DRI2: Drop sarea, implement swap buffers in the X server.
[mesa.git] / src / mesa / drivers / dri / intel / intel_context.c
index 1c340f79400e5c4fc7f40a45e8b4474df9fdccdc..f0ee96f18dac7c2178ee245d760175a0632b44c4 100644 (file)
@@ -51,8 +51,6 @@
 #include "intel_chipset.h"
 #include "intel_buffers.h"
 #include "intel_tex.h"
-#include "intel_span.h"
-#include "intel_tris.h"
 #include "intel_ioctl.h"
 #include "intel_batchbuffer.h"
 #include "intel_blit.h"
@@ -61,7 +59,7 @@
 #include "intel_buffer_objects.h"
 #include "intel_fbo.h"
 #include "intel_decode.h"
-#include "intel_bufmgr_ttm.h"
+#include "intel_bufmgr.h"
 
 #include "drirenderbuffer.h"
 #include "vblank.h"
 int INTEL_DEBUG = (0);
 #endif
 
+#define need_GL_NV_point_sprite
 #define need_GL_ARB_multisample
 #define need_GL_ARB_point_parameters
 #define need_GL_ARB_texture_compression
 #define need_GL_ARB_vertex_buffer_object
 #define need_GL_ARB_vertex_program
 #define need_GL_ARB_window_pos
+#define need_GL_ARB_occlusion_query
 #define need_GL_EXT_blend_color
 #define need_GL_EXT_blend_equation_separate
 #define need_GL_EXT_blend_func_separate
@@ -87,18 +87,22 @@ int INTEL_DEBUG = (0);
 #define need_GL_EXT_multi_draw_arrays
 #define need_GL_EXT_secondary_color
 #define need_GL_NV_vertex_program
-#include "extension_helper.h"
-
+#define need_GL_ATI_separate_stencil
+#define need_GL_EXT_point_parameters
+#define need_GL_VERSION_2_0
+#define need_GL_VERSION_2_1
+#define need_GL_ARB_shader_objects
+#define need_GL_ARB_vertex_shader
 
-#define DRIVER_DATE                     "20061102"
-
-_glthread_Mutex lockMutex;
-static GLboolean lockMutexInit = GL_FALSE;
+#include "extension_helper.h"
 
+#define DRIVER_DATE                     "20080716"
+#define DRIVER_DATE_GEM                 "GEM " DRIVER_DATE
 
 static const GLubyte *
 intelGetString(GLcontext * ctx, GLenum name)
 {
+   const struct intel_context *const intel = intel_context(ctx);
    const char *chipset;
    static char buffer[128];
 
@@ -108,7 +112,7 @@ intelGetString(GLcontext * ctx, GLenum name)
       break;
 
    case GL_RENDERER:
-      switch (intel_context(ctx)->intelScreen->deviceID) {
+      switch (intel->intelScreen->deviceID) {
       case PCI_CHIP_845_G:
          chipset = "Intel(R) 845G";
          break;
@@ -148,12 +152,42 @@ intelGetString(GLcontext * ctx, GLenum name)
       case PCI_CHIP_Q33_G:
         chipset = "Intel(R) Q33";
         break;
+      case PCI_CHIP_I965_Q:
+        chipset = "Intel(R) 965Q";
+        break;
+      case PCI_CHIP_I965_G:
+      case PCI_CHIP_I965_G_1:
+        chipset = "Intel(R) 965G";
+        break;
+      case PCI_CHIP_I946_GZ:
+        chipset = "Intel(R) 946GZ";
+        break;
+      case PCI_CHIP_I965_GM:
+        chipset = "Intel(R) 965GM";
+        break;
+      case PCI_CHIP_I965_GME:
+        chipset = "Intel(R) 965GME/GLE";
+        break;
+      case PCI_CHIP_GM45_GM:
+        chipset = "Mobile IntelĀ® GM45 Express Chipset";
+        break; 
+      case PCI_CHIP_IGD_E_G:
+        chipset = "Intel(R) Integrated Graphics Device";
+        break;
+      case PCI_CHIP_G45_G:
+         chipset = "Intel(R) G45/G43";
+         break;
+      case PCI_CHIP_Q45_G:
+         chipset = "Intel(R) Q45/Q43";
+         break;
       default:
          chipset = "Unknown Intel Chipset";
          break;
       }
 
-      (void) driGetRendererString(buffer, chipset, DRIVER_DATE, 0);
+      (void) driGetRendererString(buffer, chipset, 
+                                 (intel->ttm) ? DRIVER_DATE_GEM : DRIVER_DATE,
+                                 0);
       return (GLubyte *) buffer;
 
    default:
@@ -161,22 +195,166 @@ intelGetString(GLcontext * ctx, GLenum name)
    }
 }
 
