radeonsi: add a HUD query for getting an average GFX BO list size
[mesa.git] / src / gallium / drivers / radeon / radeon_winsys.h
1 /*
2 * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com>
3 * Copyright 2010 Marek Olšák <maraeo@gmail.com>
4 *
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * on the rights to use, copy, modify, merge, publish, distribute, sub
9 * license, and/or sell copies of the Software, and to permit persons to whom
10 * the Software is furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice (including the next
13 * paragraph) shall be included in all copies or substantial portions of the
14 * Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
20 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
21 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
22 * USE OR OTHER DEALINGS IN THE SOFTWARE. */
23
24 #ifndef RADEON_WINSYS_H
25 #define RADEON_WINSYS_H
26
27 /* The public winsys interface header for the radeon driver. */
28
29 #include "pipebuffer/pb_buffer.h"
30
31 #include "amd/common/ac_gpu_info.h"
32 #include "amd/common/ac_surface.h"
33
34 #define RADEON_FLUSH_ASYNC (1 << 0)
35 #define RADEON_FLUSH_END_OF_FRAME (1 << 1)
36
37 /* Tiling flags. */
38 enum radeon_bo_layout {
39 RADEON_LAYOUT_LINEAR = 0,
40 RADEON_LAYOUT_TILED,
41 RADEON_LAYOUT_SQUARETILED,
42
43 RADEON_LAYOUT_UNKNOWN
44 };
45
46 enum radeon_bo_domain { /* bitfield */
47 RADEON_DOMAIN_GTT = 2,
48 RADEON_DOMAIN_VRAM = 4,
49 RADEON_DOMAIN_VRAM_GTT = RADEON_DOMAIN_VRAM | RADEON_DOMAIN_GTT
50 };
51
52 enum radeon_bo_flag { /* bitfield */
53 RADEON_FLAG_GTT_WC = (1 << 0),
54 RADEON_FLAG_CPU_ACCESS = (1 << 1),
55 RADEON_FLAG_NO_CPU_ACCESS = (1 << 2),
56 RADEON_FLAG_HANDLE = (1 << 3), /* the buffer must not be suballocated */
57 RADEON_FLAG_SPARSE = (1 << 4),
58 };
59
60 enum radeon_bo_usage { /* bitfield */
61 RADEON_USAGE_READ = 2,
62 RADEON_USAGE_WRITE = 4,
63 RADEON_USAGE_READWRITE = RADEON_USAGE_READ | RADEON_USAGE_WRITE,
64
65 /* The winsys ensures that the CS submission will be scheduled after
66 * previously flushed CSs referencing this BO in a conflicting way.
67 */
68 RADEON_USAGE_SYNCHRONIZED = 8
69 };
70
71 #define RADEON_SPARSE_PAGE_SIZE (64 * 1024)
72
73 enum ring_type {
74 RING_GFX = 0,
75 RING_COMPUTE,
76 RING_DMA,
77 RING_UVD,
78 RING_VCE,
79 RING_UVD_ENC,
80 RING_VCN_DEC,
81 RING_LAST,
82 };
83
84 enum radeon_value_id {
85 RADEON_REQUESTED_VRAM_MEMORY,
86 RADEON_REQUESTED_GTT_MEMORY,
87 RADEON_MAPPED_VRAM,
88 RADEON_MAPPED_GTT,
89 RADEON_BUFFER_WAIT_TIME_NS,
90 RADEON_NUM_MAPPED_BUFFERS,
91 RADEON_TIMESTAMP,
92 RADEON_NUM_GFX_IBS,
93 RADEON_NUM_SDMA_IBS,
94 RADEON_GFX_BO_LIST_COUNTER, /* number of BOs submitted in gfx IBs */
95 RADEON_NUM_BYTES_MOVED,
96 RADEON_NUM_EVICTIONS,
97 RADEON_NUM_VRAM_CPU_PAGE_FAULTS,
98 RADEON_VRAM_USAGE,
99 RADEON_VRAM_VIS_USAGE,
100 RADEON_GTT_USAGE,
