intel/fs: Define is_send() convenience IR helper.
[mesa.git] / src / intel / vulkan / anv_private.h
index 9e59d79de7acf3649532aa2df2813edb0f28efec..3aa6d1922f955c88682900c47467aa95c8ffb20a 100644 (file)
@@ -57,6 +57,7 @@
 #include "util/u_vector.h"
 #include "util/u_math.h"
 #include "util/vma.h"
+#include "util/xmlconfig.h"
 #include "vk_alloc.h"
 #include "vk_debug_report.h"
 
@@ -631,7 +632,10 @@ union anv_free_list {
       /* A simple count that is incremented every time the head changes. */
       uint32_t count;
    };
-   uint64_t u64;
+   /* Make sure it's aligned to 64 bits. This will make atomic operations
+    * faster on 32 bit platforms.
+    */
+   uint64_t u64 __attribute__ ((aligned (8)));
 };
 
 #define ANV_FREE_LIST_EMPTY ((union anv_free_list) { { UINT32_MAX, 0 } })
@@ -642,7 +646,10 @@ struct anv_block_state {
          uint32_t next;
          uint32_t end;
       };
-      uint64_t u64;
+      /* Make sure it's aligned to 64 bits. This will make atomic operations
+       * faster on 32 bit platforms.
+       */
+      uint64_t u64 __attribute__ ((aligned (8)));
    };
 };
 
@@ -729,7 +736,7 @@ struct anv_fixed_size_state_pool {
 };
 
 #define ANV_MIN_STATE_SIZE_LOG2 6
-#define ANV_MAX_STATE_SIZE_LOG2 20
+#define ANV_MAX_STATE_SIZE_LOG2 21
 
 #define ANV_STATE_BUCKETS (ANV_MAX_STATE_SIZE_LOG2 - ANV_MIN_STATE_SIZE_LOG2 + 1)
 
@@ -961,6 +968,7 @@ struct anv_physical_device {
     bool                                        has_bindless_samplers;
 
     struct anv_device_extension_table           supported_extensions;
+    struct anv_physical_device_dispatch_table   dispatch;
 
     uint32_t                                    eu_total;
     uint32_t                                    subslice_total;
@@ -1009,6 +1017,9 @@ struct anv_instance {
     bool                                        pipeline_cache_enabled;
 
     struct vk_debug_report_instance             debug_report_callbacks;
+
+    struct driOptionCache                       dri_options;
+    struct driOptionCache                       available_dri_options;
 };
 
 VkResult anv_init_wsi(struct anv_physical_device *physical_device);
@@ -1054,6 +1065,8 @@ anv_pipeline_cache_upload_kernel(struct anv_pipeline_cache *cache,
                                  uint32_t constant_data_size,
                                  const struct brw_stage_prog_data *prog_data,
                                  uint32_t prog_data_size,
+                                 const struct brw_compile_stats *stats,
+                                 uint32_t num_stats,
                                  const struct nir_xfb_info *xfb_info,
                                  const struct anv_pipeline_bind_map *bind_map);
 
@@ -1072,6 +1085,8 @@ anv_device_upload_kernel(struct anv_device *device,
                          uint32_t constant_data_size,
                          const struct brw_stage_prog_data *prog_data,
                          uint32_t prog_data_size,
+                         const struct brw_compile_stats *stats,
+                         uint32_t num_stats,
                          const struct nir_xfb_info *xfb_info,
                          const struct anv_pipeline_bind_map *bind_map);
 
@@ -1135,6 +1150,8 @@ struct anv_device {
 
     struct anv_state                            border_colors;
 
+    struct anv_state                            slice_hash;
+
     struct anv_queue                            queue;
 
     struct anv_scratch_pool                     scratch_pool;
@@ -1500,6 +1517,11 @@ _anv_combine_address(struct anv_batch *batch, void *location,
 #define GEN11_MOCS GEN9_MOCS
 #define GEN11_EXTERNAL_MOCS GEN9_EXTERNAL_MOCS
 
+/* TigerLake MOCS */
+#define GEN12_MOCS GEN9_MOCS
+/* TC=1/LLC Only, LeCC=1/Uncacheable, LRUM=0, L3CC=1/Uncacheable */
+#define GEN12_EXTERNAL_MOCS (3 << 1)
+
 struct anv_device_memory {
    struct list_head                             link;
 
