2 * Copyright © 2016 Broadcom
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
24 #ifndef V3D_COMPILER_H
25 #define V3D_COMPILER_H
34 #include "util/macros.h"
35 #include "common/v3d_debug.h"
36 #include "common/v3d_device_info.h"
37 #include "compiler/nir/nir.h"
38 #include "util/list.h"
39 #include "util/u_math.h"
41 #include "qpu/qpu_instr.h"
42 #include "pipe/p_state.h"
44 #define V3D_MAX_TEXTURE_SAMPLERS 32
45 #define V3D_MAX_SAMPLES 4
46 #define V3D_MAX_FS_INPUTS 64
47 #define V3D_MAX_VS_INPUTS 64
51 struct v3d_fs_inputs
{
53 * Array of the meanings of the VPM inputs this shader needs.
55 * It doesn't include those that aren't part of the VPM, like
56 * point/line coordinates.
58 struct v3d_varying_slot
*input_slots
;
63 /** An unused source or destination register. */
66 /** A physical register, such as the W coordinate payload. */
68 /** One of the regsiters for fixed function interactions. */
72 * A virtual register, that will be allocated to actual accumulator
73 * or physical registers later.
81 * VPM reads use this with an index value to say what part of the VPM
87 * Stores an immediate value in the index field that will be used
88 * directly by qpu_load_imm().
93 * Stores an immediate value in the index field that can be turned
94 * into a small immediate field by qpu_encode_small_immediate().
100 * A reference to a QPU register or a virtual temp register.
107 static inline struct qreg
vir_reg(enum qfile file
, uint32_t index
)
109 return (struct qreg
){file
, index
};
113 * A reference to an actual register at the QPU level, for register
122 /** Entry in qblock->instructions */
123 struct list_head link
;
126 * The instruction being wrapped. Its condition codes, pack flags,
127 * signals, etc. will all be used, with just the register references
128 * being replaced by the contents of qinst->dst and qinst->src[].
130 struct v3d_qpu_instr qpu
;
132 /* Pre-register-allocation references to src/dst registers */
135 bool cond_is_exec_mask
;
136 bool has_implicit_uniform
;
139 /* After vir_to_qpu.c: If instr reads a uniform, which uniform from
140 * the uncompiled stream it is.
145 enum quniform_contents
{
147 * Indicates that a constant 32-bit value is copied from the program's
152 * Indicates that the program's uniform contents are used as an index
153 * into the GL uniform storage.
158 * Scaling factors from clip coordinates to relative to the viewport
161 * This is used by the coordinate and vertex shaders to produce the
162 * 32-bit entry consisting of 2 16-bit fields with 12.4 signed fixed
163 * point offsets from the viewport ccenter.
165 QUNIFORM_VIEWPORT_X_SCALE
,
166 QUNIFORM_VIEWPORT_Y_SCALE
,
169 QUNIFORM_VIEWPORT_Z_OFFSET
,
170 QUNIFORM_VIEWPORT_Z_SCALE
,
172 QUNIFORM_USER_CLIP_PLANE
,
175 * A reference to a V3D 3.x texture config parameter 0 uniform.
177 * This is a uniform implicitly loaded with a QPU_W_TMU* write, which
178 * defines texture type, miplevels, and such. It will be found as a
179 * parameter to the first QOP_TEX_[STRB] instruction in a sequence.
