[flang-commits] [flang] [mlir] [Flang] [OpenMP] [Semantics] [MLIR] [Lowering] Add lowering support for IS_DEVICE_PTR and HAS_DEVICE_ADDR clauses on OMP TARGET directive. (PR #74187)

Raghu Maddhipatla via flang-commits flang-commits at lists.llvm.org
Mon Mar 4 15:17:13 PST 2024


https://github.com/raghavendhra updated https://github.com/llvm/llvm-project/pull/74187

>From 11105ed97e31ecd5e28924157f47eb4ea250174e Mon Sep 17 00:00:00 2001
From: Raghu Maddhipatla <Raghu.Maddhipatla at amd.com>
Date: Wed, 25 Oct 2023 18:19:08 -0500
Subject: [PATCH 1/4] [Flang] [OpenMP] [Semantics] [MLIR] [Lowering] Add
 lowering support for IS_DEVICE_PTR and HAS_DEVICE_ADDR clauses on OMP TARGET
 directive.

---
 flang/lib/Lower/OpenMP/ClauseProcessor.cpp    | 29 +++++++++++++
 flang/lib/Lower/OpenMP/ClauseProcessor.h      | 14 +++++++
 flang/lib/Lower/OpenMP/OpenMP.cpp             | 16 ++++++--
 flang/test/Lower/OpenMP/FIR/target.f90        | 41 ++++++++++++++++++-
 mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td | 18 ++++++--
 mlir/test/Dialect/OpenMP/invalid.mlir         |  2 +-
 mlir/test/Dialect/OpenMP/ops.mlir             |  8 ++--
 7 files changed, 115 insertions(+), 13 deletions(-)

diff --git a/flang/lib/Lower/OpenMP/ClauseProcessor.cpp b/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
index a41f8312a28c9e..17a93057bd10a2 100644
--- a/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
+++ b/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
@@ -1031,6 +1031,35 @@ bool ClauseProcessor::processUseDevicePtr(
                            useDeviceLocs, useDeviceSymbols);
       });
 }
+
+bool ClauseProcessor::processIsDevicePtr(
+    llvm::SmallVectorImpl<mlir::Value> &operands,
+    llvm::SmallVectorImpl<mlir::Type> &isDeviceTypes,
+    llvm::SmallVectorImpl<mlir::Location> &isDeviceLocs,
+    llvm::SmallVectorImpl<const Fortran::semantics::Symbol *> &isDeviceSymbols)
+    const {
+  return findRepeatableClause<ClauseTy::IsDevicePtr>(
+      [&](const ClauseTy::IsDevicePtr *devPtrClause,
+          const Fortran::parser::CharBlock &) {
+        addUseDeviceClause(converter, devPtrClause->v, operands, isDeviceTypes,
+                           isDeviceLocs, isDeviceSymbols);
+      });
+}
+
+bool ClauseProcessor::processHasDeviceAddr(
+    llvm::SmallVectorImpl<mlir::Value> &operands,
+    llvm::SmallVectorImpl<mlir::Type> &isDeviceTypes,
+    llvm::SmallVectorImpl<mlir::Location> &isDeviceLocs,
+    llvm::SmallVectorImpl<const Fortran::semantics::Symbol *> &isDeviceSymbols)
+    const {
+  return findRepeatableClause<ClauseTy::HasDeviceAddr>(
+      [&](const ClauseTy::HasDeviceAddr *devAddrClause,
+          const Fortran::parser::CharBlock &) {
+        addUseDeviceClause(converter, devAddrClause->v, operands, isDeviceTypes,
+                           isDeviceLocs, isDeviceSymbols);
+      });
+}
+
 } // namespace omp
 } // namespace lower
 } // namespace Fortran
diff --git a/flang/lib/Lower/OpenMP/ClauseProcessor.h b/flang/lib/Lower/OpenMP/ClauseProcessor.h
index 11aff0be25053d..7e36c84c4662fe 100644
--- a/flang/lib/Lower/OpenMP/ClauseProcessor.h
+++ b/flang/lib/Lower/OpenMP/ClauseProcessor.h
@@ -143,6 +143,20 @@ class ClauseProcessor {
                       llvm::SmallVectorImpl<const Fortran::semantics::Symbol *>
                           &useDeviceSymbols) const;
 
