Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

add TaggedExpression #688

Open
wants to merge 19 commits into
base: main
Choose a base branch
from
Open

add TaggedExpression #688

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
19 commits
Select commit Hold shift + click to select a range
9233385
add WithTag
matthiasdiener Oct 11, 2022
679d269
Merge branch 'main' into withtag
matthiasdiener Oct 18, 2022
45e14e7
rename to TaggedExpression
matthiasdiener Oct 18, 2022
50adbc4
better tags propagation
matthiasdiener Oct 18, 2022
06ab987
add to memaccess
matthiasdiener Oct 18, 2022
d50f908
Merge branch 'main' into withtag
matthiasdiener Oct 18, 2022
ee395c1
Merge branch 'main' into withtag
matthiasdiener Oct 27, 2022
04d0739
Merge branch 'main' into withtag
matthiasdiener Nov 9, 2022
9ae8936
spelling fixes
matthiasdiener Nov 9, 2022
0bbbaec
provide default value for tags in map_ functions
matthiasdiener Nov 9, 2022
9529588
better doc
matthiasdiener Nov 9, 2022
9e24cd6
attempt to fix doctest
matthiasdiener Nov 9, 2022
e72dec0
add Sync
matthiasdiener Nov 9, 2022
c7f9ad0
Merge branch 'main' into withtag
matthiasdiener Nov 11, 2022
df7ed9b
no default tags arguments
matthiasdiener Nov 11, 2022
857e268
lint fix
matthiasdiener Nov 11, 2022
d5f3f5d
small arg fixes
matthiasdiener Nov 11, 2022
3d7df9f
Add some types to loopy.translation_unit
inducer Nov 28, 2022
6142a3b
Stats: fix tags passing, add types
inducer Nov 28, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
42 changes: 21 additions & 21 deletions doc/tutorial.rst
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1566,7 +1566,7 @@ information provided. Now we will count the operations:

>>> op_map = lp.get_op_map(knl, subgroup_size=32)
>>> print(op_map)
Op(np:dtype('float32'), add, subgroup, "stats_knl"): ...
Op(np:dtype('float32'), add, subgroup, "stats_knl", None): ...

Each line of output will look roughly like::

Expand Down Expand Up @@ -1628,7 +1628,7 @@ together into keys containing only the specified fields:

>>> op_map_dtype = op_map.group_by('dtype')
>>> print(op_map_dtype)
Op(np:dtype('float32'), None, None): ...
Op(np:dtype('float32'), None, None, None): ...
<BLANKLINE>
>>> f32op_count = op_map_dtype[lp.Op(dtype=np.float32)
... ].eval_with_dict(param_dict)
Expand All @@ -1654,7 +1654,7 @@ we'll continue using the kernel from the previous example:

>>> mem_map = lp.get_mem_access_map(knl, subgroup_size=32)
>>> print(mem_map)
MemAccess(global, np:dtype('float32'), {}, {}, load, a, None, subgroup, 'stats_knl'): ...
MemAccess(global, np:dtype('float32'), {}, {}, load, a, None, subgroup, 'stats_knl', None): ...
<BLANKLINE>

Each line of output will look roughly like::
Expand Down Expand Up @@ -1725,13 +1725,13 @@ using :func:`loopy.ToCountMap.to_bytes` and :func:`loopy.ToCountMap.group_by`:

