1 /**********************************************************
2 * Copyright 2008-2009 VMware, Inc. All rights reserved.
4 * Permission is hereby granted, free of charge, to any person
5 * obtaining a copy of this software and associated documentation
6 * files (the "Software"), to deal in the Software without
7 * restriction, including without limitation the rights to use, copy,
8 * modify, merge, publish, distribute, sublicense, and/or sell copies
9 * of the Software, and to permit persons to whom the Software is
10 * furnished to do so, subject to the following conditions:
12 * The above copyright notice and this permission notice shall be
13 * included in all copies or substantial portions of the Software.
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
16 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
17 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
18 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
19 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
20 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
21 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
24 **********************************************************/
28 #include "pipe/p_state.h"
29 #include "pipe/p_defines.h"
30 #include "util/u_inlines.h"
31 #include "os/os_thread.h"
32 #include "util/u_format.h"
33 #include "util/u_math.h"
34 #include "util/u_memory.h"
36 #include "svga_screen.h"
37 #include "svga_context.h"
38 #include "svga_resource_texture.h"
39 #include "svga_resource_buffer.h"
40 #include "svga_sampler_view.h"
41 #include "svga_winsys.h"
42 #include "svga_debug.h"
45 /* XXX: This isn't a real hardware flag, but just a hack for kernel to
46 * know about primary surfaces. Find a better way to accomplish this.
48 #define SVGA3D_SURFACE_HINT_SCANOUT (1 << 9)
52 * Helper function and arrays
56 svga_translate_format(struct svga_screen
*ss
,
57 enum pipe_format format
)
61 case PIPE_FORMAT_B8G8R8A8_UNORM
:
62 return SVGA3D_A8R8G8B8
;
63 case PIPE_FORMAT_B8G8R8X8_UNORM
:
64 return SVGA3D_X8R8G8B8
;
66 /* Required for GL2.1:
68 case PIPE_FORMAT_B8G8R8A8_SRGB
:
69 return SVGA3D_A8R8G8B8
;
71 case PIPE_FORMAT_B5G6R5_UNORM
:
73 case PIPE_FORMAT_B5G5R5A1_UNORM
:
74 return SVGA3D_A1R5G5B5
;
75 case PIPE_FORMAT_B4G4R4A4_UNORM
:
76 return SVGA3D_A4R4G4B4
;
79 /* XXX: Doesn't seem to work properly.
80 case PIPE_FORMAT_Z32_UNORM:
83 case PIPE_FORMAT_Z16_UNORM
:
85 case PIPE_FORMAT_S8_USCALED_Z24_UNORM
:
86 return ss
->depth
.s8z24
;
87 case PIPE_FORMAT_X8Z24_UNORM
:
88 return ss
->depth
.x8z24
;
90 case PIPE_FORMAT_A8_UNORM
:
92 case PIPE_FORMAT_L8_UNORM
:
93 return SVGA3D_LUMINANCE8
;
95 case PIPE_FORMAT_DXT1_RGB
:
96 case PIPE_FORMAT_DXT1_RGBA
:
98 case PIPE_FORMAT_DXT3_RGBA
:
100 case PIPE_FORMAT_DXT5_RGBA
:
104 return SVGA3D_FORMAT_INVALID
;
110 svga_translate_format_render(struct svga_screen
*ss
,
111 enum pipe_format format
)
114 case PIPE_FORMAT_B8G8R8A8_UNORM
:
115 case PIPE_FORMAT_B8G8R8X8_UNORM
:
116 case PIPE_FORMAT_B5G5R5A1_UNORM
:
117 case PIPE_FORMAT_B4G4R4A4_UNORM
:
118 case PIPE_FORMAT_B5G6R5_UNORM
:
119 case PIPE_FORMAT_S8_USCALED_Z24_UNORM
:
120 case PIPE_FORMAT_X8Z24_UNORM
:
121 case PIPE_FORMAT_Z32_UNORM
:
122 case PIPE_FORMAT_Z16_UNORM
:
123 case PIPE_FORMAT_L8_UNORM
:
124 return svga_translate_format(ss
, format
);
127 return SVGA3D_FORMAT_INVALID
