remove final imports.h and imports.c bits
[mesa.git] / src / mesa / state_tracker / st_context.c
index 7e721572e0a26b375c3103b6c8dec0c20170fbec..0a6e9951a1db228df307318167e1d0e2ac0a9253 100644 (file)
@@ -25,7 +25,7 @@
  *
  **************************************************************************/
 
-#include "main/imports.h"
+
 #include "main/accum.h"
 #include "main/api_exec.h"
 #include "main/context.h"
@@ -57,6 +57,7 @@
 #include "st_cb_memoryobjects.h"
 #include "st_cb_msaa.h"
 #include "st_cb_perfmon.h"
+#include "st_cb_perfquery.h"
 #include "st_cb_program.h"
 #include "st_cb_queryobj.h"
 #include "st_cb_readpixels.h"
 #include "st_shader_cache.h"
 #include "st_vdpau.h"
 #include "st_texture.h"
+#include "st_util.h"
 #include "pipe/p_context.h"
+#include "util/u_cpu_detect.h"
 #include "util/u_inlines.h"
 #include "util/u_upload_mgr.h"
 #include "util/u_vbuf.h"
+#include "util/u_memory.h"
 #include "cso_cache/cso_context.h"
+#include "compiler/glsl/glsl_parser_extras.h"
 
 
 DEBUG_GET_ONCE_BOOL_OPTION(mesa_mvp_dp4, "MESA_MVP_DP4", FALSE)
@@ -101,6 +106,9 @@ st_Enable(struct gl_context *ctx, GLenum cap, GLboolean state)
    case GL_DEBUG_OUTPUT_SYNCHRONOUS:
       st_update_debug_callback(st);
       break;
+   case GL_BLACKHOLE_RENDER_INTEL:
+      st->pipe->set_frontend_noop(st->pipe, ctx->IntelBlackholeRender);
+      break;
    default:
       break;
    }
@@ -131,21 +139,21 @@ st_query_memory_info(struct gl_context *ctx, struct gl_memory_info *out)
 }
 
 
-uint64_t
+static uint64_t
 st_get_active_states(struct gl_context *ctx)
 {
-   struct st_vertex_program *vp =
-      st_vertex_program(ctx->VertexProgram._Current);
-   struct st_common_program *tcp =
-      st_common_program(ctx->TessCtrlProgram._Current);
-   struct st_common_program *tep =
-      st_common_program(ctx->TessEvalProgram._Current);
-   struct st_common_program *gp =
-      st_common_program(ctx->GeometryProgram._Current);
-   struct st_fragment_program *fp =
-      st_fragment_program(ctx->FragmentProgram._Current);
-   struct st_compute_program *cp =
-      st_compute_program(ctx->ComputeProgram._Current);
+   struct st_program *vp =
+      st_program(ctx->VertexProgram._Current);
+   struct st_program *tcp =
+      st_program(ctx->TessCtrlProgram._Current);
+   struct st_program *tep =
+      st_program(ctx->TessEvalProgram._Current);
+   struct st_program *gp =
+      st_program(ctx->GeometryProgram._Current);
+   struct st_program *fp =
+      st_program(ctx->FragmentProgram._Current);
+   struct st_program *cp =
+      st_program(ctx->ComputeProgram._Current);
    uint64_t active_shader_states = 0;
 
    if (vp)
@@ -172,7 +180,7 @@ st_invalidate_buffers(struct st_context *st)
    st->dirty |= ST_NEW_BLEND |
                 ST_NEW_DSA |
                 ST_NEW_FB_STATE |
-                ST_NEW_SAMPLE_MASK |
+                ST_NEW_SAMPLE_STATE |
                 ST_NEW_SAMPLE_SHADING |
                 ST_NEW_FS_STATE |
                 ST_NEW_POLY_STIPPLE |
@@ -183,6 +191,14 @@ st_invalidate_buffers(struct st_context *st)
 }
 
 
+static inline bool
+st_vp_uses_current_values(const struct gl_context *ctx)
+{
+   const uint64_t inputs = ctx->VertexProgram._Current->info.inputs_read;
+   return _mesa_draw_current_bits(ctx) & inputs;
+}
+
+
 /**
  * Called via ctx->Driver.UpdateState()
  */
@@ -216,6 +232,10 @@ st_invalidate_state(struct gl_context *ctx)
                     _NEW_POINT))
       st->dirty |= ST_NEW_RASTERIZER;
 
+   if ((new_state & _NEW_LIGHT) &&
+       (st->lower_flatshade || st->lower_two_sided_color))
+      st->dirty |= ST_NEW_FS_STATE;
+
    if (new_state & _NEW_PROJECTION &&
        st_user_clip_planes_enabled(ctx))
       st->dirty |= ST_NEW_CLIP_STATE;
@@ -223,12 +243,26 @@ st_invalidate_state(struct gl_context *ctx)
    if (new_state & _NEW_PIXEL)
       st->dirty |= ST_NEW_PIXEL_TRANSFER;
 
-   if (new_state & _NEW_CURRENT_ATTRIB)
+   if (new_state & _NEW_CURRENT_ATTRIB && st_vp_uses_current_values(ctx))
       st->dirty |= ST_NEW_VERTEX_ARRAYS;
 