181 QUNIFORM_TEXTURE_CONFIG_P0_0
,
182 QUNIFORM_TEXTURE_CONFIG_P0_1
,
183 QUNIFORM_TEXTURE_CONFIG_P0_2
,
184 QUNIFORM_TEXTURE_CONFIG_P0_3
,
185 QUNIFORM_TEXTURE_CONFIG_P0_4
,
186 QUNIFORM_TEXTURE_CONFIG_P0_5
,
187 QUNIFORM_TEXTURE_CONFIG_P0_6
,
188 QUNIFORM_TEXTURE_CONFIG_P0_7
,
189 QUNIFORM_TEXTURE_CONFIG_P0_8
,
190 QUNIFORM_TEXTURE_CONFIG_P0_9
,
191 QUNIFORM_TEXTURE_CONFIG_P0_10
,
192 QUNIFORM_TEXTURE_CONFIG_P0_11
,
193 QUNIFORM_TEXTURE_CONFIG_P0_12
,
194 QUNIFORM_TEXTURE_CONFIG_P0_13
,
195 QUNIFORM_TEXTURE_CONFIG_P0_14
,
196 QUNIFORM_TEXTURE_CONFIG_P0_15
,
197 QUNIFORM_TEXTURE_CONFIG_P0_16
,
198 QUNIFORM_TEXTURE_CONFIG_P0_17
,
199 QUNIFORM_TEXTURE_CONFIG_P0_18
,
200 QUNIFORM_TEXTURE_CONFIG_P0_19
,
201 QUNIFORM_TEXTURE_CONFIG_P0_20
,
202 QUNIFORM_TEXTURE_CONFIG_P0_21
,
203 QUNIFORM_TEXTURE_CONFIG_P0_22
,
204 QUNIFORM_TEXTURE_CONFIG_P0_23
,
205 QUNIFORM_TEXTURE_CONFIG_P0_24
,
206 QUNIFORM_TEXTURE_CONFIG_P0_25
,
207 QUNIFORM_TEXTURE_CONFIG_P0_26
,
208 QUNIFORM_TEXTURE_CONFIG_P0_27
,
209 QUNIFORM_TEXTURE_CONFIG_P0_28
,
210 QUNIFORM_TEXTURE_CONFIG_P0_29
,
211 QUNIFORM_TEXTURE_CONFIG_P0_30
,
212 QUNIFORM_TEXTURE_CONFIG_P0_31
,
213 QUNIFORM_TEXTURE_CONFIG_P0_32
,
216 * A reference to a V3D 3.x texture config parameter 1 uniform.
218 * This is a uniform implicitly loaded with a QPU_W_TMU* write, which
219 * has the pointer to the indirect texture state. Our data[] field
220 * will have a packed p1 value, but the address field will be just
221 * which texture unit's texture should be referenced.
223 QUNIFORM_TEXTURE_CONFIG_P1
,
225 /* A a V3D 4.x texture config parameter. The high 8 bits will be
226 * which texture or sampler is being sampled, and the driver must
227 * replace the address field with the appropriate address.
229 QUNIFORM_TMU_CONFIG_P0
,
230 QUNIFORM_TMU_CONFIG_P1
,
232 QUNIFORM_TEXTURE_FIRST_LEVEL
,
234 QUNIFORM_TEXTURE_WIDTH
,
235 QUNIFORM_TEXTURE_HEIGHT
,
236 QUNIFORM_TEXTURE_DEPTH
,
237 QUNIFORM_TEXTURE_ARRAY_SIZE
,
238 QUNIFORM_TEXTURE_LEVELS
,
242 QUNIFORM_TEXRECT_SCALE_X
,
243 QUNIFORM_TEXRECT_SCALE_Y
,
245 QUNIFORM_TEXTURE_BORDER_COLOR
,
250 QUNIFORM_SAMPLE_MASK
,
253 * Returns the the offset of the scratch buffer for register spilling.
255 QUNIFORM_SPILL_OFFSET
,
256 QUNIFORM_SPILL_SIZE_PER_THREAD
,
259 struct v3d_varying_slot
{
260 uint8_t slot_and_component
;
263 static inline struct v3d_varying_slot
264 v3d_slot_from_slot_and_component(uint8_t slot
, uint8_t component
)
266 assert(slot
< 255 / 4);
267 return (struct v3d_varying_slot
){ (slot
<< 2) + component
};
270 static inline uint8_t v3d_slot_get_slot(struct v3d_varying_slot slot
)
272 return slot
.slot_and_component
>> 2;
275 static inline uint8_t v3d_slot_get_component(struct v3d_varying_slot slot
)
277 return slot
.slot_and_component
& 3;
280 struct v3d_ubo_range
{
282 * offset in bytes from the start of the ubo where this range is
285 * Only set once used is set.
