#include <stdio.h>
-#include "radeon/drm/radeon_winsys.h"
+#include "radeon/radeon_winsys.h"
#include "util/u_blitter.h"
-#include "util/u_double_list.h"
+#include "util/list.h"
#include "util/u_range.h"
#include "util/u_slab.h"
#include "util/u_suballoc.h"
#define R600_QUERY_NUM_BYTES_MOVED (PIPE_QUERY_DRIVER_SPECIFIC + 5)
#define R600_QUERY_VRAM_USAGE (PIPE_QUERY_DRIVER_SPECIFIC + 6)
#define R600_QUERY_GTT_USAGE (PIPE_QUERY_DRIVER_SPECIFIC + 7)
+#define R600_QUERY_GPU_TEMPERATURE (PIPE_QUERY_DRIVER_SPECIFIC + 8)
+#define R600_QUERY_CURRENT_GPU_SCLK (PIPE_QUERY_DRIVER_SPECIFIC + 9)
+#define R600_QUERY_CURRENT_GPU_MCLK (PIPE_QUERY_DRIVER_SPECIFIC + 10)
+#define R600_QUERY_GPU_LOAD (PIPE_QUERY_DRIVER_SPECIFIC + 11)
+#define R600_QUERY_NUM_COMPILATIONS (PIPE_QUERY_DRIVER_SPECIFIC + 12)
+#define R600_QUERY_NUM_SHADERS_CREATED (PIPE_QUERY_DRIVER_SPECIFIC + 13)
-/* read caches */
-#define R600_CONTEXT_INV_VERTEX_CACHE (1 << 0)
-#define R600_CONTEXT_INV_TEX_CACHE (1 << 1)
-#define R600_CONTEXT_INV_CONST_CACHE (1 << 2)
-#define R600_CONTEXT_INV_SHADER_CACHE (1 << 3)
-/* read-write caches */
-#define R600_CONTEXT_STREAMOUT_FLUSH (1 << 8)
-#define R600_CONTEXT_FLUSH_AND_INV (1 << 9)
-#define R600_CONTEXT_FLUSH_AND_INV_CB_META (1 << 10)
-#define R600_CONTEXT_FLUSH_AND_INV_DB_META (1 << 11)
-#define R600_CONTEXT_FLUSH_AND_INV_DB (1 << 12)
-#define R600_CONTEXT_FLUSH_AND_INV_CB (1 << 13)
-#define R600_CONTEXT_FLUSH_WITH_INV_L2 (1 << 14)
-/* engine synchronization */
-#define R600_CONTEXT_PS_PARTIAL_FLUSH (1 << 16)
-#define R600_CONTEXT_WAIT_3D_IDLE (1 << 17)
-#define R600_CONTEXT_WAIT_CP_DMA_IDLE (1 << 18)
-#define R600_CONTEXT_VGT_FLUSH (1 << 19)
-#define R600_CONTEXT_VGT_STREAMOUT_SYNC (1 << 20)
-#define R600_CONTEXT_CS_PARTIAL_FLUSH (1 << 21)
-/* other flags */
-#define R600_CONTEXT_FLAG_COMPUTE (1u << 31)
+#define R600_CONTEXT_STREAMOUT_FLUSH (1u << 0)
+#define R600_CONTEXT_PRIVATE_FLAG (1u << 1)
/* special primitive types */
#define R600_PRIM_RECTANGLE_LIST PIPE_PRIM_MAX
#define DBG_GS (1 << 7)
#define DBG_PS (1 << 8)
#define DBG_CS (1 << 9)
+#define DBG_TCS (1 << 10)
+#define DBG_TES (1 << 11)
+#define DBG_NO_IR (1 << 12)
+#define DBG_NO_TGSI (1 << 13)
+#define DBG_NO_ASM (1 << 14)
+/* Bits 21-31 are reserved for the r600g driver. */
/* features */
-#define DBG_NO_ASYNC_DMA (1 << 10)
-#define DBG_HYPERZ (1 << 11)
-#define DBG_NO_DISCARD_RANGE (1 << 12)
-#define DBG_NO_2D_TILING (1 << 13)
-#define DBG_NO_TILING (1 << 14)
-#define DBG_SWITCH_ON_EOP (1 << 15)
-#define DBG_FORCE_DMA (1 << 16)
-/* The maximum allowed bit is 20. */
+#define DBG_NO_ASYNC_DMA (1llu << 32)
+#define DBG_NO_HYPERZ (1llu << 33)
+#define DBG_NO_DISCARD_RANGE (1llu << 34)
+#define DBG_NO_2D_TILING (1llu << 35)
+#define DBG_NO_TILING (1llu << 36)
+#define DBG_SWITCH_ON_EOP (1llu << 37)
+#define DBG_FORCE_DMA (1llu << 38)
+#define DBG_PRECOMPILE (1llu << 39)
+#define DBG_INFO (1llu << 40)
+#define DBG_NO_WC (1llu << 41)
+#define DBG_CHECK_VM (1llu << 42)
+#define DBG_NO_DCC (1llu << 43)
+#define DBG_NO_DCC_CLEAR (1llu << 44)
#define R600_MAP_BUFFER_ALIGNMENT 64
struct r600_common_context;
+struct radeon_shader_reloc {
+ char *name;
+ uint64_t offset;
+};
+
struct radeon_shader_binary {
/** Shader code */
unsigned char *code;
unsigned char *config;
unsigned config_size;
+ /** The number of bytes of config information for each global symbol.