+   if (st->clamp_frag_depth_in_shader && (new_state & _NEW_VIEWPORT)) {
+      if (ctx->GeometryProgram._Current)
+         st->dirty |= ST_NEW_GS_CONSTANTS;
+      else if (ctx->TessEvalProgram._Current)
+         st->dirty |= ST_NEW_TES_CONSTANTS;
+      else
+         st->dirty |= ST_NEW_VS_CONSTANTS;
+      st->dirty |= ST_NEW_FS_CONSTANTS;
+   }
+
    /* Update the vertex shader if ctx->Light._ClampVertexColor was changed. */
-   if (st->clamp_vert_color_in_shader && (new_state & _NEW_LIGHT))
+   if (st->clamp_vert_color_in_shader && (new_state & _NEW_LIGHT)) {
       st->dirty |= ST_NEW_VS_STATE;
+      if (st->ctx->API == API_OPENGL_COMPAT && ctx->Version >= 32) {
+         st->dirty |= ST_NEW_GS_STATE | ST_NEW_TES_STATE;
+      }
+   }
 
    /* Which shaders are dirty will be determined manually. */
    if (new_state & _NEW_PROGRAM) {
@@ -251,6 +285,168 @@ st_invalidate_state(struct gl_context *ctx)
 }
 
 
+/*
+ * In some circumstances (such as running google-chrome) the state
+ * tracker may try to delete a resource view from a context different
+ * than when it was created.  We don't want to do that.
+ *
+ * In that situation, st_texture_release_all_sampler_views() calls this
+ * function to transfer the sampler view reference to this context (expected
+ * to be the context which created the view.)
+ */
+void
+st_save_zombie_sampler_view(struct st_context *st,
+                            struct pipe_sampler_view *view)
+{
+   struct st_zombie_sampler_view_node *entry;
+
+   assert(view->context == st->pipe);
+
+   entry = MALLOC_STRUCT(st_zombie_sampler_view_node);
+   if (!entry)
+      return;
+
+   entry->view = view;
+
+   /* We need a mutex since this function may be called from one thread
+    * while free_zombie_resource_views() is called from another.
+    */
+   simple_mtx_lock(&st->zombie_sampler_views.mutex);
+   list_addtail(&entry->node, &st->zombie_sampler_views.list.node);
+   simple_mtx_unlock(&st->zombie_sampler_views.mutex);
+}
+
+
+/*
+ * Since OpenGL shaders may be shared among contexts, we can wind up
+ * with variants of a shader created with different contexts.
+ * When we go to destroy a gallium shader, we want to free it with the
+ * same context that it was created with, unless the driver reports
+ * PIPE_CAP_SHAREABLE_SHADERS = TRUE.
+ */
+void
+st_save_zombie_shader(struct st_context *st,
+                      enum pipe_shader_type type,
+                      struct pipe_shader_state *shader)
+{
+   struct st_zombie_shader_node *entry;
+
+   /* we shouldn't be here if the driver supports shareable shaders */
+   assert(!st->has_shareable_shaders);
+
+   entry = MALLOC_STRUCT(st_zombie_shader_node);
+   if (!entry)
+      return;
+
+   entry->shader = shader;
+   entry->type = type;
+
+   /* We need a mutex since this function may be called from one thread
+    * while free_zombie_shaders() is called from another.
+    */
+   simple_mtx_lock(&st->zombie_shaders.mutex);
+   list_addtail(&entry->node, &st->zombie_shaders.list.node);
+   simple_mtx_unlock(&st->zombie_shaders.mutex);
+}
+
+
+/*
+ * Free any zombie sampler views that may be attached to this context.
+ */
+static void
+free_zombie_sampler_views(struct st_context *st)
+{
+   struct st_zombie_sampler_view_node *entry, *next;
+
+   if (list_is_empty(&st->zombie_sampler_views.list.node)) {
+      return;
+   }
+
+   simple_mtx_lock(&st->zombie_sampler_views.mutex);
+
+   LIST_FOR_EACH_ENTRY_SAFE(entry, next,
+                            &st->zombie_sampler_views.list.node, node) {
+      list_del(&entry->node);  // remove this entry from the list
+
+      assert(entry->view->context == st->pipe);
+      pipe_sampler_view_reference(&entry->view, NULL);
+
+      free(entry);
+   }
+
+   assert(list_is_empty(&st->zombie_sampler_views.list.node));
+
+   simple_mtx_unlock(&st->zombie_sampler_views.mutex);
+}
+
+
+/*
+ * Free any zombie shaders that may be attached to this context.
+ */
+static void
+free_zombie_shaders(struct st_context *st)
+{
+   struct st_zombie_shader_node *entry, *next;
+
+   if (list_is_empty(&st->zombie_shaders.list.node)) {
+      return;
+   }
+
+   simple_mtx_lock(&st->zombie_shaders.mutex);
+
+   LIST_FOR_EACH_ENTRY_SAFE(entry, next,
+                            &st->zombie_shaders.list.node, node) {
+      list_del(&entry->node);  // remove this entry from the list
+
+      switch (entry->type) {
+      case PIPE_SHADER_VERTEX:
+         st->pipe->bind_vs_state(st->pipe, NULL);
+         st->pipe->delete_vs_state(st->pipe, entry->shader);
+         break;
+      case PIPE_SHADER_FRAGMENT:
+         st->pipe->bind_fs_state(st->pipe, NULL);
+         st->pipe->delete_fs_state(st->pipe, entry->shader);
+         break;
+      case PIPE_SHADER_GEOMETRY:
+         st->pipe->bind_gs_state(st->pipe, NULL);
+         st->pipe->delete_gs_state(st->pipe, entry->shader);
+         break;
+      case PIPE_SHADER_TESS_CTRL:
+         st->pipe->bind_tcs_state(st->pipe, NULL);
+         st->pipe->delete_tcs_state(st->pipe, entry->shader);
+         break;
+      case PIPE_SHADER_TESS_EVAL:
+         st->pipe->bind_tes_state(st->pipe, NULL);
+         st->pipe->delete_tes_state(st->pipe, entry->shader);
+         break;
+      case PIPE_SHADER_COMPUTE:
+         st->pipe->bind_compute_state(st->pipe, NULL);
+         st->pipe->delete_compute_state(st->pipe, entry->shader);
+         break;
+      default:
+         unreachable("invalid shader type in free_zombie_shaders()");
+      }
+      free(entry);
+   }
+
+   assert(list_is_empty(&st->zombie_shaders.list.node));
+
+   simple_mtx_unlock(&st->zombie_shaders.mutex);
+}
+
+
+/*
+ * This function is called periodically to free any zombie objects
+ * which are attached to this context.
+ */
+void
+st_context_free_zombie_objects(struct st_context *st)
+{
+   free_zombie_sampler_views(st);
+   free_zombie_shaders(st);
+}
+
+
 static void
 st_destroy_context_priv(struct st_context *st, bool destroy_pipe)
 {
@@ -268,12 +464,13 @@ st_destroy_context_priv(struct st_context *st, bool destroy_pipe)
    st_destroy_bound_image_handles(st);
 