290 * offset in bytes from the start of the gallium uniforms where the
295 /** size in bytes of this ubo range */
304 uint8_t return_channels
;
307 unsigned compare_mode
:1;
308 unsigned compare_func
:3;
314 uint16_t msaa_width
, msaa_height
;
317 } tex
[V3D_MAX_TEXTURE_SAMPLERS
];
327 bool point_coord_upper_left
;
330 bool sample_coverage
;
331 bool sample_alpha_to_coverage
;
332 bool sample_alpha_to_one
;
334 bool shade_model_flat
;
336 uint8_t swap_color_rb
;
337 /* Mask of which render targets need to be written as 32-bit floats */
338 uint8_t f32_color_rb
;
339 uint8_t alpha_test_func
;
340 uint8_t logicop_func
;
341 uint32_t point_sprite_mask
;
343 struct pipe_rt_blend_state blend
;
349 struct v3d_varying_slot fs_inputs
[V3D_MAX_FS_INPUTS
];
350 uint8_t num_fs_inputs
;
353 bool per_vertex_point_size
;
357 /** A basic block of VIR intructions. */
359 struct list_head link
;
361 struct list_head instructions
;
363 struct set
*predecessors
;
364 struct qblock
*successors
[2];
368 /* Instruction IPs for the first and last instruction of the block.
369 * Set by qpu_schedule.c.
371 uint32_t start_qpu_ip
;
374 /* Instruction IP for the branch instruction of the block. Set by
377 uint32_t branch_qpu_ip
;
379 /** Offset within the uniform stream at the start of the block. */
380 uint32_t start_uniform
;
381 /** Offset within the uniform stream of the branch instruction */
382 uint32_t branch_uniform
;
384 /** @{ used by v3d_vir_live_variables.c */
387 BITSET_WORD
*live_in
;
388 BITSET_WORD
*live_out
;
389 int start_ip
, end_ip
;
393 /** Which util/list.h add mode we should use when inserting an instruction. */
394 enum vir_cursor_mode
{
400 * Tracking structure for where new instructions should be inserted. Create
401 * with one of the vir_after_inst()-style helper functions.
403 * This does not protect against removal of the block or instruction, so we
404 * have an assert in instruction removal to try to catch it.
407 enum vir_cursor_mode mode
;
408 struct list_head
*link
;
411 static inline struct vir_cursor
412 vir_before_inst(struct qinst
*inst
)
414 return (struct vir_cursor
){ vir_cursor_addtail
, &inst
->link
};
417 static inline struct vir_cursor
418 vir_after_inst(struct qinst
*inst
)
420 return (struct vir_cursor
){ vir_cursor_add
, &inst
->link
};
423 static inline struct vir_cursor
424 vir_before_block(struct qblock
*block
)
426 return (struct vir_cursor
){ vir_cursor_add
, &block
->instructions
};
429 static inline struct vir_cursor
430 vir_after_block(struct qblock
*block
)
432 return (struct vir_cursor
){ vir_cursor_addtail
, &block
->instructions
};
436 * Compiler state saved across compiler invocations, for any expensive global
439 struct v3d_compiler
{
440 const struct v3d_device_info
*devinfo
;
441 struct ra_regs
*regs
;
442 unsigned int reg_class_phys
[3];
443 unsigned int reg_class_phys_or_acc
[3];
447 const struct v3d_device_info
*devinfo
;
449 nir_function_impl
*impl
;
450 struct exec_list
*cf_node_list
;
451 const struct v3d_compiler
*compiler
;
454 * Mapping from nir_register * or nir_ssa_def * to array of struct
455 * qreg for the values.
457 struct hash_table
*def_ht
;
459 /* For each temp, the instruction generating its value. */
461 uint32_t defs_array_size
;
464 * Inputs to the shader, arranged by TGSI declaration order.
466 * Not all fragment shader QFILE_VARY reads are present in this array.
469 struct qreg
*outputs
;
470 bool msaa_per_sample_output
;
471 struct qreg color_reads
[V3D_MAX_SAMPLES
];
472 struct qreg sample_colors
[V3D_MAX_SAMPLES
];
473 uint32_t inputs_array_size
;
474 uint32_t outputs_array_size
;
475 uint32_t uniforms_array_size
;
477 /* Booleans for whether the corresponding QFILE_VARY[i] is
478 * flat-shaded. This includes gl_FragColor flat-shading, which is
479 * customized based on the shademodel_flat shader key.