+void
+intel_update_renderbuffers(__DRIcontext *context, __DRIdrawable *drawable)
+{
+   struct intel_framebuffer *intel_fb = drawable->driverPrivate;
+   struct intel_renderbuffer *rb;
+   struct intel_region *region, *depth_region;
+   struct intel_context *intel = context->driverPrivate;
+   __DRIbuffer *buffers;
+   __DRIscreen *screen;
+   int i, count;
+   unsigned int attachments[10];
+   uint32_t name;
+   const char *region_name;
+
+   if (INTEL_DEBUG & DEBUG_DRI)
+      fprintf(stderr, "enter %s, drawable %p\n", __func__, drawable);
+
+   screen = intel->intelScreen->driScrnPriv;
+
+   i = 0;
+   if (intel_fb->color_rb[0])
+      attachments[i++] = __DRI_BUFFER_FRONT_LEFT;
+   if (intel_fb->color_rb[1])
+      attachments[i++] = __DRI_BUFFER_BACK_LEFT;
+   if (intel_get_renderbuffer(&intel_fb->Base, BUFFER_DEPTH))
+      attachments[i++] = __DRI_BUFFER_DEPTH;
+   if (intel_get_renderbuffer(&intel_fb->Base, BUFFER_STENCIL))
+      attachments[i++] = __DRI_BUFFER_STENCIL;
+
+   buffers = (*screen->dri2.loader->getBuffers)(drawable,
+                                               &drawable->w,
+                                               &drawable->h,
+                                               attachments, i,
+                                               &count,
+                                               drawable->loaderPrivate);
+
+   drawable->x = 0;
+   drawable->y = 0;
+   drawable->backX = 0;
+   drawable->backY = 0;
+   drawable->numClipRects = 1;
+   drawable->pClipRects[0].x1 = 0;
+   drawable->pClipRects[0].y1 = 0;
+   drawable->pClipRects[0].x2 = drawable->w;
+   drawable->pClipRects[0].y2 = drawable->h;
+   drawable->numBackClipRects = 1;
+   drawable->pBackClipRects[0].x1 = 0;
+   drawable->pBackClipRects[0].y1 = 0;
+   drawable->pBackClipRects[0].x2 = drawable->w;
+   drawable->pBackClipRects[0].y2 = drawable->h;
+
+   depth_region = NULL;
+   for (i = 0; i < count; i++) {
+       switch (buffers[i].attachment) {
+       case __DRI_BUFFER_FRONT_LEFT:
+          rb = intel_fb->color_rb[0];
+          region_name = "dri2 front buffer";
+          break;
+
+       case __DRI_BUFFER_BACK_LEFT:
+          rb = intel_fb->color_rb[1];
+          region_name = "dri2 back buffer";
+          break;
+
+       case __DRI_BUFFER_DEPTH:
+          rb = intel_get_renderbuffer(&intel_fb->Base, BUFFER_DEPTH);
+          region_name = "dri2 depth buffer";
+          break;
+
+       case __DRI_BUFFER_STENCIL:
+          rb = intel_get_renderbuffer(&intel_fb->Base, BUFFER_STENCIL);
+          region_name = "dri2 stencil buffer";
+          break;
+
+       case __DRI_BUFFER_ACCUM:
+       default:
+          fprintf(stderr,
+                  "unhandled buffer attach event, attacment type %d\n",
+                  buffers[i].attachment);
+          return;
+       }
+
+       if (rb->region) {
+         intel_bo_flink(rb->region->buffer, &name);
+         if (name == buffers[i].name)
+            continue;
+       }
+
+       if (INTEL_DEBUG & DEBUG_DRI)
+         fprintf(stderr,
+                 "attaching buffer %d, at %d, cpp %d, pitch %d\n",
+                 buffers[i].name, buffers[i].attachment,
+                 buffers[i].cpp, buffers[i].pitch);
+       
+       if (buffers[i].attachment == __DRI_BUFFER_STENCIL && depth_region) {
+         if (INTEL_DEBUG & DEBUG_DRI)
+            fprintf(stderr, "(reusing depth buffer as stencil)\n");
+          region = depth_region;
+       }
+       else
+          region = intel_region_alloc_for_handle(intel, buffers[i].cpp,
+                                                buffers[i].pitch / buffers[i].cpp,
+                                                drawable->h,
+                                                buffers[i].name,
+                                                region_name);
+
+       if (buffers[i].attachment == __DRI_BUFFER_DEPTH)
+         depth_region = region;
+
+       intel_renderbuffer_set_region(rb, region);
+       intel_region_release(&region);
+   }
+
+   driUpdateFramebufferSize(&intel->ctx, drawable);
+}
+
+static void
+intel_viewport(GLcontext *ctx, GLint x, GLint y, GLsizei w, GLsizei h)
+{
+    struct intel_context *intel = intel_context(ctx);
+    __DRIcontext *driContext = intel->driContext;
+
+    if (!driContext->driScreenPriv->dri2.enabled)
+       return;
+
+    intel_update_renderbuffers(driContext, driContext->driDrawablePriv);
+    if (driContext->driDrawablePriv != driContext->driReadablePriv)
+       intel_update_renderbuffers(driContext, driContext->driReadablePriv);
+
+    ctx->Driver.Viewport = NULL;
+    intel->driDrawable = driContext->driDrawablePriv;
+    intelWindowMoved(intel);
+    intel_draw_buffer(ctx, intel->ctx.DrawBuffer);
+    ctx->Driver.Viewport = intel_viewport;
+}
 
