radv: Add thread for timeline syncobj submission.
[mesa.git] / src / amd / vulkan / radv_private.h
index 52790aea23c0997f116000233b9df944e2bfdf42..4fa1ac3e08923cd7c42159ddcf888332683afc11 100644 (file)
@@ -95,18 +95,6 @@ typedef uint32_t xcb_window_t;
 #define RADV_SUPPORT_ANDROID_HARDWARE_BUFFER 0
 #endif
 
-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
@@ -337,7 +325,6 @@ struct radv_instance {
 
        uint64_t debug_flags;
        uint64_t perftest_flags;
-       uint8_t num_sc_threads;
 
        struct vk_debug_report_instance             debug_report_callbacks;
 
@@ -358,12 +345,6 @@ struct radv_instance {
        bool enable_mrt_output_nan_fixup;
 };
 
-static inline
-bool radv_device_use_secure_compile(struct radv_instance *instance)
-{
-   return instance->num_sc_threads;
-}
-
 VkResult radv_init_wsi(struct radv_physical_device *physical_device);
 void radv_finish_wsi(struct radv_physical_device *physical_device);
 
@@ -694,6 +675,8 @@ 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 {
@@ -730,6 +713,13 @@ 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 {
@@ -743,36 +733,6 @@ 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;
-
-       /* 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_secure_compile_state {
-       struct radv_secure_compile_process *secure_compile_processes;
-       uint32_t secure_compile_thread_counter;
-       mtx_t secure_compile_mutex;
-
-       /* Unique process ID used to build name for FIFO file descriptor */
-       char *uid;
-};
-
 #define RADV_BORDER_COLOR_COUNT       4096
 #define RADV_BORDER_COLOR_BUFFER_SIZE (sizeof(VkClearColorValue) * RADV_BORDER_COLOR_COUNT)
 
@@ -859,8 +819,6 @@ struct radv_device {
 
        struct radv_device_border_color_data border_color_data;
 
-       struct radv_secure_compile_state *sc_state;
-
        /* Condition variable for legacy timelines, to notify waiters when a
         * new point gets submitted. */
        pthread_cond_t timeline_cond;
@@ -991,42 +949,62 @@ 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 {
@@ -1067,6 +1045,8 @@ enum radv_cmd_flush_bits {
 struct radv_vertex_binding {
        struct radv_buffer *                          buffer;
        VkDeviceSize                                 offset;
+       VkDeviceSize size;
+       VkDeviceSize stride;
 };
 
 struct radv_streamout_binding {
@@ -1150,6 +1130,22 @@ 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;
@@ -1163,6 +1159,16 @@ 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;
@@ -1221,9 +1227,6 @@ 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.
  *
@@ -1426,7 +1429,8 @@ 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,
@@ -1510,6 +1514,9 @@ 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,
@@ -1635,7 +1642,6 @@ 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;
 };
@@ -1680,11 +1686,13 @@ 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;
@@ -2365,13 +2373,32 @@ void radv_initialize_fmask(struct radv_cmd_buffer *cmd_buffer,
                           struct radv_image *image,
                           const VkImageSubresourceRange *range);
 
+typedef enum {
+       RADV_FENCE_NONE,
+       RADV_FENCE_WINSYS,
+       RADV_FENCE_SYNCOBJ,
+       RADV_FENCE_WSI,
+} radv_fence_kind;
+
+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 radeon_winsys_fence *fence;
-       struct wsi_fence *fence_wsi;
-
-       uint32_t syncobj;
-       uint32_t temp_syncobj;
+       struct radv_fence_part permanent;
+       struct radv_fence_part temporary;
 };
 
 /* radv_nir_to_llvm.c */
@@ -2521,6 +2548,61 @@ 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;
+       }
+}
+
 #define RADV_DEFINE_HANDLE_CASTS(__radv_type, __VkType)                \
                                                                \
        static inline struct __radv_type *                      \