radv: add initial trap handler support with RADV_TRAP_HANDLER=1
[mesa.git] / src / amd / vulkan / radv_private.h
index 6b54a8ad60011d04923494e156b45fe40dcf07bd..e8383c910b2e5ce6b1fc42c474fc17a30e1c12c7 100644 (file)
@@ -51,6 +51,7 @@
 #include "util/xmlconfig.h"
 #include "vk_alloc.h"
 #include "vk_debug_report.h"
+#include "vk_object.h"
 
 #include "radv_radeon_winsys.h"
 #include "ac_binary.h"
@@ -94,30 +95,6 @@ typedef uint32_t xcb_window_t;
 #define RADV_SUPPORT_ANDROID_HARDWARE_BUFFER 0
 #endif
 
-
-struct gfx10_format {
-    unsigned img_format:9;
-
-    /* Various formats are only supported with workarounds for vertex fetch,
-     * and some 32_32_32 formats are supported natively, but only for buffers
-     * (possibly with some image support, actually, but no filtering). */
-    bool buffers_only:1;
-};
-
-#include "gfx10_format_table.h"
-
-enum radv_secure_compile_type {
-       RADV_SC_TYPE_INIT_SUCCESS,
-       RADV_SC_TYPE_INIT_FAILURE,
-       RADV_SC_TYPE_COMPILE_PIPELINE,
-       RADV_SC_TYPE_COMPILE_PIPELINE_FINISHED,
-       RADV_SC_TYPE_READ_DISK_CACHE,
-       RADV_SC_TYPE_WRITE_DISK_CACHE,
-       RADV_SC_TYPE_FORK_DEVICE,
-       RADV_SC_TYPE_DESTROY_DEVICE,
-       RADV_SC_TYPE_COUNT
-};
-
 #define radv_printflike(a, b) __attribute__((__format__(__printf__, a, b)))
 
 static inline uint32_t
@@ -217,10 +194,24 @@ radv_clear_mask(uint32_t *inout_mask, uint32_t clear_mask)
 struct radv_image_view;
 struct radv_instance;
 
-VkResult __vk_errorf(struct radv_instance *instance, VkResult error, const char *file, int line, const char *format, ...);
-
-#define vk_error(instance, error) __vk_errorf(instance, error, __FILE__, __LINE__, NULL);
-#define vk_errorf(instance, error, format, ...) __vk_errorf(instance, error, __FILE__, __LINE__, format, ## __VA_ARGS__);
+VkResult __vk_errorv(struct radv_instance *instance, const void *object,
+                    VkDebugReportObjectTypeEXT type, VkResult error,
+                    const char *file, int line, const char *format,
+                    va_list args);
+
+VkResult __vk_errorf(struct radv_instance *instance, const void *object,
+                    VkDebugReportObjectTypeEXT type, VkResult error,
+                    const char *file, int line, const char *format, ...)
+       radv_printflike(7, 8);
+
+#define vk_error(instance, error) \
+       __vk_errorf(instance, NULL, \
+                   VK_DEBUG_REPORT_OBJECT_TYPE_UNKNOWN_EXT, \
+                   error, __FILE__, __LINE__, NULL);
+#define vk_errorf(instance, error, format, ...) \
+       __vk_errorf(instance, NULL, \
+                   VK_DEBUG_REPORT_OBJECT_TYPE_UNKNOWN_EXT, \
+                   error, __FILE__, __LINE__, format, ## __VA_ARGS__);
 
 void __radv_finishme(const char *file, int line, const char *format, ...)
        radv_printflike(3, 4);
@@ -305,9 +296,6 @@ struct radv_physical_device {
        /* Whether DCC should be enabled for MSAA textures. */
        bool dcc_msaa_allowed;
 
-       /* Whether to enable the AMD_shader_ballot extension */
-       bool use_shader_ballot;
-
        /* Whether to enable NGG. */
        bool use_ngg;
 
@@ -322,8 +310,8 @@ struct radv_physical_device {
        uint8_t cs_wave_size;
        uint8_t ge_wave_size;
 
-       /* Whether to use the experimental compiler backend */
-       bool use_aco;
+       /* Whether to use the LLVM compiler backend */
+       bool use_llvm;
 
        /* This is the drivers on-disk cache used as a fallback as opposed to
         * the pipeline cache defined by apps.
@@ -340,18 +328,19 @@ struct radv_physical_device {
 };
 
 struct radv_instance {
-       VK_LOADER_DATA                              _loader_data;
+       struct vk_object_base                       base;
 
        VkAllocationCallbacks                       alloc;
 
        uint32_t                                    apiVersion;
 
