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.
25 #include <nouveau_drm.h>
26 #include "util/u_format.h"
27 #include "util/u_format_s3tc.h"
28 #include "pipe/p_screen.h"
30 #include "nv50/nv50_context.h"
31 #include "nv50/nv50_screen.h"
33 #include "nouveau_vp3_video.h"
35 #include "nv_object.xml.h"
37 /* affected by LOCAL_WARPS_LOG_ALLOC / LOCAL_WARPS_NO_CLAMP */
38 #define LOCAL_WARPS_ALLOC 32
39 /* affected by STACK_WARPS_LOG_ALLOC / STACK_WARPS_NO_CLAMP */
40 #define STACK_WARPS_ALLOC 32
42 #define THREADS_IN_WARP 32
44 #define ONE_TEMP_SIZE (4/*vector*/ * sizeof(float))
47 nv50_screen_is_format_supported(struct pipe_screen
*pscreen
,
48 enum pipe_format format
,
49 enum pipe_texture_target target
,
50 unsigned sample_count
,
55 if (!(0x117 & (1 << sample_count
))) /* 0, 1, 2, 4 or 8 */
57 if (sample_count
== 8 && util_format_get_blocksizebits(format
) >= 128)
60 if (!util_format_is_supported(format
, bindings
))
64 case PIPE_FORMAT_Z16_UNORM
:
65 if (nv50_screen(pscreen
)->tesla
->oclass
< NVA0_3D_CLASS
)
72 /* transfers & shared are always supported */
73 bindings
&= ~(PIPE_BIND_TRANSFER_READ
|
74 PIPE_BIND_TRANSFER_WRITE
|
77 return (nv50_format_table
[format
].usage
& bindings
) == bindings
;
81 nv50_screen_get_param(struct pipe_screen
*pscreen
, enum pipe_cap param
)
83 const uint16_t class_3d
= nouveau_screen(pscreen
)->class_3d
;
84 struct nouveau_device
*dev
= nouveau_screen(pscreen
)->device
;
87 /* non-boolean caps */
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_MAX_TEXTURE_BUFFER_SIZE
:
104 case PIPE_CAP_GLSL_FEATURE_LEVEL
:
106 case PIPE_CAP_MAX_RENDER_TARGETS
:
108 case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS
:
110 case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS
:
112 case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS
:
113 case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS
:
115 case PIPE_CAP_MAX_GEOMETRY_OUTPUT_VERTICES
:
116 case PIPE_CAP_MAX_GEOMETRY_TOTAL_OUTPUT_COMPONENTS
:
118 case PIPE_CAP_MAX_VERTEX_STREAMS
:
120 case PIPE_CAP_MAX_VERTEX_ATTRIB_STRIDE
:
122 case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT
:
124 case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT
:
125 return 1; /* 256 for binding as RT, but that's not possible in GL */
126 case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT
:
127 return NOUVEAU_MIN_BUFFER_MAP_ALIGN
;
128 case PIPE_CAP_MAX_VIEWPORTS
:
129 return NV50_MAX_VIEWPORTS
;
130 case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK
:
131 return PIPE_QUIRK_TEXTURE_BORDER_COLOR_SWIZZLE_NV50
;
132 case PIPE_CAP_ENDIANNESS
:
133 return PIPE_ENDIAN_LITTLE
;
134 case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS
:
135 return (class_3d
>= NVA3_3D_CLASS
) ? 4 : 0;
138 case PIPE_CAP_TEXTURE_MIRROR_CLAMP
:
139 case PIPE_CAP_TEXTURE_SWIZZLE
:
140 case PIPE_CAP_TEXTURE_SHADOW_MAP
:
141 case PIPE_CAP_NPOT_TEXTURES
:
142 case PIPE_CAP_MIXED_FRAMEBUFFER_SIZES
:
143 case PIPE_CAP_ANISOTROPIC_FILTER
:
144 case PIPE_CAP_TEXTURE_BUFFER_OBJECTS
:
145 case PIPE_CAP_BUFFER_MAP_PERSISTENT_COHERENT
:
146 case PIPE_CAP_TWO_SIDED_STENCIL
:
147 case PIPE_CAP_DEPTH_CLIP_DISABLE
:
148 case PIPE_CAP_POINT_SPRITE
:
150 case PIPE_CAP_FRAGMENT_COLOR_CLAMPED
:
151 case PIPE_CAP_VERTEX_COLOR_UNCLAMPED
:
152 case PIPE_CAP_VERTEX_COLOR_CLAMPED
:
153 case PIPE_CAP_QUERY_TIMESTAMP
:
154 case PIPE_CAP_QUERY_TIME_ELAPSED
:
155 case PIPE_CAP_OCCLUSION_QUERY
:
156 case PIPE_CAP_BLEND_EQUATION_SEPARATE
:
157 case PIPE_CAP_INDEP_BLEND_ENABLE
:
158 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT
:
159 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER
:
160 case PIPE_CAP_PRIMITIVE_RESTART
:
161 case PIPE_CAP_TGSI_INSTANCEID
:
162 case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR
:
163 case PIPE_CAP_MIXED_COLORBUFFER_FORMATS
:
164 case PIPE_CAP_CONDITIONAL_RENDER
:
165 case PIPE_CAP_TEXTURE_BARRIER
:
166 case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION
:
167 case PIPE_CAP_START_INSTANCE
:
168 case PIPE_CAP_USER_CONSTANT_BUFFERS
:
169 case PIPE_CAP_USER_INDEX_BUFFERS
:
170 case PIPE_CAP_USER_VERTEX_BUFFERS
:
171 case PIPE_CAP_TEXTURE_MULTISAMPLE
:
172 case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER
:
173 case PIPE_CAP_TGSI_FS_FINE_DERIVATIVE
:
174 case PIPE_CAP_SAMPLER_VIEW_TARGET
:
175 case PIPE_CAP_CONDITIONAL_RENDER_INVERTED
:
176 case PIPE_CAP_CLIP_HALFZ
:
177 case PIPE_CAP_POLYGON_OFFSET_CLAMP
:
179 case PIPE_CAP_SEAMLESS_CUBE_MAP
