[flang][OpenMP] Map basic local specifiers to private clauses (#142735)
Starts the effort to map `do concurrent` locality specifiers to OpenMP clauses. This PR adds support for basic specifiers (no `init` or `copy` regions yet).
This commit is contained in:
@@ -7,9 +7,11 @@
|
||||
//===----------------------------------------------------------------------===//
|
||||
|
||||
#include "flang/Optimizer/Builder/FIRBuilder.h"
|
||||
#include "flang/Optimizer/Builder/Todo.h"
|
||||
#include "flang/Optimizer/Dialect/FIROps.h"
|
||||
#include "flang/Optimizer/OpenMP/Passes.h"
|
||||
#include "flang/Optimizer/OpenMP/Utils.h"
|
||||
#include "flang/Support/OpenMP-utils.h"
|
||||
#include "mlir/Analysis/SliceAnalysis.h"
|
||||
#include "mlir/Dialect/OpenMP/OpenMPDialect.h"
|
||||
#include "mlir/IR/IRMapping.h"
|
||||
@@ -308,10 +310,47 @@ private:
|
||||
fir::DoConcurrentLoopOp loop, mlir::IRMapping &mapper,
|
||||
const mlir::omp::LoopNestOperands &clauseOps,
|
||||
bool isComposite) const {
|
||||
mlir::omp::WsloopOperands wsloopClauseOps;
|
||||
|
||||
auto wsloopOp = rewriter.create<mlir::omp::WsloopOp>(loop.getLoc());
|
||||
// For `local` (and `local_init`) opernads, emit corresponding `private`
|
||||
// clauses and attach these clauses to the workshare loop.
|
||||
if (!loop.getLocalOperands().empty())
|
||||
for (auto [op, sym, arg] : llvm::zip_equal(
|
||||
loop.getLocalOperands(),
|
||||
loop.getLocalSymsAttr().getAsRange<mlir::SymbolRefAttr>(),
|
||||
loop.getRegionLocalArgs())) {
|
||||
auto localizer = mlir::SymbolTable::lookupNearestSymbolFrom<
|
||||
fir::LocalitySpecifierOp>(loop, sym);
|
||||
if (localizer.getLocalitySpecifierType() ==
|
||||
fir::LocalitySpecifierType::LocalInit)
|
||||
TODO(localizer.getLoc(),
|
||||
"local_init conversion is not supported yet");
|
||||
|
||||
if (!localizer.getInitRegion().empty())
|
||||
TODO(localizer.getLoc(),
|
||||
"non-empty `init` regions are not supported yet");
|
||||
|
||||
auto oldIP = rewriter.saveInsertionPoint();
|
||||
rewriter.setInsertionPointAfter(localizer);
|
||||
auto privatizer = rewriter.create<mlir::omp::PrivateClauseOp>(
|
||||
localizer.getLoc(), sym.getLeafReference().str() + ".omp",
|
||||
localizer.getTypeAttr().getValue(),
|
||||
mlir::omp::DataSharingClauseType::Private);
|
||||
rewriter.restoreInsertionPoint(oldIP);
|
||||
|
||||
wsloopClauseOps.privateVars.push_back(op);
|
||||
wsloopClauseOps.privateSyms.push_back(
|
||||
mlir::SymbolRefAttr::get(privatizer));
|
||||
}
|
||||
|
||||
auto wsloopOp =
|
||||
rewriter.create<mlir::omp::WsloopOp>(loop.getLoc(), wsloopClauseOps);
|
||||
wsloopOp.setComposite(isComposite);
|
||||
rewriter.createBlock(&wsloopOp.getRegion());
|
||||
|
||||
Fortran::common::openmp::EntryBlockArgs wsloopArgs;
|
||||
wsloopArgs.priv.vars = wsloopClauseOps.privateVars;
|
||||
Fortran::common::openmp::genEntryBlock(rewriter, wsloopArgs,
|
||||
wsloopOp.getRegion());
|
||||
|
||||
auto loopNestOp =
|
||||
rewriter.create<mlir::omp::LoopNestOp>(loop.getLoc(), clauseOps);
|
||||
@@ -324,6 +363,18 @@ private:
|
||||
rewriter.setInsertionPointToEnd(&loopNestOp.getRegion().back());
|
||||
rewriter.create<mlir::omp::YieldOp>(loop->getLoc());
|
||||
|
||||
// `local` region arguments are transferred/cloned from the `do concurrent`
|
||||
// loop to the loopnest op when the region is cloned above. Instead, these
|
||||
// region arguments should be on the workshare loop's region.
