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.
24 #include "r600_formats.h"
25 #include "r600_opcodes.h"
26 #include "r600_shader.h"
29 #include "sb/sb_public.h"
31 #include "pipe/p_shader_tokens.h"
32 #include "tgsi/tgsi_info.h"
33 #include "tgsi/tgsi_parse.h"
34 #include "tgsi/tgsi_scan.h"
35 #include "tgsi/tgsi_dump.h"
36 #include "util/u_bitcast.h"
37 #include "util/u_memory.h"
38 #include "util/u_math.h"
43 Why CAYMAN got loops for lots of instructions is explained here.
45 -These 8xx t-slot only ops are implemented in all vector slots.
46 MUL_LIT, FLT_TO_UINT, INT_TO_FLT, UINT_TO_FLT
47 These 8xx t-slot only opcodes become vector ops, with all four
48 slots expecting the arguments on sources a and b. Result is
49 broadcast to all channels.
50 MULLO_INT, MULHI_INT, MULLO_UINT, MULHI_UINT, MUL_64
51 These 8xx t-slot only opcodes become vector ops in the z, y, and
53 EXP_IEEE, LOG_IEEE/CLAMPED, RECIP_IEEE/CLAMPED/FF/INT/UINT/_64/CLAMPED_64
54 RECIPSQRT_IEEE/CLAMPED/FF/_64/CLAMPED_64
57 The w slot may have an independent co-issued operation, or if the
58 result is required to be in the w slot, the opcode above may be
59 issued in the w slot as well.
60 The compiler must issue the source argument to slots z, y, and x
63 /* Contents of r0 on entry to various shaders
69 GS - r0.xyw, r1.xyz = per-vertex offsets
75 .w = tess factor base.
77 TES - .x = TessCoord.x
79 - .z = RelPatchID (??)
82 PS - face_gpr.z = SampleMask
85 #define R600_SHADER_BUFFER_INFO_SEL (512 + R600_BUFFER_INFO_OFFSET / 16)
86 static int r600_shader_from_tgsi(struct r600_context
*rctx
,
87 struct r600_pipe_shader
*pipeshader
,
88 union r600_shader_key key
);
90 static void r600_add_gpr_array(struct r600_shader
*ps
, int start_gpr
,
91 int size
, unsigned comp_mask
) {
96 if (ps
->num_arrays
== ps
->max_arrays
) {
98 ps
->arrays
= realloc(ps
->arrays
, ps
->max_arrays
*
99 sizeof(struct r600_shader_array
));
102 int n
= ps
->num_arrays
;
105 ps
->arrays
[n
].comp_mask
= comp_mask
;
106 ps
->arrays
[n
].gpr_start
= start_gpr
;
107 ps
->arrays
[n
].gpr_count
= size
;
110 static void r600_dump_streamout(struct pipe_stream_output_info
*so
)
114 fprintf(stderr
, "STREAMOUT\n");
115 for (i
= 0; i
< so
->num_outputs
; i
++) {
116 unsigned mask
= ((1 << so
->output
[i
].num_components
) - 1) <<
117 so
->output
[i
].start_component
;
118 fprintf(stderr
, " %i: MEM_STREAM%d_BUF%i[%i..%i] <- OUT[%i].%s%s%s%s%s\n",
120 so
->output
[i
].stream
,
121 so
->output
[i
].output_buffer
,
122 so
->output
[i
].dst_offset
, so
->output
[i
].dst_offset
+ so
->output
[i
].num_components
- 1,
123 so
->output
[i
].register_index
,
128 so
->output
[i
].dst_offset
< so
->output
[i
].start_component
? " (will lower)" : "");
132 static int store_shader(struct pipe_context
*ctx
,
133 struct r600_pipe_shader
*shader
)
135 struct r600_context
*rctx
= (struct r600_context
*)ctx
;
138 if (shader
->bo
== NULL
) {
139 shader
->bo
= (struct r600_resource
*)
140 pipe_buffer_create(ctx
->screen
, 0, PIPE_USAGE_IMMUTABLE
, shader
->shader
.bc
.ndw
* 4);
141 if (shader
->bo
== NULL
) {
144 ptr
= r600_buffer_map_sync_with_rings(&rctx
->b
, shader
->bo
, PIPE_TRANSFER_WRITE
);
145 if (R600_BIG_ENDIAN
) {
146 for (i
= 0; i
< shader
->shader
.bc
.ndw
; ++i
) {
147 ptr
[i
] = util_cpu_to_le32(shader
->shader
.bc
.bytecode
[i
]);
150 memcpy(ptr
, shader
->shader
.bc
.bytecode
, shader
->shader
.bc
.ndw
* sizeof(*ptr
));
152 rctx
->b
.ws
->buffer_unmap(shader
->bo
->buf
);
158 int r600_pipe_shader_create(struct pipe_context
*ctx
,
159 struct r600_pipe_shader
*shader
,
160 union r600_shader_key key
)
162 struct r600_context
*rctx
= (struct r600_context
*)ctx
;
163 struct r600_pipe_shader_selector
*sel
= shader
->selector
;
165 bool dump
= r600_can_dump_shader(&rctx
->screen
->b
,
166 tgsi_get_processor_type(sel
->tokens
));
167 unsigned use_sb
= !(rctx
->screen
->b
.debug_flags
& DBG_NO_SB
);
168 unsigned sb_disasm
= use_sb
|| (rctx
->screen
->b
.debug_flags
& DBG_SB_DISASM
);
169 unsigned export_shader
;
171 shader
->shader
.bc
.isa
= rctx
->isa
;
174 fprintf(stderr
, "--------------------------------------------------------------\n");
175 tgsi_dump(sel
->tokens
, 0);
177 if (sel
->so
.num_outputs
) {
178 r600_dump_streamout(&sel
->so
);
181 r
= r600_shader_from_tgsi(rctx
, shader
, key
);
183 R600_ERR("translation from TGSI failed !\n");
186 if (shader
->shader
.processor_type
== PIPE_SHADER_VERTEX
) {
187 /* only disable for vertex shaders in tess paths */
191 use_sb
&= (shader
->shader
.processor_type
!= PIPE_SHADER_TESS_CTRL
);
192 use_sb
&= (shader
->shader
.processor_type
!= PIPE_SHADER_TESS_EVAL
);
194 /* disable SB for shaders using doubles */
195 use_sb
&= !shader
->shader
.uses_doubles
;
197 /* Check if the bytecode has already been built. */
198 if (!shader
->shader
.bc
.bytecode
) {
199 r
= r600_bytecode_build(&shader
->shader
.bc
);
201 R600_ERR("building bytecode failed !\n");
206 if (dump
&& !sb_disasm
) {
207 fprintf(stderr
, "--------------------------------------------------------------\n");
208 r600_bytecode_disasm(&shader
->shader
.bc
);
209 fprintf(stderr
, "______________________________________________________________\n");
210 } else if ((dump
&& sb_disasm
) || use_sb
) {
211 r
= r600_sb_bytecode_process(rctx
, &shader
->shader
.bc
, &shader
->shader
,
214 R600_ERR("r600_sb_bytecode_process failed !\n");
219 if (shader
->gs_copy_shader
) {
222 r
= r600_sb_bytecode_process(rctx
, &shader
->gs_copy_shader
->shader
.bc
,
223 &shader
->gs_copy_shader
->shader
, dump
, 0);
228 if ((r
= store_shader(ctx
, shader
->gs_copy_shader
)))
232 /* Store the shader in a buffer. */
233 if ((r
= store_shader(ctx
, shader
)))
237 switch (shader
->shader
.processor_type
) {
238 case PIPE_SHADER_TESS_CTRL
:
239 evergreen_update_hs_state(ctx
, shader
);
241 case PIPE_SHADER_TESS_EVAL
:
243 evergreen_update_es_state(ctx
, shader
);
245 evergreen_update_vs_state(ctx
, shader
);
247 case PIPE_SHADER_GEOMETRY
:
248 if (rctx
->b
.chip_class
>= EVERGREEN
) {
249 evergreen_update_gs_state(ctx
, shader
);
250 evergreen_update_vs_state(ctx
, shader
->gs_copy_shader
);
252 r600_update_gs_state(ctx
, shader
);
253 r600_update_vs_state(ctx
, shader
->gs_copy_shader
);
256 case PIPE_SHADER_VERTEX
:
257 export_shader
= key
.vs
.as_es
;
258 if (rctx
->b
.chip_class
>= EVERGREEN
) {
260 evergreen_update_ls_state(ctx
, shader
);
261 else if (key
.vs
.as_es
)
262 evergreen_update_es_state(ctx
, shader
);
264 evergreen_update_vs_state(ctx
, shader
);
267 r600_update_es_state(ctx
, shader
);
269 r600_update_vs_state(ctx
, shader
);
272 case PIPE_SHADER_FRAGMENT
:
273 if (rctx
->b
.chip_class
>= EVERGREEN
) {
274 evergreen_update_ps_state(ctx
, shader
);
276 r600_update_ps_state(ctx
, shader
);
286 r600_pipe_shader_destroy(ctx
, shader
);
290 void r600_pipe_shader_destroy(struct pipe_context
*ctx
, struct r600_pipe_shader
*shader
)
292 r600_resource_reference(&shader
->bo
, NULL
);
293 r600_bytecode_clear(&shader
->shader
.bc
);
294 r600_release_command_buffer(&shader
->command_buffer
);
298 * tgsi -> r600 shader
300 struct r600_shader_tgsi_instruction
;
302 struct r600_shader_src
{
309 boolean kc_rel
; /* true if cache bank is indexed */
318 struct r600_shader_ctx
{
319 struct tgsi_shader_info info
;
320 struct tgsi_parse_context parse
;
321 const struct tgsi_token
*tokens
;
323 unsigned file_offset
[TGSI_FILE_COUNT
];
325 const struct r600_shader_tgsi_instruction
*inst_info
;
326 struct r600_bytecode
*bc
;
327 struct r600_shader
*shader
;
328 struct r600_shader_src src
[4];
331 uint32_t max_driver_temp_used
;
332 /* needed for evergreen interpolation */
333 struct eg_interp eg_interpolators
[6]; // indexed by Persp/Linear * 3 + sample/center/centroid
334 /* evergreen/cayman also store sample mask in face register */
336 /* sample id is .w component stored in fixed point position register */
337 int fixed_pt_position_gpr
;
339 boolean clip_vertex_write
;
341 unsigned edgeflag_output
;
344 int next_ring_offset
;
345 int gs_out_ring_offset
;
347 struct r600_shader
*gs_for_vs
;
348 int gs_export_gpr_tregs
[4];
349 const struct pipe_stream_output_info
*gs_stream_output_info
;
350 unsigned enabled_stream_buffers_mask
;
351 unsigned tess_input_info
; /* temp with tess input offsets */
352 unsigned tess_output_info
; /* temp with tess input offsets */
355 struct r600_shader_tgsi_instruction
{
357 int (*process
)(struct r600_shader_ctx
*ctx
);
360 static int emit_gs_ring_writes(struct r600_shader_ctx
*ctx
, const struct pipe_stream_output_info
*so
, int stream
, bool ind
);
361 static const struct r600_shader_tgsi_instruction r600_shader_tgsi_instruction
[], eg_shader_tgsi_instruction
[], cm_shader_tgsi_instruction
[];
362 static int tgsi_helper_tempx_replicate(struct r600_shader_ctx
*ctx
);
363 static inline void callstack_push(struct r600_shader_ctx
*ctx
, unsigned reason
);
364 static void fc_pushlevel(struct r600_shader_ctx
*ctx
, int type
);
365 static int tgsi_else(struct r600_shader_ctx
*ctx
);
366 static int tgsi_endif(struct r600_shader_ctx
*ctx
);
367 static int tgsi_bgnloop(struct r600_shader_ctx
*ctx
);
368 static int tgsi_endloop(struct r600_shader_ctx
*ctx
);
369 static int tgsi_loop_brk_cont(struct r600_shader_ctx
*ctx
);
370 static int tgsi_fetch_rel_const(struct r600_shader_ctx
*ctx
,
371 unsigned int cb_idx
, unsigned cb_rel
, unsigned int offset
, unsigned ar_chan
,
372 unsigned int dst_reg
);
373 static void r600_bytecode_src(struct r600_bytecode_alu_src
*bc_src
,
374 const struct r600_shader_src
*shader_src
,
376 static int do_lds_fetch_values(struct r600_shader_ctx
*ctx
, unsigned temp_reg
,
379 static int tgsi_last_instruction(unsigned writemask
)
383 for (i
= 0; i
< 4; i
++) {
384 if (writemask
& (1 << i
)) {
391 static int tgsi_is_supported(struct r600_shader_ctx
*ctx
)
393 struct tgsi_full_instruction
*i
= &ctx
->parse
.FullToken
.FullInstruction
;
396 if (i
->Instruction
.NumDstRegs
> 1 && i
->Instruction
.Opcode
!= TGSI_OPCODE_DFRACEXP
) {
397 R600_ERR("too many dst (%d)\n", i
->Instruction
.NumDstRegs
);
400 if (i
->Instruction
.Predicate
) {
401 R600_ERR("predicate unsupported\n");
405 if (i
->Instruction
.Label
) {
406 R600_ERR("label unsupported\n");
410 for (j
= 0; j
< i
->Instruction
.NumSrcRegs
; j
++) {
411 if (i
->Src
[j
].Register
.Dimension
) {
412 switch (i
->Src
[j
].Register
.File
) {
413 case TGSI_FILE_CONSTANT
:
415 case TGSI_FILE_INPUT
:
416 if (ctx
->type
== PIPE_SHADER_GEOMETRY
||
417 ctx
->type
== PIPE_SHADER_TESS_CTRL
||
418 ctx
->type
== PIPE_SHADER_TESS_EVAL
)
420 case TGSI_FILE_OUTPUT
:
421 if (ctx
->type
== PIPE_SHADER_TESS_CTRL
)
424 R600_ERR("unsupported src %d (file %d, dimension %d)\n", j
,
425 i
->Src
[j
].Register
.File
,
426 i
->Src
[j
].Register
.Dimension
);
431 for (j
= 0; j
< i
->Instruction
.NumDstRegs
; j
++) {
432 if (i
->Dst
[j
].Register
.Dimension
) {
433 if (ctx
->type
== PIPE_SHADER_TESS_CTRL
)
435 R600_ERR("unsupported dst (dimension)\n");
442 int eg_get_interpolator_index(unsigned interpolate
, unsigned location
)
444 if (interpolate
== TGSI_INTERPOLATE_COLOR
||
445 interpolate
== TGSI_INTERPOLATE_LINEAR
||
446 interpolate
== TGSI_INTERPOLATE_PERSPECTIVE
)
448 int is_linear
= interpolate
== TGSI_INTERPOLATE_LINEAR
;
452 case TGSI_INTERPOLATE_LOC_CENTER
:
455 case TGSI_INTERPOLATE_LOC_CENTROID
:
458 case TGSI_INTERPOLATE_LOC_SAMPLE
:
463 return is_linear
* 3 + loc
;
469 static void evergreen_interp_assign_ij_index(struct r600_shader_ctx
*ctx
,
472 int i
= eg_get_interpolator_index(
473 ctx
->shader
->input
[input
].interpolate
,
474 ctx
->shader
->input
[input
].interpolate_location
);
476 ctx
->shader
->input
[input
].ij_index
= ctx
->eg_interpolators
[i
].ij_index
;
479 static int evergreen_interp_alu(struct r600_shader_ctx
*ctx
, int input
)
482 struct r600_bytecode_alu alu
;
483 int gpr
= 0, base_chan
= 0;
484 int ij_index
= ctx
->shader
->input
[input
].ij_index
;
486 /* work out gpr and base_chan from index */
488 base_chan
= (2 * (ij_index
% 2)) + 1;
490 for (i
= 0; i
< 8; i
++) {
491 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
494 alu
.op
= ALU_OP2_INTERP_ZW
;
496 alu
.op
= ALU_OP2_INTERP_XY
;
498 if ((i
> 1) && (i
< 6)) {
499 alu
.dst
.sel
= ctx
->shader
->input
[input
].gpr
;
503 alu
.dst
.chan
= i
% 4;
505 alu
.src
[0].sel
= gpr
;
506 alu
.src
[0].chan
= (base_chan
- (i
% 2));
508 alu
.src
[1].sel
= V_SQ_ALU_SRC_PARAM_BASE
+ ctx
->shader
->input
[input
].lds_pos
;
510 alu
.bank_swizzle_force
= SQ_ALU_VEC_210
;
513 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
520 static int evergreen_interp_flat(struct r600_shader_ctx
*ctx
, int input
)
523 struct r600_bytecode_alu alu
;
525 for (i
= 0; i
< 4; i
++) {
526 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
528 alu
.op
= ALU_OP1_INTERP_LOAD_P0
;
530 alu
.dst
.sel
= ctx
->shader
->input
[input
].gpr
;
535 alu
.src
[0].sel
= V_SQ_ALU_SRC_PARAM_BASE
+ ctx
->shader
->input
[input
].lds_pos
;
540 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
548 * Special export handling in shaders
550 * shader export ARRAY_BASE for EXPORT_POS:
553 * 62, 63 are clip distance vectors
555 * The use of the values exported in 61-63 are controlled by PA_CL_VS_OUT_CNTL:
556 * VS_OUT_MISC_VEC_ENA - enables the use of all fields in export 61
557 * USE_VTX_POINT_SIZE - point size in the X channel of export 61
558 * USE_VTX_EDGE_FLAG - edge flag in the Y channel of export 61
559 * USE_VTX_RENDER_TARGET_INDX - render target index in the Z channel of export 61
560 * USE_VTX_VIEWPORT_INDX - viewport index in the W channel of export 61
561 * USE_VTX_KILL_FLAG - kill flag in the Z channel of export 61 (mutually
562 * exclusive from render target index)
563 * VS_OUT_CCDIST0_VEC_ENA/VS_OUT_CCDIST1_VEC_ENA - enable clip distance vectors
566 * shader export ARRAY_BASE for EXPORT_PIXEL:
568 * 61 computed Z vector
570 * The use of the values exported in the computed Z vector are controlled
571 * by DB_SHADER_CONTROL:
572 * Z_EXPORT_ENABLE - Z as a float in RED
573 * STENCIL_REF_EXPORT_ENABLE - stencil ref as int in GREEN
574 * COVERAGE_TO_MASK_ENABLE - alpha to mask in ALPHA
575 * MASK_EXPORT_ENABLE - pixel sample mask in BLUE
576 * DB_SOURCE_FORMAT - export control restrictions
581 /* Map name/sid pair from tgsi to the 8-bit semantic index for SPI setup */
582 static int r600_spi_sid(struct r600_shader_io
* io
)
584 int index
, name
= io
->name
;
586 /* These params are handled differently, they don't need
587 * semantic indices, so we'll use 0 for them.
589 if (name
== TGSI_SEMANTIC_POSITION
||
590 name
== TGSI_SEMANTIC_PSIZE
||
591 name
== TGSI_SEMANTIC_EDGEFLAG
||
592 name
== TGSI_SEMANTIC_FACE
||
593 name
== TGSI_SEMANTIC_SAMPLEMASK
)
596 if (name
== TGSI_SEMANTIC_GENERIC
) {
597 /* For generic params simply use sid from tgsi */
600 /* For non-generic params - pack name and sid into 8 bits */
601 index
= 0x80 | (name
<<3) | (io
->sid
);
604 /* Make sure that all really used indices have nonzero value, so
605 * we can just compare it to 0 later instead of comparing the name
606 * with different values to detect special cases. */
613 /* we need this to get a common lds index for vs/tcs/tes input/outputs */
614 int r600_get_lds_unique_index(unsigned semantic_name
, unsigned index
)
616 switch (semantic_name
) {
617 case TGSI_SEMANTIC_POSITION
:
619 case TGSI_SEMANTIC_PSIZE
:
621 case TGSI_SEMANTIC_CLIPDIST
:
624 case TGSI_SEMANTIC_GENERIC
:
626 return 4 + index
- 9;
628 /* same explanation as in the default statement,
629 * the only user hitting this is st/nine.
633 /* patch indices are completely separate and thus start from 0 */
634 case TGSI_SEMANTIC_TESSOUTER
:
636 case TGSI_SEMANTIC_TESSINNER
:
638 case TGSI_SEMANTIC_PATCH
:
642 /* Don't fail here. The result of this function is only used
643 * for LS, TCS, TES, and GS, where legacy GL semantics can't
644 * occur, but this function is called for all vertex shaders
645 * before it's known whether LS will be compiled or not.
651 /* turn input into interpolate on EG */
652 static int evergreen_interp_input(struct r600_shader_ctx
*ctx
, int index
)
656 if (ctx
->shader
->input
[index
].spi_sid
) {
657 ctx
->shader
->input
[index
].lds_pos
= ctx
->shader
->nlds
++;
658 if (ctx
->shader
->input
[index
].interpolate
> 0) {
659 evergreen_interp_assign_ij_index(ctx
, index
);
660 r
= evergreen_interp_alu(ctx
, index
);
662 r
= evergreen_interp_flat(ctx
, index
);
668 static int select_twoside_color(struct r600_shader_ctx
*ctx
, int front
, int back
)
670 struct r600_bytecode_alu alu
;
672 int gpr_front
= ctx
->shader
->input
[front
].gpr
;
673 int gpr_back
= ctx
->shader
->input
[back
].gpr
;
675 for (i
= 0; i
< 4; i
++) {
676 memset(&alu
, 0, sizeof(alu
));
677 alu
.op
= ALU_OP3_CNDGT
;
680 alu
.dst
.sel
= gpr_front
;
681 alu
.src
[0].sel
= ctx
->face_gpr
;
682 alu
.src
[1].sel
= gpr_front
;
683 alu
.src
[2].sel
= gpr_back
;
690 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
697 /* execute a single slot ALU calculation */
698 static int single_alu_op2(struct r600_shader_ctx
*ctx
, int op
,
699 int dst_sel
, int dst_chan
,
700 int src0_sel
, unsigned src0_chan_val
,
701 int src1_sel
, unsigned src1_chan_val
)
703 struct r600_bytecode_alu alu
;
706 if (ctx
->bc
->chip_class
== CAYMAN
&& op
== ALU_OP2_MULLO_INT
) {
707 for (i
= 0; i
< 4; i
++) {
708 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
710 alu
.src
[0].sel
= src0_sel
;
711 if (src0_sel
== V_SQ_ALU_SRC_LITERAL
)
712 alu
.src
[0].value
= src0_chan_val
;
714 alu
.src
[0].chan
= src0_chan_val
;
715 alu
.src
[1].sel
= src1_sel
;
716 if (src1_sel
== V_SQ_ALU_SRC_LITERAL
)
717 alu
.src
[1].value
= src1_chan_val
;
719 alu
.src
[1].chan
= src1_chan_val
;
720 alu
.dst
.sel
= dst_sel
;
722 alu
.dst
.write
= i
== dst_chan
;
724 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
731 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
733 alu
.src
[0].sel
= src0_sel
;
734 if (src0_sel
== V_SQ_ALU_SRC_LITERAL
)
735 alu
.src
[0].value
= src0_chan_val
;
737 alu
.src
[0].chan
= src0_chan_val
;
738 alu
.src
[1].sel
= src1_sel
;
739 if (src1_sel
== V_SQ_ALU_SRC_LITERAL
)
740 alu
.src
[1].value
= src1_chan_val
;
742 alu
.src
[1].chan
= src1_chan_val
;
743 alu
.dst
.sel
= dst_sel
;
744 alu
.dst
.chan
= dst_chan
;
747 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
753 /* execute a single slot ALU calculation */
754 static int single_alu_op3(struct r600_shader_ctx
*ctx
, int op
,
755 int dst_sel
, int dst_chan
,
756 int src0_sel
, unsigned src0_chan_val
,
757 int src1_sel
, unsigned src1_chan_val
,
758 int src2_sel
, unsigned src2_chan_val
)
760 struct r600_bytecode_alu alu
;
763 /* validate this for other ops */
764 assert(op
== ALU_OP3_MULADD_UINT24
);
765 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
767 alu
.src
[0].sel
= src0_sel
;
768 if (src0_sel
== V_SQ_ALU_SRC_LITERAL
)
769 alu
.src
[0].value
= src0_chan_val
;
771 alu
.src
[0].chan
= src0_chan_val
;
772 alu
.src
[1].sel
= src1_sel
;
773 if (src1_sel
== V_SQ_ALU_SRC_LITERAL
)
774 alu
.src
[1].value
= src1_chan_val
;
776 alu
.src
[1].chan
= src1_chan_val
;
777 alu
.src
[2].sel
= src2_sel
;
778 if (src2_sel
== V_SQ_ALU_SRC_LITERAL
)
779 alu
.src
[2].value
= src2_chan_val
;
781 alu
.src
[2].chan
= src2_chan_val
;
782 alu
.dst
.sel
= dst_sel
;
783 alu
.dst
.chan
= dst_chan
;
786 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
792 /* put it in temp_reg.x */
793 static int get_lds_offset0(struct r600_shader_ctx
*ctx
,
795 int temp_reg
, bool is_patch_var
)
799 /* MUL temp.x, patch_stride (input_vals.x), rel_patch_id (r0.y (tcs)) */
801 Dimension - patch0_offset (input_vals.z),
802 Non-dim - patch0_data_offset (input_vals.w)
804 r
= single_alu_op3(ctx
, ALU_OP3_MULADD_UINT24
,
806 ctx
->tess_output_info
, 0,
808 ctx
->tess_output_info
, is_patch_var
? 3 : 2);
814 static inline int get_address_file_reg(struct r600_shader_ctx
*ctx
, int index
)
816 return index
> 0 ? ctx
->bc
->index_reg
[index
- 1] : ctx
->bc
->ar_reg
;
819 static int r600_get_temp(struct r600_shader_ctx
*ctx
)
821 return ctx
->temp_reg
+ ctx
->max_driver_temp_used
++;
824 static int vs_add_primid_output(struct r600_shader_ctx
*ctx
, int prim_id_sid
)
827 i
= ctx
->shader
->noutput
++;
828 ctx
->shader
->output
[i
].name
= TGSI_SEMANTIC_PRIMID
;
829 ctx
->shader
->output
[i
].sid
= 0;
830 ctx
->shader
->output
[i
].gpr
= 0;
831 ctx
->shader
->output
[i
].interpolate
= TGSI_INTERPOLATE_CONSTANT
;
832 ctx
->shader
->output
[i
].write_mask
= 0x4;
833 ctx
->shader
->output
[i
].spi_sid
= prim_id_sid
;
838 static int tgsi_barrier(struct r600_shader_ctx
*ctx
)
840 struct r600_bytecode_alu alu
;
843 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
844 alu
.op
= ctx
->inst_info
->op
;
847 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
853 static int tgsi_declaration(struct r600_shader_ctx
*ctx
)
855 struct tgsi_full_declaration
*d
= &ctx
->parse
.FullToken
.FullDeclaration
;
856 int r
, i
, j
, count
= d
->Range
.Last
- d
->Range
.First
+ 1;
858 switch (d
->Declaration
.File
) {
859 case TGSI_FILE_INPUT
:
860 for (j
= 0; j
< count
; j
++) {
861 i
= ctx
->shader
->ninput
+ j
;
862 assert(i
< ARRAY_SIZE(ctx
->shader
->input
));
863 ctx
->shader
->input
[i
].name
= d
->Semantic
.Name
;
864 ctx
->shader
->input
[i
].sid
= d
->Semantic
.Index
+ j
;
865 ctx
->shader
->input
[i
].interpolate
= d
->Interp
.Interpolate
;
866 ctx
->shader
->input
[i
].interpolate_location
= d
->Interp
.Location
;
867 ctx
->shader
->input
[i
].gpr
= ctx
->file_offset
[TGSI_FILE_INPUT
] + d
->Range
.First
+ j
;
868 if (ctx
->type
== PIPE_SHADER_FRAGMENT
) {
869 ctx
->shader
->input
[i
].spi_sid
= r600_spi_sid(&ctx
->shader
->input
[i
]);
870 switch (ctx
->shader
->input
[i
].name
) {
871 case TGSI_SEMANTIC_FACE
:
872 if (ctx
->face_gpr
!= -1)
873 ctx
->shader
->input
[i
].gpr
= ctx
->face_gpr
; /* already allocated by allocate_system_value_inputs */
875 ctx
->face_gpr
= ctx
->shader
->input
[i
].gpr
;
877 case TGSI_SEMANTIC_COLOR
:
880 case TGSI_SEMANTIC_POSITION
:
881 ctx
->fragcoord_input
= i
;
883 case TGSI_SEMANTIC_PRIMID
:
884 /* set this for now */
885 ctx
->shader
->gs_prim_id_input
= true;
886 ctx
->shader
->ps_prim_id_input
= i
;
889 if (ctx
->bc
->chip_class
>= EVERGREEN
) {
890 if ((r
= evergreen_interp_input(ctx
, i
)))
893 } else if (ctx
->type
== PIPE_SHADER_GEOMETRY
) {
894 /* FIXME probably skip inputs if they aren't passed in the ring */
895 ctx
->shader
->input
[i
].ring_offset
= ctx
->next_ring_offset
;
896 ctx
->next_ring_offset
+= 16;
897 if (ctx
->shader
->input
[i
].name
== TGSI_SEMANTIC_PRIMID
)
898 ctx
->shader
->gs_prim_id_input
= true;
901 ctx
->shader
->ninput
+= count
;
903 case TGSI_FILE_OUTPUT
:
904 for (j
= 0; j
< count
; j
++) {
905 i
= ctx
->shader
->noutput
+ j
;
906 assert(i
< ARRAY_SIZE(ctx
->shader
->output
));
907 ctx
->shader
->output
[i
].name
= d
->Semantic
.Name
;
908 ctx
->shader
->output
[i
].sid
= d
->Semantic
.Index
+ j
;
909 ctx
->shader
->output
[i
].gpr
= ctx
->file_offset
[TGSI_FILE_OUTPUT
] + d
->Range
.First
+ j
;
910 ctx
->shader
->output
[i
].interpolate
= d
->Interp
.Interpolate
;
911 ctx
->shader
->output
[i
].write_mask
= d
->Declaration
.UsageMask
;
912 if (ctx
->type
== PIPE_SHADER_VERTEX
||
913 ctx
->type
== PIPE_SHADER_GEOMETRY
||
914 ctx
->type
== PIPE_SHADER_TESS_EVAL
) {
915 ctx
->shader
->output
[i
].spi_sid
= r600_spi_sid(&ctx
->shader
->output
[i
]);
916 switch (d
->Semantic
.Name
) {
917 case TGSI_SEMANTIC_CLIPDIST
:
918 ctx
->shader
->clip_dist_write
|= d
->Declaration
.UsageMask
<<
919 ((d
->Semantic
.Index
+ j
) << 2);
921 case TGSI_SEMANTIC_PSIZE
:
922 ctx
->shader
->vs_out_misc_write
= 1;
923 ctx
->shader
->vs_out_point_size
= 1;
925 case TGSI_SEMANTIC_EDGEFLAG
:
926 ctx
->shader
->vs_out_misc_write
= 1;
927 ctx
->shader
->vs_out_edgeflag
= 1;
928 ctx
->edgeflag_output
= i
;
930 case TGSI_SEMANTIC_VIEWPORT_INDEX
:
931 ctx
->shader
->vs_out_misc_write
= 1;
932 ctx
->shader
->vs_out_viewport
= 1;
934 case TGSI_SEMANTIC_LAYER
:
935 ctx
->shader
->vs_out_misc_write
= 1;
936 ctx
->shader
->vs_out_layer
= 1;
938 case TGSI_SEMANTIC_CLIPVERTEX
:
939 ctx
->clip_vertex_write
= TRUE
;
943 if (ctx
->type
== PIPE_SHADER_GEOMETRY
) {
944 ctx
->gs_out_ring_offset
+= 16;
946 } else if (ctx
->type
== PIPE_SHADER_FRAGMENT
) {
947 switch (d
->Semantic
.Name
) {
948 case TGSI_SEMANTIC_COLOR
:
949 ctx
->shader
->nr_ps_max_color_exports
++;
954 ctx
->shader
->noutput
+= count
;
956 case TGSI_FILE_TEMPORARY
:
957 if (ctx
->info
.indirect_files
& (1 << TGSI_FILE_TEMPORARY
)) {
958 if (d
->Array
.ArrayID
) {
959 r600_add_gpr_array(ctx
->shader
,
960 ctx
->file_offset
[TGSI_FILE_TEMPORARY
] +
962 d
->Range
.Last
- d
->Range
.First
+ 1, 0x0F);
967 case TGSI_FILE_CONSTANT
:
968 case TGSI_FILE_SAMPLER
:
969 case TGSI_FILE_SAMPLER_VIEW
:
970 case TGSI_FILE_ADDRESS
:
973 case TGSI_FILE_SYSTEM_VALUE
:
974 if (d
->Semantic
.Name
== TGSI_SEMANTIC_SAMPLEMASK
||
975 d
->Semantic
.Name
== TGSI_SEMANTIC_SAMPLEID
||
976 d
->Semantic
.Name
== TGSI_SEMANTIC_SAMPLEPOS
) {
977 break; /* Already handled from allocate_system_value_inputs */
978 } else if (d
->Semantic
.Name
== TGSI_SEMANTIC_INSTANCEID
) {
979 if (!ctx
->native_integers
) {
980 struct r600_bytecode_alu alu
;
981 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
983 alu
.op
= ALU_OP1_INT_TO_FLT
;
992 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
996 } else if (d
->Semantic
.Name
== TGSI_SEMANTIC_VERTEXID
)
998 else if (d
->Semantic
.Name
== TGSI_SEMANTIC_INVOCATIONID
)
1000 else if (d
->Semantic
.Name
== TGSI_SEMANTIC_TESSINNER
||
1001 d
->Semantic
.Name
== TGSI_SEMANTIC_TESSOUTER
) {
1002 int param
= r600_get_lds_unique_index(d
->Semantic
.Name
, 0);
1003 int dreg
= d
->Semantic
.Name
== TGSI_SEMANTIC_TESSINNER
? 3 : 2;
1004 unsigned temp_reg
= r600_get_temp(ctx
);
1006 r
= get_lds_offset0(ctx
, 2, temp_reg
, true);
1010 r
= single_alu_op2(ctx
, ALU_OP2_ADD_INT
,
1013 V_SQ_ALU_SRC_LITERAL
, param
* 16);
1017 do_lds_fetch_values(ctx
, temp_reg
, dreg
);
1019 else if (d
->Semantic
.Name
== TGSI_SEMANTIC_TESSCOORD
) {
1023 for (i
= 0; i
< 2; i
++) {
1024 struct r600_bytecode_alu alu
;
1025 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
1026 alu
.op
= ALU_OP1_MOV
;
1028 alu
.src
[0].chan
= 0 + i
;
1030 alu
.dst
.chan
= 0 + i
;
1032 alu
.last
= (i
== 1) ? 1 : 0;
1033 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
1036 /* ADD r1.z, 1.0f, -r0.x */
1037 struct r600_bytecode_alu alu
;
1038 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
1039 alu
.op
= ALU_OP2_ADD
;
1040 alu
.src
[0].sel
= V_SQ_ALU_SRC_1
;
1042 alu
.src
[1].chan
= 0;
1048 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
1051 /* ADD r1.z, r1.z, -r1.y */
1052 alu
.op
= ALU_OP2_ADD
;
1054 alu
.src
[0].chan
= 2;
1056 alu
.src
[1].chan
= 1;
1062 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
1068 R600_ERR("unsupported file %d declaration\n", d
->Declaration
.File
);
1074 static int allocate_system_value_inputs(struct r600_shader_ctx
*ctx
, int gpr_offset
)
1076 struct tgsi_parse_context parse
;
1080 unsigned name
, alternate_name
;
1082 { false, &ctx
->face_gpr
, TGSI_SEMANTIC_SAMPLEMASK
, ~0u }, /* lives in Front Face GPR.z */
1084 { false, &ctx
->fixed_pt_position_gpr
, TGSI_SEMANTIC_SAMPLEID
, TGSI_SEMANTIC_SAMPLEPOS
} /* SAMPLEID is in Fixed Point Position GPR.w */
1086 int i
, k
, num_regs
= 0;
1088 if (tgsi_parse_init(&parse
, ctx
->tokens
) != TGSI_PARSE_OK
) {
1092 /* need to scan shader for system values and interpolateAtSample/Offset/Centroid */
1093 while (!tgsi_parse_end_of_tokens(&parse
)) {
1094 tgsi_parse_token(&parse
);
1096 if (parse
.FullToken
.Token
.Type
== TGSI_TOKEN_TYPE_INSTRUCTION
) {
1097 const struct tgsi_full_instruction
*inst
= &parse
.FullToken
.FullInstruction
;
1098 if (inst
->Instruction
.Opcode
== TGSI_OPCODE_INTERP_SAMPLE
||
1099 inst
->Instruction
.Opcode
== TGSI_OPCODE_INTERP_OFFSET
||
1100 inst
->Instruction
.Opcode
== TGSI_OPCODE_INTERP_CENTROID
)
1102 int interpolate
, location
, k
;
1104 if (inst
->Instruction
.Opcode
== TGSI_OPCODE_INTERP_SAMPLE
) {
1105 location
= TGSI_INTERPOLATE_LOC_CENTER
;
1106 inputs
[1].enabled
= true; /* needs SAMPLEID */
1107 } else if (inst
->Instruction
.Opcode
== TGSI_OPCODE_INTERP_OFFSET
) {
1108 location
= TGSI_INTERPOLATE_LOC_CENTER
;
1109 /* Needs sample positions, currently those are always available */
1111 location
= TGSI_INTERPOLATE_LOC_CENTROID
;
1114 interpolate
= ctx
->info
.input_interpolate
[inst
->Src
[0].Register
.Index
];
1115 k
= eg_get_interpolator_index(interpolate
, location
);
1116 ctx
->eg_interpolators
[k
].enabled
= true;
1118 } else if (parse
.FullToken
.Token
.Type
== TGSI_TOKEN_TYPE_DECLARATION
) {
1119 struct tgsi_full_declaration
*d
= &parse
.FullToken
.FullDeclaration
;
1120 if (d
->Declaration
.File
== TGSI_FILE_SYSTEM_VALUE
) {
1121 for (k
= 0; k
< ARRAY_SIZE(inputs
); k
++) {
1122 if (d
->Semantic
.Name
== inputs
[k
].name
||
1123 d
->Semantic
.Name
== inputs
[k
].alternate_name
) {
1124 inputs
[k
].enabled
= true;
1131 tgsi_parse_free(&parse
);
1133 for (i
= 0; i
< ARRAY_SIZE(inputs
); i
++) {
1134 boolean enabled
= inputs
[i
].enabled
;
1135 int *reg
= inputs
[i
].reg
;
1136 unsigned name
= inputs
[i
].name
;
1139 int gpr
= gpr_offset
+ num_regs
++;
1141 // add to inputs, allocate a gpr
1142 k
= ctx
->shader
->ninput
++;
1143 ctx
->shader
->input
[k
].name
= name
;
1144 ctx
->shader
->input
[k
].sid
= 0;
1145 ctx
->shader
->input
[k
].interpolate
= TGSI_INTERPOLATE_CONSTANT
;
1146 ctx
->shader
->input
[k
].interpolate_location
= TGSI_INTERPOLATE_LOC_CENTER
;
1147 *reg
= ctx
->shader
->input
[k
].gpr
= gpr
;
1151 return gpr_offset
+ num_regs
;
1155 * for evergreen we need to scan the shader to find the number of GPRs we need to
1156 * reserve for interpolation and system values
1158 * we need to know if we are going to emit
1159 * any sample or centroid inputs
1160 * if perspective and linear are required
1162 static int evergreen_gpr_count(struct r600_shader_ctx
*ctx
)
1166 struct tgsi_parse_context parse
;
1168 memset(&ctx
->eg_interpolators
, 0, sizeof(ctx
->eg_interpolators
));
1170 for (i
= 0; i
< ctx
->info
.num_inputs
; i
++) {
1172 /* skip position/face/mask/sampleid */
1173 if (ctx
->info
.input_semantic_name
[i
] == TGSI_SEMANTIC_POSITION
||
1174 ctx
->info
.input_semantic_name
[i
] == TGSI_SEMANTIC_FACE
||
1175 ctx
->info
.input_semantic_name
[i
] == TGSI_SEMANTIC_SAMPLEMASK
||
1176 ctx
->info
.input_semantic_name
[i
] == TGSI_SEMANTIC_SAMPLEID
)
1179 k
= eg_get_interpolator_index(
1180 ctx
->info
.input_interpolate
[i
],
1181 ctx
->info
.input_interpolate_loc
[i
]);
1183 ctx
->eg_interpolators
[k
].enabled
= TRUE
;
1186 if (tgsi_parse_init(&parse
, ctx
->tokens
) != TGSI_PARSE_OK
) {
1190 /* need to scan shader for system values and interpolateAtSample/Offset/Centroid */
1191 while (!tgsi_parse_end_of_tokens(&parse
)) {
1192 tgsi_parse_token(&parse
);
1194 if (parse
.FullToken
.Token
.Type
== TGSI_TOKEN_TYPE_INSTRUCTION
) {
1195 const struct tgsi_full_instruction
*inst
= &parse
.FullToken
.FullInstruction
;
1196 if (inst
->Instruction
.Opcode
== TGSI_OPCODE_INTERP_SAMPLE
||
1197 inst
->Instruction
.Opcode
== TGSI_OPCODE_INTERP_OFFSET
||
1198 inst
->Instruction
.Opcode
== TGSI_OPCODE_INTERP_CENTROID
)
1200 int interpolate
, location
, k
;
1202 if (inst
->Instruction
.Opcode
== TGSI_OPCODE_INTERP_SAMPLE
) {
1203 location
= TGSI_INTERPOLATE_LOC_CENTER
;
1204 } else if (inst
->Instruction
.Opcode
== TGSI_OPCODE_INTERP_OFFSET
) {
1205 location
= TGSI_INTERPOLATE_LOC_CENTER
;
1207 location
= TGSI_INTERPOLATE_LOC_CENTROID
;
1210 interpolate
= ctx
->info
.input_interpolate
[inst
->Src
[0].Register
.Index
];
1211 k
= eg_get_interpolator_index(interpolate
, location
);
1212 ctx
->eg_interpolators
[k
].enabled
= true;
1217 tgsi_parse_free(&parse
);
1219 /* assign gpr to each interpolator according to priority */
1221 for (i
= 0; i
< ARRAY_SIZE(ctx
->eg_interpolators
); i
++) {
1222 if (ctx
->eg_interpolators
[i
].enabled
) {
1223 ctx
->eg_interpolators
[i
].ij_index
= num_baryc
;
1228 /* XXX PULL MODEL and LINE STIPPLE */
1230 num_baryc
= (num_baryc
+ 1) >> 1;
1231 return allocate_system_value_inputs(ctx
, num_baryc
);
1234 /* sample_id_sel == NULL means fetch for current sample */
1235 static int load_sample_position(struct r600_shader_ctx
*ctx
, struct r600_shader_src
*sample_id
, int chan_sel
)
1237 struct r600_bytecode_vtx vtx
;
1240 assert(ctx
->fixed_pt_position_gpr
!= -1);
1242 t1
= r600_get_temp(ctx
);
1244 memset(&vtx
, 0, sizeof(struct r600_bytecode_vtx
));
1245 vtx
.op
= FETCH_OP_VFETCH
;
1246 vtx
.buffer_id
= R600_BUFFER_INFO_CONST_BUFFER
;
1247 vtx
.fetch_type
= SQ_VTX_FETCH_NO_INDEX_OFFSET
;
1248 if (sample_id
== NULL
) {
1249 vtx
.src_gpr
= ctx
->fixed_pt_position_gpr
; // SAMPLEID is in .w;
1253 struct r600_bytecode_alu alu
;
1255 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
1256 alu
.op
= ALU_OP1_MOV
;
1257 r600_bytecode_src(&alu
.src
[0], sample_id
, chan_sel
);
1261 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
1268 vtx
.mega_fetch_count
= 16;
1274 vtx
.data_format
= FMT_32_32_32_32_FLOAT
;
1275 vtx
.num_format_all
= 2;
1276 vtx
.format_comp_all
= 1;
1277 vtx
.use_const_fields
= 0;
1278 vtx
.offset
= 1; // first element is size of buffer
1279 vtx
.endian
= r600_endian_swap(32);
1280 vtx
.srf_mode_all
= 1; /* SRF_MODE_NO_ZERO */
1282 r
= r600_bytecode_add_vtx(ctx
->bc
, &vtx
);
1289 static void tgsi_src(struct r600_shader_ctx
*ctx
,
1290 const struct tgsi_full_src_register
*tgsi_src
,
1291 struct r600_shader_src
*r600_src
)
1293 memset(r600_src
, 0, sizeof(*r600_src
));
1294 r600_src
->swizzle
[0] = tgsi_src
->Register
.SwizzleX
;
1295 r600_src
->swizzle
[1] = tgsi_src
->Register
.SwizzleY
;
1296 r600_src
->swizzle
[2] = tgsi_src
->Register
.SwizzleZ
;
1297 r600_src
->swizzle
[3] = tgsi_src
->Register
.SwizzleW
;
1298 r600_src
->neg
= tgsi_src
->Register
.Negate
;
1299 r600_src
->abs
= tgsi_src
->Register
.Absolute
;
1301 if (tgsi_src
->Register
.File
== TGSI_FILE_IMMEDIATE
) {
1303 if ((tgsi_src
->Register
.SwizzleX
== tgsi_src
->Register
.SwizzleY
) &&
1304 (tgsi_src
->Register
.SwizzleX
== tgsi_src
->Register
.SwizzleZ
) &&
1305 (tgsi_src
->Register
.SwizzleX
== tgsi_src
->Register
.SwizzleW
)) {
1307 index
= tgsi_src
->Register
.Index
* 4 + tgsi_src
->Register
.SwizzleX
;
1308 r600_bytecode_special_constants(ctx
->literals
[index
], &r600_src
->sel
, &r600_src
->neg
, r600_src
->abs
);
1309 if (r600_src
->sel
!= V_SQ_ALU_SRC_LITERAL
)
1312 index
= tgsi_src
->Register
.Index
;
1313 r600_src
->sel
= V_SQ_ALU_SRC_LITERAL
;
1314 memcpy(r600_src
->value
, ctx
->literals
+ index
* 4, sizeof(r600_src
->value
));
1315 } else if (tgsi_src
->Register
.File
== TGSI_FILE_SYSTEM_VALUE
) {
1316 if (ctx
->info
.system_value_semantic_name
[tgsi_src
->Register
.Index
] == TGSI_SEMANTIC_SAMPLEMASK
) {
1317 r600_src
->swizzle
[0] = 2; // Z value
1318 r600_src
->swizzle
[1] = 2;
1319 r600_src
->swizzle
[2] = 2;
1320 r600_src
->swizzle
[3] = 2;
1321 r600_src
->sel
= ctx
->face_gpr
;
1322 } else if (ctx
->info
.system_value_semantic_name
[tgsi_src
->Register
.Index
] == TGSI_SEMANTIC_SAMPLEID
) {
1323 r600_src
->swizzle
[0] = 3; // W value
1324 r600_src
->swizzle
[1] = 3;
1325 r600_src
->swizzle
[2] = 3;
1326 r600_src
->swizzle
[3] = 3;
1327 r600_src
->sel
= ctx
->fixed_pt_position_gpr
;
1328 } else if (ctx
->info
.system_value_semantic_name
[tgsi_src
->Register
.Index
] == TGSI_SEMANTIC_SAMPLEPOS
) {
1329 r600_src
->swizzle
[0] = 0;
1330 r600_src
->swizzle
[1] = 1;
1331 r600_src
->swizzle
[2] = 4;
1332 r600_src
->swizzle
[3] = 4;
1333 r600_src
->sel
= load_sample_position(ctx
, NULL
, -1);
1334 } else if (ctx
->info
.system_value_semantic_name
[tgsi_src
->Register
.Index
] == TGSI_SEMANTIC_INSTANCEID
) {
1335 r600_src
->swizzle
[0] = 3;
1336 r600_src
->swizzle
[1] = 3;
1337 r600_src
->swizzle
[2] = 3;
1338 r600_src
->swizzle
[3] = 3;
1340 } else if (ctx
->info
.system_value_semantic_name
[tgsi_src
->Register
.Index
] == TGSI_SEMANTIC_VERTEXID
) {
1341 r600_src
->swizzle
[0] = 0;
1342 r600_src
->swizzle
[1] = 0;
1343 r600_src
->swizzle
[2] = 0;
1344 r600_src
->swizzle
[3] = 0;
1346 } else if (ctx
->type
!= PIPE_SHADER_TESS_CTRL
&& ctx
->info
.system_value_semantic_name
[tgsi_src
->Register
.Index
] == TGSI_SEMANTIC_INVOCATIONID
) {
1347 r600_src
->swizzle
[0] = 3;
1348 r600_src
->swizzle
[1] = 3;
1349 r600_src
->swizzle
[2] = 3;
1350 r600_src
->swizzle
[3] = 3;
1352 } else if (ctx
->info
.system_value_semantic_name
[tgsi_src
->Register
.Index
] == TGSI_SEMANTIC_INVOCATIONID
) {
1353 r600_src
->swizzle
[0] = 2;
1354 r600_src
->swizzle
[1] = 2;
1355 r600_src
->swizzle
[2] = 2;
1356 r600_src
->swizzle
[3] = 2;
1358 } else if (ctx
->info
.system_value_semantic_name
[tgsi_src
->Register
.Index
] == TGSI_SEMANTIC_TESSCOORD
) {
1360 } else if (ctx
->info
.system_value_semantic_name
[tgsi_src
->Register
.Index
] == TGSI_SEMANTIC_TESSINNER
) {
1362 } else if (ctx
->info
.system_value_semantic_name
[tgsi_src
->Register
.Index
] == TGSI_SEMANTIC_TESSOUTER
) {
1364 } else if (ctx
->info
.system_value_semantic_name
[tgsi_src
->Register
.Index
] == TGSI_SEMANTIC_VERTICESIN
) {
1365 if (ctx
->type
== PIPE_SHADER_TESS_CTRL
) {
1366 r600_src
->sel
= ctx
->tess_input_info
;
1367 r600_src
->swizzle
[0] = 2;
1368 r600_src
->swizzle
[1] = 2;
1369 r600_src
->swizzle
[2] = 2;
1370 r600_src
->swizzle
[3] = 2;
1372 r600_src
->sel
= ctx
->tess_input_info
;
1373 r600_src
->swizzle
[0] = 3;
1374 r600_src
->swizzle
[1] = 3;
1375 r600_src
->swizzle
[2] = 3;
1376 r600_src
->swizzle
[3] = 3;
1378 } else if (ctx
->type
== PIPE_SHADER_TESS_CTRL
&& ctx
->info
.system_value_semantic_name
[tgsi_src
->Register
.Index
] == TGSI_SEMANTIC_PRIMID
) {
1380 r600_src
->swizzle
[0] = 0;
1381 r600_src
->swizzle
[1] = 0;
1382 r600_src
->swizzle
[2] = 0;
1383 r600_src
->swizzle
[3] = 0;
1384 } else if (ctx
->type
== PIPE_SHADER_TESS_EVAL
&& ctx
->info
.system_value_semantic_name
[tgsi_src
->Register
.Index
] == TGSI_SEMANTIC_PRIMID
) {
1386 r600_src
->swizzle
[0] = 3;
1387 r600_src
->swizzle
[1] = 3;
1388 r600_src
->swizzle
[2] = 3;
1389 r600_src
->swizzle
[3] = 3;
1392 if (tgsi_src
->Register
.Indirect
)
1393 r600_src
->rel
= V_SQ_REL_RELATIVE
;
1394 r600_src
->sel
= tgsi_src
->Register
.Index
;
1395 r600_src
->sel
+= ctx
->file_offset
[tgsi_src
->Register
.File
];
1397 if (tgsi_src
->Register
.File
== TGSI_FILE_CONSTANT
) {
1398 if (tgsi_src
->Register
.Dimension
) {
1399 r600_src
->kc_bank
= tgsi_src
->Dimension
.Index
;
1400 if (tgsi_src
->Dimension
.Indirect
) {
1401 r600_src
->kc_rel
= 1;
1407 static int tgsi_fetch_rel_const(struct r600_shader_ctx
*ctx
,
1408 unsigned int cb_idx
, unsigned cb_rel
, unsigned int offset
, unsigned ar_chan
,
1409 unsigned int dst_reg
)
1411 struct r600_bytecode_vtx vtx
;
1412 unsigned int ar_reg
;
1416 struct r600_bytecode_alu alu
;
1418 memset(&alu
, 0, sizeof(alu
));
1420 alu
.op
= ALU_OP2_ADD_INT
;
1421 alu
.src
[0].sel
= ctx
->bc
->ar_reg
;
1422 alu
.src
[0].chan
= ar_chan
;
1424 alu
.src
[1].sel
= V_SQ_ALU_SRC_LITERAL
;
1425 alu
.src
[1].value
= offset
;
1427 alu
.dst
.sel
= dst_reg
;
1428 alu
.dst
.chan
= ar_chan
;
1432 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
1437 ar_reg
= ctx
->bc
->ar_reg
;
1440 memset(&vtx
, 0, sizeof(vtx
));
1441 vtx
.buffer_id
= cb_idx
;
1442 vtx
.fetch_type
= SQ_VTX_FETCH_NO_INDEX_OFFSET
;
1443 vtx
.src_gpr
= ar_reg
;
1444 vtx
.src_sel_x
= ar_chan
;
1445 vtx
.mega_fetch_count
= 16;
1446 vtx
.dst_gpr
= dst_reg
;
1447 vtx
.dst_sel_x
= 0; /* SEL_X */
1448 vtx
.dst_sel_y
= 1; /* SEL_Y */
1449 vtx
.dst_sel_z
= 2; /* SEL_Z */
1450 vtx
.dst_sel_w
= 3; /* SEL_W */
1451 vtx
.data_format
= FMT_32_32_32_32_FLOAT
;
1452 vtx
.num_format_all
= 2; /* NUM_FORMAT_SCALED */
1453 vtx
.format_comp_all
= 1; /* FORMAT_COMP_SIGNED */
1454 vtx
.endian
= r600_endian_swap(32);
1455 vtx
.buffer_index_mode
= cb_rel
; // cb_rel ? V_SQ_CF_INDEX_0 : V_SQ_CF_INDEX_NONE;
1457 if ((r
= r600_bytecode_add_vtx(ctx
->bc
, &vtx
)))
1463 static int fetch_gs_input(struct r600_shader_ctx
*ctx
, struct tgsi_full_src_register
*src
, unsigned int dst_reg
)
1465 struct r600_bytecode_vtx vtx
;
1467 unsigned index
= src
->Register
.Index
;
1468 unsigned vtx_id
= src
->Dimension
.Index
;
1469 int offset_reg
= vtx_id
/ 3;
1470 int offset_chan
= vtx_id
% 3;
1473 /* offsets of per-vertex data in ESGS ring are passed to GS in R0.x, R0.y,
1474 * R0.w, R1.x, R1.y, R1.z (it seems R0.z is used for PrimitiveID) */
1476 if (offset_reg
== 0 && offset_chan
== 2)
1479 if (src
->Dimension
.Indirect
|| src
->Register
.Indirect
)
1480 t2
= r600_get_temp(ctx
);
1482 if (src
->Dimension
.Indirect
) {
1484 struct r600_bytecode_alu alu
;
1487 addr_reg
= get_address_file_reg(ctx
, src
->DimIndirect
.Index
);
1488 if (src
->DimIndirect
.Index
> 0) {
1489 r
= single_alu_op2(ctx
, ALU_OP1_MOV
,
1497 we have to put the R0.x/y/w into Rt.x Rt+1.x Rt+2.x then index reg from Rt.
