2 * Copyright 2010 Jerome Glisse <glisse@freedesktop.org>
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 * on the rights to use, copy, modify, merge, publish, distribute, sub
8 * license, and/or sell copies of the Software, and to permit persons to whom
9 * the 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 NON-INFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
19 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
20 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
21 * USE OR OTHER DEALINGS IN THE SOFTWARE.
23 #include "pipe/p_shader_tokens.h"
24 #include "tgsi/tgsi_info.h"
25 #include "tgsi/tgsi_parse.h"
26 #include "tgsi/tgsi_scan.h"
27 #include "tgsi/tgsi_dump.h"
28 #include "util/u_format.h"
29 #include "r600_pipe.h"
32 #include "r600_formats.h"
33 #include "r600_opcodes.h"
40 Why CAYMAN got loops for lots of instructions is explained here.
42 -These 8xx t-slot only ops are implemented in all vector slots.
43 MUL_LIT, FLT_TO_UINT, INT_TO_FLT, UINT_TO_FLT
44 These 8xx t-slot only opcodes become vector ops, with all four
45 slots expecting the arguments on sources a and b. Result is
46 broadcast to all channels.
47 MULLO_INT, MULHI_INT, MULLO_UINT, MULHI_UINT
48 These 8xx t-slot only opcodes become vector ops in the z, y, and
50 EXP_IEEE, LOG_IEEE/CLAMPED, RECIP_IEEE/CLAMPED/FF/INT/UINT/_64/CLAMPED_64
51 RECIPSQRT_IEEE/CLAMPED/FF/_64/CLAMPED_64
54 The w slot may have an independent co-issued operation, or if the
55 result is required to be in the w slot, the opcode above may be
56 issued in the w slot as well.
57 The compiler must issue the source argument to slots z, y, and x
60 static int r600_pipe_shader(struct pipe_context
*ctx
, struct r600_pipe_shader
*shader
)
62 struct r600_pipe_context
*rctx
= (struct r600_pipe_context
*)ctx
;
63 struct r600_shader
*rshader
= &shader
->shader
;
68 if (shader
->bo
== NULL
) {
69 shader
->bo
= (struct r600_resource
*)
70 pipe_buffer_create(ctx
->screen
, PIPE_BIND_CUSTOM
, PIPE_USAGE_IMMUTABLE
, rshader
->bc
.ndw
* 4);
71 if (shader
->bo
== NULL
) {
74 ptr
= (uint32_t*)rctx
->ws
->buffer_map(shader
->bo
->buf
, rctx
->ctx
.cs
, PIPE_TRANSFER_WRITE
);
75 if (R600_BIG_ENDIAN
) {
76 for (i
= 0; i
< rshader
->bc
.ndw
; ++i
) {
77 ptr
[i
] = bswap_32(rshader
->bc
.bytecode
[i
]);
80 memcpy(ptr
, rshader
->bc
.bytecode
, rshader
->bc
.ndw
* sizeof(*ptr
));
82 rctx
->ws
->buffer_unmap(shader
->bo
->buf
);
85 switch (rshader
->processor_type
) {
86 case TGSI_PROCESSOR_VERTEX
:
87 if (rctx
->chip_class
>= EVERGREEN
) {
88 evergreen_pipe_shader_vs(ctx
, shader
);
90 r600_pipe_shader_vs(ctx
, shader
);
93 case TGSI_PROCESSOR_FRAGMENT
:
94 if (rctx
->chip_class
>= EVERGREEN
) {
95 evergreen_pipe_shader_ps(ctx
, shader
);
97 r600_pipe_shader_ps(ctx
, shader
);
106 static int r600_shader_from_tgsi(struct r600_pipe_context
* rctx
, struct r600_pipe_shader
*pipeshader
);
108 int r600_pipe_shader_create(struct pipe_context
*ctx
, struct r600_pipe_shader
*shader
)
110 static int dump_shaders
= -1;
111 struct r600_pipe_context
*rctx
= (struct r600_pipe_context
*)ctx
;
114 /* Would like some magic "get_bool_option_once" routine.
116 if (dump_shaders
== -1)
117 dump_shaders
= debug_get_bool_option("R600_DUMP_SHADERS", FALSE
);
120 fprintf(stderr
, "--------------------------------------------------------------\n");
121 tgsi_dump(shader
->tokens
, 0);
123 if (shader
->so
.num_outputs
) {
125 fprintf(stderr
, "STREAMOUT\n");
126 for (i
= 0; i
< shader
->so
.num_outputs
; i
++) {
127 unsigned mask
= ((1 << shader
->so
.output
[i
].num_components
) - 1) <<
128 shader
->so
.output
[i
].start_component
;
129 fprintf(stderr
, " %i: MEM_STREAM0_BUF%i OUT[%i].%s%s%s%s\n", i
,
130 shader
->so
.output
[i
].output_buffer
, shader
->so
.output
[i
].register_index
,
131 mask
& 1 ? "x" : "_",
132 (mask
>> 1) & 1 ? "y" : "_",
133 (mask
>> 2) & 1 ? "z" : "_",
134 (mask
>> 3) & 1 ? "w" : "_");
138 r
= r600_shader_from_tgsi(rctx
, shader
);
140 R600_ERR("translation from TGSI failed !\n");
143 r
= r600_bytecode_build(&shader
->shader
.bc
);
145 R600_ERR("building bytecode failed !\n");
149 r600_bytecode_dump(&shader
->shader
.bc
);
150 fprintf(stderr
, "______________________________________________________________\n");
152 return r600_pipe_shader(ctx
, shader
);
155 void r600_pipe_shader_destroy(struct pipe_context
*ctx
, struct r600_pipe_shader
*shader
)
157 pipe_resource_reference((struct pipe_resource
**)&shader
->bo
, NULL
);
158 r600_bytecode_clear(&shader
->shader
.bc
);
160 memset(&shader
->shader
,0,sizeof(struct r600_shader
));
164 * tgsi -> r600 shader
166 struct r600_shader_tgsi_instruction
;
168 struct r600_shader_src
{
177 struct r600_shader_ctx
{
178 struct tgsi_shader_info info
;
179 struct tgsi_parse_context parse
;
180 const struct tgsi_token
*tokens
;
182 unsigned file_offset
[TGSI_FILE_COUNT
];
184 struct r600_shader_tgsi_instruction
*inst_info
;
185 struct r600_bytecode
*bc
;
186 struct r600_shader
*shader
;
187 struct r600_shader_src src
[4];
190 u32 max_driver_temp_used
;
191 /* needed for evergreen interpolation */
192 boolean input_centroid
;
193 boolean input_linear
;
194 boolean input_perspective
;
200 struct r600_shader_tgsi_instruction
{
201 unsigned tgsi_opcode
;
203 unsigned r600_opcode
;
204 int (*process
)(struct r600_shader_ctx
*ctx
);
207 static struct r600_shader_tgsi_instruction r600_shader_tgsi_instruction
[], eg_shader_tgsi_instruction
[], cm_shader_tgsi_instruction
[];
208 static int tgsi_helper_tempx_replicate(struct r600_shader_ctx
*ctx
);
210 static int tgsi_is_supported(struct r600_shader_ctx
*ctx
)
212 struct tgsi_full_instruction
*i
= &ctx
->parse
.FullToken
.FullInstruction
;
215 if (i
->Instruction
.NumDstRegs
> 1) {
216 R600_ERR("too many dst (%d)\n", i
->Instruction
.NumDstRegs
);
219 if (i
->Instruction
.Predicate
) {
220 R600_ERR("predicate unsupported\n");
224 if (i
->Instruction
.Label
) {
225 R600_ERR("label unsupported\n");
229 for (j
= 0; j
< i
->Instruction
.NumSrcRegs
; j
++) {
230 if (i
->Src
[j
].Register
.Dimension
) {
231 R600_ERR("unsupported src %d (dimension %d)\n", j
,
232 i
->Src
[j
].Register
.Dimension
);
236 for (j
= 0; j
< i
->Instruction
.NumDstRegs
; j
++) {
237 if (i
->Dst
[j
].Register
.Dimension
) {
238 R600_ERR("unsupported dst (dimension)\n");
245 static int evergreen_interp_alu(struct r600_shader_ctx
*ctx
, int input
)
248 struct r600_bytecode_alu alu
;
249 int gpr
= 0, base_chan
= 0;
252 if (ctx
->shader
->input
[input
].interpolate
== TGSI_INTERPOLATE_PERSPECTIVE
) {
254 if (ctx
->shader
->input
[input
].centroid
)
256 } else if (ctx
->shader
->input
[input
].interpolate
== TGSI_INTERPOLATE_LINEAR
) {
258 /* if we have perspective add one */
259 if (ctx
->input_perspective
) {
261 /* if we have perspective centroid */
262 if (ctx
->input_centroid
)
265 if (ctx
->shader
->input
[input
].centroid
)
269 /* work out gpr and base_chan from index */
271 base_chan
= (2 * (ij_index
% 2)) + 1;
273 for (i
= 0; i
< 8; i
++) {
274 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
277 alu
.inst
= EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INTERP_ZW
;
279 alu
.inst
= EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INTERP_XY
;
281 if ((i
> 1) && (i
< 6)) {
282 alu
.dst
.sel
= ctx
->shader
->input
[input
].gpr
;
286 alu
.dst
.chan
= i
% 4;
288 alu
.src
[0].sel
= gpr
;
289 alu
.src
[0].chan
= (base_chan
- (i
% 2));
291 alu
.src
[1].sel
= V_SQ_ALU_SRC_PARAM_BASE
+ ctx
->shader
->input
[input
].lds_pos
;
293 alu
.bank_swizzle_force
= SQ_ALU_VEC_210
;
296 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
303 static int evergreen_interp_flat(struct r600_shader_ctx
*ctx
, int input
)
306 struct r600_bytecode_alu alu
;
308 for (i
= 0; i
< 4; i
++) {
309 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
311 alu
.inst
= EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INTERP_LOAD_P0
;
313 alu
.dst
.sel
= ctx
->shader
->input
[input
].gpr
;
318 alu
.src
[0].sel
= V_SQ_ALU_SRC_PARAM_BASE
+ ctx
->shader
->input
[input
].lds_pos
;
323 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
331 * Special export handling in shaders
333 * shader export ARRAY_BASE for EXPORT_POS:
336 * 62, 63 are clip distance vectors
338 * The use of the values exported in 61-63 are controlled by PA_CL_VS_OUT_CNTL:
339 * VS_OUT_MISC_VEC_ENA - enables the use of all fields in export 61
340 * USE_VTX_POINT_SIZE - point size in the X channel of export 61
341 * USE_VTX_EDGE_FLAG - edge flag in the Y channel of export 61
342 * USE_VTX_RENDER_TARGET_INDX - render target index in the Z channel of export 61
343 * USE_VTX_VIEWPORT_INDX - viewport index in the W channel of export 61
344 * USE_VTX_KILL_FLAG - kill flag in the Z channel of export 61 (mutually
345 * exclusive from render target index)
346 * VS_OUT_CCDIST0_VEC_ENA/VS_OUT_CCDIST1_VEC_ENA - enable clip distance vectors
349 * shader export ARRAY_BASE for EXPORT_PIXEL:
351 * 61 computed Z vector
353 * The use of the values exported in the computed Z vector are controlled
354 * by DB_SHADER_CONTROL:
355 * Z_EXPORT_ENABLE - Z as a float in RED
356 * STENCIL_REF_EXPORT_ENABLE - stencil ref as int in GREEN
357 * COVERAGE_TO_MASK_ENABLE - alpha to mask in ALPHA
358 * MASK_EXPORT_ENABLE - pixel sample mask in BLUE
359 * DB_SOURCE_FORMAT - export control restrictions
364 /* Map name/sid pair from tgsi to the 8-bit semantic index for SPI setup */
365 static int r600_spi_sid(struct r600_shader_io
* io
)
367 int index
, name
= io
->name
;
369 /* These params are handled differently, they don't need
370 * semantic indices, so we'll use 0 for them.
372 if (name
== TGSI_SEMANTIC_POSITION
||
373 name
== TGSI_SEMANTIC_PSIZE
||
374 name
== TGSI_SEMANTIC_FACE
)
377 if (name
== TGSI_SEMANTIC_GENERIC
) {
378 /* For generic params simply use sid from tgsi */
381 /* For non-generic params - pack name and sid into 8 bits */
382 index
= 0x80 | (name
<<3) | (io
->sid
);
385 /* Make sure that all really used indices have nonzero value, so
386 * we can just compare it to 0 later instead of comparing the name
387 * with different values to detect special cases. */
394 /* turn input into interpolate on EG */
395 static int evergreen_interp_input(struct r600_shader_ctx
*ctx
, int index
)
399 if (ctx
->shader
->input
[index
].spi_sid
) {
400 ctx
->shader
->input
[index
].lds_pos
= ctx
->shader
->nlds
++;
401 if (ctx
->shader
->input
[index
].interpolate
> 0) {
402 r
= evergreen_interp_alu(ctx
, index
);
404 r
= evergreen_interp_flat(ctx
, index
);
410 static int select_twoside_color(struct r600_shader_ctx
*ctx
, int front
, int back
)
412 struct r600_bytecode_alu alu
;
414 int gpr_front
= ctx
->shader
->input
[front
].gpr
;
415 int gpr_back
= ctx
->shader
->input
[back
].gpr
;
417 for (i
= 0; i
< 4; i
++) {
418 memset(&alu
, 0, sizeof(alu
));
419 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP3_SQ_OP3_INST_CNDGT
);
422 alu
.dst
.sel
= gpr_front
;
423 alu
.src
[0].sel
= ctx
->face_gpr
;
424 alu
.src
[1].sel
= gpr_front
;
425 alu
.src
[2].sel
= gpr_back
;
432 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
439 static int tgsi_declaration(struct r600_shader_ctx
*ctx
)
441 struct tgsi_full_declaration
*d
= &ctx
->parse
.FullToken
.FullDeclaration
;
445 switch (d
->Declaration
.File
) {
446 case TGSI_FILE_INPUT
:
447 i
= ctx
->shader
->ninput
++;
448 ctx
->shader
->input
[i
].name
= d
->Semantic
.Name
;
449 ctx
->shader
->input
[i
].sid
= d
->Semantic
.Index
;
450 ctx
->shader
->input
[i
].spi_sid
= r600_spi_sid(&ctx
->shader
->input
[i
]);
451 ctx
->shader
->input
[i
].interpolate
= d
->Declaration
.Interpolate
;
452 ctx
->shader
->input
[i
].centroid
= d
->Declaration
.Centroid
;
453 ctx
->shader
->input
[i
].gpr
= ctx
->file_offset
[TGSI_FILE_INPUT
] + d
->Range
.First
;
454 if (ctx
->type
== TGSI_PROCESSOR_FRAGMENT
) {
455 if (ctx
->shader
->input
[i
].name
== TGSI_SEMANTIC_FACE
)
456 ctx
->face_gpr
= ctx
->shader
->input
[i
].gpr
;
457 else if (ctx
->shader
->input
[i
].name
== TGSI_SEMANTIC_COLOR
)
459 if (ctx
->bc
->chip_class
>= EVERGREEN
) {
460 r
= evergreen_interp_input(ctx
, i
);
466 case TGSI_FILE_OUTPUT
:
467 i
= ctx
->shader
->noutput
++;
468 ctx
->shader
->output
[i
].name
= d
->Semantic
.Name
;
469 ctx
->shader
->output
[i
].sid
= d
->Semantic
.Index
;
470 ctx
->shader
->output
[i
].spi_sid
= r600_spi_sid(&ctx
->shader
->output
[i
]);
471 ctx
->shader
->output
[i
].gpr
= ctx
->file_offset
[TGSI_FILE_OUTPUT
] + d
->Range
.First
;
472 ctx
->shader
->output
[i
].interpolate
= d
->Declaration
.Interpolate
;
473 ctx
->shader
->output
[i
].write_mask
= d
->Declaration
.UsageMask
;
474 if (ctx
->type
== TGSI_PROCESSOR_VERTEX
) {
475 switch (d
->Semantic
.Name
) {
476 case TGSI_SEMANTIC_CLIPDIST
:
477 ctx
->shader
->clip_dist_write
|= d
->Declaration
.UsageMask
<< (d
->Semantic
.Index
<< 2);
479 case TGSI_SEMANTIC_PSIZE
:
480 ctx
->shader
->vs_out_misc_write
= 1;
485 case TGSI_FILE_CONSTANT
:
486 case TGSI_FILE_TEMPORARY
:
487 case TGSI_FILE_SAMPLER
:
488 case TGSI_FILE_ADDRESS
:
491 case TGSI_FILE_SYSTEM_VALUE
:
492 if (d
->Semantic
.Name
== TGSI_SEMANTIC_INSTANCEID
) {
493 struct r600_bytecode_alu alu
;
494 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
496 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_INT_TO_FLT
);
505 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
508 } else if (d
->Semantic
.Name
== TGSI_SEMANTIC_VERTEXID
)
511 R600_ERR("unsupported file %d declaration\n", d
->Declaration
.File
);
517 static int r600_get_temp(struct r600_shader_ctx
*ctx
)
519 return ctx
->temp_reg
+ ctx
->max_driver_temp_used
++;
523 * for evergreen we need to scan the shader to find the number of GPRs we need to
524 * reserve for interpolation.
