fs_reg(class fs_visitor *v, const struct glsl_type *type);
bool equals(const fs_reg &r) const;
+ bool is_zero() const;
+ bool is_one() const;
/** Register file: ARF, GRF, MRF, IMM. */
enum register_file file;
uint32_t u;
float f;
} imm;
+
+ fs_reg *reladdr;
};
static const fs_reg reg_undef;
static const fs_reg reg_null_f(ARF, BRW_ARF_NULL, BRW_REGISTER_TYPE_F);
static const fs_reg reg_null_d(ARF, BRW_ARF_NULL, BRW_REGISTER_TYPE_D);
+class ip_record : public exec_node {
+public:
+ static void* operator new(size_t size, void *ctx)
+ {
+ void *node;
+
+ node = rzalloc_size(ctx, size);
+ assert(node != NULL);
+
+ return node;
+ }
+
+ ip_record(int ip)
+ {
+ this->ip = ip;
+ }
+
+ int ip;
+};
+
class fs_inst : public backend_instruction {
public:
/* Callers of this ralloc-based new need not call delete. It's
bool overwrites_reg(const fs_reg ®);
bool is_tex();
bool is_math();
+ bool is_send_from_grf();
fs_reg dst;
fs_reg src[3];
bool saturate;
int conditional_mod; /**< BRW_CONDITIONAL_* */
+ /* Chooses which flag subregister (f0.0 or f0.1) is used for conditional
+ * mod and predication.
+ */
+ uint8_t flag_subreg;
+
int mlen; /**< SEND message length */
int base_mrf; /**< First MRF in the SEND message, if mlen is nonzero. */
uint32_t texture_offset; /**< Texture offset bitfield */
bool shadow_compare;
bool force_uncompressed;
bool force_sechalf;
+ bool force_writemask_all;
uint32_t offset; /* spill/unspill offset */
/** @{
/** @} */
};
+/**
+ * The fragment shader front-end.
+ *
+ * Translates either GLSL IR or Mesa IR (for ARB_fragment_program) into FS IR.
+ */
class fs_visitor : public backend_visitor
{
public:
- fs_visitor(struct brw_wm_compile *c, struct gl_shader_program *prog,
- struct brw_shader *shader);
+ fs_visitor(struct brw_context *brw,
+ struct brw_wm_compile *c,
+ struct gl_shader_program *prog,
+ struct gl_fragment_program *fp,
+ unsigned dispatch_width);
~fs_visitor();
fs_reg *variable_storage(ir_variable *var);
void swizzle_result(ir_texture *ir, fs_reg orig_val, int sampler);
+ bool can_do_source_mods(fs_inst *inst);
+
fs_inst *emit(fs_inst inst);
fs_inst *emit(fs_inst *inst);
+ void emit(exec_list list);
fs_inst *emit(enum opcode opcode);
fs_inst *emit(enum opcode opcode, fs_reg dst);
fs_inst *AND(fs_reg dst, fs_reg src0, fs_reg src1);
fs_inst *OR(fs_reg dst, fs_reg src0, fs_reg src1);
fs_inst *XOR(fs_reg dst, fs_reg src0, fs_reg src1);
+ fs_inst *IF(uint32_t predicate);
+ fs_inst *IF(fs_reg src0, fs_reg src1, uint32_t condition);
fs_inst *CMP(fs_reg dst, fs_reg src0, fs_reg src1,
uint32_t condition);
fs_inst *end,
fs_reg reg);
+ exec_list VARYING_PULL_CONSTANT_LOAD(fs_reg dst, fs_reg surf_index,
+ fs_reg offset);
+
bool run();
- void setup_paramvalues_refs();
+ void setup_payload_gen4();
+ void setup_payload_gen6();
void assign_curb_setup();
void calculate_urb_setup();
void assign_urb_setup();
void spill_reg(int spill_reg);
void split_virtual_grfs();
void compact_virtual_grfs();
+ void move_uniform_array_access_to_pull_constants();
void setup_pull_constants();
void calculate_live_intervals();
