blob: ad078cdce59f7627e973c18adf10e0155c68ca58 [file] [log] [blame]
Name Strings
Timothy Lottes (timothy.lottes 'at'
Timothy Lottes, AMD
Graham Sellers, AMD
Daniel Rakos, AMD
Jeannot Breton, NVIDIA
Pat Brown, NVIDIA
Eric Werness, NVIDIA
Mark Kilgard, NVIDIA
Jeff Bolz, NVIDIA
Copyright (c) 2015 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 26, 2015.
Ratified by the Khronos Board of Promoters on August 7, 2015.
Last Modified Date: 03/18/2017
Revision: 8
ARB Extension #183
This extension is written against Revision 5 of the version 4.50 of the
OpenGL Shading Language Specification, dated January 30, 2015.
This extension requires GL_ARB_gpu_shader_int64.
This extension provides the ability for a group of invocations which
execute in lockstep to do limited forms of cross-invocation communication
via a group broadcast of a invocation value, or broadcast of a bitarray
representing a predicate value from each invocation in the group.
New Procedures and Functions
New Tokens
IP Status
Modifications to the OpenGL Shading Language Specification, Version 4.50
Including the following line in a shader can be used to control the
language features described in this extension:
#extension GL_ARB_shader_ballot : <behavior>
where <behavior> is as specified in section 3.3.
New preprocessor #defines are added to the OpenGL Shading Language:
#define GL_ARB_shader_ballot 1
Additions to Chapter 7 of the OpenGL Shading Language Specification
(Built-in Variables)
Modify Section 7.4, Built-In Uniform State, p. 133
(Add to the list of built-in uniform variable declaration)
uniform uint gl_SubGroupSizeARB;
(Add this paragraph at the end of this section)
A sub-group is a collection of invocations which execute in lockstep.
The variable <gl_SubGroupSizeARB> is the maximum number of invocations
in a sub-group. The maximum <gl_SubGroupSizeARB> supported in this
extension is 64.
Modify Section 7.1, Built-in Languages Variable, p. 110
(Add to the list of built-in variables for the compute, vertex, geometry,
tessellation control, tessellation evaluation and fragment languages)
in uint gl_SubGroupInvocationARB;
in uint64_t gl_SubGroupEqMaskARB;
in uint64_t gl_SubGroupGeMaskARB;
in uint64_t gl_SubGroupGtMaskARB;
in uint64_t gl_SubGroupLeMaskARB;
in uint64_t gl_SubGroupLtMaskARB;
(Add those paragraphs at the end of this section)
The variable <gl_SubGroupInvocationARB> holds the index of the invocation within
sub-group. This variable is in the range 0 to <gl_SubGroupSizeARB>-1, where
<gl_SubGroupSizeARB> is the total number of invocations in a sub-group.
The <gl_SubGroup??MaskARB> variables provide a bitmask for all invocations,
with one bit per invocation starting with the least significant bit,
according to the following table,
variable equation for bit values
-------------------- ------------------------------------
gl_SubGroupEqMaskARB bit index == gl_SubGroupInvocationARB
gl_SubGroupGeMaskARB bit index >= gl_SubGroupInvocationARB
gl_SubGroupGtMaskARB bit index > gl_SubGroupInvocationARB
gl_SubGroupLeMaskARB bit index <= gl_SubGroupInvocationARB
gl_SubGroupLtMaskARB bit index < gl_SubGroupInvocationARB
Additions to Chapter 8 of the OpenGL Shading Language Specification
(Built-in Functions)
Add Section 8.18, Shader Invocation Group Functions
uint64_t ballotARB(bool value);
The function ballotARB() returns a bitfield containing the result of
evaluating the expression <value> in all active invocations in the
sub-group. An active invocation is one that is executing the ballotARB()
call. The sub-group may have inactive invocations for example due to
exit of the shader, or divergent branching. Sub-groups of up to 64
invocations may be represented by the return value of ballotARB(). Bits
for each invocation are packed in least significant bit ordering. If
<value> evaluates to true for an active invocation then the corresponding
bit is set to one in the result, otherwise it is zero. Bits corresponding
to invocations that are not active or that do not exist in the sub group
(because, for example, they are at bit positions beyond the sub-group
size) are set to zero. The following trivial assumptions can be made:
* ballotARB(true) returns bitfield where the corresponding bits are
set for all active invocations in the sub-group.
* ballotARB(false) returns zero.
genType readInvocationARB(genType value, uint invocationIndex);
genIType readInvocationARB(genIType value, uint invocationIndex);
genUType readInvocationARB(genUType value, uint invocationIndex);
genType readFirstInvocationARB(genType value);
genIType readFirstInvocationARB(genIType value);
genUType readFirstInvocationARB(genUType value);
The function readInvocationARB() returns the <value> from a given
<invocationIndex> to all active invocations in the sub-group.
The <invocationIndex> must be the same for all active invocations
in the sub-group otherwise results are undefined.
The function readFirstInvocationARB() returns the <value> from the first
active invocation to all active invocations in the sub-group.
1) How are the values of gl_SubGroup??MaskARB defined?
RESOLVED. Earlier versions of this specification defined a bitmask
such as "LtMask" ("less than mask") as having bits set if
"gl_SubGroupInvocationARB < bit index". However, this was reversed
from the definition in GL_NV_shader_thread_group that these built-ins
were derived from, and also mismatched a recent Vulkan/SPIR-V extension.
Fortunately, all known implementations of this extension had implemented
"wrong" behavior (matching the sense of the original built-ins in
GL_NV_shader_thread_group), so the best thing to do is change the
definition in the spec.
Revision History
Rev Date Author Changes
--- ---------- -------- ---------------------------------------------
8 03/18/2017 jbolz Reversed the sense of the comparison in the
definition of gl_SubGroup??MaskARB.
7 08/25/2015 nhenning Add ARB suffix on documentation for
readInvocation and readFirstInvocation
6 07/31/2015 pdaniell Add ARB suffix on the readInvocation and
readFirstInvocation functions.
5 07/30/2015 pdaniell Update the function definition syntax to use
our standard gen*Type conventions.
4 06/23/2015 tlottes More precise spec language.
3 06/22/2015 tlottes Deferred GPU processor another spec.
Cleaned up spec language.
2 04/20/2015 tlottes Updated spec language.
1 03/09/2015 tlottes Initial revision based on AMD_gcn_shader and