hr);
    }
  }
}

/**
 * Prepare a display region for rendering (set up scissor region and viewport)
 */
void DXGraphicsStateGuardian9::
prepare_display_region(DisplayRegionPipelineReader *dr) {
  nassertv(dr != nullptr);
  GraphicsStateGuardian::prepare_display_region(dr);

  int l, u, w, h;
  dr->get_region_pixels_i(l, u, w, h);

  // Create the viewport
  D3DVIEWPORT9 vp = { (DWORD)l, (DWORD)u, (DWORD)w, (DWORD)h, 0.0f, 1.0f };
  _current_viewport = vp;
  HRESULT hr = _d3d_device->SetViewport(&_current_viewport);
  if (FAILED(hr)) {
    dxgsg9_cat.error()
      << "_screen->_swap_chain = " << _screen->_swap_chain << " _swap_chain = " << _swap_chain << "\n";
    dxgsg9_cat.error()
      << "SetViewport(" << l << ", " << u << ", " << w << ", " << h
      << ") failed" << D3DERRORSTRING(hr);

    D3DVIEWPORT9 vp_old;
    _d3d_device->GetViewport(&vp_old);
    dxgsg9_cat.error()
      << "GetViewport(" << vp_old.X << ", " << vp_old.Y << ", " << vp_old.Width << ", "
      << vp_old.Height << ") returned: Trying to set that vp---->\n";
    hr = _d3d_device->SetViewport(&vp_old);
    set_render_state(D3DRS_SCISSORTESTENABLE, FALSE);

    if (FAILED(hr)) {
      dxgsg9_cat.error() << "Failed again\n";
      throw_event("panda3d-render-error");
      nassertv(false);
    }
  }

  if (_screen->_can_direct_disable_color_writes) {
    set_render_state(D3DRS_COLORWRITEENABLE, _color_write_mask);
  }
}

/**
 * Given a lens, calculates the appropriate projection matrix for use with
 * this gsg.  Note that the projection matrix depends a lot upon the
 * coordinate system of the rendering API.
 *
 * The return value is a TransformState if the lens is acceptable, NULL if it
 * is not.
 */
CPT(TransformState) DXGraphicsStateGuardian9::
calc_projection_mat(const Lens *lens) {
  if (lens == nullptr) {
    return nullptr;
  }

  if (!lens->is_linear()) {
    return nullptr;
  }

  // DirectX also uses a Z range of 0 to 1, whereas the Panda convention is
  // for the projection matrix to produce a Z range of -1 to 1.  We have to
  // rescale to compensate.
  static const LMatrix4 rescale_mat
    (1, 0, 0, 0,
     0, 1, 0, 0,
     0, 0, 0.5, 0,
     0, 0, 0.5, 1);

  LMatrix4 result =
    LMatrix4::convert_mat(CS_yup_left, _current_lens->get_coordinate_system()) *
    lens->get_projection_mat(_current_stereo_channel) *
    rescale_mat;

  if (_scene_setup->get_inverted()) {
    // If the scene is supposed to be inverted, then invert the projection
    // matrix.
    result *= LMatrix4::scale_mat(1.0f, -1.0f, 1.0f);
  }

  return TransformState::make_mat(result);
}

/**
 * Makes the current lens (whichever lens was most recently specified with
 * set_scene()) active, so that it will transform future rendered geometry.
 * Normally this is only called from the draw process, and usually it is
 * called by set_scene().
 *
 * The return value is true if the lens is acceptable, false if it is not.
 */
bool DXGraphicsStateGuardian9::
prepare_lens() {
  LMatrix4f mat = LCAST(float, _projection_mat->get_mat());
  HRESULT hr =
    _d3d_device->SetTransform(D3DTS_PROJECTION,
                              (D3DMATRIX*)mat.get_data());
  return SUCCEEDED(hr);
}

/**
 * Called before each frame is rendered, to allow the GSG a chance to do any
 * internal cleanup before beginning the frame.
 *
 * The return value is true if successful (in which case the frame will be
 * drawn and end_frame() will be called later), or false if unsuccessful (in
 * which case nothing will be drawn and end_frame() will not be called).
 */
bool DXGraphicsStateGuardian9::
begin_frame(Thread *current_thread) {

  GraphicsStateGuardian::begin_frame(current_thread);

  if (_d3d_device == nullptr) {
    dxgsg9_cat.debug()
      << this << "::begin_frame(): no device.\n";
    return false;
  }

  HRESULT hr = _d3d_device->BeginScene();

  if (FAILED(hr)) {
    if (hr == D3DERR_DEVICELOST) {
      if (dxgsg9_cat.is_debug()) {
        dxgsg9_cat.debug()
          << "BeginScene returns D3DERR_DEVICELOST" << endl;
      }

      check_cooperative_level();

    } else {
      dxgsg9_cat.error()
        << "BeginScene failed, unhandled error hr == "
        << D3DERRORSTRING(hr) << endl;
      throw_event("panda3d-render-error");
    }
    return false;
  }

  if (_current_properties->get_srgb_color()) {
    set_render_state(D3DRS_SRGBWRITEENABLE, TRUE);
  } else {
    set_render_state(D3DRS_SRGBWRITEENABLE, FALSE);
  }

  return true;
}

/**
 * Called between begin_frame() and end_frame() to mark the beginning of
 * drawing commands for a "scene" (usually a particular DisplayRegion) within
 * a frame.  All 3-D drawing commands, except the clear operation, must be
 * enclosed within begin_scene() .. end_scene().
 *
 * The return value is true if successful (in which case the scene will be
 * drawn and end_scene() will be called later), or false if unsuccessful (in
 * which case nothing will be drawn and end_scene() will not be called).
 */
bool DXGraphicsStateGuardian9::
begin_scene() {
  if (!GraphicsStateGuardian::begin_scene()) {
    return false;
  }

/*
  HRESULT hr = _d3d_device->BeginScene();

  if (FAILED(hr)) {
    if (hr == D3DERR_DEVICELOST) {
      if (dxgsg9_cat.is_debug()) {
        dxgsg9_cat.debug()
          << "BeginScene returns D3DERR_DEVICELOST" << endl;
      }

      check_cooperative_level();

    } else {
      dxgsg9_cat.error()
        << "BeginScene failed, unhandled error hr == "
        << D3DERRORSTRING(hr) << endl;
      throw_event("panda3d-render-error");
    }
    return false;
  }
*/

  return true;
}

/**
 * Called between begin_frame() and end_frame() to mark the end of drawing
 * commands for a "scene" (usually a particular DisplayRegion) within a frame.
 * All 3-D drawing commands, except the clear operation, must be enclosed
 * within begin_scene() .. end_scene().
 */
void DXGraphicsStateGuardian9::
end_scene() {
  GraphicsStateGuardian::end_scene();

  if (_vertex_array_shader_context != 0) {
    _vertex_array_shader_context->disable_shader_vertex_arrays(this);
    _vertex_array_shader = nullptr;
    _vertex_array_shader_context = nullptr;
  }
  if (_texture_binding_shader_context != 0) {
    _texture_binding_shader_context->disable_shader_texture_bindings(this);
    _texture_binding_shader = nullptr;
    _texture_binding_shader_context = nullptr;
  }
  if (_current_shader_context != 0) {
    _current_shader_context->unbind(this);
    _current_shader = nullptr;
    _current_shader_context = nullptr;
  }

  _dlights.clear();

/*
  HRESULT hr = _d3d_device->EndScene();

  if (FAILED(hr)) {
    if (hr == D3DERR_DEVICELOST) {
      if (dxgsg9_cat.is_debug()) {
        dxgsg9_cat.debug()
          << "EndScene returns DeviceLost\n";
      }
      check_cooperative_level();

    } else {
      dxgsg9_cat.error()
        << "EndScene failed, unhandled error hr == " << D3DERRORSTRING(hr);
      throw_event("panda3d-render-error");
    }
    return;
  }
*/

}

/**
 * Called after each frame is rendered, to allow the GSG a chance to do any
 * internal cleanup after rendering the frame, and before the window flips.
 */
void DXGraphicsStateGuardian9::
end_frame(Thread *current_thread) {

  HRESULT hr = _d3d_device->EndScene();

  if (FAILED(hr)) {
    if (hr == D3DERR_DEVICELOST) {
      if (dxgsg9_cat.is_debug()) {
        dxgsg9_cat.debug()
          << "EndScene returns DeviceLost\n";
      }
      check_cooperative_level();

    } else {
      dxgsg9_cat.error()
        << "EndScene failed, unhandled error hr == " << D3DERRORSTRING(hr);
      throw_event("panda3d-render-error");
    }
    return;
  }

#if defined(DO_PSTATS)
  if (_texmgrmem_total_pcollector.is_active()) {
#define TICKS_PER_GETTEXINFO (2.5*1000)   // 2.5 second interval
    static DWORD last_tick_count = 0;
    DWORD cur_tick_count = GetTickCount();

    if (cur_tick_count - last_tick_count > TICKS_PER_GETTEXINFO) {
      last_tick_count = cur_tick_count;
      report_texmgr_stats();
    }
  }
#endif

  // Note: regular GraphicsWindow::end_frame is being called, but we override
  // gsg::end_frame, so need to explicitly call it here (currently it's an
  // empty fn)
  GraphicsStateGuardian::end_frame(current_thread);
}

/**
 * Called before a sequence of draw_primitive() functions are called, this
 * should prepare the vertex data for rendering.  It returns true if the
 * vertices are ok, false to abort this group of primitives.
 */
bool DXGraphicsStateGuardian9::
begin_draw_primitives(const GeomPipelineReader *geom_reader,
                      const GeomVertexDataPipelineReader *data_reader,
                      bool force) {
  if (!GraphicsStateGuardian::begin_draw_primitives(geom_reader, data_reader, force)) {
    return false;
  }
  nassertr(_data_reader != nullptr, false);

  const GeomVertexFormat *format = _data_reader->get_format();

  const GeomVertexAnimationSpec &animation =
    data_reader->get_format()->get_animation();
  if (animation.get_animation_type() == Geom::AT_hardware) {
    // Set up vertex blending.
    switch (animation.get_num_transforms()) {
    case 1:
      // The MSDN docs suggest we should use D3DVBF_0WEIGHTS here, but that
      // doesn't seem to work at all.  On the other hand, D3DVBF_DISABLE
      // *does* work, because it disables special handling, meaning only the
      // world matrix affects these vertices--and by accident or design, the
      // first matrix, D3DTS_WORLDMATRIX(0), *is* the world matrix.
      set_render_state(D3DRS_VERTEXBLEND, D3DVBF_DISABLE);
      break;
    case 2:
      set_render_state(D3DRS_VERTEXBLEND, D3DVBF_1WEIGHTS);
      break;
    case 3:
      set_render_state(D3DRS_VERTEXBLEND, D3DVBF_2WEIGHTS);
      break;
    case 4:
      set_render_state(D3DRS_VERTEXBLEND, D3DVBF_3WEIGHTS);
      break;
    }

    if (animation.get_indexed_transforms()) {
      // Set up indexed vertex blending.
      set_render_state(D3DRS_INDEXEDVERTEXBLENDENABLE, TRUE);
    } else {
      set_render_state(D3DRS_INDEXEDVERTEXBLENDENABLE, FALSE);
    }

    const TransformTable *table = data_reader->get_transform_table();
    if (table != nullptr) {
      for (size_t i = 0; i < table->get_num_transforms(); ++i) {
        LMatrix4 mat;
        table->get_transform(i)->mult_matrix(mat, _internal_transform->get_mat());
        const D3DMATRIX *d3d_mat = (const D3DMATRIX *)mat.get_data();
        _d3d_device->SetTransform(D3DTS_WORLDMATRIX(i), d3d_mat);
      }

      // Setting the first animation matrix steps on the world matrix, so we
      // have to set a flag to reload the world matrix later.
      _transform_stale = true;
    }
    _vertex_blending_enabled = true;

  } else {
    // We're not using vertex blending.
    if (_vertex_blending_enabled) {
      set_render_state(D3DRS_INDEXEDVERTEXBLENDENABLE, FALSE);
      set_render_state(D3DRS_VERTEXBLEND, D3DVBF_DISABLE);
      _vertex_blending_enabled = false;
    }

    if (_transform_stale && !_data_reader->is_vertex_transformed()) {
      const D3DMATRIX *d3d_mat = (const D3DMATRIX *)_internal_transform->get_mat().get_data();
      _d3d_device->SetTransform(D3DTS_WORLD, d3d_mat);
      _transform_stale = false;
    }
  }

  if (_data_reader->is_vertex_transformed()) {
    // If the vertex data claims to be already transformed into clip
    // coordinates, wipe out the current projection and modelview matrix (so
    // we don't attempt to transform it again).

    // It's tempting just to use the D3DFVF_XYZRHW specification on these
    // vertices, but that turns out to be a bigger hammer than we want: that
    // also prevents lighting calculations and user clip planes.
    _d3d_device->SetTransform(D3DTS_WORLD, &_d3d_ident_mat);
    static const LMatrix4f rescale_mat
      (1, 0, 0, 0,
       0, 1, 0, 0,
       0, 0, 0.5, 0,
       0, 0, 0.5, 1);
    _transform_stale = true;

    _d3d_device->SetTransform(D3DTS_PROJECTION, (const D3DMATRIX *)rescale_mat.get_data());
  }

  if (_current_shader_context == 0 /*|| !_current_shader_context->uses_custom_vertex_arrays()*/) {
    // No shader, or a non-Cg shader.
    if (_vertex_array_shader_context != 0) {
      _vertex_array_shader_context->disable_shader_vertex_arrays(this);
    }
    if (!update_standard_vertex_arrays(force)) {
      return false;
    }
  } else {
    // Cg shader.
    if (_vertex_array_shader_context == 0) {
      disable_standard_vertex_arrays();
      if (!_current_shader_context->update_shader_vertex_arrays(nullptr, this, force)) {
        return false;
      }
    } else {
      if (!_current_shader_context->
          update_shader_vertex_arrays(_vertex_array_shader_context, this, force)) {
        return false;
      }
    }
  }

  _vertex_array_shader = _current_shader;
  _vertex_array_shader_context = _current_shader_context;

  return true;
}

/**
 * Binds vertex buffers as stream sources and sets the correct FVF format for
 * fixed-function rendering.  Used only when the standard (non-shader)
 * pipeline is about to be used - dxShaderContexts are responsible for setting
 * up their own vertex arrays.
 */
bool DXGraphicsStateGuardian9::
update_standard_vertex_arrays(bool force) {

  int fvf = 0;
  HRESULT hr;

  int number_of_arrays = _data_reader->get_num_arrays();
  for ( int array_index = 0; array_index < number_of_arrays; ++array_index ) {
    const GeomVertexArrayDataHandle* array_reader = _data_reader->get_array_reader( array_index );
    if ( array_reader == nullptr ) {
      dxgsg9_cat.error() << "Unable to get reader for array " << array_index << "\n";
      return false;
    }

    // Get the vertex buffer for this array.
    DXVertexBufferContext9* dvbc;
    if (!setup_array_data(dvbc, array_reader, force)) {
      dxgsg9_cat.error() << "Unable to setup vertex buffer for array " << array_index << "\n";
      return false;
    }

    // Bind this array as the data source for the corresponding stream.
    const GeomVertexArrayFormat* array_format = array_reader->get_array_format();
    hr = _d3d_device->SetStreamSource( array_index, dvbc->_vbuffer, 0, array_format->get_stride() );
    if (FAILED(hr)) {
      dxgsg9_cat.error() << "SetStreamSource failed" << D3DERRORSTRING(hr);
      return false;
    }

    // Update our combined set of FVF flags
    fvf |= dvbc->_fvf;
  }

  hr = _d3d_device->SetFVF( fvf );
  if (FAILED(hr)) {
    dxgsg9_cat.error() << "SetFVF failed" << D3DERRORSTRING(hr);
    return false;
  }

  return true;
}

/**
 * Unbinds all of the streams that are currently enabled.  dxShaderContexts
 * are responsible for setting up their own streams, but before they can do
 * so, the standard streams need to be disabled to get them "out of the way."
 * Called only from begin_draw_primitives.
 */
void DXGraphicsStateGuardian9::
disable_standard_vertex_arrays() {
  for ( int array_index = 0; array_index < _num_bound_streams; ++array_index )
  {
    _d3d_device->SetStreamSource( array_index, nullptr, 0, 0 );
  }
  _num_bound_streams = 0;
}

/**
 * Draws a series of disconnected triangles.
 */
bool DXGraphicsStateGuardian9::
draw_triangles(const GeomPrimitivePipelineReader *reader, bool force) {
  // PStatTimer timer(_draw_primitive_pcollector);

  _vertices_tri_pcollector.add_level(reader->get_num_vertices());
  _primitive_batches_tri_pcollector.add_level(1);

  if (reader->is_indexed()) {
    int min_vertex = dx_broken_max_index ? 0 : reader->get_min_vertex();
    int max_vertex = reader->get_max_vertex();

    // Indexed, vbuffers.
    IndexBufferContext *ibc = ((GeomPrimitive *)(reader->get_object()))->prepare_now(get_prepared_objects(), this);
    nassertr(ibc != nullptr, false);
    if (!apply_index_buffer(ibc, reader, force)) {
      return false;
    }

    _d3d_device->DrawIndexedPrimitive( D3DPT_TRIANGLELIST,
                                       0,
                                       min_vertex, max_vertex - min_vertex + 1,
                                       0, reader->get_num_primitives() );

    #if 0
    // Indexed, client arrays.
    const unsigned char *index_pointer = reader->get_read_pointer(force);
    if (index_pointer == nullptr) {
      return false;
    }
    D3DFORMAT index_type = get_index_type(reader->get_index_type());
    const unsigned char *vertex_pointer = _data_reader->get_array_reader(0)->get_read_pointer(force);
    if (vertex_pointer == nullptr) {
      return false;
    }

    draw_indexed_primitive_up( D3DPT_TRIANGLELIST,
                               min_vertex, max_vertex,
                               reader->get_num_primitives(),
                               index_pointer, index_type, vertex_pointer,
                               _data_reader->get_format()->get_array(0)->get_stride() );
    #endif
  } else {
    // Nonindexed, vbuffers.
    _d3d_device->DrawPrimitive( D3DPT_TRIANGLELIST,
                                reader->get_first_vertex(),
                                reader->get_num_primitives() );

    #if 0
    // Nonindexed, client arrays.
    const unsigned char *vertex_pointer = _data_reader->get_array_reader(0)->get_read_pointer(force);
    if (vertex_pointer == nullptr) {
      return false;
    }

    draw_primitive_up(D3DPT_TRIANGLELIST, reader->get_num_primitives(),
    reader->get_first_vertex(),
    reader->get_num_vertices(), vertex_pointer,
    _data_reader->get_format()->get_array(0)->get_stride());
    #endif
  }

  return true;
}

/**
 * Draws a series of triangle strips.
 */
bool DXGraphicsStateGuardian9::
draw_tristrips(const GeomPrimitivePipelineReader *reader, bool force) {
  // PStatTimer timer(_draw_primitive_pcollector);

  if (connect_triangle_strips && _current_fill_mode != RenderModeAttrib::M_wireframe) {
    // One long triangle strip, connected by the degenerate vertices that have
    // already been set up within the primitive.
    _vertices_tristrip_pcollector.add_level(reader->get_num_vertices());
    _primitive_batches_tristrip_pcollector.add_level(1);

    if (reader->is_indexed()) {
      int min_vertex = dx_broken_max_index ? 0 : reader->get_min_vertex();
      int max_vertex = reader->get_max_vertex();

      // Indexed, vbuffers, one long triangle strip.
      IndexBufferContext *ibc = ((GeomPrimitive *)(reader->get_object()))->prepare_now(get_prepared_objects(), this);
      nassertr(ibc != nullptr, false);
      if (!apply_index_buffer(ibc, reader, force)) {
        return false;
      }

      _d3d_device->DrawIndexedPrimitive( D3DPT_TRIANGLESTRIP,
                                         0,
                                         min_vertex, max_vertex - min_vertex + 1,
                                         0, reader->get_num_vertices() - 2 );

      #if 0
      // Indexed, client arrays, one long triangle strip.
      const unsigned char *index_pointer = reader->get_read_pointer(force);
      if (index_pointer == nullptr) {
        return false;
      }
      D3DFORMAT index_type = get_index_type(reader->get_index_type());
      const unsigned char *vertex_pointer = _data_reader->get_array_reader(0)->get_read_pointer(force);
      if (vertex_pointer == nullptr) {
        return false;
      }

      draw_indexed_primitive_up
        (D3DPT_TRIANGLESTRIP,
         min_vertex, max_vertex,
         reader->get_num_vertices() - 2,
         index_pointer, index_type, vertex_pointer,
         _data_reader->get_format()->get_array(0)->get_stride());
      #endif
    } else {
      // Nonindexed, vbuffers, one long triangle strip.
      _d3d_device->DrawPrimitive( D3DPT_TRIANGLESTRIP,
                                  reader->get_first_vertex(),
                                  reader->get_num_vertices() - 2 );

      #if 0
      // Indexed, client arrays, one long triangle strip.
      const unsigned char *vertex_pointer = _data_reader->get_array_reader(0)->get_read_pointer(force);
      if (vertex_pointer == nullptr) {
        return false;
      }
      draw_primitive_up(D3DPT_TRIANGLESTRIP,
      reader->get_num_vertices() - 2,
      reader->get_first_vertex(),
      reader->get_num_vertices(), vertex_pointer,
      _data_reader->get_format()->get_array(0)->get_stride());
      #endif
    }

  } else {
    // Send the individual triangle strips, stepping over the degenerate
    // vertices.
    CPTA_int ends = reader->get_ends();
    _primitive_batches_tristrip_pcollector.add_level(ends.size());

    if (reader->is_indexed()) {
      CPTA_int ends = reader->get_ends();
      int index_stride = reader->get_index_stride();
      _primitive_batches_tristrip_pcollector.add_level(ends.size());

      GeomVertexReader mins(reader->get_mins(), 0);
      GeomVertexReader maxs(reader->get_maxs(), 0);
      nassertr(reader->get_mins()->get_num_rows() == (int)ends.size() &&
               reader->get_maxs()->get_num_rows() == (int)ends.size(), false);

      // Indexed, vbuffers, individual triangle strips.
      IndexBufferContext *ibc = ((GeomPrimitive *)(reader->get_object()))->prepare_now(get_prepared_objects(), this);
      nassertr(ibc != nullptr, false);
      if (!apply_index_buffer(ibc, reader, force)) {
        return false;
      }

      unsigned int start = 0;
      for (size_t i = 0; i < ends.size(); i++) {
        _vertices_tristrip_pcollector.add_level(ends[i] - start);
        unsigned int min = mins.get_data1i();
        unsigned int max = maxs.get_data1i();
        _d3d_device->DrawIndexedPrimitive( D3DPT_TRIANGLESTRIP,
                                           0,
                                           min, max - min + 1,
                                           start, ends[i] - start - 2 );
        start = ends[i] + 2;
      }