    for (i = 0; i < ARRAY_SIZE(st->state.frag_sampler_views); i++) {
-      pipe_sampler_view_release(st->pipe,
-                                &st->state.frag_sampler_views[i]);
+      pipe_sampler_view_reference(&st->state.vert_sampler_views[i], NULL);
+      pipe_sampler_view_reference(&st->state.frag_sampler_views[i], NULL);
    }
 
    /* free glReadPixels cache data */
    st_invalidate_readpix_cache(st);
+   util_throttle_deinit(st->pipe->screen, &st->throttle);
 
    cso_destroy_context(st->cso_context);
 
@@ -315,7 +512,12 @@ st_init_driver_flags(struct st_context *st)
    f->NewFramebufferSRGB = ST_NEW_FB_STATE;
    f->NewScissorRect = ST_NEW_SCISSOR;
    f->NewScissorTest = ST_NEW_SCISSOR | ST_NEW_RASTERIZER;
-   f->NewAlphaTest = ST_NEW_DSA;
+
+   if (st->lower_alpha_test)
+      f->NewAlphaTest = ST_NEW_FS_STATE;
+   else
+      f->NewAlphaTest = ST_NEW_DSA;
+
    f->NewBlend = ST_NEW_BLEND;
    f->NewBlendColor = ST_NEW_BLEND_COLOR;
    f->NewColorMask = ST_NEW_BLEND;
@@ -323,9 +525,10 @@ st_init_driver_flags(struct st_context *st)
    f->NewLogicOp = ST_NEW_BLEND;
    f->NewStencil = ST_NEW_DSA;
    f->NewMultisampleEnable = ST_NEW_BLEND | ST_NEW_RASTERIZER |
-                             ST_NEW_SAMPLE_MASK | ST_NEW_SAMPLE_SHADING;
+                             ST_NEW_SAMPLE_STATE | ST_NEW_SAMPLE_SHADING;
    f->NewSampleAlphaToXEnable = ST_NEW_BLEND;
-   f->NewSampleMask = ST_NEW_SAMPLE_MASK;
+   f->NewSampleMask = ST_NEW_SAMPLE_STATE;
+   f->NewSampleLocations = ST_NEW_SAMPLE_STATE;
    f->NewSampleShading = ST_NEW_SAMPLE_SHADING;
 
    /* This depends on what the gallium driver wants. */
@@ -338,12 +541,29 @@ st_init_driver_flags(struct st_context *st)
 
    f->NewClipControl = ST_NEW_VIEWPORT | ST_NEW_RASTERIZER;
    f->NewClipPlane = ST_NEW_CLIP_STATE;
-   f->NewClipPlaneEnable = ST_NEW_RASTERIZER;
-   f->NewDepthClamp = ST_NEW_RASTERIZER;
+
+   if (st->clamp_frag_depth_in_shader) {
+      f->NewClipControl |= ST_NEW_VS_STATE | ST_NEW_GS_STATE |
+                           ST_NEW_TES_STATE;
+
+      f->NewDepthClamp = ST_NEW_FS_STATE | ST_NEW_VS_STATE |
+                         ST_NEW_GS_STATE | ST_NEW_TES_STATE;
+   } else {
+      f->NewDepthClamp = ST_NEW_RASTERIZER;
+   }
+
+   if (st->lower_ucp)
+      f->NewClipPlaneEnable = ST_NEW_VS_STATE;
+   else
+      f->NewClipPlaneEnable = ST_NEW_RASTERIZER;
+
    f->NewLineState = ST_NEW_RASTERIZER;
    f->NewPolygonState = ST_NEW_RASTERIZER;
    f->NewPolygonStipple = ST_NEW_POLY_STIPPLE;
    f->NewViewport = ST_NEW_VIEWPORT;
+   f->NewNvConservativeRasterization = ST_NEW_RASTERIZER;
+   f->NewNvConservativeRasterizationParams = ST_NEW_RASTERIZER;
+   f->NewIntelConservativeRasterization = ST_NEW_RASTERIZER;
 }
 