 /**
  * Extension strings exported by the intel driver.
+ *
+ * Extensions supported by all chips supported by i830_dri, i915_dri, or
+ * i965_dri.
  */
-const struct dri_extension card_extensions[] = {
+static const struct dri_extension card_extensions[] = {
    {"GL_ARB_multisample", GL_ARB_multisample_functions},
    {"GL_ARB_multitexture", NULL},
    {"GL_ARB_point_parameters", GL_ARB_point_parameters_functions},
+   {"GL_NV_point_sprite", GL_NV_point_sprite_functions},
    {"GL_ARB_texture_border_clamp", NULL},
    {"GL_ARB_texture_compression", GL_ARB_texture_compression_functions},
    {"GL_ARB_texture_cube_map", NULL},
    {"GL_ARB_texture_env_add", NULL},
    {"GL_ARB_texture_env_combine", NULL},
+   {"GL_ARB_texture_env_crossbar", NULL},
    {"GL_ARB_texture_env_dot3", NULL},
    {"GL_ARB_texture_mirrored_repeat", NULL},
+   {"GL_ARB_texture_non_power_of_two",   NULL },
    {"GL_ARB_texture_rectangle", NULL},
+   {"GL_NV_texture_rectangle", NULL},
+   {"GL_EXT_texture_rectangle", NULL},
+   {"GL_ARB_point_parameters", NULL}, 
    {"GL_ARB_vertex_buffer_object", GL_ARB_vertex_buffer_object_functions},
    {"GL_ARB_vertex_program", GL_ARB_vertex_program_functions},
    {"GL_ARB_window_pos", GL_ARB_window_pos_functions},
@@ -185,10 +363,12 @@ const struct dri_extension card_extensions[] = {
     GL_EXT_blend_equation_separate_functions},
    {"GL_EXT_blend_func_separate", GL_EXT_blend_func_separate_functions},
    {"GL_EXT_blend_minmax", GL_EXT_blend_minmax_functions},
+   {"GL_EXT_blend_logic_op", NULL},
    {"GL_EXT_blend_subtract", NULL},
    {"GL_EXT_cull_vertex", GL_EXT_cull_vertex_functions},
    {"GL_EXT_fog_coord", GL_EXT_fog_coord_functions},
    {"GL_EXT_multi_draw_arrays", GL_EXT_multi_draw_arrays_functions},
+   {"GL_ATI_separate_stencil", GL_ATI_separate_stencil_functions},
 #if 1                           /* XXX FBO temporary? */
    {"GL_EXT_packed_depth_stencil", NULL},
 #endif
@@ -210,60 +390,109 @@ const struct dri_extension card_extensions[] = {
    {NULL, NULL}
 };
 
-const struct dri_extension ttm_extensions[] = {
+static const struct dri_extension brw_extensions[] = {
+   { "GL_ARB_shading_language_100",       GL_VERSION_2_0_functions},
+   { "GL_ARB_shading_language_120",       GL_VERSION_2_1_functions},
+   { "GL_ARB_shader_objects",             GL_ARB_shader_objects_functions},
+   { "GL_ARB_vertex_shader",              GL_ARB_vertex_shader_functions},
+   { "GL_ARB_point_sprite",              NULL},
+   { "GL_ARB_fragment_shader",            NULL },
+   { "GL_ARB_draw_buffers",               NULL },
+   { "GL_ARB_depth_texture",              NULL },
+   { "GL_ARB_fragment_program",           NULL },
+   { "GL_ARB_shadow",                     NULL },
+   { "GL_EXT_shadow_funcs",               NULL },
+   { "GL_ARB_fragment_program_shadow",    NULL },
+   /* ARB extn won't work if not enabled */
+   { "GL_SGIX_depth_texture",             NULL },
+   { "GL_EXT_texture_sRGB",              NULL},
+   { NULL,                                NULL }
+};
+
+static const struct dri_extension arb_oc_extensions[] = {
+   {"GL_ARB_occlusion_query",            GL_ARB_occlusion_query_functions},
+   {NULL, NULL}
+};
+
+static const struct dri_extension ttm_extensions[] = {
    {"GL_EXT_framebuffer_object", GL_EXT_framebuffer_object_functions},
    {"GL_ARB_pixel_buffer_object", NULL},
    {NULL, NULL}
 };
 
-extern const struct tnl_pipeline_stage _intel_render_stage;
-
-static const struct tnl_pipeline_stage *intel_pipeline[] = {
-   &_tnl_vertex_transform_stage,
-   &_tnl_vertex_cull_stage,
-   &_tnl_normal_transform_stage,
-   &_tnl_lighting_stage,
-   &_tnl_fog_coordinate_stage,
-   &_tnl_texgen_stage,
-   &_tnl_texture_transform_stage,
-   &_tnl_point_attenuation_stage,
-   &_tnl_vertex_program_stage,
-#if 1
-   &_intel_render_stage,        /* ADD: unclipped rastersetup-to-dma */
-#endif
-   &_tnl_render_stage,
-   0,
-};
+/**
+ * Initializes potential list of extensions if ctx == NULL, or actually enables
+ * extensions for a context.
+ */
+void intelInitExtensions(GLcontext *ctx, GLboolean enable_imaging)
+{
+   struct intel_context *intel = ctx?intel_context(ctx):NULL;
 
+   /* Disable imaging extension until convolution is working in teximage paths.
+    */
+   enable_imaging = GL_FALSE;
+
+   driInitExtensions(ctx, card_extensions, enable_imaging);
+
+   if (intel == NULL || intel->ttm)
+      driInitExtensions(ctx, ttm_extensions, GL_FALSE);
+
+   if (intel == NULL || 
+       (IS_965(intel->intelScreen->deviceID) && 
+       intel->intelScreen->drmMinor >= 8))
+      driInitExtensions(ctx, arb_oc_extensions, GL_FALSE);
+
+   if (intel == NULL || IS_965(intel->intelScreen->deviceID))
+      driInitExtensions(ctx, brw_extensions, GL_FALSE);
+}
 
