blob: 5e4e947c1062d7ef21f1bad535ac9c01bd2ba23b [file] [log] [blame]
Name Strings
Graham Sellers (graham.sellers 'at'
Jeff Bolz, NVIDIA
Daniel Koch, NVIDIA
Jon Leech
Mark Kilgard, NVIDIA
Copyright (c) 2013 The Khronos Group Inc. Copyright terms at
Specification Update Policy
Khronos-approved extension specifications are updated in response to
issues and bugs prioritized by the Khronos OpenGL Working Group. For
extensions which have been promoted to a core Specification, fixes will
first appear in the latest version of that core Specification, and will
eventually be backported to the extension document. This policy is
described in more detail at
Complete. Approved by the ARB on June 3, 2013.
Ratified by the Khronos Board of Promoters on July 19, 2013.
Last Modified Date: April 20, 2015
Revision: 25
ARB Extension #144
This extension is written against version 4.3 of the Core Profile OpenGL
Specification, dated August 6, 2012.
The definition of this extension is affected by the presence of
OpenGL has long supported buffer objects as a means of storing data
that may be used to source vertex attributes, pixel data for textures,
uniforms and other elements. In un-extended GL, buffer data stores
are mutable - that is, they may be de-allocated or resized while they
are in use. The GL_ARB_texture_storage extension added immutable storage
for texture object (and was subsequently incorporated into OpenGL 4.2).
This extension further applies the concept of immutable storage to
buffer objects. If an implementation is aware of a buffer's immutability,
it may be able to make certain assumptions or apply particular
optimizations in order to increase performance or reliability.
Furthermore, this extension allows applications to pass additional
information about a requested allocation to the implementation which it
may use to select memory heaps, caching behavior or allocation strategies.
Finally, this extension introduces the concept of persistent client
mappings of buffer objects, which allow clients to retain pointers to a
buffer's data store returned as the result of a mapping, and to issue
drawing commands while those mappings are in place.
New Procedures and Functions
void BufferStorage(enum target,
sizeiptr size,
const void * data,
bitfield flags);
When EXT_direct_state_access is present:
void NamedBufferStorageEXT(uint buffer,
sizeiptr size,
const void * data,
bitfield flags);
New Tokens
Accepted in the <flags> parameter of BufferStorage and
MAP_READ_BIT 0x0001 (existing)
MAP_WRITE_BIT 0x0002 (existing)
Accepted as part of the <access> parameter to MapBufferRange:
Accepted by the <pname> parameter of GetBufferParameter{i|i64}v:
Accepted by the <barriers> parameter of MemoryBarrier:
IP Status
No known IP claims.
Additions to Chapter 2 of the OpenGL Core Profile Specification, Version 4.3,
"OpenGL Fundamentals"
Insert before the last line of Section 2.5.2, "Buffer Objects", p. 26:
Under certain circumstances, the data store of a buffer object may
be shared between the client and server and accessed simultaneously
by both.
Additions to Chapter 6 of the OpenGL Core Profile Specification, Version 4.3,
"Buffer Objects"
Modify Section 6.2, "Creating and Modifying Buffer Object Data Stores",
p. 57 as follows:
The data store of a buffer object is created by calling
void BufferStorage(enum target,
sizeiptr size,
const void * data,
bitfield flags);
with <target> set to one of the targets listed in Table 6.1, <size> set to
the size of the data store in basic machine units and <flags> containing
a bit-field describing the intended usage of the data store. The data
store of the buffer object bound to <target> is allocated as a result of
a call to this function and cannot be de-allocated until the buffer is
deleted with a call to DeleteBuffers. Such a store may not be
re-allocated through further calls to BufferStorage or BufferData.
<data> specifies the address in client memory of the data that should
be used to initialize the buffer's data store. If <data> is NULL, the
data store of the buffer is created, but contains undefined data.
Otherwise, <data> should point to an array of at least <size> basic
machine units.
