1 /**************************************************************************
3 * Copyright 2006 VMware, Inc.
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the
8 * "Software"), to deal in the Software without restriction, including
9 * without limitation the rights to use, copy, modify, merge, publish,
10 * distribute, sub license, and/or sell copies of the Software, and to
11 * permit persons to whom the Software is furnished to do so, subject to
12 * the following conditions:
14 * The above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial portions
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
21 * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26 **************************************************************************/
29 #include "main/enums.h"
30 #include "main/imports.h"
31 #include "main/macros.h"
32 #include "main/mtypes.h"
33 #include "main/fbobject.h"
34 #include "main/framebuffer.h"
35 #include "main/renderbuffer.h"
36 #include "main/context.h"
37 #include "main/teximage.h"
38 #include "main/image.h"
39 #include "main/condrender.h"
40 #include "util/hash_table.h"
43 #include "swrast/swrast.h"
44 #include "drivers/common/meta.h"
46 #include "intel_batchbuffer.h"
47 #include "intel_buffers.h"
48 #include "intel_blit.h"
49 #include "intel_fbo.h"
50 #include "intel_mipmap_tree.h"
51 #include "intel_image.h"
52 #include "intel_screen.h"
53 #include "intel_tex.h"
54 #include "brw_context.h"
56 #define FILE_DEBUG_FLAG DEBUG_FBO
59 * Create a new framebuffer object.
61 static struct gl_framebuffer
*
62 intel_new_framebuffer(struct gl_context
* ctx
, GLuint name
)
64 /* Only drawable state in intel_framebuffer at this time, just use Mesa's
67 return _mesa_new_framebuffer(ctx
, name
);
71 /** Called by gl_renderbuffer::Delete() */
73 intel_delete_renderbuffer(struct gl_context
*ctx
, struct gl_renderbuffer
*rb
)
75 struct intel_renderbuffer
*irb
= intel_renderbuffer(rb
);
79 intel_miptree_release(&irb
->mt
);
80 intel_miptree_release(&irb
->singlesample_mt
);
82 _mesa_delete_renderbuffer(ctx
, rb
);
86 * \brief Downsample a winsys renderbuffer from mt to singlesample_mt.
88 * If the miptree needs no downsample, then skip.
91 intel_renderbuffer_downsample(struct brw_context
*brw
,
92 struct intel_renderbuffer
*irb
)
94 if (!irb
->need_downsample
)
96 intel_miptree_updownsample(brw
, irb
->mt
, irb
->singlesample_mt
);
97 irb
->need_downsample
= false;
101 * \brief Upsample a winsys renderbuffer from singlesample_mt to mt.
103 * The upsample is done unconditionally.
106 intel_renderbuffer_upsample(struct brw_context
*brw
,
107 struct intel_renderbuffer
*irb
)
109 assert(!irb
->need_downsample
);
111 intel_miptree_updownsample(brw
, irb
->singlesample_mt
, irb
->mt
);
115 * \see dd_function_table::MapRenderbuffer
118 intel_map_renderbuffer(struct gl_context
*ctx
,
119 struct gl_renderbuffer
*rb
,
120 GLuint x
, GLuint y
, GLuint w
, GLuint h
,
125 struct brw_context
*brw
= brw_context(ctx
);
126 struct swrast_renderbuffer
*srb
= (struct swrast_renderbuffer
*)rb
;
127 struct intel_renderbuffer
*irb
= intel_renderbuffer(rb
);
128 struct intel_mipmap_tree
*mt
;
133 /* this is a malloc'd renderbuffer (accum buffer), not an irb */
134 GLint bpp
= _mesa_get_format_bytes(rb
->Format
);
135 GLint rowStride
= srb
->RowStride
;
136 *out_map
= (GLubyte
*) srb
->Buffer
+ y
* rowStride
+ x
* bpp
;
137 *out_stride
= rowStride
;
141 intel_prepare_render(brw
);
143 /* The MapRenderbuffer API should always return a single-sampled mapping.
144 * The case we are asked to map multisampled RBs is in glReadPixels() (or
145 * swrast paths like glCopyTexImage()) from a window-system MSAA buffer,
146 * and GL expects an automatic resolve to happen.
