2 * Copyright 2010 Christoph Bumiller
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20 * OTHER DEALINGS IN THE SOFTWARE.
23 #include "pipe/p_defines.h"
25 #include "nvc0/nvc0_context.h"
27 #include "codegen/nv50_ir_driver.h"
28 #include "nvc0/nve4_compute.h"
30 /* NOTE: Using a[0x270] in FP may cause an error even if we're using less than
31 * 124 scalar varying values.
34 nvc0_shader_input_address(unsigned sn
, unsigned si
, unsigned ubase
)
37 case NV50_SEMANTIC_TESSFACTOR
: return 0x000 + si
* 0x4;
38 case TGSI_SEMANTIC_PRIMID
: return 0x060;
39 case TGSI_SEMANTIC_PSIZE
: return 0x06c;
40 case TGSI_SEMANTIC_POSITION
: return 0x070;
41 case TGSI_SEMANTIC_GENERIC
: return ubase
+ si
* 0x10;
42 case TGSI_SEMANTIC_FOG
: return 0x2e8;
43 case TGSI_SEMANTIC_COLOR
: return 0x280 + si
* 0x10;
44 case TGSI_SEMANTIC_BCOLOR
: return 0x2a0 + si
* 0x10;
45 case NV50_SEMANTIC_CLIPDISTANCE
: return 0x2c0 + si
* 0x4;
46 case TGSI_SEMANTIC_CLIPDIST
: return 0x2c0 + si
* 0x10;
47 case TGSI_SEMANTIC_CLIPVERTEX
: return 0x270;
48 case TGSI_SEMANTIC_PCOORD
: return 0x2e0;
49 case NV50_SEMANTIC_TESSCOORD
: return 0x2f0;
50 case TGSI_SEMANTIC_INSTANCEID
: return 0x2f8;
51 case TGSI_SEMANTIC_VERTEXID
: return 0x2fc;
52 case TGSI_SEMANTIC_TEXCOORD
: return 0x300 + si
* 0x10;
53 case TGSI_SEMANTIC_FACE
: return 0x3fc;
55 assert(!"invalid TGSI input semantic");
61 nvc0_shader_output_address(unsigned sn
, unsigned si
, unsigned ubase
)
64 case NV50_SEMANTIC_TESSFACTOR
: return 0x000 + si
* 0x4;
65 case TGSI_SEMANTIC_PRIMID
: return 0x060;
66 case TGSI_SEMANTIC_LAYER
: return 0x064;
67 case TGSI_SEMANTIC_VIEWPORT_INDEX
:return 0x068;
68 case TGSI_SEMANTIC_PSIZE
: return 0x06c;
69 case TGSI_SEMANTIC_POSITION
: return 0x070;
70 case TGSI_SEMANTIC_GENERIC
: return ubase
+ si
* 0x10;
71 case TGSI_SEMANTIC_FOG
: return 0x2e8;
72 case TGSI_SEMANTIC_COLOR
: return 0x280 + si
* 0x10;
73 case TGSI_SEMANTIC_BCOLOR
: return 0x2a0 + si
* 0x10;
74 case NV50_SEMANTIC_CLIPDISTANCE
: return 0x2c0 + si
* 0x4;
75 case TGSI_SEMANTIC_CLIPDIST
: return 0x2c0 + si
* 0x10;
76 case TGSI_SEMANTIC_CLIPVERTEX
: return 0x270;
77 case TGSI_SEMANTIC_TEXCOORD
: return 0x300 + si
* 0x10;
78 case TGSI_SEMANTIC_EDGEFLAG
: return ~0;
80 assert(!"invalid TGSI output semantic");
86 nvc0_vp_assign_input_slots(struct nv50_ir_prog_info
*info
)
90 for (n
= 0, i
= 0; i
< info
->numInputs
; ++i
) {
91 switch (info
->in
[i
].sn
) {
92 case TGSI_SEMANTIC_INSTANCEID
: /* for SM4 only, in TGSI they're SVs */
93 case TGSI_SEMANTIC_VERTEXID
:
94 info
->in
[i
].mask
= 0x1;
96 nvc0_shader_input_address(info
