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 "util/u_format.h"
24 #include "util/u_format_s3tc.h"
25 #include "pipe/p_screen.h"
27 #include "nv50/nv50_context.h"
28 #include "nv50/nv50_screen.h"
30 #include "nouveau_vp3_video.h"
32 #include "nv_object.xml.h"
35 #ifndef NOUVEAU_GETPARAM_GRAPH_UNITS
36 # define NOUVEAU_GETPARAM_GRAPH_UNITS 13
39 /* affected by LOCAL_WARPS_LOG_ALLOC / LOCAL_WARPS_NO_CLAMP */
40 #define LOCAL_WARPS_ALLOC 32
41 /* affected by STACK_WARPS_LOG_ALLOC / STACK_WARPS_NO_CLAMP */
42 #define STACK_WARPS_ALLOC 32
44 #define THREADS_IN_WARP 32
46 #define ONE_TEMP_SIZE (4/*vector*/ * sizeof(float))
49 nv50_screen_is_format_supported(struct pipe_screen
*pscreen
,
50 enum pipe_format format
,
51 enum pipe_texture_target target
,
52 unsigned sample_count
,
57 if (!(0x117 & (1 << sample_count
))) /* 0, 1, 2, 4 or 8 */
59 if (sample_count
== 8 && util_format_get_blocksizebits(format
) >= 128)
62 if (!util_format_is_supported(format
, bindings
))
66 case PIPE_FORMAT_Z16_UNORM
:
67 if (nv50_screen(pscreen
)->tesla
->oclass
< NVA0_3D_CLASS
)
74 /* transfers & shared are always supported */
75 bindings
&= ~(PIPE_BIND_TRANSFER_READ
|
76 PIPE_BIND_TRANSFER_WRITE
|
79 return (nv50_format_table
[format
].usage
& bindings
) == bindings
;
83 nv50_screen_get_param(struct pipe_screen
*pscreen
, enum pipe_cap param
)
85 const uint16_t class_3d
= nouveau_screen(pscreen
)->class_3d
;
88 case PIPE_CAP_MAX_TEXTURE_2D_LEVELS
:
90 case PIPE_CAP_MAX_TEXTURE_3D_LEVELS
:
92 case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS
:
94 case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS
:
96 case PIPE_CAP_MIN_TEXTURE_GATHER_OFFSET
:
97 case PIPE_CAP_MIN_TEXEL_OFFSET
:
99 case PIPE_CAP_MAX_TEXTURE_GATHER_OFFSET
:
100 case PIPE_CAP_MAX_TEXEL_OFFSET
:
102 case PIPE_CAP_TEXTURE_MIRROR_CLAMP
:
103 case PIPE_CAP_TEXTURE_SWIZZLE
:
104 case PIPE_CAP_TEXTURE_SHADOW_MAP
:
105 case PIPE_CAP_NPOT_TEXTURES
:
106 case PIPE_CAP_MIXED_FRAMEBUFFER_SIZES
:
107 case PIPE_CAP_ANISOTROPIC_FILTER
:
108 case PIPE_CAP_TEXTURE_BUFFER_OBJECTS
:
109 case PIPE_CAP_BUFFER_MAP_PERSISTENT_COHERENT
:
111 case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE
:
113 case PIPE_CAP_SEAMLESS_CUBE_MAP
:
114 return 1; /* nv50_screen(pscreen)->tesla->oclass >= NVA0_3D_CLASS; */
115 case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE
:
117 case PIPE_CAP_CUBE_MAP_ARRAY
:
118 return nv50_screen(pscreen
)->tesla
->oclass
>= NVA3_3D_CLASS
;
119 case PIPE_CAP_TWO_SIDED_STENCIL
:
120 case PIPE_CAP_DEPTH_CLIP_DISABLE
:
121 case PIPE_CAP_POINT_SPRITE
:
125 case PIPE_CAP_GLSL_FEATURE_LEVEL
:
127 case PIPE_CAP_MAX_RENDER_TARGETS
:
129 case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS
:
131 case PIPE_CAP_FRAGMENT_COLOR_CLAMPED
:
132 case PIPE_CAP_VERTEX_COLOR_UNCLAMPED
:
133 case PIPE_CAP_VERTEX_COLOR_CLAMPED
:
135 case PIPE_CAP_QUERY_TIMESTAMP
:
136 case PIPE_CAP_QUERY_TIME_ELAPSED
:
137 case PIPE_CAP_OCCLUSION_QUERY
:
139 case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS
:
141 case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS
:
142 case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS
:
144 case PIPE_CAP_MAX_GEOMETRY_OUTPUT_VERTICES
:
145 case PIPE_CAP_MAX_GEOMETRY_TOTAL_OUTPUT_COMPONENTS
:
147 case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME
:
148 return (class_3d
>= NVA0_3D_CLASS
) ? 1 : 0;
149 case PIPE_CAP_BLEND_EQUATION_SEPARATE
:
150 case PIPE_CAP_INDEP_BLEND_ENABLE
:
152 case PIPE_CAP_INDEP_BLEND_FUNC
:
153 return nv50_screen(pscreen
)->tesla
->oclass
>= NVA3_3D_CLASS
;
154 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT
:
155 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER
:
157 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT
:
158 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER
:
160 case PIPE_CAP_SHADER_STENCIL_EXPORT
:
162 case PIPE_CAP_PRIMITIVE_RESTART
:
163 case PIPE_CAP_TGSI_INSTANCEID
:
164 case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR
:
165 case PIPE_CAP_MIXED_COLORBUFFER_FORMATS
:
166 case PIPE_CAP_CONDITIONAL_RENDER
:
167 case PIPE_CAP_TEXTURE_BARRIER
:
168 case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION
:
