Page MenuHomePhabricator

jdoerfert (Johannes Doerfert)
Argonne National Laboratory

Projects

User does not belong to any projects.

User Details

User Since
Jan 30 2014, 6:27 AM (416 w, 9 h)

Recent Activity

Today

jdoerfert added inline comments to D117806: [OpenMP] Change default visibility to protected for device declarations.
Thu, Jan 20, 12:34 PM · Restricted Project
jdoerfert added a comment to D117806: [OpenMP] Change default visibility to protected for device declarations.

Does this still work with AMDGPU as we need it to?

Thu, Jan 20, 12:19 PM · Restricted Project
jdoerfert added inline comments to D113119: [OpenMP] Introduce asynchronous malloc and free.
Thu, Jan 20, 10:33 AM · Restricted Project
jdoerfert accepted D117777: [OpenMP] Don't pass empty files to nvlink.

LG

Thu, Jan 20, 8:14 AM · Restricted Project
jdoerfert added a comment to D117777: [OpenMP] Don't pass empty files to nvlink.

test?

Thu, Jan 20, 7:03 AM · Restricted Project

Yesterday

jdoerfert added a comment to D117694: [UpdateTestChecks] Add --filter and --filter-out options.

I like this. No real concern, maybe someone else want's to take a look, otherwise I can accept this.

Wed, Jan 19, 9:46 PM · Restricted Project
jdoerfert committed rG6f2ee1ca5e39: [OpenMP][AMDGPU] Optimize the linked in math libraries (authored by jdoerfert).
[OpenMP][AMDGPU] Optimize the linked in math libraries
Wed, Jan 19, 9:36 PM
jdoerfert closed D116906: [OpenMP][AMDGPU] Optimize the linked in math libraries.
Wed, Jan 19, 9:36 PM · Restricted Project
jdoerfert committed rGb0789a1b12cd: [OpenMP] Avoid costly shadow map traversals whenever possible (authored by jdoerfert).
[OpenMP] Avoid costly shadow map traversals whenever possible
Wed, Jan 19, 8:15 PM
jdoerfert committed rG1e447d03e2f6: [OpenMP] Introduce an environment variable to disable atomic map clauses (authored by jdoerfert).
[OpenMP] Introduce an environment variable to disable atomic map clauses
Wed, Jan 19, 8:15 PM
jdoerfert closed D113124: [OpenMP] Avoid costly shadow map traversals whenever possible.
Wed, Jan 19, 8:15 PM · Restricted Project
jdoerfert closed D117627: [OpenMP] Introduce an environment variable to disable atomic map clauses.
Wed, Jan 19, 8:15 PM · Restricted Project
jdoerfert committed rGdd75a6b2ae5c: [DWARF][FIX] Try not to crash for nvptx with missing debug information (authored by jdoerfert).
[DWARF][FIX] Try not to crash for nvptx with missing debug information
Wed, Jan 19, 4:41 PM
jdoerfert committed rGb4a7559844d0: [OpenMP][FIX] Replace ICVs only with values valid at the getter position (authored by jdoerfert).
[OpenMP][FIX] Replace ICVs only with values valid at the getter position
Wed, Jan 19, 4:41 PM
jdoerfert closed D116959: [DWARF] Try not to crash for codes with missing debug information.
Wed, Jan 19, 4:40 PM · Restricted Project
jdoerfert updated the diff for D116959: [DWARF] Try not to crash for codes with missing debug information.

Add run line

Wed, Jan 19, 2:34 PM · Restricted Project
jdoerfert updated the diff for D113119: [OpenMP] Introduce asynchronous malloc and free.

Fix issues, sync w/o pending work

Wed, Jan 19, 11:31 AM · Restricted Project
jdoerfert added a comment to D116959: [DWARF] Try not to crash for codes with missing debug information.

From the earlier comments I'm assuming this is good to go. I'll commit stuff later and intend to include this if I don't hear anything to the contrary

Yep, sounds good to me. The DebugLoc on the load instruction is necessary, I take it? (I guess it could go on a ret instruction and have the @test function be void()?) That might point to another solution/direction - something is checking for debug locations, and maybe that something should be the one checking for DISubprogram on the function... *looks through the code* I don't immediately see where the existence of a !dbg on an instruction makes a difference here. (but simplified down the original godbolt to this: https://godbolt.org/z/dzGhPP51x and definitely removing the !dbg attachment for the instruction avoids the bug... )

Wed, Jan 19, 11:08 AM · Restricted Project
jdoerfert added a comment to D116959: [DWARF] Try not to crash for codes with missing debug information.

