#define VG(x)
#endif
+#include "c11/threads.h"
#include <amdgpu.h>
#include "compiler/shader_enums.h"
#include "util/macros.h"
#include "util/list.h"
+#include "util/xmlconfig.h"
#include "main/macros.h"
#include "vk_alloc.h"
#include "vk_debug_report.h"
#include "ac_nir_to_llvm.h"
#include "ac_gpu_info.h"
#include "ac_surface.h"
+#include "ac_llvm_build.h"
+#include "ac_llvm_util.h"
#include "radv_descriptor_set.h"
#include "radv_extensions.h"
+#include "radv_cs.h"
#include <llvm-c/TargetMachine.h>
#include "radv_entrypoints.h"
#include "wsi_common.h"
+#include "wsi_common_display.h"
#define ATI_VENDOR_ID 0x1002
#define MAX_DYNAMIC_STORAGE_BUFFERS 8
#define MAX_DYNAMIC_BUFFERS (MAX_DYNAMIC_UNIFORM_BUFFERS + MAX_DYNAMIC_STORAGE_BUFFERS)
#define MAX_SAMPLES_LOG2 4
-#define NUM_META_FS_KEYS 13
+#define NUM_META_FS_KEYS 12
#define RADV_MAX_DRM_DEVICES 8
#define MAX_VIEWS 8
+#define MAX_SO_STREAMS 4
+#define MAX_SO_BUFFERS 4
+#define MAX_SO_OUTPUTS 64
+#define MAX_INLINE_UNIFORM_BLOCK_SIZE (4ull * 1024 * 1024)
+#define MAX_INLINE_UNIFORM_BLOCK_COUNT 64
#define NUM_DEPTH_CLEAR_PIPELINES 3
*/
#define RADV_BUFFER_OPS_CS_THRESHOLD 4096
+#define RADV_BUFFER_UPDATE_THRESHOLD 1024
+
enum radv_mem_heap {
RADV_MEM_HEAP_VRAM,
RADV_MEM_HEAP_VRAM_CPU_ACCESS,
* propagating errors. Might be useful to plug in a stack trace here.
*/
-VkResult __vk_errorf(VkResult error, const char *file, int line, const char *format, ...);
+struct radv_instance;
-#ifdef DEBUG
-#define vk_error(error) __vk_errorf(error, __FILE__, __LINE__, NULL);
-#define vk_errorf(error, format, ...) __vk_errorf(error, __FILE__, __LINE__, format, ## __VA_ARGS__);
-#else
-#define vk_error(error) error
-#define vk_errorf(error, format, ...) error
-#endif
+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__);
void __radv_finishme(const char *file, int line, const char *format, ...)
radv_printflike(3, 4);
void radv_loge(const char *format, ...) radv_printflike(1, 2);
void radv_loge_v(const char *format, va_list va);
+void radv_logi(const char *format, ...) radv_printflike(1, 2);
+void radv_logi_v(const char *format, va_list va);
/**
* Print a FINISHME message, including its source location.
struct radeon_winsys *ws;
struct radeon_info rad_info;
- char path[20];
char name[VK_MAX_PHYSICAL_DEVICE_NAME_SIZE];
uint8_t driver_uuid[VK_UUID_SIZE];
uint8_t device_uuid[VK_UUID_SIZE];
uint8_t cache_uuid[VK_UUID_SIZE];
int local_fd;
+ int master_fd;
struct wsi_device wsi_device;
bool has_rbplus; /* if RB+ register exist */
bool has_out_of_order_rast;
bool out_of_order_rast_allowed;
+ /* Whether DCC should be enabled for MSAA textures. */
+ bool dcc_msaa_allowed;
+
+ /* Whether LOAD_CONTEXT_REG packets are supported. */
+ bool has_load_ctx_reg_pkt;
+
/* This is the drivers on-disk cache used as a fallback as opposed to
* the pipeline cache defined by apps.
