[PATCH] D97164: [flang] Add -J and -module-dir to f18 driver

Mon Feb 22 14:33:58 PST 2021

sscalpone added subscribers: schweitz, jeanPerier, klausler, PeteSteinfeld, svedanayagam.
sscalpone added a comment.

Thanks.

Sam, as for implementing these, I think for each intrinsic we should implement the full f18 spec, instead of doing ½ now and ½ later.  Also, let’s make sure we have tests submitted as we go, and not after the fact; best in a pre-commit and no later than with the first comment.

I’m saying, let’s use f77, f95, etc to guide our implementation order, but not the depth of implementation.  My preference would be to hold off on any instrinsics that we aren’t fully ready to implement until we are, instead of doing a partial job.  Tests are essential.

From: Jean Perier <jperier at nvidia.com>
Date: Monday, February 22, 2021 at 9:18 AM
To: Steve Scalpone <sscalpone at nvidia.com>, Samuel Vedanayagam <svedanayagam at nvidia.com>, Eric Schweitz <eschweitz at nvidia.com>, Val Donaldson <vdonaldson at nvidia.com>
Subject: RE: F18 <https://reviews.llvm.org/F18> Status 2/11/21

Thanks Steve,

I did update the table with the lowering status for all these intrinsics.
Out of the 357:

- 96 are lowered/nothing to do in lowering (front-end takes care of it/rewrites it another intrinsic)
- 16 are in progress (mainly shape inquiry related).
- 79 are GCC extensions that semantics does not support. I think Peter plan to support some (there are some TODOs left in semantics intrinsics.cpp), but I am not sure if all makes sense.
- 166 are TODO (I did not filled the API status, except for the 35 coarray related ones, where I am sure there is no runtime API yet).

From: Steve Scalpone <sscalpone at nvidia.com>
Sent: Monday, February 22, 2021 8:17 AM
To: Jean Perier <jperier at nvidia.com>; Samuel Vedanayagam <svedanayagam at nvidia.com>; Eric Schweitz <eschweitz at nvidia.com>; Val Donaldson <vdonaldson at nvidia.com>
Subject: Re: F18 <https://reviews.llvm.org/F18> Status 2/11/21

Jean’s sheet is here:
https://docs.google.com/spreadsheets/d/1ctIPb4ADTuLDgwKzZNCThzNnbOcwWG5tA6UTuA9tO_c/edit?usp=sharing

Plus a bunch more intrinsics for ~350 total!

From: Jean Perier <jperier at nvidia.com<mailto:jperier at nvidia.com>>
Date: Monday, February 15, 2021 at 10:14 AM
To: Samuel Vedanayagam <svedanayagam at nvidia.com<mailto:svedanayagam at nvidia.com>>, Eric Schweitz <eschweitz at nvidia.com<mailto:eschweitz at nvidia.com>>, Val Donaldson <vdonaldson at nvidia.com<mailto:vdonaldson at nvidia.com>>
Cc: Steve Scalpone <sscalpone at nvidia.com<mailto:sscalpone at nvidia.com>>
Subject: RE: F18 <https://reviews.llvm.org/F18> Status 2/11/21

Attached is the F95 <https://reviews.llvm.org/F95> intrinsic lowering status.

- 57 of the 115 generic intrinsic needed for F95 <https://reviews.llvm.org/F95> have partial (7) or no implementation at all (50).

Out of the 57:

- 13 are character related. I believe Peter K. wrote some of the runtime for that already, but we need to plug into it and see what is missing.
- 15 are scalar numeric intrinsics (mainly bit manipulations and floating point representation manipulation)
- 20 are array transformationals (MATMUL, RESHAPE, TRANSPOSE…. here lies the biggest work in my opinion).
- 4 are shape inquiry related (in progress, they are a bit standing apart because the front-end is rewriting them to array ctors and descriptor inquiries).
- 5 are system support related (time measurement and random generation related).

From: Samuel Vedanayagam <svedanayagam at nvidia.com<mailto:svedanayagam at nvidia.com>>
Sent: Thursday, February 11, 2021 9:17 PM
To: Eric Schweitz <eschweitz at nvidia.com<mailto:eschweitz at nvidia.com>>; Val Donaldson <vdonaldson at nvidia.com<mailto:vdonaldson at nvidia.com>>; Jean Perier <jperier at nvidia.com<mailto:jperier at nvidia.com>>
Cc: Steve Scalpone <sscalpone at nvidia.com<mailto:sscalpone at nvidia.com>>
Subject: F18 <https://reviews.llvm.org/F18> Status 2/11/21

Val
Code cleanup & merge on hold, as Eric wants this broken up into smaller pieces.
Corrays:

- Reading documentation, looking at the constraints (looks like about 40 have to be implemented).

- Looking to see if we can piggy back on existing checkers to implement the constraints.

Eric
Uploading changes.
TACO – code changes required for the upload, opaque pointers an issue for Mehdi.
Will work on this for the next two days.   Starting from next week, will split time 50/50 between uploading work and other work.
PR reviews.

Jean
Descriptor support work:

- Finished FIR rebox definition

- Finished code gen

- Adding more tests

Next:

- List of intrinsics for F95 <https://reviews.llvm.org/F95>

- PR for present intrinsic implementation

From: Samuel Vedanayagam
Subject: F18 <https://reviews.llvm.org/F18> Status 2/8/21

Update from Monday.  Let me know if I did not get anything correctly.

Val
Associate statement – One case left, need support array slicing to proceed
Select statement – Character case left, need FIR OP to proceed
Co-arrays – next thing to work on
Currently working on code cleanup – remove dead code from a merge, PFT changes

Jean
3 PR’s pending:

- Optional, Contiguous FIR attributes

- New IR box

- Add FIR ops

Eric
Fixed bug with array expressions
Reviewed Jean’s PR’s
Shuffling code
New FIR OP’s – next thing to work on

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D97164/new/

https://reviews.llvm.org/D97164