bool opt_algebraic();
bool remove_dead_constants();
bool remove_duplicate_mrf_writes();
bool virtual_grf_interferes(int a, int b);
- void schedule_instructions();
+ void schedule_instructions(bool post_reg_alloc);
void fail(const char *msg, ...);
void push_force_uncompressed();
void push_force_sechalf();
void pop_force_sechalf();
- void generate_code();
- void generate_fb_write(fs_inst *inst);
- void generate_pixel_xy(struct brw_reg dst, bool is_x);
- void generate_linterp(fs_inst *inst, struct brw_reg dst,
- struct brw_reg *src);
- void generate_tex(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
- void generate_math1_gen7(fs_inst *inst,
- struct brw_reg dst,
- struct brw_reg src);
- void generate_math2_gen7(fs_inst *inst,
- struct brw_reg dst,
- struct brw_reg src0,
- struct brw_reg src1);
- void generate_math1_gen6(fs_inst *inst,
- struct brw_reg dst,
- struct brw_reg src);
- void generate_math2_gen6(fs_inst *inst,
- struct brw_reg dst,
- struct brw_reg src0,
- struct brw_reg src1);
- void generate_math_gen4(fs_inst *inst,
- struct brw_reg dst,
- struct brw_reg src);
- void generate_discard(fs_inst *inst);
- void generate_ddx(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
- void generate_ddy(fs_inst *inst, struct brw_reg dst, struct brw_reg src,
- bool negate_value);
- void generate_spill(fs_inst *inst, struct brw_reg src);
- void generate_unspill(fs_inst *inst, struct brw_reg dst);
- void generate_pull_constant_load(fs_inst *inst, struct brw_reg dst,
- struct brw_reg index,
- struct brw_reg offset);
- void generate_mov_dispatch_to_flags();
-
void emit_dummy_fs();
fs_reg *emit_fragcoord_interpolation(ir_variable *ir);
fs_inst *emit_linterp(const fs_reg &attr, const fs_reg &interp,
fs_reg shadow_comp, fs_reg lod, fs_reg lod2);
fs_inst *emit_texture_gen7(ir_texture *ir, fs_reg dst, fs_reg coordinate,
fs_reg shadow_comp, fs_reg lod, fs_reg lod2);
+ fs_reg fix_math_operand(fs_reg src);
fs_inst *emit_math(enum opcode op, fs_reg dst, fs_reg src0);
fs_inst *emit_math(enum opcode op, fs_reg dst, fs_reg src0, fs_reg src1);
void emit_minmax(uint32_t conditionalmod, fs_reg dst,
void emit_color_write(int target, int index, int first_color_mrf);
void emit_fb_writes();
+
+ void emit_shader_time_begin();
+ void emit_shader_time_end();
+ void emit_shader_time_write(enum shader_time_shader_type type,
+ fs_reg value);
+
bool try_rewrite_rhs_to_dst(ir_assignment *ir,
fs_reg dst,
fs_reg src,
void resolve_ud_negate(fs_reg *reg);
void resolve_bool_comparison(ir_rvalue *rvalue, fs_reg *reg);
+ fs_reg get_timestamp();
+
struct brw_reg interp_reg(int location, int channel);
- int setup_uniform_values(int loc, const glsl_type *type);
+ void setup_uniform_values(ir_variable *ir);
void setup_builtin_uniform_values(ir_variable *ir);
int implied_mrf_writes(fs_inst *inst);
+ void dump_instructions();
+ void dump_instruction(fs_inst *inst);
+
const struct gl_fragment_program *fp;
struct brw_wm_compile *c;
+ unsigned int sanity_param_count;
- /* Delayed setup of c->prog_data.params[] due to realloc of
- * ParamValues[] during compile.
