blob: f7bae4d8310ee6fd378dbbb1283e46e85a133220 [file] [log] [blame]
Name
KHR_lock_surface2
Name Strings
EGL_KHR_lock_surface2
Contributors
Mark Callow
Gary King
Jon Leech
Marko Lukat
Alon Or-bach
Tim Renouf
Contacts
Jon Leech (jon 'at' alumni.caltech.edu)
Notice
Copyright (c) 2006-2013 The Khronos Group Inc. Copyright terms at
http://www.khronos.org/registry/speccopyright.html
Status
This extension is obsolete and has been replaced by
EGL_KHR_lock_surface3. Khronos recommends implementers who support this
extension also implement lock_surface3, and begin transitioning
developers to using that extension. See issue 21 for the reason.
Complete. Version 2 approved by the Khronos Board of Promoters on
May 28, 2010.
Implemented by Antix Labs.
Version
Version 3, December 4, 2013
Number
EGL Extension #16
Dependencies
Requires EGL 1.0 and EGL_KHR_lock_surface version 18.
This extension is written against the wording of the EGL 1.3
and EGL 1.4 Specifications. Unless otherwise specified, each change
applies to both specifications. Unless otherwise specified, a page
number refers to the EGL 1.3 specification.
This extension is written against the wording of EGL_KHR_lock_surface
version 18.
Overview
This extension slightly modifies and clarifies some semantic aspects
of the EGL_KHR_lock_surface extension, in a way that is backwards
compatible for applications.
The extension is presented here as the full text of the
EGL_KHR_lock_surface extension (minus the Status, Version, Number and
Dependencies sections at the start) as modified by the changes made by
this EGL_KHR_lock_surface2 extension. A diff utility can be used between
EGL_KHR_lock_surface version 18 and this EGL_KHR_lock_surface2 extension
to show the exact changes.
An application which needs to tell whether the implementation supports
EGL_KHR_lock_surface2, or just the original EGL_KHR_lock_surface, can
use eglQueryString with EGL_EXTENSIONS to query the list of
implemented extensions.
The changes over EGL_KHR_lock_surface can be summarized as follows:
* EGL_KHR_lock_surface had the EGL_MAP_PRESERVE_PIXELS_KHR attribute on
eglLockSurfaceKHR, but failed to point out how the surface attribute
EGL_SWAP_BEHAVIOR would interact with lock surface rendering.
EGL_KHR_lock_surface2 specifies that the locked buffer contains the
back buffer pixels if EGL_SWAP_BEHAVIOR is EGL_BUFFER_PRESERVED
_or_ if EGL_MAP_PRESERVE_PIXELS_KHR is EGL_TRUE, and provides a way to
set EGL_SWAP_BEHAVIOR on creation of a lockable window surface,
even if EGL_SWAP_BEHAVIOR is not otherwise modifiable.
EGL_SWAP_BEHAVIOR now defaults to EGL_BUFFER_PRESERVED for a
lockable surface.
* EGL_KHR_lock_surface failed to specify its interaction with the
EGL requirement that a context be current at eglSwapBuffers; no
context is used for lock surface rendering. EGL_KHR_lock_surface2
relaxes that requirement for a lockable window surface, in a way
that is anticipated to apply to all window surfaces in a future
version of EGL.
* Wording in EGL_KHR_lock_surface could be read to imply that almost
all surface attributes are invariant for a lockable surface.
EGL_KHR_lock_surface2 clarifies the wording.
* EGL_KHR_lock_surface2 clarifies what is returned when
the attribute EGL_MATCH_FORMAT_KHR is queried, especially when
one of the "inexact" formats was used to choose the config.
* EGL_KHR_lock_surface did not specify when a surface could change size.
EGL_KHR_lock_surface2 specifies that a surface cannot change size
when it is locked.
* EGL_KHR_lock_surface2 adds the config attribute
EGL_BITMAP_PIXEL_SIZE_KHR, to allow an application to dynamically
detect pixel layout for a format with a "hole", such as RGBU8888
(where "U" means "unused").
New Tokens
Accepted by the <attribute> parameter of eglQuerySurface:
EGL_BITMAP_PIXEL_SIZE_KHR 0x3110
Full text of EGL_KHR_lock_surface plus EGL_KHR_lock_surface2:
Overview
This extension allows mapping color buffers of EGL surfaces into the
client address space. This is useful primarily for software
rendering on low-end devices which do not support EGL client
rendering APIs, although it may be implemented efficiently on more
capable devices as well.
