SKL: correct the pipe control struct.

From BDW, pipe control need 6 DW, correct it. Also affect BDW.

Signed-off-by: Yang Rong <rong.r.yang@intel.com>

2 files changed, 77 insertions, 2 deletions

diff --git a/src/intel/intel_gpgpu.c b/src/intel/intel_gpgpu.c
index 4c095b9c..cd45ff9f 100644
--- a/src/intel/intel_gpgpu.c
+++ b/src/intel/intel_gpgpu.c
@@ -626,6 +626,25 @@ intel_gpgpu_pipe_control_gen75(intel_gpgpu_t *gpgpu)
 }
 
 static void
+intel_gpgpu_pipe_control_gen8(intel_gpgpu_t *gpgpu)
+{
+  gen8_pipe_control_t* pc = (gen8_pipe_control_t*)
+    intel_batchbuffer_alloc_space(gpgpu->batch, sizeof(gen8_pipe_control_t));
+  memset(pc, 0, sizeof(*pc));
+  pc->dw0.length = SIZEOF32(gen8_pipe_control_t) - 2;
+  pc->dw0.instruction_subopcode = GEN7_PIPE_CONTROL_SUBOPCODE_3D_CONTROL;
+  pc->dw0.instruction_opcode = GEN7_PIPE_CONTROL_OPCODE_3D_CONTROL;
+  pc->dw0.instruction_pipeline = GEN7_PIPE_CONTROL_3D;
+  pc->dw0.instruction_type = GEN7_PIPE_CONTROL_INSTRUCTION_GFX;
+  pc->dw1.render_target_cache_flush_enable = 1;
+  pc->dw1.texture_cache_invalidation_enable = 1;
+  pc->dw1.cs_stall = 1;
+  pc->dw1.dc_flush_enable = 1;
+  //pc->dw1.instruction_cache_invalidate_enable = 1;
+  ADVANCE_BATCH(gpgpu->batch);
+}
+
+static void
 intel_gpgpu_set_L3_gen7(intel_gpgpu_t *gpgpu, uint32_t use_slm)
 {
   BEGIN_BATCH(gpgpu->batch, 9);
@@ -2007,7 +2026,7 @@ intel_set_gpgpu_callbacks(int device_id)
     intel_gpgpu_load_curbe_buffer = intel_gpgpu_load_curbe_buffer_gen8;
     intel_gpgpu_load_idrt = intel_gpgpu_load_idrt_gen8;
     cl_gpgpu_bind_sampler = (cl_gpgpu_bind_sampler_cb *) intel_gpgpu_bind_sampler_gen8;
-    intel_gpgpu_pipe_control = intel_gpgpu_pipe_control_gen7;
+    intel_gpgpu_pipe_control = intel_gpgpu_pipe_control_gen8;
     return;
   }
   if (IS_SKYLAKE(device_id)) {
@@ -2025,7 +2044,7 @@ intel_set_gpgpu_callbacks(int device_id)
     intel_gpgpu_load_curbe_buffer = intel_gpgpu_load_curbe_buffer_gen8;
     intel_gpgpu_load_idrt = intel_gpgpu_load_idrt_gen8;
     cl_gpgpu_bind_sampler = (cl_gpgpu_bind_sampler_cb *) intel_gpgpu_bind_sampler_gen8;
-    intel_gpgpu_pipe_control = intel_gpgpu_pipe_control_gen7;
+    intel_gpgpu_pipe_control = intel_gpgpu_pipe_control_gen8;
     return;
   }
 
diff --git a/src/intel/intel_structs.h b/src/intel/intel_structs.h
index 258fbb9b..b4635f49 100644
--- a/src/intel/intel_structs.h
+++ b/src/intel/intel_structs.h
@@ -498,6 +498,62 @@ typedef struct gen6_pipe_control
   } dw4;
 } gen6_pipe_control_t;
 
+typedef struct gen8_pipe_control
+{
+  struct {
+    uint32_t length : BITFIELD_RANGE(0, 7);
+    uint32_t reserved : BITFIELD_RANGE(8, 15);
+    uint32_t instruction_subopcode : BITFIELD_RANGE(16, 23);
+    uint32_t instruction_opcode : BITFIELD_RANGE(24, 26);
+    uint32_t instruction_pipeline : BITFIELD_RANGE(27, 28);
+    uint32_t instruction_type : BITFIELD_RANGE(29, 31);
+  } dw0;
+
+  struct {
+    uint32_t depth_cache_flush_enable : BITFIELD_BIT(0);
+    uint32_t stall_at_pixel_scoreboard : BITFIELD_BIT(1);
+    uint32_t state_cache_invalidation_enable : BITFIELD_BIT(2);
+    uint32_t constant_cache_invalidation_enable : BITFIELD_BIT(3);
+    uint32_t vf_cache_invalidation_enable : BITFIELD_BIT(4);
+    uint32_t dc_flush_enable : BITFIELD_BIT(5);
+    uint32_t protected_memory_app_id : BITFIELD_BIT(6);
+    uint32_t pipe_control_flush_enable : BITFIELD_BIT(7);
+    uint32_t notify_enable : BITFIELD_BIT(8);
+    uint32_t indirect_state_pointers_disable : BITFIELD_BIT(9);
+    uint32_t texture_cache_invalidation_enable : BITFIELD_BIT(10);
+    uint32_t instruction_cache_invalidate_enable : BITFIELD_BIT(11);
+    uint32_t render_target_cache_flush_enable : BITFIELD_BIT(12);
+    uint32_t depth_stall_enable : BITFIELD_BIT(13);
+    uint32_t post_sync_operation : BITFIELD_RANGE(14, 15);
+    uint32_t generic_media_state_clear : BITFIELD_BIT(16);
+    uint32_t synchronize_gfdt_surface : BITFIELD_BIT(17);
+    uint32_t tlb_invalidate : BITFIELD_BIT(18);
+    uint32_t global_snapshot_count_reset : BITFIELD_BIT(19);
+    uint32_t cs_stall : BITFIELD_BIT(20);
+    uint32_t store_data_index : BITFIELD_BIT(21);
+    uint32_t protected_memory_enable : BITFIELD_BIT(22);
+    uint32_t reserved : BITFIELD_RANGE(23, 31);
+  } dw1;
+
+  struct {
+    uint32_t reserved : BITFIELD_RANGE(0, 1);
+    uint32_t destination_address_type : BITFIELD_BIT(2);
+    uint32_t address : BITFIELD_RANGE(3, 31);
+  } dw2;
+
+  struct {
+    uint32_t data;
+  } dw3;
+
+  struct {
+    uint32_t data;
+  } dw4;
+
+  struct {
+    uint32_t data;
+  } dw5;
+} gen8_pipe_control_t;
+
 typedef struct gen6_sampler_state
 {
   struct {