anv: Don't place scratch buffers above the 32-bit boundary

[mesa.git] / src / intel / vulkan / genX_blorp_exec.c

diff --git a/src/intel/vulkan/genX_blorp_exec.c b/src/intel/vulkan/genX_blorp_exec.c
index b617e781573f856944365ed491544a8f34c9cb0f..7f22b677ea69a38ce79df625ef8292d5023fce49 100644 (file)
--- a/src/intel/vulkan/genX_blorp_exec.c
+++ b/src/intel/vulkan/genX_blorp_exec.c
@@ -57,13 +57,15 @@ blorp_surface_reloc(struct blorp_batch *batch, uint32_t ss_offset,
                     struct blorp_address address, uint32_t delta)
 {
    struct anv_cmd_buffer *cmd_buffer = batch->driver_batch;
-   anv_reloc_list_add(&cmd_buffer->surface_relocs, &cmd_buffer->pool->alloc,
-                      ss_offset, address.buffer, address.offset + delta);
+   VkResult result =
+      anv_reloc_list_add(&cmd_buffer->surface_relocs, &cmd_buffer->pool->alloc,
+                         ss_offset, address.buffer, address.offset + delta);
+   if (result != VK_SUCCESS)
+      anv_batch_set_error(&cmd_buffer->batch, result);
 }
 
 static void *
 blorp_alloc_dynamic_state(struct blorp_batch *batch,
-                          enum aub_state_struct_type type,
                           uint32_t size,
                           uint32_t alignment,
                           uint32_t *offset)
@@ -86,9 +88,13 @@ blorp_alloc_binding_table(struct blorp_batch *batch, unsigned num_entries,
    struct anv_cmd_buffer *cmd_buffer = batch->driver_batch;
 
    uint32_t state_offset;
-   struct anv_state bt_state =
+   struct anv_state bt_state;
+
+   VkResult result =
       anv_cmd_buffer_alloc_blorp_binding_table(cmd_buffer, num_entries,
-                                               &state_offset);
+                                               &state_offset, &bt_state);
+   if (result != VK_SUCCESS)
+      return;
 
    uint32_t *bt_map = bt_state.map;
    *bt_offset = bt_state.offset;
@@ -101,8 +107,7 @@ blorp_alloc_binding_table(struct blorp_batch *batch, unsigned num_entries,
       surface_maps[i] = surface_state.map;
    }
 
-   if (!cmd_buffer->device->info.has_llc)
-      anv_state_flush(bt_state);
+   anv_state_flush(cmd_buffer->device, bt_state);
 }
 
 static void *
@@ -110,8 +115,21 @@ blorp_alloc_vertex_buffer(struct blorp_batch *batch, uint32_t size,
                           struct blorp_address *addr)
 {
    struct anv_cmd_buffer *cmd_buffer = batch->driver_batch;
+
+   /* From the Skylake PRM, 3DSTATE_VERTEX_BUFFERS:
+    *
+    *    "The VF cache needs to be invalidated before binding and then using
+    *    Vertex Buffers that overlap with any previously bound Vertex Buffer
+    *    (at a 64B granularity) since the last invalidation.  A VF cache
+    *    invalidate is performed by setting the "VF Cache Invalidation Enable"
+    *    bit in PIPE_CONTROL."
+    *
+    * This restriction first appears in the Skylake PRM but the internal docs
+    * also list it as being an issue on Broadwell.  In order to avoid this
+    * problem, we align all vertex buffer allocations to 64 bytes.
+    */
    struct anv_state vb_state =
-      anv_cmd_buffer_alloc_dynamic_state(cmd_buffer, size, 16);
+      anv_cmd_buffer_alloc_dynamic_state(cmd_buffer, size, 64);
 
    *addr = (struct blorp_address) {
       .buffer = &cmd_buffer->device->dynamic_state_block_pool.bo,
@@ -144,9 +162,6 @@ blorp_emit_urb_config(struct blorp_batch *batch, unsigned vs_entry_size)
                         entry_size);
 }
 
-void genX(blorp_exec)(struct blorp_batch *batch,
-                      const struct blorp_params *params);
-
 void
 genX(blorp_exec)(struct blorp_batch *batch,
                  const struct blorp_params *params)
@@ -170,6 +185,11 @@ genX(blorp_exec)(struct blorp_batch *batch,
     */
    genX(cmd_buffer_enable_pma_fix)(cmd_buffer, false);
 
+   /* Disable VF statistics */
+   blorp_emit(batch, GENX(3DSTATE_VF_STATISTICS), vf) {
+      vf.StatisticsEnable = false;
+   }
+
    blorp_exec(batch, params);
 
    cmd_buffer->state.vb_dirty = ~0;