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 "common/v3d_limits.h"
38 #include "compiler/nir/nir.h"
39 #include "util/list.h"
40 #include "util/u_math.h"
42 #include "qpu/qpu_instr.h"
43 #include "pipe/p_state.h"
47 struct v3d_fs_inputs
{
49 * Array of the meanings of the VPM inputs this shader needs.
51 * It doesn't include those that aren't part of the VPM, like
52 * point/line coordinates.
54 struct v3d_varying_slot
*input_slots
;
59 /** An unused source or destination register. */
62 /** A physical register, such as the W coordinate payload. */
64 /** One of the regsiters for fixed function interactions. */
68 * A virtual register, that will be allocated to actual accumulator
69 * or physical registers later.
74 * VPM reads use this with an index value to say what part of the VPM
80 * Stores an immediate value in the index field that will be used
81 * directly by qpu_load_imm().
86 * Stores an immediate value in the index field that can be turned
87 * into a small immediate field by qpu_encode_small_immediate().
93 * A reference to a QPU register or a virtual temp register.
100 static inline struct qreg
vir_reg(enum qfile file
, uint32_t index
)
102 return (struct qreg
){file
, index
};
105 static inline struct qreg
vir_magic_reg(uint32_t index
)
107 return (struct qreg
){QFILE_MAGIC
, index
};
110 static inline struct qreg
vir_nop_reg(void)
112 return (struct qreg
){QFILE_NULL
, 0};
116 * A reference to an actual register at the QPU level, for register
126 /** Entry in qblock->instructions */
127 struct list_head link
;
130 * The instruction being wrapped. Its condition codes, pack flags,
131 * signals, etc. will all be used, with just the register references
132 * being replaced by the contents of qinst->dst and qinst->src[].
134 struct v3d_qpu_instr qpu
;
136 /* Pre-register-allocation references to src/dst registers */
141 /* If the instruction reads a uniform (other than through src[i].file
142 * == QFILE_UNIF), that uniform's index in c->uniform_contents. ~0
148 enum quniform_contents
{
150 * Indicates that a constant 32-bit value is copied from the program's
155 * Indicates that the program's uniform contents are used as an index
156 * into the GL uniform storage.
161 * Scaling factors from clip coordinates to relative to the viewport
164 * This is used by the coordinate and vertex shaders to produce the
165 * 32-bit entry consisting of 2 16-bit fields with 12.4 signed fixed
166 * point offsets from the viewport ccenter.
168 QUNIFORM_VIEWPORT_X_SCALE
,
169 QUNIFORM_VIEWPORT_Y_SCALE
,
172 QUNIFORM_VIEWPORT_Z_OFFSET
,
173 QUNIFORM_VIEWPORT_Z_SCALE
,
175 QUNIFORM_USER_CLIP_PLANE
,
178 * A reference to a V3D 3.x texture config parameter 0 uniform.
180 * This is a uniform implicitly loaded with a QPU_W_TMU* write, which
181 * defines texture type, miplevels, and such. It will be found as a
182 * parameter to the first QOP_TEX_[STRB] instruction in a sequence.
184 QUNIFORM_TEXTURE_CONFIG_P0_0
,
185 QUNIFORM_TEXTURE_CONFIG_P0_1
,
186 QUNIFORM_TEXTURE_CONFIG_P0_2
,
187 QUNIFORM_TEXTURE_CONFIG_P0_3
,
188 QUNIFORM_TEXTURE_CONFIG_P0_4
,
189 QUNIFORM_TEXTURE_CONFIG_P0_5
,
190 QUNIFORM_TEXTURE_CONFIG_P0_6
,
191 QUNIFORM_TEXTURE_CONFIG_P0_7
,
192 QUNIFORM_TEXTURE_CONFIG_P0_8
,
193 QUNIFORM_TEXTURE_CONFIG_P0_9
,
194 QUNIFORM_TEXTURE_CONFIG_P0_10
,
195 QUNIFORM_TEXTURE_CONFIG_P0_11
,
196 QUNIFORM_TEXTURE_CONFIG_P0_12
,
197 QUNIFORM_TEXTURE_CONFIG_P0_13
,
198 QUNIFORM_TEXTURE_CONFIG_P0_14
,
199 QUNIFORM_TEXTURE_CONFIG_P0_15
,
200 QUNIFORM_TEXTURE_CONFIG_P0_16
,
201 QUNIFORM_TEXTURE_CONFIG_P0_17
,
202 QUNIFORM_TEXTURE_CONFIG_P0_18
,
203 QUNIFORM_TEXTURE_CONFIG_P0_19
,
204 QUNIFORM_TEXTURE_CONFIG_P0_20
,
205 QUNIFORM_TEXTURE_CONFIG_P0_21
,
206 QUNIFORM_TEXTURE_CONFIG_P0_22
,
207 QUNIFORM_TEXTURE_CONFIG_P0_23
,
208 QUNIFORM_TEXTURE_CONFIG_P0_24
,
209 QUNIFORM_TEXTURE_CONFIG_P0_25
,
210 QUNIFORM_TEXTURE_CONFIG_P0_26
,
211 QUNIFORM_TEXTURE_CONFIG_P0_27
,
212 QUNIFORM_TEXTURE_CONFIG_P0_28
,
213 QUNIFORM_TEXTURE_CONFIG_P0_29
,
214 QUNIFORM_TEXTURE_CONFIG_P0_30
,
215 QUNIFORM_TEXTURE_CONFIG_P0_31
,
216 QUNIFORM_TEXTURE_CONFIG_P0_32
,
219 * A reference to a V3D 3.x texture config parameter 1 uniform.
