#include "brw_wm.h"
#include "brw_shader.h"
}
+#include "gen8_generator.h"
#include "glsl/glsl_types.h"
#include "glsl/ir.h"
+#define MAX_SAMPLER_MESSAGE_SIZE 11
+
class bblock_t;
namespace {
struct acp_entry;
}
-enum register_file {
- BAD_FILE,
- ARF,
- GRF,
- MRF,
- IMM,
- FIXED_HW_REG, /* a struct brw_reg */
- UNIFORM, /* prog_data->params[reg] */
-};
+namespace brw {
+ class fs_live_variables;
+}
class fs_reg {
public:
- /* Callers of this ralloc-based new need not call delete. It's
- * easier to just ralloc_free 'ctx' (or any of its ancestors). */
- static void* operator new(size_t size, void *ctx)
- {
- void *node;
-
- node = ralloc_size(ctx, size);
- assert(node != NULL);
-
- return node;
- }
+ DECLARE_RALLOC_CXX_OPERATORS(fs_reg)
void init();
bool equals(const fs_reg &r) const;
bool is_zero() const;
bool is_one() const;
+ bool is_null() const;
+ bool is_valid_3src() const;
+ bool is_contiguous() const;
+ fs_reg retype(uint32_t type);
+ fs_reg &apply_stride(unsigned stride);
- /** Register file: ARF, GRF, MRF, IMM. */
+ /** Register file: GRF, MRF, IMM. */
enum register_file file;
/**
- * Register number. For ARF/MRF, it's the hardware register. For
+ * Register number. For MRF, it's the hardware register. For
* GRF, it's a virtual register number until register allocation
*/
int reg;
/**
- * For virtual registers, this is a hardware register offset from
- * the start of the register block (for example, a constant index
- * in an array access).
+ * Offset from the start of the contiguous register block.
+ *
+ * For pre-register-allocation GRFs, this is in units of a float per pixel
+ * (1 hardware register for SIMD8 mode, or 2 registers for SIMD16 mode).
+ * For uniforms, this is in units of 1 float.
*/
int reg_offset;
/** Register type. BRW_REGISTER_TYPE_* */
int type;
bool negate;
bool abs;
- bool sechalf;
struct brw_reg fixed_hw_reg;
- int smear; /* -1, or a channel of the reg to smear to all channels. */
+
+ /** Smear a channel of the reg to all channels. */
+ fs_reg &set_smear(unsigned subreg);
/** Value for file == IMM */
union {
float f;
} imm;
+ /**
+ * Offset in bytes from the start of the register. Values up to a
+ * backend_reg::reg_offset unit are valid.
+ */
+ int subreg_offset;
+
+ /** Register region horizontal stride */
+ int stride;
+
fs_reg *reladdr;
};
+static inline fs_reg
+offset(fs_reg reg, unsigned delta)
+{
+ assert(delta == 0 || (reg.file != HW_REG && reg.file != IMM));
+ reg.reg_offset += delta;
+ return reg;
+}
+
+static inline fs_reg
+byte_offset(fs_reg reg, unsigned delta)
+{
+ assert(delta == 0 || (reg.file != HW_REG && reg.file != IMM));
+ reg.subreg_offset += delta;
+ return reg;
+}
+
+/**
+ * Get either of the 8-component halves of a 16-component register.
+ *
+ * Note: this also works if \c reg represents a SIMD16 pair of registers.
