2 * Copyright 2006 VMware, Inc.
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the
7 * "Software"), to deal in the Software without restriction, including
8 * without limitation the rights to use, copy, modify, merge, publish,
9 * distribute, sublicense, and/or sell copies of the Software, and to
10 * permit persons to whom the Software is furnished to do so, subject to
11 * the following conditions:
13 * The above copyright notice and this permission notice (including the
14 * next paragraph) shall be included in all copies or substantial portions
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
18 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
19 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
20 * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
21 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
22 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
23 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26 #include "main/enums.h"
27 #include "main/imports.h"
28 #include "main/macros.h"
29 #include "main/mtypes.h"
30 #include "main/fbobject.h"
31 #include "main/framebuffer.h"
32 #include "main/renderbuffer.h"
33 #include "main/context.h"
34 #include "main/teximage.h"
35 #include "main/image.h"
36 #include "main/condrender.h"
37 #include "util/hash_table.h"
40 #include "swrast/swrast.h"
41 #include "drivers/common/meta.h"
43 #include "intel_batchbuffer.h"
44 #include "intel_buffers.h"
45 #include "intel_blit.h"
46 #include "intel_fbo.h"
47 #include "intel_mipmap_tree.h"
48 #include "intel_image.h"
49 #include "intel_screen.h"
50 #include "intel_tex.h"
51 #include "brw_context.h"
52 #include "brw_defines.h"
54 #define FILE_DEBUG_FLAG DEBUG_FBO
56 /** Called by gl_renderbuffer::Delete() */
58 intel_delete_renderbuffer(struct gl_context
*ctx
, struct gl_renderbuffer
*rb
)
60 struct intel_renderbuffer
*irb
= intel_renderbuffer(rb
);
64 intel_miptree_release(&irb
->mt
);
65 intel_miptree_release(&irb
->singlesample_mt
);
67 _mesa_delete_renderbuffer(ctx
, rb
);
71 * \brief Downsample a winsys renderbuffer from mt to singlesample_mt.
73 * If the miptree needs no downsample, then skip.
76 intel_renderbuffer_downsample(struct brw_context
*brw
,
77 struct intel_renderbuffer
*irb
)
79 if (!irb
->need_downsample
)
81 intel_miptree_updownsample(brw
, irb
->mt
, irb
->singlesample_mt
);
82 irb
->need_downsample
= false;
86 * \brief Upsample a winsys renderbuffer from singlesample_mt to mt.
88 * The upsample is done unconditionally.
91 intel_renderbuffer_upsample(struct brw_context
*brw
,
92 struct intel_renderbuffer
*irb
)
94 assert(!irb
->need_downsample
);
96 intel_miptree_updownsample(brw
, irb
->singlesample_mt
, irb
->mt
);
100 * \see dd_function_table::MapRenderbuffer
103 intel_map_renderbuffer(struct gl_context
*ctx
,
104 struct gl_renderbuffer
*rb
,
105 GLuint x
, GLuint y
, GLuint w
, GLuint h
,
110 struct brw_context
*brw
= brw_context(ctx
);
111 struct swrast_renderbuffer
*srb
= (struct swrast_renderbuffer
*)rb
;
112 struct intel_renderbuffer
*irb
= intel_renderbuffer(rb
);
113 struct intel_mipmap_tree
*mt
;
118 /* this is a malloc'd renderbuffer (accum buffer), not an irb */
119 GLint bpp
= _mesa_get_format_bytes(rb
->Format
);
120 GLint rowStride
= srb
->RowStride
;
121 *out_map
= (GLubyte
*) srb
->Buffer
+ y
* rowStride
+ x
* bpp
;
122 *out_stride
= rowStride
;
126 intel_prepare_render(brw
);
128 /* The MapRenderbuffer API should always return a single-sampled mapping.
129 * The case we are asked to map multisampled RBs is in glReadPixels() (or
130 * swrast paths like glCopyTexImage()) from a window-system MSAA buffer,
131 * and GL expects an automatic resolve to happen.
133 * If it's a color miptree, there is a ->singlesample_mt which wraps the
134 * actual window system renderbuffer (which we may resolve to at any time),
135 * while the miptree itself is our driver-private allocation. If it's a
136 * depth or stencil miptree, we have a private MSAA buffer and no shared
137 * singlesample buffer, and since we don't expect anybody to ever actually
138 * resolve it, we just make a temporary singlesample buffer now when we
141 if (rb
->NumSamples
> 1) {
142 if (!irb
->singlesample_mt
) {
143 irb
->singlesample_mt
=
144 intel_miptree_create_for_renderbuffer(brw
, irb
->mt
->format
,
145 rb
->Width
, rb
->Height
,
147 if (!irb
->singlesample_mt
)
149 irb
->singlesample_mt_is_tmp
= true;
150 irb
->need_downsample
= true;
153 intel_renderbuffer_downsample(brw
, irb
);
154 mt
= irb
->singlesample_mt
;
156 irb
->need_map_upsample
= mode
& GL_MAP_WRITE_BIT
;
161 /* For a window-system renderbuffer, we need to flip the mapping we receive
162 * upside-down. So we need to ask for a rectangle on flipped vertically, and
163 * we then return a pointer to the bottom of it with a negative stride.
