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 BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
18 * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF
19 * OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
23 #include "pipe/p_defines.h"
25 #include "nvc0_context.h"
27 #include "nv50/codegen/nv50_ir_driver.h"
29 /* If only they told use the actual semantic instead of just GENERIC ... */
31 nvc0_mesa_varying_hack(struct nv50_ir_varying
*var
)
35 if (var
->sn
!= TGSI_SEMANTIC_GENERIC
)
38 if (var
->si
<= 7) /* gl_TexCoord */
39 for (c
= 0; c
< 4; ++c
)
40 var
->slot
[c
] = (0x300 + var
->si
* 0x10 + c
* 0x4) / 4;
42 if (var
->si
== 9) /* gl_PointCoord */
43 for (c
= 0; c
< 4; ++c
)
44 var
->slot
[c
] = (0x2e0 + c
* 0x4) / 4;
47 for (c
= 0; c
< 4; ++c
) /* move down user varyings (first has index 8) */
48 var
->slot
[c
] -= 0x80 / 4;
50 NOUVEAU_ERR("too many varyings / invalid location: %u !\n", var
->si
);
51 for (c
= 0; c
< 4; ++c
)
52 var
->slot
[c
] = (0x270 + c
* 0x4) / 4; /* catch invalid indices */
57 nvc0_shader_input_address(unsigned sn
, unsigned si
, unsigned ubase
)
60 case NV50_SEMANTIC_TESSFACTOR
: return 0x000 + si
* 0x4;
61 case TGSI_SEMANTIC_PRIMID
: return 0x060;
62 case TGSI_SEMANTIC_PSIZE
: return 0x06c;
63 case TGSI_SEMANTIC_POSITION
: return 0x070;
64 case TGSI_SEMANTIC_GENERIC
: return ubase
+ si
* 0x10;
65 case TGSI_SEMANTIC_FOG
: return 0x270;
66 case TGSI_SEMANTIC_COLOR
: return 0x280 + si
* 0x10;
67 case TGSI_SEMANTIC_BCOLOR
: return 0x2a0 + si
* 0x10;
68 case NV50_SEMANTIC_CLIPDISTANCE
: return 0x2c0 + si
* 0x4;
69 case TGSI_SEMANTIC_CLIPDIST
: return 0x2c0 + si
* 0x10;
70 case TGSI_SEMANTIC_CLIPVERTEX
: return 0x260;
71 case NV50_SEMANTIC_POINTCOORD
: return 0x2e0;
72 case NV50_SEMANTIC_TESSCOORD
: return 0x2f0;
73 case TGSI_SEMANTIC_INSTANCEID
: return 0x2f8;
74 case TGSI_SEMANTIC_VERTEXID
: return 0x2fc;
75 case NV50_SEMANTIC_TEXCOORD
: return 0x300 + si
* 0x10;
76 case TGSI_SEMANTIC_FACE
: return 0x3fc;
77 case NV50_SEMANTIC_INVOCATIONID
: return ~0;
79 assert(!"invalid TGSI input semantic");
85 nvc0_shader_output_address(unsigned sn
, unsigned si
, unsigned ubase
)
88 case NV50_SEMANTIC_TESSFACTOR
: return 0x000 + si
* 0x4;
89 case TGSI_SEMANTIC_PRIMID
: return 0x060;
90 case NV50_SEMANTIC_LAYER
: return 0x064;
91 case NV50_SEMANTIC_VIEWPORTINDEX
: return 0x068;
92 case TGSI_SEMANTIC_PSIZE
: return 0x06c;
93 case TGSI_SEMANTIC_POSITION
: return 0x070;
94 case TGSI_SEMANTIC_GENERIC
: return ubase
+ si
* 0x10;
95 case TGSI_SEMANTIC_FOG
: return 0x270;
96 case TGSI_SEMANTIC_COLOR
: return 0x280 + si
* 0x10;
97 case TGSI_SEMANTIC_BCOLOR
: return 0x2a0 + si