169 case PIPE_CAP_START_INSTANCE
:
171 case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS
:
172 return 0; /* state trackers will know better */
173 case PIPE_CAP_USER_CONSTANT_BUFFERS
:
174 case PIPE_CAP_USER_INDEX_BUFFERS
:
175 case PIPE_CAP_USER_VERTEX_BUFFERS
:
177 case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT
:
179 case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT
:
180 return 1; /* 256 for binding as RT, but that's not possible in GL */
181 case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT
:
182 return NOUVEAU_MIN_BUFFER_MAP_ALIGN
;
183 case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY
:
184 case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY
:
185 case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY
:
186 case PIPE_CAP_TGSI_TEXCOORD
:
188 case PIPE_CAP_TEXTURE_MULTISAMPLE
:
190 case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER
:
192 case PIPE_CAP_QUERY_PIPELINE_STATISTICS
:
194 case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK
:
195 return PIPE_QUIRK_TEXTURE_BORDER_COLOR_SWIZZLE_NV50
;
196 case PIPE_CAP_ENDIANNESS
:
197 return PIPE_ENDIAN_LITTLE
;
198 case PIPE_CAP_TGSI_VS_LAYER
:
199 case PIPE_CAP_TEXTURE_GATHER_SM5
:
200 case PIPE_CAP_FAKE_SW_MSAA
:
201 case PIPE_CAP_TEXTURE_GATHER_OFFSETS
:
203 case PIPE_CAP_MAX_VIEWPORTS
:
204 return NV50_MAX_VIEWPORTS
;
205 case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS
:
206 return (class_3d
>= NVA3_3D_CLASS
) ? 4 : 0;
207 case PIPE_CAP_TEXTURE_QUERY_LOD
:
208 case PIPE_CAP_SAMPLE_SHADING
:
209 return class_3d
>= NVA3_3D_CLASS
;
211 NOUVEAU_ERR("unknown PIPE_CAP %d\n", param
);
217 nv50_screen_get_shader_param(struct pipe_screen
*pscreen
, unsigned shader
,
218 enum pipe_shader_cap param
)
221 case PIPE_SHADER_VERTEX
:
222 case PIPE_SHADER_GEOMETRY
:
223 case PIPE_SHADER_FRAGMENT
:
230 case PIPE_SHADER_CAP_MAX_INSTRUCTIONS
:
231 case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS
:
232 case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS
:
233 case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS
:
235 case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH
:
237 case PIPE_SHADER_CAP_MAX_INPUTS
:
238 if (shader
== PIPE_SHADER_VERTEX
)
241 case PIPE_SHADER_CAP_MAX_CONSTS
:
243 case PIPE_SHADER_CAP_MAX_CONST_BUFFERS
:
244 return NV50_MAX_PIPE_CONSTBUFS
;
245 case PIPE_SHADER_CAP_MAX_ADDRS
:
247 case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR
:
248 case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR
:
249 return shader
!= PIPE_SHADER_FRAGMENT
;
250 case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR
:
251 case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR
:
253 case PIPE_SHADER_CAP_MAX_PREDS
:
255 case PIPE_SHADER_CAP_MAX_TEMPS
:
256 return nv50_screen(pscreen
)->max_tls_space
/ ONE_TEMP_SIZE
;
257 case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED
:
259 case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED
:
261 case PIPE_SHADER_CAP_SUBROUTINES
:
262 return 0; /* please inline, or provide function declarations */
263 case PIPE_SHADER_CAP_INTEGERS
:
265 case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS
:
266 /* The chip could handle more sampler views than samplers */
267 case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS
:
268 return MIN2(32, PIPE_MAX_SAMPLERS
);
270 NOUVEAU_ERR("unknown PIPE_SHADER_CAP %d\n", param
);
276 nv50_screen_get_paramf(struct pipe_screen
*pscreen
, enum pipe_capf param
)
279 case PIPE_CAPF_MAX_LINE_WIDTH
:
280 case PIPE_CAPF_MAX_LINE_WIDTH_AA
:
282 case PIPE_CAPF_MAX_POINT_WIDTH
:
283 case PIPE_CAPF_MAX_POINT_WIDTH_AA
:
285 case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY
:
287 case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS
:
290 NOUVEAU_ERR("unknown PIPE_CAP %d\n", param
);
296 nv50_screen_destroy(struct pipe_screen
*pscreen
)
298 struct nv50_screen
*screen
= nv50_screen(pscreen
);
300 if (!nouveau_drm_screen_unref(&screen
->base
))
303 if (screen
->base
.fence
.current
) {
304 struct nouveau_fence
*current
= NULL
;
306 /* nouveau_fence_wait will create a new current fence, so wait on the
307 * _current_ one, and remove both.
