5 * Copyright (c) 2013 Connor Abbott (connor@abbott.cx)
6 * Copyright (c) 2018 Alyssa Rosenzweig (alyssa@rosenzweig.io)
8 * Permission is hereby granted, free of charge, to any person obtaining a copy
9 * of this software and associated documentation files (the "Software"), to deal
10 * in the Software without restriction, including without limitation the rights
11 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12 * copies of the Software, and to permit persons to whom the Software is
13 * furnished to do so, subject to the following conditions:
15 * The above copyright notice and this permission notice shall be included in
16 * all copies or substantial portions of the Software.
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
21 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
33 #define MIDGARD_DBG_MSGS 0x0001
34 #define MIDGARD_DBG_SHADERS 0x0002
36 extern int midgard_debug
;
39 midgard_word_type_alu
,
40 midgard_word_type_load_store
,
41 midgard_word_type_texture
,
42 midgard_word_type_unknown
58 midgard_alu_op_fadd
= 0x10,
59 midgard_alu_op_fmul
= 0x14,
60 midgard_alu_op_fmin
= 0x28,
61 midgard_alu_op_fmax
= 0x2C,
62 midgard_alu_op_fmov
= 0x30,
63 midgard_alu_op_froundeven
= 0x34,
64 midgard_alu_op_ftrunc
= 0x35,
65 midgard_alu_op_ffloor
= 0x36,
66 midgard_alu_op_fceil
= 0x37,
67 midgard_alu_op_ffma
= 0x38,
68 midgard_alu_op_fdot3
= 0x3C,
69 midgard_alu_op_fdot3r
= 0x3D,
70 midgard_alu_op_fdot4
= 0x3E,
71 midgard_alu_op_freduce
= 0x3F,
72 midgard_alu_op_iadd
= 0x40,
73 midgard_alu_op_ishladd
= 0x41,
74 midgard_alu_op_isub
= 0x46,
75 midgard_alu_op_imul
= 0x58,
76 midgard_alu_op_imin
= 0x60,
77 midgard_alu_op_umin
= 0x61,
78 midgard_alu_op_imax
= 0x62,
79 midgard_alu_op_umax
= 0x63,
80 midgard_alu_op_iasr
= 0x68,
81 midgard_alu_op_ilsr
= 0x69,
82 midgard_alu_op_ishl
= 0x6E,
83 midgard_alu_op_iand
= 0x70,
84 midgard_alu_op_ior
= 0x71,
85 midgard_alu_op_inot
= 0x72,
86 midgard_alu_op_iandnot
= 0x74, /* (a, b) -> a & ~b, used for not/b2f */
87 midgard_alu_op_ixor
= 0x76,
88 midgard_alu_op_ilzcnt
= 0x78, /* Number of zeroes on left. 31 - ilzcnt(x) = findMSB(x) */
89 midgard_alu_op_ibitcount8
= 0x7A, /* Counts bits in 8-bit increments */
90 midgard_alu_op_imov
= 0x7B,
91 midgard_alu_op_iabs
= 0x7C,
92 midgard_alu_op_feq
= 0x80,
93 midgard_alu_op_fne
= 0x81,
94 midgard_alu_op_flt
= 0x82,
95 midgard_alu_op_fle
= 0x83,
96 midgard_alu_op_fball_eq
= 0x88,
97 midgard_alu_op_bball_eq
= 0x89,
98 midgard_alu_op_fball_lt
= 0x8A, /* all(lessThan(.., ..)) */
99 midgard_alu_op_fball_lte
= 0x8B, /* all(lessThanEqual(.., ..)) */
100 midgard_alu_op_bbany_neq
= 0x90, /* used for bvec4(1) */
101 midgard_alu_op_fbany_neq
= 0x91, /* bvec4(0) also */
102 midgard_alu_op_fbany_lt
= 0x92, /* any(lessThan(.., ..)) */
103 midgard_alu_op_fbany_lte
= 0x93, /* any(lessThanEqual(.., ..)) */
104 midgard_alu_op_f2i
= 0x99,
105 midgard_alu_op_f2u8
= 0x9C,
106 midgard_alu_op_f2u
= 0x9D,
108 midgard_alu_op_ieq
= 0xA0,
109 midgard_alu_op_ine
= 0xA1,
110 midgard_alu_op_ult
= 0xA2,
111 midgard_alu_op_ule
= 0xA3,
112 midgard_alu_op_ilt
= 0xA4,
113 midgard_alu_op_ile
= 0xA5,
114 midgard_alu_op_iball_eq
= 0xA8,
