uint8_t *ra_reg_to_grf;
/**
- * ra class for the aligned pairs we use for PLN, which doesn't
+ * ra class for the aligned barycentrics we use for PLN, which doesn't
* appear in *classes.
*/
- int aligned_pairs_class;
+ int aligned_bary_class;
} fs_reg_sets[3];
void (*shader_debug_log)(void *, const char *str, ...) PRINTFLIKE(2, 3);
* whether nir_opt_large_constants will be run.
*/
bool supports_shader_constants;
+
+ /**
+ * Whether or not the driver wants uniform params to be compacted by the
+ * back-end compiler.
+ */
+ bool compact_params;
+
+ /**
+ * Whether or not the driver wants variable group size to be lowered by the
+ * back-end compiler.
+ */
+ bool lower_variable_group_size;
};
/**
float scale_factors[32];
};
+/** An enum representing what kind of input gl_SubgroupSize is. */
+enum PACKED brw_subgroup_size_type
+{
+ BRW_SUBGROUP_SIZE_API_CONSTANT, /**< Default Vulkan behavior */
+ BRW_SUBGROUP_SIZE_UNIFORM, /**< OpenGL behavior */
+ BRW_SUBGROUP_SIZE_VARYING, /**< VK_EXT_subgroup_size_control */
+
+ /* These enums are specifically chosen so that the value of the enum is
+ * also the subgroup size. If any new values are added, they must respect
+ * this invariant.
+ */
+ BRW_SUBGROUP_SIZE_REQUIRE_8 = 8, /**< VK_EXT_subgroup_size_control */
+ BRW_SUBGROUP_SIZE_REQUIRE_16 = 16, /**< VK_EXT_subgroup_size_control */
+ BRW_SUBGROUP_SIZE_REQUIRE_32 = 32, /**< VK_EXT_subgroup_size_control */
+};
+
+struct brw_base_prog_key {
+ unsigned program_string_id;
+
+ enum brw_subgroup_size_type subgroup_size_type;
+
+ struct brw_sampler_prog_key_data tex;
+};
+
/**
* The VF can't natively handle certain types of attributes, such as GL_FIXED
* or most 10_10_10_2 types. These flags enable various VS workarounds to
/** The program key for Vertex Shaders. */
struct brw_vs_prog_key {
- unsigned program_string_id;
+ struct brw_base_prog_key base;
/**
* Per-attribute workaround flags
* the VUE, even if they aren't written by the vertex shader.
*/
uint8_t point_coord_replace;
-
- struct brw_sampler_prog_key_data tex;
};
/** The program key for Tessellation Control Shaders. */
struct brw_tcs_prog_key
{
- unsigned program_string_id;
+ struct brw_base_prog_key base;
GLenum tes_primitive_mode;
uint64_t outputs_written;
bool quads_workaround;
-
- struct brw_sampler_prog_key_data tex;
};
/** The program key for Tessellation Evaluation Shaders. */
struct brw_tes_prog_key
{
- unsigned program_string_id;
+ struct brw_base_prog_key base;
/** A bitfield of per-patch inputs read. */
uint32_t patch_inputs_read;
/** A bitfield of per-vertex inputs read. */
uint64_t inputs_read;
- struct brw_sampler_prog_key_data tex;
+ /**
+ * How many user clipping planes are being uploaded to the tessellation
+ * evaluation shader as push constants.
+ *
+ * These are used for lowering legacy gl_ClipVertex/gl_Position clipping to
+ * clip distances.
+ */
+ unsigned nr_userclip_plane_consts:4;
};
/** The program key for Geometry Shaders. */
struct brw_gs_prog_key
{
- unsigned program_string_id;
+ struct brw_base_prog_key base;
- struct brw_sampler_prog_key_data tex;
+ /**
+ * How many user clipping planes are being uploaded to the geometry shader
+ * as push constants.
+ *
+ * These are used for lowering legacy gl_ClipVertex/gl_Position clipping to
+ * clip distances.