;
133 svga_transfer_dma_band(struct svga_context
*svga
,
134 struct svga_transfer
*st
,
135 SVGA3dTransferType transfer
,
136 unsigned y
, unsigned h
, unsigned srcy
,
137 SVGA3dSurfaceDMAFlags flags
)
139 struct svga_texture
*texture
= svga_texture(st
->base
.resource
);
143 box
.x
= st
->base
.box
.x
;
145 box
.z
= st
->base
.box
.z
;
146 box
.w
= st
->base
.box
.width
;
153 if (st
->base
.resource
->target
== PIPE_TEXTURE_CUBE
) {
154 st
->face
= st
->base
.box
.z
;
160 SVGA_DBG(DEBUG_DMA
, "dma %s sid %p, face %u, (%u, %u, %u) - (%u, %u, %u), %ubpp\n",
161 transfer
== SVGA3D_WRITE_HOST_VRAM
? "to" : "from",
167 st
->base
.box
.x
+ st
->base
.box
.width
,
170 util_format_get_blocksize(texture
->b
.b
.format
) * 8 /
171 (util_format_get_blockwidth(texture
->b
.b
.format
)*util_format_get_blockheight(texture
->b
.b
.format
)));
173 ret
= SVGA3D_SurfaceDMA(svga
->swc
, st
, transfer
, &box
, 1, flags
);
175 svga_context_flush(svga
, NULL
);
176 ret
= SVGA3D_SurfaceDMA(svga
->swc
, st
, transfer
, &box
, 1, flags
);
177 assert(ret
== PIPE_OK
);
183 svga_transfer_dma(struct svga_context
*svga
,
184 struct svga_transfer
*st
,
185 SVGA3dTransferType transfer
,
186 SVGA3dSurfaceDMAFlags flags
)
188 struct svga_texture
*texture
= svga_texture(st
->base
.resource
);
189 struct svga_screen
*screen
= svga_screen(texture
->b
.b
.screen
);
190 struct svga_winsys_screen
*sws
= screen
->sws
;
191 struct pipe_fence_handle
*fence
= NULL
;
193 if (transfer
== SVGA3D_READ_HOST_VRAM
) {
194 SVGA_DBG(DEBUG_PERF
, "%s: readback transfer\n", __FUNCTION__
);
197 /* Ensure any pending operations on host surfaces are queued on the command
200 svga_surfaces_flush( svga
);
203 /* Do the DMA transfer in a single go */
205 svga_transfer_dma_band(svga
, st
, transfer
,
206 st
->base
.box
.y
, st
->base
.box
.height
, 0,
209 if(transfer
== SVGA3D_READ_HOST_VRAM
) {
210 svga_context_flush(svga
, &fence
);
211 sws
->fence_finish(sws
, fence
, 0);
212 sws
->fence_reference(sws
, &fence
, NULL
);
217 unsigned blockheight
= util_format_get_blockheight(st
->base
.resource
->format
);
218 h
= st
->hw_nblocksy
* blockheight
;
220 for(y
= 0; y
< st
->base
.box
.height
; y
+= h
) {
221 unsigned offset
, length
;
224 if (y
+ h
> st
->base
.box
.height
)
225 h
= st
->base
.box
.height
- y
;
227 /* Transfer band must be aligned to pixel block boundaries */
228 assert(y
% blockheight
== 0);
229 assert(h
% blockheight
== 0);
231 offset
= y
* st
->base
.stride
/ blockheight
;
232 length
= h
* st
->base
.stride
/ blockheight
;
234 sw
= (uint8_t *)st
->swbuf
+ offset
;
236 if (transfer
== SVGA3D_WRITE_HOST_VRAM
) {
237 unsigned usage
= PIPE_TRANSFER_WRITE
;
239 /* Wait for the previous DMAs to complete */
240 /* TODO: keep one DMA (at half the size) in the background */
242 svga_context_flush(svga
, NULL
);
243 usage
|= PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE
;
246 hw
= sws
->buffer_map(sws
, st
->hwbuf
, usage
);
249 memcpy(hw
, sw
, length
);
250 sws
->buffer_unmap(sws
, st
->hwbuf
);
254 svga_transfer_dma_band(svga
, st
, transfer
, y
, h
, srcy
, flags
);
257 * Prevent the texture contents to be discarded on the next band
261 flags
.discard
= FALSE
;
263 if(transfer
== SVGA3D_READ_HOST_VRAM
) {
264 svga_context_flush(svga
, &fence
);
265 sws
->fence_finish(sws
, fence
, 0);
