projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
intel/fs: Add support for bindless image load/store/atomic
[mesa.git] / src / intel / compiler / brw_eu_defines.h
1 /*
2 Copyright (C) Intel Corp. 2006. All Rights Reserved.
3 Intel funded Tungsten Graphics to
4 develop this 3D driver.
5
6 Permission is hereby granted, free of charge, to any person obtaining
7 a copy of this software and associated documentation files (the
8 "Software"), to deal in the Software without restriction, including
9 without limitation the rights to use, copy, modify, merge, publish,
10 distribute, sublicense, and/or sell copies of the Software, and to
11 permit persons to whom the Software is furnished to do so, subject to
12 the following conditions:
13
14 The above copyright notice and this permission notice (including the
15 next paragraph) shall be included in all copies or substantial
16 portions of the Software.
17
18 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
19 EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
21 IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
22 LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
23 OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
24 WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
25
26 **********************************************************************/
27 /*
28 * Authors:
29 * Keith Whitwell <keithw@vmware.com>
30 */
31
32 #ifndef BRW_EU_DEFINES_H
33 #define BRW_EU_DEFINES_H
34
35 #include "util/macros.h"
36
37 /* The following hunk, up-to "Execution Unit" is used by both the
38 * intel/compiler and i965 codebase. */
39
40 #define INTEL_MASK(high, low) (((1u<<((high)-(low)+1))-1)<<(low))
41 /* Using the GNU statement expression extension */
42 #define SET_FIELD(value, field) \
43 ({ \
44 uint32_t fieldval = (uint32_t)(value) << field ## _SHIFT; \
45 assert((fieldval & ~ field ## _MASK) == 0); \
46 fieldval & field ## _MASK; \
47 })
48
49 #define SET_BITS(value, high, low) \
50 ({ \
51 const uint32_t fieldval = (uint32_t)(value) << (low); \
52 assert((fieldval & ~INTEL_MASK(high, low)) == 0); \
53 fieldval & INTEL_MASK(high, low); \
54 })
55
56 #define GET_BITS(data, high, low) ((data & INTEL_MASK((high), (low))) >> (low))
57 #define GET_FIELD(word, field) (((word) & field ## _MASK) >> field ## _SHIFT)
58
59 #define _3DPRIM_POINTLIST 0x01
60 #define _3DPRIM_LINELIST 0x02
61 #define _3DPRIM_LINESTRIP 0x03
62 #define _3DPRIM_TRILIST 0x04
63 #define _3DPRIM_TRISTRIP 0x05
64 #define _3DPRIM_TRIFAN 0x06
65 #define _3DPRIM_QUADLIST 0x07
66 #define _3DPRIM_QUADSTRIP 0x08
67 #define _3DPRIM_LINELIST_ADJ 0x09 /* G45+ */
68 #define _3DPRIM_LINESTRIP_ADJ 0x0A /* G45+ */
69 #define _3DPRIM_TRILIST_ADJ 0x0B /* G45+ */
70 #define _3DPRIM_TRISTRIP_ADJ 0x0C /* G45+ */
71 #define _3DPRIM_TRISTRIP_REVERSE 0x0D
72 #define _3DPRIM_POLYGON 0x0E
73 #define _3DPRIM_RECTLIST 0x0F
74 #define _3DPRIM_LINELOOP 0x10
75 #define _3DPRIM_POINTLIST_BF 0x11
76 #define _3DPRIM_LINESTRIP_CONT 0x12
77 #define _3DPRIM_LINESTRIP_BF 0x13
78 #define _3DPRIM_LINESTRIP_CONT_BF 0x14
79 #define _3DPRIM_TRIFAN_NOSTIPPLE 0x16
80 #define _3DPRIM_PATCHLIST(n) ({ assert(n > 0 && n <= 32); 0x20 + (n - 1); })
81
82 /* Bitfields for the URB_WRITE message, DW2 of message header: */
83 #define URB_WRITE_PRIM_END 0x1
84 #define URB_WRITE_PRIM_START 0x2
85 #define URB_WRITE_PRIM_TYPE_SHIFT 2
86
87 #define BRW_SPRITE_POINT_ENABLE 16
88
89 # define GEN7_GS_CONTROL_DATA_FORMAT_GSCTL_CUT 0
90 # define GEN7_GS_CONTROL_DATA_FORMAT_GSCTL_SID 1
91
92 /* Execution Unit (EU) defines
93 */
94
95 #define BRW_ALIGN_1 0
96 #define BRW_ALIGN_16 1
97
98 #define BRW_ADDRESS_DIRECT 0
99 #define BRW_ADDRESS_REGISTER_INDIRECT_REGISTER 1
100
101 #define BRW_CHANNEL_X 0
102 #define BRW_CHANNEL_Y 1
103 #define BRW_CHANNEL_Z 2
104 #define BRW_CHANNEL_W 3
105
106 enum brw_compression {
107 BRW_COMPRESSION_NONE = 0,
108 BRW_COMPRESSION_2NDHALF = 1,
109 BRW_COMPRESSION_COMPRESSED = 2,
110 };
111
112 #define GEN6_COMPRESSION_1Q 0
113 #define GEN6_COMPRESSION_2Q 1
114 #define GEN6_COMPRESSION_3Q 2
115 #define GEN6_COMPRESSION_4Q 3
116 #define GEN6_COMPRESSION_1H 0
117 #define GEN6_COMPRESSION_2H 2
118
119 enum PACKED brw_conditional_mod {
120 BRW_CONDITIONAL_NONE = 0,
121 BRW_CONDITIONAL_Z = 1,
122 BRW_CONDITIONAL_NZ = 2,
123 BRW_CONDITIONAL_EQ = 1, /* Z */
124 BRW_CONDITIONAL_NEQ = 2, /* NZ */
125 BRW_CONDITIONAL_G = 3,
126 BRW_CONDITIONAL_GE = 4,
127 BRW_CONDITIONAL_L = 5,
128 BRW_CONDITIONAL_LE = 6,
129 BRW_CONDITIONAL_R = 7, /* Gen <= 5 */
130 BRW_CONDITIONAL_O = 8,
131 BRW_CONDITIONAL_U = 9,
132 };
133
134 #define BRW_DEBUG_NONE 0
135 #define BRW_DEBUG_BREAKPOINT 1
136
137 #define BRW_DEPENDENCY_NORMAL 0
138 #define BRW_DEPENDENCY_NOTCLEARED 1
139 #define BRW_DEPENDENCY_NOTCHECKED 2
140 #define BRW_DEPENDENCY_DISABLE 3
141
142 enum PACKED brw_execution_size {
143 BRW_EXECUTE_1 = 0,
144 BRW_EXECUTE_2 = 1,
145 BRW_EXECUTE_4 = 2,
146 BRW_EXECUTE_8 = 3,
147 BRW_EXECUTE_16 = 4,
148 BRW_EXECUTE_32 = 5,
149 };
150
151 enum PACKED brw_horizontal_stride {
152 BRW_HORIZONTAL_STRIDE_0 = 0,
153 BRW_HORIZONTAL_STRIDE_1 = 1,
154 BRW_HORIZONTAL_STRIDE_2 = 2,
155 BRW_HORIZONTAL_STRIDE_4 = 3,
156 };
157
158 enum PACKED gen10_align1_3src_src_horizontal_stride {
159 BRW_ALIGN1_3SRC_SRC_HORIZONTAL_STRIDE_0 = 0,
160 BRW_ALIGN1_3SRC_SRC_HORIZONTAL_STRIDE_1 = 1,
161 BRW_ALIGN1_3SRC_SRC_HORIZONTAL_STRIDE_2 = 2,
162 BRW_ALIGN1_3SRC_SRC_HORIZONTAL_STRIDE_4 = 3,
163 };
164
165 enum PACKED gen10_align1_3src_dst_horizontal_stride {
166 BRW_ALIGN1_3SRC_DST_HORIZONTAL_STRIDE_1 = 0,
167 BRW_ALIGN1_3SRC_DST_HORIZONTAL_STRIDE_2 = 1,
168 };
169
170 #define BRW_INSTRUCTION_NORMAL 0
171 #define BRW_INSTRUCTION_SATURATE 1
172
173 #define BRW_MASK_ENABLE 0
174 #define BRW_MASK_DISABLE 1
175
176 /** @{
177 *
178 * Gen6 has replaced "mask enable/disable" with WECtrl, which is
179 * effectively the same but much simpler to think about. Now, there
180 * are two contributors ANDed together to whether channels are
181 * executed: The predication on the instruction, and the channel write
182 * enable.
183 */
184 /**
185 * This is the default value. It means that a channel's write enable is set
186 * if the per-channel IP is pointing at this instruction.
187 */
188 #define BRW_WE_NORMAL 0
189 /**
190 * This is used like BRW_MASK_DISABLE, and causes all channels to have
191 * their write enable set. Note that predication still contributes to
192 * whether the channel actually gets written.