+  bool
+  processIsDevicePtr(llvm::SmallVectorImpl<mlir::Value> &operands,
+                     llvm::SmallVectorImpl<mlir::Type> &isDeviceTypes,
+                     llvm::SmallVectorImpl<mlir::Location> &isDeviceLocs,
+                     llvm::SmallVectorImpl<const Fortran::semantics::Symbol *>
+                     &isDeviceSymbols) const;
+
+  bool
+  processHasDeviceAddr(llvm::SmallVectorImpl<mlir::Value> &operands,
+                       llvm::SmallVectorImpl<mlir::Type> &isDeviceTypes,
+                       llvm::SmallVectorImpl<mlir::Location> &isDeviceLocs,
+                       llvm::SmallVectorImpl<const Fortran::semantics::Symbol *>
+                       &isDeviceSymbols) const;
+
   template <typename T>
   bool processMotionClauses(Fortran::lower::StatementContext &stmtCtx,
                             llvm::SmallVectorImpl<mlir::Value> &mapOperands);
diff --git a/flang/lib/Lower/OpenMP/OpenMP.cpp b/flang/lib/Lower/OpenMP/OpenMP.cpp
index 90fc1f80f57ab7..9d4ddaf36db159 100644
--- a/flang/lib/Lower/OpenMP/OpenMP.cpp
+++ b/flang/lib/Lower/OpenMP/OpenMP.cpp
@@ -1052,6 +1052,10 @@ genTargetOp(Fortran::lower::AbstractConverter &converter,
   llvm::SmallVector<mlir::Type> mapSymTypes;
   llvm::SmallVector<mlir::Location> mapSymLocs;
   llvm::SmallVector<const Fortran::semantics::Symbol *> mapSymbols;
+  llvm::SmallVector<mlir::Value> devicePtrOperands, deviceAddrOperands;
+  llvm::SmallVector<mlir::Type> useDeviceTypes;
+  llvm::SmallVector<mlir::Location> useDeviceLocs;
+  llvm::SmallVector<const Fortran::semantics::Symbol *> useDeviceSymbols;
 
   ClauseProcessor cp(converter, semaCtx, clauseList);
   cp.processIf(Fortran::parser::OmpIfClause::DirectiveNameModifier::Target,
@@ -1062,11 +1066,12 @@ genTargetOp(Fortran::lower::AbstractConverter &converter,
   cp.processNowait(nowaitAttr);
   cp.processMap(currentLocation, directive, stmtCtx, mapOperands, &mapSymTypes,
                 &mapSymLocs, &mapSymbols);
-
+  cp.processIsDevicePtr(devicePtrOperands, useDeviceTypes, useDeviceLocs,
+                        useDeviceSymbols);
+  cp.processHasDeviceAddr(deviceAddrOperands, useDeviceTypes, useDeviceLocs,
+                          useDeviceSymbols);
   cp.processTODO<Fortran::parser::OmpClause::Private,
                  Fortran::parser::OmpClause::Firstprivate,
-                 Fortran::parser::OmpClause::IsDevicePtr,
-                 Fortran::parser::OmpClause::HasDeviceAddr,
                  Fortran::parser::OmpClause::Reduction,
                  Fortran::parser::OmpClause::InReduction,
                  Fortran::parser::OmpClause::Allocate,
@@ -1150,7 +1155,8 @@ genTargetOp(Fortran::lower::AbstractConverter &converter,
           ? nullptr
           : mlir::ArrayAttr::get(converter.getFirOpBuilder().getContext(),
                                  dependTypeOperands),
-      dependOperands, nowaitAttr, mapOperands);
+      dependOperands, nowaitAttr, devicePtrOperands,
+      deviceAddrOperands, mapOperands);
 
