st/mesa: add a second pipeline for compute
[mesa.git] / src / mesa / state_tracker / st_context.c
index 0245fd92b3eb32b18f158e13e88fb383e397e8ef..b73b0abd55a19f03a71a8cab3451e76c8f34924c 100644 (file)
@@ -1,6 +1,6 @@
 /**************************************************************************
  * 
- * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * Copyright 2007 VMware, Inc.
  * All Rights Reserved.
  * 
  * Permission is hereby granted, free of charge, to any person obtaining a
@@ -18,7 +18,7 @@
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
- * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 
 #include "main/imports.h"
 #include "main/accum.h"
+#include "main/api_exec.h"
 #include "main/context.h"
 #include "main/samplerobj.h"
 #include "main/shaderobj.h"
+#include "main/version.h"
+#include "main/vtxfmt.h"
+#include "main/hash.h"
 #include "program/prog_cache.h"
 #include "vbo/vbo.h"
 #include "glapi/glapi.h"
 #include "st_cb_bufferobjects.h"
 #include "st_cb_clear.h"
 #include "st_cb_condrender.h"
+#include "st_cb_copyimage.h"
 #include "st_cb_drawpixels.h"
 #include "st_cb_rasterpos.h"
 #include "st_cb_drawtex.h"
 #include "st_cb_eglimage.h"
 #include "st_cb_fbo.h"
 #include "st_cb_feedback.h"
+#include "st_cb_msaa.h"
+#include "st_cb_perfmon.h"
 #include "st_cb_program.h"
 #include "st_cb_queryobj.h"
 #include "st_cb_readpixels.h"
@@ -61,6 +68,8 @@
 #include "st_extensions.h"
 #include "st_gen_mipmap.h"
 #include "st_program.h"
+#include "st_vdpau.h"
+#include "st_texture.h"
 #include "pipe/p_context.h"
 #include "util/u_inlines.h"
 #include "util/u_upload_mgr.h"
 DEBUG_GET_ONCE_BOOL_OPTION(mesa_mvp_dp4, "MESA_MVP_DP4", FALSE)
 
 
+/**
+ * Called via ctx->Driver.Enable()
+ */
+static void st_Enable(struct gl_context * ctx, GLenum cap, GLboolean state)
+{
+   struct st_context *st = st_context(ctx);
+
+   switch (cap) {
+   case GL_DEBUG_OUTPUT:
+      st_enable_debug_output(st, state);
+      break;
+   default:
+      break;
+   }
+}
+
+
+/**
+ * Called via ctx->Driver.QueryMemoryInfo()
+ */
+static void
+st_query_memory_info(struct gl_context *ctx, struct gl_memory_info *out)
+{
+   struct pipe_screen *screen = st_context(ctx)->pipe->screen;
+   struct pipe_memory_info info;
+
+   assert(screen->query_memory_info);
+   if (!screen->query_memory_info)
+      return;
+
+   screen->query_memory_info(screen, &info);
+
+   out->total_device_memory = info.total_device_memory;
+   out->avail_device_memory = info.avail_device_memory;
+   out->total_staging_memory = info.total_staging_memory;
+   out->avail_staging_memory = info.avail_staging_memory;
+   out->device_memory_evicted = info.device_memory_evicted;
+   out->nr_device_memory_evictions = info.nr_device_memory_evictions;
+}
+
+
 /**
  * Called via ctx->Driver.UpdateState()
  */
-void st_invalidate_state(struct gl_context * ctx, GLuint new_state)
+void st_invalidate_state(struct gl_context * ctx, GLbitfield new_state)
 {
    struct st_context *st = st_context(ctx);
 
@@ -88,8 +138,11 @@ void st_invalidate_state(struct gl_context * ctx, GLuint new_state)
       st->dirty.st |= ST_NEW_VERTEX_PROGRAM;
    }
 
+   /* Invalidate render and compute pipelines. */
    st->dirty.mesa |= new_state;
    st->dirty.st |= ST_NEW_MESA;
+   st->dirty_cp.mesa |= new_state;
+   st->dirty_cp.st |= ST_NEW_MESA;
 
    /* This is the only core Mesa module we depend upon.
     * No longer use swrast, swsetup, tnl.
@@ -98,25 +151,55 @@ void st_invalidate_state(struct gl_context * ctx, GLuint new_state)
 }
 