:
180 return 1; /* class_3d >= NVA0_3D_CLASS; */
181 /* supported on nva0+ */
182 case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME
:
183 return class_3d
>= NVA0_3D_CLASS
;
184 /* supported on nva3+ */
185 case PIPE_CAP_CUBE_MAP_ARRAY
:
186 case PIPE_CAP_INDEP_BLEND_FUNC
:
187 case PIPE_CAP_TEXTURE_QUERY_LOD
:
188 case PIPE_CAP_SAMPLE_SHADING
:
189 return class_3d
>= NVA3_3D_CLASS
;
191 /* unsupported caps */
192 case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE
:
193 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT
:
194 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER
:
195 case PIPE_CAP_SHADER_STENCIL_EXPORT
:
196 case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS
:
197 case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY
:
198 case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY
:
199 case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY
:
200 case PIPE_CAP_TGSI_TEXCOORD
:
201 case PIPE_CAP_QUERY_PIPELINE_STATISTICS
:
202 case PIPE_CAP_TGSI_VS_LAYER_VIEWPORT
:
203 case PIPE_CAP_TEXTURE_GATHER_SM5
:
204 case PIPE_CAP_FAKE_SW_MSAA
:
205 case PIPE_CAP_TEXTURE_GATHER_OFFSETS
:
206 case PIPE_CAP_TGSI_VS_WINDOW_SPACE_POSITION
:
207 case PIPE_CAP_COMPUTE
:
208 case PIPE_CAP_DRAW_INDIRECT
:
209 case PIPE_CAP_VERTEXID_NOBASE
:
210 case PIPE_CAP_MULTISAMPLE_Z_RESOLVE
: /* potentially supported on some hw */
213 case PIPE_CAP_VENDOR_ID
:
215 case PIPE_CAP_DEVICE_ID
: {
217 if (nouveau_getparam(dev
, NOUVEAU_GETPARAM_PCI_DEVICE
, &device_id
)) {
218 NOUVEAU_ERR("NOUVEAU_GETPARAM_PCI_DEVICE failed.\n");
223 case PIPE_CAP_ACCELERATED
:
225 case PIPE_CAP_VIDEO_MEMORY
:
226 return dev
->vram_size
>> 20;
231 NOUVEAU_ERR("unknown PIPE_CAP %d\n", param
);
236 nv50_screen_get_shader_param(struct pipe_screen
*pscreen
, unsigned shader
,
237 enum pipe_shader_cap param
)
240 case PIPE_SHADER_VERTEX
:
241 case PIPE_SHADER_GEOMETRY
:
242 case PIPE_SHADER_FRAGMENT
:
249 case PIPE_SHADER_CAP_MAX_INSTRUCTIONS
:
250 case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS
:
251 case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS
:
252 case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS
:
254 case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH
:
256 case PIPE_SHADER_CAP_MAX_INPUTS
:
257 if (shader
== PIPE_SHADER_VERTEX
)
260 case PIPE_SHADER_CAP_MAX_OUTPUTS
:
262 case PIPE_SHADER_CAP_MAX_CONST_BUFFER_SIZE
:
264 case PIPE_SHADER_CAP_MAX_CONST_BUFFERS
:
265 return NV50_MAX_PIPE_CONSTBUFS
;
266 case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR
:
267 case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR
:
268 return shader
!= PIPE_SHADER_FRAGMENT
;
269 case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR
:
270 case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR
:
272 case PIPE_SHADER_CAP_MAX_PREDS
:
274 case PIPE_SHADER_CAP_MAX_TEMPS
:
275 return nv50_screen(pscreen
)->max_tls_space
/ ONE_TEMP_SIZE
;
276 case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED
:
278 case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED
:
280 case PIPE_SHADER_CAP_SUBROUTINES
:
281 return 0; /* please inline, or provide function declarations */
282 case PIPE_SHADER_CAP_INTEGERS
:
284 case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS
:
285 /* The chip could handle more sampler views than samplers */
286 case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS
:
287 return MIN2(32, PIPE_MAX_SAMPLERS
);
289 NOUVEAU_ERR("unknown PIPE_SHADER_CAP %d\n", param
);
295 nv50_screen_get_paramf(struct pipe_screen
*pscreen
, enum pipe_capf param
)
298 case PIPE_CAPF_MAX_LINE_WIDTH
:
299 case PIPE_CAPF_MAX_LINE_WIDTH_AA
:
301 case PIPE_CAPF_MAX_POINT_WIDTH
:
302 case PIPE_CAPF_MAX_POINT_WIDTH_AA
:
304 case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY
:
306 case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS
:
308 case PIPE_CAPF_GUARD_BAND_LEFT
:
309 case PIPE_CAPF_GUARD_BAND_TOP
:
311 case PIPE_CAPF_GUARD_BAND_RIGHT
:
312 case PIPE_CAPF_GUARD_BAND_BOTTOM
:
313 return 0.0f
; /* that or infinity */
316 NOUVEAU_ERR("unknown PIPE_CAPF %d\n", param
);
321 nv50_screen_destroy(struct pipe_screen
*pscreen
)
323 struct nv50_screen
*screen
= nv50_screen(pscreen
);
325 if (!nouveau_drm_screen_unref(&screen
->base
))
328 if (screen
->base
.fence
.current
) {
329 struct nouveau_fence
*current
= NULL
;
331 /* nouveau_fence_wait will create a new current fence, so wait on the
332 * _current_ one, and remove both.