- */
- int param_index[MAX_UNIFORMS * 4];
- int param_offset[MAX_UNIFORMS * 4];
+ int param_size[MAX_UNIFORMS * 4];
int *virtual_grf_sizes;
int virtual_grf_count;
fs_reg pixel_w;
fs_reg delta_x[BRW_WM_BARYCENTRIC_INTERP_MODE_COUNT];
fs_reg delta_y[BRW_WM_BARYCENTRIC_INTERP_MODE_COUNT];
- fs_reg reg_null_cmp;
+ fs_reg shader_start_time;
int grf_used;
+ const unsigned dispatch_width; /**< 8 or 16 */
+
int force_uncompressed_stack;
int force_sechalf_stack;
};
+/**
+ * The fragment shader code generator.
+ *
+ * Translates FS IR to actual i965 assembly code.
+ */
+class fs_generator
+{
+public:
+ fs_generator(struct brw_context *brw,
+ struct brw_wm_compile *c,
+ struct gl_shader_program *prog,
+ struct gl_fragment_program *fp,
+ bool dual_source_output);
+ ~fs_generator();
+
+ const unsigned *generate_assembly(exec_list *simd8_instructions,
+ exec_list *simd16_instructions,
+ unsigned *assembly_size);
+
+private:
+ void generate_code(exec_list *instructions);
+ void generate_fb_write(fs_inst *inst);
+ void generate_pixel_xy(struct brw_reg dst, bool is_x);
+ void generate_linterp(fs_inst *inst, struct brw_reg dst,
+ struct brw_reg *src);
+ void generate_tex(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
+ void generate_math1_gen7(fs_inst *inst,
+ struct brw_reg dst,
+ struct brw_reg src);
+ void generate_math2_gen7(fs_inst *inst,
+ struct brw_reg dst,
+ struct brw_reg src0,
+ struct brw_reg src1);
+ void generate_math1_gen6(fs_inst *inst,
+ struct brw_reg dst,
+ struct brw_reg src);
+ void generate_math2_gen6(fs_inst *inst,
+ struct brw_reg dst,
+ struct brw_reg src0,
+ struct brw_reg src1);
+ void generate_math_gen4(fs_inst *inst,
+ struct brw_reg dst,
+ struct brw_reg src);
+ void generate_ddx(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
+ void generate_ddy(fs_inst *inst, struct brw_reg dst, struct brw_reg src,
+ bool negate_value);
+ void generate_spill(fs_inst *inst, struct brw_reg src);
+ void generate_unspill(fs_inst *inst, struct brw_reg dst);
+ void generate_uniform_pull_constant_load(fs_inst *inst, struct brw_reg dst,
+ struct brw_reg index,
+ struct brw_reg offset);
+ void generate_uniform_pull_constant_load_gen7(fs_inst *inst,
+ struct brw_reg dst,
+ struct brw_reg surf_index,
+ struct brw_reg offset);
+ void generate_varying_pull_constant_load(fs_inst *inst, struct brw_reg dst,
+ struct brw_reg index);
+ void generate_varying_pull_constant_load_gen7(fs_inst *inst,
+ struct brw_reg dst,
+ struct brw_reg index,
+ struct brw_reg offset);
+ void generate_mov_dispatch_to_flags(fs_inst *inst);
+ void generate_set_global_offset(fs_inst *inst,
+ struct brw_reg dst,
+ struct brw_reg src,
+ struct brw_reg offset);
+ void generate_discard_jump(fs_inst *inst);
+
+ void patch_discard_jumps_to_fb_writes();
+
+ struct brw_context *brw;
+ struct intel_context *intel;
+ struct gl_context *ctx;
+
+ struct brw_compile *p;
+ struct brw_wm_compile *c;
+
+ struct gl_shader_program *prog;
+ struct gl_shader *shader;
+ const struct gl_fragment_program *fp;
+
+ unsigned dispatch_width; /**< 8 or 16 */
+
+ exec_list discard_halt_patches;
+ bool dual_source_output;
+ void *mem_ctx;
+};
+
bool brw_do_channel_expressions(struct exec_list *instructions);
bool brw_do_vector_splitting(struct exec_list *instructions);
bool brw_fs_precompile(struct gl_context *ctx, struct gl_shader_program *prog);