->in
[i
].sn
, 0, 0) / 4;
101 for (c
= 0; c
< 4; ++c
)
102 info
->in
[i
].slot
[c
] = (0x80 + n
* 0x10 + c
* 0x4) / 4;
110 nvc0_sp_assign_input_slots(struct nv50_ir_prog_info
*info
)
112 unsigned ubase
= MAX2(0x80, 0x20 + info
->numPatchConstants
* 0x10);
116 for (i
= 0; i
< info
->numInputs
; ++i
) {
117 offset
= nvc0_shader_input_address(info
->in
[i
].sn
,
118 info
->in
[i
].si
, ubase
);
119 if (info
->in
[i
].patch
&& offset
>= 0x20)
120 offset
= 0x20 + info
->in
[i
].si
* 0x10;
122 if (info
->in
[i
].sn
== NV50_SEMANTIC_TESSCOORD
)
123 info
->in
[i
].mask
&= 3;
125 for (c
= 0; c
< 4; ++c
)
126 info
->in
[i
].slot
[c
] = (offset
+ c
* 0x4) / 4;
133 nvc0_fp_assign_output_slots(struct nv50_ir_prog_info
*info
)
135 unsigned count
= info
->prop
.fp
.numColourResults
* 4;
138 for (i
= 0; i
< info
->numOutputs
; ++i
)
139 if (info
->out
[i
].sn
== TGSI_SEMANTIC_COLOR
)
140 for (c
= 0; c
< 4; ++c
)
141 info
->out
[i
].slot
[c
] = info
->out
[i
].si
* 4 + c
;
143 if (info
->io
.sampleMask
< PIPE_MAX_SHADER_OUTPUTS
)
144 info
->out
[info
->io
.sampleMask
].slot
[0] = count
++;
146 if (info
->target
>= 0xe0)
147 count
++; /* on Kepler, depth is always last colour reg + 2 */
149 if (info
->io
.fragDepth
< PIPE_MAX_SHADER_OUTPUTS
)
150 info
->out
[info
->io
.fragDepth
].slot
[2] = count
;
156 nvc0_sp_assign_output_slots(struct nv50_ir_prog_info
*info
)
158 unsigned ubase
= MAX2(0x80, 0x20 + info
->numPatchConstants
* 0x10);
162 for (i
= 0; i
< info
->numOutputs
; ++i
) {
163 offset
= nvc0_shader_output_address(info
->out
[i
].sn
,
164 info
->out
[i
].si
, ubase
);
165 if (info
->out
[i
].patch
&& offset
>= 0x20)
166 offset
= 0x20 + info
->out
[i
].si
* 0x10;
168 for (c
= 0; c
< 4; ++c
)
169 info
->out
[i
].slot
[c
] = (offset
+ c
* 0x4) / 4;
176 nvc0_program_assign_varying_slots(struct nv50_ir_prog_info
*info
)
180 if (info
->type
== PIPE_SHADER_VERTEX
)
181 ret
= nvc0_vp_assign_input_slots(info
);
183 ret
= nvc0_sp_assign_input_slots(info
);
187 if (info
->type
== PIPE_SHADER_FRAGMENT
)
188 ret
= nvc0_fp_assign_output_slots(info
);
190 ret
= nvc0_sp_assign_output_slots(info
);
195 nvc0_vtgp_hdr_update_oread(struct nvc0_program
*vp
, uint8_t slot
)
197 uint8_t min
= (vp
->hdr
[4] >> 12) & 0xff;
198 uint8_t max
= (vp
->hdr
[4] >> 24);
200 min
= MIN2(min
, slot
);
201 max
= MAX2(max
, slot
);
203 vp
->hdr
[4] = (max
<< 24) | (min
<< 12);
206 /* Common part of header generation for VP, TCP, TEP and GP. */
208 nvc0_vtgp_gen_header(struct nvc0_program
*vp
, struct nv50_ir_prog_info
*info
)
212 for (i
= 0; i
< info
->numInputs
; ++i
) {