 
@@ -361,10 +581,6 @@ st_create_context_priv(struct gl_context *ctx, struct pipe_context *pipe,
 
    st->ctx = ctx;
    st->pipe = pipe;
-
-   /* state tracker needs the VBO module */
-   _vbo_CreateContext(ctx);
-
    st->dirty = ST_ALL_STATES_MASK;
 
    st->can_bind_const_buffer_as_vertex =
@@ -376,13 +592,12 @@ st_create_context_priv(struct gl_context *ctx, struct pipe_context *pipe,
     * profile, so that u_vbuf is bypassed completely if there is nothing else
     * to do.
     */
-   unsigned vbuf_flags =
-      ctx->API == API_OPENGL_CORE ? U_VBUF_FLAG_NO_USER_VBOS : 0;
-   st->cso_context = cso_create_context(pipe, vbuf_flags);
+   unsigned cso_flags =
+      ctx->API == API_OPENGL_CORE ? CSO_NO_USER_VERTEX_BUFFERS : 0;
+   st->cso_context = cso_create_context(pipe, cso_flags);
 
    st_init_atoms(st);
    st_init_clear(st);
-   st_init_draw(st);
    st_init_pbo_helpers(st);
 
    /* Choose texture target for glDrawPixels, glBitmap, renderbuffers */
@@ -394,54 +609,50 @@ st_create_context_priv(struct gl_context *ctx, struct pipe_context *pipe,
    /* Setup vertex element info for 'struct st_util_vertex'.
     */
    {
-      const unsigned slot = cso_get_aux_vertex_buffer_slot(st->cso_context);
-
-      /* If this assertion ever fails all state tracker calls to
-       * cso_get_aux_vertex_buffer_slot() should be audited.  This
-       * particular call would have to be moved to just before each
-       * drawing call.
-       */
-      assert(slot == 0);
-
       STATIC_ASSERT(sizeof(struct st_util_vertex) == 9 * sizeof(float));
 
       memset(&st->util_velems, 0, sizeof(st->util_velems));
-      st->util_velems[0].src_offset = 0;
-      st->util_velems[0].vertex_buffer_index = slot;
-      st->util_velems[0].src_format = PIPE_FORMAT_R32G32B32_FLOAT;
-      st->util_velems[1].src_offset = 3 * sizeof(float);
-      st->util_velems[1].vertex_buffer_index = slot;
-      st->util_velems[1].src_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
-      st->util_velems[2].src_offset = 7 * sizeof(float);
-      st->util_velems[2].vertex_buffer_index = slot;
-      st->util_velems[2].src_format = PIPE_FORMAT_R32G32_FLOAT;
+      st->util_velems.velems[0].src_offset = 0;
+      st->util_velems.velems[0].vertex_buffer_index = 0;
+      st->util_velems.velems[0].src_format = PIPE_FORMAT_R32G32B32_FLOAT;
+      st->util_velems.velems[1].src_offset = 3 * sizeof(float);
+      st->util_velems.velems[1].vertex_buffer_index = 0;
+      st->util_velems.velems[1].src_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
+      st->util_velems.velems[2].src_offset = 7 * sizeof(float);
+      st->util_velems.velems[2].vertex_buffer_index = 0;
+      st->util_velems.velems[2].src_format = PIPE_FORMAT_R32G32_FLOAT;
    }
 
-   /* we want all vertex data to be placed in buffer objects */
-   vbo_use_buffer_objects(ctx);
-
-
-   /* make sure that no VBOs are left mapped when we're drawing. */
-   vbo_always_unmap_buffers(ctx);
-
    /* Need these flags:
     */
    ctx->FragmentProgram._MaintainTexEnvProgram = GL_TRUE;
-
    ctx->VertexProgram._MaintainTnlProgram = GL_TRUE;
 
    if (no_error)
       ctx->Const.ContextFlags |= GL_CONTEXT_FLAG_NO_ERROR_BIT_KHR;
 
