2 * Copyright © 2014 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
34 #include "util/macros.h"
35 #include "compiler/nir/nir.h"
36 #include "util/list.h"
37 #include "util/u_math.h"
39 #include "vc4_screen.h"
40 #include "vc4_qpu_defines.h"
41 #include "kernel/vc4_packet.h"
42 #include "pipe/p_state.h"
54 * Stores an immediate value in the index field that can be turned
55 * into a small immediate field by qpu_encode_small_immediate().
104 QOP_TLB_STENCIL_SETUP
,
107 QOP_TLB_COLOR_WRITE_MS
,
118 /** Texture x coordinate parameter write */
120 /** Texture y coordinate parameter write */
122 /** Texture border color parameter or cube map z coordinate write */
124 /** Texture LOD bias parameter write */
128 * Texture-unit 4-byte read with address provided direct in S
131 * The first operand is the offset from the start of the UBO, and the
132 * second is the uniform that has the UBO's base pointer.
137 * Signal of texture read being necessary and then reading r4 into
143 struct queued_qpu_inst
{
144 struct list_head link
;
149 struct list_head link
;
160 * Coordinate shader, runs during binning, before the VS, and just
168 enum quniform_contents
{
170 * Indicates that a constant 32-bit value is copied from the program's
175 * Indicates that the program's uniform contents are used as an index
176 * into the GL uniform storage.
181 * Scaling factors from clip coordinates to relative to the viewport
184 * This is used by the coordinate and vertex shaders to produce the
185 * 32-bit entry consisting of 2 16-bit fields with 12.4 signed fixed
186 * point offsets from the viewport ccenter.
188 QUNIFORM_VIEWPORT_X_SCALE
,
189 QUNIFORM_VIEWPORT_Y_SCALE
,
192 QUNIFORM_VIEWPORT_Z_OFFSET
,
193 QUNIFORM_VIEWPORT_Z_SCALE
,
195 QUNIFORM_USER_CLIP_PLANE
,
198 * A reference to a texture config parameter 0 uniform.
200 * This is a uniform implicitly loaded with a QPU_W_TMU* write, which
201 * defines texture type, miplevels, and such. It will be found as a
202 * parameter to the first QOP_TEX_[STRB] instruction in a sequence.
204 QUNIFORM_TEXTURE_CONFIG_P0
,
207 * A reference to a texture config parameter 1 uniform.
209 * This is a uniform implicitly loaded with a QPU_W_TMU* write, which
210 * defines texture width, height, filters, and wrap modes. It will be
211 * found as a parameter to the second QOP_TEX_[STRB] instruction in a
214 QUNIFORM_TEXTURE_CONFIG_P1
,
216 /** A reference to a texture config parameter 2 cubemap stride uniform */
217 QUNIFORM_TEXTURE_CONFIG_P2
,
219 QUNIFORM_TEXTURE_MSAA_ADDR
,
223 QUNIFORM_TEXRECT_SCALE_X
,
224 QUNIFORM_TEXRECT_SCALE_Y
,
226 QUNIFORM_TEXTURE_BORDER_COLOR
,
228 QUNIFORM_BLEND_CONST_COLOR_X
,
229 QUNIFORM_BLEND_CONST_COLOR_Y
,
230 QUNIFORM_BLEND_CONST_COLOR_Z
,
231 QUNIFORM_BLEND_CONST_COLOR_W
,
232 QUNIFORM_BLEND_CONST_COLOR_RGBA
,
233 QUNIFORM_BLEND_CONST_COLOR_AAAA
,
238 QUNIFORM_SAMPLE_MASK
,
241 struct vc4_varying_slot
{
246 struct vc4_compiler_ubo_range
{
248 * offset in bytes from the start of the ubo where this range is
251 * Only set once used is set.
256 * offset in bytes from the start of the gallium uniforms where the
261 /** size in bytes of this ubo range */
265 * Set if this range is used by the shader for indirect uniforms
272 struct vc4_uncompiled_shader
*shader_state
;
274 enum pipe_format format
;
278 unsigned compare_mode
:1;
279 unsigned compare_func
:3;
284 uint16_t msaa_width
, msaa_height
;
287 } tex
[VC4_MAX_TEXTURE_SAMPLERS
];
293 enum pipe_format color_format
;
295 bool stencil_enabled
;
296 bool stencil_twoside
;
297 bool stencil_full_writemasks
;
301 bool point_coord_upper_left
;
304 bool sample_coverage
;
305 bool sample_alpha_to_coverage
;
306 bool sample_alpha_to_one
;
307 uint8_t alpha_test_func
;
308 uint8_t logicop_func
;
309 uint32_t point_sprite_mask
;
311 struct pipe_rt_blend_state blend
;
318 * This is a proxy for the array of FS input semantics, which is
319 * larger than we would want to put in the key.