148 * If it's a color miptree, there is a ->singlesample_mt which wraps the
149 * actual window system renderbuffer (which we may resolve to at any time),
150 * while the miptree itself is our driver-private allocation. If it's a
151 * depth or stencil miptree, we have a private MSAA buffer and no shared
152 * singlesample buffer, and since we don't expect anybody to ever actually
153 * resolve it, we just make a temporary singlesample buffer now when we
156 if (rb
->NumSamples
> 1) {
157 if (!irb
->singlesample_mt
) {
158 irb
->singlesample_mt
=
159 intel_miptree_create_for_renderbuffer(brw
, irb
->mt
->format
,
160 rb
->Width
, rb
->Height
,
162 if (!irb
->singlesample_mt
)
164 irb
->singlesample_mt_is_tmp
= true;
165 irb
->need_downsample
= true;
168 intel_renderbuffer_downsample(brw
, irb
);
169 mt
= irb
->singlesample_mt
;
171 irb
->need_map_upsample
= mode
& GL_MAP_WRITE_BIT
;
176 /* For a window-system renderbuffer, we need to flip the mapping we receive
177 * upside-down. So we need to ask for a rectangle on flipped vertically, and
178 * we then return a pointer to the bottom of it with a negative stride.
181 y
= rb
->Height
- y
- h
;
184 intel_miptree_map(brw
, mt
, irb
->mt_level
, irb
->mt_layer
,
185 x
, y
, w
, h
, mode
, &map
, &stride
);
188 map
+= (h
- 1) * stride
;
192 DBG("%s: rb %d (%s) mt mapped: (%d, %d) (%dx%d) -> %p/%"PRIdPTR
"\n",
193 __func__
, rb
->Name
, _mesa_get_format_name(rb
->Format
),
194 x
, y
, w
, h
, map
, stride
);
197 *out_stride
= stride
;
206 * \see dd_function_table::UnmapRenderbuffer
209 intel_unmap_renderbuffer(struct gl_context
*ctx
,
210 struct gl_renderbuffer
*rb
)
212 struct brw_context
*brw
= brw_context(ctx
);
213 struct swrast_renderbuffer
*srb
= (struct swrast_renderbuffer
*)rb
;
214 struct intel_renderbuffer
*irb
= intel_renderbuffer(rb
);
215 struct intel_mipmap_tree
*mt
;
217 DBG("%s: rb %d (%s)\n", __func__
,
218 rb
->Name
, _mesa_get_format_name(rb
->Format
));
221 /* this is a malloc'd renderbuffer (accum buffer) */
226 if (rb
->NumSamples
> 1) {
227 mt
= irb
->singlesample_mt
;
232 intel_miptree_unmap(brw
, mt
, irb
->mt_level
, irb
->mt_layer
);
234 if (irb
->need_map_upsample
) {
235 intel_renderbuffer_upsample(brw
, irb
);
236 irb
->need_map_upsample
= false;
239 if (irb
->singlesample_mt_is_tmp
)
240 intel_miptree_release(&irb
->singlesample_mt
);
245 * Round up the requested multisample count to the next supported sample size.
248 intel_quantize_num_samples(struct intel_screen
*intel
, unsigned num_samples
)
250 const int *msaa_modes
= intel_supported_msaa_modes(intel
);
251 int quantized_samples
= 0;
253 for (int i
= 0; msaa_modes
[i
] != -1; ++i
) {
254 if (msaa_modes
[i
] >= num_samples
)
255 quantized_samples
= msaa_modes
[i
];
260 return quantized_samples
;
264 intel_renderbuffer_format(struct gl_context
* ctx
, GLenum internalFormat
)
266 struct brw_context
*brw
= brw_context(ctx
);
268 switch (internalFormat
) {
270 /* Use the same format-choice logic as for textures.
271 * Renderbuffers aren't any different from textures for us,
272 * except they're less useful because you can't texture with
275 return ctx
->Driver
.ChooseTextureFormat(ctx
, GL_TEXTURE_2D
,
279 case GL_STENCIL_INDEX
:
280 case GL_STENCIL_INDEX1_EXT
:
281 case GL_STENCIL_INDEX4_EXT
:
282 case GL_STENCIL_INDEX8_EXT
:
283 case GL_STENCIL_INDEX16_EXT
:
284 /* These aren't actual texture formats, so force them here. */
285 if (brw
->has_separate_stencil
) {
286 return MESA_FORMAT_S_UINT8
;
288 assert(!brw
->must_use_separate_stencil
);
289 return MESA_FORMAT_Z24_UNORM_S8_UINT
;
295 intel_alloc_private_renderbuffer_storage(struct gl_context
* ctx
, struct gl_renderbuffer
*rb
,
296 GLenum internalFormat
,
297 GLuint width
, GLuint height
)
299 struct brw_context
*brw
= brw_context(ctx
);
300 struct intel_screen
*screen
= brw
->intelScreen
;
301 struct intel_renderbuffer
*irb
= intel_renderbuffer(rb
);
303 assert(rb
->Format
!= MESA_FORMAT_NONE
);
305 rb
->NumSamples
= intel_quantize_num_samples(screen
, rb
->NumSamples
);
308 rb
->_BaseFormat
= _mesa_base_fbo_format(ctx
, internalFormat
);
310 intel_miptree_release(&irb
->mt
);
312 DBG("%s: %s: %s (%dx%d)\n", __func__
,
313 _mesa_enum_to_string(internalFormat
),
314 _mesa_get_format_name(rb
->Format
), width
, height
);
316 if (width
== 0 || height
== 0)
319 irb
->mt
= intel_miptree_create_for_renderbuffer(brw
, rb
->Format
,
325 irb
->layer_count
= 1;
331 * Called via glRenderbufferStorageEXT() to set the format and allocate
332 * storage for a user-created renderbuffer.