213 if (info
->in
[i
].patch
)
215 for (c
= 0; c
< 4; ++c
) {
216 a
= info
->in
[i
].slot
[c
];
217 if (info
->in
[i
].mask
& (1 << c
)) {
218 if (info
->in
[i
].sn
!= NV50_SEMANTIC_TESSCOORD
)
219 vp
->hdr
[5 + a
/ 32] |= 1 << (a
% 32);
221 nvc0_vtgp_hdr_update_oread(vp
, info
->in
[i
].slot
[c
]);
226 for (i
= 0; i
< info
->numOutputs
; ++i
) {
227 if (info
->out
[i
].patch
)
229 for (c
= 0; c
< 4; ++c
) {
230 if (!(info
->out
[i
].mask
& (1 << c
)))
232 assert(info
->out
[i
].slot
[c
] >= 0x40 / 4);
233 a
= info
->out
[i
].slot
[c
] - 0x40 / 4;
234 vp
->hdr
[13 + a
/ 32] |= 1 << (a
% 32);
235 if (info
->out
[i
].oread
)
236 nvc0_vtgp_hdr_update_oread(vp
, info
->out
[i
].slot
[c
]);
240 for (i
= 0; i
< info
->numSysVals
; ++i
) {
241 switch (info
->sv
[i
].sn
) {
242 case TGSI_SEMANTIC_PRIMID
:
243 vp
->hdr
[5] |= 1 << 24;
245 case TGSI_SEMANTIC_INSTANCEID
:
246 vp
->hdr
[10] |= 1 << 30;
248 case TGSI_SEMANTIC_VERTEXID
:
249 vp
->hdr
[10] |= 1 << 31;
256 vp
->vp
.clip_enable
= info
->io
.clipDistanceMask
;
257 for (i
= 0; i
< 8; ++i
)
258 if (info
->io
.cullDistanceMask
& (1 << i
))
259 vp
->vp
.clip_mode
|= 1 << (i
* 4);
261 if (info
->io
.genUserClip
< 0)
262 vp
->vp
.num_ucps
= PIPE_MAX_CLIP_PLANES
+ 1; /* prevent rebuilding */
268 nvc0_vp_gen_header(struct nvc0_program
*vp
, struct nv50_ir_prog_info
*info
)
270 vp
->hdr
[0] = 0x20061 | (1 << 10);
271 vp
->hdr
[4] = 0xff000;
273 vp
->hdr
[18] = info
->io
.clipDistanceMask
;
275 return nvc0_vtgp_gen_header(vp
, info
);
278 #if defined(PIPE_SHADER_HULL) || defined(PIPE_SHADER_DOMAIN)
280 nvc0_tp_get_tess_mode(struct nvc0_program
*tp
, struct nv50_ir_prog_info
*info
)
282 if (info
->prop
.tp
.outputPrim
== PIPE_PRIM_MAX
) {
283 tp
->tp
.tess_mode
= ~0;
286 switch (info
->prop
.tp
.domain
) {
287 case PIPE_PRIM_LINES
:
288 tp
->tp
.tess_mode
= NVC0_3D_TESS_MODE_PRIM_ISOLINES
;
290 case PIPE_PRIM_TRIANGLES
:
291 tp
->tp
.tess_mode
= NVC0_3D_TESS_MODE_PRIM_TRIANGLES
;
292 if (info
->prop
.tp
.winding
> 0)
293 tp
->tp
.tess_mode
|= NVC0_3D_TESS_MODE_CW
;
295 case PIPE_PRIM_QUADS
:
296 tp
->tp
.tess_mode
= NVC0_3D_TESS_MODE_PRIM_QUADS
;
299 tp
->tp
.tess_mode
= ~0;
302 if (info
->prop
.tp
.outputPrim
!= PIPE_PRIM_POINTS
)
303 tp
->tp
.tess_mode
|= NVC0_3D_TESS_MODE_CONNECTED
;
305 switch (info
->prop
.tp
.partitioning
) {
306 case PIPE_TESS_PART_INTEGER
:
307 case PIPE_TESS_PART_POW2
:
308 tp
->tp
.tess_mode
|= NVC0_3D_TESS_MODE_SPACING_EQUAL
;
310 case PIPE_TESS_PART_FRACT_ODD
:
311 tp
->tp
.tess_mode
|= NVC0_3D_TESS_MODE_SPACING_FRACTIONAL_ODD
;
313 case PIPE_TESS_PART_FRACT_EVEN
