#include "main/macros.h"
#include "main/mtypes.h"
#include "brw_structs.h"
-#include "brw_compiler.h"
-#include "intel_aub.h"
+#include "compiler/brw_compiler.h"
#include "isl/isl.h"
#include "blorp/blorp.h"
-#include <intel_bufmgr.h>
+#include <brw_bufmgr.h>
-#include "intel_debug.h"
+#include "common/gen_debug.h"
#include "intel_screen.h"
#include "intel_tex_obj.h"
-#include "intel_resolve_map.h"
#ifdef __cplusplus
extern "C" {
BRW_STATE_GEOMETRY_PROGRAM,
BRW_STATE_TESS_PROGRAMS,
BRW_STATE_VERTEX_PROGRAM,
- BRW_STATE_CURBE_OFFSETS,
BRW_STATE_REDUCED_PRIMITIVE,
BRW_STATE_PATCH_PRIMITIVE,
BRW_STATE_PRIMITIVE,
#define BRW_NEW_GEOMETRY_PROGRAM (1ull << BRW_STATE_GEOMETRY_PROGRAM)
#define BRW_NEW_TESS_PROGRAMS (1ull << BRW_STATE_TESS_PROGRAMS)
#define BRW_NEW_VERTEX_PROGRAM (1ull << BRW_STATE_VERTEX_PROGRAM)
-#define BRW_NEW_CURBE_OFFSETS (1ull << BRW_STATE_CURBE_OFFSETS)
#define BRW_NEW_REDUCED_PRIMITIVE (1ull << BRW_STATE_REDUCED_PRIMITIVE)
#define BRW_NEW_PATCH_PRIMITIVE (1ull << BRW_STATE_PATCH_PRIMITIVE)
#define BRW_NEW_PRIMITIVE (1ull << BRW_STATE_PRIMITIVE)
};
-struct brw_sf_prog_data {
- GLuint urb_read_length;
- GLuint total_grf;
-
- /* Each vertex may have upto 12 attributes, 4 components each,
- * except WPOS which requires only 2. (11*4 + 2) == 44 ==> 11
- * rows.
- *
- * Actually we use 4 for each, so call it 12 rows.
- */
- GLuint urb_entry_size;
-};
-
-
-struct brw_clip_prog_data {
- GLuint curb_read_length; /* user planes? */
- GLuint clip_mode;
- GLuint urb_read_length;
- GLuint total_grf;
-};
-
struct brw_ff_gs_prog_data {
GLuint urb_read_length;
GLuint total_grf;
struct brw_context *brw;
struct brw_cache_item **items;
- drm_intel_bo *bo;
+ struct brw_bo *bo;
+ void *map;
GLuint size, n_items;
uint32_t next_offset;
bool bo_used_by_gpu;
};
-
/* Considered adding a member to this struct to document which flags
* an update might raise so that ordering of the state atoms can be
* checked or derived at runtime. Dropped the idea in favor of having
struct brw_vertex_buffer {
/** Buffer object containing the uploaded vertex data */
- drm_intel_bo *bo;
+ struct brw_bo *bo;
uint32_t offset;
uint32_t size;
/** Byte stride between elements in the uploaded array */
struct gl_query_object Base;
/** Last query BO associated with this query. */
- drm_intel_bo *bo;
+ struct brw_bo *bo;
/** Last index in bo with query data for this object. */
int last_index;
struct intel_batchbuffer {
/** Current batchbuffer being queued up. */
- drm_intel_bo *bo;
+ struct brw_bo *bo;
/** Last BO submitted to the hardware. Used for glFinish(). */
- drm_intel_bo *last_bo;
+ struct brw_bo *last_bo;
#ifdef DEBUG
uint16_t emit, total;
bool needs_sol_reset;
bool state_base_address_emitted;
+ struct drm_i915_gem_relocation_entry *relocs;
+ int reloc_count;
+ int reloc_array_size;
+ /** The validation list */
+ struct drm_i915_gem_exec_object2 *exec_objects;
+ struct brw_bo **exec_bos;
+ int exec_count;
+ int exec_array_size;
+ /** The amount of aperture space (in bytes) used by all exec_bos */
+ int aperture_space;
+
struct {
uint32_t *map_next;
int reloc_count;
+ int exec_count;
} saved;
+
+ /** Map from batch offset to brw_state_batch data (with DEBUG_BATCH) */
+ struct hash_table *state_batch_sizes;
};
#define BRW_MAX_XFB_STREAMS 4
struct gl_transform_feedback_object base;
/** A buffer to hold SO_WRITE_OFFSET(n) values while paused. */
- drm_intel_bo *offset_bo;
+ struct brw_bo *offset_bo;
/** If true, SO_WRITE_OFFSET(n) should be reset to zero at next use. */
bool zero_offsets;
* @{
*/
uint64_t prims_generated[BRW_MAX_XFB_STREAMS];
- drm_intel_bo *prim_count_bo;
+ struct brw_bo *prim_count_bo;
unsigned prim_count_buffer_index; /**< in number of uint64_t units */
/** @} */
* unless you're taking additional measures to synchronize thread execution
* across slot size changes.