*/
VkPhysicalDeviceMemoryProperties memory_properties;
enum radv_mem_type mem_type_indices[RADV_MEM_TYPE_COUNT];
+ drmPciBusInfo bus_info;
+
struct radv_device_extension_table supported_extensions;
};
struct vk_debug_report_instance debug_report_callbacks;
struct radv_instance_extension_table enabled_extensions;
+
+ struct driOptionCache dri_options;
+ struct driOptionCache available_dri_options;
};
VkResult radv_init_wsi(struct radv_physical_device *physical_device);
struct radv_pipeline_key {
uint32_t instance_rate_inputs;
+ uint32_t instance_rate_divisors[MAX_VERTEX_ATTRIBS];
+ uint8_t vertex_attribute_formats[MAX_VERTEX_ATTRIBS];
+ uint32_t vertex_attribute_bindings[MAX_VERTEX_ATTRIBS];
+ uint32_t vertex_attribute_offsets[MAX_VERTEX_ATTRIBS];
+ uint32_t vertex_attribute_strides[MAX_VERTEX_ATTRIBS];
+ uint64_t vertex_alpha_adjust;
+ uint32_t vertex_post_shuffle;
unsigned tess_input_vertices;
uint32_t col_format;
uint32_t is_int8;
uint32_t is_int10;
uint8_t log2_ps_iter_samples;
- uint8_t log2_num_samples;
- uint32_t multisample : 1;
+ uint8_t num_samples;
uint32_t has_multiview_view_index : 1;
+ uint32_t optimisations_disabled : 1;
};
void
struct radv_device *device);
void
radv_pipeline_cache_finish(struct radv_pipeline_cache *cache);
-void
+bool
radv_pipeline_cache_load(struct radv_pipeline_cache *cache,
const void *data, size_t size);
radv_create_shader_variants_from_pipeline_cache(struct radv_device *device,
struct radv_pipeline_cache *cache,
const unsigned char *sha1,
- struct radv_shader_variant **variants);
+ struct radv_shader_variant **variants,
+ bool *found_in_application_cache);
void
radv_pipeline_cache_insert_shaders(struct radv_device *device,
struct radv_pipeline_cache cache;
+ /*
+ * For on-demand pipeline creation, makes sure that
+ * only one thread tries to build a pipeline at the same time.
+ */
+ mtx_t mtx;
+
/**
* Use array element `i` for images with `2^i` samples.
*/
VkPipelineLayout clear_color_p_layout;
VkPipelineLayout clear_depth_p_layout;
+
+ /* Optimized compute fast HTILE clear for stencil or depth only. */
+ VkPipeline clear_htile_mask_pipeline;
+ VkPipelineLayout clear_htile_mask_p_layout;
+ VkDescriptorSetLayout clear_htile_mask_ds_layout;
+
struct {
VkRenderPass render_pass[NUM_META_FS_KEYS][RADV_META_DST_LAYOUT_COUNT];
} blit;
struct {
- VkRenderPass render_passes[NUM_META_FS_KEYS][RADV_META_DST_LAYOUT_COUNT];
+ VkPipelineLayout p_layouts[5];
+ VkDescriptorSetLayout ds_layouts[5];
+ VkPipeline pipelines[5][NUM_META_FS_KEYS];
- VkPipelineLayout p_layouts[3];
- VkDescriptorSetLayout ds_layouts[3];
- VkPipeline pipelines[3][NUM_META_FS_KEYS];
+ VkPipeline depth_only_pipeline[5];
- VkRenderPass depth_only_rp[RADV_BLIT_DS_LAYOUT_COUNT];
- VkPipeline depth_only_pipeline[3];
+ VkPipeline stencil_only_pipeline[5];
+ } blit2d[1 + MAX_SAMPLES_LOG2];
- VkRenderPass stencil_only_rp[RADV_BLIT_DS_LAYOUT_COUNT];
- VkPipeline stencil_only_pipeline[3];
- } blit2d;
+ VkRenderPass blit2d_render_passes[NUM_META_FS_KEYS][RADV_META_DST_LAYOUT_COUNT];
+ VkRenderPass blit2d_depth_only_rp[RADV_BLIT_DS_LAYOUT_COUNT];
+ VkRenderPass blit2d_stencil_only_rp[RADV_BLIT_DS_LAYOUT_COUNT];
struct {
VkPipelineLayout img_p_layout;
