e03ce30c7db372c7dd0f5c3b90bfb2a6b46a4ac4 to main · Andreas Klöckner / loopy · GitLab

Commits on Source 6283

76c59c5c Whitespace changes.
Oct 25, 2011
cc5a489a IndexVariableFinder: Add flag include_reduction_inames.
Oct 25, 2011
0026eb8b Use loopy's own SubstitutionMapper (not pymbolic's substitute) in realize_cse().
Oct 25, 2011
c619c54f Fix for isl 0.08: set intersect doesn't align spaces automatically--do it manually.
Oct 25, 2011
9b8de2d6 Only test that reduction depends on *one* of its inames, not all of them.
Oct 25, 2011
16290f2b Create default/empty assumptions as parameter domain.
Oct 25, 2011
44c1beaa Guess which iname should be l.0 by ranking, not by pointing at one.
Oct 25, 2011
e352f730 Split off preprocess_kernel from scheduling.
Oct 25, 2011
b201a4e2 Consolidate MEMO.
Oct 25, 2011
63893d99 More MEMO hacking.
Oct 25, 2011
45657b41 Factor out loopy.isl_helpers.duplicate_axes from CSE realization.
Oct 25, 2011
b4e8ee54 Split out kernel preprocessing into separate file.
Oct 25, 2011
a721b3cf Fix dependencies of main reduction instruction: Must depend on all reduction axes.
Oct 25, 2011
3213b9cb Make a parse syntax for the 'None' tag.
Oct 25, 2011
fa54e763 Add iname duplication from parsed instructions.
Oct 25, 2011
76becd30 Add spectral-element tests.
Oct 25, 2011
94107a5e Allow array declaration from within an instruction.
Oct 26, 2011
020203ae Give a good error message for missing variables in get_problems().
Oct 26, 2011
426c0319 Implement reduction iname uniquification.
Oct 28, 2011
4fc947f6 Fix storage order in SEM example.
Oct 28, 2011
3f8f1593 Look at individual variable writes/reads to see if barriers are needed.
Oct 29, 2011
7a368750 Give user control over whether reduction inames are duplicated.
Oct 29, 2011
94eec370 Rename idempotent->boostable. Be more restrictive in marking insns boostable.
Oct 29, 2011
17ddb114 Allow unroll of sequential loops.
Oct 29, 2011
4853b6d5 Fix image arguments.
Oct 29, 2011
480223d2 Add transpose test.
Oct 29, 2011
487e9c3a Revive fancy_matmul. Fix assert child_iname <= parent_iname condition.
Oct 29, 2011
ca93d84e A few fixes. Some code shifting. Loosen up owed_barriers checking.
Oct 29, 2011
b8c98645 Add user interface for dim length prescription, test for workgroup prescribed too small.
Oct 29, 2011
e3c575dd Some test shuffling.
Oct 29, 2011
5d6b9aea Various fixes, keep insn dependencies as sets.
Oct 30, 2011
5b5cacd1 Implement dimension joining.
Oct 30, 2011
73951d70 Make a better implementation of duplicate_axes().
Oct 30, 2011
724bf1c3 Allow multiple references to a CSE with different indices in each.
Oct 31, 2011
96eca21e Variety of (mostly CSE-related) bug fixes.
Oct 31, 2011
28e5c125 For forced workgroup sizes: check that at least one iname maps to them.
Oct 31, 2011
43cc6dbd Use straight integer division if isl can show the operands are nonnegative.
Oct 31, 2011
36e2516a A zoo of bug fixes.
Oct 31, 2011
c4692f49 Automated testing.
Oct 31, 2011
60dfccd1 Scheduler: add debug mode.
Nov 01, 2011
5da45e0f Scheduler: Fix loop_priority.
Nov 01, 2011
3806860a Make it ok for boostable instructions to not depend on all hw axes.
Nov 01, 2011
1a1ed4dd Make temp. variable shapes tuples of ints (not PwAffs).
Nov 01, 2011
b78f196d Expose loopy.generate_code.
Nov 01, 2011
d6b79855 Be a bit less boring when duplicating inames.
Nov 01, 2011
93793276 Fix print_highlighted_code() if pygments is not installed.
Nov 01, 2011
4f6b046c Fix/add timer to automated tests.
Nov 01, 2011
fde0acae Remove old, unused code from CSE generation.
Nov 01, 2011
b061dbfa Don't try to adjust the storage shape of private variables.
Nov 01, 2011
1c7b8b23 Better error messages when (attempting to) duplicate inames that don't exist.