221 * This is a uniform implicitly loaded with a QPU_W_TMU* write, which
222 * has the pointer to the indirect texture state. Our data[] field
223 * will have a packed p1 value, but the address field will be just
224 * which texture unit's texture should be referenced.
226 QUNIFORM_TEXTURE_CONFIG_P1
,
228 /* A V3D 4.x texture config parameter. The high 8 bits will be
229 * which texture or sampler is being sampled, and the driver must
230 * replace the address field with the appropriate address.
232 QUNIFORM_TMU_CONFIG_P0
,
233 QUNIFORM_TMU_CONFIG_P1
,
235 QUNIFORM_IMAGE_TMU_CONFIG_P0
,
237 QUNIFORM_TEXTURE_FIRST_LEVEL
,
239 QUNIFORM_TEXTURE_WIDTH
,
240 QUNIFORM_TEXTURE_HEIGHT
,
241 QUNIFORM_TEXTURE_DEPTH
,
242 QUNIFORM_TEXTURE_ARRAY_SIZE
,
243 QUNIFORM_TEXTURE_LEVELS
,
247 QUNIFORM_TEXRECT_SCALE_X
,
248 QUNIFORM_TEXRECT_SCALE_Y
,
250 /* Returns the base offset of the SSBO given by the data value. */
251 QUNIFORM_SSBO_OFFSET
,
253 /* Returns the size of the SSBO given by the data value. */
254 QUNIFORM_GET_BUFFER_SIZE
,
256 /* Sizes (in pixels) of a shader image given by the data value. */
257 QUNIFORM_IMAGE_WIDTH
,
258 QUNIFORM_IMAGE_HEIGHT
,
259 QUNIFORM_IMAGE_DEPTH
,
260 QUNIFORM_IMAGE_ARRAY_SIZE
,
264 /* Number of workgroups passed to glDispatchCompute in the dimension
265 * selected by the data value.
267 QUNIFORM_NUM_WORK_GROUPS
,
270 * Returns the the offset of the scratch buffer for register spilling.
272 QUNIFORM_SPILL_OFFSET
,
273 QUNIFORM_SPILL_SIZE_PER_THREAD
,
276 * Returns the offset of the shared memory for compute shaders.
278 * This will be accessed using TMU general memory operations, so the
279 * L2T cache will effectively be the shared memory area.
281 QUNIFORM_SHARED_OFFSET
,
284 static inline uint32_t v3d_unit_data_create(uint32_t unit
, uint32_t value
)
286 return unit
<< 24 | value
;
289 static inline uint32_t v3d_unit_data_get_unit(uint32_t data
)
294 static inline uint32_t v3d_unit_data_get_offset(uint32_t data
)
296 return data
& 0xffffff;
299 struct v3d_varying_slot
{
300 uint8_t slot_and_component
;
303 static inline struct v3d_varying_slot
304 v3d_slot_from_slot_and_component(uint8_t slot
, uint8_t component
)
306 assert(slot
< 255 / 4);
307 return (struct v3d_varying_slot
){ (slot
<< 2) + component
};
310 static inline uint8_t v3d_slot_get_slot(struct v3d_varying_slot slot
)
312 return slot
.slot_and_component
>> 2;
315 static inline uint8_t v3d_slot_get_component(struct v3d_varying_slot slot
)
317 return slot
.slot_and_component
& 3;
320 struct v3d_ubo_range
{
322 * offset in bytes from the start of the ubo where this range is
325 * Only set once used is set.