334 nouveau_fence_ref(screen
->base
.fence
.current
, ¤t
);
335 nouveau_fence_wait(current
);
336 nouveau_fence_ref(NULL
, ¤t
);
337 nouveau_fence_ref(NULL
, &screen
->base
.fence
.current
);
339 if (screen
->base
.pushbuf
)
340 screen
->base
.pushbuf
->user_priv
= NULL
;
343 nv50_blitter_destroy(screen
);
345 nouveau_bo_ref(NULL
, &screen
->code
);
346 nouveau_bo_ref(NULL
, &screen
->tls_bo
);
347 nouveau_bo_ref(NULL
, &screen
->stack_bo
);
348 nouveau_bo_ref(NULL
, &screen
->txc
);
349 nouveau_bo_ref(NULL
, &screen
->uniforms
);
350 nouveau_bo_ref(NULL
, &screen
->fence
.bo
);
352 nouveau_heap_destroy(&screen
->vp_code_heap
);
353 nouveau_heap_destroy(&screen
->gp_code_heap
);
354 nouveau_heap_destroy(&screen
->fp_code_heap
);
356 FREE(screen
->tic
.entries
);
358 nouveau_object_del(&screen
->tesla
);
359 nouveau_object_del(&screen
->eng2d
);
360 nouveau_object_del(&screen
->m2mf
);
361 nouveau_object_del(&screen
->sync
);
363 nouveau_screen_fini(&screen
->base
);
369 nv50_screen_fence_emit(struct pipe_screen
*pscreen
, u32
*sequence
)
371 struct nv50_screen
*screen
= nv50_screen(pscreen
);
372 struct nouveau_pushbuf
*push
= screen
->base
.pushbuf
;
374 /* we need to do it after possible flush in MARK_RING */
375 *sequence
= ++screen
->base
.fence
.sequence
;
377 PUSH_DATA (push
, NV50_FIFO_PKHDR(NV50_3D(QUERY_ADDRESS_HIGH
), 4));
378 PUSH_DATAh(push
, screen
->fence
.bo
->offset
);
379 PUSH_DATA (push
, screen
->fence
.bo
->offset
);
380 PUSH_DATA (push
, *sequence
);
381 PUSH_DATA (push
, NV50_3D_QUERY_GET_MODE_WRITE_UNK0
|
382 NV50_3D_QUERY_GET_UNK4
|
383 NV50_3D_QUERY_GET_UNIT_CROP
|
384 NV50_3D_QUERY_GET_TYPE_QUERY
|
385 NV50_3D_QUERY_GET_QUERY_SELECT_ZERO
|
386 NV50_3D_QUERY_GET_SHORT
);
390 nv50_screen_fence_update(struct pipe_screen
*pscreen
)
392 return nv50_screen(pscreen
)->fence
.map
[0];
396 nv50_screen_init_hwctx(struct nv50_screen
*screen
)
398 struct nouveau_pushbuf
*push
= screen
->base
.pushbuf
;
399 struct nv04_fifo
*fifo
;
402 fifo
= (struct nv04_fifo
*)screen
->base
.channel
->data
;
404 BEGIN_NV04(push
, SUBC_M2MF(NV01_SUBCHAN_OBJECT
), 1);
405 PUSH_DATA (push
, screen
->m2mf
->handle
);
406 BEGIN_NV04(push
, SUBC_M2MF(NV03_M2MF_DMA_NOTIFY
), 3);
407 PUSH_DATA (push
, screen
->sync
->handle
);
408 PUSH_DATA (push
, fifo
->vram
);
409 PUSH_DATA (push
, fifo
->vram
);
411 BEGIN_NV04(push
, SUBC_2D(NV01_SUBCHAN_OBJECT
), 1);
412 PUSH_DATA (push
, screen
->eng2d
->handle
);
413 BEGIN_NV04(push
, NV50_2D(DMA_NOTIFY
), 4);
414 PUSH_DATA (push
, screen
->sync
->handle
);
415 PUSH_DATA (push
, fifo
->vram
);
416 PUSH_DATA (push
, fifo
->vram
);
417 PUSH_DATA (push
, fifo
->vram
);
418 BEGIN_NV04(push
, NV50_2D(OPERATION
), 1);
419 PUSH_DATA (push
, NV50_2D_OPERATION_SRCCOPY
);
420 BEGIN_NV04(push
, NV50_2D(CLIP_ENABLE
), 1);
422 BEGIN_NV04(push
, NV50_2D(COLOR_KEY_ENABLE
), 1);
424 BEGIN_NV04(push
, SUBC_2D(0x0888), 1);
426 BEGIN_NV04(push
, NV50_2D(COND_MODE
), 1);
427 PUSH_DATA (push
, NV50_2D_COND_MODE_ALWAYS
);
429 BEGIN_NV04(push