+   ctx->Const.PackedDriverUniformStorage =
+      screen->get_param(screen, PIPE_CAP_PACKED_UNIFORMS);
+
+   ctx->Const.BitmapUsesRed =
+      screen->is_format_supported(screen, PIPE_FORMAT_R8_UNORM,
+                                  PIPE_TEXTURE_2D, 0, 0,
+                                  PIPE_BIND_SAMPLER_VIEW);
+
    st->has_stencil_export =
       screen->get_param(screen, PIPE_CAP_SHADER_STENCIL_EXPORT);
-   st->has_shader_model3 = screen->get_param(screen, PIPE_CAP_SM3);
    st->has_etc1 = screen->is_format_supported(screen, PIPE_FORMAT_ETC1_RGB8,
-                                              PIPE_TEXTURE_2D, 0,
+                                              PIPE_TEXTURE_2D, 0, 0,
                                               PIPE_BIND_SAMPLER_VIEW);
    st->has_etc2 = screen->is_format_supported(screen, PIPE_FORMAT_ETC2_RGB8,
-                                              PIPE_TEXTURE_2D, 0,
+                                              PIPE_TEXTURE_2D, 0, 0,
                                               PIPE_BIND_SAMPLER_VIEW);
+   st->has_astc_2d_ldr =
+      screen->is_format_supported(screen, PIPE_FORMAT_ASTC_4x4_SRGB,
+                                  PIPE_TEXTURE_2D, 0, 0, PIPE_BIND_SAMPLER_VIEW);
+   st->has_astc_5x5_ldr =
+      screen->is_format_supported(screen, PIPE_FORMAT_ASTC_5x5_SRGB,
+                                  PIPE_TEXTURE_2D, 0, 0, PIPE_BIND_SAMPLER_VIEW);
    st->prefer_blit_based_texture_transfer = screen->get_param(screen,
                               PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER);
    st->force_persample_in_shader =
@@ -461,21 +672,58 @@ st_create_context_priv(struct gl_context *ctx, struct pipe_context *pipe,
       screen->get_param(screen, PIPE_CAP_TGSI_PACK_HALF_FLOAT);
    st->has_multi_draw_indirect =
       screen->get_param(screen, PIPE_CAP_MULTI_DRAW_INDIRECT);
+   st->has_single_pipe_stat =
+      screen->get_param(screen, PIPE_CAP_QUERY_PIPELINE_STATISTICS_SINGLE);
+   st->has_indep_blend_func =
+      screen->get_param(screen, PIPE_CAP_INDEP_BLEND_FUNC);
+   st->needs_rgb_dst_alpha_override =
+      screen->get_param(screen, PIPE_CAP_RGB_OVERRIDE_DST_ALPHA_BLEND);
+   st->lower_flatshade =
+      !screen->get_param(screen, PIPE_CAP_FLATSHADE);
+   st->lower_alpha_test =
+      !screen->get_param(screen, PIPE_CAP_ALPHA_TEST);
+   st->lower_point_size =
+      !screen->get_param(screen, PIPE_CAP_POINT_SIZE_FIXED);
+   st->lower_two_sided_color =
+      !screen->get_param(screen, PIPE_CAP_TWO_SIDED_COLOR);
+   st->lower_ucp =
+      !screen->get_param(screen, PIPE_CAP_CLIP_PLANES);
+   st->allow_st_finalize_nir_twice = screen->finalize_nir != NULL;
 
    st->has_hw_atomics =
       screen->get_shader_param(screen, PIPE_SHADER_FRAGMENT,
                                PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTERS)
       ? true : false;
 
+   util_throttle_init(&st->throttle,
+                      screen->get_param(screen,
+                                        PIPE_CAP_MAX_TEXTURE_UPLOAD_MEMORY_BUDGET));
+
    /* GL limits and extensions */
    st_init_limits(pipe->screen, &ctx->Const, &ctx->Extensions);
    st_init_extensions(pipe->screen, &ctx->Const,
                       &ctx->Extensions, &st->options, ctx->API);
 
+   /* FIXME: add support for geometry and tessellation shaders for
+    * lower_point_size
+    */
+   assert(!ctx->Extensions.OES_geometry_shader || !st->lower_point_size);
+   assert(!ctx->Extensions.ARB_tessellation_shader || !st->lower_point_size);
+
+   /* FIXME: add support for geometry and tessellation shaders for
+    * lower_ucp
+    */
+   assert(!ctx->Extensions.OES_geometry_shader || !st->lower_ucp);
+   assert(!ctx->Extensions.ARB_tessellation_shader || !st->lower_ucp);
+
    if (st_have_perfmon(st)) {
       ctx->Extensions.AMD_performance_monitor = GL_TRUE;
    }
 
+   if (st_have_perfquery(st)) {
+      ctx->Extensions.INTEL_performance_query = GL_TRUE;
+   }
+
    /* Enable shader-based fallbacks for ARB_color_buffer_float if needed. */
    if (screen->get_param(screen, PIPE_CAP_VERTEX_COLOR_UNCLAMPED)) {
       if (!screen->get_param(screen, PIPE_CAP_VERTEX_COLOR_CLAMPED)) {
@@ -497,6 +745,9 @@ st_create_context_priv(struct gl_context *ctx, struct pipe_context *pipe,
       }
    }
 
+   if (screen->get_param(screen, PIPE_CAP_DEPTH_CLIP_DISABLE) == 2)
+      st->clamp_frag_depth_in_shader = true;
+
    /* called after _mesa_create_context/_mesa_init_point, fix default user
     * settable max point size up
     */
@@ -506,9 +757,16 @@ st_create_context_priv(struct gl_context *ctx, struct pipe_context *pipe,
     * is not supported
     */
    ctx->Const.ShaderCompilerOptions[MESA_SHADER_VERTEX].EmitNoSat =
-      !st->has_shader_model3;
+      !screen->get_param(screen, PIPE_CAP_VERTEX_SHADER_SATURATE);
+
+   ctx->Const.ShaderCompilerOptions[MESA_SHADER_VERTEX].PositionAlwaysInvariant = options->vs_position_always_invariant;
+
+   enum pipe_shader_ir preferred_ir = (enum pipe_shader_ir)
+      screen->get_shader_param(screen, PIPE_SHADER_VERTEX,
+                               PIPE_SHADER_CAP_PREFERRED_IR);
+   ctx->Const.UseNIRGLSLLinker = preferred_ir == PIPE_SHADER_IR_NIR;
 