+       char *                                      applicationName;
+       uint32_t                                    applicationVersion;
        char *                                      engineName;
        uint32_t                                    engineVersion;
 
        uint64_t debug_flags;
        uint64_t perftest_flags;
-       uint8_t num_sc_threads;
 
        struct vk_debug_report_instance             debug_report_callbacks;
 
@@ -365,13 +354,12 @@ struct radv_instance {
 
        struct driOptionCache dri_options;
        struct driOptionCache available_dri_options;
-};
 
-static inline
-bool radv_device_use_secure_compile(struct radv_instance *instance)
-{
-   return instance->num_sc_threads;
-}
+       /**
+        * Workarounds for game bugs.
+        */
+       bool enable_mrt_output_nan_fixup;
+};
 
 VkResult radv_init_wsi(struct radv_physical_device *physical_device);
 void radv_finish_wsi(struct radv_physical_device *physical_device);
@@ -384,8 +372,10 @@ bool radv_physical_device_extension_supported(struct radv_physical_device *dev,
 struct cache_entry;
 
 struct radv_pipeline_cache {
-       struct radv_device *                          device;
+       struct vk_object_base                        base;
+       struct radv_device *                         device;
        pthread_mutex_t                              mutex;
+       VkPipelineCacheCreateFlags                   flags;
 
        uint32_t                                     total_size;
        uint32_t                                     table_size;
@@ -411,6 +401,7 @@ struct radv_pipeline_key {
        uint32_t is_int10;
        uint8_t log2_ps_iter_samples;
        uint8_t num_samples;
+       bool is_dual_src;
        uint32_t has_multiview_view_index : 1;
        uint32_t optimisations_disabled : 1;
        uint8_t topology;
@@ -700,6 +691,8 @@ struct radv_meta_state {
 
 #define RADV_MAX_QUEUE_FAMILIES 3
 
+struct radv_deferred_queue_submission;
+
 enum ring_type radv_queue_family_to_ring(int f);
 
 struct radv_queue {
@@ -736,6 +729,13 @@ struct radv_queue {
 
        struct list_head pending_submissions;
        pthread_mutex_t pending_mutex;
+
+       pthread_mutex_t thread_mutex;
+       pthread_cond_t thread_cond;
+       struct radv_deferred_queue_submission *thread_submission;
+       pthread_t submission_thread;
+       bool thread_exit;
+       bool thread_running;
 };
 
 struct radv_bo_list {
@@ -749,40 +749,22 @@ VkResult radv_bo_list_add(struct radv_device *device,
 void radv_bo_list_remove(struct radv_device *device,
                         struct radeon_winsys_bo *bo);
 
-struct radv_secure_compile_process {
-       /* Secure process file descriptors. Used to communicate between the
-        * user facing device and the idle forked device used to fork a clean
-        * process for each new pipeline compile.
-        */
-       int fd_secure_input;
-       int fd_secure_output;
+#define RADV_BORDER_COLOR_COUNT       4096
+#define RADV_BORDER_COLOR_BUFFER_SIZE (sizeof(VkClearColorValue) * RADV_BORDER_COLOR_COUNT)
 
-       /* FIFO file descriptors used to communicate between the user facing
-        * device and the secure process that does the actual secure compile.
-        */
-       int fd_server;
-       int fd_client;
-
-       /* Secure compile process id */
-       pid_t sc_pid;
-
-       /* Is the secure compile process currently in use by a thread */
-       bool in_use;
-};
+struct radv_device_border_color_data {
+       bool                     used[RADV_BORDER_COLOR_COUNT];
 
-struct radv_secure_compile_state {
-       struct radv_secure_compile_process *secure_compile_processes;
-       uint32_t secure_compile_thread_counter;
-       mtx_t secure_compile_mutex;
+       struct radeon_winsys_bo *bo;
+       VkClearColorValue       *colors_gpu_ptr;
 
-       /* Unique process ID used to build name for FIFO file descriptor */
-       char *uid;
+       /* Mutex is required to guarantee vkCreateSampler thread safety
+        * given that we are writing to a buffer and checking color occupation */
+       pthread_mutex_t          mutex;
 };
 
 struct radv_device {
-       VK_LOADER_DATA                              _loader_data;
-
-       VkAllocationCallbacks                       alloc;
+       struct vk_device vk;
 
        struct radv_instance *                       instance;
        struct radeon_winsys *ws;
@@ -851,7 +833,7 @@ struct radv_device {
        /* Whether anisotropy is forced with RADV_TEX_ANISO (-1 is disabled). */
        int force_aniso;
 
-       struct radv_secure_compile_state *sc_state;
+       struct radv_device_border_color_data border_color_data;
 