193 */
194 #define BRW_WE_ALL 1
195 /** @} */
196
197 enum opcode {
198 /* These are the actual hardware opcodes. */
199 BRW_OPCODE_ILLEGAL = 0,
200 BRW_OPCODE_MOV = 1,
201 BRW_OPCODE_SEL = 2,
202 BRW_OPCODE_MOVI = 3, /**< G45+ */
203 BRW_OPCODE_NOT = 4,
204 BRW_OPCODE_AND = 5,
205 BRW_OPCODE_OR = 6,
206 BRW_OPCODE_XOR = 7,
207 BRW_OPCODE_SHR = 8,
208 BRW_OPCODE_SHL = 9,
209 BRW_OPCODE_DIM = 10, /**< Gen7.5 only */ /* Reused */
210 BRW_OPCODE_SMOV = 10, /**< Gen8+ */ /* Reused */
211 /* Reserved - 11 */
212 BRW_OPCODE_ASR = 12,
213 /* Reserved - 13-15 */
214 BRW_OPCODE_CMP = 16,
215 BRW_OPCODE_CMPN = 17,
216 BRW_OPCODE_CSEL = 18, /**< Gen8+ */
217 BRW_OPCODE_F32TO16 = 19, /**< Gen7 only */
218 BRW_OPCODE_F16TO32 = 20, /**< Gen7 only */
219 /* Reserved - 21-22 */
220 BRW_OPCODE_BFREV = 23, /**< Gen7+ */
221 BRW_OPCODE_BFE = 24, /**< Gen7+ */
222 BRW_OPCODE_BFI1 = 25, /**< Gen7+ */
223 BRW_OPCODE_BFI2 = 26, /**< Gen7+ */
224 /* Reserved - 27-31 */
225 BRW_OPCODE_JMPI = 32,
226 BRW_OPCODE_BRD = 33, /**< Gen7+ */
227 BRW_OPCODE_IF = 34,
228 BRW_OPCODE_IFF = 35, /**< Pre-Gen6 */ /* Reused */
229 BRW_OPCODE_BRC = 35, /**< Gen7+ */ /* Reused */
230 BRW_OPCODE_ELSE = 36,
231 BRW_OPCODE_ENDIF = 37,
232 BRW_OPCODE_DO = 38, /**< Pre-Gen6 */ /* Reused */
233 BRW_OPCODE_CASE = 38, /**< Gen6 only */ /* Reused */
234 BRW_OPCODE_WHILE = 39,
235 BRW_OPCODE_BREAK = 40,
236 BRW_OPCODE_CONTINUE = 41,
237 BRW_OPCODE_HALT = 42,
238 BRW_OPCODE_CALLA = 43, /**< Gen7.5+ */
239 BRW_OPCODE_MSAVE = 44, /**< Pre-Gen6 */ /* Reused */
240 BRW_OPCODE_CALL = 44, /**< Gen6+ */ /* Reused */
241 BRW_OPCODE_MREST = 45, /**< Pre-Gen6 */ /* Reused */
242 BRW_OPCODE_RET = 45, /**< Gen6+ */ /* Reused */
243 BRW_OPCODE_PUSH = 46, /**< Pre-Gen6 */ /* Reused */
244 BRW_OPCODE_FORK = 46, /**< Gen6 only */ /* Reused */
245 BRW_OPCODE_GOTO = 46, /**< Gen8+ */ /* Reused */
246 BRW_OPCODE_POP = 47, /**< Pre-Gen6 */
247 BRW_OPCODE_WAIT = 48,
248 BRW_OPCODE_SEND = 49,
249 BRW_OPCODE_SENDC = 50,
250 BRW_OPCODE_SENDS = 51, /**< Gen9+ */
251 BRW_OPCODE_SENDSC = 52, /**< Gen9+ */
252 /* Reserved 53-55 */
253 BRW_OPCODE_MATH = 56, /**< Gen6+ */
254 /* Reserved 57-63 */
255 BRW_OPCODE_ADD = 64,
256 BRW_OPCODE_MUL = 65,
257 BRW_OPCODE_AVG = 66,
258 BRW_OPCODE_FRC = 67,
259 BRW_OPCODE_RNDU = 68,
260 BRW_OPCODE_RNDD = 69,
261 BRW_OPCODE_RNDE = 70,
262 BRW_OPCODE_RNDZ = 71,
263 BRW_OPCODE_MAC = 72,
264 BRW_OPCODE_MACH = 73,
265 BRW_OPCODE_LZD = 74,
266 BRW_OPCODE_FBH = 75, /**< Gen7+ */
267 BRW_OPCODE_FBL = 76, /**< Gen7+ */
268 BRW_OPCODE_CBIT = 77, /**< Gen7+ */
269 BRW_OPCODE_ADDC = 78, /**< Gen7+ */
270 BRW_OPCODE_SUBB = 79, /**< Gen7+ */
271 BRW_OPCODE_SAD2 = 80,
272 BRW_OPCODE_SADA2 = 81,
273 /* Reserved 82-83 */
274 BRW_OPCODE_DP4 = 84,
275 BRW_OPCODE_DPH = 85,
276 BRW_OPCODE_DP3 = 86,
277 BRW_OPCODE_DP2 = 87,
278 /* Reserved 88 */
279 BRW_OPCODE_LINE = 89,
280 BRW_OPCODE_PLN = 90, /**< G45+ */
281 BRW_OPCODE_MAD = 91, /**< Gen6+ */
282 BRW_OPCODE_LRP = 92, /**< Gen6+ */
283 BRW_OPCODE_MADM = 93, /**< Gen8+ */
284 /* Reserved 94-124 */
285 BRW_OPCODE_NENOP = 125, /**< G45 only */
286 BRW_OPCODE_NOP = 126,
287 /* Reserved 127 */
288
289 /* These are compiler backend opcodes that get translated into other
290 * instructions.
291 */
292 FS_OPCODE_FB_WRITE = 128,
293
294 /**
295 * Same as FS_OPCODE_FB_WRITE but expects its arguments separately as
296 * individual sources instead of as a single payload blob. The
297 * position/ordering of the arguments are defined by the enum
298 * fb_write_logical_srcs.
299 */
300 FS_OPCODE_FB_WRITE_LOGICAL,
301
302 FS_OPCODE_REP_FB_WRITE,
303
304 FS_OPCODE_FB_READ,
305 FS_OPCODE_FB_READ_LOGICAL,
306
307 SHADER_OPCODE_RCP,
308 SHADER_OPCODE_RSQ,
309 SHADER_OPCODE_SQRT,
310 SHADER_OPCODE_EXP2,
311 SHADER_OPCODE_LOG2,
312 SHADER_OPCODE_POW,
313 SHADER_OPCODE_INT_QUOTIENT,
314 SHADER_OPCODE_INT_REMAINDER,
315 SHADER_OPCODE_SIN,
316 SHADER_OPCODE_COS,
317
318 /**
319 * A generic "send" opcode. The first two sources are the message
320 * descriptor and extended message descriptor respectively. The third
321 * and optional fourth sources are the message payload
322 */
323 SHADER_OPCODE_SEND,
324
325 /**
326 * Texture sampling opcodes.
327 *
328 * LOGICAL opcodes are eventually translated to the matching non-LOGICAL
329 * opcode but instead of taking a single payload blob they expect their
330 * arguments separately as individual sources. The position/ordering of the
331 * arguments are defined by the enum tex_logical_srcs.
332 */
333 SHADER_OPCODE_TEX,
334 SHADER_OPCODE_TEX_LOGICAL,
335 SHADER_OPCODE_TXD,
336 SHADER_OPCODE_TXD_LOGICAL,
337 SHADER_OPCODE_TXF,
338 SHADER_OPCODE_TXF_LOGICAL,
339 SHADER_OPCODE_TXF_LZ,
340 SHADER_OPCODE_TXL,
341 SHADER_OPCODE_TXL_LOGICAL,
342 SHADER_OPCODE_TXL_LZ,
343 SHADER_OPCODE_TXS,
344 SHADER_OPCODE_TXS_LOGICAL,
345 FS_OPCODE_TXB,
346 FS_OPCODE_TXB_LOGICAL,
347 SHADER_OPCODE_TXF_CMS,
348 SHADER_OPCODE_TXF_CMS_LOGICAL,
349 SHADER_OPCODE_TXF_CMS_W,
350 SHADER_OPCODE_TXF_CMS_W_LOGICAL,
351 SHADER_OPCODE_TXF_UMS,
352 SHADER_OPCODE_TXF_UMS_LOGICAL,
353 SHADER_OPCODE_TXF_MCS,
354 SHADER_OPCODE_TXF_MCS_LOGICAL,
355 SHADER_OPCODE_LOD,
356 SHADER_OPCODE_LOD_LOGICAL,
357 SHADER_OPCODE_TG4,
358 SHADER_OPCODE_TG4_LOGICAL,
359 SHADER_OPCODE_TG4_OFFSET,
360 SHADER_OPCODE_TG4_OFFSET_LOGICAL,
361 SHADER_OPCODE_SAMPLEINFO,
362 SHADER_OPCODE_SAMPLEINFO_LOGICAL,
363
364 SHADER_OPCODE_IMAGE_SIZE_LOGICAL,
365
366 /**
367 * Combines multiple sources of size 1 into a larger virtual GRF.
368 * For example, parameters for a send-from-GRF message. Or, updating
369 * channels of a size 4 VGRF used to store vec4s such as texturing results.
370 *
371 * This will be lowered into MOVs from each source to consecutive offsets
372 * of the destination VGRF.
373 *
374 * src[0] may be BAD_FILE. If so, the lowering pass skips emitting the MOV,
375 * but still reserves the first channel of the destination VGRF. This can be
376 * used to reserve space for, say, a message header set up by the generators.
