1 files changed, 44 insertions, 50 deletions

diff --git a/src/mesa/drivers/dri/i965/brw_vs.c b/src/mesa/drivers/dri/i965/brw_vs.c
index c53cb49b612..0c60bde511e 100644
--- a/src/mesa/drivers/dri/i965/brw_vs.c
+++ b/src/mesa/drivers/dri/i965/brw_vs.c
@@ -98,10 +98,12 @@ brw_codegen_vs_prog(struct brw_context *brw,
    struct brw_stage_prog_data *stage_prog_data = &prog_data.base.base;
    void *mem_ctx;
    int i;
-   struct gl_shader *vs = NULL;
+   struct brw_shader *vs = NULL;
+   bool start_busy = false;
+   double start_time = 0;
 
    if (prog)
-      vs = prog->_LinkedShaders[MESA_SHADER_VERTEX];
+      vs = (struct brw_shader *) prog->_LinkedShaders[MESA_SHADER_VERTEX];
 
    memset(&prog_data, 0, sizeof(prog_data));
 
@@ -121,16 +123,16 @@ brw_codegen_vs_prog(struct brw_context *brw,
        * case being a float value that gets blown up to a vec4, so be
        * conservative here.
        */
-      param_count = vs->num_uniform_components * 4 +
-                    vs->NumImages * BRW_IMAGE_PARAM_SIZE;
-      stage_prog_data->nr_image_params = vs->NumImages;
+      param_count = vs->base.num_uniform_components * 4 +
+                    vs->base.NumImages * BRW_IMAGE_PARAM_SIZE;
+      stage_prog_data->nr_image_params = vs->base.NumImages;
    } else {
       param_count = vp->program.Base.Parameters->NumParameters * 4;
    }
    /* vec4_visitor::setup_uniform_clipplane_values() also uploads user clip
     * planes as uniforms.
     */
-   param_count += key->base.nr_userclip_plane_consts * 4;
+   param_count += key->nr_userclip_plane_consts * 4;
 
    stage_prog_data->param =
       rzalloc_array(NULL, const gl_constant_value *, param_count);
@@ -172,19 +174,26 @@ brw_codegen_vs_prog(struct brw_context *brw,
     * distance varying slots whenever clipping is enabled, even if the vertex
     * shader doesn't write to gl_ClipDistance.
     */
-   if (key->base.userclip_active) {
+   if (key->nr_userclip_plane_consts > 0) {
       outputs_written |= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST0);
       outputs_written |= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST1);
    }
 
    brw_compute_vue_map(brw->intelScreen->devinfo,
-                       &prog_data.base.vue_map, outputs_written);
+                       &prog_data.base.vue_map, outputs_written,
+                       prog ? prog->SeparateShader : false);
 
    if (0) {
       _mesa_fprint_program_opt(stderr, &vp->program.Base, PROG_PRINT_DEBUG,
 			       true);
    }
 
+   if (unlikely(brw->perf_debug)) {
+      start_busy = (brw->batch.last_bo &&
+                    drm_intel_bo_busy(brw->batch.last_bo));
+      start_time = get_time();
+   }
+
    /* Emit GEN4 code.
     */
    program = brw_vs_emit(brw, mem_ctx, key, &prog_data,
@@ -194,6 +203,17 @@ brw_codegen_vs_prog(struct brw_context *brw,
       return false;
    }
 
+   if (unlikely(brw->perf_debug) && vs) {
+      if (vs->compiled_once) {
+         brw_vs_debug_recompile(brw, prog, key);
+      }
+      if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
+         perf_debug("VS compile took %.03f ms and stalled the GPU\n",
+                    (get_time() - start_time) * 1000);
+      }
+      vs->compiled_once = true;
+   }
+
    /* Scratch space is used for register spilling */
    if (prog_data.base.base.total_scratch) {
       brw_get_scratch_bo(brw, &brw->vs.base.scratch_bo,
@@ -237,7 +257,7 @@ brw_vs_debug_recompile(struct brw_context *brw,
          if (c->cache_id == BRW_CACHE_VS_PROG) {
             old_key = c->key;
 
-            if (old_key->base.program_string_id == key->base.program_string_id)
+            if (old_key->program_string_id == key->program_string_id)
                break;
          }
       }
@@ -257,12 +277,9 @@ brw_vs_debug_recompile(struct brw_context *brw,
                          key->gl_attrib_wa_flags[i]);
    }
 