:
314 tp
->tp
.tess_mode
|= NVC0_3D_TESS_MODE_SPACING_FRACTIONAL_EVEN
;
317 assert(!"invalid tessellator partitioning");
323 #ifdef PIPE_SHADER_HULL
325 nvc0_tcp_gen_header(struct nvc0_program
*tcp
, struct nv50_ir_prog_info
*info
)
327 unsigned opcs
= 6; /* output patch constants (at least the TessFactors) */
329 tcp
->tp
.input_patch_size
= info
->prop
.tp
.inputPatchSize
;
331 if (info
->numPatchConstants
)
332 opcs
= 8 + info
->numPatchConstants
* 4;
334 tcp
->hdr
[0] = 0x20061 | (2 << 10);
336 tcp
->hdr
[1] = opcs
<< 24;
337 tcp
->hdr
[2] = info
->prop
.tp
.outputPatchSize
<< 24;
339 tcp
->hdr
[4] = 0xff000; /* initial min/max parallel output read address */
341 nvc0_vtgp_gen_header(tcp
, info
);
343 nvc0_tp_get_tess_mode(tcp
, info
);
349 #ifdef PIPE_SHADER_DOMAIN
351 nvc0_tep_gen_header(struct nvc0_program
*tep
, struct nv50_ir_prog_info
*info
)
353 tep
->tp
.input_patch_size
= ~0;
355 tep
->hdr
[0] = 0x20061 | (3 << 10);
356 tep
->hdr
[4] = 0xff000;
358 nvc0_vtgp_gen_header(tep
, info
);
360 nvc0_tp_get_tess_mode(tep
, info
);
362 tep
->hdr
[18] |= 0x3 << 12; /* ? */
369 nvc0_gp_gen_header(struct nvc0_program
*gp
, struct nv50_ir_prog_info
*info
)
371 gp
->hdr
[0] = 0x20061 | (4 << 10);
373 gp
->hdr
[2] = MIN2(info
->prop
.gp
.instanceCount
, 32) << 24;
375 switch (info
->prop
.gp
.outputPrim
) {
376 case PIPE_PRIM_POINTS
:
377 gp
->hdr
[3] = 0x01000000;
378 gp
->hdr
[0] |= 0xf0000000;
380 case PIPE_PRIM_LINE_STRIP
:
381 gp
->hdr
[3] = 0x06000000;
382 gp
->hdr
[0] |= 0x10000000;
384 case PIPE_PRIM_TRIANGLE_STRIP
:
385 gp
->hdr
[3] = 0x07000000;
386 gp
->hdr
[0] |= 0x10000000;
393 gp
->hdr
[4] = info
->prop
.gp
.maxVertices
& 0x1ff;
395 return nvc0_vtgp_gen_header(gp
, info
);
398 #define NVC0_INTERP_FLAT (1 << 0)
399 #define NVC0_INTERP_PERSPECTIVE (2 << 0)
400 #define NVC0_INTERP_LINEAR (3 << 0)
401 #define NVC0_INTERP_CENTROID (1 << 2)
404 nvc0_hdr_interp_mode(const struct nv50_ir_varying
*var
)
407 return NVC0_INTERP_LINEAR
;
409 return NVC0_INTERP_FLAT
;
410 return NVC0_INTERP_PERSPECTIVE
;
414 nvc0_fp_gen_header(struct nvc0_program
*fp
, struct nv50_ir_prog_info
*info
)
418 /* just 00062 on Kepler */
419 fp
->hdr
[0] = 0x20062 | (5 << 10);
420 fp
->hdr
[5] = 0x80000000; /* getting a trap if FRAG_COORD_UMASK.w = 0 */
422 if (info
->prop
.fp
.usesDiscard
)
423 fp
->hdr
[0] |= 0x8000;
424 if (info
->prop
.fp
.numColourResults
> 1)
425 fp
->hdr
[0] |= 0x4000;
426 if (info
->io
.sampleMask
< PIPE_MAX_SHADER_OUTPUTS
)
428 if (info
->prop
.fp
.writesDepth
) {
430 fp
->flags
[0] = 0x11; /* deactivate ZCULL */
433 for (i
= 0; i