526 * we need to know if we are going to emit
527 * any centroid inputs
528 * if perspective and linear are required
530 static int evergreen_gpr_count(struct r600_shader_ctx
*ctx
)
535 ctx
->input_linear
= FALSE
;
536 ctx
->input_perspective
= FALSE
;
537 ctx
->input_centroid
= FALSE
;
538 ctx
->num_interp_gpr
= 1;
540 /* any centroid inputs */
541 for (i
= 0; i
< ctx
->info
.num_inputs
; i
++) {
542 /* skip position/face */
543 if (ctx
->info
.input_semantic_name
[i
] == TGSI_SEMANTIC_POSITION
||
544 ctx
->info
.input_semantic_name
[i
] == TGSI_SEMANTIC_FACE
)
546 if (ctx
->info
.input_interpolate
[i
] == TGSI_INTERPOLATE_LINEAR
)
547 ctx
->input_linear
= TRUE
;
548 if (ctx
->info
.input_interpolate
[i
] == TGSI_INTERPOLATE_PERSPECTIVE
)
549 ctx
->input_perspective
= TRUE
;
550 if (ctx
->info
.input_centroid
[i
])
551 ctx
->input_centroid
= TRUE
;
555 /* ignoring sample for now */
556 if (ctx
->input_perspective
)
558 if (ctx
->input_linear
)
560 if (ctx
->input_centroid
)
563 ctx
->num_interp_gpr
+= (num_baryc
+ 1) >> 1;
565 /* TODO PULL MODEL and LINE STIPPLE, FIXED PT POS */
566 return ctx
->num_interp_gpr
;
569 static void tgsi_src(struct r600_shader_ctx
*ctx
,
570 const struct tgsi_full_src_register
*tgsi_src
,
571 struct r600_shader_src
*r600_src
)
573 memset(r600_src
, 0, sizeof(*r600_src
));
574 r600_src
->swizzle
[0] = tgsi_src
->Register
.SwizzleX
;
575 r600_src
->swizzle
[1] = tgsi_src
->Register
.SwizzleY
;
576 r600_src
->swizzle
[2] = tgsi_src
->Register
.SwizzleZ
;
577 r600_src
->swizzle
[3] = tgsi_src
->Register
.SwizzleW
;
578 r600_src
->neg
= tgsi_src
->Register
.Negate
;
579 r600_src
->abs
= tgsi_src
->Register
.Absolute
;
581 if (tgsi_src
->Register
.File
== TGSI_FILE_IMMEDIATE
) {
583 if ((tgsi_src
->Register
.SwizzleX
== tgsi_src
->Register
.SwizzleY
) &&
584 (tgsi_src
->Register
.SwizzleX
== tgsi_src
->Register
.SwizzleZ
) &&
585 (tgsi_src
->Register
.SwizzleX
== tgsi_src
->Register
.SwizzleW
)) {
587 index
= tgsi_src
->Register
.Index
* 4 + tgsi_src
->Register
.SwizzleX
;
588 r600_bytecode_special_constants(ctx
->literals
[index
], &r600_src
->sel
, &r600_src
->neg
);
589 if (r600_src
->sel
!= V_SQ_ALU_SRC_LITERAL
)
592 index
= tgsi_src
->Register
.Index
;
593 r600_src
->sel
= V_SQ_ALU_SRC_LITERAL
;
594 memcpy(r600_src
->value
, ctx
->literals
+ index
* 4, sizeof(r600_src
->value
));
595 } else if (tgsi_src
->Register
.File
== TGSI_FILE_SYSTEM_VALUE
) {
596 if (ctx
->info
.system_value_semantic_name
[tgsi_src
->Register
.Index
] == TGSI_SEMANTIC_INSTANCEID
) {
597 r600_src
->swizzle
[0] = 3;
598 r600_src
->swizzle
[1] = 3;
599 r600_src
->swizzle
[2] = 3;
600 r600_src
->swizzle
[3] = 3;
602 } else if (ctx
->info
.system_value_semantic_name
[tgsi_src
->Register
.Index
] == TGSI_SEMANTIC_VERTEXID
) {
603 r600_src
->swizzle
[0] = 0;
604 r600_src
->swizzle
[1] = 0;
605 r600_src
->swizzle
[2] = 0;
606 r600_src
->swizzle
[3] = 0;
610 if (tgsi_src
->Register
.Indirect
)
611 r600_src
->rel
= V_SQ_REL_RELATIVE
;
612 r600_src
->sel
= tgsi_src
->Register
.Index
;
613 r600_src
->sel
+= ctx
->file_offset
[tgsi_src
->Register
.File
];
617 static int tgsi_fetch_rel_const(struct r600_shader_ctx
*ctx
, unsigned int offset
, unsigned int dst_reg
)
619 struct r600_bytecode_vtx vtx
;
624 struct r600_bytecode_alu alu
;
626 memset(&alu
, 0, sizeof(alu
));
628 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_ADD_INT
);
629 alu
.src
[0].sel
= ctx
->bc
->ar_reg
;
631 alu
.src
[1].sel
= V_SQ_ALU_SRC_LITERAL
;
632 alu
.src
[1].value
= offset
;
634 alu
.dst
.sel
= dst_reg
;
638 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
643 ar_reg
= ctx
->bc
->ar_reg
;
646 memset(&vtx
, 0, sizeof(vtx
));
647 vtx
.fetch_type
= 2; /* VTX_FETCH_NO_INDEX_OFFSET */
648 vtx
.src_gpr
= ar_reg
;
649 vtx
.mega_fetch_count
= 16;
650 vtx
.dst_gpr
= dst_reg
;
651 vtx
.dst_sel_x
= 0; /* SEL_X */
652 vtx
.dst_sel_y
= 1; /* SEL_Y */
653 vtx
.dst_sel_z
= 2; /* SEL_Z */
654 vtx
.dst_sel_w
= 3; /* SEL_W */
655 vtx
.data_format
= FMT_32_32_32_32_FLOAT
;
656 vtx
.num_format_all
= 2; /* NUM_FORMAT_SCALED */
657 vtx
.format_comp_all
= 1; /* FORMAT_COMP_SIGNED */
658 vtx
.srf_mode_all
= 1; /* SRF_MODE_NO_ZERO */
659 vtx
.endian
= r600_endian_swap(32);
661 if ((r
= r600_bytecode_add_vtx(ctx
->bc
, &vtx
)))
667 static int tgsi_split_constant(struct r600_shader_ctx
*ctx
)
669 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
670 struct r600_bytecode_alu alu
;
671 int i
, j
, k
, nconst
, r
;
673 for (i
= 0, nconst
= 0; i
< inst
->Instruction
.NumSrcRegs
; i
++) {
674 if (inst
->Src
[i
].Register
.File
== TGSI_FILE_CONSTANT
) {
677 tgsi_src(ctx
, &inst
->Src
[i
], &ctx
->src
[i
]);
679 for (i
= 0, j
= nconst
- 1; i
< inst
->Instruction
.NumSrcRegs
; i
++) {
680 if (inst
->Src
[i
].Register
.File
!= TGSI_FILE_CONSTANT
) {
684 if (ctx
->src
[i
].rel
) {
685 int treg
= r600_get_temp(ctx
);
686 if ((r
= tgsi_fetch_rel_const(ctx
, ctx
->src
[i
].sel
- 512, treg
)))
689 ctx
->src
[i
].sel
= treg
;
693 int treg
= r600_get_temp(ctx
);
694 for (k
= 0; k
< 4; k
++) {
695 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
696 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV
);
697 alu
.src
[0].sel
= ctx
->src
[i
].sel
;
699 alu
.src
[0].rel
= ctx
->src
[i
].rel
;
705 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
709 ctx
->src
[i
].sel
= treg
;
717 /* need to move any immediate into a temp - for trig functions which use literal for PI stuff */
718 static int tgsi_split_literal_constant(struct r600_shader_ctx
*ctx
)
720 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
721 struct r600_bytecode_alu alu
;
722 int i
, j
, k
, nliteral
, r
;
724 for (i
= 0, nliteral
= 0; i
< inst
->Instruction
.NumSrcRegs
; i
++) {
725 if (ctx
->src
[i
].sel
== V_SQ_ALU_SRC_LITERAL
) {
729 for (i
= 0, j
= nliteral
- 1; i
< inst
->Instruction
.NumSrcRegs
; i
++) {
730 if (j
> 0 && ctx
->src
[i
].sel
== V_SQ_ALU_SRC_LITERAL
) {
731 int treg
= r600_get_temp(ctx
);
732 for (k
= 0; k
< 4; k
++) {
733 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
734 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV
);
735 alu
.src
[0].sel
= ctx
->src
[i
].sel
;
737 alu
.src
[0].value
= ctx
->src
[i
].value
[k
];
743 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
747 ctx
->src
[i
].sel
= treg
;
754 static int process_twoside_color_inputs(struct r600_shader_ctx
*ctx
)
756 int i
, r
, count
= ctx
->shader
->ninput
;
758 /* additional inputs will be allocated right after the existing inputs,
759 * we won't need them after the color selection, so we don't need to
760 * reserve these gprs for the rest of the shader code and to adjust
761 * output offsets etc. */
762 int gpr
= ctx
->file_offset
[TGSI_FILE_INPUT
] +
763 ctx
->info
.file_max
[TGSI_FILE_INPUT
] + 1;
765 if (ctx
->face_gpr
== -1) {
766 i
= ctx
->shader
->ninput
++;
767 ctx
->shader
->input
[i
].name
= TGSI_SEMANTIC_FACE
;
768 ctx
->shader
->input
[i
].spi_sid
= 0;
769 ctx
->shader
->input
[i
].gpr
= gpr
++;
770 ctx
->face_gpr
= ctx
->shader
->input
[i
].gpr
;
773 for (i
= 0; i
< count
; i
++) {
774 if (ctx
->shader
->input
[i
].name
== TGSI_SEMANTIC_COLOR
) {
775 int ni
= ctx
->shader
->ninput
++;
776 memcpy(&ctx
->shader
->input
[ni
],&ctx
->shader
->input
[i
], sizeof(struct r600_shader_io
));
777 ctx
->shader
->input
[ni
].name
= TGSI_SEMANTIC_BCOLOR
;
778 ctx
->shader
->input
[ni
].spi_sid
= r600_spi_sid(&ctx
->shader
->input
[ni
]);
779 ctx
->shader
->input
[ni
].gpr
= gpr
++;
781 if (ctx
->bc
->chip_class
>= EVERGREEN
) {
782 r
= evergreen_interp_input(ctx
, ni
);
787 r
= select_twoside_color(ctx
, i
, ni
);
795 static int r600_shader_from_tgsi(struct r600_pipe_context
* rctx
, struct r600_pipe_shader
*pipeshader
)
797 struct r600_shader
*shader
= &pipeshader
->shader
;
798 struct tgsi_token
*tokens
= pipeshader
->tokens
;
799 struct pipe_stream_output_info so
= pipeshader
->so
;
800 struct tgsi_full_immediate
*immediate
;
801 struct tgsi_full_property
*property
;
802 struct r600_shader_ctx ctx
;
803 struct r600_bytecode_output output
[32];
804 unsigned output_done
, noutput
;
806 int i
, j
, r
= 0, pos0
;
808 ctx
.bc
= &shader
->bc
;
810 r600_bytecode_init(ctx
.bc
, rctx
->chip_class
, rctx
->family
);
812 tgsi_scan_shader(tokens
, &ctx
.info
);
813 tgsi_parse_init(&ctx
.parse
, tokens
);
814 ctx
.type
= ctx
.parse
.FullHeader
.Processor
.Processor
;
815 shader
->processor_type
= ctx
.type
;
816 ctx
.bc
->type
= shader
->processor_type
;
821 shader
->two_side
= (ctx
.type
== TGSI_PROCESSOR_FRAGMENT
) && rctx
->two_side
;
823 shader
->clamp_color
= (((ctx
.type
== TGSI_PROCESSOR_FRAGMENT
) && rctx
->clamp_fragment_color
) ||
824 ((ctx
.type
== TGSI_PROCESSOR_VERTEX
) && rctx
->clamp_vertex_color
));
826 shader
->nr_cbufs
= rctx
->nr_cbufs
;
828 /* register allocations */
829 /* Values [0,127] correspond to GPR[0..127].
830 * Values [128,159] correspond to constant buffer bank 0
831 * Values [160,191] correspond to constant buffer bank 1
832 * Values [256,511] correspond to cfile constants c[0..255]. (Gone on EG)
833 * Values [256,287] correspond to constant buffer bank 2 (EG)
834 * Values [288,319] correspond to constant buffer bank 3 (EG)
835 * Other special values are shown in the list below.
836 * 244 ALU_SRC_1_DBL_L: special constant 1.0 double-float, LSW. (RV670+)
837 * 245 ALU_SRC_1_DBL_M: special constant 1.0 double-float, MSW. (RV670+)
838 * 246 ALU_SRC_0_5_DBL_L: special constant 0.5 double-float, LSW. (RV670+)
839 * 247 ALU_SRC_0_5_DBL_M: special constant 0.5 double-float, MSW. (RV670+)
840 * 248 SQ_ALU_SRC_0: special constant 0.0.
841 * 249 SQ_ALU_SRC_1: special constant 1.0 float.
842 * 250 SQ_ALU_SRC_1_INT: special constant 1 integer.
843 * 251 SQ_ALU_SRC_M_1_INT: special constant -1 integer.
844 * 252 SQ_ALU_SRC_0_5: special constant 0.5 float.
845 * 253 SQ_ALU_SRC_LITERAL: literal constant.
846 * 254 SQ_ALU_SRC_PV: previous vector result.
847 * 255 SQ_ALU_SRC_PS: previous scalar result.
849 for (i
= 0; i
< TGSI_FILE_COUNT
; i
++) {
850 ctx
.file_offset
[i
] = 0;
852 if (ctx
.type
== TGSI_PROCESSOR_VERTEX
) {
853 ctx
.file_offset
[TGSI_FILE_INPUT
] = 1;
854 if (ctx
.bc
->chip_class
>= EVERGREEN
) {
855 r600_bytecode_add_cfinst(ctx
.bc
, EG_V_SQ_CF_WORD1_SQ_CF_INST_CALL_FS
);
857 r600_bytecode_add_cfinst(ctx
.bc
, V_SQ_CF_WORD1_SQ_CF_INST_CALL_FS
);
860 if (ctx
.type
== TGSI_PROCESSOR_FRAGMENT
&& ctx
.bc
->chip_class
>= EVERGREEN
) {
861 ctx
.file_offset
[TGSI_FILE_INPUT
] = evergreen_gpr_count(&ctx
);
863 ctx
.file_offset
[TGSI_FILE_OUTPUT
] = ctx
.file_offset
[TGSI_FILE_INPUT
] +
864 ctx
.info
.file_max
[TGSI_FILE_INPUT
] + 1;
865 ctx
.file_offset
[TGSI_FILE_TEMPORARY
] = ctx
.file_offset
[TGSI_FILE_OUTPUT
] +
866 ctx
.info
.file_max
[TGSI_FILE_OUTPUT
] + 1;
868 /* Outside the GPR range. This will be translated to one of the
869 * kcache banks later. */
870 ctx
.file_offset
[TGSI_FILE_CONSTANT
] = 512;
872 ctx
.file_offset
[TGSI_FILE_IMMEDIATE
] = V_SQ_ALU_SRC_LITERAL
;
873 ctx
.bc
->ar_reg
= ctx
.file_offset
[TGSI_FILE_TEMPORARY
] +
874 ctx
.info
.file_max
[TGSI_FILE_TEMPORARY
] + 1;
875 ctx
.temp_reg
= ctx
.bc
->ar_reg
+ 1;
879 shader
->fs_write_all
= FALSE
;
880 while (!tgsi_parse_end_of_tokens(&ctx
.parse
)) {
881 tgsi_parse_token(&ctx
.parse
);
882 switch (ctx
.parse
.FullToken
.Token
.Type
) {
883 case TGSI_TOKEN_TYPE_IMMEDIATE
:
884 immediate
= &ctx
.parse
.FullToken
.FullImmediate
;
885 ctx
.literals
= realloc(ctx
.literals
, (ctx
.nliterals
+ 1) * 16);
886 if(ctx
.literals
== NULL
) {
890 ctx
.literals
[ctx
.nliterals
* 4 + 0] = immediate
->u
[0].Uint
;
891 ctx
.literals
[ctx
.nliterals
* 4 + 1] = immediate
->u
[1].Uint
;
892 ctx
.literals
[ctx
.nliterals
* 4 + 2] = immediate
->u
[2].Uint
;
893 ctx
.literals
[ctx
.nliterals
* 4 + 3] = immediate
->u
[3].Uint
;
896 case TGSI_TOKEN_TYPE_DECLARATION
:
897 r
= tgsi_declaration(&ctx
);
901 case TGSI_TOKEN_TYPE_INSTRUCTION
:
903 case TGSI_TOKEN_TYPE_PROPERTY
:
904 property
= &ctx
.parse
.FullToken
.FullProperty
;
905 switch (property
->Property
.PropertyName
) {
906 case TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS
:
907 if (property
->u
[0].Data
== 1)
908 shader
->fs_write_all
= TRUE
;
910 case TGSI_PROPERTY_VS_PROHIBIT_UCPS
:
911 if (property
->u
[0].Data
== 1)
912 shader
->vs_prohibit_ucps
= TRUE
;
917 R600_ERR("unsupported token type %d\n", ctx
.parse
.FullToken
.Token
.Type
);
923 if (shader
->two_side
&& ctx
.colors_used
) {
924 if ((r
= process_twoside_color_inputs(&ctx
)))
928 tgsi_parse_init(&ctx
.parse
, tokens
);
929 while (!tgsi_parse_end_of_tokens(&ctx
.parse
)) {
930 tgsi_parse_token(&ctx
.parse
);
931 switch (ctx
.parse
.FullToken
.Token
.Type
) {
932 case TGSI_TOKEN_TYPE_INSTRUCTION
:
933 r
= tgsi_is_supported(&ctx
);
936 ctx
.max_driver_temp_used
= 0;
937 /* reserve first tmp for everyone */
940 opcode
= ctx
.parse
.FullToken
.FullInstruction
.Instruction
.Opcode
;
941 if ((r
= tgsi_split_constant(&ctx
)))
943 if ((r
= tgsi_split_literal_constant(&ctx
)))
945 if (ctx
.bc
->chip_class
== CAYMAN
)
946 ctx
.inst_info
= &cm_shader_tgsi_instruction
[opcode
];
947 else if (ctx
.bc
->chip_class
>= EVERGREEN
)
948 ctx
.inst_info
= &eg_shader_tgsi_instruction
[opcode
];
950 ctx
.inst_info
= &r600_shader_tgsi_instruction
[opcode
];
951 r
= ctx
.inst_info
->process(&ctx
);
960 noutput
= shader
->noutput
;
962 /* clamp color outputs */
963 if (shader
->clamp_color
) {
964 for (i
= 0; i
< noutput
; i
++) {
965 if (shader
->output
[i
].name
== TGSI_SEMANTIC_COLOR
||
966 shader
->output
[i
].name
== TGSI_SEMANTIC_BCOLOR
) {
969 for (j
= 0; j
< 4; j
++) {
970 struct r600_bytecode_alu alu
;
971 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
974 alu
.inst
= BC_INST(ctx
.bc
, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV
);
975 alu
.dst
.sel
= shader
->output
[i
].gpr
;
979 alu
.src
[0].sel
= alu
.dst
.sel
;
985 r
= r600_bytecode_add_alu(ctx
.bc
, &alu
);
993 /* Add stream outputs. */
994 if (ctx
.type
== TGSI_PROCESSOR_VERTEX
&& so
.num_outputs
) {
995 for (i
= 0; i
< so
.num_outputs
; i
++) {
996 struct r600_bytecode_output output
;
998 if (so
.output
[i
].output_buffer
>= 4) {
999 R600_ERR("exceeded the max number of stream output buffers, got: %d\n",
1000 so
.output
[i
].output_buffer
);
1004 if (so
.output
[i
].start_component
) {
1005 R600_ERR("stream_output - start_component cannot be non-zero\n");
1010 memset(&output
, 0, sizeof(struct r600_bytecode_output
));
1011 output
.gpr
= shader
->output
[so
.output
[i
].register_index
].gpr
;
1012 output
.elem_size
= 0;
1013 output
.array_base
= so
.output
[i
].dst_offset
;
1014 output
.type
= V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_WRITE
;
1015 output
.burst_count
= 1;
1017 /* array_size is an upper limit for the burst_count
1018 * with MEM_STREAM instructions */
1019 output
.array_size
= 0xFFF;
1020 output
.comp_mask
= (1 << so
.output
[i
].num_components
) - 1;
1021 if (ctx
.bc
->chip_class
>= EVERGREEN
) {
1022 switch (so
.output
[i
].output_buffer
) {
1024 output
.inst
= EG_V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_MEM_STREAM0_BUF0
;
1027 output
.inst
= EG_V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_MEM_STREAM0_BUF1
;
1030 output
.inst
= EG_V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_MEM_STREAM0_BUF2
;
1033 output
.inst
= EG_V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_MEM_STREAM0_BUF3
;
1037 switch (so
.output
[i
].output_buffer
) {
1039 output
.inst
= V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_MEM_STREAM0
;
1042 output
.inst
= V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_MEM_STREAM1
;
1045 output
.inst
= V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_MEM_STREAM2
;
1048 output
.inst
= V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_MEM_STREAM3
;
1052 r
= r600_bytecode_add_output(ctx
.bc
, &output
);
1061 for (i
= 0, pos0
= 0; i
< noutput
; i
++) {
1062 memset(&output
[i
+j
], 0, sizeof(struct r600_bytecode_output
));
1063 output
[i
+ j
].gpr
= shader
->output
[i
].gpr
;
1064 output
[i
+ j
].elem_size
= 3;
1065 output
[i
+ j
].swizzle_x
= 0;
1066 output
[i
+ j
].swizzle_y
= 1;
1067 output
[i
+ j
].swizzle_z
= 2;
1068 output
[i
+ j
].swizzle_w
= 3;
1069 output
[i
+ j
].burst_count
= 1;
1070 output
[i
+ j
].barrier
= 1;
1071 output
[i
+ j
].type
= V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PARAM
;
1072 output
[i
+ j
].array_base
= i
+j
- pos0
;
1073 output
[i
+ j
].inst
= BC_INST(ctx
.bc
, V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_EXPORT
);
1075 case TGSI_PROCESSOR_VERTEX
:
1076 switch (shader
->output
[i
].name
) {
1077 case TGSI_SEMANTIC_POSITION
:
1078 output
[i
+ j
].array_base
= 60;
1079 output
[i
+ j
].type
= V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_POS
;
1080 /* position doesn't count in array_base */
1084 case TGSI_SEMANTIC_PSIZE
:
1085 output
[i
+ j
].array_base
= 61;
1086 output
[i
+ j
].type
= V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_POS
;
1087 /* position doesn't count in array_base */
1091 case TGSI_SEMANTIC_CLIPDIST
:
1092 /* array base for enabled OUT_MISC_VEC & CCDIST[0|1]_VEC
1093 * vectors is allocated sequentially, starting from 61 */
1094 output
[i
+ j
].array_base
= 61 + shader
->output
[i
].sid
1095 /* +1 if OUT_MISC_VEC is enabled */
1096 + shader
->vs_out_misc_write
1097 /* -1 if OUT_CCDIST0_VEC is disabled */
1098 - (((shader
->clip_dist_write
& 0xF) == 0)? 1 : 0);
1099 output
[i
+ j
].type
= V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_POS
;
1102 /* duplicate it as PARAM to pass to the pixel shader */
1103 memcpy(&output
[i
+j
], &output
[i
+j
-1], sizeof(struct r600_bytecode_output
));
1104 output
[i
+ j
].array_base
= i
+j
-pos0
;
1105 output
[i
+ j
].type
= V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PARAM
;
1109 case TGSI_PROCESSOR_FRAGMENT
:
1110 if (shader
->output
[i
].name
== TGSI_SEMANTIC_COLOR
) {
1111 output
[i
+ j
].array_base
= shader
->output
[i
].sid
;
1112 output
[i
+ j
].type
= V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PIXEL
;
1113 if (shader
->fs_write_all
&& (rctx
->chip_class
>= EVERGREEN
)) {
1114 for (j
= 1; j
< shader
->nr_cbufs
; j
++) {
1115 memset(&output
[i
+ j
], 0, sizeof(struct r600_bytecode_output
));
1116 output
[i
+ j
].gpr
= shader
->output
[i
].gpr
;
1117 output
[i
+ j
].elem_size
= 3;
1118 output
[i
+ j
].swizzle_x
= 0;
1119 output
[i
+ j
].swizzle_y
= 1;
1120 output
[i
+ j
].swizzle_z
= 2;
1121 output
[i
+ j
].swizzle_w
= 3;
1122 output
[i
+ j
].burst_count