-   found |= key_debug(brw, "user clip flags",
-                      old_key->base.userclip_active, key->base.userclip_active);
-
-   found |= key_debug(brw, "user clipping planes as push constants",
-                      old_key->base.nr_userclip_plane_consts,
-                      key->base.nr_userclip_plane_consts);
+   found |= key_debug(brw, "legacy user clipping",
+                      old_key->nr_userclip_plane_consts,
+                      key->nr_userclip_plane_consts);
 
    found |= key_debug(brw, "copy edgeflag",
                       old_key->copy_edgeflag, key->copy_edgeflag);
@@ -271,29 +288,13 @@ brw_vs_debug_recompile(struct brw_context *brw,
    found |= key_debug(brw, "vertex color clamping",
                       old_key->clamp_vertex_color, key->clamp_vertex_color);
 
-   found |= brw_debug_recompile_sampler_key(brw, &old_key->base.tex,
-                                            &key->base.tex);
+   found |= brw_debug_recompile_sampler_key(brw, &old_key->tex, &key->tex);
 
    if (!found) {
       perf_debug("  Something else\n");
    }
 }
 
-
-void
-brw_setup_vue_key_clip_info(struct brw_context *brw,
-                            struct brw_vue_prog_key *key,
-                            bool program_uses_clip_distance)
-{
-   struct gl_context *ctx = &brw->ctx;
-
-   key->userclip_active = (ctx->Transform.ClipPlanesEnabled != 0);
-   if (key->userclip_active && !program_uses_clip_distance) {
-      key->nr_userclip_plane_consts
-         = _mesa_logbase2(ctx->Transform.ClipPlanesEnabled) + 1;
-   }
-}
-
 static bool
 brw_vs_state_dirty(struct brw_context *brw)
 {
@@ -324,9 +325,14 @@ brw_vs_populate_key(struct brw_context *brw,
    /* Just upload the program verbatim for now.  Always send it all
     * the inputs it asks for, whether they are varying or not.
     */
-   key->base.program_string_id = vp->id;
-   brw_setup_vue_key_clip_info(brw, &key->base,
-                               vp->program.Base.UsesClipDistanceOut);
+   key->program_string_id = vp->id;
+
+   if (ctx->Transform.ClipPlanesEnabled != 0 &&
+       ctx->API == API_OPENGL_COMPAT &&
+       !vp->program.Base.UsesClipDistanceOut) {
+      key->nr_userclip_plane_consts =
+         _mesa_logbase2(ctx->Transform.ClipPlanesEnabled) + 1;
+   }
 
    /* _NEW_POLYGON */
    if (brw->gen < 6) {
@@ -350,7 +356,7 @@ brw_vs_populate_key(struct brw_context *brw,
 
    /* _NEW_TEXTURE */
    brw_populate_sampler_prog_key_data(ctx, prog, brw->vs.base.sampler_count,
-                                      &key->base.tex);
+                                      &key->tex);
 
    /* BRW_NEW_VS_ATTRIB_WORKAROUNDS */
    memcpy(key->gl_attrib_wa_flags, brw->vb.attrib_wa_flags,
@@ -381,19 +387,6 @@ brw_upload_vs_prog(struct brw_context *brw)
       assert(success);
    }
    brw->vs.base.prog_data = &brw->vs.prog_data->base.base;
-
-   if (memcmp(&brw->vs.prog_data->base.vue_map, &brw->vue_map_geom_out,
-              sizeof(brw->vue_map_geom_out)) != 0) {
-      brw->vue_map_vs = brw->vs.prog_data->base.vue_map;
-      brw->ctx.NewDriverState |= BRW_NEW_VUE_MAP_VS;
-      if (brw->gen < 6) {
-         /* No geometry shader support, so the VS VUE map is the VUE map for
-          * the output of the "geometry" portion of the pipeline.
-          */
-         brw->vue_map_geom_out = brw->vue_map_vs;
-         brw->ctx.NewDriverState |= BRW_NEW_VUE_MAP_GEOM_OUT;
-      }
-   }
 }
 
 bool
@@ -412,7 +405,8 @@ brw_vs_precompile(struct gl_context *ctx,
 
    memset(&key, 0, sizeof(key));
 
-   brw_vue_setup_prog_key_for_precompile(ctx, &key.base, bvp->id, &vp->Base);
+   brw_setup_tex_for_precompile(brw, &key.tex, prog);
+   key.program_string_id = bvp->id;
    key.clamp_vertex_color =
       (prog->OutputsWritten & (VARYING_BIT_COL0 | VARYING_BIT_COL1 |
                                VARYING_BIT_BFC0 | VARYING_BIT_BFC1));