      #if 0
      // Indexed, client arrays, individual triangle strips.
      int stride = _data_reader->get_format()->get_array(0)->get_stride();
      const unsigned char *index_pointer = reader->get_read_pointer(force);
      if (index_pointer == nullptr) {
        return false;
      }
      D3DFORMAT index_type = get_index_type(reader->get_index_type());
      const unsigned char *vertex_pointer = _data_reader->get_array_reader(0)->get_read_pointer(force);
      if (vertex_pointer == nullptr) {
        return false;
      }

      unsigned int start = 0;
      for (size_t i = 0; i < ends.size(); i++) {
        _vertices_tristrip_pcollector.add_level(ends[i] - start);
        unsigned int min = mins.get_data1i();
        unsigned int max = maxs.get_data1i();
        draw_indexed_primitive_up
          (D3DPT_TRIANGLESTRIP,
           min, max,
           ends[i] - start - 2,
           index_pointer + start * index_stride, index_type,
           vertex_pointer, stride);

        start = ends[i] + 2;
      }
      #endif
    } else {
      unsigned int first_vertex = reader->get_first_vertex();

      // Nonindexed, vbuffers, individual triangle strips.
      unsigned int start = 0;
      for (size_t i = 0; i < ends.size(); i++) {
        _vertices_tristrip_pcollector.add_level(ends[i] - start);
        _d3d_device->DrawPrimitive( D3DPT_TRIANGLESTRIP,
                                    first_vertex + start,
                                    ends[i] - start - 2 );
        start = ends[i] + 2;
      }

      #if 0
      // Nonindexed, client arrays, individual triangle strips.
      const unsigned char *vertex_pointer = _data_reader->get_array_reader(0)->get_read_pointer(force);
      if (vertex_pointer == nullptr) {
        return false;
      }
      int stride = _data_reader->get_format()->get_array(0)->get_stride();

      unsigned int start = 0;
      for (size_t i = 0; i < ends.size(); i++) {
        _vertices_tristrip_pcollector.add_level(ends[i] - start);
        draw_primitive_up(D3DPT_TRIANGLESTRIP, ends[i] - start - 2,
        first_vertex + start,
        ends[i] - start,
        vertex_pointer, stride);

        start = ends[i] + 2;
      }
      #endif
    }
  }
  return true;
}

/**
 * Draws a series of triangle fans.
 */
bool DXGraphicsStateGuardian9::
draw_trifans(const GeomPrimitivePipelineReader *reader, bool force) {
  // PStatTimer timer(_draw_primitive_pcollector);

  CPTA_int ends = reader->get_ends();
  _primitive_batches_trifan_pcollector.add_level(ends.size());

  if (reader->is_indexed()) {
    int min_vertex = dx_broken_max_index ? 0 : reader->get_min_vertex();
    int max_vertex = reader->get_max_vertex();

    // Send the individual triangle fans.  There's no connecting fans with
    // degenerate vertices, so no worries about that.
    int index_stride = reader->get_index_stride();

    GeomVertexReader mins(reader->get_mins(), 0);
    GeomVertexReader maxs(reader->get_maxs(), 0);
    nassertr(reader->get_mins()->get_num_rows() == (int)ends.size() &&
             reader->get_maxs()->get_num_rows() == (int)ends.size(), false);

    // Indexed, vbuffers.
    IndexBufferContext *ibc = ((GeomPrimitive *)(reader->get_object()))->prepare_now(get_prepared_objects(), this);
    nassertr(ibc != nullptr, false);
    if (!apply_index_buffer(ibc, reader, force)) {
      return false;
    }

    unsigned int start = 0;
    for (size_t i = 0; i < ends.size(); i++) {
      _vertices_trifan_pcollector.add_level(ends[i] - start);
      unsigned int min = mins.get_data1i();
      unsigned int max = maxs.get_data1i();
      _d3d_device->DrawIndexedPrimitive( D3DPT_TRIANGLEFAN,
                                         0,
                                         min, max - min + 1,
                                         start, ends[i] - start - 2 );
      start = ends[i];
    }

    #if 0
    // Indexed, client arrays.
    int stride = _data_reader->get_format()->get_array(0)->get_stride();
    const unsigned char *index_pointer = reader->get_read_pointer(force);
    if (index_pointer == nullptr) {
      return false;
    }
    D3DFORMAT index_type = get_index_type(reader->get_index_type());
    const unsigned char *vertex_pointer = _data_reader->get_array_reader(0)->get_read_pointer(force);
    if (vertex_pointer == nullptr) {
      return false;
    }

    unsigned int start = 0;
    for (size_t i = 0; i < ends.size(); i++) {
      _vertices_trifan_pcollector.add_level(ends[i] - start);
      unsigned int min = mins.get_data1i();
      unsigned int max = maxs.get_data1i();
      draw_indexed_primitive_up
        (D3DPT_TRIANGLEFAN,
         min, max,
         ends[i] - start - 2,
         index_pointer + start * index_stride, index_type,
         vertex_pointer, stride);

      start = ends[i];
    }
    #endif
  } else {
    unsigned int first_vertex = reader->get_first_vertex();

    // Nonindexed, vbuffers.
    unsigned int start = 0;
    for (size_t i = 0; i < ends.size(); i++) {
      _vertices_trifan_pcollector.add_level(ends[i] - start);
      _d3d_device->DrawPrimitive( D3DPT_TRIANGLEFAN,
                                  first_vertex + start,
                                  ends[i] - start - 2 );
      start = ends[i];
    }

    #if 0
    // Nonindexed, client arrays.
    const unsigned char *vertex_pointer = _data_reader->get_array_reader(0)->get_read_pointer(force);
    if (vertex_pointer == nullptr) {
      return false;
    }
    int stride = _data_reader->get_format()->get_array(0)->get_stride();

    unsigned int start = 0;
    for (size_t i = 0; i < ends.size(); i++) {
      _vertices_trifan_pcollector.add_level(ends[i] - start);
      draw_primitive_up(D3DPT_TRIANGLEFAN,
      ends[i] - start - 2,
      first_vertex,
      ends[i] - start,
      vertex_pointer, stride);
      start = ends[i];
    }
    #endif
  }
  return true;
}

/**
 * Draws a series of disconnected line segments.
 */
bool DXGraphicsStateGuardian9::
draw_lines(const GeomPrimitivePipelineReader *reader, bool force) {
  // PStatTimer timer(_draw_primitive_pcollector);

  _vertices_other_pcollector.add_level(reader->get_num_vertices());
  _primitive_batches_other_pcollector.add_level(1);

  if (reader->is_indexed()) {
    int min_vertex = dx_broken_max_index ? 0 : reader->get_min_vertex();
    int max_vertex = reader->get_max_vertex();

    // Indexed, vbuffers.
    IndexBufferContext *ibc = ((GeomPrimitive *)(reader->get_object()))->prepare_now(get_prepared_objects(), this);
    nassertr(ibc != nullptr, false);
    if (!apply_index_buffer(ibc, reader, force)) {
      return false;
    }

    _d3d_device->DrawIndexedPrimitive( D3DPT_LINELIST,
                                       0,
                                       min_vertex, max_vertex - min_vertex + 1,
                                       0, reader->get_num_primitives() );

    #if 0
    // Indexed, client arrays.
    const unsigned char *index_pointer = reader->get_read_pointer(force);
    if (index_pointer == nullptr) {
      return false;
    }
    D3DFORMAT index_type = get_index_type(reader->get_index_type());
    const unsigned char *vertex_pointer = _data_reader->get_array_reader(0)->get_read_pointer(force);
    if (vertex_pointer == nullptr) {
      return false;
    }

    draw_indexed_primitive_up
      (D3DPT_LINELIST,
       min_vertex, max_vertex,
       reader->get_num_primitives(),
       index_pointer, index_type, vertex_pointer,
       _data_reader->get_format()->get_array(0)->get_stride());
    #endif
  } else {
    // Nonindexed, vbuffers.
    _d3d_device->DrawPrimitive( D3DPT_LINELIST,
                                reader->get_first_vertex(),
                                reader->get_num_primitives() );

    #if 0
    // Nonindexed, client arrays.
    const unsigned char *vertex_pointer = _data_reader->get_array_reader(0)->get_read_pointer(force);
    if (vertex_pointer == nullptr) {
      return false;
    }
    draw_primitive_up(D3DPT_LINELIST, reader->get_num_primitives(),
    reader->get_first_vertex(),
    reader->get_num_vertices(), vertex_pointer,
    _data_reader->get_format()->get_array(0)->get_stride());
    #endif
  }
  return true;
}

/**
 * Draws a series of line strips.
 */
bool DXGraphicsStateGuardian9::
draw_linestrips(const GeomPrimitivePipelineReader *reader, bool force) {
  return false;
}

/**
 * Draws a series of disconnected points.
 */
bool DXGraphicsStateGuardian9::
draw_points(const GeomPrimitivePipelineReader *reader, bool force) {
  // PStatTimer timer(_draw_primitive_pcollector);

  _vertices_other_pcollector.add_level(reader->get_num_vertices());
  _primitive_batches_other_pcollector.add_level(1);

  // The munger should have protected us from indexed points--DirectX doesn't
  // support them.
  nassertr(!reader->is_indexed(), false);

  // Nonindexed, vbuffers.
  _d3d_device->DrawPrimitive( D3DPT_POINTLIST,
                              reader->get_first_vertex(),
                              reader->get_num_primitives() );

  #if 0
  // Nonindexed, client arrays.
  const unsigned char *vertex_pointer = _data_reader->get_array_reader(0)->get_read_pointer(force);
  if (vertex_pointer == nullptr) {
    return false;
  }
  draw_primitive_up(D3DPT_POINTLIST, reader->get_num_primitives(),
                    reader->get_first_vertex(),
                    reader->get_num_vertices(), vertex_pointer,
                    _data_reader->get_format()->get_array(0)->get_stride());
  #endif

  return true;
}

/**
 * Called after a sequence of draw_primitive() functions are called, this
 * should do whatever cleanup is appropriate.
 */
void DXGraphicsStateGuardian9::
end_draw_primitives() {
  // Turn off vertex blending--it seems to cause problems if we leave it on.
  if (_vertex_blending_enabled) {
    set_render_state(D3DRS_INDEXEDVERTEXBLENDENABLE, FALSE);
    set_render_state(D3DRS_VERTEXBLEND, D3DVBF_DISABLE);
    _vertex_blending_enabled = false;
  }

  if (_data_reader->is_vertex_transformed()) {
    // Restore the projection matrix that we wiped out above.
    LMatrix4f mat = LCAST(float, _projection_mat->get_mat());
    _d3d_device->SetTransform(D3DTS_PROJECTION,
                              (D3DMATRIX*)mat.get_data());
  }

  GraphicsStateGuardian::end_draw_primitives();
}

/**
 * Copy the pixels within the indicated display region from the framebuffer
 * into texture memory.
 *
 * If z > -1, it is the cube map index into which to copy.
 */
bool DXGraphicsStateGuardian9::
framebuffer_copy_to_texture(Texture *tex, int view, int z,
                            const DisplayRegion *dr, const RenderBuffer &rb) {
  set_read_buffer(rb);

  int orig_x = tex->get_x_size();
  int orig_y = tex->get_y_size();

  HRESULT hr;
  int xo, yo, w, h;
  dr->get_region_pixels_i(xo, yo, w, h);
  tex->set_size_padded(w, h);

  // must use a render target type texture for StretchRect
  tex->set_render_to_texture(true);

  TextureContext *tc = tex->prepare_now(view, get_prepared_objects(), this);
  if (tc == nullptr) {
    return false;
  }
  DXTextureContext9 *dtc = DCAST(DXTextureContext9, tc);
  if (!dtc->create_texture(*_screen)) {
    // Oops, we can't re-create the texture for some reason.
    dxgsg9_cat.error()
      << "Unable to re-create texture " << *dtc->get_texture() << endl;
    return false;
  }

  if (tex->get_texture_type() != Texture::TT_2d_texture) {
    // For a specialty texture like a cube map, go the slow route through RAM
    // for now.
    return do_framebuffer_copy_to_ram(tex, view, z, dr, rb, true);
  }
  nassertr(dtc->get_d3d_2d_texture() != nullptr, false);

  IDirect3DSurface9 *tex_level_0;
  hr = dtc->get_d3d_2d_texture()->GetSurfaceLevel(0, &tex_level_0);
  if (FAILED(hr)) {
    dxgsg9_cat.error() << "GetSurfaceLev failed in copy_texture" << D3DERRORSTRING(hr);
    return false;
  }

  // If the texture is the wrong size, we need to do something about it.
  D3DSURFACE_DESC texdesc;
  hr = tex_level_0->GetDesc(&texdesc);
  if (FAILED(hr)) {
    dxgsg9_cat.error() << "GetDesc failed in copy_texture" << D3DERRORSTRING(hr);
    SAFE_RELEASE(tex_level_0);
    return false;
  }
  if ((texdesc.Width != tex->get_x_size())||(texdesc.Height != tex->get_y_size())) {
    if ((orig_x != tex->get_x_size()) || (orig_y != tex->get_y_size())) {
      // Texture might be wrong size because we resized it and need to
      // recreate.
      SAFE_RELEASE(tex_level_0);
      if (!dtc->create_texture(*_screen)) {
        // Oops, we can't re-create the texture for some reason.
        dxgsg9_cat.error()
          << "Unable to re-create texture " << *dtc->get_texture() << endl;
        return false;
      }
      hr = dtc->get_d3d_2d_texture()->GetSurfaceLevel(0, &tex_level_0);
      if (FAILED(hr)) {
        dxgsg9_cat.error() << "GetSurfaceLev failed in copy_texture" << D3DERRORSTRING(hr);
        return false;
      }
      hr = tex_level_0->GetDesc(&texdesc);
      if (FAILED(hr)) {
        dxgsg9_cat.error() << "GetDesc 2 failed in copy_texture" << D3DERRORSTRING(hr);
        SAFE_RELEASE(tex_level_0);
        return false;
      }
    }
    if ((texdesc.Width != tex->get_x_size())||(texdesc.Height != tex->get_y_size())) {
      // If it's still the wrong size, it's because driver can't create size
      // that we want.  In that case, there's no helping it, we have to give
      // up.
      dxgsg9_cat.error()
        << "Unable to copy to texture, texture is wrong size: " << *dtc->get_texture() << endl;
      SAFE_RELEASE(tex_level_0);
      return false;
    }
  }

  DWORD render_target_index;
  IDirect3DSurface9 *render_target;

  /* ***** DX9 GetRenderTarget, assume only one render target so index = 0 */
  render_target_index = 0;

  hr = _d3d_device->GetRenderTarget(render_target_index, &render_target);
  if (FAILED(hr)) {
    dxgsg9_cat.error()
      << "GetRenderTarget failed in framebuffer_copy_to_texture"
      << D3DERRORSTRING(hr);
    SAFE_RELEASE(tex_level_0);
    return false;
  }

  RECT src_rect;

  src_rect.left = xo;
  src_rect.right = xo+w;
  src_rect.top = yo;
  src_rect.bottom = yo+h;

// THE DX8 WAY hr = _d3d_device->CopyRects(render_target, &src_rect, 1,
// tex_level_0, 0);

// DX9
  D3DTEXTUREFILTERTYPE filter;

  filter = D3DTEXF_POINT;

  bool okflag = true;
  hr = _d3d_device->StretchRect(render_target, &src_rect,
                                tex_level_0, &src_rect,
                                filter);
  if (FAILED(hr)) {
    dxgsg9_cat.debug()
      << "StretchRect failed in framebuffer_copy_to_texture"
      << D3DERRORSTRING(hr);
    okflag = false;
  }

  SAFE_RELEASE(render_target);
  SAFE_RELEASE(tex_level_0);

  if (okflag) {
    dtc->mark_loaded();
    dtc->enqueue_lru(&_prepared_objects->_graphics_memory_lru);

  } else {
    // The copy failed.  Fall back to copying it to RAM and back.  Terribly
    // slow, but what are you going to do?
    return do_framebuffer_copy_to_ram(tex, view, z, dr, rb, true);
  }

  return true;
}


/**
 * Copy the pixels within the indicated display region from the framebuffer
 * into system memory, not texture memory.  Returns true on success, false on
 * failure.
 *
 * This completely redefines the ram image of the indicated texture.
 */
bool DXGraphicsStateGuardian9::
framebuffer_copy_to_ram(Texture *tex, int view, int z,
                        const DisplayRegion *dr, const RenderBuffer &rb) {
  return do_framebuffer_copy_to_ram(tex, view, z, dr, rb, false);
}

/**
 * This is the implementation of framebuffer_copy_to_ram(); it adds one
 * additional parameter, which should be true if the framebuffer is to be
 * inverted during the copy (as in the same way it copies to texture memory).
 */
bool DXGraphicsStateGuardian9::
do_framebuffer_copy_to_ram(Texture *tex, int view, int z,
                           const DisplayRegion *dr, const RenderBuffer &rb,
                           bool inverted) {
  set_read_buffer(rb);

  RECT rect;
  nassertr(tex != nullptr && dr != nullptr, false);

  int xo, yo, w, h;
  dr->get_region_pixels_i(xo, yo, w, h);

  Texture::Format format = tex->get_format();
  Texture::ComponentType component_type = tex->get_component_type();

  switch (format) {
  case Texture::F_depth_stencil:
    // Sorry, not (yet?) supported in pandadx.
    return false;

  default:
    format = Texture::F_rgb;
    component_type = Texture::T_unsigned_byte;
  }

  Texture::TextureType texture_type;
  if (z >= 0) {
    texture_type = Texture::TT_cube_map;
  } else {
    texture_type = Texture::TT_2d_texture;
  }

  if (tex->get_x_size() != w || tex->get_y_size() != h ||
      tex->get_component_type() != component_type ||
      tex->get_format() != format ||
      tex->get_texture_type() != texture_type) {
    // Re-setup the texture; its properties have changed.
    tex->setup_texture(texture_type, w, h, tex->get_z_size(),
                       component_type, format);
  }

  rect.top = yo;
  rect.left = xo;
  rect.right = xo + w;
  rect.bottom = yo + h;
  bool copy_inverted = false;

  IDirect3DSurface9 *temp_surface = nullptr;
  HRESULT hr;

  // Note if you try to grab the backbuffer and full-screen anti-aliasing is
  // on, the backbuffer might be larger than the window size.  For screenshots
  // it's safer to get the front buffer.
  if (_cur_read_pixel_buffer & RenderBuffer::T_back) {
    DWORD render_target_index;
    IDirect3DSurface9 *backbuffer = nullptr;
    // GetRenderTarget() seems to be a little more reliable than
    // GetBackBuffer().  Might just be related to the swap_chain thing.

    render_target_index = 0;
    hr = _d3d_device->GetRenderTarget(render_target_index, &backbuffer);

    if (FAILED(hr)) {
      dxgsg9_cat.error() << "GetRenderTarget failed" << D3DERRORSTRING(hr);
      return false;
    }

    // Since we might not be able to Lock the back buffer, we will need to
    // copy it to a temporary surface of the appropriate type first.
    D3DPOOL pool;
    D3DSURFACE_DESC surface_description;

    backbuffer -> GetDesc (&surface_description);

    pool = D3DPOOL_SYSTEMMEM;
    hr = _d3d_device->CreateOffscreenPlainSurface(
                                                  surface_description.Width,
                                                  surface_description.Height,
                                                  surface_description.Format,
                                                  pool,
                                                  &temp_surface,
                                                  nullptr);
    if (FAILED(hr)) {
      dxgsg9_cat.error()
        << "CreateImageSurface failed in copy_pixel_buffer()"
        << D3DERRORSTRING(hr);
      backbuffer->Release();
      return false;
    }

    // Now we must copy from the backbuffer to our temporary surface.
    hr = _d3d_device -> GetRenderTargetData (backbuffer, temp_surface);
    if (FAILED(hr)) {
      dxgsg9_cat.error() << "GetRenderTargetData failed" << D3DERRORSTRING(hr);
      temp_surface->Release();
      backbuffer->Release();
      return false;
    }

    copy_inverted = true;

    RELEASE(backbuffer, dxgsg9, "backbuffer", RELEASE_ONCE);

  } else if (_cur_read_pixel_buffer & RenderBuffer::T_front) {

    if (_screen->_presentation_params.Windowed) {
      // GetFrontBuffer() retrieves the entire desktop for a monitor, so we
      // need to reserve space for that.