+ */
+ unsigned config_size_per_symbol;
+
/** Constant data accessed by the shader. This will be uploaded
* into a constant buffer. */
unsigned char *rodata;
unsigned rodata_size;
- /** Set to 1 if the disassembly for this binary has been dumped to
- * stderr. */
- int disassembled;
+ /** List of symbol offsets for the shader */
+ uint64_t *global_symbol_offsets;
+ unsigned global_symbol_count;
+
+ struct radeon_shader_reloc *relocs;
+ unsigned reloc_count;
+
+ /** Disassembled shader in a string. */
+ char *disasm_string;
};
struct r600_resource {
* the unsynchronized map flag and expect the driver to figure it out.
*/
struct util_range valid_buffer_range;
+
+ /* For buffers only. This indicates that a write operation has been
+ * performed by TC L2, but the cache hasn't been flushed.
+ * Any hw block which doesn't use or bypasses TC L2 should check this
+ * flag and flush the cache before using the buffer.
+ *
+ * For example, TC L2 must be flushed if a buffer which has been
+ * modified by a shader store instruction is about to be used as
+ * an index buffer. The reason is that VGT DMA index fetching doesn't
+ * use TC L2.
+ */
+ bool TC_L2_dirty;
};
struct r600_transfer {
unsigned pitch_override;
bool is_depth;
unsigned dirty_level_mask; /* each bit says if that mipmap is compressed */
+ unsigned stencil_dirty_level_mask; /* each bit says if that mipmap is compressed */
struct r600_texture *flushed_depth_texture;
boolean is_flushing_texture;
- struct radeon_surface surface;
+ struct radeon_surf surface;
/* Colorbuffer compression and fast clear. */
struct r600_fmask_info fmask;
struct r600_cmask_info cmask;
struct r600_resource *cmask_buffer;
+ struct r600_resource *dcc_buffer;
unsigned cb_color_info; /* fast clear enable bit */
unsigned color_clear_value[2];
float depth_clear_value;
bool non_disp_tiling; /* R600-Cayman only */
- unsigned mipmap_shift;
};
struct r600_surface {
unsigned cb_color_dim; /* EG only */
unsigned cb_color_pitch; /* EG and later */
unsigned cb_color_slice; /* EG and later */
+ unsigned cb_dcc_base; /* VI and later */
unsigned cb_color_attrib; /* EG and later */
+ unsigned cb_dcc_control; /* VI and later */
unsigned cb_color_fmask; /* CB_COLORn_FMASK (EG and later) or CB_COLORn_FRAG (r600) */
unsigned cb_color_fmask_slice; /* EG and later */
unsigned cb_color_cmask; /* CB_COLORn_TILE (r600 only) */
enum chip_class chip_class;
struct radeon_info info;
struct r600_tiling_info tiling_info;
- unsigned debug_flags;
+ uint64_t debug_flags;
bool has_cp_dma;
bool has_streamout;
struct r600_resource *trace_bo;
uint32_t *trace_ptr;
unsigned cs_count;
+
+ /* This must be in the screen, because UE4 uses one context for
+ * compilation and another one for rendering.