1498 at least this is what fglrx seems to do. */
1499 for (i
= 0; i
< 3; i
++) {
1500 treg
[i
] = r600_get_temp(ctx
);
1502 r600_add_gpr_array(ctx
->shader
, treg
[0], 3, 0x0F);
1504 for (i
= 0; i
< 3; i
++) {
1505 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
1506 alu
.op
= ALU_OP1_MOV
;
1508 alu
.src
[0].chan
= i
== 2 ? 3 : i
;
1509 alu
.dst
.sel
= treg
[i
];
1513 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
1517 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
1518 alu
.op
= ALU_OP1_MOV
;
1519 alu
.src
[0].sel
= treg
[0];
1524 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
1531 if (src
->Register
.Indirect
) {
1533 unsigned first
= ctx
->info
.input_array_first
[src
->Indirect
.ArrayID
];
1535 addr_reg
= get_address_file_reg(ctx
, src
->Indirect
.Index
);
1537 /* pull the value from index_reg */
1538 r
= single_alu_op2(ctx
, ALU_OP2_ADD_INT
,
1541 V_SQ_ALU_SRC_LITERAL
, first
);
1544 r
= single_alu_op3(ctx
, ALU_OP3_MULADD_UINT24
,
1547 V_SQ_ALU_SRC_LITERAL
, 4,
1548 offset_reg
, offset_chan
);
1553 index
= src
->Register
.Index
- first
;
1556 memset(&vtx
, 0, sizeof(vtx
));
1557 vtx
.buffer_id
= R600_GS_RING_CONST_BUFFER
;
1558 vtx
.fetch_type
= SQ_VTX_FETCH_NO_INDEX_OFFSET
;
1559 vtx
.src_gpr
= offset_reg
;
1560 vtx
.src_sel_x
= offset_chan
;
1561 vtx
.offset
= index
* 16; /*bytes*/
1562 vtx
.mega_fetch_count
= 16;
1563 vtx
.dst_gpr
= dst_reg
;
1564 vtx
.dst_sel_x
= 0; /* SEL_X */
1565 vtx
.dst_sel_y
= 1; /* SEL_Y */
1566 vtx
.dst_sel_z
= 2; /* SEL_Z */
1567 vtx
.dst_sel_w
= 3; /* SEL_W */
1568 if (ctx
->bc
->chip_class
>= EVERGREEN
) {
1569 vtx
.use_const_fields
= 1;
1571 vtx
.data_format
= FMT_32_32_32_32_FLOAT
;
1574 if ((r
= r600_bytecode_add_vtx(ctx
->bc
, &vtx
)))
1580 static int tgsi_split_gs_inputs(struct r600_shader_ctx
*ctx
)
1582 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
1585 for (i
= 0; i
< inst
->Instruction
.NumSrcRegs
; i
++) {
1586 struct tgsi_full_src_register
*src
= &inst
->Src
[i
];
1588 if (src
->Register
.File
== TGSI_FILE_INPUT
) {
1589 if (ctx
->shader
->input
[src
->Register
.Index
].name
== TGSI_SEMANTIC_PRIMID
) {
1590 /* primitive id is in R0.z */
1591 ctx
->src
[i
].sel
= 0;
1592 ctx
->src
[i
].swizzle
[0] = 2;
1595 if (src
->Register
.File
== TGSI_FILE_INPUT
&& src
->Register
.Dimension
) {
1596 int treg
= r600_get_temp(ctx
);
1598 fetch_gs_input(ctx
, src
, treg
);
1599 ctx
->src
[i
].sel
= treg
;
1600 ctx
->src
[i
].rel
= 0;
1607 /* Tessellation shaders pass outputs to the next shader using LDS.
1609 * LS outputs = TCS(HS) inputs
1610 * TCS(HS) outputs = TES(DS) inputs
1612 * The LDS layout is:
1613 * - TCS inputs for patch 0
1614 * - TCS inputs for patch 1
1615 * - TCS inputs for patch 2 = get_tcs_in_current_patch_offset (if RelPatchID==2)
1617 * - TCS outputs for patch 0 = get_tcs_out_patch0_offset
1618 * - Per-patch TCS outputs for patch 0 = get_tcs_out_patch0_patch_data_offset
1619 * - TCS outputs for patch 1
1620 * - Per-patch TCS outputs for patch 1
1621 * - TCS outputs for patch 2 = get_tcs_out_current_patch_offset (if RelPatchID==2)
1622 * - Per-patch TCS outputs for patch 2 = get_tcs_out_current_patch_data_offset (if RelPatchID==2)
1625 * All three shaders VS(LS), TCS, TES share the same LDS space.
1627 /* this will return with the dw address in temp_reg.x */
1628 static int r600_get_byte_address(struct r600_shader_ctx
*ctx
, int temp_reg
,
1629 const struct tgsi_full_dst_register
*dst
,
1630 const struct tgsi_full_src_register
*src
,
1631 int stride_bytes_reg
, int stride_bytes_chan
)
1633 struct tgsi_full_dst_register reg
;
1634 ubyte
*name
, *index
, *array_first
;
1637 struct tgsi_shader_info
*info
= &ctx
->info
;
1638 /* Set the register description. The address computation is the same
1639 * for sources and destinations. */
1641 reg
.Register
.File
= src
->Register
.File
;
1642 reg
.Register
.Index
= src
->Register
.Index
;
1643 reg
.Register
.Indirect
= src
->Register
.Indirect
;
1644 reg
.Register
.Dimension
= src
->Register
.Dimension
;
1645 reg
.Indirect
= src
->Indirect
;
1646 reg
.Dimension
= src
->Dimension
;
1647 reg
.DimIndirect
= src
->DimIndirect
;
1651 /* If the register is 2-dimensional (e.g. an array of vertices
1652 * in a primitive), calculate the base address of the vertex. */
1653 if (reg
.Register
.Dimension
) {
1655 if (reg
.Dimension
.Indirect
) {
1657 assert (reg
.DimIndirect
.File
== TGSI_FILE_ADDRESS
);
1659 addr_reg
= get_address_file_reg(ctx
, reg
.DimIndirect
.Index
);
1660 /* pull the value from index_reg */
1664 sel
= V_SQ_ALU_SRC_LITERAL
;
1665 chan
= reg
.Dimension
.Index
;
1668 r
= single_alu_op3(ctx
, ALU_OP3_MULADD_UINT24
,
1670 stride_bytes_reg
, stride_bytes_chan
,
1677 if (reg
.Register
.File
== TGSI_FILE_INPUT
) {
1678 name
= info
->input_semantic_name
;
1679 index
= info
->input_semantic_index
;
1680 array_first
= info
->input_array_first
;
1681 } else if (reg
.Register
.File
== TGSI_FILE_OUTPUT
) {
1682 name
= info
->output_semantic_name
;
1683 index
= info
->output_semantic_index
;
1684 array_first
= info
->output_array_first
;
1689 if (reg
.Register
.Indirect
) {
1692 /* Add the relative address of the element. */
1693 if (reg
.Indirect
.ArrayID
)
1694 first
= array_first
[reg
.Indirect
.ArrayID
];
1696 first
= reg
.Register
.Index
;
1698 addr_reg
= get_address_file_reg(ctx
, reg
.Indirect
.Index
);
1700 /* pull the value from index_reg */
1701 r
= single_alu_op3(ctx
, ALU_OP3_MULADD_UINT24
,
1703 V_SQ_ALU_SRC_LITERAL
, 16,
1709 param
= r600_get_lds_unique_index(name
[first
],
1713 param
= r600_get_lds_unique_index(name
[reg
.Register
.Index
],
1714 index
[reg
.Register
.Index
]);
1717 /* add to base_addr - passed in temp_reg.x */
1719 r
= single_alu_op2(ctx
, ALU_OP2_ADD_INT
,
1722 V_SQ_ALU_SRC_LITERAL
, param
* 16);
1730 static int do_lds_fetch_values(struct r600_shader_ctx
*ctx
, unsigned temp_reg
,
1733 struct r600_bytecode_alu alu
;
1736 if ((ctx
->bc
->cf_last
->ndw
>>1) >= 0x60)
1737 ctx
->bc
->force_add_cf
= 1;
1738 for (i
= 1; i
< 4; i
++) {
1739 r
= single_alu_op2(ctx
, ALU_OP2_ADD_INT
,
1742 V_SQ_ALU_SRC_LITERAL
, 4 * i
);
1746 for (i
= 0; i
< 4; i
++) {
1747 /* emit an LDS_READ_RET */
1748 memset(&alu
, 0, sizeof(alu
));
1749 alu
.op
= LDS_OP1_LDS_READ_RET
;
1750 alu
.src
[0].sel
= temp_reg
;
1751 alu
.src
[0].chan
= i
;
1752 alu
.src
[1].sel
= V_SQ_ALU_SRC_0
;
1753 alu
.src
[2].sel
= V_SQ_ALU_SRC_0
;
1755 alu
.is_lds_idx_op
= true;
1757 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
1761 for (i
= 0; i
< 4; i
++) {
1762 /* then read from LDS_OQ_A_POP */
1763 memset(&alu
, 0, sizeof(alu
));
1765 alu
.op
= ALU_OP1_MOV
;
1766 alu
.src
[0].sel
= EG_V_SQ_ALU_SRC_LDS_OQ_A_POP
;
1767 alu
.src
[0].chan
= 0;
1768 alu
.dst
.sel
= dst_reg
;
1772 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
1779 static int fetch_tes_input(struct r600_shader_ctx
*ctx
, struct tgsi_full_src_register
*src
, unsigned int dst_reg
)
1782 unsigned temp_reg
= r600_get_temp(ctx
);
1784 r
= get_lds_offset0(ctx
, 2, temp_reg
,
1785 src
->Register
.Dimension
? false : true);
1789 /* the base address is now in temp.x */
1790 r
= r600_get_byte_address(ctx
, temp_reg
,
1791 NULL
, src
, ctx
->tess_output_info
, 1);
1795 r
= do_lds_fetch_values(ctx
, temp_reg
, dst_reg
);
1801 static int fetch_tcs_input(struct r600_shader_ctx
*ctx
, struct tgsi_full_src_register
*src
, unsigned int dst_reg
)
1804 unsigned temp_reg
= r600_get_temp(ctx
);
1806 /* t.x = ips * r0.y */
1807 r
= single_alu_op2(ctx
, ALU_OP2_MUL_UINT24
,
1809 ctx
->tess_input_info
, 0,
1815 /* the base address is now in temp.x */
1816 r
= r600_get_byte_address(ctx
, temp_reg
,
1817 NULL
, src
, ctx
->tess_input_info
, 1);
1821 r
= do_lds_fetch_values(ctx
, temp_reg
, dst_reg
);
1827 static int fetch_tcs_output(struct r600_shader_ctx
*ctx
, struct tgsi_full_src_register
*src
, unsigned int dst_reg
)
1830 unsigned temp_reg
= r600_get_temp(ctx
);
1832 r
= get_lds_offset0(ctx
, 1, temp_reg
,
1833 src
->Register
.Dimension
? false : true);
1836 /* the base address is now in temp.x */
1837 r
= r600_get_byte_address(ctx
, temp_reg
,
1839 ctx
->tess_output_info
, 1);
1843 r
= do_lds_fetch_values(ctx
, temp_reg
, dst_reg
);
1849 static int tgsi_split_lds_inputs(struct r600_shader_ctx
*ctx
)
1851 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
1854 for (i
= 0; i
< inst
->Instruction
.NumSrcRegs
; i
++) {
1855 struct tgsi_full_src_register
*src
= &inst
->Src
[i
];
1857 if (ctx
->type
== PIPE_SHADER_TESS_EVAL
&& src
->Register
.File
== TGSI_FILE_INPUT
) {
1858 int treg
= r600_get_temp(ctx
);
1859 fetch_tes_input(ctx
, src
, treg
);
1860 ctx
->src
[i
].sel
= treg
;
1861 ctx
->src
[i
].rel
= 0;
1863 if (ctx
->type
== PIPE_SHADER_TESS_CTRL
&& src
->Register
.File
== TGSI_FILE_INPUT
) {
1864 int treg
= r600_get_temp(ctx
);
1865 fetch_tcs_input(ctx
, src
, treg
);
1866 ctx
->src
[i
].sel
= treg
;
1867 ctx
->src
[i
].rel
= 0;
1869 if (ctx
->type
== PIPE_SHADER_TESS_CTRL
&& src
->Register
.File
== TGSI_FILE_OUTPUT
) {
1870 int treg
= r600_get_temp(ctx
);
1871 fetch_tcs_output(ctx
, src
, treg
);
1872 ctx
->src
[i
].sel
= treg
;
1873 ctx
->src
[i
].rel
= 0;
1879 static int tgsi_split_constant(struct r600_shader_ctx
*ctx
)
1881 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
1882 struct r600_bytecode_alu alu
;
1883 int i
, j
, k
, nconst
, r
;
1885 for (i
= 0, nconst
= 0; i
< inst
->Instruction
.NumSrcRegs
; i
++) {
1886 if (inst
->Src
[i
].Register
.File
== TGSI_FILE_CONSTANT
) {
1889 tgsi_src(ctx
, &inst
->Src
[i
], &ctx
->src
[i
]);
1891 for (i
= 0, j
= nconst
- 1; i
< inst
->Instruction
.NumSrcRegs
; i
++) {
1892 if (inst
->Src
[i
].Register
.File
!= TGSI_FILE_CONSTANT
) {
1896 if (ctx
->src
[i
].rel
) {
1897 int chan
= inst
->Src
[i
].Indirect
.Swizzle
;
1898 int treg
= r600_get_temp(ctx
);
1899 if ((r
= tgsi_fetch_rel_const(ctx
, ctx
->src
[i
].kc_bank
, ctx
->src
[i
].kc_rel
, ctx
->src
[i
].sel
- 512, chan
, treg
)))
1902 ctx
->src
[i
].kc_bank
= 0;
1903 ctx
->src
[i
].kc_rel
= 0;
1904 ctx
->src
[i
].sel
= treg
;
1905 ctx
->src
[i
].rel
= 0;
1908 int treg
= r600_get_temp(ctx
);
1909 for (k
= 0; k
< 4; k
++) {
1910 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
1911 alu
.op
= ALU_OP1_MOV
;
1912 alu
.src
[0].sel
= ctx
->src
[i
].sel
;
1913 alu
.src
[0].chan
= k
;
1914 alu
.src
[0].rel
= ctx
->src
[i
].rel
;
1915 alu
.src
[0].kc_bank
= ctx
->src
[i
].kc_bank
;
1916 alu
.src
[0].kc_rel
= ctx
->src
[i
].kc_rel
;
1922 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
1926 ctx
->src
[i
].sel
= treg
;
1934 /* need to move any immediate into a temp - for trig functions which use literal for PI stuff */
1935 static int tgsi_split_literal_constant(struct r600_shader_ctx
*ctx
)
1937 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
1938 struct r600_bytecode_alu alu
;
1939 int i
, j
, k
, nliteral
, r
;
1941 for (i
= 0, nliteral
= 0; i
< inst
->Instruction
.NumSrcRegs
; i
++) {
1942 if (ctx
->src
[i
].sel
== V_SQ_ALU_SRC_LITERAL
) {
1946 for (i
= 0, j
= nliteral
- 1; i
< inst
->Instruction
.NumSrcRegs
; i
++) {
1947 if (j
> 0 && ctx
->src
[i
].sel
== V_SQ_ALU_SRC_LITERAL
) {
1948 int treg
= r600_get_temp(ctx
);
1949 for (k
= 0; k
< 4; k
++) {
1950 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
1951 alu
.op
= ALU_OP1_MOV
;
1952 alu
.src
[0].sel
= ctx
->src
[i
].sel
;
1953 alu
.src
[0].chan
= k
;
1954 alu
.src
[0].value
= ctx
->src
[i
].value
[k
];
1960 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
1964 ctx
->src
[i
].sel
= treg
;
1971 static int process_twoside_color_inputs(struct r600_shader_ctx
*ctx
)
1973 int i
, r
, count
= ctx
->shader
->ninput
;
1975 for (i
= 0; i
< count
; i
++) {
1976 if (ctx
->shader
->input
[i
].name
== TGSI_SEMANTIC_COLOR
) {
1977 r
= select_twoside_color(ctx
, i
, ctx
->shader
->input
[i
].back_color_input
);
1985 static int emit_streamout(struct r600_shader_ctx
*ctx
, struct pipe_stream_output_info
*so
,
1986 int stream
, unsigned *stream_item_size
)
1988 unsigned so_gpr
[PIPE_MAX_SHADER_OUTPUTS
];
1989 unsigned start_comp
[PIPE_MAX_SHADER_OUTPUTS
];
1992 /* Sanity checking. */
1993 if (so
->num_outputs
> PIPE_MAX_SO_OUTPUTS
) {
1994 R600_ERR("Too many stream outputs: %d\n", so
->num_outputs
);
1998 for (i
= 0; i
< so
->num_outputs
; i
++) {
1999 if (so
->output
[i
].output_buffer
>= 4) {
2000 R600_ERR("Exceeded the max number of stream output buffers, got: %d\n",
2001 so
->output
[i
].output_buffer
);
2007 /* Initialize locations where the outputs are stored. */
2008 for (i
= 0; i
< so
->num_outputs
; i
++) {
2010 so_gpr
[i
] = ctx
->shader
->output
[so
->output
[i
].register_index
].gpr
;
2011 start_comp
[i
] = so
->output
[i
].start_component
;
2012 /* Lower outputs with dst_offset < start_component.
2014 * We can only output 4D vectors with a write mask, e.g. we can
2015 * only output the W component at offset 3, etc. If we want
2016 * to store Y, Z, or W at buffer offset 0, we need to use MOV
2017 * to move it to X and output X. */
2018 if (so
->output
[i
].dst_offset
< so
->output
[i
].start_component
) {
2019 unsigned tmp
= r600_get_temp(ctx
);
2021 for (j
= 0; j
< so
->output
[i
].num_components
; j
++) {
2022 struct r600_bytecode_alu alu
;
2023 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2024 alu
.op
= ALU_OP1_MOV
;
2025 alu
.src
[0].sel
= so_gpr
[i
];
2026 alu
.src
[0].chan
= so
->output
[i
].start_component
+ j
;
2031 if (j
== so
->output
[i
].num_components
- 1)
2033 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
2042 /* Write outputs to buffers. */
2043 for (i
= 0; i
< so
->num_outputs
; i
++) {
2044 struct r600_bytecode_output output
;
2046 if (stream
!= -1 && stream
!= so
->output
[i
].output_buffer
)
2049 memset(&output
, 0, sizeof(struct r600_bytecode_output
));
2050 output
.gpr
= so_gpr
[i
];
2051 output
.elem_size
= so
->output
[i
].num_components
- 1;
2052 if (output
.elem_size
== 2)
2053 output
.elem_size
= 3; // 3 not supported, write 4 with junk at end
2054 output
.array_base
= so
->output
[i
].dst_offset
- start_comp
[i
];
2055 output
.type
= V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_WRITE
;
2056 output
.burst_count
= 1;
2057 /* array_size is an upper limit for the burst_count
2058 * with MEM_STREAM instructions */
2059 output
.array_size
= 0xFFF;
2060 output
.comp_mask
= ((1 << so
->output
[i
].num_components
) - 1) << start_comp
[i
];
2062 if (ctx
->bc
->chip_class
>= EVERGREEN
) {
2063 switch (so
->output
[i
].output_buffer
) {
2065 output
.op
= CF_OP_MEM_STREAM0_BUF0
;
2068 output
.op
= CF_OP_MEM_STREAM0_BUF1
;
2071 output
.op
= CF_OP_MEM_STREAM0_BUF2
;
2074 output
.op
= CF_OP_MEM_STREAM0_BUF3
;
2077 output
.op
+= so
->output
[i
].stream
* 4;
2078 assert(output
.op
>= CF_OP_MEM_STREAM0_BUF0
&& output
.op
<= CF_OP_MEM_STREAM3_BUF3
);
2079 ctx
->enabled_stream_buffers_mask
|= (1 << so
->output
[i
].output_buffer
) << so
->output
[i
].stream
* 4;
2081 switch (so
->output
[i
].output_buffer
) {
2083 output
.op
= CF_OP_MEM_STREAM0
;
2086 output
.op
= CF_OP_MEM_STREAM1
;
2089 output
.op
= CF_OP_MEM_STREAM2
;
2092 output
.op
= CF_OP_MEM_STREAM3
;
2095 ctx
->enabled_stream_buffers_mask
|= 1 << so
->output
[i
].output_buffer
;
2097 r
= r600_bytecode_add_output(ctx
->bc
, &output
);
2106 static void convert_edgeflag_to_int(struct r600_shader_ctx
*ctx
)
2108 struct r600_bytecode_alu alu
;
2111 if (!ctx
->shader
->vs_out_edgeflag
)
2114 reg
= ctx
->shader
->output
[ctx
->edgeflag_output
].gpr
;
2116 /* clamp(x, 0, 1) */
2117 memset(&alu
, 0, sizeof(alu
));
2118 alu
.op
= ALU_OP1_MOV
;
2119 alu
.src
[0].sel
= reg
;
2124 r600_bytecode_add_alu(ctx
->bc
, &alu
);
2126 memset(&alu
, 0, sizeof(alu
));
2127 alu
.op
= ALU_OP1_FLT_TO_INT
;
2128 alu
.src
[0].sel
= reg
;
2132 r600_bytecode_add_alu(ctx
->bc
, &alu
);
2135 static int generate_gs_copy_shader(struct r600_context
*rctx
,
2136 struct r600_pipe_shader
*gs
,
2137 struct pipe_stream_output_info
*so
)
2139 struct r600_shader_ctx ctx
= {};
2140 struct r600_shader
*gs_shader
= &gs
->shader
;
2141 struct r600_pipe_shader
*cshader
;
2142 int ocnt
= gs_shader
->noutput
;
2143 struct r600_bytecode_alu alu
;
2144 struct r600_bytecode_vtx vtx
;
2145 struct r600_bytecode_output output
;
2146 struct r600_bytecode_cf
*cf_jump
, *cf_pop
,
2147 *last_exp_pos
= NULL
, *last_exp_param
= NULL
;
2148 int i
, j
, next_clip_pos
= 61, next_param
= 0;
2150 bool only_ring_0
= true;
2151 cshader
= calloc(1, sizeof(struct r600_pipe_shader
));
2155 memcpy(cshader
->shader
.output
, gs_shader
->output
, ocnt
*
2156 sizeof(struct r600_shader_io
));
2158 cshader
->shader
.noutput
= ocnt
;
2160 ctx
.shader
= &cshader
->shader
;
2161 ctx
.bc
= &ctx
.shader
->bc
;
2162 ctx
.type
= ctx
.bc
->type
= PIPE_SHADER_VERTEX
;
2164 r600_bytecode_init(ctx
.bc
, rctx
->b
.chip_class
, rctx
->b
.family
,
2165 rctx
->screen
->has_compressed_msaa_texturing
);
2167 ctx
.bc
->isa
= rctx
->isa
;
2170 memset(cshader
->shader
.ring_item_sizes
, 0, sizeof(cshader
->shader
.ring_item_sizes
));
2172 /* R0.x = R0.x & 0x3fffffff */
2173 memset(&alu
, 0, sizeof(alu
));
2174 alu
.op
= ALU_OP2_AND_INT
;
2175 alu
.src
[1].sel
= V_SQ_ALU_SRC_LITERAL
;
2176 alu
.src
[1].value
= 0x3fffffff;
2178 r600_bytecode_add_alu(ctx
.bc
, &alu
);
2180 /* R0.y = R0.x >> 30 */
2181 memset(&alu
, 0, sizeof(alu
));
2182 alu
.op
= ALU_OP2_LSHR_INT
;
2183 alu
.src
[1].sel
= V_SQ_ALU_SRC_LITERAL
;
2184 alu
.src
[1].value
= 0x1e;
2188 r600_bytecode_add_alu(ctx
.bc
, &alu
);
2190 /* fetch vertex data from GSVS ring */
2191 for (i
= 0; i
< ocnt
; ++i
) {
2192 struct r600_shader_io
*out
= &ctx
.shader
->output
[i
];
2195 out
->ring_offset
= i
* 16;
2197 memset(&vtx
, 0, sizeof(vtx
));
2198 vtx
.op
= FETCH_OP_VFETCH
;
2199 vtx
.buffer_id
= R600_GS_RING_CONST_BUFFER
;
2200 vtx
.fetch_type
= SQ_VTX_FETCH_NO_INDEX_OFFSET
;
2201 vtx
.mega_fetch_count
= 16;
2202 vtx
.offset
= out
->ring_offset
;
2203 vtx
.dst_gpr
= out
->gpr
;
2209 if (rctx
->b
.chip_class
>= EVERGREEN
) {
2210 vtx
.use_const_fields
= 1;
2212 vtx
.data_format
= FMT_32_32_32_32_FLOAT
;
2215 r600_bytecode_add_vtx(ctx
.bc
, &vtx
);
2217 ctx
.temp_reg
= i
+ 1;
2218 for (ring
= 3; ring
>= 0; --ring
) {
2219 bool enabled
= false;
2220 for (i
= 0; i
< so
->num_outputs
; i
++) {
2221 if (so
->output
[i
].stream
== ring
) {
2224 only_ring_0
= false;
2228 if (ring
!= 0 && !enabled
) {
2229 cshader
->shader
.ring_item_sizes
[ring
] = 0;
2234 // Patch up jump label
2235 r600_bytecode_add_cfinst(ctx
.bc
, CF_OP_POP
);
2236 cf_pop
= ctx
.bc
->cf_last
;
2238 cf_jump
->cf_addr
= cf_pop
->id
+ 2;
2239 cf_jump
->pop_count
= 1;
2240 cf_pop
->cf_addr
= cf_pop
->id
+ 2;
2241 cf_pop
->pop_count
= 1;
2244 /* PRED_SETE_INT __, R0.y, ring */
2245 memset(&alu
, 0, sizeof(alu
));
2246 alu
.op
= ALU_OP2_PRED_SETE_INT
;
2247 alu
.src
[0].chan
= 1;
2248 alu
.src
[1].sel
= V_SQ_ALU_SRC_LITERAL
;
2249 alu
.src
[1].value
= ring
;
2250 alu
.execute_mask
= 1;
2251 alu
.update_pred
= 1;
2253 r600_bytecode_add_alu_type(ctx
.bc
, &alu
, CF_OP_ALU_PUSH_BEFORE
);
2255 r600_bytecode_add_cfinst(ctx
.bc
, CF_OP_JUMP
);
2256 cf_jump
= ctx
.bc
->cf_last
;
2259 emit_streamout(&ctx
, so
, only_ring_0
? -1 : ring
, &cshader
->shader
.ring_item_sizes
[ring
]);
2260 cshader
->shader
.ring_item_sizes
[ring
] = ocnt
* 16;
2263 /* bc adds nops - copy it */
2264 if (ctx
.bc
->chip_class
== R600
) {
2265 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2266 alu
.op
= ALU_OP0_NOP
;
2268 r600_bytecode_add_alu(ctx
.bc
, &alu
);
2270 r600_bytecode_add_cfinst(ctx
.bc
, CF_OP_NOP
);
2273 /* export vertex data */
2274 /* XXX factor out common code with r600_shader_from_tgsi ? */
2275 for (i
= 0; i
< ocnt
; ++i
) {
2276 struct r600_shader_io
*out
= &ctx
.shader
->output
[i
];
2277 bool instream0
= true;
2278 if (out
->name
== TGSI_SEMANTIC_CLIPVERTEX
)
2281 for (j
= 0; j
< so
->num_outputs
; j
++) {
2282 if (so
->output
[j
].register_index
== i
) {
2283 if (so
->output
[j
].stream
== 0)
2285 if (so
->output
[j
].stream
> 0)
2291 memset(&output
, 0, sizeof(output
));
2292 output
.gpr
= out
->gpr
;
2293 output
.elem_size
= 3;
2294 output
.swizzle_x
= 0;
2295 output
.swizzle_y
= 1;
2296 output
.swizzle_z
= 2;
2297 output
.swizzle_w
= 3;
2298 output
.burst_count
= 1;
2299 output
.type
= V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PARAM
;
2300 output
.op
= CF_OP_EXPORT
;
2301 switch (out
->name
) {
2302 case TGSI_SEMANTIC_POSITION
:
2303 output
.array_base
= 60;
2304 output
.type
= V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_POS
;
2307 case TGSI_SEMANTIC_PSIZE
:
2308 output
.array_base
= 61;
2309 if (next_clip_pos
== 61)
2311 output
.type
= V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_POS
;
2312 output
.swizzle_y
= 7;
2313 output
.swizzle_z
= 7;
2314 output
.swizzle_w
= 7;
2315 ctx
.shader
->vs_out_misc_write
= 1;
2316 ctx
.shader
->vs_out_point_size
= 1;
2318 case TGSI_SEMANTIC_LAYER
:
2320 /* duplicate it as PARAM to pass to the pixel shader */
2321 output
.array_base
= next_param
++;
2322 r600_bytecode_add_output(ctx
.bc
, &output
);
2323 last_exp_param
= ctx
.bc
->cf_last
;
2325 output
.array_base
= 61;
2326 if (next_clip_pos
== 61)
2328 output
.type
= V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_POS
;
2329 output
.swizzle_x
= 7;
2330 output
.swizzle_y
= 7;
2331 output
.swizzle_z
= 0;
2332 output
.swizzle_w
= 7;
2333 ctx
.shader
->vs_out_misc_write
= 1;
2334 ctx
.shader
->vs_out_layer
= 1;
2336 case TGSI_SEMANTIC_VIEWPORT_INDEX
:
2338 /* duplicate it as PARAM to pass to the pixel shader */
2339 output
.array_base
= next_param
++;
2340 r600_bytecode_add_output(ctx
.bc
, &output
);
2341 last_exp_param
= ctx
.bc
->cf_last
;
2343 output
.array_base
= 61;
2344 if (next_clip_pos
== 61)
2346 output
.type
= V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_POS
;
2347 ctx
.shader
->vs_out_misc_write
= 1;
2348 ctx
.shader
->vs_out_viewport
= 1;
2349 output
.swizzle_x
= 7;
2350 output
.swizzle_y
= 7;
2351 output
.swizzle_z
= 7;
2352 output
.swizzle_w
= 0;
2354 case TGSI_SEMANTIC_CLIPDIST
:
2355 /* spi_sid is 0 for clipdistance outputs that were generated
2356 * for clipvertex - we don't need to pass them to PS */
2357 ctx
.shader
->clip_dist_write
= gs
->shader
.clip_dist_write
;
2359 /* duplicate it as PARAM to pass to the pixel shader */
2360 output
.array_base
= next_param
++;
2361 r600_bytecode_add_output(ctx
.bc
, &output
);
2362 last_exp_param
= ctx
.bc
->cf_last
;
2364 output
.array_base
= next_clip_pos
++;
2365 output
.type
= V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_POS
;
2367 case TGSI_SEMANTIC_FOG
:
2368 output
.swizzle_y
= 4; /* 0 */
2369 output
.swizzle_z
= 4; /* 0 */
2370 output
.swizzle_w
= 5; /* 1 */
2373 output
.array_base
= next_param
++;
2376 r600_bytecode_add_output(ctx
.bc
, &output
);
2377 if (output
.type
== V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PARAM
)
2378 last_exp_param
= ctx
.bc
->cf_last
;
2380 last_exp_pos
= ctx
.bc
->cf_last
;
2383 if (!last_exp_pos
) {
2384 memset(&output
, 0, sizeof(output
));