101 RADEON_GPU_TEMPERATURE, /* DRM 2.42.0 */
102 RADEON_CURRENT_SCLK,
103 RADEON_CURRENT_MCLK,
104 RADEON_GPU_RESET_COUNTER, /* DRM 2.43.0 */
105 RADEON_CS_THREAD_TIME,
106 };
107
108 /* Each group of four has the same priority. */
109 enum radeon_bo_priority {
110 RADEON_PRIO_FENCE = 0,
111 RADEON_PRIO_TRACE,
112 RADEON_PRIO_SO_FILLED_SIZE,
113 RADEON_PRIO_QUERY,
114
115 RADEON_PRIO_IB1 = 4, /* main IB submitted to the kernel */
116 RADEON_PRIO_IB2, /* IB executed with INDIRECT_BUFFER */
117 RADEON_PRIO_DRAW_INDIRECT,
118 RADEON_PRIO_INDEX_BUFFER,
119
120 RADEON_PRIO_VCE = 8,
121 RADEON_PRIO_UVD,
122 RADEON_PRIO_SDMA_BUFFER,
123 RADEON_PRIO_SDMA_TEXTURE,
124
125 RADEON_PRIO_CP_DMA = 12,
126
127 RADEON_PRIO_CONST_BUFFER = 16,
128 RADEON_PRIO_DESCRIPTORS,
129 RADEON_PRIO_BORDER_COLORS,
130
131 RADEON_PRIO_SAMPLER_BUFFER = 20,
132 RADEON_PRIO_VERTEX_BUFFER,
133
134 RADEON_PRIO_SHADER_RW_BUFFER = 24,
135 RADEON_PRIO_COMPUTE_GLOBAL,
136
137 RADEON_PRIO_SAMPLER_TEXTURE = 28,
138 RADEON_PRIO_SHADER_RW_IMAGE,
139
140 RADEON_PRIO_SAMPLER_TEXTURE_MSAA = 32,
141
142 RADEON_PRIO_COLOR_BUFFER = 36,
143
144 RADEON_PRIO_DEPTH_BUFFER = 40,
145
146 RADEON_PRIO_COLOR_BUFFER_MSAA = 44,
147
148 RADEON_PRIO_DEPTH_BUFFER_MSAA = 48,
149
150 RADEON_PRIO_CMASK = 52,
151 RADEON_PRIO_DCC,
152 RADEON_PRIO_HTILE,
153 RADEON_PRIO_SHADER_BINARY, /* the hw can't hide instruction cache misses */
154
155 RADEON_PRIO_SHADER_RINGS = 56,
156
157 RADEON_PRIO_SCRATCH_BUFFER = 60,
158 /* 63 is the maximum value */
159 };
160
161 struct winsys_handle;
162 struct radeon_winsys_ctx;
163
164 struct radeon_winsys_cs_chunk {
165 unsigned cdw; /* Number of used dwords. */
166 unsigned max_dw; /* Maximum number of dwords. */
167 uint32_t *buf; /* The base pointer of the chunk. */
168 };
169
170 struct radeon_winsys_cs {
171 struct radeon_winsys_cs_chunk current;
172 struct radeon_winsys_cs_chunk *prev;
173 unsigned num_prev; /* Number of previous chunks. */
174 unsigned max_prev; /* Space in array pointed to by prev. */
175 unsigned prev_dw; /* Total number of dwords in previous chunks. */
176
177 /* Memory usage of the buffer list. These are always 0 for CE and preamble
178 * IBs. */
179 uint64_t used_vram;
180 uint64_t used_gart;
181 };
182
183 /* Tiling info for display code, DRI sharing, and other data. */
184 struct radeon_bo_metadata {
185 /* Tiling flags describing the texture layout for display code
186 * and DRI sharing.
187 */
188 union {
189 struct {
190 enum radeon_bo_layout microtile;
191 enum radeon_bo_layout macrotile;
192 unsigned pipe_config;
193 unsigned bankw;
194 unsigned bankh;
195 unsigned tile_split;
196 unsigned mtilea;
197 unsigned num_banks;
198 unsigned stride;
199 bool scanout;
200 } legacy;
201
202 struct {
203 /* surface flags */
204 unsigned swizzle_mode:5;
205 } gfx9;
206 } u;
207
208 /* Additional metadata associated with the buffer, in bytes.