      // We have to use GetMonitorInfo(), since this GSG may not be for the
      // primary monitor.
      MONITORINFO minfo;
      minfo.cbSize = sizeof(MONITORINFO);
      GetMonitorInfo(_screen->_monitor, &minfo);

      w = RECT_XSIZE(minfo.rcMonitor);
      h = RECT_YSIZE(minfo.rcMonitor);

      // set rect to client area of window in scrn coords
      ClientToScreen(_screen->_window, (POINT*)&rect.left);
      ClientToScreen(_screen->_window, (POINT*)&rect.right);
    }

    // For GetFrontBuffer(), we need a temporary surface of type A8R8G8B8.
    // Unlike GetBackBuffer(), GetFrontBuffer() implicitly performs a copy.
    hr = _d3d_device->CreateOffscreenPlainSurface(w, h, D3DFMT_A8R8G8B8, D3DPOOL_SCRATCH, &temp_surface, nullptr);
    if (FAILED(hr)) {
      dxgsg9_cat.error()
        << "CreateImageSurface failed in copy_pixel_buffer()"
        << D3DERRORSTRING(hr);
      return false;
    }

    UINT swap_chain;

    swap_chain = 0;
    hr = _d3d_device->GetFrontBufferData(swap_chain,temp_surface);

    if (hr == D3DERR_DEVICELOST) {
      dxgsg9_cat.error()
        << "copy_pixel_buffer failed: device lost\n";
      temp_surface->Release();
      return false;
    }

    copy_inverted = true;

  } else {
    dxgsg9_cat.error()
      << "copy_pixel_buffer: unhandled current_read_pixel_buffer type\n";
    temp_surface->Release();
    return false;
  }

  if (inverted) {
    copy_inverted = !copy_inverted;
  }
  DXTextureContext9::d3d_surface_to_texture(rect, temp_surface,
                                            copy_inverted, tex, view, z);

  RELEASE(temp_surface, dxgsg9, "temp_surface", RELEASE_ONCE);

  nassertr(tex->has_ram_image(), false);
  return true;
}

void DXGraphicsStateGuardian9::reset_render_states (void)
{
  int index;
  int maximum_texture_stages;

  maximum_texture_stages = D3D_MAXTEXTURESTAGES;

  // set to invalid values so that the state will always be set the first time
  memset (_render_state_array, -1, sizeof (_render_state_array));
  memset (_texture_stage_states_array, -1, sizeof (_texture_stage_states_array));

  // states that may be set intially to -1 by the user, so set it to D3D's
  // default value
  _render_state_array [D3DRS_FOGCOLOR] = 0;
  _render_state_array [D3DRS_AMBIENT] = 0;

  // set to D3D default values or invalid values so that the state will always
  // be set the first time
  memset (_texture_render_states_array, 0, sizeof (_texture_render_states_array));

  // states that may be set intially to 0 by the user, so set it to D3D's
  // default value
  for (index = 0; index < MAXIMUM_TEXTURES; index++) {
    TextureRenderStates *texture_render_states;

    texture_render_states = &_texture_render_states_array [index];
    texture_render_states -> state_array [D3DSAMP_MAGFILTER] = D3DTEXF_POINT;
    texture_render_states -> state_array [D3DSAMP_MINFILTER] = D3DTEXF_POINT;
    texture_render_states -> state_array [D3DSAMP_MAXANISOTROPY] = 1;
  }
  _num_active_texture_stages = 0;

  set_render_state(D3DRS_NORMALIZENORMALS, false);

  _last_fvf = 0;
}

/**
 * Resets all internal state as if the gsg were newly created.  The
 * GraphicsWindow pointer represents a typical window that might be used for
 * this context; it may be required to set up the frame buffer properly the
 * first time.
 */
void DXGraphicsStateGuardian9::
reset() {
  GraphicsStateGuardian::reset();

  // Build _inv_state_mask as a mask of 1's where we don't care, and 0's where
  // we do care, about the state.
  _inv_state_mask.clear_bit(ShaderAttrib::get_class_slot());
  _inv_state_mask.clear_bit(AlphaTestAttrib::get_class_slot());
  _inv_state_mask.clear_bit(ClipPlaneAttrib::get_class_slot());
  _inv_state_mask.clear_bit(ColorAttrib::get_class_slot());
  _inv_state_mask.clear_bit(ColorScaleAttrib::get_class_slot());
  _inv_state_mask.clear_bit(CullFaceAttrib::get_class_slot());
  _inv_state_mask.clear_bit(DepthOffsetAttrib::get_class_slot());
  _inv_state_mask.clear_bit(DepthTestAttrib::get_class_slot());
  _inv_state_mask.clear_bit(DepthWriteAttrib::get_class_slot());
  _inv_state_mask.clear_bit(RenderModeAttrib::get_class_slot());
  _inv_state_mask.clear_bit(RescaleNormalAttrib::get_class_slot());
  _inv_state_mask.clear_bit(ShadeModelAttrib::get_class_slot());
  _inv_state_mask.clear_bit(TransparencyAttrib::get_class_slot());
  _inv_state_mask.clear_bit(ColorWriteAttrib::get_class_slot());
  _inv_state_mask.clear_bit(ColorBlendAttrib::get_class_slot());
  _inv_state_mask.clear_bit(TextureAttrib::get_class_slot());
  _inv_state_mask.clear_bit(TexGenAttrib::get_class_slot());
  _inv_state_mask.clear_bit(TexMatrixAttrib::get_class_slot());
  _inv_state_mask.clear_bit(MaterialAttrib::get_class_slot());
  _inv_state_mask.clear_bit(LightAttrib::get_class_slot());
  _inv_state_mask.clear_bit(StencilAttrib::get_class_slot());
  _inv_state_mask.clear_bit(FogAttrib::get_class_slot());
  _inv_state_mask.clear_bit(ScissorAttrib::get_class_slot());

  // D3DRS_POINTSPRITEENABLE doesn't seem to support remapping the texture
  // coordinates via a texture matrix, so we don't advertise
  // GR_point_sprite_tex_matrix.
  _supported_geom_rendering =
    Geom::GR_point | Geom::GR_point_uniform_size |
    Geom::GR_point_perspective | Geom::GR_point_sprite |
    Geom::GR_indexed_other |
    Geom::GR_triangle_strip | Geom::GR_triangle_fan |
    Geom::GR_flat_first_vertex |
    Geom::GR_render_mode_wireframe | Geom::GR_render_mode_point;

  // overwrite gsg defaults with these values

  HRESULT hr;

  // make sure gsg passes all current state down to us
  // set_state_and_transform(RenderState::make_empty(),
  // TransformState::make_identity()); want gsg to pass all state settings
  // down so any non-matching defaults we set here get overwritten

  nassertv(_screen->_d3d9 != nullptr);

  if (_d3d_device == nullptr) {
    return;
  }

  D3DCAPS9 d3d_caps;
  _d3d_device->GetDeviceCaps(&d3d_caps);

  _vertex_shader_version_major = D3DSHADER_VERSION_MAJOR (d3d_caps.VertexShaderVersion);
  _vertex_shader_version_minor = D3DSHADER_VERSION_MINOR (d3d_caps.VertexShaderVersion);
  _pixel_shader_version_major = D3DSHADER_VERSION_MAJOR (d3d_caps.PixelShaderVersion);
  _pixel_shader_version_minor = D3DSHADER_VERSION_MINOR (d3d_caps.PixelShaderVersion);

  _supports_hlsl = (_pixel_shader_version_major != 0);

  _vertex_shader_profile = (char *) D3DXGetVertexShaderProfile (_d3d_device);
  _pixel_shader_profile = (char *) D3DXGetPixelShaderProfile (_d3d_device);

  _vertex_shader_maximum_constants = d3d_caps.MaxVertexShaderConst;

  switch (_pixel_shader_version_major)
  {
    case 0:
      _shader_model = SM_00;
      break;
    case 1:
      _shader_model = SM_11;
      break;
    case 2:
      // minimim specification for pixel shader 2.0 is 96 instruction slots
      _shader_model = SM_20;
      if (d3d_caps.PS20Caps.NumInstructionSlots >= 512) {
        _shader_model = SM_2X;
      }
      break;
    case 3:
      _shader_model = SM_30;
      break;
    case 4:
      _shader_model = SM_40;
      break;
    case 5:
    default:
      _shader_model = SM_50;
      break;
  }

  _auto_detect_shader_model = _shader_model;

#ifdef HAVE_CG
  set_cg_device(_d3d_device);

  _cg_context = cgCreateContext();

  if (cgD3D9IsProfileSupported(CG_PROFILE_PS_2_0) &&
      cgD3D9IsProfileSupported(CG_PROFILE_VS_2_0)) {
    _supports_basic_shaders = true;
    _shader_caps._active_vprofile = (int)cgD3D9GetLatestVertexProfile();
    _shader_caps._active_fprofile = (int)cgD3D9GetLatestPixelProfile();
    _shader_caps._ultimate_vprofile = (int)CG_PROFILE_VS_3_0;
    _shader_caps._ultimate_fprofile = (int)CG_PROFILE_PS_3_0;
/*
    _shader_caps._active_vprofile = (int)CG_PROFILE_VS_2_0;
    _shader_caps._active_fprofile = (int)CG_PROFILE_PS_2_0;
    _shader_caps._ultimate_vprofile = (int)CG_PROFILE_VS_2_0;
    _shader_caps._ultimate_fprofile = (int)CG_PROFILE_PS_2_0;
*/
  }

  if (dxgsg9_cat.is_debug()) {
    CGprofile vertex_profile;
    CGprofile pixel_profile;

    vertex_profile = cgD3D9GetLatestVertexProfile();
    pixel_profile = cgD3D9GetLatestPixelProfile();

    const char *vertex_profile_str =
      cgGetProfileString(vertex_profile);
    const char *pixel_profile_str =
      cgGetProfileString(pixel_profile);

    if (vertex_profile_str == nullptr) {
      vertex_profile_str = "(null)";
    }
    if (pixel_profile_str == nullptr) {
      pixel_profile_str = "(null)";
    }

    dxgsg9_cat.debug()
      << "\nCg latest vertex profile = " << vertex_profile_str << "  id = " << vertex_profile
      << "\nCg latest pixel profile = " << pixel_profile_str << "  id = " << pixel_profile
      << "\nshader model = " << _shader_model
      << "\n";
  }
#endif

  _supports_stream_offset = (d3d_caps.DevCaps2 & D3DDEVCAPS2_STREAMOFFSET) != 0;
  _screen->_supports_dynamic_textures = ((d3d_caps.Caps2 & D3DCAPS2_DYNAMICTEXTURES) != 0);
  _screen->_supports_automatic_mipmap_generation = ((d3d_caps.Caps2 & D3DCAPS2_CANAUTOGENMIPMAP) != 0);

  if (support_stencil) {
    int min_stencil = D3DSTENCILCAPS_ZERO | D3DSTENCILCAPS_REPLACE | D3DSTENCILCAPS_INCR | D3DSTENCILCAPS_DECR;
    if ((d3d_caps.StencilCaps & min_stencil) == min_stencil) {
      if (dxgsg9_cat.is_debug()) {
        dxgsg9_cat.debug()
          << "Checking for stencil; mode = "
          << D3DFormatStr(_screen->_presentation_params.AutoDepthStencilFormat)
          << "\n";
      }
      switch (_screen->_presentation_params.AutoDepthStencilFormat) {
        // These are the only formats that support stencil.
      case D3DFMT_D15S1:
      case D3DFMT_D24S8:
      case D3DFMT_D24X4S4:
        _supports_stencil = true;
        if (dxgsg9_cat.is_debug()) {
          dxgsg9_cat.debug()
            << "Stencils supported.\n";
        }
        break;

      default:
        if (dxgsg9_cat.is_debug()) {
          dxgsg9_cat.debug()
            << "Stencils NOT supported.\n";
        }
      }
    }
  }

  _supports_stencil_wrap = (d3d_caps.StencilCaps & D3DSTENCILCAPS_INCR) && (d3d_caps.StencilCaps & D3DSTENCILCAPS_DECR);
  _supports_two_sided_stencil = ((d3d_caps.StencilCaps & D3DSTENCILCAPS_TWOSIDED) != 0);

  _max_color_targets = d3d_caps.NumSimultaneousRTs;

  _supports_depth_bias = ((d3d_caps.RasterCaps & (D3DPRASTERCAPS_DEPTHBIAS | D3DPRASTERCAPS_SLOPESCALEDEPTHBIAS)) == (D3DPRASTERCAPS_DEPTHBIAS | D3DPRASTERCAPS_SLOPESCALEDEPTHBIAS));

  _supports_gamma_calibration = ((d3d_caps.Caps2 & D3DCAPS2_CANCALIBRATEGAMMA) != 0);

  // Test for occlusion query support
  hr = _d3d_device->CreateQuery(D3DQUERYTYPE_OCCLUSION, nullptr);
  _supports_occlusion_query = !FAILED(hr);

  if (dxgsg9_cat.is_error()) {
    dxgsg9_cat.debug()
      << "\nHwTransformAndLight = " << ((d3d_caps.DevCaps & D3DDEVCAPS_HWTRANSFORMANDLIGHT) != 0)
      << "\nMaxTextureWidth = " << d3d_caps.MaxTextureWidth
      << "\nMaxTextureHeight = " << d3d_caps.MaxTextureHeight
      << "\nMaxVolumeExtent = " << d3d_caps.MaxVolumeExtent
      << "\nMaxTextureAspectRatio = " << d3d_caps.MaxTextureAspectRatio
      << "\nTexCoordCount = " << (d3d_caps.FVFCaps & D3DFVFCAPS_TEXCOORDCOUNTMASK)
      << "\nMaxTextureBlendStages = " << d3d_caps.MaxTextureBlendStages
      << "\nMaxSimultaneousTextures = " << d3d_caps.MaxSimultaneousTextures
      << "\nMaxActiveLights = " << d3d_caps.MaxActiveLights
      << "\nMaxUserClipPlanes = " << d3d_caps.MaxUserClipPlanes
      << "\nMaxVertexBlendMatrices = " << d3d_caps.MaxVertexBlendMatrices
      << "\nMaxVertexBlendMatrixIndex = " << d3d_caps.MaxVertexBlendMatrixIndex
      << "\nMaxPointSize = " << d3d_caps.MaxPointSize
      << "\nMaxPrimitiveCount = " << d3d_caps.MaxPrimitiveCount
      << "\nMaxVertexIndex = " << d3d_caps.MaxVertexIndex
      << "\nMaxStreams = " << d3d_caps.MaxStreams
      << "\nMaxStreamStride = " << d3d_caps.MaxStreamStride
      << "\nD3DTEXOPCAPS_MULTIPLYADD = " << ((d3d_caps.TextureOpCaps & D3DTEXOPCAPS_MULTIPLYADD) != 0)
      << "\nD3DTEXOPCAPS_LERP = " << ((d3d_caps.TextureOpCaps & D3DTEXOPCAPS_LERP) != 0)
      << "\nD3DPMISCCAPS_TSSARGTEMP = " << ((d3d_caps.PrimitiveMiscCaps & D3DPMISCCAPS_TSSARGTEMP) != 0)
      << "\nD3DPRASTERCAPS_DEPTHBIAS = " << ((d3d_caps.RasterCaps & D3DPRASTERCAPS_DEPTHBIAS) != 0)
      << "\nD3DPRASTERCAPS_SLOPESCALEDEPTHBIAS = " << ((d3d_caps.RasterCaps & D3DPRASTERCAPS_SLOPESCALEDEPTHBIAS) != 0)
      << "\nVertexShaderVersion = " << _vertex_shader_version_major << "." << _vertex_shader_version_minor
      << "\nPixelShaderVersion = " << _pixel_shader_version_major << "." << _pixel_shader_version_minor
      << "\nMaxVertexShaderConst = " << _vertex_shader_maximum_constants
      << "\nsupports_stream_offset = " << _supports_stream_offset
      << "\nsupports_dynamic_textures = " << _screen->_supports_dynamic_textures
      << "\nsupports_automatic_mipmap_generation = " << _screen->_supports_automatic_mipmap_generation
      << "\nsupports_stencil_wrap = " << _supports_stencil_wrap
      << "\nsupports_two_sided_stencil = " << _supports_two_sided_stencil
      << "\nsupports_occlusion_query = " << _supports_occlusion_query
      << "\nsupports_gamma_calibration = " << _supports_gamma_calibration
      << "\nMaxAnisotropy = " << d3d_caps.MaxAnisotropy
      << "\nNumSimultaneousRTs = " << d3d_caps.NumSimultaneousRTs
      << "\nD3DPMISCCAPS_MRTPOSTPIXELSHADERBLENDING = " << ((d3d_caps.PrimitiveMiscCaps & D3DPMISCCAPS_MRTPOSTPIXELSHADERBLENDING) != 0)
      << "\nDirectX SDK version " DIRECTX_SDK_VERSION
      << "\n";
  }

  // OVERRIDE SUPPORT SINCE IT DOES NOT WORK WELL
  _screen->_supports_automatic_mipmap_generation = false;

  reset_render_states();

  _max_vertices_per_array = d3d_caps.MaxVertexIndex;
  _max_vertices_per_primitive = d3d_caps.MaxPrimitiveCount;

  _max_texture_stages = d3d_caps.MaxSimultaneousTextures;

  _max_texture_dimension = min(d3d_caps.MaxTextureWidth, d3d_caps.MaxTextureHeight);

  _supports_tex_non_pow2 = !(d3d_caps.TextureCaps & D3DPTEXTURECAPS_POW2);

  _supports_texture_combine = ((d3d_caps.TextureOpCaps & D3DTEXOPCAPS_LERP) != 0);
  _supports_texture_saved_result = ((d3d_caps.PrimitiveMiscCaps & D3DPMISCCAPS_TSSARGTEMP) != 0);
  _supports_texture_constant_color = ((d3d_caps.PrimitiveMiscCaps & D3DPMISCCAPS_PERSTAGECONSTANT) != 0);
  _supports_texture_dot3 = true;

  if (_supports_texture_constant_color) {
    _constant_color_operand = D3DTA_CONSTANT;
  } else {
    _constant_color_operand = D3DTA_TFACTOR;
  }

  _screen->_managed_textures = _gsg_managed_textures;
  _screen->_managed_vertex_buffers = _gsg_managed_vertex_buffers;
  _screen->_managed_index_buffers = _gsg_managed_index_buffers;

  UINT available_texture_memory;

  available_texture_memory = _d3d_device->GetAvailableTextureMem ( );
  if (dxgsg9_cat.is_debug()) {
    dxgsg9_cat.debug() << "*** GetAvailableTextureMem = " <<  available_texture_memory << "\n";
  }
  _available_texture_memory = available_texture_memory;

  // check for render to texture support
  D3DDEVICE_CREATION_PARAMETERS creation_parameters;

  _supports_render_texture = false;
  _screen->_render_to_texture_d3d_format = D3DFMT_UNKNOWN;
  _screen->_framebuffer_d3d_format = D3DFMT_UNKNOWN;

  #define TOTAL_RENDER_TO_TEXTURE_FORMATS 3

  D3DFORMAT render_to_texture_formats [TOTAL_RENDER_TO_TEXTURE_FORMATS] =
  {
    D3DFMT_A8R8G8B8,  // check for this format first
    D3DFMT_X8R8G8B8,
    D3DFMT_UNKNOWN,   // place holder for _screen->_display_mode.Format
  };

  render_to_texture_formats [TOTAL_RENDER_TO_TEXTURE_FORMATS - 1] = _screen->_display_mode.Format;

  hr = _d3d_device->GetCreationParameters (&creation_parameters);
  if (SUCCEEDED (hr)) {
    _screen->_framebuffer_d3d_format = _screen->_display_mode.Format;

    int index;
    for (index = 0; index < TOTAL_RENDER_TO_TEXTURE_FORMATS; index++) {
      hr = _screen->_d3d9->CheckDeviceFormat (
          creation_parameters.AdapterOrdinal,
          creation_parameters.DeviceType,
          _screen->_display_mode.Format,
          D3DUSAGE_RENDERTARGET,
          D3DRTYPE_TEXTURE,
          render_to_texture_formats [index]);
      if (SUCCEEDED (hr)) {
        _screen->_render_to_texture_d3d_format = render_to_texture_formats [index];
        _supports_render_texture = true;
      }
      if (_supports_render_texture) {
        break;
      }
    }
  }
  if (dxgsg9_cat.is_debug()) {
    dxgsg9_cat.debug() << "Render to Texture Support = " << _supports_render_texture << "\n";
  }

  // override default config setting since it is really supported or not ???
  // support_render_texture = _supports_render_texture;

  _supports_3d_texture = ((d3d_caps.TextureCaps & D3DPTEXTURECAPS_VOLUMEMAP) != 0);
  if (_supports_3d_texture) {
    _max_3d_texture_dimension = d3d_caps.MaxVolumeExtent;
  }
  _supports_cube_map = ((d3d_caps.TextureCaps & D3DPTEXTURECAPS_CUBEMAP) != 0);
  if (_supports_cube_map) {
    _max_cube_map_dimension = _max_texture_dimension;
  }

  _max_lights = (int)d3d_caps.MaxActiveLights;
  _max_clip_planes = (int)d3d_caps.MaxUserClipPlanes;
  _max_vertex_transforms = d3d_caps.MaxVertexBlendMatrices;
  _max_vertex_transform_indices = d3d_caps.MaxVertexBlendMatrixIndex;

  set_render_state(D3DRS_AMBIENT, 0x0);

  _clip_plane_bits = 0;
  set_render_state(D3DRS_CLIPPLANEENABLE , 0x0);

  set_render_state(D3DRS_CLIPPING, true);

  set_render_state(D3DRS_SHADEMODE, D3DSHADE_GOURAUD);

  set_render_state(D3DRS_ZWRITEENABLE, TRUE);

/* ***** DX9 ??? D3DRS_EDGEANTIALIAS NOT IN DX9 */
// set_render_state(D3DRS_EDGEANTIALIAS, false);

  set_render_state(D3DRS_ZENABLE, D3DZB_FALSE);

  set_render_state(D3DRS_ALPHABLENDENABLE, FALSE);

  set_render_state(D3DRS_FOGENABLE, FALSE);

  _has_scene_graph_color = false;

  _last_testcooplevel_result = D3D_OK;

  if (dxgsg9_cat.is_debug()) {
    dxgsg9_cat.debug() << "Supported texture formats:\n";
  }

  for(int i = 0; i < MAX_POSSIBLE_TEXFMTS; i++) {
    // look for all possible DX9 texture fmts
    D3DFORMAT_FLAG fmtflag = D3DFORMAT_FLAG(1 << i);
    hr = _screen->_d3d9->CheckDeviceFormat(_screen->_card_id, D3DDEVTYPE_HAL, _screen->_display_mode.Format,
                                          0x0, D3DRTYPE_TEXTURE, g_D3DFORMATmap[fmtflag]);
    if (SUCCEEDED(hr)) {
      if (dxgsg9_cat.is_debug()) {
        dxgsg9_cat.debug() << "  " << D3DFormatStr(g_D3DFORMATmap[fmtflag]) << "\n";
      }
      _screen->_supported_tex_formats_mask |= fmtflag;
    }
  }

  _supports_depth_stencil = ((_screen->_supported_tex_formats_mask & D24S8_FLAG) != 0);
  _supports_depth_texture = _supports_depth_stencil ||
    (_screen->_supported_tex_formats_mask & D16_FLAG) != 0 ||
    (_screen->_supported_tex_formats_mask & D32_FLAG) != 0 ||
    (_screen->_supported_tex_formats_mask & D24X8_FLAG) != 0;

  // In DirectX 9, all built-in depth formats use shadow map filtering.  Some
  // drivers (GeForce 8000+, Radeon HD 4000+, Intel G45+) expose a FourCC
  // format called "INTZ" that allows access to the actual depth.  We don't
  // have a flag to indicate this, though.
  _supports_shadow_filter = _supports_depth_texture;

  // check if compressed textures are supported
  #define CHECK_COMPRESSED_FMT(mode, fmt) \
  if (_screen->_supported_tex_formats_mask & fmt##_FLAG) {\
    if (dxgsg9_cat.is_debug()) {\
      dxgsg9_cat.debug() << "Compressed texture format " << #fmt << " supported\n";\
    }\
    _supports_compressed_texture = true;\
    _compressed_texture_formats.set_bit(Texture::mode);\
  }

  if (_screen->_intel_compressed_texture_bug) {
    dxgsg9_cat.info()
      << "Buggy Intel driver detected; disabling compressed textures.\n";
    _screen->_supported_tex_formats_mask &=
      ~(DXT1_FLAG | DXT2_FLAG | DXT3_FLAG | DXT4_FLAG | DXT5_FLAG);

  } else {
    // Check for available compressed formats normally.
    CHECK_COMPRESSED_FMT(CM_dxt1, DXT1);
    CHECK_COMPRESSED_FMT(CM_dxt2, DXT2);
    CHECK_COMPRESSED_FMT(CM_dxt3, DXT3);
    CHECK_COMPRESSED_FMT(CM_dxt4, DXT4);
    CHECK_COMPRESSED_FMT(CM_dxt5, DXT5);
    CHECK_COMPRESSED_FMT(CM_rgtc, ATI1);
    CHECK_COMPRESSED_FMT(CM_rgtc, ATI2);
  }