 
-/**
- * Check for multisample env var override.
- */
-int
-st_get_msaa(void)
+static void
+st_destroy_context_priv(struct st_context *st)
 {
-   const char *msaa = _mesa_getenv("__GL_FSAA_MODE");
-   if (msaa)
-      return atoi(msaa);
-   return 0;
+   uint shader, i;
+
+   st_destroy_atoms( st );
+   st_destroy_draw( st );
+   st_destroy_clear(st);
+   st_destroy_bitmap(st);
+   st_destroy_drawpix(st);
+   st_destroy_drawtex(st);
+   st_destroy_perfmon(st);
+   st_destroy_pbo_upload(st);
+
+   for (shader = 0; shader < ARRAY_SIZE(st->state.sampler_views); shader++) {
+      for (i = 0; i < ARRAY_SIZE(st->state.sampler_views[0]); i++) {
+         pipe_sampler_view_release(st->pipe,
+                                   &st->state.sampler_views[shader][i]);
+      }
+   }
+
+   if (st->default_texture) {
+      st->ctx->Driver.DeleteTexture(st->ctx, st->default_texture);
+      st->default_texture = NULL;
+   }
+
+   u_upload_destroy(st->uploader);
+   if (st->indexbuf_uploader) {
+      u_upload_destroy(st->indexbuf_uploader);
+   }
+   if (st->constbuf_uploader) {
+      u_upload_destroy(st->constbuf_uploader);
+   }
+
+   cso_destroy_context(st->cso_context);
+   free( st );
 }
 
 
 static struct st_context *
-st_create_context_priv( struct gl_context *ctx, struct pipe_context *pipe )
+st_create_context_priv( struct gl_context *ctx, struct pipe_context *pipe,
+               const struct st_config_options *options)
 {
+   struct pipe_screen *screen = pipe->screen;
    uint i;
    struct st_context *st = ST_CALLOC_STRUCT( st_context );
    
+   st->options = *options;
+
    ctx->st = st;
 
    st->ctx = ctx;
@@ -128,29 +211,51 @@ st_create_context_priv( struct gl_context *ctx, struct pipe_context *pipe )
    /* state tracker needs the VBO module */
    _vbo_CreateContext(ctx);
 
+   /* Initialize render and compute pipelines flags */
    st->dirty.mesa = ~0;
    st->dirty.st = ~0;
+   st->dirty_cp.mesa = ~0;
+   st->dirty_cp.st = ~0;
+
+   /* Create upload manager for vertex data for glBitmap, glDrawPixels,
+    * glClear, etc.
+    */
+   st->uploader = u_upload_create(st->pipe, 65536, PIPE_BIND_VERTEX_BUFFER,
+                                  PIPE_USAGE_STREAM);
+
+   if (!screen->get_param(screen, PIPE_CAP_USER_INDEX_BUFFERS)) {
+      st->indexbuf_uploader = u_upload_create(st->pipe, 128 * 1024,
+                                              PIPE_BIND_INDEX_BUFFER,
+                                              PIPE_USAGE_STREAM);
+   }
+
+   if (!screen->get_param(screen, PIPE_CAP_USER_CONSTANT_BUFFERS))
+      st->constbuf_uploader = u_upload_create(pipe, 128 * 1024,
+                                              PIPE_BIND_CONSTANT_BUFFER,
+                                              PIPE_USAGE_STREAM);
 
-   st->uploader = u_upload_create(st->pipe, 65536, 4, PIPE_BIND_VERTEX_BUFFER);
    st->cso_context = cso_create_context(pipe);
 
    st_init_atoms( st );
-   st_init_bitmap(st);
    st_init_clear(st);
    st_init_draw( st );
-   st_init_generate_mipmap(st);
-   st_init_blit(st);
+   st_init_pbo_upload(st);
 
-   if(pipe->screen->get_param(pipe->screen, PIPE_CAP_NPOT_TEXTURES))
+   /* Choose texture target for glDrawPixels, glBitmap, renderbuffers */
+   if (pipe->screen->get_param(pipe->screen, PIPE_CAP_NPOT_TEXTURES))
       st->internal_target = PIPE_TEXTURE_2D;
    else
       st->internal_target = PIPE_TEXTURE_RECT;
 