        /* Condition variable for legacy timelines, to notify waiters when a
         * new point gets submitted. */
@@ -865,13 +847,36 @@ struct radv_device {
        uint32_t thread_trace_buffer_size;
        int thread_trace_start_frame;
 
+       /* Trap handler. */
+       struct radv_shader_variant *trap_handler_shader;
+       struct radeon_winsys_bo *tma_bo; /* Trap Memory Address */
+       uint32_t *tma_ptr;
+
        /* Overallocation. */
        bool overallocation_disallowed;
        uint64_t allocated_memory_size[VK_MAX_MEMORY_HEAPS];
        mtx_t overallocation_mutex;
+
+       /* Track the number of device loss occurs. */
+       int lost;
 };
 
+VkResult _radv_device_set_lost(struct radv_device *device,
+                              const char *file, int line,
+                              const char *msg, ...)
+       radv_printflike(4, 5);
+
+#define radv_device_set_lost(dev, ...) \
+       _radv_device_set_lost(dev, __FILE__, __LINE__, __VA_ARGS__)
+
+static inline bool
+radv_device_is_lost(const struct radv_device *device)
+{
+       return unlikely(p_atomic_read(&device->lost));
+}
+
 struct radv_device_memory {
+       struct vk_object_base                        base;
        struct radeon_winsys_bo                      *bo;
        /* for dedicated allocations */
        struct radv_image                            *image;
@@ -893,6 +898,7 @@ struct radv_descriptor_range {
 };
 
 struct radv_descriptor_set {
+       struct vk_object_base base;
        const struct radv_descriptor_set_layout *layout;
        uint32_t size;
        uint32_t buffer_count;
@@ -918,6 +924,7 @@ struct radv_descriptor_pool_entry {
 };
 
 struct radv_descriptor_pool {
+       struct vk_object_base base;
        struct radeon_winsys_bo *bo;
        uint8_t *mapped_ptr;
        uint64_t current_offset;
@@ -959,12 +966,14 @@ struct radv_descriptor_update_template_entry {
 };
 
 struct radv_descriptor_update_template {
+       struct vk_object_base base;
        uint32_t entry_count;
        VkPipelineBindPoint bind_point;
        struct radv_descriptor_update_template_entry entry[0];
 };
 
 struct radv_buffer {
+       struct vk_object_base                        base;
        VkDeviceSize                                 size;
 
        VkBufferUsageFlags                           usage;
@@ -978,42 +987,62 @@ struct radv_buffer {
 };
 
 enum radv_dynamic_state_bits {
-       RADV_DYNAMIC_VIEWPORT             = 1 << 0,
-       RADV_DYNAMIC_SCISSOR              = 1 << 1,
-       RADV_DYNAMIC_LINE_WIDTH           = 1 << 2,
-       RADV_DYNAMIC_DEPTH_BIAS           = 1 << 3,
-       RADV_DYNAMIC_BLEND_CONSTANTS      = 1 << 4,
-       RADV_DYNAMIC_DEPTH_BOUNDS         = 1 << 5,
-       RADV_DYNAMIC_STENCIL_COMPARE_MASK = 1 << 6,
-       RADV_DYNAMIC_STENCIL_WRITE_MASK   = 1 << 7,
-       RADV_DYNAMIC_STENCIL_REFERENCE    = 1 << 8,
-       RADV_DYNAMIC_DISCARD_RECTANGLE    = 1 << 9,
-       RADV_DYNAMIC_SAMPLE_LOCATIONS     = 1 << 10,
-       RADV_DYNAMIC_LINE_STIPPLE         = 1 << 11,
-       RADV_DYNAMIC_ALL                  = (1 << 12) - 1,
+       RADV_DYNAMIC_VIEWPORT                           = 1 << 0,
+       RADV_DYNAMIC_SCISSOR                            = 1 << 1,
+       RADV_DYNAMIC_LINE_WIDTH                         = 1 << 2,
+       RADV_DYNAMIC_DEPTH_BIAS                         = 1 << 3,
+       RADV_DYNAMIC_BLEND_CONSTANTS                    = 1 << 4,
+       RADV_DYNAMIC_DEPTH_BOUNDS                       = 1 << 5,
+       RADV_DYNAMIC_STENCIL_COMPARE_MASK               = 1 << 6,
+       RADV_DYNAMIC_STENCIL_WRITE_MASK                 = 1 << 7,
+       RADV_DYNAMIC_STENCIL_REFERENCE                  = 1 << 8,
+       RADV_DYNAMIC_DISCARD_RECTANGLE                  = 1 << 9,
+       RADV_DYNAMIC_SAMPLE_LOCATIONS                   = 1 << 10,
+       RADV_DYNAMIC_LINE_STIPPLE                       = 1 << 11,
+       RADV_DYNAMIC_CULL_MODE                          = 1 << 12,
+       RADV_DYNAMIC_FRONT_FACE                         = 1 << 13,
+       RADV_DYNAMIC_PRIMITIVE_TOPOLOGY                 = 1 << 14,
+       RADV_DYNAMIC_DEPTH_TEST_ENABLE                  = 1 << 15,
+       RADV_DYNAMIC_DEPTH_WRITE_ENABLE                 = 1 << 16,
+       RADV_DYNAMIC_DEPTH_COMPARE_OP                   = 1 << 17,
+       RADV_DYNAMIC_DEPTH_BOUNDS_TEST_ENABLE           = 1 << 18,
+       RADV_DYNAMIC_STENCIL_TEST_ENABLE                = 1 << 19,
+       RADV_DYNAMIC_STENCIL_OP                         = 1 << 20,
+       RADV_DYNAMIC_VERTEX_INPUT_BINDING_STRIDE        = 1 << 21,
+       RADV_DYNAMIC_ALL                                = (1 << 22) - 1,
 };
 
