- User Since
- Jan 12 2016, 6:13 AM (255 w, 13 h)
May 6 2018
Oct 8 2017
This should be it.
Sep 6 2017
Sep 5 2017
@jvesely I would like to get an ACK before I tackle the tests.
Apr 19 2017
That was only a partial diff, should be correct now.
This should be it.
Apr 18 2017
Sorry for the delay, updated patch, now passes tests, addressed all the comments. I would appreciate if you could check if I got all the R_ and A_ prefixes correctly (it's just variable naming, but there is some logic behind it -- register vs address, I presume?) in f16 tests.
Mar 9 2017
TODO: TESTS NEED TO BE UPDATED
Feb 22 2017
@arsenm Can you eventually push this into 4.0 branch as well?
Feb 18 2017
Feb 15 2017
Better version: handle log10 as well, define constants locally, handle f16, do not pretend to handle f64. Tests incoming.
Feb 14 2017
Feb 9 2017
To the extent of my knowledge and testing, this looks good.
Whoa, this seems to fix the remaining part of https://bugs.freedesktop.org/show_bug.cgi?id=99552
Jan 25 2017
Jan 22 2017
Updated per @arsenm's comment.
Jan 20 2017
Addressed all the issues mentioned. Please review.
Jan 18 2017
Better approach. I have hardcode the processor names in AMDGPUSubtarget::initializeSubtargetDependencies() because getGeneration() will return SI before ParseSubtargetFeatures() finishes, and I have to provide Feature String for it to execute.
Jan 17 2017
Dec 22 2016
You just wait until a maintainer picks it up.
Dec 16 2016
LGTM as well
Dec 12 2016
Jul 21 2016
Addressed both concerns.
Jul 18 2016
Specifically handle only kernels.
May 29 2016
Updated patch. Single element structs are coerced to its element, and there are tests for structs of different sizes, structs of arrays, structs containing structs. Arrays of structs are disallowed by clang in kernels.
Non-kernel functions are not specifically handled, should they be? How to decide?
May 27 2016
@Anastasia I looked into introducing a separate getOpenCLVersion() function (or perhaps three - major version, minor version and version string). This would have to be used in lib/CodeGen/TargetInfo.cpp and lib/Parse/ParseDecl.cpp, and I am undecided on where should one put this code. One option would be in Parse/Parser.h inside class Parser, and then TargetInfo.cpp would have to include Parser.h, unless we decide to declare it inside AST/ASTContext.h.
May 14 2016
May 12 2016
Now with 100% more tests.
May 11 2016
Could we solve that at a later point? There is one more place where such code is already used, but this would enlarge the scope of this patch.
Please, can anyone push this?
May 9 2016
Thanks for the reviews!
May 6 2016
Make that int const as well.
I am neither aware how to convert ints to StringRef nor how to concatenate StringRefs. Apologies if I missed something in the API.
@Anastasia would you still prefer to make VerSpec a std::string?
Replace Twine with std::string and use llvm::to_string().
May 3 2016
May 1 2016
Fixed wrong search and replace in tests.
Apr 27 2016
GCC 6.1 (GCC 6 final release in GCC's funky release numbering scheme) does not have this issue anymore: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70528
Apr 4 2016
Apr 2 2016
Is GCC 6 being stricter than older versions? Is this behavior standard-compliant?
Mar 31 2016
LGTM and 32-bit version of the instruction tested on Tonga, works.
Mar 29 2016
Works now. Since there were no changes to MUBUF_Atomic multiclass, this addresses the concerns by @nhaehnle.
Merged Jan's approach. It would be nice if this works, because it's a much smaller diff.
Mar 14 2016
Mar 10 2016
Updated per comments by @arsenm.
Mar 8 2016
@rnk can you re-review?
Mar 2 2016
Sorry for the delay, I only read "ehsan accepted this revision" and did not notice there were comments.
Feb 25 2016
@arsenm is this mergeable? Any other changes I should make?
Posting updated version.
Feb 24 2016
Yes, I will post the updated patch per arsenm's suggestions tomorrow and one with tests over the weekend. I didn't find time to write the tests, sorry.
Feb 22 2016
rnk, ehsan, any comments? Anything else I should do?
Feb 18 2016
Is there anything else I need to do for this to land?
Feb 17 2016
Handle Windows first, output full path. Elsewhere, output -l and library name.
Feb 16 2016
It does not do anything special on any OS, but I guess it should. I'm happy to adjust it for Windows if you give me the specification.
Fixed all comments except intrinsics, will upload a new revision with tests and without intrinsics chages when I have tests ready.
Feb 15 2016
Please ignore changes in lib/Target/AMDGPU/SIIntrinsics.td, that's unrelated and was uploaded by accident. Will fix on reupload.
Feb 11 2016
Mark as done.
Updated per arsenm's comments.
Feb 10 2016
Cool, I can mark comments as done.
I see. Thanks for the info.
Updated per Matt's suggestions.
Feb 9 2016
Jan 12 2016
Works for me. Needs rebase to current master.