 static const struct dri_debug_control debug_control[] = {
-   {"tex", DEBUG_TEXTURE},
-   {"state", DEBUG_STATE},
-   {"ioctl", DEBUG_IOCTL},
-   {"blit", DEBUG_BLIT},
-   {"mip", DEBUG_MIPTREE},
-   {"fall", DEBUG_FALLBACKS},
-   {"verb", DEBUG_VERBOSE},
-   {"bat", DEBUG_BATCH},
-   {"pix", DEBUG_PIXEL},
-   {"buf", DEBUG_BUFMGR},
-   {"reg", DEBUG_REGION},
-   {"fbo", DEBUG_FBO},
-   {"lock", DEBUG_LOCK},
-   {"sync", DEBUG_SYNC},
-   {NULL, 0}
+   { "tex",   DEBUG_TEXTURE},
+   { "state", DEBUG_STATE},
+   { "ioctl", DEBUG_IOCTL},
+   { "blit",  DEBUG_BLIT},
+   { "mip",   DEBUG_MIPTREE},
+   { "fall",  DEBUG_FALLBACKS},
+   { "verb",  DEBUG_VERBOSE},
+   { "bat",   DEBUG_BATCH},
+   { "pix",   DEBUG_PIXEL},
+   { "buf",   DEBUG_BUFMGR},
+   { "reg",   DEBUG_REGION},
+   { "fbo",   DEBUG_FBO},
+   { "lock",  DEBUG_LOCK},
+   { "sync",  DEBUG_SYNC},
+   { "prim",  DEBUG_PRIMS },
+   { "vert",  DEBUG_VERTS },
+   { "dri",   DEBUG_DRI },
+   { "dma",   DEBUG_DMA },
+   { "san",   DEBUG_SANITY },
+   { "sleep", DEBUG_SLEEP },
+   { "stats", DEBUG_STATS },
+   { "tile",  DEBUG_TILE },
+   { "sing",  DEBUG_SINGLE_THREAD },
+   { "thre",  DEBUG_SINGLE_THREAD },
+   { "wm",    DEBUG_WM },
+   { "urb",   DEBUG_URB },
+   { "vs",    DEBUG_VS },
+   { NULL,    0 }
 };
 
 
 static void
 intelInvalidateState(GLcontext * ctx, GLuint new_state)
 {
+    struct intel_context *intel = intel_context(ctx);
+
    _swrast_InvalidateState(ctx, new_state);
    _swsetup_InvalidateState(ctx, new_state);
    _vbo_InvalidateState(ctx, new_state);
    _tnl_InvalidateState(ctx, new_state);
    _tnl_invalidate_vertex_state(ctx, new_state);
-   intel_context(ctx)->NewGLState |= new_state;
+
+   intel->NewGLState |= new_state;
+
+   if (intel->vtbl.invalidate_state)
+      intel->vtbl.invalidate_state( intel, new_state );
 }
 
 
@@ -275,27 +504,71 @@ intelFlush(GLcontext * ctx)
    if (intel->Fallback)
       _swrast_flush(ctx);
 
-   INTEL_FIREVERTICES(intel);
+   if (!IS_965(intel->intelScreen->deviceID))
+      INTEL_FIREVERTICES(intel);
+
+   /* Emit a flush so that any frontbuffer rendering that might have occurred
+    * lands onscreen in a timely manner, even if the X Server doesn't trigger
+    * a flush for us.
+    */
+   intel_batchbuffer_emit_mi_flush(intel->batch);
 
    if (intel->batch->map != intel->batch->ptr)
       intel_batchbuffer_flush(intel->batch);
-
-   /* XXX: Need to do an MI_FLUSH here.
-    */
 }
 
 void
 intelFinish(GLcontext * ctx)
 {
-   struct intel_context *intel = intel_context(ctx);
+   struct gl_framebuffer *fb = ctx->DrawBuffer;
+   int i;
+
    intelFlush(ctx);
-   if (intel->batch->last_fence) {
-      dri_fence_wait(intel->batch->last_fence);
-      dri_fence_unreference(intel->batch->last_fence);
-      intel->batch->last_fence = NULL;
+
+   for (i = 0; i < fb->_NumColorDrawBuffers; i++) {
+       struct intel_renderbuffer *irb;
+
+       irb = intel_renderbuffer(fb->_ColorDrawBuffers[i]);
+
+       if (irb->region)
+         dri_bo_wait_rendering(irb->region->buffer);
+   }
+   if (fb->_DepthBuffer) {
+      /* XXX: Wait on buffer idle */
    }
 }
 
+static void
+intelBeginQuery(GLcontext *ctx, GLenum target, struct gl_query_object *q)
+{
+       struct intel_context *intel = intel_context( ctx );
+       struct drm_i915_mmio io = {
+               .read_write = I915_MMIO_READ,
+               .reg = MMIO_REGS_PS_DEPTH_COUNT,
+               .data = &q->Result 
+       };
+       intel->stats_wm++;
+       intelFinish(&intel->ctx);
+       drmCommandWrite(intel->driFd, DRM_I915_MMIO, &io, sizeof(io));
+}
+
+static void
+intelEndQuery(GLcontext *ctx, GLenum target, struct gl_query_object *q)
+{
+       struct intel_context *intel = intel_context( ctx );
+       GLuint64EXT tmp;        
+       struct drm_i915_mmio io = {
+               .read_write = I915_MMIO_READ,
+               .reg = MMIO_REGS_PS_DEPTH_COUNT,
+               .data = &tmp
+       };
+       intelFinish(&intel->ctx);
+       drmCommandWrite(intel->driFd, DRM_I915_MMIO, &io, sizeof(io));
+       q->Result = tmp - q->Result;
+       q->Ready = GL_TRUE;
+       intel->stats_wm--;
+}
+
 /** Driver-specific fence emit implementation for the fake memory manager. */
 static unsigned int
 intel_fence_emit(void *private)
@@ -327,31 +600,51 @@ static GLboolean
 intel_init_bufmgr(struct intel_context *intel)
 {
    intelScreenPrivate *intelScreen = intel->intelScreen;
-   GLboolean ttm_disable = getenv("INTEL_NO_TTM") != NULL;
+   GLboolean gem_disable = getenv("INTEL_NO_GEM") != NULL;
+   int gem_kernel = 0;
+   GLboolean gem_supported;
+   struct drm_i915_getparam gp;
+
+   gp.param = I915_PARAM_HAS_GEM;
+   gp.value = &gem_kernel;
 