*/
- drm_intel_bo *scratch_bo;
+ struct brw_bo *scratch_bo;
/**
* Scratch slot size allocated for each thread in the buffer object given
/* In this case whether to draw or not depends on the result of an
* MI_PREDICATE command so the predicate enable bit needs to be checked.
*/
- BRW_PREDICATE_STATE_USE_BIT
+ BRW_PREDICATE_STATE_USE_BIT,
+ /* In this case, either MI_PREDICATE doesn't exist or we lack the
+ * necessary kernel features to use it. Stall for the query result.
+ */
+ BRW_PREDICATE_STATE_STALL_FOR_QUERY,
};
struct shader_times;
struct gen_l3_config;
enum brw_query_kind {
+ OA_COUNTERS,
PIPELINE_STATS
};
{
enum brw_query_kind kind;
const char *name;
+ const char *guid;
struct brw_perf_query_counter *counters;
int n_counters;
size_t data_size;
+
+ /* OA specific */
+ uint64_t oa_metrics_set_id;
+ int oa_format;
+
+ /* For indexing into the accumulator[] ... */
+ int gpu_time_offset;
+ int gpu_clock_offset;
+ int a_offset;
+ int b_offset;
+ int c_offset;
};
/**
uint32_t width, uint32_t height,
uint32_t tile_x, uint32_t tile_y);
+ /**
+ * Emit an MI_REPORT_PERF_COUNT command packet.
+ *
+ * This asks the GPU to write a report of the current OA counter values
+ * into @bo at the given offset and containing the given @report_id
+ * which we can cross-reference when parsing the report (gen7+ only).
+ */
+ void (*emit_mi_report_perf_count)(struct brw_context *brw,
+ struct brw_bo *bo,
+ uint32_t offset_in_bytes,
+ uint32_t report_id);
} vtbl;
- dri_bufmgr *bufmgr;
+ struct brw_bufmgr *bufmgr;
- drm_intel_context *hw_ctx;
+ uint32_t hw_ctx;
/** BO for post-sync nonzero writes for gen6 workaround. */
- drm_intel_bo *workaround_bo;
+ struct brw_bo *workaround_bo;
uint8_t pipe_controls_since_last_cs_stall;
/**
- * Set of drm_intel_bo * that have been rendered to within this batchbuffer
+ * Set of struct brw_bo * that have been rendered to within this batchbuffer
* and would need flushing before being used from another cache domain that
* isn't coherent with it (i.e. the sampler).
*/
bool no_batch_wrap;
struct {
- drm_intel_bo *bo;
+ struct brw_bo *bo;
+ void *map;
uint32_t next_offset;
} upload;
bool front_buffer_dirty;
/** Framerate throttling: @{ */
- drm_intel_bo *throttle_batch[2];
+ struct brw_bo *throttle_batch[2];
/* Limit the number of outstanding SwapBuffers by waiting for an earlier
* frame of rendering to complete. This gives a very precise cap to the
bool has_negative_rhw_bug;
bool has_pln;
bool no_simd8;
- bool use_rep_send;
- bool use_resource_streamer;
/**
* Some versions of Gen hardware don't do centroid interpolation correctly
*/
bool needs_unlit_centroid_workaround;
+ /** Derived stencil states. */
+ bool stencil_enabled;
+ bool stencil_two_sided;
+ bool stencil_write_enabled;
+ /** Derived polygon state. */
+ bool polygon_front_bit; /**< 0=GL_CCW, 1=GL_CW */
+
struct isl_device isl_dev;
struct blorp_context blorp;
* Buffer and offset used for GL_ARB_shader_draw_parameters
* (for now, only gl_BaseVertex).