@@ -1947,14 +1969,57 @@ enum anv_cmd_dirty_bits {
    ANV_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK      = 1 << 6, /* VK_DYNAMIC_STATE_STENCIL_COMPARE_MASK */
    ANV_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK        = 1 << 7, /* VK_DYNAMIC_STATE_STENCIL_WRITE_MASK */
    ANV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE         = 1 << 8, /* VK_DYNAMIC_STATE_STENCIL_REFERENCE */
-   ANV_CMD_DIRTY_DYNAMIC_ALL                       = (1 << 9) - 1,
    ANV_CMD_DIRTY_PIPELINE                          = 1 << 9,
    ANV_CMD_DIRTY_INDEX_BUFFER                      = 1 << 10,
    ANV_CMD_DIRTY_RENDER_TARGETS                    = 1 << 11,
    ANV_CMD_DIRTY_XFB_ENABLE                        = 1 << 12,
+   ANV_CMD_DIRTY_DYNAMIC_LINE_STIPPLE              = 1 << 13, /* VK_DYNAMIC_STATE_LINE_STIPPLE_EXT */
 };
 typedef uint32_t anv_cmd_dirty_mask_t;
 
+#define ANV_CMD_DIRTY_DYNAMIC_ALL                  \
+   (ANV_CMD_DIRTY_DYNAMIC_VIEWPORT |               \
+    ANV_CMD_DIRTY_DYNAMIC_SCISSOR |                \
+    ANV_CMD_DIRTY_DYNAMIC_LINE_WIDTH |             \
+    ANV_CMD_DIRTY_DYNAMIC_DEPTH_BIAS |             \
+    ANV_CMD_DIRTY_DYNAMIC_BLEND_CONSTANTS |        \
+    ANV_CMD_DIRTY_DYNAMIC_DEPTH_BOUNDS |           \
+    ANV_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK |   \
+    ANV_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK |     \
+    ANV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE |      \
+    ANV_CMD_DIRTY_DYNAMIC_LINE_STIPPLE)
+
+static inline enum anv_cmd_dirty_bits
+anv_cmd_dirty_bit_for_vk_dynamic_state(VkDynamicState vk_state)
+{
+   switch (vk_state) {
+   case VK_DYNAMIC_STATE_VIEWPORT:
+      return ANV_CMD_DIRTY_DYNAMIC_VIEWPORT;
+   case VK_DYNAMIC_STATE_SCISSOR:
+      return ANV_CMD_DIRTY_DYNAMIC_SCISSOR;
+   case VK_DYNAMIC_STATE_LINE_WIDTH:
+      return ANV_CMD_DIRTY_DYNAMIC_LINE_WIDTH;
+   case VK_DYNAMIC_STATE_DEPTH_BIAS:
+      return ANV_CMD_DIRTY_DYNAMIC_DEPTH_BIAS;
+   case VK_DYNAMIC_STATE_BLEND_CONSTANTS:
+      return ANV_CMD_DIRTY_DYNAMIC_BLEND_CONSTANTS;
+   case VK_DYNAMIC_STATE_DEPTH_BOUNDS:
+      return ANV_CMD_DIRTY_DYNAMIC_DEPTH_BOUNDS;
+   case VK_DYNAMIC_STATE_STENCIL_COMPARE_MASK:
+      return ANV_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK;
+   case VK_DYNAMIC_STATE_STENCIL_WRITE_MASK:
+      return ANV_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK;
+   case VK_DYNAMIC_STATE_STENCIL_REFERENCE:
+      return ANV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE;
+   case VK_DYNAMIC_STATE_LINE_STIPPLE_EXT:
+      return ANV_CMD_DIRTY_DYNAMIC_LINE_STIPPLE;
+   default:
+      assert(!"Unsupported dynamic state");
+      return 0;
+   }
+}
+
+
 enum anv_pipe_bits {
    ANV_PIPE_DEPTH_CACHE_FLUSH_BIT            = (1 << 0),
    ANV_PIPE_STALL_AT_SCOREBOARD_BIT          = (1 << 1),
@@ -2217,13 +2282,18 @@ struct anv_dynamic_state {
       uint32_t                                  front;
       uint32_t                                  back;
    } stencil_reference;
+
+   struct {
+      uint32_t                                  factor;
+      uint16_t                                  pattern;
+   } line_stipple;
 };
 
 extern const struct anv_dynamic_state default_dynamic_state;
 
-void anv_dynamic_state_copy(struct anv_dynamic_state *dest,
-                            const struct anv_dynamic_state *src,
-                            uint32_t copy_mask);
+uint32_t anv_dynamic_state_copy(struct anv_dynamic_state *dest,
+                                const struct anv_dynamic_state *src,
+                                uint32_t copy_mask);
 
 struct anv_surface_state {
    struct anv_state state;
@@ -2273,6 +2343,7 @@ struct anv_attachment_state {
     * have not been cleared yet when multiview is active.
     */
    uint32_t                                     pending_clear_views;
+   struct anv_image_view *                      image_view;
 };
 