-   /* If we've got a new enough DDX that's initializing TTM and giving us
+   (void) drmCommandWriteRead(intel->driFd, DRM_I915_GETPARAM, &gp, sizeof(gp));
+
+   /* If we've got a new enough DDX that's initializing GEM and giving us
     * object handles for the shared buffers, use that.
     */
    intel->ttm = GL_FALSE;
-   if (!ttm_disable &&
-       intel->intelScreen->driScrnPriv->ddx_version.minor >= 9 &&
-       intel->intelScreen->drmMinor >= 11 &&
-       intel->intelScreen->front.bo_handle != -1)
-   {
-      intel->bufmgr = intel_bufmgr_ttm_init(intel->driFd,
-                                           DRM_FENCE_TYPE_EXE,
-                                           DRM_FENCE_TYPE_EXE |
-                                           DRM_I915_FENCE_TYPE_RW,
+   if (intel->intelScreen->driScrnPriv->dri2.enabled)
+       gem_supported = GL_TRUE;
+   else if (intel->intelScreen->driScrnPriv->ddx_version.minor >= 9 &&
+           gem_kernel &&
+           intel->intelScreen->front.bo_handle != -1)
+       gem_supported = GL_TRUE;
+   else
+       gem_supported = GL_FALSE;
+
+   if (!gem_disable && gem_supported) {
+      int bo_reuse_mode;
+      intel->bufmgr = intel_bufmgr_gem_init(intel->driFd,
                                            BATCH_SZ);
       if (intel->bufmgr != NULL)
         intel->ttm = GL_TRUE;
+
+      bo_reuse_mode = driQueryOptioni(&intel->optionCache, "bo_reuse");
+      switch (bo_reuse_mode) {
+      case DRI_CONF_BO_REUSE_DISABLED:
+        break;
+      case DRI_CONF_BO_REUSE_ALL:
+        intel_bufmgr_gem_enable_reuse(intel->bufmgr);
+        break;
+      }
    }
    /* Otherwise, use the classic buffer manager. */
    if (intel->bufmgr == NULL) {
-      if (ttm_disable) {
-        fprintf(stderr, "TTM buffer manager disabled.  Using classic.\n");
+      if (gem_disable) {
+        fprintf(stderr, "GEM disabled.  Using classic.\n");
       } else {
-        fprintf(stderr, "Failed to initialize TTM buffer manager.  "
+        fprintf(stderr, "Failed to initialize GEM.  "
                 "Falling back to classic.\n");
       }
 
@@ -361,14 +654,17 @@ intel_init_bufmgr(struct intel_context *intel)
         return GL_FALSE;
       }
 
-      intel->bufmgr = dri_bufmgr_fake_init(intelScreen->tex.offset,
-                                          intelScreen->tex.map,
-                                          intelScreen->tex.size,
-                                          intel_fence_emit,
-                                          intel_fence_wait,
-                                          intel);
+      intel->bufmgr = intel_bufmgr_fake_init(intelScreen->tex.offset,
+                                            intelScreen->tex.map,
+                                            intelScreen->tex.size,
+                                            intel_fence_emit,
+                                            intel_fence_wait,
+                                            intel);
    }
 
+   /* XXX bufmgr should be per-screen, not per-context */
+   intelScreen->ttm = intel->ttm;
+
    return GL_TRUE;
 }
 
@@ -381,15 +677,20 @@ intelInitDriverFunctions(struct dd_function_table *functions)
    functions->Finish = intelFinish;
    functions->GetString = intelGetString;
    functions->UpdateState = intelInvalidateState;
+   functions->Viewport = intel_viewport;
+
    functions->CopyColorTable = _swrast_CopyColorTable;
    functions->CopyColorSubTable = _swrast_CopyColorSubTable;
    functions->CopyConvolutionFilter1D = _swrast_CopyConvolutionFilter1D;
    functions->CopyConvolutionFilter2D = _swrast_CopyConvolutionFilter2D;
 
+   functions->BeginQuery = intelBeginQuery;
+   functions->EndQuery = intelEndQuery;
+
    intelInitTextureFuncs(functions);
-   intelInitPixelFuncs(functions);
    intelInitStateFuncs(functions);
    intelInitBufferFuncs(functions);
+   intelInitPixelFuncs(functions);
 }
 
 
@@ -404,19 +705,21 @@ intelInitContext(struct intel_context *intel,
    GLcontext *shareCtx = (GLcontext *) sharedContextPrivate;
    __DRIscreenPrivate *sPriv = driContextPriv->driScreenPriv;
    intelScreenPrivate *intelScreen = (intelScreenPrivate *) sPriv->private;
-   struct drm_i915_sarea *saPriv = (struct drm_i915_sarea *)
+   volatile struct drm_i915_sarea *saPriv = (struct drm_i915_sarea *)
       (((GLubyte *) sPriv->pSAREA) + intelScreen->sarea_priv_offset);
    int fthrottle_mode;
 
-   if (!_mesa_initialize_context(&intel->ctx,
-                                 mesaVis, shareCtx,
-                                 functions, (void *) intel))
+   if (!_mesa_initialize_context(&intel->ctx, mesaVis, shareCtx,
+                                 functions, (void *) intel)) {
+      _mesa_printf("%s: failed to init mesa context\n", __FUNCTION__);
       return GL_FALSE;
+   }
 
    driContextPriv->driverPrivate = intel;
    intel->intelScreen = intelScreen;
    intel->driScreen = sPriv;
    intel->sarea = saPriv;
+   intel->driContext = driContextPriv;
 