335 intel_alloc_renderbuffer_storage(struct gl_context
* ctx
, struct gl_renderbuffer
*rb
,
336 GLenum internalFormat
,
337 GLuint width
, GLuint height
)
339 rb
->Format
= intel_renderbuffer_format(ctx
, internalFormat
);
340 return intel_alloc_private_renderbuffer_storage(ctx
, rb
, internalFormat
, width
, height
);
344 intel_image_target_renderbuffer_storage(struct gl_context
*ctx
,
345 struct gl_renderbuffer
*rb
,
348 struct brw_context
*brw
= brw_context(ctx
);
349 struct intel_renderbuffer
*irb
;
353 screen
= brw
->intelScreen
->driScrnPriv
;
354 image
= screen
->dri2
.image
->lookupEGLImage(screen
, image_handle
,
355 screen
->loaderPrivate
);
359 if (image
->planar_format
&& image
->planar_format
->nplanes
> 1) {
360 _mesa_error(ctx
, GL_INVALID_OPERATION
,
361 "glEGLImageTargetRenderbufferStorage(planar buffers are not "
362 "supported as render targets.");
366 /* __DRIimage is opaque to the core so it has to be checked here */
367 switch (image
->format
) {
368 case MESA_FORMAT_R8G8B8A8_UNORM
:
369 _mesa_error(ctx
, GL_INVALID_OPERATION
,
370 "glEGLImageTargetRenderbufferStorage(unsupported image format");
377 irb
= intel_renderbuffer(rb
);
378 intel_miptree_release(&irb
->mt
);
380 /* Disable creation of the miptree's aux buffers because the driver exposes
381 * no EGL API to manage them. That is, there is no API for resolving the aux
382 * buffer's content to the main buffer nor for invalidating the aux buffer's
385 irb
->mt
= intel_miptree_create_for_bo(brw
,
393 MIPTREE_LAYOUT_DISABLE_AUX
);
397 rb
->InternalFormat
= image
->internal_format
;
398 rb
->Width
= image
->width
;
399 rb
->Height
= image
->height
;
400 rb
->Format
= image
->format
;
401 rb
->_BaseFormat
= _mesa_get_format_base_format(image
->format
);
402 rb
->NeedsFinishRenderTexture
= true;
403 irb
->layer_count
= 1;
407 * Called by _mesa_resize_framebuffer() for each hardware renderbuffer when a
408 * window system framebuffer is resized.
410 * Any actual buffer reallocations for hardware renderbuffers (which would
411 * have triggered _mesa_resize_framebuffer()) were done by
412 * intel_process_dri2_buffer().
415 intel_alloc_window_storage(struct gl_context
* ctx
, struct gl_renderbuffer
*rb
,
416 GLenum internalFormat
, GLuint width
, GLuint height
)
418 assert(rb
->Name
== 0);
421 rb
->InternalFormat
= internalFormat
;
426 /** Dummy function for gl_renderbuffer::AllocStorage() */
428 intel_nop_alloc_storage(struct gl_context
* ctx
, struct gl_renderbuffer
*rb
,
429 GLenum internalFormat
, GLuint width
, GLuint height
)
431 _mesa_problem(ctx
, "intel_nop_alloc_storage should never be called.");
436 * Create a new intel_renderbuffer which corresponds to an on-screen window,
437 * not a user-created renderbuffer.
439 * \param num_samples must be quantized.