309 nouveau_fence_ref(screen
->base
.fence
.current
, ¤t
);
310 nouveau_fence_wait(current
);
311 nouveau_fence_ref(NULL
, ¤t
);
312 nouveau_fence_ref(NULL
, &screen
->base
.fence
.current
);
314 if (screen
->base
.pushbuf
)
315 screen
->base
.pushbuf
->user_priv
= NULL
;
318 nv50_blitter_destroy(screen
);
320 nouveau_bo_ref(NULL
, &screen
->code
);
321 nouveau_bo_ref(NULL
, &screen
->tls_bo
);
322 nouveau_bo_ref(NULL
, &screen
->stack_bo
);
323 nouveau_bo_ref(NULL
, &screen
->txc
);
324 nouveau_bo_ref(NULL
, &screen
->uniforms
);
325 nouveau_bo_ref(NULL
, &screen
->fence
.bo
);
327 nouveau_heap_destroy(&screen
->vp_code_heap
);
328 nouveau_heap_destroy(&screen
->gp_code_heap
);
329 nouveau_heap_destroy(&screen
->fp_code_heap
);
331 FREE(screen
->tic
.entries
);
333 nouveau_object_del(&screen
->tesla
);
334 nouveau_object_del(&screen
->eng2d
);
335 nouveau_object_del(&screen
->m2mf
);
336 nouveau_object_del(&screen
->sync
);
338 nouveau_screen_fini(&screen
->base
);
344 nv50_screen_fence_emit(struct pipe_screen
*pscreen
, u32
*sequence
)
346 struct nv50_screen
*screen
= nv50_screen(pscreen
);
347 struct nouveau_pushbuf
*push
= screen
->base
.pushbuf
;
349 /* we need to do it after possible flush in MARK_RING */
350 *sequence
= ++screen
->base
.fence
.sequence
;
352 PUSH_DATA (push
, NV50_FIFO_PKHDR(NV50_3D(QUERY_ADDRESS_HIGH
), 4));
353 PUSH_DATAh(push
, screen
->fence
.bo
->offset
);
354 PUSH_DATA (push
, screen
->fence
.bo
->offset
);
355 PUSH_DATA (push
, *sequence
);
356 PUSH_DATA (push
, NV50_3D_QUERY_GET_MODE_WRITE_UNK0
|
357 NV50_3D_QUERY_GET_UNK4
|
358 NV50_3D_QUERY_GET_UNIT_CROP
|
359 NV50_3D_QUERY_GET_TYPE_QUERY
|
360 NV50_3D_QUERY_GET_QUERY_SELECT_ZERO
|
361 NV50_3D_QUERY_GET_SHORT
);
365 nv50_screen_fence_update(struct pipe_screen
*pscreen
)
367 return nv50_screen(pscreen
)->fence
.map
[0];
371 nv50_screen_init_hwctx(struct nv50_screen
*screen
)
373 struct nouveau_pushbuf
*push
= screen
->base
.pushbuf
;
374 struct nv04_fifo
*fifo
;
377 fifo
= (struct nv04_fifo
*)screen
->base
.channel
->data
;
379 BEGIN_NV04(push
, SUBC_M2MF(NV01_SUBCHAN_OBJECT
), 1);
380 PUSH_DATA (push
, screen
->m2mf
->handle
);
381 BEGIN_NV04(push
, SUBC_M2MF(NV03_M2MF_DMA_NOTIFY
), 3);
382 PUSH_DATA (push
, screen
->sync
->handle
);
383 PUSH_DATA (push
, fifo
->vram
);
384 PUSH_DATA (push
, fifo
->vram
);
386 BEGIN_NV04(push
, SUBC_2D(NV01_SUBCHAN_OBJECT
), 1);
387 PUSH_DATA (push
, screen
->eng2d
->handle
);
388 BEGIN_NV04(push
, NV50_2D(DMA_NOTIFY
), 4);
389 PUSH_DATA (push
, screen
->sync
->handle
);
390 PUSH_DATA (push
, fifo
->vram
);
391 PUSH_DATA (push
, fifo
->vram
);
392 PUSH_DATA (push
, fifo
->vram
);
393 BEGIN_NV04(push
, NV50_2D(OPERATION
), 1);
394 PUSH_DATA (push
, NV50_2D_OPERATION_SRCCOPY
);
395 BEGIN_NV04(push
, NV50_2D(CLIP_ENABLE
), 1);
397 BEGIN_NV04(push
, NV50_2D(COLOR_KEY_ENABLE
), 1);
399 BEGIN_NV04(push
, SUBC_2D(0x0888), 1);
401 BEGIN_NV04(push