166 y
= rb
->Height
- y
- h
;
169 intel_miptree_map(brw
, mt
, irb
->mt_level
, irb
->mt_layer
,
170 x
, y
, w
, h
, mode
, &map
, &stride
);
173 map
+= (h
- 1) * stride
;
177 DBG("%s: rb %d (%s) mt mapped: (%d, %d) (%dx%d) -> %p/%"PRIdPTR
"\n",
178 __func__
, rb
->Name
, _mesa_get_format_name(rb
->Format
),
179 x
, y
, w
, h
, map
, stride
);
182 *out_stride
= stride
;
191 * \see dd_function_table::UnmapRenderbuffer
194 intel_unmap_renderbuffer(struct gl_context
*ctx
,
195 struct gl_renderbuffer
*rb
)
197 struct brw_context
*brw
= brw_context(ctx
);
198 struct swrast_renderbuffer
*srb
= (struct swrast_renderbuffer
*)rb
;
199 struct intel_renderbuffer
*irb
= intel_renderbuffer(rb
);
200 struct intel_mipmap_tree
*mt
;
202 DBG("%s: rb %d (%s)\n", __func__
,
203 rb
->Name
, _mesa_get_format_name(rb
->Format
));
206 /* this is a malloc'd renderbuffer (accum buffer) */
211 if (rb
->NumSamples
> 1) {
212 mt
= irb
->singlesample_mt
;
217 intel_miptree_unmap(brw
, mt
, irb
->mt_level
, irb
->mt_layer
);
219 if (irb
->need_map_upsample
) {
220 intel_renderbuffer_upsample(brw
, irb
);
221 irb
->need_map_upsample
= false;
224 if (irb
->singlesample_mt_is_tmp
)
225 intel_miptree_release(&irb
->singlesample_mt
);
230 * Round up the requested multisample count to the next supported sample size.
233 intel_quantize_num_samples(struct intel_screen
*intel
, unsigned num_samples
)
235 const int *msaa_modes
= intel_supported_msaa_modes(intel
);
236 int quantized_samples
= 0;
238 for (int i
= 0; msaa_modes
[i
] != -1; ++i
) {
239 if (msaa_modes
[i
] >= num_samples
)
240 quantized_samples
= msaa_modes
[i
];
245 return quantized_samples
;
249 intel_renderbuffer_format(struct gl_context
* ctx
, GLenum internalFormat
)
251 struct brw_context
*brw
= brw_context(ctx
);
253 switch (internalFormat
) {
255 /* Use the same format-choice logic as for textures.
256 * Renderbuffers aren't any different from textures for us,
257 * except they're less useful because you can't texture with
260 return ctx
->Driver
.ChooseTextureFormat(ctx
, GL_TEXTURE_2D
,
264 case GL_STENCIL_INDEX
:
265 case GL_STENCIL_INDEX1_EXT
:
266 case GL_STENCIL_INDEX4_EXT
:
267 case GL_STENCIL_INDEX8_EXT
:
268 case GL_STENCIL_INDEX16_EXT
:
269 /* These aren't actual texture formats, so force them here. */
270 if (brw
->has_separate_stencil
) {
271 return MESA_FORMAT_S_UINT8
;
273 assert(!brw
->must_use_separate_stencil
);
274 return MESA_FORMAT_Z24_UNORM_S8_UINT
;
280 intel_alloc_private_renderbuffer_storage(struct gl_context
* ctx
, struct gl_renderbuffer
*rb
,
281 GLenum internalFormat
,
282 GLuint width
, GLuint height
)
284 struct brw_context
*brw
= brw_context(ctx
);
285 struct intel_screen
*screen
= brw
->screen
;
286 struct intel_renderbuffer
*irb
= intel_renderbuffer(rb
);
288 assert(rb
->Format
!= MESA_FORMAT_NONE
);
290 rb
->NumSamples
= intel_quantize_num_samples(screen
, rb
->NumSamples
);
293 rb
->_BaseFormat
= _mesa_get_format_base_format(rb
->Format
);
295 intel_miptree_release(&irb
->mt
);
297 DBG("%s: %s: %s (%dx%d)\n", __func__
,
298 _mesa_enum_to_string(internalFormat
),
299 _mesa_get_format_name(rb
->Format
), width
, height
);
301 if (width
== 0 || height
== 0)
304 irb
->mt
= intel_miptree_create_for_renderbuffer(brw
, rb
->Format
,
310 irb
->layer_count
= 1;
316 * Called via glRenderbufferStorageEXT() to set the format and allocate
317 * storage for a user-created renderbuffer.