321 uint64_t compiled_fs_id
;
323 enum pipe_format attr_formats
[8];
325 bool per_vertex_point_size
;
329 struct vc4_context
*vc4
;
331 nir_function_impl
*impl
;
332 struct exec_list
*cf_node_list
;
335 * Mapping from nir_register * or nir_ssa_def * to array of struct
336 * qreg for the values.
338 struct hash_table
*def_ht
;
340 /* For each temp, the instruction generating its value. */
342 uint32_t defs_array_size
;
345 * Inputs to the shader, arranged by TGSI declaration order.
347 * Not all fragment shader QFILE_VARY reads are present in this array.
350 struct qreg
*outputs
;
351 bool msaa_per_sample_output
;
352 struct qreg color_reads
[VC4_MAX_SAMPLES
];
353 struct qreg sample_colors
[VC4_MAX_SAMPLES
];
354 uint32_t inputs_array_size
;
355 uint32_t outputs_array_size
;
356 uint32_t uniforms_array_size
;
358 struct vc4_compiler_ubo_range
*ubo_ranges
;
359 uint32_t ubo_ranges_array_size
;
360 /** Number of uniform areas declared in ubo_ranges. */
361 uint32_t num_uniform_ranges
;
362 /** Number of uniform areas used for indirect addressed loads. */
363 uint32_t num_ubo_ranges
;
364 uint32_t next_ubo_dst_offset
;
366 struct qreg line_x
, point_x
, point_y
;
369 uint8_t vattr_sizes
[8];
371 /* Bitfield for whether a given channel of a sampler needs sRGB
374 uint8_t tex_srgb_decode
[VC4_MAX_TEXTURE_SAMPLERS
];
377 * Array of the VARYING_SLOT_* of all FS QFILE_VARY reads.
379 * This includes those that aren't part of the VPM varyings, like
380 * point/line coordinates.
382 struct vc4_varying_slot
*input_slots
;
383 uint32_t num_input_slots
;
384 uint32_t input_slots_array_size
;
387 * An entry per outputs[] in the VS indicating what the VARYING_SLOT_*
388 * of the output is. Used to emit from the VS in the order that the
391 struct vc4_varying_slot
*output_slots
;
393 struct pipe_shader_state
*shader_state
;
395 struct vc4_fs_key
*fs_key
;
396 struct vc4_vs_key
*vs_key
;
398 uint32_t *uniform_data
;
399 enum quniform_contents
*uniform_contents
;
400 uint32_t uniform_array_size
;
401 uint32_t num_uniforms
;
402 uint32_t num_outputs
;
403 uint32_t num_texture_samples
;
404 uint32_t output_position_index
;
405 uint32_t output_color_index
;
406 uint32_t output_point_size_index
;
407 uint32_t output_sample_mask_index
;
412 struct list_head instructions
;
413 uint32_t immediates
[1024];
415 struct list_head qpu_inst_list
;
417 uint32_t qpu_inst_count
;
418 uint32_t qpu_inst_size
;
425 /* Special nir_load_input intrinsic index for loading the current TLB
428 #define VC4_NIR_TLB_COLOR_READ_INPUT 2000000000
430 #define VC4_NIR_MS_MASK_OUTPUT 2000000000
432 /* Special offset for nir_load_uniform values to get a QUNIFORM_*
433 * state-dependent value.