209 * The maximum size is 64 * 4. This is opaque for the winsys & kernel.
210 * Supported by amdgpu only.
211 */
212 uint32_t size_metadata;
213 uint32_t metadata[64];
214 };
215
216 enum radeon_feature_id {
217 RADEON_FID_R300_HYPERZ_ACCESS, /* ZMask + HiZ */
218 RADEON_FID_R300_CMASK_ACCESS,
219 };
220
221 struct radeon_bo_list_item {
222 uint64_t bo_size;
223 uint64_t vm_address;
224 uint64_t priority_usage; /* mask of (1 << RADEON_PRIO_*) */
225 };
226
227 struct radeon_winsys {
228 /**
229 * The screen object this winsys was created for
230 */
231 struct pipe_screen *screen;
232
233 /**
234 * Decrement the winsys reference count.
235 *
236 * \param ws The winsys this function is called for.
237 * \return True if the winsys and screen should be destroyed.
238 */
239 bool (*unref)(struct radeon_winsys *ws);
240
241 /**
242 * Destroy this winsys.
243 *
244 * \param ws The winsys this function is called from.
245 */
246 void (*destroy)(struct radeon_winsys *ws);
247
248 /**
249 * Query an info structure from winsys.
250 *
251 * \param ws The winsys this function is called from.
252 * \param info Return structure
253 */
254 void (*query_info)(struct radeon_winsys *ws,
255 struct radeon_info *info);
256
257 /**************************************************************************
258 * Buffer management. Buffer attributes are mostly fixed over its lifetime.
259 *
260 * Remember that gallium gets to choose the interface it needs, and the
261 * window systems must then implement that interface (rather than the
262 * other way around...).
263 *************************************************************************/
264
265 /**
266 * Create a buffer object.
267 *
268 * \param ws The winsys this function is called from.
269 * \param size The size to allocate.
270 * \param alignment An alignment of the buffer in memory.
271 * \param use_reusable_pool Whether the cache buffer manager should be used.
272 * \param domain A bitmask of the RADEON_DOMAIN_* flags.
273 * \return The created buffer object.
274 */
275 struct pb_buffer *(*buffer_create)(struct radeon_winsys *ws,
276 uint64_t size,
277 unsigned alignment,
278 enum radeon_bo_domain domain,
279 enum radeon_bo_flag flags);
280
281 /**
282 * Map the entire data store of a buffer object into the client's address
283 * space.
284 *
285 * \param buf A winsys buffer object to map.
286 * \param cs A command stream to flush if the buffer is referenced by it.
287 * \param usage A bitmask of the PIPE_TRANSFER_* flags.
288 * \return The pointer at the beginning of the buffer.
289 */
290 void *(*buffer_map)(struct pb_buffer *buf,
291 struct radeon_winsys_cs *cs,
292 enum pipe_transfer_usage usage);
293
294 /**
295 * Unmap a buffer object from the client's address space.
296 *
297 * \param buf A winsys buffer object to unmap.
298 */
299 void (*buffer_unmap)(struct pb_buffer *buf);
300
301 /**
302 * Wait for the buffer and return true if the buffer is not used
303 * by the device.
304 *
305 * The timeout of 0 will only return the status.
306 * The timeout of PIPE_TIMEOUT_INFINITE will always wait until the buffer
307 * is idle.
308 */
309 bool (*buffer_wait)(struct pb_buffer *buf, uint64_t timeout,
310 enum radeon_bo_usage usage);
311
312 /**
313 * Return buffer metadata.
314 * (tiling info for display code, DRI sharing, and other data)
315 *
316 * \param buf A winsys buffer object to get the flags from.
317 * \param md Metadata
318 */
319 void (*buffer_get_metadata)(struct pb_buffer *buf,
320 struct radeon_bo_metadata *md);
321
322 /**
323 * Set buffer metadata.
324 * (tiling info for display code, DRI sharing, and other data)
325 *
326 * \param buf A winsys buffer object to set the flags for.