 enum radv_cmd_dirty_bits {
        /* Keep the dynamic state dirty bits in sync with
         * enum radv_dynamic_state_bits */
-       RADV_CMD_DIRTY_DYNAMIC_VIEWPORT                  = 1 << 0,
-       RADV_CMD_DIRTY_DYNAMIC_SCISSOR                   = 1 << 1,
-       RADV_CMD_DIRTY_DYNAMIC_LINE_WIDTH                = 1 << 2,
-       RADV_CMD_DIRTY_DYNAMIC_DEPTH_BIAS                = 1 << 3,
-       RADV_CMD_DIRTY_DYNAMIC_BLEND_CONSTANTS           = 1 << 4,
-       RADV_CMD_DIRTY_DYNAMIC_DEPTH_BOUNDS              = 1 << 5,
-       RADV_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK      = 1 << 6,
-       RADV_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK        = 1 << 7,
-       RADV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE         = 1 << 8,
-       RADV_CMD_DIRTY_DYNAMIC_DISCARD_RECTANGLE         = 1 << 9,
-       RADV_CMD_DIRTY_DYNAMIC_SAMPLE_LOCATIONS          = 1 << 10,
-       RADV_CMD_DIRTY_DYNAMIC_LINE_STIPPLE              = 1 << 11,
-       RADV_CMD_DIRTY_DYNAMIC_ALL                       = (1 << 12) - 1,
-       RADV_CMD_DIRTY_PIPELINE                          = 1 << 12,
-       RADV_CMD_DIRTY_INDEX_BUFFER                      = 1 << 13,
-       RADV_CMD_DIRTY_FRAMEBUFFER                       = 1 << 14,
-       RADV_CMD_DIRTY_VERTEX_BUFFER                     = 1 << 15,
-       RADV_CMD_DIRTY_STREAMOUT_BUFFER                  = 1 << 16,
+       RADV_CMD_DIRTY_DYNAMIC_VIEWPORT                         = 1 << 0,
+       RADV_CMD_DIRTY_DYNAMIC_SCISSOR                          = 1 << 1,
+       RADV_CMD_DIRTY_DYNAMIC_LINE_WIDTH                       = 1 << 2,
+       RADV_CMD_DIRTY_DYNAMIC_DEPTH_BIAS                       = 1 << 3,
+       RADV_CMD_DIRTY_DYNAMIC_BLEND_CONSTANTS                  = 1 << 4,
+       RADV_CMD_DIRTY_DYNAMIC_DEPTH_BOUNDS                     = 1 << 5,
+       RADV_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK             = 1 << 6,
+       RADV_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK               = 1 << 7,
+       RADV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE                = 1 << 8,
+       RADV_CMD_DIRTY_DYNAMIC_DISCARD_RECTANGLE                = 1 << 9,
+       RADV_CMD_DIRTY_DYNAMIC_SAMPLE_LOCATIONS                 = 1 << 10,
+       RADV_CMD_DIRTY_DYNAMIC_LINE_STIPPLE                     = 1 << 11,
+       RADV_CMD_DIRTY_DYNAMIC_CULL_MODE                        = 1 << 12,
+       RADV_CMD_DIRTY_DYNAMIC_FRONT_FACE                       = 1 << 13,
+       RADV_CMD_DIRTY_DYNAMIC_PRIMITIVE_TOPOLOGY               = 1 << 14,
+       RADV_CMD_DIRTY_DYNAMIC_DEPTH_TEST_ENABLE                = 1 << 15,
+       RADV_CMD_DIRTY_DYNAMIC_DEPTH_WRITE_ENABLE               = 1 << 16,
+       RADV_CMD_DIRTY_DYNAMIC_DEPTH_COMPARE_OP                 = 1 << 17,
+       RADV_CMD_DIRTY_DYNAMIC_DEPTH_BOUNDS_TEST_ENABLE         = 1 << 18,
+       RADV_CMD_DIRTY_DYNAMIC_STENCIL_TEST_ENABLE              = 1 << 19,
+       RADV_CMD_DIRTY_DYNAMIC_STENCIL_OP                       = 1 << 20,
+       RADV_CMD_DIRTY_DYNAMIC_VERTEX_INPUT_BINDING_STRIDE      = 1 << 21,
+       RADV_CMD_DIRTY_DYNAMIC_ALL                              = (1 << 22) - 1,
+       RADV_CMD_DIRTY_PIPELINE                                 = 1 << 22,
+       RADV_CMD_DIRTY_INDEX_BUFFER                             = 1 << 23,
+       RADV_CMD_DIRTY_FRAMEBUFFER                              = 1 << 24,
+       RADV_CMD_DIRTY_VERTEX_BUFFER                            = 1 << 25,
+       RADV_CMD_DIRTY_STREAMOUT_BUFFER                         = 1 << 26,
 };
 