-   if (!ctx->Extensions.ARB_gpu_shader5) {
+   if (ctx->Const.GLSLVersion < 400) {
       for (i = 0; i < MESA_SHADER_STAGES; i++)
          ctx->Const.ShaderCompilerOptions[i].EmitNoIndirectSampler = true;
    }
@@ -516,16 +774,29 @@ st_create_context_priv(struct gl_context *ctx, struct pipe_context *pipe,
    /* Set which shader types can be compiled at link time. */
    st->shader_has_one_variant[MESA_SHADER_VERTEX] =
          st->has_shareable_shaders &&
-         !st->clamp_vert_color_in_shader;
+         !st->clamp_frag_depth_in_shader &&
+         !st->clamp_vert_color_in_shader &&
+         !st->lower_point_size &&
+         !st->lower_ucp;
 
    st->shader_has_one_variant[MESA_SHADER_FRAGMENT] =
          st->has_shareable_shaders &&
+         !st->lower_flatshade &&
+         !st->lower_alpha_test &&
          !st->clamp_frag_color_in_shader &&
-         !st->force_persample_in_shader;
+         !st->clamp_frag_depth_in_shader &&
+         !st->force_persample_in_shader &&
+         !st->lower_two_sided_color;
 
    st->shader_has_one_variant[MESA_SHADER_TESS_CTRL] = st->has_shareable_shaders;
-   st->shader_has_one_variant[MESA_SHADER_TESS_EVAL] = st->has_shareable_shaders;
-   st->shader_has_one_variant[MESA_SHADER_GEOMETRY] = st->has_shareable_shaders;
+   st->shader_has_one_variant[MESA_SHADER_TESS_EVAL] =
+         st->has_shareable_shaders &&
+         !st->clamp_frag_depth_in_shader &&
+         !st->clamp_vert_color_in_shader;
+   st->shader_has_one_variant[MESA_SHADER_GEOMETRY] =
+         st->has_shareable_shaders &&
+         !st->clamp_frag_depth_in_shader &&
+         !st->clamp_vert_color_in_shader;
    st->shader_has_one_variant[MESA_SHADER_COMPUTE] = st->has_shareable_shaders;
 
    st->bitmap.cache.empty = true;
@@ -541,127 +812,27 @@ st_create_context_priv(struct gl_context *ctx, struct pipe_context *pipe,
       return NULL;
    }
 
+   /* This must be done after extensions are initialized to enable persistent
+    * mappings immediately.
+    */
+   _vbo_CreateContext(ctx, true);
+
    _mesa_initialize_dispatch_tables(ctx);
    _mesa_initialize_vbo_vtxfmt(ctx);
    st_init_driver_flags(st);
 
    /* Initialize context's winsys buffers list */
-   LIST_INITHEAD(&st->winsys_buffers);
-
-   return st;
-}
-
-
-struct st_context *
-st_create_context(gl_api api, struct pipe_context *pipe,
-                  const struct gl_config *visual,
-                  struct st_context *share,
-                  const struct st_config_options *options,
-                  bool no_error)
-{
-   struct gl_context *ctx;
-   struct gl_context *shareCtx = share ? share->ctx : NULL;
-   struct dd_function_table funcs;
-   struct st_context *st;
-
-   memset(&funcs, 0, sizeof(funcs));
-   st_init_driver_functions(pipe->screen, &funcs);
-
-   ctx = calloc(1, sizeof(struct gl_context));
-   if (!ctx)
-      return NULL;
-
-   if (!_mesa_initialize_context(ctx, api, visual, shareCtx, &funcs)) {
-      free(ctx);
-      return NULL;
-   }
-
-   st_debug_init();
+   list_inithead(&st->winsys_buffers);
 
-   if (pipe->screen->get_disk_shader_cache &&
-       !(ST_DEBUG & DEBUG_TGSI))
-      ctx->Cache = pipe->screen->get_disk_shader_cache(pipe->screen);
-
-   /* XXX: need a capability bit in gallium to query if the pipe
-    * driver prefers DP4 or MUL/MAD for vertex transformation.
-    */
-   if (debug_get_option_mesa_mvp_dp4())
-      ctx->Const.ShaderCompilerOptions[MESA_SHADER_VERTEX].OptimizeForAOS = GL_TRUE;
-
-   st = st_create_context_priv(ctx, pipe, options, no_error);
-   if (!st) {
-      _mesa_destroy_context(ctx);
-   }
+   list_inithead(&st->zombie_sampler_views.list.node);
+   simple_mtx_init(&st->zombie_sampler_views.mutex, mtx_plain);
+   list_inithead(&st->zombie_shaders.list.node);
+   simple_mtx_init(&st->zombie_shaders.mutex, mtx_plain);
 