320 intel_alloc_renderbuffer_storage(struct gl_context
* ctx
, struct gl_renderbuffer
*rb
,
321 GLenum internalFormat
,
322 GLuint width
, GLuint height
)
324 rb
->Format
= intel_renderbuffer_format(ctx
, internalFormat
);
325 return intel_alloc_private_renderbuffer_storage(ctx
, rb
, internalFormat
, width
, height
);
329 intel_image_target_renderbuffer_storage(struct gl_context
*ctx
,
330 struct gl_renderbuffer
*rb
,
333 struct brw_context
*brw
= brw_context(ctx
);
334 struct intel_renderbuffer
*irb
;
335 __DRIscreen
*dri_screen
= brw
->screen
->driScrnPriv
;
338 image
= dri_screen
->dri2
.image
->lookupEGLImage(dri_screen
, image_handle
,
339 dri_screen
->loaderPrivate
);
343 if (image
->planar_format
&& image
->planar_format
->nplanes
> 1) {
344 _mesa_error(ctx
, GL_INVALID_OPERATION
,
345 "glEGLImageTargetRenderbufferStorage(planar buffers are not "
346 "supported as render targets.)");
350 /* __DRIimage is opaque to the core so it has to be checked here */
351 if (!brw
->mesa_format_supports_render
[image
->format
]) {
352 _mesa_error(ctx
, GL_INVALID_OPERATION
,
353 "glEGLImageTargetRenderbufferStorage(unsupported image format)");
357 irb
= intel_renderbuffer(rb
);
358 intel_miptree_release(&irb
->mt
);
360 /* Disable creation of the miptree's aux buffers because the driver exposes
361 * no EGL API to manage them. That is, there is no API for resolving the aux
362 * buffer's content to the main buffer nor for invalidating the aux buffer's
365 irb
->mt
= intel_miptree_create_for_dri_image(brw
, image
, GL_TEXTURE_2D
,
366 ISL_COLORSPACE_NONE
, false);
370 rb
->InternalFormat
= image
->internal_format
;
371 rb
->Width
= image
->width
;
372 rb
->Height
= image
->height
;
373 rb
->Format
= image
->format
;
374 rb
->_BaseFormat
= _mesa_get_format_base_format(image
->format
);
375 rb
->NeedsFinishRenderTexture
= true;
376 irb
->layer_count
= 1;
380 * Called by _mesa_resize_framebuffer() for each hardware renderbuffer when a
381 * window system framebuffer is resized.
383 * Any actual buffer reallocations for hardware renderbuffers (which would
384 * have triggered _mesa_resize_framebuffer()) were done by
385 * intel_process_dri2_buffer().
388 intel_alloc_window_storage(struct gl_context
* ctx
, struct gl_renderbuffer
*rb
,
389 GLenum internalFormat
, GLuint width
, GLuint height
)
392 assert(rb
->Name
== 0);
395 rb
->InternalFormat
= internalFormat
;
400 /** Dummy function for gl_renderbuffer::AllocStorage() */
402 intel_nop_alloc_storage(struct gl_context
* ctx
, struct gl_renderbuffer
*rb
,
403 GLenum internalFormat
, GLuint width
, GLuint height
)
406 (void) internalFormat
;
409 _mesa_problem(ctx
, "intel_nop_alloc_storage should never be called.");
414 * Create an intel_renderbuffer for a __DRIdrawable. This function is
415 * unrelated to GL renderbuffers (that is, those created by
416 * glGenRenderbuffers).
418 * \param num_samples must be quantized.
420 struct intel_renderbuffer
*
421 intel_create_winsys_renderbuffer(struct intel_screen
*screen
,
422 mesa_format format
, unsigned num_samples
)
424 struct intel_renderbuffer
*irb
= CALLOC_STRUCT(intel_renderbuffer
);
428 struct gl_renderbuffer
*rb
= &irb
->Base
.Base
;
429 irb
->layer_count
= 1;
431 _mesa_init_renderbuffer(rb
, 0);
432 rb
->ClassID
= INTEL_RB_CLASS
;
433 rb
->NumSamples
= num_samples
;
435 /* The base format and internal format must be derived from the user-visible
436 * format (that is, the gl_config's format), even if we internally use
437 * choose a different format for the renderbuffer. Otherwise, rendering may
438 * use incorrect channel write masks.