  #undef CHECK_COMPRESSED_FMT

  _screen->_supports_rgba16f_texture_format = false;
  hr = _screen->_d3d9->CheckDeviceFormat(_screen->_card_id, D3DDEVTYPE_HAL, _screen->_display_mode.Format, 0x0, D3DRTYPE_TEXTURE, D3DFMT_A16B16G16R16F);
  if (SUCCEEDED(hr)){
    _screen->_supports_rgba16f_texture_format = true;
  }
  _screen->_supports_rgba32_texture_format = false;
  hr = _screen->_d3d9->CheckDeviceFormat(_screen->_card_id, D3DDEVTYPE_HAL, _screen->_display_mode.Format, 0x0, D3DRTYPE_TEXTURE, D3DFMT_A32B32G32R32F);
  if (SUCCEEDED(hr)){
    _screen->_supports_rgba32_texture_format = true;
  }

  // s3 virge drivers sometimes give crap values for these
  if (_screen->_d3dcaps.MaxTextureWidth == 0)
    _screen->_d3dcaps.MaxTextureWidth = 256;

  if (_screen->_d3dcaps.MaxTextureHeight == 0)
    _screen->_d3dcaps.MaxTextureHeight = 256;

  if (_screen->_d3dcaps.RasterCaps & D3DPRASTERCAPS_FOGTABLE) {
    // Watch out for drivers that emulate per-pixel fog with per-vertex fog
    // (Riva128, Matrox Millen G200).  Some of these require gouraud-shading
    // to be set to work, as if you were using vertex fog
    _do_fog_type = PerPixelFog;
  } else {
    // every card is going to have vertex fog, since it's implemented in d3d
    // runtime.
    nassertv((_screen->_d3dcaps.RasterCaps & D3DPRASTERCAPS_FOGVERTEX) != 0);

    // vertex fog may look crappy if you have large polygons in the foreground
    // and they get clipped, so you may want to disable it

    if (dx_no_vertex_fog) {
      _do_fog_type = None;
    } else {
      _do_fog_type = PerVertexFog;

      // range-based fog only works with vertex fog in dx78
      if (dx_use_rangebased_fog && (_screen->_d3dcaps.RasterCaps & D3DPRASTERCAPS_FOGRANGE)) {
        set_render_state(D3DRS_RANGEFOGENABLE, true);
      }
    }
  }

  _screen->_can_direct_disable_color_writes = ((_screen->_d3dcaps.PrimitiveMiscCaps & D3DPMISCCAPS_COLORWRITEENABLE) != 0);

  // Lighting, let's turn it off initially.
  set_render_state(D3DRS_LIGHTING, false);

  // turn on dithering if the rendertarget is < 8bitscolor channel
  bool dither_enabled = ((!dx_no_dithering) && IS_16BPP_DISPLAY_FORMAT(_screen->_presentation_params.BackBufferFormat)
       && (_screen->_d3dcaps.RasterCaps & D3DPRASTERCAPS_DITHER));
  set_render_state(D3DRS_DITHERENABLE, dither_enabled);

  set_render_state(D3DRS_CLIPPING, true);

  // Stencil test is off by default
  set_render_state(D3DRS_STENCILENABLE, FALSE);
  if (_supports_two_sided_stencil) {
    set_render_state(D3DRS_TWOSIDEDSTENCILMODE, FALSE);
  }

  // Antialiasing.
/* ***** DX9 ??? D3DRS_EDGEANTIALIAS NOT IN DX9 */
// set_render_state(D3DRS_EDGEANTIALIAS, FALSE);

  _current_fill_mode = RenderModeAttrib::M_filled;
  set_render_state(D3DRS_FILLMODE, D3DFILL_SOLID);

  // must do SetTSS here because redundant states are filtered out by our code
  // based on current values above, so initial conditions must be correct
  set_texture_stage_state(0, D3DTSS_COLOROP, D3DTOP_DISABLE);  // disables texturing

  _cull_face_mode = CullFaceAttrib::M_cull_none;
  set_render_state(D3DRS_CULLMODE, D3DCULL_NONE);

  set_render_state(D3DRS_ALPHAFUNC, D3DCMP_ALWAYS);
  set_render_state(D3DRS_ALPHAREF, 255);
  set_render_state(D3DRS_ALPHATESTENABLE, FALSE);

  // this is a new DX8 state that lets you do additional operations other than
  // ADD (e.g.  subtractmaxmin) must check
  // (_screen->_d3dcaps.PrimitiveMiscCaps & D3DPMISCCAPS_BLENDOP) (yes on
  // GF2Radeon8500, no on TNT)
  set_render_state(D3DRS_BLENDOP, D3DBLENDOP_ADD);

  _current_shader = nullptr;
  _current_shader_context = nullptr;
  _vertex_array_shader = nullptr;
  _vertex_array_shader_context = nullptr;
  _texture_binding_shader = nullptr;
  _texture_binding_shader_context = nullptr;

  PRINT_REFCNT(dxgsg9, _d3d_device);

  // Now that the GSG has been initialized, make it available for
  // optimizations.
  add_gsg(this);
}

/**
 *
 */
void DXGraphicsStateGuardian9::
apply_fog(Fog *fog) {
  if (_do_fog_type == None)
    return;

  Fog::Mode panda_fogmode = fog->get_mode();
  D3DFOGMODE d3dfogmode = get_fog_mode_type(panda_fogmode);

  set_render_state((D3DRENDERSTATETYPE)_do_fog_type, d3dfogmode);

  const LColor &fog_colr = fog->get_color();
  set_render_state(D3DRS_FOGCOLOR,
                              MY_D3DRGBA(fog_colr[0], fog_colr[1], fog_colr[2], 0.0f));  // Alpha bits are not used

  // do we need to adjust fog startend values based on
  // D3DPRASTERCAPS_WFOGD3DPRASTERCAPS_ZFOG ? if not WFOG, then docs say we
  // need to adjust values to range [0, 1]

  switch (panda_fogmode) {
  case Fog::M_linear:
    {
      PN_stdfloat onset, opaque;
      fog->get_linear_range(onset, opaque);

      set_render_state(D3DRS_FOGSTART,
                                   *((LPDWORD) (&onset)));
      set_render_state(D3DRS_FOGEND,
                                   *((LPDWORD) (&opaque)));
    }
    break;
  case Fog::M_exponential:
  case Fog::M_exponential_squared:
    {
      // Exponential fog is always camera-relative.
      PN_stdfloat fog_density = fog->get_exp_density();
      set_render_state(D3DRS_FOGDENSITY,
                                   *((LPDWORD) (&fog_density)));
    }
    break;
  }
}

/**
 * Sends the indicated transform matrix to the graphics API to be applied to
 * future vertices.
 *
 * This transform is the internal_transform, already converted into the GSG's
 * internal coordinate system.
 */
void DXGraphicsStateGuardian9::
do_issue_transform() {
  const TransformState *transform = _internal_transform;
  DO_PSTATS_STUFF(_transform_state_pcollector.add_level(1));

  if (_current_shader_context) {
// _current_shader_context->issue_transform(this);
    _current_shader_context->issue_parameters(this, Shader::SSD_transform);

// ??? NO NEED TO SET THE D3D TRANSFORM VIA SetTransform SINCE THE TRANSFORM
// IS ONLY USED IN THE SHADER
    LMatrix4f mat = LCAST(float, transform->get_mat());
    const D3DMATRIX *d3d_mat = (const D3DMATRIX *)mat.get_data();
    _d3d_device->SetTransform(D3DTS_WORLD, d3d_mat);

  }
  else {
    LMatrix4f mat = LCAST(float, transform->get_mat());
    const D3DMATRIX *d3d_mat = (const D3DMATRIX *)mat.get_data();
    _d3d_device->SetTransform(D3DTS_WORLD, d3d_mat);

// DEBUG PRINT
/*
    const PN_stdfloat *data;
    data = &d3d_mat -> _11;
        dxgsg9_cat.debug ( ) << "do_issue_transform\n" <<
          data[ 0] << " " << data[ 1] << " " << data[ 2] << " " << data[ 3] << "\n" <<
          data[ 4] << " " << data[ 5] << " " << data[ 6] << " " << data[ 7] << "\n" <<
          data[ 8] << " " << data[ 9] << " " << data[10] << " " << data[11] << "\n" <<
          data[12] << " " << data[13] << " " << data[14] << " " << data[15] << "\n";
*/

  }

  _transform_stale = false;
}

/**
 *
 */
void DXGraphicsStateGuardian9::
do_issue_alpha_test() {
  if (_target_shader->get_flag(ShaderAttrib::F_subsume_alpha_test)) {
    set_render_state(D3DRS_ALPHATESTENABLE, FALSE);
  } else {
    const AlphaTestAttrib *target_alpha_test = DCAST(AlphaTestAttrib, _target_rs->get_attrib_def(AlphaTestAttrib::get_class_slot()));
    AlphaTestAttrib::PandaCompareFunc mode = target_alpha_test->get_mode();
    if (mode == AlphaTestAttrib::M_none) {
      set_render_state(D3DRS_ALPHATESTENABLE, FALSE);
    } else {
      // AlphaTestAttrib::PandaCompareFunc === D3DCMPFUNC
      set_render_state(D3DRS_ALPHAFUNC, (D3DCMPFUNC)mode);
      set_render_state(D3DRS_ALPHAREF, (UINT) (target_alpha_test->get_reference_alpha()*255.0f));  //d3d uses 0x0-0xFF, not a float
      set_render_state(D3DRS_ALPHATESTENABLE, TRUE);
    }
  }
}

/**
 *
 */
void DXGraphicsStateGuardian9::
do_issue_shader() {

  DXShaderContext9 *context = 0;
  Shader *shader = 0;
  if (_target_shader) {
    shader = (Shader *)(_target_shader->get_shader());
  }
  if (shader) {
    context = (DXShaderContext9 *)(shader->prepare_now(get_prepared_objects(), this));
  }

  if (context == 0 || (context && context -> valid (this) == false)) {
    if (_current_shader_context != 0) {
      _current_shader_context->unbind(this);
      _current_shader = 0;
      _current_shader_context = 0;
      disable_standard_texture_bindings();
    }
    return;
  }

  if (context != _current_shader_context) {
    // Use a completely different shader than before.  Unbind old shader, bind
    // the new one.
    if (_current_shader_context != 0) {
      _current_shader_context->unbind(this);
      _current_shader_context = 0;
      _current_shader = 0;
      disable_standard_texture_bindings();
    }
    if (context != 0) {
      context->bind(this);
      _current_shader = shader;
      _current_shader_context = context;
    }
  } else {
    // Use the same shader as before, but with new input arguments.
    context->issue_parameters(this, Shader::SSD_shaderinputs);
  }
}

/**
 *
 */
void DXGraphicsStateGuardian9::
do_issue_render_mode() {
  const RenderModeAttrib *target_render_mode;
  _target_rs->get_attrib_def(target_render_mode);
  RenderModeAttrib::Mode mode = target_render_mode->get_mode();

  switch (mode) {
  case RenderModeAttrib::M_unchanged:
  case RenderModeAttrib::M_filled:
  case RenderModeAttrib::M_filled_flat:
    set_render_state(D3DRS_FILLMODE, D3DFILL_SOLID);
    break;

  case RenderModeAttrib::M_wireframe:
    set_render_state(D3DRS_FILLMODE, D3DFILL_WIREFRAME);
    break;

  case RenderModeAttrib::M_point:
    set_render_state(D3DRS_FILLMODE, D3DFILL_POINT);
    break;

  default:
    dxgsg9_cat.error()
      << "Unknown render mode " << (int)mode << endl;
  }

  // This might also specify the point size.
  PN_stdfloat point_size = target_render_mode->get_thickness();
  set_render_state(D3DRS_POINTSIZE, *((DWORD*)&point_size));

  if (target_render_mode->get_perspective()) {
    set_render_state(D3DRS_POINTSCALEENABLE, TRUE);

    LVector3 height(0.0f, point_size, 1.0f);
    height = height * _projection_mat->get_mat();
    PN_stdfloat s = height[1] / point_size;

    PN_stdfloat zero = 0.0f;
    PN_stdfloat one_over_s2 = 1.0f / (s * s);
    set_render_state(D3DRS_POINTSCALE_A, *((DWORD*)&zero));
    set_render_state(D3DRS_POINTSCALE_B, *((DWORD*)&zero));
    set_render_state(D3DRS_POINTSCALE_C, *((DWORD*)&one_over_s2));

  } else {
    set_render_state(D3DRS_POINTSCALEENABLE, FALSE);
  }

  _current_fill_mode = mode;
}

/**
 *
 */
void DXGraphicsStateGuardian9::
do_issue_rescale_normal() {
  RescaleNormalAttrib::Mode mode = RescaleNormalAttrib::M_none;

  const RescaleNormalAttrib *target_rescale_normal;
  if (_target_rs->get_attrib(target_rescale_normal)) {
    mode = target_rescale_normal->get_mode();
  }

  switch (mode) {
  case RescaleNormalAttrib::M_none:
    set_render_state(D3DRS_NORMALIZENORMALS, false);
    break;

  case RescaleNormalAttrib::M_rescale:
  case RescaleNormalAttrib::M_normalize:
    set_render_state(D3DRS_NORMALIZENORMALS, true);
    break;

  default:
    dxgsg9_cat.error()
      << "Unknown rescale_normal mode " << (int)mode << endl;
  }
}

/**
 *
 */
void DXGraphicsStateGuardian9::
do_issue_depth_test() {
  const DepthTestAttrib *target_depth_test = DCAST(DepthTestAttrib, _target_rs->get_attrib_def(DepthTestAttrib::get_class_slot()));
  DepthTestAttrib::PandaCompareFunc mode = target_depth_test->get_mode();
  if (mode == DepthTestAttrib::M_none) {
    set_render_state(D3DRS_ZENABLE, D3DZB_FALSE);
  } else {
    set_render_state(D3DRS_ZENABLE, D3DZB_TRUE);
    set_render_state(D3DRS_ZFUNC, (D3DCMPFUNC) mode);
  }
}

/**
 *
 */
void DXGraphicsStateGuardian9::
do_issue_depth_write() {
  const DepthWriteAttrib *target_depth_write = DCAST(DepthWriteAttrib, _target_rs->get_attrib_def(DepthWriteAttrib::get_class_slot()));
  DepthWriteAttrib::Mode mode = target_depth_write->get_mode();
  if (mode == DepthWriteAttrib::M_on) {
    set_render_state(D3DRS_ZWRITEENABLE, TRUE);
  } else {
    set_render_state(D3DRS_ZWRITEENABLE, FALSE);
  }
}

/**
 *
 */
void DXGraphicsStateGuardian9::
do_issue_cull_face() {
  const CullFaceAttrib *target_cull_face = DCAST(CullFaceAttrib, _target_rs->get_attrib_def(CullFaceAttrib::get_class_slot()));
  _cull_face_mode = target_cull_face->get_effective_mode();

  switch (_cull_face_mode) {
  case CullFaceAttrib::M_cull_none:
    set_render_state(D3DRS_CULLMODE, D3DCULL_NONE);

// printf ("------------------- D3DCULL_NONE\n");

    break;
  case CullFaceAttrib::M_cull_clockwise:
    set_render_state(D3DRS_CULLMODE, D3DCULL_CW);

// printf ("------------------- D3DCULL_CW -- CLOCKWISE \n");

    break;
  case CullFaceAttrib::M_cull_counter_clockwise:
    set_render_state(D3DRS_CULLMODE, D3DCULL_CCW);

// printf ("------------------- D3DCULL_CCW\n");

    break;
  default:
    dxgsg9_cat.error()
      << "invalid cull face mode " << (int)_cull_face_mode << endl;
    break;
  }
}

/**
 *
 */
void DXGraphicsStateGuardian9::
do_issue_fog() {
  const FogAttrib *target_fog = DCAST(FogAttrib, _target_rs->get_attrib_def(FogAttrib::get_class_slot()));
  if (!target_fog->is_off()) {
    set_render_state(D3DRS_FOGENABLE, TRUE);
    Fog *fog = target_fog->get_fog();
    nassertv(fog != nullptr);
    apply_fog(fog);
  } else {
    set_render_state(D3DRS_FOGENABLE, FALSE);
  }
}

/**
 *
 */
void DXGraphicsStateGuardian9::
do_issue_depth_offset() {
  const DepthOffsetAttrib *target_depth_offset = DCAST(DepthOffsetAttrib, _target_rs->get_attrib_def(DepthOffsetAttrib::get_class_slot()));
  int offset = target_depth_offset->get_offset();

  if (_supports_depth_bias && !dx_broken_depth_bias) {
    set_render_state(D3DRS_DEPTHBIAS, offset);
    set_render_state(D3DRS_SLOPESCALEDEPTHBIAS, offset);

  } else {
    // DirectX depth bias isn't directly supported by the driver.  Cheese a
    // depth bias effect by sliding the viewport backward a bit.
    static const PN_stdfloat bias_scale = dx_depth_bias_scale;
    D3DVIEWPORT9 vp = _current_viewport;
    vp.MinZ -= bias_scale * offset;
    vp.MaxZ -= bias_scale * offset;
    _d3d_device->SetViewport(&vp);
  }
}

/**
 *
 */
void DXGraphicsStateGuardian9::
do_issue_shade_model() {
  const ShadeModelAttrib *target_shade_model = DCAST(ShadeModelAttrib, _target_rs->get_attrib_def(ShadeModelAttrib::get_class_slot()));
  switch (target_shade_model->get_mode()) {
  case ShadeModelAttrib::M_smooth:
    set_render_state(D3DRS_SHADEMODE, D3DSHADE_GOURAUD);
    break;

  case ShadeModelAttrib::M_flat:
    set_render_state(D3DRS_SHADEMODE, D3DSHADE_FLAT);
    break;
  }
}

/**
 * Simultaneously resets the render state and the transform state.
 *
 * This transform specified is the "internal" net transform, already converted
 * into the GSG's internal coordinate space by composing it to
 * get_cs_transform().  (Previously, this used to be the "external" net
 * transform, with the assumption that that GSG would convert it internally,
 * but that is no longer the case.)
 *
 * Special case: if (state==NULL), then the target state is already stored in
 * _target.
 */
void DXGraphicsStateGuardian9::
set_state_and_transform(const RenderState *target,
                        const TransformState *transform) {
#ifndef NDEBUG
  if (gsg_cat.is_spam()) {
    gsg_cat.spam() << "Setting GSG state to " << (void *)target << ":\n";
    target->write(gsg_cat.spam(false), 2);
  }
#endif
  _state_pcollector.add_level(1);
  PStatTimer timer1(_draw_set_state_pcollector);

  if (transform != _internal_transform) {
    // PStatTimer timer(_draw_set_state_transform_pcollector);
    _state_pcollector.add_level(1);
    _internal_transform = transform;
    do_issue_transform();
  }

  if (target == _state_rs && (_state_mask | _inv_state_mask).is_all_on()) {
    return;
  }
  _target_rs = target;

  determine_target_shader();

  int alpha_test_slot = AlphaTestAttrib::get_class_slot();
  if (_target_rs->get_attrib(alpha_test_slot) != _state_rs->get_attrib(alpha_test_slot) ||
      !_state_mask.get_bit(alpha_test_slot)) {
    // PStatTimer timer(_draw_set_state_alpha_test_pcollector);
    do_issue_alpha_test();
    _state_mask.set_bit(alpha_test_slot);
  }

  int clip_plane_slot = ClipPlaneAttrib::get_class_slot();
  if (_target_rs->get_attrib(clip_plane_slot) != _state_rs->get_attrib(clip_plane_slot) ||
      !_state_mask.get_bit(clip_plane_slot)) {
    // PStatTimer timer(_draw_set_state_clip_plane_pcollector);
    do_issue_clip_plane();
    _state_mask.set_bit(clip_plane_slot);
  }

  int color_slot = ColorAttrib::get_class_slot();
  int color_scale_slot = ColorScaleAttrib::get_class_slot();
  if (_target_rs->get_attrib(color_slot) != _state_rs->get_attrib(color_slot) ||
      _target_rs->get_attrib(color_scale_slot) != _state_rs->get_attrib(color_scale_slot) ||
      !_state_mask.get_bit(color_slot) ||
      !_state_mask.get_bit(color_scale_slot)) {
    // PStatTimer timer(_draw_set_state_color_pcollector);
    do_issue_color();
    do_issue_color_scale();
    _state_mask.set_bit(color_slot);
    _state_mask.set_bit(color_scale_slot);
    if (_current_shader_context) {
      _current_shader_context->issue_parameters(this, Shader::SSD_color);
      _current_shader_context->issue_parameters(this, Shader::SSD_colorscale);
    }
  }

  int cull_face_slot = CullFaceAttrib::get_class_slot();
  if (_target_rs->get_attrib(cull_face_slot) != _state_rs->get_attrib(cull_face_slot) ||
      !_state_mask.get_bit(cull_face_slot)) {
    // PStatTimer timer(_draw_set_state_cull_face_pcollector);
    do_issue_cull_face();
    _state_mask.set_bit(cull_face_slot);
  }

  int depth_offset_slot = DepthOffsetAttrib::get_class_slot();
  if (_target_rs->get_attrib(depth_offset_slot) != _state_rs->get_attrib(depth_offset_slot) ||
      !_state_mask.get_bit(depth_offset_slot)) {
    // PStatTimer timer(_draw_set_state_depth_offset_pcollector);
    do_issue_depth_offset();
    _state_mask.set_bit(depth_offset_slot);
  }

  int depth_test_slot = DepthTestAttrib::get_class_slot();
  if (_target_rs->get_attrib(depth_test_slot) != _state_rs->get_attrib(depth_test_slot) ||
      !_state_mask.get_bit(depth_test_slot)) {
    // PStatTimer timer(_draw_set_state_depth_test_pcollector);
    do_issue_depth_test();
    _state_mask.set_bit(depth_test_slot);
  }

  int depth_write_slot = DepthWriteAttrib::get_class_slot();
  if (_target_rs->get_attrib(depth_write_slot) != _state_rs->get_attrib(depth_write_slot) ||
      !_state_mask.get_bit(depth_write_slot)) {
    // PStatTimer timer(_draw_set_state_depth_write_pcollector);
    do_issue_depth_write();
    _state_mask.set_bit(depth_write_slot);
  }

  int render_mode_slot = RenderModeAttrib::get_class_slot();
  if (_target_rs->get_attrib(render_mode_slot) != _state_rs->get_attrib(render_mode_slot) ||
      !_state_mask.get_bit(render_mode_slot)) {
    // PStatTimer timer(_draw_set_state_render_mode_pcollector);
    do_issue_render_mode();
    _state_mask.set_bit(render_mode_slot);
  }

  int rescale_normal_slot = RescaleNormalAttrib::get_class_slot();
  if (_target_rs->get_attrib(rescale_normal_slot) != _state_rs->get_attrib(rescale_normal_slot) ||
      !_state_mask.get_bit(rescale_normal_slot)) {
    // PStatTimer timer(_draw_set_state_rescale_normal_pcollector);
    do_issue_rescale_normal();
    _state_mask.set_bit(rescale_normal_slot);
  }

  int shade_model_slot = ShadeModelAttrib::get_class_slot();
  if (_target_rs->get_attrib(shade_model_slot) != _state_rs->get_attrib(shade_model_slot) ||
      !_state_mask.get_bit(shade_model_slot)) {
    // PStatTimer timer(_draw_set_state_shade_model_pcollector);
    do_issue_shade_model();
    _state_mask.set_bit(shade_model_slot);
  }

  int transparency_slot = TransparencyAttrib::get_class_slot();
  int color_write_slot = ColorWriteAttrib::get_class_slot();
  int color_blend_slot = ColorBlendAttrib::get_class_slot();
  if (_target_rs->get_attrib(transparency_slot) != _state_rs->get_attrib(transparency_slot) ||
      _target_rs->get_attrib(color_write_slot) != _state_rs->get_attrib(color_write_slot) ||
      _target_rs->get_attrib(color_blend_slot) != _state_rs->get_attrib(color_blend_slot) ||
      !_state_mask.get_bit(transparency_slot) ||
      !_state_mask.get_bit(color_write_slot) ||
      !_state_mask.get_bit(color_blend_slot) ||
      (_target_shader->get_flag(ShaderAttrib::F_disable_alpha_write) !=
       _state_shader->get_flag(ShaderAttrib::F_disable_alpha_write))) {
    // PStatTimer timer(_draw_set_state_blending_pcollector);
    do_issue_blending();
    _state_mask.set_bit(transparency_slot);
    _state_mask.set_bit(color_write_slot);
    _state_mask.set_bit(color_blend_slot);
  }

  if (_target_shader != _state_shader) {
    // PStatTimer timer(_draw_set_state_shader_pcollector);
    do_issue_shader();
    _state_shader = _target_shader;
    _state_mask.clear_bit(TextureAttrib::get_class_slot());
  }

  int texture_slot = TextureAttrib::get_class_slot();
  int tex_matrix_slot = TexMatrixAttrib::get_class_slot();
  int tex_gen_slot = TexGenAttrib::get_class_slot();
  if (_target_rs->get_attrib(texture_slot) != _state_rs->get_attrib(texture_slot) ||
      _target_rs->get_attrib(tex_matrix_slot) != _state_rs->get_attrib(tex_matrix_slot) ||
      _target_rs->get_attrib(tex_gen_slot) != _state_rs->get_attrib(tex_gen_slot) ||
      !_state_mask.get_bit(texture_slot) ||
      !_state_mask.get_bit(tex_matrix_slot) ||
      !_state_mask.get_bit(tex_gen_slot)) {
    // PStatTimer timer(_draw_set_state_texture_pcollector);
    determine_target_texture();
    do_issue_texture();