, NV50_2D(COND_MODE
), 1);
402 PUSH_DATA (push
, NV50_2D_COND_MODE_ALWAYS
);
404 BEGIN_NV04(push
, SUBC_3D(NV01_SUBCHAN_OBJECT
), 1);
405 PUSH_DATA (push
, screen
->tesla
->handle
);
407 BEGIN_NV04(push
, NV50_3D(COND_MODE
), 1);
408 PUSH_DATA (push
, NV50_3D_COND_MODE_ALWAYS
);
410 BEGIN_NV04(push
, NV50_3D(DMA_NOTIFY
), 1);
411 PUSH_DATA (push
, screen
->sync
->handle
);
412 BEGIN_NV04(push
, NV50_3D(DMA_ZETA
), 11);
413 for (i
= 0; i
< 11; ++i
)
414 PUSH_DATA(push
, fifo
->vram
);
415 BEGIN_NV04(push
, NV50_3D(DMA_COLOR(0)), NV50_3D_DMA_COLOR__LEN
);
416 for (i
= 0; i
< NV50_3D_DMA_COLOR__LEN
; ++i
)
417 PUSH_DATA(push
, fifo
->vram
);
419 BEGIN_NV04(push
, NV50_3D(REG_MODE
), 1);
420 PUSH_DATA (push
, NV50_3D_REG_MODE_STRIPED
);
421 BEGIN_NV04(push
, NV50_3D(UNK1400_LANES
), 1);
422 PUSH_DATA (push
, 0xf);
424 if (debug_get_bool_option("NOUVEAU_SHADER_WATCHDOG", TRUE
)) {
425 BEGIN_NV04(push
, NV50_3D(WATCHDOG_TIMER
), 1);
426 PUSH_DATA (push
, 0x18);
429 BEGIN_NV04(push
, NV50_3D(RT_CONTROL
), 1);
432 BEGIN_NV04(push
, NV50_3D(CSAA_ENABLE
), 1);
434 BEGIN_NV04(push
, NV50_3D(MULTISAMPLE_ENABLE
), 1);
436 BEGIN_NV04(push
, NV50_3D(MULTISAMPLE_MODE
), 1);
437 PUSH_DATA (push
, NV50_3D_MULTISAMPLE_MODE_MS1
);
438 BEGIN_NV04(push
, NV50_3D(MULTISAMPLE_CTRL
), 1);
440 BEGIN_NV04(push
, NV50_3D(LINE_LAST_PIXEL
), 1);
442 BEGIN_NV04(push
, NV50_3D(BLEND_SEPARATE_ALPHA
), 1);
445 if (screen
->tesla
->oclass
>= NVA0_3D_CLASS
) {
446 BEGIN_NV04(push
, SUBC_3D(NVA0_3D_TEX_MISC
), 1);
447 PUSH_DATA (push
, NVA0_3D_TEX_MISC_SEAMLESS_CUBE_MAP
);
450 BEGIN_NV04(push
, NV50_3D(SCREEN_Y_CONTROL
), 1);
452 BEGIN_NV04(push
, NV50_3D(WINDOW_OFFSET_X
), 2);
455 BEGIN_NV04(push
, NV50_3D(ZCULL_REGION
), 1);
456 PUSH_DATA (push
, 0x3f);
458 BEGIN_NV04(push
, NV50_3D(VP_ADDRESS_HIGH
), 2);
459 PUSH_DATAh(push
, screen
->code
->offset
+ (0 << NV50_CODE_BO_SIZE_LOG2
));
460 PUSH_DATA (push
, screen
->code
->offset
+ (0 << NV50_CODE_BO_SIZE_LOG2
));
462 BEGIN_NV04(push
, NV50_3D(FP_ADDRESS_HIGH
), 2);
463 PUSH_DATAh(push
, screen
->code
->offset
+ (1 << NV50_CODE_BO_SIZE_LOG2
));
464 PUSH_DATA (push
, screen
->code
->offset
+ (1 << NV50_CODE_BO_SIZE_LOG2
));
466 BEGIN_NV04(push
, NV50_3D(GP_ADDRESS_HIGH
), 2);
467 PUSH_DATAh(push
, screen
->code
->offset
+ (2 << NV50_CODE_BO_SIZE_LOG2
));
468 PUSH_DATA (push
, screen
->code
->offset
+ (2 << NV50_CODE_BO_SIZE_LOG2
));
470 BEGIN_NV04(push
, NV50_3D(LOCAL_ADDRESS_HIGH
), 3);
471 PUSH_DATAh(push
, screen
->tls_bo
->offset
);
472 PUSH_DATA (push
, screen
->tls_bo
->offset
);
473 PUSH_DATA (push
, util_logbase2(screen
->cur_tls_space
/ 8));
475 BEGIN_NV04(push
, NV50_3D(STACK_ADDRESS_HIGH
), 3);
476 PUSH_DATAh(push
, screen
->stack_bo
->offset
);
477 PUSH_DATA (push
, screen
->stack_bo
->offset
);
480 BEGIN_NV04(push
, NV50_3D(CB_DEF_ADDRESS_HIGH
), 3);
481 PUSH_DATAh(push
, screen
->uniforms
->offset
+ (0 << 16));
