Skip to content

Loopy sprint read write output args in slate#2033

Merged
sv2518 merged 4 commits into
loopy-sprintfrom
loopy-sprint-RWargsInSlate
Apr 21, 2021
Merged

Loopy sprint read write output args in slate#2033
sv2518 merged 4 commits into
loopy-sprintfrom
loopy-sprint-RWargsInSlate

Conversation

@sv2518

@sv2518 sv2518 commented Apr 21, 2021

Copy link
Copy Markdown
Contributor

This adapting to 714b68579b784a869d789e2c639348e1753727f9.

sv2518 added 2 commits April 21, 2021 11:57
@sv2518
Output args of TSFC kernels are read and write.
c94d83c
@sv2518
Output args of TSFC kernels are read and write. We adapted to that in…
0da7b99 
… PyOP2 here 714b68579b784a869d789e2c639348e1753727f9, so we have adapt the slatewrapper to have its output read and write as well.
@sv2518 sv2518 requested a review from wence- April 21, 2021 10:03
@sv2518

sv2518 commented Apr 21, 2021

Copy link
Copy Markdown
Contributor Author

There is still something missing, will need to add the output arg to the variables of the Slate CallInstructions too.

wence-
wence- reviewed Apr 21, 2021
View reviewed changes
Comment thread firedrake/slate/slac/compiler.py
idx = make_indices(len(shape))
indexed_gem_expr = gem.Indexed(gem_expr, idx)
args = ([loopy.GlobalArg("output", shape=shape, dtype=scalar_type)]
args = ([loopy.GlobalArg("output", shape=shape, dtype=scalar_type, is_output=True, is_input=True)]

@wence- wence- Apr 21, 2021

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Should TSFC be setting these inout parameters on its globalargs too?

@sv2518 sv2518 Apr 21, 2021

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I would think so yes, but want so see if there is a test actually failing because it's not there. The test would fail with a KeyError in the Kernel matching.

@sv2518
Output args of TSFC kernels are read and write. We adapted to that in…
af2f959 
… PyOP2 here 714b68579b784a869d789e2c639348e1753727f9, so we have adapt the slatewrapper to have its output read and write as well. And also the call to the kernel containing the Slate instructions.
@sv2518 sv2518 merged commit c4346b1 into loopy-sprint Apr 21, 2021
@sv2518 sv2518 deleted the loopy-sprint-RWargsInSlate branch April 21, 2021 15:53
sv2518 added a commit that referenced this pull request May 10, 2021
@sv2518
Squashed commit of the following:
ebd2fa6 
commit c53cc67
Author: Sophia Vorderwuelbecke <sv2518@ic.ac.uk>
Date:   Mon May 10 11:59:53 2021 +0200

    Lint.

commit 8a46d6c
Merge: 2f7da5f aa1647a
Author: Sophia Vorderwuelbecke <sv2518@ic.ac.uk>
Date:   Mon May 10 11:52:18 2021 +0200

    Merge branch 'master' into loopy-sprint

commit 2f7da5f
Author: Sophia Vorderwuelbecke <sv2518@ic.ac.uk>
Date:   Mon May 10 11:10:40 2021 +0200

    Lint.

commit 524121b
Author: Sophia Vorderwuelbecke <sv2518@ic.ac.uk>
Date:   Tue Apr 27 11:42:58 2021 +0200

    Fix my bug in extrusion utils.

commit 259c3a2
Author: Sophia Vorderwuelbecke <sv2518@ic.ac.uk>
Date:   Tue Apr 27 10:26:30 2021 +0200

    Fix assumption on the layer in extrusion kernel. Only make additional assumption if there is more than one layer height.

commit db0a907
Author: Sophia Vorderwuelbecke <sv2518@ic.ac.uk>
Date:   Mon Apr 26 22:58:15 2021 +0200

    Add import.

commit 0a87fbc
Author: Sophia Vorderwuelbecke <sv2518@ic.ac.uk>
Date:   Mon Apr 26 22:57:41 2021 +0200

    ? Fix remaining failing extrusion tests. Why are all the related tests now failing on assertion with a value that is double of what it is supposed to be? e.g. tests/extrusion/test_kernel_intrhs.py

commit d1b25c9
Author: Sophia Vorderwuelbecke <sv2518@ic.ac.uk>
Date:   Mon Apr 26 22:51:22 2021 +0200

    Add additional assumptions on the layer arg. Fixes tests/extrusion/test_layer_height_mesh_volume.py::test_extrude_radial_mesh_volume

commit b413015
Author: Sophia Vorderwuelbecke <sv2518@ic.ac.uk>
Date:   Mon Apr 26 13:42:05 2021 +0200

    JENKINS run with uptodate loopy branch.