441 struct intel_renderbuffer
*
442 intel_create_renderbuffer(mesa_format format
, unsigned num_samples
)
444 struct intel_renderbuffer
*irb
;
445 struct gl_renderbuffer
*rb
;
447 GET_CURRENT_CONTEXT(ctx
);
449 irb
= CALLOC_STRUCT(intel_renderbuffer
);
451 _mesa_error(ctx
, GL_OUT_OF_MEMORY
, "creating renderbuffer");
455 rb
= &irb
->Base
.Base
;
456 irb
->layer_count
= 1;
458 _mesa_init_renderbuffer(rb
, 0);
459 rb
->ClassID
= INTEL_RB_CLASS
;
460 rb
->_BaseFormat
= _mesa_get_format_base_format(format
);
462 rb
->InternalFormat
= rb
->_BaseFormat
;
463 rb
->NumSamples
= num_samples
;
465 /* intel-specific methods */
466 rb
->Delete
= intel_delete_renderbuffer
;
467 rb
->AllocStorage
= intel_alloc_window_storage
;
473 * Private window-system buffers (as opposed to ones shared with the display
474 * server created with intel_create_renderbuffer()) are most similar in their
475 * handling to user-created renderbuffers, but they have a resize handler that
476 * may be called at intel_update_renderbuffers() time.
478 * \param num_samples must be quantized.
480 struct intel_renderbuffer
*
481 intel_create_private_renderbuffer(mesa_format format
, unsigned num_samples
)
483 struct intel_renderbuffer
*irb
;
485 irb
= intel_create_renderbuffer(format
, num_samples
);
486 irb
->Base
.Base
.AllocStorage
= intel_alloc_private_renderbuffer_storage
;
492 * Create a new renderbuffer object.
493 * Typically called via glBindRenderbufferEXT().
495 static struct gl_renderbuffer
*
496 intel_new_renderbuffer(struct gl_context
* ctx
, GLuint name
)
498 struct intel_renderbuffer
*irb
;
499 struct gl_renderbuffer
*rb
;
501 irb
= CALLOC_STRUCT(intel_renderbuffer
);
503 _mesa_error(ctx
, GL_OUT_OF_MEMORY
, "creating renderbuffer");
507 rb
= &irb
->Base
.Base
;
509 _mesa_init_renderbuffer(rb
, name
);
510 rb
->ClassID
= INTEL_RB_CLASS
;
512 /* intel-specific methods */
513 rb
->Delete
= intel_delete_renderbuffer
;
514 rb
->AllocStorage
= intel_alloc_renderbuffer_storage
;
515 /* span routines set in alloc_storage function */
521 intel_renderbuffer_update_wrapper(struct brw_context
*brw
,
522 struct intel_renderbuffer
*irb
,
523 struct gl_texture_image
*image
,
527 struct gl_renderbuffer
*rb
= &irb
->Base
.Base
;
528 struct intel_texture_image
*intel_image
= intel_texture_image(image
);
529 struct intel_mipmap_tree
*mt
= intel_image
->mt
;
530 int level
= image
->Level
;
532 rb
->AllocStorage
= intel_nop_alloc_storage
;
534 /* adjust for texture view parameters */
535 layer
+= image
->TexObject
->MinLayer
;
536 level
+= image
->TexObject
->MinLevel
;
538 intel_miptree_check_level_layer(mt
, level
, layer
);
539 irb
->mt_level
= level
;
541 int layer_multiplier
;
542 switch (mt
->msaa_layout
) {
543 case INTEL_MSAA_LAYOUT_UMS
:
544 case INTEL_MSAA_LAYOUT_CMS
:
545 layer_multiplier
= mt
->num_samples
;
549 layer_multiplier
= 1;
552 irb
->mt_layer
= layer_multiplier
* layer
;
555 irb
->layer_count
= 1;
556 } else if (image
->TexObject
->NumLayers
> 0) {
557 irb
->layer_count
= image
->TexObject
->NumLayers
;
559 irb
->layer_count
= mt
->level
[level
].depth
/ layer_multiplier
;
562 intel_miptree_reference(&irb
->mt
, mt
);
564 intel_renderbuffer_set_draw_offset(irb
);
566 if (intel_miptree_wants_hiz_buffer(brw
, mt
)) {
567 intel_miptree_alloc_hiz(brw
, mt
);
576 intel_renderbuffer_set_draw_offset(struct intel_renderbuffer
*irb
)
578 unsigned int dst_x
, dst_y
;
580 /* compute offset of the particular 2D image within the texture region */
581 intel_miptree_get_image_offset(irb
->mt
,
591 * Called by glFramebufferTexture[123]DEXT() (and other places) to
592 * prepare for rendering into texture memory. This might be called
593 * many times to choose different texture levels, cube faces, etc
594 * before intel_finish_render_texture() is ever called.
597 intel_render_texture(struct gl_context
* ctx
,
598 struct gl_framebuffer
*fb
,
599 struct gl_renderbuffer_attachment
*att
)
601 struct brw_context
*brw
= brw_context(ctx
);
602 struct gl_renderbuffer
*rb
= att
->Renderbuffer
;
603 struct intel_renderbuffer
*irb
= intel_renderbuffer(rb
);
604 struct gl_texture_image
*image
= rb
->TexImage
;
605 struct intel_texture_image
*intel_image
= intel_texture_image(image
);
606 struct intel_mipmap_tree
*mt
= intel_image
->mt
;
611 if (att
->CubeMapFace
> 0) {
612 assert(att
->Zoffset
== 0);
613 layer
= att
->CubeMapFace
;
615 layer
= att
->Zoffset
;
618 if (!intel_image
->mt
) {
619 /* Fallback on drawing to a texture that doesn't have a miptree
620 * (has a border, width/height 0, etc.)