440 rb
->_BaseFormat
= _mesa_get_format_base_format(format
);
441 rb
->InternalFormat
= rb
->_BaseFormat
;
444 if (!screen
->mesa_format_supports_render
[rb
->Format
]) {
445 /* The glRenderbufferStorage paths in core Mesa detect if the driver
446 * does not support the user-requested format, and then searches for
447 * a falback format. The DRI code bypasses core Mesa, though. So we do
448 * the fallbacks here.
450 * We must support MESA_FORMAT_R8G8B8X8 on Android because the Android
451 * framework requires HAL_PIXEL_FORMAT_RGBX8888 winsys surfaces.
453 rb
->Format
= _mesa_format_fallback_rgbx_to_rgba(rb
->Format
);
454 assert(screen
->mesa_format_supports_render
[rb
->Format
]);
457 /* intel-specific methods */
458 rb
->Delete
= intel_delete_renderbuffer
;
459 rb
->AllocStorage
= intel_alloc_window_storage
;
465 * Private window-system buffers (as opposed to ones shared with the display
466 * server created with intel_create_winsys_renderbuffer()) are most similar in their
467 * handling to user-created renderbuffers, but they have a resize handler that
468 * may be called at intel_update_renderbuffers() time.
470 * \param num_samples must be quantized.
472 struct intel_renderbuffer
*
473 intel_create_private_renderbuffer(struct intel_screen
*screen
,
474 mesa_format format
, unsigned num_samples
)
476 struct intel_renderbuffer
*irb
;
478 irb
= intel_create_winsys_renderbuffer(screen
, format
, num_samples
);
479 irb
->Base
.Base
.AllocStorage
= intel_alloc_private_renderbuffer_storage
;
485 * Create a new renderbuffer object.
486 * Typically called via glBindRenderbufferEXT().
488 static struct gl_renderbuffer
*
489 intel_new_renderbuffer(struct gl_context
* ctx
, GLuint name
)
491 struct intel_renderbuffer
*irb
;
492 struct gl_renderbuffer
*rb
;
494 irb
= CALLOC_STRUCT(intel_renderbuffer
);
496 _mesa_error(ctx
, GL_OUT_OF_MEMORY
, "creating renderbuffer");
500 rb
= &irb
->Base
.Base
;
502 _mesa_init_renderbuffer(rb
, name
);
503 rb
->ClassID
= INTEL_RB_CLASS
;
505 /* intel-specific methods */
506 rb
->Delete
= intel_delete_renderbuffer
;
507 rb
->AllocStorage
= intel_alloc_renderbuffer_storage
;
508 /* span routines set in alloc_storage function */
514 intel_renderbuffer_update_wrapper(struct brw_context
*brw
,
515 struct intel_renderbuffer
*irb
,
516 struct gl_texture_image
*image
,
520 struct gl_renderbuffer
*rb
= &irb
->Base
.Base
;
521 struct intel_texture_image
*intel_image
= intel_texture_image(image
);
522 struct intel_mipmap_tree
*mt
= intel_image
->mt
;
523 int level
= image
->Level
;
525 rb
->AllocStorage
= intel_nop_alloc_storage
;
527 /* adjust for texture view parameters */
528 layer
+= image
->TexObject
->MinLayer
;
529 level
+= image
->TexObject
->MinLevel
;
531 intel_miptree_check_level_layer(mt
, level
, layer
);
532 irb
->mt_level
= level
;
534 int layer_multiplier
;
535 switch (mt
->msaa_layout
) {
536 case INTEL_MSAA_LAYOUT_UMS
:
537 case INTEL_MSAA_LAYOUT_CMS
:
538 layer_multiplier
= MAX2(mt
->num_samples
, 1);
542 layer_multiplier
= 1;
545 irb
->mt_layer
= layer_multiplier
* layer
;
548 irb
->layer_count
= 1;
549 } else if (mt
->target
!= GL_TEXTURE_3D
&& image
->TexObject
->NumLayers
> 0) {
550 irb
->layer_count
= image
->TexObject
->NumLayers
;
552 irb
->layer_count
= mt
->level
[level
].depth
/ layer_multiplier
;
555 intel_miptree_reference(&irb
->mt
, mt
);
557 intel_renderbuffer_set_draw_offset(irb
);
559 if (mt
->aux_usage
== ISL_AUX_USAGE_HIZ
&& !mt
->hiz_buf
) {
560 intel_miptree_alloc_hiz(brw
, mt
);
569 intel_renderbuffer_set_draw_offset(struct intel_renderbuffer
*irb
)
571 unsigned int dst_x
, dst_y
;
573 /* compute offset of the particular 2D image within the texture region */
574 intel_miptree_get_image_offset(irb
->mt
,
584 * Called by glFramebufferTexture[123]DEXT() (and other places) to
585 * prepare for rendering into texture memory. This might be called
586 * many times to choose different texture levels, cube faces, etc
587 * before intel_finish_render_texture() is ever called.