482 PUSH_DATA (push
, screen
->uniforms
->offset
+ (0 << 16));
483 PUSH_DATA (push
, (NV50_CB_PVP
<< 16) | 0x0000);
485 BEGIN_NV04(push
, NV50_3D(CB_DEF_ADDRESS_HIGH
), 3);
486 PUSH_DATAh(push
, screen
->uniforms
->offset
+ (1 << 16));
487 PUSH_DATA (push
, screen
->uniforms
->offset
+ (1 << 16));
488 PUSH_DATA (push
, (NV50_CB_PGP
<< 16) | 0x0000);
490 BEGIN_NV04(push
, NV50_3D(CB_DEF_ADDRESS_HIGH
), 3);
491 PUSH_DATAh(push
, screen
->uniforms
->offset
+ (2 << 16));
492 PUSH_DATA (push
, screen
->uniforms
->offset
+ (2 << 16));
493 PUSH_DATA (push
, (NV50_CB_PFP
<< 16) | 0x0000);
495 BEGIN_NV04(push
, NV50_3D(CB_DEF_ADDRESS_HIGH
), 3);
496 PUSH_DATAh(push
, screen
->uniforms
->offset
+ (3 << 16));
497 PUSH_DATA (push
, screen
->uniforms
->offset
+ (3 << 16));
498 PUSH_DATA (push
, (NV50_CB_AUX
<< 16) | (NV50_CB_AUX_SIZE
& 0xffff));
500 BEGIN_NI04(push
, NV50_3D(SET_PROGRAM_CB
), 3);
501 PUSH_DATA (push
, (NV50_CB_AUX
<< 12) | 0xf01);
502 PUSH_DATA (push
, (NV50_CB_AUX
<< 12) | 0xf21);
503 PUSH_DATA (push
, (NV50_CB_AUX
<< 12) | 0xf31);
505 /* return { 0.0, 0.0, 0.0, 0.0 } on out-of-bounds vtxbuf access */
506 BEGIN_NV04(push
, NV50_3D(CB_ADDR
), 1);
507 PUSH_DATA (push
, (NV50_CB_AUX_RUNOUT_OFFSET
<< (8 - 2)) | NV50_CB_AUX
);
508 BEGIN_NI04(push
, NV50_3D(CB_DATA(0)), 4);
509 PUSH_DATAf(push
, 0.0f
);
510 PUSH_DATAf(push
, 0.0f
);
511 PUSH_DATAf(push
, 0.0f
);
512 PUSH_DATAf(push
, 0.0f
);
513 BEGIN_NV04(push
, NV50_3D(VERTEX_RUNOUT_ADDRESS_HIGH
), 2);
514 PUSH_DATAh(push
, screen
->uniforms
->offset
+ (3 << 16) + NV50_CB_AUX_RUNOUT_OFFSET
);
515 PUSH_DATA (push
, screen
->uniforms
->offset
+ (3 << 16) + NV50_CB_AUX_RUNOUT_OFFSET
);
517 nv50_upload_ms_info(push
);
519 /* max TIC (bits 4:8) & TSC bindings, per program type */
520 for (i
= 0; i
< 3; ++i
) {
521 BEGIN_NV04(push
, NV50_3D(TEX_LIMITS(i
)), 1);
522 PUSH_DATA (push
, 0x54);
525 BEGIN_NV04(push
, NV50_3D(TIC_ADDRESS_HIGH
), 3);
526 PUSH_DATAh(push
, screen
->txc
->offset
);
527 PUSH_DATA (push
, screen
->txc
->offset
);
528 PUSH_DATA (push
, NV50_TIC_MAX_ENTRIES
- 1);
530 BEGIN_NV04(push
, NV50_3D(TSC_ADDRESS_HIGH
), 3);
531 PUSH_DATAh(push
, screen
->txc
->offset
+ 65536);
532 PUSH_DATA (push
, screen
->txc
->offset
+ 65536);
533 PUSH_DATA (push
, NV50_TSC_MAX_ENTRIES
- 1);
535 BEGIN_NV04(push
, NV50_3D(LINKED_TSC
), 1);
538 BEGIN_NV04(push
, NV50_3D(CLIP_RECTS_EN
), 1);
540 BEGIN_NV04(push
, NV50_3D(CLIP_RECTS_MODE
), 1);
541 PUSH_DATA (push
, NV50_3D_CLIP_RECTS_MODE_INSIDE_ANY
);
542 BEGIN_NV04(push
, NV50_3D(CLIP_RECT_HORIZ(0)), 8 * 2);
543 for (i
= 0; i
< 8 * 2; ++i
)
545 BEGIN_NV04(push
, NV50_3D(CLIPID_ENABLE
), 1);
548 BEGIN_NV04(push
, NV50_3D(VIEWPORT_TRANSFORM_EN
), 1);
550 for (i
= 0; i
< NV50_MAX_VIEWPORTS
; i
++) {
551 BEGIN_NV04(push
, NV50_3D(DEPTH_RANGE_NEAR(i
)), 2);
552 PUSH_DATAf(push
, 0.0f
);
553 PUSH_DATAf(push
, 1.0f
);
554 BEGIN_NV04(push
, NV50_3D(VIEWPORT_HORIZ(i
