i965: Add support for EXT_timer_query on Ironlake.

We could potentially do this on G45 as well, though the units are
different.  On 965, the timestamp is tied to hclk, which would make
supporting it harder.

3 files changed, 68 insertions, 24 deletions

diff --git a/docs/relnotes-7.9.html b/docs/relnotes-7.9.html
index f7d5016dc6..6122f8c34e 100644
--- a/docs/relnotes-7.9.html
+++ b/docs/relnotes-7.9.html
@@ -34,6 +34,7 @@ tbd
 
 <h2>New features</h2>
 <ul>
+<li>GL_EXT_timer_query extension (i965 driver only)
 </ul>
 
 
diff --git a/src/mesa/drivers/dri/i965/brw_queryobj.c b/src/mesa/drivers/dri/i965/brw_queryobj.c
index 3f47a68049..7cb812b521 100644
--- a/src/mesa/drivers/dri/i965/brw_queryobj.c
+++ b/src/mesa/drivers/dri/i965/brw_queryobj.c
@@ -55,11 +55,15 @@ brw_queryobj_get_results(struct brw_query_object *query)
    if (query->bo == NULL)
       return;
 
-   /* Map and count the pixels from the current query BO */
    dri_bo_map(query->bo, GL_FALSE);
    results = query->bo->virtual;
-   for (i = query->first_index; i <= query->last_index; i++) {
-      query->Base.Result += results[i * 2 + 1] - results[i * 2];
+   if (query->Base.Target == GL_TIME_ELAPSED_EXT) {
+      query->Base.Result += 1000 * ((results[1] >> 32) - (results[0] >> 32));
+   } else {
+      /* Map and count the pixels from the current query BO */
+      for (i = query->first_index; i <= query->last_index; i++) {
+	 query->Base.Result += results[i * 2 + 1] - results[i * 2];
+      }
    }
    dri_bo_unmap(query->bo);
 
@@ -98,14 +102,31 @@ brw_begin_query(GLcontext *ctx, struct gl_query_object *q)
    struct intel_context *intel = intel_context(ctx);
    struct brw_query_object *query = (struct brw_query_object *)q;
 
-   /* Reset our driver's tracking of query state. */
-   dri_bo_unreference(query->bo);
-   query->bo = NULL;
-   query->first_index = -1;
-   query->last_index = -1;
-
-   brw->query.obj = query;
-   intel->stats_wm++;
+   if (query->Base.Target == GL_TIME_ELAPSED_EXT) {
+      dri_bo_unreference(query->bo);
+      query->bo = drm_intel_bo_alloc(intel->bufmgr, "timer query",
+				     4096, 4096);
+
+      BEGIN_BATCH(4);
+      OUT_BATCH(_3DSTATE_PIPE_CONTROL |
+		PIPE_CONTROL_WRITE_TIMESTAMP);
+      OUT_RELOC(query->bo,
+		I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
+		PIPE_CONTROL_GLOBAL_GTT_WRITE |
+		0);
+      OUT_BATCH(0);
+      OUT_BATCH(0);
+      ADVANCE_BATCH();
+   } else {
+      /* Reset our driver's tracking of query state. */
+      dri_bo_unreference(query->bo);
+      query->bo = NULL;
+      query->first_index = -1;
+      query->last_index = -1;
+
+      brw->query.obj = query;
+      intel->stats_wm++;
+   }
 }
 
 /**
@@ -118,21 +139,36 @@ brw_end_query(GLcontext *ctx, struct gl_query_object *q)
    struct intel_context *intel = intel_context(ctx);
    struct brw_query_object *query = (struct brw_query_object *)q;
 
-   /* Flush the batchbuffer in case it has writes to our query BO.
-    * Have later queries write to a new query BO so that further rendering
-    * doesn't delay the collection of our results.
-    */
-   if (query->bo) {
-      brw_emit_query_end(brw);
-      intel_batchbuffer_flush(intel->batch);
+   if (query->Base.Target == GL_TIME_ELAPSED_EXT) {
+      BEGIN_BATCH(4);
+      OUT_BATCH(_3DSTATE_PIPE_CONTROL |
+		PIPE_CONTROL_WRITE_TIMESTAMP);
+      OUT_RELOC(query->bo,
+		I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
+		PIPE_CONTROL_GLOBAL_GTT_WRITE |
+		8);
+      OUT_BATCH(0);
+      OUT_BATCH(0);
+      ADVANCE_BATCH();
 
-      dri_bo_unreference(brw->query.bo);
-      brw->query.bo = NULL;
+      intel_batchbuffer_flush(intel->batch);
+   } else {
+      /* Flush the batchbuffer in case it has writes to our query BO.
+       * Have later queries write to a new query BO so that further rendering
+       * doesn't delay the collection of our results.
+       */
+      if (query->bo) {
+	 brw_emit_query_end(brw);
+	 intel_batchbuffer_flush(intel->batch);
+
+	 dri_bo_unreference(brw->query.bo);
+	 brw->query.bo = NULL;
+      }
+
+      brw->query.obj = NULL;
+
+      intel->stats_wm--;
    }
-
-   brw->query.obj = NULL;
-
-   intel->stats_wm--;
 }
 
 static void brw_wait_query(GLcontext *ctx, struct gl_query_object *q)
diff --git a/src/mesa/drivers/dri/intel/intel_extensions.c b/src/mesa/drivers/dri/intel/intel_extensions.c
index 9c2083873a..edba1fc2f2 100644
--- a/src/mesa/drivers/dri/intel/intel_extensions.c
+++ b/src/mesa/drivers/dri/intel/intel_extensions.c
@@ -57,6 +57,7 @@
 #define need_GL_EXT_provoking_vertex
 #define need_GL_EXT_secondary_color
 #define need_GL_EXT_stencil_two_side
+#define need_GL_EXT_timer_query
 #define need_GL_APPLE_vertex_array_object
 #define need_GL_APPLE_object_purgeable
 #define need_GL_ATI_separate_stencil
@@ -182,6 +183,9 @@ static const struct dri_extension brw_extensions[] = {
    { NULL,                                NULL }
 };
 
+static const struct dri_extension ironlake_extensions[] = {
+   { "GL_EXT_timer_query",                GL_EXT_timer_query_functions },
+};
 
 static const struct dri_extension arb_oq_extensions[] = {
    { "GL_ARB_occlusion_query",            GL_ARB_occlusion_query_functions },
@@ -207,6 +211,9 @@ intelInitExtensions(GLcontext *ctx)
     */
    driInitExtensions(ctx, card_extensions, GL_FALSE);
 
+   if (intel->gen >= 5)
+      driInitExtensions(ctx, ironlake_extensions, GL_FALSE);
+
    if (intel->gen >= 4)
       driInitExtensions(ctx, brw_extensions, GL_FALSE);