 enum radv_cmd_flush_bits {
@@ -1054,6 +1083,8 @@ enum radv_cmd_flush_bits {
 struct radv_vertex_binding {
        struct radv_buffer *                          buffer;
        VkDeviceSize                                 offset;
+       VkDeviceSize size;
+       VkDeviceSize stride;
 };
 
 struct radv_streamout_binding {
@@ -1137,6 +1168,22 @@ struct radv_dynamic_state {
                uint32_t                                  back;
        } stencil_write_mask;
 
+       struct {
+               struct {
+                       VkStencilOp fail_op;
+                       VkStencilOp pass_op;
+                       VkStencilOp depth_fail_op;
+                       VkCompareOp compare_op;
+               } front;
+
+               struct {
+                       VkStencilOp fail_op;
+                       VkStencilOp pass_op;
+                       VkStencilOp depth_fail_op;
+                       VkCompareOp compare_op;
+               } back;
+       } stencil_op;
+
        struct {
                uint32_t                                  front;
                uint32_t                                  back;
@@ -1150,6 +1197,16 @@ struct radv_dynamic_state {
                uint32_t factor;
                uint16_t pattern;
        } line_stipple;
+
+       VkCullModeFlags cull_mode;
+       VkFrontFace front_face;
+       unsigned primitive_topology;
+
+       bool depth_test_enable;
+       bool depth_write_enable;
+       VkCompareOp depth_compare_op;
+       bool depth_bounds_test_enable;
+       bool stencil_test_enable;
 };
 
 extern const struct radv_dynamic_state default_dynamic_state;
@@ -1208,9 +1265,6 @@ radv_initialise_ds_surface(struct radv_device *device,
                           struct radv_ds_buffer_info *ds,
                           struct radv_image_view *iview);
 
-bool
-radv_sc_read(int fd, void *buf, size_t size, bool timeout);
-
 /**
  * Attachment state when recording a renderpass instance.
  *
@@ -1317,6 +1371,7 @@ struct radv_cmd_state {
 };
 
 struct radv_cmd_pool {
+       struct vk_object_base                        base;
        VkAllocationCallbacks                        alloc;
        struct list_head                             cmd_buffers;
        struct list_head                             free_cmd_buffers;
@@ -1340,7 +1395,7 @@ enum radv_cmd_buffer_status {
 };
 
 struct radv_cmd_buffer {
-       VK_LOADER_DATA                               _loader_data;
+       struct vk_object_base                         base;
 
        struct radv_device *                          device;
 
@@ -1399,7 +1454,7 @@ bool radv_cmd_buffer_uses_mec(struct radv_cmd_buffer *cmd_buffer);
 
 void si_emit_graphics(struct radv_device *device,
                      struct radeon_cmdbuf *cs);
-void si_emit_compute(struct radv_physical_device *physical_device,
+void si_emit_compute(struct radv_device *device,
                      struct radeon_cmdbuf *cs);
 
 void cik_create_gfx_config(struct radv_device *device);
@@ -1412,7 +1467,8 @@ void si_write_scissors(struct radeon_cmdbuf *cs, int first,
 uint32_t si_get_ia_multi_vgt_param(struct radv_cmd_buffer *cmd_buffer,
                                   bool instanced_draw, bool indirect_draw,
                                   bool count_from_stream_output,
-                                  uint32_t draw_vertex_count);
+                                  uint32_t draw_vertex_count,
+                                  unsigned topology);
 void si_cs_emit_write_event_eop(struct radeon_cmdbuf *cs,
                                enum chip_class chip_class,
                                bool is_mec,
@@ -1496,6 +1552,9 @@ void radv_cmd_buffer_trace_emit(struct radv_cmd_buffer *cmd_buffer);
 bool radv_get_memory_fd(struct radv_device *device,
                        struct radv_device_memory *memory,
                        int *pFD);
+void radv_free_memory(struct radv_device *device,
+                     const VkAllocationCallbacks* pAllocator,
+                     struct radv_device_memory *mem);
 
