Sahasrabuddhe, Sameer
2014-Nov-14 18:17 UTC
[LLVMdev] memory scopes in atomic instructions
<html>
<head>
<meta http-equiv="content-type" content="text/html;
charset=utf-8">
</head>
<body bgcolor="#FFFFFF" text="#000000">
Hi all,<br>
<br>
OpenCL 2.0 introduced the notion of memory scope in atomic
operations to global memory. These scopes are a hint to the
underlying platform to optimize how synchronization is achieved.
HSAIL also has a notion of memory scopes that is compatible with
OpenCL 2.0. Currently, the LLVM IR uses a binary value
(SingleThread/CrossThread) to represent synchronization scope on
atomic instructions. This makes it difficult to translate OpenCL 2.0
atomic operations to LLVM IR, and also to implement HSAIL memory
scopes in the proposed HSAIL backend for LLVM.<br>
<br>
We would like to enhance the representation of memory scopes in LLVM
IR to allow more values than just the current two. The intention of
this email is to invite comments before we start prototyping. Here's
what we have in mind:<br>
<ol>
<li>Update the synchronization scope field in atomic instructions
from a single bit to a wider field, say 32-bit unsigned integer.
</li>
<li>Retain the current default of zero as "system scope",
replacing the current "cross thread" scope.<br>
</li>
<li>All other values are target-defined.</li>
<li>The use of "single thread scope" is not clear. If it
is
required in target-independent transforms, then it could be
encoded as just "1", or as "all ones" in the wider
field. The
latter option is a bit weird, because most targets will have
very few scopes. But it is useful in case the next point is
included in LLVM IR.</li>
<li>Possibly add the following constraint on memory scopes:
"The
scope represented by a larger value is nested inside (is a
proper subset of) the scope represented by a smaller value."
This would also imply that the value used for single-thread
scope must be the largest value used by the target.<br>
This constraint on "nesting" is easily satisfied by HSAIL (and
also OpenCL), where synchronization scopes increase from a
single work-item to the entire system. But it is conceivable
that other targets do not have this constraint. For example, a
platform may define synchronization scopes in terms of
overlapping sets instead of proper subsets. <br>
</li>
<li>The impact of this change is limited to increasing the number
of bits used to store synchronization scope. Future
optimizations on atomics may need to interpret scopes in
target-defined ways. When the synchronization scopes of two
atomic instructions do not match, these optimizations must query
the target for validity. <br>
</li>
</ol>
<b>Relation with SPIR: </b>SPIR defines an enumeration for
memory
scopes, but it does not support LLVM atomic instructions. So memory
scopes in SPIR are independent of the representation finally chosen
in LLVM IR. A compiler that translates SPIR to native LLVM IR will
have to translate memory scopes wherever appropriate. <br>
<br>
Sameer.<br>
</body>
</html>
On Fri, Nov 14, 2014 at 11:47:45PM +0530, Sahasrabuddhe, Sameer wrote: Can you send a plain-text version of this email. It's easier to read and reply to. -Tom> _______________________________________________ > LLVM Developers mailing list > LLVMdev at cs.uiuc.edu http://llvm.cs.uiuc.edu > http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
Hi Sameer, I support this proposal, and have discussed more or less the same idea with various people in the past. Regarding point #5, I believe address spaces may already provide the functionality needed to express overlapping constraints. I’m also not aware of any systems that would really need that functionality anyways. —Owen> On Nov 14, 2014, at 10:17 AM, Sahasrabuddhe, Sameer <Sameer.Sahasrabuddhe at amd.com> wrote: > > Hi all, > > OpenCL 2.0 introduced the notion of memory scope in atomic operations to global memory. These scopes are a hint to the underlying platform to optimize how synchronization is achieved. HSAIL also has a notion of memory scopes that is compatible with OpenCL 2.0. Currently, the LLVM IR uses a binary value (SingleThread/CrossThread) to represent synchronization scope on atomic instructions. This makes it difficult to translate OpenCL 2.0 atomic operations to LLVM IR, and also to implement HSAIL memory scopes in the proposed HSAIL backend for LLVM. > > We would like to enhance the representation of memory scopes in LLVM IR to allow more values than just the current two. The intention of this email is to invite comments before we start prototyping. Here's what we have in mind: > Update the synchronization scope field in atomic instructions from a single bit to a wider field, say 32-bit unsigned integer. > Retain the current default of zero as "system scope", replacing the current "cross thread" scope. > All other values are target-defined. > The use of "single thread scope" is not clear. If it is required in target-independent transforms, then it could be encoded as just "1", or as "all ones" in the wider field. The latter option is a bit weird, because most targets will have very few scopes. But it is useful in case the next point is included in LLVM IR. > Possibly add the following constraint on memory scopes: "The scope represented by a larger value is nested inside (is a proper subset of) the scope represented by a smaller value." This would also imply that the value used for single-thread scope must be the largest value used by the target. > This constraint on "nesting" is easily satisfied by HSAIL (and also OpenCL), where synchronization scopes increase from a single work-item to the entire system. But it is conceivable that other targets do not have this constraint. For example, a platform may define synchronization scopes in terms of overlapping sets instead of proper subsets. > The impact of this change is limited to increasing the number of bits used to store synchronization scope. Future optimizations on atomics may need to interpret scopes in target-defined ways. When the synchronization scopes of two atomic instructions do not match, these optimizations must query the target for validity. > Relation with SPIR: SPIR defines an enumeration for memory scopes, but it does not support LLVM atomic instructions. So memory scopes in SPIR are independent of the representation finally chosen in LLVM IR. A compiler that translates SPIR to native LLVM IR will have to translate memory scopes wherever appropriate. > > Sameer. > _______________________________________________ > LLVM Developers mailing list > LLVMdev at cs.uiuc.edu http://llvm.cs.uiuc.edu > http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev-------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20141114/4816418e/attachment.html>