    return st;
 }
 
 
-/**
- * Callback to release the sampler view attached to a texture object.
- * Called by _mesa_HashWalk().
- */
-static void
-destroy_tex_sampler_cb(GLuint id, void *data, void *userData)
-{
-   struct gl_texture_object *texObj = (struct gl_texture_object *) data;
-   struct st_context *st = (struct st_context *) userData;
-
-   st_texture_release_sampler_view(st, st_texture_object(texObj));
-}
-
-
-void
-st_destroy_context(struct st_context *st)
-{
-   struct gl_context *ctx = st->ctx;
-   struct st_framebuffer *stfb, *next;
-
-   GET_CURRENT_CONTEXT(curctx);
-
-   if (curctx == NULL) {
-      /* No current context, but we need one to release
-       * renderbuffer surface when we release framebuffer.
-       * So temporarily bind the context.
-       */
-      _mesa_make_current(ctx, NULL, NULL);
-   }
-
-   /* This must be called first so that glthread has a chance to finish */
-   _mesa_glthread_destroy(ctx);
-
-   _mesa_HashWalk(ctx->Shared->TexObjects, destroy_tex_sampler_cb, st);
-
-   st_reference_fragprog(st, &st->fp, NULL);
-   st_reference_prog(st, &st->gp, NULL);
-   st_reference_vertprog(st, &st->vp, NULL);
-   st_reference_prog(st, &st->tcp, NULL);
-   st_reference_prog(st, &st->tep, NULL);
-   st_reference_compprog(st, &st->cp, NULL);
-
-   /* release framebuffer in the winsys buffers list */
-   LIST_FOR_EACH_ENTRY_SAFE_REV(stfb, next, &st->winsys_buffers, head) {
-      st_framebuffer_reference(&stfb, NULL);
-   }
-
-   pipe_sampler_view_reference(&st->pixel_xfer.pixelmap_sampler_view, NULL);
-   pipe_resource_reference(&st->pixel_xfer.pixelmap_texture, NULL);
-
-   _vbo_DestroyContext(ctx);
-
-   st_destroy_program_variants(st);
-
-   _mesa_free_context_data(ctx);
-
-   /* This will free the st_context too, so 'st' must not be accessed
-    * afterwards. */
-   st_destroy_context_priv(st, true);
-   st = NULL;
-
-   free(ctx);
-}
-
-
 static void
 st_emit_string_marker(struct gl_context *ctx, const GLchar *string, GLsizei len)
 {
@@ -705,13 +876,13 @@ st_get_driver_uuid(struct gl_context *ctx, char *uuid)
 }
 
 
-void
+static void
 st_init_driver_functions(struct pipe_screen *screen,
                          struct dd_function_table *functions)
 {
-   _mesa_init_shader_object_functions(functions);
    _mesa_init_sampler_object_functions(functions);
 
+   st_init_draw_functions(functions);
    st_init_blit_functions(functions);
    st_init_bufferobject_functions(screen, functions);
    st_init_clear_functions(functions);
@@ -729,6 +900,7 @@ st_init_driver_functions(struct pipe_screen *screen,
    st_init_memoryobject_functions(functions);
    st_init_msaa_functions(functions);
    st_init_perfmon_functions(functions);
+   st_init_perfquery_functions(functions);
    st_init_program_functions(functions);
    st_init_query_functions(functions);
    st_init_cond_render_functions(functions);
@@ -758,6 +930,182 @@ st_init_driver_functions(struct pipe_screen *screen,
 