|
||||
for (auto [wsloopArg, loopNestArg] :
|
||||
llvm::zip_equal(wsloopOp.getRegion().getArguments(),
|
||||
loopNestOp.getRegion().getArguments().drop_front(
|
||||
clauseOps.loopLowerBounds.size())))
|
||||
rewriter.replaceAllUsesWith(loopNestArg, wsloopArg);
|
||||
|
||||
for (unsigned i = 0; i < loop.getLocalVars().size(); ++i)
|
||||
loopNestOp.getRegion().eraseArgument(clauseOps.loopLowerBounds.size());
|
||||
|
||||
return loopNestOp;
|
||||
}
|
||||
|
||||
|
||||
@@ -0,0 +1,48 @@
|
||||
// Tests mapping `local` locality specifier to `private` clauses for a simple
|
||||
// case (not `init` or `copy` regions).
|
||||
|
||||
// RUN: fir-opt --omp-do-concurrent-conversion="map-to=host" %s | FileCheck %s
|
||||
|
||||
fir.local {type = local} @_QFlocal_spec_translationElocal_var_private_f32 : f32
|
||||
|
||||
func.func @_QPlocal_spec_translation() {
|
||||
%3 = fir.alloca f32 {bindc_name = "local_var", uniq_name = "_QFlocal_spec_translationElocal_var"}
|
||||
%4:2 = hlfir.declare %3 {uniq_name = "_QFlocal_spec_translationElocal_var"} : (!fir.ref<f32>) -> (!fir.ref<f32>, !fir.ref<f32>)
|
||||
|
||||
%c4_i32 = arith.constant 4 : index
|
||||
%c11_i32 = arith.constant 11 : index
|
||||
%c1 = arith.constant 1 : index
|
||||
|
||||
fir.do_concurrent {
|
||||
%7 = fir.alloca i32 {bindc_name = "i"}
|
||||
%8:2 = hlfir.declare %7 {uniq_name = "_QFlocal_spec_translationEi"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
|
||||
|
||||
fir.do_concurrent.loop (%arg0) = (%c4_i32) to (%c11_i32) step (%c1)
|
||||
local(@_QFlocal_spec_translationElocal_var_private_f32 %4#0 -> %arg1 : !fir.ref<f32>) {
|
||||
%9 = fir.convert %arg0 : (index) -> i32
|
||||
fir.store %9 to %8#0 : !fir.ref<i32>
|
||||
|
||||
%10:2 = hlfir.declare %arg1 {uniq_name = "_QFlocal_spec_translationElocal_var"} : (!fir.ref<f32>) -> (!fir.ref<f32>, !fir.ref<f32>)
|
||||
%cst = arith.constant 4.200000e+01 : f32
|
||||
hlfir.assign %cst to %10#0 : f32, !fir.ref<f32>
|
||||
}
|
||||
}
|
||||
return
|
||||
}
|
||||
|
||||
// CHECK: omp.private {type = private} @[[PRIVATIZER:.*local_spec_translationElocal_var.*.omp]] : f32
|
||||
|
||||
// CHECK: func.func @_QPlocal_spec_translation
|
||||
// CHECK: %[[LOCAL_VAR:.*]] = fir.alloca f32 {bindc_name = "local_var", {{.*}}}
|
||||
// CHECK: %[[LOCAL_VAR_DECL:.*]]:2 = hlfir.declare %[[LOCAL_VAR]]
|
||||
// CHECK: omp.parallel {
|
||||
// CHECK: omp.wsloop private(@[[PRIVATIZER]] %[[LOCAL_VAR_DECL]]#0 -> %[[LOCAL_ARG:.*]] : !fir.ref<f32>) {
|
||||
// CHECK: omp.loop_nest {{.*}} {
|
||||
// CHECK: %[[PRIV_DECL:.*]]:2 = hlfir.declare %[[LOCAL_ARG]]
|
||||
// CHECK: %[[C42:.*]] = arith.constant
|
||||
// CHECK: hlfir.assign %[[C42]] to %[[PRIV_DECL]]#0
|
||||
// CHECK: omp.yield
|
||||
// CHECK: }
|
||||
// CHECK: }
|
||||
// CHECK: omp.terminator
|
||||
// CHECK: }
|
||||
Reference in New Issue
Block a user