New Types
None
New Procedures and Functions
EGLBoolean eglLockSurfaceKHR(EGLDisplay dpy, EGLSurface surface,
const EGLint *attrib_list);
EGLBoolean eglUnlockSurfaceKHR(EGLDisplay dpy,
EGLSurface surface);
New Tokens
Returned in the EGL_SURFACE_TYPE bitmask attribute of EGLConfigs:
EGL_LOCK_SURFACE_BIT_KHR 0x0080
EGL_OPTIMAL_FORMAT_BIT_KHR 0x0100
Accepted as an attribute name in the <attrib_list> argument of
eglChooseConfig, and the <attribute> argument of eglGetConfigAttrib:
EGL_MATCH_FORMAT_KHR 0x3043
Accepted as attribute values for the EGL_MATCH_FORMAT_KHR attribute
of eglChooseConfig, and defined as possible values of that attribute
when querying it:
EGL_FORMAT_RGB_565_EXACT_KHR 0x30C0
EGL_FORMAT_RGBA_8888_EXACT_KHR 0x30C2
Accepted as attribute values for the EGL_MATCH_FORMAT_KHR attribute
of eglChooseConfig:
EGL_FORMAT_RGB_565_KHR 0x30C1
EGL_FORMAT_RGBA_8888_KHR 0x30C3
Accepted as attribute names in the <attrib_list> argument of
eglLockSurfaceKHR:
EGL_MAP_PRESERVE_PIXELS_KHR 0x30C4
EGL_LOCK_USAGE_HINT_KHR 0x30C5
Accepted as bit values in the EGL_LOCK_USAGE_HINT_KHR bitmask attribute
of eglLockSurfaceKHR:
EGL_READ_SURFACE_BIT_KHR 0x0001
EGL_WRITE_SURFACE_BIT_KHR 0x0002
Accepted by the <attribute> parameter of eglQuerySurface:
EGL_BITMAP_POINTER_KHR 0x30C6
EGL_BITMAP_PITCH_KHR 0x30C7
EGL_BITMAP_ORIGIN_KHR 0x30C8
EGL_BITMAP_PIXEL_RED_OFFSET_KHR 0x30C9
EGL_BITMAP_PIXEL_GREEN_OFFSET_KHR 0x30CA
EGL_BITMAP_PIXEL_BLUE_OFFSET_KHR 0x30CB
EGL_BITMAP_PIXEL_ALPHA_OFFSET_KHR 0x30CC
EGL_BITMAP_PIXEL_LUMINANCE_OFFSET_KHR 0x30CD
EGL_BITMAP_PIXEL_SIZE_KHR 0x3110
Returns in the *<value> parameter of eglQuerySurface when
<attribute> is EGL_BITMAP_ORIGIN_KHR:
EGL_LOWER_LEFT_KHR 0x30CE
EGL_UPPER_LEFT_KHR 0x30CF
Additions to Chapter 2 of the EGL 1.3 Specification (EGL Operation)
Add to the end of section 2.2.2:
Finally, some surfaces may be <locked>, which allows the
implementation to map buffers of that surface into client memory
for use by software renderers(fn). Locked surfaces cannot be
used for any other purpose. When a locked surface is <unlocked>,
any changes to the mapped buffer(s) are reflected in the actual
graphics or system memory containing the surface.
[fn: on implementations not supporting mapping graphics
memory, or which do not wish to take the stability and
security risks that entail, mapping may be done using
copy-out and copy-in behavior.]
Additions to Chapter 3 of the EGL 1.3 Specification (EGL Functions and Errors)
Add to the description of the EGL_BAD_ACCESS error in section 3.1:
"... or, a surface is locked)."
Add to table 3.2 ("Types of surfaces supported by an EGLConfig")
EGL Token Name Description
-------------------- ------------------------------------
EGL_LOCK_SURFACE_BIT_KHR EGLConfig allows locking surfaces
EGL_OPTIMAL_FORMAT_BIT_KHR This format is considered optimal
(preferred) when locking / mapping /
unlocking is being done.
Change the first paragraph under "Other EGLConfig Attribute
Descriptions" on p. 16:
"EGL_SURFACE_TYPE is a mask indicating both the surface types
that can be created by the corresponding EGLConfig (the config
is said to <support> those surface types), and the optional
behaviors such surfaces may allow. The valid bit settings are
shown in Table 3.2."
Add a new paragraph following the second paragraph of the same
section:
"If EGL_LOCK_SURFACE_BIT_KHR is set in EGL_SURFACE_TYPE, then
a surface created from the EGLConfig may be locked, mapped into
client memory, and unlocked. Locking is described in section
3.5.6. If EGL_OPTIMAL_FORMAT_BIT_KHR is set in
EGL_SURFACE_TYPE, then the surface is considered optimal (by
the implementation) from a performance standpoint when buffer
mapping is being done."
Replace the second paragraph of section 3.3 "EGL Versioning":
"The EGL_CLIENT_APIS string describes which client rendering APIs
are supported. It is zero-terminated and contains a
space-separated list of API names, which may include
``OpenGL_ES'' if OpenGL ES is supported, and ``OpenVG'' if
OpenVG is supported. If no client APIs are supported, then the
empty string is returned."
Insert a new paragraph and table in section 3.4.1 "Querying
Configurations", following the description of
EGL_MATCH_NATIVE_PIXMAP on page 21:
"If EGL_MATCH_FORMAT_KHR is specified in <attrib_list>, it must
be followed by one of the attribute values EGL_DONT_CARE,
EGL_NONE, or one of the format tokens in table
[locksurf.format].
When EGL_MATCH_FORMAT_KHR has the value EGL_NONE, only configs
which cannot be locked or mapped will match. Such configs must
not have the EGL_LOCK_SURFACE_BIT_KHR set in EGL_SURFACE_TYPE.
When EGL_MATCH_FORMAT_KHR has the value EGL_DONT_CARE, it is
ignored.
When EGL_MATCH_FORMAT_KHR has one of the values in table
[locksurf.format], only EGLConfigs describing surfaces whose
color buffers have the specified format, when mapped with
eglLockSurface, will match this attribute. In this case, the
EGL_<component>_SIZE attributes of resulting configs must agree
with the specific component sizes specified by the format."
Specific Format Name Description
-------------------- -----------
EGL_FORMAT_RGB_565_EXACT_KHR RGB565 fields in order from MSB to LSB within a 16-bit integer
EGL_FORMAT_RGB_565_KHR RGB565 fields in implementation-chosen order within a 16-bit integer
EGL_FORMAT_RGBA_8888_EXACT_KHR RGBA8888 fields in B, G, R, A byte order in memory
EGL_FORMAT_RGBA_8888_KHR RGBA8888 fields in implementation-chosen order within a 32-bit integer
------------------------------------------------------------------------------------------------------
Table [locksurf.format]: Specific formats for mapped pixels.
Add to table 3.4 ("Default values and match critera for EGLConfig
attributes") on page 22:
Attribute Default Selection Sort Sort
Criteria Order Priority
-------------------- ------------- --------- ----- --------
EGL_MATCH_FORMAT_KHR EGL_DONT_CARE Exact None -
Add EGL_MATCH_FORMAT_KHR to the last paragraph in section 3.4.1 on
p. 23, describing attributes not used for sorting EGLConfigs.