* 0x10;
98 case NV50_SEMANTIC_CLIPDISTANCE
: return 0x2c0 + si
* 0x4;
99 case TGSI_SEMANTIC_CLIPDIST
: return 0x2c0 + si
* 0x10;
100 case TGSI_SEMANTIC_CLIPVERTEX
: return 0x260;
101 case NV50_SEMANTIC_TEXCOORD
: return 0x300 + si
* 0x10;
102 case TGSI_SEMANTIC_EDGEFLAG
: return ~0;
104 assert(!"invalid TGSI output semantic");
110 nvc0_vp_assign_input_slots(struct nv50_ir_prog_info
*info
)
114 for (n
= 0, i
= 0; i
< info
->numInputs
; ++i
) {
115 switch (info
->in
[i
].sn
) {
116 case TGSI_SEMANTIC_INSTANCEID
: /* for SM4 only, in TGSI they're SVs */
117 case TGSI_SEMANTIC_VERTEXID
:
118 info
->in
[i
].mask
= 0x1;
119 info
->in
[i
].slot
[0] =
120 nvc0_shader_input_address(info
->in
[i
].sn
, 0, 0) / 4;
125 for (c
= 0; c
< 4; ++c
)
126 info
->in
[i
].slot
[c
] = (0x80 + n
* 0x10 + c
* 0x4) / 4;
134 nvc0_sp_assign_input_slots(struct nv50_ir_prog_info
*info
)
136 unsigned ubase
= MAX2(0x80, 0x20 + info
->numPatchConstants
* 0x10);
140 for (i
= 0; i
< info
->numInputs
; ++i
) {
141 offset
= nvc0_shader_input_address(info
->in
[i
].sn
,
142 info
->in
[i
].si
, ubase
);
143 if (info
->in
[i
].patch
&& offset
>= 0x20)
144 offset
= 0x20 + info
->in
[i
].si
* 0x10;
146 if (info
->in
[i
].sn
== NV50_SEMANTIC_TESSCOORD
)
147 info
->in
[i
].mask
&= 3;
149 for (c
= 0; c
< 4; ++c
)
150 info
->in
[i
].slot
[c
] = (offset
+ c
* 0x4) / 4;
152 nvc0_mesa_varying_hack(&info
->in
[i
]);
159 nvc0_fp_assign_output_slots(struct nv50_ir_prog_info
*info
)
161 unsigned count
= info
->prop
.fp
.numColourResults
* 4;
164 for (i
= 0; i
< info
->numOutputs
; ++i
)
165 if (info
->out
[i
].sn
== TGSI_SEMANTIC_COLOR
)
166 for (c
= 0; c
< 4; ++c
)
167 info
->out
[i
].slot
[c
] = info
->out
[i
].si
* 4 + c
;
169 if (info
->io
.sampleMask
< PIPE_MAX_SHADER_OUTPUTS
)
170 info
->out
[info
->io
.sampleMask
].slot
[0] = count
++;
172 if (info
->target
>= 0xe0)
173 count
++; /* on Kepler, depth is always last colour reg + 2 */
175 if (info
->io
.fragDepth
< PIPE_MAX_SHADER_OUTPUTS
)
176 info
->out
[info
->io
.fragDepth
].slot
[2] = count
;
182 nvc0_sp_assign_output_slots(struct nv50_ir_prog_info
*info
)
184 unsigned ubase
= MAX2(0x80, 0x20 + info
->numPatchConstants
* 0x10);
188 for (i
= 0; i
< info
->numOutputs
; ++i
) {
189 offset
= nvc0_shader_output_address(info
->out
[i
].sn
,
190 info
->out
[i
].si
, ubase
);
191 if (info
->out
[i
].patch
&& offset
>= 0x20)
192 offset
= 0x20 + info
->out
[i
].si
* 0x10;
194 for (c
= 0; c
< 4; ++c
)
195 info
->out
[i
].slot
[c
] = (offset
+ c
* 0x4) / 4;
197 nvc0_mesa_varying_hack(&info
->out
[i
]);
204 nvc0_program_assign_varying_slots(struct nv50_ir_prog_info
*info
)
208 if (info
->type