<flags> is the bitwise OR of flags describing the intended usage
of the buffer object's data store by the application. Valid flags and
their meanings are as follows:
DYNAMIC_STORAGE_BIT The contents of the data store may be
updated after creation through calls to BufferSubData. If this bit is not
set, the buffer content may not be directly updated by the client. The
<data> argument may be used to specify the initial content of the buffer's
data store regardless of the presence of the DYNAMIC_STORAGE_BIT.
Regardless of the presence of this bit, buffers may always be updated
with server-side calls such as CopyBufferSubData and ClearBufferSubData.
MAP_READ_BIT The buffer's data store may be mapped by the client for
read access and a pointer in the client's address space obtained that may
be read from.
MAP_WRITE_BIT The buffer's data store may be mapped by the client for
write access and a pointer in the client's address space obtained that may
be written to.
MAP_PERSISTENT_BIT The client may request that the server read from
or write to the buffer while it is mapped. The client's pointer to the
data store remains valid so long as the data store is mapped, even during
execution of drawing or dispatch commands.
MAP_COHERENT_BIT Shared access to buffers that are simultaneously
mapped for client access and are used by the server will be coherent, so
long as that mapping is performed using MapBufferRange. That is, data
written to the store by either the client or server will be immediately
visible to the other with no further action taken by the application. In
- If MAP_COHERENT_BIT is not set and the client performs a write
followed by a call to one of the FlushMapped*BufferRange commands
with a range including the written range, then in subsequent
commands the server will see the writes.
- If MAP_COHERENT_BIT is set and the client performs a write, then in
subsequent commands the server will see the writes.
- If MAP_COHERENT_BIT is not set and the server performs a write, the
application must call MemoryBarrier with the
CLIENT_MAPPED_BUFFER_BARRIER_BIT set and then call FenceSync with
SYNC_GPU_COMMANDS_COMPLETE (or Finish). Then the CPU will see the
writes after the sync is complete.
- If MAP_COHERENT_BIT is set and the server does a write, the app must
call FenceSync with SYNC_GPU_COMMANDS_COMPLETE (or Finish). Then the
CPU will see the writes after the sync is complete.
CLIENT_STORAGE_BIT When all other criteria for the buffer storage
allocation are met, this bit may be used by an implementation to determine
whether to use storage that is local to the server or to the client to
serve as the backing store for the buffer.
If <flags> contains MAP_PERSISTENT_BIT, it must also contain at least one
It is an error to specify MAP_COHERENT_BIT without also specifying
BufferStorage deletes any existing data store, and sets the values of
the buffer object's state variables as shown in table 6.3.
If any portion of the buffer object is mapped in the current context or
any context current to another thread, it is as though UnmapBuffer (see
section 6.3.1) is executed in each such context prior to deleting the
existing data store.
Name | Value for | Value for
| BufferData | BufferStorage
BUFFER_SIZE | <size> | <size>
Table 6.3: Buffer object state after calling BufferData or
A mutable data store may be allocated for a buffer object by calling
void BufferData(...)
<include the remainder of Section 6.2 as written, and then append>.
Calling BufferData is equivalent to calling BufferStorage with
<target>, <size> and <data> as specified, and <flags> set to the logical
use the value of <usage> parameter to BufferData as a hint to further
determine the intended use of the buffer. However, BufferStorage allocates
immutable storage whereas BufferData allocates mutable storage. Thus, when
a buffer's data store is allocated through a call to BufferData, the
buffer's BUFFER_IMMUTABLE_STORAGE flags is set to FALSE.
Add the following errors:
An INVALID_OPERATION error is generated by BufferData and BufferStorage
if the BUFFER_IMMUTABLE_STORAGE flag of the buffer bound to <target> is
set to TRUE.
An INVALID_OPERATION error is generated by BufferSubData if the
BUFFER_IMMUTABLE_STORAGE flag of the buffer bound to <target> is TRUE
and the value of BUFFER_STORAGE_FLAGS for the buffer does not have
The command:
void NamedBufferStorageEXT(uint buffer,
sizeiptr size,
const void * data,
bitfield flags);
behaves similarly to BufferStorage, except that the buffer whose storage
is to be defined is specified by <buffer> rather than by the current
binding to <target>.