327 * \param md Metadata
328 */
329 void (*buffer_set_metadata)(struct pb_buffer *buf,
330 struct radeon_bo_metadata *md);
331
332 /**
333 * Get a winsys buffer from a winsys handle. The internal structure
334 * of the handle is platform-specific and only a winsys should access it.
335 *
336 * \param ws The winsys this function is called from.
337 * \param whandle A winsys handle pointer as was received from a state
338 * tracker.
339 * \param stride The returned buffer stride in bytes.
340 */
341 struct pb_buffer *(*buffer_from_handle)(struct radeon_winsys *ws,
342 struct winsys_handle *whandle,
343 unsigned *stride, unsigned *offset);
344
345 /**
346 * Get a winsys buffer from a user pointer. The resulting buffer can't
347 * be exported. Both pointer and size must be page aligned.
348 *
349 * \param ws The winsys this function is called from.
350 * \param pointer User pointer to turn into a buffer object.
351 * \param Size Size in bytes for the new buffer.
352 */
353 struct pb_buffer *(*buffer_from_ptr)(struct radeon_winsys *ws,
354 void *pointer, uint64_t size);
355
356 /**
357 * Whether the buffer was created from a user pointer.
358 *
359 * \param buf A winsys buffer object
360 * \return whether \p buf was created via buffer_from_ptr
361 */
362 bool (*buffer_is_user_ptr)(struct pb_buffer *buf);
363
364 /**
365 * Get a winsys handle from a winsys buffer. The internal structure
366 * of the handle is platform-specific and only a winsys should access it.
367 *
368 * \param buf A winsys buffer object to get the handle from.
369 * \param whandle A winsys handle pointer.
370 * \param stride A stride of the buffer in bytes, for texturing.
371 * \return true on success.
372 */
373 bool (*buffer_get_handle)(struct pb_buffer *buf,
374 unsigned stride, unsigned offset,
375 unsigned slice_size,
376 struct winsys_handle *whandle);
377
378 /**
379 * Change the commitment of a (64KB-page aligned) region of the given
380 * sparse buffer.
381 *
382 * \warning There is no automatic synchronization with command submission.
383 *
384 * \note Only implemented by the amdgpu winsys.
385 *
386 * \return false on out of memory or other failure, true on success.
387 */
388 bool (*buffer_commit)(struct pb_buffer *buf,
389 uint64_t offset, uint64_t size,
390 bool commit);
391
392 /**
393 * Return the virtual address of a buffer.
394 *
395 * When virtual memory is not in use, this is the offset relative to the
396 * relocation base (non-zero for sub-allocated buffers).
397 *
398 * \param buf A winsys buffer object
399 * \return virtual address
400 */
401 uint64_t (*buffer_get_virtual_address)(struct pb_buffer *buf);
402
403 /**
404 * Return the offset of this buffer relative to the relocation base.
405 * This is only non-zero for sub-allocated buffers.
406 *
407 * This is only supported in the radeon winsys, since amdgpu uses virtual
408 * addresses in submissions even for the video engines.
409 *
410 * \param buf A winsys buffer object
411 * \return the offset for relocations
412 */
413 unsigned (*buffer_get_reloc_offset)(struct pb_buffer *buf);
414
415 /**
416 * Query the initial placement of the buffer from the kernel driver.
417 */
418 enum radeon_bo_domain (*buffer_get_initial_domain)(struct pb_buffer *buf);
419
420 /**************************************************************************
421 * Command submission.
422 *
423 * Each pipe context should create its own command stream and submit
424 * commands independently of other contexts.
425 *************************************************************************/
426
427 /**
428 * Create a command submission context.
429 * Various command streams can be submitted to the same context.
430 */
431 struct radeon_winsys_ctx *(*ctx_create)(struct radeon_winsys *ws);
432
433 /**
434 * Destroy a context.
435 */
436 void (*ctx_destroy)(struct radeon_winsys_ctx *ctx);
437
438 /**
439 * Query a GPU reset status.
440 */
441 enum pipe_reset_status (*ctx_query_reset_status)(struct radeon_winsys_ctx *ctx);
442
443 /**
444 * Create a command stream.