= 1;
1123 output
[i
+ j
].barrier
= 1;
1124 output
[i
+ j
].array_base
= shader
->output
[i
].sid
+ j
;
1125 output
[i
+ j
].inst
= BC_INST(ctx
.bc
, V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_EXPORT
);
1126 output
[i
+ j
].type
= V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PIXEL
;
1128 j
= shader
->nr_cbufs
-1;
1130 } else if (shader
->output
[i
].name
== TGSI_SEMANTIC_POSITION
) {
1131 output
[i
+ j
].array_base
= 61;
1132 output
[i
+ j
].swizzle_x
= 2;
1133 output
[i
+ j
].swizzle_y
= 7;
1134 output
[i
+ j
].swizzle_z
= output
[i
+ j
].swizzle_w
= 7;
1135 output
[i
+ j
].type
= V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PIXEL
;
1136 } else if (shader
->output
[i
].name
== TGSI_SEMANTIC_STENCIL
) {
1137 output
[i
+ j
].array_base
= 61;
1138 output
[i
+ j
].swizzle_x
= 7;
1139 output
[i
+ j
].swizzle_y
= 1;
1140 output
[i
+ j
].swizzle_z
= output
[i
+ j
].swizzle_w
= 7;
1141 output
[i
+ j
].type
= V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PIXEL
;
1143 R600_ERR("unsupported fragment output name %d\n", shader
->output
[i
].name
);
1149 R600_ERR("unsupported processor type %d\n", ctx
.type
);
1155 /* add fake param output for vertex shader if no param is exported */
1156 if (ctx
.type
== TGSI_PROCESSOR_VERTEX
) {
1157 for (i
= 0, pos0
= 0; i
< noutput
; i
++) {
1158 if (output
[i
].type
== V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PARAM
) {
1164 memset(&output
[i
], 0, sizeof(struct r600_bytecode_output
));
1166 output
[i
].elem_size
= 3;
1167 output
[i
].swizzle_x
= 7;
1168 output
[i
].swizzle_y
= 7;
1169 output
[i
].swizzle_z
= 7;
1170 output
[i
].swizzle_w
= 7;
1171 output
[i
].burst_count
= 1;
1172 output
[i
].barrier
= 1;
1173 output
[i
].type
= V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PARAM
;
1174 output
[i
].array_base
= 0;
1175 output
[i
].inst
= BC_INST(ctx
.bc
, V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_EXPORT
);
1179 /* add fake pixel export */
1180 if (ctx
.type
== TGSI_PROCESSOR_FRAGMENT
&& !noutput
) {
1181 memset(&output
[0], 0, sizeof(struct r600_bytecode_output
));
1183 output
[0].elem_size
= 3;
1184 output
[0].swizzle_x
= 7;
1185 output
[0].swizzle_y
= 7;
1186 output
[0].swizzle_z
= 7;
1187 output
[0].swizzle_w
= 7;
1188 output
[0].burst_count
= 1;
1189 output
[0].barrier
= 1;
1190 output
[0].type
= V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PIXEL
;
1191 output
[0].array_base
= 0;
1192 output
[0].inst
= BC_INST(ctx
.bc
, V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_EXPORT
);
1195 /* set export done on last export of each type */
1196 for (i
= noutput
- 1, output_done
= 0; i
>= 0; i
--) {
1197 if (ctx
.bc
->chip_class
< CAYMAN
) {
1198 if (i
== (noutput
- 1)) {
1199 output
[i
].end_of_program
= 1;
1202 if (!(output_done
& (1 << output
[i
].type
))) {
1203 output_done
|= (1 << output
[i
].type
);
1204 output
[i
].inst
= BC_INST(ctx
.bc
, V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_EXPORT_DONE
);
1207 /* add output to bytecode */
1208 for (i
= 0; i
< noutput
; i
++) {
1209 r
= r600_bytecode_add_output(ctx
.bc
, &output
[i
]);
1213 /* add program end */
1214 if (ctx
.bc
->chip_class
== CAYMAN
)
1215 cm_bytecode_add_cf_end(ctx
.bc
);
1218 tgsi_parse_free(&ctx
.parse
);
1222 tgsi_parse_free(&ctx
.parse
);
1226 static int tgsi_unsupported(struct r600_shader_ctx
*ctx
)
1228 R600_ERR("%s tgsi opcode unsupported\n",
1229 tgsi_get_opcode_name(ctx
->inst_info
->tgsi_opcode
));
1233 static int tgsi_end(struct r600_shader_ctx
*ctx
)
1238 static void r600_bytecode_src(struct r600_bytecode_alu_src
*bc_src
,
1239 const struct r600_shader_src
*shader_src
,
1242 bc_src
->sel
= shader_src
->sel
;
1243 bc_src
->chan
= shader_src
->swizzle
[chan
];
1244 bc_src
->neg
= shader_src
->neg
;
1245 bc_src
->abs
= shader_src
->abs
;
1246 bc_src
->rel
= shader_src
->rel
;
1247 bc_src
->value
= shader_src
->value
[bc_src
->chan
];
1250 static void r600_bytecode_src_set_abs(struct r600_bytecode_alu_src
*bc_src
)
1256 static void r600_bytecode_src_toggle_neg(struct r600_bytecode_alu_src
*bc_src
)
1258 bc_src
->neg
= !bc_src
->neg
;
1261 static void tgsi_dst(struct r600_shader_ctx
*ctx
,
1262 const struct tgsi_full_dst_register
*tgsi_dst
,
1264 struct r600_bytecode_alu_dst
*r600_dst
)
1266 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
1268 r600_dst
->sel
= tgsi_dst
->Register
.Index
;
1269 r600_dst
->sel
+= ctx
->file_offset
[tgsi_dst
->Register
.File
];
1270 r600_dst
->chan
= swizzle
;
1271 r600_dst
->write
= 1;
1272 if (tgsi_dst
->Register
.Indirect
)
1273 r600_dst
->rel
= V_SQ_REL_RELATIVE
;
1274 if (inst
->Instruction
.Saturate
) {
1275 r600_dst
->clamp
= 1;
1279 static int tgsi_last_instruction(unsigned writemask
)
1283 for (i
= 0; i
< 4; i
++) {
1284 if (writemask
& (1 << i
)) {
1291 static int tgsi_op2_s(struct r600_shader_ctx
*ctx
, int swap
, int trans_only
)
1293 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
1294 struct r600_bytecode_alu alu
;
1296 int lasti
= tgsi_last_instruction(inst
->Dst
[0].Register
.WriteMask
);
1298 for (i
= 0; i
< lasti
+ 1; i
++) {
1299 if (!(inst
->Dst
[0].Register
.WriteMask
& (1 << i
)))
1302 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
1303 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
1305 alu
.inst
= ctx
->inst_info
->r600_opcode
;
1307 for (j
= 0; j
< inst
->Instruction
.NumSrcRegs
; j
++) {
1308 r600_bytecode_src(&alu
.src
[j
], &ctx
->src
[j
], i
);
1311 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[1], i
);
1312 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[0], i
);
1314 /* handle some special cases */
1315 switch (ctx
->inst_info
->tgsi_opcode
) {
1316 case TGSI_OPCODE_SUB
:
1317 r600_bytecode_src_toggle_neg(&alu
.src
[1]);
1319 case TGSI_OPCODE_ABS
:
1320 r600_bytecode_src_set_abs(&alu
.src
[0]);
1325 if (i
== lasti
|| trans_only
) {
1328 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
1335 static int tgsi_op2(struct r600_shader_ctx
*ctx
)
1337 return tgsi_op2_s(ctx
, 0, 0);
1340 static int tgsi_op2_swap(struct r600_shader_ctx
*ctx
)
1342 return tgsi_op2_s(ctx
, 1, 0);
1345 static int tgsi_op2_trans(struct r600_shader_ctx
*ctx
)
1347 return tgsi_op2_s(ctx
, 0, 1);
1350 static int tgsi_ineg(struct r600_shader_ctx
*ctx
)
1352 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
1353 struct r600_bytecode_alu alu
;
1355 int lasti
= tgsi_last_instruction(inst
->Dst
[0].Register
.WriteMask
);
1357 for (i
= 0; i
< lasti
+ 1; i
++) {
1359 if (!(inst
->Dst
[0].Register
.WriteMask
& (1 << i
)))
1361 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
1362 alu
.inst
= ctx
->inst_info
->r600_opcode
;
1364 alu
.src
[0].sel
= V_SQ_ALU_SRC_0
;
1366 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[0], i
);
1368 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
1373 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
1381 static int cayman_emit_float_instr(struct r600_shader_ctx
*ctx
)
1383 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
1385 struct r600_bytecode_alu alu
;
1386 int last_slot
= (inst
->Dst
[0].Register
.WriteMask
& 0x8) ? 4 : 3;
1388 for (i
= 0 ; i
< last_slot
; i
++) {
1389 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
1390 alu
.inst
= ctx
->inst_info
->r600_opcode
;
1391 for (j
= 0; j
< inst
->Instruction
.NumSrcRegs
; j
++) {
1392 r600_bytecode_src(&alu
.src
[j
], &ctx
->src
[j
], 0);
1394 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
1395 alu
.dst
.write
= (inst
->Dst
[0].Register
.WriteMask
>> i
) & 1;
1397 if (i
== last_slot
- 1)
1399 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
1407 * r600 - trunc to -PI..PI range
1408 * r700 - normalize by dividing by 2PI
1411 static int tgsi_setup_trig(struct r600_shader_ctx
*ctx
)
1413 static float half_inv_pi
= 1.0 /(3.1415926535 * 2);
1414 static float double_pi
= 3.1415926535 * 2;
1415 static float neg_pi
= -3.1415926535;
1418 struct r600_bytecode_alu alu
;
1420 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
1421 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP3_SQ_OP3_INST_MULADD
);
1425 alu
.dst
.sel
= ctx
->temp_reg
;
1428 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 0);
1430 alu
.src
[1].sel
= V_SQ_ALU_SRC_LITERAL
;
1431 alu
.src
[1].chan
= 0;
1432 alu
.src
[1].value
= *(uint32_t *)&half_inv_pi
;
1433 alu
.src
[2].sel
= V_SQ_ALU_SRC_0_5
;
1434 alu
.src
[2].chan
= 0;
1436 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
1440 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
1441 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FRACT
);
1444 alu
.dst
.sel
= ctx
->temp_reg
;
1447 alu
.src
[0].sel
= ctx
->temp_reg
;
1448 alu
.src
[0].chan
= 0;
1450 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
1454 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
1455 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP3_SQ_OP3_INST_MULADD
);
1459 alu
.dst
.sel
= ctx
->temp_reg
;
1462 alu
.src
[0].sel
= ctx
->temp_reg
;
1463 alu
.src
[0].chan
= 0;
1465 alu
.src
[1].sel
= V_SQ_ALU_SRC_LITERAL
;
1466 alu
.src
[1].chan
= 0;
1467 alu
.src
[2].sel
= V_SQ_ALU_SRC_LITERAL
;
1468 alu
.src
[2].chan
= 0;
1470 if (ctx
->bc
->chip_class
== R600
) {
1471 alu
.src
[1].value
= *(uint32_t *)&double_pi
;
1472 alu
.src
[2].value
= *(uint32_t *)&neg_pi
;
1474 alu
.src
[1].sel
= V_SQ_ALU_SRC_1
;
1475 alu
.src
[2].sel
= V_SQ_ALU_SRC_0_5
;
1480 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
1486 static int cayman_trig(struct r600_shader_ctx
*ctx
)
1488 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
1489 struct r600_bytecode_alu alu
;
1490 int last_slot
= (inst
->Dst
[0].Register
.WriteMask
& 0x8) ? 4 : 3;
1493 r
= tgsi_setup_trig(ctx
);
1498 for (i
= 0; i
< last_slot
; i
++) {
1499 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
1500 alu
.inst
= ctx
->inst_info
->r600_opcode
;
1503 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
1504 alu
.dst
.write
= (inst
->Dst
[0].Register
.WriteMask
>> i
) & 1;
1506 alu
.src
[0].sel
= ctx
->temp_reg
;
1507 alu
.src
[0].chan
= 0;
1508 if (i
== last_slot
- 1)
1510 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
1517 static int tgsi_trig(struct r600_shader_ctx
*ctx
)
1519 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
1520 struct r600_bytecode_alu alu
;
1522 int lasti
= tgsi_last_instruction(inst
->Dst
[0].Register
.WriteMask
);
1524 r
= tgsi_setup_trig(ctx
);
1528 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
1529 alu
.inst
= ctx
->inst_info
->r600_opcode
;
1531 alu
.dst
.sel
= ctx
->temp_reg
;
1534 alu
.src
[0].sel
= ctx
->temp_reg
;
1535 alu
.src
[0].chan
= 0;
1537 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
1541 /* replicate result */
1542 for (i
= 0; i
< lasti
+ 1; i
++) {
1543 if (!(inst
->Dst
[0].Register
.WriteMask
& (1 << i
)))
1546 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
1547 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV
);
1549 alu
.src
[0].sel
= ctx
->temp_reg
;
1550 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
1553 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
1560 static int tgsi_scs(struct r600_shader_ctx
*ctx
)
1562 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
1563 struct r600_bytecode_alu alu
;
1566 /* We'll only need the trig stuff if we are going to write to the
1567 * X or Y components of the destination vector.
1569 if (likely(inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_XY
)) {
1570 r
= tgsi_setup_trig(ctx
);
1576 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_X
) {
1577 if (ctx
->bc
->chip_class
== CAYMAN
) {
1578 for (i
= 0 ; i
< 3; i
++) {
1579 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
1580 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_COS
);
1581 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
1587 alu
.src
[0].sel
= ctx
->temp_reg
;
1588 alu
.src
[0].chan
= 0;
1591 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
1596 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
1597 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_COS
);
1598 tgsi_dst(ctx
, &inst
->Dst
[0], 0, &alu
.dst
);
1600 alu
.src
[0].sel
= ctx
->temp_reg
;
1601 alu
.src
[0].chan
= 0;
1603 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
1610 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Y
) {
1611 if (ctx
->bc
->chip_class
== CAYMAN
) {
1612 for (i
= 0 ; i
< 3; i
++) {
1613 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
1614 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SIN
);
1615 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
1620 alu
.src
[0].sel
= ctx
->temp_reg
;
1621 alu
.src
[0].chan
= 0;
1624 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
1629 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
1630 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SIN
);
1631 tgsi_dst(ctx
, &inst
->Dst
[0], 1, &alu
.dst
);
1633 alu
.src
[0].sel
= ctx
->temp_reg
;
1634 alu
.src
[0].chan
= 0;
1636 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
1643 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Z
) {
1644 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
1646 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV
);
1648 tgsi_dst(ctx
, &inst
->Dst
[0], 2, &alu
.dst
);
1650 alu
.src
[0].sel
= V_SQ_ALU_SRC_0
;
1651 alu
.src
[0].chan
= 0;
1655 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
1661 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_W
) {
1662 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
1664 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV
);
1666 tgsi_dst(ctx
, &inst
->Dst
[0], 3, &alu
.dst
);
1668 alu
.src
[0].sel
= V_SQ_ALU_SRC_1
;
1669 alu
.src
[0].chan
= 0;
1673 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
1681 static int tgsi_kill(struct r600_shader_ctx
*ctx
)
1683 struct r600_bytecode_alu alu
;
1686 for (i
= 0; i
< 4; i
++) {
1687 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
1688 alu
.inst
= ctx
->inst_info
->r600_opcode
;
1692 alu
.src
[0].sel
= V_SQ_ALU_SRC_0
;
1694 if (ctx
->inst_info
->tgsi_opcode
== TGSI_OPCODE_KILP
) {
1695 alu
.src
[1].sel
= V_SQ_ALU_SRC_1
;
1698 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[0], i
);
1703 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
1708 /* kill must be last in ALU */
1709 ctx
->bc
->force_add_cf
= 1;
1710 ctx
->shader
->uses_kill
= TRUE
;
1714 static int tgsi_lit(struct r600_shader_ctx
*ctx
)
1716 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
1717 struct r600_bytecode_alu alu
;
1720 /* tmp.x = max(src.y, 0.0) */
1721 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
1722 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MAX
);
1723 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 1);
1724 alu
.src
[1].sel
= V_SQ_ALU_SRC_0
; /*0.0*/
1725 alu
.src
[1].chan
= 1;
1727 alu
.dst
.sel
= ctx
->temp_reg
;
1732 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
1736 if (inst
->Dst
[0].Register
.WriteMask
& (1 << 2))
1742 if (ctx
->bc
->chip_class
== CAYMAN
) {
1743 for (i
= 0; i
< 3; i
++) {
1744 /* tmp.z = log(tmp.x) */
1745 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
1746 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_LOG_CLAMPED
);
1747 alu
.src
[0].sel
= ctx
->temp_reg
;
1748 alu
.src
[0].chan
= 0;
1749 alu
.dst
.sel
= ctx
->temp_reg
;
1757 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
1762 /* tmp.z = log(tmp.x) */
1763 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
1764 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_LOG_CLAMPED
);
1765 alu
.src
[0].sel
= ctx
->temp_reg
;
1766 alu
.src
[0].chan
= 0;
1767 alu
.dst
.sel
= ctx
->temp_reg
;
1771 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
1776 chan
= alu
.dst
.chan
;
1779 /* tmp.x = amd MUL_LIT(tmp.z, src.w, src.x ) */
1780 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
1781 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP3_SQ_OP3_INST_MUL_LIT
);
1782 alu
.src
[0].sel
= sel
;
1783 alu
.src
[0].chan
= chan
;
1784 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[0], 3);
1785 r600_bytecode_src(&alu
.src
[2], &ctx
->src
[0], 0);
1786 alu
.dst
.sel
= ctx
->temp_reg
;
1791 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
1795 if (ctx
->bc
->chip_class
== CAYMAN
) {
1796 for (i
= 0; i
< 3; i
++) {
1797 /* dst.z = exp(tmp.x) */
1798 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
1799 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_EXP_IEEE
);
1800 alu
.src
[0].sel
= ctx
->temp_reg
;
1801 alu
.src
[0].chan
= 0;
1802 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
1808 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
1813 /* dst.z = exp(tmp.x) */
1814 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
1815 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_EXP_IEEE
);
1816 alu
.src
[0].sel
= ctx
->temp_reg
;
1817 alu
.src
[0].chan
= 0;
1818 tgsi_dst(ctx
, &inst
->Dst
[0], 2, &alu
.dst
);
1820 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
1827 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
1828 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV
);
1829 alu
.src
[0].sel
= V_SQ_ALU_SRC_1
; /*1.0*/
1830 alu
.src
[0].chan
= 0;
1831 tgsi_dst(ctx
, &inst
->Dst
[0], 0, &alu
.dst
);
1832 alu
.dst
.write
= (inst
->Dst
[0].Register
.WriteMask
>> 0) & 1;
1833 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
1837 /* dst.y = max(src.x, 0.0) */
1838 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
1839 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MAX
);
1840 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 0);
1841 alu
.src
[1].sel
= V_SQ_ALU_SRC_0
; /*0.0*/
1842 alu
.src
[1].chan
= 0;
1843 tgsi_dst(ctx
, &inst
->Dst
[0], 1, &alu
.dst
);
1844 alu
.dst
.write
= (inst
->Dst
[0].Register
.WriteMask
>> 1) & 1;
1845 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
1850 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
1851 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV
);
1852 alu
.src
[0].sel
= V_SQ_ALU_SRC_1
;
1853 alu
.src
[0].chan
= 0;
1854 tgsi_dst(ctx
, &inst
->Dst
[0], 3, &alu
.dst
);
1855 alu
.dst
.write
= (inst
->Dst
[0].Register
.WriteMask
>> 3) & 1;
1857 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
1864 static int tgsi_rsq(struct r600_shader_ctx
*ctx
)
1866 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
1867 struct r600_bytecode_alu alu
;
1870 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
1873 * For state trackers other than OpenGL, we'll want to use
1874 * _RECIPSQRT_IEEE instead.