 static inline void
 radv_emit_shader_pointer_head(struct radeon_cmdbuf *cs,
@@ -1554,6 +1613,7 @@ void radv_unaligned_dispatch(
        uint32_t                                    z);
 
 struct radv_event {
+       struct vk_object_base base;
        struct radeon_winsys_bo *bo;
        uint64_t *map;
 };
@@ -1564,7 +1624,7 @@ struct radv_shader_module;
 #define RADV_HASH_SHADER_CS_WAVE32           (1 << 1)
 #define RADV_HASH_SHADER_PS_WAVE32           (1 << 2)
 #define RADV_HASH_SHADER_GE_WAVE32           (1 << 3)
-#define RADV_HASH_SHADER_ACO                 (1 << 4)
+#define RADV_HASH_SHADER_LLVM                (1 << 4)
 
 void
 radv_hash_shaders(unsigned char *hash,
@@ -1599,7 +1659,6 @@ unsigned radv_format_meta_fs_key(VkFormat format);
 
 struct radv_multisample_state {
        uint32_t db_eqaa;
-       uint32_t pa_sc_line_cntl;
        uint32_t pa_sc_mode_cntl_0;
        uint32_t pa_sc_mode_cntl_1;
        uint32_t pa_sc_aa_config;
@@ -1612,15 +1671,10 @@ struct radv_prim_vertex_count {
        uint8_t incr;
 };
 
-struct radv_vertex_elements_info {
-       uint32_t format_size[MAX_VERTEX_ATTRIBS];
-};
-
 struct radv_ia_multi_vgt_param_helpers {
        uint32_t base;
        bool partial_es_wave;
        uint8_t primgroup_size;
-       bool wd_switch_on_eop;
        bool ia_switch_on_eoi;
        bool partial_vs_wave;
 };
@@ -1633,6 +1687,7 @@ struct radv_binning_state {
 #define SI_GS_PER_ES 128
 
 struct radv_pipeline {
+       struct vk_object_base                         base;
        struct radv_device *                          device;
        struct radv_dynamic_state                     dynamic_state;
 
@@ -1647,8 +1702,6 @@ struct radv_pipeline {
        uint32_t                                  ctx_cs_hash;
        struct radeon_cmdbuf                      ctx_cs;
 
-       struct radv_vertex_elements_info             vertex_elements;
-
        uint32_t                                     binding_stride[MAX_VBS];
        uint8_t                                      num_vertex_bindings;
 
@@ -1664,11 +1717,13 @@ struct radv_pipeline {
                        uint32_t vtx_base_sgpr;
                        struct radv_ia_multi_vgt_param_helpers ia_multi_vgt_param;
                        uint8_t vtx_emit_num;
-                       struct radv_prim_vertex_count prim_vertex_count;
                        bool can_use_guardband;
                        uint32_t needed_dynamic_state;
                        bool disable_out_of_order_rast_for_occlusion;
-                       uint8_t topology;
+                       unsigned tess_patch_control_points;
+                       unsigned pa_su_sc_mode_cntl;
+                       unsigned db_depth_control;
+                       bool uses_dynamic_stride;
 
                        /* Used for rbplus */
                        uint32_t col_format;
@@ -1703,7 +1758,7 @@ struct radv_userdata_info *radv_lookup_user_sgpr(struct radv_pipeline *pipeline,
                                                 gl_shader_stage stage,
                                                 int idx);
 
-struct radv_shader_variant *radv_get_shader(struct radv_pipeline *pipeline,
+struct radv_shader_variant *radv_get_shader(const struct radv_pipeline *pipeline,
                                            gl_shader_stage stage);
 
 struct radv_graphics_pipeline_create_info {
@@ -1769,6 +1824,7 @@ struct radv_image_plane {
 };
 
 struct radv_image {
+       struct vk_object_base base;
        VkImageType type;
        /* The original VkFormat provided by the client.  This may not match any
         * of the actual surface formats.
@@ -1790,13 +1846,9 @@ struct radv_image {
        /* Set when bound */
        struct radeon_winsys_bo *bo;
        VkDeviceSize offset;
-       uint64_t dcc_offset;
-       uint64_t htile_offset;
        bool tc_compatible_htile;
        bool tc_compatible_cmask;
 