445 *
446 * \param ctx The submission context
447 * \param ring_type The ring type (GFX, DMA, UVD)
448 * \param flush Flush callback function associated with the command stream.
449 * \param user User pointer that will be passed to the flush callback.
450 */
451 struct radeon_winsys_cs *(*cs_create)(struct radeon_winsys_ctx *ctx,
452 enum ring_type ring_type,
453 void (*flush)(void *ctx, unsigned flags,
454 struct pipe_fence_handle **fence),
455 void *flush_ctx);
456
457 /**
458 * Add a constant engine IB to a graphics CS. This makes the graphics CS
459 * from "cs_create" a group of two IBs that share a buffer list and are
460 * flushed together.
461 *
462 * The returned constant CS is only a stream for writing packets to the new
463 * IB. Calling other winsys functions with it is not allowed, not even
464 * "cs_destroy".
465 *
466 * In order to add buffers and check memory usage, use the graphics CS.
467 * In order to flush it, use the graphics CS, which will flush both IBs.
468 * Destroying the graphics CS will destroy both of them.
469 *
470 * \param cs The graphics CS from "cs_create" that will hold the buffer
471 * list and will be used for flushing.
472 */
473 struct radeon_winsys_cs *(*cs_add_const_ib)(struct radeon_winsys_cs *cs);
474
475 /**
476 * Add a constant engine preamble IB to a graphics CS. This add an extra IB
477 * in similar manner to cs_add_const_ib. This should always be called after
478 * cs_add_const_ib.
479 *
480 * The returned IB is a constant engine IB that only gets flushed if the
481 * context changed.
482 *
483 * \param cs The graphics CS from "cs_create" that will hold the buffer
484 * list and will be used for flushing.
485 */
486 struct radeon_winsys_cs *(*cs_add_const_preamble_ib)(struct radeon_winsys_cs *cs);
487 /**
488 * Destroy a command stream.
489 *
490 * \param cs A command stream to destroy.
491 */
492 void (*cs_destroy)(struct radeon_winsys_cs *cs);
493
494 /**
495 * Add a buffer. Each buffer used by a CS must be added using this function.
496 *
497 * \param cs Command stream
498 * \param buf Buffer
499 * \param usage Whether the buffer is used for read and/or write.
500 * \param domain Bitmask of the RADEON_DOMAIN_* flags.
501 * \param priority A higher number means a greater chance of being
502 * placed in the requested domain. 15 is the maximum.
503 * \return Buffer index.
504 */
505 unsigned (*cs_add_buffer)(struct radeon_winsys_cs *cs,
506 struct pb_buffer *buf,
507 enum radeon_bo_usage usage,
508 enum radeon_bo_domain domain,
509 enum radeon_bo_priority priority);
510
511 /**
512 * Return the index of an already-added buffer.
513 *
514 * Not supported on amdgpu. Drivers with GPUVM should not care about
515 * buffer indices.
516 *
517 * \param cs Command stream
518 * \param buf Buffer
519 * \return The buffer index, or -1 if the buffer has not been added.
520 */
521 int (*cs_lookup_buffer)(struct radeon_winsys_cs *cs,
522 struct pb_buffer *buf);
523
524 /**
525 * Return true if there is enough memory in VRAM and GTT for the buffers
526 * added so far. If the validation fails, all buffers which have
527 * been added since the last call of cs_validate will be removed and
528 * the CS will be flushed (provided there are still any buffers).
529 *
530 * \param cs A command stream to validate.
531 */
532 bool (*cs_validate)(struct radeon_winsys_cs *cs);
533
534 /**
535 * Check whether the given number of dwords is available in the IB.
536 * Optionally chain a new chunk of the IB if necessary and supported.
537 *
538 * \param cs A command stream.
539 * \param dw Number of CS dwords requested by the caller.
540 */
541 bool (*cs_check_space)(struct radeon_winsys_cs *cs, unsigned dw);
542
543 /**
544 * Return the buffer list.
545 *
546 * This is the buffer list as passed to the kernel, i.e. it only contains
547 * the parent buffers of sub-allocated buffers.
548 *
549 * \param cs Command stream
550 * \param list Returned buffer list. Set to NULL to query the count only.