*/
- drm_intel_bo *draw_params_bo;
+ struct brw_bo *draw_params_bo;
uint32_t draw_params_offset;
/**
* draw parameters.
*/
int gl_drawid;
- drm_intel_bo *draw_id_bo;
+ struct brw_bo *draw_id_bo;
uint32_t draw_id_offset;
} draw;
* an indirect call, and num_work_groups_offset is valid. Otherwise,
* num_work_groups is set based on glDispatchCompute.
*/
- drm_intel_bo *num_work_groups_bo;
+ struct brw_bo *num_work_groups_bo;
GLintptr num_work_groups_offset;
const GLuint *num_work_groups;
} compute;
const struct _mesa_index_buffer *ib;
/* Updates are signaled by BRW_NEW_INDEX_BUFFER. */
- drm_intel_bo *bo;
+ struct brw_bo *bo;
uint32_t size;
- GLuint type;
+ unsigned index_size;
/* Offset to index buffer index to use in CMD_3D_PRIM so that we can
* avoid re-uploading the IB packet over and over if we're actually
} urb;
- /* BRW_NEW_CURBE_OFFSETS:
- */
+ /* BRW_NEW_PUSH_CONSTANT_ALLOCATION */
struct {
GLuint wm_start; /**< pos of first wm const in CURBE buffer */
GLuint wm_size; /**< number of float[4] consts, multiple of 16 */
* Pointer to the (intel_upload.c-generated) BO containing the uniforms
* for upload to the CURBE.
*/
- drm_intel_bo *curbe_bo;
+ struct brw_bo *curbe_bo;
/** Offset within curbe_bo of space for current curbe entry */
GLuint curbe_offset;
} curbe;
struct {
struct brw_stage_state base;
-
- /**
- * True if the 3DSTATE_HS command most recently emitted to the 3D
- * pipeline enabled the HS; false otherwise.
- */
- bool enabled;
} tcs;
struct {
struct brw_stage_state base;
-
- /**
- * True if the 3DSTATE_DS command most recently emitted to the 3D
- * pipeline enabled the DS; false otherwise.
- */
- bool enabled;
} tes;
struct {
uint32_t prog_offset;
uint32_t state_offset;
uint32_t vp_offset;
- bool viewport_transform_enable;
} sf;
struct {
* Buffer object used in place of multisampled null render targets on
* Gen6. See brw_emit_null_surface_state().
*/
- drm_intel_bo *multisampled_null_render_target_bo;
+ struct brw_bo *multisampled_null_render_target_bo;
uint32_t fast_clear_op;
float offset_clamp;
struct brw_stage_state base;
} cs;
- /* RS hardware binding table */
- struct {
- drm_intel_bo *bo;
- uint32_t next_offset;
- } hw_bt_pool;
-
struct {
uint32_t state_offset;
uint32_t blend_state_offset;
} predicate;
struct {
+ /* Variables referenced in the XML meta data for OA performance
+ * counters, e.g in the normalization equations.
+ *
+ * All uint64_t for consistent operand types in generated code
+ */
+ struct {
+ uint64_t timestamp_frequency; /** $GpuTimestampFrequency */
+ uint64_t n_eus; /** $EuCoresTotalCount */
+ uint64_t n_eu_slices; /** $EuSlicesTotalCount */
+ uint64_t n_eu_sub_slices; /** $EuSubslicesTotalCount */
+ uint64_t eu_threads_count; /** $EuThreadsCount */
+ uint64_t slice_mask; /** $SliceMask */
+ uint64_t subslice_mask; /** $SubsliceMask */
+ uint64_t gt_min_freq; /** $GpuMinFrequency */
+ uint64_t gt_max_freq; /** $GpuMaxFrequency */
+ } sys_vars;
+
+ /* OA metric sets, indexed by GUID, as know by Mesa at build time,
+ * to cross-reference with the GUIDs of configs advertised by the
+ * kernel at runtime
+ */
+ struct hash_table *oa_metrics_table;
+
struct brw_perf_query_info *queries;
int n_queries;
+ /* The i915 perf stream we open to setup + enable the OA counters */
+ int oa_stream_fd;
+
+ /* An i915 perf stream fd gives exclusive access to the OA unit that will
+ * report counter snapshots for a specific counter set/profile in a
+ * specific layout/format so we can only start OA queries that are
+ * compatible with the currently open fd...