, SUBC_3D(NV01_SUBCHAN_OBJECT
), 1);
430 PUSH_DATA (push
, screen
->tesla
->handle
);
432 BEGIN_NV04(push
, NV50_3D(COND_MODE
), 1);
433 PUSH_DATA (push
, NV50_3D_COND_MODE_ALWAYS
);
435 BEGIN_NV04(push
, NV50_3D(DMA_NOTIFY
), 1);
436 PUSH_DATA (push
, screen
->sync
->handle
);
437 BEGIN_NV04(push
, NV50_3D(DMA_ZETA
), 11);
438 for (i
= 0; i
< 11; ++i
)
439 PUSH_DATA(push
, fifo
->vram
);
440 BEGIN_NV04(push
, NV50_3D(DMA_COLOR(0)), NV50_3D_DMA_COLOR__LEN
);
441 for (i
= 0; i
< NV50_3D_DMA_COLOR__LEN
; ++i
)
442 PUSH_DATA(push
, fifo
->vram
);
444 BEGIN_NV04(push
, NV50_3D(REG_MODE
), 1);
445 PUSH_DATA (push
, NV50_3D_REG_MODE_STRIPED
);
446 BEGIN_NV04(push
, NV50_3D(UNK1400_LANES
), 1);
447 PUSH_DATA (push
, 0xf);
449 if (debug_get_bool_option("NOUVEAU_SHADER_WATCHDOG", TRUE
)) {
450 BEGIN_NV04(push
, NV50_3D(WATCHDOG_TIMER
), 1);
451 PUSH_DATA (push
, 0x18);
454 BEGIN_NV04(push
, NV50_3D(ZETA_COMP_ENABLE
), 1);
455 PUSH_DATA(push
, screen
->base
.device
->drm_version
>= 0x01000101);
457 BEGIN_NV04(push
, NV50_3D(RT_COMP_ENABLE(0)), 8);
458 for (i
= 0; i
< 8; ++i
)
459 PUSH_DATA(push
, screen
->base
.device
->drm_version
>= 0x01000101);
461 BEGIN_NV04(push
, NV50_3D(RT_CONTROL
), 1);
464 BEGIN_NV04(push
, NV50_3D(CSAA_ENABLE
), 1);
466 BEGIN_NV04(push
, NV50_3D(MULTISAMPLE_ENABLE
), 1);
468 BEGIN_NV04(push
, NV50_3D(MULTISAMPLE_MODE
), 1);
469 PUSH_DATA (push
, NV50_3D_MULTISAMPLE_MODE_MS1
);
470 BEGIN_NV04(push
, NV50_3D(MULTISAMPLE_CTRL
), 1);
472 BEGIN_NV04(push
, NV50_3D(PRIM_RESTART_WITH_DRAW_ARRAYS
), 1);
474 BEGIN_NV04(push
, NV50_3D(BLEND_SEPARATE_ALPHA
), 1);
477 if (screen
->tesla
->oclass
>= NVA0_3D_CLASS
) {
478 BEGIN_NV04(push
, SUBC_3D(NVA0_3D_TEX_MISC
), 1);
479 PUSH_DATA (push
, NVA0_3D_TEX_MISC_SEAMLESS_CUBE_MAP
);
482 BEGIN_NV04(push
, NV50_3D(SCREEN_Y_CONTROL
), 1);
484 BEGIN_NV04(push
, NV50_3D(WINDOW_OFFSET_X
), 2);
487 BEGIN_NV04(push
, NV50_3D(ZCULL_REGION
), 1);
488 PUSH_DATA (push
, 0x3f);
490 BEGIN_NV04(push
, NV50_3D(VP_ADDRESS_HIGH
), 2);
491 PUSH_DATAh(push
, screen
->code
->offset
+ (0 << NV50_CODE_BO_SIZE_LOG2
));
492 PUSH_DATA (push
, screen
->code
->offset
+ (0 << NV50_CODE_BO_SIZE_LOG2
));
494 BEGIN_NV04(push
, NV50_3D(FP_ADDRESS_HIGH
), 2);
495 PUSH_DATAh(push
, screen
->code
->offset
+ (1 << NV50_CODE_BO_SIZE_LOG2
));
496 PUSH_DATA (push
, screen
->code
->offset
+ (1 << NV50_CODE_BO_SIZE_LOG2
));
498 BEGIN_NV04(push
, NV50_3D(GP_ADDRESS_HIGH
), 2);
499 PUSH_DATAh(push
, screen
->code
->offset
+ (2 << NV50_CODE_BO_SIZE_LOG2
));
500 PUSH_DATA (push
, screen
->code
->offset
+ (2 << NV50_CODE_BO_SIZE_LOG2
));
502 BEGIN_NV04(push
, NV50_3D(LOCAL_ADDRESS_HIGH
), 3);
503 PUSH_DATAh(push
, screen
->tls_bo
->offset
);
504 PUSH_DATA (push
, screen
->tls_bo
->offset
);
505 PUSH_DATA (push
, util_logbase2(screen
->cur_tls_space
/ 8));
507 BEGIN_NV04(push
, NV50_3D(STACK_ADDRESS_HIGH
), 3);
508 PUSH_DATAh(push
, screen
->stack_bo
->offset
);
509 PUSH_DATA (push
, screen
->stack_bo
->offset
);
512 BEGIN_NV04(push
, NV50_3D(CB_DEF_ADDRESS_HIGH
), 3);