551 * \return The buffer count.
552 */
553 unsigned (*cs_get_buffer_list)(struct radeon_winsys_cs *cs,
554 struct radeon_bo_list_item *list);
555
556 /**
557 * Flush a command stream.
558 *
559 * \param cs A command stream to flush.
560 * \param flags, RADEON_FLUSH_ASYNC or 0.
561 * \param fence Pointer to a fence. If non-NULL, a fence is inserted
562 * after the CS and is returned through this parameter.
563 * \return Negative POSIX error code or 0 for success.
564 * Asynchronous submissions never return an error.
565 */
566 int (*cs_flush)(struct radeon_winsys_cs *cs,
567 unsigned flags,
568 struct pipe_fence_handle **fence);
569
570 /**
571 * Create a fence before the CS is flushed.
572 * The user must flush manually to complete the initializaton of the fence.
573 * The fence must not be used before the flush.
574 */
575 struct pipe_fence_handle *(*cs_get_next_fence)(struct radeon_winsys_cs *cs);
576
577 /**
578 * Return true if a buffer is referenced by a command stream.
579 *
580 * \param cs A command stream.
581 * \param buf A winsys buffer.
582 */
583 bool (*cs_is_buffer_referenced)(struct radeon_winsys_cs *cs,
584 struct pb_buffer *buf,
585 enum radeon_bo_usage usage);
586
587 /**
588 * Request access to a feature for a command stream.
589 *
590 * \param cs A command stream.
591 * \param fid Feature ID, one of RADEON_FID_*
592 * \param enable Whether to enable or disable the feature.
593 */
594 bool (*cs_request_feature)(struct radeon_winsys_cs *cs,
595 enum radeon_feature_id fid,
596 bool enable);
597 /**
598 * Make sure all asynchronous flush of the cs have completed
599 *
600 * \param cs A command stream.
601 */
602 void (*cs_sync_flush)(struct radeon_winsys_cs *cs);
603
604 /**
605 * Wait for the fence and return true if the fence has been signalled.
606 * The timeout of 0 will only return the status.
607 * The timeout of PIPE_TIMEOUT_INFINITE will always wait until the fence
608 * is signalled.
609 */
610 bool (*fence_wait)(struct radeon_winsys *ws,
611 struct pipe_fence_handle *fence,
612 uint64_t timeout);
613
614 /**
615 * Reference counting for fences.
616 */
617 void (*fence_reference)(struct pipe_fence_handle **dst,
618 struct pipe_fence_handle *src);
619
620 /**
621 * Initialize surface
622 *
623 * \param ws The winsys this function is called from.
624 * \param tex Input texture description
625 * \param flags Bitmask of RADEON_SURF_* flags
626 * \param bpe Bytes per pixel, it can be different for Z buffers.
627 * \param mode Preferred tile mode. (linear, 1D, or 2D)
628 * \param surf Output structure
629 */
630 int (*surface_init)(struct radeon_winsys *ws,
631 const struct pipe_resource *tex,
632 unsigned flags, unsigned bpe,
633 enum radeon_surf_mode mode,
634 struct radeon_surf *surf);
635
636 uint64_t (*query_value)(struct radeon_winsys *ws,
637 enum radeon_value_id value);
638
639 bool (*read_registers)(struct radeon_winsys *ws, unsigned reg_offset,
640 unsigned num_registers, uint32_t *out);
641
642 const char* (*get_chip_name)(struct radeon_winsys *ws);
643 };
644
645 static inline bool radeon_emitted(struct radeon_winsys_cs *cs, unsigned num_dw)
646 {
647 return cs && (cs->prev_dw + cs->current.cdw > num_dw);
648 }
649
650 static inline void radeon_emit(struct radeon_winsys_cs *cs, uint32_t value)
651 {
652 cs->current.buf[cs->current.cdw++] = value;
653 }
654
655 static inline void radeon_emit_array(struct radeon_winsys_cs *cs,
656 const uint32_t *values, unsigned count)
657 {
658 memcpy(cs->current.buf + cs->current.cdw, values, count * 4);
659 cs->current.cdw += count;
660 }
661
662 #endif