== PIPE_SHADER_VERTEX
)
209 ret
= nvc0_vp_assign_input_slots(info
);
211 ret
= nvc0_sp_assign_input_slots(info
);
215 if (info
->type
== PIPE_SHADER_FRAGMENT
)
216 ret
= nvc0_fp_assign_output_slots(info
);
218 ret
= nvc0_sp_assign_output_slots(info
);
223 nvc0_vtgp_hdr_update_oread(struct nvc0_program
*vp
, uint8_t slot
)
225 uint8_t min
= (vp
->hdr
[4] >> 12) & 0xff;
226 uint8_t max
= (vp
->hdr
[4] >> 24);
228 min
= MIN2(min
, slot
);
229 max
= MAX2(max
, slot
);
231 vp
->hdr
[4] = (max
<< 24) | (min
<< 12);
234 /* Common part of header generation for VP, TCP, TEP and GP. */
236 nvc0_vtgp_gen_header(struct nvc0_program
*vp
, struct nv50_ir_prog_info
*info
)
240 for (i
= 0; i
< info
->numInputs
; ++i
) {
241 if (info
->in
[i
].patch
)
243 for (c
= 0; c
< 4; ++c
) {
244 a
= info
->in
[i
].slot
[c
];
245 if (info
->in
[i
].mask
& (1 << c
)) {
246 if (info
->in
[i
].sn
!= NV50_SEMANTIC_TESSCOORD
)
247 vp
->hdr
[5 + a
/ 32] |= 1 << (a
% 32);
249 nvc0_vtgp_hdr_update_oread(vp
, info
->in
[i
].slot
[c
]);
254 for (i
= 0; i
< info
->numOutputs
; ++i
) {
255 if (info
->out
[i
].patch
)
257 for (c
= 0; c
< 4; ++c
) {
258 if (!(info
->out
[i
].mask
& (1 << c
)))
260 assert(info
->out
[i
].slot
[c
] >= 0x40 / 4);
261 a
= info
->out
[i
].slot
[c
] - 0x40 / 4;
262 vp
->hdr
[13 + a
/ 32] |= 1 << (a
% 32);
263 if (info
->out
[i
].oread
)
264 nvc0_vtgp_hdr_update_oread(vp
, info
->out
[i
].slot
[c
]);
268 for (i
= 0; i
< info
->numSysVals
; ++i
) {
269 switch (info
->sv
[i
].sn
) {
270 case TGSI_SEMANTIC_PRIMID
:
271 vp
->hdr
[5] |= 1 << 24;
273 case TGSI_SEMANTIC_INSTANCEID
:
274 vp
->hdr
[10] |= 1 << 30;
276 case TGSI_SEMANTIC_VERTEXID
:
277 vp
->hdr
[10] |= 1 << 31;
284 vp
->vp
.clip_enable
= info
->io
.clipDistanceMask
;
285 for (i
= 0; i
< 8; ++i
)
286 if (info
->io
.cullDistanceMask
& (1 << i
))
287 vp
->vp
.clip_mode
|= 1 << (i
* 4);
289 if (info
->io
.genUserClip
< 0)
290 vp
->vp
.num_ucps
= PIPE_MAX_CLIP_PLANES
+ 1; /* prevent rebuilding */
296 nvc0_vp_gen_header(struct nvc0_program
*vp
, struct nv50_ir_prog_info
*info
)
298 vp
->hdr
[0] = 0x20061 | (1 << 10);
299 vp
->hdr
[4] = 0xff000;
301 vp
->hdr
[18] = info
->io
.clipDistanceMask
;
303 return nvc0_vtgp_gen_header(vp
, info
);
306 #if defined(PIPE_SHADER_HULL) || defined(PIPE_SHADER_DOMAIN)
308 nvc0_tp_get_tess_mode(struct nvc0_program
*tp
, struct nv50_ir_prog_info
*info
)
310 if (info
->prop
.tp
.outputPrim
== PIPE_PRIM_MAX
) {
311 tp
->tp
.tess_mode
= ~0;
314 switch (info
->prop
.tp
.domain
) {
315 case PIPE_PRIM_LINES
:
316 tp
->tp
.tess_mode
= NVC0_3D_TESS_MODE_PRIM_ISOLINES
;
318 case PIPE_PRIM_TRIANGLES
:
319 tp
->tp
.tess_mode