VkPipeline pipeline;
VkPipeline pipeline_3d;
} btoi;
+ struct {
+ VkPipelineLayout img_p_layout;
+ VkDescriptorSetLayout img_ds_layout;
+ VkPipeline pipeline;
+ } btoi_r32g32b32;
struct {
VkPipelineLayout img_p_layout;
VkDescriptorSetLayout img_ds_layout;
VkPipeline pipeline;
VkPipeline pipeline_3d;
} itoi;
+ struct {
+ VkPipelineLayout img_p_layout;
+ VkDescriptorSetLayout img_ds_layout;
+ VkPipeline pipeline;
+ } itoi_r32g32b32;
struct {
VkPipelineLayout img_p_layout;
VkDescriptorSetLayout img_ds_layout;
VkPipeline pipeline;
VkPipeline pipeline_3d;
} cleari;
+ struct {
+ VkPipelineLayout img_p_layout;
+ VkDescriptorSetLayout img_ds_layout;
+ VkPipeline pipeline;
+ } cleari_r32g32b32;
struct {
VkPipelineLayout p_layout;
VkPipelineLayout p_layout;
VkPipeline occlusion_query_pipeline;
VkPipeline pipeline_statistics_query_pipeline;
+ VkPipeline tfb_query_pipeline;
} query;
+
+ struct {
+ VkDescriptorSetLayout ds_layout;
+ VkPipelineLayout p_layout;
+ VkPipeline pipeline[MAX_SAMPLES_LOG2];
+ } fmask_expand;
};
/* queue types */
struct radeon_winsys_bo *esgs_ring_bo;
struct radeon_winsys_bo *gsvs_ring_bo;
struct radeon_winsys_bo *tess_rings_bo;
- struct radeon_winsys_cs *initial_preamble_cs;
- struct radeon_winsys_cs *initial_full_flush_preamble_cs;
- struct radeon_winsys_cs *continue_preamble_cs;
+ struct radeon_cmdbuf *initial_preamble_cs;
+ struct radeon_cmdbuf *initial_full_flush_preamble_cs;
+ struct radeon_cmdbuf *continue_preamble_cs;
+};
+
+struct radv_bo_list {
+ struct radv_winsys_bo_list list;
+ unsigned capacity;
+ pthread_mutex_t mutex;
};
struct radv_device {
struct radv_queue *queues[RADV_MAX_QUEUE_FAMILIES];
int queue_count[RADV_MAX_QUEUE_FAMILIES];
- struct radeon_winsys_cs *empty_cs[RADV_MAX_QUEUE_FAMILIES];
+ struct radeon_cmdbuf *empty_cs[RADV_MAX_QUEUE_FAMILIES];
bool always_use_syncobj;
- bool llvm_supports_spill;
bool has_distributed_tess;
bool pbb_allowed;
bool dfsm_allowed;
uint64_t dmesg_timestamp;
struct radv_device_extension_table enabled_extensions;
+
+ /* Whether the driver uses a global BO list. */
+ bool use_global_bo_list;
+
+ struct radv_bo_list bo_list;
+
+ /* Whether anisotropy is forced with RADV_TEX_ANISO (-1 is disabled). */
+ int force_aniso;
};
struct radv_device_memory {
RADV_CMD_DIRTY_INDEX_BUFFER = 1 << 11,
RADV_CMD_DIRTY_FRAMEBUFFER = 1 << 12,
RADV_CMD_DIRTY_VERTEX_BUFFER = 1 << 13,
+ RADV_CMD_DIRTY_STREAMOUT_BUFFER = 1 << 14,
};
enum radv_cmd_flush_bits {
RADV_CMD_FLAG_PS_PARTIAL_FLUSH = 1 << 10,
RADV_CMD_FLAG_CS_PARTIAL_FLUSH = 1 << 11,
RADV_CMD_FLAG_VGT_FLUSH = 1 << 12,
+ /* Pipeline query controls. */
+ RADV_CMD_FLAG_START_PIPELINE_STATS = 1 << 13,
+ RADV_CMD_FLAG_STOP_PIPELINE_STATS = 1 << 14,
+ RADV_CMD_FLAG_VGT_STREAMOUT_SYNC = 1 << 15,
RADV_CMD_FLUSH_AND_INV_FRAMEBUFFER = (RADV_CMD_FLAG_FLUSH_AND_INV_CB |
RADV_CMD_FLAG_FLUSH_AND_INV_CB_META |
VkDeviceSize offset;
};
+struct radv_streamout_binding {
+ struct radv_buffer *buffer;
+ VkDeviceSize offset;
+ VkDeviceSize size;
+};
+
+struct radv_streamout_state {
+ /* Mask of bound streamout buffers. */
+ uint8_t enabled_mask;
+
+ /* External state that comes from the last vertex stage, it must be
+ * set explicitely when binding a new graphics pipeline.