-       uint64_t cmask_offset;
-       uint64_t fmask_offset;
        uint64_t clear_value_offset;
        uint64_t fce_pred_offset;
        uint64_t dcc_pred_offset;
@@ -1844,7 +1896,7 @@ bool radv_layout_dcc_compressed(const struct radv_device *device,
 static inline bool
 radv_image_has_cmask(const struct radv_image *image)
 {
-       return image->cmask_offset;
+       return image->planes[0].surface.cmask_offset;
 }
 
 /**
@@ -1853,7 +1905,7 @@ radv_image_has_cmask(const struct radv_image *image)
 static inline bool
 radv_image_has_fmask(const struct radv_image *image)
 {
-       return image->fmask_offset;
+       return image->planes[0].surface.fmask_offset;
 }
 
 /**
@@ -2007,6 +2059,7 @@ union radv_descriptor {
 };
 
 struct radv_image_view {
+       struct vk_object_base base;
        struct radv_image *image; /**< VkImageViewCreateInfo::image */
        struct radeon_winsys_bo *bo;
 
@@ -2085,6 +2138,7 @@ void radv_image_view_init(struct radv_image_view *view,
 VkFormat radv_get_aspect_format(struct radv_image *image, VkImageAspectFlags mask);
 
 struct radv_sampler_ycbcr_conversion {
+       struct vk_object_base base;
        VkFormat format;
        VkSamplerYcbcrModelConversion ycbcr_model;
        VkSamplerYcbcrRange ycbcr_range;
@@ -2094,6 +2148,7 @@ struct radv_sampler_ycbcr_conversion {
 };
 
 struct radv_buffer_view {
+       struct vk_object_base base;
        struct radeon_winsys_bo *bo;
        VkFormat vk_format;
        uint64_t range; /**< VkBufferViewCreateInfo::range */
@@ -2147,11 +2202,14 @@ radv_image_extent_compare(const struct radv_image *image,
 }
 
 struct radv_sampler {
+       struct vk_object_base base;
        uint32_t state[4];
        struct radv_sampler_ycbcr_conversion *ycbcr_sampler;
+       uint32_t border_color_slot;
 };
 
 struct radv_framebuffer {
+       struct vk_object_base                        base;
        uint32_t                                     width;
        uint32_t                                     height;
        uint32_t                                     layers;
@@ -2224,6 +2282,7 @@ struct radv_render_pass_attachment {
 };
 
 struct radv_render_pass {
+       struct vk_object_base                        base;
        uint32_t                                     attachment_count;
        uint32_t                                     subpass_count;
        struct radv_subpass_attachment *             subpass_attachments;
@@ -2236,6 +2295,7 @@ VkResult radv_device_init_meta(struct radv_device *device);
 void radv_device_finish_meta(struct radv_device *device);
 
 struct radv_query_pool {
+       struct vk_object_base base;
        struct radeon_winsys_bo *bo;
        uint32_t stride;
        uint32_t availability_offset;
@@ -2249,6 +2309,7 @@ typedef enum {
        RADV_SEMAPHORE_NONE,
        RADV_SEMAPHORE_WINSYS,
        RADV_SEMAPHORE_SYNCOBJ,
+       RADV_SEMAPHORE_TIMELINE_SYNCOBJ,
        RADV_SEMAPHORE_TIMELINE,
 } radv_semaphore_kind;
 
@@ -2289,16 +2350,25 @@ struct radv_timeline {
        struct list_head waiters;
 };
 
+struct radv_timeline_syncobj {
+       /* Keep syncobj first, so common-code can just handle this as
+        * non-timeline syncobj. */
+       uint32_t syncobj;
+       uint64_t max_point; /* max submitted point. */
+};
+
 struct radv_semaphore_part {
        radv_semaphore_kind kind;
        union {
                uint32_t syncobj;
                struct radeon_winsys_sem *ws_sem;
                struct radv_timeline timeline;
+               struct radv_timeline_syncobj timeline_syncobj;
        };
 };
 
 struct radv_semaphore {
+       struct vk_object_base base;
        struct radv_semaphore_part permanent;
        struct radv_semaphore_part temporary;
 };
@@ -2342,12 +2412,32 @@ void radv_initialize_fmask(struct radv_cmd_buffer *cmd_buffer,
                           struct radv_image *image,
                           const VkImageSubresourceRange *range);
 
-struct radv_fence {
-       struct radeon_winsys_fence *fence;
-       struct wsi_fence *fence_wsi;
+typedef enum {
+       RADV_FENCE_NONE,
+       RADV_FENCE_WINSYS,
+       RADV_FENCE_SYNCOBJ,
+       RADV_FENCE_WSI,
+} radv_fence_kind;
 