513 PUSH_DATAh(push
, screen
->uniforms
->offset
+ (0 << 16));
514 PUSH_DATA (push
, screen
->uniforms
->offset
+ (0 << 16));
515 PUSH_DATA (push
, (NV50_CB_PVP
<< 16) | 0x0000);
517 BEGIN_NV04(push
, NV50_3D(CB_DEF_ADDRESS_HIGH
), 3);
518 PUSH_DATAh(push
, screen
->uniforms
->offset
+ (1 << 16));
519 PUSH_DATA (push
, screen
->uniforms
->offset
+ (1 << 16));
520 PUSH_DATA (push
, (NV50_CB_PGP
<< 16) | 0x0000);
522 BEGIN_NV04(push
, NV50_3D(CB_DEF_ADDRESS_HIGH
), 3);
523 PUSH_DATAh(push
, screen
->uniforms
->offset
+ (2 << 16));
524 PUSH_DATA (push
, screen
->uniforms
->offset
+ (2 << 16));
525 PUSH_DATA (push
, (NV50_CB_PFP
<< 16) | 0x0000);
527 BEGIN_NV04(push
, NV50_3D(CB_DEF_ADDRESS_HIGH
), 3);
528 PUSH_DATAh(push
, screen
->uniforms
->offset
+ (3 << 16));
529 PUSH_DATA (push
, screen
->uniforms
->offset
+ (3 << 16));
530 PUSH_DATA (push
, (NV50_CB_AUX
<< 16) | (NV50_CB_AUX_SIZE
& 0xffff));
532 BEGIN_NI04(push
, NV50_3D(SET_PROGRAM_CB
), 3);
533 PUSH_DATA (push
, (NV50_CB_AUX
<< 12) | 0xf01);
534 PUSH_DATA (push
, (NV50_CB_AUX
<< 12) | 0xf21);
535 PUSH_DATA (push
, (NV50_CB_AUX
<< 12) | 0xf31);
537 /* return { 0.0, 0.0, 0.0, 0.0 } on out-of-bounds vtxbuf access */
538 BEGIN_NV04(push
, NV50_3D(CB_ADDR
), 1);
539 PUSH_DATA (push
, (NV50_CB_AUX_RUNOUT_OFFSET
<< (8 - 2)) | NV50_CB_AUX
);
540 BEGIN_NI04(push
, NV50_3D(CB_DATA(0)), 4);
541 PUSH_DATAf(push
, 0.0f
);
542 PUSH_DATAf(push
, 0.0f
);
543 PUSH_DATAf(push
, 0.0f
);
544 PUSH_DATAf(push
, 0.0f
);
545 BEGIN_NV04(push
, NV50_3D(VERTEX_RUNOUT_ADDRESS_HIGH
), 2);
546 PUSH_DATAh(push
, screen
->uniforms
->offset
+ (3 << 16) + NV50_CB_AUX_RUNOUT_OFFSET
);
547 PUSH_DATA (push
, screen
->uniforms
->offset
+ (3 << 16) + NV50_CB_AUX_RUNOUT_OFFSET
);
549 nv50_upload_ms_info(push
);
551 /* max TIC (bits 4:8) & TSC bindings, per program type */
552 for (i
= 0; i
< 3; ++i
) {
553 BEGIN_NV04(push
, NV50_3D(TEX_LIMITS(i
)), 1);
554 PUSH_DATA (push
, 0x54);
557 BEGIN_NV04(push
, NV50_3D(TIC_ADDRESS_HIGH
), 3);
558 PUSH_DATAh(push
, screen
->txc
->offset
);
559 PUSH_DATA (push
, screen
->txc
->offset
);
560 PUSH_DATA (push
, NV50_TIC_MAX_ENTRIES
- 1);
562 BEGIN_NV04(push
, NV50_3D(TSC_ADDRESS_HIGH
), 3);
563 PUSH_DATAh(push
, screen
->txc
->offset
+ 65536);
564 PUSH_DATA (push
, screen
->txc
->offset
+ 65536);
565 PUSH_DATA (push
, NV50_TSC_MAX_ENTRIES
- 1);
567 BEGIN_NV04(push
, NV50_3D(LINKED_TSC
), 1);
570 BEGIN_NV04(push
, NV50_3D(CLIP_RECTS_EN
), 1);
572 BEGIN_NV04(push
, NV50_3D(CLIP_RECTS_MODE
), 1);
573 PUSH_DATA (push
, NV50_3D_CLIP_RECTS_MODE_INSIDE_ANY
);
574 BEGIN_NV04(push
, NV50_3D(CLIP_RECT_HORIZ(0)), 8 * 2);
575 for (i
= 0; i
< 8 * 2; ++i
)
577 BEGIN_NV04(push
, NV50_3D(CLIPID_ENABLE
), 1);
580 BEGIN_NV04(push
, NV50_3D(VIEWPORT_TRANSFORM_EN
), 1);
582 for (i
= 0; i
< NV50_MAX_VIEWPORTS
; i
++) {
583 BEGIN_NV04(push
, NV50_3D(DEPTH_RANGE_NEAR(i
)), 2);
584 PUSH_DATAf(push
, 0.0f
);
585 PUSH_DATAf(push
, 1.0f
);
586 BEGIN_NV04(push
, NV50_3D(VIEWPORT_HORIZ(i
)), 2);
587 PUSH_DATA (push
, 8192 << 16);