590 intel_render_texture(struct gl_context
* ctx
,
591 struct gl_framebuffer
*fb
,
592 struct gl_renderbuffer_attachment
*att
)
594 struct brw_context
*brw
= brw_context(ctx
);
595 struct gl_renderbuffer
*rb
= att
->Renderbuffer
;
596 struct intel_renderbuffer
*irb
= intel_renderbuffer(rb
);
597 struct gl_texture_image
*image
= rb
->TexImage
;
598 struct intel_texture_image
*intel_image
= intel_texture_image(image
);
599 struct intel_mipmap_tree
*mt
= intel_image
->mt
;
604 if (att
->CubeMapFace
> 0) {
605 assert(att
->Zoffset
== 0);
606 layer
= att
->CubeMapFace
;
608 layer
= att
->Zoffset
;
611 if (!intel_image
->mt
) {
612 /* Fallback on drawing to a texture that doesn't have a miptree
613 * (has a border, width/height 0, etc.)
615 _swrast_render_texture(ctx
, fb
, att
);
619 intel_miptree_check_level_layer(mt
, att
->TextureLevel
, layer
);
621 if (!intel_renderbuffer_update_wrapper(brw
, irb
, image
, layer
, att
->Layered
)) {
622 _swrast_render_texture(ctx
, fb
, att
);
626 DBG("Begin render %s texture tex=%u w=%d h=%d d=%d refcount=%d\n",
627 _mesa_get_format_name(image
->TexFormat
),
628 att
->Texture
->Name
, image
->Width
, image
->Height
, image
->Depth
,
633 #define fbo_incomplete(fb, ...) do { \
634 static GLuint msg_id = 0; \
635 if (unlikely(ctx->Const.ContextFlags & GL_CONTEXT_FLAG_DEBUG_BIT)) { \
636 _mesa_gl_debug(ctx, &msg_id, \
637 MESA_DEBUG_SOURCE_API, \
638 MESA_DEBUG_TYPE_OTHER, \
639 MESA_DEBUG_SEVERITY_MEDIUM, \
643 fb->_Status = GL_FRAMEBUFFER_UNSUPPORTED; \
647 * Do additional "completeness" testing of a framebuffer object.
650 intel_validate_framebuffer(struct gl_context
*ctx
, struct gl_framebuffer
*fb
)
652 struct brw_context
*brw
= brw_context(ctx
);
653 struct intel_renderbuffer
*depthRb
=
654 intel_get_renderbuffer(fb
, BUFFER_DEPTH
);
655 struct intel_renderbuffer
*stencilRb
=
656 intel_get_renderbuffer(fb
, BUFFER_STENCIL
);
657 struct intel_mipmap_tree
*depth_mt
= NULL
, *stencil_mt
= NULL
;
660 DBG("%s() on fb %p (%s)\n", __func__
,
661 fb
, (fb
== ctx
->DrawBuffer
? "drawbuffer" :
662 (fb
== ctx
->ReadBuffer
? "readbuffer" : "other buffer")));
665 depth_mt
= depthRb
->mt
;
667 stencil_mt
= stencilRb
->mt
;
668 if (stencil_mt
->stencil_mt
)
669 stencil_mt
= stencil_mt
->stencil_mt
;
672 if (depth_mt
&& stencil_mt
) {
674 unsigned d_width
, d_height
, d_depth
;
675 unsigned s_width
, s_height
, s_depth
;
677 if (depth_mt
->surf
.size
> 0) {
678 d_width
= depth_mt
->surf
.phys_level0_sa
.width
;
679 d_height
= depth_mt
->surf
.phys_level0_sa
.height
;
680 d_depth
= depth_mt
->surf
.dim
== ISL_SURF_DIM_3D
?
681 depth_mt
->surf
.phys_level0_sa
.depth
:
682 depth_mt
->surf
.phys_level0_sa
.array_len
;
684 d_width
= depth_mt
->physical_width0
;
685 d_height
= depth_mt
->physical_height0
;
686 d_depth
= depth_mt
->physical_depth0
;
689 if (stencil_mt
->surf
.size
> 0) {
690 s_width
= stencil_mt
->surf
.phys_level0_sa
.width
;
691 s_height
= stencil_mt
->surf
.phys_level0_sa
.height
;
692 s_depth
= stencil_mt
->surf
.dim
== ISL_SURF_DIM_3D
?
