Skip to content

[NFC][mlir][OpenMP] Remove mentions of target from generic loop rewrite #124528

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

Merged
merged 1 commit into from
Jan 27, 2025
Merged

[NFC][mlir][OpenMP] Remove mentions of target from generic loop rewrite #124528

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
27 changes: 11 additions & 16 deletions flang/lib/Optimizer/OpenMP/GenericLoopConversion.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -29,11 +29,7 @@ namespace {
class GenericLoopConversionPattern
: public mlir::OpConversionPattern<mlir::omp::LoopOp> {
public:
enum class GenericLoopCombinedInfo {
Standalone,
TargetTeamsLoop,
TargetParallelLoop
};
enum class GenericLoopCombinedInfo { Standalone, TeamsLoop, ParallelLoop };

using mlir::OpConversionPattern<mlir::omp::LoopOp>::OpConversionPattern;

Expand All @@ -55,10 +51,11 @@ class GenericLoopConversionPattern
case GenericLoopCombinedInfo::Standalone:
rewriteStandaloneLoop(loopOp, rewriter);
break;
case GenericLoopCombinedInfo::TargetParallelLoop:
llvm_unreachable("not yet implemented: `parallel loop` direcitve");
case GenericLoopCombinedInfo::ParallelLoop:
llvm_unreachable(
"not yet implemented: Combined `parallel loop` directive");
break;
case GenericLoopCombinedInfo::TargetTeamsLoop:
case GenericLoopCombinedInfo::TeamsLoop:
rewriteToDistributeParallelDo(loopOp, rewriter);
break;
}
Expand All @@ -74,10 +71,10 @@ class GenericLoopConversionPattern
switch (combinedInfo) {
case GenericLoopCombinedInfo::Standalone:
break;
case GenericLoopCombinedInfo::TargetParallelLoop:
case GenericLoopCombinedInfo::ParallelLoop:
return loopOp.emitError(
"not yet implemented: Combined `omp target parallel loop` directive");
case GenericLoopCombinedInfo::TargetTeamsLoop:
"not yet implemented: Combined `parallel loop` directive");
case GenericLoopCombinedInfo::TeamsLoop:
break;
}

Expand All @@ -99,7 +96,7 @@ class GenericLoopConversionPattern
if (!loopOp.getReductionVars().empty())
return todo("reduction");

// TODO For `target teams loop`, check similar constrains to what is checked
// TODO For `teams loop`, check similar constrains to what is checked
// by `TeamsLoopChecker` in SemaOpenMP.cpp.
return mlir::success();
}
Expand All @@ -111,13 +108,11 @@ class GenericLoopConversionPattern
GenericLoopCombinedInfo result = GenericLoopCombinedInfo::Standalone;

if (auto teamsOp = mlir::dyn_cast_if_present<mlir::omp::TeamsOp>(parentOp))
if (mlir::isa_and_present<mlir::omp::TargetOp>(teamsOp->getParentOp()))
result = GenericLoopCombinedInfo::TargetTeamsLoop;
result = GenericLoopCombinedInfo::TeamsLoop;

if (auto parallelOp =
mlir::dyn_cast_if_present<mlir::omp::ParallelOp>(parentOp))
if (mlir::isa_and_present<mlir::omp::TargetOp>(parallelOp->getParentOp()))
result = GenericLoopCombinedInfo::TargetParallelLoop;
result = GenericLoopCombinedInfo::ParallelLoop;

return result;
}
Expand Down
66 changes: 28 additions & 38 deletions flang/test/Transforms/generic-loop-rewriting-todo.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,37 +1,31 @@
// RUN: fir-opt --omp-generic-loop-conversion -verify-diagnostics %s

func.func @_QPtarget_parallel_loop() {
omp.target {
omp.parallel {
%c0 = arith.constant 0 : i32
%c10 = arith.constant 10 : i32
%c1 = arith.constant 1 : i32
// expected-error@below {{not yet implemented: Combined `omp target parallel loop` directive}}
omp.loop {
omp.loop_nest (%arg3) : i32 = (%c0) to (%c10) inclusive step (%c1) {
omp.yield
}
func.func @_QPparallel_loop() {
omp.parallel {
%c0 = arith.constant 0 : i32
%c10 = arith.constant 10 : i32
%c1 = arith.constant 1 : i32
// expected-error@below {{not yet implemented: Combined `parallel loop` directive}}
omp.loop {
omp.loop_nest (%arg3) : i32 = (%c0) to (%c10) inclusive step (%c1) {
omp.yield
}
omp.terminator
}
omp.terminator
}
return
}

func.func @_QPtarget_loop_bind() {
omp.target {
omp.teams {
%c0 = arith.constant 0 : i32
%c10 = arith.constant 10 : i32
%c1 = arith.constant 1 : i32
// expected-error@below {{not yet implemented: Unhandled clause bind in omp.loop operation}}
omp.loop bind(thread) {
omp.loop_nest (%arg3) : i32 = (%c0) to (%c10) inclusive step (%c1) {
omp.yield
}
func.func @_QPloop_bind() {
omp.teams {
%c0 = arith.constant 0 : i32
%c10 = arith.constant 10 : i32
%c1 = arith.constant 1 : i32
// expected-error@below {{not yet implemented: Unhandled clause bind in omp.loop operation}}
omp.loop bind(thread) {
omp.loop_nest (%arg3) : i32 = (%c0) to (%c10) inclusive step (%c1) {
omp.yield
}
omp.terminator
}
omp.terminator
}
Expand All @@ -48,22 +42,18 @@ omp.declare_reduction @add_reduction_i32 : i32 init {
omp.yield(%0 : i32)
}

func.func @_QPtarget_loop_order() {
func.func @_QPloop_order() {
omp.teams {
%c0 = arith.constant 0 : i32
%c10 = arith.constant 10 : i32
%c1 = arith.constant 1 : i32
%sum = fir.alloca i32 {bindc_name = "i", uniq_name = "_QFtest_orderEi"}

omp.target {
omp.teams {
%c0 = arith.constant 0 : i32
%c10 = arith.constant 10 : i32
%c1 = arith.constant 1 : i32
%sum = fir.alloca i32 {bindc_name = "i", uniq_name = "_QFtest_orderEi"}