588 PUSH_DATA (push
, 8192 << 16);
591 BEGIN_NV04(push
, NV50_3D(VIEW_VOLUME_CLIP_CTRL
), 1);
592 #ifdef NV50_SCISSORS_CLIPPING
593 PUSH_DATA (push
, 0x0000);
595 PUSH_DATA (push
, 0x1080);
598 BEGIN_NV04(push
, NV50_3D(CLEAR_FLAGS
), 1);
599 PUSH_DATA (push
, NV50_3D_CLEAR_FLAGS_CLEAR_RECT_VIEWPORT
);
601 /* We use scissors instead of exact view volume clipping,
602 * so they're always enabled.
604 for (i
= 0; i
< NV50_MAX_VIEWPORTS
; i
++) {
605 BEGIN_NV04(push
, NV50_3D(SCISSOR_ENABLE(i
)), 3);
607 PUSH_DATA (push
, 8192 << 16);
608 PUSH_DATA (push
, 8192 << 16);
611 BEGIN_NV04(push
, NV50_3D(RASTERIZE_ENABLE
), 1);
613 BEGIN_NV04(push
, NV50_3D(POINT_RASTER_RULES
), 1);
614 PUSH_DATA (push
, NV50_3D_POINT_RASTER_RULES_OGL
);
615 BEGIN_NV04(push
, NV50_3D(FRAG_COLOR_CLAMP_EN
), 1);
616 PUSH_DATA (push
, 0x11111111);
617 BEGIN_NV04(push
, NV50_3D(EDGEFLAG
), 1);
620 BEGIN_NV04(push
, NV50_3D(VB_ELEMENT_BASE
), 1);
622 if (screen
->base
.class_3d
>= NV84_3D_CLASS
) {
623 BEGIN_NV04(push
, SUBC_3D(NV84_3D_VERTEX_ID_BASE
), 1);
630 static int nv50_tls_alloc(struct nv50_screen
*screen
, unsigned tls_space
,
633 struct nouveau_device
*dev
= screen
->base
.device
;
636 screen
->cur_tls_space
= util_next_power_of_two(tls_space
/ ONE_TEMP_SIZE
) *
638 if (nouveau_mesa_debug
)
639 debug_printf("allocating space for %u temps\n",
640 util_next_power_of_two(tls_space
/ ONE_TEMP_SIZE
));
641 *tls_size
= screen
->cur_tls_space
* util_next_power_of_two(screen
->TPs
) *
642 screen
->MPsInTP
* LOCAL_WARPS_ALLOC
* THREADS_IN_WARP
;
644 ret
= nouveau_bo_new(dev
, NOUVEAU_BO_VRAM
, 1 << 16,
645 *tls_size
, NULL
, &screen
->tls_bo
);
647 NOUVEAU_ERR("Failed to allocate local bo: %d\n", ret
);
654 int nv50_tls_realloc(struct nv50_screen
*screen
, unsigned tls_space
)
656 struct nouveau_pushbuf
*push
= screen
->base
.pushbuf
;
660 if (tls_space
< screen
->cur_tls_space
)
662 if (tls_space
> screen
->max_tls_space
) {
663 /* fixable by limiting number of warps (LOCAL_WARPS_LOG_ALLOC /
664 * LOCAL_WARPS_NO_CLAMP) */
665 NOUVEAU_ERR("Unsupported number of temporaries (%u > %u). Fixable if someone cares.\n",
666 (unsigned)(tls_space
/ ONE_TEMP_SIZE
),
667 (unsigned)(screen
->max_tls_space
/ ONE_TEMP_SIZE
));
671 nouveau_bo_ref(NULL
, &screen
->tls_bo
);
672 ret
= nv50_tls_alloc(screen
, tls_space
, &tls_size
);
676 BEGIN_NV04(push
, NV50_3D(LOCAL_ADDRESS_HIGH
), 3);
677 PUSH_DATAh(push
, screen
->tls_bo
->offset
);
678 PUSH_DATA (push
, screen
->tls_bo
->offset
);
679 PUSH_DATA (push
, util_logbase2(screen
->cur_tls_space
/ 8));
685 nv50_screen_create(struct nouveau_device
*dev
)
687 struct nv50_screen
*screen
;
688 struct pipe_screen
*pscreen
;
689 struct nouveau_object
*chan
;
691 uint32_t tesla_class
;
695 screen
= CALLOC_STRUCT(nv50_screen
);
698 pscreen
= &screen
->base
.base
;
700 ret
= nouveau_screen_init(&screen
->base
, dev
);
702 NOUVEAU_ERR("nouveau_screen_init failed: %d\n", ret
);
706 /* TODO: Prevent FIFO prefetch before transfer of index buffers and
707 * admit them to VRAM.