+ */
+ uint16_t stride_in_dw[MAX_SO_BUFFERS];
+ uint32_t enabled_stream_buffers_mask; /* stream0 buffers0-3 in 4 LSB */
+
+ /* State of VGT_STRMOUT_BUFFER_(CONFIG|END) */
+ uint32_t hw_enabled_mask;
+
+ /* State of VGT_STRMOUT_(CONFIG|EN) */
+ bool streamout_enabled;
+};
+
struct radv_viewport_state {
uint32_t count;
VkViewport viewports[MAX_VIEWPORTS];
uint32_t valid;
struct radv_push_descriptor_set push_set;
bool push_dirty;
+ uint32_t dynamic_buffers[4 * MAX_DYNAMIC_BUFFERS];
};
struct radv_cmd_state {
const struct radv_subpass * subpass;
struct radv_dynamic_state dynamic;
struct radv_attachment_state * attachments;
+ struct radv_streamout_state streamout;
VkRect2D render_area;
/* Index buffer */
enum radv_cmd_flush_bits flush_bits;
unsigned active_occlusion_queries;
bool perfect_occlusion_queries_enabled;
+ unsigned active_pipeline_queries;
float offset_scale;
uint32_t trace_id;
uint32_t last_ia_multi_vgt_param;
uint32_t last_num_instances;
uint32_t last_first_instance;
uint32_t last_vertex_offset;
+
+ /* Whether CP DMA is busy/idle. */
+ bool dma_is_busy;
+
+ /* Conditional rendering info. */
+ int predication_type; /* -1: disabled, 0: normal, 1: inverted */
+ uint64_t predication_va;
+
+ bool context_roll_without_scissor_emitted;
};
struct radv_cmd_pool {
VkCommandBufferUsageFlags usage_flags;
VkCommandBufferLevel level;
enum radv_cmd_buffer_status status;
- struct radeon_winsys_cs *cs;
+ struct radeon_cmdbuf *cs;
struct radv_cmd_state state;
struct radv_vertex_binding vertex_bindings[MAX_VBS];
+ struct radv_streamout_binding streamout_bindings[MAX_SO_BUFFERS];
uint32_t queue_family_index;
uint8_t push_constants[MAX_PUSH_CONSTANTS_SIZE];
- uint32_t dynamic_buffers[4 * MAX_DYNAMIC_BUFFERS];
VkShaderStageFlags push_constant_stages;
struct radv_descriptor_set meta_push_descriptors;
VkResult record_result;
- int ring_offsets_idx; /* just used for verification */
- uint32_t gfx9_fence_offset;
- struct radeon_winsys_bo *gfx9_fence_bo;
+ uint64_t gfx9_fence_va;
uint32_t gfx9_fence_idx;
+ uint64_t gfx9_eop_bug_va;
/**
* Whether a query pool has been resetted and we have to flush caches.