2386 output
.elem_size
= 3;
2387 output
.swizzle_x
= 7;
2388 output
.swizzle_y
= 7;
2389 output
.swizzle_z
= 7;
2390 output
.swizzle_w
= 7;
2391 output
.burst_count
= 1;
2393 output
.op
= CF_OP_EXPORT
;
2394 output
.array_base
= 60;
2395 output
.type
= V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_POS
;
2396 r600_bytecode_add_output(ctx
.bc
, &output
);
2397 last_exp_pos
= ctx
.bc
->cf_last
;
2400 if (!last_exp_param
) {
2401 memset(&output
, 0, sizeof(output
));
2403 output
.elem_size
= 3;
2404 output
.swizzle_x
= 7;
2405 output
.swizzle_y
= 7;
2406 output
.swizzle_z
= 7;
2407 output
.swizzle_w
= 7;
2408 output
.burst_count
= 1;
2410 output
.op
= CF_OP_EXPORT
;
2411 output
.array_base
= next_param
++;
2412 output
.type
= V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PARAM
;
2413 r600_bytecode_add_output(ctx
.bc
, &output
);
2414 last_exp_param
= ctx
.bc
->cf_last
;
2417 last_exp_pos
->op
= CF_OP_EXPORT_DONE
;
2418 last_exp_param
->op
= CF_OP_EXPORT_DONE
;
2420 r600_bytecode_add_cfinst(ctx
.bc
, CF_OP_POP
);
2421 cf_pop
= ctx
.bc
->cf_last
;
2423 cf_jump
->cf_addr
= cf_pop
->id
+ 2;
2424 cf_jump
->pop_count
= 1;
2425 cf_pop
->cf_addr
= cf_pop
->id
+ 2;
2426 cf_pop
->pop_count
= 1;
2428 if (ctx
.bc
->chip_class
== CAYMAN
)
2429 cm_bytecode_add_cf_end(ctx
.bc
);
2431 r600_bytecode_add_cfinst(ctx
.bc
, CF_OP_NOP
);
2432 ctx
.bc
->cf_last
->end_of_program
= 1;
2435 gs
->gs_copy_shader
= cshader
;
2436 cshader
->enabled_stream_buffers_mask
= ctx
.enabled_stream_buffers_mask
;
2440 return r600_bytecode_build(ctx
.bc
);
2443 static int emit_inc_ring_offset(struct r600_shader_ctx
*ctx
, int idx
, bool ind
)
2446 struct r600_bytecode_alu alu
;
2449 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2450 alu
.op
= ALU_OP2_ADD_INT
;
2451 alu
.src
[0].sel
= ctx
->gs_export_gpr_tregs
[idx
];
2452 alu
.src
[1].sel
= V_SQ_ALU_SRC_LITERAL
;
2453 alu
.src
[1].value
= ctx
->gs_out_ring_offset
>> 4;
2454 alu
.dst
.sel
= ctx
->gs_export_gpr_tregs
[idx
];
2457 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
2464 static int emit_gs_ring_writes(struct r600_shader_ctx
*ctx
, const struct pipe_stream_output_info
*so
, int stream
, bool ind
)
2466 struct r600_bytecode_output output
;
2467 int i
, k
, ring_offset
;
2468 int effective_stream
= stream
== -1 ? 0 : stream
;
2471 for (i
= 0; i
< ctx
->shader
->noutput
; i
++) {
2472 if (ctx
->gs_for_vs
) {
2473 /* for ES we need to lookup corresponding ring offset expected by GS
2474 * (map this output to GS input by name and sid) */
2475 /* FIXME precompute offsets */
2477 for(k
= 0; k
< ctx
->gs_for_vs
->ninput
; ++k
) {
2478 struct r600_shader_io
*in
= &ctx
->gs_for_vs
->input
[k
];
2479 struct r600_shader_io
*out
= &ctx
->shader
->output
[i
];
2480 if (in
->name
== out
->name
&& in
->sid
== out
->sid
)
2481 ring_offset
= in
->ring_offset
;
2484 if (ring_offset
== -1)
2487 ring_offset
= idx
* 16;
2491 if (stream
> 0 && ctx
->shader
->output
[i
].name
== TGSI_SEMANTIC_POSITION
)
2493 /* next_ring_offset after parsing input decls contains total size of
2494 * single vertex data, gs_next_vertex - current vertex index */
2496 ring_offset
+= ctx
->gs_out_ring_offset
* ctx
->gs_next_vertex
;
2498 memset(&output
, 0, sizeof(struct r600_bytecode_output
));
2499 output
.gpr
= ctx
->shader
->output
[i
].gpr
;
2500 output
.elem_size
= 3;
2501 output
.comp_mask
= 0xF;
2502 output
.burst_count
= 1;
2505 output
.type
= V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_WRITE_IND
;
2507 output
.type
= V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_WRITE
;
2512 output
.op
= CF_OP_MEM_RING
; break;
2514 output
.op
= CF_OP_MEM_RING1
; break;
2516 output
.op
= CF_OP_MEM_RING2
; break;
2518 output
.op
= CF_OP_MEM_RING3
; break;
2522 output
.array_base
= ring_offset
>> 2; /* in dwords */
2523 output
.array_size
= 0xfff;
2524 output
.index_gpr
= ctx
->gs_export_gpr_tregs
[effective_stream
];
2526 output
.array_base
= ring_offset
>> 2; /* in dwords */
2527 r600_bytecode_add_output(ctx
->bc
, &output
);
2530 ++ctx
->gs_next_vertex
;
2535 static int r600_fetch_tess_io_info(struct r600_shader_ctx
*ctx
)
2538 struct r600_bytecode_vtx vtx
;
2539 int temp_val
= ctx
->temp_reg
;
2540 /* need to store the TCS output somewhere */
2541 r
= single_alu_op2(ctx
, ALU_OP1_MOV
,
2543 V_SQ_ALU_SRC_LITERAL
, 0,
2548 /* used by VS/TCS */
2549 if (ctx
->tess_input_info
) {
2550 /* fetch tcs input values into resv space */
2551 memset(&vtx
, 0, sizeof(struct r600_bytecode_vtx
));
2552 vtx
.op
= FETCH_OP_VFETCH
;
2553 vtx
.buffer_id
= R600_LDS_INFO_CONST_BUFFER
;
2554 vtx
.fetch_type
= SQ_VTX_FETCH_NO_INDEX_OFFSET
;
2555 vtx
.mega_fetch_count
= 16;
2556 vtx
.data_format
= FMT_32_32_32_32
;
2557 vtx
.num_format_all
= 2;
2558 vtx
.format_comp_all
= 1;
2559 vtx
.use_const_fields
= 0;
2560 vtx
.endian
= r600_endian_swap(32);
2561 vtx
.srf_mode_all
= 1;
2563 vtx
.dst_gpr
= ctx
->tess_input_info
;
2568 vtx
.src_gpr
= temp_val
;
2571 r
= r600_bytecode_add_vtx(ctx
->bc
, &vtx
);
2576 /* used by TCS/TES */
2577 if (ctx
->tess_output_info
) {
2578 /* fetch tcs output values into resv space */
2579 memset(&vtx
, 0, sizeof(struct r600_bytecode_vtx
));
2580 vtx
.op
= FETCH_OP_VFETCH
;
2581 vtx
.buffer_id
= R600_LDS_INFO_CONST_BUFFER
;
2582 vtx
.fetch_type
= SQ_VTX_FETCH_NO_INDEX_OFFSET
;
2583 vtx
.mega_fetch_count
= 16;
2584 vtx
.data_format
= FMT_32_32_32_32
;
2585 vtx
.num_format_all
= 2;
2586 vtx
.format_comp_all
= 1;
2587 vtx
.use_const_fields
= 0;
2588 vtx
.endian
= r600_endian_swap(32);
2589 vtx
.srf_mode_all
= 1;
2591 vtx
.dst_gpr
= ctx
->tess_output_info
;
2596 vtx
.src_gpr
= temp_val
;
2599 r
= r600_bytecode_add_vtx(ctx
->bc
, &vtx
);
2606 static int emit_lds_vs_writes(struct r600_shader_ctx
*ctx
)
2611 /* fetch tcs input values into input_vals */
2612 ctx
->tess_input_info
= r600_get_temp(ctx
);
2613 ctx
->tess_output_info
= 0;
2614 r
= r600_fetch_tess_io_info(ctx
);
2618 temp_reg
= r600_get_temp(ctx
);
2619 /* dst reg contains LDS address stride * idx */
2620 /* MUL vertexID, vertex_dw_stride */
2621 r
= single_alu_op2(ctx
, ALU_OP2_MUL_UINT24
,
2623 ctx
->tess_input_info
, 1,
2624 0, 1); /* rel id in r0.y? */
2628 for (i
= 0; i
< ctx
->shader
->noutput
; i
++) {
2629 struct r600_bytecode_alu alu
;
2630 int param
= r600_get_lds_unique_index(ctx
->shader
->output
[i
].name
, ctx
->shader
->output
[i
].sid
);
2633 r
= single_alu_op2(ctx
, ALU_OP2_ADD_INT
,
2636 V_SQ_ALU_SRC_LITERAL
, param
* 16);
2641 r
= single_alu_op2(ctx
, ALU_OP2_ADD_INT
,
2643 temp_reg
, param
? 1 : 0,
2644 V_SQ_ALU_SRC_LITERAL
, 8);
2649 for (j
= 0; j
< 2; j
++) {
2650 int chan
= (j
== 1) ? 2 : (param
? 1 : 0);
2651 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2652 alu
.op
= LDS_OP3_LDS_WRITE_REL
;
2653 alu
.src
[0].sel
= temp_reg
;
2654 alu
.src
[0].chan
= chan
;
2655 alu
.src
[1].sel
= ctx
->shader
->output
[i
].gpr
;
2656 alu
.src
[1].chan
= j
* 2;
2657 alu
.src
[2].sel
= ctx
->shader
->output
[i
].gpr
;
2658 alu
.src
[2].chan
= (j
* 2) + 1;
2662 alu
.is_lds_idx_op
= true;
2663 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
2671 static int r600_store_tcs_output(struct r600_shader_ctx
*ctx
)
2673 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
2674 const struct tgsi_full_dst_register
*dst
= &inst
->Dst
[0];
2676 int temp_reg
= r600_get_temp(ctx
);
2677 struct r600_bytecode_alu alu
;
2678 unsigned write_mask
= dst
->Register
.WriteMask
;
2680 if (inst
->Dst
[0].Register
.File
!= TGSI_FILE_OUTPUT
)
2683 r
= get_lds_offset0(ctx
, 1, temp_reg
, dst
->Register
.Dimension
? false : true);
2687 /* the base address is now in temp.x */
2688 r
= r600_get_byte_address(ctx
, temp_reg
,
2689 &inst
->Dst
[0], NULL
, ctx
->tess_output_info
, 1);
2694 lasti
= tgsi_last_instruction(write_mask
);
2695 for (i
= 1; i
<= lasti
; i
++) {
2697 if (!(write_mask
& (1 << i
)))
2699 r
= single_alu_op2(ctx
, ALU_OP2_ADD_INT
,
2702 V_SQ_ALU_SRC_LITERAL
, 4 * i
);
2707 for (i
= 0; i
<= lasti
; i
++) {
2708 if (!(write_mask
& (1 << i
)))
2711 if ((i
== 0 && ((write_mask
& 3) == 3)) ||
2712 (i
== 2 && ((write_mask
& 0xc) == 0xc))) {
2713 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2714 alu
.op
= LDS_OP3_LDS_WRITE_REL
;
2715 alu
.src
[0].sel
= temp_reg
;
2716 alu
.src
[0].chan
= i
;
2718 alu
.src
[1].sel
= dst
->Register
.Index
;
2719 alu
.src
[1].sel
+= ctx
->file_offset
[dst
->Register
.File
];
2720 alu
.src
[1].chan
= i
;
2722 alu
.src
[2].sel
= dst
->Register
.Index
;
2723 alu
.src
[2].sel
+= ctx
->file_offset
[dst
->Register
.File
];
2724 alu
.src
[2].chan
= i
+ 1;
2728 alu
.is_lds_idx_op
= true;
2729 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
2735 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
2736 alu
.op
= LDS_OP2_LDS_WRITE
;
2737 alu
.src
[0].sel
= temp_reg
;
2738 alu
.src
[0].chan
= i
;
2740 alu
.src
[1].sel
= dst
->Register
.Index
;
2741 alu
.src
[1].sel
+= ctx
->file_offset
[dst
->Register
.File
];
2742 alu
.src
[1].chan
= i
;
2744 alu
.src
[2].sel
= V_SQ_ALU_SRC_0
;
2747 alu
.is_lds_idx_op
= true;
2748 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
2755 static int r600_tess_factor_read(struct r600_shader_ctx
*ctx
,
2759 unsigned temp_reg
= r600_get_temp(ctx
);
2760 unsigned name
= ctx
->shader
->output
[output_idx
].name
;
2761 int dreg
= ctx
->shader
->output
[output_idx
].gpr
;
2764 param
= r600_get_lds_unique_index(name
, 0);
2765 r
= get_lds_offset0(ctx
, 1, temp_reg
, true);
2769 r
= single_alu_op2(ctx
, ALU_OP2_ADD_INT
,
2772 V_SQ_ALU_SRC_LITERAL
, param
* 16);
2776 do_lds_fetch_values(ctx
, temp_reg
, dreg
);
2780 static int r600_emit_tess_factor(struct r600_shader_ctx
*ctx
)
2783 int stride
, outer_comps
, inner_comps
;
2784 int tessinner_idx
= -1, tessouter_idx
= -1;
2786 int temp_reg
= r600_get_temp(ctx
);
2787 int treg
[3] = {-1, -1, -1};
2788 struct r600_bytecode_alu alu
;
2789 struct r600_bytecode_cf
*cf_jump
, *cf_pop
;
2791 /* only execute factor emission for invocation 0 */
2792 /* PRED_SETE_INT __, R0.x, 0 */
2793 memset(&alu
, 0, sizeof(alu
));
2794 alu
.op
= ALU_OP2_PRED_SETE_INT
;
2795 alu
.src
[0].chan
= 2;
2796 alu
.src
[1].sel
= V_SQ_ALU_SRC_LITERAL
;
2797 alu
.execute_mask
= 1;
2798 alu
.update_pred
= 1;
2800 r600_bytecode_add_alu_type(ctx
->bc
, &alu
, CF_OP_ALU_PUSH_BEFORE
);
2802 r600_bytecode_add_cfinst(ctx
->bc
, CF_OP_JUMP
);
2803 cf_jump
= ctx
->bc
->cf_last
;
2805 treg
[0] = r600_get_temp(ctx
);
2806 switch (ctx
->shader
->tcs_prim_mode
) {
2807 case PIPE_PRIM_LINES
:
2808 stride
= 8; /* 2 dwords, 1 vec2 store */
2812 case PIPE_PRIM_TRIANGLES
:
2813 stride
= 16; /* 4 dwords, 1 vec4 store */
2816 treg
[1] = r600_get_temp(ctx
);
2818 case PIPE_PRIM_QUADS
:
2819 stride
= 24; /* 6 dwords, 2 stores (vec4 + vec2) */
2822 treg
[1] = r600_get_temp(ctx
);
2823 treg
[2] = r600_get_temp(ctx
);
2830 /* R0 is InvocationID, RelPatchID, PatchID, tf_base */
2831 /* TF_WRITE takes index in R.x, value in R.y */
2832 for (i
= 0; i
< ctx
->shader
->noutput
; i
++) {
2833 if (ctx
->shader
->output
[i
].name
== TGSI_SEMANTIC_TESSINNER
)
2835 if (ctx
->shader
->output
[i
].name
== TGSI_SEMANTIC_TESSOUTER
)
2839 if (tessouter_idx
== -1)
2842 if (tessinner_idx
== -1 && inner_comps
)
2845 if (tessouter_idx
!= -1) {
2846 r
= r600_tess_factor_read(ctx
, tessouter_idx
);
2851 if (tessinner_idx
!= -1) {
2852 r
= r600_tess_factor_read(ctx
, tessinner_idx
);
2857 /* r.x = tf_base(r0.w) + relpatchid(r0.y) * tf_stride */
2858 /* r.x = relpatchid(r0.y) * tf_stride */
2860 /* multiply incoming r0.y * stride - t.x = r0.y * stride */
2861 /* add incoming r0.w to it: t.x = t.x + r0.w */
2862 r
= single_alu_op3(ctx
, ALU_OP3_MULADD_UINT24
,
2865 V_SQ_ALU_SRC_LITERAL
, stride
,
2870 for (i
= 0; i
< outer_comps
+ inner_comps
; i
++) {
2871 int out_idx
= i
>= outer_comps
? tessinner_idx
: tessouter_idx
;
2872 int out_comp
= i
>= outer_comps
? i
- outer_comps
: i
;
2874 r
= single_alu_op2(ctx
, ALU_OP2_ADD_INT
,
2875 treg
[i
/ 2], (2 * (i
% 2)),
2877 V_SQ_ALU_SRC_LITERAL
, 4 * i
);
2880 r
= single_alu_op2(ctx
, ALU_OP1_MOV
,
2881 treg
[i
/ 2], 1 + (2 * (i
%2)),
2882 ctx
->shader
->output
[out_idx
].gpr
, out_comp
,
2887 for (i
= 0; i
< outer_comps
+ inner_comps
; i
++) {
2888 struct r600_bytecode_gds gds
;
2890 memset(&gds
, 0, sizeof(struct r600_bytecode_gds
));
2891 gds
.src_gpr
= treg
[i
/ 2];
2892 gds
.src_sel_x
= 2 * (i
% 2);
2893 gds
.src_sel_y
= 1 + (2 * (i
% 2));
2899 gds
.op
= FETCH_OP_TF_WRITE
;
2900 r
= r600_bytecode_add_gds(ctx
->bc
, &gds
);
2905 // Patch up jump label
2906 r600_bytecode_add_cfinst(ctx
->bc
, CF_OP_POP
);
2907 cf_pop
= ctx
->bc
->cf_last
;
2909 cf_jump
->cf_addr
= cf_pop
->id
+ 2;
2910 cf_jump
->pop_count
= 1;
2911 cf_pop
->cf_addr
= cf_pop
->id
+ 2;
2912 cf_pop
->pop_count
= 1;
2917 static int r600_shader_from_tgsi(struct r600_context
*rctx
,
2918 struct r600_pipe_shader
*pipeshader
,
2919 union r600_shader_key key
)
2921 struct r600_screen
*rscreen
= rctx
->screen
;
2922 struct r600_shader
*shader
= &pipeshader
->shader
;
2923 struct tgsi_token
*tokens
= pipeshader
->selector
->tokens
;
2924 struct pipe_stream_output_info so
= pipeshader
->selector
->so
;
2925 struct tgsi_full_immediate
*immediate
;
2926 struct r600_shader_ctx ctx
;
2927 struct r600_bytecode_output output
[32];
2928 unsigned output_done
, noutput
;
2931 int next_param_base
= 0, next_clip_base
;
2932 int max_color_exports
= MAX2(key
.ps
.nr_cbufs
, 1);
2934 bool ring_outputs
= false;
2935 bool lds_outputs
= false;
2936 bool lds_inputs
= false;
2937 bool pos_emitted
= false;
2939 ctx
.bc
= &shader
->bc
;
2940 ctx
.shader
= shader
;
2941 ctx
.native_integers
= true;
2943 r600_bytecode_init(ctx
.bc
, rscreen
->b
.chip_class
, rscreen
->b
.family
,
2944 rscreen
->has_compressed_msaa_texturing
);
2945 ctx
.tokens
= tokens
;
2946 tgsi_scan_shader(tokens
, &ctx
.info
);
2947 shader
->indirect_files
= ctx
.info
.indirect_files
;
2949 shader
->uses_doubles
= ctx
.info
.uses_doubles
;
2951 indirect_gprs
= ctx
.info
.indirect_files
& ~((1 << TGSI_FILE_CONSTANT
) | (1 << TGSI_FILE_SAMPLER
));
2952 tgsi_parse_init(&ctx
.parse
, tokens
);
2953 ctx
.type
= ctx
.info
.processor
;
2954 shader
->processor_type
= ctx
.type
;
2955 ctx
.bc
->type
= shader
->processor_type
;
2958 case PIPE_SHADER_VERTEX
:
2959 shader
->vs_as_gs_a
= key
.vs
.as_gs_a
;
2960 shader
->vs_as_es
= key
.vs
.as_es
;
2961 shader
->vs_as_ls
= key
.vs
.as_ls
;
2962 if (shader
->vs_as_es
)
2963 ring_outputs
= true;
2964 if (shader
->vs_as_ls
)
2967 case PIPE_SHADER_GEOMETRY
:
2968 ring_outputs
= true;
2970 case PIPE_SHADER_TESS_CTRL
:
2971 shader
->tcs_prim_mode
= key
.tcs
.prim_mode
;
2975 case PIPE_SHADER_TESS_EVAL
:
2976 shader
->tes_as_es
= key
.tes
.as_es
;
2978 if (shader
->tes_as_es
)
2979 ring_outputs
= true;
2981 case PIPE_SHADER_FRAGMENT
:
2982 shader
->two_side
= key
.ps
.color_two_side
;
2988 if (shader
->vs_as_es
|| shader
->tes_as_es
) {
2989 ctx
.gs_for_vs
= &rctx
->gs_shader
->current
->shader
;
2991 ctx
.gs_for_vs
= NULL
;
2994 ctx
.next_ring_offset
= 0;
2995 ctx
.gs_out_ring_offset
= 0;
2996 ctx
.gs_next_vertex
= 0;
2997 ctx
.gs_stream_output_info
= &so
;
3000 ctx
.fixed_pt_position_gpr
= -1;
3001 ctx
.fragcoord_input
= -1;
3002 ctx
.colors_used
= 0;
3003 ctx
.clip_vertex_write
= 0;
3005 shader
->nr_ps_color_exports
= 0;
3006 shader
->nr_ps_max_color_exports
= 0;
3009 /* register allocations */
3010 /* Values [0,127] correspond to GPR[0..127].
3011 * Values [128,159] correspond to constant buffer bank 0
3012 * Values [160,191] correspond to constant buffer bank 1
3013 * Values [256,511] correspond to cfile constants c[0..255]. (Gone on EG)
3014 * Values [256,287] correspond to constant buffer bank 2 (EG)
3015 * Values [288,319] correspond to constant buffer bank 3 (EG)
3016 * Other special values are shown in the list below.
3017 * 244 ALU_SRC_1_DBL_L: special constant 1.0 double-float, LSW. (RV670+)
3018 * 245 ALU_SRC_1_DBL_M: special constant 1.0 double-float, MSW. (RV670+)
3019 * 246 ALU_SRC_0_5_DBL_L: special constant 0.5 double-float, LSW. (RV670+)
3020 * 247 ALU_SRC_0_5_DBL_M: special constant 0.5 double-float, MSW. (RV670+)
3021 * 248 SQ_ALU_SRC_0: special constant 0.0.
3022 * 249 SQ_ALU_SRC_1: special constant 1.0 float.
3023 * 250 SQ_ALU_SRC_1_INT: special constant 1 integer.
3024 * 251 SQ_ALU_SRC_M_1_INT: special constant -1 integer.
3025 * 252 SQ_ALU_SRC_0_5: special constant 0.5 float.
3026 * 253 SQ_ALU_SRC_LITERAL: literal constant.
3027 * 254 SQ_ALU_SRC_PV: previous vector result.
3028 * 255 SQ_ALU_SRC_PS: previous scalar result.
3030 for (i
= 0; i
< TGSI_FILE_COUNT
; i
++) {
3031 ctx
.file_offset
[i
] = 0;
3034 if (ctx
.type
== PIPE_SHADER_VERTEX
) {
3035 ctx
.file_offset
[TGSI_FILE_INPUT
] = 1;
3036 r600_bytecode_add_cfinst(ctx
.bc
, CF_OP_CALL_FS
);
3038 if (ctx
.type
== PIPE_SHADER_FRAGMENT
) {
3039 if (ctx
.bc
->chip_class
>= EVERGREEN
)
3040 ctx
.file_offset
[TGSI_FILE_INPUT
] = evergreen_gpr_count(&ctx
);
3042 ctx
.file_offset
[TGSI_FILE_INPUT
] = allocate_system_value_inputs(&ctx
, ctx
.file_offset
[TGSI_FILE_INPUT
]);
3044 if (ctx
.type
== PIPE_SHADER_GEOMETRY
) {
3045 /* FIXME 1 would be enough in some cases (3 or less input vertices) */
3046 ctx
.file_offset
[TGSI_FILE_INPUT
] = 2;
3048 if (ctx
.type
== PIPE_SHADER_TESS_CTRL
)
3049 ctx
.file_offset
[TGSI_FILE_INPUT
] = 1;
3050 if (ctx
.type
== PIPE_SHADER_TESS_EVAL
) {
3051 bool add_tesscoord
= false, add_tess_inout
= false;
3052 ctx
.file_offset
[TGSI_FILE_INPUT
] = 1;
3053 for (i
= 0; i
< PIPE_MAX_SHADER_INPUTS
; i
++) {
3054 /* if we have tesscoord save one reg */
3055 if (ctx
.info
.system_value_semantic_name
[i
] == TGSI_SEMANTIC_TESSCOORD
)
3056 add_tesscoord
= true;
3057 if (ctx
.info
.system_value_semantic_name
[i
] == TGSI_SEMANTIC_TESSINNER
||
3058 ctx
.info
.system_value_semantic_name
[i
] == TGSI_SEMANTIC_TESSOUTER
)
3059 add_tess_inout
= true;
3061 if (add_tesscoord
|| add_tess_inout
)
3062 ctx
.file_offset
[TGSI_FILE_INPUT
]++;
3064 ctx
.file_offset
[TGSI_FILE_INPUT
]+=2;
3067 ctx
.file_offset
[TGSI_FILE_OUTPUT
] =
3068 ctx
.file_offset
[TGSI_FILE_INPUT
] +
3069 ctx
.info
.file_max
[TGSI_FILE_INPUT
] + 1;
3070 ctx
.file_offset
[TGSI_FILE_TEMPORARY
] = ctx
.file_offset
[TGSI_FILE_OUTPUT
] +
3071 ctx
.info
.file_max
[TGSI_FILE_OUTPUT
] + 1;
3073 /* Outside the GPR range. This will be translated to one of the
3074 * kcache banks later. */
3075 ctx
.file_offset
[TGSI_FILE_CONSTANT
] = 512;
3077 ctx
.file_offset
[TGSI_FILE_IMMEDIATE
] = V_SQ_ALU_SRC_LITERAL
;
3078 ctx
.bc
->ar_reg
= ctx
.file_offset
[TGSI_FILE_TEMPORARY
] +
3079 ctx
.info
.file_max
[TGSI_FILE_TEMPORARY
] + 1;
3080 ctx
.bc
->index_reg
[0] = ctx
.bc
->ar_reg
+ 1;
3081 ctx
.bc
->index_reg
[1] = ctx
.bc
->ar_reg
+ 2;
3083 if (ctx
.type
== PIPE_SHADER_TESS_CTRL
) {
3084 ctx
.tess_input_info
= ctx
.bc
->ar_reg
+ 3;
3085 ctx
.tess_output_info
= ctx
.bc
->ar_reg
+ 4;
3086 ctx
.temp_reg
= ctx
.bc
->ar_reg
+ 5;
3087 } else if (ctx
.type
== PIPE_SHADER_TESS_EVAL
) {
3088 ctx
.tess_input_info
= 0;
3089 ctx
.tess_output_info
= ctx
.bc
->ar_reg
+ 3;
3090 ctx
.temp_reg
= ctx
.bc
->ar_reg
+ 4;
3091 } else if (ctx
.type
== PIPE_SHADER_GEOMETRY
) {
3092 ctx
.gs_export_gpr_tregs
[0] = ctx
.bc
->ar_reg
+ 3;
3093 ctx
.gs_export_gpr_tregs
[1] = ctx
.bc
->ar_reg
+ 4;
3094 ctx
.gs_export_gpr_tregs
[2] = ctx
.bc
->ar_reg
+ 5;
3095 ctx
.gs_export_gpr_tregs
[3] = ctx
.bc
->ar_reg
+ 6;
3096 ctx
.temp_reg
= ctx
.bc
->ar_reg
+ 7;
3098 ctx
.temp_reg
= ctx
.bc
->ar_reg
+ 3;
3101 shader
->max_arrays
= 0;
3102 shader
->num_arrays
= 0;
3103 if (indirect_gprs
) {
3105 if (ctx
.info
.indirect_files
& (1 << TGSI_FILE_INPUT
)) {
3106 r600_add_gpr_array(shader
, ctx
.file_offset
[TGSI_FILE_INPUT
],
3107 ctx
.file_offset
[TGSI_FILE_OUTPUT
] -
3108 ctx
.file_offset
[TGSI_FILE_INPUT
],
3111 if (ctx
.info
.indirect_files
& (1 << TGSI_FILE_OUTPUT
)) {
3112 r600_add_gpr_array(shader
, ctx
.file_offset
[TGSI_FILE_OUTPUT
],
3113 ctx
.file_offset
[TGSI_FILE_TEMPORARY
] -
3114 ctx
.file_offset
[TGSI_FILE_OUTPUT
],
3120 ctx
.literals
= NULL
;
3122 shader
->fs_write_all
= ctx
.info
.properties
[TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS
] &&
3123 ctx
.info
.colors_written
== 1;
3124 shader
->vs_position_window_space
= ctx
.info
.properties
[TGSI_PROPERTY_VS_WINDOW_SPACE_POSITION
];
3125 shader
->ps_conservative_z
= (uint8_t)ctx
.info
.properties
[TGSI_PROPERTY_FS_DEPTH_LAYOUT
];
3127 if (shader
->vs_as_gs_a
)
3128 vs_add_primid_output(&ctx
, key
.vs
.prim_id_out
);
3130 if (ctx
.type
== PIPE_SHADER_TESS_EVAL
)
3131 r600_fetch_tess_io_info(&ctx
);
3133 while (!tgsi_parse_end_of_tokens(&ctx
.parse
)) {
3134 tgsi_parse_token(&ctx
.parse
);
3135 switch (ctx
.parse
.FullToken
.Token
.Type
) {
3136 case TGSI_TOKEN_TYPE_IMMEDIATE
:
3137 immediate
= &ctx
.parse
.FullToken
.FullImmediate
;
3138 ctx
.literals
= realloc(ctx
.literals
, (ctx
.nliterals
+ 1) * 16);
3139 if(ctx
.literals
== NULL
) {
3143 ctx
.literals
[ctx
.nliterals
* 4 + 0] = immediate
->u
[0].Uint
;
3144 ctx
.literals
[ctx
.nliterals
* 4 + 1] = immediate
->u
[1].Uint
;
3145 ctx
.literals
[ctx
.nliterals
* 4 + 2] = immediate
->u
[2].Uint
;
3146 ctx
.literals
[ctx
.nliterals
* 4 + 3] = immediate
->u
[3].Uint
;
3149 case TGSI_TOKEN_TYPE_DECLARATION
:
3150 r
= tgsi_declaration(&ctx
);
3154 case TGSI_TOKEN_TYPE_INSTRUCTION
:
3155 case TGSI_TOKEN_TYPE_PROPERTY
:
3158 R600_ERR("unsupported token type %d\n", ctx
.parse
.FullToken
.Token
.Type
);
3164 shader
->ring_item_sizes
[0] = ctx
.next_ring_offset
;
3165 shader
->ring_item_sizes
[1] = 0;
3166 shader
->ring_item_sizes
[2] = 0;
3167 shader
->ring_item_sizes
[3] = 0;
3169 /* Process two side if needed */
3170 if (shader
->two_side
&& ctx
.colors_used
) {
3171 int i
, count
= ctx
.shader
->ninput
;
3172 unsigned next_lds_loc
= ctx
.shader
->nlds
;
3174 /* additional inputs will be allocated right after the existing inputs,
3175 * we won't need them after the color selection, so we don't need to
3176 * reserve these gprs for the rest of the shader code and to adjust
3177 * output offsets etc. */
3178 int gpr
= ctx
.file_offset
[TGSI_FILE_INPUT
] +
3179 ctx
.info
.file_max
[TGSI_FILE_INPUT
] + 1;
3181 /* if two sided and neither face or sample mask is used by shader, ensure face_gpr is emitted */
3182 if (ctx
.face_gpr
== -1) {
3183 i
= ctx
.shader
->ninput
++;
3184 ctx
.shader
->input
[i
].name
= TGSI_SEMANTIC_FACE
;
3185 ctx
.shader
->input
[i
].spi_sid
= 0;
3186 ctx
.shader
->input
[i
].gpr
= gpr
++;
3187 ctx
.face_gpr
= ctx
.shader
->input
[i
].gpr
;
3190 for (i
= 0; i
< count
; i
++) {
3191 if (ctx
.shader
->input
[i
].name
== TGSI_SEMANTIC_COLOR
) {
3192 int ni
= ctx
.shader
->ninput
++;
3193 memcpy(&ctx
.shader
->input
[ni
],&ctx
.shader
->input
[i
], sizeof(struct r600_shader_io
));
3194 ctx
.shader
->input
[ni
].name
= TGSI_SEMANTIC_BCOLOR
;
3195 ctx
.shader
->input
[ni
].spi_sid
= r600_spi_sid(&ctx
.shader
->input
[ni
]);
3196 ctx
.shader
->input
[ni
].gpr
= gpr
++;
3197 // TGSI to LLVM needs to know the lds position of inputs.