1876 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_RECIPSQRT_CLAMPED
);
1878 for (i
= 0; i
< inst
->Instruction
.NumSrcRegs
; i
++) {
1879 r600_bytecode_src(&alu
.src
[i
], &ctx
->src
[i
], 0);
1880 r600_bytecode_src_set_abs(&alu
.src
[i
]);
1882 alu
.dst
.sel
= ctx
->temp_reg
;
1885 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
1888 /* replicate result */
1889 return tgsi_helper_tempx_replicate(ctx
);
1892 static int tgsi_helper_tempx_replicate(struct r600_shader_ctx
*ctx
)
1894 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
1895 struct r600_bytecode_alu alu
;
1898 for (i
= 0; i
< 4; i
++) {
1899 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
1900 alu
.src
[0].sel
= ctx
->temp_reg
;
1901 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV
);
1903 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
1904 alu
.dst
.write
= (inst
->Dst
[0].Register
.WriteMask
>> i
) & 1;
1907 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
1914 static int tgsi_trans_srcx_replicate(struct r600_shader_ctx
*ctx
)
1916 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
1917 struct r600_bytecode_alu alu
;
1920 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
1921 alu
.inst
= ctx
->inst_info
->r600_opcode
;
1922 for (i
= 0; i
< inst
->Instruction
.NumSrcRegs
; i
++) {
1923 r600_bytecode_src(&alu
.src
[i
], &ctx
->src
[i
], 0);
1925 alu
.dst
.sel
= ctx
->temp_reg
;
1928 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
1931 /* replicate result */
1932 return tgsi_helper_tempx_replicate(ctx
);
1935 static int cayman_pow(struct r600_shader_ctx
*ctx
)
1937 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
1939 struct r600_bytecode_alu alu
;
1940 int last_slot
= (inst
->Dst
[0].Register
.WriteMask
& 0x8) ? 4 : 3;
1942 for (i
= 0; i
< 3; i
++) {
1943 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
1944 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_LOG_IEEE
);
1945 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 0);
1946 alu
.dst
.sel
= ctx
->temp_reg
;
1951 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
1957 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
1958 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MUL
);
1959 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[1], 0);
1960 alu
.src
[1].sel
= ctx
->temp_reg
;
1961 alu
.dst
.sel
= ctx
->temp_reg
;
1964 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
1968 for (i
= 0; i
< last_slot
; i
++) {
1969 /* POW(a,b) = EXP2(b * LOG2(a))*/
1970 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
1971 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_EXP_IEEE
);
1972 alu
.src
[0].sel
= ctx
->temp_reg
;
1974 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
1975 alu
.dst
.write
= (inst
->Dst
[0].Register
.WriteMask
>> i
) & 1;
1976 if (i
== last_slot
- 1)
1978 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
1985 static int tgsi_pow(struct r600_shader_ctx
*ctx
)
1987 struct r600_bytecode_alu alu
;
1991 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
1992 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_LOG_IEEE
);
1993 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 0);
1994 alu
.dst
.sel
= ctx
->temp_reg
;
1997 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
2001 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2002 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MUL
);
2003 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[1], 0);
2004 alu
.src
[1].sel
= ctx
->temp_reg
;
2005 alu
.dst
.sel
= ctx
->temp_reg
;
2008 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
2011 /* POW(a,b) = EXP2(b * LOG2(a))*/
2012 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2013 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_EXP_IEEE
);
2014 alu
.src
[0].sel
= ctx
->temp_reg
;
2015 alu
.dst
.sel
= ctx
->temp_reg
;
2018 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
2021 return tgsi_helper_tempx_replicate(ctx
);
2024 static int tgsi_divmod(struct r600_shader_ctx
*ctx
, int mod
, int signed_op
)
2026 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
2027 struct r600_bytecode_alu alu
;
2029 unsigned write_mask
= inst
->Dst
[0].Register
.WriteMask
;
2030 int tmp0
= ctx
->temp_reg
;
2031 int tmp1
= r600_get_temp(ctx
);
2032 int tmp2
= r600_get_temp(ctx
);
2036 * we need to represent src1 as src2*q + r, where q - quotient, r - remainder
2038 * 1. tmp0.x = rcp (src2) = 2^32/src2 + e, where e is rounding error
2039 * 2. tmp0.z = lo (tmp0.x * src2)
2040 * 3. tmp0.w = -tmp0.z
2041 * 4. tmp0.y = hi (tmp0.x * src2)
2042 * 5. tmp0.z = (tmp0.y == 0 ? tmp0.w : tmp0.z) = abs(lo(rcp*src2))
2043 * 6. tmp0.w = hi (tmp0.z * tmp0.x) = e, rounding error
2044 * 7. tmp1.x = tmp0.x - tmp0.w
2045 * 8. tmp1.y = tmp0.x + tmp0.w
2046 * 9. tmp0.x = (tmp0.y == 0 ? tmp1.y : tmp1.x)
2047 * 10. tmp0.z = hi(tmp0.x * src1) = q
2048 * 11. tmp0.y = lo (tmp0.z * src2) = src2*q = src1 - r
2050 * 12. tmp0.w = src1 - tmp0.y = r
2051 * 13. tmp1.x = tmp0.w >= src2 = r >= src2 (uint comparison)
2052 * 14. tmp1.y = src1 >= tmp0.y = r >= 0 (uint comparison)
2056 * 15. tmp1.z = tmp0.z + 1 = q + 1
2057 * 16. tmp1.w = tmp0.z - 1 = q - 1
2061 * 15. tmp1.z = tmp0.w - src2 = r - src2
2062 * 16. tmp1.w = tmp0.w + src2 = r + src2
2066 * 17. tmp1.x = tmp1.x & tmp1.y
2068 * DIV: 18. tmp0.z = tmp1.x==0 ? tmp0.z : tmp1.z
2069 * MOD: 18. tmp0.z = tmp1.x==0 ? tmp0.w : tmp1.z
2071 * 19. tmp0.z = tmp1.y==0 ? tmp1.w : tmp0.z
2072 * 20. dst = src2==0 ? MAX_UINT : tmp0.z
2076 * Same as unsigned, using abs values of the operands,
2077 * and fixing the sign of the result in the end.
2080 for (i
= 0; i
< 4; i
++) {
2081 if (!(write_mask
& (1<<i
)))
2086 /* tmp2.x = -src0 */
2087 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2088 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SUB_INT
);
2094 alu
.src
[0].sel
= V_SQ_ALU_SRC_0
;
2096 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[0], i
);
2099 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
2102 /* tmp2.y = -src1 */
2103 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2104 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SUB_INT
);
2110 alu
.src
[0].sel
= V_SQ_ALU_SRC_0
;
2112 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[1], i
);
2115 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
2118 /* tmp2.z sign bit is set if src0 and src2 signs are different */
2119 /* it will be a sign of the quotient */
2122 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2123 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_XOR_INT
);
2129 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], i
);
2130 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[1], i
);
2133 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
2137 /* tmp2.x = |src0| */
2138 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2139 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP3_SQ_OP3_INST_CNDGE_INT
);
2146 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], i
);
2147 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[0], i
);
2148 alu
.src
[2].sel
= tmp2
;
2149 alu
.src
[2].chan
= 0;
2152 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
2155 /* tmp2.y = |src1| */
2156 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2157 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP3_SQ_OP3_INST_CNDGE_INT
);
2164 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[1], i
);
2165 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[1], i
);
2166 alu
.src
[2].sel
= tmp2
;
2167 alu
.src
[2].chan
= 1;
2170 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
2175 /* 1. tmp0.x = rcp_u (src2) = 2^32/src2 + e, where e is rounding error */
2176 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2177 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_RECIP_UINT
);
2184 alu
.src
[0].sel
= tmp2
;
2185 alu
.src
[0].chan
= 1;
2187 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[1], i
);
2191 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
2194 /* 2. tmp0.z = lo (tmp0.x * src2) */
2195 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2196 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MULLO_UINT
);
2202 alu
.src
[0].sel
= tmp0
;
2203 alu
.src
[0].chan
= 0;
2205 alu
.src
[1].sel
= tmp2
;
2206 alu
.src
[1].chan
= 1;
2208 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[1], i
);
2212 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
2215 /* 3. tmp0.w = -tmp0.z */
2216 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2217 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SUB_INT
);
2223 alu
.src
[0].sel
= V_SQ_ALU_SRC_0
;
2224 alu
.src
[1].sel
= tmp0
;
2225 alu
.src
[1].chan
= 2;
2228 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
2231 /* 4. tmp0.y = hi (tmp0.x * src2) */
2232 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2233 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MULHI_UINT
);
2239 alu
.src
[0].sel
= tmp0
;
2240 alu
.src
[0].chan
= 0;
2243 alu
.src
[1].sel
= tmp2
;
2244 alu
.src
[1].chan
= 1;
2246 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[1], i
);
2250 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
2253 /* 5. tmp0.z = (tmp0.y == 0 ? tmp0.w : tmp0.z) = abs(lo(rcp*src)) */
2254 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2255 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP3_SQ_OP3_INST_CNDE_INT
);
2262 alu
.src
[0].sel
= tmp0
;
2263 alu
.src
[0].chan
= 1;
2264 alu
.src
[1].sel
= tmp0
;
2265 alu
.src
[1].chan
= 3;
2266 alu
.src
[2].sel
= tmp0
;
2267 alu
.src
[2].chan
= 2;
2270 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
2273 /* 6. tmp0.w = hi (tmp0.z * tmp0.x) = e, rounding error */
2274 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2275 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MULHI_UINT
);
2281 alu
.src
[0].sel
= tmp0
;
2282 alu
.src
[0].chan
= 2;
2284 alu
.src
[1].sel
= tmp0
;
2285 alu
.src
[1].chan
= 0;
2288 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
2291 /* 7. tmp1.x = tmp0.x - tmp0.w */
2292 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2293 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SUB_INT
);
2299 alu
.src
[0].sel
= tmp0
;
2300 alu
.src
[0].chan
= 0;
2301 alu
.src
[1].sel
= tmp0
;
2302 alu
.src
[1].chan
= 3;
2305 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
2308 /* 8. tmp1.y = tmp0.x + tmp0.w */
2309 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2310 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_ADD_INT
);
2316 alu
.src
[0].sel
= tmp0
;
2317 alu
.src
[0].chan
= 0;
2318 alu
.src
[1].sel
= tmp0
;
2319 alu
.src
[1].chan
= 3;
2322 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
2325 /* 9. tmp0.x = (tmp0.y == 0 ? tmp1.y : tmp1.x) */
2326 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2327 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP3_SQ_OP3_INST_CNDE_INT
);
2334 alu
.src
[0].sel
= tmp0
;
2335 alu
.src
[0].chan
= 1;
2336 alu
.src
[1].sel
= tmp1
;
2337 alu
.src
[1].chan
= 1;
2338 alu
.src
[2].sel
= tmp1
;
2339 alu
.src
[2].chan
= 0;
2342 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
2345 /* 10. tmp0.z = hi(tmp0.x * src1) = q */
2346 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2347 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MULHI_UINT
);
2353 alu
.src
[0].sel
= tmp0
;
2354 alu
.src
[0].chan
= 0;
2357 alu
.src
[1].sel
= tmp2
;
2358 alu
.src
[1].chan
= 0;
2360 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[0], i
);
2364 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
2367 /* 11. tmp0.y = lo (src2 * tmp0.z) = src2*q = src1 - r */
2368 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2369 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MULLO_UINT
);
2376 alu
.src
[0].sel
= tmp2
;
2377 alu
.src
[0].chan
= 1;
2379 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[1], i
);
2382 alu
.src
[1].sel
= tmp0
;
2383 alu
.src
[1].chan
= 2;
2386 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
2389 /* 12. tmp0.w = src1 - tmp0.y = r */
2390 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2391 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SUB_INT
);
2398 alu
.src
[0].sel
= tmp2
;
2399 alu
.src
[0].chan
= 0;
2401 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], i
);
2404 alu
.src
[1].sel
= tmp0
;
2405 alu
.src
[1].chan
= 1;
2408 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
2411 /* 13. tmp1.x = tmp0.w >= src2 = r >= src2 */
2412 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2413 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SETGE_UINT
);
2419 alu
.src
[0].sel
= tmp0
;
2420 alu
.src
[0].chan
= 3;
2422 alu
.src
[1].sel
= tmp2
;
2423 alu
.src
[1].chan
= 1;
2425 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[1], i
);
2429 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
2432 /* 14. tmp1.y = src1 >= tmp0.y = r >= 0 */
2433 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2434 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SETGE_UINT
);
2441 alu
.src
[0].sel
= tmp2
;
2442 alu
.src
[0].chan
= 0;
2444 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], i
);
2447 alu
.src
[1].sel
= tmp0
;
2448 alu
.src
[1].chan
= 1;
2451 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
2454 if (mod
) { /* UMOD */
2456 /* 15. tmp1.z = tmp0.w - src2 = r - src2 */
2457 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2458 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SUB_INT
);
2464 alu
.src
[0].sel
= tmp0
;
2465 alu
.src
[0].chan
= 3;
2468 alu
.src
[1].sel
= tmp2
;
2469 alu
.src
[1].chan
= 1;
2471 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[1], i
);
2475 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
2478 /* 16. tmp1.w = tmp0.w + src2 = r + src2 */
2479 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2480 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_ADD_INT
);
2486 alu
.src
[0].sel
= tmp0
;
2487 alu
.src
[0].chan
= 3;
2489 alu
.src
[1].sel
= tmp2
;
2490 alu
.src
[1].chan
= 1;
2492 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[1], i
);
2496 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
2501 /* 15. tmp1.z = tmp0.z + 1 = q + 1 DIV */
2502 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2503 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_ADD_INT
);
2509 alu
.src
[0].sel
= tmp0
;
2510 alu
.src
[0].chan
= 2;
2511 alu
.src
[1].sel
= V_SQ_ALU_SRC_1_INT
;
2514 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
2517 /* 16. tmp1.w = tmp0.z - 1 = q - 1 */
2518 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2519 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_ADD_INT
);
2525 alu
.src
[0].sel
= tmp0
;
2526 alu
.src
[0].chan
= 2;
2527 alu
.src
[1].sel
= V_SQ_ALU_SRC_M_1_INT
;
2530 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
2535 /* 17. tmp1.x = tmp1.x & tmp1.y */
2536 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2537 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_AND_INT
);
2543 alu
.src
[0].sel
= tmp1
;
2544 alu
.src
[0].chan
= 0;
2545 alu
.src
[1].sel
= tmp1
;
2546 alu
.src
[1].chan
= 1;
2549 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
2552 /* 18. tmp0.z = tmp1.x==0 ? tmp0.z : tmp1.z DIV */
2553 /* 18. tmp0.z = tmp1.x==0 ? tmp0.w : tmp1.z MOD */
2554 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2555 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP3_SQ_OP3_INST_CNDE_INT
);
2562 alu
.src
[0].sel
= tmp1
;
2563 alu
.src
[0].chan
= 0;
2564 alu
.src
[1].sel
= tmp0
;
2565 alu
.src
[1].chan
= mod
? 3 : 2;
2566 alu
.src
[2].sel
= tmp1
;
2567 alu
.src
[2].chan
= 2;
2570 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
2573 /* 19. tmp0.z = tmp1.y==0 ? tmp1.w : tmp0.z */
2574 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2575 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP3_SQ_OP3_INST_CNDE_INT
);
2583 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
2586 alu
.src
[0].sel
= tmp1
;
2587 alu
.src
[0].chan
= 1;
2588 alu
.src
[1].sel
= tmp1
;
2589 alu
.src
[1].chan
= 3;
2590 alu
.src
[2].sel
= tmp0
;
2591 alu
.src
[2].chan
= 2;
2594 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
2599 /* fix the sign of the result */
2603 /* tmp0.x = -tmp0.z */
2604 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2605 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SUB_INT
);
2611 alu
.src
[0].sel
= V_SQ_ALU_SRC_0
;
2612 alu
.src
[1].sel
= tmp0
;
2613 alu
.src
[1].chan
= 2;
2616 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
2619 /* sign of the remainder is the same as the sign of src0 */
2620 /* tmp0.x = src0>=0 ? tmp0.z : tmp0.x */
2621 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2622 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP3_SQ_OP3_INST_CNDGE_INT
);
2625 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
2627 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], i
);
2628 alu
.src
[1].sel
= tmp0
;
2629 alu
.src
[1].chan
= 2;
2630 alu
.src
[2].sel
= tmp0
;
2631 alu
.src
[2].chan
= 0;
2634 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
2639 /* tmp0.x = -tmp0.z */
2640 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2641 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SUB_INT
);
2647 alu
.src
[0].sel
= V_SQ_ALU_SRC_0
;
2648 alu
.src
[1].sel
= tmp0
;
2649 alu
.src
[1].chan
= 2;
2652 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
2655 /* fix the quotient sign (same as the sign of src0*src1) */
2656 /* tmp0.x = tmp2.z>=0 ? tmp0.z : tmp0.x */