481 uint32_t flat_shade_flags
[BITSET_WORDS(V3D_MAX_FS_INPUTS
)];
483 struct v3d_ubo_range
*ubo_ranges
;
484 bool *ubo_range_used
;
485 uint32_t ubo_ranges_array_size
;
486 /** Number of uniform areas tracked in ubo_ranges. */
487 uint32_t num_ubo_ranges
;
488 uint32_t next_ubo_dst_offset
;
490 /* State for whether we're executing on each channel currently. 0 if
491 * yes, otherwise a block number + 1 that the channel jumped to.
495 struct qreg line_x
, point_x
, point_y
;
498 * Instance ID, which comes in before the vertex attribute payload if
499 * the shader record requests it.
504 * Vertex ID, which comes in before the vertex attribute payload
505 * (after Instance ID) if the shader record requests it.
509 /* Fragment shader payload regs. */
510 struct qreg payload_w
, payload_w_centroid
, payload_z
;
512 uint8_t vattr_sizes
[V3D_MAX_VS_INPUTS
];
513 uint32_t num_vpm_writes
;
515 /* Size in bytes of registers that have been spilled. This is how much
516 * space needs to be available in the spill BO per thread per QPU.
519 /* Shader-db stats for register spilling. */
520 uint32_t spills
, fills
;
522 * Register spilling's per-thread base address, shared between each
523 * spill/fill's addressing calculations.
525 struct qreg spill_base
;
526 /* Bit vector of which temps may be spilled */
527 BITSET_WORD
*spillable
;
530 * Array of the VARYING_SLOT_* of all FS QFILE_VARY reads.
532 * This includes those that aren't part of the VPM varyings, like
533 * point/line coordinates.
535 struct v3d_varying_slot input_slots
[V3D_MAX_FS_INPUTS
];
538 * An entry per outputs[] in the VS indicating what the VARYING_SLOT_*
539 * of the output is. Used to emit from the VS in the order that the
542 struct v3d_varying_slot
*output_slots
;
544 struct pipe_shader_state
*shader_state
;
546 struct v3d_fs_key
*fs_key
;
547 struct v3d_vs_key
*vs_key
;
549 /* Live ranges of temps. */
550 int *temp_start
, *temp_end
;
551 bool live_intervals_valid
;
553 uint32_t *uniform_data
;
554 enum quniform_contents
*uniform_contents
;
555 uint32_t uniform_array_size
;
556 uint32_t num_uniforms
;
557 uint32_t num_outputs
;
558 uint32_t output_position_index
;
559 nir_variable
*output_color_var
[4];
560 uint32_t output_point_size_index
;
561 uint32_t output_sample_mask_index
;
566 struct vir_cursor cursor
;
567 struct list_head blocks
;
568 int next_block_index
;
569 struct qblock
*cur_block
;
570 struct qblock
*loop_cont_block
;
571 struct qblock
*loop_break_block
;
574 uint32_t qpu_inst_count
;
575 uint32_t qpu_inst_size
;
577 /* For the FS, the number of varying inputs not counting the
578 * point/line varyings payload
583 * Number of inputs from num_inputs remaining to be queued to the read
586 uint32_t num_inputs_remaining
;
588 /* Number of inputs currently in the read FIFO for the VS/CS */
589 uint32_t num_inputs_in_fifo
;
591 /** Next offset in the VPM to read from in the VS/CS */
592 uint32_t vpm_read_offset
;
597 /* Set to compile program in in 1x, 2x, or 4x threaded mode, where
598 * SIG_THREAD_SWITCH is used to hide texturing latency at the cost of
599 * limiting ourselves to the part of the physical reg space.
601 * On V3D 3.x, 2x or 4x divide the physical reg space by 2x or 4x. On
602 * V3D 4.x, all shaders are 2x threaded, and 4x only divides the
603 * physical reg space in half.
606 struct qinst
*last_thrsw
;
607 bool last_thrsw_at_top_level
;
612 struct v3d_uniform_list
{
613 enum quniform_contents
*contents
;
618 struct v3d_prog_data
{
619 struct v3d_uniform_list uniforms
;
621 struct v3d_ubo_range
*ubo_ranges
;
622 uint32_t num_ubo_ranges
;
629 /* For threads > 1, whether the program should be dispatched in the
630 * after-final-THRSW state.