< info
->numInputs
; ++i
) {
434 m
= nvc0_hdr_interp_mode(&info
->in
[i
]);
435 for (c
= 0; c
< 4; ++c
) {
436 if (!(info
->in
[i
].mask
& (1 << c
)))
438 a
= info
->in
[i
].slot
[c
];
439 if (info
->in
[i
].slot
[0] >= (0x060 / 4) &&
440 info
->in
[i
].slot
[0] <= (0x07c / 4)) {
441 fp
->hdr
[5] |= 1 << (24 + (a
- 0x060 / 4));
443 if (info
->in
[i
].slot
[0] >= (0x2c0 / 4) &&
444 info
->in
[i
].slot
[0] <= (0x2fc / 4)) {
445 fp
->hdr
[14] |= (1 << (a
- 0x280 / 4)) & 0x07ff0000;
447 if (info
->in
[i
].slot
[c
] < (0x040 / 4) ||
448 info
->in
[i
].slot
[c
] > (0x380 / 4))
451 if (info
->in
[i
].slot
[0] >= (0x300 / 4))
453 fp
->hdr
[4 + a
/ 32] |= m
<< (a
% 32);
458 for (i
= 0; i
< info
->numOutputs
; ++i
) {
459 if (info
->out
[i
].sn
== TGSI_SEMANTIC_COLOR
)
460 fp
->hdr
[18] |= info
->out
[i
].mask
<< info
->out
[i
].slot
[0];
463 fp
->fp
.early_z
= info
->prop
.fp
.earlyFragTests
;
468 static struct nvc0_transform_feedback_state
*
469 nvc0_program_create_tfb_state(const struct nv50_ir_prog_info
*info
,
470 const struct pipe_stream_output_info
*pso
)
472 struct nvc0_transform_feedback_state
*tfb
;
475 tfb
= MALLOC_STRUCT(nvc0_transform_feedback_state
);
478 for (b
= 0; b
< 4; ++b
) {
479 tfb
->stride
[b
] = pso
->stride
[b
] * 4;
480 tfb
->varying_count
[b
] = 0;
482 memset(tfb
->varying_index
, 0xff, sizeof(tfb
->varying_index
)); /* = skip */
484 for (i
= 0; i
< pso
->num_outputs
; ++i
) {
485 unsigned s
= pso
->output
[i
].start_component
;
486 unsigned p
= pso
->output
[i
].dst_offset
;
487 b
= pso
->output
[i
].output_buffer
;
489 for (c
= 0; c
< pso
->output
[i
].num_components
; ++c
)
490 tfb
->varying_index
[b
][p
++] =
491 info
->out
[pso
->output
[i
].register_index
].slot
[s
+ c
];
493 tfb
->varying_count
[b
] = MAX2(tfb
->varying_count
[b
], p
);
495 for (b
= 0; b
< 4; ++b
) // zero unused indices (looks nicer)
496 for (c
= tfb
->varying_count
[b
]; c
& 3; ++c
)
497 tfb
->varying_index
[b
][c
] = 0;
504 nvc0_program_dump(struct nvc0_program
*prog
)
508 if (prog
->type
!= PIPE_SHADER_COMPUTE
) {
509 for (pos
= 0; pos
< sizeof(prog
->hdr
) / sizeof(prog
->hdr
[0]); ++pos
)
510 debug_printf("HDR[%02"PRIxPTR
"] = 0x%08x\n",
511 pos
* sizeof(prog
->hdr
[0]), prog
->hdr
[pos
]);
513 debug_printf("shader binary code (0x%x bytes):", prog
->code_size
);
514 for (pos
= 0; pos
< prog
->code_size
/ 4; ++pos
) {
517 debug_printf("%08x ", prog
->code
[pos
]);
524 nvc0_program_translate(struct nvc0_program
*prog
, uint16_t chipset
)
526 struct nv50_ir_prog_info
*info
;
529 info
= CALLOC_STRUCT(nv50_ir_prog_info
);
533 info
->type