// expected-error@below {{not yet implemented: Unhandled clause reduction in omp.loop operation}}
omp.loop reduction(@add_reduction_i32 %sum -> %arg2 : !fir.ref<i32>) {
omp.loop_nest (%arg3) : i32 = (%c0) to (%c10) inclusive step (%c1) {
omp.yield
}
// expected-error@below {{not yet implemented: Unhandled clause reduction in omp.loop operation}}
omp.loop reduction(@add_reduction_i32 %sum -> %arg2 : !fir.ref<i32>) {
omp.loop_nest (%arg3) : i32 = (%c0) to (%c10) inclusive step (%c1) {
omp.yield
}
omp.terminator
}
omp.terminator
}
Expand Down
53 changes: 23 additions & 30 deletions flang/test/Transforms/generic-loop-rewriting.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,55 +1,48 @@
// RUN: fir-opt --omp-generic-loop-conversion %s | FileCheck %s

omp.private {type = private} @_QFtarget_teams_loopEi_private_ref_i32 : !fir.ref<i32> alloc {
omp.private {type = private} @_QFteams_loopEi_private_ref_i32 : !fir.ref<i32> alloc {
^bb0(%arg0: !fir.ref<i32>):
omp.yield(%arg0 : !fir.ref<i32>)
}

func.func @_QPtarget_teams_loop() {
func.func @_QPteams_loop() {
%i = fir.alloca i32
%i_map = omp.map.info var_ptr(%i : !fir.ref<i32>, i32) map_clauses(implicit, exit_release_or_enter_alloc) capture(ByCopy) -> !fir.ref<i32> {name = "i"}
omp.target map_entries(%i_map -> %arg0 : !fir.ref<i32>) {
omp.teams {
%c0 = arith.constant 0 : i32
%c10 = arith.constant 10 : i32
%c1 = arith.constant 1 : i32
omp.loop private(@_QFtarget_teams_loopEi_private_ref_i32 %arg0 -> %arg2 : !fir.ref<i32>) {
omp.loop_nest (%arg3) : i32 = (%c0) to (%c10) inclusive step (%c1) {
fir.store %arg3 to %arg2 : !fir.ref<i32>
omp.yield
}
omp.teams {
%c0 = arith.constant 0 : i32
%c10 = arith.constant 10 : i32
%c1 = arith.constant 1 : i32
omp.loop private(@_QFteams_loopEi_private_ref_i32 %i -> %arg2 : !fir.ref<i32>) {
omp.loop_nest (%arg3) : i32 = (%c0) to (%c10) inclusive step (%c1) {
fir.store %arg3 to %arg2 : !fir.ref<i32>
omp.yield
}
omp.terminator
}
omp.terminator
}
return
}

// CHECK-LABEL: func.func @_QPtarget_teams_loop
// CHECK: omp.target map_entries(
// CHECK-SAME: %{{.*}} -> %[[I_ARG:[^[:space:]]+]] : {{.*}}) {
//
// CHECK: omp.teams {
// CHECK-LABEL: func.func @_QPteams_loop
// CHECK: %[[I:.*]] = fir.alloca i32
// CHECK: omp.teams {
//
// TODO we probably need to move the `loop_nest` bounds ops from the `teams`
// region to the `parallel` region to avoid making these values `shared`. We can
// find the backward slices of these bounds that are within the `teams` region
// and move these slices to the `parallel` op.

// CHECK: %[[LB:.*]] = arith.constant 0 : i32
// CHECK: %[[UB:.*]] = arith.constant 10 : i32
// CHECK: %[[STEP:.*]] = arith.constant 1 : i32
// CHECK: %[[LB:.*]] = arith.constant 0 : i32
// CHECK: %[[UB:.*]] = arith.constant 10 : i32
// CHECK: %[[STEP:.*]] = arith.constant 1 : i32
//
// CHECK: omp.parallel private(@{{.*}} %[[I_ARG]]
// CHECK-SAME: -> %[[I_PRIV_ARG:[^[:space:]]+]] : !fir.ref<i32>) {
// CHECK: omp.distribute {
// CHECK: omp.wsloop {
// CHECK: omp.parallel private(@{{.*}} %[[I]]
// CHECK-SAME: -> %[[I_PRIV_ARG:[^[:space:]]+]] : !fir.ref<i32>) {
// CHECK: omp.distribute {
// CHECK: omp.wsloop {
//
// CHECK: omp.loop_nest (%{{.*}}) : i32 =
// CHECK-SAME: (%[[LB]]) to (%[[UB]]) inclusive step (%[[STEP]]) {
// CHECK: fir.store %{{.*}} to %[[I_PRIV_ARG]] : !fir.ref<i32>
// CHECK: }
// CHECK: omp.loop_nest (%{{.*}}) : i32 =
// CHECK-SAME: (%[[LB]]) to (%[[UB]]) inclusive step (%[[STEP]]) {
// CHECK: fir.store %{{.*}} to %[[I_PRIV_ARG]] : !fir.ref<i32>
// CHECK: }
// CHECK: }
// CHECK: }
Expand Down
Loading