435 #define VC4_NIR_STATE_UNIFORM_OFFSET 1000000000
437 struct vc4_compile
*qir_compile_init(void);
438 void qir_compile_destroy(struct vc4_compile
*c
);
439 struct qinst
*qir_inst(enum qop op
, struct qreg dst
,
440 struct qreg src0
, struct qreg src1
);
441 struct qinst
*qir_inst4(enum qop op
, struct qreg dst
,
446 void qir_remove_instruction(struct vc4_compile
*c
, struct qinst
*qinst
);
447 struct qreg
qir_uniform(struct vc4_compile
*c
,
448 enum quniform_contents contents
,
450 void qir_schedule_instructions(struct vc4_compile
*c
);
451 void qir_reorder_uniforms(struct vc4_compile
*c
);
453 void qir_emit(struct vc4_compile
*c
, struct qinst
*inst
);
454 static inline struct qinst
*
455 qir_emit_nodef(struct vc4_compile
*c
, struct qinst
*inst
)
457 list_addtail(&inst
->link
, &c
->instructions
);
461 struct qreg
qir_get_temp(struct vc4_compile
*c
);
462 int qir_get_op_nsrc(enum qop qop
);
463 bool qir_reg_equals(struct qreg a
, struct qreg b
);
464 bool qir_has_side_effects(struct vc4_compile
*c
, struct qinst
*inst
);
465 bool qir_has_side_effect_reads(struct vc4_compile
*c
, struct qinst
*inst
);
466 bool qir_is_multi_instruction(struct qinst
*inst
);
467 bool qir_is_mul(struct qinst
*inst
);
468 bool qir_is_raw_mov(struct qinst
*inst
);
469 bool qir_is_tex(struct qinst
*inst
);
470 bool qir_is_float_input(struct qinst
*inst
);
471 bool qir_depends_on_flags(struct qinst
*inst
);
472 bool qir_writes_r4(struct qinst
*inst
);
473 struct qreg
qir_follow_movs(struct vc4_compile
*c
, struct qreg reg
);
475 void qir_dump(struct vc4_compile
*c
);
476 void qir_dump_inst(struct vc4_compile
*c
, struct qinst
*inst
);
477 const char *qir_get_stage_name(enum qstage stage
);
479 void qir_optimize(struct vc4_compile
*c
);
480 bool qir_opt_algebraic(struct vc4_compile
*c
);
481 bool qir_opt_constant_folding(struct vc4_compile
*c
);
482 bool qir_opt_copy_propagation(struct vc4_compile
*c
);
483 bool qir_opt_cse(struct vc4_compile
*c
);
484 bool qir_opt_dead_code(struct vc4_compile
*c
);
485 bool qir_opt_small_immediates(struct vc4_compile
*c
);
486 bool qir_opt_vpm(struct vc4_compile
*c
);
487 void vc4_nir_lower_blend(struct vc4_compile
*c
);
488 void vc4_nir_lower_io(struct vc4_compile
*c
);
489 nir_ssa_def
*vc4_nir_get_state_uniform(struct nir_builder
*b
,
490 enum quniform_contents contents
);
491 nir_ssa_def
*vc4_nir_get_swizzled_channel(struct nir_builder
*b
,
492 nir_ssa_def
**srcs
, int swiz
);
493 void vc4_nir_lower_txf_ms(struct vc4_compile
*c
);
494 void qir_lower_uniforms(struct vc4_compile
*c
);
496 uint32_t qpu_schedule_instructions(struct vc4_compile
*c
);
498 void qir_SF(struct vc4_compile
*c
, struct qreg src
);
500 static inline struct qreg
501 qir_uniform_ui(struct vc4_compile
*c
, uint32_t ui
)
503 return qir_uniform(c
, QUNIFORM_CONSTANT
, ui
);
506 static inline struct qreg
507 qir_uniform_f(struct vc4_compile
*c
, float f
)
509 return qir_uniform(c
, QUNIFORM_CONSTANT
, fui(f
));
512 #define QIR_ALU0(name) \
513 static inline struct qreg \
514 qir_##name(struct vc4_compile *c) \
516 struct qreg t = qir_get_temp(c); \
517 qir_emit(c, qir_inst(QOP_##name, t, c->undef, c->undef)); \
521 #define QIR_ALU1(name) \
522 static inline struct qreg \
523 qir_##name(struct vc4_compile *c, struct qreg a) \