    /* Dri stuff */
    intel->hHWContext = driContextPriv->hHWContext;
@@ -426,6 +729,9 @@ intelInitContext(struct intel_context *intel,
    intel->width = intelScreen->width;
    intel->height = intelScreen->height;
 
+   driParseConfigFiles(&intel->optionCache, &intelScreen->optionCache,
+                       intel->driScreen->myNum,
+                      IS_965(intelScreen->deviceID) ? "i965" : "i915");
    if (intelScreen->deviceID == PCI_CHIP_I865_G)
       intel->maxBatchSize = 4096;
    else
@@ -434,14 +740,6 @@ intelInitContext(struct intel_context *intel,
    if (!intel_init_bufmgr(intel))
       return GL_FALSE;
 
-   if (!lockMutexInit) {
-      lockMutexInit = GL_TRUE;
-      _glthread_INIT_MUTEX(lockMutex);
-   }
-
-   driParseConfigFiles(&intel->optionCache, &intelScreen->optionCache,
-                       intel->driScreen->myNum, "i915");
-
    ctx->Const.MaxTextureMaxAnisotropy = 2.0;
 
    /* This doesn't yet catch all non-conformant rendering, but it's a
@@ -451,11 +749,20 @@ intelInitContext(struct intel_context *intel,
       intel->strict_conformance = 1;
    }
 
-   ctx->Const.MinLineWidth = 1.0;
-   ctx->Const.MinLineWidthAA = 1.0;
-   ctx->Const.MaxLineWidth = 3.0;
-   ctx->Const.MaxLineWidthAA = 3.0;
-   ctx->Const.LineWidthGranularity = 1.0;
+   if (intel->strict_conformance) {
+      ctx->Const.MinLineWidth = 1.0;
+      ctx->Const.MinLineWidthAA = 1.0;
+      ctx->Const.MaxLineWidth = 1.0;
+      ctx->Const.MaxLineWidthAA = 1.0;
+      ctx->Const.LineWidthGranularity = 1.0;
+   }
+   else {
+      ctx->Const.MinLineWidth = 1.0;
+      ctx->Const.MinLineWidthAA = 1.0;
+      ctx->Const.MaxLineWidth = 5.0;
+      ctx->Const.MaxLineWidthAA = 5.0;
+      ctx->Const.LineWidthGranularity = 0.5;
+   }
 
    ctx->Const.MinPointSize = 1.0;
    ctx->Const.MinPointSizeAA = 1.0;
@@ -475,15 +782,12 @@ intelInitContext(struct intel_context *intel,
    _vbo_CreateContext(ctx);
    _tnl_CreateContext(ctx);
    _swsetup_CreateContext(ctx);
-
-   /* Install the customized pipeline: */
-   _tnl_destroy_pipeline(ctx);
-   _tnl_install_pipeline(ctx, intel_pipeline);
-
    /* Configure swrast to match hardware characteristics: */
    _swrast_allow_pixel_fog(ctx, GL_FALSE);
    _swrast_allow_vertex_fog(ctx, GL_TRUE);
 
+   intel->hw_stencil = mesaVis->stencilBits && mesaVis->depthBits == 24;
    intel->hw_stipple = 1;
 
    /* XXX FBO: this doesn't seem to be used anywhere */
@@ -500,10 +804,8 @@ intelInitContext(struct intel_context *intel,
       break;
    }
 
-   /* Initialize swrast, tnl driver tables: */
-   intelInitSpanFuncs(ctx);
-   intelInitTriFuncs(ctx);
-
+   if (IS_965(intelScreen->deviceID))
+      intel->polygon_offset_scale /= 0xffff;
 
    intel->RenderIndex = ~0;
 
@@ -517,22 +819,21 @@ intelInitContext(struct intel_context *intel,
 
    _math_matrix_ctr(&intel->ViewportMatrix);
 
-   /* Disable imaging extension until convolution is working in
-    * teximage paths:
-    */
-   driInitExtensions(ctx, card_extensions,
-/*                   GL_TRUE, */
-                     GL_FALSE);
+   if (IS_965(intelScreen->deviceID) && !intel->intelScreen->irq_active) {
+      _mesa_printf("IRQs not active.  Exiting\n");
+      exit(1);
+   }
 
-   if (intel->ttm)
-      driInitExtensions(ctx, ttm_extensions, GL_FALSE);
+   intelInitExtensions(ctx, GL_FALSE);
+
+   INTEL_DEBUG = driParseDebugString(getenv("INTEL_DEBUG"), debug_control);
+   if (INTEL_DEBUG & DEBUG_BUFMGR)
+      dri_bufmgr_set_debug(intel->bufmgr, GL_TRUE);
 
    if (!sPriv->dri2.enabled)
       intel_recreate_static_regions(intel);
 
    intel->batch = intel_batchbuffer_alloc(intel);
-   intel->last_swap_fence = NULL;
-   intel->first_swap_fence = NULL;
 
    intel_bufferobj_init(intel);
    intel_fbo_init(intel);
@@ -547,15 +848,10 @@ intelInitContext(struct intel_context *intel,
 
    intel->prim.primitive = ~0;
 