+ */
+ unsigned num_compilations;
+ /* Along with ST_DEBUG=precompile, this should show if applications
+ * are loading shaders on demand. This is a monotonic counter.
+ */
+ unsigned num_shaders_created;
+
+ /* GPU load thread. */
+ pipe_mutex gpu_load_mutex;
+ pipe_thread gpu_load_thread;
+ unsigned gpu_load_counter_busy;
+ unsigned gpu_load_counter_idle;
+ volatile unsigned gpu_load_stop_thread; /* bool */
+
+ char renderer_string[64];
};
/* This encapsulates a state or an operation which can emitted into the GPU
struct r600_atom {
void (*emit)(struct r600_common_context *ctx, struct r600_atom *state);
unsigned num_dw;
- bool dirty;
+ unsigned short id;
};
struct r600_so_target {
/* The buffer where BUFFER_FILLED_SIZE is stored. */
struct r600_resource *buf_filled_size;
unsigned buf_filled_size_offset;
+ bool buf_filled_size_valid;
unsigned stride_in_dw;
};
/* External state which comes from the vertex shader,
* it must be set explicitly when binding a shader. */
unsigned *stride_in_dw;
+ unsigned enabled_stream_buffers_mask; /* stream0 buffers0-3 in 4 LSB */
+
+ /* The state of VGT_STRMOUT_BUFFER_(CONFIG|EN). */
+ unsigned hw_enabled_mask;
/* The state of VGT_STRMOUT_(CONFIG|EN). */
struct r600_atom enable_atom;
struct r600_ring {
struct radeon_winsys_cs *cs;
- bool flushing;
void (*flush)(void *ctx, unsigned flags,
struct pipe_fence_handle **fence);
};
-struct r600_rings {
- struct r600_ring gfx;
- struct r600_ring dma;
-};
-
struct r600_common_context {
struct pipe_context b; /* base class */
struct r600_common_screen *screen;
struct radeon_winsys *ws;
+ struct radeon_winsys_ctx *ctx;
enum radeon_family family;
enum chip_class chip_class;
- struct r600_rings rings;
+ struct r600_ring gfx;
+ struct r600_ring dma;
+ struct pipe_fence_handle *last_sdma_fence;
unsigned initial_gfx_cs_size;
+ unsigned gpu_reset_counter;
struct u_upload_mgr *uploader;
struct u_suballocator *allocator_so_filled_size;
/* Queries. */
/* The list of active queries. Only one query of each type can be active. */
int num_occlusion_queries;
- int num_pipelinestat_queries;
/* Keep track of non-timer queries, because they should be suspended
* during context flushing.
- * The timer queries (TIME_ELAPSED) shouldn't be suspended. */
+ * The timer queries (TIME_ELAPSED) shouldn't be suspended for blits,
+ * but they should be suspended between IBs. */
struct list_head active_nontimer_queries;
+ struct list_head active_timer_queries;
unsigned num_cs_dw_nontimer_queries_suspend;
+ unsigned num_cs_dw_timer_queries_suspend;
/* If queries have been suspended. */
- bool nontimer_queries_suspended;
+ bool queries_suspended_for_flush;
/* Additional hardware info. */
unsigned backend_mask;
unsigned max_db; /* for OQ */
struct pipe_query *current_render_cond;
unsigned current_render_cond_mode;
boolean current_render_cond_cond;
- boolean predicate_drawing;
+ bool predicate_drawing;
+ bool render_cond_force_off; /* for u_blitter */
/* For context flushing. */
struct pipe_query *saved_render_cond;
boolean saved_render_cond_cond;
const struct pipe_box *src_box);
void (*clear_buffer)(struct pipe_context *ctx, struct pipe_resource *dst,
- unsigned offset, unsigned size, unsigned value);
+ unsigned offset, unsigned size, unsigned value,
+ bool is_framebuffer);
void (*blit_decompress_depth)(struct pipe_context *ctx,
struct r600_texture *texture,
/* This ensures there is enough space in the command stream. */
void (*need_gfx_cs_space)(struct pipe_context *ctx, unsigned num_dw,
bool include_draw_vbo);
+
+ void (*set_atom_dirty)(struct r600_common_context *ctx,
+ struct r600_atom *atom, bool dirty);