    _state_texture = _target_texture;
    _state_mask.set_bit(texture_slot);
    _state_mask.set_bit(tex_matrix_slot);
    _state_mask.set_bit(tex_gen_slot);
  }

  int material_slot = MaterialAttrib::get_class_slot();
  if (_target_rs->get_attrib(material_slot) != _state_rs->get_attrib(material_slot) ||
      !_state_mask.get_bit(material_slot)) {
    // PStatTimer timer(_draw_set_state_material_pcollector);
    do_issue_material();
    _state_mask.set_bit(material_slot);
    if (_current_shader_context) {
      _current_shader_context->issue_parameters(this, Shader::SSD_material);
    }
  }

  int light_slot = LightAttrib::get_class_slot();
  if (_target_rs->get_attrib(light_slot) != _state_rs->get_attrib(light_slot) ||
      !_state_mask.get_bit(light_slot)) {
    // PStatTimer timer(_draw_set_state_light_pcollector);
    do_issue_light();
    _state_mask.set_bit(light_slot);
  }

  int stencil_slot = StencilAttrib::get_class_slot();
  if (_target_rs->get_attrib(stencil_slot) != _state_rs->get_attrib(stencil_slot) ||
      !_state_mask.get_bit(stencil_slot)) {
    // PStatTimer timer(_draw_set_state_stencil_pcollector);
    do_issue_stencil();
    _state_mask.set_bit(stencil_slot);
  }

  int fog_slot = FogAttrib::get_class_slot();
  if (_target_rs->get_attrib(fog_slot) != _state_rs->get_attrib(fog_slot) ||
      !_state_mask.get_bit(fog_slot)) {
    // PStatTimer timer(_draw_set_state_fog_pcollector);
    do_issue_fog();
    _state_mask.set_bit(fog_slot);
    if (_current_shader_context) {
      _current_shader_context->issue_parameters(this, Shader::SSD_fog);
    }
  }

  int scissor_slot = ScissorAttrib::get_class_slot();
  if (_target_rs->get_attrib(scissor_slot) != _state_rs->get_attrib(scissor_slot) ||
      !_state_mask.get_bit(scissor_slot)) {
    // PStatTimer timer(_draw_set_state_scissor_pcollector);
    do_issue_scissor();
    _state_mask.set_bit(scissor_slot);
  }

  _state_rs = _target_rs;
}

/**
 * Called the first time a particular light has been bound to a given id
 * within a frame, this should set up the associated hardware light with the
 * light's properties.
 */
void DXGraphicsStateGuardian9::
bind_light(PointLight *light_obj, const NodePath &light, int light_id) {
  // Get the light in "world coordinates" (actually, view coordinates).  This
  // means the light in the coordinate space of the camera, converted to DX's
  // coordinate system.
  CPT(TransformState) transform = light.get_transform(_scene_setup->get_camera_path());
  const LMatrix4 &light_mat = transform->get_mat();
  LMatrix4 rel_mat = light_mat * LMatrix4::convert_mat(CS_yup_left, CS_default);
  LPoint3f pos = LCAST(float, light_obj->get_point() * rel_mat);

  D3DCOLORVALUE black;
  black.r = black.g = black.b = black.a = 0.0f;
  D3DLIGHT9 alight;
  alight.Type =  D3DLIGHT_POINT;
  alight.Diffuse  = get_light_color(light_obj);
  alight.Ambient  =  black ;
  LColorf color = LCAST(float, light_obj->get_specular_color());
  alight.Specular = *(D3DCOLORVALUE *)(color.get_data());

  // Position needs to specify x, y, z, and w w == 1 implies non-infinite
  // position
  alight.Position = *(D3DVECTOR *)pos.get_data();

  alight.Range =  __D3DLIGHT_RANGE_MAX;
  alight.Falloff =  1.0f;

  const LVecBase3 &att = light_obj->get_attenuation();
  alight.Attenuation0 = att[0];
  alight.Attenuation1 = att[1];
  alight.Attenuation2 = att[2];

  HRESULT hr = _d3d_device->SetLight(light_id, &alight);
  if (FAILED(hr)) {
    wdxdisplay9_cat.warning()
      << "Could not set light properties for " << light
      << " to id " << light_id << ": " << D3DERRORSTRING(hr) << "\n";
  }
}

/**
 * Called the first time a particular light has been bound to a given id
 * within a frame, this should set up the associated hardware light with the
 * light's properties.
 */
void DXGraphicsStateGuardian9::
bind_light(DirectionalLight *light_obj, const NodePath &light, int light_id) {
  static PStatCollector _draw_set_state_light_bind_directional_pcollector("Draw:Set State:Light:Bind:Directional");
  // PStatTimer timer(_draw_set_state_light_bind_directional_pcollector);

  std::pair<DirectionalLights::iterator, bool> lookup = _dlights.insert(DirectionalLights::value_type(light, D3DLIGHT9()));
  D3DLIGHT9 &fdata = (*lookup.first).second;
  if (lookup.second) {
    // Get the light in "world coordinates" (actually, view coordinates).
    // This means the light in the coordinate space of the camera, converted
    // to DX's coordinate system.
    CPT(TransformState) transform = light.get_transform(_scene_setup->get_camera_path());
    const LMatrix4 &light_mat = transform->get_mat();
    LMatrix4 rel_mat = light_mat * LMatrix4::convert_mat(CS_yup_left, CS_default);
    LVector3f dir = LCAST(float, light_obj->get_direction() * rel_mat);

    D3DCOLORVALUE black;
    black.r = black.g = black.b = black.a = 0.0f;

    ZeroMemory(&fdata, sizeof(D3DLIGHT9));

    fdata.Type =  D3DLIGHT_DIRECTIONAL;
    fdata.Ambient  =  black ;
    LColorf color = LCAST(float, light_obj->get_specular_color());
    fdata.Specular = *(D3DCOLORVALUE *)(color.get_data());

    fdata.Direction = *(D3DVECTOR *)dir.get_data();

    fdata.Range =  __D3DLIGHT_RANGE_MAX;
    fdata.Falloff =  1.0f;

    fdata.Attenuation0 = 1.0f;       // constant
    fdata.Attenuation1 = 0.0f;       // linear
    fdata.Attenuation2 = 0.0f;       // quadratic
  }

  // We have to reset the Diffuse color at each call, because it might have
  // changed independently of the light object itself (due to
  // color_scale_via_lighting being in effect).
  fdata.Diffuse  = get_light_color(light_obj);

  HRESULT hr = _d3d_device->SetLight(light_id, &fdata);
  if (FAILED(hr)) {
    wdxdisplay9_cat.warning()
      << "Could not set light properties for " << light
      << " to id " << light_id << ": " << D3DERRORSTRING(hr) << "\n";
  }
}

/**
 * Called the first time a particular light has been bound to a given id
 * within a frame, this should set up the associated hardware light with the
 * light's properties.
 */
void DXGraphicsStateGuardian9::
bind_light(Spotlight *light_obj, const NodePath &light, int light_id) {
  Lens *lens = light_obj->get_lens();
  nassertv(lens != nullptr);

  // Get the light in "world coordinates" (actually, view coordinates).  This
  // means the light in the coordinate space of the camera, converted to DX's
  // coordinate system.
  CPT(TransformState) transform = light.get_transform(_scene_setup->get_camera_path());
  const LMatrix4 &light_mat = transform->get_mat();
  LMatrix4 rel_mat = light_mat * LMatrix4::convert_mat(CS_yup_left, CS_default);
  LPoint3f pos = LCAST(float, lens->get_nodal_point() * rel_mat);
  LVector3f dir = LCAST(float, lens->get_view_vector() * rel_mat);

  D3DCOLORVALUE black;
  black.r = black.g = black.b = black.a = 0.0f;

  D3DLIGHT9  alight;
  ZeroMemory(&alight, sizeof(D3DLIGHT9));

  alight.Type =  D3DLIGHT_SPOT;
  alight.Ambient  =  black ;
  alight.Diffuse  = get_light_color(light_obj);
  LColorf color = LCAST(float, light_obj->get_specular_color());
  alight.Specular = *(D3DCOLORVALUE *)(color.get_data());

  alight.Position = *(D3DVECTOR *)pos.get_data();

  alight.Direction = *(D3DVECTOR *)dir.get_data();

  alight.Range =  __D3DLIGHT_RANGE_MAX;

  // I determined this formular empirically.  It seems to mostly approximate
  // the OpenGL spotlight equation, for a reasonable range of values for FOV.
  PN_stdfloat fov = lens->get_hfov();
  alight.Falloff =  light_obj->get_exponent() * (fov * fov * fov) / 1620000.0f;

  alight.Theta =  0.0f;
  alight.Phi = deg_2_rad(fov);

  const LVecBase3 &att = light_obj->get_attenuation();
  alight.Attenuation0 = att[0];
  alight.Attenuation1 = att[1];
  alight.Attenuation2 = att[2];

  HRESULT hr = _d3d_device->SetLight(light_id, &alight);
  if (FAILED(hr)) {
    wdxdisplay9_cat.warning()
      << "Could not set light properties for " << light
      << " to id " << light_id << ": " << D3DERRORSTRING(hr) << "\n";
  }
}

/**
 * Maps from the Geom's internal numeric type symbols to DirectX's.
 */
D3DFORMAT DXGraphicsStateGuardian9::
get_index_type(Geom::NumericType numeric_type) {
  switch (numeric_type) {
  // NT_uint8 is automatically promoted to uint16.
  case Geom::NT_uint8:
  case Geom::NT_uint16:
    return D3DFMT_INDEX16;

  case Geom::NT_uint32:
    return D3DFMT_INDEX32;
  }

  dxgsg9_cat.error()
    << "Invalid index NumericType value (" << (int)numeric_type << ")\n";
  return D3DFMT_INDEX16;
}

/**
 *
 */
void DXGraphicsStateGuardian9::
do_issue_material() {
  static Material empty;
  const Material *material;
  const MaterialAttrib *target_material = DCAST(MaterialAttrib, _target_rs->get_attrib_def(MaterialAttrib::get_class_slot()));
  if (target_material->is_off()) {
    material = &empty;
  } else {
    material = target_material->get_material();
  }

  D3DMATERIAL9 cur_material;
  LColorf color = LCAST(float, material->get_diffuse());
  cur_material.Diffuse = *(D3DCOLORVALUE *)(color.get_data());
  color = LCAST(float, material->get_ambient());
  cur_material.Ambient = *(D3DCOLORVALUE *)(color.get_data());
  color = LCAST(float, material->get_specular());
  cur_material.Specular = *(D3DCOLORVALUE *)(color.get_data());
  color = LCAST(float, material->get_emission());
  cur_material.Emissive = *(D3DCOLORVALUE *)(color.get_data());
  cur_material.Power = material->get_shininess();

  if (material->has_diffuse() || material->has_base_color()) {
    // If the material specifies an diffuse color, use it.
    set_render_state(D3DRS_DIFFUSEMATERIALSOURCE, D3DMCS_MATERIAL);
  } else {
    // Otherwise, the diffuse color comes from the object color.
    if (_has_material_force_color) {
      color = LCAST(float, _material_force_color);
      cur_material.Diffuse = *(D3DCOLORVALUE *)color.get_data();
      set_render_state(D3DRS_DIFFUSEMATERIALSOURCE, D3DMCS_MATERIAL);
    } else {
      set_render_state(D3DRS_DIFFUSEMATERIALSOURCE, D3DMCS_COLOR1);
    }
  }
  if (material->has_ambient() || material->has_base_color()) {
    // If the material specifies an ambient color, use it.
    set_render_state(D3DRS_AMBIENTMATERIALSOURCE, D3DMCS_MATERIAL);
  } else {
    // Otherwise, the ambient color comes from the object color.
    if (_has_material_force_color) {
      color = LCAST(float, _material_force_color);
      cur_material.Ambient = *(D3DCOLORVALUE *)color.get_data();
      set_render_state(D3DRS_AMBIENTMATERIALSOURCE, D3DMCS_MATERIAL);
    } else {
      set_render_state(D3DRS_AMBIENTMATERIALSOURCE, D3DMCS_COLOR1);
    }
  }

  if (material->has_specular() || material->has_base_color()) {
    set_render_state(D3DRS_SPECULARENABLE, TRUE);
  } else {
    set_render_state(D3DRS_SPECULARENABLE, FALSE);
  }

  if (material->get_local()) {
    set_render_state(D3DRS_LOCALVIEWER, TRUE);
  } else {
    set_render_state(D3DRS_LOCALVIEWER, FALSE);
  }

  _d3d_device->SetMaterial(&cur_material);
}

/**
 *
 */
void DXGraphicsStateGuardian9::
do_issue_texture() {
  DO_PSTATS_STUFF(_texture_state_pcollector.add_level(1));

  if (_texture_binding_shader_context==0) {
    if (_current_shader_context==0) {
      update_standard_texture_bindings();
    } else {
      disable_standard_texture_bindings();
      _current_shader_context->update_shader_texture_bindings(nullptr,this);
    }
  } else {
    if (_current_shader_context==0) {
      _texture_binding_shader_context->disable_shader_texture_bindings(this);
      update_standard_texture_bindings();
    } else {
      _current_shader_context->
        update_shader_texture_bindings(_texture_binding_shader_context,this);
    }
  }
  _texture_binding_shader = _current_shader;
  _texture_binding_shader_context = _current_shader_context;
}

/**
 *
 */
void DXGraphicsStateGuardian9::
disable_standard_texture_bindings() {
  // Disable the texture stages that are no longer used.
  for (int i = 0; i < _num_active_texture_stages; i++) {
    HRESULT hr;

    hr = _d3d_device -> SetTexture (i, nullptr);
    if (FAILED (hr)) {
      dxgsg9_cat.error()
        << "SetTexture ("
        << i
        << ", NULL) failed "
        << D3DERRORSTRING(hr);
    }
    set_texture_stage_state(i, D3DTSS_COLOROP, D3DTOP_DISABLE);
  }

  _num_active_texture_stages = 0;
}

/**
 *
 */
void DXGraphicsStateGuardian9::
update_standard_texture_bindings() {
  DO_PSTATS_STUFF(_texture_state_pcollector.add_level(1));

  int num_stages = _target_texture->get_num_on_ff_stages();
  int num_old_stages = _max_texture_stages;
  if (_state_texture != nullptr) {
    num_old_stages = _state_texture->get_num_on_ff_stages();
  }

  nassertv(num_stages <= _max_texture_stages &&
           _num_active_texture_stages <= _max_texture_stages);

  _texture_involves_color_scale = false;

  // We have to match up the texcoord stage index to the order written out by
  // the DXGeomMunger.  This means the texcoord names are written in the order
  // indicated by the TextureAttrib.

  int si;
  for (si = 0; si < num_stages; si++) {
    TextureStage *stage = _target_texture->get_on_ff_stage(si);
    int texcoord_index = _target_texture->get_ff_tc_index(si);

    Texture *texture = _target_texture->get_on_texture(stage);
    nassertv(texture != nullptr);
    const SamplerState &sampler = _target_texture->get_on_sampler(stage);

    // We always reissue every stage in DX, just in case the texcoord index or
    // texgen mode or some other property has changed.
    int view = get_current_tex_view_offset() + stage->get_tex_view_offset();
    TextureContext *tc = texture->prepare_now(view, _prepared_objects, this);
    apply_texture(si, tc, sampler);
    set_texture_blend_mode(si, stage);

    int texcoord_dimensions = 2;

    CPT(TransformState) tex_mat = TransformState::make_identity();
    const TexMatrixAttrib *target_tex_matrix = DCAST(TexMatrixAttrib, _target_rs->get_attrib_def(TexMatrixAttrib::get_class_slot()));
    if (target_tex_matrix->has_stage(stage)) {
      tex_mat = target_tex_matrix->get_transform(stage);
    }

    // Issue the texgen mode.
    TexGenAttrib::Mode mode = _target_tex_gen->get_mode(stage);
    bool any_point_sprite = false;

    switch (mode) {
    case TexGenAttrib::M_off:
    case TexGenAttrib::M_unused2:
      set_texture_stage_state(si, D3DTSS_TEXCOORDINDEX, texcoord_index);
      break;

    case TexGenAttrib::M_eye_sphere_map:
      {
        set_texture_stage_state(si, D3DTSS_TEXCOORDINDEX,
                                texcoord_index | D3DTSS_TCI_CAMERASPACEREFLECTIONVECTOR);
        // This texture matrix, applied on top of the texcoord computed by
        // D3DTSS_TCI_CAMERASPACEREFLECTIONVECTOR, approximates the effect
        // produced by OpenGL's GL_SPHERE_MAP.
        static CPT(TransformState) sphere_map =
          TransformState::make_mat(LMatrix4(0.33, 0.0f, 0.0f, 0.0f,
                                             0.0f, 0.33, 0.0f, 0.0f,
                                             0.0f, 0.0f, 1.0f, 0.0f,
                                             0.5f, 0.5f, 0.0f, 1.0f));
        tex_mat = tex_mat->compose(sphere_map);
        texcoord_dimensions = 3;
      }
      break;

    case TexGenAttrib::M_world_cube_map:
      // To achieve world reflection vector, we must transform camera
      // coordinates to world coordinates; i.e.  apply the camera transform.
      // In the case of a vector, we should not apply the pos component of the
      // transform.
      {
        set_texture_stage_state(si, D3DTSS_TEXCOORDINDEX,
                                texcoord_index | D3DTSS_TCI_CAMERASPACEREFLECTIONVECTOR);
        texcoord_dimensions = 3;
        CPT(TransformState) camera_transform = _scene_setup->get_camera_transform()->compose(_inv_cs_transform);
        tex_mat = tex_mat->compose(camera_transform->set_pos(LVecBase3::zero()));
      }
      break;

    case TexGenAttrib::M_eye_cube_map:
      set_texture_stage_state(si, D3DTSS_TEXCOORDINDEX,
                              texcoord_index | D3DTSS_TCI_CAMERASPACEREFLECTIONVECTOR);
      tex_mat = tex_mat->compose(_inv_cs_transform);
      texcoord_dimensions = 3;
      break;

    case TexGenAttrib::M_world_normal:
      // To achieve world normal, we must transform camera coordinates to
      // world coordinates; i.e.  apply the camera transform.  In the case of
      // a normal, we should not apply the pos component of the transform.
      {
        set_texture_stage_state(si, D3DTSS_TEXCOORDINDEX,
                                texcoord_index | D3DTSS_TCI_CAMERASPACENORMAL);
        texcoord_dimensions = 3;
        CPT(TransformState) camera_transform = _scene_setup->get_camera_transform()->compose(_inv_cs_transform);
        tex_mat = tex_mat->compose(camera_transform->set_pos(LVecBase3::zero()));
      }
      break;

    case TexGenAttrib::M_eye_normal:
      set_texture_stage_state(si, D3DTSS_TEXCOORDINDEX,
                              texcoord_index | D3DTSS_TCI_CAMERASPACENORMAL);
      texcoord_dimensions = 3;
      tex_mat = tex_mat->compose(_inv_cs_transform);
      break;

    case TexGenAttrib::M_world_position:
      // To achieve world position, we must transform camera coordinates to
      // world coordinates; i.e.  apply the camera transform.
      {
        set_texture_stage_state(si, D3DTSS_TEXCOORDINDEX,
                                texcoord_index | D3DTSS_TCI_CAMERASPACEPOSITION);
        texcoord_dimensions = 3;
        CPT(TransformState) camera_transform = _scene_setup->get_camera_transform()->compose(_inv_cs_transform);
        tex_mat = tex_mat->compose(camera_transform);
      }
      break;

    case TexGenAttrib::M_eye_position:
      set_texture_stage_state(si, D3DTSS_TEXCOORDINDEX,
                              texcoord_index | D3DTSS_TCI_CAMERASPACEPOSITION);
      texcoord_dimensions = 3;
      tex_mat = tex_mat->compose(_inv_cs_transform);
      break;

    case TexGenAttrib::M_point_sprite:
      set_texture_stage_state(si, D3DTSS_TEXCOORDINDEX, texcoord_index);
      any_point_sprite = true;
      break;

    case TexGenAttrib::M_constant:
      // To generate a constant UV(w) coordinate everywhere, we use
      // CAMERASPACEPOSITION coordinates, but we construct a special matrix
      // that flattens the existing values to zero and then adds our desired
      // value.