622 _swrast_render_texture(ctx
, fb
, att
);
626 intel_miptree_check_level_layer(mt
, att
->TextureLevel
, layer
);
628 if (!intel_renderbuffer_update_wrapper(brw
, irb
, image
, layer
, att
->Layered
)) {
629 _swrast_render_texture(ctx
, fb
, att
);
633 DBG("Begin render %s texture tex=%u w=%d h=%d d=%d refcount=%d\n",
634 _mesa_get_format_name(image
->TexFormat
),
635 att
->Texture
->Name
, image
->Width
, image
->Height
, image
->Depth
,
640 #define fbo_incomplete(fb, ...) do { \
641 static GLuint msg_id = 0; \
642 if (unlikely(ctx->Const.ContextFlags & GL_CONTEXT_FLAG_DEBUG_BIT)) { \
643 _mesa_gl_debug(ctx, &msg_id, \
644 MESA_DEBUG_SOURCE_API, \
645 MESA_DEBUG_TYPE_OTHER, \
646 MESA_DEBUG_SEVERITY_MEDIUM, \
650 fb->_Status = GL_FRAMEBUFFER_UNSUPPORTED; \
654 * Do additional "completeness" testing of a framebuffer object.
657 intel_validate_framebuffer(struct gl_context
*ctx
, struct gl_framebuffer
*fb
)
659 struct brw_context
*brw
= brw_context(ctx
);
660 struct intel_renderbuffer
*depthRb
=
661 intel_get_renderbuffer(fb
, BUFFER_DEPTH
);
662 struct intel_renderbuffer
*stencilRb
=
663 intel_get_renderbuffer(fb
, BUFFER_STENCIL
);
664 struct intel_mipmap_tree
*depth_mt
= NULL
, *stencil_mt
= NULL
;
667 DBG("%s() on fb %p (%s)\n", __func__
,
668 fb
, (fb
== ctx
->DrawBuffer
? "drawbuffer" :
669 (fb
== ctx
->ReadBuffer
? "readbuffer" : "other buffer")));
672 depth_mt
= depthRb
->mt
;
674 stencil_mt
= stencilRb
->mt
;
675 if (stencil_mt
->stencil_mt
)
676 stencil_mt
= stencil_mt
->stencil_mt
;
679 if (depth_mt
&& stencil_mt
) {
681 /* For gen >= 6, we are using the lod/minimum-array-element fields
682 * and supporting layered rendering. This means that we must restrict
683 * the depth & stencil attachments to match in various more retrictive
684 * ways. (width, height, depth, LOD and layer)
686 if (depth_mt
->physical_width0
!= stencil_mt
->physical_width0
||
687 depth_mt
->physical_height0
!= stencil_mt
->physical_height0
||
688 depth_mt
->physical_depth0
!= stencil_mt
->physical_depth0
||
689 depthRb
->mt_level
!= stencilRb
->mt_level
||
690 depthRb
->mt_layer
!= stencilRb
->mt_layer
) {
692 "FBO incomplete: depth and stencil must match in"
693 "width, height, depth, LOD and layer\n");
696 if (depth_mt
== stencil_mt
) {
697 /* For true packed depth/stencil (not faked on prefers-separate-stencil
698 * hardware) we need to be sure they're the same level/layer, since
699 * we'll be emitting a single packet describing the packed setup.
701 if (depthRb
->mt_level
!= stencilRb
->mt_level
||
702 depthRb
->mt_layer
!= stencilRb
->mt_layer
) {
704 "FBO incomplete: depth image level/layer %d/%d != "
705 "stencil image %d/%d\n",
709 stencilRb
->mt_layer
);
712 if (!brw
->has_separate_stencil
) {
713 fbo_incomplete(fb
, "FBO incomplete: separate stencil "
716 if (stencil_mt
->format
!= MESA_FORMAT_S_UINT8
) {
717 fbo_incomplete(fb
, "FBO incomplete: separate stencil is %s "
719 _mesa_get_format_name(stencil_mt
->format
));
721 if (brw
->gen
< 7 && !intel_renderbuffer_has_hiz(depthRb
)) {
722 /* Before Gen7, separate depth and stencil buffers can be used
723 * only if HiZ is enabled. From the Sandybridge PRM, Volume 2,
724 * Part 1, Bit 3DSTATE_DEPTH_BUFFER.SeparateStencilBufferEnable:
725 * [DevSNB]: This field must be set to the same value (enabled
726 * or disabled) as Hierarchical Depth Buffer Enable.