3198 // Non LLVM path computes it later (in process_twoside_color)
3199 ctx
.shader
->input
[ni
].lds_pos
= next_lds_loc
++;
3200 ctx
.shader
->input
[i
].back_color_input
= ni
;
3201 if (ctx
.bc
->chip_class
>= EVERGREEN
) {
3202 if ((r
= evergreen_interp_input(&ctx
, ni
)))
3209 if (shader
->fs_write_all
&& rscreen
->b
.chip_class
>= EVERGREEN
)
3210 shader
->nr_ps_max_color_exports
= 8;
3212 if (ctx
.fragcoord_input
>= 0) {
3213 if (ctx
.bc
->chip_class
== CAYMAN
) {
3214 for (j
= 0 ; j
< 4; j
++) {
3215 struct r600_bytecode_alu alu
;
3216 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3217 alu
.op
= ALU_OP1_RECIP_IEEE
;
3218 alu
.src
[0].sel
= shader
->input
[ctx
.fragcoord_input
].gpr
;
3219 alu
.src
[0].chan
= 3;
3221 alu
.dst
.sel
= shader
->input
[ctx
.fragcoord_input
].gpr
;
3223 alu
.dst
.write
= (j
== 3);
3225 if ((r
= r600_bytecode_add_alu(ctx
.bc
, &alu
)))
3229 struct r600_bytecode_alu alu
;
3230 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3231 alu
.op
= ALU_OP1_RECIP_IEEE
;
3232 alu
.src
[0].sel
= shader
->input
[ctx
.fragcoord_input
].gpr
;
3233 alu
.src
[0].chan
= 3;
3235 alu
.dst
.sel
= shader
->input
[ctx
.fragcoord_input
].gpr
;
3239 if ((r
= r600_bytecode_add_alu(ctx
.bc
, &alu
)))
3244 if (ctx
.type
== PIPE_SHADER_GEOMETRY
) {
3245 struct r600_bytecode_alu alu
;
3248 /* GS thread with no output workaround - emit a cut at start of GS */
3249 if (ctx
.bc
->chip_class
== R600
)
3250 r600_bytecode_add_cfinst(ctx
.bc
, CF_OP_CUT_VERTEX
);
3252 for (j
= 0; j
< 4; j
++) {
3253 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3254 alu
.op
= ALU_OP1_MOV
;
3255 alu
.src
[0].sel
= V_SQ_ALU_SRC_LITERAL
;
3256 alu
.src
[0].value
= 0;
3257 alu
.dst
.sel
= ctx
.gs_export_gpr_tregs
[j
];
3260 r
= r600_bytecode_add_alu(ctx
.bc
, &alu
);
3266 if (ctx
.type
== PIPE_SHADER_TESS_CTRL
)
3267 r600_fetch_tess_io_info(&ctx
);
3269 if (shader
->two_side
&& ctx
.colors_used
) {
3270 if ((r
= process_twoside_color_inputs(&ctx
)))
3274 tgsi_parse_init(&ctx
.parse
, tokens
);
3275 while (!tgsi_parse_end_of_tokens(&ctx
.parse
)) {
3276 tgsi_parse_token(&ctx
.parse
);
3277 switch (ctx
.parse
.FullToken
.Token
.Type
) {
3278 case TGSI_TOKEN_TYPE_INSTRUCTION
:
3279 r
= tgsi_is_supported(&ctx
);
3282 ctx
.max_driver_temp_used
= 0;
3283 /* reserve first tmp for everyone */
3284 r600_get_temp(&ctx
);
3286 opcode
= ctx
.parse
.FullToken
.FullInstruction
.Instruction
.Opcode
;
3287 if ((r
= tgsi_split_constant(&ctx
)))
3289 if ((r
= tgsi_split_literal_constant(&ctx
)))
3291 if (ctx
.type
== PIPE_SHADER_GEOMETRY
) {
3292 if ((r
= tgsi_split_gs_inputs(&ctx
)))
3294 } else if (lds_inputs
) {
3295 if ((r
= tgsi_split_lds_inputs(&ctx
)))
3298 if (ctx
.bc
->chip_class
== CAYMAN
)
3299 ctx
.inst_info
= &cm_shader_tgsi_instruction
[opcode
];
3300 else if (ctx
.bc
->chip_class
>= EVERGREEN
)
3301 ctx
.inst_info
= &eg_shader_tgsi_instruction
[opcode
];
3303 ctx
.inst_info
= &r600_shader_tgsi_instruction
[opcode
];
3304 r
= ctx
.inst_info
->process(&ctx
);
3308 if (ctx
.type
== PIPE_SHADER_TESS_CTRL
) {
3309 r
= r600_store_tcs_output(&ctx
);
3319 /* Reset the temporary register counter. */
3320 ctx
.max_driver_temp_used
= 0;
3322 noutput
= shader
->noutput
;
3324 if (!ring_outputs
&& ctx
.clip_vertex_write
) {
3325 unsigned clipdist_temp
[2];
3327 clipdist_temp
[0] = r600_get_temp(&ctx
);
3328 clipdist_temp
[1] = r600_get_temp(&ctx
);
3330 /* need to convert a clipvertex write into clipdistance writes and not export
3331 the clip vertex anymore */
3333 memset(&shader
->output
[noutput
], 0, 2*sizeof(struct r600_shader_io
));
3334 shader
->output
[noutput
].name
= TGSI_SEMANTIC_CLIPDIST
;
3335 shader
->output
[noutput
].gpr
= clipdist_temp
[0];
3337 shader
->output
[noutput
].name
= TGSI_SEMANTIC_CLIPDIST
;
3338 shader
->output
[noutput
].gpr
= clipdist_temp
[1];
3341 /* reset spi_sid for clipvertex output to avoid confusing spi */
3342 shader
->output
[ctx
.cv_output
].spi_sid
= 0;
3344 shader
->clip_dist_write
= 0xFF;
3346 for (i
= 0; i
< 8; i
++) {
3350 for (j
= 0; j
< 4; j
++) {
3351 struct r600_bytecode_alu alu
;
3352 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3353 alu
.op
= ALU_OP2_DOT4
;
3354 alu
.src
[0].sel
= shader
->output
[ctx
.cv_output
].gpr
;
3355 alu
.src
[0].chan
= j
;
3357 alu
.src
[1].sel
= 512 + i
;
3358 alu
.src
[1].kc_bank
= R600_BUFFER_INFO_CONST_BUFFER
;
3359 alu
.src
[1].chan
= j
;
3361 alu
.dst
.sel
= clipdist_temp
[oreg
];
3363 alu
.dst
.write
= (j
== ochan
);
3366 r
= r600_bytecode_add_alu(ctx
.bc
, &alu
);
3373 /* Add stream outputs. */
3374 if (so
.num_outputs
) {
3376 if (!lds_outputs
&& !ring_outputs
&& ctx
.type
== PIPE_SHADER_VERTEX
)
3378 if (!ring_outputs
&& ctx
.type
== PIPE_SHADER_TESS_EVAL
)
3381 emit_streamout(&ctx
, &so
, -1, NULL
);
3383 pipeshader
->enabled_stream_buffers_mask
= ctx
.enabled_stream_buffers_mask
;
3384 convert_edgeflag_to_int(&ctx
);
3386 if (ctx
.type
== PIPE_SHADER_TESS_CTRL
)
3387 r600_emit_tess_factor(&ctx
);
3390 if (ctx
.type
== PIPE_SHADER_VERTEX
) {
3391 if (ctx
.shader
->noutput
)
3392 emit_lds_vs_writes(&ctx
);
3394 } else if (ring_outputs
) {
3395 if (shader
->vs_as_es
|| shader
->tes_as_es
) {
3396 ctx
.gs_export_gpr_tregs
[0] = r600_get_temp(&ctx
);
3397 ctx
.gs_export_gpr_tregs
[1] = -1;
3398 ctx
.gs_export_gpr_tregs
[2] = -1;
3399 ctx
.gs_export_gpr_tregs
[3] = -1;
3401 emit_gs_ring_writes(&ctx
, &so
, -1, FALSE
);
3405 next_clip_base
= shader
->vs_out_misc_write
? 62 : 61;
3407 for (i
= 0, j
= 0; i
< noutput
; i
++, j
++) {
3408 memset(&output
[j
], 0, sizeof(struct r600_bytecode_output
));
3409 output
[j
].gpr
= shader
->output
[i
].gpr
;
3410 output
[j
].elem_size
= 3;
3411 output
[j
].swizzle_x
= 0;
3412 output
[j
].swizzle_y
= 1;
3413 output
[j
].swizzle_z
= 2;
3414 output
[j
].swizzle_w
= 3;
3415 output
[j
].burst_count
= 1;
3416 output
[j
].type
= -1;
3417 output
[j
].op
= CF_OP_EXPORT
;
3419 case PIPE_SHADER_VERTEX
:
3420 case PIPE_SHADER_TESS_EVAL
:
3421 switch (shader
->output
[i
].name
) {
3422 case TGSI_SEMANTIC_POSITION
:
3423 output
[j
].array_base
= 60;
3424 output
[j
].type
= V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_POS
;
3428 case TGSI_SEMANTIC_PSIZE
:
3429 output
[j
].array_base
= 61;
3430 output
[j
].swizzle_y
= 7;
3431 output
[j
].swizzle_z
= 7;
3432 output
[j
].swizzle_w
= 7;
3433 output
[j
].type
= V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_POS
;
3436 case TGSI_SEMANTIC_EDGEFLAG
:
3437 output
[j
].array_base
= 61;
3438 output
[j
].swizzle_x
= 7;
3439 output
[j
].swizzle_y
= 0;
3440 output
[j
].swizzle_z
= 7;
3441 output
[j
].swizzle_w
= 7;
3442 output
[j
].type
= V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_POS
;
3445 case TGSI_SEMANTIC_LAYER
:
3446 /* spi_sid is 0 for outputs that are
3447 * not consumed by PS */
3448 if (shader
->output
[i
].spi_sid
) {
3449 output
[j
].array_base
= next_param_base
++;
3450 output
[j
].type
= V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PARAM
;
3452 memcpy(&output
[j
], &output
[j
-1], sizeof(struct r600_bytecode_output
));
3454 output
[j
].array_base
= 61;
3455 output
[j
].swizzle_x
= 7;
3456 output
[j
].swizzle_y
= 7;
3457 output
[j
].swizzle_z
= 0;
3458 output
[j
].swizzle_w
= 7;
3459 output
[j
].type
= V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_POS
;
3462 case TGSI_SEMANTIC_VIEWPORT_INDEX
:
3463 /* spi_sid is 0 for outputs that are
3464 * not consumed by PS */
3465 if (shader
->output
[i
].spi_sid
) {
3466 output
[j
].array_base
= next_param_base
++;
3467 output
[j
].type
= V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PARAM
;
3469 memcpy(&output
[j
], &output
[j
-1], sizeof(struct r600_bytecode_output
));
3471 output
[j
].array_base
= 61;
3472 output
[j
].swizzle_x
= 7;
3473 output
[j
].swizzle_y
= 7;
3474 output
[j
].swizzle_z
= 7;
3475 output
[j
].swizzle_w
= 0;
3476 output
[j
].type
= V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_POS
;
3479 case TGSI_SEMANTIC_CLIPVERTEX
:
3482 case TGSI_SEMANTIC_CLIPDIST
:
3483 output
[j
].array_base
= next_clip_base
++;
3484 output
[j
].type
= V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_POS
;
3486 /* spi_sid is 0 for clipdistance outputs that were generated
3487 * for clipvertex - we don't need to pass them to PS */
3488 if (shader
->output
[i
].spi_sid
) {
3490 /* duplicate it as PARAM to pass to the pixel shader */
3491 memcpy(&output
[j
], &output
[j
-1], sizeof(struct r600_bytecode_output
));
3492 output
[j
].array_base
= next_param_base
++;
3493 output
[j
].type
= V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PARAM
;
3496 case TGSI_SEMANTIC_FOG
:
3497 output
[j
].swizzle_y
= 4; /* 0 */
3498 output
[j
].swizzle_z
= 4; /* 0 */
3499 output
[j
].swizzle_w
= 5; /* 1 */
3501 case TGSI_SEMANTIC_PRIMID
:
3502 output
[j
].swizzle_x
= 2;
3503 output
[j
].swizzle_y
= 4; /* 0 */
3504 output
[j
].swizzle_z
= 4; /* 0 */
3505 output
[j
].swizzle_w
= 4; /* 0 */
3510 case PIPE_SHADER_FRAGMENT
:
3511 if (shader
->output
[i
].name
== TGSI_SEMANTIC_COLOR
) {
3512 /* never export more colors than the number of CBs */
3513 if (shader
->output
[i
].sid
>= max_color_exports
) {
3518 output
[j
].swizzle_w
= key
.ps
.alpha_to_one
? 5 : 3;
3519 output
[j
].array_base
= shader
->output
[i
].sid
;
3520 output
[j
].type
= V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PIXEL
;
3521 shader
->nr_ps_color_exports
++;
3522 if (shader
->fs_write_all
&& (rscreen
->b
.chip_class
>= EVERGREEN
)) {
3523 for (k
= 1; k
< max_color_exports
; k
++) {
3525 memset(&output
[j
], 0, sizeof(struct r600_bytecode_output
));
3526 output
[j
].gpr
= shader
->output
[i
].gpr
;
3527 output
[j
].elem_size
= 3;
3528 output
[j
].swizzle_x
= 0;
3529 output
[j
].swizzle_y
= 1;
3530 output
[j
].swizzle_z
= 2;
3531 output
[j
].swizzle_w
= key
.ps
.alpha_to_one
? 5 : 3;
3532 output
[j
].burst_count
= 1;
3533 output
[j
].array_base
= k
;
3534 output
[j
].op
= CF_OP_EXPORT
;
3535 output
[j
].type
= V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PIXEL
;
3536 shader
->nr_ps_color_exports
++;
3539 } else if (shader
->output
[i
].name
== TGSI_SEMANTIC_POSITION
) {
3540 output
[j
].array_base
= 61;
3541 output
[j
].swizzle_x
= 2;
3542 output
[j
].swizzle_y
= 7;
3543 output
[j
].swizzle_z
= output
[j
].swizzle_w
= 7;
3544 output
[j
].type
= V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PIXEL
;
3545 } else if (shader
->output
[i
].name
== TGSI_SEMANTIC_STENCIL
) {
3546 output
[j
].array_base
= 61;
3547 output
[j
].swizzle_x
= 7;
3548 output
[j
].swizzle_y
= 1;
3549 output
[j
].swizzle_z
= output
[j
].swizzle_w
= 7;
3550 output
[j
].type
= V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PIXEL
;
3551 } else if (shader
->output
[i
].name
== TGSI_SEMANTIC_SAMPLEMASK
) {
3552 output
[j
].array_base
= 61;
3553 output
[j
].swizzle_x
= 7;
3554 output
[j
].swizzle_y
= 7;
3555 output
[j
].swizzle_z
= 0;
3556 output
[j
].swizzle_w
= 7;
3557 output
[j
].type
= V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PIXEL
;
3559 R600_ERR("unsupported fragment output name %d\n", shader
->output
[i
].name
);
3564 case PIPE_SHADER_TESS_CTRL
:
3567 R600_ERR("unsupported processor type %d\n", ctx
.type
);
3572 if (output
[j
].type
==-1) {
3573 output
[j
].type
= V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PARAM
;
3574 output
[j
].array_base
= next_param_base
++;
3578 /* add fake position export */
3579 if ((ctx
.type
== PIPE_SHADER_VERTEX
|| ctx
.type
== PIPE_SHADER_TESS_EVAL
) && pos_emitted
== false) {
3580 memset(&output
[j
], 0, sizeof(struct r600_bytecode_output
));
3582 output
[j
].elem_size
= 3;
3583 output
[j
].swizzle_x
= 7;
3584 output
[j
].swizzle_y
= 7;
3585 output
[j
].swizzle_z
= 7;
3586 output
[j
].swizzle_w
= 7;
3587 output
[j
].burst_count
= 1;
3588 output
[j
].type
= V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_POS
;
3589 output
[j
].array_base
= 60;
3590 output
[j
].op
= CF_OP_EXPORT
;
3594 /* add fake param output for vertex shader if no param is exported */
3595 if ((ctx
.type
== PIPE_SHADER_VERTEX
|| ctx
.type
== PIPE_SHADER_TESS_EVAL
) && next_param_base
== 0) {
3596 memset(&output
[j
], 0, sizeof(struct r600_bytecode_output
));
3598 output
[j
].elem_size
= 3;
3599 output
[j
].swizzle_x
= 7;
3600 output
[j
].swizzle_y
= 7;
3601 output
[j
].swizzle_z
= 7;
3602 output
[j
].swizzle_w
= 7;
3603 output
[j
].burst_count
= 1;
3604 output
[j
].type
= V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PARAM
;
3605 output
[j
].array_base
= 0;
3606 output
[j
].op
= CF_OP_EXPORT
;
3610 /* add fake pixel export */
3611 if (ctx
.type
== PIPE_SHADER_FRAGMENT
&& shader
->nr_ps_color_exports
== 0) {
3612 memset(&output
[j
], 0, sizeof(struct r600_bytecode_output
));
3614 output
[j
].elem_size
= 3;
3615 output
[j
].swizzle_x
= 7;
3616 output
[j
].swizzle_y
= 7;
3617 output
[j
].swizzle_z
= 7;
3618 output
[j
].swizzle_w
= 7;
3619 output
[j
].burst_count
= 1;
3620 output
[j
].type
= V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PIXEL
;
3621 output
[j
].array_base
= 0;
3622 output
[j
].op
= CF_OP_EXPORT
;
3624 shader
->nr_ps_color_exports
++;
3629 /* set export done on last export of each type */
3630 for (i
= noutput
- 1, output_done
= 0; i
>= 0; i
--) {
3631 if (!(output_done
& (1 << output
[i
].type
))) {
3632 output_done
|= (1 << output
[i
].type
);
3633 output
[i
].op
= CF_OP_EXPORT_DONE
;
3636 /* add output to bytecode */
3637 for (i
= 0; i
< noutput
; i
++) {
3638 r
= r600_bytecode_add_output(ctx
.bc
, &output
[i
]);
3644 /* add program end */
3645 if (ctx
.bc
->chip_class
== CAYMAN
)
3646 cm_bytecode_add_cf_end(ctx
.bc
);
3648 const struct cf_op_info
*last
= NULL
;
3650 if (ctx
.bc
->cf_last
)
3651 last
= r600_isa_cf(ctx
.bc
->cf_last
->op
);
3653 /* alu clause instructions don't have EOP bit, so add NOP */
3654 if (!last
|| last
->flags
& CF_ALU
|| ctx
.bc
->cf_last
->op
== CF_OP_LOOP_END
|| ctx
.bc
->cf_last
->op
== CF_OP_CALL_FS
|| ctx
.bc
->cf_last
->op
== CF_OP_POP
|| ctx
.bc
->cf_last
->op
== CF_OP_GDS
)
3655 r600_bytecode_add_cfinst(ctx
.bc
, CF_OP_NOP
);
3657 ctx
.bc
->cf_last
->end_of_program
= 1;
3660 /* check GPR limit - we have 124 = 128 - 4
3661 * (4 are reserved as alu clause temporary registers) */
3662 if (ctx
.bc
->ngpr
> 124) {
3663 R600_ERR("GPR limit exceeded - shader requires %d registers\n", ctx
.bc
->ngpr
);
3668 if (ctx
.type
== PIPE_SHADER_GEOMETRY
) {
3669 if ((r
= generate_gs_copy_shader(rctx
, pipeshader
, &so
)))
3674 tgsi_parse_free(&ctx
.parse
);
3678 tgsi_parse_free(&ctx
.parse
);
3682 static int tgsi_unsupported(struct r600_shader_ctx
*ctx
)
3684 const unsigned tgsi_opcode
=
3685 ctx
->parse
.FullToken
.FullInstruction
.Instruction
.Opcode
;
3686 R600_ERR("%s tgsi opcode unsupported\n",
3687 tgsi_get_opcode_name(tgsi_opcode
));
3691 static int tgsi_end(struct r600_shader_ctx
*ctx
)
3696 static void r600_bytecode_src(struct r600_bytecode_alu_src
*bc_src
,
3697 const struct r600_shader_src
*shader_src
,
3700 bc_src
->sel
= shader_src
->sel
;
3701 bc_src
->chan
= shader_src
->swizzle
[chan
];
3702 bc_src
->neg
= shader_src
->neg
;
3703 bc_src
->abs
= shader_src
->abs
;
3704 bc_src
->rel
= shader_src
->rel
;
3705 bc_src
->value
= shader_src
->value
[bc_src
->chan
];
3706 bc_src
->kc_bank
= shader_src
->kc_bank
;
3707 bc_src
->kc_rel
= shader_src
->kc_rel
;
3710 static void r600_bytecode_src_set_abs(struct r600_bytecode_alu_src
*bc_src
)
3716 static void r600_bytecode_src_toggle_neg(struct r600_bytecode_alu_src
*bc_src
)
3718 bc_src
->neg
= !bc_src
->neg
;
3721 static void tgsi_dst(struct r600_shader_ctx
*ctx
,
3722 const struct tgsi_full_dst_register
*tgsi_dst
,
3724 struct r600_bytecode_alu_dst
*r600_dst
)
3726 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
3728 r600_dst
->sel
= tgsi_dst
->Register
.Index
;
3729 r600_dst
->sel
+= ctx
->file_offset
[tgsi_dst
->Register
.File
];
3730 r600_dst
->chan
= swizzle
;
3731 r600_dst
->write
= 1;
3732 if (inst
->Instruction
.Saturate
) {
3733 r600_dst
->clamp
= 1;
3735 if (ctx
->type
== PIPE_SHADER_TESS_CTRL
) {
3736 if (tgsi_dst
->Register
.File
== TGSI_FILE_OUTPUT
) {
3740 if (tgsi_dst
->Register
.Indirect
)
3741 r600_dst
->rel
= V_SQ_REL_RELATIVE
;
3745 static int tgsi_op2_64_params(struct r600_shader_ctx
*ctx
, bool singledest
, bool swap
)
3747 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
3748 unsigned write_mask
= inst
->Dst
[0].Register
.WriteMask
;
3749 struct r600_bytecode_alu alu
;
3750 int i
, j
, r
, lasti
= tgsi_last_instruction(write_mask
);
3754 switch (write_mask
) {
3772 lasti
= tgsi_last_instruction(write_mask
);
3773 for (i
= 0; i
<= lasti
; i
++) {
3775 if (!(write_mask
& (1 << i
)))
3778 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3781 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
3783 alu
.dst
.sel
= ctx
->temp_reg
;
3787 if (i
== 1 || i
== 3)
3790 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
3792 alu
.op
= ctx
->inst_info
->op
;
3793 if (ctx
->parse
.FullToken
.FullInstruction
.Instruction
.Opcode
== TGSI_OPCODE_DABS
) {
3794 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], i
);
3796 for (j
= 0; j
< inst
->Instruction
.NumSrcRegs
; j
++) {
3797 r600_bytecode_src(&alu
.src
[j
], &ctx
->src
[j
], fp64_switch(i
));
3800 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[1], fp64_switch(i
));
3801 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[0], fp64_switch(i
));
3804 /* handle some special cases */
3805 if (i
== 1 || i
== 3) {
3806 switch (ctx
->parse
.FullToken
.FullInstruction
.Instruction
.Opcode
) {
3807 case TGSI_OPCODE_SUB
:
3808 r600_bytecode_src_toggle_neg(&alu
.src
[1]);
3810 case TGSI_OPCODE_DABS
:
3811 r600_bytecode_src_set_abs(&alu
.src
[0]);
3820 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3826 write_mask
= inst
->Dst
[0].Register
.WriteMask
;
3828 /* move result from temp to dst */
3829 for (i
= 0; i
<= lasti
; i
++) {
3830 if (!(write_mask
& (1 << i
)))
3833 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3834 alu
.op
= ALU_OP1_MOV
;
3835 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
3836 alu
.src
[0].sel
= ctx
->temp_reg
;
3837 alu
.src
[0].chan
= use_tmp
- 1;
3838 alu
.last
= (i
== lasti
);
3840 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3848 static int tgsi_op2_64(struct r600_shader_ctx
*ctx
)
3850 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
3851 unsigned write_mask
= inst
->Dst
[0].Register
.WriteMask
;
3852 /* confirm writemasking */
3853 if ((write_mask
& 0x3) != 0x3 &&
3854 (write_mask
& 0xc) != 0xc) {
3855 fprintf(stderr
, "illegal writemask for 64-bit: 0x%x\n", write_mask
);
3858 return tgsi_op2_64_params(ctx
, false, false);
3861 static int tgsi_op2_64_single_dest(struct r600_shader_ctx
*ctx
)
3863 return tgsi_op2_64_params(ctx
, true, false);
3866 static int tgsi_op2_64_single_dest_s(struct r600_shader_ctx
*ctx
)
3868 return tgsi_op2_64_params(ctx
, true, true);
3871 static int tgsi_op3_64(struct r600_shader_ctx
*ctx
)
3873 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
3874 struct r600_bytecode_alu alu
;
3877 int tmp
= r600_get_temp(ctx
);
3879 for (i
= 0; i
< lasti
+ 1; i
++) {
3881 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3882 alu
.op
= ctx
->inst_info
->op
;
3883 for (j
= 0; j
< inst
->Instruction
.NumSrcRegs
; j
++) {
3884 r600_bytecode_src(&alu
.src
[j
], &ctx
->src
[j
], i
== 3 ? 0 : 1);
3887 if (inst
->Dst
[0].Register
.WriteMask
& (1 << i
))
3888 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
3897 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3904 static int tgsi_op2_s(struct r600_shader_ctx
*ctx
, int swap
, int trans_only
)
3906 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
3907 struct r600_bytecode_alu alu
;
3908 unsigned write_mask
= inst
->Dst
[0].Register
.WriteMask
;
3909 int i
, j
, r
, lasti
= tgsi_last_instruction(write_mask
);
3910 /* use temp register if trans_only and more than one dst component */
3911 int use_tmp
= trans_only
&& (write_mask
^ (1 << lasti
));
3913 for (i
= 0; i
<= lasti
; i
++) {
3914 if (!(write_mask
& (1 << i
)))
3917 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3919 alu
.dst
.sel
= ctx
->temp_reg
;
3923 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
3925 alu
.op
= ctx
->inst_info
->op
;
3927 for (j
= 0; j
< inst
->Instruction
.NumSrcRegs
; j
++) {
3928 r600_bytecode_src(&alu
.src
[j
], &ctx
->src
[j
], i
);
3931 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[1], i
);
3932 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[0], i
);
3934 /* handle some special cases */
3935 switch (inst
->Instruction
.Opcode
) {
3936 case TGSI_OPCODE_SUB
:
3937 r600_bytecode_src_toggle_neg(&alu
.src
[1]);
3942 if (i
== lasti
|| trans_only
) {
3945 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3951 /* move result from temp to dst */
3952 for (i
= 0; i
<= lasti
; i
++) {
3953 if (!(write_mask
& (1 << i
)))
3956 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3957 alu
.op
= ALU_OP1_MOV
;
3958 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
3959 alu
.src
[0].sel
= ctx
->temp_reg
;
3960 alu
.src
[0].chan
= i
;
3961 alu
.last
= (i
== lasti
);
3963 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
3971 static int tgsi_op2(struct r600_shader_ctx
*ctx
)
3973 return tgsi_op2_s(ctx
, 0, 0);
3976 static int tgsi_op2_swap(struct r600_shader_ctx
*ctx
)
3978 return tgsi_op2_s(ctx
, 1, 0);
3981 static int tgsi_op2_trans(struct r600_shader_ctx
*ctx
)
3983 return tgsi_op2_s(ctx
, 0, 1);
3986 static int tgsi_ineg(struct r600_shader_ctx
*ctx
)
3988 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
3989 struct r600_bytecode_alu alu
;
3991 int lasti
= tgsi_last_instruction(inst
->Dst
[0].Register
.WriteMask
);
3993 for (i
= 0; i
< lasti
+ 1; i
++) {
3995 if (!(inst
->Dst
[0].Register
.WriteMask
& (1 << i
)))
3997 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
3998 alu
.op
= ctx
->inst_info
->op
;
4000 alu
.src
[0].sel
= V_SQ_ALU_SRC_0
;
4002 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[0], i
);
4004 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
4009 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4017 static int tgsi_dneg(struct r600_shader_ctx
*ctx
)
4019 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
4020 struct r600_bytecode_alu alu
;
4022 int lasti
= tgsi_last_instruction(inst
->Dst
[0].Register
.WriteMask
);
4024 for (i
= 0; i
< lasti
+ 1; i
++) {
4026 if (!(inst
->Dst
[0].Register
.WriteMask
& (1 << i
)))
4028 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4029 alu
.op
= ALU_OP1_MOV
;
4031 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], i
);
4033 if (i
== 1 || i
== 3)
4034 r600_bytecode_src_toggle_neg(&alu
.src
[0]);
4035 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
4040 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4048 static int tgsi_dfracexp(struct r600_shader_ctx
*ctx
)
4050 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
4051 struct r600_bytecode_alu alu
;
4052 unsigned write_mask
= inst
->Dst
[0].Register
.WriteMask
;
4054 int firsti
= write_mask
== 0xc ? 2 : 0;
4056 for (i
= 0; i
<= 3; i
++) {
4057 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4058 alu
.op
= ctx
->inst_info
->op
;
4060 alu
.dst
.sel
= ctx
->temp_reg
;
4063 for (j
= 0; j
< inst
->Instruction
.NumSrcRegs
; j
++) {
4064 r600_bytecode_src(&alu
.src
[j
], &ctx
->src
[j
], fp64_switch(i
));
4070 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4075 /* MOV first two channels to writemask dst0 */
4076 for (i
= 0; i
<= 1; i
++) {
4077 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4078 alu
.op
= ALU_OP1_MOV
;
4079 alu
.src
[0].chan
= i
+ 2;
4080 alu
.src
[0].sel
= ctx
->temp_reg
;
4082 tgsi_dst(ctx
, &inst
->Dst
[0], firsti
+ i
, &alu
.dst
);
4083 alu
.dst
.write
= (inst
->Dst
[0].Register
.WriteMask
>> (firsti
+ i
)) & 1;
4085 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4090 for (i
= 0; i
<= 3; i
++) {
4091 if (inst
->Dst
[1].Register
.WriteMask
& (1 << i
)) {
4092 /* MOV third channels to writemask dst1 */
4093 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4094 alu
.op
= ALU_OP1_MOV
;
4095 alu
.src
[0].chan
= 1;
4096 alu
.src
[0].sel
= ctx
->temp_reg
;
4098 tgsi_dst(ctx
, &inst
->Dst
[1], i
, &alu
.dst
);
4100 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4110 static int egcm_int_to_double(struct r600_shader_ctx
*ctx
)
4112 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
4113 struct r600_bytecode_alu alu
;
4115 int lasti
= tgsi_last_instruction(inst
->Dst
[0].Register
.WriteMask
);
4117 assert(inst
->Instruction
.Opcode
== TGSI_OPCODE_I2D
||
4118 inst
->Instruction
.Opcode
== TGSI_OPCODE_U2D
);
4120 for (i
= 0; i
<= (lasti
+1)/2; i
++) {
4121 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4122 alu
.op
= ctx
->inst_info
->op
;
4124 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], i
);
4125 alu
.dst
.sel
= ctx
->temp_reg
;
4130 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4135 for (i
= 0; i
<= lasti
; i
++) {
4136 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4137 alu
.op
= ALU_OP1_FLT32_TO_FLT64
;
4139 alu
.src
[0].chan
= i
/2;
4141 alu
.src
[0].sel
= ctx
->temp_reg
;
4143 alu
.src
[0].sel
= V_SQ_ALU_SRC_LITERAL
;
4144 alu
.src
[0].value
= 0x0;
4146 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
4147 alu
.last
= i
== lasti
;
4149 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4157 static int egcm_double_to_int(struct r600_shader_ctx
*ctx
)
4159 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
4160 struct r600_bytecode_alu alu
;
4162 int lasti
= tgsi_last_instruction(inst
->Dst
[0].Register
.WriteMask
);
4164 assert(inst
->Instruction
.Opcode
== TGSI_OPCODE_D2I
||
4165 inst
->Instruction
.Opcode
== TGSI_OPCODE_D2U
);
4167 for (i
= 0; i
<= lasti
; i
++) {
4168 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4169 alu
.op
= ALU_OP1_FLT64_TO_FLT32
;
4171 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], fp64_switch(i
));
4173 alu
.dst
.sel
= ctx
->temp_reg
;
4174 alu
.dst
.write
= i
%2 == 0;
4175 alu
.last
= i
== lasti
;
4177 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4182 for (i
= 0; i
<= (lasti
+1)/2; i
++) {
4183 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4184 alu
.op
= ctx
->inst_info
->op
;
4186 alu
.src
[0].chan
= i
*2;
4187 alu
.src
[0].sel
= ctx
->temp_reg
;
4188 tgsi_dst(ctx
, &inst
->Dst
[0], 0, &alu
.dst
);
4191 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4199 static int cayman_emit_double_instr(struct r600_shader_ctx
*ctx
)
4201 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
4203 struct r600_bytecode_alu alu
;
4205 int lasti
= tgsi_last_instruction(inst
->Dst
[0].Register
.WriteMask
);
4206 int t1
= ctx
->temp_reg
;
4208 /* these have to write the result to X/Y by the looks of it */
4209 for (i
= 0 ; i
< last_slot
; i
++) {
4210 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4211 alu
.op
= ctx
->inst_info
->op
;
4213 /* should only be one src regs */
4214 assert (inst
->Instruction
.NumSrcRegs
== 1);
4216 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 1);
4217 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[0], 0);
4219 /* RSQ should take the absolute value of src */
4220 if (ctx
->parse
.FullToken
.FullInstruction
.Instruction
.Opcode
== TGSI_OPCODE_DRSQ
||
4221 ctx
->parse
.FullToken
.FullInstruction
.Instruction
.Opcode
== TGSI_OPCODE_DSQRT
) {
4222 r600_bytecode_src_set_abs(&alu
.src
[1]);
4226 alu
.dst
.write
= (i
== 0 || i
== 1);
4228 if (ctx
->bc
->chip_class
!= CAYMAN
|| i
== last_slot
- 1)
4230 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4235 for (i
= 0 ; i
<= lasti
; i
++) {
4236 if (!(inst
->Dst
[0].Register
.WriteMask
& (1 << i
)))
4238 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4239 alu
.op
= ALU_OP1_MOV
;
4240 alu
.src
[0].sel
= t1
;
4241 alu
.src
[0].chan
= (i
== 0 || i
== 2) ? 0 : 1;
4242 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
4246 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4253 static int cayman_emit_float_instr(struct r600_shader_ctx
*ctx
)
4255 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
4257 struct r600_bytecode_alu alu
;
4258 int last_slot
= (inst
->Dst
[0].Register
.WriteMask
& 0x8) ? 4 : 3;
4260 for (i
= 0 ; i
< last_slot
; i
++) {
4261 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4262 alu
.op
= ctx
->inst_info
->op
;
4263 for (j
= 0; j
< inst
->Instruction
.NumSrcRegs
; j
++) {
4264 r600_bytecode_src(&alu
.src
[j
], &ctx
->src
[j
], 0);
4266 /* RSQ should take the absolute value of src */
4267 if (inst
->Instruction
.Opcode
== TGSI_OPCODE_RSQ
) {
4268 r600_bytecode_src_set_abs(&alu
.src
[j
]);
4271 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
4272 alu
.dst
.write
= (inst
->Dst
[0].Register
.WriteMask
>> i
) & 1;
4274 if (i
== last_slot
- 1)
4276 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4283 static int cayman_mul_int_instr(struct r600_shader_ctx
*ctx
)
4285 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
4287 struct r600_bytecode_alu alu
;
4288 int lasti
= tgsi_last_instruction(inst
->Dst
[0].Register
.WriteMask
);
4289 int t1
= ctx
->temp_reg
;
4291 for (k
= 0; k
<= lasti
; k
++) {
4292 if (!(inst
->Dst
[0].Register
.WriteMask
& (1 << k
)))
4295 for (i
= 0 ; i
< 4; i
++) {
4296 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4297 alu
.op
= ctx
->inst_info
->op
;
4298 for (j
= 0; j
< inst
->Instruction
.NumSrcRegs
; j
++) {
4299 r600_bytecode_src(&alu
.src
[j
], &ctx
->src
[j
], k
);
4303 alu
.dst
.write
= (i
== k
);
4306 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4312 for (i
= 0 ; i
<= lasti
; i
++) {
4313 if (!(inst
->Dst
[0].Register
.WriteMask
& (1 << i
)))
4315 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4316 alu
.op
= ALU_OP1_MOV
;
4317 alu
.src
[0].sel
= t1
;
4318 alu
.src
[0].chan
= i
;
4319 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
4323 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4332 static int cayman_mul_double_instr(struct r600_shader_ctx
*ctx
)
4334 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
4336 struct r600_bytecode_alu alu
;
4337 int lasti
= tgsi_last_instruction(inst
->Dst
[0].Register
.WriteMask
);
4338 int t1
= ctx
->temp_reg
;
4340 for (k
= 0; k
< 2; k
++) {
4341 if (!(inst
->Dst
[0].Register
.WriteMask
& (0x3 << (k
* 2))))
4344 for (i
= 0; i
< 4; i
++) {
4345 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4346 alu
.op
= ctx
->inst_info
->op
;
4347 for (j
= 0; j
< inst
->Instruction
.NumSrcRegs
; j
++) {
4348 r600_bytecode_src(&alu
.src
[j
], &ctx
->src
[j
], k
* 2 + ((i
== 3) ? 0 : 1));
4355 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4361 for (i
= 0; i
<= lasti
; i
++) {
4362 if (!(inst
->Dst
[0].Register
.WriteMask
& (1 << i
)))
4364 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4365 alu
.op
= ALU_OP1_MOV
;
4366 alu
.src
[0].sel
= t1
;
4367 alu
.src
[0].chan
= i
;
4368 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
4372 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4381 * r600 - trunc to -PI..PI range
4382 * r700 - normalize by dividing by 2PI
4385 static int tgsi_setup_trig(struct r600_shader_ctx
*ctx
)
4388 struct r600_bytecode_alu alu
;
4390 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4391 alu
.op
= ALU_OP3_MULADD
;
4395 alu
.dst
.sel
= ctx
->temp_reg
;
4398 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 0);
4400 alu
.src
[1].sel
= V_SQ_ALU_SRC_LITERAL
;
4401 alu
.src
[1].chan
= 0;
4402 alu
.src
[1].value
= u_bitcast_f2u(0.5f
* M_1_PI
);
4403 alu
.src
[2].sel
= V_SQ_ALU_SRC_0_5
;
4404 alu
.src
[2].chan
= 0;
4406 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4410 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4411 alu
.op
= ALU_OP1_FRACT
;
4414 alu
.dst
.sel
= ctx
->temp_reg
;
4417 alu
.src
[0].sel
= ctx
->temp_reg
;
4418 alu
.src
[0].chan
= 0;
4420 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4424 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4425 alu
.op
= ALU_OP3_MULADD
;
4429 alu
.dst
.sel
= ctx
->temp_reg
;
4432 alu
.src
[0].sel
= ctx
->temp_reg
;
4433 alu
.src
[0].chan
= 0;
4435 alu
.src
[1].sel
= V_SQ_ALU_SRC_LITERAL
;
4436 alu
.src
[1].chan
= 0;
4437 alu
.src
[2].sel
= V_SQ_ALU_SRC_LITERAL
;
4438 alu
.src
[2].chan
= 0;
4440 if (ctx
->bc
->chip_class
== R600
) {
4441 alu
.src
[1].value
= u_bitcast_f2u(2.0f
* M_PI
);
4442 alu
.src
[2].value
= u_bitcast_f2u(-M_PI
);
4444 alu
.src
[1].sel
= V_SQ_ALU_SRC_1
;
4445 alu
.src
[2].sel
= V_SQ_ALU_SRC_0_5
;
4450 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4456 static int cayman_trig(struct r600_shader_ctx
*ctx
)
4458 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
4459 struct r600_bytecode_alu alu
;
4460 int last_slot
= (inst
->Dst
[0].Register
.WriteMask
& 0x8) ? 4 : 3;
4463 r
= tgsi_setup_trig(ctx
);
4468 for (i
= 0; i
< last_slot
; i
++) {
4469 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4470 alu
.op
= ctx
->inst_info
->op
;
4473 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
4474 alu
.dst
.write
= (inst
->Dst
[0].Register
.WriteMask
>> i
) & 1;
4476 alu
.src
[0].sel
= ctx
->temp_reg
;
4477 alu
.src
[0].chan
= 0;
4478 if (i
== last_slot
- 1)
4480 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4487 static int tgsi_trig(struct r600_shader_ctx
*ctx
)
4489 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
4490 struct r600_bytecode_alu alu
;
4492 int lasti
= tgsi_last_instruction(inst
->Dst
[0].Register
.WriteMask
);
4494 r
= tgsi_setup_trig(ctx
);
4498 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4499 alu
.op
= ctx
->inst_info
->op
;
4501 alu
.dst
.sel
= ctx
->temp_reg
;
4504 alu
.src
[0].sel
= ctx
->temp_reg
;
4505 alu
.src
[0].chan
= 0;
4507 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4511 /* replicate result */
4512 for (i
= 0; i
< lasti
+ 1; i
++) {
4513 if (!(inst
->Dst
[0].Register
.WriteMask
& (1 << i
)))
4516 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4517 alu
.op
= ALU_OP1_MOV
;
4519 alu
.src
[0].sel
= ctx
->temp_reg
;
4520 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
4523 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4530 static int tgsi_scs(struct r600_shader_ctx
*ctx
)
4532 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
4533 struct r600_bytecode_alu alu
;
4536 /* We'll only need the trig stuff if we are going to write to the
4537 * X or Y components of the destination vector.
4539 if (likely(inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_XY
)) {
4540 r
= tgsi_setup_trig(ctx
);
4546 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_X
) {
4547 if (ctx
->bc
->chip_class
== CAYMAN
) {
4548 for (i
= 0 ; i
< 3; i
++) {
4549 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4550 alu
.op
= ALU_OP1_COS
;
4551 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
4557 alu
.src
[0].sel
= ctx
->temp_reg
;
4558 alu
.src
[0].chan
= 0;
4561 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4566 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4567 alu
.op
= ALU_OP1_COS
;
4568 tgsi_dst(ctx
, &inst
->Dst
[0], 0, &alu
.dst
);
4570 alu
.src
[0].sel
= ctx
->temp_reg
;
4571 alu
.src
[0].chan
= 0;
4573 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4580 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Y
) {
4581 if (ctx
->bc
->chip_class
== CAYMAN
) {
4582 for (i
= 0 ; i
< 3; i
++) {
4583 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4584 alu
.op
= ALU_OP1_SIN
;
4585 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
4590 alu
.src
[0].sel
= ctx
->temp_reg
;
4591 alu
.src
[0].chan
= 0;
4594 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4599 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4600 alu
.op
= ALU_OP1_SIN
;
4601 tgsi_dst(ctx
, &inst
->Dst
[0], 1, &alu
.dst
);
4603 alu
.src
[0].sel
= ctx
->temp_reg
;
4604 alu
.src
[0].chan
= 0;
4606 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4613 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Z
) {
4614 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4616 alu
.op
= ALU_OP1_MOV
;
4618 tgsi_dst(ctx
, &inst
->Dst
[0], 2, &alu
.dst
);
4620 alu
.src
[0].sel
= V_SQ_ALU_SRC_0
;
4621 alu
.src
[0].chan
= 0;
4625 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4631 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_W
) {
4632 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4634 alu
.op
= ALU_OP1_MOV
;
4636 tgsi_dst(ctx
, &inst
->Dst
[0], 3, &alu
.dst
);
4638 alu
.src
[0].sel
= V_SQ_ALU_SRC_1
;
4639 alu
.src
[0].chan
= 0;
4643 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4651 static int tgsi_kill(struct r600_shader_ctx
*ctx
)
4653 const struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
4654 struct r600_bytecode_alu alu
;
4657 for (i
= 0; i
< 4; i
++) {
4658 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4659 alu
.op
= ctx
->inst_info
->op
;
4663 alu
.src
[0].sel
= V_SQ_ALU_SRC_0
;
4665 if (inst
->Instruction
.Opcode
== TGSI_OPCODE_KILL
) {
4666 alu
.src
[1].sel
= V_SQ_ALU_SRC_1
;
4669 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[0], i
);
4674 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4679 /* kill must be last in ALU */
4680 ctx
->bc
->force_add_cf
= 1;
4681 ctx
->shader
->uses_kill
= TRUE
;
4685 static int tgsi_lit(struct r600_shader_ctx
*ctx
)
4687 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
4688 struct r600_bytecode_alu alu
;
4691 /* tmp.x = max(src.y, 0.0) */
4692 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4693 alu
.op
= ALU_OP2_MAX
;
4694 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 1);
4695 alu
.src
[1].sel
= V_SQ_ALU_SRC_0
; /*0.0*/
4696 alu
.src
[1].chan
= 1;
4698 alu
.dst
.sel
= ctx
->temp_reg
;
4703 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4707 if (inst
->Dst
[0].Register
.WriteMask
& (1 << 2))
4713 if (ctx
->bc
->chip_class
== CAYMAN
) {
4714 for (i
= 0; i
< 3; i
++) {
4715 /* tmp.z = log(tmp.x) */
4716 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4717 alu
.op
= ALU_OP1_LOG_CLAMPED
;
4718 alu
.src
[0].sel
= ctx
->temp_reg
;
4719 alu
.src
[0].chan
= 0;
4720 alu
.dst
.sel
= ctx
->temp_reg
;
4728 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4733 /* tmp.z = log(tmp.x) */
4734 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4735 alu
.op
= ALU_OP1_LOG_CLAMPED
;
4736 alu
.src
[0].sel
= ctx
->temp_reg
;
4737 alu
.src
[0].chan
= 0;
4738 alu
.dst
.sel
= ctx
->temp_reg
;
4742 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4747 chan
= alu
.dst
.chan
;
4750 /* tmp.x = amd MUL_LIT(tmp.z, src.w, src.x ) */
4751 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4752 alu
.op
= ALU_OP3_MUL_LIT
;
4753 alu
.src
[0].sel
= sel
;
4754 alu
.src
[0].chan
= chan
;
4755 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[0], 3);
4756 r600_bytecode_src(&alu
.src
[2], &ctx
->src
[0], 0);
4757 alu
.dst
.sel
= ctx
->temp_reg
;
4762 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4766 if (ctx
->bc
->chip_class
== CAYMAN
) {
4767 for (i
= 0; i
< 3; i
++) {
4768 /* dst.z = exp(tmp.x) */
4769 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4770 alu
.op
= ALU_OP1_EXP_IEEE
;
4771 alu
.src
[0].sel
= ctx
->temp_reg
;
4772 alu
.src
[0].chan
= 0;
4773 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
4779 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4784 /* dst.z = exp(tmp.x) */
4785 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4786 alu
.op
= ALU_OP1_EXP_IEEE
;
4787 alu
.src
[0].sel
= ctx
->temp_reg
;
4788 alu
.src
[0].chan
= 0;
4789 tgsi_dst(ctx
, &inst
->Dst
[0], 2, &alu
.dst
);
4791 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4798 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4799 alu
.op
= ALU_OP1_MOV
;
4800 alu
.src
[0].sel
= V_SQ_ALU_SRC_1
; /*1.0*/
4801 alu
.src
[0].chan
= 0;
4802 tgsi_dst(ctx
, &inst
->Dst
[0], 0, &alu
.dst
);
4803 alu
.dst
.write
= (inst
->Dst
[0].Register
.WriteMask
>> 0) & 1;
4804 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4808 /* dst.y = max(src.x, 0.0) */
4809 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4810 alu
.op
= ALU_OP2_MAX
;
4811 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 0);
4812 alu
.src
[1].sel
= V_SQ_ALU_SRC_0
; /*0.0*/
4813 alu
.src
[1].chan
= 0;
4814 tgsi_dst(ctx
, &inst
->Dst
[0], 1, &alu
.dst
);
4815 alu
.dst
.write
= (inst
->Dst
[0].Register
.WriteMask
>> 1) & 1;
4816 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4821 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4822 alu
.op
= ALU_OP1_MOV
;
4823 alu
.src
[0].sel
= V_SQ_ALU_SRC_1
;
4824 alu
.src
[0].chan
= 0;
4825 tgsi_dst(ctx
, &inst
->Dst
[0], 3, &alu
.dst
);
4826 alu
.dst
.write
= (inst
->Dst
[0].Register
.WriteMask
>> 3) & 1;
4828 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4835 static int tgsi_rsq(struct r600_shader_ctx
*ctx
)
4837 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
4838 struct r600_bytecode_alu alu
;
4841 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4844 * For state trackers other than OpenGL, we'll want to use
4845 * _RECIPSQRT_IEEE instead.
4847 alu
.op
= ALU_OP1_RECIPSQRT_CLAMPED
;
4849 for (i
= 0; i
< inst
->Instruction
.NumSrcRegs
; i
++) {
4850 r600_bytecode_src(&alu
.src
[i
], &ctx
->src
[i
], 0);
4851 r600_bytecode_src_set_abs(&alu
.src
[i
]);
4853 alu
.dst
.sel
= ctx
->temp_reg
;
4856 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4859 /* replicate result */
4860 return tgsi_helper_tempx_replicate(ctx
);
4863 static int tgsi_helper_tempx_replicate(struct r600_shader_ctx
*ctx
)
4865 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
4866 struct r600_bytecode_alu alu
;
4869 for (i
= 0; i
< 4; i
++) {
4870 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4871 alu
.src
[0].sel
= ctx
->temp_reg
;
4872 alu
.op
= ALU_OP1_MOV
;
4874 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
4875 alu
.dst
.write
= (inst
->Dst
[0].Register
.WriteMask
>> i
) & 1;
4878 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4885 static int tgsi_trans_srcx_replicate(struct r600_shader_ctx
*ctx
)
4887 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
4888 struct r600_bytecode_alu alu
;
4891 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4892 alu
.op
= ctx
->inst_info
->op
;
4893 for (i
= 0; i
< inst
->Instruction
.NumSrcRegs
; i
++) {
4894 r600_bytecode_src(&alu
.src
[i
], &ctx
->src
[i
], 0);
4896 alu
.dst
.sel
= ctx
->temp_reg
;
4899 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4902 /* replicate result */
4903 return tgsi_helper_tempx_replicate(ctx
);
4906 static int cayman_pow(struct r600_shader_ctx
*ctx
)
4908 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
4910 struct r600_bytecode_alu alu
;
4911 int last_slot
= (inst
->Dst
[0].Register
.WriteMask
& 0x8) ? 4 : 3;
4913 for (i
= 0; i
< 3; i
++) {
4914 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4915 alu
.op
= ALU_OP1_LOG_IEEE
;
4916 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 0);
4917 alu
.dst
.sel
= ctx
->temp_reg
;
4922 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4928 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4929 alu
.op
= ALU_OP2_MUL
;
4930 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[1], 0);
4931 alu
.src
[1].sel
= ctx
->temp_reg
;
4932 alu
.dst
.sel
= ctx
->temp_reg
;
4935 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4939 for (i
= 0; i
< last_slot
; i
++) {
4940 /* POW(a,b) = EXP2(b * LOG2(a))*/
4941 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4942 alu
.op
= ALU_OP1_EXP_IEEE
;
4943 alu
.src
[0].sel
= ctx
->temp_reg
;
4945 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
4946 alu
.dst
.write
= (inst
->Dst
[0].Register
.WriteMask
>> i
) & 1;
4947 if (i
== last_slot
- 1)
4949 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4956 static int tgsi_pow(struct r600_shader_ctx
*ctx
)
4958 struct r600_bytecode_alu alu
;
4962 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4963 alu
.op
= ALU_OP1_LOG_IEEE
;
4964 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 0);
4965 alu
.dst
.sel
= ctx
->temp_reg
;
4968 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4972 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4973 alu
.op
= ALU_OP2_MUL
;
4974 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[1], 0);
4975 alu
.src
[1].sel
= ctx
->temp_reg
;
4976 alu
.dst
.sel
= ctx
->temp_reg
;
4979 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4982 /* POW(a,b) = EXP2(b * LOG2(a))*/
4983 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
4984 alu
.op
= ALU_OP1_EXP_IEEE
;
4985 alu
.src
[0].sel
= ctx
->temp_reg
;
4986 alu
.dst
.sel
= ctx
->temp_reg
;
4989 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
4992 return tgsi_helper_tempx_replicate(ctx
);
4995 static int tgsi_divmod(struct r600_shader_ctx
*ctx
, int mod
, int signed_op
)
4997 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
4998 struct r600_bytecode_alu alu
;
5000 unsigned write_mask
= inst
->Dst
[0].Register
.WriteMask
;
5001 int tmp0
= ctx
->temp_reg
;
5002 int tmp1
= r600_get_temp(ctx
);
5003 int tmp2
= r600_get_temp(ctx
);
5004 int tmp3
= r600_get_temp(ctx
);
5007 * we need to represent src1 as src2*q + r, where q - quotient, r - remainder
5009 * 1. tmp0.x = rcp (src2) = 2^32/src2 + e, where e is rounding error
5010 * 2. tmp0.z = lo (tmp0.x * src2)
5011 * 3. tmp0.w = -tmp0.z
5012 * 4. tmp0.y = hi (tmp0.x * src2)
5013 * 5. tmp0.z = (tmp0.y == 0 ? tmp0.w : tmp0.z) = abs(lo(rcp*src2))
5014 * 6. tmp0.w = hi (tmp0.z * tmp0.x) = e, rounding error
5015 * 7. tmp1.x = tmp0.x - tmp0.w
5016 * 8. tmp1.y = tmp0.x + tmp0.w
5017 * 9. tmp0.x = (tmp0.y == 0 ? tmp1.y : tmp1.x)
5018 * 10. tmp0.z = hi(tmp0.x * src1) = q
5019 * 11. tmp0.y = lo (tmp0.z * src2) = src2*q = src1 - r
5021 * 12. tmp0.w = src1 - tmp0.y = r
5022 * 13. tmp1.x = tmp0.w >= src2 = r >= src2 (uint comparison)
5023 * 14. tmp1.y = src1 >= tmp0.y = r >= 0 (uint comparison)
5027 * 15. tmp1.z = tmp0.z + 1 = q + 1
5028 * 16. tmp1.w = tmp0.z - 1 = q - 1
5032 * 15. tmp1.z = tmp0.w - src2 = r - src2
5033 * 16. tmp1.w = tmp0.w + src2 = r + src2
5037 * 17. tmp1.x = tmp1.x & tmp1.y
5039 * DIV: 18. tmp0.z = tmp1.x==0 ? tmp0.z : tmp1.z
5040 * MOD: 18. tmp0.z = tmp1.x==0 ? tmp0.w : tmp1.z
5042 * 19. tmp0.z = tmp1.y==0 ? tmp1.w : tmp0.z
5043 * 20. dst = src2==0 ? MAX_UINT : tmp0.z
5047 * Same as unsigned, using abs values of the operands,
5048 * and fixing the sign of the result in the end.