Add a new paragraph to the end of section 3.4.3 "Querying Configuration
Attributes":
"Querying the EGL_MATCH_FORMAT_KHR attribute results in EGL_NONE
for an EGLConfig that is not lockable, one of the "exact" formats
(EGL_FORMAT_RGB_565_EXACT_KHR, EGL_FORMAT_RGBA_8888_EXACT_KHR)
if the color buffer matches that format when mapped with
eglLockSurface, or for any other format a value that is not
EGL_NONE or EGL_DONT_CARE but is otherwise undefined. In particular,
the color buffer format matching one of the "inexact" formats
does not guarantee that that EGL_FORMAT_* value is returned."
In section 3.5.1 "Creating On-Screen Rendering Surfaces", add the
following to the paragraph that lists the attributes that can be set
in attrib_list:
"... and EGL_SWAP_BEHAVIOR."
and add a new penultimate paragraph:
"EGL_SWAP_BEHAVIOR specifies the initial value of the
EGL_SWAP_BEHAVIOR surface attribute (section 3.5.6), and is thus
either EGL_BUFFER_PRESERVED or EGL_BUFFER_DESTROYED. This setting
of EGL_SWAP_BEHAVIOR at surface creation time is supported only
for a lockable surface, i.e. where the EGLConfig has
EGL_LOCK_SURFACE_BIT_KHR set in EGL_SURFACE_TYPE."
In EGL 1.4, also add the following text to that same paragraph:
"For such a lockable surface, whether it is possible to change
the EGL_SWAP_BEHAVIOR attribute after surface creation is
determined by EGL_SWAP_BEHAVIOR_PRESERVED_BIT in the
EGL_SURFACE_TYPE EGLConfig attribute."
Add a new section following the current section 3.5.5:
"3.5.6 Locking and Mapping Rendering Surfaces
A rendering surface may be <locked> by calling
EGLBoolean eglLockSurfaceKHR(EGLDisplay dpy,
EGLSurface surface,
const EGLint *attrib_list);
While a surface is locked, only two operations can be performed
on it. First, a surface attribute may be queried using
eglQuerySurface. This includes the case of querying
EGL_BITMAP_POINTER_KHR, which causes the surface to be
<mapped> (if not already mapped) and gives
a pointer into client memory corresponding to the memory
of the mapped buffer. Second, the surface may be unlocked. Any
attempts to use a locked surface in other EGL APIs will fail and
generate an EGL_BAD_ACCESS error.
While a surface is locked, its dimensions (the values of the EGL_WIDTH
and EGL_HEIGHT surface attributes) do not change. They may change
at any other time, therefore an application must query these
attributes <after> the call to eglLockSurfaceKHR to ensure that it has
the correct size of the mapped buffer.
<attrib_list> specifies additional parameters affecting the locking
operation. The list has the same structure as described for
eglChooseConfig. Attributes that may be defined are shown in table
[locksurf.attr], together with their default values if not specified
in <attrib_list>, and possible values which may be specified for
them in <attrib_list>.
Attribute Name Type Default Value Possible Values
----------------------- ------- ------------- -------------------------
EGL_MAP_PRESERVE_PIXELS_KHR boolean EGL_FALSE EGL_TRUE / EGL_FALSE
EGL_LOCK_USAGE_HINT_KHR bitmask EGL_READ_SURFACE_BIT_KHR | Any combination of
EGL_WRITE_SURFACE_BIT_KHR EGL_READ_SURFACE_BIT_KHR
and EGL_WRITE_SURFACE_BIT_KHR
--------------------------------------------------------------
Table [locksurf.attr]: eglLockSurfaceKHR attribute names,
default values, and possible values.
On failure, the surface is unaffected and eglLockSurfaceKHR
returns EGL_FALSE. An EGL_BAD_ACCESS error is generated if any
of these condition, are true:
* <surface> was created with an EGLConfig whose
EGL_SURFACE_TYPE attribute does not contain
EGL_LOCK_SURFACE_BIT_KHR.
* <surface> is already locked.
* Any client API is current to <surface>.
An EGL_BAD_ATTRIBUTE error is generated if an attribute or
attribute value not described in table [locksurf.attr] is
specified.
Mapping Buffers of a Locked Surface
-----------------------------------
The color buffer of a locked surface can be <mapped> by calling
eglQuerySurface (see section 3.5.7) with <attribute>
EGL_BITMAP_POINTER_KHR(fn). The query returns a pointer to a
buffer in client memory corresponding to the color buffer of
<surface>. In the case of a back-buffered surface, color buffer
refers to the back buffer.
[fn: "mapped" only means that the pointer returned is
intended to *correspond* to graphics memory. Implementation
are not required to return an actual pointer into graphics
memory, and often will not.]
The contents of the mapped buffer are initially undefined(fn)
unless either the EGL_MAP_PRESERVE_PIXELS_KHR attribute of
eglLockSurfaceKHR is set to EGL_TRUE, or (for a window surface)
the EGL_SWAP_BEHAVIOR surface attribute is set to
EGL_BUFFER_PRESERVE, in which case the contents
of the buffer are taken from the contents of <surface>'s color
buffer. The default value of EGL_MAP_PRESERVE_PIXELS_KHR is
EGL_FALSE.
[fn: In order to avoid pipeline stalls and readback delays on
accelerated implementations, we do not mandate that the
current contents of a color buffer appear when it's mapped
to client memory, unless the EGL_MAP_PRESERVE_PIXELS_KHR
flag is set or (for a window surface) EGL_SWAP_BEHAVIOR is
set to EGL_BUFFER_PRESERVE. Applications using mapped
buffers which are not
preserved must write to every pixel of the buffer before
unlocking the surface. This constraint is considered
acceptable for the intended usage scenario (full-frame
software renderers). Such an application may lock-render-unlock
multiple times per frame (i.e. per eglSwapBuffers) by setting
EGL_MAP_PRESERVE_PIXELS_KHR to EGL_TRUE for the second and
subsequent locks.