+ */
+ int current_oa_metrics_set_id;
+ int current_oa_format;
+
+ /* List of buffers containing OA reports */
+ struct exec_list sample_buffers;
+
+ /* Cached list of empty sample buffers */
+ struct exec_list free_sample_buffers;
+
+ int n_active_oa_queries;
int n_active_pipeline_stats_queries;
+
+ /* The number of queries depending on running OA counters which
+ * extends beyond brw_end_perf_query() since we need to wait until
+ * the last MI_RPC command has parsed by the GPU.
+ *
+ * Accurate accounting is important here as emitting an
+ * MI_REPORT_PERF_COUNT command while the OA unit is disabled will
+ * effectively hang the gpu.
+ */
+ int n_oa_users;
+
+ /* To help catch an spurious problem with the hardware or perf
+ * forwarding samples, we emit each MI_REPORT_PERF_COUNT command
+ * with a unique ID that we can explicitly check for...
+ */
+ int next_query_start_report_id;
+
+ /**
+ * An array of queries whose results haven't yet been assembled
+ * based on the data in buffer objects.
+ *
+ * These may be active, or have already ended. However, the
+ * results have not been requested.
+ */
+ struct brw_perf_query_object **unaccumulated;
+ int unaccumulated_elements;
+ int unaccumulated_array_size;
+
+ /* The total number of query objects so we can relinquish
+ * our exclusive access to perf if the application deletes
+ * all of its objects. (NB: We only disable perf while
+ * there are no active queries)
+ */
+ int n_query_instances;
} perfquery;
int num_atoms[BRW_NUM_PIPELINES];
const struct brw_tracked_state render_atoms[76];
const struct brw_tracked_state compute_atoms[11];
- /* If (INTEL_DEBUG & DEBUG_BATCH) */
- struct {
- uint32_t offset;
- uint32_t size;
- enum aub_state_struct_type type;
- int index;
- } *state_batch_list;
- int state_batch_count;
-
- uint32_t render_target_format[MESA_FORMAT_COUNT];
- bool format_supported_as_render_target[MESA_FORMAT_COUNT];
+ const enum isl_format *mesa_to_isl_render_format;
+ const bool *mesa_format_supports_render;
/* PrimitiveRestart */
struct {
* brw_workaround_depthstencil_alignment().
*/
struct {
- struct intel_mipmap_tree *depth_mt;
- struct intel_mipmap_tree *stencil_mt;
-
/* Inter-tile (page-aligned) byte offsets. */
- uint32_t depth_offset, hiz_offset, stencil_offset;
- /* Intra-tile x,y offsets for drawing to depth/stencil/hiz */
+ uint32_t depth_offset;
+ /* Intra-tile x,y offsets for drawing to combined depth-stencil. Only
+ * used for Gen < 6.