330 * offset in bytes from the start of the gallium uniforms where the
335 /** size in bytes of this ubo range */
344 uint8_t return_channels
;
348 } tex
[V3D_MAX_TEXTURE_SAMPLERS
];
358 bool point_coord_upper_left
;
361 bool sample_coverage
;
362 bool sample_alpha_to_coverage
;
363 bool sample_alpha_to_one
;
365 bool shade_model_flat
;
366 /* Mask of which color render targets are present. */
368 uint8_t swap_color_rb
;
369 /* Mask of which render targets need to be written as 32-bit floats */
370 uint8_t f32_color_rb
;
371 /* Masks of which render targets need to be written as ints/uints.
372 * Used by gallium to work around lost information in TGSI.
374 uint8_t int_color_rb
;
375 uint8_t uint_color_rb
;
376 uint8_t alpha_test_func
;
377 uint8_t logicop_func
;
378 uint32_t point_sprite_mask
;
380 struct pipe_rt_blend_state blend
;
386 struct v3d_varying_slot fs_inputs
[V3D_MAX_FS_INPUTS
];
387 uint8_t num_fs_inputs
;
390 bool per_vertex_point_size
;
394 /** A basic block of VIR intructions. */
396 struct list_head link
;
398 struct list_head instructions
;
400 struct set
*predecessors
;
401 struct qblock
*successors
[2];
405 /* Instruction IPs for the first and last instruction of the block.
406 * Set by qpu_schedule.c.
408 uint32_t start_qpu_ip
;
411 /* Instruction IP for the branch instruction of the block. Set by
414 uint32_t branch_qpu_ip
;
416 /** Offset within the uniform stream at the start of the block. */
417 uint32_t start_uniform
;
418 /** Offset within the uniform stream of the branch instruction */
419 uint32_t branch_uniform
;
421 /** @{ used by v3d_vir_live_variables.c */
426 BITSET_WORD
*live_in
;
427 BITSET_WORD
*live_out
;
428 int start_ip
, end_ip
;
432 /** Which util/list.h add mode we should use when inserting an instruction. */
433 enum vir_cursor_mode
{
439 * Tracking structure for where new instructions should be inserted. Create
440 * with one of the vir_after_inst()-style helper functions.
442 * This does not protect against removal of the block or instruction, so we
443 * have an assert in instruction removal to try to catch it.
446 enum vir_cursor_mode mode
;
447 struct list_head
*link
;
450 static inline struct vir_cursor
451 vir_before_inst(struct qinst
*inst
)
453 return (struct vir_cursor
){ vir_cursor_addtail
, &inst
->link
};
456 static inline struct vir_cursor
457 vir_after_inst(struct qinst
*inst
)
459 return (struct vir_cursor
){ vir_cursor_add
, &inst
->link
};
462 static inline struct vir_cursor
463 vir_before_block(struct qblock
*block
)
465 return (struct vir_cursor
){ vir_cursor_add
, &block
->instructions
};
468 static inline struct vir_cursor
469 vir_after_block(struct qblock
*block
)
471 return (struct vir_cursor
){ vir_cursor_addtail
, &block
->instructions
};
475 * Compiler state saved across compiler invocations, for any expensive global
478 struct v3d_compiler
{
479 const struct v3d_device_info
*devinfo
;
480 struct ra_regs
*regs
;
481 unsigned int reg_class_any
[3];
482 unsigned int reg_class_r5
[3];
483 unsigned int reg_class_phys
[3];
484 unsigned int reg_class_phys_or_acc
[3];
488 const struct v3d_device_info
*devinfo
;
490 nir_function_impl
*impl
;
491 struct exec_list
*cf_node_list
;
492 const struct v3d_compiler
*compiler
;
494 void (*debug_output
)(const char *msg
,
495 void *debug_output_data
);
496 void *debug_output_data
;
499 * Mapping from nir_register * or nir_ssa_def * to array of struct
500 * qreg for the values.
502 struct hash_table
*def_ht
;
504 /* For each temp, the instruction generating its value. */
506 uint32_t defs_array_size
;
509 * Inputs to the shader, arranged by TGSI declaration order.
511 * Not all fragment shader QFILE_VARY reads are present in this array.
514 struct qreg
*outputs
;
515 bool msaa_per_sample_output
;
516 struct qreg color_reads
[V3D_MAX_SAMPLES
];
517 struct qreg sample_colors
[V3D_MAX_SAMPLES
];
518 uint32_t inputs_array_size
;
519 uint32_t outputs_array_size
;
520 uint32_t uniforms_array_size
;
522 /* Booleans for whether the corresponding QFILE_VARY[i] is
523 * flat-shaded. This includes gl_FragColor flat-shading, which is
524 * customized based on the shademodel_flat shader key.