377 */
378 SHADER_OPCODE_LOAD_PAYLOAD,
379
380 /**
381 * Packs a number of sources into a single value. Unlike LOAD_PAYLOAD, this
382 * acts intra-channel, obtaining the final value for each channel by
383 * combining the sources values for the same channel, the first source
384 * occupying the lowest bits and the last source occupying the highest
385 * bits.
386 */
387 FS_OPCODE_PACK,
388
389 SHADER_OPCODE_SHADER_TIME_ADD,
390
391 /**
392 * Typed and untyped surface access opcodes.
393 *
394 * LOGICAL opcodes are eventually translated to the matching non-LOGICAL
395 * opcode but instead of taking a single payload blob they expect their
396 * arguments separately as individual sources:
397 *
398 * Source 0: [required] Surface coordinates.
399 * Source 1: [optional] Operation source.
400 * Source 2: [required] Surface index.
401 * Source 3: [required] Number of coordinate components (as UD immediate).
402 * Source 4: [required] Opcode-specific control immediate, same as source 2
403 * of the matching non-LOGICAL opcode.
404 */
405 VEC4_OPCODE_UNTYPED_ATOMIC,
406 SHADER_OPCODE_UNTYPED_ATOMIC_LOGICAL,
407 SHADER_OPCODE_UNTYPED_ATOMIC_FLOAT_LOGICAL,
408 VEC4_OPCODE_UNTYPED_SURFACE_READ,
409 SHADER_OPCODE_UNTYPED_SURFACE_READ_LOGICAL,
410 VEC4_OPCODE_UNTYPED_SURFACE_WRITE,
411 SHADER_OPCODE_UNTYPED_SURFACE_WRITE_LOGICAL,
412
413 /**
414 * Untyped A64 surface access opcodes.
415 *
416 * Source 0: 64-bit address
417 * Source 1: Operational source
418 * Source 2: [required] Opcode-specific control immediate, same as source 2
419 * of the matching non-LOGICAL opcode.
420 */
421 SHADER_OPCODE_A64_UNTYPED_READ_LOGICAL,
422 SHADER_OPCODE_A64_UNTYPED_WRITE_LOGICAL,
423 SHADER_OPCODE_A64_BYTE_SCATTERED_READ_LOGICAL,
424 SHADER_OPCODE_A64_BYTE_SCATTERED_WRITE_LOGICAL,
425 SHADER_OPCODE_A64_UNTYPED_ATOMIC_LOGICAL,
426 SHADER_OPCODE_A64_UNTYPED_ATOMIC_INT64_LOGICAL,
427 SHADER_OPCODE_A64_UNTYPED_ATOMIC_FLOAT_LOGICAL,
428
429 SHADER_OPCODE_TYPED_ATOMIC_LOGICAL,
430 SHADER_OPCODE_TYPED_SURFACE_READ_LOGICAL,
431 SHADER_OPCODE_TYPED_SURFACE_WRITE_LOGICAL,
432
433 SHADER_OPCODE_RND_MODE,
434
435 /**
436 * Byte scattered write/read opcodes.
437 *
438 * LOGICAL opcodes are eventually translated to the matching non-LOGICAL
439 * opcode, but instead of taking a single payload blog they expect their
440 * arguments separately as individual sources, like untyped write/read.
441 */
442 SHADER_OPCODE_BYTE_SCATTERED_READ_LOGICAL,
443 SHADER_OPCODE_BYTE_SCATTERED_WRITE_LOGICAL,
444
445 SHADER_OPCODE_MEMORY_FENCE,
446
447 SHADER_OPCODE_GEN4_SCRATCH_READ,
448 SHADER_OPCODE_GEN4_SCRATCH_WRITE,
449 SHADER_OPCODE_GEN7_SCRATCH_READ,
450
451 /**
452 * Gen8+ SIMD8 URB Read messages.
453 */
454 SHADER_OPCODE_URB_READ_SIMD8,
455 SHADER_OPCODE_URB_READ_SIMD8_PER_SLOT,
456
457 SHADER_OPCODE_URB_WRITE_SIMD8,
458 SHADER_OPCODE_URB_WRITE_SIMD8_PER_SLOT,
459 SHADER_OPCODE_URB_WRITE_SIMD8_MASKED,
460 SHADER_OPCODE_URB_WRITE_SIMD8_MASKED_PER_SLOT,
461
462 /**
463 * Return the index of an arbitrary live channel (i.e. one of the channels
464 * enabled in the current execution mask) and assign it to the first
465 * component of the destination. Expected to be used as input for the
466 * BROADCAST pseudo-opcode.
467 */
468 SHADER_OPCODE_FIND_LIVE_CHANNEL,
469
470 /**
471 * Pick the channel from its first source register given by the index
472 * specified as second source. Useful for variable indexing of surfaces.
473 *
474 * Note that because the result of this instruction is by definition
475 * uniform and it can always be splatted to multiple channels using a
476 * scalar regioning mode, only the first channel of the destination region
477 * is guaranteed to be updated, which implies that BROADCAST instructions
478 * should usually be marked force_writemask_all.
479 */
480 SHADER_OPCODE_BROADCAST,
481
482 /* Pick the channel from its first source register given by the index
483 * specified as second source.
484 *
485 * This is similar to the BROADCAST instruction except that it takes a
486 * dynamic index and potentially puts a different value in each output
487 * channel.
488 */
489 SHADER_OPCODE_SHUFFLE,
490
491 /* Select between src0 and src1 based on channel enables.
492 *
493 * This instruction copies src0 into the enabled channels of the
494 * destination and copies src1 into the disabled channels.
495 */
496 SHADER_OPCODE_SEL_EXEC,
497
498 /* This turns into an align16 mov from src0 to dst with a swizzle
499 * provided as an immediate in src1.
500 */
501 SHADER_OPCODE_QUAD_SWIZZLE,
502
503 /* Take every Nth element in src0 and broadcast it to the group of N
504 * channels in which it lives in the destination. The offset within the
505 * cluster is given by src1 and the cluster size is given by src2.
506 */
507 SHADER_OPCODE_CLUSTER_BROADCAST,
508
509 SHADER_OPCODE_GET_BUFFER_SIZE,
510
511 SHADER_OPCODE_INTERLOCK,
512
513 VEC4_OPCODE_MOV_BYTES,
514 VEC4_OPCODE_PACK_BYTES,
515 VEC4_OPCODE_UNPACK_UNIFORM,
516 VEC4_OPCODE_DOUBLE_TO_F32,
517 VEC4_OPCODE_DOUBLE_TO_D32,
518 VEC4_OPCODE_DOUBLE_TO_U32,
519 VEC4_OPCODE_TO_DOUBLE,
520 VEC4_OPCODE_PICK_LOW_32BIT,
521 VEC4_OPCODE_PICK_HIGH_32BIT,
522 VEC4_OPCODE_SET_LOW_32BIT,
523 VEC4_OPCODE_SET_HIGH_32BIT,
524
525 FS_OPCODE_DDX_COARSE,
526 FS_OPCODE_DDX_FINE,
527 /**
528 * Compute dFdy(), dFdyCoarse(), or dFdyFine().
529 */
530 FS_OPCODE_DDY_COARSE,
531 FS_OPCODE_DDY_FINE,
532 FS_OPCODE_LINTERP,
533 FS_OPCODE_PIXEL_X,
534 FS_OPCODE_PIXEL_Y,
535 FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD,
536 FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD_GEN7,
537 FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_GEN4,
538 FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_LOGICAL,
539 FS_OPCODE_DISCARD_JUMP,
540 FS_OPCODE_SET_SAMPLE_ID,
541 FS_OPCODE_PACK_HALF_2x16_SPLIT,
542 FS_OPCODE_PLACEHOLDER_HALT,
543 FS_OPCODE_INTERPOLATE_AT_SAMPLE,
544 FS_OPCODE_INTERPOLATE_AT_SHARED_OFFSET,
545 FS_OPCODE_INTERPOLATE_AT_PER_SLOT_OFFSET,
546
547 VS_OPCODE_URB_WRITE,
548 VS_OPCODE_PULL_CONSTANT_LOAD,
549 VS_OPCODE_PULL_CONSTANT_LOAD_GEN7,
550 VS_OPCODE_SET_SIMD4X2_HEADER_GEN9,
551
552 VS_OPCODE_UNPACK_FLAGS_SIMD4X2,
553
554 /**
555 * Write geometry shader output data to the URB.
556 *
557 * Unlike VS_OPCODE_URB_WRITE, this opcode doesn't do an implied move from
558 * R0 to the first MRF. This allows the geometry shader to override the
559 * "Slot {0,1} Offset" fields in the message header.
560 */
561 GS_OPCODE_URB_WRITE,
562
563 /**
564 * Write geometry shader output data to the URB and request a new URB
565 * handle (gen6).
566 *
567 * This opcode doesn't do an implied move from R0 to the first MRF.
568 */
569 GS_OPCODE_URB_WRITE_ALLOCATE,
570
571 /**
572 * Terminate the geometry shader thread by doing an empty URB write.
573 *
574 * This opcode doesn't do an implied move from R0 to the first MRF. This
575 * allows the geometry shader to override the "GS Number of Output Vertices
576 * for Slot {0,1}" fields in the message header.
