anv: implement Wa_14015814527 for task shaders

After using task shader, we need to emit a zero URB state and a
nullprim (empty pipe control) before rendering with primitives.

After this, a normal URB state needs to be returned, this will
happen when pipeline batch is emitted during pipeline switch.

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20334>

src/intel/vulkan/anv_genX.h

@@ -83,6 +83,8 @@ void genX(flush_pipeline_select_3d)(struct anv_cmd_buffer *cmd_buffer);
 void genX(flush_pipeline_select_gpgpu)(struct anv_cmd_buffer *cmd_buffer);
 void genX(emit_pipeline_select)(struct anv_batch *batch, uint32_t pipeline);
 
+void genX(apply_task_urb_workaround)(struct anv_cmd_buffer *cmd_buffer);
+
 enum anv_pipe_bits
 genX(emit_apply_pipe_flushes)(struct anv_batch *batch,
                               struct anv_device *device,

src/intel/vulkan/anv_private.h

@@ -2571,6 +2571,7 @@ struct anv_cmd_graphics_state {
    VkShaderStageFlags push_constant_stages;
 
    uint32_t primitive_topology;
+   bool used_task_shader;
 
    struct anv_buffer *index_buffer;
    uint32_t index_type; /**< 3DSTATE_INDEX_BUFFER.IndexFormat */

src/intel/vulkan/genX_blorp_exec.c

@@ -294,6 +294,9 @@ blorp_exec_on_render(struct blorp_batch *batch,
 
    genX(flush_pipeline_select_3d)(cmd_buffer);
 
+   /* Wa_14015814527 */
+   genX(apply_task_urb_workaround)(cmd_buffer);
+
    /* Apply any outstanding flushes in case pipeline select haven't. */
    genX(cmd_buffer_apply_pipe_flushes)(cmd_buffer);
 

src/intel/vulkan/genX_cmd_buffer.c

@@ -3244,6 +3244,19 @@ genX(cmd_buffer_flush_gfx_state)(struct anv_cmd_buffer *cmd_buffer)
 
    genX(flush_pipeline_select_3d)(cmd_buffer);
 
+   /* Wa_14015814527
+    *
+    * Apply task URB workaround when switching from task to primitive.
+    */
+   if (cmd_buffer->state.gfx.dirty & ANV_CMD_DIRTY_PIPELINE) {
+      if (anv_pipeline_is_primitive(pipeline)) {
+         genX(apply_task_urb_workaround)(cmd_buffer);
+      } else if (anv_pipeline_has_stage(cmd_buffer->state.gfx.pipeline,
+                                        MESA_SHADER_TASK)) {
+         cmd_buffer->state.gfx.used_task_shader = true;
+      }
+   }
+
    /* Apply any pending pipeline flushes we may have.  We want to apply them
     * now because, if any of those flushes are for things like push constants,
     * the GPU will read the state at weird times.
@@ -3696,6 +3709,12 @@ genX(EndCommandBuffer)(
     */
    genX(cmd_buffer_enable_pma_fix)(cmd_buffer, false);
 
+   /* Wa_14015814527
+    *
+    * Apply task URB workaround in the end of primary or secondary cmd_buffer.
+    */
+   genX(apply_task_urb_workaround)(cmd_buffer);
+
    genX(cmd_buffer_apply_pipe_flushes)(cmd_buffer);
 
    emit_isp_disable(cmd_buffer);
@@ -3729,6 +3748,12 @@ genX(CmdExecuteCommands)(
    if (!primary->state.gfx.object_preemption)
       genX(cmd_buffer_set_preemption)(primary, true);
 
+   /* Wa_14015814527
+    *
+    * Apply task URB workaround before secondary cmd buffers.
+    */
+   genX(apply_task_urb_workaround)(primary);
+
    /* The secondary command buffer doesn't know which textures etc. have been
     * flushed prior to their execution.  Apply those flushes now.
     */

src/intel/vulkan/genX_gpu_memcpy.c

@@ -298,6 +298,10 @@ genX(cmd_buffer_so_memcpy)(struct anv_cmd_buffer *cmd_buffer,
 #if GFX_VER == 9
    genX(cmd_buffer_set_binding_for_gfx8_vb_flush)(cmd_buffer, 32, src, size);
 #endif
+
+   /* Wa_14015814527 */
+   genX(apply_task_urb_workaround)(cmd_buffer);
+
    genX(cmd_buffer_apply_pipe_flushes)(cmd_buffer);
 
    genX(flush_pipeline_select_3d)(cmd_buffer);

src/intel/vulkan/genX_state.c

@@ -995,3 +995,44 @@ VkResult genX(CreateSampler)(
 
    return VK_SUCCESS;
 }
+
+/* Wa_14015814527
+ *
+ * Check if task shader was utilized within cmd_buffer, if so
+ * commit empty URB states and null prim.
+ */
+void
+genX(apply_task_urb_workaround)(struct anv_cmd_buffer *cmd_buffer)
+{
+#if GFX_VERx10 != 125
+   return;
+#else
+   if (cmd_buffer->state.current_pipeline != _3D ||
+       !cmd_buffer->state.gfx.used_task_shader)
+      return;
+
+   cmd_buffer->state.gfx.used_task_shader = false;
+
+   /* Wa_14015821291 mentions that WA below is not required if we have
+    * a pipeline flush going on. It will get flushed during
+    * cmd_buffer_flush_state before draw.
+    */
+   if ((cmd_buffer->state.pending_pipe_bits & ANV_PIPE_CS_STALL_BIT))
+      return;
+
+   for (int i = 0; i <= MESA_SHADER_GEOMETRY; i++) {
+      anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_URB_VS), urb) {
+         urb._3DCommandSubOpcode += i;
+      }
+   }
+
+   anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_URB_ALLOC_MESH), zero);
+   anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_URB_ALLOC_TASK), zero);
+
+   /* Issue 'nullprim' to commit the state. */
+   anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), pc) {
+      pc.PostSyncOperation = WriteImmediateData;
+      pc.Address = cmd_buffer->device->workaround_address;
+   }
+#endif
+}