709 screen
->base
.vidmem_bindings
|= PIPE_BIND_CONSTANT_BUFFER
|
710 PIPE_BIND_VERTEX_BUFFER
;
711 screen
->base
.sysmem_bindings
|=
712 PIPE_BIND_VERTEX_BUFFER
| PIPE_BIND_INDEX_BUFFER
;
714 screen
->base
.pushbuf
->user_priv
= screen
;
715 screen
->base
.pushbuf
->rsvd_kick
= 5;
717 chan
= screen
->base
.channel
;
719 pscreen
->destroy
= nv50_screen_destroy
;
720 pscreen
->context_create
= nv50_create
;
721 pscreen
->is_format_supported
= nv50_screen_is_format_supported
;
722 pscreen
->get_param
= nv50_screen_get_param
;
723 pscreen
->get_shader_param
= nv50_screen_get_shader_param
;
724 pscreen
->get_paramf
= nv50_screen_get_paramf
;
726 nv50_screen_init_resource_functions(pscreen
);
728 if (screen
->base
.device
->chipset
< 0x84 ||
729 debug_get_bool_option("NOUVEAU_PMPEG", FALSE
)) {
731 nouveau_screen_init_vdec(&screen
->base
);
732 } else if (screen
->base
.device
->chipset
< 0x98 ||
733 screen
->base
.device
->chipset
== 0xa0) {
735 screen
->base
.base
.get_video_param
= nv84_screen_get_video_param
;
736 screen
->base
.base
.is_video_format_supported
= nv84_screen_video_supported
;
739 screen
->base
.base
.get_video_param
= nouveau_vp3_screen_get_video_param
;
740 screen
->base
.base
.is_video_format_supported
= nouveau_vp3_screen_video_supported
;
743 ret
= nouveau_bo_new(dev
, NOUVEAU_BO_GART
| NOUVEAU_BO_MAP
, 0, 4096,
744 NULL
, &screen
->fence
.bo
);
746 NOUVEAU_ERR("Failed to allocate fence bo: %d\n", ret
);
750 nouveau_bo_map(screen
->fence
.bo
, 0, NULL
);
751 screen
->fence
.map
= screen
->fence
.bo
->map
;
752 screen
->base
.fence
.emit
= nv50_screen_fence_emit
;
753 screen
->base
.fence
.update
= nv50_screen_fence_update
;
755 ret
= nouveau_object_new(chan
, 0xbeef0301, NOUVEAU_NOTIFIER_CLASS
,
756 &(struct nv04_notify
){ .length
= 32 },
757 sizeof(struct nv04_notify
), &screen
->sync
);
759 NOUVEAU_ERR("Failed to allocate notifier: %d\n", ret
);
763 ret
= nouveau_object_new(chan
, 0xbeef5039, NV50_M2MF_CLASS
,
764 NULL
, 0, &screen
->m2mf
);
766 NOUVEAU_ERR("Failed to allocate PGRAPH context for M2MF: %d\n", ret
);
770 ret
= nouveau_object_new(chan
, 0xbeef502d, NV50_2D_CLASS
,
771 NULL
, 0, &screen
->eng2d
);
773 NOUVEAU_ERR("Failed to allocate PGRAPH context for 2D: %d\n", ret
);
777 switch (dev
->chipset
& 0xf0) {
779 tesla_class
= NV50_3D_CLASS
;
783 tesla_class
= NV84_3D_CLASS
;
786 switch (dev
->chipset
) {
790 tesla_class
= NVA0_3D_CLASS
;
793 tesla_class
= NVAF_3D_CLASS
;
796 tesla_class
= NVA3_3D_CLASS
;
801 NOUVEAU_ERR("Not a known NV50 chipset: NV%02x\n", dev
->chipset
);
804 screen
->base
.class_3d
= tesla_class
;
806 ret
= nouveau_object_new(chan
, 0xbeef5097, tesla_class
,
807 NULL
, 0, &screen
->tesla
);
809 NOUVEAU_ERR("Failed to allocate PGRAPH context for 3D: %d\n", ret
);
813 /* This over-allocates by a page. The GP, which would execute at the end of
814 * the last page, would trigger faults. The going theory is that it
815 * prefetches up to a certain amount.