   genBodyOfTargetOp(converter, semaCtx, eval, genNested, targetOp, mapSymTypes,
                     mapSymLocs, mapSymbols, currentLocation);
@@ -1757,6 +1763,8 @@ genOMP(Fortran::lower::AbstractConverter &converter,
         !std::get_if<Fortran::parser::OmpClause::Map>(&clause.u) &&
         !std::get_if<Fortran::parser::OmpClause::UseDevicePtr>(&clause.u) &&
         !std::get_if<Fortran::parser::OmpClause::UseDeviceAddr>(&clause.u) &&
+        !std::get_if<Fortran::parser::OmpClause::IsDevicePtr>(&clause.u) &&
+        !std::get_if<Fortran::parser::OmpClause::HasDeviceAddr>(&clause.u) &&
         !std::get_if<Fortran::parser::OmpClause::ThreadLimit>(&clause.u) &&
         !std::get_if<Fortran::parser::OmpClause::NumTeams>(&clause.u)) {
       TODO(clauseLocation, "OpenMP Block construct clause");
diff --git a/flang/test/Lower/OpenMP/FIR/target.f90 b/flang/test/Lower/OpenMP/FIR/target.f90
index 06772771647de8..2e03e8064d80ce 100644
--- a/flang/test/Lower/OpenMP/FIR/target.f90
+++ b/flang/test/Lower/OpenMP/FIR/target.f90
@@ -506,4 +506,43 @@ subroutine omp_target_parallel_do
    !CHECK: omp.terminator
    !CHECK: }
    !$omp end target parallel do
- end subroutine omp_target_parallel_do
+end subroutine omp_target_parallel_do
+
+!===============================================================================
+! Target `is_device_ptr` clause
+!===============================================================================
+
+!CHECK-LABEL: func.func @_QPomp_target_is_device_ptr() {
+subroutine omp_target_is_device_ptr
+   use iso_c_binding, only : c_ptr, c_loc
+   !CHECK: %[[DEV_PTR:.*]] = fir.alloca !fir.type<_QM__fortran_builtinsT__builtin_c_ptr{__address:i64}> {bindc_name = "a", uniq_name = "_QFomp_target_is_device_ptrEa"}
+   type(c_ptr) :: a
+   !CHECK %[[VAL_0:.*]] = fir.alloca i32 {bindc_name = "b", fir.target, uniq_name = "_QFomp_target_is_device_ptrEb"}
+   integer, target :: b
+   !CHECK: %[[MAP_0:.*]] = omp.map_info var_ptr(%[[DEV_PTR:.*]] : !fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_c_ptr{__address:i64}>>, !fir.type<_QM__fortran_builtinsT__builtin_c_ptr{__address:i64}>) map_clauses(tofrom) capture(ByRef) -> !fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_c_ptr{__address:i64}>> {name = "a"}
+   !CHECK: %[[MAP_1:.*]] = omp.map_info var_ptr(%[[VAL_0:.*]] : !fir.ref<i32>, i32)   map_clauses(tofrom) capture(ByRef) -> !fir.ref<i32> {name = "b"}
+   !CHECK: omp.target is_device_ptr(%[[DEV_PTR:.*]] : !fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_c_ptr{__address:i64}>>) map_entries(%[[MAP_0:.*]], %[[MAP_1:.*]] : !fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_c_ptr{__address:i64}>>, !fir.ref<i32>, !fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_c_ptr{__address:i64}>>) {
+   !$omp target map(tofrom: a,b) is_device_ptr(a)
+      !CHECK: {{.*}} = fir.coordinate_of %[[DEV_PTR:.*]], {{.*}} : (!fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_c_ptr{__address:i64}>>, !fir.field) -> !fir.ref<i64>
+      a = c_loc(b)
+   !CHECK: omp.terminator
+   !$omp end target
+   !CHECK: }
+end subroutine omp_target_is_device_ptr
+
+ !===============================================================================
+ ! Target `has_device_addr` clause
+ !===============================================================================
+
+ !CHECK-LABEL: func.func @_QPomp_target_has_device_addr() {
+ subroutine omp_target_has_device_addr
+   integer, pointer :: a
+   !CHECK: %[[VAL_0:.*]] = fir.alloca !fir.box<!fir.ptr<i32>> {bindc_name = "a", uniq_name = "_QFomp_target_has_device_addrEa"}
+   !CHECK: omp.target has_device_addr(%[[VAL_0:.*]] : !fir.ref<!fir.box<!fir.ptr<i32>>>) map_entries({{.*}} -> {{.*}}, {{.*}} -> {{.*}} : !fir.llvm_ptr<!fir.ref<i32>>, !fir.ref<!fir.box<!fir.ptr<i32>>>) {
+   !$omp target has_device_addr(a)
+   !CHECK: {{.*}} = fir.load %[[VAL_0:.*]] : !fir.ref<!fir.box<!fir.ptr<i32>>>
+      a = 10
+   !CHECK: omp.terminator
+   !$omp end target
+   !CHECK: }
+end subroutine omp_target_has_device_addr
diff --git a/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td b/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
index 39ff49f0ccf541..9d25fa73b9c490 100644
--- a/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
+++ b/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
@@ -1619,14 +1619,23 @@ def TargetOp : OpenMP_Op<"target",[IsolatedFromAbove, MapClauseOwningOpInterface
 
     The optional $thread_limit specifies the limit on the number of threads
 
-    The optional $nowait elliminates the implicit barrier so the parent task can make progress
+    The optional $nowait eliminates the implicit barrier so the parent task can make progress
     even if the target task is not yet completed.
 
