def test_k_cache_detection_basic():
testee = VerticalLoopFactory(
loop_order=LoopOrder.FORWARD,
sections__0__horizontal_executions__0__body=[
AssignStmtFactory(
left__name="foo",
right__name="foo",
right__offset__k=1,
),
AssignStmtFactory(
left__name="bar",
right__name="foo",
right__offset__k=-1,
),
AssignStmtFactory(
left__name="baz",
right__name="baz",
right__offset__i=1,
right__offset__k=1,
),
AssignStmtFactory(
left__name="foo",
right__name="baz",
right__offset__j=1,
right__offset__k=-1,
),
],
)
transformed = KCacheDetection().visit(testee)
assert {c.name for c in transformed.caches} == {"foo"}
assert all(isinstance(cache, KCache) for cache in transformed.caches)
def _optimize_oir(self, oir):
oir = GreedyMerging().visit(oir)
oir = AdjacentLoopMerging().visit(oir)
oir = LocalTemporariesToScalars().visit(oir)
oir = WriteBeforeReadTemporariesToScalars().visit(oir)
oir = OnTheFlyMerging().visit(oir)
oir = NoFieldAccessPruning().visit(oir)
oir = IJCacheDetection().visit(oir)
oir = KCacheDetection().visit(oir)
oir = PruneKCacheFills().visit(oir)
oir = PruneKCacheFlushes().visit(oir)
return oir
def _optimize_oir(self, oir):
oir = optimize_horizontal_executions(oir, GraphMerging)
oir = AdjacentLoopMerging().visit(oir)
oir = LocalTemporariesToScalars().visit(oir)
oir = WriteBeforeReadTemporariesToScalars().visit(oir)
oir = OnTheFlyMerging().visit(oir)
oir = MaskStmtMerging().visit(oir)
oir = NoFieldAccessPruning().visit(oir)
oir = IJCacheDetection().visit(oir)
oir = KCacheDetection().visit(oir)
oir = PruneKCacheFills().visit(oir)
oir = PruneKCacheFlushes().visit(oir)
return oir
def test_k_cache_detection_single_access_point():
testee = VerticalLoopFactory(
loop_order=LoopOrder.FORWARD,
sections__0__horizontal_executions=[
HorizontalExecutionFactory(
body=[AssignStmtFactory(left__name="foo", right__name="bar")]),
HorizontalExecutionFactory(body=[
AssignStmtFactory(
left__name="bar", right__name="baz", right__offset__k=1)
]),
],
)
transformed = KCacheDetection().visit(testee)
assert not transformed.caches