= NVC0_3D_TESS_MODE_PRIM_TRIANGLES
;
320 if (info
->prop
.tp
.winding
> 0)
321 tp
->tp
.tess_mode
|= NVC0_3D_TESS_MODE_CW
;
323 case PIPE_PRIM_QUADS
:
324 tp
->tp
.tess_mode
= NVC0_3D_TESS_MODE_PRIM_QUADS
;
327 tp
->tp
.tess_mode
= ~0;
330 if (info
->prop
.tp
.outputPrim
!= PIPE_PRIM_POINTS
)
331 tp
->tp
.tess_mode
|= NVC0_3D_TESS_MODE_CONNECTED
;
333 switch (info
->prop
.tp
.partitioning
) {
334 case PIPE_TESS_PART_INTEGER
:
335 case PIPE_TESS_PART_POW2
:
336 tp
->tp
.tess_mode
|= NVC0_3D_TESS_MODE_SPACING_EQUAL
;
338 case PIPE_TESS_PART_FRACT_ODD
:
339 tp
->tp
.tess_mode
|= NVC0_3D_TESS_MODE_SPACING_FRACTIONAL_ODD
;
341 case PIPE_TESS_PART_FRACT_EVEN
:
342 tp
->tp
.tess_mode
|= NVC0_3D_TESS_MODE_SPACING_FRACTIONAL_EVEN
;
345 assert(!"invalid tessellator partitioning");
351 #ifdef PIPE_SHADER_HULL
353 nvc0_tcp_gen_header(struct nvc0_program
*tcp
, struct nv50_ir_prog_info
*info
)
355 unsigned opcs
= 6; /* output patch constants (at least the TessFactors) */
357 tcp
->tp
.input_patch_size
= info
->prop
.tp
.inputPatchSize
;
359 if (info
->numPatchConstants
)
360 opcs
= 8 + info
->numPatchConstants
* 4;
362 tcp
->hdr
[0] = 0x20061 | (2 << 10);
364 tcp
->hdr
[1] = opcs
<< 24;
365 tcp
->hdr
[2] = info
->prop
.tp
.outputPatchSize
<< 24;
367 tcp
->hdr
[4] = 0xff000; /* initial min/max parallel output read address */
369 nvc0_vtgp_gen_header(tcp
, info
);
371 nvc0_tp_get_tess_mode(tcp
, info
);
377 #ifdef PIPE_SHADER_DOMAIN
379 nvc0_tep_gen_header(struct nvc0_program
*tep
, struct nv50_ir_prog_info
*info
)
381 tep
->tp
.input_patch_size
= ~0;
383 tep
->hdr
[0] = 0x20061 | (3 << 10);
384 tep
->hdr
[4] = 0xff000;
386 nvc0_vtgp_gen_header(tep
, info
);
388 nvc0_tp_get_tess_mode(tep
, info
);
390 tep
->hdr
[18] |= 0x3 << 12; /* ? */
397 nvc0_gp_gen_header(struct nvc0_program
*gp
, struct nv50_ir_prog_info
*info
)
399 gp
->hdr
[0] = 0x20061 | (4 << 10);
401 gp
->hdr
[2] = MIN2(info
->prop
.gp
.instanceCount
, 32) << 24;
403 switch (info
->prop
.gp
.outputPrim
) {
404 case PIPE_PRIM_POINTS
:
405 gp
->hdr
[3] = 0x01000000;
406 gp
->hdr
[0] |= 0xf0000000;
408 case PIPE_PRIM_LINE_STRIP
:
409 gp
->hdr
[3] = 0x06000000;
410 gp
->hdr
[0] |= 0x10000000;
412 case PIPE_PRIM_TRIANGLE_STRIP
:
413 gp
->hdr
[3] = 0x07000000;
414 gp
->hdr
[0] |= 0x10000000;
421 gp
->hdr
[4] = info
->prop
.gp
.maxVertices
& 0x1ff;
423 return nvc0_vtgp_gen_header(gp
, info
);
426 #define NVC0_INTERP_FLAT (1 << 0)
427 #define NVC0_INTERP_PERSPECTIVE (2 << 0)
428 #define NVC0_INTERP_LINEAR (3 << 0)
429 #define NVC0_INTERP_CENTROID (1 << 2)
432 nvc0_hdr_interp_mode(const struct nv50_ir_varying
*var
)
435 return NVC0_INTERP_LINEAR
;
437 return NVC0_INTERP_FLAT