526 uint32_t flat_shade_flags
[BITSET_WORDS(V3D_MAX_FS_INPUTS
)];
528 uint32_t noperspective_flags
[BITSET_WORDS(V3D_MAX_FS_INPUTS
)];
530 uint32_t centroid_flags
[BITSET_WORDS(V3D_MAX_FS_INPUTS
)];
535 struct v3d_ubo_range
*ubo_ranges
;
536 bool *ubo_range_used
;
537 uint32_t ubo_ranges_array_size
;
538 /** Number of uniform areas tracked in ubo_ranges. */
539 uint32_t num_ubo_ranges
;
540 uint32_t next_ubo_dst_offset
;
542 /* State for whether we're executing on each channel currently. 0 if
543 * yes, otherwise a block number + 1 that the channel jumped to.
546 bool in_control_flow
;
548 struct qreg line_x
, point_x
, point_y
;
551 * Instance ID, which comes in before the vertex attribute payload if
552 * the shader record requests it.
557 * Vertex ID, which comes in before the vertex attribute payload
558 * (after Instance ID) if the shader record requests it.
562 /* Fragment shader payload regs. */
563 struct qreg payload_w
, payload_w_centroid
, payload_z
;
565 struct qreg cs_payload
[2];
566 struct qreg cs_shared_offset
;
567 int local_invocation_index_bits
;
569 uint8_t vattr_sizes
[V3D_MAX_VS_INPUTS
/ 4];
570 uint32_t vpm_output_size
;
572 /* Size in bytes of registers that have been spilled. This is how much
573 * space needs to be available in the spill BO per thread per QPU.
576 /* Shader-db stats */
577 uint32_t spills
, fills
, loops
;
579 * Register spilling's per-thread base address, shared between each
580 * spill/fill's addressing calculations.
582 struct qreg spill_base
;
583 /* Bit vector of which temps may be spilled */
584 BITSET_WORD
*spillable
;
587 * Array of the VARYING_SLOT_* of all FS QFILE_VARY reads.
589 * This includes those that aren't part of the VPM varyings, like
590 * point/line coordinates.
592 struct v3d_varying_slot input_slots
[V3D_MAX_FS_INPUTS
];
595 * An entry per outputs[] in the VS indicating what the VARYING_SLOT_*
596 * of the output is. Used to emit from the VS in the order that the
599 struct v3d_varying_slot
*output_slots
;
601 struct pipe_shader_state
*shader_state
;
603 struct v3d_fs_key
*fs_key
;
604 struct v3d_vs_key
*vs_key
;
606 /* Live ranges of temps. */
607 int *temp_start
, *temp_end
;
608 bool live_intervals_valid
;
610 uint32_t *uniform_data
;
611 enum quniform_contents
*uniform_contents
;
612 uint32_t uniform_array_size
;
613 uint32_t num_uniforms
;
614 uint32_t output_position_index
;
615 nir_variable
*output_color_var
[4];
616 uint32_t output_sample_mask_index
;
621 struct vir_cursor cursor
;
622 struct list_head blocks
;
623 int next_block_index
;
624 struct qblock
*cur_block
;
625 struct qblock
*loop_cont_block
;
626 struct qblock
*loop_break_block
;
629 uint32_t qpu_inst_count
;
630 uint32_t qpu_inst_size
;
632 /* For the FS, the number of varying inputs not counting the
633 * point/line varyings payload
638 * Number of inputs from num_inputs remaining to be queued to the read
641 uint32_t num_inputs_remaining
;
643 /* Number of inputs currently in the read FIFO for the VS/CS */
644 uint32_t num_inputs_in_fifo
;
646 /** Next offset in the VPM to read from in the VS/CS */
647 uint32_t vpm_read_offset
;
652 /* Set to compile program in in 1x, 2x, or 4x threaded mode, where
653 * SIG_THREAD_SWITCH is used to hide texturing latency at the cost of
654 * limiting ourselves to the part of the physical reg space.
656 * On V3D 3.x, 2x or 4x divide the physical reg space by 2x or 4x. On
657 * V3D 4.x, all shaders are 2x threaded, and 4x only divides the
658 * physical reg space in half.
661 struct qinst
*last_thrsw
;
662 bool last_thrsw_at_top_level
;
667 struct v3d_uniform_list
{
668 enum quniform_contents
*contents
;
673 struct v3d_prog_data
{
674 struct v3d_uniform_list uniforms
;
676 struct v3d_ubo_range
*ubo_ranges
;
677 uint32_t num_ubo_ranges
;
683 /* For threads > 1, whether the program should be dispatched in the
684 * after-final-THRSW state.