525 struct qreg t = qir_get_temp(c); \
526 qir_emit(c, qir_inst(QOP_##name, t, a, c->undef)); \
529 static inline struct qinst * \
530 qir_##name##_dest(struct vc4_compile *c, struct qreg dest, \
533 return qir_emit_nodef(c, qir_inst(QOP_##name, dest, a, \
537 #define QIR_ALU2(name) \
538 static inline struct qreg \
539 qir_##name(struct vc4_compile *c, struct qreg a, struct qreg b) \
541 struct qreg t = qir_get_temp(c); \
542 qir_emit(c, qir_inst(QOP_##name, t, a, b)); \
546 qir_##name##_dest(struct vc4_compile *c, struct qreg dest, \
547 struct qreg a, struct qreg b) \
549 qir_emit_nodef(c, qir_inst(QOP_##name, dest, a, b)); \
552 #define QIR_NODST_1(name) \
553 static inline struct qinst * \
554 qir_##name(struct vc4_compile *c, struct qreg a) \
556 struct qinst *inst = qir_inst(QOP_##name, c->undef, \
562 #define QIR_NODST_2(name) \
563 static inline struct qinst * \
564 qir_##name(struct vc4_compile *c, struct qreg a, struct qreg b) \
566 struct qinst *inst = qir_inst(QOP_##name, c->undef, \
572 #define QIR_PACK(name) \
573 static inline struct qreg \
574 qir_##name(struct vc4_compile *c, struct qreg dest, struct qreg a) \
576 qir_emit_nodef(c, qir_inst(QOP_##name, dest, a, c->undef)); \
620 QIR_NODST_2(TEX_DIRECT
)
625 QIR_ALU0(FRAG_REV_FLAG
)
627 QIR_ALU0(TLB_COLOR_READ
)
628 QIR_NODST_1(TLB_COLOR_WRITE
)
629 QIR_NODST_1(TLB_COLOR_WRITE_MS
)
630 QIR_NODST_1(TLB_Z_WRITE
)
631 QIR_NODST_1(TLB_STENCIL_SETUP
)
634 static inline struct qreg
635 qir_SEL(struct vc4_compile
*c
, uint8_t cond
, struct qreg src0
, struct qreg src1
)
637 struct qreg t
= qir_get_temp(c
);
638 struct qinst
*a
= qir_MOV_dest(c
, t
, src0
);
639 struct qinst
*b
= qir_MOV_dest(c
, t
, src1
);
645 static inline struct qreg
646 qir_UNPACK_8_F(struct vc4_compile
*c
, struct qreg src
, int i
)
648 struct qreg t
= qir_FMOV(c
, src
);
649 c
->defs
[t
.index
]->src
[0].pack
= QPU_UNPACK_8A
+ i
;
653 static inline struct qreg
654 qir_UNPACK_8_I(struct vc4_compile
*c
, struct qreg src
, int i
)
656 struct qreg t
= qir_MOV(c
, src
);
657 c
->defs
[t
.index
]->src
[0].pack
= QPU_UNPACK_8A
+ i
;
661 static inline struct qreg
662 qir_UNPACK_16_F(struct vc4_compile
*c
, struct qreg src
, int i
)
664 struct qreg t
= qir_FMOV(c
, src
);
665 c
->defs
[t
.index
]->src
[0].pack
= QPU_UNPACK_16A
+ i
;
669 static inline struct qreg
670 qir_UNPACK_16_I(struct vc4_compile
*c
, struct qreg src
, int i
)
672 struct qreg t
= qir_MOV(c
, src
);
673 c
->defs
[t
.index
]->src
[0].pack
= QPU_UNPACK_16A
+ i
;
678 qir_PACK_8_F(struct vc4_compile
*c
, struct qreg dest
, struct qreg val
, int chan
)
681 dest
.pack
= QPU_PACK_MUL_8A
+ chan
;
682 qir_emit(c
, qir_inst(QOP_MMOV
, dest
, val
, c
->undef
));
683 if (dest
.file
== QFILE_TEMP
)
684 c
->defs
[dest
.index
] = NULL
;
687 static inline struct qreg
688 qir_PACK_8888_F(struct vc4_compile
*c
, struct qreg val
)
690 struct qreg dest
= qir_MMOV(c
, val
);
691 c
->defs
[dest
.index
]->dst
.pack
= QPU_PACK_MUL_8888
;
695 static inline struct qreg
696 qir_POW(struct vc4_compile
*c
, struct qreg x
, struct qreg y
)
698 return qir_EXP2(c
, qir_FMUL(c
,
704 qir_VPM_WRITE(struct vc4_compile
*c
, struct qreg val
)
706 static const struct qreg vpm
= { QFILE_VPM
, 0 };
707 qir_emit(c
, qir_inst(QOP_MOV
, vpm
, val
, c
->undef
));
710 #endif /* VC4_QIR_H */