115 midgard_alu_op_ball
= 0xA9,
116 midgard_alu_op_uball_lt
= 0xAA,
117 midgard_alu_op_uball_lte
= 0xAB,
118 midgard_alu_op_iball_lt
= 0xAC,
119 midgard_alu_op_iball_lte
= 0xAD,
120 midgard_alu_op_ibany_eq
= 0xB0,
121 midgard_alu_op_ibany_neq
= 0xB1,
122 midgard_alu_op_ubany_lt
= 0xB2,
123 midgard_alu_op_ubany_lte
= 0xB3,
124 midgard_alu_op_ibany_lt
= 0xB4, /* any(lessThan(.., ..)) */
125 midgard_alu_op_ibany_lte
= 0xB5, /* any(lessThanEqual(.., ..)) */
126 midgard_alu_op_i2f
= 0xB8,
127 midgard_alu_op_u2f
= 0xBC,
128 midgard_alu_op_icsel
= 0xC1,
129 midgard_alu_op_fcsel_i
= 0xC4,
130 midgard_alu_op_fcsel
= 0xC5,
131 midgard_alu_op_fround
= 0xC6,
132 midgard_alu_op_fatan_pt2
= 0xE8,
133 midgard_alu_op_fpow_pt1
= 0xEC,
134 midgard_alu_op_frcp
= 0xF0,
135 midgard_alu_op_frsqrt
= 0xF2,
136 midgard_alu_op_fsqrt
= 0xF3,
137 midgard_alu_op_fexp2
= 0xF4,
138 midgard_alu_op_flog2
= 0xF5,
139 midgard_alu_op_fsin
= 0xF6,
140 midgard_alu_op_fcos
= 0xF7,
141 midgard_alu_op_fatan2_pt1
= 0xF9,
145 midgard_outmod_none
= 0,
146 midgard_outmod_pos
= 1,
147 midgard_outmod_int
= 2,
148 midgard_outmod_sat
= 3
152 midgard_reg_mode_quarter
= 0,
153 midgard_reg_mode_half
= 1,
154 midgard_reg_mode_full
= 2,
155 midgard_reg_mode_double
= 3 /* TODO: verify */
159 midgard_dest_override_lower
= 0,
160 midgard_dest_override_upper
= 1,
161 midgard_dest_override_none
= 2
162 } midgard_dest_override
;
165 midgard_int_sign_extend
= 0,
166 midgard_int_zero_extend
= 1,
167 midgard_int_normal
= 2,
168 midgard_int_reserved
= 3
171 #define MIDGARD_FLOAT_MOD_ABS (1 << 0)
172 #define MIDGARD_FLOAT_MOD_NEG (1 << 1)
175 __attribute__((__packed__
))
177 /* Either midgard_int_mod or from midgard_float_mod_*, depending on the
181 /* replicate lower half if dest = half, or low/high half selection if
185 bool rep_high
: 1; /* unused if dest = full */
186 bool half
: 1; /* only matters if dest = full */
187 unsigned swizzle
: 8;
189 midgard_vector_alu_src
;
192 __attribute__((__packed__
))
194 midgard_alu_op op
: 8;
195 midgard_reg_mode reg_mode
: 2;
198 midgard_dest_override dest_override
: 2;
199 midgard_outmod outmod
: 2;
205 __attribute__((__packed__
))
209 bool full
: 1; /* 0 = half, 1 = full */
210 unsigned component
: 3;
212 midgard_scalar_alu_src
;
215 __attribute__((__packed__
))
217 midgard_alu_op op
: 8;
220 unsigned unknown
: 1;
221 midgard_outmod outmod
: 2;
222 bool output_full
: 1;
223 unsigned output_component
: 3;
228 __attribute__((__packed__
))
230 unsigned src1_reg
: 5;
231 unsigned src2_reg
: 5;
232 unsigned out_reg
: 5;
237 /* In addition to conditional branches and jumps (unconditional branches),
238 * Midgard implements a bit of fixed function functionality used in fragment
239 * shaders via specially crafted branches. These have special branch opcodes,
240 * which perform a fixed-function operation and/or use the results of a
241 * fixed-function operation as the branch condition. */
244 /* Regular branches */
245 midgard_jmp_writeout_op_branch_uncond
= 1,
246 midgard_jmp_writeout_op_branch_cond
= 2,
248 /* In a fragment shader, execute a discard_if instruction, with the