689 struct v3d_vs_prog_data
{
690 struct v3d_prog_data base
;
692 bool uses_iid
, uses_vid
;
694 /* Number of components read from each vertex attribute. */
695 uint8_t vattr_sizes
[V3D_MAX_VS_INPUTS
/ 4];
697 /* Total number of components read, for the shader state record. */
698 uint32_t vpm_input_size
;
700 /* Total number of components written, for the shader state record. */
701 uint32_t vpm_output_size
;
703 /* Set if there should be separate VPM segments for input and output.
704 * If unset, vpm_input_size will be 0.
706 bool separate_segments
;
708 /* Value to be programmed in VCM_CACHE_SIZE. */
709 uint8_t vcm_cache_size
;
712 struct v3d_fs_prog_data
{
713 struct v3d_prog_data base
;
715 struct v3d_varying_slot input_slots
[V3D_MAX_FS_INPUTS
];
717 /* Array of flat shade flags.
719 * Each entry is only 24 bits (high 8 bits 0), to match the hardware
722 uint32_t flat_shade_flags
[((V3D_MAX_FS_INPUTS
- 1) / 24) + 1];
724 uint32_t noperspective_flags
[((V3D_MAX_FS_INPUTS
- 1) / 24) + 1];
726 uint32_t centroid_flags
[((V3D_MAX_FS_INPUTS
- 1) / 24) + 1];
735 vir_has_uniform(struct qinst
*inst
)
737 return inst
->uniform
!= ~0;
740 /* Special nir_load_input intrinsic index for loading the current TLB
743 #define V3D_NIR_TLB_COLOR_READ_INPUT 2000000000
745 #define V3D_NIR_MS_MASK_OUTPUT 2000000000
747 extern const nir_shader_compiler_options v3d_nir_options
;
749 const struct v3d_compiler
*v3d_compiler_init(const struct v3d_device_info
*devinfo
);
750 void v3d_compiler_free(const struct v3d_compiler
*compiler
);
751 void v3d_optimize_nir(struct nir_shader
*s
);
753 uint64_t *v3d_compile(const struct v3d_compiler
*compiler
,
755 struct v3d_prog_data
**prog_data
,
757 void (*debug_output
)(const char *msg
,
758 void *debug_output_data
),
759 void *debug_output_data
,
760 int program_id
, int variant_id
,
761 uint32_t *final_assembly_size
);
763 void v3d_nir_to_vir(struct v3d_compile
*c
);
765 void vir_compile_destroy(struct v3d_compile
*c
);
766 const char *vir_get_stage_name(struct v3d_compile
*c
);
767 struct qblock
*vir_new_block(struct v3d_compile
*c
);
768 void vir_set_emit_block(struct v3d_compile
*c
, struct qblock
*block
);
769 void vir_link_blocks(struct qblock
*predecessor
, struct qblock
*successor
);
770 struct qblock
*vir_entry_block(struct v3d_compile
*c
);
771 struct qblock
*vir_exit_block(struct v3d_compile
*c
);
772 struct qinst
*vir_add_inst(enum v3d_qpu_add_op op
, struct qreg dst
,
773 struct qreg src0
, struct qreg src1
);
774 struct qinst
*vir_mul_inst(enum v3d_qpu_mul_op op
, struct qreg dst
,
775 struct qreg src0
, struct qreg src1
);
776 struct qinst
*vir_branch_inst(struct v3d_compile
*c
,
777 enum v3d_qpu_branch_cond cond
);
778 void vir_remove_instruction(struct v3d_compile
*c
, struct qinst
*qinst
);
779 uint32_t vir_get_uniform_index(struct v3d_compile
*c
,
780 enum quniform_contents contents
,
782 struct qreg
vir_uniform(struct v3d_compile
*c
,
783 enum quniform_contents contents
,
785 void vir_schedule_instructions(struct v3d_compile
*c
);
786 struct v3d_qpu_instr
v3d_qpu_nop(void);
788 struct qreg
vir_emit_def(struct v3d_compile
*c
, struct qinst
*inst
);
789 struct qinst
*vir_emit_nondef(struct v3d_compile
*c
, struct qinst
*inst
);
790 void vir_set_cond(struct qinst
*inst
, enum v3d_qpu_cond cond
);
791 void vir_set_pf(struct qinst
*inst
, enum v3d_qpu_pf pf
);
792 void vir_set_uf(struct qinst
*inst