Sahasrabuddhe, Sameer
2014-Nov-14 19:09 UTC
[LLVMdev] memory scopes in atomic instructions
On 11/15/2014 12:08 AM, Tom Stellard wrote:> Can you send a plain-text version of this email. It's easier to read > and reply to.Sorry about that! Here's the plain text (I hope!): Hi all, OpenCL 2.0 introduced the notion of memory scope in atomic operations to global memory. These scopes are a hint to the underlying platform to optimize how synchronization is achieved. HSAIL also has a notion of memory scopes that is compatible with OpenCL 2.0. Currently, the LLVM IR uses a binary value (SingleThread/CrossThread) to represent synchronization scope on atomic instructions. This makes it difficult to translate OpenCL 2.0 atomic operations to LLVM IR, and also to implement HSAIL memory scopes in the proposed HSAIL backend for LLVM. We would like to enhance the representation of memory scopes in LLVM IR to allow more values than just the current two. The intention of this email is to invite comments before we start prototyping. Here's what we have in mind: 1. Update the synchronization scope field in atomic instructions from a single bit to a wider field, say 32-bit unsigned integer. 2. Retain the current default of zero as "system scope", replacing the current "cross thread" scope. 3. All other values are target-defined. 4. The use of "single thread scope" is not clear. If it is required in target-independent transforms, then it could be encoded as just "1", or as "all ones" in the wider field. The latter option is a bit weird, because most targets will have very few scopes. But it is useful in case the next point is included in LLVM IR. 5. Possibly add the following constraint on memory scopes: "The scope represented by a larger value is nested inside (is a proper subset of) the scope represented by a smaller value." This would also imply that the value used for single-thread scope must be the largest value used by the target. This constraint on "nesting" is easily satisfied by HSAIL (and also OpenCL), where synchronization scopes increase from a single work-item to the entire system. But it is conceivable that other targets do not have this constraint. For example, a platform may define synchronization scopes in terms of overlapping sets instead of proper subsets. 6. The impact of this change is limited to increasing the number of bits used to store synchronization scope. Future optimizations on atomics may need to interpret scopes in target-defined ways. When the synchronization scopes of two atomic instructions do not match, these optimizations must query the target for validity. *Relation with SPIR: *SPIR defines an enumeration for memory scopes, but it does not support LLVM atomic instructions. So memory scopes in SPIR are independent of the representation finally chosen in LLVM IR. A compiler that translates SPIR to native LLVM IR will have to translate memory scopes wherever appropriate. Sameer.
Sahasrabuddhe, Sameer
2014-Nov-17 05:03 UTC
[LLVMdev] memory scopes in atomic instructions
Copying #5 here for reference: > 5. Possibly add the following constraint on memory scopes: "The scope > represented by a larger value is nested inside (is a proper subset > of) the scope represented by a smaller value." This would also imply > that the value used for single-thread scope must be the largest > value used by the target. > This constraint on "nesting" is easily satisfied by HSAIL (and also > OpenCL), where synchronization scopes increase from a single > work-item to the entire system. But it is conceivable that other > targets do not have this constraint. For example, a platform may > define synchronization scopes in terms of overlapping sets instead > of proper subsets. On 11/15/2014 12:20 AM, Owen Anderson wrote:> I support this proposal, and have discussed more or less the same idea > with various people in the past.Thanks! Good to know that.> Regarding point #5, I believe address spaces may already provide the > functionality needed to express overlapping constraints.I am not sure how address spaces can solve the need for such a constraint. Memory scopes are orthogonal to address spaces. An agent that atomically accesses a shared location in one address space may specify a different memory scope on different instructions.> I’m also not aware of any systems that would really need that > functionality anyways.Agreed that there are no known systems with overlapping memory scopes, but it might premature to just dismiss the possibility. For example, there could be a system with a fixed number of agents, say four. Each agent can choose to synchronize with one, two or all three peers on different instructions. The resulting memory scopes cannot be ordered as proper subsets. Sameer.
Possibly Parallel Threads
- [LLVMdev] memory scopes in atomic instructions
- [LLVMdev] memory scopes in atomic instructions
- [LLVMdev] memory scopes in atomic instructions
- [LLVMdev] [RFC][PATCH][OPENCL] synchronization scopes redux
- [LLVMdev] [RFC][PATCH][OPENCL] synchronization scopes redux