bool radv_cmd_buffer_uses_mec(struct radv_cmd_buffer *cmd_buffer);
-void si_init_compute(struct radv_cmd_buffer *cmd_buffer);
-void si_init_config(struct radv_cmd_buffer *cmd_buffer);
+void si_emit_graphics(struct radv_physical_device *physical_device,
+ struct radeon_cmdbuf *cs);
+void si_emit_compute(struct radv_physical_device *physical_device,
+ struct radeon_cmdbuf *cs);
void cik_create_gfx_config(struct radv_device *device);
-void si_write_viewport(struct radeon_winsys_cs *cs, int first_vp,
+void si_write_viewport(struct radeon_cmdbuf *cs, int first_vp,
int count, const VkViewport *viewports);
-void si_write_scissors(struct radeon_winsys_cs *cs, int first,
+void si_write_scissors(struct radeon_cmdbuf *cs, int first,
int count, const VkRect2D *scissors,
const VkViewport *viewports, bool can_use_guardband);
uint32_t si_get_ia_multi_vgt_param(struct radv_cmd_buffer *cmd_buffer,
bool instanced_draw, bool indirect_draw,
uint32_t draw_vertex_count);
-void si_cs_emit_write_event_eop(struct radeon_winsys_cs *cs,
- bool predicated,
+void si_cs_emit_write_event_eop(struct radeon_cmdbuf *cs,
enum chip_class chip_class,
bool is_mec,
unsigned event, unsigned event_flags,
unsigned data_sel,
uint64_t va,
- uint32_t old_fence,
- uint32_t new_fence);
-
-void si_emit_wait_fence(struct radeon_winsys_cs *cs,
- bool predicated,
- uint64_t va, uint32_t ref,
- uint32_t mask);
-void si_cs_emit_cache_flush(struct radeon_winsys_cs *cs,
+ uint32_t new_fence,
+ uint64_t gfx9_eop_bug_va);
+
+void radv_cp_wait_mem(struct radeon_cmdbuf *cs, uint32_t op, uint64_t va,
+ uint32_t ref, uint32_t mask);
+void si_cs_emit_cache_flush(struct radeon_cmdbuf *cs,
enum chip_class chip_class,
uint32_t *fence_ptr, uint64_t va,
bool is_mec,
- enum radv_cmd_flush_bits flush_bits);
+ enum radv_cmd_flush_bits flush_bits,
+ uint64_t gfx9_eop_bug_va);
void si_emit_cache_flush(struct radv_cmd_buffer *cmd_buffer);
-void si_emit_set_predication_state(struct radv_cmd_buffer *cmd_buffer, uint64_t va);
+void si_emit_set_predication_state(struct radv_cmd_buffer *cmd_buffer,
+ bool inverted, uint64_t va);
void si_cp_dma_buffer_copy(struct radv_cmd_buffer *cmd_buffer,
uint64_t src_va, uint64_t dest_va,
uint64_t size);
unsigned size);
void si_cp_dma_clear_buffer(struct radv_cmd_buffer *cmd_buffer, uint64_t va,
uint64_t size, unsigned value);
+void si_cp_dma_wait_for_idle(struct radv_cmd_buffer *cmd_buffer);
+
void radv_set_db_count_control(struct radv_cmd_buffer *cmd_buffer);
bool
radv_cmd_buffer_upload_alloc(struct radv_cmd_buffer *cmd_buffer,
void **ptr);
void
radv_cmd_buffer_set_subpass(struct radv_cmd_buffer *cmd_buffer,
- const struct radv_subpass *subpass,
- bool transitions);
+ const struct radv_subpass *subpass);
bool
radv_cmd_buffer_upload_data(struct radv_cmd_buffer *cmd_buffer,
unsigned size, unsigned alignmnet,
void radv_cmd_buffer_resolve_subpass(struct radv_cmd_buffer *cmd_buffer);
void radv_cmd_buffer_resolve_subpass_cs(struct radv_cmd_buffer *cmd_buffer);
void radv_cmd_buffer_resolve_subpass_fs(struct radv_cmd_buffer *cmd_buffer);
-void radv_cayman_emit_msaa_sample_locs(struct radeon_winsys_cs *cs, int nr_samples);
+void radv_cayman_emit_msaa_sample_locs(struct radeon_cmdbuf *cs, int nr_samples);
unsigned radv_cayman_get_maxdist(int log_samples);
void radv_device_init_msaa(struct radv_device *device);