+ */
+static inline fs_reg
+half(const fs_reg ®, unsigned idx)
+{
+ assert(idx < 2);
+ assert(idx == 0 || (reg.file != HW_REG && reg.file != IMM));
+ return byte_offset(reg, 8 * idx * reg.stride * type_sz(reg.type));
+}
+
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);
+static const fs_reg reg_null_f(retype(brw_null_reg(), BRW_REGISTER_TYPE_F));
+static const fs_reg reg_null_d(retype(brw_null_reg(), BRW_REGISTER_TYPE_D));
+static const fs_reg reg_null_ud(retype(brw_null_reg(), BRW_REGISTER_TYPE_UD));
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;
- }
+ DECLARE_RALLOC_CXX_OPERATORS(ip_record)
ip_record(int ip)
{
class fs_inst : public backend_instruction {
public:
- /* Callers of this ralloc-based new need not call delete. It's
- * easier to just ralloc_free 'ctx' (or any of its ancestors). */
- static void* operator new(size_t size, void *ctx)
- {
- void *node;
-
- node = rzalloc_size(ctx, size);
- assert(node != NULL);
-
- return node;
- }
+ DECLARE_RALLOC_CXX_OPERATORS(fs_inst)
void init();
bool equals(fs_inst *inst);
bool overwrites_reg(const fs_reg ®);
- bool is_tex();
- bool is_math();
- bool is_control_flow();
bool is_send_from_grf();
+ bool is_partial_write();
+ int regs_read(fs_visitor *v, int arg);
+
+ bool reads_flag();
+ bool writes_flag();
fs_reg dst;
fs_reg src[3];
void visit(ir_call *ir);
void visit(ir_function *ir);
void visit(ir_function_signature *ir);
+ void visit(ir_emit_vertex *);
+ void visit(ir_end_primitive *);
+ uint32_t gather_channel(ir_texture *ir, int sampler);
void swizzle_result(ir_texture *ir, fs_reg orig_val, int sampler);
bool can_do_source_mods(fs_inst *inst);
uint32_t condition);
fs_inst *LRP(fs_reg dst, fs_reg a, fs_reg y, fs_reg x);
fs_inst *DEP_RESOLVE_MOV(int grf);
+ fs_inst *BFREV(fs_reg dst, fs_reg value);
+ fs_inst *BFE(fs_reg dst, fs_reg bits, fs_reg offset, fs_reg value);
+ fs_inst *BFI1(fs_reg dst, fs_reg bits, fs_reg offset);
+ fs_inst *BFI2(fs_reg dst, fs_reg bfi1_dst, fs_reg insert, fs_reg base);
+ fs_inst *FBH(fs_reg dst, fs_reg value);
+ fs_inst *FBL(fs_reg dst, fs_reg value);
+ fs_inst *CBIT(fs_reg dst, fs_reg value);
+ fs_inst *MAD(fs_reg dst, fs_reg c, fs_reg b, fs_reg a);
+ fs_inst *ADDC(fs_reg dst, fs_reg src0, fs_reg src1);
+ fs_inst *SUBB(fs_reg dst, fs_reg src0, fs_reg src1);
+ fs_inst *SEL(fs_reg dst, fs_reg src0, fs_reg src1);
int type_size(const struct glsl_type *type);
fs_inst *get_instruction_generating_reg(fs_inst *start,
uint32_t const_offset);
bool run();
+ void assign_binding_table_offsets();
void setup_payload_gen4();
void setup_payload_gen6();
void assign_curb_setup();
void calculate_urb_setup();
void assign_urb_setup();
- bool assign_regs();
+ bool assign_regs(bool allow_spilling);
void assign_regs_trivial();
+ void get_used_mrfs(bool *mrf_used);
void setup_payload_interference(struct ra_graph *g, int payload_reg_count,
int first_payload_node);
void setup_mrf_hack_interference(struct ra_graph *g,
void compact_virtual_grfs();
void move_uniform_array_access_to_pull_constants();
void setup_pull_constants();
+ void invalidate_live_intervals();
void calculate_live_intervals();
+ void calculate_register_pressure();
bool opt_algebraic();
bool opt_cse();
bool opt_cse_local(bblock_t *block, exec_list *aeb);
bool opt_copy_propagate_local(void *mem_ctx, bblock_t *block,
exec_list *acp);
bool register_coalesce();
- bool register_coalesce_2();
bool compute_to_mrf();
bool dead_code_eliminate();
+ bool dead_code_eliminate_local();
bool remove_dead_constants();