     The `depends` and `depend_vars` arguments are variadic lists of values
     that specify the dependencies of this particular target task in relation to
     other tasks.
 
-    TODO:  is_device_ptr, defaultmap, in_reduction
+    The optional $is_device_ptr indicates list items are device pointers
+
+    The optional $has_device_addr indicates that list items already have device
+    addresses, so may be directly accessed from target device. May include array
+    sections.
+
+    The optional $map_operands maps data from the task’s environment to the
+    device environment.
+
+    TODO:  defaultmap, in_reduction
 
   }];
 
@@ -1636,8 +1645,9 @@ def TargetOp : OpenMP_Op<"target",[IsolatedFromAbove, MapClauseOwningOpInterface
                        OptionalAttr<TaskDependArrayAttr>:$depends,
                        Variadic<OpenMP_PointerLikeType>:$depend_vars,
                        UnitAttr:$nowait,
+                       Variadic<OpenMP_PointerLikeType>:$is_device_ptr,
+                       Variadic<OpenMP_PointerLikeType>:$has_device_addr,
                        Variadic<AnyType>:$map_operands);
-
   let regions = (region AnyRegion:$region);
 
   let assemblyFormat = [{
@@ -1645,6 +1655,8 @@ def TargetOp : OpenMP_Op<"target",[IsolatedFromAbove, MapClauseOwningOpInterface
     | `device` `(` $device `:` type($device) `)`
     | `thread_limit` `(` $thread_limit `:` type($thread_limit) `)`
     | `nowait` $nowait
+    | `is_device_ptr` `(` $is_device_ptr `:` type($is_device_ptr) `)`
+    | `has_device_addr` `(` $has_device_addr `:` type($has_device_addr) `)`
     | `map_entries` `(` custom<MapEntries>($map_operands, type($map_operands)) `)`
     | `depend` `(` custom<DependVarList>($depend_vars, type($depend_vars), $depends) `)`
     ) $region attr-dict
diff --git a/mlir/test/Dialect/OpenMP/invalid.mlir b/mlir/test/Dialect/OpenMP/invalid.mlir
index d9261b89e24e3a..cae80390da1457 100644
--- a/mlir/test/Dialect/OpenMP/invalid.mlir
+++ b/mlir/test/Dialect/OpenMP/invalid.mlir
@@ -1790,7 +1790,7 @@ func.func @omp_target_depend(%data_var: memref<i32>) {
   // expected-error @below {{op expected as many depend values as depend variables}}
     "omp.target"(%data_var) ({
       "omp.terminator"() : () -> ()
-    }) {depends = [], operandSegmentSizes = array<i32: 0, 0, 0, 1, 0>} : (memref<i32>) -> ()
+    }) {depends = [], operandSegmentSizes = array<i32: 0, 0, 0, 1, 0, 0, 0>} : (memref<i32>) -> ()
    "func.return"() : () -> ()
 }
 
diff --git a/mlir/test/Dialect/OpenMP/ops.mlir b/mlir/test/Dialect/OpenMP/ops.mlir
index 211ff0ff9272ec..916aa36104537c 100644
--- a/mlir/test/Dialect/OpenMP/ops.mlir
+++ b/mlir/test/Dialect/OpenMP/ops.mlir
@@ -510,22 +510,22 @@ return
 
 
 // CHECK-LABEL: omp_target
-func.func @omp_target(%if_cond : i1, %device : si32,  %num_threads : i32, %map1: memref<?xi32>, %map2: memref<?xi32>) -> () {
+func.func @omp_target(%if_cond : i1, %device : si32,  %num_threads : i32, %device_ptr: memref<i32>, %device_addr: memref<?xi32>, %map1: memref<?xi32>, %map2: memref<?xi32>) -> () {
 
     // Test with optional operands; if_expr, device, thread_limit, private, firstprivate and nowait.
     // CHECK: omp.target if({{.*}}) device({{.*}}) thread_limit({{.*}}) nowait
     "omp.target"(%if_cond, %device, %num_threads) ({
        // CHECK: omp.terminator
        omp.terminator
-    }) {nowait, operandSegmentSizes = array<i32: 1,1,1,0,0>} : ( i1, si32, i32 ) -> ()
+    }) {nowait, operandSegmentSizes = array<i32: 1,1,1,0,0,0,0>} : ( i1, si32, i32 ) -> ()
 