Note that EGL_SWAP_BEHAVIOR also controls whether the color
buffer contents are preserved over a call to eglSwapBuffers.]
The EGL_LOCK_USAGE_HINT_KHR attribute of eglLockSurfaceKHR is a
bitmask describing the intended use of the mapped buffer. If the
mask contains EGL_READ_SURFACE_BIT_KHR, data will be read from
the mapped buffer. If the mask contains
EGL_WRITE_SURFACE_BIT_KHR, data will be written to the mapped
buffer. Implementations must support both reading and writing to
a mapped buffer regardless of the value of
EGL_LOCK_USAGE_HINT_KHR, but performance may be better if the
hint is consistent with the actual usage of the buffer. The
default value of EGL_LOCK_USAGE_HINT_KHR hints that both reads
and writes to the mapped buffer will be done.
Other attributes of the mapped buffer describe the format of
pixels it contains, including its pitch (EGL_BITMAP_PITCH_KHR),
origin (EGL_BITMAP_ORIGIN_KHR), and the bit location of each color
component within a pixel (EGL_BITMAP_PIXEL_x_OFFSET_KHR). These
attributes may be queried using eglQuerySurface, and are described
in more detail in section 3.5.7.
The EGL_BITMAP_POINTER_KHR and EGL_BITMAP_PITCH_KHR attributes
of a locked surface may change following successive calls to
eglLockSurfaceKHR(fn), so they must be queried each time a
buffer is mapped. Other attributes of a mapped buffer (listed in
the paragraph above) are invariant and need be queried only once
following surface creation.
[fn: The pointer and pitch of a mapped buffer may change due
to display mode changes, for example.]
Mapping will not suceed if client memory to map the surface into
cannot be allocated. In this case, querying eglQuerySurface with
<attribute> EGL_BITMAP_POINTER_KHR will fail and generate an EGL
error.
Unlocking Surfaces
------------------
A rendering surface may be <unlocked> by calling
EGLBoolean eglUnlockSurfaceKHR(EGLDisplay dpy,
EGLSurface surface);
Any mapped buffers of <surface> become unmapped following
eglUnlockSurfaceKHR. Any changes made to mapped buffers of
<surface> which it was locked are reflected in the surface after
unlocking(fn).
[fn: This language enables implementations to either map
video memory, or copy from a separate buffer in client
memory.]
If <surface> was created with an EGLConfig whose
EGL_SURFACE_TYPE attribute contains EGL_OPTIMAL_FORMAT_BIT_KHR,
then the surface is considered optimal (by the implementation)
from a performance standpoint when buffer mapping is being
done(fn).
[fn: This often means that the format of all mapped buffers
corresponds directly to the format of those buffers in
<surface>, so no format conversions are required during
unmapping. This results in a high-performance software
rendering path. But "optimal format" is really just a hint
from EGL that this config is preferred, whatever the actual
reason.]
On failure, eglUnlockSurfaceKHR returns EGL_FALSE. An
EGL_BAD_ACCESS error is generated if any of these conditions are
true:
* <surface> is already unlocked.
* A display mode change occurred while the surface was locked,
and the implementation was unable to reflect mapped buffer
state(fn). In this case, <surface> will still be unlocked.
However, the contents of the previously mapped buffers of
<surface> become undefined, rather than reflecting changes
made in the mapped buffers in client memory.
[fn: Usually this may only occur with window surfaces which
have been mapped. EGL does not have an event mechanism to
indicate display mode changes. If such a mechanism exists
(using native platform events or the OpenKODE event system),
applications should respond to mode changes by regenerating
all visible window content, including re-doing any software
rendering overlapping the mode change.]"
Add to table 3.5 ("Queryable surface attributes and types")
Attribute Type Description
--------- ---- -----------
EGL_BITMAP_POINTER_KHR pointer Address of a mapped color buffer (MCB).
EGL_BITMAP_PITCH_KHR integer Number of bytes between the start of
adjacent rows in an MCB.
EGL_BITMAP_ORIGIN_KHR enum Bitmap origin & direction
EGL_BITMAP_PIXEL_x_OFFSET_KHR integer Bit location of each color buffer
component within a pixel in an MCB.
EGL_BITMAP_PIXEL_SIZE_KHR integer Bits per pixel
In EGL 1.4 only, in the description of eglSurfaceAttrib properties
that can be set in section 3.5.6 (renumbered to 3.5.7), on page 35,
add to the first paragraph describing EGL_SWAP_BEHAVIOR:
"The value of EGL_SWAP_BEHAVIOR also affects the semantics of
eglLockSurfaceKHR for a lockable window surface. See section 3.5.6."
In EGL 1.4 only, in the description of eglSurfaceAttrib properties
that can be set in section 3.5.6 (renumbered to 3.5.7), on page 35,
change the paragraph concerning the initial value of EGL_SWAP_BEHAVIOR
to:
"The initial value of EGL_SWAP_BEHAVIOR is chosen by the
implementation, except for a lockable window surface (i.e. where the
EGLConfig has both EGL_LOCK_SURFACE_BIT_KHR and EGL_WINDOW_BIT set in
EGL_SURFACE_TYPE), where the default is EGL_BUFFER_PRESERVED, but it
may be overridden by specifying EGL_SWAP_BEHAVIOR to
eglCreateWindowSurface."
In EGL 1.3 only, in the description of eglQuerySurface properties
that can be queried in section 3.5.6 (renumbered to 3.5.7), on page 33,
add to the paragraph describing EGL_SWAP_BEHAVIOR:
"The value of EGL_SWAP_BEHAVIOR also affects the semantics of
eglLockSurfaceKHR for a lockable window surface. See section 3.5.6.
For a lockable window surface (one whose EGLConfig has both
EGL_LOCK_SURFACE_BIT_KHR and EGL_WINDOW_BIT set in EGL_SURFACE_TYPE),
the value of this attribute may be set in the eglCreateWindowSurface
call, and if not set there defaults to EGL_BUFFER_PRESERVED. See
section 3.5.1. The default for a non-lockable surface is chosen by the
implementation."