Add the following error:
An INVALID_OPERATION error is generated by NamedBufferStorageEXT if
the BUFFER_IMMUTABLE_STORAGE flag of <buffer> is set to TRUE.
Append to Table 6.2, "Buffer object parameters and their values":
| | | Initial | Legal |
| Name | Type | Value | Values |
| BUFFER_STORAGE_FLAGS | int | 0 | See section 6.2 |
Append to Table 6.3, "Buffer object initial state":
+---------------------------+-------------------------------------- +
| Name | Value |
| BUFFER_IMMUTABLE_STORAGE | TRUE if the buffer's storage is |
| | immutable, FALSE otherwise |
Modify Section 6.3, "Mapping and Unmapping Buffer Data"
Add to the bulleted list describing flags that modify buffer mappings,
* MAP_PERSISTENT_BIT indicates that it is not an error for the GL to
read data from or write data to the buffer while it is mapped (see
section 6.3.2). If this bit is set, the value of
BUFFER_STORAGE_FLAGS for the buffer being mapped must include
* MAP_COHERENT_BIT indicates that the mapping should be performed
coherently. That is, such a mapping follows the rules set forth in
section 6.2, "Creating and Modifying Buffer Object Data Stores".
If the MAP_COHERENT_BIT is set and the buffer's BUFFER_STORAGE_FLAGS
does not include MAP_COHERENT_BIT, the error INVALID_OPERATION is
Modify Section 6.3.2, "Effects of Mapping Buffers on Other GL Commands"
to read:
An INVALID_OPERATION error is generated by most, but not all GL
commands when an attempt is detected by such a command to read data from
or write data to a mapped buffer object unless it was allocated with the
by a call to BufferStorage with MAP_PERSISTENT_BIT set in
Any command which does not detect these attempts, and performs such an
invalid read or write, has undefined results and may result in GL
interruption or termination.
Add the following to the description of FlushMappedBufferRange:
If a buffer range is mapped with both the MAP_PERSISTENT_BIT and
MAP_FLUSH_EXPLICIT_BIT set, then FlushMappedBufferRange may be called to
ensure that data written by the client into the flushed region becomes
visible to the server. Data written to a coherent store will always
become visible to the server after an unspecified period of time.
Modify Section 6.8, "Buffer Object State", p. 70:
Add the following required state to a buffer object:
..., a boolean indicating whether or not buffer storage is
immutable, an unsigned integer storing the flags with which it was
allocated, ...
Additions to Chapter 7 of the OpenGL Core Profile Specification, Version 4.3,
"Programs and Shaders"
Add to the list of flags accepted by the <barriers> parameter to
MemoryBarrier in Section 7.12.2, "Shader Memory Access Synchronization":
* CLIENT_MAPPED_BUFFER_BARRIER_BIT: Access by the client to persistent
mapped regions of buffer objects will reflect data written by shaders
prior to the barrier. Note that this may cause additional
synchronization operations.
New State
Append to Table 23.6, "Buffer Object State", p. 511:
| Get Value | Type | Get Command | Initial Value | Description | Sec. |
| BUFFER_IMMUTABLE_STORAGE | B | GetBufferParameteriv | FALSE | TRUE if buffer's data store is | 6 |
| | | | | immutable, FALSE otherwise | |
| BUFFER_STORAGE_FLAGS | Z+ | GetBufferParameteriv | 0 | The buffer object's storage | 6 |
| | | | | flags. | |
New Implementation Dependent State
INVALID_OPERATION is generated by BufferStorage if zero is bound to
INVALID_OPERATION is generated by BufferStorage, NamedBufferStorageEXT
and BufferData if the buffer object already owns an immutable data
INVALID_VALUE is generated by BufferStorage and NamedBufferStorageEXT
if <size> is less than or equal to zero.