577 */
578 GS_OPCODE_THREAD_END,
579
580 /**
581 * Set the "Slot {0,1} Offset" fields of a URB_WRITE message header.
582 *
583 * - dst is the MRF containing the message header.
584 *
585 * - src0.x indicates which portion of the URB should be written to (e.g. a
586 * vertex number)
587 *
588 * - src1 is an immediate multiplier which will be applied to src0
589 * (e.g. the size of a single vertex in the URB).
590 *
591 * Note: the hardware will apply this offset *in addition to* the offset in
592 * vec4_instruction::offset.
593 */
594 GS_OPCODE_SET_WRITE_OFFSET,
595
596 /**
597 * Set the "GS Number of Output Vertices for Slot {0,1}" fields of a
598 * URB_WRITE message header.
599 *
600 * - dst is the MRF containing the message header.
601 *
602 * - src0.x is the vertex count. The upper 16 bits will be ignored.
603 */
604 GS_OPCODE_SET_VERTEX_COUNT,
605
606 /**
607 * Set DWORD 2 of dst to the value in src.
608 */
609 GS_OPCODE_SET_DWORD_2,
610
611 /**
612 * Prepare the dst register for storage in the "Channel Mask" fields of a
613 * URB_WRITE message header.
614 *
615 * DWORD 4 of dst is shifted left by 4 bits, so that later,
616 * GS_OPCODE_SET_CHANNEL_MASKS can OR DWORDs 0 and 4 together to form the
617 * final channel mask.
618 *
619 * Note: since GS_OPCODE_SET_CHANNEL_MASKS ORs DWORDs 0 and 4 together to
620 * form the final channel mask, DWORDs 0 and 4 of the dst register must not
621 * have any extraneous bits set prior to execution of this opcode (that is,
622 * they should be in the range 0x0 to 0xf).
623 */
624 GS_OPCODE_PREPARE_CHANNEL_MASKS,
625
626 /**
627 * Set the "Channel Mask" fields of a URB_WRITE message header.
628 *
629 * - dst is the MRF containing the message header.
630 *
631 * - src.x is the channel mask, as prepared by
632 * GS_OPCODE_PREPARE_CHANNEL_MASKS. DWORDs 0 and 4 are OR'ed together to
633 * form the final channel mask.
634 */
635 GS_OPCODE_SET_CHANNEL_MASKS,
636
637 /**
638 * Get the "Instance ID" fields from the payload.
639 *
640 * - dst is the GRF for gl_InvocationID.
641 */
642 GS_OPCODE_GET_INSTANCE_ID,
643
644 /**
645 * Send a FF_SYNC message to allocate initial URB handles (gen6).
646 *
647 * - dst will be used as the writeback register for the FF_SYNC operation.
648 *
649 * - src0 is the number of primitives written.
650 *
651 * - src1 is the value to hold in M0.0: number of SO vertices to write
652 * and number of SO primitives needed. Its value will be overwritten
653 * with the SVBI values if transform feedback is enabled.
654 *
655 * Note: This opcode uses an implicit MRF register for the ff_sync message
656 * header, so the caller is expected to set inst->base_mrf and initialize
657 * that MRF register to r0. This opcode will also write to this MRF register
658 * to include the allocated URB handle so it can then be reused directly as
659 * the header in the URB write operation we are allocating the handle for.
660 */
661 GS_OPCODE_FF_SYNC,
662
663 /**
664 * Move r0.1 (which holds PrimitiveID information in gen6) to a separate
665 * register.
666 *
667 * - dst is the GRF where PrimitiveID information will be moved.
668 */
669 GS_OPCODE_SET_PRIMITIVE_ID,
670
671 /**
672 * Write transform feedback data to the SVB by sending a SVB WRITE message.
673 * Used in gen6.
674 *
675 * - dst is the MRF register containing the message header.
676 *
677 * - src0 is the register where the vertex data is going to be copied from.
678 *
679 * - src1 is the destination register when write commit occurs.
680 */
681 GS_OPCODE_SVB_WRITE,
682
683 /**
684 * Set destination index in the SVB write message payload (M0.5). Used
685 * in gen6 for transform feedback.
686 *
687 * - dst is the header to save the destination indices for SVB WRITE.
688 * - src is the register that holds the destination indices value.
689 */
690 GS_OPCODE_SVB_SET_DST_INDEX,
691
692 /**
693 * Prepare Mx.0 subregister for being used in the FF_SYNC message header.
694 * Used in gen6 for transform feedback.
695 *
696 * - dst will hold the register with the final Mx.0 value.
697 *
698 * - src0 has the number of vertices emitted in SO (NumSOVertsToWrite)
699 *
700 * - src1 has the number of needed primitives for SO (NumSOPrimsNeeded)
701 *
702 * - src2 is the value to hold in M0: number of SO vertices to write
703 * and number of SO primitives needed.
704 */
705 GS_OPCODE_FF_SYNC_SET_PRIMITIVES,
706
707 /**
708 * Terminate the compute shader.
709 */
710 CS_OPCODE_CS_TERMINATE,
711
712 /**
713 * GLSL barrier()
714 */
715 SHADER_OPCODE_BARRIER,
716
717 /**
718 * Calculate the high 32-bits of a 32x32 multiply.
719 */
720 SHADER_OPCODE_MULH,
721
722 /**
723 * A MOV that uses VxH indirect addressing.
724 *
725 * Source 0: A register to start from (HW_REG).
726 * Source 1: An indirect offset (in bytes, UD GRF).
727 * Source 2: The length of the region that could be accessed (in bytes,
728 * UD immediate).
729 */
730 SHADER_OPCODE_MOV_INDIRECT,
731
732 VEC4_OPCODE_URB_READ,
733 TCS_OPCODE_GET_INSTANCE_ID,
734 TCS_OPCODE_URB_WRITE,
735 TCS_OPCODE_SET_INPUT_URB_OFFSETS,
736 TCS_OPCODE_SET_OUTPUT_URB_OFFSETS,
737 TCS_OPCODE_GET_PRIMITIVE_ID,
738 TCS_OPCODE_CREATE_BARRIER_HEADER,
739 TCS_OPCODE_SRC0_010_IS_ZERO,
740 TCS_OPCODE_RELEASE_INPUT,
741 TCS_OPCODE_THREAD_END,
742
743 TES_OPCODE_GET_PRIMITIVE_ID,
744 TES_OPCODE_CREATE_INPUT_READ_HEADER,
745 TES_OPCODE_ADD_INDIRECT_URB_OFFSET,
746 };
747
748 enum brw_urb_write_flags {
749 BRW_URB_WRITE_NO_FLAGS = 0,
750
751 /**
752 * Causes a new URB entry to be allocated, and its address stored in the
753 * destination register (gen < 7).
754 */
755 BRW_URB_WRITE_ALLOCATE = 0x1,
756
757 /**
758 * Causes the current URB entry to be deallocated (gen < 7).
759 */
760 BRW_URB_WRITE_UNUSED = 0x2,
761
762 /**
763 * Causes the thread to terminate.
764 */
765 BRW_URB_WRITE_EOT = 0x4,
766
767 /**
768 * Indicates that the given URB entry is complete, and may be sent further
769 * down the 3D pipeline (gen < 7).
770 */
771 BRW_URB_WRITE_COMPLETE = 0x8,
772
773 /**
774 * Indicates that an additional offset (which may be different for the two
775 * vec4 slots) is stored in the message header (gen == 7).
776 */
777 BRW_URB_WRITE_PER_SLOT_OFFSET = 0x10,
778
779 /**
780 * Indicates that the channel masks in the URB_WRITE message header should
781 * not be overridden to 0xff (gen == 7).
782 */
783 BRW_URB_WRITE_USE_CHANNEL_MASKS = 0x20,
784
785 /**
786 * Indicates that the data should be sent to the URB using the
787 * URB_WRITE_OWORD message rather than URB_WRITE_HWORD (gen == 7). This
788 * causes offsets to be interpreted as multiples of an OWORD instead of an
789 * HWORD, and only allows one OWORD to be written.
790 */
791 BRW_URB_WRITE_OWORD = 0x40,
792
793 /**
794 * Convenient combination of flags: end the thread while simultaneously
795 * marking the given URB entry as complete.
796 */
797 BRW_URB_WRITE_EOT_COMPLETE = BRW_URB_WRITE_EOT | BRW_URB_WRITE_COMPLETE,
798
799 /**
800 * Convenient combination of flags: mark the given URB entry as complete
801 * and simultaneously allocate a new one.