>>> bytes_map = mem_map.to_bytes()
>>> print(bytes_map)
MemAccess(global, np:dtype('float32'), {}, {}, load, a, None, subgroup, 'stats_knl'): ...
MemAccess(global, np:dtype('float32'), {}, {}, load, a, None, subgroup, 'stats_knl', None): ...
<BLANKLINE>
>>> global_ld_st_bytes = bytes_map.filter_by(mtype=['global']
... ).group_by('direction')
>>> print(global_ld_st_bytes)
MemAccess(None, None, None, None, load, None, None, None, None): ...
MemAccess(None, None, None, None, store, None, None, None, None): ...
MemAccess(None, None, None, None, load, None, None, None, None, None): ...
MemAccess(None, None, None, None, store, None, None, None, None, None): ...
<BLANKLINE>
>>> loaded = global_ld_st_bytes[lp.MemAccess(direction='load')
... ].eval_with_dict(param_dict)
Expand Down Expand Up @@ -1768,12 +1768,12 @@ this time.
... outer_tag="l.1", inner_tag="l.0")
>>> mem_map = lp.get_mem_access_map(knl_consec, subgroup_size=32)
>>> print(mem_map)
MemAccess(global, np:dtype('float32'), {0: 1, 1: 128}, {}, load, a, None, workitem, 'stats_knl'): ...
MemAccess(global, np:dtype('float32'), {0: 1, 1: 128}, {}, load, b, None, workitem, 'stats_knl'): ...
MemAccess(global, np:dtype('float32'), {0: 1, 1: 128}, {}, store, c, None, workitem, 'stats_knl'): ...
MemAccess(global, np:dtype('float64'), {0: 1, 1: 128}, {}, load, g, None, workitem, 'stats_knl'): ...
MemAccess(global, np:dtype('float64'), {0: 1, 1: 128}, {}, load, h, None, workitem, 'stats_knl'): ...
MemAccess(global, np:dtype('float64'), {0: 1, 1: 128}, {}, store, e, None, workitem, 'stats_knl'): ...
MemAccess(global, np:dtype('float32'), {0: 1, 1: 128}, {}, load, a, None, workitem, 'stats_knl', None): ...
MemAccess(global, np:dtype('float32'), {0: 1, 1: 128}, {}, load, b, None, workitem, 'stats_knl', None): ...
MemAccess(global, np:dtype('float32'), {0: 1, 1: 128}, {}, store, c, None, workitem, 'stats_knl', None): ...
MemAccess(global, np:dtype('float64'), {0: 1, 1: 128}, {}, load, g, None, workitem, 'stats_knl', None): ...
MemAccess(global, np:dtype('float64'), {0: 1, 1: 128}, {}, load, h, None, workitem, 'stats_knl', None): ...
MemAccess(global, np:dtype('float64'), {0: 1, 1: 128}, {}, store, e, None, workitem, 'stats_knl', None): ...
<BLANKLINE>

With this parallelization, consecutive work-items will access consecutive array
Expand Down Expand Up @@ -1813,12 +1813,12 @@ we'll switch the inner and outer tags in our parallelization of the kernel:
... outer_tag="l.0", inner_tag="l.1")
>>> mem_map = lp.get_mem_access_map(knl_nonconsec, subgroup_size=32)
>>> print(mem_map)
MemAccess(global, np:dtype('float32'), {0: 128, 1: 1}, {}, load, a, None, workitem, 'stats_knl'): ...
MemAccess(global, np:dtype('float32'), {0: 128, 1: 1}, {}, load, b, None, workitem, 'stats_knl'): ...
MemAccess(global, np:dtype('float32'), {0: 128, 1: 1}, {}, store, c, None, workitem, 'stats_knl'): ...
MemAccess(global, np:dtype('float64'), {0: 128, 1: 1}, {}, load, g, None, workitem, 'stats_knl'): ...
MemAccess(global, np:dtype('float64'), {0: 128, 1: 1}, {}, load, h, None, workitem, 'stats_knl'): ...
MemAccess(global, np:dtype('float64'), {0: 128, 1: 1}, {}, store, e, None, workitem, 'stats_knl'): ...
MemAccess(global, np:dtype('float32'), {0: 128, 1: 1}, {}, load, a, None, workitem, 'stats_knl', None): ...
MemAccess(global, np:dtype('float32'), {0: 128, 1: 1}, {}, load, b, None, workitem, 'stats_knl', None): ...
MemAccess(global, np:dtype('float32'), {0: 128, 1: 1}, {}, store, c, None, workitem, 'stats_knl', None): ...
MemAccess(global, np:dtype('float64'), {0: 128, 1: 1}, {}, load, g, None, workitem, 'stats_knl', None): ...
MemAccess(global, np:dtype('float64'), {0: 128, 1: 1}, {}, load, h, None, workitem, 'stats_knl', None): ...
MemAccess(global, np:dtype('float64'), {0: 128, 1: 1}, {}, store, e, None, workitem, 'stats_knl', None): ...
<BLANKLINE>

With this parallelization, consecutive work-items will access *nonconsecutive*
Expand Down Expand Up @@ -1871,7 +1871,7 @@ kernel from the previous example:

>>> sync_map = lp.get_synchronization_map(knl)
>>> print(sync_map)
Sync(kernel_launch, stats_knl): [l, m, n] -> { 1 }
Sync(kernel_launch, stats_knl, None): [l, m, n] -> { 1 }
<BLANKLINE>

We can evaluate this polynomial using :meth:`islpy.PwQPolynomial.eval_with_dict`:
Expand Down Expand Up @@ -1931,8 +1931,8 @@ count the barriers using :func:`loopy.get_synchronization_map`:

>>> sync_map = lp.get_synchronization_map(knl)
>>> print(sync_map)
Sync(barrier_local, loopy_kernel): { 1000 }
Sync(kernel_launch, loopy_kernel): { 1 }
Sync(barrier_local, loopy_kernel, None): { 1000 }
Sync(kernel_launch, loopy_kernel, None): { 1 }
<BLANKLINE>

Based on the kernel code printed above, we would expect each work-item to
Expand Down
Loading