-   for (i = 0; i < 3; i++) {
+   /* Vertex element objects used for drawing rectangles for glBitmap,
+    * glDrawPixels, glClear, etc.
+    */
+   for (i = 0; i < ARRAY_SIZE(st->velems_util_draw); i++) {
       memset(&st->velems_util_draw[i], 0, sizeof(struct pipe_vertex_element));
       st->velems_util_draw[i].src_offset = i * 4 * sizeof(float);
       st->velems_util_draw[i].instance_divisor = 0;
-      st->velems_util_draw[i].vertex_buffer_index = 0;
+      st->velems_util_draw[i].vertex_buffer_index =
+            cso_get_aux_vertex_buffer_slot(st->cso_context);
       st->velems_util_draw[i].src_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
    }
 
@@ -167,107 +272,188 @@ st_create_context_priv( struct gl_context *ctx, struct pipe_context *pipe )
 
    st->ctx->VertexProgram._MaintainTnlProgram = GL_TRUE;
 
-   st->pixel_xfer.cache = _mesa_new_program_cache();
-
-   st->force_msaa = st_get_msaa();
+   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_BIND_SAMPLER_VIEW);
+   st->has_etc2 = screen->is_format_supported(screen, PIPE_FORMAT_ETC2_RGB8,
+                                              PIPE_TEXTURE_2D, 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 =
+      screen->get_param(screen, PIPE_CAP_SAMPLE_SHADING) &&
+      !screen->get_param(screen, PIPE_CAP_FORCE_PERSAMPLE_INTERP);
+   st->has_shareable_shaders = screen->get_param(screen,
+                                                 PIPE_CAP_SHAREABLE_SHADERS);
+   st->needs_texcoord_semantic =
+      screen->get_param(screen, PIPE_CAP_TGSI_TEXCOORD);
+   st->apply_texture_swizzle_to_border_color =
+      !!(screen->get_param(screen, PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK) &
+         (PIPE_QUIRK_TEXTURE_BORDER_COLOR_SWIZZLE_NV50 |
+          PIPE_QUIRK_TEXTURE_BORDER_COLOR_SWIZZLE_R600));
+   st->has_time_elapsed =
+      screen->get_param(screen, PIPE_CAP_QUERY_TIME_ELAPSED);
+   st->has_half_float_packing =
+      screen->get_param(screen, PIPE_CAP_TGSI_PACK_HALF_FLOAT);
+   st->has_multi_draw_indirect =
+      screen->get_param(screen, PIPE_CAP_MULTI_DRAW_INDIRECT);
 
    /* GL limits and extensions */
-   st_init_limits(st);
-   st_init_extensions(st);
+   st_init_limits(st->pipe->screen, &ctx->Const, &ctx->Extensions);
+   st_init_extensions(st->pipe->screen, &ctx->Const,
+                      &ctx->Extensions, &st->options, ctx->Mesa_DXTn);
+
+   if (st_have_perfmon(st)) {
+      ctx->Extensions.AMD_performance_monitor = 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)) {
+         st->clamp_vert_color_in_shader = GL_TRUE;
+      }
+
+      if (!screen->get_param(screen, PIPE_CAP_FRAGMENT_COLOR_CLAMPED)) {
+         st->clamp_frag_color_in_shader = GL_TRUE;
+      }
+
+      /* For drivers which cannot do color clamping, it's better to just
+       * disable ARB_color_buffer_float in the core profile, because
+       * the clamping is deprecated there anyway. */
+      if (ctx->API == API_OPENGL_CORE &&
+          (st->clamp_frag_color_in_shader || st->clamp_vert_color_in_shader)) {
+         st->clamp_vert_color_in_shader = GL_FALSE;
+         st->clamp_frag_color_in_shader = GL_FALSE;
+         ctx->Extensions.ARB_color_buffer_float = GL_FALSE;
+      }
+   }
+
+   /* called after _mesa_create_context/_mesa_init_point, fix default user
+    * settable max point size up
+    */
+   st->ctx->Point.MaxSize = MAX2(ctx->Const.MaxPointSize,
+                                 ctx->Const.MaxPointSizeAA);
+   /* For vertex shaders, make sure not to emit saturate when SM 3.0 is not supported */
+   ctx->Const.ShaderCompilerOptions[MESA_SHADER_VERTEX].EmitNoSat = !st->has_shader_model3;
+
+   if (!ctx->Extensions.ARB_gpu_shader5) {
+      for (i = 0; i < MESA_SHADER_STAGES; i++)
+         ctx->Const.ShaderCompilerOptions[i].EmitNoIndirectSampler = true;
+   }
+
+   /* 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->shader_has_one_variant[MESA_SHADER_FRAGMENT] =
+         st->has_shareable_shaders &&
+         !st->clamp_frag_color_in_shader &&
+         !st->force_persample_in_shader;
+
+   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;
+
+   _mesa_compute_version(ctx);
+
+   if (ctx->Version == 0) {
+      /* This can happen when a core profile was requested, but the driver
+       * does not support some features of GL 3.1 or later.
+       */
+      st_destroy_context_priv(st);
+      return NULL;
+   }
+
+   _mesa_initialize_dispatch_tables(ctx);
+   _mesa_initialize_vbo_vtxfmt(ctx);
 
