intel: Use a CPU map of the batch on LLC-sharing architectures.

Before, we were keeping a CPU-only buffer to accumulate the batchbuffer in,
which was an improvement over mapping the batch through the GTT directly
(since any readback or other failure to stream through write combining
correctly would hurt).  However, on LLC-sharing architectures we can do better
by mapping the batch directly, which reduces the cache footprint of the
application since we no longer have this extra copy of a batchbuffer around.

Improves performance of GLBenchmark 2.1 offscreen on IVB by 3.5% +/- 0.4%
(n=21).  Improves Lightsmark performance by 1.1 +/- 0.1% (n=76).  Improves
cairo-gl performance by 1.9% +/- 1.4% (n=57).

No statistically significant difference in GLB2.1 on SNB (n=37).  Improves
cairo-gl performance by 2.1% +/- 0.1% (n=278).

4 files changed, 24 insertions, 9 deletions

diff --git a/src/mesa/drivers/dri/intel/intel_batchbuffer.c b/src/mesa/drivers/dri/intel/intel_batchbuffer.c
index d36dacc6109..8c6524e71af 100644
--- a/src/mesa/drivers/dri/intel/intel_batchbuffer.c
+++ b/src/mesa/drivers/dri/intel/intel_batchbuffer.c
@@ -70,2 +70,7 @@ intel_batchbuffer_init(struct intel_context *intel)
    }
+
+   if (!intel->has_llc) {
+      intel->batch.cpu_map = malloc(intel->maxBatchSize);
+      intel->batch.map = intel->batch.cpu_map;
+   }
 }
@@ -85,2 +90,6 @@ intel_batchbuffer_reset(struct intel_context *intel)
 					intel->maxBatchSize, 4096);
+   if (intel->has_llc) {
+      drm_intel_bo_map(intel->batch.bo, true);
+      intel->batch.map = intel->batch.bo->virtual;
+   }
 
@@ -116,2 +125,3 @@ intel_batchbuffer_free(struct intel_context *intel)
 {
+   free(intel->batch.cpu_map);
    drm_intel_bo_unreference(intel->batch.last_bo);
@@ -170,8 +180,12 @@ do_flush_locked(struct intel_context *intel)
 
-   ret = drm_intel_bo_subdata(batch->bo, 0, 4*batch->used, batch->map);
-   if (ret == 0 && batch->state_batch_offset != batch->bo->size) {
-      ret = drm_intel_bo_subdata(batch->bo,
-				 batch->state_batch_offset,
-				 batch->bo->size - batch->state_batch_offset,
-				 (char *)batch->map + batch->state_batch_offset);
+   if (intel->has_llc) {
+      drm_intel_bo_unmap(batch->bo);
+   } else {
+      ret = drm_intel_bo_subdata(batch->bo, 0, 4*batch->used, batch->map);
+      if (ret == 0 && batch->state_batch_offset != batch->bo->size) {
+	 ret = drm_intel_bo_subdata(batch->bo,
+				    batch->state_batch_offset,
+				    batch->bo->size - batch->state_batch_offset,
+				    (char *)batch->map + batch->state_batch_offset);
+      }
    }
diff --git a/src/mesa/drivers/dri/intel/intel_batchbuffer.h b/src/mesa/drivers/dri/intel/intel_batchbuffer.h
index bae65553d08..39e7d26851d 100644
--- a/src/mesa/drivers/dri/intel/intel_batchbuffer.h
+++ b/src/mesa/drivers/dri/intel/intel_batchbuffer.h
@@ -114,3 +114,3 @@ intel_batchbuffer_require_space(struct intel_context *intel,
 #ifdef DEBUG
-   assert(sz < sizeof(intel->batch.map) - BATCH_RESERVED);
+   assert(sz < intel->maxBatchSize - BATCH_RESERVED);
 #endif
diff --git a/src/mesa/drivers/dri/intel/intel_context.c b/src/mesa/drivers/dri/intel/intel_context.c
index 3aa35e6d7f5..39460334b43 100644
--- a/src/mesa/drivers/dri/intel/intel_context.c
+++ b/src/mesa/drivers/dri/intel/intel_context.c
@@ -710,3 +710,3 @@ intelInitContext(struct intel_context *intel,
    else
-      intel->maxBatchSize = sizeof(intel->batch.map);
+      intel->maxBatchSize = BATCH_SZ;
 
diff --git a/src/mesa/drivers/dri/intel/intel_context.h b/src/mesa/drivers/dri/intel/intel_context.h
index 80e4cac131d..af49ab137c3 100644
--- a/src/mesa/drivers/dri/intel/intel_context.h
+++ b/src/mesa/drivers/dri/intel/intel_context.h
@@ -131,3 +131,4 @@ struct intel_batchbuffer {
    uint16_t used, reserved_space;
-   uint32_t map[8192];
+   uint32_t *map;
+   uint32_t *cpu_map;
 #define BATCH_SZ (8192*sizeof(uint32_t))