)), 2);
555 PUSH_DATA (push
, 8192 << 16);
556 PUSH_DATA (push
, 8192 << 16);
559 BEGIN_NV04(push
, NV50_3D(VIEW_VOLUME_CLIP_CTRL
), 1);
560 #ifdef NV50_SCISSORS_CLIPPING
561 PUSH_DATA (push
, 0x0000);
563 PUSH_DATA (push
, 0x1080);
566 BEGIN_NV04(push
, NV50_3D(CLEAR_FLAGS
), 1);
567 PUSH_DATA (push
, NV50_3D_CLEAR_FLAGS_CLEAR_RECT_VIEWPORT
);
569 /* We use scissors instead of exact view volume clipping,
570 * so they're always enabled.
572 for (i
= 0; i
< NV50_MAX_VIEWPORTS
; i
++) {
573 BEGIN_NV04(push
, NV50_3D(SCISSOR_ENABLE(i
)), 3);
575 PUSH_DATA (push
, 8192 << 16);
576 PUSH_DATA (push
, 8192 << 16);
579 BEGIN_NV04(push
, NV50_3D(RASTERIZE_ENABLE
), 1);
581 BEGIN_NV04(push
, NV50_3D(POINT_RASTER_RULES
), 1);
582 PUSH_DATA (push
, NV50_3D_POINT_RASTER_RULES_OGL
);
583 BEGIN_NV04(push
, NV50_3D(FRAG_COLOR_CLAMP_EN
), 1);
584 PUSH_DATA (push
, 0x11111111);
585 BEGIN_NV04(push
, NV50_3D(EDGEFLAG
), 1);
591 static int nv50_tls_alloc(struct nv50_screen
*screen
, unsigned tls_space
,
594 struct nouveau_device
*dev
= screen
->base
.device
;
597 screen
->cur_tls_space
= util_next_power_of_two(tls_space
/ ONE_TEMP_SIZE
) *
599 if (nouveau_mesa_debug
)
600 debug_printf("allocating space for %u temps\n",
601 util_next_power_of_two(tls_space
/ ONE_TEMP_SIZE
));
602 *tls_size
= screen
->cur_tls_space
* util_next_power_of_two(screen
->TPs
) *
603 screen
->MPsInTP
* LOCAL_WARPS_ALLOC
* THREADS_IN_WARP
;
605 ret
= nouveau_bo_new(dev
, NOUVEAU_BO_VRAM
, 1 << 16,
606 *tls_size
, NULL
, &screen
->tls_bo
);
608 NOUVEAU_ERR("Failed to allocate local bo: %d\n", ret
);
615 int nv50_tls_realloc(struct nv50_screen
*screen
, unsigned tls_space
)
617 struct nouveau_pushbuf
*push
= screen
->base
.pushbuf
;
621 if (tls_space
< screen
->cur_tls_space
)
623 if (tls_space
> screen
->max_tls_space
) {
624 /* fixable by limiting number of warps (LOCAL_WARPS_LOG_ALLOC /
625 * LOCAL_WARPS_NO_CLAMP) */
626 NOUVEAU_ERR("Unsupported number of temporaries (%u > %u). Fixable if someone cares.\n",
627 (unsigned)(tls_space
/ ONE_TEMP_SIZE
),
628 (unsigned)(screen
->max_tls_space
/ ONE_TEMP_SIZE
));
632 nouveau_bo_ref(NULL
, &screen
->tls_bo
);
633 ret
= nv50_tls_alloc(screen
, tls_space
, &tls_size
);
637 BEGIN_NV04(push
, NV50_3D(LOCAL_ADDRESS_HIGH
), 3);
638 PUSH_DATAh(push
, screen
->tls_bo
->offset
);
639 PUSH_DATA (push
, screen
->tls_bo
->offset
);
640 PUSH_DATA (push
, util_logbase2(screen
->cur_tls_space
/ 8));
646 nv50_screen_create(struct nouveau_device
*dev
)
648 struct nv50_screen
*screen
;
649 struct pipe_screen
*pscreen
;
650 struct nouveau_object
*chan
;
652 uint32_t tesla_class
;
656 screen
= CALLOC_STRUCT(nv50_screen
);
659 pscreen
= &screen
->base
.base
;
661 ret
= nouveau_screen_init(&screen
->base
, dev
);
663 NOUVEAU_ERR("nouveau_screen_init failed: %d\n", ret
);
667 /* TODO: Prevent FIFO prefetch before transfer of index buffers and
668 * admit them to VRAM.