2657 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2658 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP3_SQ_OP3_INST_CNDGE_INT
);
2661 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
2663 alu
.src
[0].sel
= tmp2
;
2664 alu
.src
[0].chan
= 2;
2665 alu
.src
[1].sel
= tmp0
;
2666 alu
.src
[1].chan
= 2;
2667 alu
.src
[2].sel
= tmp0
;
2668 alu
.src
[2].chan
= 0;
2671 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
2679 static int tgsi_udiv(struct r600_shader_ctx
*ctx
)
2681 return tgsi_divmod(ctx
, 0, 0);
2684 static int tgsi_umod(struct r600_shader_ctx
*ctx
)
2686 return tgsi_divmod(ctx
, 1, 0);
2689 static int tgsi_idiv(struct r600_shader_ctx
*ctx
)
2691 return tgsi_divmod(ctx
, 0, 1);
2694 static int tgsi_imod(struct r600_shader_ctx
*ctx
)
2696 return tgsi_divmod(ctx
, 1, 1);
2700 static int tgsi_f2i(struct r600_shader_ctx
*ctx
)
2702 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
2703 struct r600_bytecode_alu alu
;
2705 unsigned write_mask
= inst
->Dst
[0].Register
.WriteMask
;
2706 int last_inst
= tgsi_last_instruction(write_mask
);
2708 for (i
= 0; i
< 4; i
++) {
2709 if (!(write_mask
& (1<<i
)))
2712 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2713 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_TRUNC
);
2715 alu
.dst
.sel
= ctx
->temp_reg
;
2719 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], i
);
2722 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
2727 for (i
= 0; i
< 4; i
++) {
2728 if (!(write_mask
& (1<<i
)))
2731 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2732 alu
.inst
= ctx
->inst_info
->r600_opcode
;
2734 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
2736 alu
.src
[0].sel
= ctx
->temp_reg
;
2737 alu
.src
[0].chan
= i
;
2741 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
2749 static int tgsi_iabs(struct r600_shader_ctx
*ctx
)
2751 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
2752 struct r600_bytecode_alu alu
;
2754 unsigned write_mask
= inst
->Dst
[0].Register
.WriteMask
;
2755 int last_inst
= tgsi_last_instruction(write_mask
);
2758 for (i
= 0; i
< 4; i
++) {
2759 if (!(write_mask
& (1<<i
)))
2762 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2763 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SUB_INT
);
2765 alu
.dst
.sel
= ctx
->temp_reg
;
2769 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[0], i
);
2770 alu
.src
[0].sel
= V_SQ_ALU_SRC_0
;
2774 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
2779 /* dst = (src >= 0 ? src : tmp) */
2780 for (i
= 0; i
< 4; i
++) {
2781 if (!(write_mask
& (1<<i
)))
2784 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2785 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP3_SQ_OP3_INST_CNDGE_INT
);
2789 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
2791 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], i
);
2792 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[0], i
);
2793 alu
.src
[2].sel
= ctx
->temp_reg
;
2794 alu
.src
[2].chan
= i
;
2798 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
2805 static int tgsi_issg(struct r600_shader_ctx
*ctx
)
2807 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
2808 struct r600_bytecode_alu alu
;
2810 unsigned write_mask
= inst
->Dst
[0].Register
.WriteMask
;
2811 int last_inst
= tgsi_last_instruction(write_mask
);
2813 /* tmp = (src >= 0 ? src : -1) */
2814 for (i
= 0; i
< 4; i
++) {
2815 if (!(write_mask
& (1<<i
)))
2818 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2819 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP3_SQ_OP3_INST_CNDGE_INT
);
2822 alu
.dst
.sel
= ctx
->temp_reg
;
2826 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], i
);
2827 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[0], i
);
2828 alu
.src
[2].sel
= V_SQ_ALU_SRC_M_1_INT
;
2832 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
2837 /* dst = (tmp > 0 ? 1 : tmp) */
2838 for (i
= 0; i
< 4; i
++) {
2839 if (!(write_mask
& (1<<i
)))
2842 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2843 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP3_SQ_OP3_INST_CNDGT_INT
);
2847 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
2849 alu
.src
[0].sel
= ctx
->temp_reg
;
2850 alu
.src
[0].chan
= i
;
2852 alu
.src
[1].sel
= V_SQ_ALU_SRC_1_INT
;
2854 alu
.src
[2].sel
= ctx
->temp_reg
;
2855 alu
.src
[2].chan
= i
;
2859 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
2868 static int tgsi_ssg(struct r600_shader_ctx
*ctx
)
2870 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
2871 struct r600_bytecode_alu alu
;
2874 /* tmp = (src > 0 ? 1 : src) */
2875 for (i
= 0; i
< 4; i
++) {
2876 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2877 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP3_SQ_OP3_INST_CNDGT
);
2880 alu
.dst
.sel
= ctx
->temp_reg
;
2883 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], i
);
2884 alu
.src
[1].sel
= V_SQ_ALU_SRC_1
;
2885 r600_bytecode_src(&alu
.src
[2], &ctx
->src
[0], i
);
2889 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
2894 /* dst = (-tmp > 0 ? -1 : tmp) */
2895 for (i
= 0; i
< 4; i
++) {
2896 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2897 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP3_SQ_OP3_INST_CNDGT
);
2899 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
2901 alu
.src
[0].sel
= ctx
->temp_reg
;
2902 alu
.src
[0].chan
= i
;
2905 alu
.src
[1].sel
= V_SQ_ALU_SRC_1
;
2908 alu
.src
[2].sel
= ctx
->temp_reg
;
2909 alu
.src
[2].chan
= i
;
2913 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
2920 static int tgsi_helper_copy(struct r600_shader_ctx
*ctx
, struct tgsi_full_instruction
*inst
)
2922 struct r600_bytecode_alu alu
;
2925 for (i
= 0; i
< 4; i
++) {
2926 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2927 if (!(inst
->Dst
[0].Register
.WriteMask
& (1 << i
))) {
2928 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
);
2931 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV
);
2932 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
2933 alu
.src
[0].sel
= ctx
->temp_reg
;
2934 alu
.src
[0].chan
= i
;
2939 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
2946 static int tgsi_op3(struct r600_shader_ctx
*ctx
)
2948 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
2949 struct r600_bytecode_alu alu
;
2951 int lasti
= tgsi_last_instruction(inst
->Dst
[0].Register
.WriteMask
);
2953 for (i
= 0; i
< lasti
+ 1; i
++) {
2954 if (!(inst
->Dst
[0].Register
.WriteMask
& (1 << i
)))
2957 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2958 alu
.inst
= ctx
->inst_info
->r600_opcode
;
2959 for (j
= 0; j
< inst
->Instruction
.NumSrcRegs
; j
++) {
2960 r600_bytecode_src(&alu
.src
[j
], &ctx
->src
[j
], i
);
2963 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
2970 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
2977 static int tgsi_dp(struct r600_shader_ctx
*ctx
)
2979 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
2980 struct r600_bytecode_alu alu
;
2983 for (i
= 0; i
< 4; i
++) {
2984 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2985 alu
.inst
= ctx
->inst_info
->r600_opcode
;
2986 for (j
= 0; j
< inst
->Instruction
.NumSrcRegs
; j
++) {
2987 r600_bytecode_src(&alu
.src
[j
], &ctx
->src
[j
], i
);
2990 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
2992 alu
.dst
.write
= (inst
->Dst
[0].Register
.WriteMask
>> i
) & 1;
2993 /* handle some special cases */
2994 switch (ctx
->inst_info
->tgsi_opcode
) {
2995 case TGSI_OPCODE_DP2
:
2997 alu
.src
[0].sel
= alu
.src
[1].sel
= V_SQ_ALU_SRC_0
;
2998 alu
.src
[0].chan
= alu
.src
[1].chan
= 0;
3001 case TGSI_OPCODE_DP3
:
3003 alu
.src
[0].sel
= alu
.src
[1].sel
= V_SQ_ALU_SRC_0
;
3004 alu
.src
[0].chan
= alu
.src
[1].chan
= 0;
3007 case TGSI_OPCODE_DPH
:
3009 alu
.src
[0].sel
= V_SQ_ALU_SRC_1
;
3010 alu
.src
[0].chan
= 0;
3020 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3027 static inline boolean
tgsi_tex_src_requires_loading(struct r600_shader_ctx
*ctx
,
3030 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
3031 return (inst
->Src
[index
].Register
.File
!= TGSI_FILE_TEMPORARY
&&
3032 inst
->Src
[index
].Register
.File
!= TGSI_FILE_INPUT
) ||
3033 ctx
->src
[index
].neg
|| ctx
->src
[index
].abs
;
3036 static inline unsigned tgsi_tex_get_src_gpr(struct r600_shader_ctx
*ctx
,
3039 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
3040 return ctx
->file_offset
[inst
->Src
[index
].Register
.File
] + inst
->Src
[index
].Register
.Index
;
3043 static int tgsi_tex(struct r600_shader_ctx
*ctx
)
3045 static float one_point_five
= 1.5f
;
3046 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
3047 struct r600_bytecode_tex tex
;
3048 struct r600_bytecode_alu alu
;
3052 /* Texture fetch instructions can only use gprs as source.
3053 * Also they cannot negate the source or take the absolute value */
3054 const boolean src_requires_loading
= tgsi_tex_src_requires_loading(ctx
, 0);
3055 boolean src_loaded
= FALSE
;
3056 unsigned sampler_src_reg
= 1;
3057 u8 offset_x
= 0, offset_y
= 0, offset_z
= 0;
3059 src_gpr
= tgsi_tex_get_src_gpr(ctx
, 0);
3061 if (inst
->Instruction
.Opcode
== TGSI_OPCODE_TXF
) {
3062 /* get offset values */
3063 if (inst
->Texture
.NumOffsets
) {
3064 assert(inst
->Texture
.NumOffsets
== 1);
3066 offset_x
= ctx
->literals
[inst
->TexOffsets
[0].Index
+ inst
->TexOffsets
[0].SwizzleX
] << 1;
3067 offset_y
= ctx
->literals
[inst
->TexOffsets
[0].Index
+ inst
->TexOffsets
[0].SwizzleY
] << 1;
3068 offset_z
= ctx
->literals
[inst
->TexOffsets
[0].Index
+ inst
->TexOffsets
[0].SwizzleZ
] << 1;
3070 } else if (inst
->Instruction
.Opcode
== TGSI_OPCODE_TXD
) {
3071 /* TGSI moves the sampler to src reg 3 for TXD */
3072 sampler_src_reg
= 3;
3074 for (i
= 1; i
< 3; i
++) {
3075 /* set gradients h/v */
3076 memset(&tex
, 0, sizeof(struct r600_bytecode_tex
));
3077 tex
.inst
= (i
== 1) ? SQ_TEX_INST_SET_GRADIENTS_H
:
3078 SQ_TEX_INST_SET_GRADIENTS_V
;
3079 tex
.sampler_id
= tgsi_tex_get_src_gpr(ctx
, sampler_src_reg
);
3080 tex
.resource_id
= tex
.sampler_id
+ R600_MAX_CONST_BUFFERS
;
3082 if (tgsi_tex_src_requires_loading(ctx
, i
)) {
3083 tex
.src_gpr
= r600_get_temp(ctx
);
3089 for (j
= 0; j
< 4; j
++) {
3090 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3091 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV
);
3092 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[i
], j
);
3093 alu
.dst
.sel
= tex
.src_gpr
;
3098 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3104 tex
.src_gpr
= tgsi_tex_get_src_gpr(ctx
, i
);
3105 tex
.src_sel_x
= ctx
->src
[i
].swizzle
[0];
3106 tex
.src_sel_y
= ctx
->src
[i
].swizzle
[1];
3107 tex
.src_sel_z
= ctx
->src
[i
].swizzle
[2];
3108 tex
.src_sel_w
= ctx
->src
[i
].swizzle
[3];
3109 tex
.src_rel
= ctx
->src
[i
].rel
;
3111 tex
.dst_gpr
= ctx
->temp_reg
; /* just to avoid confusing the asm scheduler */
3112 tex
.dst_sel_x
= tex
.dst_sel_y
= tex
.dst_sel_z
= tex
.dst_sel_w
= 7;
3113 if (inst
->Texture
.Texture
!= TGSI_TEXTURE_RECT
) {
3114 tex
.coord_type_x
= 1;
3115 tex
.coord_type_y
= 1;
3116 tex
.coord_type_z
= 1;
3117 tex
.coord_type_w
= 1;
3119 r
= r600_bytecode_add_tex(ctx
->bc
, &tex
);
3123 } else if (inst
->Instruction
.Opcode
== TGSI_OPCODE_TXP
) {
3125 /* Add perspective divide */
3126 if (ctx
->bc
->chip_class
== CAYMAN
) {
3128 for (i
= 0; i
< 3; i
++) {
3129 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3130 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_RECIP_IEEE
);
3131 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 3);
3133 alu
.dst
.sel
= ctx
->temp_reg
;
3139 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3146 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3147 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_RECIP_IEEE
);
3148 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 3);
3150 alu
.dst
.sel
= ctx
->temp_reg
;
3151 alu
.dst
.chan
= out_chan
;
3154 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3159 for (i
= 0; i
< 3; i
++) {
3160 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3161 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MUL
);
3162 alu
.src
[0].sel
= ctx
->temp_reg
;
3163 alu
.src
[0].chan
= out_chan
;
3164 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[0], i
);
3165 alu
.dst
.sel
= ctx
->temp_reg
;
3168 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3172 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3173 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV
);
3174 alu
.src
[0].sel
= V_SQ_ALU_SRC_1
;
3175 alu
.src
[0].chan
= 0;
3176 alu
.dst
.sel
= ctx
->temp_reg
;
3180 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3184 src_gpr
= ctx
->temp_reg
;
3187 if (inst
->Texture
.Texture
== TGSI_TEXTURE_CUBE
) {
3188 static const unsigned src0_swizzle
[] = {2, 2, 0, 1};
3189 static const unsigned src1_swizzle
[] = {1, 0, 2, 2};
3191 /* tmp1.xyzw = CUBE(R0.zzxy, R0.yxzz) */
3192 for (i
= 0; i
< 4; i
++) {
3193 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3194 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_CUBE
);
3195 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], src0_swizzle
[i
]);
3196 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[0], src1_swizzle
[i
]);
3197 alu
.dst
.sel
= ctx
->temp_reg
;
3202 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3207 /* tmp1.z = RCP_e(|tmp1.z|) */
3208 if (ctx
->bc
->chip_class
== CAYMAN
) {
3209 for (i
= 0; i
< 3; i
++) {
3210 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3211 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_RECIP_IEEE
);
3212 alu
.src
[0].sel
= ctx
->temp_reg
;
3213 alu
.src
[0].chan
= 2;
3215 alu
.dst
.sel
= ctx
->temp_reg
;
3221 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3226 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3227 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_RECIP_IEEE
);
3228 alu
.src
[0].sel
= ctx
->temp_reg
;
3229 alu
.src
[0].chan
= 2;
3231 alu
.dst
.sel
= ctx
->temp_reg
;
3235 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3240 /* MULADD R0.x, R0.x, PS1, (0x3FC00000, 1.5f).x
3241 * MULADD R0.y, R0.y, PS1, (0x3FC00000, 1.5f).x
3242 * muladd has no writemask, have to use another temp
3244 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3245 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP3_SQ_OP3_INST_MULADD
);
3248 alu
.src
[0].sel
= ctx
->temp_reg
;
3249 alu
.src
[0].chan
= 0;
3250 alu
.src
[1].sel
= ctx
->temp_reg
;
3251 alu
.src
[1].chan
= 2;
3253 alu
.src
[2].sel
= V_SQ_ALU_SRC_LITERAL
;
3254 alu
.src
[2].chan
= 0;
3255 alu
.src
[2].value
= *(uint32_t *)&one_point_five
;
3257 alu
.dst
.sel
= ctx
->temp_reg
;
3261 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3265 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3266 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP3_SQ_OP3_INST_MULADD
);
3269 alu
.src
[0].sel
= ctx
->temp_reg
;
3270 alu
.src
[0].chan
= 1;
3271 alu
.src
[1].sel
= ctx
->temp_reg
;
3272 alu
.src
[1].chan
= 2;
3274 alu
.src
[2].sel
= V_SQ_ALU_SRC_LITERAL
;
3275 alu
.src
[2].chan
= 0;
3276 alu
.src
[2].value
= *(uint32_t *)&one_point_five
;
3278 alu
.dst
.sel
= ctx
->temp_reg
;
3283 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3288 src_gpr
= ctx
->temp_reg
;
3291 if (src_requires_loading
&& !src_loaded
) {
3292 for (i
= 0; i
< 4; i
++) {
3293 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3294 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV
);
3295 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], i
);
3296 alu
.dst
.sel
= ctx
->temp_reg
;
3301 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3306 src_gpr
= ctx
->temp_reg
;
3309 opcode
= ctx
->inst_info
->r600_opcode
;
3310 if (inst
->Texture
.Texture
== TGSI_TEXTURE_SHADOW1D
||
3311 inst
->Texture
.Texture
== TGSI_TEXTURE_SHADOW2D
||
3312 inst
->Texture
.Texture
== TGSI_TEXTURE_SHADOWRECT
||
3313 inst
->Texture
.Texture
== TGSI_TEXTURE_SHADOW1D_ARRAY
||
3314 inst
->Texture
.Texture
== TGSI_TEXTURE_SHADOW2D_ARRAY
) {
3316 case SQ_TEX_INST_SAMPLE
:
3317 opcode
= SQ_TEX_INST_SAMPLE_C
;
3319 case SQ_TEX_INST_SAMPLE_L
:
3320 opcode
= SQ_TEX_INST_SAMPLE_C_L
;
3322 case SQ_TEX_INST_SAMPLE_LB
:
3323 opcode
= SQ_TEX_INST_SAMPLE_C_LB
;
3325 case SQ_TEX_INST_SAMPLE_G
:
3326 opcode
= SQ_TEX_INST_SAMPLE_C_G
;
3331 memset(&tex
, 0, sizeof(struct r600_bytecode_tex
));
3334 tex
.sampler_id
= tgsi_tex_get_src_gpr(ctx
, sampler_src_reg
);
3335 tex
.resource_id
= tex
.sampler_id
+ R600_MAX_CONST_BUFFERS
;
3336 tex
.src_gpr
= src_gpr
;
3337 tex
.dst_gpr
= ctx
->file_offset
[inst
->Dst
[0].Register
.File
] + inst
->Dst
[0].Register
.Index
;
3338 tex
.dst_sel_x
= (inst
->Dst
[0].Register
.WriteMask
& 1) ? 0 : 7;
3339 tex
.dst_sel_y
= (inst
->Dst
[0].Register
.WriteMask
& 2) ? 1 : 7;
3340 tex
.dst_sel_z
= (inst
->Dst
[0].Register
.WriteMask
& 4) ? 2 : 7;
3341 tex
.dst_sel_w
= (inst
->Dst
[0].Register
.WriteMask
& 8) ? 3 : 7;
3348 tex
.src_sel_x
= ctx
->src
[0].swizzle
[0];
3349 tex
.src_sel_y
= ctx
->src
[0].swizzle
[1];
3350 tex
.src_sel_z
= ctx
->src
[0].swizzle
[2];
3351 tex
.src_sel_w
= ctx
->src
[0].swizzle
[3];
3352 tex
.src_rel
= ctx
->src
[0].rel
;
3355 if (inst
->Texture
.Texture
== TGSI_TEXTURE_CUBE
) {
3362 if (inst
->Texture
.Texture
!= TGSI_TEXTURE_RECT
&&
3363 inst
->Texture
.Texture
!= TGSI_TEXTURE_SHADOWRECT
) {
3364 tex
.coord_type_x
= 1;
3365 tex
.coord_type_y
= 1;
3367 tex
.coord_type_z
= 1;
3368 tex
.coord_type_w
= 1;
3370 tex
.offset_x
= offset_x
;
3371 tex
.offset_y
= offset_y
;
3372 tex
.offset_z
= offset_z
;
3374 /* Put the depth for comparison in W.
3375 * TGSI_TEXTURE_SHADOW2D_ARRAY already has the depth in W.