693 stencil_mt
->surf
.phys_level0_sa
.depth
:
694 stencil_mt
->surf
.phys_level0_sa
.array_len
;
696 s_width
= stencil_mt
->physical_width0
;
697 s_height
= stencil_mt
->physical_height0
;
698 s_depth
= stencil_mt
->physical_depth0
;
701 /* For gen >= 6, we are using the lod/minimum-array-element fields
702 * and supporting layered rendering. This means that we must restrict
703 * the depth & stencil attachments to match in various more retrictive
704 * ways. (width, height, depth, LOD and layer)
706 if (d_width
!= s_width
||
707 d_height
!= s_height
||
708 d_depth
!= s_depth
||
709 depthRb
->mt_level
!= stencilRb
->mt_level
||
710 depthRb
->mt_layer
!= stencilRb
->mt_layer
) {
712 "FBO incomplete: depth and stencil must match in"
713 "width, height, depth, LOD and layer\n");
716 if (depth_mt
== stencil_mt
) {
717 /* For true packed depth/stencil (not faked on prefers-separate-stencil
718 * hardware) we need to be sure they're the same level/layer, since
719 * we'll be emitting a single packet describing the packed setup.
721 if (depthRb
->mt_level
!= stencilRb
->mt_level
||
722 depthRb
->mt_layer
!= stencilRb
->mt_layer
) {
724 "FBO incomplete: depth image level/layer %d/%d != "
725 "stencil image %d/%d\n",
729 stencilRb
->mt_layer
);
732 if (!brw
->has_separate_stencil
) {
733 fbo_incomplete(fb
, "FBO incomplete: separate stencil "
736 if (stencil_mt
->format
!= MESA_FORMAT_S_UINT8
) {
737 fbo_incomplete(fb
, "FBO incomplete: separate stencil is %s "
739 _mesa_get_format_name(stencil_mt
->format
));
741 if (brw
->gen
< 7 && !intel_renderbuffer_has_hiz(depthRb
)) {
742 /* Before Gen7, separate depth and stencil buffers can be used
743 * only if HiZ is enabled. From the Sandybridge PRM, Volume 2,
744 * Part 1, Bit 3DSTATE_DEPTH_BUFFER.SeparateStencilBufferEnable:
745 * [DevSNB]: This field must be set to the same value (enabled
746 * or disabled) as Hierarchical Depth Buffer Enable.
748 fbo_incomplete(fb
, "FBO incomplete: separate stencil "
754 for (i
= 0; i
< ARRAY_SIZE(fb
->Attachment
); i
++) {
755 struct gl_renderbuffer
*rb
;
756 struct intel_renderbuffer
*irb
;
758 if (fb
->Attachment
[i
].Type
== GL_NONE
)
761 /* A supported attachment will have a Renderbuffer set either
762 * from being a Renderbuffer or being a texture that got the
763 * intel_wrap_texture() treatment.
765 rb
= fb
->Attachment
[i
].Renderbuffer
;
767 fbo_incomplete(fb
, "FBO incomplete: attachment without "
772 if (fb
->Attachment
[i
].Type
== GL_TEXTURE
) {
773 if (rb
->TexImage
->Border
) {
774 fbo_incomplete(fb
, "FBO incomplete: texture with border\n");
779 irb
= intel_renderbuffer(rb
);
781 fbo_incomplete(fb
, "FBO incomplete: software rendering "
786 if (!brw_render_target_supported(brw
, rb
)) {
787 fbo_incomplete(fb
, "FBO incomplete: Unsupported HW "
788 "texture/renderbuffer format attached: %s\n",
789 _mesa_get_format_name(intel_rb_format(irb
)));
795 * Try to do a glBlitFramebuffer using glCopyTexSubImage2D
796 * We can do this when the dst renderbuffer is actually a texture and
797 * there is no scaling, mirroring or scissoring.
799 * \return new buffer mask indicating the buffers left to blit using the
803 intel_blit_framebuffer_with_blitter(struct gl_context
*ctx
,
804 const struct gl_framebuffer
*readFb
,
805 const struct gl_framebuffer
*drawFb
,
806 GLint srcX0
, GLint srcY0
,
807 GLint srcX1
, GLint srcY1
,
808 GLint dstX0
, GLint dstY0
,
809 GLint dstX1
, GLint dstY1
,
812 struct brw_context
*brw
= brw_context(ctx
);
814 /* Sync up the state of window system buffers. We need to do this before
815 * we go looking for the buffers.
817 intel_prepare_render(brw
);
819 if (mask
& GL_COLOR_BUFFER_BIT
) {
821 struct gl_renderbuffer
*src_rb
= readFb
->_ColorReadBuffer
;
822 struct intel_renderbuffer
*src_irb
= intel_renderbuffer(src_rb
);
825 perf_debug("glBlitFramebuffer(): missing src renderbuffer. "
826 "Falling back to software rendering.\n");
830 /* If the source and destination are the same size with no mirroring,
831 * the rectangles are within the size of the texture and there is no
832 * scissor, then we can probably use the blit engine.