      // The only reason we need to specify CAMERASPACEPOSITION at all,
      // instead of using whatever texture coordinates (if any) happen to be
      // on the vertices, is because we need to guarantee that there are 3-d
      // texture coordinates, because of the 3-component texture coordinate in
      // get_constant_value().
      {
        set_texture_stage_state(si, D3DTSS_TEXCOORDINDEX,
                                texcoord_index | D3DTSS_TCI_CAMERASPACEPOSITION);
        texcoord_dimensions = 3;

        const LTexCoord3 &v = _target_tex_gen->get_constant_value(stage);
        CPT(TransformState) squash =
          TransformState::make_pos_hpr_scale(v, LVecBase3::zero(),
                                             LVecBase3::zero());
        tex_mat = tex_mat->compose(squash);
      }
      break;
    }

    set_render_state(D3DRS_POINTSPRITEENABLE, any_point_sprite);

    if (!tex_mat->is_identity()) {
      if (/*tex_mat->is_2d() &&*/ texcoord_dimensions <= 2) {
        // For 2-d texture coordinates, we have to reorder the matrix.
        LMatrix4 m = tex_mat->get_mat();
        LMatrix4f mf;
        mf.set(m(0, 0), m(0, 1), m(0, 3), 0.0f,
               m(1, 0), m(1, 1), m(1, 3), 0.0f,
               m(3, 0), m(3, 1), m(3, 3), 0.0f,
               0.0f, 0.0f, 0.0f, 1.0f);
        _d3d_device->SetTransform(get_tex_mat_sym(si), (D3DMATRIX *)mf.get_data());
        set_texture_stage_state(si, D3DTSS_TEXTURETRANSFORMFLAGS,
                                D3DTTFF_COUNT2);
      } else {
        LMatrix4f mf = LCAST(float, tex_mat->get_mat());
        _d3d_device->SetTransform(get_tex_mat_sym(si), (D3DMATRIX *)mf.get_data());
        DWORD transform_flags = texcoord_dimensions;
        if (mf.get_col(3) != LVecBase4f(0.0f, 0.0f, 0.0f, 1.0f)) {
          // If we have a projected texture matrix, we also need to set
          // D3DTTFF_COUNT4.
          transform_flags = D3DTTFF_COUNT4 | D3DTTFF_PROJECTED;
        }
        set_texture_stage_state(si, D3DTSS_TEXTURETRANSFORMFLAGS,
                                transform_flags);
      }

    } else {
      set_texture_stage_state(si, D3DTSS_TEXTURETRANSFORMFLAGS,
                              D3DTTFF_DISABLE);
      // For some reason, "disabling" texture coordinate transforms doesn't
      // seem to be sufficient.  We'll load an identity matrix to underscore
      // the point.
      _d3d_device->SetTransform(get_tex_mat_sym(si), &_d3d_ident_mat);
    }
  }

  // Disable the texture stages that are no longer used.
  for (si = num_stages; si < _num_active_texture_stages; si++) {
    set_texture_stage_state(si, D3DTSS_COLOROP, D3DTOP_DISABLE);
    _d3d_device->SetTexture(si, nullptr);
  }

  // Save the count of texture stages for next time.
  _num_active_texture_stages = num_stages;
}

/**
 * Called after any of the things that might change blending state have
 * changed, this function is responsible for setting the appropriate color
 * blending mode based on the current properties.
 */
void DXGraphicsStateGuardian9::
do_issue_blending() {
  // Handle the color_write attrib.  If color_write is off, then all the other
  // blending-related stuff doesn't matter.  If the device doesn't support
  // color-write, we use blending tricks to effectively disable color write.
  const ColorWriteAttrib *target_color_write = DCAST(ColorWriteAttrib, _target_rs->get_attrib_def(ColorWriteAttrib::get_class_slot()));
  unsigned int color_channels =
    target_color_write->get_channels() & _color_write_mask;
  if (_target_shader->get_flag(ShaderAttrib::F_disable_alpha_write)) {
    color_channels &= ~(ColorWriteAttrib::C_alpha);
  }
  if (color_channels == ColorWriteAttrib::C_off) {
    if (_screen->_can_direct_disable_color_writes) {
      set_render_state(D3DRS_ALPHABLENDENABLE, FALSE);
      set_render_state(D3DRS_COLORWRITEENABLE, (DWORD)0x0);
    } else {
      set_render_state(D3DRS_ALPHABLENDENABLE, TRUE);
      set_render_state(D3DRS_SRCBLEND, D3DBLEND_ZERO);
      set_render_state(D3DRS_DESTBLEND, D3DBLEND_ONE);
    }
    return;
  } else {
    if (_screen->_can_direct_disable_color_writes) {
      set_render_state(D3DRS_COLORWRITEENABLE, color_channels);
    }
  }

  const ColorBlendAttrib *color_blend;
  _target_rs->get_attrib_def(color_blend);
  ColorBlendAttrib::Mode color_blend_mode = color_blend->get_mode();

  const TransparencyAttrib *target_transparency;
  _target_rs->get_attrib_def(target_transparency);
  TransparencyAttrib::Mode transparency_mode = target_transparency->get_mode();

  // Is there a color blend set?
  if (color_blend_mode != ColorBlendAttrib::M_none) {
    set_render_state(D3DRS_ALPHABLENDENABLE, TRUE);
    set_render_state(D3DRS_SEPARATEALPHABLENDENABLE, TRUE);
    set_render_state(D3DRS_BLENDOP, get_blend_mode(color_blend_mode));
    set_render_state(D3DRS_BLENDOPALPHA, get_blend_mode(color_blend->get_alpha_mode()));
    set_render_state(D3DRS_SRCBLEND, get_blend_func(color_blend->get_operand_a()));
    set_render_state(D3DRS_DESTBLEND, get_blend_func(color_blend->get_operand_b()));
    set_render_state(D3DRS_SRCBLENDALPHA, get_blend_func(color_blend->get_alpha_operand_a()));
    set_render_state(D3DRS_DESTBLENDALPHA, get_blend_func(color_blend->get_alpha_operand_b()));
    return;
  }

  // No color blend; is there a transparency set?
  switch (transparency_mode) {
  case TransparencyAttrib::M_none:
  case TransparencyAttrib::M_binary:
    break;

  case TransparencyAttrib::M_alpha:
  case TransparencyAttrib::M_multisample:
  case TransparencyAttrib::M_multisample_mask:
  case TransparencyAttrib::M_dual:
    set_render_state(D3DRS_ALPHABLENDENABLE, TRUE);
    if (old_alpha_blend) {
      set_render_state(D3DRS_SEPARATEALPHABLENDENABLE, FALSE);
    } else {
      set_render_state(D3DRS_SEPARATEALPHABLENDENABLE, TRUE);
      set_render_state(D3DRS_BLENDOPALPHA, D3DBLENDOP_ADD);
      set_render_state(D3DRS_SRCBLENDALPHA, D3DBLEND_ONE);
      set_render_state(D3DRS_DESTBLENDALPHA, D3DBLEND_INVSRCALPHA);
    }
    set_render_state(D3DRS_BLENDOP, D3DBLENDOP_ADD);
    set_render_state(D3DRS_SRCBLEND, D3DBLEND_SRCALPHA);
    set_render_state(D3DRS_DESTBLEND, D3DBLEND_INVSRCALPHA);
    return;

  case TransparencyAttrib::M_premultiplied_alpha:
    set_render_state(D3DRS_ALPHABLENDENABLE, TRUE);
    set_render_state(D3DRS_SEPARATEALPHABLENDENABLE, FALSE);
    set_render_state(D3DRS_BLENDOP, D3DBLENDOP_ADD);
    set_render_state(D3DRS_SRCBLEND, D3DBLEND_ONE);
    set_render_state(D3DRS_DESTBLEND, D3DBLEND_INVSRCALPHA);
    return;

  default:
    dxgsg9_cat.error()
      << "invalid transparency mode " << (int)transparency_mode << endl;
    break;
  }

  // Nothing's set, so disable blending.
  set_render_state(D3DRS_ALPHABLENDENABLE, FALSE);
}

/**
 * Called by clear_state_and_transform() to ensure that the current modelview
 * and projection matrices are properly loaded in the graphics state, after a
 * callback might have mucked them up.
 */
void DXGraphicsStateGuardian9::
reissue_transforms() {
  prepare_lens();
  do_issue_transform();
}

/**
 * Intended to be overridden by a derived class to enable or disable the use
 * of lighting overall.  This is called by issue_light() according to whether
 * any lights are in use or not.
 */
void DXGraphicsStateGuardian9::
enable_lighting(bool enable) {
  set_render_state(D3DRS_LIGHTING, (DWORD)enable);
}

/**
 * Intended to be overridden by a derived class to indicate the color of the
 * ambient light that should be in effect.  This is called by issue_light()
 * after all other lights have been enabled or disabled.
 */
void DXGraphicsStateGuardian9::
set_ambient_light(const LColor &color) {
  LColor c = color;
  c.set(c[0] * _light_color_scale[0],
        c[1] * _light_color_scale[1],
        c[2] * _light_color_scale[2],
        c[3] * _light_color_scale[3]);

  set_render_state(D3DRS_AMBIENT, LColor_to_D3DCOLOR(c));
}

/**
 * Intended to be overridden by a derived class to enable the indicated light
 * id.  A specific Light will already have been bound to this id via
 * bind_light().
 */
void DXGraphicsStateGuardian9::
enable_light(int light_id, bool enable) {
  HRESULT hr = _d3d_device->LightEnable(light_id, enable);

  if (FAILED(hr)) {
    wdxdisplay9_cat.warning()
      << "Could not enable light " << light_id << ": "
      << D3DERRORSTRING(hr) << "\n";
  }
}

/**
 * Intended to be overridden by a derived class to enable the indicated
 * clip_plane id.  A specific PlaneNode will already have been bound to this
 * id via bind_clip_plane().
 */
void DXGraphicsStateGuardian9::
enable_clip_plane(int plane_id, bool enable) {
  if (enable) {
    _clip_plane_bits |= ((DWORD)1 << plane_id);
  } else {
    _clip_plane_bits &= ~((DWORD)1 << plane_id);
  }
  set_render_state(D3DRS_CLIPPLANEENABLE, _clip_plane_bits);
}

/**
 * Called the first time a particular clip_plane has been bound to a given id
 * within a frame, this should set up the associated hardware clip_plane with
 * the clip_plane's properties.
 */
void DXGraphicsStateGuardian9::
bind_clip_plane(const NodePath &plane, int plane_id) {
  // Get the plane in "world coordinates" (actually, view coordinates).  This
  // means the plane in the coordinate space of the camera, converted to DX's
  // coordinate system.
  CPT(TransformState) transform = plane.get_transform(_scene_setup->get_camera_path());
  const LMatrix4 &plane_mat = transform->get_mat();
  LMatrix4 rel_mat = plane_mat * LMatrix4::convert_mat(CS_yup_left, CS_default);
  const PlaneNode *plane_node;
  DCAST_INTO_V(plane_node, plane.node());
  LPlanef world_plane = LCAST(float, plane_node->get_plane() * rel_mat);

  HRESULT hr = _d3d_device->SetClipPlane(plane_id, world_plane.get_data());
  if (FAILED(hr)) {
    wdxdisplay9_cat.warning()
      << "Could not set clip plane for " << plane
      << " to id " << plane_id << ": " << D3DERRORSTRING(hr) << "\n";
  }
}

/**
 * This is called by the associated GraphicsWindow when close_window() is
 * called.  It should null out the _win pointer and possibly free any open
 * resources associated with the GSG.
 */
void DXGraphicsStateGuardian9::
close_gsg() {
  GraphicsStateGuardian::close_gsg();

  if (_prepared_objects.is_null()) {
    return;
  }

  if (dxgsg9_cat.is_debug()) {
    dxgsg9_cat.debug()
      << "Closing GSG, prepared_objects count = "
      << _prepared_objects->get_ref_count() << "\n";
  }

  // Unlike in OpenGL, in DX9 it is safe to try to explicitly release any
  // textures here.  And it may even be a good idea.
  if (_prepared_objects->get_ref_count() == 1) {
    release_all();

    // Now we need to actually delete all of the objects we just released.
    Thread *current_thread = Thread::get_current_thread();
    _prepared_objects->begin_frame(this, current_thread);
    _prepared_objects->end_frame(current_thread);
  }
}

/**
 * Frees some memory that was explicitly allocated within the dxgsg.
 */
void DXGraphicsStateGuardian9::
free_nondx_resources() {
#ifdef HAVE_CG
  if (_cg_context) {
    cgDestroyContext(_cg_context);
    _cg_context = 0;
  }
#endif
}

/**
 * setup for re-calling dx_init(), this is not the final exit cleanup routine
 * (see dx_cleanup)
 */
void DXGraphicsStateGuardian9::
free_d3d_device() {
  // dont want a full reset of gsg, just a state clear
  _state_rs = RenderState::make_empty();
  _state_mask.clear();

  // want gsg to pass all state settings through

  _dx_is_ready = false;

  if (_d3d_device != nullptr) {
    for(int i = 0; i < D3D_MAXTEXTURESTAGES; i++) {
      // d3d should release this stuff internally anyway, but whatever
      _d3d_device->SetTexture(i, nullptr);
    }
  }

  release_all();

  if (_d3d_device != nullptr) {
    RELEASE(_d3d_device, dxgsg9, "d3dDevice", RELEASE_DOWN_TO_ZERO);
  }

  free_nondx_resources();

  // obviously we dont release ID3D9, just ID3DDevice9
}

/**
 * Sets up the glDrawBuffer to render into the buffer indicated by the
 * RenderBuffer object.  This only sets up the color bits; it does not affect
 * the depth, stencil, accum layers.
 */
void DXGraphicsStateGuardian9::
set_draw_buffer(const RenderBuffer &rb) {
  dxgsg9_cat.fatal() << "DX set_draw_buffer unimplemented!!!";
  return;
}

/**
 * Vestigial analog of glReadBuffer
 */
void DXGraphicsStateGuardian9::
set_read_buffer(const RenderBuffer &rb) {
  if (rb._buffer_type & RenderBuffer::T_front) {
    _cur_read_pixel_buffer = RenderBuffer::T_front;
  } else  if (rb._buffer_type & RenderBuffer::T_back) {
    _cur_read_pixel_buffer = RenderBuffer::T_back;
  } else  if (rb._buffer_type & RenderBuffer::T_aux_rgba_ALL) {
    _cur_read_pixel_buffer = RenderBuffer::T_back;
  } else {
    dxgsg9_cat.error() << "Invalid or unimplemented Argument to set_read_buffer!\n";
  }
  return;
}

/**
 * Returns the array of four floats that should be issued as the light's
 * color, as scaled by the current value of _light_color_scale, in the case of
 * color_scale_via_lighting.
 */
const D3DCOLORVALUE &DXGraphicsStateGuardian9::
get_light_color(Light *light) const {
  LColor c = light->get_color();
  static LColorf cf;
  cf.set(c[0] * _light_color_scale[0],
         c[1] * _light_color_scale[1],
         c[2] * _light_color_scale[2],
         c[3] * _light_color_scale[3]);
  return *(D3DCOLORVALUE *)cf.get_data();
}

/**
 * Maps from ColorBlendAttrib::Mode to D3DBLENDOP vaule.
 */
D3DBLENDOP DXGraphicsStateGuardian9::
get_blend_mode(ColorBlendAttrib::Mode mode) {
  switch (mode) {
  case ColorBlendAttrib::M_add:
    return D3DBLENDOP_ADD;

  case ColorBlendAttrib::M_subtract:
    return D3DBLENDOP_SUBTRACT;

  case ColorBlendAttrib::M_inv_subtract:
    return D3DBLENDOP_REVSUBTRACT;

  case ColorBlendAttrib::M_min:
    return D3DBLENDOP_MIN;

  case ColorBlendAttrib::M_max:
    return D3DBLENDOP_MAX;
  }

  dxgsg9_cat.error()
    << "Unknown color blend mode " << (int)mode << endl;
  return D3DBLENDOP_ADD;
}

/**
 * Maps from ColorBlendAttrib::Operand to D3DBLEND value.
 */
D3DBLEND DXGraphicsStateGuardian9::
get_blend_func(ColorBlendAttrib::Operand operand) {
  switch (operand) {
  case ColorBlendAttrib::O_zero:
    return D3DBLEND_ZERO;

  case ColorBlendAttrib::O_one:
    return D3DBLEND_ONE;

  case ColorBlendAttrib::O_incoming_color:
    return D3DBLEND_SRCCOLOR;

  case ColorBlendAttrib::O_one_minus_incoming_color:
    return D3DBLEND_INVSRCCOLOR;

  case ColorBlendAttrib::O_fbuffer_color:
    return D3DBLEND_DESTCOLOR;

  case ColorBlendAttrib::O_one_minus_fbuffer_color:
    return D3DBLEND_INVDESTCOLOR;

  case ColorBlendAttrib::O_incoming_alpha:
    return D3DBLEND_SRCALPHA;

  case ColorBlendAttrib::O_one_minus_incoming_alpha:
    return D3DBLEND_INVSRCALPHA;

  case ColorBlendAttrib::O_fbuffer_alpha:
    return D3DBLEND_DESTALPHA;

  case ColorBlendAttrib::O_one_minus_fbuffer_alpha:
    return D3DBLEND_INVDESTALPHA;

  case ColorBlendAttrib::O_constant_color:
    // Not supported by DX.
    return D3DBLEND_SRCCOLOR;

  case ColorBlendAttrib::O_one_minus_constant_color:
    // Not supported by DX.
    return D3DBLEND_INVSRCCOLOR;

  case ColorBlendAttrib::O_constant_alpha:
    // Not supported by DX.
    return D3DBLEND_SRCALPHA;

  case ColorBlendAttrib::O_one_minus_constant_alpha:
    // Not supported by DX.
    return D3DBLEND_INVSRCALPHA;

  case ColorBlendAttrib::O_incoming_color_saturate:
    return D3DBLEND_SRCALPHASAT;

  case ColorBlendAttrib::O_incoming1_color:
    return (D3DBLEND)16; //D3DBLEND_SRCCOLOR2;

  case ColorBlendAttrib::O_one_minus_incoming1_color:
    return (D3DBLEND)17; //D3DBLEND_INVSRCCOLOR2;

  case ColorBlendAttrib::O_incoming1_alpha:
    // Not supported by DX9.
    return (D3DBLEND)18;

  case ColorBlendAttrib::O_one_minus_incoming1_alpha:
    // Not supported by DX9.
    return (D3DBLEND)19;
  }

  dxgsg9_cat.error()
    << "Unknown color blend operand " << (int)operand << endl;
  return D3DBLEND_ZERO;
}

/**
 * Reports the DX texture manager's activity to PStats.
 */
void DXGraphicsStateGuardian9::
report_texmgr_stats() {

#ifdef DO_PSTATS
  HRESULT hr;
  hr = 0;

#ifdef TEXMGRSTATS_USES_GETAVAILVIDMEM
  DWORD dwTexTotal, dwTexFree, dwVidTotal, dwVidFree;

  if (_total_texmem_pcollector.is_active()) {
    DDSCAPS2 ddsCaps;

    ZeroMemory(&ddsCaps, sizeof(ddsCaps));

    ddsCaps.dwCaps = DDSCAPS_VIDEOMEMORY | DDSCAPS_PRIMARYSURFACE | DDSCAPS_3DDEVICE;
    if (FAILED( hr = _d3d_device->GetAvailableVidMem(&ddsCaps, &dwVidTotal, &dwVidFree))) {
      dxgsg9_cat.fatal() << "report_texmgr GetAvailableVidMem for VIDMEM failed : result = " << D3DERRORSTRING(hr);
      throw_event("panda3d-render-error");
      return;
    }

    ddsCaps.dwCaps = DDSCAPS_TEXTURE;
    if (FAILED( hr = _d3d_device->GetAvailableVidMem(&ddsCaps, &dwTexTotal, &dwTexFree))) {
      dxgsg9_cat.fatal() << "report_texmgr GetAvailableVidMem for TEXTURE failed : result = " << D3DERRORSTRING(hr);
      throw_event("panda3d-render-error");
      return;
    }
  }
#endif  // TEXMGRSTATS_USES_GETAVAILVIDMEM

  D3DDEVINFO_RESOURCEMANAGER all_resource_stats;
  ZeroMemory(&all_resource_stats, sizeof(D3DDEVINFO_RESOURCEMANAGER));

/* ***** DX9, GetInfo ( ) NOT IN DX9 */
/*
  if (!_tex_stats_retrieval_impossible) {
    hr = _d3d_device->GetInfo(D3DDEVINFOID_RESOURCEMANAGER, &all_resource_stats, sizeof(D3DDEVINFO_RESOURCEMANAGER));
    if (hr != D3D_OK) {
      if (hr == S_FALSE) {
        static int PrintedMsg = 2;
        if (PrintedMsg>0) {
          if (dxgsg9_cat.is_debug()) {
            dxgsg9_cat.debug()
              << "texstats GetInfo() requires debug DX DLLs to be installed!!\n";
          }
          ZeroMemory(&all_resource_stats, sizeof(D3DDEVINFO_RESOURCEMANAGER));
          _tex_stats_retrieval_impossible = true;
        }
      } else {
        dxgsg9_cat.error() << "GetInfo(RESOURCEMANAGER) failed to get tex stats: result = " << D3DERRORSTRING(hr);
        return;
      }
    }
  }
*/

  // Tell PStats about the state of the texture memory.

  if (_texmgrmem_total_pcollector.is_active()) {
    // report zero if no debug dlls, to signal this info is invalid
    _texmgrmem_total_pcollector.set_level(all_resource_stats.stats[D3DRTYPE_TEXTURE].TotalBytes);
    _texmgrmem_resident_pcollector.set_level(all_resource_stats.stats[D3DRTYPE_TEXTURE].WorkingSetBytes);
  }
#ifdef TEXMGRSTATS_USES_GETAVAILVIDMEM
  if (_total_texmem_pcollector.is_active()) {
    _total_texmem_pcollector.set_level(dwTexTotal);
    _used_texmem_pcollector.set_level(dwTexTotal - dwTexFree);
  }
#endif  // TEXMGRSTATS_USES_GETAVAILVIDMEM
#endif  // DO_PSTATS
}

/**
 *
 */
void DXGraphicsStateGuardian9::
set_context(DXScreenData *new_context) {
  nassertv(new_context != nullptr);
  _screen = new_context;
  _d3d_device = _screen->_d3d_device;   //copy this one field for speed of deref
  _swap_chain = _screen->_swap_chain;   //copy this one field for speed of deref

  _screen->_dxgsg9 = this;
  set_cg_device(_d3d_device);
}

/**
 * Set render target to the backbuffer of current swap chain.
 */
void DXGraphicsStateGuardian9::
set_render_target() {
  if (_d3d_device == nullptr) {
    return;
  }