    return st;
 }
 
+static void st_init_driver_flags(struct gl_driver_flags *f)
+{
+   f->NewArray = ST_NEW_VERTEX_ARRAYS;
+   f->NewRasterizerDiscard = ST_NEW_RASTERIZER;
+   f->NewUniformBuffer = ST_NEW_UNIFORM_BUFFER;
+   f->NewDefaultTessLevels = ST_NEW_TESS_STATE;
+   f->NewTextureBuffer = ST_NEW_SAMPLER_VIEWS;
+   f->NewAtomicBuffer = ST_NEW_ATOMIC_BUFFER;
+   f->NewShaderStorageBuffer = ST_NEW_STORAGE_BUFFER;
+}
 
 struct st_context *st_create_context(gl_api api, struct pipe_context *pipe,
                                      const struct gl_config *visual,
-                                     struct st_context *share)
+                                     struct st_context *share,
+                                     const struct st_config_options *options)
 {
    struct gl_context *ctx;
    struct gl_context *shareCtx = share ? share->ctx : NULL;
    struct dd_function_table funcs;
-
-   /* Sanity checks */
-   assert(MESA_SHADER_VERTEX == PIPE_SHADER_VERTEX);
-   assert(MESA_SHADER_FRAGMENT == PIPE_SHADER_FRAGMENT);
-   assert(MESA_SHADER_GEOMETRY == PIPE_SHADER_GEOMETRY);
+   struct st_context *st;
 
    memset(&funcs, 0, sizeof(funcs));
-   st_init_driver_functions(&funcs);
+   st_init_driver_functions(pipe->screen, &funcs);
 
-   ctx = _mesa_create_context(api, visual, shareCtx, &funcs, NULL);
+   ctx = _mesa_create_context(api, visual, shareCtx, &funcs);
    if (!ctx) {
       return NULL;
    }
 
+   st_init_driver_flags(&ctx->DriverFlags);
+
    /* 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())
-      _mesa_set_mvp_with_dp4( ctx, GL_TRUE );
+      ctx->Const.ShaderCompilerOptions[MESA_SHADER_VERTEX].OptimizeForAOS = GL_TRUE;
+
+   st = st_create_context_priv(ctx, pipe, options);
+   if (!st) {
+      _mesa_destroy_context(ctx);
+   }
 
-   return st_create_context_priv(ctx, pipe);
+   return st;
 }
 
 
-static void st_destroy_context_priv( struct st_context *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)
 {
-   uint i;
-
-   st_destroy_atoms( st );
-   st_destroy_draw( st );
-   st_destroy_generate_mipmap(st);
-   st_destroy_blit(st);
-   st_destroy_clear(st);
-   st_destroy_bitmap(st);
-   st_destroy_drawpix(st);
-   st_destroy_drawtex(st);
-
-   /* Unreference any user vertex buffers. */
-   for (i = 0; i < st->num_user_attribs; i++) {
-      pipe_resource_reference(&st->user_attrib[i].buffer, NULL);
-   }
-
-   for (i = 0; i < Elements(st->state.sampler_views); i++) {
-      pipe_sampler_view_reference(&st->state.sampler_views[i], NULL);
-   }
+   struct gl_texture_object *texObj = (struct gl_texture_object *) data;
+   struct st_context *st = (struct st_context *) userData;
 