249 * corresponding condition code. Terminates the shader, so generally
250 * set the branch target to out of the shader */
251 midgard_jmp_writeout_op_discard
= 4,
253 /* Branch if the tilebuffer is not yet ready. At the beginning of a
254 * fragment shader that reads from the tile buffer, for instance via
255 * ARM_shader_framebuffer_fetch or EXT_pixel_local_storage, this branch
256 * operation should be used as a loop. An instruction like
257 * "br.tilebuffer.always -1" does the trick, corresponding to
258 * "while(!is_tilebuffer_ready) */
259 midgard_jmp_writeout_op_tilebuffer_pending
= 6,
261 /* In a fragment shader, try to write out the value pushed to r0 to the
262 * tilebuffer, subject to unknown state in r1.z and r1.w. If this
263 * succeeds, the shader terminates. If it fails, it branches to the
264 * specified branch target. Generally, this should be used in a loop to
265 * itself, acting as "do { write(r0); } while(!write_successful);" */
266 midgard_jmp_writeout_op_writeout
= 7,
267 } midgard_jmp_writeout_op
;
270 midgard_condition_write0
= 0,
272 /* These condition codes denote a conditional branch on FALSE and on
273 * TRUE respectively */
274 midgard_condition_false
= 1,
275 midgard_condition_true
= 2,
277 /* This condition code always branches. For a pure branch, the
278 * unconditional branch coding should be used instead, but for
279 * fixed-function branch opcodes, this is still useful */
280 midgard_condition_always
= 3,
284 __attribute__((__packed__
))
286 midgard_jmp_writeout_op op
: 3; /* == branch_uncond */
287 unsigned dest_tag
: 4; /* tag of branch destination */
288 unsigned unknown
: 2;
291 midgard_branch_uncond
;
294 __attribute__((__packed__
))
296 midgard_jmp_writeout_op op
: 3; /* == branch_cond */
297 unsigned dest_tag
: 4; /* tag of branch destination */
299 midgard_condition cond
: 2;
304 __attribute__((__packed__
))
306 midgard_jmp_writeout_op op
: 3; /* == branch_cond */
307 unsigned dest_tag
: 4; /* tag of branch destination */
308 unsigned unknown
: 2;
312 midgard_branch_extended
;
315 __attribute__((__packed__
))
317 midgard_jmp_writeout_op op
: 3; /* == writeout */
318 unsigned unknown
: 13;
327 midgard_op_ld_st_noop
= 0x03,
329 /* Unclear why this is on the L/S unit, but (with an address of 0,
330 * appropriate swizzle, magic constant 0x24, and xy mask?) moves fp32 cube
331 * map coordinates in r27 to its cube map texture coordinate
332 * destination (e.g r29). 0x4 magic for loading from fp16 instead */
334 midgard_op_store_cubemap_coords
= 0x0E,
336 midgard_op_load_attr_16
= 0x95,
337 midgard_op_load_attr_32
= 0x94,
338 midgard_op_load_vary_16
= 0x99,
339 midgard_op_load_vary_32
= 0x98,
340 midgard_op_load_color_buffer_16
= 0x9D,
341 midgard_op_load_color_buffer_8
= 0xBA,
342 midgard_op_load_uniform_16
= 0xAC,
343 midgard_op_load_uniform_32
= 0xB0,
344 midgard_op_store_vary_16
= 0xD5,
345 midgard_op_store_vary_32
= 0xD4
346 } midgard_load_store_op
;
349 midgard_interp_centroid
= 1,
350 midgard_interp_default
= 2
351 } midgard_interpolation
;
354 __attribute__((__packed__
))
356 unsigned zero1
: 4; /* Always zero */
358 /* Varying qualifiers, zero if not a varying */