  LPDIRECT3DSURFACE9 back = nullptr, stencil = nullptr;

  UINT swap_chain;

  /* ***** DX9 swap_chain ??? */
  swap_chain = 0;

  if (!_swap_chain)  //maybe fullscreen mode or main/single window
    _d3d_device->GetBackBuffer(swap_chain, 0, D3DBACKBUFFER_TYPE_MONO, &back);
  else
    _swap_chain->GetBackBuffer(0, D3DBACKBUFFER_TYPE_MONO, &back);

  // wdxdisplay9_cat.debug() << "swapchain is " << _swap_chain << "\n";
  // wdxdisplay9_cat.debug() << "back buffer is " << back << "\n";

  _d3d_device->GetDepthStencilSurface(&stencil);

// _d3d_device->SetRenderTarget(back, stencil);
  DWORD render_target_index;
  render_target_index = 0;
  _d3d_device->SetRenderTarget(render_target_index, back);

  if (back) {
    back->Release();
  }
  if (stencil) {
    stencil->Release();
  }
}

/**
 *
 */
void DXGraphicsStateGuardian9::
set_texture_blend_mode(int i, const TextureStage *stage) {
  switch (stage->get_mode()) {
  case TextureStage::M_modulate:
  case TextureStage::M_modulate_glow:
  case TextureStage::M_modulate_gloss:
    // emulates GL_MODULATE glTexEnv mode
    set_texture_stage_state(i, D3DTSS_COLOROP, D3DTOP_MODULATE);
    set_texture_stage_state(i, D3DTSS_COLORARG1, D3DTA_TEXTURE);
    set_texture_stage_state(i, D3DTSS_COLORARG2, D3DTA_CURRENT);
    set_texture_stage_state(i, D3DTSS_ALPHAOP, D3DTOP_MODULATE);
    set_texture_stage_state(i, D3DTSS_ALPHAARG1, D3DTA_TEXTURE);
    set_texture_stage_state(i, D3DTSS_ALPHAARG2, D3DTA_CURRENT);
    break;

  case TextureStage::M_decal:
    // emulates GL_DECAL glTexEnv mode
    set_texture_stage_state(i, D3DTSS_COLOROP, D3DTOP_BLENDTEXTUREALPHA);
    set_texture_stage_state(i, D3DTSS_COLORARG1, D3DTA_TEXTURE);
    set_texture_stage_state(i, D3DTSS_COLORARG2, D3DTA_CURRENT);

    set_texture_stage_state(i, D3DTSS_ALPHAOP, D3DTOP_SELECTARG1);
    set_texture_stage_state(i, D3DTSS_ALPHAARG1, D3DTA_CURRENT);
    break;

  case TextureStage::M_replace:
    set_texture_stage_state(i, D3DTSS_COLOROP, D3DTOP_SELECTARG1);
    set_texture_stage_state(i, D3DTSS_COLORARG1, D3DTA_TEXTURE);

    set_texture_stage_state(i, D3DTSS_ALPHAOP, D3DTOP_SELECTARG1);
    set_texture_stage_state(i, D3DTSS_ALPHAARG1, D3DTA_TEXTURE);
    break;

  case TextureStage::M_add:
    set_texture_stage_state(i, D3DTSS_COLOROP, D3DTOP_ADD);
    set_texture_stage_state(i, D3DTSS_COLORARG1, D3DTA_TEXTURE);
    set_texture_stage_state(i, D3DTSS_COLORARG2, D3DTA_CURRENT);

    set_texture_stage_state(i, D3DTSS_ALPHAOP, D3DTOP_MODULATE);
    set_texture_stage_state(i, D3DTSS_ALPHAARG1, D3DTA_TEXTURE);
    set_texture_stage_state(i, D3DTSS_ALPHAARG2, D3DTA_CURRENT);
    break;

  case TextureStage::M_blend:
  case TextureStage::M_blend_color_scale:
    {
      set_texture_stage_state(i, D3DTSS_COLOROP, D3DTOP_LERP);
      set_texture_stage_state(i, D3DTSS_COLORARG0, D3DTA_TEXTURE);
      set_texture_stage_state(i, D3DTSS_COLORARG2, D3DTA_CURRENT);
      set_texture_stage_state(i, D3DTSS_COLORARG1, _constant_color_operand);

      set_texture_stage_state(i, D3DTSS_ALPHAOP, D3DTOP_MODULATE);
      set_texture_stage_state(i, D3DTSS_ALPHAARG1, D3DTA_TEXTURE);
      set_texture_stage_state(i, D3DTSS_ALPHAARG2, D3DTA_CURRENT);
    }
    break;

  case TextureStage::M_combine:
    // M_combine mode begins a collection of more sophisticated modes, which
    // match up more closely with DirectX's built-in modes.
    set_texture_stage_state
      (i, D3DTSS_COLOROP,
       get_texture_operation(stage->get_combine_rgb_mode(),
                             stage->get_rgb_scale()));

    switch (stage->get_num_combine_rgb_operands()) {
    case 3:
      set_texture_stage_state
        (i, D3DTSS_COLORARG0,
         get_texture_argument(stage->get_combine_rgb_source2(),
                              stage->get_combine_rgb_operand2()));
      // fall through

    case 2:
      set_texture_stage_state
        (i, D3DTSS_COLORARG2,
         get_texture_argument(stage->get_combine_rgb_source1(),
                              stage->get_combine_rgb_operand1()));
      // fall through

    case 1:
      set_texture_stage_state
        (i, D3DTSS_COLORARG1,
         get_texture_argument(stage->get_combine_rgb_source0(),
                              stage->get_combine_rgb_operand0()));
      // fall through

    default:
      break;
    }

    set_texture_stage_state
      (i, D3DTSS_ALPHAOP,
       get_texture_operation(stage->get_combine_alpha_mode(),
                             stage->get_alpha_scale()));

    switch (stage->get_num_combine_alpha_operands()) {
    case 3:
      set_texture_stage_state
        (i, D3DTSS_ALPHAARG0,
         get_texture_argument(stage->get_combine_alpha_source2(),
                              stage->get_combine_alpha_operand2()));
      // fall through

    case 2:
      set_texture_stage_state
        (i, D3DTSS_ALPHAARG2,
         get_texture_argument(stage->get_combine_alpha_source1(),
                              stage->get_combine_alpha_operand1()));
      // fall through

    case 1:
      set_texture_stage_state
        (i, D3DTSS_ALPHAARG1,
         get_texture_argument(stage->get_combine_alpha_source0(),
                              stage->get_combine_alpha_operand0()));
      // fall through

    default:
      break;
    }
    break;

  default:
    dxgsg9_cat.error()
      << "Unknown texture mode " << (int)stage->get_mode() << endl;
    break;
  }

  if (stage->get_saved_result()) {
    set_texture_stage_state(i, D3DTSS_RESULTARG, D3DTA_TEMP);
  } else {
    set_texture_stage_state(i, D3DTSS_RESULTARG, D3DTA_CURRENT);
  }

  if (stage->uses_color() || stage->involves_color_scale()) {
    // Set up the constant color for this stage.

    D3DCOLOR constant_color;
    if (stage->involves_color_scale() && _color_scale_enabled) {
      LColor color = stage->get_color();
      color.set(color[0] * _current_color_scale[0],
                color[1] * _current_color_scale[1],
                color[2] * _current_color_scale[2],
                color[3] * _current_color_scale[3]);
      _texture_involves_color_scale = true;
      constant_color = LColor_to_D3DCOLOR(color);
    } else {
      constant_color = LColor_to_D3DCOLOR(stage->get_color());
    }
    if (_supports_texture_constant_color) {
      set_texture_stage_state(i, D3DTSS_CONSTANT, constant_color);
    } else {
      // This device doesn't supoprt a per-stage constant color, so we have to
      // fall back to a single constant color for the overall texture
      // pipeline.
      set_render_state(D3DRS_TEXTUREFACTOR, constant_color);
    }
  }
}

/**
 * Clean up the DirectX environment, accounting for exit()
 */
void DXGraphicsStateGuardian9::
dx_cleanup() {
  if (!_d3d_device) {
    return;
  }

  free_nondx_resources();
  PRINT_REFCNT(dxgsg9, _d3d_device);

  // Do a safe check for releasing the D3DDEVICE. RefCount should be zero.  if
  // we're called from exit(), _d3d_device may already have been released
  RELEASE(_d3d_device, dxgsg9, "d3dDevice", RELEASE_DOWN_TO_ZERO);
  _screen->_d3d_device = nullptr;

  // Releasing pD3D is now the responsibility of the GraphicsPipe destructor
}

/**
 * This function checks current device's framebuffer dimension against passed
 * p_presentation_params backbuffer dimension to determine a device reset if
 * there is only one window or it is the main window or fullscreen mode then,
 * it resets the device.  Finally it returns the new DXScreenData through
 * parameter screen
 */
HRESULT DXGraphicsStateGuardian9::
reset_d3d_device(D3DPRESENT_PARAMETERS *presentation_params,
                 DXScreenData **screen) {
  HRESULT hr;

  nassertr(IS_VALID_PTR(presentation_params), E_FAIL);
  nassertr(IS_VALID_PTR(_screen->_d3d9), E_FAIL);
  nassertr(IS_VALID_PTR(_d3d_device), E_FAIL);

  // for windowed mode make sure our format matches the desktop fmt, in case
  // the desktop mode has been changed
  _screen->_d3d9->GetAdapterDisplayMode(_screen->_card_id, &_screen->_display_mode);
  presentation_params->BackBufferFormat = _screen->_display_mode.Format;

  // here we have to look at the _presentation_reset frame buffer dimension if
  // current window's dimension is bigger than _presentation_reset we have to
  // reset the device before creating new swapchain.  inorder to reset
  // properly, we need to release all swapchains

  if (true || !(_screen->_swap_chain)
      || (_presentation_reset.BackBufferWidth < presentation_params->BackBufferWidth)
      || (_presentation_reset.BackBufferHeight < presentation_params->BackBufferHeight)) {
    if (wdxdisplay9_cat.is_debug()) {
      wdxdisplay9_cat.debug()
        << "swap_chain = " << _screen->_swap_chain << " _presentation_reset = "
        << _presentation_reset.BackBufferWidth << "x" << _presentation_reset.BackBufferHeight
        << " presentation_params = "
        << presentation_params->BackBufferWidth << "x" << presentation_params->BackBufferHeight << "\n";
    }

    get_engine()->reset_all_windows(false);// reset old swapchain by releasing

    if (_screen->_swap_chain) {  //other windows might be using bigger buffers
      _presentation_reset.BackBufferWidth = max(_presentation_reset.BackBufferWidth, presentation_params->BackBufferWidth);
      _presentation_reset.BackBufferHeight = max(_presentation_reset.BackBufferHeight, presentation_params->BackBufferHeight);

    } else {  // single window, must reset to the new presentation_params dimension
      _presentation_reset.BackBufferWidth = presentation_params->BackBufferWidth;
      _presentation_reset.BackBufferHeight = presentation_params->BackBufferHeight;
    }

    // Calling this forces all of the textures and vbuffers to be regenerated,
    // a prerequisite to calling Reset().
    release_all();

    // Just to be extra-conservative for now, we'll go ahead and release the
    // vbuffers and ibuffers at least; they're relatively cheap to replace.
    release_all_vertex_buffers();
    release_all_index_buffers();

    if (_white_vbuffer != nullptr) {
      _white_vbuffer->Release();
      _white_vbuffer = nullptr;
    }

    // must be called before reset
    Thread *current_thread = Thread::get_current_thread();
    _prepared_objects->begin_frame(this, current_thread);

    // release graphics buffer surfaces
    {
      wdxGraphicsBuffer9 *graphics_buffer;
      std::list <wdxGraphicsBuffer9 **>::iterator graphics_buffer_iterator;

      for (graphics_buffer_iterator = _graphics_buffer_list.begin( ); graphics_buffer_iterator != _graphics_buffer_list.end( ); graphics_buffer_iterator++)
      {
        graphics_buffer = **graphics_buffer_iterator;
        if (graphics_buffer -> _color_backing_store)
        {
          graphics_buffer -> _color_backing_store -> Release ( );
          graphics_buffer -> _color_backing_store = 0;
        }
        if (graphics_buffer -> _depth_backing_store)
        {
          graphics_buffer -> _depth_backing_store -> Release ( );
          graphics_buffer -> _depth_backing_store = 0;
        }
      }
    }

    mark_new();
    hr = _d3d_device->Reset(&_presentation_reset);
    if (FAILED(hr) && hr != D3DERR_DEVICELOST) {
      return hr;
    }

    get_engine()->reset_all_windows(true);// reset with new swapchains by creating
    if (screen) {
      *screen = nullptr;
    }

    if (presentation_params != &_screen->_presentation_params) {
      memcpy(&_screen->_presentation_params, presentation_params, sizeof(D3DPRESENT_PARAMETERS));
    }

    return hr;
  }

  // release the old swapchain and create a new one
  if (_screen && _screen->_swap_chain) {
    _screen->_swap_chain->Release();
    wdxdisplay9_cat.debug()
      << "swap chain " << _screen->_swap_chain << " is released\n";
    _screen->_swap_chain = nullptr;
    hr = _d3d_device->CreateAdditionalSwapChain(presentation_params, &_screen->_swap_chain);
  }
  if (SUCCEEDED(hr)) {
    if (presentation_params != &_screen->_presentation_params) {
      memcpy(&_screen->_presentation_params, presentation_params, sizeof(D3DPRESENT_PARAMETERS));
    }
    if (screen) {
      *screen = _screen;
    }
  }
  return hr;
}

/**
 *
 */
bool DXGraphicsStateGuardian9::
check_cooperative_level() {
  bool bDoReactivateWindow = false;
  if (_d3d_device == nullptr) {
    return false;
  }

  HRESULT hr = _d3d_device->TestCooperativeLevel();

  if (SUCCEEDED(hr)) {
    nassertr(SUCCEEDED(_last_testcooplevel_result), false);
    return true;
  }

  switch (hr) {
  case D3DERR_DEVICENOTRESET:
    _dx_is_ready = false;

    // call this just in case
    _prepared_objects->begin_frame(this, Thread::get_current_thread());

    hr = reset_d3d_device(&_screen->_presentation_params);
    if (FAILED(hr)) {
      // I think this shouldnt fail unless I've screwed up the
      // _presentation_params from the original working ones somehow
      dxgsg9_cat.error()
        << "check_cooperative_level Reset() failed, hr = " << D3DERRORSTRING(hr);
    }

    hr = _d3d_device->TestCooperativeLevel();
    if (FAILED(hr)) {
      // internal chk, shouldnt fail
      dxgsg9_cat.error()
        << "TestCooperativeLevel following Reset() failed, hr = " << D3DERRORSTRING(hr);

    }

    _dx_is_ready = TRUE;
    break;

  case D3DERR_DEVICELOST:
    // sleep while the device is lost to free up the CPU
    Sleep (10);

    if (SUCCEEDED(_last_testcooplevel_result)) {
      if (_dx_is_ready) {
        _dx_is_ready = false;
        if (dxgsg9_cat.is_debug()) {
          dxgsg9_cat.debug() << "D3D Device was Lost, waiting...\n";
        }
      }
    }
  }

  _last_testcooplevel_result = hr;
  return SUCCEEDED(hr);
}

/**
 * redraw primary buffer
 */
void DXGraphicsStateGuardian9::
show_frame() {
  if (_d3d_device == nullptr) {
    return;
  }

  HRESULT hr;

  if (_swap_chain) {
    DWORD flags;
    flags = 0;

    hr = _swap_chain->Present(nullptr, nullptr, (HWND)nullptr, nullptr, flags);
  } else {
    hr = _d3d_device->Present(nullptr, nullptr, (HWND)nullptr, nullptr);
  }

  if (FAILED(hr)) {
    if (hr == D3DERR_DEVICELOST) {
      check_cooperative_level();
    } else {
      dxgsg9_cat.error()
        << "show_frame() - Present() failed" << D3DERRORSTRING(hr);
      throw_event("panda3d-render-error");
    }
  }
}

/**
 *
 */
bool DXGraphicsStateGuardian9::
create_swap_chain(DXScreenData *new_context) {
  // Instead of creating a device and rendering as d3ddevice->present() we
  // should render using SwapChain->present(). This is done to support
  // multiple windows rendering.  For that purpose, we need to set additional
  // swap chains here.

  HRESULT hr;
  hr = new_context->_d3d_device->CreateAdditionalSwapChain(&new_context->_presentation_params, &new_context->_swap_chain);
  if (FAILED(hr)) {
    wdxdisplay9_cat.debug() << "Swapchain creation failed :"<<D3DERRORSTRING(hr)<<"\n";
    return false;
  }
  return true;
}

/**
 * Release the swap chain on this DXScreenData
 */
bool DXGraphicsStateGuardian9::
release_swap_chain(DXScreenData *new_context) {
  HRESULT hr;
  if (new_context->_swap_chain) {
    hr = new_context->_swap_chain->Release();
    if (FAILED(hr)) {
      wdxdisplay9_cat.debug() << "Swapchain release failed:" << D3DERRORSTRING(hr) << "\n";
      return false;
    }
  }
  return true;
}

/**
 * copies the PresReset from passed DXScreenData
 */
void DXGraphicsStateGuardian9::
copy_pres_reset(DXScreenData *screen) {
  memcpy(&_presentation_reset, &_screen->_presentation_params, sizeof(D3DPRESENT_PARAMETERS));
}

/**
 *
 */
D3DTEXTUREFILTERTYPE DXGraphicsStateGuardian9::
get_d3d_min_type(SamplerState::FilterType filter_type) {
  switch (filter_type) {
  case SamplerState::FT_nearest:
    return D3DTEXF_POINT;

  case SamplerState::FT_linear:
    return D3DTEXF_LINEAR;

  case SamplerState::FT_nearest_mipmap_nearest:
    return D3DTEXF_POINT;

  case SamplerState::FT_linear_mipmap_nearest:
    return D3DTEXF_LINEAR;

  case SamplerState::FT_nearest_mipmap_linear:
    return D3DTEXF_POINT;

  case SamplerState::FT_linear_mipmap_linear:
    return D3DTEXF_LINEAR;

  case SamplerState::FT_shadow:
  case SamplerState::FT_default:
    return D3DTEXF_LINEAR;
  }

  dxgsg9_cat.error()
    << "Invalid FilterType value (" << (int)filter_type << ")\n";
  return D3DTEXF_POINT;
}

/**
 *
 */
D3DTEXTUREFILTERTYPE DXGraphicsStateGuardian9::
get_d3d_mip_type(SamplerState::FilterType filter_type) {
  switch (filter_type) {
  case SamplerState::FT_nearest:
    return D3DTEXF_NONE;

  case SamplerState::FT_linear:
    return D3DTEXF_NONE;

  case SamplerState::FT_nearest_mipmap_nearest:
    return D3DTEXF_POINT;

  case SamplerState::FT_linear_mipmap_nearest:
    return D3DTEXF_POINT;

  case SamplerState::FT_nearest_mipmap_linear:
    return D3DTEXF_LINEAR;

  case SamplerState::FT_linear_mipmap_linear:
    return D3DTEXF_LINEAR;

  case SamplerState::FT_shadow:
  case SamplerState::FT_default:
    return D3DTEXF_NONE;
  }

  dxgsg9_cat.error()
    << "Invalid FilterType value (" << (int)filter_type << ")\n";
  return D3DTEXF_NONE;
}

/**
 * Returns the D3DTEXTUREOP value corresponding to the indicated
 * TextureStage::CombineMode enumerated type.
 */
D3DTEXTUREOP DXGraphicsStateGuardian9::
get_texture_operation(TextureStage::CombineMode mode, int scale) {
  switch (mode) {
  case TextureStage::CM_undefined:
  case TextureStage::CM_replace:
    return D3DTOP_SELECTARG1;

  case TextureStage::CM_modulate:
    if (scale < 2) {
      return D3DTOP_MODULATE;
    } else if (scale < 4) {
      return D3DTOP_MODULATE2X;
    } else {
      return D3DTOP_MODULATE4X;
    }

  case TextureStage::CM_add:
    return D3DTOP_ADD;

  case TextureStage::CM_add_signed:
    if (scale < 2) {
      return D3DTOP_ADDSIGNED;
    } else {
      return D3DTOP_ADDSIGNED2X;
    }

  case TextureStage::CM_interpolate:
    return D3DTOP_LERP;

  case TextureStage::CM_subtract:
    return D3DTOP_SUBTRACT;

  case TextureStage::CM_dot3_rgb:
  case TextureStage::CM_dot3_rgba:
    return D3DTOP_DOTPRODUCT3;
  }

  dxgsg9_cat.error()
    << "Invalid TextureStage::CombineMode value (" << (int)mode << ")\n";
  return D3DTOP_DISABLE;
}

/**
 * Returns the D3DTA value corresponding to the indicated
 * TextureStage::CombineSource and TextureStage::CombineOperand enumerated
 * types.
 */
DWORD DXGraphicsStateGuardian9::
get_texture_argument(TextureStage::CombineSource source,
                     TextureStage::CombineOperand operand) const {
  switch (source) {
  case TextureStage::CS_undefined:
  case TextureStage::CS_texture:
    return D3DTA_TEXTURE | get_texture_argument_modifier(operand);

  case TextureStage::CS_constant:
  case TextureStage::CS_constant_color_scale:
    return _constant_color_operand | get_texture_argument_modifier(operand);

  case TextureStage::CS_primary_color:
    return D3DTA_DIFFUSE | get_texture_argument_modifier(operand);

  case TextureStage::CS_previous:
    return D3DTA_CURRENT | get_texture_argument_modifier(operand);

  case TextureStage::CS_last_saved_result:
    return D3DTA_TEMP | get_texture_argument_modifier(operand);
  }
  dxgsg9_cat.error()
    << "Invalid TextureStage::CombineSource value (" << (int)source << ")\n";
  return D3DTA_CURRENT;
}

/**
 * Returns the extra bits that modify the D3DTA argument, according to the
 * indicated TextureStage::CombineOperand enumerated type.
 */
DWORD DXGraphicsStateGuardian9::
get_texture_argument_modifier(TextureStage::CombineOperand operand) {
  switch (operand) {
  case TextureStage::CO_src_color:
    return 0;

  case TextureStage::CO_one_minus_src_color:
    return D3DTA_COMPLEMENT;

  case TextureStage::CO_src_alpha:
    return D3DTA_ALPHAREPLICATE;

  case TextureStage::CO_one_minus_src_alpha:
    return D3DTA_ALPHAREPLICATE | D3DTA_COMPLEMENT;

  case TextureStage::CO_undefined:
    break;
  }
  dxgsg9_cat.error()
    << "Invalid TextureStage::CombineOperand value (" << (int)operand << ")\n";
  return 0;
}