802 */
803 BRW_URB_WRITE_ALLOCATE_COMPLETE =
804 BRW_URB_WRITE_ALLOCATE | BRW_URB_WRITE_COMPLETE,
805 };
806
807 enum fb_write_logical_srcs {
808 FB_WRITE_LOGICAL_SRC_COLOR0, /* REQUIRED */
809 FB_WRITE_LOGICAL_SRC_COLOR1, /* for dual source blend messages */
810 FB_WRITE_LOGICAL_SRC_SRC0_ALPHA,
811 FB_WRITE_LOGICAL_SRC_SRC_DEPTH, /* gl_FragDepth */
812 FB_WRITE_LOGICAL_SRC_DST_DEPTH, /* GEN4-5: passthrough from thread */
813 FB_WRITE_LOGICAL_SRC_SRC_STENCIL, /* gl_FragStencilRefARB */
814 FB_WRITE_LOGICAL_SRC_OMASK, /* Sample Mask (gl_SampleMask) */
815 FB_WRITE_LOGICAL_SRC_COMPONENTS, /* REQUIRED */
816 FB_WRITE_LOGICAL_NUM_SRCS
817 };
818
819 enum tex_logical_srcs {
820 /** Texture coordinates */
821 TEX_LOGICAL_SRC_COORDINATE,
822 /** Shadow comparator */
823 TEX_LOGICAL_SRC_SHADOW_C,
824 /** dPdx if the operation takes explicit derivatives, otherwise LOD value */
825 TEX_LOGICAL_SRC_LOD,
826 /** dPdy if the operation takes explicit derivatives */
827 TEX_LOGICAL_SRC_LOD2,
828 /** Min LOD */
829 TEX_LOGICAL_SRC_MIN_LOD,
830 /** Sample index */
831 TEX_LOGICAL_SRC_SAMPLE_INDEX,
832 /** MCS data */
833 TEX_LOGICAL_SRC_MCS,
834 /** REQUIRED: Texture surface index */
835 TEX_LOGICAL_SRC_SURFACE,
836 /** Texture sampler index */
837 TEX_LOGICAL_SRC_SAMPLER,
838 /** Texture surface bindless handle */
839 TEX_LOGICAL_SRC_SURFACE_HANDLE,
840 /** Texture sampler bindless handle */
841 TEX_LOGICAL_SRC_SAMPLER_HANDLE,
842 /** Texel offset for gathers */
843 TEX_LOGICAL_SRC_TG4_OFFSET,
844 /** REQUIRED: Number of coordinate components (as UD immediate) */
845 TEX_LOGICAL_SRC_COORD_COMPONENTS,
846 /** REQUIRED: Number of derivative components (as UD immediate) */
847 TEX_LOGICAL_SRC_GRAD_COMPONENTS,
848
849 TEX_LOGICAL_NUM_SRCS,
850 };
851
852 enum surface_logical_srcs {
853 /** Surface binding table index */
854 SURFACE_LOGICAL_SRC_SURFACE,
855 /** Surface bindless handle */
856 SURFACE_LOGICAL_SRC_SURFACE_HANDLE,
857 /** Surface address; could be multi-dimensional for typed opcodes */
858 SURFACE_LOGICAL_SRC_ADDRESS,
859 /** Data to be written or used in an atomic op */
860 SURFACE_LOGICAL_SRC_DATA,
861 /** Surface number of dimensions. Affects the size of ADDRESS */
862 SURFACE_LOGICAL_SRC_IMM_DIMS,
863 /** Per-opcode immediate argument. For atomics, this is the atomic opcode */
864 SURFACE_LOGICAL_SRC_IMM_ARG,
865
866 SURFACE_LOGICAL_NUM_SRCS
867 };
868
869 #ifdef __cplusplus
870 /**
871 * Allow brw_urb_write_flags enums to be ORed together.
872 */
873 inline brw_urb_write_flags
874 operator|(brw_urb_write_flags x, brw_urb_write_flags y)
875 {
876 return static_cast<brw_urb_write_flags>(static_cast<int>(x) |
877 static_cast<int>(y));
878 }
879 #endif
880
881 enum PACKED brw_predicate {
882 BRW_PREDICATE_NONE = 0,
883 BRW_PREDICATE_NORMAL = 1,
884 BRW_PREDICATE_ALIGN1_ANYV = 2,
885 BRW_PREDICATE_ALIGN1_ALLV = 3,
886 BRW_PREDICATE_ALIGN1_ANY2H = 4,
887 BRW_PREDICATE_ALIGN1_ALL2H = 5,
888 BRW_PREDICATE_ALIGN1_ANY4H = 6,
889 BRW_PREDICATE_ALIGN1_ALL4H = 7,
890 BRW_PREDICATE_ALIGN1_ANY8H = 8,
891 BRW_PREDICATE_ALIGN1_ALL8H = 9,
892 BRW_PREDICATE_ALIGN1_ANY16H = 10,
893 BRW_PREDICATE_ALIGN1_ALL16H = 11,
894 BRW_PREDICATE_ALIGN1_ANY32H = 12,
895 BRW_PREDICATE_ALIGN1_ALL32H = 13,
896 BRW_PREDICATE_ALIGN16_REPLICATE_X = 2,
897 BRW_PREDICATE_ALIGN16_REPLICATE_Y = 3,
898 BRW_PREDICATE_ALIGN16_REPLICATE_Z = 4,
899 BRW_PREDICATE_ALIGN16_REPLICATE_W = 5,
900 BRW_PREDICATE_ALIGN16_ANY4H = 6,
901 BRW_PREDICATE_ALIGN16_ALL4H = 7,
902 };
903
904 enum PACKED brw_reg_file {
905 BRW_ARCHITECTURE_REGISTER_FILE = 0,
906 BRW_GENERAL_REGISTER_FILE = 1,
907 BRW_MESSAGE_REGISTER_FILE = 2,
908 BRW_IMMEDIATE_VALUE = 3,
909
910 ARF = BRW_ARCHITECTURE_REGISTER_FILE,
911 FIXED_GRF = BRW_GENERAL_REGISTER_FILE,
912 MRF = BRW_MESSAGE_REGISTER_FILE,
913 IMM = BRW_IMMEDIATE_VALUE,
914
915 /* These are not hardware values */
916 VGRF,
917 ATTR,
918 UNIFORM, /* prog_data->params[reg] */
919 BAD_FILE,
920 };
921
922 enum PACKED gen10_align1_3src_reg_file {
923 BRW_ALIGN1_3SRC_GENERAL_REGISTER_FILE = 0,
924 BRW_ALIGN1_3SRC_IMMEDIATE_VALUE = 1, /* src0, src2 */
925 BRW_ALIGN1_3SRC_ACCUMULATOR = 1, /* dest, src1 */
926 };
927
928 /* CNL adds Align1 support for 3-src instructions. Bit 35 of the instruction
929 * word is "Execution Datatype" which controls whether the instruction operates
930 * on float or integer types. The register arguments have fields that offer
931 * more fine control their respective types.
932 */
933 enum PACKED gen10_align1_3src_exec_type {
934 BRW_ALIGN1_3SRC_EXEC_TYPE_INT = 0,
935 BRW_ALIGN1_3SRC_EXEC_TYPE_FLOAT = 1,
936 };
937
938 #define BRW_ARF_NULL 0x00
939 #define BRW_ARF_ADDRESS 0x10
940 #define BRW_ARF_ACCUMULATOR 0x20
941 #define BRW_ARF_FLAG 0x30
942 #define BRW_ARF_MASK 0x40
943 #define BRW_ARF_MASK_STACK 0x50
944 #define BRW_ARF_MASK_STACK_DEPTH 0x60
945 #define BRW_ARF_STATE 0x70
946 #define BRW_ARF_CONTROL 0x80
947 #define BRW_ARF_NOTIFICATION_COUNT 0x90
948 #define BRW_ARF_IP 0xA0
949 #define BRW_ARF_TDR 0xB0
950 #define BRW_ARF_TIMESTAMP 0xC0
951
952 #define BRW_MRF_COMPR4 (1 << 7)
953
954 #define BRW_AMASK 0
955 #define BRW_IMASK 1
956 #define BRW_LMASK 2
957 #define BRW_CMASK 3
958
959
960
961 #define BRW_THREAD_NORMAL 0
962 #define BRW_THREAD_ATOMIC 1
963 #define BRW_THREAD_SWITCH 2
964
965 enum PACKED brw_vertical_stride {
966 BRW_VERTICAL_STRIDE_0 = 0,
967 BRW_VERTICAL_STRIDE_1 = 1,
968 BRW_VERTICAL_STRIDE_2 = 2,
969 BRW_VERTICAL_STRIDE_4 = 3,
970 BRW_VERTICAL_STRIDE_8 = 4,
971 BRW_VERTICAL_STRIDE_16 = 5,
972 BRW_VERTICAL_STRIDE_32 = 6,
973 BRW_VERTICAL_STRIDE_ONE_DIMENSIONAL = 0xF,
974 };
975
976 enum PACKED gen10_align1_3src_vertical_stride {
977 BRW_ALIGN1_3SRC_VERTICAL_STRIDE_0 = 0,
978 BRW_ALIGN1_3SRC_VERTICAL_STRIDE_2 = 1,
979 BRW_ALIGN1_3SRC_VERTICAL_STRIDE_4 = 2,
980 BRW_ALIGN1_3SRC_VERTICAL_STRIDE_8 = 3,
981 };
982
983 enum PACKED brw_width {
984 BRW_WIDTH_1 = 0,
985 BRW_WIDTH_2 = 1,
986 BRW_WIDTH_4 = 2,
987 BRW_WIDTH_8 = 3,
988 BRW_WIDTH_16 = 4,
989 };
990
991 /**
992 * Message target: Shared Function ID for where to SEND a message.
993 *
994 * These are enumerated in the ISA reference under "send - Send Message".