5051 for (i
= 0; i
< 4; i
++) {
5052 if (!(write_mask
& (1<<i
)))
5057 /* tmp2.x = -src0 */
5058 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
5059 alu
.op
= ALU_OP2_SUB_INT
;
5065 alu
.src
[0].sel
= V_SQ_ALU_SRC_0
;
5067 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[0], i
);
5070 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
5073 /* tmp2.y = -src1 */
5074 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
5075 alu
.op
= ALU_OP2_SUB_INT
;
5081 alu
.src
[0].sel
= V_SQ_ALU_SRC_0
;
5083 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[1], i
);
5086 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
5089 /* tmp2.z sign bit is set if src0 and src2 signs are different */
5090 /* it will be a sign of the quotient */
5093 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
5094 alu
.op
= ALU_OP2_XOR_INT
;
5100 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], i
);
5101 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[1], i
);
5104 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
5108 /* tmp2.x = |src0| */
5109 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
5110 alu
.op
= ALU_OP3_CNDGE_INT
;
5117 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], i
);
5118 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[0], i
);
5119 alu
.src
[2].sel
= tmp2
;
5120 alu
.src
[2].chan
= 0;
5123 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
5126 /* tmp2.y = |src1| */
5127 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
5128 alu
.op
= ALU_OP3_CNDGE_INT
;
5135 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[1], i
);
5136 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[1], i
);
5137 alu
.src
[2].sel
= tmp2
;
5138 alu
.src
[2].chan
= 1;
5141 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
5146 /* 1. tmp0.x = rcp_u (src2) = 2^32/src2 + e, where e is rounding error */
5147 if (ctx
->bc
->chip_class
== CAYMAN
) {
5148 /* tmp3.x = u2f(src2) */
5149 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
5150 alu
.op
= ALU_OP1_UINT_TO_FLT
;
5157 alu
.src
[0].sel
= tmp2
;
5158 alu
.src
[0].chan
= 1;
5160 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[1], i
);
5164 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
5167 /* tmp0.x = recip(tmp3.x) */
5168 for (j
= 0 ; j
< 3; j
++) {
5169 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
5170 alu
.op
= ALU_OP1_RECIP_IEEE
;
5174 alu
.dst
.write
= (j
== 0);
5176 alu
.src
[0].sel
= tmp3
;
5177 alu
.src
[0].chan
= 0;
5181 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
5185 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
5186 alu
.op
= ALU_OP2_MUL
;
5188 alu
.src
[0].sel
= tmp0
;
5189 alu
.src
[0].chan
= 0;
5191 alu
.src
[1].sel
= V_SQ_ALU_SRC_LITERAL
;
5192 alu
.src
[1].value
= 0x4f800000;
5197 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
5201 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
5202 alu
.op
= ALU_OP1_FLT_TO_UINT
;
5208 alu
.src
[0].sel
= tmp3
;
5209 alu
.src
[0].chan
= 0;
5212 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
5216 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
5217 alu
.op
= ALU_OP1_RECIP_UINT
;
5224 alu
.src
[0].sel
= tmp2
;
5225 alu
.src
[0].chan
= 1;
5227 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[1], i
);
5231 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
5235 /* 2. tmp0.z = lo (tmp0.x * src2) */
5236 if (ctx
->bc
->chip_class
== CAYMAN
) {
5237 for (j
= 0 ; j
< 4; j
++) {
5238 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
5239 alu
.op
= ALU_OP2_MULLO_UINT
;
5243 alu
.dst
.write
= (j
== 2);
5245 alu
.src
[0].sel
= tmp0
;
5246 alu
.src
[0].chan
= 0;
5248 alu
.src
[1].sel
= tmp2
;
5249 alu
.src
[1].chan
= 1;
5251 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[1], i
);
5254 alu
.last
= (j
== 3);
5255 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
5259 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
5260 alu
.op
= ALU_OP2_MULLO_UINT
;
5266 alu
.src
[0].sel
= tmp0
;
5267 alu
.src
[0].chan
= 0;
5269 alu
.src
[1].sel
= tmp2
;
5270 alu
.src
[1].chan
= 1;
5272 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[1], i
);
5276 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
5280 /* 3. tmp0.w = -tmp0.z */
5281 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
5282 alu
.op
= ALU_OP2_SUB_INT
;
5288 alu
.src
[0].sel
= V_SQ_ALU_SRC_0
;
5289 alu
.src
[1].sel
= tmp0
;
5290 alu
.src
[1].chan
= 2;
5293 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
5296 /* 4. tmp0.y = hi (tmp0.x * src2) */
5297 if (ctx
->bc
->chip_class
== CAYMAN
) {
5298 for (j
= 0 ; j
< 4; j
++) {
5299 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
5300 alu
.op
= ALU_OP2_MULHI_UINT
;
5304 alu
.dst
.write
= (j
== 1);
5306 alu
.src
[0].sel
= tmp0
;
5307 alu
.src
[0].chan
= 0;
5310 alu
.src
[1].sel
= tmp2
;
5311 alu
.src
[1].chan
= 1;
5313 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[1], i
);
5315 alu
.last
= (j
== 3);
5316 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
5320 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
5321 alu
.op
= ALU_OP2_MULHI_UINT
;
5327 alu
.src
[0].sel
= tmp0
;
5328 alu
.src
[0].chan
= 0;
5331 alu
.src
[1].sel
= tmp2
;
5332 alu
.src
[1].chan
= 1;
5334 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[1], i
);
5338 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
5342 /* 5. tmp0.z = (tmp0.y == 0 ? tmp0.w : tmp0.z) = abs(lo(rcp*src)) */
5343 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
5344 alu
.op
= ALU_OP3_CNDE_INT
;
5351 alu
.src
[0].sel
= tmp0
;
5352 alu
.src
[0].chan
= 1;
5353 alu
.src
[1].sel
= tmp0
;
5354 alu
.src
[1].chan
= 3;
5355 alu
.src
[2].sel
= tmp0
;
5356 alu
.src
[2].chan
= 2;
5359 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
5362 /* 6. tmp0.w = hi (tmp0.z * tmp0.x) = e, rounding error */
5363 if (ctx
->bc
->chip_class
== CAYMAN
) {
5364 for (j
= 0 ; j
< 4; j
++) {
5365 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
5366 alu
.op
= ALU_OP2_MULHI_UINT
;
5370 alu
.dst
.write
= (j
== 3);
5372 alu
.src
[0].sel
= tmp0
;
5373 alu
.src
[0].chan
= 2;
5375 alu
.src
[1].sel
= tmp0
;
5376 alu
.src
[1].chan
= 0;
5378 alu
.last
= (j
== 3);
5379 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
5383 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
5384 alu
.op
= ALU_OP2_MULHI_UINT
;
5390 alu
.src
[0].sel
= tmp0
;
5391 alu
.src
[0].chan
= 2;
5393 alu
.src
[1].sel
= tmp0
;
5394 alu
.src
[1].chan
= 0;
5397 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
5401 /* 7. tmp1.x = tmp0.x - tmp0.w */
5402 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
5403 alu
.op
= ALU_OP2_SUB_INT
;
5409 alu
.src
[0].sel
= tmp0
;
5410 alu
.src
[0].chan
= 0;
5411 alu
.src
[1].sel
= tmp0
;
5412 alu
.src
[1].chan
= 3;
5415 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
5418 /* 8. tmp1.y = tmp0.x + tmp0.w */
5419 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
5420 alu
.op
= ALU_OP2_ADD_INT
;
5426 alu
.src
[0].sel
= tmp0
;
5427 alu
.src
[0].chan
= 0;
5428 alu
.src
[1].sel
= tmp0
;
5429 alu
.src
[1].chan
= 3;
5432 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
5435 /* 9. tmp0.x = (tmp0.y == 0 ? tmp1.y : tmp1.x) */
5436 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
5437 alu
.op
= ALU_OP3_CNDE_INT
;
5444 alu
.src
[0].sel
= tmp0
;
5445 alu
.src
[0].chan
= 1;
5446 alu
.src
[1].sel
= tmp1
;
5447 alu
.src
[1].chan
= 1;
5448 alu
.src
[2].sel
= tmp1
;
5449 alu
.src
[2].chan
= 0;
5452 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
5455 /* 10. tmp0.z = hi(tmp0.x * src1) = q */
5456 if (ctx
->bc
->chip_class
== CAYMAN
) {
5457 for (j
= 0 ; j
< 4; j
++) {
5458 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
5459 alu
.op
= ALU_OP2_MULHI_UINT
;
5463 alu
.dst
.write
= (j
== 2);
5465 alu
.src
[0].sel
= tmp0
;
5466 alu
.src
[0].chan
= 0;
5469 alu
.src
[1].sel
= tmp2
;
5470 alu
.src
[1].chan
= 0;
5472 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[0], i
);
5475 alu
.last
= (j
== 3);
5476 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
5480 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
5481 alu
.op
= ALU_OP2_MULHI_UINT
;
5487 alu
.src
[0].sel
= tmp0
;
5488 alu
.src
[0].chan
= 0;
5491 alu
.src
[1].sel
= tmp2
;
5492 alu
.src
[1].chan
= 0;
5494 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[0], i
);
5498 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
5502 /* 11. tmp0.y = lo (src2 * tmp0.z) = src2*q = src1 - r */
5503 if (ctx
->bc
->chip_class
== CAYMAN
) {
5504 for (j
= 0 ; j
< 4; j
++) {
5505 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
5506 alu
.op
= ALU_OP2_MULLO_UINT
;
5510 alu
.dst
.write
= (j
== 1);
5513 alu
.src
[0].sel
= tmp2
;
5514 alu
.src
[0].chan
= 1;
5516 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[1], i
);
5519 alu
.src
[1].sel
= tmp0
;
5520 alu
.src
[1].chan
= 2;
5522 alu
.last
= (j
== 3);
5523 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
5527 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
5528 alu
.op
= ALU_OP2_MULLO_UINT
;
5535 alu
.src
[0].sel
= tmp2
;
5536 alu
.src
[0].chan
= 1;
5538 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[1], i
);
5541 alu
.src
[1].sel
= tmp0
;
5542 alu
.src
[1].chan
= 2;
5545 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
5549 /* 12. tmp0.w = src1 - tmp0.y = r */
5550 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
5551 alu
.op
= ALU_OP2_SUB_INT
;
5558 alu
.src
[0].sel
= tmp2
;
5559 alu
.src
[0].chan
= 0;
5561 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], i
);
5564 alu
.src
[1].sel
= tmp0
;
5565 alu
.src
[1].chan
= 1;
5568 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
5571 /* 13. tmp1.x = tmp0.w >= src2 = r >= src2 */
5572 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
5573 alu
.op
= ALU_OP2_SETGE_UINT
;
5579 alu
.src
[0].sel
= tmp0
;
5580 alu
.src
[0].chan
= 3;
5582 alu
.src
[1].sel
= tmp2
;
5583 alu
.src
[1].chan
= 1;
5585 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[1], i
);
5589 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
5592 /* 14. tmp1.y = src1 >= tmp0.y = r >= 0 */
5593 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
5594 alu
.op
= ALU_OP2_SETGE_UINT
;
5601 alu
.src
[0].sel
= tmp2
;
5602 alu
.src
[0].chan
= 0;
5604 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], i
);
5607 alu
.src
[1].sel
= tmp0
;
5608 alu
.src
[1].chan
= 1;
5611 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
5614 if (mod
) { /* UMOD */
5616 /* 15. tmp1.z = tmp0.w - src2 = r - src2 */
5617 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
5618 alu
.op
= ALU_OP2_SUB_INT
;
5624 alu
.src
[0].sel
= tmp0
;
5625 alu
.src
[0].chan
= 3;
5628 alu
.src
[1].sel
= tmp2
;
5629 alu
.src
[1].chan
= 1;
5631 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[1], i
);
5635 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
5638 /* 16. tmp1.w = tmp0.w + src2 = r + src2 */
5639 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
5640 alu
.op
= ALU_OP2_ADD_INT
;
5646 alu
.src
[0].sel
= tmp0
;
5647 alu
.src
[0].chan
= 3;
5649 alu
.src
[1].sel
= tmp2
;
5650 alu
.src
[1].chan
= 1;
5652 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[1], i
);
5656 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
5661 /* 15. tmp1.z = tmp0.z + 1 = q + 1 DIV */
5662 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
5663 alu
.op
= ALU_OP2_ADD_INT
;
5669 alu
.src
[0].sel
= tmp0
;
5670 alu
.src
[0].chan
= 2;
5671 alu
.src
[1].sel
= V_SQ_ALU_SRC_1_INT
;
5674 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
5677 /* 16. tmp1.w = tmp0.z - 1 = q - 1 */
5678 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
5679 alu
.op
= ALU_OP2_ADD_INT
;
5685 alu
.src
[0].sel
= tmp0
;
5686 alu
.src
[0].chan
= 2;
5687 alu
.src
[1].sel
= V_SQ_ALU_SRC_M_1_INT
;
5690 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
5695 /* 17. tmp1.x = tmp1.x & tmp1.y */
5696 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
5697 alu
.op
= ALU_OP2_AND_INT
;
5703 alu
.src
[0].sel
= tmp1
;
5704 alu
.src
[0].chan
= 0;
5705 alu
.src
[1].sel
= tmp1
;
5706 alu
.src
[1].chan
= 1;
5709 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
5712 /* 18. tmp0.z = tmp1.x==0 ? tmp0.z : tmp1.z DIV */
5713 /* 18. tmp0.z = tmp1.x==0 ? tmp0.w : tmp1.z MOD */
5714 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
5715 alu
.op
= ALU_OP3_CNDE_INT
;
5722 alu
.src
[0].sel
= tmp1
;
5723 alu
.src
[0].chan
= 0;
5724 alu
.src
[1].sel
= tmp0
;
5725 alu
.src
[1].chan
= mod
? 3 : 2;
5726 alu
.src
[2].sel
= tmp1
;
5727 alu
.src
[2].chan
= 2;
5730 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
5733 /* 19. tmp0.z = tmp1.y==0 ? tmp1.w : tmp0.z */
5734 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
5735 alu
.op
= ALU_OP3_CNDE_INT
;
5743 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
5746 alu
.src
[0].sel
= tmp1
;
5747 alu
.src
[0].chan
= 1;
5748 alu
.src
[1].sel
= tmp1
;
5749 alu
.src
[1].chan
= 3;
5750 alu
.src
[2].sel
= tmp0
;
5751 alu
.src
[2].chan
= 2;
5754 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
5759 /* fix the sign of the result */
5763 /* tmp0.x = -tmp0.z */
5764 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
5765 alu
.op
= ALU_OP2_SUB_INT
;
5771 alu
.src
[0].sel
= V_SQ_ALU_SRC_0
;
5772 alu
.src
[1].sel
= tmp0
;
5773 alu
.src
[1].chan
= 2;
5776 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
5779 /* sign of the remainder is the same as the sign of src0 */
5780 /* tmp0.x = src0>=0 ? tmp0.z : tmp0.x */
5781 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
5782 alu
.op
= ALU_OP3_CNDGE_INT
;
5785 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
5787 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], i
);
5788 alu
.src
[1].sel
= tmp0
;
5789 alu
.src
[1].chan
= 2;
5790 alu
.src
[2].sel
= tmp0
;
5791 alu
.src
[2].chan
= 0;
5794 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
5799 /* tmp0.x = -tmp0.z */
5800 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
5801 alu
.op
= ALU_OP2_SUB_INT
;
5807 alu
.src
[0].sel
= V_SQ_ALU_SRC_0
;
5808 alu
.src
[1].sel
= tmp0
;
5809 alu
.src
[1].chan
= 2;
5812 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
5815 /* fix the quotient sign (same as the sign of src0*src1) */
5816 /* tmp0.x = tmp2.z>=0 ? tmp0.z : tmp0.x */
5817 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
5818 alu
.op
= ALU_OP3_CNDGE_INT
;
5821 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
5823 alu
.src
[0].sel
= tmp2
;
5824 alu
.src
[0].chan
= 2;
5825 alu
.src
[1].sel
= tmp0
;
5826 alu
.src
[1].chan
= 2;
5827 alu
.src
[2].sel
= tmp0
;
5828 alu
.src
[2].chan
= 0;
5831 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
5839 static int tgsi_udiv(struct r600_shader_ctx
*ctx
)
5841 return tgsi_divmod(ctx
, 0, 0);
5844 static int tgsi_umod(struct r600_shader_ctx
*ctx
)
5846 return tgsi_divmod(ctx
, 1, 0);
5849 static int tgsi_idiv(struct r600_shader_ctx
*ctx
)
5851 return tgsi_divmod(ctx
, 0, 1);
5854 static int tgsi_imod(struct r600_shader_ctx
*ctx
)
5856 return tgsi_divmod(ctx
, 1, 1);
5860 static int tgsi_f2i(struct r600_shader_ctx
*ctx
)
5862 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
5863 struct r600_bytecode_alu alu
;
5865 unsigned write_mask
= inst
->Dst
[0].Register
.WriteMask
;
5866 int last_inst
= tgsi_last_instruction(write_mask
);
5868 for (i
= 0; i
< 4; i
++) {
5869 if (!(write_mask
& (1<<i
)))
5872 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
5873 alu
.op
= ALU_OP1_TRUNC
;
5875 alu
.dst
.sel
= ctx
->temp_reg
;
5879 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], i
);
5882 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
5887 for (i
= 0; i
< 4; i
++) {
5888 if (!(write_mask
& (1<<i
)))
5891 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
5892 alu
.op
= ctx
->inst_info
->op
;
5894 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
5896 alu
.src
[0].sel
= ctx
->temp_reg
;
5897 alu
.src
[0].chan
= i
;
5899 if (i
== last_inst
|| alu
.op
== ALU_OP1_FLT_TO_UINT
)
5901 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
5909 static int tgsi_iabs(struct r600_shader_ctx
*ctx
)
5911 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
5912 struct r600_bytecode_alu alu
;
5914 unsigned write_mask
= inst
->Dst
[0].Register
.WriteMask
;
5915 int last_inst
= tgsi_last_instruction(write_mask
);
5918 for (i
= 0; i
< 4; i
++) {
5919 if (!(write_mask
& (1<<i
)))
5922 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
5923 alu
.op
= ALU_OP2_SUB_INT
;
5925 alu
.dst
.sel
= ctx
->temp_reg
;
5929 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[0], i
);
5930 alu
.src
[0].sel
= V_SQ_ALU_SRC_0
;
5934 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
5939 /* dst = (src >= 0 ? src : tmp) */
5940 for (i
= 0; i
< 4; i
++) {
5941 if (!(write_mask
& (1<<i
)))
5944 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
5945 alu
.op
= ALU_OP3_CNDGE_INT
;
5949 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
5951 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], i
);
5952 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[0], i
);
5953 alu
.src
[2].sel
= ctx
->temp_reg
;
5954 alu
.src
[2].chan
= i
;
5958 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
5965 static int tgsi_issg(struct r600_shader_ctx
*ctx
)
5967 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
5968 struct r600_bytecode_alu alu
;
5970 unsigned write_mask
= inst
->Dst
[0].Register
.WriteMask
;
5971 int last_inst
= tgsi_last_instruction(write_mask
);
5973 /* tmp = (src >= 0 ? src : -1) */
5974 for (i
= 0; i
< 4; i
++) {
5975 if (!(write_mask
& (1<<i
)))
5978 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
5979 alu
.op
= ALU_OP3_CNDGE_INT
;
5982 alu
.dst
.sel
= ctx
->temp_reg
;
5986 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], i
);
5987 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[0], i
);
5988 alu
.src
[2].sel
= V_SQ_ALU_SRC_M_1_INT
;
5992 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
5997 /* dst = (tmp > 0 ? 1 : tmp) */
5998 for (i
= 0; i
< 4; i
++) {
5999 if (!(write_mask
& (1<<i
)))
6002 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
6003 alu
.op
= ALU_OP3_CNDGT_INT
;
6007 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
6009 alu
.src
[0].sel
= ctx
->temp_reg
;
6010 alu
.src
[0].chan
= i
;
6012 alu
.src
[1].sel
= V_SQ_ALU_SRC_1_INT
;
6014 alu
.src
[2].sel
= ctx
->temp_reg
;
6015 alu
.src
[2].chan
= i
;
6019 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
6028 static int tgsi_ssg(struct r600_shader_ctx
*ctx
)
6030 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
6031 struct r600_bytecode_alu alu
;
6034 /* tmp = (src > 0 ? 1 : src) */
6035 for (i
= 0; i
< 4; i
++) {
6036 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
6037 alu
.op
= ALU_OP3_CNDGT
;
6040 alu
.dst
.sel
= ctx
->temp_reg
;
6043 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], i
);
6044 alu
.src
[1].sel
= V_SQ_ALU_SRC_1
;
6045 r600_bytecode_src(&alu
.src
[2], &ctx
->src
[0], i
);
6049 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
6054 /* dst = (-tmp > 0 ? -1 : tmp) */
6055 for (i
= 0; i
< 4; i
++) {
6056 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
6057 alu
.op
= ALU_OP3_CNDGT
;
6059 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
6061 alu
.src
[0].sel
= ctx
->temp_reg
;
6062 alu
.src
[0].chan
= i
;
6065 alu
.src
[1].sel
= V_SQ_ALU_SRC_1
;
6068 alu
.src
[2].sel
= ctx
->temp_reg
;
6069 alu
.src
[2].chan
= i
;
6073 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
6080 static int tgsi_bfi(struct r600_shader_ctx
*ctx
)
6082 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
6083 struct r600_bytecode_alu alu
;
6086 unsigned write_mask
= inst
->Dst
[0].Register
.WriteMask
;
6087 int last_inst
= tgsi_last_instruction(write_mask
);
6091 for (i
= 0; i
< 4; i
++) {
6092 if (!(write_mask
& (1<<i
)))
6095 /* create mask tmp */
6096 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
6097 alu
.op
= ALU_OP2_BFM_INT
;
6101 alu
.last
= i
== last_inst
;
6103 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[3], i
);
6104 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[2], i
);
6106 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
6111 t2
= r600_get_temp(ctx
);
6113 for (i
= 0; i
< 4; i
++) {
6114 if (!(write_mask
& (1<<i
)))
6117 /* shift insert left */
6118 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
6119 alu
.op
= ALU_OP2_LSHL_INT
;
6123 alu
.last
= i
== last_inst
;
6125 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[1], i
);
6126 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[2], i
);
6128 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
6133 for (i
= 0; i
< 4; i
++) {
6134 if (!(write_mask
& (1<<i
)))
6137 /* actual bitfield insert */
6138 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
6139 alu
.op
= ALU_OP3_BFI_INT
;
6141 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
6144 alu
.last
= i
== last_inst
;
6146 alu
.src
[0].sel
= t1
;
6147 alu
.src
[0].chan
= i
;
6148 alu
.src
[1].sel
= t2
;
6149 alu
.src
[1].chan
= i
;
6150 r600_bytecode_src(&alu
.src
[2], &ctx
->src
[0], i
);
6152 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
6160 static int tgsi_msb(struct r600_shader_ctx
*ctx
)
6162 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
6163 struct r600_bytecode_alu alu
;
6166 unsigned write_mask
= inst
->Dst
[0].Register
.WriteMask
;
6167 int last_inst
= tgsi_last_instruction(write_mask
);
6169 assert(ctx
->inst_info
->op
== ALU_OP1_FFBH_INT
||
6170 ctx
->inst_info
->op
== ALU_OP1_FFBH_UINT
);
6174 /* bit position is indexed from lsb by TGSI, and from msb by the hardware */
6175 for (i
= 0; i
< 4; i
++) {
6176 if (!(write_mask
& (1<<i
)))
6179 /* t1 = FFBH_INT / FFBH_UINT */
6180 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
6181 alu
.op
= ctx
->inst_info
->op
;
6185 alu
.last
= i
== last_inst
;
6187 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], i
);
6189 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
6194 t2
= r600_get_temp(ctx
);
6196 for (i
= 0; i
< 4; i
++) {
6197 if (!(write_mask
& (1<<i
)))
6201 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
6202 alu
.op
= ALU_OP2_SUB_INT
;
6206 alu
.last
= i
== last_inst
;
6208 alu
.src
[0].sel
= V_SQ_ALU_SRC_LITERAL
;
6209 alu
.src
[0].value
= 31;
6210 alu
.src
[1].sel
= t1
;
6211 alu
.src
[1].chan
= i
;
6213 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
6218 for (i
= 0; i
< 4; i
++) {
6219 if (!(write_mask
& (1<<i
)))
6222 /* result = t1 >= 0 ? t2 : t1 */
6223 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
6224 alu
.op
= ALU_OP3_CNDGE_INT
;
6226 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
6229 alu
.last
= i
== last_inst
;
6231 alu
.src
[0].sel
= t1
;
6232 alu
.src
[0].chan
= i
;
6233 alu
.src
[1].sel
= t2
;
6234 alu
.src
[1].chan
= i
;
6235 alu
.src
[2].sel
= t1
;
6236 alu
.src
[2].chan
= i
;
6238 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
6246 static int tgsi_interp_egcm(struct r600_shader_ctx
*ctx
)
6248 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
6249 struct r600_bytecode_alu alu
;
6250 int r
, i
= 0, k
, interp_gpr
, interp_base_chan
, tmp
, lasti
;
6254 assert(inst
->Src
[0].Register
.File
== TGSI_FILE_INPUT
);
6256 input
= inst
->Src
[0].Register
.Index
;
6258 /* Interpolators have been marked for use already by allocate_system_value_inputs */
6259 if (inst
->Instruction
.Opcode
== TGSI_OPCODE_INTERP_OFFSET
||
6260 inst
->Instruction
.Opcode
== TGSI_OPCODE_INTERP_SAMPLE
) {
6261 location
= TGSI_INTERPOLATE_LOC_CENTER
; /* sample offset will be added explicitly */
6264 location
= TGSI_INTERPOLATE_LOC_CENTROID
;
6267 k
= eg_get_interpolator_index(ctx
->shader
->input
[input
].interpolate
, location
);
6270 interp_gpr
= ctx
->eg_interpolators
[k
].ij_index
/ 2;
6271 interp_base_chan
= 2 * (ctx
->eg_interpolators
[k
].ij_index
% 2);
6273 /* NOTE: currently offset is not perspective correct */
6274 if (inst
->Instruction
.Opcode
== TGSI_OPCODE_INTERP_OFFSET
||
6275 inst
->Instruction
.Opcode
== TGSI_OPCODE_INTERP_SAMPLE
) {
6276 int sample_gpr
= -1;
6277 int gradientsH
, gradientsV
;
6278 struct r600_bytecode_tex tex
;
6280 if (inst
->Instruction
.Opcode
== TGSI_OPCODE_INTERP_SAMPLE
) {
6281 sample_gpr
= load_sample_position(ctx
, &ctx
->src
[1], ctx
->src
[1].swizzle
[0]);
6284 gradientsH
= r600_get_temp(ctx
);
6285 gradientsV
= r600_get_temp(ctx
);
6286 for (i
= 0; i
< 2; i
++) {
6287 memset(&tex
, 0, sizeof(struct r600_bytecode_tex
));
6288 tex
.op
= i
== 0 ? FETCH_OP_GET_GRADIENTS_H
: FETCH_OP_GET_GRADIENTS_V
;
6289 tex
.src_gpr
= interp_gpr
;
6290 tex
.src_sel_x
= interp_base_chan
+ 0;
6291 tex
.src_sel_y
= interp_base_chan
+ 1;
6294 tex
.dst_gpr
= i
== 0 ? gradientsH
: gradientsV
;
6299 tex
.inst_mod
= 1; // Use per pixel gradient calculation
6301 tex
.resource_id
= tex
.sampler_id
;
6302 r
= r600_bytecode_add_tex(ctx
->bc
, &tex
);
6307 for (i
= 0; i
< 2; i
++) {
6308 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
6309 alu
.op
= ALU_OP3_MULADD
;
6311 alu
.src
[0].sel
= gradientsH
;
6312 alu
.src
[0].chan
= i
;
6313 if (inst
->Instruction
.Opcode
== TGSI_OPCODE_INTERP_SAMPLE
) {
6314 alu
.src
[1].sel
= sample_gpr
;
6315 alu
.src
[1].chan
= 2;
6318 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[1], 0);
6320 alu
.src
[2].sel
= interp_gpr
;
6321 alu
.src
[2].chan
= interp_base_chan
+ i
;
6322 alu
.dst
.sel
= ctx
->temp_reg
;
6326 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
6331 for (i
= 0; i
< 2; i
++) {
6332 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
6333 alu
.op
= ALU_OP3_MULADD
;
6335 alu
.src
[0].sel
= gradientsV
;
6336 alu
.src
[0].chan
= i
;
6337 if (inst
->Instruction
.Opcode
== TGSI_OPCODE_INTERP_SAMPLE
) {
6338 alu
.src
[1].sel
= sample_gpr
;
6339 alu
.src
[1].chan
= 3;
6342 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[1], 1);
6344 alu
.src
[2].sel
= ctx
->temp_reg
;
6345 alu
.src
[2].chan
= i
;
6346 alu
.dst
.sel
= ctx
->temp_reg
;
6350 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
6356 tmp
= r600_get_temp(ctx
);
6357 for (i
= 0; i
< 8; i
++) {
6358 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
6359 alu
.op
= i
< 4 ? ALU_OP2_INTERP_ZW
: ALU_OP2_INTERP_XY
;
6362 if ((i
> 1 && i
< 6)) {
6368 alu
.dst
.chan
= i
% 4;
6370 if (inst
->Instruction
.Opcode
== TGSI_OPCODE_INTERP_OFFSET
||
6371 inst
->Instruction
.Opcode
== TGSI_OPCODE_INTERP_SAMPLE
) {
6372 alu
.src
[0].sel
= ctx
->temp_reg
;
6373 alu
.src
[0].chan
= 1 - (i
% 2);
6375 alu
.src
[0].sel
= interp_gpr
;
6376 alu
.src
[0].chan
= interp_base_chan
+ 1 - (i
% 2);
6378 alu
.src
[1].sel
= V_SQ_ALU_SRC_PARAM_BASE
+ ctx
->shader
->input
[input
].lds_pos
;
6379 alu
.src
[1].chan
= 0;
6381 alu
.last
= i
% 4 == 3;
6382 alu
.bank_swizzle_force
= SQ_ALU_VEC_210
;
6384 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
6389 // INTERP can't swizzle dst
6390 lasti
= tgsi_last_instruction(inst
->Dst
[0].Register
.WriteMask
);
6391 for (i
= 0; i
<= lasti
; i
++) {
6392 if (!(inst
->Dst
[0].Register
.WriteMask
& (1 << i
)))
6395 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
6396 alu
.op
= ALU_OP1_MOV
;
6397 alu
.src
[0].sel
= tmp
;
6398 alu
.src
[0].chan
= ctx
->src
[0].swizzle
[i
];
6399 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
6401 alu
.last
= i
== lasti
;
6402 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
6411 static int tgsi_helper_copy(struct r600_shader_ctx
*ctx
, struct tgsi_full_instruction
*inst
)
6413 struct r600_bytecode_alu alu
;
6416 for (i
= 0; i
< 4; i
++) {
6417 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
6418 if (!(inst
->Dst
[0].Register
.WriteMask
& (1 << i
))) {
6419 alu
.op
= ALU_OP0_NOP
;
6422 alu
.op
= ALU_OP1_MOV
;
6423 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
6424 alu
.src
[0].sel
= ctx
->temp_reg
;
6425 alu
.src
[0].chan
= i
;
6430 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
6437 static int tgsi_make_src_for_op3(struct r600_shader_ctx
*ctx
,
6438 unsigned temp
, int chan
,
6439 struct r600_bytecode_alu_src
*bc_src
,
6440 const struct r600_shader_src
*shader_src
)
6442 struct r600_bytecode_alu alu
;
6445 r600_bytecode_src(bc_src
, shader_src
, chan
);
6447 /* op3 operands don't support abs modifier */
6449 assert(temp
!=0); /* we actually need the extra register, make sure it is allocated. */
6450 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
6451 alu
.op
= ALU_OP1_MOV
;
6453 alu
.dst
.chan
= chan
;
6456 alu
.src
[0] = *bc_src
;
6457 alu
.last
= true; // sufficient?
6458 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
6462 memset(bc_src
, 0, sizeof(*bc_src
));
6464 bc_src
->chan
= chan
;
6469 static int tgsi_op3(struct r600_shader_ctx
*ctx
)
6471 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
6472 struct r600_bytecode_alu alu
;
6474 int lasti
= tgsi_last_instruction(inst
->Dst
[0].Register
.WriteMask
);
6477 for (j
= 0; j
< inst
->Instruction
.NumSrcRegs
; j
++) {
6479 if (ctx
->src
[j
].abs
)
6480 temp_regs
[j
] = r600_get_temp(ctx
);
6482 for (i
= 0; i
< lasti
+ 1; i
++) {
6483 if (!(inst
->Dst
[0].Register
.WriteMask
& (1 << i
)))
6486 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
6487 alu
.op
= ctx
->inst_info
->op
;
6488 for (j
= 0; j
< inst
->Instruction
.NumSrcRegs
; j
++) {
6489 r
= tgsi_make_src_for_op3(ctx
, temp_regs
[j
], i
, &alu
.src
[j
], &ctx
->src
[j
]);
6494 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
6501 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
6508 static int tgsi_dp(struct r600_shader_ctx
*ctx
)
6510 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
6511 struct r600_bytecode_alu alu
;
6514 for (i
= 0; i
< 4; i
++) {
6515 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
6516 alu
.op
= ctx
->inst_info
->op
;
6517 for (j
= 0; j
< inst
->Instruction
.NumSrcRegs
; j
++) {
6518 r600_bytecode_src(&alu
.src
[j
], &ctx
->src
[j
], i
);
6521 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
6523 alu
.dst
.write
= (inst
->Dst
[0].Register
.WriteMask
>> i
) & 1;
6524 /* handle some special cases */
6525 switch (inst
->Instruction
.Opcode
) {
6526 case TGSI_OPCODE_DP2
:
6528 alu
.src
[0].sel
= alu
.src
[1].sel
= V_SQ_ALU_SRC_0
;
6529 alu
.src
[0].chan
= alu
.src
[1].chan
= 0;
6532 case TGSI_OPCODE_DP3
:
6534 alu
.src
[0].sel
= alu
.src
[1].sel
= V_SQ_ALU_SRC_0
;
6535 alu
.src
[0].chan
= alu
.src
[1].chan
= 0;
6538 case TGSI_OPCODE_DPH
:
6540 alu
.src
[0].sel
= V_SQ_ALU_SRC_1
;
6541 alu
.src
[0].chan
= 0;
6551 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
6558 static inline boolean
tgsi_tex_src_requires_loading(struct r600_shader_ctx
*ctx
,
6561 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
6562 return (inst
->Src
[index
].Register
.File
!= TGSI_FILE_TEMPORARY
&&
6563 inst
->Src
[index
].Register
.File
!= TGSI_FILE_INPUT
&&
6564 inst
->Src
[index
].Register
.File
!= TGSI_FILE_OUTPUT
) ||
6565 ctx
->src
[index
].neg
|| ctx
->src
[index
].abs
||
6566 (inst
->Src
[index
].Register
.File
== TGSI_FILE_INPUT
&& ctx
->type
== PIPE_SHADER_GEOMETRY
);
6569 static inline unsigned tgsi_tex_get_src_gpr(struct r600_shader_ctx
*ctx
,
6572 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
6573 return ctx
->file_offset
[inst
->Src
[index
].Register
.File
] + inst
->Src
[index
].Register
.Index
;
6576 static int do_vtx_fetch_inst(struct r600_shader_ctx
*ctx
, boolean src_requires_loading
)
6578 struct r600_bytecode_vtx vtx
;
6579 struct r600_bytecode_alu alu
;
6580 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
6582 int id
= tgsi_tex_get_src_gpr(ctx
, 1);
6584 src_gpr
= tgsi_tex_get_src_gpr(ctx
, 0);
6585 if (src_requires_loading
) {
6586 for (i
= 0; i
< 4; i
++) {
6587 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
6588 alu
.op
= ALU_OP1_MOV
;
6589 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], i
);
6590 alu
.dst
.sel
= ctx
->temp_reg
;
6595 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
6599 src_gpr
= ctx
->temp_reg
;
6602 memset(&vtx
, 0, sizeof(vtx
));
6603 vtx
.op
= FETCH_OP_VFETCH
;
6604 vtx
.buffer_id
= id
+ R600_MAX_CONST_BUFFERS
;
6605 vtx
.fetch_type
= SQ_VTX_FETCH_NO_INDEX_OFFSET
;
6606 vtx
.src_gpr
= src_gpr
;
6607 vtx
.mega_fetch_count
= 16;
6608 vtx
.dst_gpr
= ctx
->file_offset
[inst
->Dst
[0].Register
.File
] + inst
->Dst
[0].Register
.Index
;
6609 vtx
.dst_sel_x
= (inst
->Dst
[0].Register
.WriteMask
& 1) ? 0 : 7; /* SEL_X */
6610 vtx
.dst_sel_y
= (inst
->Dst
[0].Register
.WriteMask
& 2) ? 1 : 7; /* SEL_Y */
6611 vtx
.dst_sel_z
= (inst
->Dst
[0].Register
.WriteMask
& 4) ? 2 : 7; /* SEL_Z */
6612 vtx
.dst_sel_w
= (inst
->Dst
[0].Register
.WriteMask
& 8) ? 3 : 7; /* SEL_W */
6613 vtx
.use_const_fields
= 1;
6615 if ((r
= r600_bytecode_add_vtx(ctx
->bc
, &vtx
)))
6618 if (ctx
->bc
->chip_class
>= EVERGREEN
)
6621 for (i
= 0; i
< 4; i
++) {
6622 int lasti
= tgsi_last_instruction(inst
->Dst
[0].Register
.WriteMask
);
6623 if (!(inst
->Dst
[0].Register
.WriteMask
& (1 << i
)))
6626 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
6627 alu
.op
= ALU_OP2_AND_INT
;
6630 alu
.dst
.sel
= vtx
.dst_gpr
;
6633 alu
.src
[0].sel
= vtx
.dst_gpr
;
6634 alu
.src
[0].chan
= i
;
6636 alu
.src
[1].sel
= R600_SHADER_BUFFER_INFO_SEL
;
6637 alu
.src
[1].sel
+= (id
* 2);
6638 alu
.src
[1].chan
= i
% 4;
6639 alu
.src
[1].kc_bank
= R600_BUFFER_INFO_CONST_BUFFER
;
6643 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
6648 if (inst
->Dst
[0].Register
.WriteMask
& 3) {
6649 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
6650 alu
.op
= ALU_OP2_OR_INT
;
6653 alu
.dst
.sel
= vtx
.dst_gpr
;
6656 alu
.src
[0].sel
= vtx
.dst_gpr
;
6657 alu
.src
[0].chan
= 3;
6659 alu
.src
[1].sel
= R600_SHADER_BUFFER_INFO_SEL
+ (id
* 2) + 1;
6660 alu
.src
[1].chan
= 0;
6661 alu
.src
[1].kc_bank
= R600_BUFFER_INFO_CONST_BUFFER
;
6664 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
6671 static int r600_do_buffer_txq(struct r600_shader_ctx
*ctx
)
6673 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
6674 struct r600_bytecode_alu alu
;
6676 int id
= tgsi_tex_get_src_gpr(ctx
, 1);
6678 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
6679 alu
.op
= ALU_OP1_MOV
;
6680 alu
.src
[0].sel
= R600_SHADER_BUFFER_INFO_SEL
;
6681 if (ctx
->bc
->chip_class
>= EVERGREEN
) {
6682 /* channel 0 or 2 of each word */
6683 alu
.src
[0].sel
+= (id
/ 2);
6684 alu
.src
[0].chan
= (id
% 2) * 2;
6686 /* r600 we have them at channel 2 of the second dword */
6687 alu
.src
[0].sel
+= (id
* 2) + 1;
6688 alu
.src
[0].chan
= 1;
6690 alu
.src
[0].kc_bank
= R600_BUFFER_INFO_CONST_BUFFER
;
6691 tgsi_dst(ctx
, &inst
->Dst
[0], 0, &alu
.dst
);
6693 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
6699 static int tgsi_tex(struct r600_shader_ctx
*ctx
)
6701 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
6702 struct r600_bytecode_tex tex
;
6703 struct r600_bytecode_alu alu
;
6707 bool read_compressed_msaa
= ctx
->bc
->has_compressed_msaa_texturing
&&
6708 inst
->Instruction
.Opcode
== TGSI_OPCODE_TXF
&&
6709 (inst
->Texture
.Texture
== TGSI_TEXTURE_2D_MSAA
||
6710 inst
->Texture
.Texture
== TGSI_TEXTURE_2D_ARRAY_MSAA
);
6712 bool txf_add_offsets
= inst
->Texture
.NumOffsets
&&
6713 inst
->Instruction
.Opcode
== TGSI_OPCODE_TXF
&&
6714 inst
->Texture
.Texture
!= TGSI_TEXTURE_BUFFER
;
6716 /* Texture fetch instructions can only use gprs as source.