-#if DO_DEBUG
-   INTEL_DEBUG = driParseDebugString(getenv("INTEL_DEBUG"), debug_control);
-   if (INTEL_DEBUG & DEBUG_BUFMGR)
-      dri_bufmgr_set_debug(intel->bufmgr, GL_TRUE);
-#endif
-
+   /* Force all software fallbacks */
    if (driQueryOptionb(&intel->optionCache, "no_rast")) {
       fprintf(stderr, "disabling 3D rasterization\n");
-      FALLBACK(intel, INTEL_FALLBACK_USER, 1);
+      intel->no_rast = 1;
    }
 
    /* Disable all hardware rendering (skip emitting batches and fences/waits
@@ -589,17 +885,7 @@ intelDestroyContext(__DRIcontextPrivate * driContextPriv)
       intel->Fallback = 0;      /* don't call _swrast_Flush later */
 
       intel_batchbuffer_free(intel->batch);
-
-      if (intel->last_swap_fence) {
-        dri_fence_wait(intel->last_swap_fence);
-        dri_fence_unreference(intel->last_swap_fence);
-        intel->last_swap_fence = NULL;
-      }
-      if (intel->first_swap_fence) {
-        dri_fence_wait(intel->first_swap_fence);
-        dri_fence_unreference(intel->first_swap_fence);
-        intel->first_swap_fence = NULL;
-      }
+      free(intel->prim.vb);
 
       if (release_texture_heaps) {
          /* This share group is about to go away, free our private
@@ -627,6 +913,7 @@ intelMakeCurrent(__DRIcontextPrivate * driContextPriv,
                  __DRIdrawablePrivate * driDrawPriv,
                  __DRIdrawablePrivate * driReadPriv)
 {
+   __DRIscreenPrivate *psp = driDrawPriv->driScreenPriv;
 
    if (driContextPriv) {
       struct intel_context *intel =
@@ -635,6 +922,9 @@ intelMakeCurrent(__DRIcontextPrivate * driContextPriv,
         (struct intel_framebuffer *) driDrawPriv->driverPrivate;
       GLframebuffer *readFb = (GLframebuffer *) driReadPriv->driverPrivate;
 
+    intel_update_renderbuffers(driContextPriv, driDrawPriv);
+    if (driDrawPriv != driReadPriv)
+       intel_update_renderbuffers(driContextPriv, driReadPriv);
 
       /* XXX FBO temporary fix-ups! */
       /* if the renderbuffers don't have regions, init them from the context */
@@ -652,10 +942,12 @@ intelMakeCurrent(__DRIcontextPrivate * driContextPriv,
            intel_renderbuffer_set_region(intel_fb->color_rb[1],
                                          intel->back_region);
          }
+#if 0
          if (intel_fb->color_rb[2]) {
            intel_renderbuffer_set_region(intel_fb->color_rb[2],
                                          intel->third_region);
          }
+#endif
          if (irbDepth) {
            intel_renderbuffer_set_region(irbDepth, intel->depth_region);
          }
@@ -677,6 +969,9 @@ intelMakeCurrent(__DRIcontextPrivate * driContextPriv,
        */
       if (intel->ctx.DrawBuffer == &intel_fb->Base) {
 
+        if (intel->driReadDrawable != driReadPriv)
+           intel->driReadDrawable = driReadPriv;
+
         if (intel->driDrawable != driDrawPriv) {
            if (driDrawPriv->swap_interval == (unsigned)-1) {
               int i;
@@ -685,7 +980,7 @@ intelMakeCurrent(__DRIcontextPrivate * driContextPriv,
                  ? driGetDefaultVBlankFlags(&intel->optionCache)
                 : VBLANK_FLAG_NO_IRQ;
 
-              (*dri_interface->getUST) (&intel_fb->swap_ust);
+              (*psp->systemTime->getUST) (&intel_fb->swap_ust);
               driDrawableInitVBlank(driDrawPriv);
               intel_fb->vbl_waited = driDrawPriv->vblSeq;
 
@@ -713,10 +1008,11 @@ intelContendedLock(struct intel_context *intel, GLuint flags)
 {
    __DRIdrawablePrivate *dPriv = intel->driDrawable;
    __DRIscreenPrivate *sPriv = intel->driScreen;
-   struct drm_i915_sarea *sarea = intel->sarea;
-   int drawable_changed = 0;
+   volatile struct drm_i915_sarea *sarea = intel->sarea;
+   int me = intel->hHWContext;
 
    drmGetLock(intel->driFd, intel->hHWContext, flags);
+   intel->locked = 1;
 
    if (INTEL_DEBUG & DEBUG_LOCK)
       _mesa_printf("%s - got contended lock\n", __progname);
@@ -726,11 +1022,15 @@ intelContendedLock(struct intel_context *intel, GLuint flags)
     * NOTE: This releases and regains the hw lock, so all state
     * checking must be done *after* this call:
     */
-   if (dPriv) {
-      if (sPriv->dri2.enabled)
-        drawable_changed = __driParseEvents(sPriv, dPriv);
-      else
-        DRI_VALIDATE_DRAWABLE_INFO(sPriv, dPriv);
+   if (dPriv)
+       DRI_VALIDATE_DRAWABLE_INFO(sPriv, dPriv);
+
+   if (sarea && sarea->ctxOwner != me) {
+      if (INTEL_DEBUG & DEBUG_BUFMGR) {
+        fprintf(stderr, "Lost Context: sarea->ctxOwner %x me %x\n",
+                sarea->ctxOwner, me);
+      }
+      sarea->ctxOwner = me;
    }
 