INVALID_VALUE is generated by BufferStorage and NamedBufferStorageEXT if
<flags> contains MAP_PERSISTENT_BIT but does not contain
at least one of MAP_READ_BIT or
INVALID_VALUE is generated by BufferStorage and NamedBufferStorageEXT if
<flags> contains MAP_COHERENT_BIT, but does not also
INVALID_OPERATION is generated by MapBufferRange if any of MAP_READ_BIT,
<access>, but the same bit is not included in the buffer's storage
INVALID_OPERATION is generated by MapBufferRange if MAP_PERSISTENT_BIT
is included in <access> but MAP_PERSISTENT_BIT is not
included in the buffer's storage flags, or if MAP_COHERENT_BIT is included
in <access> but
OUT_OF_MEMORY is generated by BufferStorage and NamedBufferStorageEXT if
the GL is not able to allocate a data store with the properties requested
in <flags>.
*REMOVE* all errors generated by any command should they detect access to
a mapped buffer and replace with language such as:
INVALID_OPERATION is generated by <command> if the buffer is currently
mapped by MapBuffer{Range} unless it was mapped with the
MAP_PERSISTENT_BIT included in <access>.
Dependencies on GL_EXT_direct_state_access
If GL_EXT_direct_state_access is not supported, remove all references to
Conformance Tests
Usage Examples
Example 1: Updating the content of a buffer which does not have the
DYNAMIC flag set:
// Allocate two buffers, one of which will be our 'staging buffer'.
GLuint bufs[2];
glGenBuffers(2, &bufs[0]);
// Client can map this buffer for write.
// One could possibly make this mapping persistent.
glBindBuffer(GL_COPY_READ_BUFFER, bufs[0]);
glBufferStorage(GL_COPY_READ_BUFFER, size, NULL,
// Client cannot read or write this buffer, server can do both.
glBindBuffer(GL_COPY_WRITE_BUFFER, bufs[1]);
glBufferStorage(GL_COPY_WRITE_BUFFER, size, NULL, 0);
// Now, map the staging buffer to put data into it.
void * data = glMapBufferRange(GL_COPY_READ_BUFFER, 0, size,
memcpy(data, source_data, size);
// Copy from the staging buffer to the server-side buffer.
glCopyBufferSubData(GL_COPY_READ_BUFFER, GL_COPY_WRITE_BUFFER, 0, 0, size);
Example 2: Read from framebuffer into a buffer mapped into client's
address space:
// Create buffer, allocate storage, and create a persistent map.
GLuint pbo;
glGenBuffers(1, &pbo);
glBindBuffer(GL_PIXEL_PACK_BUFFER, pbo);
glBufferStorage(GL_PIXEL_PACK_BUFFER, size, NULL,
void * data = glMapBufferRange(GL_PIXEL_PACK_BUFFER,
glReadPixels(0, 0, width, height, format, type, NULL);
GLsync fence = glFenceSync(GL_SYNC_GPU_COMMANDS_COMPLETE, 0);
// Do stuff to use time...
// Use the data written to the buffer
1) What is the best strategy to allow 'render while mapped'? Options I
can think of right now are (a) Allow the application to render when a
buffer is mapped, so long as the MAP_PERSISTENT_BIT
was set when it was allocated; (b) Continue to disallow 'render while
mapped', but say that a client pointer obtained from MapBuffer{Range}
remains valid even when the buffer is not mapped, so long as it has
not been destroyed; (c) add a flag to glMapBufferRange's <access>
parameter to indicate the desire to render with it.
RESOLVED: We choose a combination of (a) and (c). The application must
both create the data store with MAP_PERSISTENT_BIT set
_and_ map it with MAP_PERSISTENT_BIT set in <access>. Did the same
for coherency too.
2) The new flags don't directly map to the <usage> parameter for
glBufferData and one cannot be expressed in terms of the other. Does
that matter?