6717 * Also they cannot negate the source or take the absolute value */
6718 const boolean src_requires_loading
= (inst
->Instruction
.Opcode
!= TGSI_OPCODE_TXQ_LZ
&&
6719 inst
->Instruction
.Opcode
!= TGSI_OPCODE_TXQS
&&
6720 tgsi_tex_src_requires_loading(ctx
, 0)) ||
6721 read_compressed_msaa
|| txf_add_offsets
;
6723 boolean src_loaded
= FALSE
;
6724 unsigned sampler_src_reg
= inst
->Instruction
.Opcode
== TGSI_OPCODE_TXQ_LZ
? 0 : 1;
6725 int8_t offset_x
= 0, offset_y
= 0, offset_z
= 0;
6726 boolean has_txq_cube_array_z
= false;
6727 unsigned sampler_index_mode
;
6729 if (inst
->Instruction
.Opcode
== TGSI_OPCODE_TXQ
&&
6730 ((inst
->Texture
.Texture
== TGSI_TEXTURE_CUBE_ARRAY
||
6731 inst
->Texture
.Texture
== TGSI_TEXTURE_SHADOWCUBE_ARRAY
)))
6732 if (inst
->Dst
[0].Register
.WriteMask
& 4) {
6733 ctx
->shader
->has_txq_cube_array_z_comp
= true;
6734 has_txq_cube_array_z
= true;
6737 if (inst
->Instruction
.Opcode
== TGSI_OPCODE_TEX2
||
6738 inst
->Instruction
.Opcode
== TGSI_OPCODE_TXB2
||
6739 inst
->Instruction
.Opcode
== TGSI_OPCODE_TXL2
||
6740 inst
->Instruction
.Opcode
== TGSI_OPCODE_TG4
)
6741 sampler_src_reg
= 2;
6743 /* TGSI moves the sampler to src reg 3 for TXD */
6744 if (inst
->Instruction
.Opcode
== TGSI_OPCODE_TXD
)
6745 sampler_src_reg
= 3;
6747 sampler_index_mode
= inst
->Src
[sampler_src_reg
].Indirect
.Index
== 2 ? 2 : 0; // CF_INDEX_1 : CF_INDEX_NONE
6749 src_gpr
= tgsi_tex_get_src_gpr(ctx
, 0);
6751 if (inst
->Texture
.Texture
== TGSI_TEXTURE_BUFFER
) {
6752 if (inst
->Instruction
.Opcode
== TGSI_OPCODE_TXQ
) {
6753 ctx
->shader
->uses_tex_buffers
= true;
6754 return r600_do_buffer_txq(ctx
);
6756 else if (inst
->Instruction
.Opcode
== TGSI_OPCODE_TXF
) {
6757 if (ctx
->bc
->chip_class
< EVERGREEN
)
6758 ctx
->shader
->uses_tex_buffers
= true;
6759 return do_vtx_fetch_inst(ctx
, src_requires_loading
);
6763 if (inst
->Instruction
.Opcode
== TGSI_OPCODE_TXP
) {
6765 /* Add perspective divide */
6766 if (ctx
->bc
->chip_class
== CAYMAN
) {
6768 for (i
= 0; i
< 3; i
++) {
6769 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
6770 alu
.op
= ALU_OP1_RECIP_IEEE
;
6771 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 3);
6773 alu
.dst
.sel
= ctx
->temp_reg
;
6779 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
6786 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
6787 alu
.op
= ALU_OP1_RECIP_IEEE
;
6788 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 3);
6790 alu
.dst
.sel
= ctx
->temp_reg
;
6791 alu
.dst
.chan
= out_chan
;
6794 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
6799 for (i
= 0; i
< 3; i
++) {
6800 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
6801 alu
.op
= ALU_OP2_MUL
;
6802 alu
.src
[0].sel
= ctx
->temp_reg
;
6803 alu
.src
[0].chan
= out_chan
;
6804 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[0], i
);
6805 alu
.dst
.sel
= ctx
->temp_reg
;
6808 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
6812 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
6813 alu
.op
= ALU_OP1_MOV
;
6814 alu
.src
[0].sel
= V_SQ_ALU_SRC_1
;
6815 alu
.src
[0].chan
= 0;
6816 alu
.dst
.sel
= ctx
->temp_reg
;
6820 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
6824 src_gpr
= ctx
->temp_reg
;
6828 if ((inst
->Texture
.Texture
== TGSI_TEXTURE_CUBE
||
6829 inst
->Texture
.Texture
== TGSI_TEXTURE_CUBE_ARRAY
||
6830 inst
->Texture
.Texture
== TGSI_TEXTURE_SHADOWCUBE
||
6831 inst
->Texture
.Texture
== TGSI_TEXTURE_SHADOWCUBE_ARRAY
) &&
6832 inst
->Instruction
.Opcode
!= TGSI_OPCODE_TXQ
&&
6833 inst
->Instruction
.Opcode
!= TGSI_OPCODE_TXQ_LZ
) {
6835 static const unsigned src0_swizzle
[] = {2, 2, 0, 1};
6836 static const unsigned src1_swizzle
[] = {1, 0, 2, 2};
6838 /* tmp1.xyzw = CUBE(R0.zzxy, R0.yxzz) */
6839 for (i
= 0; i
< 4; i
++) {
6840 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
6841 alu
.op
= ALU_OP2_CUBE
;
6842 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], src0_swizzle
[i
]);
6843 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[0], src1_swizzle
[i
]);
6844 alu
.dst
.sel
= ctx
->temp_reg
;
6849 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
6854 /* tmp1.z = RCP_e(|tmp1.z|) */
6855 if (ctx
->bc
->chip_class
== CAYMAN
) {
6856 for (i
= 0; i
< 3; i
++) {
6857 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
6858 alu
.op
= ALU_OP1_RECIP_IEEE
;
6859 alu
.src
[0].sel
= ctx
->temp_reg
;
6860 alu
.src
[0].chan
= 2;
6862 alu
.dst
.sel
= ctx
->temp_reg
;
6868 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
6873 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
6874 alu
.op
= ALU_OP1_RECIP_IEEE
;
6875 alu
.src
[0].sel
= ctx
->temp_reg
;
6876 alu
.src
[0].chan
= 2;
6878 alu
.dst
.sel
= ctx
->temp_reg
;
6882 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
6887 /* MULADD R0.x, R0.x, PS1, (0x3FC00000, 1.5f).x
6888 * MULADD R0.y, R0.y, PS1, (0x3FC00000, 1.5f).x
6889 * muladd has no writemask, have to use another temp
6891 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
6892 alu
.op
= ALU_OP3_MULADD
;
6895 alu
.src
[0].sel
= ctx
->temp_reg
;
6896 alu
.src
[0].chan
= 0;
6897 alu
.src
[1].sel
= ctx
->temp_reg
;
6898 alu
.src
[1].chan
= 2;
6900 alu
.src
[2].sel
= V_SQ_ALU_SRC_LITERAL
;
6901 alu
.src
[2].chan
= 0;
6902 alu
.src
[2].value
= u_bitcast_f2u(1.5f
);
6904 alu
.dst
.sel
= ctx
->temp_reg
;
6908 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
6912 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
6913 alu
.op
= ALU_OP3_MULADD
;
6916 alu
.src
[0].sel
= ctx
->temp_reg
;
6917 alu
.src
[0].chan
= 1;
6918 alu
.src
[1].sel
= ctx
->temp_reg
;
6919 alu
.src
[1].chan
= 2;
6921 alu
.src
[2].sel
= V_SQ_ALU_SRC_LITERAL
;
6922 alu
.src
[2].chan
= 0;
6923 alu
.src
[2].value
= u_bitcast_f2u(1.5f
);
6925 alu
.dst
.sel
= ctx
->temp_reg
;
6930 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
6933 /* write initial compare value into Z component
6934 - W src 0 for shadow cube
6935 - X src 1 for shadow cube array */
6936 if (inst
->Texture
.Texture
== TGSI_TEXTURE_SHADOWCUBE
||
6937 inst
->Texture
.Texture
== TGSI_TEXTURE_SHADOWCUBE_ARRAY
) {
6938 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
6939 alu
.op
= ALU_OP1_MOV
;
6940 if (inst
->Texture
.Texture
== TGSI_TEXTURE_SHADOWCUBE_ARRAY
)
6941 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[1], 0);
6943 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 3);
6944 alu
.dst
.sel
= ctx
->temp_reg
;
6948 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
6953 if (inst
->Texture
.Texture
== TGSI_TEXTURE_CUBE_ARRAY
||
6954 inst
->Texture
.Texture
== TGSI_TEXTURE_SHADOWCUBE_ARRAY
) {
6955 if (ctx
->bc
->chip_class
>= EVERGREEN
) {
6956 int mytmp
= r600_get_temp(ctx
);
6957 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
6958 alu
.op
= ALU_OP1_MOV
;
6959 alu
.src
[0].sel
= ctx
->temp_reg
;
6960 alu
.src
[0].chan
= 3;
6961 alu
.dst
.sel
= mytmp
;
6965 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
6969 /* have to multiply original layer by 8 and add to face id (temp.w) in Z */
6970 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
6971 alu
.op
= ALU_OP3_MULADD
;
6973 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 3);
6974 alu
.src
[1].sel
= V_SQ_ALU_SRC_LITERAL
;
6975 alu
.src
[1].chan
= 0;
6976 alu
.src
[1].value
= u_bitcast_f2u(8.0f
);
6977 alu
.src
[2].sel
= mytmp
;
6978 alu
.src
[2].chan
= 0;
6979 alu
.dst
.sel
= ctx
->temp_reg
;
6983 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
6986 } else if (ctx
->bc
->chip_class
< EVERGREEN
) {
6987 memset(&tex
, 0, sizeof(struct r600_bytecode_tex
));
6988 tex
.op
= FETCH_OP_SET_CUBEMAP_INDEX
;
6989 tex
.sampler_id
= tgsi_tex_get_src_gpr(ctx
, sampler_src_reg
);
6990 tex
.resource_id
= tex
.sampler_id
+ R600_MAX_CONST_BUFFERS
;
6991 tex
.src_gpr
= r600_get_temp(ctx
);
6996 tex
.dst_sel_x
= tex
.dst_sel_y
= tex
.dst_sel_z
= tex
.dst_sel_w
= 7;
6997 tex
.coord_type_x
= 1;
6998 tex
.coord_type_y
= 1;
6999 tex
.coord_type_z
= 1;
7000 tex
.coord_type_w
= 1;
7001 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
7002 alu
.op
= ALU_OP1_MOV
;
7003 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 3);
7004 alu
.dst
.sel
= tex
.src_gpr
;
7008 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
7012 r
= r600_bytecode_add_tex(ctx
->bc
, &tex
);
7019 /* for cube forms of lod and bias we need to route things */
7020 if (inst
->Instruction
.Opcode
== TGSI_OPCODE_TXB
||
7021 inst
->Instruction
.Opcode
== TGSI_OPCODE_TXL
||
7022 inst
->Instruction
.Opcode
== TGSI_OPCODE_TXB2
||
7023 inst
->Instruction
.Opcode
== TGSI_OPCODE_TXL2
) {
7024 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
7025 alu
.op
= ALU_OP1_MOV
;
7026 if (inst
->Instruction
.Opcode
== TGSI_OPCODE_TXB2
||
7027 inst
->Instruction
.Opcode
== TGSI_OPCODE_TXL2
)
7028 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[1], 0);
7030 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 3);
7031 alu
.dst
.sel
= ctx
->temp_reg
;
7035 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
7041 src_gpr
= ctx
->temp_reg
;
7044 if (inst
->Instruction
.Opcode
== TGSI_OPCODE_TXD
) {
7045 int temp_h
= 0, temp_v
= 0;
7048 /* if we've already loaded the src (i.e. CUBE don't reload it). */
7049 if (src_loaded
== TRUE
)
7053 for (i
= start_val
; i
< 3; i
++) {
7054 int treg
= r600_get_temp(ctx
);
7063 for (j
= 0; j
< 4; j
++) {
7064 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
7065 alu
.op
= ALU_OP1_MOV
;
7066 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[i
], j
);
7072 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
7077 for (i
= 1; i
< 3; i
++) {
7078 /* set gradients h/v */
7079 memset(&tex
, 0, sizeof(struct r600_bytecode_tex
));
7080 tex
.op
= (i
== 1) ? FETCH_OP_SET_GRADIENTS_H
:
7081 FETCH_OP_SET_GRADIENTS_V
;
7082 tex
.sampler_id
= tgsi_tex_get_src_gpr(ctx
, sampler_src_reg
);
7083 tex
.sampler_index_mode
= sampler_index_mode
;
7084 tex
.resource_id
= tex
.sampler_id
+ R600_MAX_CONST_BUFFERS
;
7085 tex
.resource_index_mode
= sampler_index_mode
;
7087 tex
.src_gpr
= (i
== 1) ? temp_h
: temp_v
;
7093 tex
.dst_gpr
= r600_get_temp(ctx
); /* just to avoid confusing the asm scheduler */
7094 tex
.dst_sel_x
= tex
.dst_sel_y
= tex
.dst_sel_z
= tex
.dst_sel_w
= 7;
7095 if (inst
->Texture
.Texture
!= TGSI_TEXTURE_RECT
) {
7096 tex
.coord_type_x
= 1;
7097 tex
.coord_type_y
= 1;
7098 tex
.coord_type_z
= 1;
7099 tex
.coord_type_w
= 1;
7101 r
= r600_bytecode_add_tex(ctx
->bc
, &tex
);
7107 if (src_requires_loading
&& !src_loaded
) {
7108 for (i
= 0; i
< 4; i
++) {
7109 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
7110 alu
.op
= ALU_OP1_MOV
;
7111 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], i
);
7112 alu
.dst
.sel
= ctx
->temp_reg
;
7117 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
7122 src_gpr
= ctx
->temp_reg
;
7125 /* get offset values */
7126 if (inst
->Texture
.NumOffsets
) {
7127 assert(inst
->Texture
.NumOffsets
== 1);
7129 /* The texture offset feature doesn't work with the TXF instruction
7130 * and must be emulated by adding the offset to the texture coordinates. */
7131 if (txf_add_offsets
) {
7132 const struct tgsi_texture_offset
*off
= inst
->TexOffsets
;
7134 switch (inst
->Texture
.Texture
) {
7135 case TGSI_TEXTURE_3D
:
7136 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
7137 alu
.op
= ALU_OP2_ADD_INT
;
7138 alu
.src
[0].sel
= src_gpr
;
7139 alu
.src
[0].chan
= 2;
7140 alu
.src
[1].sel
= V_SQ_ALU_SRC_LITERAL
;
7141 alu
.src
[1].value
= ctx
->literals
[4 * off
[0].Index
+ off
[0].SwizzleZ
];
7142 alu
.dst
.sel
= src_gpr
;
7146 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
7151 case TGSI_TEXTURE_2D
:
7152 case TGSI_TEXTURE_SHADOW2D
:
7153 case TGSI_TEXTURE_RECT
:
7154 case TGSI_TEXTURE_SHADOWRECT
:
7155 case TGSI_TEXTURE_2D_ARRAY
:
7156 case TGSI_TEXTURE_SHADOW2D_ARRAY
:
7157 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
7158 alu
.op
= ALU_OP2_ADD_INT
;
7159 alu
.src
[0].sel
= src_gpr
;
7160 alu
.src
[0].chan
= 1;
7161 alu
.src
[1].sel
= V_SQ_ALU_SRC_LITERAL
;
7162 alu
.src
[1].value
= ctx
->literals
[4 * off
[0].Index
+ off
[0].SwizzleY
];
7163 alu
.dst
.sel
= src_gpr
;
7167 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
7172 case TGSI_TEXTURE_1D
:
7173 case TGSI_TEXTURE_SHADOW1D
:
7174 case TGSI_TEXTURE_1D_ARRAY
:
7175 case TGSI_TEXTURE_SHADOW1D_ARRAY
:
7176 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
7177 alu
.op
= ALU_OP2_ADD_INT
;
7178 alu
.src
[0].sel
= src_gpr
;
7179 alu
.src
[1].sel
= V_SQ_ALU_SRC_LITERAL
;
7180 alu
.src
[1].value
= ctx
->literals
[4 * off
[0].Index
+ off
[0].SwizzleX
];
7181 alu
.dst
.sel
= src_gpr
;
7184 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
7188 /* texture offsets do not apply to other texture targets */
7191 switch (inst
->Texture
.Texture
) {
7192 case TGSI_TEXTURE_3D
:
7193 offset_z
= ctx
->literals
[4 * inst
->TexOffsets
[0].Index
+ inst
->TexOffsets
[0].SwizzleZ
] << 1;
7195 case TGSI_TEXTURE_2D
:
7196 case TGSI_TEXTURE_SHADOW2D
:
7197 case TGSI_TEXTURE_RECT
:
7198 case TGSI_TEXTURE_SHADOWRECT
:
7199 case TGSI_TEXTURE_2D_ARRAY
:
7200 case TGSI_TEXTURE_SHADOW2D_ARRAY
:
7201 offset_y
= ctx
->literals
[4 * inst
->TexOffsets
[0].Index
+ inst
->TexOffsets
[0].SwizzleY
] << 1;
7203 case TGSI_TEXTURE_1D
:
7204 case TGSI_TEXTURE_SHADOW1D
:
7205 case TGSI_TEXTURE_1D_ARRAY
:
7206 case TGSI_TEXTURE_SHADOW1D_ARRAY
:
7207 offset_x
= ctx
->literals
[4 * inst
->TexOffsets
[0].Index
+ inst
->TexOffsets
[0].SwizzleX
] << 1;
7212 /* Obtain the sample index for reading a compressed MSAA color texture.
7213 * To read the FMASK, we use the ldfptr instruction, which tells us
7214 * where the samples are stored.
7215 * For uncompressed 8x MSAA surfaces, ldfptr should return 0x76543210,
7216 * which is the identity mapping. Each nibble says which physical sample
7217 * should be fetched to get that sample.
7219 * Assume src.z contains the sample index. It should be modified like this:
7220 * src.z = (ldfptr() >> (src.z * 4)) & 0xF;
7221 * Then fetch the texel with src.
7223 if (read_compressed_msaa
) {
7224 unsigned sample_chan
= 3;
7225 unsigned temp
= r600_get_temp(ctx
);
7228 /* temp.w = ldfptr() */
7229 memset(&tex
, 0, sizeof(struct r600_bytecode_tex
));
7230 tex
.op
= FETCH_OP_LD
;
7231 tex
.inst_mod
= 1; /* to indicate this is ldfptr */
7232 tex
.sampler_id
= tgsi_tex_get_src_gpr(ctx
, sampler_src_reg
);
7233 tex
.sampler_index_mode
= sampler_index_mode
;
7234 tex
.resource_id
= tex
.sampler_id
+ R600_MAX_CONST_BUFFERS
;
7235 tex
.resource_index_mode
= sampler_index_mode
;
7236 tex
.src_gpr
= src_gpr
;
7238 tex
.dst_sel_x
= 7; /* mask out these components */
7241 tex
.dst_sel_w
= 0; /* store X */
7246 tex
.offset_x
= offset_x
;
7247 tex
.offset_y
= offset_y
;
7248 tex
.offset_z
= offset_z
;
7249 r
= r600_bytecode_add_tex(ctx
->bc
, &tex
);
7253 /* temp.x = sample_index*4 */
7254 if (ctx
->bc
->chip_class
== CAYMAN
) {
7255 for (i
= 0 ; i
< 4; i
++) {
7256 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
7257 alu
.op
= ALU_OP2_MULLO_INT
;
7258 alu
.src
[0].sel
= src_gpr
;
7259 alu
.src
[0].chan
= sample_chan
;
7260 alu
.src
[1].sel
= V_SQ_ALU_SRC_LITERAL
;
7261 alu
.src
[1].value
= 4;
7264 alu
.dst
.write
= i
== 0;
7267 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
7272 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
7273 alu
.op
= ALU_OP2_MULLO_INT
;
7274 alu
.src
[0].sel
= src_gpr
;
7275 alu
.src
[0].chan
= sample_chan
;
7276 alu
.src
[1].sel
= V_SQ_ALU_SRC_LITERAL
;
7277 alu
.src
[1].value
= 4;
7282 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
7287 /* sample_index = temp.w >> temp.x */
7288 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
7289 alu
.op
= ALU_OP2_LSHR_INT
;
7290 alu
.src
[0].sel
= temp
;
7291 alu
.src
[0].chan
= 3;
7292 alu
.src
[1].sel
= temp
;
7293 alu
.src
[1].chan
= 0;
7294 alu
.dst
.sel
= src_gpr
;
7295 alu
.dst
.chan
= sample_chan
;
7298 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
7302 /* sample_index & 0xF */
7303 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
7304 alu
.op
= ALU_OP2_AND_INT
;
7305 alu
.src
[0].sel
= src_gpr
;
7306 alu
.src
[0].chan
= sample_chan
;
7307 alu
.src
[1].sel
= V_SQ_ALU_SRC_LITERAL
;
7308 alu
.src
[1].value
= 0xF;
7309 alu
.dst
.sel
= src_gpr
;
7310 alu
.dst
.chan
= sample_chan
;
7313 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
7317 /* visualize the FMASK */
7318 for (i
= 0; i
< 4; i
++) {
7319 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
7320 alu
.op
= ALU_OP1_INT_TO_FLT
;
7321 alu
.src
[0].sel
= src_gpr
;
7322 alu
.src
[0].chan
= sample_chan
;
7323 alu
.dst
.sel
= ctx
->file_offset
[inst
->Dst
[0].Register
.File
] + inst
->Dst
[0].Register
.Index
;
7327 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
7335 /* does this shader want a num layers from TXQ for a cube array? */
7336 if (has_txq_cube_array_z
) {
7337 int id
= tgsi_tex_get_src_gpr(ctx
, sampler_src_reg
);
7339 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
7340 alu
.op
= ALU_OP1_MOV
;
7342 alu
.src
[0].sel
= R600_SHADER_BUFFER_INFO_SEL
;
7343 if (ctx
->bc
->chip_class
>= EVERGREEN
) {
7344 /* channel 1 or 3 of each word */
7345 alu
.src
[0].sel
+= (id
/ 2);
7346 alu
.src
[0].chan
= ((id
% 2) * 2) + 1;
7348 /* r600 we have them at channel 2 of the second dword */
7349 alu
.src
[0].sel
+= (id
* 2) + 1;
7350 alu
.src
[0].chan
= 2;
7352 alu
.src
[0].kc_bank
= R600_BUFFER_INFO_CONST_BUFFER
;
7353 tgsi_dst(ctx
, &inst
->Dst
[0], 2, &alu
.dst
);
7355 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
7358 /* disable writemask from texture instruction */
7359 inst
->Dst
[0].Register
.WriteMask
&= ~4;
7362 opcode
= ctx
->inst_info
->op
;
7363 if (opcode
== FETCH_OP_GATHER4
&&
7364 inst
->TexOffsets
[0].File
!= TGSI_FILE_NULL
&&
7365 inst
->TexOffsets
[0].File
!= TGSI_FILE_IMMEDIATE
) {
7366 opcode
= FETCH_OP_GATHER4_O
;
7368 /* GATHER4_O/GATHER4_C_O use offset values loaded by
7369 SET_TEXTURE_OFFSETS instruction. The immediate offset values
7370 encoded in the instruction are ignored. */
7371 memset(&tex
, 0, sizeof(struct r600_bytecode_tex
));
7372 tex
.op
= FETCH_OP_SET_TEXTURE_OFFSETS
;
7373 tex
.sampler_id
= tgsi_tex_get_src_gpr(ctx
, sampler_src_reg
);
7374 tex
.sampler_index_mode
= sampler_index_mode
;
7375 tex
.resource_id
= tex
.sampler_id
+ R600_MAX_CONST_BUFFERS
;
7376 tex
.resource_index_mode
= sampler_index_mode
;
7378 tex
.src_gpr
= ctx
->file_offset
[inst
->TexOffsets
[0].File
] + inst
->TexOffsets
[0].Index
;
7379 tex
.src_sel_x
= inst
->TexOffsets
[0].SwizzleX
;
7380 tex
.src_sel_y
= inst
->TexOffsets
[0].SwizzleY
;
7381 tex
.src_sel_z
= inst
->TexOffsets
[0].SwizzleZ
;
7389 r
= r600_bytecode_add_tex(ctx
->bc
, &tex
);
7394 if (inst
->Texture
.Texture
== TGSI_TEXTURE_SHADOW1D
||
7395 inst
->Texture
.Texture
== TGSI_TEXTURE_SHADOW2D
||
7396 inst
->Texture
.Texture
== TGSI_TEXTURE_SHADOWRECT
||
7397 inst
->Texture
.Texture
== TGSI_TEXTURE_SHADOWCUBE
||
7398 inst
->Texture
.Texture
== TGSI_TEXTURE_SHADOW1D_ARRAY
||
7399 inst
->Texture
.Texture
== TGSI_TEXTURE_SHADOW2D_ARRAY
||
7400 inst
->Texture
.Texture
== TGSI_TEXTURE_SHADOWCUBE_ARRAY
) {
7402 case FETCH_OP_SAMPLE
:
7403 opcode
= FETCH_OP_SAMPLE_C
;
7405 case FETCH_OP_SAMPLE_L
:
7406 opcode
= FETCH_OP_SAMPLE_C_L
;
7408 case FETCH_OP_SAMPLE_LB
:
7409 opcode
= FETCH_OP_SAMPLE_C_LB
;
7411 case FETCH_OP_SAMPLE_G
:
7412 opcode
= FETCH_OP_SAMPLE_C_G
;
7414 /* Texture gather variants */
7415 case FETCH_OP_GATHER4
:
7416 opcode
= FETCH_OP_GATHER4_C
;
7418 case FETCH_OP_GATHER4_O
:
7419 opcode
= FETCH_OP_GATHER4_C_O
;
7424 memset(&tex
, 0, sizeof(struct r600_bytecode_tex
));
7427 tex
.sampler_id
= tgsi_tex_get_src_gpr(ctx
, sampler_src_reg
);
7428 tex
.sampler_index_mode
= sampler_index_mode
;
7429 tex
.resource_id
= tex
.sampler_id
+ R600_MAX_CONST_BUFFERS
;
7430 tex
.resource_index_mode
= sampler_index_mode
;
7431 tex
.src_gpr
= src_gpr
;
7432 tex
.dst_gpr
= ctx
->file_offset
[inst
->Dst
[0].Register
.File
] + inst
->Dst
[0].Register
.Index
;
7434 if (inst
->Instruction
.Opcode
== TGSI_OPCODE_DDX_FINE
||
7435 inst
->Instruction
.Opcode
== TGSI_OPCODE_DDY_FINE
) {
7436 tex
.inst_mod
= 1; /* per pixel gradient calculation instead of per 2x2 quad */
7439 if (inst
->Instruction
.Opcode
== TGSI_OPCODE_TG4
) {
7440 int8_t texture_component_select
= ctx
->literals
[4 * inst
->Src
[1].Register
.Index
+ inst
->Src
[1].Register
.SwizzleX
];
7441 tex
.inst_mod
= texture_component_select
;
7443 if (ctx
->bc
->chip_class
== CAYMAN
) {
7444 /* GATHER4 result order is different from TGSI TG4 */
7445 tex
.dst_sel_x
= (inst
->Dst
[0].Register
.WriteMask
& 2) ? 0 : 7;
7446 tex
.dst_sel_y
= (inst
->Dst
[0].Register
.WriteMask
& 4) ? 1 : 7;
7447 tex
.dst_sel_z
= (inst
->Dst
[0].Register
.WriteMask
& 1) ? 2 : 7;
7448 tex
.dst_sel_w
= (inst
->Dst
[0].Register
.WriteMask
& 8) ? 3 : 7;
7450 tex
.dst_sel_x
= (inst
->Dst
[0].Register
.WriteMask
& 2) ? 1 : 7;
7451 tex
.dst_sel_y
= (inst
->Dst
[0].Register
.WriteMask
& 4) ? 2 : 7;
7452 tex
.dst_sel_z
= (inst
->Dst
[0].Register
.WriteMask
& 1) ? 0 : 7;
7453 tex
.dst_sel_w
= (inst
->Dst
[0].Register
.WriteMask
& 8) ? 3 : 7;
7456 else if (inst
->Instruction
.Opcode
== TGSI_OPCODE_LODQ
) {
7457 tex
.dst_sel_x
= (inst
->Dst
[0].Register
.WriteMask
& 2) ? 1 : 7;
7458 tex
.dst_sel_y
= (inst
->Dst
[0].Register
.WriteMask
& 1) ? 0 : 7;
7462 else if (inst
->Instruction
.Opcode
== TGSI_OPCODE_TXQS
) {
7469 tex
.dst_sel_x
= (inst
->Dst
[0].Register
.WriteMask
& 1) ? 0 : 7;
7470 tex
.dst_sel_y
= (inst
->Dst
[0].Register
.WriteMask
& 2) ? 1 : 7;
7471 tex
.dst_sel_z
= (inst
->Dst
[0].Register
.WriteMask
& 4) ? 2 : 7;
7472 tex
.dst_sel_w
= (inst
->Dst
[0].Register
.WriteMask
& 8) ? 3 : 7;
7476 if (inst
->Instruction
.Opcode
== TGSI_OPCODE_TXQ_LZ
||
7477 inst
->Instruction
.Opcode
== TGSI_OPCODE_TXQS
) {
7482 } else if (src_loaded
) {
7488 tex
.src_sel_x
= ctx
->src
[0].swizzle
[0];
7489 tex
.src_sel_y
= ctx
->src
[0].swizzle
[1];
7490 tex
.src_sel_z
= ctx
->src
[0].swizzle
[2];
7491 tex
.src_sel_w
= ctx
->src
[0].swizzle
[3];
7492 tex
.src_rel
= ctx
->src
[0].rel
;
7495 if (inst
->Texture
.Texture
== TGSI_TEXTURE_CUBE
||
7496 inst
->Texture
.Texture
== TGSI_TEXTURE_SHADOWCUBE
||
7497 inst
->Texture
.Texture
== TGSI_TEXTURE_CUBE_ARRAY
||
7498 inst
->Texture
.Texture
== TGSI_TEXTURE_SHADOWCUBE_ARRAY
) {
7502 tex
.src_sel_w
= 2; /* route Z compare or Lod value into W */
7505 if (inst
->Texture
.Texture
!= TGSI_TEXTURE_RECT
&&
7506 inst
->Texture
.Texture
!= TGSI_TEXTURE_SHADOWRECT
) {
7507 tex
.coord_type_x
= 1;
7508 tex
.coord_type_y
= 1;
7510 tex
.coord_type_z
= 1;
7511 tex
.coord_type_w
= 1;
7513 tex
.offset_x
= offset_x
;
7514 tex
.offset_y
= offset_y
;
7515 if (inst
->Instruction
.Opcode
== TGSI_OPCODE_TG4
&&
7516 (inst
->Texture
.Texture
== TGSI_TEXTURE_2D_ARRAY
||
7517 inst
->Texture
.Texture
== TGSI_TEXTURE_SHADOW2D_ARRAY
)) {
7521 tex
.offset_z
= offset_z
;
7524 /* Put the depth for comparison in W.
7525 * TGSI_TEXTURE_SHADOW2D_ARRAY already has the depth in W.
7526 * Some instructions expect the depth in Z. */
7527 if ((inst
->Texture
.Texture
== TGSI_TEXTURE_SHADOW1D
||
7528 inst
->Texture
.Texture
== TGSI_TEXTURE_SHADOW2D
||
7529 inst
->Texture
.Texture
== TGSI_TEXTURE_SHADOWRECT
||
7530 inst
->Texture
.Texture
== TGSI_TEXTURE_SHADOW1D_ARRAY
) &&
7531 opcode
!= FETCH_OP_SAMPLE_C_L
&&
7532 opcode
!= FETCH_OP_SAMPLE_C_LB
) {
7533 tex
.src_sel_w
= tex
.src_sel_z
;
7536 if (inst
->Texture
.Texture
== TGSI_TEXTURE_1D_ARRAY
||
7537 inst
->Texture
.Texture
== TGSI_TEXTURE_SHADOW1D_ARRAY
) {
7538 if (opcode
== FETCH_OP_SAMPLE_C_L
||
7539 opcode
== FETCH_OP_SAMPLE_C_LB
) {
7540 /* the array index is read from Y */
7541 tex
.coord_type_y
= 0;
7543 /* the array index is read from Z */
7544 tex
.coord_type_z
= 0;
7545 tex
.src_sel_z
= tex
.src_sel_y
;
7547 } else if (inst
->Texture
.Texture
== TGSI_TEXTURE_2D_ARRAY
||
7548 inst
->Texture
.Texture
== TGSI_TEXTURE_SHADOW2D_ARRAY
||
7549 ((inst
->Texture
.Texture
== TGSI_TEXTURE_CUBE_ARRAY
||
7550 inst
->Texture
.Texture
== TGSI_TEXTURE_SHADOWCUBE_ARRAY
) &&
7551 (ctx
->bc
->chip_class
>= EVERGREEN
)))
7552 /* the array index is read from Z */
7553 tex
.coord_type_z
= 0;
7555 /* mask unused source components */
7556 if (opcode
== FETCH_OP_SAMPLE
|| opcode
== FETCH_OP_GATHER4
) {
7557 switch (inst
->Texture
.Texture
) {
7558 case TGSI_TEXTURE_2D
:
7559 case TGSI_TEXTURE_RECT
:
7563 case TGSI_TEXTURE_1D_ARRAY
:
7567 case TGSI_TEXTURE_1D
:
7575 r
= r600_bytecode_add_tex(ctx
->bc
, &tex
);
7579 /* add shadow ambient support - gallium doesn't do it yet */
7583 static int tgsi_lrp(struct r600_shader_ctx
*ctx
)
7585 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
7586 struct r600_bytecode_alu alu
;
7587 int lasti
= tgsi_last_instruction(inst
->Dst
[0].Register
.WriteMask
);
7588 unsigned i
, temp_regs
[2];
7591 /* optimize if it's just an equal balance */
7592 if (ctx
->src
[0].sel
== V_SQ_ALU_SRC_0_5
) {
7593 for (i
= 0; i
< lasti
+ 1; i
++) {
7594 if (!(inst
->Dst
[0].Register
.WriteMask
& (1 << i
)))
7597 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
7598 alu
.op
= ALU_OP2_ADD
;
7599 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[1], i
);
7600 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[2], i
);
7602 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
7607 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
7615 for (i
= 0; i
< lasti
+ 1; i
++) {
7616 if (!(inst
->Dst
[0].Register
.WriteMask
& (1 << i
)))
7619 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
7620 alu
.op
= ALU_OP2_ADD
;
7621 alu
.src
[0].sel
= V_SQ_ALU_SRC_1
;
7622 alu
.src
[0].chan
= 0;
7623 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[0], i
);
7624 r600_bytecode_src_toggle_neg(&alu
.src
[1]);
7625 alu
.dst
.sel
= ctx
->temp_reg
;
7631 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
7636 /* (1 - src0) * src2 */
7637 for (i
= 0; i
< lasti
+ 1; i
++) {
7638 if (!(inst
->Dst
[0].Register
.WriteMask
& (1 << i
)))
7641 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
7642 alu
.op
= ALU_OP2_MUL
;
7643 alu
.src
[0].sel
= ctx
->temp_reg
;
7644 alu
.src
[0].chan
= i
;
7645 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[2], i
);
7646 alu
.dst
.sel
= ctx
->temp_reg
;
7652 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
7657 /* src0 * src1 + (1 - src0) * src2 */
7658 if (ctx
->src
[0].abs
)
7659 temp_regs
[0] = r600_get_temp(ctx
);
7662 if (ctx
->src
[1].abs
)
7663 temp_regs
[1] = r600_get_temp(ctx
);
7667 for (i
= 0; i
< lasti
+ 1; i
++) {
7668 if (!(inst
->Dst
[0].Register
.WriteMask
& (1 << i
)))
7671 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
7672 alu
.op
= ALU_OP3_MULADD
;
7674 r
= tgsi_make_src_for_op3(ctx
, temp_regs
[0], i
, &alu
.src
[0], &ctx
->src
[0]);
7677 r
= tgsi_make_src_for_op3(ctx
, temp_regs
[1], i
, &alu
.src
[1], &ctx
->src
[1]);
7680 alu
.src
[2].sel
= ctx
->temp_reg
;
7681 alu
.src
[2].chan
= i
;
7683 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
7688 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
7695 static int tgsi_cmp(struct r600_shader_ctx
*ctx
)
7697 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
7698 struct r600_bytecode_alu alu
;
7700 int lasti
= tgsi_last_instruction(inst
->Dst
[0].Register
.WriteMask
);
7704 if (ctx
->src
[0].abs
&& ctx
->src
[0].neg
) {
7706 ctx
->src
[0].abs
= 0;
7707 ctx
->src
[0].neg
= 0;
7712 for (j
= 0; j
< inst
->Instruction
.NumSrcRegs
; j
++) {
7714 if (ctx
->src
[j
].abs
)
7715 temp_regs
[j
] = r600_get_temp(ctx
);
7718 for (i
= 0; i
< lasti
+ 1; i
++) {
7719 if (!(inst
->Dst
[0].Register
.WriteMask
& (1 << i
)))
7722 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
7724 r
= tgsi_make_src_for_op3(ctx
, temp_regs
[0], i
, &alu
.src
[0], &ctx
->src
[0]);
7727 r
= tgsi_make_src_for_op3(ctx
, temp_regs
[2], i
, &alu
.src
[1], &ctx
->src
[2]);
7730 r
= tgsi_make_src_for_op3(ctx
, temp_regs
[1], i
, &alu
.src
[2], &ctx
->src
[1]);
7733 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
7739 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
7746 static int tgsi_ucmp(struct r600_shader_ctx
*ctx
)
7748 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
7749 struct r600_bytecode_alu alu
;
7751 int lasti
= tgsi_last_instruction(inst
->Dst
[0].Register
.WriteMask
);
7753 for (i
= 0; i
< lasti
+ 1; i
++) {
7754 if (!(inst
->Dst
[0].Register
.WriteMask
& (1 << i
)))
7757 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
7758 alu
.op
= ALU_OP3_CNDE_INT
;
7759 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], i
);
7760 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[2], i
);
7761 r600_bytecode_src(&alu
.src
[2], &ctx
->src
[1], i
);
7762 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
7768 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
7775 static int tgsi_xpd(struct r600_shader_ctx
*ctx
)
7777 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
7778 static const unsigned int src0_swizzle
[] = {2, 0, 1};
7779 static const unsigned int src1_swizzle
[] = {1, 2, 0};
7780 struct r600_bytecode_alu alu
;
7781 uint32_t use_temp
= 0;
7784 if (inst
->Dst
[0].Register
.WriteMask
!= 0xf)
7787 for (i
= 0; i
< 4; i
++) {
7788 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
7789 alu
.op
= ALU_OP2_MUL
;
7791 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], src0_swizzle
[i
]);
7792 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[1], src1_swizzle
[i
]);
7794 alu
.src
[0].sel
= V_SQ_ALU_SRC_0
;
7795 alu
.src
[0].chan
= i
;
7796 alu
.src
[1].sel
= V_SQ_ALU_SRC_0
;
7797 alu
.src
[1].chan
= i
;
7800 alu
.dst
.sel
= ctx
->temp_reg
;
7806 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
7811 for (i
= 0; i
< 4; i
++) {
7812 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
7813 alu
.op
= ALU_OP3_MULADD
;
7816 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], src1_swizzle
[i
]);
7817 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[1], src0_swizzle
[i
]);
7819 alu
.src
[0].sel
= V_SQ_ALU_SRC_0
;
7820 alu
.src
[0].chan
= i
;
7821 alu
.src
[1].sel
= V_SQ_ALU_SRC_0
;
7822 alu
.src
[1].chan
= i
;
7825 alu
.src
[2].sel
= ctx
->temp_reg
;
7827 alu
.src
[2].chan
= i
;
7830 alu
.dst
.sel
= ctx
->temp_reg
;
7832 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
7838 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
7843 return tgsi_helper_copy(ctx
, inst
);
7847 static int tgsi_exp(struct r600_shader_ctx
*ctx
)
7849 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
7850 struct r600_bytecode_alu alu
;
7854 /* result.x = 2^floor(src); */
7855 if (inst
->Dst
[0].Register
.WriteMask
& 1) {
7856 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
7858 alu
.op
= ALU_OP1_FLOOR
;
7859 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 0);
7861 alu
.dst
.sel
= ctx
->temp_reg
;
7865 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
7869 if (ctx
->bc
->chip_class
== CAYMAN
) {
7870 for (i
= 0; i
< 3; i
++) {
7871 alu
.op
= ALU_OP1_EXP_IEEE
;
7872 alu
.src
[0].sel
= ctx
->temp_reg
;
7873 alu
.src
[0].chan
= 0;
7875 alu
.dst
.sel
= ctx
->temp_reg
;
7877 alu
.dst
.write
= i
== 0;
7879 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
7884 alu
.op
= ALU_OP1_EXP_IEEE
;
7885 alu
.src
[0].sel
= ctx
->temp_reg
;
7886 alu
.src
[0].chan
= 0;
7888 alu
.dst
.sel
= ctx
->temp_reg
;
7892 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
7898 /* result.y = tmp - floor(tmp); */
7899 if ((inst
->Dst
[0].Register
.WriteMask
>> 1) & 1) {
7900 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
7902 alu
.op
= ALU_OP1_FRACT
;
7903 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 0);
7905 alu
.dst
.sel
= ctx
->temp_reg
;
7907 r
= tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
7916 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
7921 /* result.z = RoughApprox2ToX(tmp);*/
7922 if ((inst
->Dst
[0].Register
.WriteMask
>> 2) & 0x1) {
7923 if (ctx
->bc
->chip_class
== CAYMAN
) {
7924 for (i
= 0; i
< 3; i
++) {
7925 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
7926 alu
.op
= ALU_OP1_EXP_IEEE
;
7927 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 0);
7929 alu
.dst
.sel
= ctx
->temp_reg
;
7936 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
7941 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
7942 alu
.op
= ALU_OP1_EXP_IEEE
;
7943 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 0);
7945 alu
.dst
.sel
= ctx
->temp_reg
;
7951 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
7957 /* result.w = 1.0;*/
7958 if ((inst
->Dst
[0].Register
.WriteMask
>> 3) & 0x1) {
7959 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
7961 alu
.op
= ALU_OP1_MOV
;
7962 alu
.src
[0].sel
= V_SQ_ALU_SRC_1
;
7963 alu
.src
[0].chan
= 0;
7965 alu
.dst
.sel
= ctx
->temp_reg
;
7969 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
7973 return tgsi_helper_copy(ctx
, inst
);
7976 static int tgsi_log(struct r600_shader_ctx
*ctx
)
7978 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
7979 struct r600_bytecode_alu alu
;
7983 /* result.x = floor(log2(|src|)); */
7984 if (inst
->Dst
[0].Register
.WriteMask
& 1) {
7985 if (ctx
->bc
->chip_class
== CAYMAN
) {
7986 for (i
= 0; i
< 3; i
++) {
7987 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
7989 alu
.op
= ALU_OP1_LOG_IEEE
;
7990 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 0);
7991 r600_bytecode_src_set_abs(&alu
.src
[0]);
7993 alu
.dst
.sel
= ctx
->temp_reg
;
7999 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
8005 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
8007 alu
.op
= ALU_OP1_LOG_IEEE
;
8008 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 0);
8009 r600_bytecode_src_set_abs(&alu
.src
[0]);
8011 alu
.dst
.sel
= ctx
->temp_reg
;
8015 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
8020 alu
.op
= ALU_OP1_FLOOR
;
8021 alu
.src
[0].sel
= ctx
->temp_reg
;
8022 alu
.src
[0].chan
= 0;
8024 alu
.dst
.sel
= ctx
->temp_reg
;
8029 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
8034 /* result.y = |src.x| / (2 ^ floor(log2(|src.x|))); */
8035 if ((inst
->Dst
[0].Register
.WriteMask
>> 1) & 1) {
8037 if (ctx
->bc
->chip_class
== CAYMAN
) {
8038 for (i
= 0; i
< 3; i
++) {
8039 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
8041 alu
.op
= ALU_OP1_LOG_IEEE
;
8042 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 0);
8043 r600_bytecode_src_set_abs(&alu
.src
[0]);
8045 alu
.dst
.sel
= ctx
->temp_reg
;
8052 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
8057 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
8059 alu
.op
= ALU_OP1_LOG_IEEE
;
8060 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 0);
8061 r600_bytecode_src_set_abs(&alu
.src
[0]);
8063 alu
.dst
.sel
= ctx
->temp_reg
;
8068 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
8073 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
8075 alu
.op
= ALU_OP1_FLOOR
;
8076 alu
.src
[0].sel
= ctx
->temp_reg
;
8077 alu
.src
[0].chan
= 1;
8079 alu
.dst
.sel
= ctx
->temp_reg
;
8084 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
8088 if (ctx
->bc
->chip_class
== CAYMAN
) {
8089 for (i
= 0; i
< 3; i
++) {
8090 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
8091 alu
.op
= ALU_OP1_EXP_IEEE
;
8092 alu
.src
[0].sel
= ctx
->temp_reg
;
8093 alu
.src
[0].chan
= 1;
8095 alu
.dst
.sel
= ctx
->temp_reg
;
8102 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
8107 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
8108 alu
.op
= ALU_OP1_EXP_IEEE
;
8109 alu
.src
[0].sel
= ctx
->temp_reg
;
8110 alu
.src
[0].chan
= 1;
8112 alu
.dst
.sel
= ctx
->temp_reg
;
8117 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
8122 if (ctx
->bc
->chip_class
== CAYMAN
) {
8123 for (i
= 0; i
< 3; i
++) {
8124 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
8125 alu
.op
= ALU_OP1_RECIP_IEEE
;
8126 alu
.src
[0].sel
= ctx
->temp_reg
;
8127 alu
.src
[0].chan
= 1;
8129 alu
.dst
.sel
= ctx
->temp_reg
;
8136 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
8141 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
8142 alu
.op
= ALU_OP1_RECIP_IEEE
;
8143 alu
.src
[0].sel
= ctx
->temp_reg
;
8144 alu
.src
[0].chan
= 1;
8146 alu
.dst
.sel
= ctx
->temp_reg
;
8151 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
8156 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
8158 alu
.op
= ALU_OP2_MUL
;
8160 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 0);
8161 r600_bytecode_src_set_abs(&alu
.src
[0]);
8163 alu
.src
[1].sel
= ctx
->temp_reg
;
8164 alu
.src
[1].chan
= 1;
8166 alu
.dst
.sel
= ctx
->temp_reg
;
8171 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
8176 /* result.z = log2(|src|);*/
8177 if ((inst
->Dst
[0].Register
.WriteMask
>> 2) & 1) {
8178 if (ctx
->bc
->chip_class
== CAYMAN
) {
8179 for (i
= 0; i
< 3; i
++) {
8180 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
8182 alu
.op
= ALU_OP1_LOG_IEEE
;
8183 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 0);
8184 r600_bytecode_src_set_abs(&alu
.src
[0]);
8186 alu
.dst
.sel
= ctx
->temp_reg
;
8193 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
8198 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
8200 alu
.op
= ALU_OP1_LOG_IEEE
;
8201 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 0);
8202 r600_bytecode_src_set_abs(&alu
.src
[0]);
8204 alu
.dst
.sel
= ctx
->temp_reg
;
8209 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
8215 /* result.w = 1.0; */
8216 if ((inst
->Dst
[0].Register
.WriteMask
>> 3) & 1) {
8217 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
8219 alu
.op
= ALU_OP1_MOV
;
8220 alu
.src
[0].sel
= V_SQ_ALU_SRC_1
;
8221 alu
.src
[0].chan
= 0;
8223 alu
.dst
.sel
= ctx
->temp_reg
;
8228 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
8233 return tgsi_helper_copy(ctx
, inst
);
8236 static int tgsi_eg_arl(struct r600_shader_ctx
*ctx
)
8238 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
8239 struct r600_bytecode_alu alu
;
8241 int i
, lasti
= tgsi_last_instruction(inst
->Dst
[0].Register
.WriteMask
);
8242 unsigned reg
= get_address_file_reg(ctx
, inst
->Dst
[0].Register
.Index
);
8244 assert(inst
->Dst
[0].Register
.Index
< 3);
8245 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
8247 switch (inst
->Instruction
.Opcode
) {
8248 case TGSI_OPCODE_ARL
:
8249 alu
.op
= ALU_OP1_FLT_TO_INT_FLOOR
;
8251 case TGSI_OPCODE_ARR
:
8252 alu
.op
= ALU_OP1_FLT_TO_INT
;
8254 case TGSI_OPCODE_UARL
:
8255 alu
.op
= ALU_OP1_MOV
;
8262 for (i
= 0; i
<= lasti
; ++i
) {
8263 if (!(inst
->Dst
[0].Register
.WriteMask
& (1 << i
)))
8265 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], i
);
8266 alu
.last
= i
== lasti
;
8270 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
8275 if (inst
->Dst
[0].Register
.Index
> 0)
8276 ctx
->bc
->index_loaded
[inst
->Dst
[0].Register
.Index
- 1] = 0;
8278 ctx
->bc
->ar_loaded
= 0;
8282 static int tgsi_r600_arl(struct r600_shader_ctx
*ctx
)
8284 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
8285 struct r600_bytecode_alu alu
;
8287 int i
, lasti
= tgsi_last_instruction(inst
->Dst
[0].Register
.WriteMask
);
8289 switch (inst
->Instruction
.Opcode
) {
8290 case TGSI_OPCODE_ARL
:
8291 memset(&alu
, 0, sizeof(alu
));
8292 alu
.op
= ALU_OP1_FLOOR
;
8293 alu
.dst
.sel
= ctx
->bc
->ar_reg
;
8295 for (i
= 0; i
<= lasti
; ++i
) {
8296 if (inst
->Dst
[0].Register
.WriteMask
& (1 << i
)) {
8298 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], i
);
8299 alu
.last
= i
== lasti
;
8300 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
8305 memset(&alu
, 0, sizeof(alu
));
8306 alu
.op
= ALU_OP1_FLT_TO_INT
;
8307 alu
.src
[0].sel
= ctx
->bc
->ar_reg
;
8308 alu
.dst
.sel
= ctx
->bc
->ar_reg
;
8310 /* FLT_TO_INT is trans-only on r600/r700 */
8312 for (i
= 0; i
<= lasti
; ++i
) {
8314 alu
.src
[0].chan
= i
;
8315 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
8319 case TGSI_OPCODE_ARR
:
8320 memset(&alu
, 0, sizeof(alu
));
8321 alu
.op
= ALU_OP1_FLT_TO_INT
;
8322 alu
.dst
.sel
= ctx
->bc
->ar_reg
;
8324 /* FLT_TO_INT is trans-only on r600/r700 */
8326 for (i
= 0; i
<= lasti
; ++i
) {
8327 if (inst
->Dst
[0].Register
.WriteMask
& (1 << i
)) {
8329 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], i
);
8330 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
8335 case TGSI_OPCODE_UARL
:
8336 memset(&alu
, 0, sizeof(alu
));
8337 alu
.op
= ALU_OP1_MOV
;
8338 alu
.dst
.sel
= ctx
->bc
->ar_reg
;
8340 for (i
= 0; i
<= lasti
; ++i
) {
8341 if (inst
->Dst
[0].Register
.WriteMask
& (1 << i
)) {
8343 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], i
);
8344 alu
.last
= i
== lasti
;
8345 if ((r
= r600_bytecode_add_alu(ctx
->bc
, &alu
)))
8355 ctx
->bc
->ar_loaded
= 0;
8359 static int tgsi_opdst(struct r600_shader_ctx
*ctx
)
8361 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
8362 struct r600_bytecode_alu alu
;
8365 for (i
= 0; i
< 4; i
++) {
8366 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
8368 alu
.op
= ALU_OP2_MUL
;
8369 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
8371 if (i
== 0 || i
== 3) {
8372 alu
.src
[0].sel
= V_SQ_ALU_SRC_1
;
8374 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], i
);
8377 if (i
== 0 || i
== 2) {
8378 alu
.src
[1].sel
= V_SQ_ALU_SRC_1
;
8380 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[1], i
);
8384 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
8391 static int emit_logic_pred(struct r600_shader_ctx
*ctx
, int opcode
, int alu_type
)
8393 struct r600_bytecode_alu alu
;
8396 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
8398 alu
.execute_mask
= 1;
8399 alu
.update_pred
= 1;
8401 alu
.dst
.sel
= ctx
->temp_reg
;
8405 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 0);
8406 alu
.src
[1].sel
= V_SQ_ALU_SRC_0
;
8407 alu
.src
[1].chan
= 0;
8411 r
= r600_bytecode_add_alu_type(ctx
->bc
, &alu
, alu_type
);
8417 static int pops(struct r600_shader_ctx
*ctx
, int pops
)
8419 unsigned force_pop
= ctx
->bc
->force_add_cf
;
8423 if (ctx
->bc
->cf_last
) {
8424 if (ctx
->bc
->cf_last
->op
== CF_OP_ALU
)
8426 else if (ctx
->bc
->cf_last
->op
== CF_OP_ALU_POP_AFTER
)
8431 ctx
->bc
->cf_last
->op
= CF_OP_ALU_POP_AFTER
;
8432 ctx
->bc
->force_add_cf
= 1;
8433 } else if (alu_pop
== 2) {
8434 ctx
->bc
->cf_last
->op
= CF_OP_ALU_POP2_AFTER
;
8435 ctx
->bc
->force_add_cf
= 1;
8442 r600_bytecode_add_cfinst(ctx
->bc
, CF_OP_POP
);
8443 ctx
->bc
->cf_last
->pop_count
= pops
;
8444 ctx
->bc
->cf_last
->cf_addr
= ctx
->bc
->cf_last
->id
+ 2;
8450 static inline void callstack_update_max_depth(struct r600_shader_ctx
*ctx
,
8453 struct r600_stack_info
*stack
= &ctx
->bc
->stack
;
8454 unsigned elements
, entries
;
8456 unsigned entry_size
= stack
->entry_size
;
8458 elements
= (stack
->loop
+ stack
->push_wqm
) * entry_size
;
8459 elements
+= stack
->push
;
8461 switch (ctx
->bc
->chip_class
) {
8464 /* pre-r8xx: if any non-WQM PUSH instruction is invoked, 2 elements on
8465 * the stack must be reserved to hold the current active/continue
8467 if (reason
== FC_PUSH_VPM
) {
8473 /* r9xx: any stack operation on empty stack consumes 2 additional
8478 /* FIXME: do the two elements added above cover the cases for the
8482 /* r8xx+: 2 extra elements are not always required, but one extra
8483 * element must be added for each of the following cases:
8484 * 1. There is an ALU_ELSE_AFTER instruction at the point of greatest
8486 * (Currently we don't use ALU_ELSE_AFTER.)