3376 * Some instructions expect the depth in Z. */
3377 if ((inst
->Texture
.Texture
== TGSI_TEXTURE_SHADOW1D
||
3378 inst
->Texture
.Texture
== TGSI_TEXTURE_SHADOW2D
||
3379 inst
->Texture
.Texture
== TGSI_TEXTURE_SHADOWRECT
||
3380 inst
->Texture
.Texture
== TGSI_TEXTURE_SHADOW1D_ARRAY
) &&
3381 opcode
!= SQ_TEX_INST_SAMPLE_C_L
&&
3382 opcode
!= SQ_TEX_INST_SAMPLE_C_LB
) {
3383 tex
.src_sel_w
= tex
.src_sel_z
;
3386 if (inst
->Texture
.Texture
== TGSI_TEXTURE_1D_ARRAY
||
3387 inst
->Texture
.Texture
== TGSI_TEXTURE_SHADOW1D_ARRAY
) {
3388 if (opcode
== SQ_TEX_INST_SAMPLE_C_L
||
3389 opcode
== SQ_TEX_INST_SAMPLE_C_LB
) {
3390 /* the array index is read from Y */
3391 tex
.coord_type_y
= 0;
3393 /* the array index is read from Z */
3394 tex
.coord_type_z
= 0;
3395 tex
.src_sel_z
= tex
.src_sel_y
;
3397 } else if (inst
->Texture
.Texture
== TGSI_TEXTURE_2D_ARRAY
||
3398 inst
->Texture
.Texture
== TGSI_TEXTURE_SHADOW2D_ARRAY
)
3399 /* the array index is read from Z */
3400 tex
.coord_type_z
= 0;
3402 r
= r600_bytecode_add_tex(ctx
->bc
, &tex
);
3406 /* add shadow ambient support - gallium doesn't do it yet */
3410 static int tgsi_lrp(struct r600_shader_ctx
*ctx
)
3412 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
3413 struct r600_bytecode_alu alu
;
3414 int lasti
= tgsi_last_instruction(inst
->Dst
[0].Register
.WriteMask
);
3418 /* optimize if it's just an equal balance */
3419 if (ctx
->src
[0].sel
== V_SQ_ALU_SRC_0_5
) {
3420 for (i
= 0; i
< lasti
+ 1; i
++) {
3421 if (!(inst
->Dst
[0].Register
.WriteMask
& (1 << i
)))
3424 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3425 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_ADD
);
3426 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[1], i
);
3427 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[2], i
);
3429 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
3434 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3442 for (i
= 0; i
< lasti
+ 1; i
++) {
3443 if (!(inst
->Dst
[0].Register
.WriteMask
& (1 << i
)))
3446 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3447 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_ADD
);
3448 alu
.src
[0].sel
= V_SQ_ALU_SRC_1
;
3449 alu
.src
[0].chan
= 0;
3450 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[0], i
);
3451 r600_bytecode_src_toggle_neg(&alu
.src
[1]);
3452 alu
.dst
.sel
= ctx
->temp_reg
;
3458 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3463 /* (1 - src0) * src2 */
3464 for (i
= 0; i
< lasti
+ 1; i
++) {
3465 if (!(inst
->Dst
[0].Register
.WriteMask
& (1 << i
)))
3468 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3469 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MUL
);
3470 alu
.src
[0].sel
= ctx
->temp_reg
;
3471 alu
.src
[0].chan
= i
;
3472 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[2], i
);
3473 alu
.dst
.sel
= ctx
->temp_reg
;
3479 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3484 /* src0 * src1 + (1 - src0) * src2 */
3485 for (i
= 0; i
< lasti
+ 1; i
++) {
3486 if (!(inst
->Dst
[0].Register
.WriteMask
& (1 << i
)))
3489 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3490 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP3_SQ_OP3_INST_MULADD
);
3492 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], i
);
3493 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[1], i
);
3494 alu
.src
[2].sel
= ctx
->temp_reg
;
3495 alu
.src
[2].chan
= i
;
3497 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
3502 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3509 static int tgsi_cmp(struct r600_shader_ctx
*ctx
)
3511 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
3512 struct r600_bytecode_alu alu
;
3514 int lasti
= tgsi_last_instruction(inst
->Dst
[0].Register
.WriteMask
);
3516 for (i
= 0; i
< lasti
+ 1; i
++) {
3517 if (!(inst
->Dst
[0].Register
.WriteMask
& (1 << i
)))
3520 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3521 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP3_SQ_OP3_INST_CNDGE
);
3522 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], i
);
3523 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[2], i
);
3524 r600_bytecode_src(&alu
.src
[2], &ctx
->src
[1], i
);
3525 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
3531 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3538 static int tgsi_xpd(struct r600_shader_ctx
*ctx
)
3540 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
3541 static const unsigned int src0_swizzle
[] = {2, 0, 1};
3542 static const unsigned int src1_swizzle
[] = {1, 2, 0};
3543 struct r600_bytecode_alu alu
;
3544 uint32_t use_temp
= 0;
3547 if (inst
->Dst
[0].Register
.WriteMask
!= 0xf)
3550 for (i
= 0; i
< 4; i
++) {
3551 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3552 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MUL
);
3554 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], src0_swizzle
[i
]);
3555 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[1], src1_swizzle
[i
]);
3557 alu
.src
[0].sel
= V_SQ_ALU_SRC_0
;
3558 alu
.src
[0].chan
= i
;
3559 alu
.src
[1].sel
= V_SQ_ALU_SRC_0
;
3560 alu
.src
[1].chan
= i
;
3563 alu
.dst
.sel
= ctx
->temp_reg
;
3569 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3574 for (i
= 0; i
< 4; i
++) {
3575 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3576 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP3_SQ_OP3_INST_MULADD
);
3579 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], src1_swizzle
[i
]);
3580 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[1], src0_swizzle
[i
]);
3582 alu
.src
[0].sel
= V_SQ_ALU_SRC_0
;
3583 alu
.src
[0].chan
= i
;
3584 alu
.src
[1].sel
= V_SQ_ALU_SRC_0
;
3585 alu
.src
[1].chan
= i
;
3588 alu
.src
[2].sel
= ctx
->temp_reg
;
3590 alu
.src
[2].chan
= i
;
3593 alu
.dst
.sel
= ctx
->temp_reg
;
3595 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
3601 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3606 return tgsi_helper_copy(ctx
, inst
);
3610 static int tgsi_exp(struct r600_shader_ctx
*ctx
)
3612 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
3613 struct r600_bytecode_alu alu
;
3617 /* result.x = 2^floor(src); */
3618 if (inst
->Dst
[0].Register
.WriteMask
& 1) {
3619 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3621 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FLOOR
);
3622 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 0);
3624 alu
.dst
.sel
= ctx
->temp_reg
;
3628 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3632 if (ctx
->bc
->chip_class
== CAYMAN
) {
3633 for (i
= 0; i
< 3; i
++) {
3634 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_EXP_IEEE
);
3635 alu
.src
[0].sel
= ctx
->temp_reg
;
3636 alu
.src
[0].chan
= 0;
3638 alu
.dst
.sel
= ctx
->temp_reg
;
3644 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3649 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_EXP_IEEE
);
3650 alu
.src
[0].sel
= ctx
->temp_reg
;
3651 alu
.src
[0].chan
= 0;
3653 alu
.dst
.sel
= ctx
->temp_reg
;
3657 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3663 /* result.y = tmp - floor(tmp); */
3664 if ((inst
->Dst
[0].Register
.WriteMask
>> 1) & 1) {
3665 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3667 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FRACT
);
3668 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 0);
3670 alu
.dst
.sel
= ctx
->temp_reg
;
3672 r
= tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
3681 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3686 /* result.z = RoughApprox2ToX(tmp);*/
3687 if ((inst
->Dst
[0].Register
.WriteMask
>> 2) & 0x1) {
3688 if (ctx
->bc
->chip_class
== CAYMAN
) {
3689 for (i
= 0; i
< 3; i
++) {
3690 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3691 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_EXP_IEEE
);
3692 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 0);
3694 alu
.dst
.sel
= ctx
->temp_reg
;
3701 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3706 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3707 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_EXP_IEEE
);
3708 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 0);
3710 alu
.dst
.sel
= ctx
->temp_reg
;
3716 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3722 /* result.w = 1.0;*/
3723 if ((inst
->Dst
[0].Register
.WriteMask
>> 3) & 0x1) {
3724 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3726 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV
);
3727 alu
.src
[0].sel
= V_SQ_ALU_SRC_1
;
3728 alu
.src
[0].chan
= 0;
3730 alu
.dst
.sel
= ctx
->temp_reg
;
3734 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3738 return tgsi_helper_copy(ctx
, inst
);
3741 static int tgsi_log(struct r600_shader_ctx
*ctx
)
3743 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
3744 struct r600_bytecode_alu alu
;
3748 /* result.x = floor(log2(|src|)); */
3749 if (inst
->Dst
[0].Register
.WriteMask
& 1) {
3750 if (ctx
->bc
->chip_class
== CAYMAN
) {
3751 for (i
= 0; i
< 3; i
++) {
3752 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3754 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_LOG_IEEE
);
3755 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 0);
3756 r600_bytecode_src_set_abs(&alu
.src
[0]);
3758 alu
.dst
.sel
= ctx
->temp_reg
;
3764 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3770 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3772 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_LOG_IEEE
);
3773 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 0);
3774 r600_bytecode_src_set_abs(&alu
.src
[0]);
3776 alu
.dst
.sel
= ctx
->temp_reg
;
3780 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3785 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FLOOR
);
3786 alu
.src
[0].sel
= ctx
->temp_reg
;
3787 alu
.src
[0].chan
= 0;
3789 alu
.dst
.sel
= ctx
->temp_reg
;
3794 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3799 /* result.y = |src.x| / (2 ^ floor(log2(|src.x|))); */
3800 if ((inst
->Dst
[0].Register
.WriteMask
>> 1) & 1) {
3802 if (ctx
->bc
->chip_class
== CAYMAN
) {
3803 for (i
= 0; i
< 3; i
++) {
3804 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3806 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_LOG_IEEE
);
3807 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 0);
3808 r600_bytecode_src_set_abs(&alu
.src
[0]);
3810 alu
.dst
.sel
= ctx
->temp_reg
;
3817 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3822 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3824 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_LOG_IEEE
);
3825 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 0);
3826 r600_bytecode_src_set_abs(&alu
.src
[0]);
3828 alu
.dst
.sel
= ctx
->temp_reg
;
3833 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3838 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3840 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FLOOR
);
3841 alu
.src
[0].sel
= ctx
->temp_reg
;
3842 alu
.src
[0].chan
= 1;
3844 alu
.dst
.sel
= ctx
->temp_reg
;
3849 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3853 if (ctx
->bc
->chip_class
== CAYMAN
) {
3854 for (i
= 0; i
< 3; i
++) {
3855 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3856 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_EXP_IEEE
);
3857 alu
.src
[0].sel
= ctx
->temp_reg
;
3858 alu
.src
[0].chan
= 1;
3860 alu
.dst
.sel
= ctx
->temp_reg
;
3867 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3872 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3873 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_EXP_IEEE
);
3874 alu
.src
[0].sel
= ctx
->temp_reg
;
3875 alu
.src
[0].chan
= 1;
3877 alu
.dst
.sel
= ctx
->temp_reg
;
3882 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3887 if (ctx
->bc
->chip_class
== CAYMAN
) {
3888 for (i
= 0; i
< 3; i
++) {
3889 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3890 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_RECIP_IEEE
);
3891 alu
.src
[0].sel
= ctx
->temp_reg
;
3892 alu
.src
[0].chan
= 1;
3894 alu
.dst
.sel
= ctx
->temp_reg
;
3901 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3906 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3907 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_RECIP_IEEE
);
3908 alu
.src
[0].sel
= ctx
->temp_reg
;
3909 alu
.src
[0].chan
= 1;
3911 alu
.dst
.sel
= ctx
->temp_reg
;
3916 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3921 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3923 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MUL
);
3925 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 0);
3926 r600_bytecode_src_set_abs(&alu
.src
[0]);
3928 alu
.src
[1].sel
= ctx
->temp_reg
;
3929 alu
.src
[1].chan
= 1;
3931 alu
.dst
.sel
= ctx
->temp_reg
;
3936 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3941 /* result.z = log2(|src|);*/
3942 if ((inst
->Dst
[0].Register
.WriteMask
>> 2) & 1) {
3943 if (ctx
->bc
->chip_class
== CAYMAN
) {
3944 for (i
= 0; i
< 3; i
++) {
3945 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3947 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_LOG_IEEE
);
3948 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 0);
3949 r600_bytecode_src_set_abs(&alu
.src
[0]);
3951 alu
.dst
.sel
= ctx
->temp_reg
;
3958 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3963 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3965 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_LOG_IEEE
);
3966 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 0);
3967 r600_bytecode_src_set_abs(&alu
.src
[0]);
3969 alu
.dst
.sel
= ctx
->temp_reg
;
3974 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3980 /* result.w = 1.0; */
3981 if ((inst
->Dst
[0].Register
.WriteMask
>> 3) & 1) {
3982 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3984 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV
);
3985 alu
.src
[0].sel
= V_SQ_ALU_SRC_1
;
3986 alu
.src
[0].chan
= 0;
3988 alu
.dst
.sel
= ctx
->temp_reg
;
3993 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3998 return tgsi_helper_copy(ctx
, inst
);
4001 static int tgsi_eg_arl(struct r600_shader_ctx
*ctx
)
4003 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
4004 struct r600_bytecode_alu alu
;
4007 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4009 switch (inst
->Instruction
.Opcode
) {
4010 case TGSI_OPCODE_ARL
:
4011 alu
.inst
= EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FLT_TO_INT_FLOOR
;
4013 case TGSI_OPCODE_ARR
:
4014 alu
.inst
= EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FLT_TO_INT
;
4016 case TGSI_OPCODE_UARL
:
4017 alu
.inst
= EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV
;
4024 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 0);
4026 alu
.dst
.sel
= ctx
->bc
->ar_reg
;
4028 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4032 ctx
->bc
->ar_loaded
= 0;
4035 static int tgsi_r600_arl(struct r600_shader_ctx
*ctx
)
4037 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
4038 struct r600_bytecode_alu alu
;
4041 switch (inst
->Instruction
.Opcode
) {
4042 case TGSI_OPCODE_ARL
:
4043 memset(&alu
, 0, sizeof(alu
));
4044 alu
.inst
= V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FLOOR
;
4045 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 0);
4046 alu
.dst
.sel
= ctx
->bc
->ar_reg
;
4050 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
4053 memset(&alu
, 0, sizeof(alu
));
4054 alu
.inst
= V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FLT_TO_INT
;
4055 alu
.src
[0].sel
= ctx
->bc
->ar_reg
;
4056 alu
.dst
.sel
= ctx
->bc
->ar_reg
;
4060 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
4063 case TGSI_OPCODE_ARR
:
4064 memset(&alu
, 0, sizeof(alu
));
4065 alu
.inst
= V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FLT_TO_INT
;
4066 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 0);
4067 alu
.dst
.sel
= ctx
->bc
->ar_reg
;
4071 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
4074 case TGSI_OPCODE_UARL
:
4075 memset(&alu
, 0, sizeof(alu
));
4076 alu
.inst
= V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV
;
4077 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 0);
4078 alu
.dst
.sel
= ctx
->bc
->ar_reg
;
4082 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
4090 ctx
->bc
->ar_loaded
= 0;
4094 static int tgsi_opdst(struct r600_shader_ctx
*ctx
)
4096 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
4097 struct r600_bytecode_alu alu
;
4100 for (i
= 0; i
< 4; i
++) {
4101 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4103 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MUL
);
4104 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
4106 if (i
== 0 || i
== 3) {
4107 alu
.src
[0].sel
= V_SQ_ALU_SRC_1
;
4109 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], i
);
4112 if (i
== 0 || i
== 2) {
4113 alu
.src
[1].sel
= V_SQ_ALU_SRC_1
;
4115 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[1], i
);
4119 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4126 static int emit_logic_pred(struct r600_shader_ctx
*ctx
, int opcode
)
4128 struct r600_bytecode_alu alu
;
4131 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4135 alu
.dst
.sel
= ctx
->temp_reg
;
4139 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 0);
4140 alu
.src
[1].sel
= V_SQ_ALU_SRC_0
;
4141 alu
.src
[1].chan
= 0;
4145 r
= r600_bytecode_add_alu_type(ctx
->bc
, &alu
, CTX_INST(V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_PUSH_BEFORE
));
4151 static int pops(struct r600_shader_ctx
*ctx
, int pops
)
4153 unsigned force_pop
= ctx
->bc
->force_add_cf
;
4157 if (ctx
->bc
->cf_last
) {
4158 if (ctx
->bc
->cf_last
->inst
== CTX_INST(V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU
))
4160 else if (ctx
->bc
->cf_last
->inst
== CTX_INST(V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_POP_AFTER
))
4165 ctx
->bc
->cf_last
->inst
= CTX_INST(V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_POP_AFTER
);
4166 ctx
->bc
->force_add_cf
= 1;
4167 } else if (alu_pop
== 2) {
4168 ctx
->bc
->cf_last
->inst
= CTX_INST(V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_POP2_AFTER
);
4169 ctx
->bc
->force_add_cf
= 1;
4176 r600_bytecode_add_cfinst(ctx
->bc
, CTX_INST(V_SQ_CF_WORD1_SQ_CF_INST_POP
));
4177 ctx
->bc
->cf_last
->pop_count
= pops
;
4178 ctx
->bc
->cf_last
->cf_addr
= ctx
->bc
->cf_last
->id
+ 2;
4184 static inline void callstack_decrease_current(struct r600_shader_ctx
*ctx
, unsigned reason
)
4188 ctx
->bc
->callstack
[ctx
->bc
->call_sp
].current
--;
4192 ctx
->bc
->callstack
[ctx
->bc
->call_sp
].current
-= 4;
4195 /* TOODO : for 16 vp asic should -= 2; */
4196 ctx
->bc
->callstack
[ctx
->bc
->call_sp
].current
--;
4201 static inline void callstack_check_depth(struct r600_shader_ctx
*ctx
, unsigned reason
, unsigned check_max_only
)
4203 if (check_max_only
) {
4216 if ((ctx
->bc
->callstack
[ctx
->bc
->call_sp
].current
+ diff
) >
4217 ctx
->bc
->callstack
[ctx
->bc
->call_sp
].max
) {
4218 ctx
->bc
->callstack
[ctx
->bc
->call_sp
].max
=
4219 ctx
->bc
->callstack
[ctx
->bc
->call_sp
].current
+ diff
;
4225 ctx
->bc
->callstack
[ctx
->bc
->call_sp
].current
++;
4229 ctx
->bc
->callstack
[ctx
->bc
->call_sp
].current
+= 4;
4232 ctx
->bc
->callstack
[ctx
->bc
->call_sp
].current
++;
4236 if ((ctx
->bc
->callstack
[ctx
->bc
->call_sp
].current
) >
4237 ctx
->bc
->callstack
[ctx
->bc
->call_sp
].max
) {
4238 ctx
->bc
->callstack
[ctx
->bc
->call_sp
].max
=
4239 ctx
->bc
->callstack
[ctx
->bc
->call_sp
].current
;
4243 static void fc_set_mid(struct r600_shader_ctx
*ctx
, int fc_sp
)
4245 struct r600_cf_stack_entry
*sp
= &ctx
->bc
->fc_stack
[fc_sp
];
4247 sp
->mid
= (struct r600_bytecode_cf
**)realloc((void *)sp
->mid
,
4248 sizeof(struct r600_bytecode_cf
*) * (sp
->num_mid
+ 1));
4249 sp
->mid
[sp
->num_mid
] = ctx
->bc
->cf_last
;
4253 static void fc_pushlevel(struct r600_shader_ctx
*ctx
, int type
)
4256 ctx
->bc
->fc_stack
[ctx
->bc
->fc_sp
].type
= type
;
4257 ctx
->bc
->fc_stack
[ctx
->bc
->fc_sp
].start
= ctx
->bc
->cf_last
;
4260 static void fc_poplevel(struct r600_shader_ctx
*ctx
)
4262 struct r600_cf_stack_entry
*sp
= &ctx
->bc
->fc_stack
[ctx
->bc
->fc_sp
];
4274 static int emit_return(struct r600_shader_ctx
*ctx
)
4276 r600_bytecode_add_cfinst(ctx
->bc
, CTX_INST(V_SQ_CF_WORD1_SQ_CF_INST_RETURN
));
4280 static int emit_jump_to_offset(struct r600_shader_ctx
*ctx
, int pops
, int offset
)
4283 r600_bytecode_add_cfinst(ctx
->bc
, CTX_INST(V_SQ_CF_WORD1_SQ_CF_INST_JUMP
));
4284 ctx
->bc
->cf_last
->pop_count
= pops
;
4285 /* TODO work out offset */
4289 static int emit_setret_in_loop_flag(struct r600_shader_ctx
*ctx
, unsigned flag_value
)
4294 static void emit_testflag(struct r600_shader_ctx
*ctx
)
4299 static void emit_return_on_flag(struct r600_shader_ctx
*ctx
, unsigned ifidx
)
4302 emit_jump_to_offset(ctx
, 1, 4);
4303 emit_setret_in_loop_flag(ctx
, V_SQ_ALU_SRC_0
);
4304 pops(ctx
, ifidx
+ 1);
4308 static void break_loop_on_flag(struct r600_shader_ctx
*ctx
, unsigned fc_sp
)
4312 r600_bytecode_add_cfinst(ctx
->bc
, ctx
->inst_info
->r600_opcode
);
4313 ctx
->bc
->cf_last
->pop_count
= 1;
4315 fc_set_mid(ctx
, fc_sp
);
4321 static int tgsi_if(struct r600_shader_ctx
*ctx
)
4323 emit_logic_pred(ctx
, CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_PRED_SETNE_INT
));
4325 r600_bytecode_add_cfinst(ctx
->bc
, CTX_INST(V_SQ_CF_WORD1_SQ_CF_INST_JUMP
));
4327 fc_pushlevel(ctx
, FC_IF
);
4329 callstack_check_depth(ctx
, FC_PUSH_VPM
, 0);
4333 static int tgsi_else(struct r600_shader_ctx
*ctx
)
4335 r600_bytecode_add_cfinst(ctx
->bc
, CTX_INST(V_SQ_CF_WORD1_SQ_CF_INST_ELSE
));
4336 ctx
->bc
->cf_last
->pop_count
= 1;
4338 fc_set_mid(ctx
, ctx
->bc
->fc_sp
);
4339 ctx
->bc
->fc_stack
[ctx
->bc
->fc_sp
].start
->cf_addr
= ctx
->bc
->cf_last
->id
;
4343 static int tgsi_endif(struct r600_shader_ctx
*ctx
)
4346 if (ctx
->bc
->fc_stack
[ctx
->bc
->fc_sp
].type
!= FC_IF
) {
4347 R600_ERR("if/endif unbalanced in shader\n");
4351 if (ctx
->bc
->fc_stack
[ctx
->bc
->fc_sp
].mid
== NULL
) {
4352 ctx
->bc
->fc_stack
[ctx
->bc
->fc_sp
].start
->cf_addr
= ctx
->bc
->cf_last
->id
+ 2;
4353 ctx
->bc
->fc_stack
[ctx
->bc
->fc_sp
].start
->pop_count
= 1;
4355 ctx
->bc
->fc_stack
[ctx
->bc
->fc_sp
].mid
[0]->cf_addr
= ctx
->bc
->cf_last
->id
+ 2;
4359 callstack_decrease_current(ctx
, FC_PUSH_VPM
);
4363 static int tgsi_bgnloop(struct r600_shader_ctx
*ctx
)
4365 r600_bytecode_add_cfinst(ctx
->bc
, CTX_INST(V_SQ_CF_WORD1_SQ_CF_INST_LOOP_START_NO_AL
));
4367 fc_pushlevel(ctx
, FC_LOOP
);
4369 /* check stack depth */
4370 callstack_check_depth(ctx
, FC_LOOP
, 0);
4374 static int tgsi_endloop(struct r600_shader_ctx
*ctx
)
4378 r600_bytecode_add_cfinst(ctx
->bc
, CTX_INST(V_SQ_CF_WORD1_SQ_CF_INST_LOOP_END
));
4380 if (ctx
->bc
->fc_stack
[ctx
->bc
->fc_sp
].type
!= FC_LOOP
) {
4381 R600_ERR("loop/endloop in shader code are not paired.\n");
4385 /* fixup loop pointers - from r600isa
4386 LOOP END points to CF after LOOP START,
4387 LOOP START point to CF after LOOP END
4388 BRK/CONT point to LOOP END CF
4390 ctx
->bc
->cf_last
->cf_addr
= ctx
->bc
->fc_stack
[ctx
->bc
->fc_sp
].start
->id
+ 2;
4392 ctx
->bc
->fc_stack
[ctx
->bc
->fc_sp
].start
->cf_addr
= ctx
->bc
->cf_last
->id
+ 2;
4394 for (i
= 0; i
< ctx
->bc
->fc_stack
[ctx
->bc
->fc_sp
].num_mid
; i
++) {
4395 ctx
->bc
->fc_stack
[ctx
->bc
->fc_sp
].mid
[i
]->cf_addr
= ctx
->bc
->cf_last
->id
;
4397 /* TODO add LOOPRET support */
4399 callstack_decrease_current(ctx
, FC_LOOP
);
4403 static int tgsi_loop_brk_cont(struct r600_shader_ctx
*ctx
)
4407 for (fscp
= ctx
->bc
->fc_sp
; fscp
> 0; fscp
--)
4409 if (FC_LOOP
== ctx
->bc
->fc_stack
[fscp
].type
)
4414 R600_ERR("Break not inside loop/endloop pair\n");
4418 r600_bytecode_add_cfinst(ctx
->bc
, ctx
->inst_info
->r600_opcode
);
4419 ctx
->bc
->cf_last
->pop_count
= 1;
4421 fc_set_mid(ctx
, fscp
);
4424 callstack_check_depth(ctx
, FC_PUSH_VPM
, 1);
4428 static int tgsi_umad(struct r600_shader_ctx
*ctx
)
4430 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
4431 struct r600_bytecode_alu alu
;
4433 int lasti
= tgsi_last_instruction(inst
->Dst
[0].Register
.WriteMask
);
4436 for (i
= 0; i
< lasti
+ 1; i
++) {
4437 if (!(inst
->Dst
[0].Register
.WriteMask
& (1 << i
)))
4440 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4443 alu
.dst
.sel
= ctx
->temp_reg
;
4446 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MULLO_UINT
);
4447 for (j
= 0; j
< 2; j
++) {
4448 r600_bytecode_src(&alu
.src
[j
], &ctx
->src
[j
], i
);
4452 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4458 for (i
= 0; i
< lasti
+ 1; i
++) {
4459 if (!(inst
->Dst
[0].Register
.WriteMask
& (1 << i
)))
4462 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4463 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
4465 alu
.inst
= CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_ADD_INT
);
4467 alu
.src
[0].sel
= ctx
->temp_reg
;
4468 alu
.src
[0].chan
= i
;
4470 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[2], i
);
4474 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4481 static struct r600_shader_tgsi_instruction r600_shader_tgsi_instruction
[] = {
4482 {TGSI_OPCODE_ARL
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_r600_arl
},
4483 {TGSI_OPCODE_MOV
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV
, tgsi_op2
},
4484 {TGSI_OPCODE_LIT
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_lit
},
4487 * For state trackers other than OpenGL, we'll want to use
4488 * _RECIP_IEEE instead.