-       uint32_t syncobj;
-       uint32_t temp_syncobj;
+struct radv_fence_part {
+       radv_fence_kind kind;
+
+       union {
+               /* AMDGPU winsys fence. */
+               struct radeon_winsys_fence *fence;
+
+               /* DRM syncobj handle for syncobj-based fences. */
+               uint32_t syncobj;
+
+               /* WSI fence. */
+               struct wsi_fence *fence_wsi;
+       };
+};
+
+struct radv_fence {
+       struct vk_object_base base;
+       struct radv_fence_part permanent;
+       struct radv_fence_part temporary;
 };
 
 /* radv_nir_to_llvm.c */
@@ -2371,7 +2461,7 @@ void radv_nir_shader_info_pass(const struct nir_shader *nir,
                               const struct radv_pipeline_layout *layout,
                               const struct radv_shader_variant_key *key,
                               struct radv_shader_info *info,
-                              bool use_aco);
+                              bool use_llvm);
 
 void radv_nir_shader_info_init(struct radv_shader_info *info);
 
@@ -2497,6 +2587,71 @@ void radv_cmd_buffer_begin_render_pass(struct radv_cmd_buffer *cmd_buffer,
                                       const VkRenderPassBeginInfo *pRenderPassBegin);
 void radv_cmd_buffer_end_render_pass(struct radv_cmd_buffer *cmd_buffer);
 
+static inline uint32_t si_translate_prim(unsigned topology)
+{
+       switch (topology) {
+       case VK_PRIMITIVE_TOPOLOGY_POINT_LIST:
+               return V_008958_DI_PT_POINTLIST;
+       case VK_PRIMITIVE_TOPOLOGY_LINE_LIST:
+               return V_008958_DI_PT_LINELIST;
+       case VK_PRIMITIVE_TOPOLOGY_LINE_STRIP:
+               return V_008958_DI_PT_LINESTRIP;
+       case VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST:
+               return V_008958_DI_PT_TRILIST;
+       case VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP:
+               return V_008958_DI_PT_TRISTRIP;
+       case VK_PRIMITIVE_TOPOLOGY_TRIANGLE_FAN:
+               return V_008958_DI_PT_TRIFAN;
+       case VK_PRIMITIVE_TOPOLOGY_LINE_LIST_WITH_ADJACENCY:
+               return V_008958_DI_PT_LINELIST_ADJ;
+       case VK_PRIMITIVE_TOPOLOGY_LINE_STRIP_WITH_ADJACENCY:
+               return V_008958_DI_PT_LINESTRIP_ADJ;
+       case VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST_WITH_ADJACENCY:
+               return V_008958_DI_PT_TRILIST_ADJ;
+       case VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP_WITH_ADJACENCY:
+               return V_008958_DI_PT_TRISTRIP_ADJ;
+       case VK_PRIMITIVE_TOPOLOGY_PATCH_LIST:
+               return V_008958_DI_PT_PATCH;
+       default:
+               assert(0);
+               return 0;
+       }
+}
+
+static inline uint32_t si_translate_stencil_op(enum VkStencilOp op)
+{
+       switch (op) {
+       case VK_STENCIL_OP_KEEP:
+               return V_02842C_STENCIL_KEEP;
+       case VK_STENCIL_OP_ZERO:
+               return V_02842C_STENCIL_ZERO;
+       case VK_STENCIL_OP_REPLACE:
+               return V_02842C_STENCIL_REPLACE_TEST;
+       case VK_STENCIL_OP_INCREMENT_AND_CLAMP:
+               return V_02842C_STENCIL_ADD_CLAMP;
+       case VK_STENCIL_OP_DECREMENT_AND_CLAMP:
+               return V_02842C_STENCIL_SUB_CLAMP;
+       case VK_STENCIL_OP_INVERT:
+               return V_02842C_STENCIL_INVERT;
+       case VK_STENCIL_OP_INCREMENT_AND_WRAP:
+               return V_02842C_STENCIL_ADD_WRAP;
+       case VK_STENCIL_OP_DECREMENT_AND_WRAP:
+               return V_02842C_STENCIL_SUB_WRAP;
+       default:
+               return 0;
+       }
+}
+
+/**
+ * Helper used for debugging compiler issues by enabling/disabling LLVM for a
+ * specific shader stage (developers only).
+ */
+static inline bool
+radv_use_llvm_for_stage(struct radv_device *device, UNUSED gl_shader_stage stage)
+{
+       return device->physical_device->use_llvm;
+}
+
 #define RADV_DEFINE_HANDLE_CASTS(__radv_type, __VkType)                \
                                                                \
        static inline struct __radv_type *                      \