;
438 return NVC0_INTERP_PERSPECTIVE
;
442 nvc0_fp_gen_header(struct nvc0_program
*fp
, struct nv50_ir_prog_info
*info
)
446 /* just 00062 on Kepler */
447 fp
->hdr
[0] = 0x20062 | (5 << 10);
448 fp
->hdr
[5] = 0x80000000; /* getting a trap if FRAG_COORD_UMASK.w = 0 */
450 if (info
->prop
.fp
.usesDiscard
)
451 fp
->hdr
[0] |= 0x8000;
452 if (info
->prop
.fp
.numColourResults
> 1)
453 fp
->hdr
[0] |= 0x4000;
454 if (info
->io
.sampleMask
< PIPE_MAX_SHADER_OUTPUTS
)
456 if (info
->prop
.fp
.writesDepth
) {
458 fp
->flags
[0] = 0x11; /* deactivate ZCULL */
461 for (i
= 0; i
< info
->numInputs
; ++i
) {
462 m
= nvc0_hdr_interp_mode(&info
->in
[i
]);
463 for (c
= 0; c
< 4; ++c
) {
464 if (!(info
->in
[i
].mask
& (1 << c
)))
466 a
= info
->in
[i
].slot
[c
];
467 if (info
->in
[i
].slot
[0] >= (0x060 / 4) &&
468 info
->in
[i
].slot
[0] <= (0x07c / 4)) {
469 fp
->hdr
[5] |= 1 << (24 + (a
- 0x060 / 4));
471 if (info
->in
[i
].slot
[0] >= (0x2c0 / 4) &&
472 info
->in
[i
].slot
[0] <= (0x2fc / 4)) {
473 fp
->hdr
[14] |= (1 << (a
- 0x280 / 4)) & 0x03ff0000;
475 if (info
->in
[i
].slot
[c
] < (0x040 / 4) ||
476 info
->in
[i
].slot
[c
] > (0x380 / 4))
479 if (info
->in
[i
].slot
[0] >= (0x300 / 4))
481 fp
->hdr
[4 + a
/ 32] |= m
<< (a
% 32);
486 for (i
= 0; i
< info
->numOutputs
; ++i
) {
487 if (info
->out
[i
].sn
== TGSI_SEMANTIC_COLOR
)
488 fp
->hdr
[18] |= info
->out
[i
].mask
<< info
->out
[i
].slot
[0];
491 fp
->fp
.early_z
= info
->prop
.fp
.earlyFragTests
;
496 static struct nvc0_transform_feedback_state
*
497 nvc0_program_create_tfb_state(const struct nv50_ir_prog_info
*info
,
498 const struct pipe_stream_output_info
*pso
)
500 struct nvc0_transform_feedback_state
*tfb
;
503 tfb
= MALLOC_STRUCT(nvc0_transform_feedback_state
);
506 for (b
= 0; b
< 4; ++b
) {
507 tfb
->stride
[b
] = pso
->stride
[b
] * 4;
508 tfb
->varying_count
[b
] = 0;
510 memset(tfb
->varying_index
, 0xff, sizeof(tfb
->varying_index
)); /* = skip */
512 for (i
= 0; i
< pso
->num_outputs
; ++i
) {
513 unsigned s
= pso
->output
[i
].start_component
;
514 unsigned p
= pso
->output
[i
].dst_offset
;
515 b
= pso
->output
[i
].output_buffer
;
517 for (c
= 0; c
< pso
->output
[i
].num_components
; ++c
)
518 tfb
->varying_index
[b
][p
++] =
519 info
->out
[pso
->output
[i
].register_index
].slot
[s
+ c
];
521 tfb
->varying_count
[b
] = MAX2(tfb
->varying_count
[b
], p
);
523 for (b
= 0; b
< 4; ++b
) // zero unused indices (looks nicer)
524 for (c
= tfb
->varying_count
[b
]; c
& 3; ++c
)
525 tfb
->varying_index
[b
][c
] = 0;
532 nvc0_program_dump(struct nvc0_program
*prog
)
536 for (pos
= 0; pos
< sizeof(prog
->hdr
) / sizeof(prog