From the earlier comments I'm assuming this is good to go. I'll commit stuff later and intend to include this if I don't hear anything to the contrary

Wed, Jan 19, 10:40 AM · Restricted Project
jdoerfert updated the diff for D116959: [DWARF] Try not to crash for codes with missing debug information.

Address comments, add test

Wed, Jan 19, 10:39 AM · Restricted Project
jdoerfert updated the diff for D113119: [OpenMP] Introduce asynchronous malloc and free.

Ensure atomicity (via events) of async allocations

Wed, Jan 19, 10:00 AM · Restricted Project
jdoerfert updated the diff for D113124: [OpenMP] Avoid costly shadow map traversals whenever possible.

Fix errors, uninitialized iterator and wrong TPR (obj was used, ptr is needed).

Wed, Jan 19, 9:45 AM · Restricted Project
jdoerfert added a comment to D116959: [DWARF] Try not to crash for codes with missing debug information.

Finally reduced the reproducer: https://godbolt.org/z/dzKf9devd

Wed, Jan 19, 9:03 AM · Restricted Project
jdoerfert accepted D117634: [OpenMP] Expand short verisions of OpenMP offloading triples.

LG

Wed, Jan 19, 8:26 AM · Restricted Project, Restricted Project, Restricted Project
jdoerfert updated the diff for D117627: [OpenMP] Introduce an environment variable to disable atomic map clauses.

Typo

Wed, Jan 19, 8:11 AM · Restricted Project
jdoerfert added inline comments to D117627: [OpenMP] Introduce an environment variable to disable atomic map clauses.
Wed, Jan 19, 8:10 AM · Restricted Project
jdoerfert updated the diff for D117627: [OpenMP] Introduce an environment variable to disable atomic map clauses.

Address comments

Wed, Jan 19, 8:07 AM · Restricted Project
jdoerfert added a comment to D117634: [OpenMP] Expand short verisions of OpenMP offloading triples.

Can we have a test? Similar to a/the test introduced with the libdevice handling.

Wed, Jan 19, 7:58 AM · Restricted Project, Restricted Project, Restricted Project

Tue, Jan 18

jdoerfert added inline comments to D117627: [OpenMP] Introduce an environment variable to disable atomic map clauses.
Tue, Jan 18, 5:52 PM · Restricted Project
jdoerfert added inline comments to D117627: [OpenMP] Introduce an environment variable to disable atomic map clauses.
Tue, Jan 18, 5:37 PM · Restricted Project
jdoerfert updated the diff for D117627: [OpenMP] Introduce an environment variable to disable atomic map clauses.

Address comments, use lock guard

Tue, Jan 18, 5:37 PM · Restricted Project
jdoerfert requested review of D117627: [OpenMP] Introduce an environment variable to disable atomic map clauses.
Tue, Jan 18, 5:02 PM · Restricted Project
jdoerfert added inline comments to D113359: [Libomptarget][WIP] Introduce VGPU Plugin.
Tue, Jan 18, 12:50 PM · Restricted Project, Restricted Project, Restricted Project
jdoerfert accepted D117203: [Attributor] Remove notional overindexing check.

LG, see comment

Tue, Jan 18, 9:57 AM · Restricted Project
jdoerfert accepted D117571: [LangRef] Clarify that inaccessiblememonly functions are allowed noalias returns.

LG

Tue, Jan 18, 9:53 AM · Restricted Project
jdoerfert accepted D117362: [OpenMP] Remove hidden visibility for declare target variables.

LG, let's hope this solves all the AMDGPU issues we've seen.

Tue, Jan 18, 7:58 AM · Restricted Project
jdoerfert accepted D117526: [Libomptarget] Fix external visibility for internal variables.

LG, even though we need to get rid of the non THREAD_LOCAL globals asap.

Tue, Jan 18, 7:53 AM · Restricted Project

Thu, Jan 13

jdoerfert updated the diff for D113119: [OpenMP] Introduce asynchronous malloc and free.

put sync for delete back

Thu, Jan 13, 4:15 PM · Restricted Project
jdoerfert updated the diff for D113119: [OpenMP] Introduce asynchronous malloc and free.

Address comment

Thu, Jan 13, 4:07 PM · Restricted Project
jdoerfert added a comment to D117241: [Attributor] Share code for abstract interpretation of allocation sizes with getObjectSize [NFC-ish].

LG, minor nits below though

Thu, Jan 13, 3:35 PM · Restricted Project
jdoerfert accepted D117242: [Attributor] Generalize heap to stack to any allocator with relevant properties.