Nov 02, 2011
a0b5f0ac Minor variable rename.
Nov 02, 2011
0b8736ce An instruction cannot lose iname dependencies by CSE realization.
Nov 02, 2011
2ed5c11c Minor fix to storage shape adjustment.
Nov 02, 2011
b6a68c1e Don't fail in automatic axis assignment if there are no local axes.
Nov 02, 2011
6c3d59be Be less boring in assigning instruction names.
Nov 02, 2011
f0c9980a Only assign axis 0 based on real array access. Document, speed up auto axis assignment.
Nov 02, 2011
3953ff82 New syntax for CSEs and reduction iname duplication.
Nov 02, 2011
be5f165f Add a force flag to tag_dimensions.
Nov 02, 2011
7690a127 Add a switch for annotation in the CCodeMapper.
Nov 02, 2011
04cd0ac6 Generate the shapes of ILP accumulators in the correct type.
Nov 02, 2011
984ca646 Some mucking around with the image ILP test.
Nov 02, 2011
bba4f50a Remove manual test from transpose.
Nov 02, 2011
d25ec07f Add Tim's SEM tests.
Nov 02, 2011
e64156dc Merge branch 'master' of http://git.tiker.net/trees/loopy
Nov 02, 2011
3afae933 Support image arguments in automated tests.
Nov 02, 2011
1eeffebe Fix code generation for floordiv.
Nov 02, 2011
6efeee7c Use static value for lower bound in hw axis setup.
Nov 02, 2011
78ec206c Change mechanism for specifying default tag in CSEs/prefetches.
Nov 02, 2011
f868f6f4 Defer decision on whether variables are local to preprocessing.
Nov 02, 2011
dfb03bb6 Be a bit laxer about PwAffs and Affs.
Nov 02, 2011
2c5b310a Be more rigorous about length-1 axes (don't insert them, don't ignore them).
Nov 02, 2011
38b23406 Fix linalg tests, move some of them towards the automated tests.
Nov 02, 2011
f2887fc1 Merge branch 'master' of git://localhost/loopy
Nov 02, 2011
a2eab227 New syntax in SEM test.
Nov 02, 2011
26107a94 Merge branch 'master' of git://localhost/loopy
Nov 02, 2011
0ec159e1 Make sure domain of map_space is correctly ordered in finding CSE lead index domain.
Nov 03, 2011
b37aaaf7 Deal with @-signs of iname-duplicating reductions in more places.
Nov 03, 2011
d183034b Refactor CSE handling to allow the user to specify the lead expression.
Nov 03, 2011
9956ff05 Find insn iname deps by fixed point iteration. Dot dependency graphing. Schedule improvements.
Nov 03, 2011
9d7b6802 Scheduler: Better notion of 'useful' for boostable instructions.
Nov 03, 2011
0d31c6a8 Make it ok to retag l.auto inames.
Nov 04, 2011
b4019d56 Add variable substitution.
Nov 05, 2011
0a9f6ed2 Add support for constant array arguments.
Nov 05, 2011
0171e5d8 Add simple FEM assembly test without prefetch.
Nov 05, 2011
da8d7c8e Some code rearrangement.
Nov 07, 2011
98395146 Scheduler: be less eager about entering loops, more eager about leaving them.
Nov 07, 2011
30e44717 Barrier Insertion: Be less strict in dep checking when checking for pre-barriers.
Nov 07, 2011
5485e575 Rewrite CSEs in terms of unification templates.
Nov 07, 2011
65312cd5 CSEs: Remove Gaussian elimination/affine eqn solving. Turns out not to be needed.
Nov 07, 2011
d72f97d2 Scheduler debugging.
Nov 08, 2011
bb11a67a A few CSE fixes, plus new error checks.
Nov 08, 2011
b8a175df Add remove_cses().
Nov 08, 2011
38ba6c3c Improve scheduler heuristics. Limit boostability to specific inames.
Nov 08, 2011
803791f3 Use loop nest maps to simplify scheduler.
Nov 08, 2011
618cc903 Pick not just axis 0, but all auto axes by lowest available stride.
Nov 08, 2011
8d248de6 Improve dep barrier insertion before loops.
Nov 08, 2011
715fd101 Test updates.
Nov 08, 2011
8f088865 SEM test updates.
Nov 08, 2011
ee602dc2 Merge git://github.com/inducer/loopy
Nov 08, 2011
6ef2c660 Add edit_code flag to automated testing harness.
Nov 08, 2011

6,183 additional commits have been omitted to prevent performance issues.