635 struct v3d_vs_prog_data
{
636 struct v3d_prog_data base
;
638 bool uses_iid
, uses_vid
;
640 /* Number of components read from each vertex attribute. */
641 uint8_t vattr_sizes
[32];
643 /* Total number of components read, for the shader state record. */
644 uint32_t vpm_input_size
;
646 /* Total number of components written, for the shader state record. */
647 uint32_t vpm_output_size
;
650 struct v3d_fs_prog_data
{
651 struct v3d_prog_data base
;
653 struct v3d_varying_slot input_slots
[V3D_MAX_FS_INPUTS
];
655 /* Array of flat shade flags.
657 * Each entry is only 24 bits (high 8 bits 0), to match the hardware
660 uint32_t flat_shade_flags
[((V3D_MAX_FS_INPUTS
- 1) / 24) + 1];
666 /* Special nir_load_input intrinsic index for loading the current TLB
669 #define V3D_NIR_TLB_COLOR_READ_INPUT 2000000000
671 #define V3D_NIR_MS_MASK_OUTPUT 2000000000
673 extern const nir_shader_compiler_options v3d_nir_options
;
675 const struct v3d_compiler
*v3d_compiler_init(const struct v3d_device_info
*devinfo
);
676 void v3d_compiler_free(const struct v3d_compiler
*compiler
);
677 void v3d_optimize_nir(struct nir_shader
*s
);
679 uint64_t *v3d_compile_vs(const struct v3d_compiler
*compiler
,
680 struct v3d_vs_key
*key
,
681 struct v3d_vs_prog_data
*prog_data
,
683 int program_id
, int variant_id
,
684 uint32_t *final_assembly_size
);
686 uint64_t *v3d_compile_fs(const struct v3d_compiler
*compiler
,
687 struct v3d_fs_key
*key
,
688 struct v3d_fs_prog_data
*prog_data
,
690 int program_id
, int variant_id
,
691 uint32_t *final_assembly_size
);
693 void v3d_nir_to_vir(struct v3d_compile
*c
);
695 void vir_compile_destroy(struct v3d_compile
*c
);
696 const char *vir_get_stage_name(struct v3d_compile
*c
);
697 struct qblock
*vir_new_block(struct v3d_compile
*c
);
698 void vir_set_emit_block(struct v3d_compile
*c
, struct qblock
*block
);
699 void vir_link_blocks(struct qblock
*predecessor
, struct qblock
*successor
);
700 struct qblock
*vir_entry_block(struct v3d_compile
*c
);
701 struct qblock
*vir_exit_block(struct v3d_compile
*c
);
702 struct qinst
*vir_add_inst(enum v3d_qpu_add_op op
, struct qreg dst
,
703 struct qreg src0
, struct qreg src1
);
704 struct qinst
*vir_mul_inst(enum v3d_qpu_mul_op op
, struct qreg dst
,
705 struct qreg src0
, struct qreg src1
);
706 struct qinst
*vir_branch_inst(enum v3d_qpu_branch_cond cond
, struct qreg src0
);
707 void vir_remove_instruction(struct v3d_compile
*c
, struct qinst
*qinst
);
708 struct qreg
vir_uniform(struct v3d_compile
*c
,
709 enum quniform_contents contents
,
711 void vir_schedule_instructions(struct v3d_compile
*c
);
712 struct v3d_qpu_instr
v3d_qpu_nop(void);
714 struct qreg
vir_emit_def(struct v3d_compile
*c
, struct qinst
*inst
);
715 struct qinst
*vir_emit_nondef(struct v3d_compile
*c
, struct qinst
*inst
);
716 void vir_set_cond(struct qinst
*inst
, enum v3d_qpu_cond cond
);
717 void vir_set_pf(struct qinst
*inst
, enum v3d_qpu_pf pf
);
718 void vir_set_unpack(struct qinst
*inst
, int src
,
719 enum v3d_qpu_input_unpack unpack
);