Add to the description of eglQuerySurface properties in section
3.5.6 (renumbered to 3.5.7) on page 33:
"Properties of a bitmap surface which may be queried include:
* EGL_BITMAP_POINTER_KHR, which maps the color buffer of a
locked surface and returns the address in client memory of
the mapped buffer.
* EGL_BITMAP_PITCH_KHR, which returns the number of bytes
between successive rows of a mapped buffer.
* EGL_BITMAP_ORIGIN_KHR, which describes the way in which a
mapped color buffer is displayed on the screen. Possible
values are either EGL_LOWER_LEFT_KHR or EGL_UPPER_LEFT_KHR,
indicating that the first pixel of the mapped buffer
corresponds to the lower left or upper left of a visible
window, respectively.
* EGL_BITMAP_PIXEL_<x>_OFFSET_KHR, which describes the bit
location of the least significant bit of each color
component of a pixel within a mapped buffer. <x> is one of
RED, GREEN, BLUE, ALPHA, or LUMINANCE.
The offset for a color component should be treated as the
number of bits to left shift the component value to place it
within a n-bit
integer containing the pixel, where n is the number of bits
per pixel. If a color
component does not exist in the mapped buffer, then the bit
offset of that component is zero. If a color component
does exist but not in a single contiguous range of bits,
then the value of the attribute is EGL_UNKNOWN.
* EGL_BITMAP_PIXEL_SIZE_KHR, which returns the number of bits
per pixel, assumed to be least significant leftmost if there
are multiple pixels per byte. The attribute takes the value
EGL_UNKNOWN if this assumption is not true, or if pixels are not
laid out left to right in memory (for example pairs of 16-bit
pixels are swapped in memory).
In addition to these attributes, the number of bits for each
color component of a pixel within a mapped buffer is obtained by
querying the EGL_<x>_SIZE attribute of the EGLConfig used to
create the surface, where <x> is <x> is one of RED, GREEN, BLUE,
ALPHA, or LUMINANCE.
Querying EGL_BITMAP_POINTER_KHR and EGL_BITMAP_PITCH_KHR is only
allowed when <surface> is mapped (see section 3.5.6). Querying
either of these attributes for the first time after calling
eglLockSurfaceKHR causes the color buffer of the locked surface
to be mapped. Querying them again before unlocking the surface
will return the same values as the first time. However, after
calling eglUnlockSurfaceKHR, these properties become undefined.
After a second call to eglLockSurfaceKHR, these properties may
again be queried, but their values may have changed.
Other properties of the mapped color buffer of a surface
(in the list above) are
invariant, and need be queried only once following surface
creation. If <surface> was created with an EGLConfig whose
EGL_SURFACE_TYPE attribute does not contain
EGL_LOCK_SURFACE_BIT_KHR, queries of EGL_BITMAP_ORIGIN_KHR,
EGL_BITMAP_PIXEL_<x>_OFFSET_KHR and EGL_BITMAP_PIXEL_SIZE_KHR
return undefined values."
Add to the description of eglQuerySurface errors in the last
paragraph of section 3.5.6 (renumbered to 3.5.7) on page 34:
"... If <attribute> is either EGL_BITMAP_POINTER_KHR or
EGL_BITMAP_PITCH_KHR, and either <surface> is not locked using
eglLockSurfaceKHR, or <surface> is locked but mapping fails,
then an EGL_BAD_ACCESS error is generated."
In section 3.9.3 Posting Semantics on page 46, append to the first
paragraph:
"This restriction does not apply to lockable surfaces; for such
a surface, eglSwapBuffers and eglCopyBuffers may be called for
a surface not bound to any client API context(fn).
[fn: Normally this would only be done when using methods other
than client API rendering to specify the color buffer contents,
such as software rendering to a locked surface.]"
and replace the second paragraph ("If <dpy> and <surface> ... not be
executed until posting is completed.") with:
"If <surface> is bound to a current client API context for the calling
thread, eglSwapBuffers and eglCopyBuffers perform an implicit flush
operation on the context (glFlush for an OpenGL or OpenGL ES context,
vgFlush for an OpenVG context). Subsequent client API commands can be
issued immediately, but will not be executed until posting is
completed.
If <surface> is current to a client API context in any thread other
than the calling thread, eglSwapBuffers and eglCopyBuffers will fail.
and append the following sentence to the eglSwapInterval paragraph:
"The swap interval has no effect on an eglSwapBuffers for a surface
not bound to a current client API context."
In 3.9.4 Posting Errors, change the sentence "If <surface> is not bound
to the calling thread's current context, an EGL_BAD_SURFACE error is
generated." to:
"If <surface> is bound to a current context in a thread other
than the calling thread, an EGL_BAD_SURFACE error is generated."
Issues
1) What is the rationale for this extension?
Software renderers on low-end implementations need an efficient way
to draw pixel data to the screen. High-end implementations must
support the same interface for compatibility, while not compromising
the accelerability of OpenGL ES and OpenVG client APIs using
dedicated graphics hardware and memory.
Using lock/unlock semantics enables low-end implementations to
expose pointers directly into display memory (as extremely dangerous
as that may be), while high-end implementations may choose to create
backing store in client memory when mapping a buffer, and copy it to
graphics memory when the surface is unlocked. Making the initial
contents of a mapped buffer undefined means that no readbacks from
graphics memory are required, avoiding pipeline stalls.
This extension is not intended to support mixed-mode (client API and
software) rendering. Since mapped buffer contents are undefined,
unless the buffer is explicitly preserved (which may be unacceptably
expensive on many implementations), applications doing software
rendering must touch every pixel of mapped buffers at least once
before unlocking the surface.
2) Do we need to know if locked surfaces are "fast" or "native"?