995 * In particular, see the following tables:
996 * - G45 PRM, Volume 4, Table 14-15 "Message Descriptor Definition"
997 * - Sandybridge PRM, Volume 4 Part 2, Table 8-16 "Extended Message Descriptor"
998 * - Ivybridge PRM, Volume 1 Part 1, section 3.2.7 "GPE Function IDs"
999 */
1000 enum brw_message_target {
1001 BRW_SFID_NULL = 0,
1002 BRW_SFID_MATH = 1, /* Only valid on Gen4-5 */
1003 BRW_SFID_SAMPLER = 2,
1004 BRW_SFID_MESSAGE_GATEWAY = 3,
1005 BRW_SFID_DATAPORT_READ = 4,
1006 BRW_SFID_DATAPORT_WRITE = 5,
1007 BRW_SFID_URB = 6,
1008 BRW_SFID_THREAD_SPAWNER = 7,
1009 BRW_SFID_VME = 8,
1010
1011 GEN6_SFID_DATAPORT_SAMPLER_CACHE = 4,
1012 GEN6_SFID_DATAPORT_RENDER_CACHE = 5,
1013 GEN6_SFID_DATAPORT_CONSTANT_CACHE = 9,
1014
1015 GEN7_SFID_DATAPORT_DATA_CACHE = 10,
1016 GEN7_SFID_PIXEL_INTERPOLATOR = 11,
1017 HSW_SFID_DATAPORT_DATA_CACHE_1 = 12,
1018 HSW_SFID_CRE = 13,
1019 };
1020
1021 #define GEN7_MESSAGE_TARGET_DP_DATA_CACHE 10
1022
1023 #define BRW_SAMPLER_RETURN_FORMAT_FLOAT32 0
1024 #define BRW_SAMPLER_RETURN_FORMAT_UINT32 2
1025 #define BRW_SAMPLER_RETURN_FORMAT_SINT32 3
1026
1027 #define BRW_SAMPLER_MESSAGE_SIMD8_SAMPLE 0
1028 #define BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE 0
1029 #define BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE_BIAS 0
1030 #define BRW_SAMPLER_MESSAGE_SIMD8_KILLPIX 1
1031 #define BRW_SAMPLER_MESSAGE_SIMD4X2_SAMPLE_LOD 1
1032 #define BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE_LOD 1
1033 #define BRW_SAMPLER_MESSAGE_SIMD4X2_SAMPLE_GRADIENTS 2
1034 #define BRW_SAMPLER_MESSAGE_SIMD8_SAMPLE_GRADIENTS 2
1035 #define BRW_SAMPLER_MESSAGE_SIMD4X2_SAMPLE_COMPARE 0
1036 #define BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE_COMPARE 2
1037 #define BRW_SAMPLER_MESSAGE_SIMD8_SAMPLE_BIAS_COMPARE 0
1038 #define BRW_SAMPLER_MESSAGE_SIMD4X2_SAMPLE_LOD_COMPARE 1
1039 #define BRW_SAMPLER_MESSAGE_SIMD8_SAMPLE_LOD_COMPARE 1
1040 #define BRW_SAMPLER_MESSAGE_SIMD4X2_RESINFO 2
1041 #define BRW_SAMPLER_MESSAGE_SIMD16_RESINFO 2
1042 #define BRW_SAMPLER_MESSAGE_SIMD4X2_LD 3
1043 #define BRW_SAMPLER_MESSAGE_SIMD8_LD 3
1044 #define BRW_SAMPLER_MESSAGE_SIMD16_LD 3
1045
1046 #define GEN5_SAMPLER_MESSAGE_SAMPLE 0
1047 #define GEN5_SAMPLER_MESSAGE_SAMPLE_BIAS 1
1048 #define GEN5_SAMPLER_MESSAGE_SAMPLE_LOD 2
1049 #define GEN5_SAMPLER_MESSAGE_SAMPLE_COMPARE 3
1050 #define GEN5_SAMPLER_MESSAGE_SAMPLE_DERIVS 4
1051 #define GEN5_SAMPLER_MESSAGE_SAMPLE_BIAS_COMPARE 5
1052 #define GEN5_SAMPLER_MESSAGE_SAMPLE_LOD_COMPARE 6
1053 #define GEN5_SAMPLER_MESSAGE_SAMPLE_LD 7
1054 #define GEN7_SAMPLER_MESSAGE_SAMPLE_GATHER4 8
1055 #define GEN5_SAMPLER_MESSAGE_LOD 9
1056 #define GEN5_SAMPLER_MESSAGE_SAMPLE_RESINFO 10
1057 #define GEN6_SAMPLER_MESSAGE_SAMPLE_SAMPLEINFO 11
1058 #define GEN7_SAMPLER_MESSAGE_SAMPLE_GATHER4_C 16
1059 #define GEN7_SAMPLER_MESSAGE_SAMPLE_GATHER4_PO 17
1060 #define GEN7_SAMPLER_MESSAGE_SAMPLE_GATHER4_PO_C 18
1061 #define HSW_SAMPLER_MESSAGE_SAMPLE_DERIV_COMPARE 20
1062 #define GEN9_SAMPLER_MESSAGE_SAMPLE_LZ 24
1063 #define GEN9_SAMPLER_MESSAGE_SAMPLE_C_LZ 25
1064 #define GEN9_SAMPLER_MESSAGE_SAMPLE_LD_LZ 26
1065 #define GEN9_SAMPLER_MESSAGE_SAMPLE_LD2DMS_W 28
1066 #define GEN7_SAMPLER_MESSAGE_SAMPLE_LD_MCS 29
1067 #define GEN7_SAMPLER_MESSAGE_SAMPLE_LD2DMS 30
1068 #define GEN7_SAMPLER_MESSAGE_SAMPLE_LD2DSS 31
1069
1070 /* for GEN5 only */
1071 #define BRW_SAMPLER_SIMD_MODE_SIMD4X2 0
1072 #define BRW_SAMPLER_SIMD_MODE_SIMD8 1
1073 #define BRW_SAMPLER_SIMD_MODE_SIMD16 2
1074 #define BRW_SAMPLER_SIMD_MODE_SIMD32_64 3
1075
1076 /* GEN9 changes SIMD mode 0 to mean SIMD8D, but lets us get the SIMD4x2
1077 * behavior by setting bit 22 of dword 2 in the message header. */
1078 #define GEN9_SAMPLER_SIMD_MODE_SIMD8D 0
1079 #define GEN9_SAMPLER_SIMD_MODE_EXTENSION_SIMD4X2 (1 << 22)
1080
1081 #define BRW_DATAPORT_OWORD_BLOCK_1_OWORDLOW 0
1082 #define BRW_DATAPORT_OWORD_BLOCK_1_OWORDHIGH 1
1083 #define BRW_DATAPORT_OWORD_BLOCK_2_OWORDS 2
1084 #define BRW_DATAPORT_OWORD_BLOCK_4_OWORDS 3
1085 #define BRW_DATAPORT_OWORD_BLOCK_8_OWORDS 4
1086 #define BRW_DATAPORT_OWORD_BLOCK_DWORDS(n) \
1087 ((n) == 4 ? BRW_DATAPORT_OWORD_BLOCK_1_OWORDLOW : \
1088 (n) == 8 ? BRW_DATAPORT_OWORD_BLOCK_2_OWORDS : \
1089 (n) == 16 ? BRW_DATAPORT_OWORD_BLOCK_4_OWORDS : \
1090 (n) == 32 ? BRW_DATAPORT_OWORD_BLOCK_8_OWORDS : \
1091 (abort(), ~0))
1092
1093 #define BRW_DATAPORT_OWORD_DUAL_BLOCK_1OWORD 0
1094 #define BRW_DATAPORT_OWORD_DUAL_BLOCK_4OWORDS 2
1095
1096 #define BRW_DATAPORT_DWORD_SCATTERED_BLOCK_8DWORDS 2
1097 #define BRW_DATAPORT_DWORD_SCATTERED_BLOCK_16DWORDS 3
1098
1099 /* This one stays the same across generations. */
1100 #define BRW_DATAPORT_READ_MESSAGE_OWORD_BLOCK_READ 0
1101 /* GEN4 */
1102 #define BRW_DATAPORT_READ_MESSAGE_OWORD_DUAL_BLOCK_READ 1
1103 #define BRW_DATAPORT_READ_MESSAGE_MEDIA_BLOCK_READ 2
1104 #define BRW_DATAPORT_READ_MESSAGE_DWORD_SCATTERED_READ 3
1105 /* G45, GEN5 */
1106 #define G45_DATAPORT_READ_MESSAGE_RENDER_UNORM_READ 1
1107 #define G45_DATAPORT_READ_MESSAGE_OWORD_DUAL_BLOCK_READ 2
1108 #define G45_DATAPORT_READ_MESSAGE_AVC_LOOP_FILTER_READ 3
1109 #define G45_DATAPORT_READ_MESSAGE_MEDIA_BLOCK_READ 4
1110 #define G45_DATAPORT_READ_MESSAGE_DWORD_SCATTERED_READ 6
1111 /* GEN6 */
1112 #define GEN6_DATAPORT_READ_MESSAGE_RENDER_UNORM_READ 1
1113 #define GEN6_DATAPORT_READ_MESSAGE_OWORD_DUAL_BLOCK_READ 2
1114 #define GEN6_DATAPORT_READ_MESSAGE_MEDIA_BLOCK_READ 4
1115 #define GEN6_DATAPORT_READ_MESSAGE_OWORD_UNALIGN_BLOCK_READ 5
1116 #define GEN6_DATAPORT_READ_MESSAGE_DWORD_SCATTERED_READ 6
1117
1118 #define BRW_DATAPORT_READ_TARGET_DATA_CACHE 0