670 screen
->base
.vidmem_bindings
|= PIPE_BIND_CONSTANT_BUFFER
|
671 PIPE_BIND_VERTEX_BUFFER
;
672 screen
->base
.sysmem_bindings
|=
673 PIPE_BIND_VERTEX_BUFFER
| PIPE_BIND_INDEX_BUFFER
;
675 screen
->base
.pushbuf
->user_priv
= screen
;
676 screen
->base
.pushbuf
->rsvd_kick
= 5;
678 chan
= screen
->base
.channel
;
680 pscreen
->destroy
= nv50_screen_destroy
;
681 pscreen
->context_create
= nv50_create
;
682 pscreen
->is_format_supported
= nv50_screen_is_format_supported
;
683 pscreen
->get_param
= nv50_screen_get_param
;
684 pscreen
->get_shader_param
= nv50_screen_get_shader_param
;
685 pscreen
->get_paramf
= nv50_screen_get_paramf
;
687 nv50_screen_init_resource_functions(pscreen
);
689 if (screen
->base
.device
->chipset
< 0x84 ||
690 debug_get_bool_option("NOUVEAU_PMPEG", FALSE
)) {
692 nouveau_screen_init_vdec(&screen
->base
);
693 } else if (screen
->base
.device
->chipset
< 0x98 ||
694 screen
->base
.device
->chipset
== 0xa0) {
696 screen
->base
.base
.get_video_param
= nv84_screen_get_video_param
;
697 screen
->base
.base
.is_video_format_supported
= nv84_screen_video_supported
;
700 screen
->base
.base
.get_video_param
= nouveau_vp3_screen_get_video_param
;
701 screen
->base
.base
.is_video_format_supported
= nouveau_vp3_screen_video_supported
;
704 ret
= nouveau_bo_new(dev
, NOUVEAU_BO_GART
| NOUVEAU_BO_MAP
, 0, 4096,
705 NULL
, &screen
->fence
.bo
);
707 NOUVEAU_ERR("Failed to allocate fence bo: %d\n", ret
);
711 nouveau_bo_map(screen
->fence
.bo
, 0, NULL
);
712 screen
->fence
.map
= screen
->fence
.bo
->map
;
713 screen
->base
.fence
.emit
= nv50_screen_fence_emit
;
714 screen
->base
.fence
.update
= nv50_screen_fence_update
;
716 ret
= nouveau_object_new(chan
, 0xbeef0301, NOUVEAU_NOTIFIER_CLASS
,
717 &(struct nv04_notify
){ .length
= 32 },
718 sizeof(struct nv04_notify
), &screen
->sync
);
720 NOUVEAU_ERR("Failed to allocate notifier: %d\n", ret
);
724 ret
= nouveau_object_new(chan
, 0xbeef5039, NV50_M2MF_CLASS
,
725 NULL
, 0, &screen
->m2mf
);
727 NOUVEAU_ERR("Failed to allocate PGRAPH context for M2MF: %d\n", ret
);
731 ret
= nouveau_object_new(chan
, 0xbeef502d, NV50_2D_CLASS
,
732 NULL
, 0, &screen
->eng2d
);
734 NOUVEAU_ERR("Failed to allocate PGRAPH context for 2D: %d\n", ret
);
738 switch (dev
->chipset
& 0xf0) {
740 tesla_class
= NV50_3D_CLASS
;
744 tesla_class
= NV84_3D_CLASS
;
747 switch (dev
->chipset
) {
751 tesla_class
= NVA0_3D_CLASS
;
754 tesla_class
= NVAF_3D_CLASS
;
757 tesla_class
= NVA3_3D_CLASS
;
762 NOUVEAU_ERR("Not a known NV50 chipset: NV%02x\n", dev
->chipset
);
765 screen
->base
.class_3d
= tesla_class
;
767 ret
= nouveau_object_new(chan
, 0xbeef5097, tesla_class
,
768 NULL
, 0, &screen
->tesla
);
770 NOUVEAU_ERR("Failed to allocate PGRAPH context for 3D: %d\n", ret
);
774 /* This over-allocates by a page. The GP, which would execute at the end of
775 * the last page, would trigger faults. The going theory is that it
776 * prefetches up to a certain amount.