-   if (st->default_texture) {
-      st->ctx->Driver.DeleteTexture(st->ctx, st->default_texture);
-      st->default_texture = NULL;
-   }
-
-   u_upload_destroy(st->uploader);
-   free( st );
+   st_texture_release_sampler_view(st, st_texture_object(texObj));
 }
-
  
 void st_destroy_context( struct st_context *st )
 {
    struct pipe_context *pipe = st->pipe;
-   struct cso_context *cso = st->cso_context;
    struct gl_context *ctx = st->ctx;
    GLuint i;
 
-   /* need to unbind and destroy CSO objects before anything else */
-   cso_release_all(st->cso_context);
+   _mesa_HashWalk(ctx->Shared->TexObjects, destroy_tex_sampler_cb, st);
 
    st_reference_fragprog(st, &st->fp, NULL);
+   st_reference_geomprog(st, &st->gp, NULL);
    st_reference_vertprog(st, &st->vp, NULL);
+   st_reference_tesscprog(st, &st->tcp, NULL);
+   st_reference_tesseprog(st, &st->tep, NULL);
 
    /* release framebuffer surfaces */
    for (i = 0; i < PIPE_MAX_COLOR_BUFS; i++) {
       pipe_surface_reference(&st->state.framebuffer.cbufs[i], NULL);
    }
    pipe_surface_reference(&st->state.framebuffer.zsbuf, NULL);
-
-   pipe->set_index_buffer(pipe, NULL);
-
-   for (i = 0; i < PIPE_SHADER_TYPES; i++) {
-      pipe->set_constant_buffer(pipe, i, 0, NULL);
-   }
-
-   _mesa_delete_program_cache(st->ctx, st->pixel_xfer.cache);
+   pipe_sampler_view_reference(&st->pixel_xfer.pixelmap_sampler_view, NULL);
+   pipe_resource_reference(&st->pixel_xfer.pixelmap_texture, NULL);
 
    _vbo_DestroyContext(st->ctx);
 
@@ -275,27 +461,34 @@ void st_destroy_context( struct st_context *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);
-
-   cso_destroy_context(cso);
+   st = NULL;
 
    pipe->destroy( pipe );
 
    free(ctx);
 }
 
+static void
+st_emit_string_marker(struct gl_context *ctx, const GLchar *string, GLsizei len)
+{
+   struct st_context *st = ctx->st;
+   st->pipe->emit_string_marker(st->pipe, string, len);
+}
 
-void st_init_driver_functions(struct dd_function_table *functions)
+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);
 
-   functions->Accum = _mesa_accum;
-
    st_init_blit_functions(functions);
-   st_init_bufferobject_functions(functions);
+   st_init_bufferobject_functions(screen, functions);
    st_init_clear_functions(functions);
    st_init_bitmap_functions(functions);
+   st_init_copy_image_functions(functions);
    st_init_drawpixels_functions(functions);
    st_init_rasterpos_functions(functions);
 
@@ -305,18 +498,27 @@ void st_init_driver_functions(struct dd_function_table *functions)
 
    st_init_fbo_functions(functions);
    st_init_feedback_functions(functions);
+   st_init_msaa_functions(functions);
+   st_init_perfmon_functions(functions);
    st_init_program_functions(functions);
    st_init_query_functions(functions);
    st_init_cond_render_functions(functions);
    st_init_readpixels_functions(functions);
    st_init_texture_functions(functions);
    st_init_texture_barrier_functions(functions);
-   st_init_flush_functions(functions);
+   st_init_flush_functions(screen, functions);
    st_init_string_functions(functions);
    st_init_viewport_functions(functions);
 
    st_init_xformfb_functions(functions);
    st_init_syncobj_functions(functions);
 
+   st_init_vdpau_functions(functions);
+
+   if (screen->get_param(screen, PIPE_CAP_STRING_MARKER))
+      functions->EmitStringMarker = st_emit_string_marker;
+
+   functions->Enable = st_Enable;
    functions->UpdateState = st_invalidate_state;
+   functions->QueryMemoryInfo = st_query_memory_info;
 }