/**
 * Issues the DrawPrimitiveUP call to draw the indicated primitive_type from
 * the given buffer.  We add the num_vertices parameter, so we can determine
 * the size of the buffer.
 */
void DXGraphicsStateGuardian9::
draw_primitive_up(D3DPRIMITIVETYPE primitive_type,
      unsigned int primitive_count,
      unsigned int first_vertex,
      unsigned int num_vertices,
      const unsigned char *buffer, size_t stride) {

  // It appears that the common ATI driver seems to fail to draw anything in
  // the DrawPrimitiveUP() call if the address range of the buffer supplied
  // crosses over a multiple of 0x10000.  That's incredibly broken, yet it
  // undeniably appears to be true.  We'll have to hack around it.

  const unsigned char *buffer_start = buffer + stride * first_vertex;
  const unsigned char *buffer_end = buffer_start + stride * num_vertices;

  if (buffer_end - buffer_start > 0x10000) {
    // Actually, the buffer doesn't fit within the required limit anyway.  Go
    // ahead and draw it and hope for the best.
    _d3d_device->DrawPrimitiveUP(primitive_type, primitive_count,
         buffer_start, stride);

  } else if ((((uintptr_t)buffer_end ^ (uintptr_t)buffer_start) & ~0xffff) == 0) {
    // No problem; we can draw the buffer directly.
    _d3d_device->DrawPrimitiveUP(primitive_type, primitive_count,
         buffer_start, stride);

  } else {
    // We have a problem--the buffer crosses over a 0x10000 boundary.  We have
    // to copy the buffer to a temporary buffer that we can draw from.
    unsigned char *safe_buffer_start = get_safe_buffer_start();
    memcpy(safe_buffer_start, buffer_start, buffer_end - buffer_start);
    _d3d_device->DrawPrimitiveUP(primitive_type, primitive_count,
         safe_buffer_start, stride);

  }
}

/**
 * Issues the DrawIndexedPrimitiveUP call to draw the indicated primitive_type
 * from the given buffer.  As in draw_primitive_up(), above, the parameter
 * list is not exactly one-for-one with the DrawIndexedPrimitiveUP() call, but
 * it's similar (in particular, we pass max_index instead of NumVertices,
 * which always seemed ambiguous to me).
 */
void DXGraphicsStateGuardian9::
draw_indexed_primitive_up(D3DPRIMITIVETYPE primitive_type,
        unsigned int min_index, unsigned int max_index,
        unsigned int num_primitives,
        const unsigned char *index_data,
        D3DFORMAT index_type,
        const unsigned char *buffer, size_t stride) {
  // As above, we'll hack the case of the buffer crossing the 0x10000
  // boundary.
  const unsigned char *buffer_start = buffer + stride * min_index;
  const unsigned char *buffer_end = buffer + stride * (max_index + 1);

  if (buffer_end - buffer_start > 0x10000) {
    // Actually, the buffer doesn't fit within the required limit anyway.  Go
    // ahead and draw it and hope for the best.
    _d3d_device->DrawIndexedPrimitiveUP
      (primitive_type, min_index, max_index - min_index + 1, num_primitives,
       index_data, index_type, buffer, stride);

  } else if ((((uintptr_t)buffer_end ^ (uintptr_t)buffer_start) & ~0xffff) == 0) {
    // No problem; we can draw the buffer directly.
    _d3d_device->DrawIndexedPrimitiveUP
      (primitive_type, min_index, max_index - min_index + 1, num_primitives,
       index_data, index_type, buffer, stride);

  } else {
    // We have a problem--the buffer crosses over a 0x10000 boundary.  We have
    // to copy the buffer to a temporary buffer that we can draw from.
    unsigned char *safe_buffer_start = get_safe_buffer_start();
    memcpy(safe_buffer_start, buffer_start, buffer_end - buffer_start);
    _d3d_device->DrawIndexedPrimitiveUP
      (primitive_type, min_index, max_index - min_index + 1, num_primitives,
       index_data, index_type, safe_buffer_start - stride * min_index, stride);
  }
}

/**
 * This function is called after the creation of textures, vertex buffers, and
 * index buffers to check if DirectX is out of memory.  If DirectX is out of
 * memory and the LRU is being used, then page out some memory.  This function
 * is a fail-safe just in case another process allocates video memory, DirectX
 * is fragmented, or there are some borderline memory allocation cases, ...
 */
bool DXGraphicsStateGuardian9::
check_dx_allocation (HRESULT result, int allocation_size, int attempts)
{
  bool retry;

  retry = false;
  if (attempts <= 4)
  {
    switch (result) {
      case D3D_OK:
        break;

      case D3DERR_OUTOFVIDEOMEMORY:
      case E_OUTOFMEMORY:
        // increase the page out size as the number of attempts increases
        {
          size_t current_size = _prepared_objects->_graphics_memory_lru.get_total_size();
          size_t target_size = max(current_size - allocation_size * attempts, (size_t) 0);
          _prepared_objects->_graphics_memory_lru.evict_to(target_size);
          dxgsg9_cat.info()
            << "Evicted " << current_size - _prepared_objects->_graphics_memory_lru.get_total_size() << " bytes of texture memory to make room for more.\n";
          if (_prepared_objects->_graphics_memory_lru.get_total_size() < current_size) {
            retry = true;
          }
        }
        break;

      default:
        break;
    }
  }

  return retry;
}

// DX stencil code section
static int dx_stencil_comparison_function_array[] = {
  D3DCMP_NEVER,
  D3DCMP_LESS,
  D3DCMP_EQUAL,
  D3DCMP_LESSEQUAL,
  D3DCMP_GREATER,
  D3DCMP_NOTEQUAL,
  D3DCMP_GREATEREQUAL,
  D3DCMP_ALWAYS,
};

static int dx_stencil_operation_array[] = {
  D3DSTENCILOP_KEEP,
  D3DSTENCILOP_ZERO,
  D3DSTENCILOP_REPLACE,
  D3DSTENCILOP_INCR,
  D3DSTENCILOP_DECR,
  D3DSTENCILOP_INVERT,

  D3DSTENCILOP_INCRSAT,
  D3DSTENCILOP_DECRSAT,
};

/**
 * Set stencil render states.
 */
void DXGraphicsStateGuardian9::
do_issue_stencil() {
  if (!_supports_stencil) {
    return;
  }

  const StencilAttrib *stencil = DCAST(StencilAttrib, _target_rs->get_attrib(StencilAttrib::get_class_slot()));

  if (stencil != nullptr) {
    // DEBUG
    if (false) {
      dxgsg9_cat.debug() << "STENCIL STATE CHANGE\n";
      dxgsg9_cat.debug() << "\n"
        << "SRS_front_comparison_function " << stencil->get_render_state(StencilAttrib::SRS_front_comparison_function) << "\n"
        << "SRS_front_stencil_fail_operation " << stencil->get_render_state(StencilAttrib::SRS_front_stencil_fail_operation) << "\n"
        << "SRS_front_stencil_pass_z_fail_operation " << stencil->get_render_state(StencilAttrib::SRS_front_stencil_pass_z_fail_operation) << "\n"
        << "SRS_front_stencil_pass_z_pass_operation " << stencil->get_render_state(StencilAttrib::SRS_front_stencil_pass_z_pass_operation) << "\n"
        << "SRS_reference " << stencil->get_render_state(StencilAttrib::SRS_reference) << "\n"
        << "SRS_read_mask " << stencil->get_render_state(StencilAttrib::SRS_read_mask) << "\n"
        << "SRS_write_mask " << stencil->get_render_state(StencilAttrib::SRS_write_mask) << "\n"
        << "SRS_back_comparison_function " << stencil->get_render_state(StencilAttrib::SRS_back_comparison_function) << "\n"
        << "SRS_back_stencil_fail_operation " << stencil->get_render_state(StencilAttrib::SRS_back_stencil_fail_operation) << "\n"
        << "SRS_back_stencil_pass_z_fail_operation " << stencil->get_render_state(StencilAttrib::SRS_back_stencil_pass_z_fail_operation) << "\n"
        << "SRS_back_stencil_pass_z_pass_operation " << stencil->get_render_state(StencilAttrib::SRS_back_stencil_pass_z_pass_operation) << "\n";
    }

    bool on = false;

    unsigned int front_compare;
    front_compare = stencil->get_render_state(StencilAttrib::SRS_front_comparison_function);

    if (front_compare != RenderAttrib::M_none) {
      set_render_state(D3DRS_STENCILENABLE, TRUE);
      set_render_state(D3DRS_STENCILFUNC, dx_stencil_comparison_function_array[front_compare]);
      set_render_state(D3DRS_STENCILFAIL, dx_stencil_operation_array[
        stencil->get_render_state(StencilAttrib::SRS_front_stencil_fail_operation)]);
      set_render_state(D3DRS_STENCILZFAIL, dx_stencil_operation_array[
        stencil->get_render_state(StencilAttrib::SRS_front_stencil_pass_z_fail_operation)]);
      set_render_state(D3DRS_STENCILPASS, dx_stencil_operation_array[
        stencil->get_render_state(StencilAttrib::SRS_front_stencil_pass_z_pass_operation)]);
      on = true;
    } else {
      set_render_state(D3DRS_STENCILENABLE, FALSE);
    }

    if (_supports_two_sided_stencil) {
      unsigned int back_compare;
      back_compare = stencil->get_render_state(StencilAttrib::SRS_back_comparison_function);

      if (back_compare != RenderAttrib::M_none) {
        set_render_state(D3DRS_TWOSIDEDSTENCILMODE, TRUE);
        set_render_state(D3DRS_CCW_STENCILFUNC, dx_stencil_comparison_function_array[back_compare]);
        set_render_state(D3DRS_CCW_STENCILFAIL, dx_stencil_operation_array[
          stencil->get_render_state(StencilAttrib::SRS_back_stencil_fail_operation)]);
        set_render_state(D3DRS_CCW_STENCILZFAIL, dx_stencil_operation_array[
          stencil->get_render_state(StencilAttrib::SRS_back_stencil_pass_z_fail_operation)]);
        set_render_state(D3DRS_CCW_STENCILPASS, dx_stencil_operation_array[
          stencil->get_render_state(StencilAttrib::SRS_back_stencil_pass_z_pass_operation)]);
        on = true;
      } else {
        set_render_state(D3DRS_TWOSIDEDSTENCILMODE, FALSE);
      }
    }

    if (on) {
      set_render_state(D3DRS_STENCILREF, stencil->get_render_state(StencilAttrib::SRS_reference));
      set_render_state(D3DRS_STENCILMASK, stencil->get_render_state(StencilAttrib::SRS_read_mask));
      set_render_state(D3DRS_STENCILWRITEMASK, stencil->get_render_state(StencilAttrib::SRS_write_mask));
    }

    if (stencil->get_render_state(StencilAttrib::SRS_clear)) {
      _d3d_device->Clear(0, nullptr, D3DCLEAR_STENCIL, 0, 0.0f, stencil->get_render_state(StencilAttrib::SRS_clear_value));
    }

  } else {
    // DEBUG
    if (false) {
      dxgsg9_cat.debug() << "STENCIL STATE CHANGE TO OFF\n";
    }

    set_render_state(D3DRS_STENCILENABLE, FALSE);
    if (_supports_two_sided_stencil) {
      set_render_state(D3DRS_TWOSIDEDSTENCILMODE, FALSE);
    }
  }
}

/**
 *
 */
void DXGraphicsStateGuardian9::
do_issue_scissor() {
  const ScissorAttrib *target_scissor = DCAST(ScissorAttrib, _target_rs->get_attrib_def(ScissorAttrib::get_class_slot()));
  const LVecBase4 &frame = target_scissor->get_frame();

  RECT r;
  r.left = _current_viewport.X + _current_viewport.Width * frame[0];
  r.top = _current_viewport.Y + _current_viewport.Height * (1.0f - frame[3]);
  r.right = _current_viewport.X + _current_viewport.Width * frame[1];
  r.bottom = _current_viewport.Y + _current_viewport.Height * (1.0f - frame[2]);
  _d3d_device->SetScissorRect(&r);
  set_render_state(D3DRS_SCISSORTESTENABLE, TRUE);
}

/**
 * Convert DirectX framebuffer format ids into a FrameBufferProperties
 * structure.
 */
FrameBufferProperties DXGraphicsStateGuardian9::
calc_fb_properties(DWORD cformat, DWORD dformat,
                   DWORD multisampletype, DWORD multisamplequality) {
  FrameBufferProperties props;
  int index=0;
  int r=0, g=0, b=0, a=0;
  switch (cformat) {
  case D3DFMT_R8G8B8:      r=8; g=8; b=8; a=0; break;
  case D3DFMT_A8R8G8B8:    r=8; g=8; b=8; a=8; break;
  case D3DFMT_X8R8G8B8:    r=8; g=8; b=8; a=0; break;
  case D3DFMT_R5G6B5:      r=5; g=6; b=5; a=0; break;
  case D3DFMT_X1R5G5B5:    r=5; g=5; b=5; a=0; break;
  case D3DFMT_A1R5G5B5:    r=5; g=5; b=5; a=1; break;
  case D3DFMT_A4R4G4B4:    r=4; g=4; b=4; a=4; break;
  case D3DFMT_R3G3B2:      r=3; g=3; b=2; a=0; break;
  case D3DFMT_A8R3G3B2:    r=3; g=3; b=2; a=8; break;
  case D3DFMT_X4R4G4B4:    r=4; g=4; b=4; a=0; break;
  case D3DFMT_A2B10G10R10: r=10;g=10;b=10;a=2; break;
  case D3DFMT_A8P8:        index=8; a=8; break;
  case D3DFMT_P8:          index=8; a=0; break;
  default: break;
  }
  if (index > 0) {
    props.set_rgb_color(0);
    props.set_indexed_color(1);
    props.set_color_bits(index);
  } else if (r + g + b > 0) {
    props.set_rgb_color(1);
    props.set_indexed_color(0);
    props.set_rgba_bits(r, g, b, a);
  }
  props.set_alpha_bits(a);

  int depth=0;
  int stencil=0;
  switch (dformat) {
  case D3DFMT_D32:     depth=32; stencil=0; break;
  case D3DFMT_D15S1:   depth=15; stencil=1; break;
  case D3DFMT_D24S8:   depth=24; stencil=8; break;
  case D3DFMT_D16:     depth=16; stencil=0; break;
  case D3DFMT_D24X8:   depth=24; stencil=0; break;
  case D3DFMT_D24X4S4: depth=24; stencil=4; break;
  default: break;
  }
  props.set_depth_bits(depth);
  props.set_stencil_bits(stencil);
  if (multisampletype == D3DMULTISAMPLE_NONMASKABLE) {
    props.set_multisamples(2);
  } else {
    props.set_multisamples(multisampletype);
  }
  return props;
}


#define GAMMA_1 (255.0 * 256.0)

static bool _gamma_table_initialized = false;
static unsigned short _original_gamma_table [256 * 3];

void _create_gamma_table_dx9 (PN_stdfloat gamma, unsigned short *original_red_table, unsigned short *original_green_table, unsigned short *original_blue_table, unsigned short *red_table, unsigned short *green_table, unsigned short *blue_table) {
  int i;
  double gamma_correction;

  if (gamma <= 0.0) {
    // avoid divide by zero and negative exponents
    gamma = 1.0;
  }
  gamma_correction = 1.0 / (double) gamma;

  for (i = 0; i < 256; i++) {
    double r;
    double g;
    double b;

    if (original_red_table) {
      r = (double) original_red_table [i] / GAMMA_1;
      g = (double) original_green_table [i] / GAMMA_1;
      b = (double) original_blue_table [i] / GAMMA_1;
    }
    else {
      r = ((double) i / 255.0);
      g = r;
      b = r;
    }

    r = pow (r, gamma_correction);
    g = pow (g, gamma_correction);
    b = pow (b, gamma_correction);

    if (r > 1.00) {
      r = 1.0;
    }
    if (g > 1.00) {
      g = 1.0;
    }
    if (b > 1.00) {
      b = 1.0;
    }

    r = r * GAMMA_1;
    g = g * GAMMA_1;
    b = b * GAMMA_1;

    red_table [i] = r;
    green_table [i] = g;
    blue_table [i] = b;
  }
}

/**
 * Static function for getting the original gamma.
 */
bool DXGraphicsStateGuardian9::
get_gamma_table(void) {
  bool get;

  get = false;
  if (_gamma_table_initialized == false) {
    HDC hdc = GetDC(nullptr);

    if (hdc) {
      if (GetDeviceGammaRamp (hdc, (LPVOID) _original_gamma_table)) {
        _gamma_table_initialized = true;
        get = true;
      }

      ReleaseDC (nullptr, hdc);
    }
  }

  return get;
}

/**
 * Static function for setting gamma which is needed for atexit.
 */
bool DXGraphicsStateGuardian9::
static_set_gamma(bool restore, PN_stdfloat gamma) {
  bool set;
  HDC hdc = GetDC(nullptr);

  set = false;
  if (hdc) {
    unsigned short ramp [256 * 3];

    if (restore && _gamma_table_initialized) {
      _create_gamma_table_dx9 (gamma, &_original_gamma_table [0], &_original_gamma_table [256], &_original_gamma_table [512], &ramp [0], &ramp [256], &ramp [512]);
    }
    else {
      _create_gamma_table_dx9 (gamma, 0, 0, 0, &ramp [0], &ramp [256], &ramp [512]);
    }

    if (SetDeviceGammaRamp (hdc, ramp)) {
      set = true;
    }

    ReleaseDC (nullptr, hdc);
  }

  return set;
}

/**
 * Non static version of setting gamma.  Returns true on success.
 */
bool DXGraphicsStateGuardian9::
set_gamma(PN_stdfloat gamma) {
  bool set;

  set = static_set_gamma(false, gamma);
  if (set) {
    _gamma = gamma;
  }

  return set;
}

/**
 * Restore original gamma.
 */
void DXGraphicsStateGuardian9::
restore_gamma() {
  static_set_gamma(true, 1.0f);
}

/**
 * This function is passed to the atexit function.
 */
void DXGraphicsStateGuardian9::
atexit_function(void) {
  set_cg_device(nullptr);
  static_set_gamma(true, 1.0f);
}

/**
 * Returns true if this particular GSG supports the specified Cg Shader
 * Profile.
 */
bool DXGraphicsStateGuardian9::
get_supports_cg_profile(const std::string &name) const {
#ifndef HAVE_CG
  return false;
#else
  CGprofile profile = cgGetProfile(name.c_str());

  if (profile == CG_PROFILE_UNKNOWN) {
    dxgsg9_cat.error() << name <<", unknown Cg-profile\n";
    return false;
  }
  return cgD3D9IsProfileSupported(cgGetProfile(name.c_str())) != 0;
#endif  // HAVE_CG
}

/**
 * Sets the global Cg device pointer.  TODO: make this thread-safe somehow.
 * Maybe Cg is inherently not thread-safe.
 */
void DXGraphicsStateGuardian9::
set_cg_device(LPDIRECT3DDEVICE9 cg_device) {
#ifdef HAVE_CG
  if (_cg_device != cg_device) {
    cgD3D9SetDevice(cg_device);
    _cg_device = cg_device;
  }
#endif // HAVE_CG
}

/**
 * Returns a vertex buffer containing only a full-white color.
 */
LPDIRECT3DVERTEXBUFFER9 DXGraphicsStateGuardian9::
get_white_vbuffer() {
  if (_white_vbuffer != nullptr) {
    return _white_vbuffer;
  }

  LPDIRECT3DVERTEXBUFFER9 vbuffer;
  HRESULT hr;
  hr = _screen->_d3d_device->CreateVertexBuffer(sizeof(D3DCOLOR), D3DUSAGE_WRITEONLY, D3DFVF_DIFFUSE, D3DPOOL_DEFAULT, &vbuffer, nullptr);

  if (FAILED(hr)) {
    dxgsg9_cat.error()
      << "CreateVertexBuffer failed" << D3DERRORSTRING(hr);
    return nullptr;
  }

  D3DCOLOR *local_pointer;
  hr = vbuffer->Lock(0, sizeof(D3DCOLOR), (void **) &local_pointer, D3DLOCK_DISCARD);
  if (FAILED(hr)) {
    dxgsg9_cat.error()
      << "VertexBuffer::Lock failed" << D3DERRORSTRING(hr);
    return false;
  }

  *local_pointer = D3DCOLOR_ARGB(255, 255, 255, 255);

  vbuffer->Unlock();
  _white_vbuffer = vbuffer;
  return vbuffer;
}

typedef std::string KEY;

typedef struct _KEY_ELEMENT
{
  KEY key;
  int count;
  int secondary_count;

  struct _KEY_ELEMENT *next;
}
KEY_ELEMENT;

typedef struct _KEY_LIST
{
  int total_key_elements;
  KEY_ELEMENT *key_element;
}
KEY_LIST;

KEY_ELEMENT *new_key_element (KEY key, KEY_LIST *key_list)
{
  KEY_ELEMENT *key_element;

  key_element = new KEY_ELEMENT;
  key_element -> key = key;
  key_element -> count = 1;
  key_element -> secondary_count = 0;
  key_element -> next = 0;

  key_list -> total_key_elements++;

  return key_element;
}

KEY_ELEMENT *first_key_element (KEY_LIST *key_list)
{
  return key_list -> key_element;
}

KEY_ELEMENT *next_key_element (KEY_ELEMENT *key_element)
{
  return key_element -> next;
}

void delete_key_list (KEY_LIST *key_list)
{
  if (key_list)
  {
    KEY_ELEMENT *key_element;
    KEY_ELEMENT *key_element_next;

    key_element = first_key_element (key_list);
    while (key_element)
    {
      key_element_next = next_key_element (key_element);
      delete key_element;
      key_element = key_element_next;
    }

    delete key_list;
  }
}

KEY_LIST *new_key_list (void)
{
  KEY_LIST *key_list;

  key_list = new KEY_LIST;
  memset (key_list, 0, sizeof (KEY_LIST));

  return key_list;
}

KEY_ELEMENT *add_to_key_list (KEY key, KEY_LIST *key_list)
{
  KEY_ELEMENT *key_element;
  KEY_ELEMENT *last_key_element;
  KEY_ELEMENT *current_key_element;

  key_element = 0;
  last_key_element = 0;
  current_key_element = key_list -> key_element;
  if (current_key_element == 0)
  {
    key_element = new_key_element (key, key_list);
    key_list -> key_element = key_element;
  }
  else
  {
    while (current_key_element)
    {
      if (key < current_key_element -> key)
      {
        key_element = new_key_element (key, key_list);
        key_element -> next = current_key_element;

        if (last_key_element == 0)
        {
          key_list -> key_element = key_element;
        }
        else
        {
          last_key_element -> next = key_element;
        }
        break;
      }
      else
      {
        if (key > current_key_element -> key)
        {
          if (current_key_element -> next == 0)
          {
            key_element = new_key_element (key, key_list);
            current_key_element -> next = key_element;
            break;
          }
          else
          {

          }
        }
        else
        {
          current_key_element -> count++;
          break;
        }
      }

      last_key_element = current_key_element;
      current_key_element = current_key_element -> next;
    }
  }

  return key_element;
}