728 fbo_incomplete(fb
, "FBO incomplete: separate stencil "
734 for (i
= 0; i
< ARRAY_SIZE(fb
->Attachment
); i
++) {
735 struct gl_renderbuffer
*rb
;
736 struct intel_renderbuffer
*irb
;
738 if (fb
->Attachment
[i
].Type
== GL_NONE
)
741 /* A supported attachment will have a Renderbuffer set either
742 * from being a Renderbuffer or being a texture that got the
743 * intel_wrap_texture() treatment.
745 rb
= fb
->Attachment
[i
].Renderbuffer
;
747 fbo_incomplete(fb
, "FBO incomplete: attachment without "
752 if (fb
->Attachment
[i
].Type
== GL_TEXTURE
) {
753 if (rb
->TexImage
->Border
) {
754 fbo_incomplete(fb
, "FBO incomplete: texture with border\n");
759 irb
= intel_renderbuffer(rb
);
761 fbo_incomplete(fb
, "FBO incomplete: software rendering "
766 if (!brw_render_target_supported(brw
, rb
)) {
767 fbo_incomplete(fb
, "FBO incomplete: Unsupported HW "
768 "texture/renderbuffer format attached: %s\n",
769 _mesa_get_format_name(intel_rb_format(irb
)));
775 * Try to do a glBlitFramebuffer using glCopyTexSubImage2D
776 * We can do this when the dst renderbuffer is actually a texture and
777 * there is no scaling, mirroring or scissoring.
779 * \return new buffer mask indicating the buffers left to blit using the
783 intel_blit_framebuffer_with_blitter(struct gl_context
*ctx
,
784 const struct gl_framebuffer
*readFb
,
785 const struct gl_framebuffer
*drawFb
,
786 GLint srcX0
, GLint srcY0
,
787 GLint srcX1
, GLint srcY1
,
788 GLint dstX0
, GLint dstY0
,
789 GLint dstX1
, GLint dstY1
,
790 GLbitfield mask
, GLenum filter
)
792 struct brw_context
*brw
= brw_context(ctx
);
794 /* Sync up the state of window system buffers. We need to do this before
795 * we go looking for the buffers.
797 intel_prepare_render(brw
);
799 if (mask
& GL_COLOR_BUFFER_BIT
) {
801 struct gl_renderbuffer
*src_rb
= readFb
->_ColorReadBuffer
;
802 struct intel_renderbuffer
*src_irb
= intel_renderbuffer(src_rb
);
805 perf_debug("glBlitFramebuffer(): missing src renderbuffer. "
806 "Falling back to software rendering.\n");
810 /* If the source and destination are the same size with no mirroring,
811 * the rectangles are within the size of the texture and there is no
812 * scissor, then we can probably use the blit engine.
814 if (!(srcX0
- srcX1
== dstX0
- dstX1
&&
815 srcY0
- srcY1
== dstY0
- dstY1
&&
818 srcX0
>= 0 && srcX1
<= readFb
->Width
&&
819 srcY0
>= 0 && srcY1
<= readFb
->Height
&&
820 dstX0
>= 0 && dstX1
<= drawFb
->Width
&&
821 dstY0
>= 0 && dstY1
<= drawFb
->Height
&&
822 !(ctx
->Scissor
.EnableFlags
))) {
823 perf_debug("glBlitFramebuffer(): non-1:1 blit. "
824 "Falling back to software rendering.\n");
828 /* Blit to all active draw buffers. We don't do any pre-checking,
829 * because we assume that copying to MRTs is rare, and failure midway
830 * through copying is even more rare. Even if it was to occur, it's
831 * safe to let meta start the copy over from scratch, because
832 * glBlitFramebuffer completely overwrites the destination pixels, and
833 * results are undefined if any destination pixels have a dependency on
836 for (i
= 0; i
< drawFb
->_NumColorDrawBuffers
; i
++) {
837 struct gl_renderbuffer
*dst_rb
= drawFb
->_ColorDrawBuffers
[i
];
838 struct intel_renderbuffer
*dst_irb
= intel_renderbuffer(dst_rb
);
841 perf_debug("glBlitFramebuffer(): missing dst renderbuffer. "
842 "Falling back to software rendering.\n");
846 if (!intel_miptree_blit(brw
,
848 src_irb
->mt_level
, src_irb
->mt_layer
,
849 srcX0
, srcY0
, src_rb
->Name
== 0,
851 dst_irb
->mt_level
, dst_irb
->mt_layer
,
852 dstX0
, dstY0
, dst_rb
->Name
== 0,
853 dstX1
- dstX0
, dstY1
- dstY0
, GL_COPY
)) {
854 perf_debug("glBlitFramebuffer(): unknown blit failure. "
855 "Falling back to software rendering.\n");
860 mask
&= ~GL_COLOR_BUFFER_BIT
;
867 intel_blit_framebuffer(struct gl_context
*ctx
,
868 struct gl_framebuffer
*readFb
,
869 struct gl_framebuffer
*drawFb
,
870 GLint srcX0
, GLint srcY0
, GLint srcX1
, GLint srcY1
,
871 GLint dstX0
, GLint dstY0
, GLint dstX1
, GLint dstY1
,
872 GLbitfield mask
, GLenum filter
)
874 struct brw_context
*brw
= brw_context(ctx
);
876 /* Page 679 of OpenGL 4.4 spec says:
877 * "Added BlitFramebuffer to commands affected by conditional rendering in
878 * section 10.10 (Bug 9562)."