721 struct qreg
vir_get_temp(struct v3d_compile
*c
);
722 void vir_emit_last_thrsw(struct v3d_compile
*c
);
723 void vir_calculate_live_intervals(struct v3d_compile
*c
);
724 bool vir_has_implicit_uniform(struct qinst
*inst
);
725 int vir_get_implicit_uniform_src(struct qinst
*inst
);
726 int vir_get_non_sideband_nsrc(struct qinst
*inst
);
727 int vir_get_nsrc(struct qinst
*inst
);
728 bool vir_has_side_effects(struct v3d_compile
*c
, struct qinst
*inst
);
729 bool vir_get_add_op(struct qinst
*inst
, enum v3d_qpu_add_op
*op
);
730 bool vir_get_mul_op(struct qinst
*inst
, enum v3d_qpu_mul_op
*op
);
731 bool vir_is_raw_mov(struct qinst
*inst
);
732 bool vir_is_tex(struct qinst
*inst
);
733 bool vir_is_add(struct qinst
*inst
);
734 bool vir_is_mul(struct qinst
*inst
);
735 bool vir_is_float_input(struct qinst
*inst
);
736 bool vir_depends_on_flags(struct qinst
*inst
);
737 bool vir_writes_r3(const struct v3d_device_info
*devinfo
, struct qinst
*inst
);
738 bool vir_writes_r4(const struct v3d_device_info
*devinfo
, struct qinst
*inst
);
739 struct qreg
vir_follow_movs(struct v3d_compile
*c
, struct qreg reg
);
740 uint8_t vir_channels_written(struct qinst
*inst
);
741 struct qreg
ntq_get_src(struct v3d_compile
*c
, nir_src src
, int i
);
742 void ntq_store_dest(struct v3d_compile
*c
, nir_dest
*dest
, int chan
,
744 void vir_emit_thrsw(struct v3d_compile
*c
);
746 void vir_dump(struct v3d_compile
*c
);
747 void vir_dump_inst(struct v3d_compile
*c
, struct qinst
*inst
);
749 void vir_validate(struct v3d_compile
*c
);
751 void vir_optimize(struct v3d_compile
*c
);
752 bool vir_opt_algebraic(struct v3d_compile
*c
);
753 bool vir_opt_constant_folding(struct v3d_compile
*c
);
754 bool vir_opt_copy_propagate(struct v3d_compile
*c
);
755 bool vir_opt_dead_code(struct v3d_compile
*c
);
756 bool vir_opt_peephole_sf(struct v3d_compile
*c
);
757 bool vir_opt_small_immediates(struct v3d_compile
*c
);
758 bool vir_opt_vpm(struct v3d_compile
*c
);
759 void v3d_nir_lower_blend(nir_shader
*s
, struct v3d_compile
*c
);
760 void v3d_nir_lower_io(nir_shader
*s
, struct v3d_compile
*c
);
761 void v3d_nir_lower_txf_ms(nir_shader
*s
, struct v3d_compile
*c
);
762 void vir_lower_uniforms(struct v3d_compile
*c
);
764 void v3d33_vir_vpm_read_setup(struct v3d_compile
*c
, int num_components
);
765 void v3d33_vir_vpm_write_setup(struct v3d_compile
*c
);
766 void v3d33_vir_emit_tex(struct v3d_compile
*c
, nir_tex_instr
*instr
);
767 void v3d40_vir_emit_tex(struct v3d_compile
*c
, nir_tex_instr
*instr
);
769 void v3d_vir_to_qpu(struct v3d_compile
*c
, struct qpu_reg
*temp_registers
);
770 uint32_t v3d_qpu_schedule_instructions(struct v3d_compile
*c
);
771 void qpu_validate(struct v3d_compile
*c
);
772 struct qpu_reg
*v3d_register_allocate(struct v3d_compile
*c
, bool *spilled
);
773 bool vir_init_reg_sets(struct v3d_compiler
*compiler
);
775 void vir_PF(struct v3d_compile
*c
, struct qreg src
, enum v3d_qpu_pf pf
);
778 quniform_contents_is_texture_p0(enum quniform_contents contents
)
780 return (contents