, enum v3d_qpu_uf uf
);
793 void vir_set_unpack(struct qinst
*inst
, int src
,
794 enum v3d_qpu_input_unpack unpack
);
796 struct qreg
vir_get_temp(struct v3d_compile
*c
);
797 void vir_emit_last_thrsw(struct v3d_compile
*c
);
798 void vir_calculate_live_intervals(struct v3d_compile
*c
);
799 int vir_get_nsrc(struct qinst
*inst
);
800 bool vir_has_side_effects(struct v3d_compile
*c
, struct qinst
*inst
);
801 bool vir_get_add_op(struct qinst
*inst
, enum v3d_qpu_add_op
*op
);
802 bool vir_get_mul_op(struct qinst
*inst
, enum v3d_qpu_mul_op
*op
);
803 bool vir_is_raw_mov(struct qinst
*inst
);
804 bool vir_is_tex(struct qinst
*inst
);
805 bool vir_is_add(struct qinst
*inst
);
806 bool vir_is_mul(struct qinst
*inst
);
807 bool vir_writes_r3(const struct v3d_device_info
*devinfo
, struct qinst
*inst
);
808 bool vir_writes_r4(const struct v3d_device_info
*devinfo
, struct qinst
*inst
);
809 struct qreg
vir_follow_movs(struct v3d_compile
*c
, struct qreg reg
);
810 uint8_t vir_channels_written(struct qinst
*inst
);
811 struct qreg
ntq_get_src(struct v3d_compile
*c
, nir_src src
, int i
);
812 void ntq_store_dest(struct v3d_compile
*c
, nir_dest
*dest
, int chan
,
814 void vir_emit_thrsw(struct v3d_compile
*c
);
816 void vir_dump(struct v3d_compile
*c
);
817 void vir_dump_inst(struct v3d_compile
*c
, struct qinst
*inst
);
818 void vir_dump_uniform(enum quniform_contents contents
, uint32_t data
);
820 void vir_validate(struct v3d_compile
*c
);
822 void vir_optimize(struct v3d_compile
*c
);
823 bool vir_opt_algebraic(struct v3d_compile
*c
);
824 bool vir_opt_constant_folding(struct v3d_compile
*c
);
825 bool vir_opt_copy_propagate(struct v3d_compile
*c
);
826 bool vir_opt_dead_code(struct v3d_compile
*c
);
827 bool vir_opt_peephole_sf(struct v3d_compile
*c
);
828 bool vir_opt_small_immediates(struct v3d_compile
*c
);
829 bool vir_opt_vpm(struct v3d_compile
*c
);
830 void v3d_nir_lower_blend(nir_shader
*s
, struct v3d_compile
*c
);
831 void v3d_nir_lower_io(nir_shader
*s
, struct v3d_compile
*c
);
832 void v3d_nir_lower_txf_ms(nir_shader
*s
, struct v3d_compile
*c
);
833 void v3d_nir_lower_image_load_store(nir_shader
*s
);
834 void vir_lower_uniforms(struct v3d_compile
*c
);
836 void v3d33_vir_vpm_read_setup(struct v3d_compile
*c
, int num_components
);
837 void v3d33_vir_vpm_write_setup(struct v3d_compile
*c
);
838 void v3d33_vir_emit_tex(struct v3d_compile
*c
, nir_tex_instr
*instr
);
839 void v3d40_vir_emit_tex(struct v3d_compile
*c
, nir_tex_instr
*instr
);
840 void v3d40_vir_emit_image_load_store(struct v3d_compile
*c
,
841 nir_intrinsic_instr
*instr
);
843 void v3d_vir_to_qpu(struct v3d_compile
*c
, struct qpu_reg
*temp_registers
);
844 uint32_t v3d_qpu_schedule_instructions(struct v3d_compile
*c
);
845 void qpu_validate(struct v3d_compile
*c
);
846 struct qpu_reg
*v3d_register_allocate(struct v3d_compile
*c
, bool *spilled
);
847 bool vir_init_reg_sets(struct v3d_compiler
*compiler
);
849 bool v3d_gl_format_is_return_32(GLenum format
);
852 quniform_contents_is_texture_p0(enum quniform_contents contents
)
854 return (contents
>= QUNIFORM_TEXTURE_CONFIG_P0_0
&&
855 contents
< (QUNIFORM_TEXTURE_CONFIG_P0_0
+
856 V3D_MAX_TEXTURE_SAMPLERS
));
860 vir_in_nonuniform_control_flow(struct v3d_compile
*c
)
862 return c
->execute
.file
!= QFILE_NULL
;
865 static inline struct qreg
866 vir_uniform_ui(struct v3d_compile
*c
, uint32_t ui
)
868 return vir_uniform(c
