intel: Add performance debug for some common GPU stalls.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

src/mesa/drivers/dri/i965/brw_queryobj.c

@@ -139,6 +139,12 @@ brw_queryobj_get_results(struct gl_context *ctx,
    if (query->bo == NULL)
       return;
 
+   if (unlikely(INTEL_DEBUG & DEBUG_PERF)) {
+      if (drm_intel_bo_busy(query->bo)) {
+         perf_debug("Stalling on the GPU waiting for a query object.\n");
+      }
+   }
+
    drm_intel_bo_map(query->bo, false);
    results = query->bo->virtual;
    switch (query->Base.Target) {

src/mesa/drivers/dri/intel/intel_buffer_objects.c

@@ -212,7 +212,8 @@ intel_bufferobj_subdata(struct gl_context * ctx,
 	 intel_bufferobj_alloc_buffer(intel, intel_obj);
 	 drm_intel_bo_subdata(intel_obj->buffer, 0, size, data);
       } else {
-	 /* Use the blitter to upload the new data. */
+         perf_debug("Using a blit copy to avoid stalling on glBufferSubData() "
+                    "to a busy buffer object.\n");
 	 drm_intel_bo *temp_bo =
 	    drm_intel_bo_alloc(intel->bufmgr, "subdata temp", size, 64);
 
@@ -226,6 +227,11 @@ intel_bufferobj_subdata(struct gl_context * ctx,
 	 drm_intel_bo_unreference(temp_bo);
       }
    } else {
+      if (unlikely(INTEL_DEBUG & DEBUG_PERF)) {
+         if (drm_intel_bo_busy(intel_obj->buffer)) {
+            perf_debug("Stalling on the GPU in glBufferSubData().\n");
+         }
+      }
       drm_intel_bo_subdata(intel_obj->buffer, offset, size, data);
    }
 }

src/mesa/drivers/dri/intel/intel_regions.c

@@ -123,6 +123,12 @@ intel_region_map(struct intel_context *intel, struct intel_region *region,
     * flush is only needed on first map of the buffer.
     */
 
+   if (unlikely(INTEL_DEBUG & DEBUG_PERF)) {
+      if (drm_intel_bo_busy(region->bo)) {
+         perf_debug("Mapping a busy BO, causing a stall on the GPU.\n");
+      }
+   }
+
    _DBG("%s %p\n", __FUNCTION__, region);
    if (!region->map_refcount) {
       intel_flush(&intel->ctx);