817 ret
= nouveau_bo_new(dev
, NOUVEAU_BO_VRAM
, 1 << 16,
818 (3 << NV50_CODE_BO_SIZE_LOG2
) + 0x1000,
819 NULL
, &screen
->code
);
821 NOUVEAU_ERR("Failed to allocate code bo: %d\n", ret
);
825 nouveau_heap_init(&screen
->vp_code_heap
, 0, 1 << NV50_CODE_BO_SIZE_LOG2
);
826 nouveau_heap_init(&screen
->gp_code_heap
, 0, 1 << NV50_CODE_BO_SIZE_LOG2
);
827 nouveau_heap_init(&screen
->fp_code_heap
, 0, 1 << NV50_CODE_BO_SIZE_LOG2
);
829 nouveau_getparam(dev
, NOUVEAU_GETPARAM_GRAPH_UNITS
, &value
);
831 screen
->TPs
= util_bitcount(value
& 0xffff);
832 screen
->MPsInTP
= util_bitcount((value
>> 24) & 0xf);
834 stack_size
= util_next_power_of_two(screen
->TPs
) * screen
->MPsInTP
*
835 STACK_WARPS_ALLOC
* 64 * 8;
837 ret
= nouveau_bo_new(dev
, NOUVEAU_BO_VRAM
, 1 << 16, stack_size
, NULL
,
840 NOUVEAU_ERR("Failed to allocate stack bo: %d\n", ret
);
844 uint64_t size_of_one_temp
= util_next_power_of_two(screen
->TPs
) *
845 screen
->MPsInTP
* LOCAL_WARPS_ALLOC
* THREADS_IN_WARP
*
847 screen
->max_tls_space
= dev
->vram_size
/ size_of_one_temp
* ONE_TEMP_SIZE
;
848 screen
->max_tls_space
/= 2; /* half of vram */
850 /* hw can address max 64 KiB */
851 screen
->max_tls_space
= MIN2(screen
->max_tls_space
, 64 << 10);
854 unsigned tls_space
= 4/*temps*/ * ONE_TEMP_SIZE
;
855 ret
= nv50_tls_alloc(screen
, tls_space
, &tls_size
);
859 if (nouveau_mesa_debug
)
860 debug_printf("TPs = %u, MPsInTP = %u, VRAM = %"PRIu64
" MiB, tls_size = %"PRIu64
" KiB\n",
861 screen
->TPs
, screen
->MPsInTP
, dev
->vram_size
>> 20, tls_size
>> 10);
863 ret
= nouveau_bo_new(dev
, NOUVEAU_BO_VRAM
, 1 << 16, 4 << 16, NULL
,
866 NOUVEAU_ERR("Failed to allocate uniforms bo: %d\n", ret
);
870 ret
= nouveau_bo_new(dev
, NOUVEAU_BO_VRAM
, 1 << 16, 3 << 16, NULL
,
873 NOUVEAU_ERR("Failed to allocate TIC/TSC bo: %d\n", ret
);
877 screen
->tic
.entries
= CALLOC(4096, sizeof(void *));
878 screen
->tsc
.entries
= screen
->tic
.entries
+ 2048;
880 if (!nv50_blitter_create(screen
))
883 nv50_screen_init_hwctx(screen
);
885 nouveau_fence_new(&screen
->base
, &screen
->base
.fence
.current
, FALSE
);
890 nv50_screen_destroy(pscreen
);
895 nv50_screen_tic_alloc(struct nv50_screen
*screen
, void *entry
)
897 int i
= screen
->tic
.next
;
899 while (screen
->tic
.lock
[i
/ 32] & (1 << (i
% 32)))
900 i
= (i
+ 1) & (NV50_TIC_MAX_ENTRIES
- 1);
902 screen
->tic
.next
= (i
+ 1) & (NV50_TIC_MAX_ENTRIES
- 1);
904 if (screen
->tic
.entries
[i
])
905 nv50_tic_entry(screen
->tic
.entries
[i
])->id
= -1;
907 screen
->tic
.entries
[i
] = entry
;
912 nv50_screen_tsc_alloc(struct nv50_screen
*screen
, void *entry
)
914 int i
= screen
->tsc
.next
;
916 while (screen
->tsc
.lock
[i
/ 32] & (1 << (i
% 32)))
917 i
= (i
+ 1) & (NV50_TSC_MAX_ENTRIES
- 1);
919 screen
->tsc
.next
= (i
+ 1) & (NV50_TSC_MAX_ENTRIES
- 1);
921 if (screen
->tsc
.entries
[i
])
922 nv50_tsc_entry(screen
->tsc
.entries
[i
])->id
= -1;
924 screen
->tsc
.entries
[i
] = entry
;