= prog
->type
;
534 info
->target
= chipset
;
535 info
->bin
.sourceRep
= NV50_PROGRAM_IR_TGSI
;
536 info
->bin
.source
= (void *)prog
->pipe
.tokens
;
538 info
->io
.genUserClip
= prog
->vp
.num_ucps
;
539 info
->io
.ucpBase
= 256;
540 info
->io
.ucpCBSlot
= 15;
541 info
->io
.sampleInterp
= prog
->fp
.sample_interp
;
543 if (prog
->type
== PIPE_SHADER_COMPUTE
) {
544 if (chipset
>= NVISA_GK104_CHIPSET
) {
545 info
->io
.resInfoCBSlot
= 0;
546 info
->io
.texBindBase
= NVE4_CP_INPUT_TEX(0);
547 info
->io
.suInfoBase
= NVE4_CP_INPUT_SUF(0);
548 info
->prop
.cp
.gridInfoBase
= NVE4_CP_INPUT_GRID_INFO(0);
550 info
->io
.msInfoCBSlot
= 0;
551 info
->io
.msInfoBase
= NVE4_CP_INPUT_MS_OFFSETS
;
553 if (chipset
>= NVISA_GK104_CHIPSET
) {
554 info
->io
.texBindBase
= 0x20;
555 info
->io
.suInfoBase
= 0; /* TODO */
557 info
->io
.resInfoCBSlot
= 15;
558 info
->io
.sampleInfoBase
= 256 + 128;
559 info
->io
.msInfoCBSlot
= 15;
560 info
->io
.msInfoBase
= 0; /* TODO */
563 info
->assignSlots
= nvc0_program_assign_varying_slots
;
566 info
->optLevel
= debug_get_num_option("NV50_PROG_OPTIMIZE", 3);
567 info
->dbgFlags
= debug_get_num_option("NV50_PROG_DEBUG", 0);
572 ret
= nv50_ir_generate_code(info
);
574 NOUVEAU_ERR("shader translation failed: %i\n", ret
);
577 if (prog
->type
!= PIPE_SHADER_COMPUTE
)
578 FREE(info
->bin
.syms
);
580 prog
->code
= info
->bin
.code
;
581 prog
->code_size
= info
->bin
.codeSize
;
582 prog
->immd_data
= info
->immd
.buf
;
583 prog
->immd_size
= info
->immd
.bufSize
;
584 prog
->relocs
= info
->bin
.relocData
;
585 prog
->num_gprs
= MAX2(4, (info
->bin
.maxGPR
+ 1));
586 prog
->num_barriers
= info
->numBarriers
;
588 prog
->vp
.need_vertex_id
= info
->io
.vertexId
< PIPE_MAX_SHADER_INPUTS
;
590 if (info
->io
.edgeFlagOut
< PIPE_MAX_ATTRIBS
)
591 info
->out
[info
->io
.edgeFlagOut
].mask
= 0; /* for headergen */
592 prog
->vp
.edgeflag
= info
->io
.edgeFlagIn
;
594 switch (prog
->type
) {
595 case PIPE_SHADER_VERTEX
:
596 ret
= nvc0_vp_gen_header(prog
, info
);
598 #ifdef PIPE_SHADER_HULL
599 case PIPE_SHADER_HULL
:
600 ret
= nvc0_tcp_gen_header(prog
, info
);
603 #ifdef PIPE_SHADER_DOMAIN
604 case PIPE_SHADER_DOMAIN
:
605 ret
= nvc0_tep_gen_header(prog
, info
);
608 case PIPE_SHADER_GEOMETRY
:
609 ret
= nvc0_gp_gen_header(prog
, info
);
611 case PIPE_SHADER_FRAGMENT
:
612 ret
= nvc0_fp_gen_header(prog
, info
);
614 case PIPE_SHADER_COMPUTE
:
615 prog
->cp
.syms
= info
->bin
.syms
;
616 prog
->cp
.num_syms
= info
->bin
.numSyms
;
620 NOUVEAU_ERR("unknown program type: %u\n", prog
->type
);
626 if (info
->bin
.tlsSpace