, QUNIFORM_CONSTANT
, ui
);
871 static inline struct qreg
872 vir_uniform_f(struct v3d_compile
*c
, float f
)
874 return vir_uniform(c
, QUNIFORM_CONSTANT
, fui(f
));
877 #define VIR_ALU0(name, vir_inst, op) \
878 static inline struct qreg \
879 vir_##name(struct v3d_compile *c) \
881 return vir_emit_def(c, vir_inst(op, c->undef, \
882 c->undef, c->undef)); \
884 static inline struct qinst * \
885 vir_##name##_dest(struct v3d_compile *c, struct qreg dest) \
887 return vir_emit_nondef(c, vir_inst(op, dest, \
888 c->undef, c->undef)); \
891 #define VIR_ALU1(name, vir_inst, op) \
892 static inline struct qreg \
893 vir_##name(struct v3d_compile *c, struct qreg a) \
895 return vir_emit_def(c, vir_inst(op, c->undef, \
898 static inline struct qinst * \
899 vir_##name##_dest(struct v3d_compile *c, struct qreg dest, \
902 return vir_emit_nondef(c, vir_inst(op, dest, a, \
906 #define VIR_ALU2(name, vir_inst, op) \
907 static inline struct qreg \
908 vir_##name(struct v3d_compile *c, struct qreg a, struct qreg b) \
910 return vir_emit_def(c, vir_inst(op, c->undef, a, b)); \
912 static inline struct qinst * \
913 vir_##name##_dest(struct v3d_compile *c, struct qreg dest, \
914 struct qreg a, struct qreg b) \
916 return vir_emit_nondef(c, vir_inst(op, dest, a, b)); \
919 #define VIR_NODST_0(name, vir_inst, op) \
920 static inline struct qinst * \
921 vir_##name(struct v3d_compile *c) \
923 return vir_emit_nondef(c, vir_inst(op, c->undef, \
924 c->undef, c->undef)); \
927 #define VIR_NODST_1(name, vir_inst, op) \
928 static inline struct qinst * \
929 vir_##name(struct v3d_compile *c, struct qreg a) \
931 return vir_emit_nondef(c, vir_inst(op, c->undef, \
935 #define VIR_NODST_2(name, vir_inst, op) \
936 static inline struct qinst * \
937 vir_##name(struct v3d_compile *c, struct qreg a, struct qreg b) \
939 return vir_emit_nondef(c, vir_inst(op, c->undef, \
943 #define VIR_SFU(name) \
944 static inline struct qreg \
945 vir_##name(struct v3d_compile *c, struct qreg a) \
947 if (c->devinfo->ver >= 41) { \
948 return vir_emit_def(c, vir_add_inst(V3D_QPU_A_##name, \
952 vir_FMOV_dest(c, vir_reg(QFILE_MAGIC, V3D_QPU_WADDR_##name), a); \
953 return vir_FMOV(c, vir_reg(QFILE_MAGIC, V3D_QPU_WADDR_R4)); \
956 static inline struct qinst * \
957 vir_##name##_dest(struct v3d_compile *c, struct qreg dest, \
960 if (c->devinfo->ver >= 41) { \
961 return vir_emit_nondef(c, vir_add_inst(V3D_QPU_A_##name, \
965 vir_FMOV_dest(c, vir_reg(QFILE_MAGIC, V3D_QPU_WADDR_##name), a); \
966 return vir_FMOV_dest(c, dest, vir_reg(QFILE_MAGIC, V3D_QPU_WADDR_R4)); \
970 #define VIR_A_ALU2(name) VIR_ALU2(name, vir_add_inst, V3D_QPU_A_##name)
971 #define VIR_M_ALU2(name) VIR_ALU2(name, vir_mul_inst, V3D_QPU_M_##name)
972 #define VIR_A_ALU1(name) VIR_ALU1(name, vir_add_inst, V3D_QPU_A_##name)
973 #define VIR_M_ALU1(name) VIR_ALU1(name, vir_mul_inst, V3D_QPU_M_##name)
974 #define VIR_A_ALU0(name) VIR_ALU0(name, vir_add_inst, V3D_QPU_A_##name)
975 #define VIR_M_ALU0(name) VIR_ALU0(name, vir_mul_inst, V3D_QPU_M_##name)
976 #define VIR_A_NODST_2(name) VIR_NODST_2(name, vir_add_inst, V3D_QPU_A_##name)
977 #define VIR_M_NODST_2(name) VIR_NODST_2(name, vir_mul_inst, V3D_QPU_M_##name)
978 #define VIR_A_NODST_1(name) VIR_NODST_1(name, vir_add_inst, V3D_QPU_A_##name)
979 #define VIR_M_NODST_1(name) VIR_NODST_1(name, vir_mul_inst, V3D_QPU_M_##name)
980 #define VIR_A_NODST_0(name) VIR_NODST_0(name, vir_add_inst, V3D_QPU_A_##name)