RESOLVED: Yes. This is indicated with the EGL_OPTIMAL_FORMAT_BIT_KHR
of EGL_SURFACE_TYPE. However, note that there is no way to guarantee
what "fast" or "no format conversions" really means; this is little
more than an implementation hint.
3) Should we be able to map buffers other than the color buffer?
RESOLVED: Not initially. However, the <attrib_list> parameter of
eglLockSurfaceKHR supports this in the future. There is no <buffer>
attribute to eglQuerySurface, so such a layered extension would have
to either create a new naming convention (such as
EGL_BITMAP_{DEPTH,COLOR,STENCIL,ALPHA_MASK}_POINTER), or define an
extended query eglQuerySurfaceBuffer() which takes a <buffer>
parameter. It would also be tricky to support interleaved depth /
stencil formats. But the attribute list offers some future-proofing
at low cost.
4) What properties of mapped buffers can be queried?
RESOLVED: A pointer to the buffer and its pitch, both of which may
change in successive lock/unlock cycles. These may be queried only
while the underlying surface is locked, and are undefined after
unlocking. The first query following locking is the point at which
actual buffer mapping must occur.
RESOLVED: Additionally, the pixel size, origin, and color component
bitfield size and offset for each component, which are invariant
and may be queried at any time.
5) How are mode changes indicated? What happens to the mapped
buffer during a mode change?
RESOLVED: UnlockSurfaceKHR fails and raises an error if a mode
change occurred while the surface was locked (although the surface
still ends up in the unlocked state - this is necessary since
there's no way to clear the error!). If a mode change occurs while a
buffer is mapped, the implementation must still allow the
application to access mapped buffer memory, even though the contents
of the mapped buffer may not be reflected in the changed window
after unmapping.
Note: There's no convenient way to indicate mode changes while
a surface is unlocked, despite that being useful to tell apps they
have to redraw. The problem is that we don't have an event system,
and the power management functionality is overkill since the only
resources which are likely to be damaged by a mode change are
visible window contents. Fortunately, this problem is beyond the
scope of this extension.
6) Does locking a surface imply mapping its buffers?
RESOLVED: No. Locking simply places the surface in that state and
prevents it from being made current / swapped / etc. Buffers are
mapped only when their pointers or pitch are queried using
eglQuerySurface.
An interesting side effect of this resolution is that calling
eglLockSurfaceKHR immediately followed by eglUnlockSurfaceKHR DOES
NOT CHANGE THE CONTENTS OF BUFFERS, since none of them were mapped.
Likewise locking a surface, querying a buffer pointer or pitch, and
then unlocking it without changing the mapped buffer contents causes
the surface contents of the mapper buffer(s) to become undefined.
At the Atlanta F2F, there was a suggestion that eglLockSurfaceKHR
should immediately map the color buffer and return a pointer to it,
on the basis that this would make it harder for applications to
mistakenly use an old buffer pointer from a previous mapping cycle.
At the same time, people working on more powerful GPUs wanted the
lock operation to be lightweight. These are not consistent goals and
we have thus far chosen to separate the lightweight locking, and
more expensive mapping operations.
7) Can buffer contents be preserved in mapping?
RESOLVED: Yes. The default behavior is to discard / leave undefined
the mapped buffer contents, but the EGL_MAP_PRESERVE_PIXELS_KHR flag
may be specified to eglLockSurfaceKHR.
8) Should usage hints be provided during mapping?
RESOLVED: Yes, they may be provided in the EGL_LOCK_USAGE_HINT_KHR
bitmask attribute to eglLockSurfaceKHR. Implementations are required
to behave correctly no matter the value of the flag vs. the
operations actually performed, so the hint may be ignored.
9) Should we be able to lock subrectangles of a surface?
RESOLVED: No. However, the attribute list parameter of
eglLockSurfaceKHR allows a layered extension to implement this
behavior by specifying an origin and size to map within the buffer.
10) Should the BITMAP_PIXEL_<component>_OFFSET attributes belong to the
surface, or the config?
RESOLVED: Part of the surface. Configs supporting a specific format
are matched with config attribute EGL_MATCH_FORMAT_KHR, which
supports specific bit-exact formats such as
EGL_FORMAT_565_EXACT_KHR.
11) Can the pixel size in a mapped buffer be derived from the
EGL_BUFFER_SIZE attribute of the config used to create it?
RESOLVED: Yes. In principle, hardware using padding bytes in its
framebuffer storage could be a problem, and a separate
BITMAP_PIXEL_SIZE surface attribute would be needed. However, we
don't think implementations are likely to waste graphics memory and
bandwidth in this fashion.
12) How are color component locations within a pixel described?
RESOLVED: Each R, G, B, and A component has a queryable bit offset
within an integer. The size of the integer depends on the total size
of the pixel; for the 565 formats, the pixel is a 16-bit integer.
For the 8888 formats, the pixel is a 32-bit integer.
We cannot describe component locations with byte locations, since
the 565 formats have components straddling byte boundaries. However,
this means that offsets for the RGBA8888_EXACT format are different
between little- and big-endian CPUs, since the desired format is B,
G, R, A components laid out as bytes in increasing memory order.
13) Can mapped buffer contents be affected by other EGL operations?
RESOLVED: No. A locked surface only allows two operations:
unlocking, and mapping. No other EGL operations can take place while
the surface is locked (if this were not the case, then
eglSwapBuffers might destroy the contents of a mapped buffer).
It is possible that operations outside the scope of EGL could affect
a mapped color buffer. For example, if a surface's color buffer were
made up of an EGLImage, one of the EGL client APIs could draw to
that image while it was mapped. Responsibility for avoiding this
situation is in the hands of the client.
14) Can EGL_MATCH_FORMAT_KHR be queried for a config?
RESOLVED: Yes. Unlockable configs return EGL_NONE for this
attribute.
For a config whose format matches one of the EGL_FORMAT_*_EXACT_KHR
definitions, that value is returned.