+ */
+ unsigned nr_userclip_plane_consts:4;
};
enum brw_sf_primitive {
/** The program key for Fragment/Pixel Shaders. */
struct brw_wm_prog_key {
+ struct brw_base_prog_key base;
+
/* Some collection of BRW_WM_IZ_* */
uint8_t iz_lookup;
bool stats_wm:1;
uint8_t color_outputs_valid;
uint64_t input_slots_valid;
- unsigned program_string_id;
GLenum alpha_test_func; /* < For Gen4/5 MRT alpha test */
float alpha_test_ref;
-
- struct brw_sampler_prog_key_data tex;
};
struct brw_cs_prog_key {
- uint32_t program_string_id;
- struct brw_sampler_prog_key_data tex;
+ struct brw_base_prog_key base;
};
/* brw_any_prog_key is any of the keys that map to an API stage */
union brw_any_prog_key {
+ struct brw_base_prog_key base;
struct brw_vs_prog_key vs;
struct brw_tcs_prog_key tcs;
struct brw_tes_prog_key tes;
BRW_PARAM_BUILTIN_BASE_WORK_GROUP_ID_Y,
BRW_PARAM_BUILTIN_BASE_WORK_GROUP_ID_Z,
BRW_PARAM_BUILTIN_SUBGROUP_ID,
+ BRW_PARAM_BUILTIN_WORK_GROUP_SIZE_X,
+ BRW_PARAM_BUILTIN_WORK_GROUP_SIZE_Y,
+ BRW_PARAM_BUILTIN_WORK_GROUP_SIZE_Z,
};
#define BRW_PARAM_BUILTIN_CLIP_PLANE(idx, comp) \
GLuint nr_params; /**< number of float params/constants */
GLuint nr_pull_params;
+ /* zero_push_reg is a bitfield which indicates what push registers (if any)
+ * should be zeroed by SW at the start of the shader. The corresponding
+ * push_reg_mask_param specifies the param index (in 32-bit units) where
+ * the actual runtime 64-bit mask will be pushed. The shader will zero
+ * push reg i if
+ *
+ * reg_used & zero_push_reg & ~*push_reg_mask_param & (1ull << i)
+ *
+ * If this field is set, brw_compiler::compact_params must be false.
+ */
+ uint64_t zero_push_reg;
+ unsigned push_reg_mask_param;
+
unsigned curb_read_length;
unsigned total_scratch;
unsigned total_shared;
unsigned program_size;
+ /** Does this program pull from any UBO or other constant buffers? */
+ bool has_ubo_pull;
+
/**
* Register where the thread expects to find input data from the URB
* (typically uniforms, followed by vertex or fragment attributes).
*/
uint32_t *param;
uint32_t *pull_param;
+
+ /* Whether shader uses atomic operations. */
+ bool uses_atomic_load_store;
};
static inline uint32_t *
bool dispatch_16;
bool dispatch_32;
bool dual_src_blend;
- bool replicate_alpha;
bool persample_dispatch;
bool uses_pos_offset;
bool uses_omask;
*/
uint32_t flat_inputs;
+ /**
+ * The FS inputs
+ */
+ uint64_t inputs;
+
/* Mapping of VUE slots to interpolation modes.
* Used by the Gen4-5 clip/sf/wm stages.
*/
* For varying slots that are not used by the FS, the value is -1.
*/
int urb_setup[VARYING_SLOT_MAX];
+
+ /**
+ * Cache structure into the urb_setup array above that contains the
+ * attribute numbers of active varyings out of urb_setup.
+ * The actual count is stored in urb_setup_attribs_count.
+ */
+ uint8_t urb_setup_attribs[VARYING_SLOT_MAX];
+ uint8_t urb_setup_attribs_count;
};
/** Returns the SIMD width corresponding to a given KSP index
struct brw_stage_prog_data base;
unsigned local_size[3];
- unsigned simd_size;
- unsigned threads;
+ unsigned slm_size;
+
+ /* Program offsets for the 8/16/32 SIMD variants. Multiple variants are
+ * kept when using variable group size, and the right one can only be
+ * decided at dispatch time.