1003 VIR_A_NODST_2(STVPMV
)
1013 VIR_A_ALU1(LDVPMV_IN
)
1014 VIR_A_ALU1(LDVPMV_OUT
)
1023 VIR_A_ALU0(BARRIERID
)
1024 VIR_A_NODST_1(VPMSETUP
)
1025 VIR_A_NODST_0(VPMWT
)
1048 VIR_M_NODST_2(MULTOP
)
1060 static inline struct qinst
*
1061 vir_MOV_cond(struct v3d_compile
*c
, enum v3d_qpu_cond cond
,
1062 struct qreg dest
, struct qreg src
)
1064 struct qinst
*mov
= vir_MOV_dest(c
, dest
, src
);
1065 vir_set_cond(mov
, cond
);
1069 static inline struct qreg
1070 vir_SEL(struct v3d_compile
*c
, enum v3d_qpu_cond cond
,
1071 struct qreg src0
, struct qreg src1
)
1073 struct qreg t
= vir_get_temp(c
);
1074 vir_MOV_dest(c
, t
, src1
);
1075 vir_MOV_cond(c
, cond
, t
, src0
);
1079 static inline struct qinst
*
1080 vir_NOP(struct v3d_compile
*c
)
1082 return vir_emit_nondef(c
, vir_add_inst(V3D_QPU_A_NOP
,
1083 c
->undef
, c
->undef
, c
->undef
));
1086 static inline struct qreg
1087 vir_LDTMU(struct v3d_compile
*c
)
1089 if (c
->devinfo
->ver
>= 41) {
1090 struct qinst
*ldtmu
= vir_add_inst(V3D_QPU_A_NOP
, c
->undef
,
1091 c
->undef
, c
->undef
);
1092 ldtmu
->qpu
.sig
.ldtmu
= true;
1094 return vir_emit_def(c
, ldtmu
);
1096 vir_NOP(c
)->qpu
.sig
.ldtmu
= true;
1097 return vir_MOV(c
, vir_reg(QFILE_MAGIC
, V3D_QPU_WADDR_R4
));
1101 static inline struct qreg
1102 vir_UMUL(struct v3d_compile
*c
, struct qreg src0
, struct qreg src1
)
1104 vir_MULTOP(c
, src0
, src1
);
1105 return vir_UMUL24(c
, src0
, src1
);
1109 static inline struct qreg
1110 vir_LOAD_IMM(struct v3d_compile *c, uint32_t val)
1112 return vir_emit_def(c, vir_inst(QOP_LOAD_IMM, c->undef,
1113 vir_reg(QFILE_LOAD_IMM, val), c->undef));
1116 static inline struct qreg
1117 vir_LOAD_IMM_U2(struct v3d_compile *c, uint32_t val)
1119 return vir_emit_def(c, vir_inst(QOP_LOAD_IMM_U2, c->undef,
1120 vir_reg(QFILE_LOAD_IMM, val),
1123 static inline struct qreg
1124 vir_LOAD_IMM_I2(struct v3d_compile *c, uint32_t val)
1126 return vir_emit_def(c, vir_inst(QOP_LOAD_IMM_I2, c->undef,
1127 vir_reg(QFILE_LOAD_IMM, val),
1132 static inline struct qinst
*
1133 vir_BRANCH(struct v3d_compile
*c
, enum v3d_qpu_branch_cond cond
)
1135 /* The actual uniform_data value will be set at scheduling time */
1136 return vir_emit_nondef(c
, vir_branch_inst(c
, cond
));
1139 #define vir_for_each_block(block, c) \
1140 list_for_each_entry(struct qblock, block, &c->blocks, link)
1142 #define vir_for_each_block_rev(block, c) \
1143 list_for_each_entry_rev(struct qblock, block, &c->blocks, link)
1145 /* Loop over the non-NULL members of the successors array. */
1146 #define vir_for_each_successor(succ, block) \
1147 for (struct qblock *succ = block->successors[0]; \
1149 succ = (succ == block->successors[1] ? NULL : \
1150 block->successors[1]))
1152 #define vir_for_each_inst(inst, block) \
1153 list_for_each_entry(struct qinst, inst, &block->instructions, link)
1155 #define vir_for_each_inst_rev(inst, block) \
1156 list_for_each_entry_rev(struct qinst, inst, &block->instructions, link)
1158 #define vir_for_each_inst_safe(inst, block) \
1159 list_for_each_entry_safe(struct qinst, inst, &block->instructions, link)
1161 #define vir_for_each_inst_inorder(inst, c) \
1162 vir_for_each_block(_block, c) \
1163 vir_for_each_inst(inst, _block)
1165 #define vir_for_each_inst_inorder_safe(inst, c) \
1166 vir_for_each_block(_block, c) \
1167 vir_for_each_inst_safe(inst, _block)
1169 #endif /* V3D_COMPILER_H */