For a config whose format does not match one of the
EGL_FORMAT_*_EXACT_KHR definitions, but it does match one of the
inexact definitions, we do not want to specify that it returns that
value, since that precludes it returning an exact format if one were
to be defined in a future extension, or a future version of this
extension.
Therefore, for a config whose format does not match a defined
"exact" format, the attribute returns a value other than the defined
"exact" formats and EGL_NONE and EGL_DONT_CARE.
15) Is a goal of this extension to support "mixed-mode" rendering (both
software and EGL client API rendering to the same surface)?
RESOLVED: No. An implementation *can* choose to export configs
supporting creation of lockable surfaces which also support
rendering by OpenGL ES, OpenVG, or other client APIs (when the
surface is not locked). But there is nothing in the extension
requiring this, and the motivation for the extension is simply to
support software rendering.
16) Can mapping a locked surface fail?
RESOLVED: Yes, if memory can't be allocated in the client. This is
indicated by queries of EGL_BITMAP_POINTER_KHR and
EGL_BITMAP_PITCH_KHR failing and generating an EGL_BAD_ACCESS error.
17) How does the application specify that it would like the pixels in the
mapped buffer to be preserved from the state at the last unlock?
[Bug 4410]
There are actually two items that need to be configurable:
a. Whether eglSwapBuffers preserves the color buffer.
b. Whether eglLockSurfaceKHR puts the color buffer pixels into the
locked buffer.
An implementation may use two different render paths depending on whether
the application is full frame rendering or incrementally rendering, where
the full frame rendering path may involve less copying than the
incrementally rendering path. The implementation needs to know in advance
which path to use; finding out from EGL_MAP_PRESERVE_PIXELS_KHR on
eglLockSurfaceKHR whether that lock should have the color buffer pixels
from the last eglUnlockSurfaceKHR in the last frame is too late.
This need to know in advance only applies to (a).
Note that the original EGL_KHR_lock_surface extension failed to state
that EGL_SWAP_BEHAVIOR needed to be set to EGL_BUFFER_PRESERVED for
incremental rendering.
RESOLVED: EGL_MAP_PRESERVE_PIXELS_KHR continues to control (b), as in the
original EGL_KHR_lock_surface extension. EGL_SWAP_BEHAVIOR continues
to control (a), as in unextended EGL. It is possible to set
EGL_SWAP_BEHAVIOR in eglCreateWindowSurface, even on EGL versions
before 1.4, and even if the config does not have
EGL_SWAP_BEHAVIOR_PRESERVED_BIT set in EGL_SURFACE_TYPE.
For a lockable window surface, EGL_SWAP_BEHAVIOR needs to default to
EGL_BUFFER_PRESERVED, and having EGL_SWAP_BEHAVIOR set to
EGL_BUFFER_PRESERVED needs to override EGL_MAP_PRESERVE_PIXELS_KHR. This is for
compatibility with version 1 of this extension, which had EGL_SWAP_BEHAVIOR
as the sole means of controlling (a) and (b).
This resolution is backward compatible with the original
EGL_KHR_lock_surface extension as it now defaults to preserving pixels,
and thus has that default for an old application that does not know about
setting EGL_SWAP_BEHAVIOR=EGL_BUFFER_DESTROYED to optimize full frame
rendering. The downside is a possible performance reduction of an old
application that does not want pixels preserved, until it is appropriately
modified and recompiled.
18) When can a lockable surface change size? [Bug 4522]
RESOLVED: The surface cannot change size while it is locked. For any
window size change during that time, EGL defers the corresponding
surface size change until some time after the unlock.
The implication is that an application must get the surface size
_after_locking, otherwise the size may have changed in between the
attribute query and the lock.
As part of this change, using eglQuerySurface on any surface attribute
while the surface is locked has been allowed.
19) Should we provide for more than one pixel per byte, or a format with
a "hole", or a byte-swapped 16-bit format? [Bug 4513]
For the use case where the application uses EGL_FORMAT_* to
determine the pixel format, this is not an issue. For the use case
where the application is written to cope with any format by reading
the EGL_BITMAP_* attributes and adjusting its rendering at runtime,
this is an issue.
RESOLVED: We add a new attribute EGL_BITMAP_PIXEL_SIZE_KHR which gives
the pixel size in bits.
The use of this attribute assumes that multiple pixels packed into
a byte are arranged least significant on the left, and that pixels
are arranged leftmost first in memory. If either of these assumptions
is false, then EGL_BITMAP_PIXEL_SIZE_KHR is EGL_UNKNOWN to indicate
that an application attempting to use these attributes to adjust its
rendering code will not work.
We also define that any of the EGL_BITMAP_PIXEL_x_OFFSET_KHR attributes
have a value of EGL_UNKNOWN if the corresponding color component is not
a contiguous bit range.
20) EGL states that surface must be bound to the thread's current context
when posting, but there is no current context when using lock surface.
eglSwapInterval is also related to the context. [Bug 5923]
RESOLVED: The posting restrictions are relaxed, so it is allowed to
post for a surface not bound to any client API context. In this case
no eglSwapInterval setting affects an eglSwapBuffers. This relaxation
only applies to a lockable surface so that this extension continues
to be implementable as a middleware layer on top of a "real" EGL.
The relaxation applies to a surface that is both lockable and client
API renderable; the implicit flush in a post only occurs when the surface
is bound to a current API context. Posting a surface bound to a different
thread's current API context is still disallowed.
21) Why has this extension been obsoleted and replaced by
EGL_KHR_lock_surface3?
RESOLVED: Starting with the December 4, 2013 release of EGL 1.4, EGLint
is defined to be the same size as the native platform "int" type. Handle
and pointer attribute values *cannot* be represented in attribute lists
on platforms where sizeof(handle/pointer) > sizeof(int). Existing
extensions which assume this functionality are being replaced with new
extensions specifying new entry points to work around this issue. See
the latest EGL 1.4 Specification for more details.