360 unsigned is_varying
: 1; /* Always one for varying, but maybe something else? */
361 midgard_interpolation interpolation
: 2;
363 unsigned zero2
: 2; /* Always zero */
365 midgard_varying_parameter
;
368 __attribute__((__packed__
))
370 midgard_load_store_op op
: 8;
373 unsigned swizzle
: 8;
374 unsigned unknown
: 16;
376 unsigned varying_parameters
: 10;
378 unsigned address
: 9;
380 midgard_load_store_word
;
383 __attribute__((__packed__
))
386 unsigned next_type
: 4;
392 /* Texture pipeline results are in r28-r29 */
393 #define REG_TEX_BASE 28
395 /* Texture opcodes... maybe? */
396 #define TEXTURE_OP_NORMAL 0x11
397 #define TEXTURE_OP_TEXEL_FETCH 0x14
399 /* Texture format types, found in format */
400 #define TEXTURE_CUBE 0x00
401 #define TEXTURE_2D 0x02
402 #define TEXTURE_3D 0x03
405 __attribute__((__packed__
))
408 unsigned next_type
: 4;
412 unsigned unknown3
: 1;
414 /* A little obscure, but last is set for the last texture operation in
415 * a shader. cont appears to just be last's opposite (?). Yeah, I know,
416 * kind of funky.. BiOpen thinks it could do with memory hinting, or
423 unsigned has_offset
: 1;
425 /* Like in Bifrost */
428 unsigned in_reg_select
: 1;
429 unsigned in_reg_upper
: 1;
431 unsigned in_reg_swizzle_left
: 2;
432 unsigned in_reg_swizzle_right
: 2;
434 unsigned unknown1
: 2;
436 unsigned unknown8
: 4;
438 unsigned out_full
: 1;
440 /* Always 1 afaict... */
441 unsigned unknown7
: 2;
443 unsigned out_reg_select
: 1;
444 unsigned out_upper
: 1;
448 unsigned unknown2
: 2;
450 unsigned swizzle
: 8;
451 unsigned unknown4
: 8;
453 unsigned unknownA
: 4;
455 unsigned offset_unknown1
: 1;
456 unsigned offset_reg_select
: 1;
457 unsigned offset_reg_upper
: 1;
458 unsigned offset_unknown4
: 1;
459 unsigned offset_unknown5
: 1;
460 unsigned offset_unknown6
: 1;
461 unsigned offset_unknown7
: 1;
462 unsigned offset_unknown8
: 1;
463 unsigned offset_unknown9
: 1;
465 unsigned unknownB
: 3;
467 /* Texture bias or LOD, depending on whether it is executed in a
468 * fragment/vertex shader respectively. Compute as int(2^8 * biasf).
470 * For texel fetch, this is the LOD as is. */
473 unsigned unknown9
: 8;
475 unsigned texture_handle
: 16;
476 unsigned sampler_handle
: 16;
478 midgard_texture_word
;
480 /* Opcode name table */
482 static char *alu_opcode_names
[256] = {
483 [midgard_alu_op_fadd
] = "fadd",
484 [midgard_alu_op_fmul
] = "fmul",
485 [midgard_alu_op_fmin
] = "fmin",
486 [midgard_alu_op_fmax
] = "fmax",
487 [midgard_alu_op_fmov
] = "fmov",
488 [midgard_alu_op_froundeven
] = "froundeven",
489 [midgard_alu_op_ftrunc
] = "ftrunc",
490 [midgard_alu_op_ffloor
] = "ffloor",
491 [midgard_alu_op_fceil
] = "fceil",
492 [midgard_alu_op_ffma
] = "ffma",
493 [midgard_alu_op_fdot3
] = "fdot3",
494 [midgard_alu_op_fdot3r
] = "fdot3r",
495 [midgard_alu_op_fdot4
] = "fdot4",
496 [midgard_alu_op_freduce
] = "freduce",
497 [midgard_alu_op_imin
] = "imin",
498 [midgard_alu_op_umin
] = "umin",
499 [midgard_alu_op_imax
] = "imax",
500 [midgard_alu_op_umax
] = "umax",
501 [midgard_alu_op_ishl
] = "ishl",
502 [midgard_alu_op_iasr
] = "iasr",
503 [midgard_alu_op_ilsr
] = "ilsr",