    /* If the last consumer of the texture memory wasn't us, notify the fake
@@ -740,67 +1040,70 @@ intelContendedLock(struct intel_context *intel, GLuint flags)
     */
    if (!intel->ttm && sarea->texAge != intel->hHWContext) {
       sarea->texAge = intel->hHWContext;
-      dri_bufmgr_fake_contended_lock_take(intel->bufmgr);
+      intel_bufmgr_fake_contended_lock_take(intel->bufmgr);
       if (INTEL_DEBUG & DEBUG_BATCH)
         intel_decode_context_reset();
+      if (INTEL_DEBUG & DEBUG_BUFMGR)
+        fprintf(stderr, "Lost Textures: sarea->texAge %x hw context %x\n",
+                sarea->ctxOwner, intel->hHWContext);
    }
 
-   if (!sPriv->dri2.enabled) {
-      if (sarea->width != intel->width ||
-         sarea->height != intel->height) {
-        int numClipRects = intel->numClipRects;
+   if (sarea->width != intel->width || sarea->height != intel->height) {
+       int numClipRects = intel->numClipRects;
 
-        /*
-         * FIXME: Really only need to do this when drawing to a
-         * common back- or front buffer.
-         */
+       /*
+       * FIXME: Really only need to do this when drawing to a
+       * common back- or front buffer.
+       */
 
-        /*
-         * This will essentially drop the outstanding batchbuffer on the floor.
-         */
-        intel->numClipRects = 0;
+       /*
+       * This will essentially drop the outstanding batchbuffer on
+       * the floor.
+       */
+       intel->numClipRects = 0;
 
-        if (intel->Fallback)
-           _swrast_flush(&intel->ctx);
+       if (intel->Fallback)
+          _swrast_flush(&intel->ctx);
 
-        INTEL_FIREVERTICES(intel);
+       if (!IS_965(intel->intelScreen->deviceID))
+          INTEL_FIREVERTICES(intel);
 
-        if (intel->batch->map != intel->batch->ptr)
-           intel_batchbuffer_flush(intel->batch);
+       if (intel->batch->map != intel->batch->ptr)
+          intel_batchbuffer_flush(intel->batch);
 
-        intel->numClipRects = numClipRects;
+       intel->numClipRects = numClipRects;
 
-        /* force window update */
-        intel->lastStamp = 0;
+       /* force window update */
+       intel->lastStamp = 0;
 
-        intel->width = sarea->width;
-        intel->height = sarea->height;
-      }
+       intel->width = sarea->width;
+       intel->height = sarea->height;
+   }
 
-      /* Drawable changed?
-       */
-      if (dPriv && intel->lastStamp != dPriv->lastStamp) {
-        intelWindowMoved(intel);
-        intel->lastStamp = dPriv->lastStamp;
-      }
-   } else if (drawable_changed) {
-     intelWindowMoved(intel);                                                 
-     intel_draw_buffer(&intel->ctx, intel->ctx.DrawBuffer);
+   /* Drawable changed?
+    */
+   if (dPriv && intel->lastStamp != dPriv->lastStamp) {
+       intelWindowMoved(intel);
+       intel->lastStamp = dPriv->lastStamp;
    }
 }
 
 
+_glthread_DECLARE_STATIC_MUTEX(lockMutex);
 
 /* Lock the hardware and validate our state.  
  */
 void LOCK_HARDWARE( struct intel_context *intel )
 {
-    __DRIdrawablePrivate *dPriv = intel->driDrawable;
-    char __ret=0;
+    __DRIdrawable *dPriv = intel->driDrawable;
+    __DRIscreen *sPriv = intel->driScreen;
+    char __ret = 0;
     struct intel_framebuffer *intel_fb = NULL;
     struct intel_renderbuffer *intel_rb = NULL;
+
     _glthread_LOCK_MUTEX(lockMutex);
     assert(!intel->locked);
+    intel->locked = 1;
 
     if (intel->driDrawable) {
        intel_fb = intel->driDrawable->driverPrivate;
@@ -827,26 +1130,31 @@ void LOCK_HARDWARE( struct intel_context *intel )
        intel_fb->vbl_waited = vbl.reply.sequence;
     }
 
-    DRM_CAS(intel->driHwLock, intel->hHWContext,
-        (DRM_LOCK_HELD|intel->hHWContext), __ret);
+    if (!sPriv->dri2.enabled) {
+       DRM_CAS(intel->driHwLock, intel->hHWContext,
+               (DRM_LOCK_HELD|intel->hHWContext), __ret);
 
-    intel->locked = 1;
+       if (__ret)
+           intelContendedLock( intel, 0 );
+    }
 
-    if (__ret)
-        intelContendedLock( intel, 0 );
 
     if (INTEL_DEBUG & DEBUG_LOCK)
       _mesa_printf("%s - locked\n", __progname);
 }
 
 
-  /* Unlock the hardware using the global current context 
  */
+/* Unlock the hardware using the global current context 
+ */
 void UNLOCK_HARDWARE( struct intel_context *intel )
 {
+    __DRIscreen *sPriv = intel->driScreen;
+
+   intel->vtbl.note_unlock( intel );
    intel->locked = 0;
 
-   DRM_UNLOCK(intel->driFd, intel->driHwLock, intel->hHWContext);
+   if (!sPriv->dri2.enabled)
+      DRM_UNLOCK(intel->driFd, intel->driHwLock, intel->hHWContext);
 
    _glthread_UNLOCK_MUTEX(lockMutex);
 
@@ -857,6 +1165,7 @@ void UNLOCK_HARDWARE( struct intel_context *intel )
     * Nothing should be left in batch outside of LOCK/UNLOCK which references
     * cliprects.
     */
-   assert(intel->batch->cliprect_mode != REFERENCES_CLIPRECTS);
+   if (intel->batch->cliprect_mode == REFERENCES_CLIPRECTS)
+      intel_batchbuffer_flush(intel->batch);
 }