4490 {TGSI_OPCODE_RCP
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_RECIP_CLAMPED
, tgsi_trans_srcx_replicate
},
4492 {TGSI_OPCODE_RSQ
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_rsq
},
4493 {TGSI_OPCODE_EXP
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_exp
},
4494 {TGSI_OPCODE_LOG
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_log
},
4495 {TGSI_OPCODE_MUL
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MUL
, tgsi_op2
},
4496 {TGSI_OPCODE_ADD
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_ADD
, tgsi_op2
},
4497 {TGSI_OPCODE_DP3
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_DOT4
, tgsi_dp
},
4498 {TGSI_OPCODE_DP4
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_DOT4
, tgsi_dp
},
4499 {TGSI_OPCODE_DST
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_opdst
},
4500 {TGSI_OPCODE_MIN
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MIN
, tgsi_op2
},
4501 {TGSI_OPCODE_MAX
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MAX
, tgsi_op2
},
4502 {TGSI_OPCODE_SLT
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SETGT
, tgsi_op2_swap
},
4503 {TGSI_OPCODE_SGE
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SETGE
, tgsi_op2
},
4504 {TGSI_OPCODE_MAD
, 1, V_SQ_ALU_WORD1_OP3_SQ_OP3_INST_MULADD
, tgsi_op3
},
4505 {TGSI_OPCODE_SUB
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_ADD
, tgsi_op2
},
4506 {TGSI_OPCODE_LRP
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_lrp
},
4507 {TGSI_OPCODE_CND
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4509 {20, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4510 {TGSI_OPCODE_DP2A
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4512 {22, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4513 {23, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4514 {TGSI_OPCODE_FRC
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FRACT
, tgsi_op2
},
4515 {TGSI_OPCODE_CLAMP
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4516 {TGSI_OPCODE_FLR
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FLOOR
, tgsi_op2
},
4517 {TGSI_OPCODE_ROUND
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_RNDNE
, tgsi_op2
},
4518 {TGSI_OPCODE_EX2
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_EXP_IEEE
, tgsi_trans_srcx_replicate
},
4519 {TGSI_OPCODE_LG2
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_LOG_IEEE
, tgsi_trans_srcx_replicate
},
4520 {TGSI_OPCODE_POW
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_pow
},
4521 {TGSI_OPCODE_XPD
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_xpd
},
4523 {32, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4524 {TGSI_OPCODE_ABS
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV
, tgsi_op2
},
4525 {TGSI_OPCODE_RCC
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4526 {TGSI_OPCODE_DPH
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_DOT4
, tgsi_dp
},
4527 {TGSI_OPCODE_COS
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_COS
, tgsi_trig
},
4528 {TGSI_OPCODE_DDX
, 0, SQ_TEX_INST_GET_GRADIENTS_H
, tgsi_tex
},
4529 {TGSI_OPCODE_DDY
, 0, SQ_TEX_INST_GET_GRADIENTS_V
, tgsi_tex
},
4530 {TGSI_OPCODE_KILP
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_KILLGT
, tgsi_kill
}, /* predicated kill */
4531 {TGSI_OPCODE_PK2H
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4532 {TGSI_OPCODE_PK2US
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4533 {TGSI_OPCODE_PK4B
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4534 {TGSI_OPCODE_PK4UB
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4535 {TGSI_OPCODE_RFL
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4536 {TGSI_OPCODE_SEQ
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SETE
, tgsi_op2
},
4537 {TGSI_OPCODE_SFL
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4538 {TGSI_OPCODE_SGT
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SETGT
, tgsi_op2
},
4539 {TGSI_OPCODE_SIN
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SIN
, tgsi_trig
},
4540 {TGSI_OPCODE_SLE
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SETGE
, tgsi_op2_swap
},
4541 {TGSI_OPCODE_SNE
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SETNE
, tgsi_op2
},
4542 {TGSI_OPCODE_STR
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4543 {TGSI_OPCODE_TEX
, 0, SQ_TEX_INST_SAMPLE
, tgsi_tex
},
4544 {TGSI_OPCODE_TXD
, 0, SQ_TEX_INST_SAMPLE_G
, tgsi_tex
},
4545 {TGSI_OPCODE_TXP
, 0, SQ_TEX_INST_SAMPLE
, tgsi_tex
},
4546 {TGSI_OPCODE_UP2H
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4547 {TGSI_OPCODE_UP2US
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4548 {TGSI_OPCODE_UP4B
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4549 {TGSI_OPCODE_UP4UB
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4550 {TGSI_OPCODE_X2D
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4551 {TGSI_OPCODE_ARA
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4552 {TGSI_OPCODE_ARR
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_r600_arl
},
4553 {TGSI_OPCODE_BRA
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4554 {TGSI_OPCODE_CAL
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4555 {TGSI_OPCODE_RET
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4556 {TGSI_OPCODE_SSG
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_ssg
},
4557 {TGSI_OPCODE_CMP
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_cmp
},
4558 {TGSI_OPCODE_SCS
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_scs
},
4559 {TGSI_OPCODE_TXB
, 0, SQ_TEX_INST_SAMPLE_LB
, tgsi_tex
},
4560 {TGSI_OPCODE_NRM
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4561 {TGSI_OPCODE_DIV
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4562 {TGSI_OPCODE_DP2
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_DOT4
, tgsi_dp
},
4563 {TGSI_OPCODE_TXL
, 0, SQ_TEX_INST_SAMPLE_L
, tgsi_tex
},
4564 {TGSI_OPCODE_BRK
, 0, V_SQ_CF_WORD1_SQ_CF_INST_LOOP_BREAK
, tgsi_loop_brk_cont
},
4565 {TGSI_OPCODE_IF
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_if
},
4567 {75, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4568 {76, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4569 {TGSI_OPCODE_ELSE
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_else
},
4570 {TGSI_OPCODE_ENDIF
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_endif
},
4572 {79, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4573 {80, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4574 {TGSI_OPCODE_PUSHA
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4575 {TGSI_OPCODE_POPA
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4576 {TGSI_OPCODE_CEIL
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4577 {TGSI_OPCODE_I2F
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_INT_TO_FLT
, tgsi_op2_trans
},
4578 {TGSI_OPCODE_NOT
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOT_INT
, tgsi_op2
},
4579 {TGSI_OPCODE_TRUNC
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_TRUNC
, tgsi_op2
},
4580 {TGSI_OPCODE_SHL
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_LSHL_INT
, tgsi_op2_trans
},
4582 {88, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4583 {TGSI_OPCODE_AND
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_AND_INT
, tgsi_op2
},
4584 {TGSI_OPCODE_OR
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_OR_INT
, tgsi_op2
},
4585 {TGSI_OPCODE_MOD
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_imod
},
4586 {TGSI_OPCODE_XOR
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_XOR_INT
, tgsi_op2
},
4587 {TGSI_OPCODE_SAD
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4588 {TGSI_OPCODE_TXF
, 0, SQ_TEX_INST_LD
, tgsi_tex
},
4589 {TGSI_OPCODE_TXQ
, 0, SQ_TEX_INST_GET_TEXTURE_RESINFO
, tgsi_tex
},
4590 {TGSI_OPCODE_CONT
, 0, V_SQ_CF_WORD1_SQ_CF_INST_LOOP_CONTINUE
, tgsi_loop_brk_cont
},
4591 {TGSI_OPCODE_EMIT
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4592 {TGSI_OPCODE_ENDPRIM
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4593 {TGSI_OPCODE_BGNLOOP
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_bgnloop
},
4594 {TGSI_OPCODE_BGNSUB
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4595 {TGSI_OPCODE_ENDLOOP
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_endloop
},
4596 {TGSI_OPCODE_ENDSUB
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4598 {103, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4599 {104, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4600 {105, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4601 {106, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4602 {TGSI_OPCODE_NOP
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4604 {108, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4605 {109, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4606 {110, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4607 {111, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4608 {TGSI_OPCODE_NRM4
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4609 {TGSI_OPCODE_CALLNZ
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4610 {TGSI_OPCODE_IFC
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4611 {TGSI_OPCODE_BREAKC
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4612 {TGSI_OPCODE_KIL
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_KILLGT
, tgsi_kill
}, /* conditional kill */
4613 {TGSI_OPCODE_END
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_end
}, /* aka HALT */
4615 {118, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4616 {TGSI_OPCODE_F2I
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FLT_TO_INT
, tgsi_op2_trans
},
4617 {TGSI_OPCODE_IDIV
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_idiv
},
4618 {TGSI_OPCODE_IMAX
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MAX_INT
, tgsi_op2
},
4619 {TGSI_OPCODE_IMIN
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MIN_INT
, tgsi_op2
},
4620 {TGSI_OPCODE_INEG
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SUB_INT
, tgsi_ineg
},
4621 {TGSI_OPCODE_ISGE
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SETGE_INT
, tgsi_op2
},
4622 {TGSI_OPCODE_ISHR
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_ASHR_INT
, tgsi_op2_trans
},
4623 {TGSI_OPCODE_ISLT
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SETGT_INT
, tgsi_op2_swap
},
4624 {TGSI_OPCODE_F2U
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FLT_TO_UINT
, tgsi_op2
},
4625 {TGSI_OPCODE_U2F
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_UINT_TO_FLT
, tgsi_op2_trans
},
4626 {TGSI_OPCODE_UADD
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_ADD_INT
, tgsi_op2
},
4627 {TGSI_OPCODE_UDIV
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_udiv
},
4628 {TGSI_OPCODE_UMAD
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_umad
},
4629 {TGSI_OPCODE_UMAX
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MAX_UINT
, tgsi_op2
},
4630 {TGSI_OPCODE_UMIN
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MIN_UINT
, tgsi_op2
},
4631 {TGSI_OPCODE_UMOD
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_umod
},
4632 {TGSI_OPCODE_UMUL
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MULLO_UINT
, tgsi_op2_trans
},
4633 {TGSI_OPCODE_USEQ
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SETE_INT
, tgsi_op2
},
4634 {TGSI_OPCODE_USGE
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SETGE_UINT
, tgsi_op2
},
4635 {TGSI_OPCODE_USHR
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_LSHR_INT
, tgsi_op2_trans
},
4636 {TGSI_OPCODE_USLT
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SETGT_UINT
, tgsi_op2_swap
},
4637 {TGSI_OPCODE_USNE
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SETNE_INT
, tgsi_op2_swap
},
4638 {TGSI_OPCODE_SWITCH
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4639 {TGSI_OPCODE_CASE
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4640 {TGSI_OPCODE_DEFAULT
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4641 {TGSI_OPCODE_ENDSWITCH
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4642 {TGSI_OPCODE_LOAD
, 0, 0, tgsi_unsupported
},
4643 {TGSI_OPCODE_LOAD_MS
, 0, 0, tgsi_unsupported
},
4644 {TGSI_OPCODE_SAMPLE
, 0, 0, tgsi_unsupported
},
4645 {TGSI_OPCODE_SAMPLE_B
, 0, 0, tgsi_unsupported
},
4646 {TGSI_OPCODE_SAMPLE_C
, 0, 0, tgsi_unsupported
},
4647 {TGSI_OPCODE_SAMPLE_C_LZ
, 0, 0, tgsi_unsupported
},
4648 {TGSI_OPCODE_SAMPLE_D
, 0, 0, tgsi_unsupported
},
4649 {TGSI_OPCODE_SAMPLE_L
, 0, 0, tgsi_unsupported
},
4650 {TGSI_OPCODE_GATHER4
, 0, 0, tgsi_unsupported
},
4651 {TGSI_OPCODE_RESINFO
, 0, 0, tgsi_unsupported
},
4652 {TGSI_OPCODE_SAMPLE_POS
, 0, 0, tgsi_unsupported
},
4653 {TGSI_OPCODE_SAMPLE_INFO
, 0, 0, tgsi_unsupported
},
4654 {TGSI_OPCODE_UARL
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOVA_INT
, tgsi_r600_arl
},
4655 {TGSI_OPCODE_UCMP
, 0, 0, tgsi_unsupported
},
4656 {TGSI_OPCODE_IABS
, 0, 0, tgsi_iabs
},
4657 {TGSI_OPCODE_ISSG
, 0, 0, tgsi_issg
},
4658 {TGSI_OPCODE_LAST
, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4661 static struct r600_shader_tgsi_instruction eg_shader_tgsi_instruction
[] = {
4662 {TGSI_OPCODE_ARL
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_eg_arl
},
4663 {TGSI_OPCODE_MOV
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV
, tgsi_op2
},
4664 {TGSI_OPCODE_LIT
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_lit
},
4665 {TGSI_OPCODE_RCP
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_RECIP_IEEE
, tgsi_trans_srcx_replicate
},
4666 {TGSI_OPCODE_RSQ
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_RECIPSQRT_IEEE
, tgsi_rsq
},
4667 {TGSI_OPCODE_EXP
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_exp
},
4668 {TGSI_OPCODE_LOG
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_log
},
4669 {TGSI_OPCODE_MUL
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MUL
, tgsi_op2
},
4670 {TGSI_OPCODE_ADD
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_ADD
, tgsi_op2
},
4671 {TGSI_OPCODE_DP3
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_DOT4
, tgsi_dp
},
4672 {TGSI_OPCODE_DP4
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_DOT4
, tgsi_dp
},
4673 {TGSI_OPCODE_DST
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_opdst
},
4674 {TGSI_OPCODE_MIN
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MIN
, tgsi_op2
},
4675 {TGSI_OPCODE_MAX
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MAX
, tgsi_op2
},
4676 {TGSI_OPCODE_SLT
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SETGT
, tgsi_op2_swap
},
4677 {TGSI_OPCODE_SGE
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SETGE
, tgsi_op2
},
4678 {TGSI_OPCODE_MAD
, 1, EG_V_SQ_ALU_WORD1_OP3_SQ_OP3_INST_MULADD
, tgsi_op3
},
4679 {TGSI_OPCODE_SUB
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_ADD
, tgsi_op2
},
4680 {TGSI_OPCODE_LRP
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_lrp
},
4681 {TGSI_OPCODE_CND
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4683 {20, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4684 {TGSI_OPCODE_DP2A
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4686 {22, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4687 {23, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4688 {TGSI_OPCODE_FRC
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FRACT
, tgsi_op2
},
4689 {TGSI_OPCODE_CLAMP
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4690 {TGSI_OPCODE_FLR
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FLOOR
, tgsi_op2
},
4691 {TGSI_OPCODE_ROUND
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_RNDNE
, tgsi_op2
},
4692 {TGSI_OPCODE_EX2
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_EXP_IEEE
, tgsi_trans_srcx_replicate
},
4693 {TGSI_OPCODE_LG2
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_LOG_IEEE
, tgsi_trans_srcx_replicate
},
4694 {TGSI_OPCODE_POW
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_pow
},
4695 {TGSI_OPCODE_XPD
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_xpd
},
4697 {32, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4698 {TGSI_OPCODE_ABS
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV
, tgsi_op2
},
4699 {TGSI_OPCODE_RCC
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4700 {TGSI_OPCODE_DPH
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_DOT4
, tgsi_dp
},
4701 {TGSI_OPCODE_COS
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_COS
, tgsi_trig
},
4702 {TGSI_OPCODE_DDX
, 0, SQ_TEX_INST_GET_GRADIENTS_H
, tgsi_tex
},
4703 {TGSI_OPCODE_DDY
, 0, SQ_TEX_INST_GET_GRADIENTS_V
, tgsi_tex
},
4704 {TGSI_OPCODE_KILP
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_KILLGT
, tgsi_kill
}, /* predicated kill */
4705 {TGSI_OPCODE_PK2H
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4706 {TGSI_OPCODE_PK2US
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4707 {TGSI_OPCODE_PK4B
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4708 {TGSI_OPCODE_PK4UB
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4709 {TGSI_OPCODE_RFL
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4710 {TGSI_OPCODE_SEQ
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SETE
, tgsi_op2
},
4711 {TGSI_OPCODE_SFL
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4712 {TGSI_OPCODE_SGT
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SETGT
, tgsi_op2
},
4713 {TGSI_OPCODE_SIN
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SIN
, tgsi_trig
},
4714 {TGSI_OPCODE_SLE
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SETGE
, tgsi_op2_swap
},
4715 {TGSI_OPCODE_SNE
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SETNE
, tgsi_op2
},
4716 {TGSI_OPCODE_STR
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4717 {TGSI_OPCODE_TEX
, 0, SQ_TEX_INST_SAMPLE
, tgsi_tex
},
4718 {TGSI_OPCODE_TXD
, 0, SQ_TEX_INST_SAMPLE_G
, tgsi_tex
},
4719 {TGSI_OPCODE_TXP
, 0, SQ_TEX_INST_SAMPLE
, tgsi_tex
},
4720 {TGSI_OPCODE_UP2H
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4721 {TGSI_OPCODE_UP2US
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4722 {TGSI_OPCODE_UP4B
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4723 {TGSI_OPCODE_UP4UB
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4724 {TGSI_OPCODE_X2D
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4725 {TGSI_OPCODE_ARA
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4726 {TGSI_OPCODE_ARR
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_eg_arl
},
4727 {TGSI_OPCODE_BRA
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4728 {TGSI_OPCODE_CAL
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4729 {TGSI_OPCODE_RET
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4730 {TGSI_OPCODE_SSG
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_ssg
},
4731 {TGSI_OPCODE_CMP
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_cmp
},
4732 {TGSI_OPCODE_SCS
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_scs
},
4733 {TGSI_OPCODE_TXB
, 0, SQ_TEX_INST_SAMPLE_LB
, tgsi_tex
},
4734 {TGSI_OPCODE_NRM
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4735 {TGSI_OPCODE_DIV
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4736 {TGSI_OPCODE_DP2
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_DOT4
, tgsi_dp
},
4737 {TGSI_OPCODE_TXL
, 0, SQ_TEX_INST_SAMPLE_L
, tgsi_tex
},
4738 {TGSI_OPCODE_BRK
, 0, EG_V_SQ_CF_WORD1_SQ_CF_INST_LOOP_BREAK
, tgsi_loop_brk_cont
},
4739 {TGSI_OPCODE_IF
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_if
},
4741 {75, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4742 {76, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4743 {TGSI_OPCODE_ELSE