504 [midgard_alu_op_iadd
] = "iadd",
505 [midgard_alu_op_ishladd
] = "ishladd",
506 [midgard_alu_op_isub
] = "isub",
507 [midgard_alu_op_imul
] = "imul",
508 [midgard_alu_op_imov
] = "imov",
509 [midgard_alu_op_iabs
] = "iabs",
510 [midgard_alu_op_iand
] = "iand",
511 [midgard_alu_op_ior
] = "ior",
512 [midgard_alu_op_inot
] = "inot",
513 [midgard_alu_op_iandnot
] = "iandnot",
514 [midgard_alu_op_ixor
] = "ixor",
515 [midgard_alu_op_ilzcnt
] = "ilzcnt",
516 [midgard_alu_op_ibitcount8
] = "ibitcount8",
517 [midgard_alu_op_feq
] = "feq",
518 [midgard_alu_op_fne
] = "fne",
519 [midgard_alu_op_flt
] = "flt",
520 [midgard_alu_op_fle
] = "fle",
521 [midgard_alu_op_fball_eq
] = "fball_eq",
522 [midgard_alu_op_fbany_neq
] = "fbany_neq",
523 [midgard_alu_op_bball_eq
] = "bball_eq",
524 [midgard_alu_op_fball_lt
] = "fball_lt",
525 [midgard_alu_op_fball_lte
] = "fball_lte",
526 [midgard_alu_op_bbany_neq
] = "bbany_neq",
527 [midgard_alu_op_fbany_lt
] = "fbany_lt",
528 [midgard_alu_op_fbany_lte
] = "fbany_lte",
529 [midgard_alu_op_f2i
] = "f2i",
530 [midgard_alu_op_f2u
] = "f2u",
531 [midgard_alu_op_f2u8
] = "f2u8",
532 [midgard_alu_op_ieq
] = "ieq",
533 [midgard_alu_op_ine
] = "ine",
534 [midgard_alu_op_ult
] = "ult",
535 [midgard_alu_op_ule
] = "ule",
536 [midgard_alu_op_ilt
] = "ilt",
537 [midgard_alu_op_ile
] = "ile",
538 [midgard_alu_op_iball_eq
] = "iball_eq",
539 [midgard_alu_op_ball
] = "ball",
540 [midgard_alu_op_uball_lt
] = "uball_lt",
541 [midgard_alu_op_uball_lte
] = "uball_lte",
542 [midgard_alu_op_iball_lt
] = "iball_lt",
543 [midgard_alu_op_iball_lte
] = "iball_lte",
544 [midgard_alu_op_iball_eq
] = "iball_eq",
545 [midgard_alu_op_ibany_neq
] = "ibany_neq",
546 [midgard_alu_op_ubany_lt
] = "ubany_lt",
547 [midgard_alu_op_ubany_lte
] = "ubany_lte",
548 [midgard_alu_op_ibany_lt
] = "ibany_lt",
549 [midgard_alu_op_ibany_lte
] = "ibany_lte",
550 [midgard_alu_op_i2f
] = "i2f",
551 [midgard_alu_op_u2f
] = "u2f",
552 [midgard_alu_op_icsel
] = "icsel",
553 [midgard_alu_op_fcsel_i
] = "fcsel_i",
554 [midgard_alu_op_fcsel
] = "fcsel",
555 [midgard_alu_op_fround
] = "fround",
556 [midgard_alu_op_fatan_pt2
] = "fatan_pt2",
557 [midgard_alu_op_frcp
] = "frcp",
558 [midgard_alu_op_frsqrt
] = "frsqrt",
559 [midgard_alu_op_fsqrt
] = "fsqrt",
560 [midgard_alu_op_fpow_pt1
] = "fpow_pt1",
561 [midgard_alu_op_fexp2
] = "fexp2",
562 [midgard_alu_op_flog2
] = "flog2",
563 [midgard_alu_op_fsin
] = "fsin",
564 [midgard_alu_op_fcos
] = "fcos",
565 [midgard_alu_op_fatan2_pt1
] = "fatan2_pt1"
568 static char *load_store_opcode_names
[256] = {
569 [midgard_op_store_cubemap_coords
] = "st_cubemap_coords",
570 [midgard_op_load_attr_16
] = "ld_attr_16",
571 [midgard_op_load_attr_32
] = "ld_attr_32",
572 [midgard_op_load_vary_16
] = "ld_vary_16",
573 [midgard_op_load_vary_32
] = "ld_vary_32",
574 [midgard_op_load_uniform_16
] = "ld_uniform_16",
575 [midgard_op_load_uniform_32
] = "ld_uniform_32",
576 [midgard_op_load_color_buffer_8
] = "ld_color_buffer_8",
577 [midgard_op_load_color_buffer_16
] = "ld_color_buffer_16",
578 [midgard_op_store_vary_16
] = "st_vary_16",
579 [midgard_op_store_vary_32
] = "st_vary_32"