I'm going to assume the helpers do something sane. Works for me. Thx :)

Thu, Jan 13, 12:26 PM · Restricted Project
jdoerfert accepted D117230: [openmp][devicertl] Handle missing clang_tool.

LG

Thu, Jan 13, 9:38 AM · Restricted Project
jdoerfert added a comment to D117211: [openmp] Mark used variables as retain as well.

See D97446. Used no longer means "used by something so don't delete it".

I don't know why we had no multiply defined symbol errors with the old version, should have done. Maybe lucky interleaving of the breaking change to the attribute with the changes to the driver.

Tests in trunk don't use the environment variable, so didn't notice it missing / containing zero.

Thu, Jan 13, 8:55 AM · Restricted Project
jdoerfert reopened D117211: [openmp] Mark used variables as retain as well.

This might have broke nvidia and I'm not sure I understand the logic.

Thu, Jan 13, 8:50 AM · Restricted Project
jdoerfert accepted D117226: [OpenMPIRBuilder] Detect and fix ambiguous InsertPoints for createParallel..

LG, thanks!

Thu, Jan 13, 7:37 AM · Restricted Project, Restricted Project

Tue, Jan 11

jdoerfert accepted D116549: [OpenMP][Clang] Allow passing target features in ISA trait for metadirective clause.

LG, thanks for the adjustment. Hope you are happy with the result.

Tue, Jan 11, 7:33 AM · Restricted Project
jdoerfert added a comment to D116998: [LangRef] Don't allow read from sret memory after unwind.

FWIW, I'm fine with this.

Tue, Jan 11, 7:32 AM · Restricted Project

Mon, Jan 10

jdoerfert added inline comments to D116549: [OpenMP][Clang] Allow passing target features in ISA trait for metadirective clause.
Mon, Jan 10, 2:18 PM · Restricted Project
jdoerfert added a comment to D116971: [AttributorAttributes] Remove hardcoded parameters.

Is it possible to get away without adding getAllocSizeArgs to MemoryBuiltins. Based on some email conversations that I have had, getObjectSize would be prefered but in this case, we want to get the size arguments even if actual size is not a known constant. It's also used to get one of the types for the memset on line 5999. I don't fully understand what that type is doing so there may be a better way to do that.

Mon, Jan 10, 2:09 PM · Restricted Project
jdoerfert planned changes to D116959: [DWARF] Try not to crash for codes with missing debug information.

I plan to distill a test from the bug report before I move this further. Wanted to put it here to get feedback and not forget about the patch file in my local folder ;)

Mon, Jan 10, 11:15 AM · Restricted Project
jdoerfert updated the summary of D116959: [DWARF] Try not to crash for codes with missing debug information.
Mon, Jan 10, 10:36 AM · Restricted Project
jdoerfert requested review of D116959: [DWARF] Try not to crash for codes with missing debug information.
Mon, Jan 10, 10:35 AM · Restricted Project
jdoerfert committed rG7b39dccbe4d6: [Attributor][FIX] Ensure "IsExact" is false for non-exact accesses (authored by jdoerfert).
[Attributor][FIX] Ensure "IsExact" is false for non-exact accesses
Mon, Jan 10, 8:15 AM
jdoerfert added inline comments to D113359: [Libomptarget][WIP] Introduce VGPU Plugin.
Mon, Jan 10, 7:00 AM · Restricted Project, Restricted Project, Restricted Project
jdoerfert added inline comments to D116549: [OpenMP][Clang] Allow passing target features in ISA trait for metadirective clause.
Mon, Jan 10, 6:45 AM · Restricted Project

Sun, Jan 9

jdoerfert added a comment to D116906: [OpenMP][AMDGPU] Optimize the linked in math libraries.

This fixes/hides the last remaining OvO failures with upstream on an AMD MI100 compiled with -O3.

Sun, Jan 9, 7:21 PM · Restricted Project
jdoerfert abandoned D113140: [OpenMP][NFCI] Introduce the kernel environment for target regions.

Split up, main content starts with D116908.

Sun, Jan 9, 7:16 PM · Restricted Project, Restricted Project, Restricted Project
jdoerfert requested review of D116910: [OpenMP][3/3] Introduce the KernelEnvironment into Clang tests.
Sun, Jan 9, 7:16 PM · Restricted Project
jdoerfert requested review of D116909: [OpenMP][2/3] Introduce the KernelEnvironment into the middle-end.
Sun, Jan 9, 7:15 PM · Restricted Project
jdoerfert requested review of D116908: [OpenMP][1/3] Introduce the KernelEnvironment into the device runtimes.
Sun, Jan 9, 7:15 PM · Restricted Project, Restricted Project
jdoerfert requested review of D116907: [OpenMP][NFCI] Move headers into llvm/Frontend/OpenMP to be reusable.
Sun, Jan 9, 7:14 PM · Restricted Project, Restricted Project
jdoerfert added a comment to D113119: [OpenMP] Introduce asynchronous malloc and free.