267 hw
= sws
->buffer_map(sws
, st
->hwbuf
, PIPE_TRANSFER_READ
);
270 memcpy(sw
, hw
, length
);
271 sws
->buffer_unmap(sws
, st
->hwbuf
);
283 svga_texture_get_handle(struct pipe_screen
*screen
,
284 struct pipe_resource
*texture
,
285 struct winsys_handle
*whandle
)
287 struct svga_winsys_screen
*sws
= svga_winsys_screen(texture
->screen
);
290 assert(svga_texture(texture
)->key
.cachable
== 0);
291 svga_texture(texture
)->key
.cachable
= 0;
292 stride
= util_format_get_nblocksx(texture
->format
, texture
->width0
) *
293 util_format_get_blocksize(texture
->format
);
294 return sws
->surface_get_handle(sws
, svga_texture(texture
)->handle
, stride
, whandle
);
299 svga_texture_destroy(struct pipe_screen
*screen
,
300 struct pipe_resource
*pt
)
302 struct svga_screen
*ss
= svga_screen(screen
);
303 struct svga_texture
*tex
= (struct svga_texture
*)pt
;
305 ss
->texture_timestamp
++;
307 svga_sampler_view_reference(&tex
->cached_view
, NULL
);
310 DBG("%s deleting %p\n", __FUNCTION__, (void *) tex);
312 SVGA_DBG(DEBUG_DMA
, "unref sid %p (texture)\n", tex
->handle
);
313 svga_screen_surface_destroy(ss
, &tex
->key
, &tex
->handle
);
324 /* XXX: Still implementing this as if it was a screen function, but
325 * can now modify it to queue transfers on the context.
327 static struct pipe_transfer
*
328 svga_texture_get_transfer(struct pipe_context
*pipe
,
329 struct pipe_resource
*texture
,
332 const struct pipe_box
*box
)
334 struct svga_context
*svga
= svga_context(pipe
);
335 struct svga_screen
*ss
= svga_screen(pipe
->screen
);
336 struct svga_winsys_screen
*sws
= ss
->sws
;
337 struct svga_transfer
*st
;
338 unsigned nblocksx
= util_format_get_nblocksx(texture
->format
, box
->width
);
339 unsigned nblocksy
= util_format_get_nblocksy(texture
->format
, box
->height
);
341 /* We can't map texture storage directly */
342 if (usage
& PIPE_TRANSFER_MAP_DIRECTLY
)
345 assert(box
->depth
== 1);
346 st
= CALLOC_STRUCT(svga_transfer
);
350 pipe_resource_reference(&st
->base
.resource
, texture
);
351 st
->base
.level
= level
;
352 st
->base
.usage
= usage
;
354 st
->base
.stride
= nblocksx
*util_format_get_blocksize(texture
->format
);
355 st
->base
.layer_stride
= 0;
357 st
->hw_nblocksy
= nblocksy
;
359 st
->hwbuf
= svga_winsys_buffer_create(svga
,
362 st
->hw_nblocksy
*st
->base
.stride
);
363 while(!st
->hwbuf
&& (st
->hw_nblocksy
/= 2)) {
364 st
->hwbuf
= svga_winsys_buffer_create(svga
,
367 st
->hw_nblocksy
*st
->base
.stride
);
373 if(st
->hw_nblocksy
< nblocksy
) {
374 /* We couldn't allocate a hardware buffer big enough for the transfer,
375 * so allocate regular malloc memory instead */
377 debug_printf("%s: failed to allocate %u KB of DMA, "
378 "splitting into %u x %u KB DMA transfers\n",
380 (nblocksy
*st
->base
.stride
+ 1023)/1024,
381 (nblocksy
+ st
->hw_nblocksy
- 1)/st
->hw_nblocksy
,
382 (st
->hw_nblocksy
*st
->base
.stride
+ 1023)/1024);
385 st
->swbuf
= MALLOC(nblocksy
*st
->base
.stride
);
390 if (usage
& PIPE_TRANSFER_READ
) {
391 SVGA3dSurfaceDMAFlags flags
;
392 memset(&flags
, 0, sizeof flags
);
393 svga_transfer_dma(svga
, st
, SVGA3D_READ_HOST_VRAM
, flags
);
399 sws
->buffer_destroy(sws
, st
->hwbuf
);
406 /* XXX: Still implementing this as if it was a screen function, but
407 * can now modify it to queue transfers on the context.