+ */
+ unsigned prog_offset[3];
+
+ /* Bitmask indicating which program offsets are valid. */
+ unsigned prog_mask;
+
+ /* Bitmask indicating which programs have spilled. */
+ unsigned prog_spilled;
+
bool uses_barrier;
bool uses_num_work_groups;
struct {
struct brw_push_const_block cross_thread;
struct brw_push_const_block per_thread;
- struct brw_push_const_block total;
} push;
struct {
} binding_table;
};
+static inline uint32_t
+brw_cs_prog_data_prog_offset(const struct brw_cs_prog_data *prog_data,
+ unsigned dispatch_width)
+{
+ assert(dispatch_width == 8 ||
+ dispatch_width == 16 ||
+ dispatch_width == 32);
+ const unsigned index = dispatch_width / 16;
+ assert(prog_data->prog_mask & (1 << index));
+ return prog_data->prog_offset[index];
+}
+
/**
* Enum representing the i965-specific vertex results that don't correspond
* exactly to any element of gl_varying_slot. The values of this enum are
void brw_compute_vue_map(const struct gen_device_info *devinfo,
struct brw_vue_map *vue_map,
uint64_t slots_valid,
- bool separate_shader);
+ bool separate_shader,
+ uint32_t pos_slots);
void brw_compute_tess_vue_map(struct brw_vue_map *const vue_map,
uint64_t slots_valid,
/** Number vertices in output patch */
int instances;
+
+ /** Track patch count threshold */
+ int patch_count_threshold;
};
struct brw_cs_prog_data cs;
};
-#define DEFINE_PROG_DATA_DOWNCAST(stage) \
-static inline struct brw_##stage##_prog_data * \
-brw_##stage##_prog_data(struct brw_stage_prog_data *prog_data) \
-{ \
- return (struct brw_##stage##_prog_data *) prog_data; \
+#define DEFINE_PROG_DATA_DOWNCAST(stage) \
+static inline struct brw_##stage##_prog_data * \
+brw_##stage##_prog_data(struct brw_stage_prog_data *prog_data) \
+{ \
+ return (struct brw_##stage##_prog_data *) prog_data; \
+} \
+static inline const struct brw_##stage##_prog_data * \
+brw_##stage##_prog_data_const(const struct brw_stage_prog_data *prog_data) \
+{ \
+ return (const struct brw_##stage##_prog_data *) prog_data; \
}
DEFINE_PROG_DATA_DOWNCAST(vue)
DEFINE_PROG_DATA_DOWNCAST(vs)
DEFINE_PROG_DATA_DOWNCAST(sf)
#undef DEFINE_PROG_DATA_DOWNCAST
+struct brw_compile_stats {
+ uint32_t dispatch_width; /**< 0 for vec4 */
+ uint32_t instructions;
+ uint32_t sends;
+ uint32_t loops;
+ uint32_t cycles;
+ uint32_t spills;
+ uint32_t fills;
+};
+
/** @} */
struct brw_compiler *
struct brw_vs_prog_data *prog_data,
struct nir_shader *shader,
int shader_time_index,
+ struct brw_compile_stats *stats,
char **error_str);
/**
struct brw_tcs_prog_data *prog_data,
struct nir_shader *nir,
int shader_time_index,
+ struct brw_compile_stats *stats,
char **error_str);
/**
const struct brw_vue_map *input_vue_map,
struct brw_tes_prog_data *prog_data,
struct nir_shader *shader,
- struct gl_program *prog,
int shader_time_index,
+ struct brw_compile_stats *stats,
char **error_str);
/**
struct nir_shader *shader,
struct gl_program *prog,
int shader_time_index,
+ struct brw_compile_stats *stats,
char **error_str);
/**
const struct brw_wm_prog_key *key,
struct brw_wm_prog_data *prog_data,
struct nir_shader *shader,
- struct gl_program *prog,
int shader_time_index8,
int shader_time_index16,
int shader_time_index32,
bool allow_spilling,
bool use_rep_send, struct brw_vue_map *vue_map,
+ struct brw_compile_stats *stats, /**< Array of three stats */
char **error_str);
/**
struct brw_cs_prog_data *prog_data,
const struct nir_shader *shader,
int shader_time_index,
+ struct brw_compile_stats *stats,
char **error_str);
void brw_debug_key_recompile(const struct brw_compiler *c, void *log,
gl_shader_stage stage,
- const void *old_key, const void *key);
+ const struct brw_base_prog_key *old_key,
+ const struct brw_base_prog_key *key);
static inline uint32_t
encode_slm_size(unsigned gen, uint32_t bytes)
return slm_size;
}
+unsigned
+brw_cs_push_const_total_size(const struct brw_cs_prog_data *cs_prog_data,
+ unsigned threads);
+
+unsigned
+brw_cs_simd_size_for_group_size(const struct gen_device_info *devinfo,
+ const struct brw_cs_prog_data *cs_prog_data,
+ unsigned group_size);
+
/**
* Return true if the given shader stage is dispatched contiguously by the
* relevant fixed function starting from channel 0 of the SIMD thread, which
* '2^n - 1' for some n.
*/
static inline bool
-brw_stage_has_packed_dispatch(MAYBE_UNUSED const struct gen_device_info *devinfo,
+brw_stage_has_packed_dispatch(ASSERTED const struct gen_device_info *devinfo,
gl_shader_stage stage,
const struct brw_stage_prog_data *prog_data)
{
* to do a full test run with brw_fs_test_dispatch_packing() hooked up to
* the NIR front-end before changing this assertion.
*/
- assert(devinfo->gen <= 11);
+ assert(devinfo->gen <= 12);
switch (stage) {
case MESA_SHADER_FRAGMENT: {