880 if (!_mesa_check_conditional_render(ctx
))
883 mask
= brw_blorp_framebuffer(brw
, readFb
, drawFb
,
884 srcX0
, srcY0
, srcX1
, srcY1
,
885 dstX0
, dstY0
, dstX1
, dstY1
,
890 mask
= _mesa_meta_BlitFramebuffer(ctx
, readFb
, drawFb
,
891 srcX0
, srcY0
, srcX1
, srcY1
,
892 dstX0
, dstY0
, dstX1
, dstY1
,
897 if (brw
->gen
>= 8 && (mask
& GL_STENCIL_BUFFER_BIT
)) {
898 brw_meta_fbo_stencil_blit(brw_context(ctx
), readFb
, drawFb
,
899 srcX0
, srcY0
, srcX1
, srcY1
,
900 dstX0
, dstY0
, dstX1
, dstY1
);
901 mask
&= ~GL_STENCIL_BUFFER_BIT
;
906 /* Try using the BLT engine. */
907 mask
= intel_blit_framebuffer_with_blitter(ctx
, readFb
, drawFb
,
908 srcX0
, srcY0
, srcX1
, srcY1
,
909 dstX0
, dstY0
, dstX1
, dstY1
,
914 _swrast_BlitFramebuffer(ctx
, readFb
, drawFb
,
915 srcX0
, srcY0
, srcX1
, srcY1
,
916 dstX0
, dstY0
, dstX1
, dstY1
,
921 * Gen4-5 implementation of glBlitFrameBuffer().
923 * Tries BLT, Meta, then swrast.
925 * Gen4-5 have a single ring for both 3D and BLT operations, so there's no
926 * inter-ring synchronization issues like on Gen6+. It is apparently faster
927 * than using the 3D pipeline. Original Gen4 also has to rebase and copy
928 * miptree slices in order to render to unaligned locations.
931 gen4_blit_framebuffer(struct gl_context
*ctx
,
932 struct gl_framebuffer
*readFb
,
933 struct gl_framebuffer
*drawFb
,
934 GLint srcX0
, GLint srcY0
, GLint srcX1
, GLint srcY1
,
935 GLint dstX0
, GLint dstY0
, GLint dstX1
, GLint dstY1
,
936 GLbitfield mask
, GLenum filter
)
938 /* Page 679 of OpenGL 4.4 spec says:
939 * "Added BlitFramebuffer to commands affected by conditional rendering in
940 * section 10.10 (Bug 9562)."
942 if (!_mesa_check_conditional_render(ctx
))
945 mask
= intel_blit_framebuffer_with_blitter(ctx
, readFb
, drawFb
,
946 srcX0
, srcY0
, srcX1
, srcY1
,
947 dstX0
, dstY0
, dstX1
, dstY1
,
952 mask
= _mesa_meta_BlitFramebuffer(ctx
, readFb
, drawFb
,
953 srcX0
, srcY0
, srcX1
, srcY1
,
954 dstX0
, dstY0
, dstX1
, dstY1
,
959 _swrast_BlitFramebuffer(ctx
, readFb
, drawFb
,
960 srcX0
, srcY0
, srcX1
, srcY1
,
961 dstX0
, dstY0
, dstX1
, dstY1
,
966 * Does the renderbuffer have hiz enabled?