8487 * 2. There are LOOP/WQM frames on the stack when any flavor of non-WQM
8488 * PUSH instruction executed.
8490 * NOTE: it seems we also need to reserve additional element in some
8491 * other cases, e.g. when we have 4 levels of PUSH_VPM in the shader,
8492 * then STACK_SIZE should be 2 instead of 1 */
8493 if (reason
== FC_PUSH_VPM
) {
8503 /* NOTE: it seems STACK_SIZE is interpreted by hw as if entry_size is 4
8504 * for all chips, so we use 4 in the final formula, not the real entry_size
8508 entries
= (elements
+ (entry_size
- 1)) / entry_size
;
8510 if (entries
> stack
->max_entries
)
8511 stack
->max_entries
= entries
;
8514 static inline void callstack_pop(struct r600_shader_ctx
*ctx
, unsigned reason
)
8518 --ctx
->bc
->stack
.push
;
8519 assert(ctx
->bc
->stack
.push
>= 0);
8522 --ctx
->bc
->stack
.push_wqm
;
8523 assert(ctx
->bc
->stack
.push_wqm
>= 0);
8526 --ctx
->bc
->stack
.loop
;
8527 assert(ctx
->bc
->stack
.loop
>= 0);
8535 static inline void callstack_push(struct r600_shader_ctx
*ctx
, unsigned reason
)
8539 ++ctx
->bc
->stack
.push
;
8542 ++ctx
->bc
->stack
.push_wqm
;
8544 ++ctx
->bc
->stack
.loop
;
8550 callstack_update_max_depth(ctx
, reason
);
8553 static void fc_set_mid(struct r600_shader_ctx
*ctx
, int fc_sp
)
8555 struct r600_cf_stack_entry
*sp
= &ctx
->bc
->fc_stack
[fc_sp
];
8557 sp
->mid
= realloc((void *)sp
->mid
,
8558 sizeof(struct r600_bytecode_cf
*) * (sp
->num_mid
+ 1));
8559 sp
->mid
[sp
->num_mid
] = ctx
->bc
->cf_last
;
8563 static void fc_pushlevel(struct r600_shader_ctx
*ctx
, int type
)
8566 ctx
->bc
->fc_stack
[ctx
->bc
->fc_sp
].type
= type
;
8567 ctx
->bc
->fc_stack
[ctx
->bc
->fc_sp
].start
= ctx
->bc
->cf_last
;
8570 static void fc_poplevel(struct r600_shader_ctx
*ctx
)
8572 struct r600_cf_stack_entry
*sp
= &ctx
->bc
->fc_stack
[ctx
->bc
->fc_sp
];
8582 static int emit_return(struct r600_shader_ctx
*ctx
)
8584 r600_bytecode_add_cfinst(ctx
->bc
, CF_OP_RETURN
));
8588 static int emit_jump_to_offset(struct r600_shader_ctx
*ctx
, int pops
, int offset
)
8591 r600_bytecode_add_cfinst(ctx
->bc
, CF_OP_JUMP
));
8592 ctx
->bc
->cf_last
->pop_count
= pops
;
8593 /* XXX work out offset */
8597 static int emit_setret_in_loop_flag(struct r600_shader_ctx
*ctx
, unsigned flag_value
)
8602 static void emit_testflag(struct r600_shader_ctx
*ctx
)
8607 static void emit_return_on_flag(struct r600_shader_ctx
*ctx
, unsigned ifidx
)
8610 emit_jump_to_offset(ctx
, 1, 4);
8611 emit_setret_in_loop_flag(ctx
, V_SQ_ALU_SRC_0
);
8612 pops(ctx
, ifidx
+ 1);
8616 static void break_loop_on_flag(struct r600_shader_ctx
*ctx
, unsigned fc_sp
)
8620 r600_bytecode_add_cfinst(ctx
->bc
, ctx
->inst_info
->op
);
8621 ctx
->bc
->cf_last
->pop_count
= 1;
8623 fc_set_mid(ctx
, fc_sp
);
8629 static int emit_if(struct r600_shader_ctx
*ctx
, int opcode
)
8631 int alu_type
= CF_OP_ALU_PUSH_BEFORE
;
8633 /* There is a hardware bug on Cayman where a BREAK/CONTINUE followed by
8634 * LOOP_STARTxxx for nested loops may put the branch stack into a state
8635 * such that ALU_PUSH_BEFORE doesn't work as expected. Workaround this
8636 * by replacing the ALU_PUSH_BEFORE with a PUSH + ALU */
8637 if (ctx
->bc
->chip_class
== CAYMAN
&& ctx
->bc
->stack
.loop
> 1) {
8638 r600_bytecode_add_cfinst(ctx
->bc
, CF_OP_PUSH
);
8639 ctx
->bc
->cf_last
->cf_addr
= ctx
->bc
->cf_last
->id
+ 2;
8640 alu_type
= CF_OP_ALU
;
8643 emit_logic_pred(ctx
, opcode
, alu_type
);
8645 r600_bytecode_add_cfinst(ctx
->bc
, CF_OP_JUMP
);
8647 fc_pushlevel(ctx
, FC_IF
);
8649 callstack_push(ctx
, FC_PUSH_VPM
);
8653 static int tgsi_if(struct r600_shader_ctx
*ctx
)
8655 return emit_if(ctx
, ALU_OP2_PRED_SETNE
);
8658 static int tgsi_uif(struct r600_shader_ctx
*ctx
)
8660 return emit_if(ctx
, ALU_OP2_PRED_SETNE_INT
);
8663 static int tgsi_else(struct r600_shader_ctx
*ctx
)
8665 r600_bytecode_add_cfinst(ctx
->bc
, CF_OP_ELSE
);
8666 ctx
->bc
->cf_last
->pop_count
= 1;
8668 fc_set_mid(ctx
, ctx
->bc
->fc_sp
);
8669 ctx
->bc
->fc_stack
[ctx
->bc
->fc_sp
].start
->cf_addr
= ctx
->bc
->cf_last
->id
;
8673 static int tgsi_endif(struct r600_shader_ctx
*ctx
)
8676 if (ctx
->bc
->fc_stack
[ctx
->bc
->fc_sp
].type
!= FC_IF
) {
8677 R600_ERR("if/endif unbalanced in shader\n");
8681 if (ctx
->bc
->fc_stack
[ctx
->bc
->fc_sp
].mid
== NULL
) {
8682 ctx
->bc
->fc_stack
[ctx
->bc
->fc_sp
].start
->cf_addr
= ctx
->bc
->cf_last
->id
+ 2;
8683 ctx
->bc
->fc_stack
[ctx
->bc
->fc_sp
].start
->pop_count
= 1;
8685 ctx
->bc
->fc_stack
[ctx
->bc
->fc_sp
].mid
[0]->cf_addr
= ctx
->bc
->cf_last
->id
+ 2;
8689 callstack_pop(ctx
, FC_PUSH_VPM
);
8693 static int tgsi_bgnloop(struct r600_shader_ctx
*ctx
)
8695 /* LOOP_START_DX10 ignores the LOOP_CONFIG* registers, so it is not
8696 * limited to 4096 iterations, like the other LOOP_* instructions. */
8697 r600_bytecode_add_cfinst(ctx
->bc
, CF_OP_LOOP_START_DX10
);
8699 fc_pushlevel(ctx
, FC_LOOP
);
8701 /* check stack depth */
8702 callstack_push(ctx
, FC_LOOP
);
8706 static int tgsi_endloop(struct r600_shader_ctx
*ctx
)
8710 r600_bytecode_add_cfinst(ctx
->bc
, CF_OP_LOOP_END
);
8712 if (ctx
->bc
->fc_stack
[ctx
->bc
->fc_sp
].type
!= FC_LOOP
) {
8713 R600_ERR("loop/endloop in shader code are not paired.\n");
8717 /* fixup loop pointers - from r600isa
8718 LOOP END points to CF after LOOP START,
8719 LOOP START point to CF after LOOP END
8720 BRK/CONT point to LOOP END CF
8722 ctx
->bc
->cf_last
->cf_addr
= ctx
->bc
->fc_stack
[ctx
->bc
->fc_sp
].start
->id
+ 2;
8724 ctx
->bc
->fc_stack
[ctx
->bc
->fc_sp
].start
->cf_addr
= ctx
->bc
->cf_last
->id
+ 2;
8726 for (i
= 0; i
< ctx
->bc
->fc_stack
[ctx
->bc
->fc_sp
].num_mid
; i
++) {
8727 ctx
->bc
->fc_stack
[ctx
->bc
->fc_sp
].mid
[i
]->cf_addr
= ctx
->bc
->cf_last
->id
;
8729 /* XXX add LOOPRET support */
8731 callstack_pop(ctx
, FC_LOOP
);
8735 static int tgsi_loop_breakc(struct r600_shader_ctx
*ctx
)
8740 for (fscp
= ctx
->bc
->fc_sp
; fscp
> 0; fscp
--)
8742 if (FC_LOOP
== ctx
->bc
->fc_stack
[fscp
].type
)
8746 R600_ERR("BREAKC not inside loop/endloop pair\n");
8750 if (ctx
->bc
->chip_class
== EVERGREEN
&&
8751 ctx
->bc
->family
!= CHIP_CYPRESS
&&
8752 ctx
->bc
->family
!= CHIP_JUNIPER
) {
8753 /* HW bug: ALU_BREAK does not save the active mask correctly */
8758 r
= r600_bytecode_add_cfinst(ctx
->bc
, CF_OP_LOOP_BREAK
);
8761 fc_set_mid(ctx
, fscp
);
8763 return tgsi_endif(ctx
);
8765 r
= emit_logic_pred(ctx
, ALU_OP2_PRED_SETE_INT
, CF_OP_ALU_BREAK
);
8768 fc_set_mid(ctx
, fscp
);
8774 static int tgsi_loop_brk_cont(struct r600_shader_ctx
*ctx
)
8778 for (fscp
= ctx
->bc
->fc_sp
; fscp
> 0; fscp
--)
8780 if (FC_LOOP
== ctx
->bc
->fc_stack
[fscp
].type
)
8785 R600_ERR("Break not inside loop/endloop pair\n");
8789 r600_bytecode_add_cfinst(ctx
->bc
, ctx
->inst_info
->op
);
8791 fc_set_mid(ctx
, fscp
);
8796 static int tgsi_gs_emit(struct r600_shader_ctx
*ctx
)
8798 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
8799 int stream
= ctx
->literals
[inst
->Src
[0].Register
.Index
* 4 + inst
->Src
[0].Register
.SwizzleX
];
8802 if (ctx
->inst_info
->op
== CF_OP_EMIT_VERTEX
)
8803 emit_gs_ring_writes(ctx
, ctx
->gs_stream_output_info
, stream
, TRUE
);
8805 r
= r600_bytecode_add_cfinst(ctx
->bc
, ctx
->inst_info
->op
);
8807 ctx
->bc
->cf_last
->count
= stream
; // Count field for CUT/EMIT_VERTEX indicates which stream
8808 if (ctx
->inst_info
->op
== CF_OP_EMIT_VERTEX
)
8809 return emit_inc_ring_offset(ctx
, stream
, TRUE
);
8814 static int tgsi_umad(struct r600_shader_ctx
*ctx
)
8816 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
8817 struct r600_bytecode_alu alu
;
8819 int lasti
= tgsi_last_instruction(inst
->Dst
[0].Register
.WriteMask
);
8822 for (i
= 0; i
< lasti
+ 1; i
++) {
8823 if (!(inst
->Dst
[0].Register
.WriteMask
& (1 << i
)))
8826 if (ctx
->bc
->chip_class
== CAYMAN
) {
8827 for (j
= 0 ; j
< 4; j
++) {
8828 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
8830 alu
.op
= ALU_OP2_MULLO_UINT
;
8831 for (k
= 0; k
< inst
->Instruction
.NumSrcRegs
; k
++) {
8832 r600_bytecode_src(&alu
.src
[k
], &ctx
->src
[k
], i
);
8835 alu
.dst
.sel
= ctx
->temp_reg
;
8836 alu
.dst
.write
= (j
== i
);
8839 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
8844 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
8847 alu
.dst
.sel
= ctx
->temp_reg
;
8850 alu
.op
= ALU_OP2_MULLO_UINT
;
8851 for (j
= 0; j
< 2; j
++) {
8852 r600_bytecode_src(&alu
.src
[j
], &ctx
->src
[j
], i
);
8856 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
8863 for (i
= 0; i
< lasti
+ 1; i
++) {
8864 if (!(inst
->Dst
[0].Register
.WriteMask
& (1 << i
)))
8867 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
8868 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
8870 alu
.op
= ALU_OP2_ADD_INT
;
8872 alu
.src
[0].sel
= ctx
->temp_reg
;
8873 alu
.src
[0].chan
= i
;
8875 r600_bytecode_src(&alu
.src
[1], &ctx
->src
[2], i
);
8879 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
8886 static int tgsi_pk2h(struct r600_shader_ctx
*ctx
)
8888 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
8889 struct r600_bytecode_alu alu
;
8891 int lasti
= tgsi_last_instruction(inst
->Dst
[0].Register
.WriteMask
);
8893 /* temp.xy = f32_to_f16(src) */
8894 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
8895 alu
.op
= ALU_OP1_FLT32_TO_FLT16
;
8897 alu
.dst
.sel
= ctx
->temp_reg
;
8899 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 0);
8900 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
8904 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 1);
8906 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
8910 /* dst.x = temp.y * 0x10000 + temp.x */
8911 for (i
= 0; i
< lasti
+ 1; i
++) {
8912 if (!(inst
->Dst
[0].Register
.WriteMask
& (1 << i
)))
8915 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
8916 alu
.op
= ALU_OP3_MULADD_UINT24
;
8918 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
8919 alu
.last
= i
== lasti
;
8920 alu
.src
[0].sel
= ctx
->temp_reg
;
8921 alu
.src
[0].chan
= 1;
8922 alu
.src
[1].sel
= V_SQ_ALU_SRC_LITERAL
;
8923 alu
.src
[1].value
= 0x10000;
8924 alu
.src
[2].sel
= ctx
->temp_reg
;
8925 alu
.src
[2].chan
= 0;
8926 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
8934 static int tgsi_up2h(struct r600_shader_ctx
*ctx
)
8936 struct tgsi_full_instruction
*inst
= &ctx
->parse
.FullToken
.FullInstruction
;
8937 struct r600_bytecode_alu alu
;
8939 int lasti
= tgsi_last_instruction(inst
->Dst
[0].Register
.WriteMask
);
8941 /* temp.x = src.x */
8942 /* note: no need to mask out the high bits */
8943 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
8944 alu
.op
= ALU_OP1_MOV
;
8946 alu
.dst
.sel
= ctx
->temp_reg
;
8948 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 0);
8949 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
8953 /* temp.y = src.x >> 16 */
8954 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
8955 alu
.op
= ALU_OP2_LSHR_INT
;
8957 alu
.dst
.sel
= ctx
->temp_reg
;
8959 r600_bytecode_src(&alu
.src
[0], &ctx
->src
[0], 0);
8960 alu
.src
[1].sel
= V_SQ_ALU_SRC_LITERAL
;
8961 alu
.src
[1].value
= 16;
8963 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
8967 /* dst.wz = dst.xy = f16_to_f32(temp.xy) */
8968 for (i
= 0; i
< lasti
+ 1; i
++) {
8969 if (!(inst
->Dst
[0].Register
.WriteMask
& (1 << i
)))
8971 memset(&alu
, 0, sizeof(struct r600_bytecode_alu
));
8972 tgsi_dst(ctx
, &inst
->Dst
[0], i
, &alu
.dst
);
8973 alu
.op
= ALU_OP1_FLT16_TO_FLT32
;
8974 alu
.src
[0].sel
= ctx
->temp_reg
;
8975 alu
.src
[0].chan
= i
% 2;
8976 alu
.last
= i
== lasti
;
8977 r
= r600_bytecode_add_alu(ctx
->bc
, &alu
);
8985 static const struct r600_shader_tgsi_instruction r600_shader_tgsi_instruction
[] = {
8986 [TGSI_OPCODE_ARL
] = { ALU_OP0_NOP
, tgsi_r600_arl
},
8987 [TGSI_OPCODE_MOV
] = { ALU_OP1_MOV
, tgsi_op2
},
8988 [TGSI_OPCODE_LIT
] = { ALU_OP0_NOP
, tgsi_lit
},
8991 * For state trackers other than OpenGL, we'll want to use
8992 * _RECIP_IEEE instead.
8994 [TGSI_OPCODE_RCP
] = { ALU_OP1_RECIP_CLAMPED
, tgsi_trans_srcx_replicate
},
8996 [TGSI_OPCODE_RSQ
] = { ALU_OP0_NOP
, tgsi_rsq
},
8997 [TGSI_OPCODE_EXP
] = { ALU_OP0_NOP
, tgsi_exp
},
8998 [TGSI_OPCODE_LOG
] = { ALU_OP0_NOP
, tgsi_log
},
8999 [TGSI_OPCODE_MUL
] = { ALU_OP2_MUL
, tgsi_op2
},
9000 [TGSI_OPCODE_ADD
] = { ALU_OP2_ADD
, tgsi_op2
},
9001 [TGSI_OPCODE_DP3
] = { ALU_OP2_DOT4
, tgsi_dp
},
9002 [TGSI_OPCODE_DP4
] = { ALU_OP2_DOT4
, tgsi_dp
},
9003 [TGSI_OPCODE_DST
] = { ALU_OP0_NOP
, tgsi_opdst
},
9004 [TGSI_OPCODE_MIN
] = { ALU_OP2_MIN
, tgsi_op2
},
9005 [TGSI_OPCODE_MAX
] = { ALU_OP2_MAX
, tgsi_op2
},
9006 [TGSI_OPCODE_SLT
] = { ALU_OP2_SETGT
, tgsi_op2_swap
},
9007 [TGSI_OPCODE_SGE
] = { ALU_OP2_SETGE
, tgsi_op2
},
9008 [TGSI_OPCODE_MAD
] = { ALU_OP3_MULADD
, tgsi_op3
},
9009 [TGSI_OPCODE_SUB
] = { ALU_OP2_ADD
, tgsi_op2
},
9010 [TGSI_OPCODE_LRP
] = { ALU_OP0_NOP
, tgsi_lrp
},
9011 [TGSI_OPCODE_FMA
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9012 [TGSI_OPCODE_SQRT
] = { ALU_OP1_SQRT_IEEE
, tgsi_trans_srcx_replicate
},
9013 [TGSI_OPCODE_DP2A
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9014 [22] = { ALU_OP0_NOP
, tgsi_unsupported
},
9015 [23] = { ALU_OP0_NOP
, tgsi_unsupported
},
9016 [TGSI_OPCODE_FRC
] = { ALU_OP1_FRACT
, tgsi_op2
},
9017 [TGSI_OPCODE_CLAMP
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9018 [TGSI_OPCODE_FLR
] = { ALU_OP1_FLOOR
, tgsi_op2
},
9019 [TGSI_OPCODE_ROUND
] = { ALU_OP1_RNDNE
, tgsi_op2
},
9020 [TGSI_OPCODE_EX2
] = { ALU_OP1_EXP_IEEE
, tgsi_trans_srcx_replicate
},
9021 [TGSI_OPCODE_LG2
] = { ALU_OP1_LOG_IEEE
, tgsi_trans_srcx_replicate
},
9022 [TGSI_OPCODE_POW
] = { ALU_OP0_NOP
, tgsi_pow
},
9023 [TGSI_OPCODE_XPD
] = { ALU_OP0_NOP
, tgsi_xpd
},
9024 [32] = { ALU_OP0_NOP
, tgsi_unsupported
},
9025 [33] = { ALU_OP0_NOP
, tgsi_unsupported
},
9026 [34] = { ALU_OP0_NOP
, tgsi_unsupported
},
9027 [TGSI_OPCODE_DPH
] = { ALU_OP2_DOT4
, tgsi_dp
},
9028 [TGSI_OPCODE_COS
] = { ALU_OP1_COS
, tgsi_trig
},
9029 [TGSI_OPCODE_DDX
] = { FETCH_OP_GET_GRADIENTS_H
, tgsi_tex
},
9030 [TGSI_OPCODE_DDY
] = { FETCH_OP_GET_GRADIENTS_V
, tgsi_tex
},
9031 [TGSI_OPCODE_KILL
] = { ALU_OP2_KILLGT
, tgsi_kill
}, /* unconditional kill */
9032 [TGSI_OPCODE_PK2H
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9033 [TGSI_OPCODE_PK2US
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9034 [TGSI_OPCODE_PK4B
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9035 [TGSI_OPCODE_PK4UB
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9036 [44] = { ALU_OP0_NOP
, tgsi_unsupported
},
9037 [TGSI_OPCODE_SEQ
] = { ALU_OP2_SETE
, tgsi_op2
},
9038 [46] = { ALU_OP0_NOP
, tgsi_unsupported
},
9039 [TGSI_OPCODE_SGT
] = { ALU_OP2_SETGT
, tgsi_op2
},
9040 [TGSI_OPCODE_SIN
] = { ALU_OP1_SIN
, tgsi_trig
},
9041 [TGSI_OPCODE_SLE
] = { ALU_OP2_SETGE
, tgsi_op2_swap
},
9042 [TGSI_OPCODE_SNE
] = { ALU_OP2_SETNE
, tgsi_op2
},
9043 [51] = { ALU_OP0_NOP
, tgsi_unsupported
},
9044 [TGSI_OPCODE_TEX
] = { FETCH_OP_SAMPLE
, tgsi_tex
},
9045 [TGSI_OPCODE_TXD
] = { FETCH_OP_SAMPLE_G
, tgsi_tex
},
9046 [TGSI_OPCODE_TXP
] = { FETCH_OP_SAMPLE
, tgsi_tex
},
9047 [TGSI_OPCODE_UP2H
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9048 [TGSI_OPCODE_UP2US
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9049 [TGSI_OPCODE_UP4B
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9050 [TGSI_OPCODE_UP4UB
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9051 [59] = { ALU_OP0_NOP
, tgsi_unsupported
},
9052 [60] = { ALU_OP0_NOP
, tgsi_unsupported
},
9053 [TGSI_OPCODE_ARR
] = { ALU_OP0_NOP
, tgsi_r600_arl
},
9054 [62] = { ALU_OP0_NOP
, tgsi_unsupported
},
9055 [TGSI_OPCODE_CAL
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9056 [TGSI_OPCODE_RET
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9057 [TGSI_OPCODE_SSG
] = { ALU_OP0_NOP
, tgsi_ssg
},
9058 [TGSI_OPCODE_CMP
] = { ALU_OP0_NOP
, tgsi_cmp
},
9059 [TGSI_OPCODE_SCS
] = { ALU_OP0_NOP
, tgsi_scs
},
9060 [TGSI_OPCODE_TXB
] = { FETCH_OP_SAMPLE_LB
, tgsi_tex
},
9061 [69] = { ALU_OP0_NOP
, tgsi_unsupported
},
9062 [TGSI_OPCODE_DIV
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9063 [TGSI_OPCODE_DP2
] = { ALU_OP2_DOT4
, tgsi_dp
},
9064 [TGSI_OPCODE_TXL
] = { FETCH_OP_SAMPLE_L
, tgsi_tex
},
9065 [TGSI_OPCODE_BRK
] = { CF_OP_LOOP_BREAK
, tgsi_loop_brk_cont
},
9066 [TGSI_OPCODE_IF
] = { ALU_OP0_NOP
, tgsi_if
},
9067 [TGSI_OPCODE_UIF
] = { ALU_OP0_NOP
, tgsi_uif
},
9068 [76] = { ALU_OP0_NOP
, tgsi_unsupported
},
9069 [TGSI_OPCODE_ELSE
] = { ALU_OP0_NOP
, tgsi_else
},
9070 [TGSI_OPCODE_ENDIF
] = { ALU_OP0_NOP
, tgsi_endif
},
9071 [TGSI_OPCODE_DDX_FINE
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9072 [TGSI_OPCODE_DDY_FINE
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9073 [TGSI_OPCODE_PUSHA
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9074 [TGSI_OPCODE_POPA
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9075 [TGSI_OPCODE_CEIL
] = { ALU_OP1_CEIL
, tgsi_op2
},
9076 [TGSI_OPCODE_I2F
] = { ALU_OP1_INT_TO_FLT
, tgsi_op2_trans
},
9077 [TGSI_OPCODE_NOT
] = { ALU_OP1_NOT_INT
, tgsi_op2
},
9078 [TGSI_OPCODE_TRUNC
] = { ALU_OP1_TRUNC
, tgsi_op2
},
9079 [TGSI_OPCODE_SHL
] = { ALU_OP2_LSHL_INT
, tgsi_op2_trans
},
9080 [88] = { ALU_OP0_NOP
, tgsi_unsupported
},
9081 [TGSI_OPCODE_AND
] = { ALU_OP2_AND_INT
, tgsi_op2
},
9082 [TGSI_OPCODE_OR
] = { ALU_OP2_OR_INT
, tgsi_op2
},
9083 [TGSI_OPCODE_MOD
] = { ALU_OP0_NOP
, tgsi_imod
},
9084 [TGSI_OPCODE_XOR
] = { ALU_OP2_XOR_INT
, tgsi_op2
},
9085 [TGSI_OPCODE_SAD
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9086 [TGSI_OPCODE_TXF
] = { FETCH_OP_LD
, tgsi_tex
},
9087 [TGSI_OPCODE_TXQ
] = { FETCH_OP_GET_TEXTURE_RESINFO
, tgsi_tex
},
9088 [TGSI_OPCODE_CONT
] = { CF_OP_LOOP_CONTINUE
, tgsi_loop_brk_cont
},
9089 [TGSI_OPCODE_EMIT
] = { CF_OP_EMIT_VERTEX
, tgsi_gs_emit
},
9090 [TGSI_OPCODE_ENDPRIM
] = { CF_OP_CUT_VERTEX
, tgsi_gs_emit
},
9091 [TGSI_OPCODE_BGNLOOP
] = { ALU_OP0_NOP
, tgsi_bgnloop
},
9092 [TGSI_OPCODE_BGNSUB
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9093 [TGSI_OPCODE_ENDLOOP
] = { ALU_OP0_NOP
, tgsi_endloop
},
9094 [TGSI_OPCODE_ENDSUB
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9095 [TGSI_OPCODE_TXQ_LZ
] = { FETCH_OP_GET_TEXTURE_RESINFO
, tgsi_tex
},
9096 [TGSI_OPCODE_TXQS
] = { FETCH_OP_GET_NUMBER_OF_SAMPLES
, tgsi_tex
},
9097 [TGSI_OPCODE_RESQ
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9098 [106] = { ALU_OP0_NOP
, tgsi_unsupported
},
9099 [TGSI_OPCODE_NOP
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9100 [TGSI_OPCODE_FSEQ
] = { ALU_OP2_SETE_DX10
, tgsi_op2
},
9101 [TGSI_OPCODE_FSGE
] = { ALU_OP2_SETGE_DX10
, tgsi_op2
},
9102 [TGSI_OPCODE_FSLT
] = { ALU_OP2_SETGT_DX10
, tgsi_op2_swap
},
9103 [TGSI_OPCODE_FSNE
] = { ALU_OP2_SETNE_DX10
, tgsi_op2_swap
},
9104 [TGSI_OPCODE_MEMBAR
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9105 [TGSI_OPCODE_CALLNZ
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9106 [114] = { ALU_OP0_NOP
, tgsi_unsupported
},
9107 [TGSI_OPCODE_BREAKC
] = { ALU_OP0_NOP
, tgsi_loop_breakc
},
9108 [TGSI_OPCODE_KILL_IF
] = { ALU_OP2_KILLGT
, tgsi_kill
}, /* conditional kill */
9109 [TGSI_OPCODE_END
] = { ALU_OP0_NOP
, tgsi_end
}, /* aka HALT */
9110 [TGSI_OPCODE_DFMA
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9111 [TGSI_OPCODE_F2I
] = { ALU_OP1_FLT_TO_INT
, tgsi_op2_trans
},
9112 [TGSI_OPCODE_IDIV
] = { ALU_OP0_NOP
, tgsi_idiv
},
9113 [TGSI_OPCODE_IMAX
] = { ALU_OP2_MAX_INT
, tgsi_op2
},
9114 [TGSI_OPCODE_IMIN
] = { ALU_OP2_MIN_INT
, tgsi_op2
},
9115 [TGSI_OPCODE_INEG
] = { ALU_OP2_SUB_INT
, tgsi_ineg
},
9116 [TGSI_OPCODE_ISGE
] = { ALU_OP2_SETGE_INT
, tgsi_op2
},
9117 [TGSI_OPCODE_ISHR
] = { ALU_OP2_ASHR_INT
, tgsi_op2_trans
},
9118 [TGSI_OPCODE_ISLT
] = { ALU_OP2_SETGT_INT
, tgsi_op2_swap
},
9119 [TGSI_OPCODE_F2U
] = { ALU_OP1_FLT_TO_UINT
, tgsi_op2_trans
},
9120 [TGSI_OPCODE_U2F
] = { ALU_OP1_UINT_TO_FLT
, tgsi_op2_trans
},
9121 [TGSI_OPCODE_UADD
] = { ALU_OP2_ADD_INT
, tgsi_op2
},
9122 [TGSI_OPCODE_UDIV
] = { ALU_OP0_NOP
, tgsi_udiv
},
9123 [TGSI_OPCODE_UMAD
] = { ALU_OP0_NOP
, tgsi_umad
},
9124 [TGSI_OPCODE_UMAX
] = { ALU_OP2_MAX_UINT
, tgsi_op2
},
9125 [TGSI_OPCODE_UMIN
] = { ALU_OP2_MIN_UINT
, tgsi_op2
},
9126 [TGSI_OPCODE_UMOD
] = { ALU_OP0_NOP
, tgsi_umod
},
9127 [TGSI_OPCODE_UMUL
] = { ALU_OP2_MULLO_UINT
, tgsi_op2_trans
},
9128 [TGSI_OPCODE_USEQ
] = { ALU_OP2_SETE_INT
, tgsi_op2
},
9129 [TGSI_OPCODE_USGE
] = { ALU_OP2_SETGE_UINT
, tgsi_op2
},
9130 [TGSI_OPCODE_USHR
] = { ALU_OP2_LSHR_INT
, tgsi_op2_trans
},
9131 [TGSI_OPCODE_USLT
] = { ALU_OP2_SETGT_UINT
, tgsi_op2_swap
},
9132 [TGSI_OPCODE_USNE
] = { ALU_OP2_SETNE_INT
, tgsi_op2_swap
},
9133 [TGSI_OPCODE_SWITCH
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9134 [TGSI_OPCODE_CASE
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9135 [TGSI_OPCODE_DEFAULT
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9136 [TGSI_OPCODE_ENDSWITCH
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9137 [TGSI_OPCODE_SAMPLE
] = { 0, tgsi_unsupported
},
9138 [TGSI_OPCODE_SAMPLE_I
] = { 0, tgsi_unsupported
},
9139 [TGSI_OPCODE_SAMPLE_I_MS
] = { 0, tgsi_unsupported
},
9140 [TGSI_OPCODE_SAMPLE_B
] = { 0, tgsi_unsupported
},
9141 [TGSI_OPCODE_SAMPLE_C
] = { 0, tgsi_unsupported
},
9142 [TGSI_OPCODE_SAMPLE_C_LZ
] = { 0, tgsi_unsupported
},
9143 [TGSI_OPCODE_SAMPLE_D
] = { 0, tgsi_unsupported
},
9144 [TGSI_OPCODE_SAMPLE_L
] = { 0, tgsi_unsupported
},
9145 [TGSI_OPCODE_GATHER4
] = { 0, tgsi_unsupported
},
9146 [TGSI_OPCODE_SVIEWINFO
] = { 0, tgsi_unsupported
},
9147 [TGSI_OPCODE_SAMPLE_POS
] = { 0, tgsi_unsupported
},
9148 [TGSI_OPCODE_SAMPLE_INFO
] = { 0, tgsi_unsupported
},
9149 [TGSI_OPCODE_UARL
] = { ALU_OP1_MOVA_INT
, tgsi_r600_arl
},
9150 [TGSI_OPCODE_UCMP
] = { ALU_OP0_NOP
, tgsi_ucmp
},
9151 [TGSI_OPCODE_IABS
] = { 0, tgsi_iabs
},
9152 [TGSI_OPCODE_ISSG
] = { 0, tgsi_issg
},
9153 [TGSI_OPCODE_LOAD
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9154 [TGSI_OPCODE_STORE
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9155 [TGSI_OPCODE_MFENCE
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9156 [TGSI_OPCODE_LFENCE
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9157 [TGSI_OPCODE_SFENCE
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9158 [TGSI_OPCODE_BARRIER
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9159 [TGSI_OPCODE_ATOMUADD
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9160 [TGSI_OPCODE_ATOMXCHG
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9161 [TGSI_OPCODE_ATOMCAS
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9162 [TGSI_OPCODE_ATOMAND
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9163 [TGSI_OPCODE_ATOMOR
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9164 [TGSI_OPCODE_ATOMXOR
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9165 [TGSI_OPCODE_ATOMUMIN
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9166 [TGSI_OPCODE_ATOMUMAX
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9167 [TGSI_OPCODE_ATOMIMIN
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9168 [TGSI_OPCODE_ATOMIMAX
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9169 [TGSI_OPCODE_TEX2
] = { FETCH_OP_SAMPLE
, tgsi_tex
},
9170 [TGSI_OPCODE_TXB2
] = { FETCH_OP_SAMPLE_LB
, tgsi_tex
},
9171 [TGSI_OPCODE_TXL2
] = { FETCH_OP_SAMPLE_L
, tgsi_tex
},
9172 [TGSI_OPCODE_IMUL_HI
] = { ALU_OP2_MULHI_INT
, tgsi_op2_trans
},
9173 [TGSI_OPCODE_UMUL_HI
] = { ALU_OP2_MULHI_UINT
, tgsi_op2_trans
},
9174 [TGSI_OPCODE_TG4
] = { FETCH_OP_GATHER4
, tgsi_unsupported
},
9175 [TGSI_OPCODE_LODQ
] = { FETCH_OP_GET_LOD
, tgsi_unsupported
},
9176 [TGSI_OPCODE_IBFE
] = { ALU_OP3_BFE_INT
, tgsi_unsupported
},
9177 [TGSI_OPCODE_UBFE
] = { ALU_OP3_BFE_UINT
, tgsi_unsupported
},
9178 [TGSI_OPCODE_BFI
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9179 [TGSI_OPCODE_BREV
] = { ALU_OP1_BFREV_INT
, tgsi_unsupported
},
9180 [TGSI_OPCODE_POPC
] = { ALU_OP1_BCNT_INT
, tgsi_unsupported
},
9181 [TGSI_OPCODE_LSB
] = { ALU_OP1_FFBL_INT
, tgsi_unsupported
},
9182 [TGSI_OPCODE_IMSB
] = { ALU_OP1_FFBH_INT
, tgsi_unsupported
},
9183 [TGSI_OPCODE_UMSB
] = { ALU_OP1_FFBH_UINT
, tgsi_unsupported
},
9184 [TGSI_OPCODE_INTERP_CENTROID
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9185 [TGSI_OPCODE_INTERP_SAMPLE
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9186 [TGSI_OPCODE_INTERP_OFFSET
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9187 [TGSI_OPCODE_LAST
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9190 static const struct r600_shader_tgsi_instruction eg_shader_tgsi_instruction
[] = {
9191 [TGSI_OPCODE_ARL
] = { ALU_OP0_NOP
, tgsi_eg_arl
},
9192 [TGSI_OPCODE_MOV
] = { ALU_OP1_MOV
, tgsi_op2
},
9193 [TGSI_OPCODE_LIT
] = { ALU_OP0_NOP
, tgsi_lit
},
9194 [TGSI_OPCODE_RCP
] = { ALU_OP1_RECIP_IEEE
, tgsi_trans_srcx_replicate
},
9195 [TGSI_OPCODE_RSQ
] = { ALU_OP1_RECIPSQRT_IEEE
, tgsi_rsq
},
9196 [TGSI_OPCODE_EXP
] = { ALU_OP0_NOP
, tgsi_exp
},
9197 [TGSI_OPCODE_LOG
] = { ALU_OP0_NOP
, tgsi_log
},
9198 [TGSI_OPCODE_MUL
] = { ALU_OP2_MUL
, tgsi_op2
},
9199 [TGSI_OPCODE_ADD
] = { ALU_OP2_ADD
, tgsi_op2
},
9200 [TGSI_OPCODE_DP3
] = { ALU_OP2_DOT4
, tgsi_dp
},
9201 [TGSI_OPCODE_DP4
] = { ALU_OP2_DOT4
, tgsi_dp
},
9202 [TGSI_OPCODE_DST
] = { ALU_OP0_NOP
, tgsi_opdst
},
9203 [TGSI_OPCODE_MIN
] = { ALU_OP2_MIN
, tgsi_op2
},
9204 [TGSI_OPCODE_MAX
] = { ALU_OP2_MAX
, tgsi_op2
},
9205 [TGSI_OPCODE_SLT
] = { ALU_OP2_SETGT
, tgsi_op2_swap
},
9206 [TGSI_OPCODE_SGE
] = { ALU_OP2_SETGE
, tgsi_op2
},
9207 [TGSI_OPCODE_MAD
] = { ALU_OP3_MULADD
, tgsi_op3
},
9208 [TGSI_OPCODE_SUB
] = { ALU_OP2_ADD
, tgsi_op2
},
9209 [TGSI_OPCODE_LRP
] = { ALU_OP0_NOP
, tgsi_lrp
},
9210 [TGSI_OPCODE_FMA
] = { ALU_OP3_FMA
, tgsi_op3
},
9211 [TGSI_OPCODE_SQRT
] = { ALU_OP1_SQRT_IEEE
, tgsi_trans_srcx_replicate
},
9212 [TGSI_OPCODE_DP2A
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9213 [22] = { ALU_OP0_NOP
, tgsi_unsupported
},
9214 [23] = { ALU_OP0_NOP
, tgsi_unsupported
},
9215 [TGSI_OPCODE_FRC
] = { ALU_OP1_FRACT
, tgsi_op2
},
9216 [TGSI_OPCODE_CLAMP
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9217 [TGSI_OPCODE_FLR
] = { ALU_OP1_FLOOR
, tgsi_op2
},
9218 [TGSI_OPCODE_ROUND
] = { ALU_OP1_RNDNE
, tgsi_op2
},
9219 [TGSI_OPCODE_EX2
] = { ALU_OP1_EXP_IEEE
, tgsi_trans_srcx_replicate
},
9220 [TGSI_OPCODE_LG2
] = { ALU_OP1_LOG_IEEE
, tgsi_trans_srcx_replicate
},
9221 [TGSI_OPCODE_POW
] = { ALU_OP0_NOP
, tgsi_pow