->hdr
[0]); ++pos
)
537 debug_printf("HDR[%02lx] = 0x%08x\n",
538 pos
* sizeof(prog
->hdr
[0]), prog
->hdr
[pos
]);
540 debug_printf("shader binary code (0x%x bytes):", prog
->code_size
);
541 for (pos
= 0; pos
< prog
->code_size
/ 4; ++pos
) {
544 debug_printf("%08x ", prog
->code
[pos
]);
551 nvc0_program_translate(struct nvc0_program
*prog
, uint16_t chipset
)
553 struct nv50_ir_prog_info
*info
;
556 info
= CALLOC_STRUCT(nv50_ir_prog_info
);
560 info
->type
= prog
->type
;
561 info
->target
= chipset
;
562 info
->bin
.sourceRep
= NV50_PROGRAM_IR_TGSI
;
563 info
->bin
.source
= (void *)prog
->pipe
.tokens
;
565 info
->io
.genUserClip
= prog
->vp
.num_ucps
;
566 info
->io
.ucpBase
= 256;
567 info
->io
.ucpBinding
= 15;
569 info
->assignSlots
= nvc0_program_assign_varying_slots
;
572 info
->optLevel
= debug_get_num_option("NV50_PROG_OPTIMIZE", 3);
573 info
->dbgFlags
= debug_get_num_option("NV50_PROG_DEBUG", 0);
578 ret
= nv50_ir_generate_code(info
);
580 NOUVEAU_ERR("shader translation failed: %i\n", ret
);
583 FREE(info
->bin
.syms
);
585 prog
->code
= info
->bin
.code
;
586 prog
->code_size
= info
->bin
.codeSize
;
587 prog
->immd_data
= info
->immd
.buf
;
588 prog
->immd_size
= info
->immd
.bufSize
;
589 prog
->relocs
= info
->bin
.relocData
;
590 prog
->max_gpr
= MAX2(4, (info
->bin
.maxGPR
+ 1));
592 prog
->vp
.need_vertex_id
= info
->io
.vertexId
< PIPE_MAX_SHADER_INPUTS
;
594 if (info
->io
.edgeFlagOut
< PIPE_MAX_ATTRIBS
)
595 info
->out
[info
->io
.edgeFlagOut
].mask
= 0; /* for headergen */
596 prog
->vp
.edgeflag
= info
->io
.edgeFlagIn
;
598 switch (prog
->type
) {
599 case PIPE_SHADER_VERTEX
:
600 ret
= nvc0_vp_gen_header(prog
, info
);
602 #ifdef PIPE_SHADER_HULL
603 case PIPE_SHADER_HULL
:
604 ret
= nvc0_tcp_gen_header(prog
, info
);
607 #ifdef PIPE_SHADER_DOMAIN
608 case PIPE_SHADER_DOMAIN
:
609 ret
= nvc0_tep_gen_header(prog
, info
);
612 case PIPE_SHADER_GEOMETRY
:
613 ret
= nvc0_gp_gen_header(prog
, info
);
615 case PIPE_SHADER_FRAGMENT
:
616 ret
= nvc0_fp_gen_header(prog
, info
);
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
;
658 uint32_t size
= prog
->code_size
+ NVC0_SHADER_HEADER_SIZE
;
659 uint32_t lib_pos
= screen
->lib_code
->start
;
662 /* c[] bindings need to be aligned to 0x100, but we could use relocations
664 if (prog
->immd_size
) {
665 prog
->immd_base
= size
;
666 size
= align(size
, 0x40);
667 size
+= prog
->immd_size
+ 0xc0; /* add 0xc0 for align 0x40 -> 0x100 */
669 /* On Fermi, SP_START_ID must be aligned to 0x40.
670 * On Kepler, the first instruction must be aligned to 0x80 because
671 * latency information is expected only at certain positions.