834 if (!(srcX0
- srcX1
== dstX0
- dstX1
&&
835 srcY0
- srcY1
== dstY0
- dstY1
&&
838 srcX0
>= 0 && srcX1
<= readFb
->Width
&&
839 srcY0
>= 0 && srcY1
<= readFb
->Height
&&
840 dstX0
>= 0 && dstX1
<= drawFb
->Width
&&
841 dstY0
>= 0 && dstY1
<= drawFb
->Height
&&
842 !(ctx
->Scissor
.EnableFlags
))) {
843 perf_debug("glBlitFramebuffer(): non-1:1 blit. "
844 "Falling back to software rendering.\n");
848 /* Blit to all active draw buffers. We don't do any pre-checking,
849 * because we assume that copying to MRTs is rare, and failure midway
850 * through copying is even more rare. Even if it was to occur, it's
851 * safe to let meta start the copy over from scratch, because
852 * glBlitFramebuffer completely overwrites the destination pixels, and
853 * results are undefined if any destination pixels have a dependency on
856 for (i
= 0; i
< drawFb
->_NumColorDrawBuffers
; i
++) {
857 struct gl_renderbuffer
*dst_rb
= drawFb
->_ColorDrawBuffers
[i
];
858 struct intel_renderbuffer
*dst_irb
= intel_renderbuffer(dst_rb
);
861 perf_debug("glBlitFramebuffer(): missing dst renderbuffer. "
862 "Falling back to software rendering.\n");
866 if (ctx
->Color
.sRGBEnabled
&&
867 _mesa_get_format_color_encoding(src_irb
->mt
->format
) !=
868 _mesa_get_format_color_encoding(dst_irb
->mt
->format
)) {
869 perf_debug("glBlitFramebuffer() with sRGB conversion cannot be "
870 "handled by BLT path.\n");
874 if (!intel_miptree_blit(brw
,
876 src_irb
->mt_level
, src_irb
->mt_layer
,
877 srcX0
, srcY0
, src_rb
->Name
== 0,
879 dst_irb
->mt_level
, dst_irb
->mt_layer
,
880 dstX0
, dstY0
, dst_rb
->Name
== 0,
881 dstX1
- dstX0
, dstY1
- dstY0
, GL_COPY
)) {
882 perf_debug("glBlitFramebuffer(): unknown blit failure. "
883 "Falling back to software rendering.\n");
888 mask
&= ~GL_COLOR_BUFFER_BIT
;
895 intel_blit_framebuffer(struct gl_context
*ctx
,
896 struct gl_framebuffer
*readFb
,
897 struct gl_framebuffer
*drawFb
,
898 GLint srcX0
, GLint srcY0
, GLint srcX1
, GLint srcY1
,
899 GLint dstX0
, GLint dstY0
, GLint dstX1
, GLint dstY1
,
900 GLbitfield mask
, GLenum filter
)
902 struct brw_context
*brw
= brw_context(ctx
);
904 /* Page 679 of OpenGL 4.4 spec says:
905 * "Added BlitFramebuffer to commands affected by conditional rendering in
906 * section 10.10 (Bug 9562)."
908 if (!_mesa_check_conditional_render(ctx
))
912 /* On gen4-5, try BLT first.
914 * Gen4-5 have a single ring for both 3D and BLT operations, so there's
915 * no inter-ring synchronization issues like on Gen6+. It is apparently
916 * faster than using the 3D pipeline. Original Gen4 also has to rebase
917 * and copy miptree slices in order to render to unaligned locations.
919 mask
= intel_blit_framebuffer_with_blitter(ctx
, readFb
, drawFb
,
920 srcX0
, srcY0
, srcX1
, srcY1
,
921 dstX0
, dstY0
, dstX1
, dstY1
,
927 mask
= brw_blorp_framebuffer(brw
, readFb
, drawFb
,
928 srcX0
, srcY0
, srcX1
, srcY1
,
929 dstX0
, dstY0
, dstX1
, dstY1
,
934 mask
= _mesa_meta_BlitFramebuffer(ctx
, readFb
, drawFb
,
935 srcX0
, srcY0
, srcX1
, srcY1
,
936 dstX0
, dstY0
, dstX1
, dstY1
,
941 if (brw
->gen
>= 8 && (mask
& GL_STENCIL_BUFFER_BIT
)) {
942 assert(!"Invalid blit");
945 /* Try using the BLT engine. */
946 mask
= intel_blit_framebuffer_with_blitter(ctx
, readFb
, drawFb
,
947 srcX0
, srcY0
, srcX1
, srcY1
,
948 dstX0
, dstY0
, dstX1
, dstY1
,
953 _swrast_BlitFramebuffer(ctx
, readFb
, drawFb
,
954 srcX0
, srcY0
, srcX1
, srcY1
,
955 dstX0
, dstY0
, dstX1
, dstY1
,
960 * Does the renderbuffer have hiz enabled?