bool remove_duplicate_mrf_writes();
bool virtual_grf_interferes(int a, int b);
- void schedule_instructions(bool post_reg_alloc);
+ void schedule_instructions(instruction_scheduler_mode mode);
void insert_gen4_send_dependency_workarounds();
void insert_gen4_pre_send_dependency_workarounds(fs_inst *inst);
void insert_gen4_post_send_dependency_workarounds(fs_inst *inst);
void push_force_uncompressed();
void pop_force_uncompressed();
- void push_force_sechalf();
- void pop_force_sechalf();
void emit_dummy_fs();
fs_reg *emit_fragcoord_interpolation(ir_variable *ir);
fs_inst *emit_linterp(const fs_reg &attr, const fs_reg &interp,
glsl_interp_qualifier interpolation_mode,
- bool is_centroid);
+ bool is_centroid, bool is_sample);
fs_reg *emit_frontfacing_interpolation(ir_variable *ir);
+ fs_reg *emit_samplepos_setup(ir_variable *ir);
+ fs_reg *emit_sampleid_setup(ir_variable *ir);
+ fs_reg *emit_samplemaskin_setup(ir_variable *ir);
fs_reg *emit_general_interpolation(ir_variable *ir);
void emit_interpolation_setup_gen4();
void emit_interpolation_setup_gen6();
+ void compute_sample_position(fs_reg dst, fs_reg int_sample_pos);
fs_reg rescale_texcoord(ir_texture *ir, fs_reg coordinate,
bool is_rect, int sampler, int texunit);
fs_inst *emit_texture_gen4(ir_texture *ir, fs_reg dst, fs_reg coordinate,
fs_reg sample_index);
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 sample_index);
+ fs_reg sample_index, fs_reg mcs, int sampler);
+ fs_reg emit_mcs_fetch(ir_texture *ir, fs_reg coordinate, int sampler);
+ void emit_gen6_gather_wa(uint8_t wa, fs_reg dst);
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);
fs_reg src0, fs_reg src1);
bool try_emit_saturate(ir_expression *ir);
bool try_emit_mad(ir_expression *ir, int mul_arg);
+ void try_replace_with_sel();
+ bool opt_peephole_sel();
+ bool opt_peephole_predicated_break();
+ bool opt_saturate_propagation();
void emit_bool_to_cond_code(ir_rvalue *condition);
void emit_if_gen6(ir_if *ir);
- void emit_unspill(fs_inst *inst, fs_reg reg, uint32_t spill_offset);
+ void emit_unspill(fs_inst *inst, fs_reg reg, uint32_t spill_offset,
+ int count);
void emit_fragment_program_code();
void setup_fp_regs();
fs_reg dst, fs_reg src0, fs_reg src1, fs_reg one);
void emit_color_write(int target, int index, int first_color_mrf);
+ void emit_alpha_test();
void emit_fb_writes();
void emit_shader_time_begin();
void emit_shader_time_write(enum shader_time_shader_type type,
fs_reg value);
+ void emit_untyped_atomic(unsigned atomic_op, unsigned surf_index,
+ fs_reg dst, fs_reg offset, fs_reg src0,
+ fs_reg src1);
+
+ void emit_untyped_surface_read(unsigned surf_index, fs_reg dst,
+ fs_reg offset);
+
bool try_rewrite_rhs_to_dst(ir_assignment *ir,
fs_reg dst,
fs_reg src,
void setup_builtin_uniform_values(ir_variable *ir);
int implied_mrf_writes(fs_inst *inst);
- void dump_instructions();
- void dump_instruction(fs_inst *inst);
+ virtual void dump_instructions();
+ void dump_instruction(backend_instruction *inst);
+
+ void visit_atomic_counter_intrinsic(ir_call *ir);
struct gl_fragment_program *fp;
struct brw_wm_compile *c;
int *virtual_grf_sizes;
int virtual_grf_count;
int virtual_grf_array_size;
- int *virtual_grf_def;
- int *virtual_grf_use;
- bool live_intervals_valid;
+ int *virtual_grf_start;
+ int *virtual_grf_end;
+ brw::fs_live_variables *live_intervals;
+
+ int *regs_live_at_ip;
/* This is the map from UNIFORM hw_reg + reg_offset as generated by
* the visitor to the packed uniform number after
* uniform index.