commit a47c31a
Author: Sophia Vorderwuelbecke <sv2518@ic.ac.uk>
Date:   Mon Apr 26 11:58:11 2021 +0200

    Adapt to migation of match_caller_callee...

commit 2e02549
Merge: ef4254b 5e05a66
Author: Sophia Vdw <54070354+sv2518@users.noreply.github.com>
Date:   Fri Apr 23 16:40:12 2021 +0200

    Merge pull request #2040 from firedrakeproject/correct-access-mods-for-expression-kernels

    Specify is_input and is_output for loopy.GlobalArg objects

commit ef4254b
Author: Reuben Nixon-Hill <ReubenHill@users.noreply.github.com>
Date:   Fri Apr 23 15:34:05 2021 +0100

    lint

commit cebdf52
Author: Lawrence Mitchell <lawrence@wence.uk>
Date:   Fri Apr 23 15:32:45 2021 +0100

    Correct access descriptors for augmented pointwise assignment

commit 5e05a66
Author: Connor Ward <c.ward20@imperial.ac.uk>
Date:   Fri Apr 23 15:14:43 2021 +0100

    Add op2.MAX and op2.MIN to is_input and is_output

commit c3c18bd
Author: Connor Ward <c.ward20@imperial.ac.uk>
Date:   Fri Apr 23 15:06:32 2021 +0100

    Specify is_input and is_output for loopy.GlobalArg objects

commit 2f11e7b
Author: Kaushik Kulkarni <kaushikcfd@gmail.com>
Date:   Fri Apr 23 07:30:40 2021 -0500

    add lang_version=(2018, 2)

commit aa98a01
Author: Sophia Vorderwuelbecke <sv2518@ic.ac.uk>
Date:   Fri Apr 23 12:54:58 2021 +0200

    New fix for adaption to 714b68579b784a869d789e2c639348e1753727f9 in PyOP2

commit db038c7
Author: Sophia Vorderwuelbecke <sv2518@ic.ac.uk>
Date:   Fri Apr 23 12:54:15 2021 +0200

    Revert: ? This should adapt to 714b68579b784a869d789e2c639348e1753727f9...

commit 750a2f7
Author: Reuben Nixon-Hill <ReubenHill@users.noreply.github.com>
Date:   Fri Apr 23 10:48:54 2021 +0100

    Lint

commit 39123b2
Author: Reuben Nixon-Hill <ReubenHill@users.noreply.github.com>
Date:   Fri Apr 23 10:15:02 2021 +0100

    remove out of date import

commit 44750f0
Author: Sophia Vorderwuelbecke <sv2518@ic.ac.uk>
Date:   Thu Apr 22 21:48:23 2021 +0200

    ? This should adapt to 714b68579b784a869d789e2c639348e1753727f9 in PyOP2, where the matrix arg gets read and written in the kernel. The regression test regression/test_p1pc.py is now compiling but failing with the wrong answer.

commit 04afdc2
Author: Sophia Vorderwuelbecke <sv2518@ic.ac.uk>
Date:   Thu Apr 22 21:43:13 2021 +0200

    Adapt to tsfc changes.

commit e89adf1
Author: Sophia Vorderwuelbecke <sv2518@ic.ac.uk>
Date:   Thu Apr 22 14:21:55 2021 +0200

    Slate kernel builder: Make the layer argument a scalar.

commit f88493c
Author: Sophia Vorderwuelbecke <sv2518@ic.ac.uk>
Date:   Thu Apr 22 14:21:27 2021 +0200

    Slate kernel builder: Subscript the layer argument.

commit bb059c8
Author: Sophia Vorderwuelbecke <sv2518@ic.ac.uk>
Date:   Thu Apr 22 08:27:30 2021 +0200

    Match Slate wrapper with slate and tsfc kernels.

commit 1f0d8bf
Author: Sophia Vorderwuelbecke <sv2518@ic.ac.uk>
Date:   Thu Apr 22 08:24:09 2021 +0200

    Fix global arg generation for cellsizes and cellorientations. I think it would be better to safe a bit more information on the bag so that I don't need to access the orienations/size over slate_expression.ufl_domain().

commit c4346b1
Merge: 8775b9a af2f959
Author: Sophia Vdw <54070354+sv2518@users.noreply.github.com>
Date:   Wed Apr 21 17:53:25 2021 +0200

    Merge pull request #2033 from firedrakeproject/loopy-sprint-RWargsInSlate

    Loopy sprint read write output args in slate

commit af2f959
Author: Sophia Vorderwuelbecke <sv2518@ic.ac.uk>
Date:   Wed Apr 21 14:03:13 2021 +0200