RESOLVED: Most applications get <usage> wrong and they're only hints
anyway. The flags are hard and fast rules that must be followed. They
serve a different purpose. The idea here is to allow the
implementation to not have to second guess the application and to
perform less tracking, and for the application to have more control.
We define BufferData in terms of BufferStorage with the most liberal
allowed flags (essentially, anything goes), but still pass the
<usage> hint to the implementation to allow it to continue to second
guess the application.
3) Do we have all the flags we want? Are any problematic?
RESOLVED: We don't want any more flags. We don't believe any are
4) Should we include MULTI_TARGET_BIT? There are legitimate use cases where
a buffer could be used on two or three targets. However, this bit is an
'all or nothing' kind of thing.
RESOLVED: No, not at this time.
5) How do you get data into a non-dynamic buffer if you can't write to it
from the client?
RESOLVED: The server is capable of writing to buffers that were not
allocated with the DYNAMIC flag set. Therefore, it is possible to
use CopyBufferSubData to copy from a dynamic buffer to a non-dynamic
buffer. It's also possible to write to it with any other server-side
mechanism such as transform feedback, image stores and so on.
6) If a buffer is allocated without the GL_BUFFER_STORAGE_SERVER_READ_BIT
(or GL_BUFFER_STORAGE_SERVER_WRITE_BIT), what happens if an attempt is
made use the buffer in a way that may cause the server to read
(or write) to the buffer?
RESOLVED: Nuked the SERVER_READ and SERVER_WRITE bits. They didn't
serve the purpose for which they were intended.
7) Which operations are able to update buffers that are not dynamic?
Non-dynamic buffers effectively don't allow direct transfer of data
from client to server (i.e., glBufferSubData). Examples of operations
that may write to non-dynamic buffers are transform feedback, image
stores, ReadPixels, GetTexImage (PBO), CopyBufferSubData,
ClearBufferSubData - essentially anything that doesn't transfer
arbitrary amounts of data from client to server.
8) Are there any restrictions on calling GetBufferSubData on a buffer
allocated using BufferStorage?
RESOLVED: No, there are not.
9) What is the meaning of CLIENT_STORAGE_BIT? Is it one of those
silly hint things?
DISCUSSION: Unfortunately, yes, it is. For some platforms, such as UMA
systems, it's irrelevant and all memory is both server and client
accessible. The issue is, that on some platforms and for certain
combinations of flags, there may be multiple regions of memory that
can satisfy the request (visible to both server and client and coherent
to both, for example), but may have substantially different performance
characteristics for access from either. This bit essentially serves
as a hint to say that that an application will access the store more
frequently from the client than from the server. In practice,
applications will still get it wrong (like setting it all the time or
never setting it at all, for example), implementations will still have
to second guess applications and end up full of heuristics to figure out
where to put data and gobs of code to move things around based on what
applications do, and eventually it'll make no difference whether
applications set it or not. But hey, we tried.
10) Do we want to add flags for MapBufferRange for PERSISTENT and/or
COHERENT mapping? In their absence, implementations must assume that
any mapping performed on a buffer whose storage flags include the
PERSISTENT or COHERENT flags must behave appropriately.
11) Do we need language to explicitly say that flushes of non-coherent
mapped buffers need to occur on buffers mapped with the FLUSH_EXPLICIT
RESOLVED: No. The language already states that FlushMappedBufferRange
should be used to perform the flush, and this command requires that
the mapping be established with the FLUSH_EXPLICIT bit set.
12) Which functions can/cannot be used to update the content of a
non-DYNAMIC buffer? Can the buffer be the target of an update
operation at all?
RESOLVED: BufferSubData is only allowed for DYNAMIC buffers. Updates
through mappings are allowed so long as the STORAGE_MAP_WRITE_BIT is
set. Server side commands, including CopyBufferSubData,
ClearBufferSubData, ReadPixels, GetTexImage are allowed. Further,
shader writes such as image stores, SSBO, atomic counters, transform
feedback and so on are also allowed.