969 intel_renderbuffer_has_hiz(struct intel_renderbuffer
*irb
)
971 return intel_miptree_level_has_hiz(irb
->mt
, irb
->mt_level
);
975 intel_renderbuffer_resolve_hiz(struct brw_context
*brw
,
976 struct intel_renderbuffer
*irb
)
979 return intel_miptree_slice_resolve_hiz(brw
,
988 intel_renderbuffer_att_set_needs_depth_resolve(struct gl_renderbuffer_attachment
*att
)
990 struct intel_renderbuffer
*irb
= intel_renderbuffer(att
->Renderbuffer
);
993 intel_miptree_set_all_slices_need_depth_resolve(irb
->mt
, irb
->mt_level
);
995 intel_miptree_slice_set_needs_depth_resolve(irb
->mt
,
1003 intel_renderbuffer_resolve_depth(struct brw_context
*brw
,
1004 struct intel_renderbuffer
*irb
)
1007 return intel_miptree_slice_resolve_depth(brw
,
1016 intel_renderbuffer_move_to_temp(struct brw_context
*brw
,
1017 struct intel_renderbuffer
*irb
,
1020 struct gl_renderbuffer
*rb
=&irb
->Base
.Base
;
1021 struct intel_texture_image
*intel_image
= intel_texture_image(rb
->TexImage
);
1022 struct intel_mipmap_tree
*new_mt
;
1023 int width
, height
, depth
;
1025 uint32_t layout_flags
= MIPTREE_LAYOUT_ACCELERATED_UPLOAD
|
1026 MIPTREE_LAYOUT_ALLOC_ANY_TILED
;
1028 intel_miptree_get_dimensions_for_image(rb
->TexImage
, &width
, &height
, &depth
);
1030 new_mt
= intel_miptree_create(brw
, rb
->TexImage
->TexObject
->Target
,
1031 intel_image
->base
.Base
.TexFormat
,
1032 intel_image
->base
.Base
.Level
,
1033 intel_image
->base
.Base
.Level
,
1034 width
, height
, depth
,
1035 irb
->mt
->num_samples
,
1038 if (intel_miptree_wants_hiz_buffer(brw
, new_mt
)) {
1039 intel_miptree_alloc_hiz(brw
, new_mt
);
1042 intel_miptree_copy_teximage(brw
, intel_image
, new_mt
, invalidate
);
1044 intel_miptree_reference(&irb
->mt
, intel_image
->mt
);
1045 intel_renderbuffer_set_draw_offset(irb
);
1046 intel_miptree_release(&new_mt
);
1050 brw_render_cache_set_clear(struct brw_context
*brw
)
1052 struct set_entry
*entry
;
1054 set_foreach(brw
->render_cache
, entry
) {
1055 _mesa_set_remove(brw
->render_cache
, entry
);
1060 brw_render_cache_set_add_bo(struct brw_context
*brw
, drm_intel_bo
*bo
)
1062 _mesa_set_add(brw
->render_cache
, bo
);
1066 * Emits an appropriate flush for a BO if it has been rendered to within the
1067 * same batchbuffer as a read that's about to be emitted.
1069 * The GPU has separate, incoherent caches for the render cache and the
1070 * sampler cache, along with other caches. Usually data in the different
1071 * caches don't interact (e.g. we don't render to our driver-generated
1072 * immediate constant data), but for render-to-texture in FBOs we definitely
1073 * do. When a batchbuffer is flushed, the kernel will ensure that everything
1074 * necessary is flushed before another use of that BO, but for reuse from
1075 * different caches within a batchbuffer, it's all our responsibility.
1078 brw_render_cache_set_check_flush(struct brw_context
*brw
, drm_intel_bo
*bo
)
1080 if (!_mesa_set_search(brw
->render_cache
, bo
))
1083 brw_emit_mi_flush(brw
);
1087 * Do one-time context initializations related to GL_EXT_framebuffer_object.
1088 * Hook in device driver functions.
1091 intel_fbo_init(struct brw_context
*brw
)
1093 struct dd_function_table
*dd
= &brw
->ctx
.Driver
;
1094 dd
->NewFramebuffer
= intel_new_framebuffer
;
1095 dd
->NewRenderbuffer
= intel_new_renderbuffer
;
1096 dd
->MapRenderbuffer
= intel_map_renderbuffer
;
1097 dd
->UnmapRenderbuffer
= intel_unmap_renderbuffer
;
1098 dd
->RenderTexture
= intel_render_texture
;
1099 dd
->ValidateFramebuffer
= intel_validate_framebuffer
;
1101 dd
->BlitFramebuffer
= intel_blit_framebuffer
;
1103 dd
->BlitFramebuffer
= gen4_blit_framebuffer
;
1104 dd
->EGLImageTargetRenderbufferStorage
=
1105 intel_image_target_renderbuffer_storage
;
1107 brw
->render_cache
= _mesa_set_create(brw
, _mesa_hash_pointer
,
1108 _mesa_key_pointer_equal
);