778 ret
= nouveau_bo_new(dev
, NOUVEAU_BO_VRAM
, 1 << 16,
779 (3 << NV50_CODE_BO_SIZE_LOG2
) + 0x1000,
780 NULL
, &screen
->code
);
782 NOUVEAU_ERR("Failed to allocate code bo: %d\n", ret
);
786 nouveau_heap_init(&screen
->vp_code_heap
, 0, 1 << NV50_CODE_BO_SIZE_LOG2
);
787 nouveau_heap_init(&screen
->gp_code_heap
, 0, 1 << NV50_CODE_BO_SIZE_LOG2
);
788 nouveau_heap_init(&screen
->fp_code_heap
, 0, 1 << NV50_CODE_BO_SIZE_LOG2
);
790 nouveau_getparam(dev
, NOUVEAU_GETPARAM_GRAPH_UNITS
, &value
);
792 screen
->TPs
= util_bitcount(value
& 0xffff);
793 screen
->MPsInTP
= util_bitcount((value
>> 24) & 0xf);
795 stack_size
= util_next_power_of_two(screen
->TPs
) * screen
->MPsInTP
*
796 STACK_WARPS_ALLOC
* 64 * 8;
798 ret
= nouveau_bo_new(dev
, NOUVEAU_BO_VRAM
, 1 << 16, stack_size
, NULL
,
801 NOUVEAU_ERR("Failed to allocate stack bo: %d\n", ret
);
805 uint64_t size_of_one_temp
= util_next_power_of_two(screen
->TPs
) *
806 screen
->MPsInTP
* LOCAL_WARPS_ALLOC
* THREADS_IN_WARP
*
808 screen
->max_tls_space
= dev
->vram_size
/ size_of_one_temp
* ONE_TEMP_SIZE
;
809 screen
->max_tls_space
/= 2; /* half of vram */
811 /* hw can address max 64 KiB */
812 screen
->max_tls_space
= MIN2(screen
->max_tls_space
, 64 << 10);
815 unsigned tls_space
= 4/*temps*/ * ONE_TEMP_SIZE
;
816 ret
= nv50_tls_alloc(screen
, tls_space
, &tls_size
);
820 if (nouveau_mesa_debug
)
821 debug_printf("TPs = %u, MPsInTP = %u, VRAM = %"PRIu64
" MiB, tls_size = %"PRIu64
" KiB\n",
822 screen
->TPs
, screen
->MPsInTP
, dev
->vram_size
>> 20, tls_size
>> 10);
824 ret
= nouveau_bo_new(dev
, NOUVEAU_BO_VRAM
, 1 << 16, 4 << 16, NULL
,
827 NOUVEAU_ERR("Failed to allocate uniforms bo: %d\n", ret
);
831 ret
= nouveau_bo_new(dev
, NOUVEAU_BO_VRAM
, 1 << 16, 3 << 16, NULL
,
834 NOUVEAU_ERR("Failed to allocate TIC/TSC bo: %d\n", ret
);
838 screen
->tic
.entries
= CALLOC(4096, sizeof(void *));
839 screen
->tsc
.entries
= screen
->tic
.entries
+ 2048;
841 if (!nv50_blitter_create(screen
))
844 nv50_screen_init_hwctx(screen
);
846 nouveau_fence_new(&screen
->base
, &screen
->base
.fence
.current
, FALSE
);
851 nv50_screen_destroy(pscreen
);
856 nv50_screen_tic_alloc(struct nv50_screen
*screen
, void *entry
)
858 int i
= screen
->tic
.next
;
860 while (screen
->tic
.lock
[i
/ 32] & (1 << (i
% 32)))
861 i
= (i
+ 1) & (NV50_TIC_MAX_ENTRIES
- 1);
863 screen
->tic
.next
= (i
+ 1) & (NV50_TIC_MAX_ENTRIES
- 1);
865 if (screen
->tic
.entries
[i
])
866 nv50_tic_entry(screen
->tic
.entries
[i
])->id
= -1;
868 screen
->tic
.entries
[i
] = entry
;
873 nv50_screen_tsc_alloc(struct nv50_screen
*screen
, void *entry
)
875 int i
= screen
->tsc
.next
;
877 while (screen
->tsc
.lock
[i
/ 32] & (1 << (i
% 32)))
878 i
= (i
+ 1) & (NV50_TSC_MAX_ENTRIES
- 1);
880 screen
->tsc
.next
= (i
+ 1) & (NV50_TSC_MAX_ENTRIES
- 1);
882 if (screen
->tsc
.entries
[i
])
883 nv50_tsc_entry(screen
->tsc
.entries
[i
])->id
= -1;
885 screen
->tsc
.entries
[i
] = entry
;