Took me a while to get back to this. Way behind on certain things. Apologies.

Sun, Jan 9, 6:26 PM · Restricted Project
jdoerfert updated the diff for D113119: [OpenMP] Introduce asynchronous malloc and free.

Address comments

Sun, Jan 9, 6:25 PM · Restricted Project
jdoerfert requested review of D116906: [OpenMP][AMDGPU] Optimize the linked in math libraries.
Sun, Jan 9, 6:20 PM · Restricted Project
jdoerfert updated subscribers of D116879: [llvm] Allow auto-vectorization of sincos() using libmvec.
Sun, Jan 9, 4:10 PM · Restricted Project
jdoerfert committed rG4e8a02e7f49d: [Attributor][FIX] Remove assumption that doesn't have to hold (authored by jdoerfert).
[Attributor][FIX] Remove assumption that doesn't have to hold
Sun, Jan 9, 11:16 AM

Sat, Jan 8

jdoerfert committed rG37639b72a17b: [OpenMP][FIX] Emit debug declares only if debug info is available (authored by jdoerfert).
[OpenMP][FIX] Emit debug declares only if debug info is available
Sat, Jan 8, 4:13 PM
jdoerfert committed rG6c745e04fac8: [Attributor][FIX] Ensure order for multiple references into map (authored by jdoerfert).
[Attributor][FIX] Ensure order for multiple references into map
Sat, Jan 8, 4:13 PM
jdoerfert closed D116865: [OpenMP][FIX] Emit debug declares only if debug info is available.
Sat, Jan 8, 4:13 PM · Restricted Project
jdoerfert updated the diff for D116865: [OpenMP][FIX] Emit debug declares only if debug info is available.

Clang format

Sat, Jan 8, 11:59 AM · Restricted Project
jdoerfert requested review of D116865: [OpenMP][FIX] Emit debug declares only if debug info is available.
Sat, Jan 8, 11:56 AM · Restricted Project
jdoerfert reopened D96248: [OpenMP][AMDGPU] Add support for linking libomptarget bitcode.
Sat, Jan 8, 11:31 AM · Restricted Project

Fri, Jan 7

jdoerfert accepted D116746: IR: Make getRetAlign check callee function attributes.

Still inconsistent after but I guess somewhat better.

Fri, Jan 7, 8:06 AM · Restricted Project
jdoerfert added a comment to D116800: [MemoryBuiltins] Remove isNoAliasFn() in favor of isNoAliasCall().

I like this, anyone else?

Fri, Jan 7, 8:03 AM · Restricted Project

Thu, Jan 6

jdoerfert added inline comments to D116549: [OpenMP][Clang] Allow passing target features in ISA trait for metadirective clause.
Thu, Jan 6, 9:58 AM · Restricted Project

Wed, Jan 5

jdoerfert added a comment to D116609: [EarlyCSE] Allow elimination of redundant writeonly calls.

Drive by: nounwind and nosync come to mind as being somewhat important here. Also willreturn.

Wed, Jan 5, 5:43 AM · Restricted Project

Mon, Jan 3

jdoerfert added a comment to D116549: [OpenMP][Clang] Allow passing target features in ISA trait for metadirective clause.

Can you make the test check for the diagnose message? Also, do we have a test to verify an isa trait is properly handled?

Mon, Jan 3, 2:15 PM · Restricted Project
jdoerfert accepted D116540: [OpenMP] Add nec and nvidia as compiler vendors for OpenMP.
  1. I don't know why "pgi" is present here as a vendor. Should it be renamed as nvidia instead? @Meinersbur @tra

It was in the spec list and I would not remove it now.

