Allocation workaround for unused write-only dispatch results. (#13748)
The stream allocation pass is currently just looking at the top level of
each execution region and values nested under concurrency regions with
no users don't get allocated any memory. Since there are cases where
dispatch region formation can produce unused write-only results we need
to handle these even if suboptimal (wasting an allocation for
/dev/null).
Fixes #13729.
diff --git a/compiler/src/iree/compiler/Dialect/Stream/Transforms/ScheduleAllocation.cpp b/compiler/src/iree/compiler/Dialect/Stream/Transforms/ScheduleAllocation.cpp
index 51adb91..8a74652 100644
--- a/compiler/src/iree/compiler/Dialect/Stream/Transforms/ScheduleAllocation.cpp
+++ b/compiler/src/iree/compiler/Dialect/Stream/Transforms/ScheduleAllocation.cpp
@@ -199,6 +199,26 @@
// Compute ranges for all values independently (ignoring aliasing).
for (auto &op : *streamBlock) {
int start = opOrdering[&op];
+ if (auto concurrentOp = dyn_cast<IREE::Stream::AsyncConcurrentOp>(op)) {
+ // HACK: allocation planning currently only works on the top-level
+ // execute op but sometimes we need to allocate locals inside of
+ // concurrent regions. The real fix here is to make allocation planning
+ // handle arbitrary nesting but for now we do a quick walk through the
+ // regions to see if there are any locals that need to be marked live for
+ // the duration of the region.
+ concurrentOp.walk([&](Operation *op) {
+ for (auto value : op->getResults()) {
+ if (!value.getType().isa<IREE::Stream::ResourceType>()) continue;
+ if (!value.use_empty()) continue;
+ LivenessInterval interval;
+ interval.start = start;
+ interval.end = start;
+ interval.value = value;
+ interval.ordinal = -1;
+ valueIntervals[value] = interval;
+ }
+ });
+ }
for (auto value : op.getResults()) {
if (!value.getType().isa<IREE::Stream::ResourceType>()) continue;
LivenessInterval interval;
diff --git a/compiler/src/iree/compiler/Dialect/Stream/Transforms/test/schedule_allocation.mlir b/compiler/src/iree/compiler/Dialect/Stream/Transforms/test/schedule_allocation.mlir
index 86193eb..ca85637 100644
--- a/compiler/src/iree/compiler/Dialect/Stream/Transforms/test/schedule_allocation.mlir
+++ b/compiler/src/iree/compiler/Dialect/Stream/Transforms/test/schedule_allocation.mlir
@@ -471,6 +471,45 @@
// -----
+// Tests that unused dispatch results nested in concurrent regions are still
+// allocated memory.
+
+// CHECK-LABEL: @applyAsyncDispatchUnusedOp
+// CHECK-SAME: (%[[OPERAND:.+]]: !stream.resource<transient>, %[[SIZE:.+]]: index, %[[OFFSET:.+]]: index, %[[END:.+]]: index, %[[LENGTH:.+]]: index)
+func.func @applyAsyncDispatchUnusedOp(%operand: !stream.resource<transient>, %size: index, %offset: index, %end: index, %length: index) {
+ %c0 = arith.constant 0 : index
+ %c1 = arith.constant 1 : index
+ %c4 = arith.constant 4 : index
+ // CHECK: %[[PACK:.+]]:2 = stream.resource.pack
+ // CHECK: %[[ALLOCA:.+]], %[[ALLOCA_TIMEPOINT:.+]] = stream.resource.alloca uninitialized : !stream.resource<transient>{%[[PACK]]#0}
+ // CHECK: %[[TIMEPOINT:.+]] = stream.cmd.execute
+ // CHECK-SAME: await(%[[ALLOCA_TIMEPOINT]])
+ // CHECK-SAME: with(%[[OPERAND]] as %[[OPERAND_CAPTURE:.+]]: !stream.resource<transient>{%[[SIZE]]},
+ // CHECK-SAME: %[[ALLOCA]] as %[[ALLOCA_CAPTURE:.+]]: !stream.resource<transient>{%[[PACK]]#0})
+ %result, %result_timepoint = stream.async.execute with(%operand as %capture: !stream.resource<transient>{%size}) -> (%operand as !stream.resource<transient>{%size}) {
+ // CHECK: stream.cmd.concurrent
+ %concurrent = stream.async.concurrent with(%capture as %concurrent_capture: !stream.resource<transient>{%size}) -> (%capture as !stream.resource<transient>{%size}) {
+ // CHECK-NEXT: stream.cmd.dispatch @executable::@dispatch[%c1, %c1, %c1](%c4 : index) {
+ // CHECK-NEXT: rw %[[OPERAND_CAPTURE]][%[[OFFSET]] for %[[LENGTH]]] : !stream.resource<transient>{%[[SIZE]]},
+ // CHECK-NEXT: wo %[[ALLOCA_CAPTURE]][%[[PACK]]#1 for %[[SIZE]]] : !stream.resource<transient>{%[[PACK]]#0}
+ // CHECK-NEXT: }
+ %0:2 = stream.async.dispatch @executable::@dispatch[%c1, %c1, %c1](%concurrent_capture[%offset to %end for %length], %c4) : (!stream.resource<transient>{%size}, index) -> (%concurrent_capture{%size}, !stream.resource<transient>{%size})
+ // NOTE: %0#1 is unused.
+ stream.yield %0#0 : !stream.resource<transient>{%size}
+ }
+ stream.yield %concurrent : !stream.resource<transient>{%size}
+ } => !stream.timepoint
+ // CHECK: %[[DEALLOCA:.+]] = stream.resource.dealloca await(%[[TIMEPOINT]]) => %[[ALLOCA]]
+ // CHECK: %[[JOIN:.+]] = stream.timepoint.join max(%[[DEALLOCA]], %[[TIMEPOINT]])
+ // CHECK: util.optimization_barrier %[[JOIN]]
+ util.optimization_barrier %result_timepoint : !stream.timepoint
+ // CHECK: util.optimization_barrier %[[OPERAND]]
+ util.optimization_barrier %result : !stream.resource<transient>
+ return
+}
+
+// -----
+
// CHECK: stream.cmd.func private @asyncExtern(%arg0[%arg1 for %arg2]: !stream.resource<*>, %arg3: index, %arg4[%arg5 for %arg6]: !stream.resource<*>)
stream.async.func private @asyncExtern(%arg0: !stream.resource<*>, %arg1: index) -> (%arg0, !stream.resource<*>)