>= QUNIFORM_TEXTURE_CONFIG_P0_0
&&
781 contents
< (QUNIFORM_TEXTURE_CONFIG_P0_0
+
782 V3D_MAX_TEXTURE_SAMPLERS
));
785 static inline struct qreg
786 vir_uniform_ui(struct v3d_compile
*c
, uint32_t ui
)
788 return vir_uniform(c
, QUNIFORM_CONSTANT
, ui
);
791 static inline struct qreg
792 vir_uniform_f(struct v3d_compile
*c
, float f
)
794 return vir_uniform(c
, QUNIFORM_CONSTANT
, fui(f
));
797 #define VIR_ALU0(name, vir_inst, op) \
798 static inline struct qreg \
799 vir_##name(struct v3d_compile *c) \
801 return vir_emit_def(c, vir_inst(op, c->undef, \
802 c->undef, c->undef)); \
804 static inline struct qinst * \
805 vir_##name##_dest(struct v3d_compile *c, struct qreg dest) \
807 return vir_emit_nondef(c, vir_inst(op, dest, \
808 c->undef, c->undef)); \
811 #define VIR_ALU1(name, vir_inst, op) \
812 static inline struct qreg \
813 vir_##name(struct v3d_compile *c, struct qreg a) \
815 return vir_emit_def(c, vir_inst(op, c->undef, \
818 static inline struct qinst * \
819 vir_##name##_dest(struct v3d_compile *c, struct qreg dest, \
822 return vir_emit_nondef(c, vir_inst(op, dest, a, \
826 #define VIR_ALU2(name, vir_inst, op) \
827 static inline struct qreg \
828 vir_##name(struct v3d_compile *c, struct qreg a, struct qreg b) \
830 return vir_emit_def(c, vir_inst(op, c->undef, a, b)); \
832 static inline struct qinst * \
833 vir_##name##_dest(struct v3d_compile *c, struct qreg dest, \
834 struct qreg a, struct qreg b) \
836 return vir_emit_nondef(c, vir_inst(op, dest, a, b)); \
839 #define VIR_NODST_0(name, vir_inst, op) \
840 static inline struct qinst * \
841 vir_##name(struct v3d_compile *c) \
843 return vir_emit_nondef(c, vir_inst(op, c->undef, \
844 c->undef, c->undef)); \
847 #define VIR_NODST_1(name, vir_inst, op) \
848 static inline struct qinst * \
849 vir_##name(struct v3d_compile *c, struct qreg a) \
851 return vir_emit_nondef(c, vir_inst(op, c->undef, \
855 #define VIR_NODST_2(name, vir_inst, op) \
856 static inline struct qinst * \
857 vir_##name(struct v3d_compile *c, struct qreg a, struct qreg b) \
859 return vir_emit_nondef(c, vir_inst(op, c->undef, \
863 #define VIR_A_ALU2(name) VIR_ALU2(name, vir_add_inst, V3D_QPU_A_##name)
864 #define VIR_M_ALU2(name) VIR_ALU2(name, vir_mul_inst, V3D_QPU_M_##name)
865 #define VIR_A_ALU1(name) VIR_ALU1(name, vir_add_inst, V3D_QPU_A_##name)
866 #define VIR_M_ALU1(name) VIR_ALU1(name, vir_mul_inst, V3D_QPU_M_##name)
867 #define VIR_A_ALU0(name) VIR_ALU0(name, vir_add_inst, V3D_QPU_A_##name)
868 #define VIR_M_ALU0(name) VIR_ALU0(name, vir_mul_inst, V3D_QPU_M_##name)
869 #define VIR_A_NODST_2(name) VIR_NODST_2(name, vir_add_inst, V3D_QPU_A_##name)
870 #define VIR_M_NODST_2(name) VIR_NODST_2(name, vir_mul_inst, V3D_QPU_M_##name)
871 #define VIR_A_NODST_1(name) VIR_NODST_1(name, vir_add_inst, V3D_QPU_A_##name)
872 #define VIR_M_NODST_1(name) VIR_NODST_1(name, vir_mul_inst, V3D_QPU_M_##name)
873 #define VIR_A_NODST_0(name) VIR_NODST_0(name, vir_add_inst, V3D_QPU_A_##name)
906 VIR_A_ALU1(LDVPMV_IN
)
907 VIR_A_ALU1(LDVPMV_OUT
)
915 VIR_A_NODST_1(VPMSETUP