963 intel_renderbuffer_has_hiz(struct intel_renderbuffer
*irb
)
965 return intel_miptree_level_has_hiz(irb
->mt
, irb
->mt_level
);
969 intel_renderbuffer_move_to_temp(struct brw_context
*brw
,
970 struct intel_renderbuffer
*irb
,
973 struct gl_renderbuffer
*rb
=&irb
->Base
.Base
;
974 struct intel_texture_image
*intel_image
= intel_texture_image(rb
->TexImage
);
975 struct intel_mipmap_tree
*new_mt
;
976 int width
, height
, depth
;
978 uint32_t layout_flags
= MIPTREE_LAYOUT_ACCELERATED_UPLOAD
|
979 MIPTREE_LAYOUT_TILING_ANY
;
981 intel_get_image_dims(rb
->TexImage
, &width
, &height
, &depth
);
983 assert(irb
->align_wa_mt
== NULL
);
984 new_mt
= intel_miptree_create(brw
, GL_TEXTURE_2D
,
985 intel_image
->base
.Base
.TexFormat
,
988 irb
->mt
->num_samples
,
992 intel_miptree_copy_slice(brw
, intel_image
->mt
,
993 intel_image
->base
.Base
.Level
, irb
->mt_layer
,
996 intel_miptree_reference(&irb
->align_wa_mt
, new_mt
);
997 intel_miptree_release(&new_mt
);
1004 brw_render_cache_set_clear(struct brw_context
*brw
)
1006 struct set_entry
*entry
;
1008 set_foreach(brw
->render_cache
, entry
) {
1009 _mesa_set_remove(brw
->render_cache
, entry
);
1014 brw_render_cache_set_add_bo(struct brw_context
*brw
, struct brw_bo
*bo
)
1016 _mesa_set_add(brw
->render_cache
, bo
);
1020 * Emits an appropriate flush for a BO if it has been rendered to within the
1021 * same batchbuffer as a read that's about to be emitted.
1023 * The GPU has separate, incoherent caches for the render cache and the
1024 * sampler cache, along with other caches. Usually data in the different
1025 * caches don't interact (e.g. we don't render to our driver-generated
1026 * immediate constant data), but for render-to-texture in FBOs we definitely
1027 * do. When a batchbuffer is flushed, the kernel will ensure that everything
1028 * necessary is flushed before another use of that BO, but for reuse from
1029 * different caches within a batchbuffer, it's all our responsibility.
1032 brw_render_cache_set_check_flush(struct brw_context
*brw
, struct brw_bo
*bo
)
1034 if (!_mesa_set_search(brw
->render_cache
, bo
))
1037 if (brw
->gen
>= 6) {
1038 brw_emit_pipe_control_flush(brw
,
1039 PIPE_CONTROL_DEPTH_CACHE_FLUSH
|
1040 PIPE_CONTROL_RENDER_TARGET_FLUSH
|
1041 PIPE_CONTROL_CS_STALL
);
1043 brw_emit_pipe_control_flush(brw
,
1044 PIPE_CONTROL_TEXTURE_CACHE_INVALIDATE
|
1045 PIPE_CONTROL_CONST_CACHE_INVALIDATE
);
1047 brw_emit_mi_flush(brw
);
1050 brw_render_cache_set_clear(brw
);
1054 * Do one-time context initializations related to GL_EXT_framebuffer_object.
1055 * Hook in device driver functions.
1058 intel_fbo_init(struct brw_context
*brw
)
1060 struct dd_function_table
*dd
= &brw
->ctx
.Driver
;
1061 dd
->NewRenderbuffer
= intel_new_renderbuffer
;
1062 dd
->MapRenderbuffer
= intel_map_renderbuffer
;
1063 dd
->UnmapRenderbuffer
= intel_unmap_renderbuffer
;
1064 dd
->RenderTexture
= intel_render_texture
;
1065 dd
->ValidateFramebuffer
= intel_validate_framebuffer
;
1066 dd
->BlitFramebuffer
= intel_blit_framebuffer
;
1067 dd
->EGLImageTargetRenderbufferStorage
=
1068 intel_image_target_renderbuffer_storage
;
1070 brw
->render_cache
= _mesa_set_create(brw
, _mesa_hash_pointer
,
1071 _mesa_key_pointer_equal
);