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_else
},
4744 {TGSI_OPCODE_ENDIF
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_endif
},
4746 {79, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4747 {80, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4748 {TGSI_OPCODE_PUSHA
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4749 {TGSI_OPCODE_POPA
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4750 {TGSI_OPCODE_CEIL
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4751 {TGSI_OPCODE_I2F
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_INT_TO_FLT
, tgsi_op2_trans
},
4752 {TGSI_OPCODE_NOT
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOT_INT
, tgsi_op2
},
4753 {TGSI_OPCODE_TRUNC
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_TRUNC
, tgsi_op2
},
4754 {TGSI_OPCODE_SHL
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_LSHL_INT
, tgsi_op2
},
4756 {88, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4757 {TGSI_OPCODE_AND
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_AND_INT
, tgsi_op2
},
4758 {TGSI_OPCODE_OR
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_OR_INT
, tgsi_op2
},
4759 {TGSI_OPCODE_MOD
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_imod
},
4760 {TGSI_OPCODE_XOR
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_XOR_INT
, tgsi_op2
},
4761 {TGSI_OPCODE_SAD
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4762 {TGSI_OPCODE_TXF
, 0, SQ_TEX_INST_LD
, tgsi_tex
},
4763 {TGSI_OPCODE_TXQ
, 0, SQ_TEX_INST_GET_TEXTURE_RESINFO
, tgsi_tex
},
4764 {TGSI_OPCODE_CONT
, 0, EG_V_SQ_CF_WORD1_SQ_CF_INST_LOOP_CONTINUE
, tgsi_loop_brk_cont
},
4765 {TGSI_OPCODE_EMIT
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4766 {TGSI_OPCODE_ENDPRIM
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4767 {TGSI_OPCODE_BGNLOOP
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_bgnloop
},
4768 {TGSI_OPCODE_BGNSUB
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4769 {TGSI_OPCODE_ENDLOOP
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_endloop
},
4770 {TGSI_OPCODE_ENDSUB
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4772 {103, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4773 {104, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4774 {105, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4775 {106, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4776 {TGSI_OPCODE_NOP
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4778 {108, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4779 {109, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4780 {110, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4781 {111, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4782 {TGSI_OPCODE_NRM4
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4783 {TGSI_OPCODE_CALLNZ
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4784 {TGSI_OPCODE_IFC
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4785 {TGSI_OPCODE_BREAKC
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4786 {TGSI_OPCODE_KIL
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_KILLGT
, tgsi_kill
}, /* conditional kill */
4787 {TGSI_OPCODE_END
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_end
}, /* aka HALT */
4789 {118, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4790 {TGSI_OPCODE_F2I
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FLT_TO_INT
, tgsi_f2i
},
4791 {TGSI_OPCODE_IDIV
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_idiv
},
4792 {TGSI_OPCODE_IMAX
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MAX_INT
, tgsi_op2
},
4793 {TGSI_OPCODE_IMIN
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MIN_INT
, tgsi_op2
},
4794 {TGSI_OPCODE_INEG
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SUB_INT
, tgsi_ineg
},
4795 {TGSI_OPCODE_ISGE
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SETGE_INT
, tgsi_op2
},
4796 {TGSI_OPCODE_ISHR
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_ASHR_INT
, tgsi_op2
},
4797 {TGSI_OPCODE_ISLT
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SETGT_INT
, tgsi_op2_swap
},
4798 {TGSI_OPCODE_F2U
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FLT_TO_UINT
, tgsi_f2i
},
4799 {TGSI_OPCODE_U2F
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_UINT_TO_FLT
, tgsi_op2_trans
},
4800 {TGSI_OPCODE_UADD
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_ADD_INT
, tgsi_op2
},
4801 {TGSI_OPCODE_UDIV
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_udiv
},
4802 {TGSI_OPCODE_UMAD
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_umad
},
4803 {TGSI_OPCODE_UMAX
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MAX_UINT
, tgsi_op2
},
4804 {TGSI_OPCODE_UMIN
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MIN_UINT
, tgsi_op2
},
4805 {TGSI_OPCODE_UMOD
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_umod
},
4806 {TGSI_OPCODE_UMUL
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MULLO_UINT
, tgsi_op2_trans
},
4807 {TGSI_OPCODE_USEQ
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SETE_INT
, tgsi_op2
},
4808 {TGSI_OPCODE_USGE
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SETGE_UINT
, tgsi_op2
},
4809 {TGSI_OPCODE_USHR
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_LSHR_INT
, tgsi_op2
},
4810 {TGSI_OPCODE_USLT
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SETGT_UINT
, tgsi_op2_swap
},
4811 {TGSI_OPCODE_USNE
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SETNE_INT
, tgsi_op2
},
4812 {TGSI_OPCODE_SWITCH
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4813 {TGSI_OPCODE_CASE
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4814 {TGSI_OPCODE_DEFAULT
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4815 {TGSI_OPCODE_ENDSWITCH
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4816 {TGSI_OPCODE_LOAD
, 0, 0, tgsi_unsupported
},
4817 {TGSI_OPCODE_LOAD_MS
, 0, 0, tgsi_unsupported
},
4818 {TGSI_OPCODE_SAMPLE
, 0, 0, tgsi_unsupported
},
4819 {TGSI_OPCODE_SAMPLE_B
, 0, 0, tgsi_unsupported
},
4820 {TGSI_OPCODE_SAMPLE_C
, 0, 0, tgsi_unsupported
},
4821 {TGSI_OPCODE_SAMPLE_C_LZ
, 0, 0, tgsi_unsupported
},
4822 {TGSI_OPCODE_SAMPLE_D
, 0, 0, tgsi_unsupported
},
4823 {TGSI_OPCODE_SAMPLE_L
, 0, 0, tgsi_unsupported
},
4824 {TGSI_OPCODE_GATHER4
, 0, 0, tgsi_unsupported
},
4825 {TGSI_OPCODE_RESINFO
, 0, 0, tgsi_unsupported
},
4826 {TGSI_OPCODE_SAMPLE_POS
, 0, 0, tgsi_unsupported
},
4827 {TGSI_OPCODE_SAMPLE_INFO
, 0, 0, tgsi_unsupported
},
4828 {TGSI_OPCODE_UARL
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOVA_INT
, tgsi_eg_arl
},
4829 {TGSI_OPCODE_UCMP
, 0, 0, tgsi_unsupported
},
4830 {TGSI_OPCODE_IABS
, 0, 0, tgsi_iabs
},
4831 {TGSI_OPCODE_ISSG
, 0, 0, tgsi_issg
},
4832 {TGSI_OPCODE_LAST
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4835 static struct r600_shader_tgsi_instruction cm_shader_tgsi_instruction
[] = {
4836 {TGSI_OPCODE_ARL
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_eg_arl
},
4837 {TGSI_OPCODE_MOV
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV
, tgsi_op2
},
4838 {TGSI_OPCODE_LIT
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_lit
},
4839 {TGSI_OPCODE_RCP
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_RECIP_IEEE
, cayman_emit_float_instr
},
4840 {TGSI_OPCODE_RSQ
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_RECIPSQRT_IEEE
, cayman_emit_float_instr
},
4841 {TGSI_OPCODE_EXP
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_exp
},
4842 {TGSI_OPCODE_LOG
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_log
},
4843 {TGSI_OPCODE_MUL
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MUL
, tgsi_op2
},
4844 {TGSI_OPCODE_ADD
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_ADD
, tgsi_op2
},
4845 {TGSI_OPCODE_DP3
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_DOT4
, tgsi_dp
},
4846 {TGSI_OPCODE_DP4
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_DOT4
, tgsi_dp
},
4847 {TGSI_OPCODE_DST
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_opdst
},
4848 {TGSI_OPCODE_MIN
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MIN
, tgsi_op2
},
4849 {TGSI_OPCODE_MAX
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MAX
, tgsi_op2
},
4850 {TGSI_OPCODE_SLT
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SETGT
, tgsi_op2_swap
},
4851 {TGSI_OPCODE_SGE
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SETGE
, tgsi_op2
},
4852 {TGSI_OPCODE_MAD
, 1, EG_V_SQ_ALU_WORD1_OP3_SQ_OP3_INST_MULADD
, tgsi_op3
},
4853 {TGSI_OPCODE_SUB
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_ADD
, tgsi_op2
},
4854 {TGSI_OPCODE_LRP
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_lrp
},
4855 {TGSI_OPCODE_CND
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4857 {20, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4858 {TGSI_OPCODE_DP2A
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4860 {22, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4861 {23, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4862 {TGSI_OPCODE_FRC
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FRACT
, tgsi_op2
},
4863 {TGSI_OPCODE_CLAMP
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4864 {TGSI_OPCODE_FLR
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FLOOR
, tgsi_op2
},
4865 {TGSI_OPCODE_ROUND
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_RNDNE
, tgsi_op2
},
4866 {TGSI_OPCODE_EX2
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_EXP_IEEE
, cayman_emit_float_instr
},
4867 {TGSI_OPCODE_LG2
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_LOG_IEEE
, cayman_emit_float_instr
},
4868 {TGSI_OPCODE_POW
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, cayman_pow
},
4869 {TGSI_OPCODE_XPD
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_xpd
},
4871 {32, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4872 {TGSI_OPCODE_ABS
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV
, tgsi_op2
},
4873 {TGSI_OPCODE_RCC
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4874 {TGSI_OPCODE_DPH
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_DOT4
, tgsi_dp
},
4875 {TGSI_OPCODE_COS
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_COS
, cayman_trig
},
4876 {TGSI_OPCODE_DDX
, 0, SQ_TEX_INST_GET_GRADIENTS_H
, tgsi_tex
},
4877 {TGSI_OPCODE_DDY
, 0, SQ_TEX_INST_GET_GRADIENTS_V
, tgsi_tex
},
4878 {TGSI_OPCODE_KILP
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_KILLGT
, tgsi_kill
}, /* predicated kill */
4879 {TGSI_OPCODE_PK2H
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4880 {TGSI_OPCODE_PK2US
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4881 {TGSI_OPCODE_PK4B
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4882 {TGSI_OPCODE_PK4UB
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4883 {TGSI_OPCODE_RFL
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4884 {TGSI_OPCODE_SEQ
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SETE
, tgsi_op2
},
4885 {TGSI_OPCODE_SFL
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4886 {TGSI_OPCODE_SGT
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SETGT
, tgsi_op2
},
4887 {TGSI_OPCODE_SIN
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SIN
, cayman_trig
},
4888 {TGSI_OPCODE_SLE
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SETGE
, tgsi_op2_swap
},
4889 {TGSI_OPCODE_SNE
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SETNE
, tgsi_op2
},
4890 {TGSI_OPCODE_STR
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4891 {TGSI_OPCODE_TEX
, 0, SQ_TEX_INST_SAMPLE
, tgsi_tex
},
4892 {TGSI_OPCODE_TXD
, 0, SQ_TEX_INST_SAMPLE_G
, tgsi_tex
},
4893 {TGSI_OPCODE_TXP
, 0, SQ_TEX_INST_SAMPLE
, tgsi_tex
},
4894 {TGSI_OPCODE_UP2H
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4895 {TGSI_OPCODE_UP2US
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4896 {TGSI_OPCODE_UP4B
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4897 {TGSI_OPCODE_UP4UB
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4898 {TGSI_OPCODE_X2D
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4899 {TGSI_OPCODE_ARA
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4900 {TGSI_OPCODE_ARR
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_eg_arl
},
4901 {TGSI_OPCODE_BRA
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4902 {TGSI_OPCODE_CAL
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4903 {TGSI_OPCODE_RET
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4904 {TGSI_OPCODE_SSG
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_ssg
},
4905 {TGSI_OPCODE_CMP
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_cmp
},
4906 {TGSI_OPCODE_SCS
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_scs
},
4907 {TGSI_OPCODE_TXB
, 0, SQ_TEX_INST_SAMPLE_LB
, tgsi_tex
},
4908 {TGSI_OPCODE_NRM
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4909 {TGSI_OPCODE_DIV
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4910 {TGSI_OPCODE_DP2
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_DOT4
, tgsi_dp
},
4911 {TGSI_OPCODE_TXL
, 0, SQ_TEX_INST_SAMPLE_L
, tgsi_tex
},
4912 {TGSI_OPCODE_BRK
, 0, EG_V_SQ_CF_WORD1_SQ_CF_INST_LOOP_BREAK
, tgsi_loop_brk_cont
},
4913 {TGSI_OPCODE_IF
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_if
},
4915 {75, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4916 {76, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4917 {TGSI_OPCODE_ELSE
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_else
},
4918 {TGSI_OPCODE_ENDIF
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_endif
},
4920 {79, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4921 {80, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4922 {TGSI_OPCODE_PUSHA
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4923 {TGSI_OPCODE_POPA
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4924 {TGSI_OPCODE_CEIL
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4925 {TGSI_OPCODE_I2F
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4926 {TGSI_OPCODE_NOT
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOT_INT
, tgsi_op2
},
4927 {TGSI_OPCODE_TRUNC
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_TRUNC
, tgsi_op2
},
4928 {TGSI_OPCODE_SHL
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4930 {88, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4931 {TGSI_OPCODE_AND
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4932 {TGSI_OPCODE_OR
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4933 {TGSI_OPCODE_MOD
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4934 {TGSI_OPCODE_XOR
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_XOR_INT
, tgsi_op2
},
4935 {TGSI_OPCODE_SAD
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4936 {TGSI_OPCODE_TXF
, 0, SQ_TEX_INST_LD
, tgsi_tex
},
4937 {TGSI_OPCODE_TXQ
, 0, SQ_TEX_INST_GET_TEXTURE_RESINFO
, tgsi_tex
},
4938 {TGSI_OPCODE_CONT
, 0, EG_V_SQ_CF_WORD1_SQ_CF_INST_LOOP_CONTINUE
, tgsi_loop_brk_cont
},
4939 {TGSI_OPCODE_EMIT
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4940 {TGSI_OPCODE_ENDPRIM
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4941 {TGSI_OPCODE_BGNLOOP
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_bgnloop
},
4942 {TGSI_OPCODE_BGNSUB
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4943 {TGSI_OPCODE_ENDLOOP
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_endloop
},
4944 {TGSI_OPCODE_ENDSUB
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4946 {103, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4947 {104, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4948 {105, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4949 {106, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4950 {TGSI_OPCODE_NOP
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4952 {108, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4953 {109, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4954 {110, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4955 {111, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4956 {TGSI_OPCODE_NRM4
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4957 {TGSI_OPCODE_CALLNZ
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4958 {TGSI_OPCODE_IFC
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4959 {TGSI_OPCODE_BREAKC
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4960 {TGSI_OPCODE_KIL
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_KILLGT
, tgsi_kill
}, /* conditional kill */
4961 {TGSI_OPCODE_END
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_end
}, /* aka HALT */
4963 {118, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4964 {TGSI_OPCODE_F2I
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4965 {TGSI_OPCODE_IDIV
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4966 {TGSI_OPCODE_IMAX
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MAX_INT
, tgsi_op2
},
4967 {TGSI_OPCODE_IMIN
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MIN_INT
, tgsi_op2
},
4968 {TGSI_OPCODE_INEG
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4969 {TGSI_OPCODE_ISGE
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4970 {TGSI_OPCODE_ISHR
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4971 {TGSI_OPCODE_ISLT
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4972 {TGSI_OPCODE_F2U
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4973 {TGSI_OPCODE_U2F
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4974 {TGSI_OPCODE_UADD
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4975 {TGSI_OPCODE_UDIV
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4976 {TGSI_OPCODE_UMAD
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4977 {TGSI_OPCODE_UMAX
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4978 {TGSI_OPCODE_UMIN
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4979 {TGSI_OPCODE_UMOD
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4980 {TGSI_OPCODE_UMUL
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4981 {TGSI_OPCODE_USEQ
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4982 {TGSI_OPCODE_USGE
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4983 {TGSI_OPCODE_USHR
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4984 {TGSI_OPCODE_USLT
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4985 {TGSI_OPCODE_USNE
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4986 {TGSI_OPCODE_SWITCH
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4987 {TGSI_OPCODE_CASE
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4988 {TGSI_OPCODE_DEFAULT
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4989 {TGSI_OPCODE_ENDSWITCH
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},
4990 {TGSI_OPCODE_LOAD
, 0, 0, tgsi_unsupported
},
4991 {TGSI_OPCODE_LOAD_MS
, 0, 0, tgsi_unsupported
},
4992 {TGSI_OPCODE_SAMPLE
, 0, 0, tgsi_unsupported
},
4993 {TGSI_OPCODE_SAMPLE_B
, 0, 0, tgsi_unsupported
},
4994 {TGSI_OPCODE_SAMPLE_C
, 0, 0, tgsi_unsupported
},
4995 {TGSI_OPCODE_SAMPLE_C_LZ
, 0, 0, tgsi_unsupported
},
4996 {TGSI_OPCODE_SAMPLE_D
, 0, 0, tgsi_unsupported
},
4997 {TGSI_OPCODE_SAMPLE_L
, 0, 0, tgsi_unsupported
},
4998 {TGSI_OPCODE_GATHER4
, 0, 0, tgsi_unsupported
},
4999 {TGSI_OPCODE_RESINFO
, 0, 0, tgsi_unsupported
},
5000 {TGSI_OPCODE_SAMPLE_POS
, 0, 0, tgsi_unsupported
},
5001 {TGSI_OPCODE_SAMPLE_INFO
, 0, 0, tgsi_unsupported
},
5002 {TGSI_OPCODE_UARL
, 0, 0, tgsi_unsupported
},
5003 {TGSI_OPCODE_UCMP
, 0, 0, tgsi_unsupported
},
5004 {TGSI_OPCODE_LAST
, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP
, tgsi_unsupported
},