Mon, Jan 3, 9:13 AM · Restricted Project, Restricted Project

Tue, Dec 28

jdoerfert committed rG7e14e881c411: [OpenMP][OpenACC] Update test after encoding change in D113126 (authored by jdoerfert).
[OpenMP][OpenACC] Update test after encoding change in D113126
Tue, Dec 28, 11:30 PM
jdoerfert added a reverting change for rG73ece231ee0c: Revert "[OpenMP][NFCI] Embed the source location string size in the ident_t": rG944aa0421cb7: Reapply "[OpenMP][NFCI] Embed the source location string size in the ident_t".
Tue, Dec 28, 11:12 PM
jdoerfert added a reverting change for rG423ba12971ba: Revert "[OpenMP][FIX] Also update unit test after API change": rG944aa0421cb7: Reapply "[OpenMP][NFCI] Embed the source location string size in the ident_t".
Tue, Dec 28, 11:12 PM
jdoerfert committed rG944aa0421cb7: Reapply "[OpenMP][NFCI] Embed the source location string size in the ident_t" (authored by jdoerfert).
Reapply "[OpenMP][NFCI] Embed the source location string size in the ident_t"
Tue, Dec 28, 11:12 PM
jdoerfert committed rG5602c866c062: [Attributor] Look through allocated heap memory (authored by jdoerfert).
[Attributor] Look through allocated heap memory
Tue, Dec 28, 10:22 PM
jdoerfert committed rG3e0c512ce6ce: [OpenMP] Simplify all stores in the device code (authored by jdoerfert).
[OpenMP] Simplify all stores in the device code
Tue, Dec 28, 10:22 PM
jdoerfert closed D109170: [Attributor] Look through allocated heap memory.
Tue, Dec 28, 10:21 PM · Restricted Project
jdoerfert closed D109998: [OpenMP] Simplify all stores in the device code.
Tue, Dec 28, 10:21 PM · Restricted Project
jdoerfert committed rG16da21400458: [OpenMP][FIX] Also update unit test after API change (authored by jdoerfert).
[OpenMP][FIX] Also update unit test after API change
Tue, Dec 28, 10:18 PM
jdoerfert committed rG7f1eaeafe7a1: [OpenMP][NFC] Extract assumption helpers into own header file (authored by jdoerfert).
[OpenMP][NFC] Extract assumption helpers into own header file
Tue, Dec 28, 9:54 PM
jdoerfert committed rG7bfcdbcbf368: [OpenMP][NFCI] Embed the source location string size in the ident_t (authored by jdoerfert).
[OpenMP][NFCI] Embed the source location string size in the ident_t
Tue, Dec 28, 9:54 PM
jdoerfert committed rG6e2fcf8513a3: [Attributor][FIX] Ensure store uses are correlated with reloads (authored by jdoerfert).
[Attributor][FIX] Ensure store uses are correlated with reloads
Tue, Dec 28, 9:54 PM
jdoerfert committed rG9f04a0ea433b: [OpenMP][FIX] Make AAExecutionDomain deterministic (authored by jdoerfert).
[OpenMP][FIX] Make AAExecutionDomain deterministic
Tue, Dec 28, 9:54 PM
jdoerfert committed rGba70f3a5d914: [OpenMP][FIX] Make heap2shared deterministic (authored by jdoerfert).
[OpenMP][FIX] Make heap2shared deterministic
Tue, Dec 28, 9:54 PM
jdoerfert closed D113126: [OpenMP][NFCI] Embed the source location string size in the ident_t.
Tue, Dec 28, 9:54 PM · Restricted Project, Restricted Project
jdoerfert committed rG7de5da2a67e5: [OpenMP][NFC] Move address space enum into OMPConstants header (authored by jdoerfert).
[OpenMP][NFC] Move address space enum into OMPConstants header
Tue, Dec 28, 9:54 PM
jdoerfert closed D109662: [Attributor] Ensure store uses are correlated with reloads.
Tue, Dec 28, 9:54 PM · Restricted Project
jdoerfert committed rG73104ad65bc0: [OpenMP][NFC] Move headers into include folder (authored by jdoerfert).
[OpenMP][NFC] Move headers into include folder
Tue, Dec 28, 9:54 PM
jdoerfert added a comment to D113126: [OpenMP][NFCI] Embed the source location string size in the ident_t.

I'm not convinced. std::strlen can do the job. Can you explain more why need it?

Because the ident_t is on the device and you want to grab it from the host.

So you mean strlen is not available on device?

Tue, Dec 28, 9:02 PM · Restricted Project, Restricted Project
jdoerfert accepted D116340: [OpenMP][Plugin] Minor adjustments to ResourcePool.

LG

Tue, Dec 28, 12:34 PM · Restricted Project

Mon, Dec 27

jdoerfert accepted D116319: [OpenMP] Use alignment information in HeapToShared.

LG

Mon, Dec 27, 12:38 PM · Restricted Project
jdoerfert added a comment to D115888: [Attributor][Fix] Add alignment return attribute to HeapToStack.

Still LG

Mon, Dec 27, 11:39 AM · Restricted Project, Restricted Project