) {
627 assert(info
->bin
.tlsSpace
< (1 << 24));
628 prog
->hdr
[0] |= 1 << 26;
629 prog
->hdr
[1] |= info
->bin
.tlsSpace
; /* l[] size */
630 prog
->need_tls
= TRUE
;
632 /* TODO: factor 2 only needed where joinat/precont is used,
633 * and we only have to count non-uniform branches
636 if ((info->maxCFDepth * 2) > 16) {
637 prog->hdr[2] |= (((info->maxCFDepth * 2) + 47) / 48) * 0x200;
638 prog->need_tls = TRUE;
641 if (info
->io
.globalAccess
)
642 prog
->hdr
[0] |= 1 << 16;
644 if (prog
->pipe
.stream_output
.num_outputs
)
645 prog
->tfb
= nvc0_program_create_tfb_state(info
,
646 &prog
->pipe
.stream_output
);
654 nvc0_program_upload_code(struct nvc0_context
*nvc0
, struct nvc0_program
*prog
)
656 struct nvc0_screen
*screen
= nvc0
->screen
;
657 const boolean is_cp
= prog
->type
== PIPE_SHADER_COMPUTE
;
659 uint32_t size
= prog
->code_size
+ (is_cp
? 0 : NVC0_SHADER_HEADER_SIZE
);
660 uint32_t lib_pos
= screen
->lib_code
->start
;
663 /* c[] bindings need to be aligned to 0x100, but we could use relocations
665 if (prog
->immd_size
) {
666 prog
->immd_base
= size
;
667 size
= align(size
, 0x40);
668 size
+= prog
->immd_size
+ 0xc0; /* add 0xc0 for align 0x40 -> 0x100 */
670 /* On Fermi, SP_START_ID must be aligned to 0x40.
671 * On Kepler, the first instruction must be aligned to 0x80 because
672 * latency information is expected only at certain positions.
674 if (screen
->base
.class_3d
>= NVE4_3D_CLASS
)
675 size
= size
+ (is_cp
? 0x40 : 0x70);
676 size
= align(size
, 0x40);
678 ret
= nouveau_heap_alloc(screen
->text_heap
, size
, prog
, &prog
->mem
);
680 struct nouveau_heap
*heap
= screen
->text_heap
;
681 struct nouveau_heap
*iter
;
682 for (iter
= heap
; iter
&& iter
->next
!= heap
; iter
= iter
->next
) {
683 struct nvc0_program
*evict
= iter
->priv
;
685 nouveau_heap_free(&evict
->mem
);
687 debug_printf("WARNING: out of code space, evicting all shaders.\n");
688 ret
= nouveau_heap_alloc(heap
, size
, prog
, &prog
->mem
);
690 NOUVEAU_ERR("shader too large (0x%x) to fit in code space ?\n", size
);
693 IMMED_NVC0(nvc0
->base
.pushbuf
, NVC0_3D(SERIALIZE
), 0);
695 prog
->code_base
= prog
->mem
->start
;
696 prog
->immd_base
= align(prog
->mem
->start
+ prog
->immd_base
, 0x100);
697 assert((prog
->immd_size
== 0) || (prog
->immd_base
+ prog
->immd_size
<=
698 prog
->mem
->start
+ prog
->mem
->size
));
701 if (screen
->base
.class_3d
>= NVE4_3D_CLASS
) {
702 switch (prog
->mem
->start
& 0xff) {
703 case 0x40: prog
->code_base
+= 0x70; break;
704 case 0x80: prog
->code_base
+= 0x30; break;
705 case 0xc0: prog
->code_base