-void radv_set_depth_clear_regs(struct radv_cmd_buffer *cmd_buffer,
- struct radv_image *image,
- VkClearDepthStencilValue ds_clear_value,
- VkImageAspectFlags aspects);
-void radv_set_color_clear_regs(struct radv_cmd_buffer *cmd_buffer,
- struct radv_image *image,
- int idx,
- uint32_t color_values[2]);
-void radv_set_dcc_need_cmask_elim_pred(struct radv_cmd_buffer *cmd_buffer,
- struct radv_image *image,
- bool value);
+
+void radv_update_ds_clear_metadata(struct radv_cmd_buffer *cmd_buffer,
+ struct radv_image *image,
+ VkClearDepthStencilValue ds_clear_value,
+ VkImageAspectFlags aspects);
+
+void radv_update_color_clear_metadata(struct radv_cmd_buffer *cmd_buffer,
+ struct radv_image *image,
+ int cb_idx,
+ uint32_t color_values[2]);
+
+void radv_update_fce_metadata(struct radv_cmd_buffer *cmd_buffer,
+ struct radv_image *image, bool value);
+
+void radv_update_dcc_metadata(struct radv_cmd_buffer *cmd_buffer,
+ struct radv_image *image, bool value);
+
uint32_t radv_fill_buffer(struct radv_cmd_buffer *cmd_buffer,
struct radeon_winsys_bo *bo,
uint64_t offset, uint64_t size, uint32_t value);
struct radv_device_memory *memory,
int *pFD);
+static inline void
+radv_emit_shader_pointer_head(struct radeon_cmdbuf *cs,
+ unsigned sh_offset, unsigned pointer_count,
+ bool use_32bit_pointers)
+{
+ radeon_emit(cs, PKT3(PKT3_SET_SH_REG, pointer_count * (use_32bit_pointers ? 1 : 2), 0));
+ radeon_emit(cs, (sh_offset - SI_SH_REG_OFFSET) >> 2);
+}
+
+static inline void
+radv_emit_shader_pointer_body(struct radv_device *device,
+ struct radeon_cmdbuf *cs,
+ uint64_t va, bool use_32bit_pointers)
+{
+ radeon_emit(cs, va);
+
+ if (use_32bit_pointers) {
+ assert(va == 0 ||
+ (va >> 32) == device->physical_device->rad_info.address32_hi);
+ } else {
+ radeon_emit(cs, va >> 32);
+ }
+}
+
+static inline void
+radv_emit_shader_pointer(struct radv_device *device,
+ struct radeon_cmdbuf *cs,
+ uint32_t sh_offset, uint64_t va, bool global)
+{
+ bool use_32bit_pointers = !global;
+
+ radv_emit_shader_pointer_head(cs, sh_offset, 1, use_32bit_pointers);
+ radv_emit_shader_pointer_body(device, cs, va, use_32bit_pointers);
+}
+
static inline struct radv_descriptor_state *
radv_get_descriptors_state(struct radv_cmd_buffer *cmd_buffer,
VkPipelineBindPoint bind_point)
stage = __builtin_ffs(__tmp) - 1, __tmp; \
__tmp &= ~(1 << (stage)))
+extern const VkFormat radv_fs_key_format_exemplars[NUM_META_FS_KEYS];
unsigned radv_format_meta_fs_key(VkFormat format);
struct radv_multisample_state {
};
struct radv_vertex_elements_info {
- uint32_t rsrc_word3[MAX_VERTEX_ATTRIBS];
uint32_t format_size[MAX_VERTEX_ATTRIBS];
- uint32_t binding[MAX_VERTEX_ATTRIBS];
- uint32_t offset[MAX_VERTEX_ATTRIBS];
- uint32_t count;
};
struct radv_ia_multi_vgt_param_helpers {
struct radv_shader_variant *gs_copy_shader;
VkShaderStageFlags active_stages;
- struct radeon_winsys_cs cs;
+ struct radeon_cmdbuf cs;
+ 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;
uint32_t user_data_0[MESA_SHADER_STAGES];
union {
bool can_use_guardband;
uint32_t needed_dynamic_state;
bool disable_out_of_order_rast_for_occlusion;
+
+ /* Used for rbplus */
+ uint32_t col_format;
+ uint32_t cb_target_mask;
} graphics;
};
unsigned max_waves;
unsigned scratch_bytes_per_wave;
+