 /** State tracking for particular pipeline bind point
@@ -2372,6 +2443,12 @@ struct anv_cmd_state {
 
    bool                                         conditional_render_enabled;
 
+   /**
+    * Last rendering scale argument provided to
+    * genX(cmd_buffer_emit_hashing_mode)().
+    */
+   unsigned                                     current_hash_scale;
+
    /**
     * Array length is anv_cmd_state::pass::attachment_count. Array content is
     * valid only when recording a render pass instance.
@@ -2698,6 +2775,9 @@ struct anv_shader_bin {
    const struct brw_stage_prog_data *prog_data;
    uint32_t prog_data_size;
 
+   struct brw_compile_stats stats[3];
+   uint32_t num_stats;
+
    struct nir_xfb_info *xfb_info;
 
    struct anv_pipeline_bind_map bind_map;
@@ -2710,6 +2790,7 @@ anv_shader_bin_create(struct anv_device *device,
                       const void *constant_data, uint32_t constant_data_size,
                       const struct brw_stage_prog_data *prog_data,
                       uint32_t prog_data_size, const void *prog_data_param,
+                      const struct brw_compile_stats *stats, uint32_t num_stats,
                       const struct nir_xfb_info *xfb_info,
                       const struct anv_pipeline_bind_map *bind_map);
 
@@ -2731,20 +2812,38 @@ anv_shader_bin_unref(struct anv_device *device, struct anv_shader_bin *shader)
       anv_shader_bin_destroy(device, shader);
 }
 
+/* 5 possible simultaneous shader stages and FS may have up to 3 binaries */
+#define MAX_PIPELINE_EXECUTABLES 7
+
+struct anv_pipeline_executable {
+   gl_shader_stage stage;
+
+   struct brw_compile_stats stats;
+
+   char *nir;
+   char *disasm;
+};
+
 struct anv_pipeline {
    struct anv_device *                          device;
    struct anv_batch                             batch;
    uint32_t                                     batch_data[512];
    struct anv_reloc_list                        batch_relocs;
-   uint32_t                                     dynamic_state_mask;
+   anv_cmd_dirty_mask_t                         dynamic_state_mask;
    struct anv_dynamic_state                     dynamic_state;
 
+   void *                                       mem_ctx;
+
+   VkPipelineCreateFlags                        flags;
    struct anv_subpass *                         subpass;
 
    bool                                         needs_data_cache;
 
    struct anv_shader_bin *                      shaders[MESA_SHADER_STAGES];
 
+   uint32_t                                     num_executables;
+   struct anv_pipeline_executable               executables[MAX_PIPELINE_EXECUTABLES];
+
    struct {
       const struct gen_l3_config *              l3_config;
       uint32_t                                  total_size;
@@ -3602,11 +3701,18 @@ struct anv_query_pool {
 
 int anv_get_instance_entrypoint_index(const char *name);
 int anv_get_device_entrypoint_index(const char *name);
+int anv_get_physical_device_entrypoint_index(const char *name);
+
+const char *anv_get_instance_entry_name(int index);
+const char *anv_get_physical_device_entry_name(int index);
+const char *anv_get_device_entry_name(int index);
 
 bool
 anv_instance_entrypoint_is_enabled(int index, uint32_t core_version,
                                    const struct anv_instance_extension_table *instance);
-
+bool
+anv_physical_device_entrypoint_is_enabled(int index, uint32_t core_version,
+                                          const struct anv_instance_extension_table *instance);
 bool
 anv_device_entrypoint_is_enabled(int index, uint32_t core_version,
                                  const struct anv_instance_extension_table *instance,
@@ -3627,8 +3733,7 @@ enum anv_dump_action {
 void anv_dump_start(struct anv_device *device, enum anv_dump_action actions);
 void anv_dump_finish(void);
 
-void anv_dump_add_framebuffer(struct anv_cmd_buffer *cmd_buffer,
-                              struct anv_framebuffer *fb);
+void anv_dump_add_attachments(struct anv_cmd_buffer *cmd_buffer);
 
 static inline uint32_t
 anv_get_subpass_id(const struct anv_cmd_state * const cmd_state)
@@ -3728,6 +3833,9 @@ ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_ycbcr_conversion, VkSamplerYcbcrConversion)
 #  define genX(x) gen11_##x
 #  include "anv_genX.h"
 #  undef genX
+#  define genX(x) gen12_##x
+#  include "anv_genX.h"
+#  undef genX
 #endif
 
 #endif /* ANV_PRIVATE_H */