13) Why is there a gap between the MAP_WRITE_BIT and MAP_PERSISTENT_BIT
token values?
the bitfield used for MapBufferRange, which include values that
aren't relevant for BufferStorage. This allows the same tokens
to be used as flags for BufferStorage and MapBufferRange, hopefully
reducing confusion.
Revision History
Rev. Date Author Changes
---- -------- -------- -----------------------------------------
1 01/16/2013 gsellers Initial draft
2 01/21/2013 gsellers Updates
3 01/22/2013 gsellers Change static->dynamic. Remove target
restrictions. Get a little closer to expressing
BufferData in terms of BufferStorage.
4 04/22/2013 gsellers Add CLIENT_MAPPED_BUFFER_ACCESS_BIT for
Add example usage for non-dynamic buffers.
Add issue 7.
5 04/23/2013 gsellers Nuked the BUFFER_STORAGE_{READ|WRITE}_BIT
flags again.
Add another example.
Update values of tokens.
Add (and resolve) issue 8.
6 05/14/2013 gsellers Add BUFFER_STORAGE_SERVER_BIT.
Define value of <usage> for buffers allocated
with BufferStorage. Issue 9.
7 05/22/2013 gsellers Address several issues from bug 10246.
8 05/23/2013 gsellers Address issues from bug 10288.
CLIENT_STORAGE_BIT (inverting its
sense), which makes 0 'fast'.
* Clarify that DYNAMIC_BIT only affects
BufferSubData (i.e., direct, arbitrary
client->server transfers).
* Add issues 11 + 12.
9 05/28/2013 Jon Leech Fix various typos resulting from changes
in token names, tweak language to match API
spec, some paragraph reflowing, insert some
questions marked by '**' inline.
10 05/29/2013 gsellers Remove <target> parameter from
Incorporate new rules for coherency.
Add COHERENT_MAP_BIT for MapBufferRange.
11 05/30/2013 Jon Leech Fix typos including COHERENT_MAP_BIT
12 05/30/2013 gsellers Resolve issues 3 and 10. Fix typos.
Resolve issues from bug 10326.
Add (and resolve) issue 13.
13 05/30/2013 gsellers Change names of flags (again).
Use same values for MapBufferRange flags
and BufferStorage flags.
14 05/30/2013 Jon Leech Clean up language describing flags and
some indentation issues.
15 05/31/2013 Jon Leech Add BUFFER_IMMUTABLE_STORAGE to table
6.2 (Bug 10288).
16 06/06/2013 Jon Leech Change default BUFFER_IMMUTABLE_STORAGE
value in table 6.2 to FALSE, matching API
spec, since these are values when created
with BindBuffer. Fix typo from bug 10326.
17 06/27/2013 Jon Leech Add error for BufferSubData and fix
example code (Bug 10326)
18 07/03/2013 gsellers Fix language describing DYNAMIC_STORAGE_BIT
(mutated -> updated), and typo in description
of usage parameter when storage is allocated
with BufferStorage. (Bug 10471)
19 07/18/2013 gsellers Added missing values for MAP_PERSISTENT_BIT
20 07/18/2013 Jon Leech Add BufferStorage initial state to table
6.3 and add error when zero is bound to
<target> (Bug 10335).
21 07/19/2013 Jon Leech Clean up table 6.3 captions to match
API spec (Bug 10335).
22 08/15/2013 Jon Leech Remove error for BufferStorage and
NamedBufferStorageEXT if <flags> contains
MAP_WRITE_BIT but does not contain
DYNAMIC_STORAGE_BIT (Bug 10561, public Bug
23 08/16/2013 mjk Better indicate DSA entrypoints
24 06/09/2014 Jon Leech Change query commands for buffer storage
state to GetBufferParameteriv (Bug 12307).
25 04/20/2015 Jon Leech Change description of MAP_COHERENT_BIT for
buffer storage so that barriers with
to make CPU writes visible to the GPU in
this case without an explicit flush (Bug