+= 0x70; break;
707 prog
->code_base
+= 0x30;
708 assert((prog
->mem
->start
& 0xff) == 0x00);
712 code_pos
= prog
->code_base
+ NVC0_SHADER_HEADER_SIZE
;
714 if (screen
->base
.class_3d
>= NVE4_3D_CLASS
) {
715 if (prog
->mem
->start
& 0x40)
716 prog
->code_base
+= 0x40;
717 assert((prog
->code_base
& 0x7f) == 0x00);
719 code_pos
= prog
->code_base
;
723 nv50_ir_relocate_code(prog
->relocs
, prog
->code
, code_pos
, lib_pos
, 0);
726 if (debug_get_bool_option("NV50_PROG_DEBUG", FALSE
))
727 nvc0_program_dump(prog
);
731 nvc0
->base
.push_data(&nvc0
->base
, screen
->text
, prog
->code_base
,
732 NOUVEAU_BO_VRAM
, NVC0_SHADER_HEADER_SIZE
, prog
->hdr
);
733 nvc0
->base
.push_data(&nvc0
->base
, screen
->text
, code_pos
,
734 NOUVEAU_BO_VRAM
, prog
->code_size
, prog
->code
);
736 nvc0
->base
.push_data(&nvc0
->base
,
737 screen
->text
, prog
->immd_base
, NOUVEAU_BO_VRAM
,
738 prog
->immd_size
, prog
->immd_data
);
740 BEGIN_NVC0(nvc0
->base
.pushbuf
, NVC0_3D(MEM_BARRIER
), 1);
741 PUSH_DATA (nvc0
->base
.pushbuf
, 0x1011);
746 /* Upload code for builtin functions like integer division emulation. */
748 nvc0_program_library_upload(struct nvc0_context
*nvc0
)
750 struct nvc0_screen
*screen
= nvc0
->screen
;
753 const uint32_t *code
;
755 if (screen
->lib_code
)
758 nv50_ir_get_target_library(screen
->base
.device
->chipset
, &code
, &size
);
762 ret
= nouveau_heap_alloc(screen
->text_heap
, align(size
, 0x100), NULL
,
767 nvc0
->base
.push_data(&nvc0
->base
,
768 screen
->text
, screen
->lib_code
->start
, NOUVEAU_BO_VRAM
,
770 /* no need for a memory barrier, will be emitted with first program */
774 nvc0_program_destroy(struct nvc0_context
*nvc0
, struct nvc0_program
*prog
)
776 const struct pipe_shader_state pipe
= prog
->pipe
;
777 const ubyte type
= prog
->type
;
780 nouveau_heap_free(&prog
->mem
);
782 FREE(prog
->code
); /* may be 0 for hardcoded shaders */
783 FREE(prog
->immd_data
);
785 if (prog
->type
== PIPE_SHADER_COMPUTE
&& prog
->cp
.syms
)
788 if (nvc0
->state
.tfb
== prog
->tfb
)
789 nvc0
->state
.tfb
= NULL
;
793 memset(prog
, 0, sizeof(*prog
));
800 nvc0_program_symbol_offset(const struct nvc0_program
*prog
, uint32_t label
)
802 const struct nv50_ir_prog_symbol
*syms
=
803 (const struct nv50_ir_prog_symbol
*)prog
->cp
.syms
;
806 if (prog
->type
!= PIPE_SHADER_COMPUTE
)
807 base
= NVC0_SHADER_HEADER_SIZE
;
808 for (i
= 0; i
< prog
->cp
.num_syms
; ++i
)
809 if (syms
[i
].label
== label
)
810 return prog
->code_base
+ base
+ syms
[i
].offset
;
811 return prog
->code_base
; /* no symbols or symbol not found */