+ /* Not NULL if graphics pipeline uses streamout. */
+ struct radv_shader_variant *streamout_shader;
};
static inline bool radv_pipeline_has_gs(const struct radv_pipeline *pipeline)
gl_shader_stage stage,
int idx);
-struct radv_shader_variant *radv_get_vertex_shader(struct radv_pipeline *pipeline);
+struct radv_shader_variant *radv_get_shader(struct radv_pipeline *pipeline,
+ gl_shader_stage stage);
struct radv_graphics_pipeline_create_info {
bool use_rectlist;
bool radv_is_colorbuffer_format_supported(VkFormat format, bool *blendable);
bool radv_dcc_formats_compatible(VkFormat format1,
VkFormat format2);
+bool radv_device_supports_etc(struct radv_physical_device *physical_device);
struct radv_fmask_info {
uint64_t offset;
struct radv_fmask_info fmask;
struct radv_cmask_info cmask;
uint64_t clear_value_offset;
+ uint64_t fce_pred_offset;
uint64_t dcc_pred_offset;
+ /*
+ * Metadata for the TC-compat zrange workaround. If the 32-bit value
+ * stored at this offset is UINT_MAX, the driver will emit
+ * DB_Z_INFO.ZRANGE_PRECISION=0, otherwise it will skip the
+ * SET_CONTEXT_REG packet.
+ */
+ uint64_t tc_compat_zrange_offset;
+
/* For VK_ANDROID_native_buffer, the WSI image owns the memory, */
VkDeviceMemory owned_memory;
};
level < image->surface.num_dcc_levels;
}
+/**
+ * Return whether the image has CB metadata.
+ */
+static inline bool
+radv_image_has_CB_metadata(const struct radv_image *image)
+{
+ return radv_image_has_cmask(image) ||
+ radv_image_has_fmask(image) ||
+ radv_image_has_dcc(image);
+}
+
/**
* Return whether the image has HTILE metadata for depth surfaces.
*/
return radv_image_has_htile(image) && level == 0;
}
+/**
+ * Return whether the image is TC-compatible HTILE.
+ */
+static inline bool
+radv_image_is_tc_compat_htile(const struct radv_image *image)
+{
+ return radv_image_has_htile(image) && image->tc_compatible_htile;
+}
+
unsigned radv_image_queue_family_mask(const struct radv_image *image, uint32_t family, uint32_t queue_family);
static inline uint32_t
VkAccessFlags dst_access_mask;
};
+void radv_subpass_barrier(struct radv_cmd_buffer *cmd_buffer,
+ const struct radv_subpass_barrier *barrier);
+
+struct radv_subpass_attachment {
+ uint32_t attachment;
+ VkImageLayout layout;
+};
+
struct radv_subpass {
+ uint32_t attachment_count;
+ struct radv_subpass_attachment * attachments;
+
uint32_t input_count;
uint32_t color_count;
- VkAttachmentReference * input_attachments;
- VkAttachmentReference * color_attachments;
- VkAttachmentReference * resolve_attachments;
- VkAttachmentReference depth_stencil_attachment;
+ struct radv_subpass_attachment * input_attachments;
+ struct radv_subpass_attachment * color_attachments;
+ struct radv_subpass_attachment * resolve_attachments;
+ struct radv_subpass_attachment * depth_stencil_attachment;
/** Subpass has at least one resolve attachment */
bool has_resolve;
+ /** Subpass has at least one color attachment */
+ bool has_color_att;
+
struct radv_subpass_barrier start_barrier;
uint32_t view_mask;
VkAttachmentLoadOp stencil_load_op;
VkImageLayout initial_layout;
VkImageLayout final_layout;
- uint32_t view_mask;
+
+ /* The subpass id in which the attachment will be used last. */
+ uint32_t last_subpass_idx;
};
struct radv_render_pass {
uint32_t attachment_count;
uint32_t subpass_count;