+ */
uint32_t tile_x, tile_y;
} depthstencil;
} l3;
struct {
- drm_intel_bo *bo;
+ struct brw_bo *bo;
const char **names;
int *ids;
enum shader_time_shader_type *types;
__DRIdrawable *drawable);
void intel_prepare_render(struct brw_context *brw);
+void brw_predraw_resolve_inputs(struct brw_context *brw);
+
void intel_resolve_for_dri2_flush(struct brw_context *brw,
__DRIdrawable *drawable);
void brw_emit_query_end(struct brw_context *brw);
void brw_query_counter(struct gl_context *ctx, struct gl_query_object *q);
bool brw_is_query_pipelined(struct brw_query_object *query);
+uint64_t brw_timebase_scale(struct brw_context *brw, uint64_t gpu_timestamp);
+uint64_t brw_raw_timestamp_delta(struct brw_context *brw,
+ uint64_t time0, uint64_t time1);
/** gen6_queryobj.c */
void gen6_init_queryobj_functions(struct dd_function_table *functions);
-void brw_write_timestamp(struct brw_context *brw, drm_intel_bo *bo, int idx);
-void brw_write_depth_count(struct brw_context *brw, drm_intel_bo *bo, int idx);
+void brw_write_timestamp(struct brw_context *brw, struct brw_bo *bo, int idx);
+void brw_write_depth_count(struct brw_context *brw, struct brw_bo *bo, int idx);
/** hsw_queryobj.c */
void hsw_overflow_result_to_gpr0(struct brw_context *brw,
/** intel_batchbuffer.c */
void brw_load_register_mem(struct brw_context *brw,
uint32_t reg,
- drm_intel_bo *bo,
+ struct brw_bo *bo,
uint32_t read_domains, uint32_t write_domain,
uint32_t offset);
void brw_load_register_mem64(struct brw_context *brw,
uint32_t reg,
- drm_intel_bo *bo,
+ struct brw_bo *bo,
uint32_t read_domains, uint32_t write_domain,
uint32_t offset);
void brw_store_register_mem32(struct brw_context *brw,
- drm_intel_bo *bo, uint32_t reg, uint32_t offset);
+ struct brw_bo *bo, uint32_t reg, uint32_t offset);
void brw_store_register_mem64(struct brw_context *brw,
- drm_intel_bo *bo, uint32_t reg, uint32_t offset);
+ struct brw_bo *bo, uint32_t reg, uint32_t offset);
void brw_load_register_imm32(struct brw_context *brw,
uint32_t reg, uint32_t imm);
void brw_load_register_imm64(struct brw_context *brw,
uint32_t dest);
void brw_load_register_reg64(struct brw_context *brw, uint32_t src,
uint32_t dest);
-void brw_store_data_imm32(struct brw_context *brw, drm_intel_bo *bo,
+void brw_store_data_imm32(struct brw_context *brw, struct brw_bo *bo,
uint32_t offset, uint32_t imm);
-void brw_store_data_imm64(struct brw_context *brw, drm_intel_bo *bo,
+void brw_store_data_imm64(struct brw_context *brw, struct brw_bo *bo,
uint32_t offset, uint64_t imm);
-/*======================================================================
- * brw_state_dump.c
- */
-void brw_debug_batch(struct brw_context *brw);
-void brw_annotate_aub(struct brw_context *brw);
-
/*======================================================================
* intel_tex_validate.c
*/
void brwInitFragProgFuncs( struct dd_function_table *functions );
void brw_get_scratch_bo(struct brw_context *brw,
- drm_intel_bo **scratch_bo, int size);
+ struct brw_bo **scratch_bo, int size);
void brw_alloc_stage_scratch(struct brw_context *brw,
struct brw_stage_state *stage_state,
unsigned per_thread_size,
/* brw_urb.c
*/
+void brw_calculate_urb_fence(struct brw_context *brw, unsigned csize,
+ unsigned vsize, unsigned sfsize);
void brw_upload_urb_fence(struct brw_context *brw);
/* brw_curbe.c
const struct gl_vertex_array *glarray);
static inline unsigned
-brw_get_index_type(GLenum type)
+brw_get_index_type(unsigned index_size)
{
- assert((type == GL_UNSIGNED_BYTE)
- || (type == GL_UNSIGNED_SHORT)
- || (type == GL_UNSIGNED_INT));
-
- /* The possible values for type are GL_UNSIGNED_BYTE (0x1401),
- * GL_UNSIGNED_SHORT (0x1403), and GL_UNSIGNED_INT (0x1405) which we want
- * to map to scale factors of 0, 1, and 2, respectively. These scale
- * factors are then left-shfited by 8 to be in the correct position in the
- * CMD_INDEX_BUFFER packet.
- *
- * Subtracting 0x1401 gives 0, 2, and 4. Shifting left by 7 afterwards
- * gives 0x00000000, 0x00000100, and 0x00000200. These just happen to be
- * the values the need to be written in the CMD_INDEX_BUFFER packet.