410 svga_texture_transfer_map( struct pipe_context
*pipe
,
411 struct pipe_transfer
*transfer
)
413 struct svga_screen
*ss
= svga_screen(pipe
->screen
);
414 struct svga_winsys_screen
*sws
= ss
->sws
;
415 struct svga_transfer
*st
= svga_transfer(transfer
);
420 /* The wait for read transfers already happened when svga_transfer_dma
422 return sws
->buffer_map(sws
, st
->hwbuf
, transfer
->usage
);
426 /* XXX: Still implementing this as if it was a screen function, but
427 * can now modify it to queue transfers on the context.
430 svga_texture_transfer_unmap(struct pipe_context
*pipe
,
431 struct pipe_transfer
*transfer
)
433 struct svga_screen
*ss
= svga_screen(pipe
->screen
);
434 struct svga_winsys_screen
*sws
= ss
->sws
;
435 struct svga_transfer
*st
= svga_transfer(transfer
);
438 sws
->buffer_unmap(sws
, st
->hwbuf
);
443 svga_texture_transfer_destroy(struct pipe_context
*pipe
,
444 struct pipe_transfer
*transfer
)
446 struct svga_context
*svga
= svga_context(pipe
);
447 struct svga_texture
*tex
= svga_texture(transfer
->resource
);
448 struct svga_screen
*ss
= svga_screen(pipe
->screen
);
449 struct svga_winsys_screen
*sws
= ss
->sws
;
450 struct svga_transfer
*st
= svga_transfer(transfer
);
452 if (st
->base
.usage
& PIPE_TRANSFER_WRITE
) {
453 SVGA3dSurfaceDMAFlags flags
;
455 memset(&flags
, 0, sizeof flags
);
456 if (transfer
->usage
& PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE
) {
457 flags
.discard
= TRUE
;
459 if (transfer
->usage
& PIPE_TRANSFER_UNSYNCHRONIZED
) {
460 flags
.unsynchronized
= TRUE
;
463 svga_transfer_dma(svga
, st
, SVGA3D_WRITE_HOST_VRAM
, flags
);
464 ss
->texture_timestamp
++;
465 tex
->view_age
[transfer
->level
] = ++(tex
->age
);
466 if (transfer
->resource
->target
== PIPE_TEXTURE_CUBE
)
467 tex
->defined
[transfer
->box
.z
][transfer
->level
] = TRUE
;
469 tex
->defined
[0][transfer
->level
] = TRUE
;
472 pipe_resource_reference(&st
->base
.resource
, NULL
);
474 sws
->buffer_destroy(sws
, st
->hwbuf
);
482 struct u_resource_vtbl svga_texture_vtbl
=
484 svga_texture_get_handle
, /* get_handle */
485 svga_texture_destroy
, /* resource_destroy */
486 svga_texture_get_transfer
, /* get_transfer */
487 svga_texture_transfer_destroy
, /* transfer_destroy */
488 svga_texture_transfer_map
, /* transfer_map */
489 u_default_transfer_flush_region
, /* transfer_flush_region */
490 svga_texture_transfer_unmap
, /* transfer_unmap */
491 u_default_transfer_inline_write
/* transfer_inline_write */
497 struct pipe_resource
*
498 svga_texture_create(struct pipe_screen
*screen
,
499 const struct pipe_resource
*template)
501 struct svga_screen
*svgascreen
= svga_screen(screen
);
502 struct svga_texture
*tex
= CALLOC_STRUCT(svga_texture
);
507 tex
->b
.b
= *template;
508 tex
->b
.vtbl
= &svga_texture_vtbl
;
509 pipe_reference_init(&tex
->b
.b
.reference
, 1);
510 tex
->b
.b
.screen
= screen
;
512 assert(template->last_level
< SVGA_MAX_TEXTURE_LEVELS
);
513 if(template->last_level
>= SVGA_MAX_TEXTURE_LEVELS
)
517 tex
->key
.size
.width
= template->width0
;
518 tex
->key
.size
.height
= template->height0
;
519 tex
->key
.size
.depth
= template->depth0
;
521 if(template->target
== PIPE_TEXTURE_CUBE
) {
522 tex
->key
.flags
|= SVGA3D_SURFACE_CUBEMAP
;
523 tex
->key
.numFaces
= 6;
526 tex
->key
.numFaces
= 1;
529 /* XXX: Disabled for now */
530 tex
->key
.cachable
= 0;
532 if (template->bind
& PIPE_BIND_SAMPLER_VIEW
)
533 tex
->key
.flags
|= SVGA3D_SURFACE_HINT_TEXTURE
;
535 if (template->bind
& PIPE_BIND_DISPLAY_TARGET
) {
536 tex
->key
.cachable
= 0;
539 if (template->bind
& PIPE_BIND_SHARED
) {
540 tex
->key
.cachable
= 0;
543 if (template->bind
& PIPE_BIND_SCANOUT
) {
544 tex
->key
.flags
|= SVGA3D_SURFACE_HINT_SCANOUT
;
545 tex
->key
.cachable
= 0;
549 * XXX: Never pass the SVGA3D_SURFACE_HINT_RENDERTARGET hint. Mesa cannot
550 * know beforehand whether a texture will be used as a rendertarget or not
551 * and it always requests PIPE_BIND_RENDER_TARGET, therefore
552 * passing the SVGA3D_SURFACE_HINT_RENDERTARGET here defeats its purpose.