- VkAttachmentReference * subpass_attachments;
+ struct radv_subpass_attachment * subpass_attachments;
struct radv_render_pass_attachment * attachments;
struct radv_subpass_barrier end_barrier;
struct radv_subpass subpasses[0];
uint32_t temp_syncobj;
};
-VkResult radv_alloc_sem_info(struct radv_winsys_sem_info *sem_info,
- int num_wait_sems,
- const VkSemaphore *wait_sems,
- int num_signal_sems,
- const VkSemaphore *signal_sems,
- VkFence fence);
-void radv_free_sem_info(struct radv_winsys_sem_info *sem_info);
-
void radv_set_descriptor_set(struct radv_cmd_buffer *cmd_buffer,
VkPipelineBindPoint bind_point,
struct radv_descriptor_set *set,
radv_update_descriptor_set_with_template(struct radv_device *device,
struct radv_cmd_buffer *cmd_buffer,
struct radv_descriptor_set *set,
- VkDescriptorUpdateTemplateKHR descriptorUpdateTemplate,
+ VkDescriptorUpdateTemplate descriptorUpdateTemplate,
const void *pData);
void radv_meta_push_descriptor_set(struct radv_cmd_buffer *cmd_buffer,
uint32_t descriptorWriteCount,
const VkWriteDescriptorSet *pDescriptorWrites);
-void radv_initialise_cmask(struct radv_cmd_buffer *cmd_buffer,
- struct radv_image *image, uint32_t value);
void radv_initialize_dcc(struct radv_cmd_buffer *cmd_buffer,
struct radv_image *image, uint32_t value);
+void radv_initialize_fmask(struct radv_cmd_buffer *cmd_buffer,
+ struct radv_image *image);
+
struct radv_fence {
struct radeon_winsys_fence *fence;
+ struct wsi_fence *fence_wsi;
bool submitted;
bool signalled;
struct radv_shader_variant_info;
struct radv_nir_compiler_options;
-void radv_compile_gs_copy_shader(LLVMTargetMachineRef tm,
+void radv_compile_gs_copy_shader(struct ac_llvm_compiler *ac_llvm,
struct nir_shader *geom_shader,
struct ac_shader_binary *binary,
struct ac_shader_config *config,
struct radv_shader_variant_info *shader_info,
const struct radv_nir_compiler_options *option);
-void radv_compile_nir_shader(LLVMTargetMachineRef tm,
+void radv_compile_nir_shader(struct ac_llvm_compiler *ac_llvm,
struct ac_shader_binary *binary,
struct ac_shader_config *config,
struct radv_shader_variant_info *shader_info,
int nir_count,
const struct radv_nir_compiler_options *options);
+unsigned radv_nir_get_max_workgroup_size(enum chip_class chip_class,
+ const struct nir_shader *nir);
+
/* radv_shader_info.h */
struct radv_shader_info;
const struct radv_nir_compiler_options *options,
struct radv_shader_info *info);
+void radv_nir_shader_info_init(struct radv_shader_info *info);
+
struct radeon_winsys_sem;
+uint64_t radv_get_current_time(void);
+
#define RADV_DEFINE_HANDLE_CASTS(__radv_type, __VkType) \
\
static inline struct __radv_type * \
RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_pool, VkDescriptorPool)
RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_set, VkDescriptorSet)
RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_set_layout, VkDescriptorSetLayout)
-RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_update_template, VkDescriptorUpdateTemplateKHR)
+RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_descriptor_update_template, VkDescriptorUpdateTemplate)
RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_device_memory, VkDeviceMemory)
RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_fence, VkFence)
RADV_DEFINE_NONDISP_HANDLE_CASTS(radv_event, VkEvent)