+ /* The hw needs 0x00, 0x01, and 0x02 for ubyte, ushort, and uint,
+ * respectively.
*/
- return (type - 0x1401) << 7;
+ return index_size >> 1;
}
void brw_prepare_vertices(struct brw_context *brw);
/* brw_wm_surface_state.c */
-void brw_init_surface_formats(struct brw_context *brw);
void brw_create_constant_surface(struct brw_context *brw,
- drm_intel_bo *bo,
+ struct brw_bo *bo,
uint32_t offset,
uint32_t size,
uint32_t *out_offset);
void brw_create_buffer_surface(struct brw_context *brw,
- drm_intel_bo *bo,
+ struct brw_bo *bo,
uint32_t offset,
uint32_t size,
uint32_t *out_offset);
struct brw_stage_prog_data *prog_data);
/* brw_surface_formats.c */
+void intel_screen_init_surface_formats(struct intel_screen *screen);
+void brw_init_surface_formats(struct brw_context *brw);
bool brw_render_target_supported(struct brw_context *brw,
struct gl_renderbuffer *rb);
uint32_t brw_depth_format(struct brw_context *brw, mesa_format format);
/* brw_performance_query.c */
void brw_init_performance_queries(struct brw_context *brw);
-/* intel_buffer_objects.c */
-int brw_bo_map(struct brw_context *brw, drm_intel_bo *bo, int write_enable,
- const char *bo_name);
-int brw_bo_map_gtt(struct brw_context *brw, drm_intel_bo *bo,
- const char *bo_name);
-
/* intel_extensions.c */
extern void intelInitExtensions(struct gl_context *ctx);
int dstX0, int dstY0,
int width, int height);
-/* gen6_multisample_state.c */
-unsigned
-gen6_determine_sample_mask(struct brw_context *brw);
-
-void
-gen6_emit_3dstate_multisample(struct brw_context *brw,
- unsigned num_samples);
-void
-gen6_emit_3dstate_sample_mask(struct brw_context *brw, unsigned mask);
void
gen6_get_sample_position(struct gl_context *ctx,
struct gl_framebuffer *fb,
return (const struct brw_program *) p;
}
-static inline uint32_t
-brw_program_reloc(struct brw_context *brw, uint32_t state_offset,
- uint32_t prog_offset)
-{
- if (brw->gen >= 5) {
- /* Using state base address. */
- return prog_offset;
- }
-
- drm_intel_bo_emit_reloc(brw->batch.bo,
- state_offset,
- brw->cache.bo,
- prog_offset,
- I915_GEM_DOMAIN_INSTRUCTION, 0);
-
- return brw->cache.bo->offset64 + prog_offset;
-}
-
-bool brw_do_cubemap_normalize(struct exec_list *instructions);
-
static inline bool
brw_depth_writes_enabled(const struct brw_context *brw)
{
uint32_t width, uint32_t height,
uint32_t tile_x, uint32_t tile_y);
-void gen8_hiz_exec(struct brw_context *brw, struct intel_mipmap_tree *mt,
- unsigned int level, unsigned int layer, enum blorp_hiz_op op);
-
uint32_t get_hw_prim_for_gl_prim(int mode);
void
gen6_upload_push_constants(struct brw_context *brw,
const struct gl_program *prog,
const struct brw_stage_prog_data *prog_data,
- struct brw_stage_state *stage_state,
- enum aub_state_struct_type type);
+ struct brw_stage_state *stage_state);
bool
gen9_use_linear_1d_layout(const struct brw_context *brw,
void brw_emit_pipe_control_flush(struct brw_context *brw, uint32_t flags);
void brw_emit_pipe_control_write(struct brw_context *brw, uint32_t flags,
- drm_intel_bo *bo, uint32_t offset,
- uint32_t imm_lower, uint32_t imm_upper);
+ struct brw_bo *bo, uint32_t offset,
+ uint64_t imm);
+void brw_emit_end_of_pipe_sync(struct brw_context *brw, uint32_t flags);
void brw_emit_mi_flush(struct brw_context *brw);
void brw_emit_post_sync_nonzero_flush(struct brw_context *brw);
void brw_emit_depth_stall_flushes(struct brw_context *brw);