    Output args of TSFC kernels are read and write. We adapted to that in PyOP2 here 714b68579b784a869d789e2c639348e1753727f9, so we have adapt the slatewrapper to have its output read and write as well. And also the call to the kernel containing the Slate instructions.

commit 1166fd8
Author: Sophia Vorderwuelbecke <sv2518@ic.ac.uk>
Date:   Wed Apr 21 12:08:16 2021 +0200

    Argh.

commit 0da7b99
Author: Sophia Vorderwuelbecke <sv2518@ic.ac.uk>
Date:   Wed Apr 21 12:00:51 2021 +0200

    Output args of TSFC kernels are read and write. We adapted to that in PyOP2 here 714b68579b784a869d789e2c639348e1753727f9, so we have adapt the slatewrapper to have its output read and write as well.

commit c94d83c
Author: Sophia Vorderwuelbecke <sv2518@ic.ac.uk>
Date:   Wed Apr 21 11:57:24 2021 +0200

    Output args of TSFC kernels are read and write.

commit 8775b9a
Author: Sophia Vorderwuelbecke <sv2518@ic.ac.uk>
Date:   Tue Apr 20 22:16:02 2021 +0200

    Fix passing the name of the extrusion kernels into a PyOP2 Kernel.

commit f20481e
Author: Sophia Vorderwuelbecke <sv2518@ic.ac.uk>
Date:   Tue Apr 20 21:44:20 2021 +0200

    Fix passing the name of the expression kernel into a PyOP2 Kernel.

commit 26745c1
Author: Sophia Vorderwuelbecke <sv2518@ic.ac.uk>
Date:   Tue Apr 20 10:45:30 2021 +0200

    Slate Kernel Builder: fetch name from kernel callables table (rather than entrypoints) because we use make_function for Kernel generation in TSFC.

commit cb3fbbf
Author: Sophia Vorderwuelbecke <sv2518@ic.ac.uk>
Date:   Tue Apr 20 09:43:19 2021 +0200

    Compiler: Drop the workaround. We don't need to make the Slate wrapper a string Kernel anymore since we can inline multiple kernels into each other now.

commit f7a5a67
Author: Sophia Vorderwuelbecke <sv2518@ic.ac.uk>
Date:   Tue Apr 20 08:14:08 2021 +0200

    Fix cell size arg in Slate wrapper kernel.

commit d90ef48
Author: Sophia Vorderwuelbecke <sv2518@ic.ac.uk>
Date:   Mon Apr 19 23:01:40 2021 +0200

    Slate utils: Remove make_program, which became unneccessary because all Kernels are a Program now.

commit 491cc00
Author: Sophia Vorderwuelbecke <sv2518@ic.ac.uk>
Date:   Mon Apr 19 23:00:06 2021 +0200

    KernelBuilder: access kernel information of the Slate loopy kernel to generate a CallInstruction.

commit 52de378
Author: Sophia Vorderwuelbecke <sv2518@ic.ac.uk>
Date:   Mon Apr 19 22:51:58 2021 +0200

    Slate KernelBuilder: infer which args are needed for the Slate wrapper kernel over the information from the Slate wrapper bag.

commit d81f3e4
Author: Sophia Vorderwuelbecke <sv2518@ic.ac.uk>
Date:   Mon Apr 19 22:49:18 2021 +0200

    Slac: Let the compiler be in control of the wrapper kernel naming.

commit 1fca6c5
Author: Lawrence Mitchell <lawrence@wence.uk>
Date:   Mon Apr 19 16:31:03 2021 +0100

    Adapt to pyop2/loopy changes

commit 0733140
Author: Lawrence Mitchell <lawrence@wence.uk>
Date:   Mon Apr 19 16:12:27 2021 +0100

    Adapt to TSFC api change

commit ab57862
Author: Reuben Nixon-Hill <ReubenHill@users.noreply.github.com>
Date:   Mon Apr 19 15:27:38 2021 +0100

    Replace lp.register_callable_kernel with lp.merge

    knl = lp.register_callable_kernel(knl, callee) is now
    knl = lp.merge([knl, callee]) .
    Reason given for loopy breaking back-compat: earlier they had a notion
    of a special "root_kernel" which only permitted one level of nesting
    loopy kernel calls and that was too restrictive.

commit 1a54a0f
Author: Sophia Vorderwuelbecke <sv2518@ic.ac.uk>
Date:   Mon Apr 19 16:05:59 2021 +0200

    JENKINS.

commit 2ebefd3
Author: Sophia Vorderwuelbecke <sv2518@ic.ac.uk>
Date:   Mon Apr 19 15:49:19 2021 +0200

    JENKINS
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@wence- wence- wence- left review comments

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

2 participants

@sv2518 @wence-