    /* GL_ARB_get_program_binary */
    functions->GetProgramBinaryDriverSHA1 = st_get_program_binary_driver_sha1;
-   functions->ProgramBinarySerializeDriverBlob = st_serialise_tgsi_program;
-   functions->ProgramBinaryDeserializeDriverBlob = st_deserialise_tgsi_program;
+
+   enum pipe_shader_ir preferred_ir = (enum pipe_shader_ir)
+      screen->get_shader_param(screen, PIPE_SHADER_VERTEX,
+                               PIPE_SHADER_CAP_PREFERRED_IR);
+   if (preferred_ir == PIPE_SHADER_IR_NIR) {
+      functions->ShaderCacheSerializeDriverBlob =  st_serialise_nir_program;
+      functions->ProgramBinarySerializeDriverBlob =
+         st_serialise_nir_program_binary;
+      functions->ProgramBinaryDeserializeDriverBlob =
+         st_deserialise_nir_program;
+   } else {
+      functions->ShaderCacheSerializeDriverBlob =  st_serialise_tgsi_program;
+      functions->ProgramBinarySerializeDriverBlob =
+         st_serialise_tgsi_program_binary;
+      functions->ProgramBinaryDeserializeDriverBlob =
+         st_deserialise_tgsi_program;
+   }
+}
+
+
+struct st_context *
+st_create_context(gl_api api, struct pipe_context *pipe,
+                  const struct gl_config *visual,
+                  struct st_context *share,
+                  const struct st_config_options *options,
+                  bool no_error)
+{
+   struct gl_context *ctx;
+   struct gl_context *shareCtx = share ? share->ctx : NULL;
+   struct dd_function_table funcs;
+   struct st_context *st;
+
+   util_cpu_detect();
+
+   memset(&funcs, 0, sizeof(funcs));
+   st_init_driver_functions(pipe->screen, &funcs);
+
+   ctx = calloc(1, sizeof(struct gl_context));
+   if (!ctx)
+      return NULL;
+
+   if (!_mesa_initialize_context(ctx, api, visual, shareCtx, &funcs)) {
+      free(ctx);
+      return NULL;
+   }
+
+   st_debug_init();
+
+   if (pipe->screen->get_disk_shader_cache)
+      ctx->Cache = pipe->screen->get_disk_shader_cache(pipe->screen);
+
+   /* XXX: need a capability bit in gallium to query if the pipe
+    * driver prefers DP4 or MUL/MAD for vertex transformation.
+    */
+   if (debug_get_option_mesa_mvp_dp4())
+      ctx->Const.ShaderCompilerOptions[MESA_SHADER_VERTEX].OptimizeForAOS = GL_TRUE;
+
+   st = st_create_context_priv(ctx, pipe, options, no_error);
+   if (!st) {
+      _mesa_destroy_context(ctx);
+   }
+
+   return st;
+}
+
+
+/**
+ * When we destroy a context, we must examine all texture objects to
+ * find/release any sampler views created by that context.
+ *
+ * This callback is called per-texture object.  It releases all the
+ * texture's sampler views which belong to the context.
+ */
+static void
+destroy_tex_sampler_cb(GLuint id, void *data, void *userData)
+{
+   struct gl_texture_object *texObj = (struct gl_texture_object *) data;
+   struct st_context *st = (struct st_context *) userData;
+
+   st_texture_release_context_sampler_view(st, st_texture_object(texObj));
+}
+
+static void
+destroy_framebuffer_attachment_sampler_cb(GLuint id, void *data, void *userData)
+{
+   struct gl_framebuffer* glfb = (struct gl_framebuffer*) data;
+   struct st_context *st = (struct st_context *) userData;
+
+    for (unsigned i = 0; i < BUFFER_COUNT; i++) {
+      struct gl_renderbuffer_attachment *att = &glfb->Attachment[i];
+      if (att->Texture) {
+        st_texture_release_context_sampler_view(st, st_texture_object(att->Texture));
+      }
+   }
+}
+
+void
+st_destroy_context(struct st_context *st)
+{
+   struct gl_context *ctx = st->ctx;
+   struct st_framebuffer *stfb, *next;
+   struct gl_framebuffer *save_drawbuffer;
+   struct gl_framebuffer *save_readbuffer;
+
+   /* Save the current context and draw/read buffers*/
+   GET_CURRENT_CONTEXT(save_ctx);
+   if (save_ctx) {
+      save_drawbuffer = save_ctx->WinSysDrawBuffer;
+      save_readbuffer = save_ctx->WinSysReadBuffer;
+   } else {
+      save_drawbuffer = save_readbuffer = NULL;
+   }
+
+   /*
+    * We need to bind the context we're deleting so that
+    * _mesa_reference_texobj_() uses this context when deleting textures.
+    * Similarly for framebuffer objects, etc.
+    */
+   _mesa_make_current(ctx, NULL, NULL);
+
+   /* This must be called first so that glthread has a chance to finish */
+   _mesa_glthread_destroy(ctx);
+
+   _mesa_HashWalk(ctx->Shared->TexObjects, destroy_tex_sampler_cb, st);
+
+   /* For the fallback textures, free any sampler views belonging to this
+    * context.
+    */
+   for (unsigned i = 0; i < NUM_TEXTURE_TARGETS; i++) {
+      struct st_texture_object *stObj =
+         st_texture_object(ctx->Shared->FallbackTex[i]);
+      if (stObj) {
+         st_texture_release_context_sampler_view(st, stObj);
+      }
+   }
+
+   st_context_free_zombie_objects(st);
+
+   simple_mtx_destroy(&st->zombie_sampler_views.mutex);
+   simple_mtx_destroy(&st->zombie_shaders.mutex);
+
+   st_release_program(st, &st->fp);
+   st_release_program(st, &st->gp);
+   st_release_program(st, &st->vp);
+   st_release_program(st, &st->tcp);
+   st_release_program(st, &st->tep);
+   st_release_program(st, &st->cp);
+
+   /* release framebuffer in the winsys buffers list */
+   LIST_FOR_EACH_ENTRY_SAFE_REV(stfb, next, &st->winsys_buffers, head) {
+      st_framebuffer_reference(&stfb, NULL);
+   }
+
+   _mesa_HashWalk(ctx->Shared->FrameBuffers, destroy_framebuffer_attachment_sampler_cb, st);
+
+   pipe_sampler_view_reference(&st->pixel_xfer.pixelmap_sampler_view, NULL);
+   pipe_resource_reference(&st->pixel_xfer.pixelmap_texture, NULL);
+
+   _vbo_DestroyContext(ctx);
+
+   st_destroy_program_variants(st);
+
+   _mesa_free_context_data(ctx);
+
+   /* This will free the st_context too, so 'st' must not be accessed
+    * afterwards. */
+   st_destroy_context_priv(st, true);
+   st = NULL;
+
+   free(ctx);
+
+   if (save_ctx == ctx) {
+      /* unbind the context we just deleted */
+      _mesa_make_current(NULL, NULL, NULL);
+   } else {
+      /* Restore the current context and draw/read buffers (may be NULL) */
+      _mesa_make_current(save_ctx, save_drawbuffer, save_readbuffer);
+   }
 }