1119 #define BRW_DATAPORT_READ_TARGET_RENDER_CACHE 1
1120 #define BRW_DATAPORT_READ_TARGET_SAMPLER_CACHE 2
1121
1122 #define BRW_DATAPORT_RENDER_TARGET_WRITE_SIMD16_SINGLE_SOURCE 0
1123 #define BRW_DATAPORT_RENDER_TARGET_WRITE_SIMD16_SINGLE_SOURCE_REPLICATED 1
1124 #define BRW_DATAPORT_RENDER_TARGET_WRITE_SIMD8_DUAL_SOURCE_SUBSPAN01 2
1125 #define BRW_DATAPORT_RENDER_TARGET_WRITE_SIMD8_DUAL_SOURCE_SUBSPAN23 3
1126 #define BRW_DATAPORT_RENDER_TARGET_WRITE_SIMD8_SINGLE_SOURCE_SUBSPAN01 4
1127
1128 #define BRW_DATAPORT_WRITE_MESSAGE_OWORD_BLOCK_WRITE 0
1129 #define BRW_DATAPORT_WRITE_MESSAGE_OWORD_DUAL_BLOCK_WRITE 1
1130 #define BRW_DATAPORT_WRITE_MESSAGE_MEDIA_BLOCK_WRITE 2
1131 #define BRW_DATAPORT_WRITE_MESSAGE_DWORD_SCATTERED_WRITE 3
1132 #define BRW_DATAPORT_WRITE_MESSAGE_RENDER_TARGET_WRITE 4
1133 #define BRW_DATAPORT_WRITE_MESSAGE_STREAMED_VERTEX_BUFFER_WRITE 5
1134 #define BRW_DATAPORT_WRITE_MESSAGE_FLUSH_RENDER_CACHE 7
1135
1136 /* GEN6 */
1137 #define GEN6_DATAPORT_WRITE_MESSAGE_DWORD_ATOMIC_WRITE 7
1138 #define GEN6_DATAPORT_WRITE_MESSAGE_OWORD_BLOCK_WRITE 8
1139 #define GEN6_DATAPORT_WRITE_MESSAGE_OWORD_DUAL_BLOCK_WRITE 9
1140 #define GEN6_DATAPORT_WRITE_MESSAGE_MEDIA_BLOCK_WRITE 10
1141 #define GEN6_DATAPORT_WRITE_MESSAGE_DWORD_SCATTERED_WRITE 11
1142 #define GEN6_DATAPORT_WRITE_MESSAGE_RENDER_TARGET_WRITE 12
1143 #define GEN6_DATAPORT_WRITE_MESSAGE_STREAMED_VB_WRITE 13
1144 #define GEN6_DATAPORT_WRITE_MESSAGE_RENDER_TARGET_UNORM_WRITE 14
1145
1146 /* GEN7 */
1147 #define GEN7_DATAPORT_RC_MEDIA_BLOCK_READ 4
1148 #define GEN7_DATAPORT_RC_TYPED_SURFACE_READ 5
1149 #define GEN7_DATAPORT_RC_TYPED_ATOMIC_OP 6
1150 #define GEN7_DATAPORT_RC_MEMORY_FENCE 7
1151 #define GEN7_DATAPORT_RC_MEDIA_BLOCK_WRITE 10
1152 #define GEN7_DATAPORT_RC_RENDER_TARGET_WRITE 12
1153 #define GEN7_DATAPORT_RC_TYPED_SURFACE_WRITE 13
1154 #define GEN7_DATAPORT_DC_OWORD_BLOCK_READ 0
1155 #define GEN7_DATAPORT_DC_UNALIGNED_OWORD_BLOCK_READ 1
1156 #define GEN7_DATAPORT_DC_OWORD_DUAL_BLOCK_READ 2
1157 #define GEN7_DATAPORT_DC_DWORD_SCATTERED_READ 3
1158 #define GEN7_DATAPORT_DC_BYTE_SCATTERED_READ 4
1159 #define GEN7_DATAPORT_DC_UNTYPED_SURFACE_READ 5
1160 #define GEN7_DATAPORT_DC_UNTYPED_ATOMIC_OP 6
1161 #define GEN7_DATAPORT_DC_MEMORY_FENCE 7
1162 #define GEN7_DATAPORT_DC_OWORD_BLOCK_WRITE 8
1163 #define GEN7_DATAPORT_DC_OWORD_DUAL_BLOCK_WRITE 10
1164 #define GEN7_DATAPORT_DC_DWORD_SCATTERED_WRITE 11
1165 #define GEN7_DATAPORT_DC_BYTE_SCATTERED_WRITE 12
1166 #define GEN7_DATAPORT_DC_UNTYPED_SURFACE_WRITE 13
1167
1168 #define GEN7_DATAPORT_SCRATCH_READ ((1 << 18) | \
1169 (0 << 17))
1170 #define GEN7_DATAPORT_SCRATCH_WRITE ((1 << 18) | \
1171 (1 << 17))
1172 #define GEN7_DATAPORT_SCRATCH_NUM_REGS_SHIFT 12
1173
1174 #define GEN7_PIXEL_INTERPOLATOR_LOC_SHARED_OFFSET 0
1175 #define GEN7_PIXEL_INTERPOLATOR_LOC_SAMPLE 1
1176 #define GEN7_PIXEL_INTERPOLATOR_LOC_CENTROID 2
1177 #define GEN7_PIXEL_INTERPOLATOR_LOC_PER_SLOT_OFFSET 3
1178
1179 /* HSW */
1180 #define HSW_DATAPORT_DC_PORT0_OWORD_BLOCK_READ 0
1181 #define HSW_DATAPORT_DC_PORT0_UNALIGNED_OWORD_BLOCK_READ 1
1182 #define HSW_DATAPORT_DC_PORT0_OWORD_DUAL_BLOCK_READ 2
1183 #define HSW_DATAPORT_DC_PORT0_DWORD_SCATTERED_READ 3
1184 #define HSW_DATAPORT_DC_PORT0_BYTE_SCATTERED_READ 4
1185 #define HSW_DATAPORT_DC_PORT0_MEMORY_FENCE 7
1186 #define HSW_DATAPORT_DC_PORT0_OWORD_BLOCK_WRITE 8
1187 #define HSW_DATAPORT_DC_PORT0_OWORD_DUAL_BLOCK_WRITE 10
1188 #define HSW_DATAPORT_DC_PORT0_DWORD_SCATTERED_WRITE 11
1189 #define HSW_DATAPORT_DC_PORT0_BYTE_SCATTERED_WRITE 12
1190
1191 #define HSW_DATAPORT_DC_PORT1_UNTYPED_SURFACE_READ 1
1192 #define HSW_DATAPORT_DC_PORT1_UNTYPED_ATOMIC_OP 2
1193 #define HSW_DATAPORT_DC_PORT1_UNTYPED_ATOMIC_OP_SIMD4X2 3
1194 #define HSW_DATAPORT_DC_PORT1_MEDIA_BLOCK_READ 4
1195 #define HSW_DATAPORT_DC_PORT1_TYPED_SURFACE_READ 5
1196 #define HSW_DATAPORT_DC_PORT1_TYPED_ATOMIC_OP 6
1197 #define HSW_DATAPORT_DC_PORT1_TYPED_ATOMIC_OP_SIMD4X2 7
1198 #define HSW_DATAPORT_DC_PORT1_UNTYPED_SURFACE_WRITE 9
1199 #define HSW_DATAPORT_DC_PORT1_MEDIA_BLOCK_WRITE 10
1200 #define HSW_DATAPORT_DC_PORT1_ATOMIC_COUNTER_OP 11
1201 #define HSW_DATAPORT_DC_PORT1_ATOMIC_COUNTER_OP_SIMD4X2 12
1202 #define HSW_DATAPORT_DC_PORT1_TYPED_SURFACE_WRITE 13
1203 #define GEN9_DATAPORT_DC_PORT1_A64_SCATTERED_READ 0x10
1204 #define GEN8_DATAPORT_DC_PORT1_A64_UNTYPED_SURFACE_READ 0x11
1205 #define GEN8_DATAPORT_DC_PORT1_A64_UNTYPED_ATOMIC_OP 0x12
1206 #define GEN8_DATAPORT_DC_PORT1_A64_UNTYPED_SURFACE_WRITE 0x19
1207 #define GEN8_DATAPORT_DC_PORT1_A64_SCATTERED_WRITE 0x1a
1208 #define GEN9_DATAPORT_DC_PORT1_UNTYPED_ATOMIC_FLOAT_OP 0x1b
1209 #define GEN9_DATAPORT_DC_PORT1_A64_UNTYPED_ATOMIC_FLOAT_OP 0x1d
1210
1211 /* GEN9 */
1212 #define GEN9_DATAPORT_RC_RENDER_TARGET_WRITE 12
1213 #define GEN9_DATAPORT_RC_RENDER_TARGET_READ 13
1214
1215 /* A64 scattered message subtype */
1216 #define GEN8_A64_SCATTERED_SUBTYPE_BYTE 0
1217 #define GEN8_A64_SCATTERED_SUBTYPE_DWORD 1
1218 #define GEN8_A64_SCATTERED_SUBTYPE_QWORD 2
1219 #define GEN8_A64_SCATTERED_SUBTYPE_HWORD 3
1220
1221 /* Dataport special binding table indices: */
1222 #define BRW_BTI_STATELESS 255
1223 #define GEN7_BTI_SLM 254
1224 /* Note that on Gen8+ BTI 255 was redefined to be IA-coherent according to the
1225 * hardware spec, however because the DRM sets bit 4 of HDC_CHICKEN0 on BDW,
1226 * CHV and at least some pre-production steppings of SKL due to
1227 * WaForceEnableNonCoherent, HDC memory access may have been overridden by the
1228 * kernel to be non-coherent (matching the behavior of the same BTI on
1229 * pre-Gen8 hardware) and BTI 255 may actually be an alias for BTI 253.