555 if((template->bind
& PIPE_BIND_RENDER_TARGET
) &&
556 !util_format_is_s3tc(template->format
))
557 tex
->key
.flags
|= SVGA3D_SURFACE_HINT_RENDERTARGET
;
560 if(template->bind
& PIPE_BIND_DEPTH_STENCIL
)
561 tex
->key
.flags
|= SVGA3D_SURFACE_HINT_DEPTHSTENCIL
;
563 tex
->key
.numMipLevels
= template->last_level
+ 1;
565 tex
->key
.format
= svga_translate_format(svgascreen
, template->format
);
566 if(tex
->key
.format
== SVGA3D_FORMAT_INVALID
)
569 SVGA_DBG(DEBUG_DMA
, "surface_create for texture\n", tex
->handle
);
570 tex
->handle
= svga_screen_surface_create(svgascreen
, &tex
->key
);
572 SVGA_DBG(DEBUG_DMA
, " --> got sid %p (texture)\n", tex
->handle
);
574 debug_reference(&tex
->b
.b
.reference
,
575 (debug_reference_descriptor
)debug_describe_resource
, 0);
588 struct pipe_resource
*
589 svga_texture_from_handle(struct pipe_screen
*screen
,
590 const struct pipe_resource
*template,
591 struct winsys_handle
*whandle
)
593 struct svga_winsys_screen
*sws
= svga_winsys_screen(screen
);
594 struct svga_winsys_surface
*srf
;
595 struct svga_texture
*tex
;
596 enum SVGA3dSurfaceFormat format
= 0;
599 /* Only supports one type */
600 if ((template->target
!= PIPE_TEXTURE_2D
&&
601 template->target
!= PIPE_TEXTURE_RECT
) ||
602 template->last_level
!= 0 ||
603 template->depth0
!= 1) {
607 srf
= sws
->surface_from_handle(sws
, whandle
, &format
);
612 if (svga_translate_format(svga_screen(screen
), template->format
) != format
) {
613 unsigned f1
= svga_translate_format(svga_screen(screen
), template->format
);
614 unsigned f2
= format
;
616 /* It's okay for XRGB and ARGB or depth with/out stencil to get mixed up */
617 if ( !( (f1
== SVGA3D_X8R8G8B8
&& f2
== SVGA3D_A8R8G8B8
) ||
618 (f1
== SVGA3D_A8R8G8B8
&& f2
== SVGA3D_X8R8G8B8
) ||
619 (f1
== SVGA3D_Z_D24X8
&& f2
== SVGA3D_Z_D24S8
) ||
620 (f1
== SVGA3D_Z_DF24
&& f2
== SVGA3D_Z_D24S8_INT
) ) ) {
621 debug_printf("%s wrong format %u != %u\n", __FUNCTION__
, f1
, f2
);
626 tex
= CALLOC_STRUCT(svga_texture
);
630 tex
->b
.b
= *template;
631 tex
->b
.vtbl
= &svga_texture_vtbl
;
632 pipe_reference_init(&tex
->b
.b
.reference
, 1);
633 tex
->b
.b
.screen
= screen
;
635 if (format
== SVGA3D_X8R8G8B8
)
636 tex
->b
.b
.format
= PIPE_FORMAT_B8G8R8X8_UNORM
;
637 else if (format
== SVGA3D_A8R8G8B8
)
638 tex
->b
.b
.format
= PIPE_FORMAT_B8G8R8A8_UNORM
;
643 SVGA_DBG(DEBUG_DMA
, "wrap surface sid %p\n", srf
);
645 tex
->key
.cachable
= 0;