};
/* r600_buffer.c */
struct pipe_resource *r600_buffer_create(struct pipe_screen *screen,
const struct pipe_resource *templ,
unsigned alignment);
+struct pipe_resource * r600_aligned_buffer_create(struct pipe_screen *screen,
+ unsigned bind,
+ unsigned usage,
+ unsigned size,
+ unsigned alignment);
+struct pipe_resource *
+r600_buffer_from_user_memory(struct pipe_screen *screen,
+ const struct pipe_resource *templ,
+ void *user_memory);
/* r600_common_pipe.c */
void r600_draw_rectangle(struct blitter_context *blitter,
bool r600_can_dump_shader(struct r600_common_screen *rscreen,
const struct tgsi_token *tokens);
void r600_screen_clear_buffer(struct r600_common_screen *rscreen, struct pipe_resource *dst,
- unsigned offset, unsigned size, unsigned value);
+ unsigned offset, unsigned size, unsigned value,
+ bool is_framebuffer);
struct pipe_resource *r600_resource_create_common(struct pipe_screen *screen,
const struct pipe_resource *templ);
const char *r600_get_llvm_processor_name(enum radeon_family family);
void r600_need_dma_space(struct r600_common_context *ctx, unsigned num_dw);
+/* r600_gpu_load.c */
+void r600_gpu_load_kill_thread(struct r600_common_screen *rscreen);
+uint64_t r600_gpu_load_begin(struct r600_common_screen *rscreen);
+unsigned r600_gpu_load_end(struct r600_common_screen *rscreen, uint64_t begin);
+
/* r600_query.c */
void r600_query_init(struct r600_common_context *rctx);
void r600_suspend_nontimer_queries(struct r600_common_context *ctx);
void r600_resume_nontimer_queries(struct r600_common_context *ctx);
+void r600_suspend_timer_queries(struct r600_common_context *ctx);
+void r600_resume_timer_queries(struct r600_common_context *ctx);
void r600_query_init_backend_mask(struct r600_common_context *ctx);
/* r600_streamout.c */
void evergreen_do_fast_color_clear(struct r600_common_context *rctx,
struct pipe_framebuffer_state *fb,
struct r600_atom *fb_state,
- unsigned *buffers,
+ unsigned *buffers, unsigned *dirty_cbufs,
const union pipe_color_union *color);
void r600_init_screen_texture_functions(struct r600_common_screen *rscreen);
void r600_init_context_texture_functions(struct r600_common_context *rctx);
void cayman_init_msaa(struct pipe_context *ctx);
void cayman_emit_msaa_sample_locs(struct radeon_winsys_cs *cs, int nr_samples);
void cayman_emit_msaa_config(struct radeon_winsys_cs *cs, int nr_samples,
- int ps_iter_samples);
+ int ps_iter_samples, int overrast_samples);
/* Inline helpers. */
-static INLINE struct r600_resource *r600_resource(struct pipe_resource *r)
+static inline struct r600_resource *r600_resource(struct pipe_resource *r)
{
return (struct r600_resource*)r;
}
-static INLINE void
+static inline void
r600_resource_reference(struct r600_resource **ptr, struct r600_resource *res)
{
pipe_resource_reference((struct pipe_resource **)ptr,
/* else */ return 4;
}
+static inline unsigned r600_wavefront_size(enum radeon_family family)
+{
+ switch (family) {
+ case CHIP_RV610:
+ case CHIP_RS780:
+ case CHIP_RV620:
+ case CHIP_RS880:
+ return 16;
+ case CHIP_RV630:
+ case CHIP_RV635:
+ case CHIP_RV730:
+ case CHIP_RV710:
+ case CHIP_PALM:
+ case CHIP_CEDAR:
+ return 32;
+ default:
+ return 64;
+ }
+}
+
+static inline enum radeon_bo_priority
+r600_get_sampler_view_priority(struct r600_resource *res)
+{
+ if (res->b.b.target == PIPE_BUFFER)
+ return RADEON_PRIO_SAMPLER_BUFFER;
+
+ if (res->b.b.nr_samples > 1)
+ return RADEON_PRIO_SAMPLER_TEXTURE_MSAA;
+
+ return RADEON_PRIO_SAMPLER_TEXTURE;
+}
+
#define COMPUTE_DBG(rscreen, fmt, args...) \
do { \
if ((rscreen->b.debug_flags & DBG_COMPUTE)) fprintf(stderr, fmt, ##args); \