1230 */
1231 #define GEN8_BTI_STATELESS_IA_COHERENT 255
1232 #define GEN8_BTI_STATELESS_NON_COHERENT 253
1233 #define GEN9_BTI_BINDLESS 252
1234
1235 /* Dataport atomic operations for Untyped Atomic Integer Operation message
1236 * (and others).
1237 */
1238 #define BRW_AOP_AND 1
1239 #define BRW_AOP_OR 2
1240 #define BRW_AOP_XOR 3
1241 #define BRW_AOP_MOV 4
1242 #define BRW_AOP_INC 5
1243 #define BRW_AOP_DEC 6
1244 #define BRW_AOP_ADD 7
1245 #define BRW_AOP_SUB 8
1246 #define BRW_AOP_REVSUB 9
1247 #define BRW_AOP_IMAX 10
1248 #define BRW_AOP_IMIN 11
1249 #define BRW_AOP_UMAX 12
1250 #define BRW_AOP_UMIN 13
1251 #define BRW_AOP_CMPWR 14
1252 #define BRW_AOP_PREDEC 15
1253
1254 /* Dataport atomic operations for Untyped Atomic Float Operation message. */
1255 #define BRW_AOP_FMAX 1
1256 #define BRW_AOP_FMIN 2
1257 #define BRW_AOP_FCMPWR 3
1258
1259 #define BRW_MATH_FUNCTION_INV 1
1260 #define BRW_MATH_FUNCTION_LOG 2
1261 #define BRW_MATH_FUNCTION_EXP 3
1262 #define BRW_MATH_FUNCTION_SQRT 4
1263 #define BRW_MATH_FUNCTION_RSQ 5
1264 #define BRW_MATH_FUNCTION_SIN 6
1265 #define BRW_MATH_FUNCTION_COS 7
1266 #define BRW_MATH_FUNCTION_SINCOS 8 /* gen4, gen5 */
1267 #define BRW_MATH_FUNCTION_FDIV 9 /* gen6+ */
1268 #define BRW_MATH_FUNCTION_POW 10
1269 #define BRW_MATH_FUNCTION_INT_DIV_QUOTIENT_AND_REMAINDER 11
1270 #define BRW_MATH_FUNCTION_INT_DIV_QUOTIENT 12
1271 #define BRW_MATH_FUNCTION_INT_DIV_REMAINDER 13
1272 #define GEN8_MATH_FUNCTION_INVM 14
1273 #define GEN8_MATH_FUNCTION_RSQRTM 15
1274
1275 #define BRW_MATH_INTEGER_UNSIGNED 0
1276 #define BRW_MATH_INTEGER_SIGNED 1
1277
1278 #define BRW_MATH_PRECISION_FULL 0
1279 #define BRW_MATH_PRECISION_PARTIAL 1
1280
1281 #define BRW_MATH_SATURATE_NONE 0
1282 #define BRW_MATH_SATURATE_SATURATE 1
1283
1284 #define BRW_MATH_DATA_VECTOR 0
1285 #define BRW_MATH_DATA_SCALAR 1
1286
1287 #define BRW_URB_OPCODE_WRITE_HWORD 0
1288 #define BRW_URB_OPCODE_WRITE_OWORD 1
1289 #define BRW_URB_OPCODE_READ_HWORD 2
1290 #define BRW_URB_OPCODE_READ_OWORD 3
1291 #define GEN7_URB_OPCODE_ATOMIC_MOV 4
1292 #define GEN7_URB_OPCODE_ATOMIC_INC 5
1293 #define GEN8_URB_OPCODE_ATOMIC_ADD 6
1294 #define GEN8_URB_OPCODE_SIMD8_WRITE 7
1295 #define GEN8_URB_OPCODE_SIMD8_READ 8
1296
1297 #define BRW_URB_SWIZZLE_NONE 0
1298 #define BRW_URB_SWIZZLE_INTERLEAVE 1
1299 #define BRW_URB_SWIZZLE_TRANSPOSE 2
1300
1301 #define BRW_SCRATCH_SPACE_SIZE_1K 0
1302 #define BRW_SCRATCH_SPACE_SIZE_2K 1
1303 #define BRW_SCRATCH_SPACE_SIZE_4K 2
1304 #define BRW_SCRATCH_SPACE_SIZE_8K 3
1305 #define BRW_SCRATCH_SPACE_SIZE_16K 4
1306 #define BRW_SCRATCH_SPACE_SIZE_32K 5
1307 #define BRW_SCRATCH_SPACE_SIZE_64K 6
1308 #define BRW_SCRATCH_SPACE_SIZE_128K 7
1309 #define BRW_SCRATCH_SPACE_SIZE_256K 8
1310 #define BRW_SCRATCH_SPACE_SIZE_512K 9
1311 #define BRW_SCRATCH_SPACE_SIZE_1M 10
1312 #define BRW_SCRATCH_SPACE_SIZE_2M 11
1313
1314 #define BRW_MESSAGE_GATEWAY_SFID_OPEN_GATEWAY 0
1315 #define BRW_MESSAGE_GATEWAY_SFID_CLOSE_GATEWAY 1
1316 #define BRW_MESSAGE_GATEWAY_SFID_FORWARD_MSG 2
1317 #define BRW_MESSAGE_GATEWAY_SFID_GET_TIMESTAMP 3
1318 #define BRW_MESSAGE_GATEWAY_SFID_BARRIER_MSG 4
1319 #define BRW_MESSAGE_GATEWAY_SFID_UPDATE_GATEWAY_STATE 5
1320 #define BRW_MESSAGE_GATEWAY_SFID_MMIO_READ_WRITE 6
1321
1322
1323 /* Gen7 "GS URB Entry Allocation Size" is a U9-1 field, so the maximum gs_size
1324 * is 2^9, or 512. It's counted in multiples of 64 bytes.
1325 *
1326 * Identical for VS, DS, and HS.
1327 */
1328 #define GEN7_MAX_GS_URB_ENTRY_SIZE_BYTES (512*64)
1329 #define GEN7_MAX_DS_URB_ENTRY_SIZE_BYTES (512*64)
1330 #define GEN7_MAX_HS_URB_ENTRY_SIZE_BYTES (512*64)
1331 #define GEN7_MAX_VS_URB_ENTRY_SIZE_BYTES (512*64)
1332
1333 /* Gen6 "GS URB Entry Allocation Size" is defined as a number of 1024-bit
1334 * (128 bytes) URB rows and the maximum allowed value is 5 rows.
1335 */
1336 #define GEN6_MAX_GS_URB_ENTRY_SIZE_BYTES (5*128)
1337
1338 /* GS Thread Payload
1339 */
1340
1341 /* 3DSTATE_GS "Output Vertex Size" has an effective maximum of 62. It's
1342 * counted in multiples of 16 bytes.
1343 */
1344 #define GEN7_MAX_GS_OUTPUT_VERTEX_SIZE_BYTES (62*16)
1345
1346
1347 /* R0 */
1348 # define GEN7_GS_PAYLOAD_INSTANCE_ID_SHIFT 27
1349
1350 /* CR0.0[5:4] Floating-Point Rounding Modes
1351 * Skylake PRM, Volume 7 Part 1, "Control Register", page 756
1352 */
1353
1354 #define BRW_CR0_RND_MODE_MASK 0x30
1355 #define BRW_CR0_RND_MODE_SHIFT 4
1356
1357 enum PACKED brw_rnd_mode {
1358 BRW_RND_MODE_RTNE = 0, /* Round to Nearest or Even */
1359 BRW_RND_MODE_RU = 1, /* Round Up, toward +inf */
1360 BRW_RND_MODE_RD = 2, /* Round Down, toward -inf */
1361 BRW_RND_MODE_RTZ = 3, /* Round Toward Zero */
1362 BRW_RND_MODE_UNSPECIFIED, /* Unspecified rounding mode */
1363 };
1364
1365 /* MDC_DS - Data Size Message Descriptor Control Field
1366 * Skylake PRM, Volume 2d, page 129
1367 *
1368 * Specifies the number of Bytes to be read or written per Dword used at
1369 * byte_scattered read/write and byte_scaled read/write messages.
1370 */
1371 #define GEN7_BYTE_SCATTERED_DATA_ELEMENT_BYTE 0
1372 #define GEN7_BYTE_SCATTERED_DATA_ELEMENT_WORD 1
1373 #define GEN7_BYTE_SCATTERED_DATA_ELEMENT_DWORD 2
1374
1375 #endif /* BRW_EU_DEFINES_H */