},
9222 [TGSI_OPCODE_XPD
] = { ALU_OP0_NOP
, tgsi_xpd
},
9223 [32] = { ALU_OP0_NOP
, tgsi_unsupported
},
9224 [33] = { ALU_OP0_NOP
, tgsi_unsupported
},
9225 [34] = { ALU_OP0_NOP
, tgsi_unsupported
},
9226 [TGSI_OPCODE_DPH
] = { ALU_OP2_DOT4
, tgsi_dp
},
9227 [TGSI_OPCODE_COS
] = { ALU_OP1_COS
, tgsi_trig
},
9228 [TGSI_OPCODE_DDX
] = { FETCH_OP_GET_GRADIENTS_H
, tgsi_tex
},
9229 [TGSI_OPCODE_DDY
] = { FETCH_OP_GET_GRADIENTS_V
, tgsi_tex
},
9230 [TGSI_OPCODE_KILL
] = { ALU_OP2_KILLGT
, tgsi_kill
}, /* unconditional kill */
9231 [TGSI_OPCODE_PK2H
] = { ALU_OP0_NOP
, tgsi_pk2h
},
9232 [TGSI_OPCODE_PK2US
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9233 [TGSI_OPCODE_PK4B
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9234 [TGSI_OPCODE_PK4UB
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9235 [44] = { ALU_OP0_NOP
, tgsi_unsupported
},
9236 [TGSI_OPCODE_SEQ
] = { ALU_OP2_SETE
, tgsi_op2
},
9237 [46] = { ALU_OP0_NOP
, tgsi_unsupported
},
9238 [TGSI_OPCODE_SGT
] = { ALU_OP2_SETGT
, tgsi_op2
},
9239 [TGSI_OPCODE_SIN
] = { ALU_OP1_SIN
, tgsi_trig
},
9240 [TGSI_OPCODE_SLE
] = { ALU_OP2_SETGE
, tgsi_op2_swap
},
9241 [TGSI_OPCODE_SNE
] = { ALU_OP2_SETNE
, tgsi_op2
},
9242 [51] = { ALU_OP0_NOP
, tgsi_unsupported
},
9243 [TGSI_OPCODE_TEX
] = { FETCH_OP_SAMPLE
, tgsi_tex
},
9244 [TGSI_OPCODE_TXD
] = { FETCH_OP_SAMPLE_G
, tgsi_tex
},
9245 [TGSI_OPCODE_TXP
] = { FETCH_OP_SAMPLE
, tgsi_tex
},
9246 [TGSI_OPCODE_UP2H
] = { ALU_OP0_NOP
, tgsi_up2h
},
9247 [TGSI_OPCODE_UP2US
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9248 [TGSI_OPCODE_UP4B
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9249 [TGSI_OPCODE_UP4UB
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9250 [59] = { ALU_OP0_NOP
, tgsi_unsupported
},
9251 [60] = { ALU_OP0_NOP
, tgsi_unsupported
},
9252 [TGSI_OPCODE_ARR
] = { ALU_OP0_NOP
, tgsi_eg_arl
},
9253 [62] = { ALU_OP0_NOP
, tgsi_unsupported
},
9254 [TGSI_OPCODE_CAL
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9255 [TGSI_OPCODE_RET
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9256 [TGSI_OPCODE_SSG
] = { ALU_OP0_NOP
, tgsi_ssg
},
9257 [TGSI_OPCODE_CMP
] = { ALU_OP0_NOP
, tgsi_cmp
},
9258 [TGSI_OPCODE_SCS
] = { ALU_OP0_NOP
, tgsi_scs
},
9259 [TGSI_OPCODE_TXB
] = { FETCH_OP_SAMPLE_LB
, tgsi_tex
},
9260 [69] = { ALU_OP0_NOP
, tgsi_unsupported
},
9261 [TGSI_OPCODE_DIV
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9262 [TGSI_OPCODE_DP2
] = { ALU_OP2_DOT4
, tgsi_dp
},
9263 [TGSI_OPCODE_TXL
] = { FETCH_OP_SAMPLE_L
, tgsi_tex
},
9264 [TGSI_OPCODE_BRK
] = { CF_OP_LOOP_BREAK
, tgsi_loop_brk_cont
},
9265 [TGSI_OPCODE_IF
] = { ALU_OP0_NOP
, tgsi_if
},
9266 [TGSI_OPCODE_UIF
] = { ALU_OP0_NOP
, tgsi_uif
},
9267 [76] = { ALU_OP0_NOP
, tgsi_unsupported
},
9268 [TGSI_OPCODE_ELSE
] = { ALU_OP0_NOP
, tgsi_else
},
9269 [TGSI_OPCODE_ENDIF
] = { ALU_OP0_NOP
, tgsi_endif
},
9270 [TGSI_OPCODE_DDX_FINE
] = { FETCH_OP_GET_GRADIENTS_H
, tgsi_tex
},
9271 [TGSI_OPCODE_DDY_FINE
] = { FETCH_OP_GET_GRADIENTS_V
, tgsi_tex
},
9272 [TGSI_OPCODE_PUSHA
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9273 [TGSI_OPCODE_POPA
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9274 [TGSI_OPCODE_CEIL
] = { ALU_OP1_CEIL
, tgsi_op2
},
9275 [TGSI_OPCODE_I2F
] = { ALU_OP1_INT_TO_FLT
, tgsi_op2_trans
},
9276 [TGSI_OPCODE_NOT
] = { ALU_OP1_NOT_INT
, tgsi_op2
},
9277 [TGSI_OPCODE_TRUNC
] = { ALU_OP1_TRUNC
, tgsi_op2
},
9278 [TGSI_OPCODE_SHL
] = { ALU_OP2_LSHL_INT
, tgsi_op2
},
9279 [88] = { ALU_OP0_NOP
, tgsi_unsupported
},
9280 [TGSI_OPCODE_AND
] = { ALU_OP2_AND_INT
, tgsi_op2
},
9281 [TGSI_OPCODE_OR
] = { ALU_OP2_OR_INT
, tgsi_op2
},
9282 [TGSI_OPCODE_MOD
] = { ALU_OP0_NOP
, tgsi_imod
},
9283 [TGSI_OPCODE_XOR
] = { ALU_OP2_XOR_INT
, tgsi_op2
},
9284 [TGSI_OPCODE_SAD
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9285 [TGSI_OPCODE_TXF
] = { FETCH_OP_LD
, tgsi_tex
},
9286 [TGSI_OPCODE_TXQ
] = { FETCH_OP_GET_TEXTURE_RESINFO
, tgsi_tex
},
9287 [TGSI_OPCODE_CONT
] = { CF_OP_LOOP_CONTINUE
, tgsi_loop_brk_cont
},
9288 [TGSI_OPCODE_EMIT
] = { CF_OP_EMIT_VERTEX
, tgsi_gs_emit
},
9289 [TGSI_OPCODE_ENDPRIM
] = { CF_OP_CUT_VERTEX
, tgsi_gs_emit
},
9290 [TGSI_OPCODE_BGNLOOP
] = { ALU_OP0_NOP
, tgsi_bgnloop
},
9291 [TGSI_OPCODE_BGNSUB
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9292 [TGSI_OPCODE_ENDLOOP
] = { ALU_OP0_NOP
, tgsi_endloop
},
9293 [TGSI_OPCODE_ENDSUB
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9294 [TGSI_OPCODE_TXQ_LZ
] = { FETCH_OP_GET_TEXTURE_RESINFO
, tgsi_tex
},
9295 [TGSI_OPCODE_TXQS
] = { FETCH_OP_GET_NUMBER_OF_SAMPLES
, tgsi_tex
},
9296 [TGSI_OPCODE_RESQ
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9297 [106] = { ALU_OP0_NOP
, tgsi_unsupported
},
9298 [TGSI_OPCODE_NOP
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9299 [TGSI_OPCODE_FSEQ
] = { ALU_OP2_SETE_DX10
, tgsi_op2
},
9300 [TGSI_OPCODE_FSGE
] = { ALU_OP2_SETGE_DX10
, tgsi_op2
},
9301 [TGSI_OPCODE_FSLT
] = { ALU_OP2_SETGT_DX10
, tgsi_op2_swap
},
9302 [TGSI_OPCODE_FSNE
] = { ALU_OP2_SETNE_DX10
, tgsi_op2_swap
},
9303 [TGSI_OPCODE_MEMBAR
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9304 [TGSI_OPCODE_CALLNZ
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9305 [114] = { ALU_OP0_NOP
, tgsi_unsupported
},
9306 [TGSI_OPCODE_BREAKC
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9307 [TGSI_OPCODE_KILL_IF
] = { ALU_OP2_KILLGT
, tgsi_kill
}, /* conditional kill */
9308 [TGSI_OPCODE_END
] = { ALU_OP0_NOP
, tgsi_end
}, /* aka HALT */
9309 /* Refer below for TGSI_OPCODE_DFMA */
9310 [TGSI_OPCODE_F2I
] = { ALU_OP1_FLT_TO_INT
, tgsi_f2i
},
9311 [TGSI_OPCODE_IDIV
] = { ALU_OP0_NOP
, tgsi_idiv
},
9312 [TGSI_OPCODE_IMAX
] = { ALU_OP2_MAX_INT
, tgsi_op2
},
9313 [TGSI_OPCODE_IMIN
] = { ALU_OP2_MIN_INT
, tgsi_op2
},
9314 [TGSI_OPCODE_INEG
] = { ALU_OP2_SUB_INT
, tgsi_ineg
},
9315 [TGSI_OPCODE_ISGE
] = { ALU_OP2_SETGE_INT
, tgsi_op2
},
9316 [TGSI_OPCODE_ISHR
] = { ALU_OP2_ASHR_INT
, tgsi_op2
},
9317 [TGSI_OPCODE_ISLT
] = { ALU_OP2_SETGT_INT
, tgsi_op2_swap
},
9318 [TGSI_OPCODE_F2U
] = { ALU_OP1_FLT_TO_UINT
, tgsi_f2i
},
9319 [TGSI_OPCODE_U2F
] = { ALU_OP1_UINT_TO_FLT
, tgsi_op2_trans
},
9320 [TGSI_OPCODE_UADD
] = { ALU_OP2_ADD_INT
, tgsi_op2
},
9321 [TGSI_OPCODE_UDIV
] = { ALU_OP0_NOP
, tgsi_udiv
},
9322 [TGSI_OPCODE_UMAD
] = { ALU_OP0_NOP
, tgsi_umad
},
9323 [TGSI_OPCODE_UMAX
] = { ALU_OP2_MAX_UINT
, tgsi_op2
},
9324 [TGSI_OPCODE_UMIN
] = { ALU_OP2_MIN_UINT
, tgsi_op2
},
9325 [TGSI_OPCODE_UMOD
] = { ALU_OP0_NOP
, tgsi_umod
},
9326 [TGSI_OPCODE_UMUL
] = { ALU_OP2_MULLO_UINT
, tgsi_op2_trans
},
9327 [TGSI_OPCODE_USEQ
] = { ALU_OP2_SETE_INT
, tgsi_op2
},
9328 [TGSI_OPCODE_USGE
] = { ALU_OP2_SETGE_UINT
, tgsi_op2
},
9329 [TGSI_OPCODE_USHR
] = { ALU_OP2_LSHR_INT
, tgsi_op2
},
9330 [TGSI_OPCODE_USLT
] = { ALU_OP2_SETGT_UINT
, tgsi_op2_swap
},
9331 [TGSI_OPCODE_USNE
] = { ALU_OP2_SETNE_INT
, tgsi_op2
},
9332 [TGSI_OPCODE_SWITCH
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9333 [TGSI_OPCODE_CASE
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9334 [TGSI_OPCODE_DEFAULT
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9335 [TGSI_OPCODE_ENDSWITCH
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9336 [TGSI_OPCODE_SAMPLE
] = { 0, tgsi_unsupported
},
9337 [TGSI_OPCODE_SAMPLE_I
] = { 0, tgsi_unsupported
},
9338 [TGSI_OPCODE_SAMPLE_I_MS
] = { 0, tgsi_unsupported
},
9339 [TGSI_OPCODE_SAMPLE_B
] = { 0, tgsi_unsupported
},
9340 [TGSI_OPCODE_SAMPLE_C
] = { 0, tgsi_unsupported
},
9341 [TGSI_OPCODE_SAMPLE_C_LZ
] = { 0, tgsi_unsupported
},
9342 [TGSI_OPCODE_SAMPLE_D
] = { 0, tgsi_unsupported
},
9343 [TGSI_OPCODE_SAMPLE_L
] = { 0, tgsi_unsupported
},
9344 [TGSI_OPCODE_GATHER4
] = { 0, tgsi_unsupported
},
9345 [TGSI_OPCODE_SVIEWINFO
] = { 0, tgsi_unsupported
},
9346 [TGSI_OPCODE_SAMPLE_POS
] = { 0, tgsi_unsupported
},
9347 [TGSI_OPCODE_SAMPLE_INFO
] = { 0, tgsi_unsupported
},
9348 [TGSI_OPCODE_UARL
] = { ALU_OP1_MOVA_INT
, tgsi_eg_arl
},
9349 [TGSI_OPCODE_UCMP
] = { ALU_OP0_NOP
, tgsi_ucmp
},
9350 [TGSI_OPCODE_IABS
] = { 0, tgsi_iabs
},
9351 [TGSI_OPCODE_ISSG
] = { 0, tgsi_issg
},
9352 [TGSI_OPCODE_LOAD
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9353 [TGSI_OPCODE_STORE
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9354 [TGSI_OPCODE_MFENCE
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9355 [TGSI_OPCODE_LFENCE
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9356 [TGSI_OPCODE_SFENCE
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9357 [TGSI_OPCODE_BARRIER
] = { ALU_OP0_GROUP_BARRIER
, tgsi_barrier
},
9358 [TGSI_OPCODE_ATOMUADD
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9359 [TGSI_OPCODE_ATOMXCHG
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9360 [TGSI_OPCODE_ATOMCAS
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9361 [TGSI_OPCODE_ATOMAND
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9362 [TGSI_OPCODE_ATOMOR
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9363 [TGSI_OPCODE_ATOMXOR
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9364 [TGSI_OPCODE_ATOMUMIN
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9365 [TGSI_OPCODE_ATOMUMAX
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9366 [TGSI_OPCODE_ATOMIMIN
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9367 [TGSI_OPCODE_ATOMIMAX
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9368 [TGSI_OPCODE_TEX2
] = { FETCH_OP_SAMPLE
, tgsi_tex
},
9369 [TGSI_OPCODE_TXB2
] = { FETCH_OP_SAMPLE_LB
, tgsi_tex
},
9370 [TGSI_OPCODE_TXL2
] = { FETCH_OP_SAMPLE_L
, tgsi_tex
},
9371 [TGSI_OPCODE_IMUL_HI
] = { ALU_OP2_MULHI_INT
, tgsi_op2_trans
},
9372 [TGSI_OPCODE_UMUL_HI
] = { ALU_OP2_MULHI_UINT
, tgsi_op2_trans
},
9373 [TGSI_OPCODE_TG4
] = { FETCH_OP_GATHER4
, tgsi_tex
},
9374 [TGSI_OPCODE_LODQ
] = { FETCH_OP_GET_LOD
, tgsi_tex
},
9375 [TGSI_OPCODE_IBFE
] = { ALU_OP3_BFE_INT
, tgsi_op3
},
9376 [TGSI_OPCODE_UBFE
] = { ALU_OP3_BFE_UINT
, tgsi_op3
},
9377 [TGSI_OPCODE_BFI
] = { ALU_OP0_NOP
, tgsi_bfi
},
9378 [TGSI_OPCODE_BREV
] = { ALU_OP1_BFREV_INT
, tgsi_op2
},
9379 [TGSI_OPCODE_POPC
] = { ALU_OP1_BCNT_INT
, tgsi_op2
},
9380 [TGSI_OPCODE_LSB
] = { ALU_OP1_FFBL_INT
, tgsi_op2
},
9381 [TGSI_OPCODE_IMSB
] = { ALU_OP1_FFBH_INT
, tgsi_msb
},
9382 [TGSI_OPCODE_UMSB
] = { ALU_OP1_FFBH_UINT
, tgsi_msb
},
9383 [TGSI_OPCODE_INTERP_CENTROID
] = { ALU_OP0_NOP
, tgsi_interp_egcm
},
9384 [TGSI_OPCODE_INTERP_SAMPLE
] = { ALU_OP0_NOP
, tgsi_interp_egcm
},
9385 [TGSI_OPCODE_INTERP_OFFSET
] = { ALU_OP0_NOP
, tgsi_interp_egcm
},
9386 [TGSI_OPCODE_F2D
] = { ALU_OP1_FLT32_TO_FLT64
, tgsi_op2_64
},
9387 [TGSI_OPCODE_D2F
] = { ALU_OP1_FLT64_TO_FLT32
, tgsi_op2_64_single_dest
},
9388 [TGSI_OPCODE_DABS
] = { ALU_OP1_MOV
, tgsi_op2_64
},
9389 [TGSI_OPCODE_DNEG
] = { ALU_OP2_ADD_64
, tgsi_dneg
},
9390 [TGSI_OPCODE_DADD
] = { ALU_OP2_ADD_64
, tgsi_op2_64
},
9391 [TGSI_OPCODE_DMUL
] = { ALU_OP2_MUL_64
, cayman_mul_double_instr
},
9392 [TGSI_OPCODE_DMAX
] = { ALU_OP2_MAX_64
, tgsi_op2_64
},
9393 [TGSI_OPCODE_DMIN
] = { ALU_OP2_MIN_64
, tgsi_op2_64
},
9394 [TGSI_OPCODE_DSLT
] = { ALU_OP2_SETGT_64
, tgsi_op2_64_single_dest_s
},
9395 [TGSI_OPCODE_DSGE
] = { ALU_OP2_SETGE_64
, tgsi_op2_64_single_dest
},
9396 [TGSI_OPCODE_DSEQ
] = { ALU_OP2_SETE_64
, tgsi_op2_64_single_dest
},
9397 [TGSI_OPCODE_DSNE
] = { ALU_OP2_SETNE_64
, tgsi_op2_64_single_dest
},
9398 [TGSI_OPCODE_DRCP
] = { ALU_OP2_RECIP_64
, cayman_emit_double_instr
},
9399 [TGSI_OPCODE_DSQRT
] = { ALU_OP2_SQRT_64
, cayman_emit_double_instr
},
9400 [TGSI_OPCODE_DMAD
] = { ALU_OP3_FMA_64
, tgsi_op3_64
},
9401 [TGSI_OPCODE_DFMA
] = { ALU_OP3_FMA_64
, tgsi_op3_64
},
9402 [TGSI_OPCODE_DFRAC
] = { ALU_OP1_FRACT_64
, tgsi_op2_64
},
9403 [TGSI_OPCODE_DLDEXP
] = { ALU_OP2_LDEXP_64
, tgsi_op2_64
},
9404 [TGSI_OPCODE_DFRACEXP
] = { ALU_OP1_FREXP_64
, tgsi_dfracexp
},
9405 [TGSI_OPCODE_D2I
] = { ALU_OP1_FLT_TO_INT
, egcm_double_to_int
},
9406 [TGSI_OPCODE_I2D
] = { ALU_OP1_INT_TO_FLT
, egcm_int_to_double
},
9407 [TGSI_OPCODE_D2U
] = { ALU_OP1_FLT_TO_UINT
, egcm_double_to_int
},
9408 [TGSI_OPCODE_U2D
] = { ALU_OP1_UINT_TO_FLT
, egcm_int_to_double
},
9409 [TGSI_OPCODE_DRSQ
] = { ALU_OP2_RECIPSQRT_64
, cayman_emit_double_instr
},
9410 [TGSI_OPCODE_LAST
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9413 static const struct r600_shader_tgsi_instruction cm_shader_tgsi_instruction
[] = {
9414 [TGSI_OPCODE_ARL
] = { ALU_OP0_NOP
, tgsi_eg_arl
},
9415 [TGSI_OPCODE_MOV
] = { ALU_OP1_MOV
, tgsi_op2
},
9416 [TGSI_OPCODE_LIT
] = { ALU_OP0_NOP
, tgsi_lit
},
9417 [TGSI_OPCODE_RCP
] = { ALU_OP1_RECIP_IEEE
, cayman_emit_float_instr
},
9418 [TGSI_OPCODE_RSQ
] = { ALU_OP1_RECIPSQRT_IEEE
, cayman_emit_float_instr
},
9419 [TGSI_OPCODE_EXP
] = { ALU_OP0_NOP
, tgsi_exp
},
9420 [TGSI_OPCODE_LOG
] = { ALU_OP0_NOP
, tgsi_log
},
9421 [TGSI_OPCODE_MUL
] = { ALU_OP2_MUL
, tgsi_op2
},
9422 [TGSI_OPCODE_ADD
] = { ALU_OP2_ADD
, tgsi_op2
},
9423 [TGSI_OPCODE_DP3
] = { ALU_OP2_DOT4
, tgsi_dp
},
9424 [TGSI_OPCODE_DP4
] = { ALU_OP2_DOT4
, tgsi_dp
},
9425 [TGSI_OPCODE_DST
] = { ALU_OP0_NOP
, tgsi_opdst
},
9426 [TGSI_OPCODE_MIN
] = { ALU_OP2_MIN
, tgsi_op2
},
9427 [TGSI_OPCODE_MAX
] = { ALU_OP2_MAX
, tgsi_op2
},
9428 [TGSI_OPCODE_SLT
] = { ALU_OP2_SETGT
, tgsi_op2_swap
},
9429 [TGSI_OPCODE_SGE
] = { ALU_OP2_SETGE
, tgsi_op2
},
9430 [TGSI_OPCODE_MAD
] = { ALU_OP3_MULADD
, tgsi_op3
},
9431 [TGSI_OPCODE_SUB
] = { ALU_OP2_ADD
, tgsi_op2
},
9432 [TGSI_OPCODE_LRP
] = { ALU_OP0_NOP
, tgsi_lrp
},
9433 [TGSI_OPCODE_FMA
] = { ALU_OP3_FMA
, tgsi_op3
},
9434 [TGSI_OPCODE_SQRT
] = { ALU_OP1_SQRT_IEEE
, cayman_emit_float_instr
},
9435 [TGSI_OPCODE_DP2A
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9436 [22] = { ALU_OP0_NOP
, tgsi_unsupported
},
9437 [23] = { ALU_OP0_NOP
, tgsi_unsupported
},
9438 [TGSI_OPCODE_FRC
] = { ALU_OP1_FRACT
, tgsi_op2
},
9439 [TGSI_OPCODE_CLAMP
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9440 [TGSI_OPCODE_FLR
] = { ALU_OP1_FLOOR
, tgsi_op2
},
9441 [TGSI_OPCODE_ROUND
] = { ALU_OP1_RNDNE
, tgsi_op2
},
9442 [TGSI_OPCODE_EX2
] = { ALU_OP1_EXP_IEEE
, cayman_emit_float_instr
},
9443 [TGSI_OPCODE_LG2
] = { ALU_OP1_LOG_IEEE
, cayman_emit_float_instr
},
9444 [TGSI_OPCODE_POW
] = { ALU_OP0_NOP
, cayman_pow
},
9445 [TGSI_OPCODE_XPD
] = { ALU_OP0_NOP
, tgsi_xpd
},
9446 [32] = { ALU_OP0_NOP
, tgsi_unsupported
},
9447 [33] = { ALU_OP0_NOP
, tgsi_unsupported
},
9448 [34] = { ALU_OP0_NOP
, tgsi_unsupported
},
9449 [TGSI_OPCODE_DPH
] = { ALU_OP2_DOT4
, tgsi_dp
},
9450 [TGSI_OPCODE_COS
] = { ALU_OP1_COS
, cayman_trig
},
9451 [TGSI_OPCODE_DDX
] = { FETCH_OP_GET_GRADIENTS_H
, tgsi_tex
},
9452 [TGSI_OPCODE_DDY
] = { FETCH_OP_GET_GRADIENTS_V
, tgsi_tex
},
9453 [TGSI_OPCODE_KILL
] = { ALU_OP2_KILLGT
, tgsi_kill
}, /* unconditional kill */
9454 [TGSI_OPCODE_PK2H
] = { ALU_OP0_NOP
, tgsi_pk2h
},
9455 [TGSI_OPCODE_PK2US
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9456 [TGSI_OPCODE_PK4B
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9457 [TGSI_OPCODE_PK4UB
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9458 [44] = { ALU_OP0_NOP
, tgsi_unsupported
},
9459 [TGSI_OPCODE_SEQ
] = { ALU_OP2_SETE
, tgsi_op2
},
9460 [46] = { ALU_OP0_NOP
, tgsi_unsupported
},
9461 [TGSI_OPCODE_SGT
] = { ALU_OP2_SETGT
, tgsi_op2
},
9462 [TGSI_OPCODE_SIN
] = { ALU_OP1_SIN
, cayman_trig
},
9463 [TGSI_OPCODE_SLE
] = { ALU_OP2_SETGE
, tgsi_op2_swap
},
9464 [TGSI_OPCODE_SNE
] = { ALU_OP2_SETNE
, tgsi_op2
},
9465 [51] = { ALU_OP0_NOP
, tgsi_unsupported
},
9466 [TGSI_OPCODE_TEX
] = { FETCH_OP_SAMPLE
, tgsi_tex
},
9467 [TGSI_OPCODE_TXD
] = { FETCH_OP_SAMPLE_G
, tgsi_tex
},
9468 [TGSI_OPCODE_TXP
] = { FETCH_OP_SAMPLE
, tgsi_tex
},
9469 [TGSI_OPCODE_UP2H
] = { ALU_OP0_NOP
, tgsi_up2h
},
9470 [TGSI_OPCODE_UP2US
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9471 [TGSI_OPCODE_UP4B
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9472 [TGSI_OPCODE_UP4UB
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9473 [59] = { ALU_OP0_NOP
, tgsi_unsupported
},
9474 [60] = { ALU_OP0_NOP
, tgsi_unsupported
},
9475 [TGSI_OPCODE_ARR
] = { ALU_OP0_NOP
, tgsi_eg_arl
},
9476 [62] = { ALU_OP0_NOP
, tgsi_unsupported
},
9477 [TGSI_OPCODE_CAL
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9478 [TGSI_OPCODE_RET
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9479 [TGSI_OPCODE_SSG
] = { ALU_OP0_NOP
, tgsi_ssg
},
9480 [TGSI_OPCODE_CMP
] = { ALU_OP0_NOP
, tgsi_cmp
},
9481 [TGSI_OPCODE_SCS
] = { ALU_OP0_NOP
, tgsi_scs
},
9482 [TGSI_OPCODE_TXB
] = { FETCH_OP_SAMPLE_LB
, tgsi_tex
},
9483 [69] = { ALU_OP0_NOP
, tgsi_unsupported
},
9484 [TGSI_OPCODE_DIV
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9485 [TGSI_OPCODE_DP2
] = { ALU_OP2_DOT4
, tgsi_dp
},
9486 [TGSI_OPCODE_TXL
] = { FETCH_OP_SAMPLE_L
, tgsi_tex
},
9487 [TGSI_OPCODE_BRK
] = { CF_OP_LOOP_BREAK
, tgsi_loop_brk_cont
},
9488 [TGSI_OPCODE_IF
] = { ALU_OP0_NOP
, tgsi_if
},
9489 [TGSI_OPCODE_UIF
] = { ALU_OP0_NOP
, tgsi_uif
},
9490 [76] = { ALU_OP0_NOP
, tgsi_unsupported
},
9491 [TGSI_OPCODE_ELSE
] = { ALU_OP0_NOP
, tgsi_else
},
9492 [TGSI_OPCODE_ENDIF
] = { ALU_OP0_NOP
, tgsi_endif
},
9493 [TGSI_OPCODE_DDX_FINE
] = { FETCH_OP_GET_GRADIENTS_H
, tgsi_tex
},
9494 [TGSI_OPCODE_DDY_FINE
] = { FETCH_OP_GET_GRADIENTS_V
, tgsi_tex
},
9495 [TGSI_OPCODE_PUSHA
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9496 [TGSI_OPCODE_POPA
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9497 [TGSI_OPCODE_CEIL
] = { ALU_OP1_CEIL
, tgsi_op2
},
9498 [TGSI_OPCODE_I2F
] = { ALU_OP1_INT_TO_FLT
, tgsi_op2
},
9499 [TGSI_OPCODE_NOT
] = { ALU_OP1_NOT_INT
, tgsi_op2
},
9500 [TGSI_OPCODE_TRUNC
] = { ALU_OP1_TRUNC
, tgsi_op2
},
9501 [TGSI_OPCODE_SHL
] = { ALU_OP2_LSHL_INT
, tgsi_op2
},
9502 [88] = { ALU_OP0_NOP
, tgsi_unsupported
},
9503 [TGSI_OPCODE_AND
] = { ALU_OP2_AND_INT
, tgsi_op2
},
9504 [TGSI_OPCODE_OR
] = { ALU_OP2_OR_INT
, tgsi_op2
},
9505 [TGSI_OPCODE_MOD
] = { ALU_OP0_NOP
, tgsi_imod
},
9506 [TGSI_OPCODE_XOR
] = { ALU_OP2_XOR_INT
, tgsi_op2
},
9507 [TGSI_OPCODE_SAD
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9508 [TGSI_OPCODE_TXF
] = { FETCH_OP_LD
, tgsi_tex
},
9509 [TGSI_OPCODE_TXQ
] = { FETCH_OP_GET_TEXTURE_RESINFO
, tgsi_tex
},
9510 [TGSI_OPCODE_CONT
] = { CF_OP_LOOP_CONTINUE
, tgsi_loop_brk_cont
},
9511 [TGSI_OPCODE_EMIT
] = { CF_OP_EMIT_VERTEX
, tgsi_gs_emit
},
9512 [TGSI_OPCODE_ENDPRIM
] = { CF_OP_CUT_VERTEX
, tgsi_gs_emit
},
9513 [TGSI_OPCODE_BGNLOOP
] = { ALU_OP0_NOP
, tgsi_bgnloop
},
9514 [TGSI_OPCODE_BGNSUB
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9515 [TGSI_OPCODE_ENDLOOP
] = { ALU_OP0_NOP
, tgsi_endloop
},
9516 [TGSI_OPCODE_ENDSUB
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9517 [TGSI_OPCODE_TXQ_LZ
] = { FETCH_OP_GET_TEXTURE_RESINFO
, tgsi_tex
},
9518 [TGSI_OPCODE_TXQS
] = { FETCH_OP_GET_NUMBER_OF_SAMPLES
, tgsi_tex
},
9519 [TGSI_OPCODE_RESQ
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9520 [106] = { ALU_OP0_NOP
, tgsi_unsupported
},
9521 [TGSI_OPCODE_NOP
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9522 [TGSI_OPCODE_FSEQ
] = { ALU_OP2_SETE_DX10
, tgsi_op2
},
9523 [TGSI_OPCODE_FSGE
] = { ALU_OP2_SETGE_DX10
, tgsi_op2
},
9524 [TGSI_OPCODE_FSLT
] = { ALU_OP2_SETGT_DX10
, tgsi_op2_swap
},
9525 [TGSI_OPCODE_FSNE
] = { ALU_OP2_SETNE_DX10
, tgsi_op2_swap
},
9526 [TGSI_OPCODE_MEMBAR
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9527 [TGSI_OPCODE_CALLNZ
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9528 [114] = { ALU_OP0_NOP
, tgsi_unsupported
},
9529 [TGSI_OPCODE_BREAKC
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9530 [TGSI_OPCODE_KILL_IF
] = { ALU_OP2_KILLGT
, tgsi_kill
}, /* conditional kill */
9531 [TGSI_OPCODE_END
] = { ALU_OP0_NOP
, tgsi_end
}, /* aka HALT */
9532 /* Refer below for TGSI_OPCODE_DFMA */
9533 [TGSI_OPCODE_F2I
] = { ALU_OP1_FLT_TO_INT
, tgsi_op2
},
9534 [TGSI_OPCODE_IDIV
] = { ALU_OP0_NOP
, tgsi_idiv
},
9535 [TGSI_OPCODE_IMAX
] = { ALU_OP2_MAX_INT
, tgsi_op2
},
9536 [TGSI_OPCODE_IMIN
] = { ALU_OP2_MIN_INT
, tgsi_op2
},
9537 [TGSI_OPCODE_INEG
] = { ALU_OP2_SUB_INT
, tgsi_ineg
},
9538 [TGSI_OPCODE_ISGE
] = { ALU_OP2_SETGE_INT
, tgsi_op2
},
9539 [TGSI_OPCODE_ISHR
] = { ALU_OP2_ASHR_INT
, tgsi_op2
},
9540 [TGSI_OPCODE_ISLT
] = { ALU_OP2_SETGT_INT
, tgsi_op2_swap
},
9541 [TGSI_OPCODE_F2U
] = { ALU_OP1_FLT_TO_UINT
, tgsi_op2
},
9542 [TGSI_OPCODE_U2F
] = { ALU_OP1_UINT_TO_FLT
, tgsi_op2
},
9543 [TGSI_OPCODE_UADD
] = { ALU_OP2_ADD_INT
, tgsi_op2
},
9544 [TGSI_OPCODE_UDIV
] = { ALU_OP0_NOP
, tgsi_udiv
},
9545 [TGSI_OPCODE_UMAD
] = { ALU_OP0_NOP
, tgsi_umad
},
9546 [TGSI_OPCODE_UMAX
] = { ALU_OP2_MAX_UINT
, tgsi_op2
},
9547 [TGSI_OPCODE_UMIN
] = { ALU_OP2_MIN_UINT
, tgsi_op2
},
9548 [TGSI_OPCODE_UMOD
] = { ALU_OP0_NOP
, tgsi_umod
},
9549 [TGSI_OPCODE_UMUL
] = { ALU_OP2_MULLO_INT
, cayman_mul_int_instr
},
9550 [TGSI_OPCODE_USEQ
] = { ALU_OP2_SETE_INT
, tgsi_op2
},
9551 [TGSI_OPCODE_USGE
] = { ALU_OP2_SETGE_UINT
, tgsi_op2
},
9552 [TGSI_OPCODE_USHR
] = { ALU_OP2_LSHR_INT
, tgsi_op2
},
9553 [TGSI_OPCODE_USLT
] = { ALU_OP2_SETGT_UINT
, tgsi_op2_swap
},
9554 [TGSI_OPCODE_USNE
] = { ALU_OP2_SETNE_INT
, tgsi_op2
},
9555 [TGSI_OPCODE_SWITCH
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9556 [TGSI_OPCODE_CASE
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9557 [TGSI_OPCODE_DEFAULT
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9558 [TGSI_OPCODE_ENDSWITCH
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9559 [TGSI_OPCODE_SAMPLE
] = { 0, tgsi_unsupported
},
9560 [TGSI_OPCODE_SAMPLE_I
] = { 0, tgsi_unsupported
},
9561 [TGSI_OPCODE_SAMPLE_I_MS
] = { 0, tgsi_unsupported
},
9562 [TGSI_OPCODE_SAMPLE_B
] = { 0, tgsi_unsupported
},
9563 [TGSI_OPCODE_SAMPLE_C
] = { 0, tgsi_unsupported
},
9564 [TGSI_OPCODE_SAMPLE_C_LZ
] = { 0, tgsi_unsupported
},
9565 [TGSI_OPCODE_SAMPLE_D
] = { 0, tgsi_unsupported
},
9566 [TGSI_OPCODE_SAMPLE_L
] = { 0, tgsi_unsupported
},
9567 [TGSI_OPCODE_GATHER4
] = { 0, tgsi_unsupported
},
9568 [TGSI_OPCODE_SVIEWINFO
] = { 0, tgsi_unsupported
},
9569 [TGSI_OPCODE_SAMPLE_POS
] = { 0, tgsi_unsupported
},
9570 [TGSI_OPCODE_SAMPLE_INFO
] = { 0, tgsi_unsupported
},
9571 [TGSI_OPCODE_UARL
] = { ALU_OP1_MOVA_INT
, tgsi_eg_arl
},
9572 [TGSI_OPCODE_UCMP
] = { ALU_OP0_NOP
, tgsi_ucmp
},
9573 [TGSI_OPCODE_IABS
] = { 0, tgsi_iabs
},
9574 [TGSI_OPCODE_ISSG
] = { 0, tgsi_issg
},
9575 [TGSI_OPCODE_LOAD
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9576 [TGSI_OPCODE_STORE
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9577 [TGSI_OPCODE_MFENCE
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9578 [TGSI_OPCODE_LFENCE
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9579 [TGSI_OPCODE_SFENCE
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9580 [TGSI_OPCODE_BARRIER
] = { ALU_OP0_GROUP_BARRIER
, tgsi_barrier
},
9581 [TGSI_OPCODE_ATOMUADD
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9582 [TGSI_OPCODE_ATOMXCHG
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9583 [TGSI_OPCODE_ATOMCAS
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9584 [TGSI_OPCODE_ATOMAND
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9585 [TGSI_OPCODE_ATOMOR
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9586 [TGSI_OPCODE_ATOMXOR
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9587 [TGSI_OPCODE_ATOMUMIN
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9588 [TGSI_OPCODE_ATOMUMAX
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9589 [TGSI_OPCODE_ATOMIMIN
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9590 [TGSI_OPCODE_ATOMIMAX
] = { ALU_OP0_NOP
, tgsi_unsupported
},
9591 [TGSI_OPCODE_TEX2
] = { FETCH_OP_SAMPLE
, tgsi_tex
},
9592 [TGSI_OPCODE_TXB2
] = { FETCH_OP_SAMPLE_LB
, tgsi_tex
},
9593 [TGSI_OPCODE_TXL2
] = { FETCH_OP_SAMPLE_L
, tgsi_tex
},
9594 [TGSI_OPCODE_IMUL_HI
] = { ALU_OP2_MULHI_INT
, cayman_mul_int_instr
},
9595 [TGSI_OPCODE_UMUL_HI
] = { ALU_OP2_MULHI_UINT
, cayman_mul_int_instr
},
9596 [TGSI_OPCODE_TG4
] = { FETCH_OP_GATHER4
, tgsi_tex
},
9597 [TGSI_OPCODE_LODQ
] = { FETCH_OP_GET_LOD
, tgsi_tex
},
9598 [TGSI_OPCODE_IBFE
] = { ALU_OP3_BFE_INT
, tgsi_op3
},
9599 [TGSI_OPCODE_UBFE
] = { ALU_OP3_BFE_UINT
, tgsi_op3
},
9600 [TGSI_OPCODE_BFI
] = { ALU_OP0_NOP
, tgsi_bfi
},
9601 [TGSI_OPCODE_BREV
] = { ALU_OP1_BFREV_INT
, tgsi_op2
},
9602 [TGSI_OPCODE_POPC
] = { ALU_OP1_BCNT_INT
, tgsi_op2
},
9603 [TGSI_OPCODE_LSB
] = { ALU_OP1_FFBL_INT
, tgsi_op2
},
9604 [TGSI_OPCODE_IMSB
] = { ALU_OP1_FFBH_INT
, tgsi_msb
},
9605 [TGSI_OPCODE_UMSB
] = { ALU_OP1_FFBH_UINT
, tgsi_msb
},
9606 [TGSI_OPCODE_INTERP_CENTROID
] = { ALU_OP0_NOP
, tgsi_interp_egcm
},
9607 [TGSI_OPCODE_INTERP_SAMPLE
] = { ALU_OP0_NOP
, tgsi_interp_egcm
},
9608 [TGSI_OPCODE_INTERP_OFFSET
] = { ALU_OP0_NOP
, tgsi_interp_egcm
},
9609 [TGSI_OPCODE_F2D
] = { ALU_OP1_FLT32_TO_FLT64
, tgsi_op2_64
},
9610 [TGSI_OPCODE_D2F
] = { ALU_OP1_FLT64_TO_FLT32
, tgsi_op2_64_single_dest
},
9611 [TGSI_OPCODE_DABS
] = { ALU_OP1_MOV
, tgsi_op2_64
},
9612 [TGSI_OPCODE_DNEG
] = { ALU_OP2_ADD_64
, tgsi_dneg
},
9613 [TGSI_OPCODE_DADD
] = { ALU_OP2_ADD_64
, tgsi_op2_64
},
9614 [TGSI_OPCODE_DMUL
] = { ALU_OP2_MUL_64
, cayman_mul_double_instr
},
9615 [TGSI_OPCODE_DMAX
] = { ALU_OP2_MAX_64
, tgsi_op2_64
},
9616 [TGSI_OPCODE_DMIN
] = { ALU_OP2_MIN_64
, tgsi_op2_64
},
9617 [TGSI_OPCODE_DSLT
] = { ALU_OP2_SETGT_64
, tgsi_op2_64_single_dest_s
},
9618 [TGSI_OPCODE_DSGE
] = { ALU_OP2_SETGE_64
, tgsi_op2_64_single_dest
},
9619 [TGSI_OPCODE_DSEQ
] = { ALU_OP2_SETE_64
, tgsi_op2_64_single_dest
},
9620 [TGSI_OPCODE_DSNE
] = { ALU_OP2_SETNE_64
, tgsi_op2_64_single_dest
},
9621 [TGSI_OPCODE_DRCP
] = { ALU_OP2_RECIP_64
, cayman_emit_double_instr
},
9622 [TGSI_OPCODE_DSQRT
] = { ALU_OP2_SQRT_64
, cayman_emit_double_instr
},
9623 [TGSI_OPCODE_DMAD
] = { ALU_OP3_FMA_64
, tgsi_op3_64
},
9624 [TGSI_OPCODE_DFMA
] = { ALU_OP3_FMA_64
, tgsi_op3_64
},
9625 [TGSI_OPCODE_DFRAC
] = { ALU_OP1_FRACT_64
, tgsi_op2_64
},
9626 [TGSI_OPCODE_DLDEXP
] = { ALU_OP2_LDEXP_64
, tgsi_op2_64
},
9627 [TGSI_OPCODE_DFRACEXP
] = { ALU_OP1_FREXP_64
, tgsi_dfracexp
},
9628 [TGSI_OPCODE_D2I
] = { ALU_OP1_FLT_TO_INT
, egcm_double_to_int
},
9629 [TGSI_OPCODE_I2D
] = { ALU_OP1_INT_TO_FLT
, egcm_int_to_double
},
9630 [TGSI_OPCODE_D2U
] = { ALU_OP1_FLT_TO_UINT
, egcm_double_to_int
},
9631 [TGSI_OPCODE_U2D
] = { ALU_OP1_UINT_TO_FLT
, egcm_int_to_double
},
9632 [TGSI_OPCODE_DRSQ
] = { ALU_OP2_RECIPSQRT_64
, cayman_emit_double_instr
},
9633 [TGSI_OPCODE_LAST
] = { ALU_OP0_NOP
, tgsi_unsupported
},