Revision History
Version 3, 2013/12/04 - Add issue 21 explaining that the bitmap pointer
cannot be safely queried using this extension on 64-bit platforms,
and suggesting EGL_KHR_lock_surface3 instead. Change formal
parameter names from 'display' to 'dpy' to match other EGL APIs.
Version 2, 2010/03/10
Allocated extension number 16.
[Bug 6089] Clarified that it is the back buffer of a back-buffered
surface that is mapped.
[Bug 4410] Issue 17: Reinstate EGL_MAP_PRESERVE_PIXELS_KHR attribute on
eglLockSurfaceKHR as the means of configuring whether a locked
buffer contains pixel data from the color buffer, as in
EGL_KHR_lock_surface, but, for a lockable window surface,
make EGL_SWAP_BEHAVIOR=EGL_BUFFER_PRESERVED override the effect
of that attribute. Tidy wording so that behavior, and that the
default of EGL_SWAP_BEHAVIOR must be EGL_BUFFER_PRESERVED, only
applies to lockable window surfaces, not lockable pbuffer or
pixmap surfaces.
[Bug 5923] Issue 20: A config that is lockable and renderable by
an EGL client API is now allowed. Posting restrictions are relaxed
to allow posting for a lockable surface not bound to any client
API context.
[Bug 5143] Fixed typos EGL_SURFACE_TYPE_KHR. Added summary of
changes over EGL_KHR_lock_surface.
Version 1, 2010/01/19
[Bug 4410] Issue 17: Change to use EGL_SWAP_BEHAVIOR as the
primary means of determining whether the application wants pixels
preserved between lock/unlock cycles.
Allowed setting of EGL_SWAP_BEHAVIOR in
eglCreateWindowSurface for a lockable surface. Made the
default EGL_BUFFER_PRESERVED on a lockable surface to maintain
compatibility with the original lock surface extension.
[Bug 3192] Wording clarification to avoid accidentally suggesting
that almost all surface attributes are invariant.
[Bug 4518] Issue 14: Clarify the returned value when
EGL_MATCH_FORMAT_KHR is queried.
[Bug 4522] Issue 18: Clarify that a locked surface cannot change
size.
[Bug 4513] Issue 19: Added EGL_BITMAP_PIXEL_SIZE_KHR.
[Bug 5923] Issue 20: Must be no current context when posting
a lockable surface. In that case any eglSwapInterval setting
is ignored. Disallowed a config that is lockable and renderable
by an EGL client API.
Revision History of original EGL_KHR_lock_surface extension
Version 18, 2010/03/23 - Added introductory remark referring to the
EGL_KHR_lock_surface2 extension.
Version 17, 2008/10/08 - Updated status (approved as part of
OpenKODE 1.0).
Version 16, 2008/01/24 - Add issue 16 noting that mapping can fail,
and a corresponding new error condition for eglQuerySurface.
Clean up the issues list.
Version 15, 2008/01/09 - Add issue 15 noting that supporting
mixed-mode rendering is not a goal or requirement of the
extension.
Version 14, 2007/11/07 - change ARGB_8888_EXACT back to
RGBA_8888_EXACT, since the offsets are now dependent on the
endianness of the CPU. Add issue 12 describing this, and clarify
that offsets are within a 16- or 32-bit integer depending on the
format. Added issue 13 clarifying that locked buffer contents
are not affected by eglSwapBuffers, because eglSwapBuffers
cannot be issued on a mapped surface. Allow querying
EGL_MATCH_FORMAT_KHR for a config, and added related issue 14.
Version 13, 2007/05/10 - change RGBA_8888_EXACT to ARGB_8888_EXACT
to match hardware layout.
Version 12, 2007/04/06 - clarify that when EGL_MATCH_FORMAT_KHR is
EGL_DONT_CARE, it does not affect component size of selected
configs.
Version 11, 2007/04/05 - add missing KHR suffix to some tokens.
Version 10, 2007/04/05 - assign enumerant values. Add OpenKODE 1.0
Provisional disclaimer.
Version 9, 2007/03/26 - add format tokens to "New Tokens"
section. Correct description of RGBA format tokens.
Version 8, 2007/03/26 - add issue 11 noting theoretical possibility
of EGL_BUFFER_SIZE not directly corresponding to the mapped
pixel size. Add EGL_MATCH_FORMAT_KHR attribute to
eglChooseConfig, and 565 / 8888 formats for it.
Version 7, 2007/03/25 - note in issue 5 that access to a mapped
buffer must continue to work even after a mode change. Add KHR
suffix to new functions and tokens. Remove BITMAP_PIXEL_<x>_BITS
and BITMAP_PIXEL_SIZE tokens, which duplicate information in the
EGLConfig. Add issue 10 asking whether bitmap pixel offset
attributes belong to the config, or to the surface.
Version 6, 2007/02/26 - allow EGL_CLIENT_APIS string to be empty in
implementations supporting only this extension.
Version 5, 2007/02/05 - update contributor list. Changed bit offset
queries to return LSB offset, rather than MSB offset.
Version 4, 2007/02/02 - correct extension name. Change
name of FAST_UNLOCK_BIT_KHR to OPTIMAL_FORMAT_BIT_KHR.
Replace buffer_mask parameter of eglLockSurfaceKHR with an
attribute list. Add the EGL_MAP_PRESERVE_PIXELS_KHR and
EGL_LOCK_USAGE_HINT_KHR attributes per request from Gary. Add issues
7, 8, and 9 describing these attributes and how to support
locking subrects in a layered extension, by extending the
attribute list.
Version 3, 2007/02/01 - the implementation once again controls the
mapped buffer memory. There is no longer a separate bitmap
surface type; any type surface may potentially be mapped, using
lock/unlock semantics.
Version 2, 2006/12/22 - simplify by only supporting drawing from
client memory to EGL surface color buffers. Specify use of
OpenGL DrawPixels terminology. Change name of the extension to
EGL_KHR_draw_pixels, since there is no longer any "bitmap
surface" involved.
Version 1, 2006/12/14 - write up as formal spec language for
external review.