*/
int *params_remap;
+ int nr_params_remap;
struct hash_table *variable_ht;
fs_reg frag_depth;
+ fs_reg sample_mask;
fs_reg outputs[BRW_MAX_DRAW_BUFFERS];
unsigned output_components[BRW_MAX_DRAW_BUFFERS];
fs_reg dual_src_output;
int first_non_payload_grf;
/** Either BRW_MAX_GRF or GEN7_MRF_HACK_START */
int max_grf;
- int urb_setup[VARYING_SLOT_MAX];
fs_reg *fp_temp_regs;
fs_reg *fp_input_regs;
fs_reg shader_start_time;
int grf_used;
+ bool spilled_any_registers;
const unsigned dispatch_width; /**< 8 or 16 */
int force_uncompressed_stack;
- int force_sechalf_stack;
};
/**
const unsigned *generate_assembly(exec_list *simd8_instructions,
exec_list *simd16_instructions,
- unsigned *assembly_size);
+ unsigned *assembly_size,
+ FILE *dump_file = NULL);
private:
- void generate_code(exec_list *instructions);
+ void generate_code(exec_list *instructions, FILE *dump_file);
void generate_fb_write(fs_inst *inst);
+ void generate_blorp_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_math_gen4(fs_inst *inst,
struct brw_reg dst,
struct brw_reg src);
+ void generate_math_g45(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_scratch_write(fs_inst *inst, struct brw_reg src);
+ void generate_scratch_read(fs_inst *inst, struct brw_reg dst);
+ void generate_scratch_read_gen7(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);
struct brw_reg index,
struct brw_reg offset);
void generate_mov_dispatch_to_flags(fs_inst *inst);
+
+ void generate_set_omask(fs_inst *inst,
+ struct brw_reg dst,
+ struct brw_reg sample_mask);
+
+ void generate_set_sample_id(fs_inst *inst,
+ struct brw_reg dst,
+ struct brw_reg src0,
+ struct brw_reg src1);
+
void generate_set_simd4x2_offset(fs_inst *inst,
struct brw_reg dst,
struct brw_reg offset);
struct brw_reg offset,
struct brw_reg value);
+ void generate_untyped_atomic(fs_inst *inst,
+ struct brw_reg dst,
+ struct brw_reg atomic_op,
+ struct brw_reg surf_index);
+
+ void generate_untyped_surface_read(fs_inst *inst,
+ struct brw_reg dst,
+ struct brw_reg surf_index);
+
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 */
void *mem_ctx;
};
+/**
+ * The fragment shader code generator.
+ *
+ * Translates FS IR to actual i965 assembly code.
+ */
+class gen8_fs_generator : public gen8_generator
+{
+public:
+ gen8_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);
+ ~gen8_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_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(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
+ void generate_math2(fs_inst *inst, struct brw_reg dst,
+ struct brw_reg src0, struct brw_reg src1);
+ 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_scratch_write(fs_inst *inst, struct brw_reg src);
+ void generate_scratch_read(fs_inst *inst, struct brw_reg dst);
+ void generate_scratch_read_gen7(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_varying_pull_constant_load(fs_inst *inst,
+ struct brw_reg dst,
+ struct brw_reg index,
+ struct brw_reg offset);
+ void generate_mov_dispatch_to_flags(fs_inst *ir);
+ void generate_set_simd4x2_offset(fs_inst *ir,
+ struct brw_reg dst,
+ struct brw_reg offset);
+ void generate_discard_jump(fs_inst *ir);
+
+ void patch_discard_jumps_to_fb_writes();
+
+ void mark_surface_used(unsigned surf_index);
+
+ struct brw_wm_compile *c;
+ const struct gl_fragment_program *fp;
+
+ unsigned dispatch_width; /** 8 or 16 */
+
+ bool dual_source_output;
+
+ exec_list discard_halt_patches;
+};
+
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);
+
+struct brw_reg brw_reg_from_fs_reg(fs_reg *reg);