673 if (screen
->base
.class_3d
>= NVE4_3D_CLASS
)
675 size
= align(size
, 0x40);
677 ret
= nouveau_heap_alloc(screen
->text_heap
, size
, prog
, &prog
->mem
);
679 struct nouveau_heap
*heap
= screen
->text_heap
;
680 struct nouveau_heap
*iter
;
681 for (iter
= heap
; iter
&& iter
->next
!= heap
; iter
= iter
->next
) {
682 struct nvc0_program
*evict
= iter
->priv
;
684 nouveau_heap_free(&evict
->mem
);
686 debug_printf("WARNING: out of code space, evicting all shaders.\n");
687 ret
= nouveau_heap_alloc(heap
, size
, prog
, &prog
->mem
);
689 NOUVEAU_ERR("shader too large (0x%x) to fit in code space ?\n", size
);
692 IMMED_NVC0(nvc0
->base
.pushbuf
, NVC0_3D(SERIALIZE
), 0);
694 prog
->code_base
= prog
->mem
->start
;
695 prog
->immd_base
= align(prog
->mem
->start
+ prog
->immd_base
, 0x100);
696 assert((prog
->immd_size
== 0) || (prog
->immd_base
+ prog
->immd_size
<=
697 prog
->mem
->start
+ prog
->mem
->size
));
699 if (screen
->base
.class_3d
>= NVE4_3D_CLASS
) {
700 switch (prog
->mem
->start
& 0xff) {
701 case 0x40: prog
->code_base
+= 0x70; break;
702 case 0x80: prog
->code_base
+= 0x30; break;
703 case 0xc0: prog
->code_base
+= 0x70; break;
705 prog
->code_base
+= 0x30;
706 assert((prog
->mem
->start
& 0xff) == 0x00);
710 code_pos
= prog
->code_base
+ NVC0_SHADER_HEADER_SIZE
;
713 nv50_ir_relocate_code(prog
->relocs
, prog
->code
, code_pos
, lib_pos
, 0);
716 if (debug_get_bool_option("NV50_PROG_DEBUG", FALSE
))
717 nvc0_program_dump(prog
);
720 nvc0
->base
.push_data(&nvc0
->base
, screen
->text
, prog
->code_base
,
721 NOUVEAU_BO_VRAM
, NVC0_SHADER_HEADER_SIZE
, prog
->hdr
);
722 nvc0
->base
.push_data(&nvc0
->base
, screen
->text
,
723 prog
->code_base
+ NVC0_SHADER_HEADER_SIZE
,
724 NOUVEAU_BO_VRAM
, prog
->code_size
, prog
->code
);
726 nvc0
->base
.push_data(&nvc0
->base
,
727 screen
->text
, prog
->immd_base
, NOUVEAU_BO_VRAM
,
728 prog
->immd_size
, prog
->immd_data
);
730 BEGIN_NVC0(nvc0
->base
.pushbuf
, NVC0_3D(MEM_BARRIER
), 1);
731 PUSH_DATA (nvc0
->base
.pushbuf
, 0x1011);
736 /* Upload code for builtin functions like integer division emulation. */
738 nvc0_program_library_upload(struct nvc0_context
*nvc0
)
740 struct nvc0_screen
*screen
= nvc0
->screen
;
743 const uint32_t *code
;
745 if (screen
->lib_code
)
748 nv50_ir_get_target_library(screen
->base
.device
->chipset
, &code
, &size
);
752 ret
= nouveau_heap_alloc(screen
->text_heap
, align(size
, 0x100), NULL
,
757 nvc0
->base
.push_data(&nvc0
->base
,
758 screen
->text
, screen
->lib_code
->start
, NOUVEAU_BO_VRAM
,
760 /* no need for a memory barrier, will be emitted with first program */
764 nvc0_program_destroy(struct nvc0_context
*nvc0
, struct nvc0_program
*prog
)
766 const struct pipe_shader_state pipe
= prog
->pipe
;
767 const ubyte type
= prog
->type
;
770 nouveau_heap_free(&prog
->mem
);
773 FREE(prog
->immd_data
);
776 if (nvc0
->state
.tfb
== prog
->tfb
)
777 nvc0
->state
.tfb
= NULL
;
781 memset(prog
, 0, sizeof(*prog
));