)
939 VIR_M_NODST_2(MULTOP
)
944 static inline struct qinst
*
945 vir_MOV_cond(struct v3d_compile
*c
, enum v3d_qpu_cond cond
,
946 struct qreg dest
, struct qreg src
)
948 struct qinst
*mov
= vir_MOV_dest(c
, dest
, src
);
949 vir_set_cond(mov
, cond
);
953 static inline struct qreg
954 vir_SEL(struct v3d_compile
*c
, enum v3d_qpu_cond cond
,
955 struct qreg src0
, struct qreg src1
)
957 struct qreg t
= vir_get_temp(c
);
958 vir_MOV_dest(c
, t
, src1
);
959 vir_MOV_cond(c
, cond
, t
, src0
);
963 static inline struct qinst
*
964 vir_NOP(struct v3d_compile
*c
)
966 return vir_emit_nondef(c
, vir_add_inst(V3D_QPU_A_NOP
,
967 c
->undef
, c
->undef
, c
->undef
));
970 static inline struct qreg
971 vir_LDTMU(struct v3d_compile
*c
)
973 if (c
->devinfo
->ver
>= 41) {
974 struct qinst
*ldtmu
= vir_add_inst(V3D_QPU_A_NOP
, c
->undef
,
976 ldtmu
->qpu
.sig
.ldtmu
= true;
978 return vir_emit_def(c
, ldtmu
);
980 vir_NOP(c
)->qpu
.sig
.ldtmu
= true;
981 return vir_MOV(c
, vir_reg(QFILE_MAGIC
, V3D_QPU_WADDR_R4
));
985 static inline struct qreg
986 vir_UMUL(struct v3d_compile
*c
, struct qreg src0
, struct qreg src1
)
988 vir_MULTOP(c
, src0
, src1
);
989 return vir_UMUL24(c
, src0
, src1
);
993 static inline struct qreg
994 vir_LOAD_IMM(struct v3d_compile *c, uint32_t val)
996 return vir_emit_def(c, vir_inst(QOP_LOAD_IMM, c->undef,
997 vir_reg(QFILE_LOAD_IMM, val), c->undef));
1000 static inline struct qreg
1001 vir_LOAD_IMM_U2(struct v3d_compile *c, uint32_t val)
1003 return vir_emit_def(c, vir_inst(QOP_LOAD_IMM_U2, c->undef,
1004 vir_reg(QFILE_LOAD_IMM, val),
1007 static inline struct qreg
1008 vir_LOAD_IMM_I2(struct v3d_compile *c, uint32_t val)
1010 return vir_emit_def(c, vir_inst(QOP_LOAD_IMM_I2, c->undef,
1011 vir_reg(QFILE_LOAD_IMM, val),
1016 static inline struct qinst
*
1017 vir_BRANCH(struct v3d_compile
*c
, enum v3d_qpu_cond cond
)
1019 /* The actual uniform_data value will be set at scheduling time */
1020 return vir_emit_nondef(c
, vir_branch_inst(cond
, vir_uniform_ui(c
, 0)));
1023 #define vir_for_each_block(block, c) \
1024 list_for_each_entry(struct qblock, block, &c->blocks, link)
1026 #define vir_for_each_block_rev(block, c) \
1027 list_for_each_entry_rev(struct qblock, block, &c->blocks, link)
1029 /* Loop over the non-NULL members of the successors array. */
1030 #define vir_for_each_successor(succ, block) \
1031 for (struct qblock *succ = block->successors[0]; \
1033 succ = (succ == block->successors[1] ? NULL : \
1034 block->successors[1]))
1036 #define vir_for_each_inst(inst, block) \
1037 list_for_each_entry(struct qinst, inst, &block->instructions, link)
1039 #define vir_for_each_inst_rev(inst, block) \
1040 list_for_each_entry_rev(struct qinst, inst, &block->instructions, link)
1042 #define vir_for_each_inst_safe(inst, block) \
1043 list_for_each_entry_safe(struct qinst, inst, &block->instructions, link)
1045 #define vir_for_each_inst_inorder(inst, c) \
1046 vir_for_each_block(_block, c) \
1047 vir_for_each_inst(inst, _block)
1049 #endif /* V3D_COMPILER_H */