     // Test with optional map clause.
     // CHECK: %[[MAP_A:.*]] = omp.map_info var_ptr(%[[VAL_1:.*]] : memref<?xi32>, tensor<?xi32>)   map_clauses(tofrom) capture(ByRef) -> memref<?xi32> {name = ""}
     // CHECK: %[[MAP_B:.*]] = omp.map_info var_ptr(%[[VAL_2:.*]] : memref<?xi32>, tensor<?xi32>)   map_clauses(exit_release_or_enter_alloc) capture(ByRef) -> memref<?xi32> {name = ""}
-    // CHECK: omp.target map_entries(%[[MAP_A]] -> {{.*}}, %[[MAP_B]] -> {{.*}} : memref<?xi32>, memref<?xi32>) {
+    // CHECK: omp.target is_device_ptr(%[[VAL_4:.*]] : memref<i32>) has_device_addr(%[[VAL_5:.*]] : memref<?xi32>) map_entries(%[[MAP_A]] -> {{.*}}, %[[MAP_B]] -> {{.*}} : memref<?xi32>, memref<?xi32>) {
     %mapv1 = omp.map_info var_ptr(%map1 : memref<?xi32>, tensor<?xi32>)   map_clauses(tofrom) capture(ByRef) -> memref<?xi32> {name = ""}
     %mapv2 = omp.map_info var_ptr(%map2 : memref<?xi32>, tensor<?xi32>)   map_clauses(exit_release_or_enter_alloc) capture(ByRef) -> memref<?xi32> {name = ""}
-    omp.target map_entries(%mapv1 -> %arg0, %mapv2 -> %arg1 : memref<?xi32>, memref<?xi32>) {
+    omp.target map_entries(%mapv1 -> %arg0, %mapv2 -> %arg1 : memref<?xi32>, memref<?xi32>) is_device_ptr(%device_ptr : memref<i32>) has_device_addr(%device_addr : memref<?xi32>) {
     ^bb0(%arg0: memref<?xi32>, %arg1: memref<?xi32>):
       omp.terminator
     }

>From 3b2c7dc7ef6bc9b96fb5f1c1a501511c4a2bce04 Mon Sep 17 00:00:00 2001
From: Raghu Maddhipatla <Raghu.Maddhipatla at amd.com>
Date: Mon, 12 Feb 2024 09:34:20 -0600
Subject: [PATCH 2/4] Rebase patch with upstream.

---
 flang/lib/Lower/OpenMP/OpenMP.cpp      | 16 +++++++++-------
 flang/test/Lower/OpenMP/FIR/target.f90 | 14 +++++++-------
 2 files changed, 16 insertions(+), 14 deletions(-)

diff --git a/flang/lib/Lower/OpenMP/OpenMP.cpp b/flang/lib/Lower/OpenMP/OpenMP.cpp
index 9d4ddaf36db159..986e418cf5432d 100644
--- a/flang/lib/Lower/OpenMP/OpenMP.cpp
+++ b/flang/lib/Lower/OpenMP/OpenMP.cpp
@@ -1053,9 +1053,10 @@ genTargetOp(Fortran::lower::AbstractConverter &converter,
   llvm::SmallVector<mlir::Location> mapSymLocs;
   llvm::SmallVector<const Fortran::semantics::Symbol *> mapSymbols;
   llvm::SmallVector<mlir::Value> devicePtrOperands, deviceAddrOperands;
-  llvm::SmallVector<mlir::Type> useDeviceTypes;
-  llvm::SmallVector<mlir::Location> useDeviceLocs;
-  llvm::SmallVector<const Fortran::semantics::Symbol *> useDeviceSymbols;
+  llvm::SmallVector<mlir::Type> devicePtrTypes, deviceAddrTypes;
+  llvm::SmallVector<mlir::Location> devicePtrLocs, deviceAddrLocs;
+  llvm::SmallVector<const Fortran::semantics::Symbol *> devicePtrSymbols,
+                                                        deviceAddrSymbols;
 
   ClauseProcessor cp(converter, semaCtx, clauseList);
   cp.processIf(Fortran::parser::OmpIfClause::DirectiveNameModifier::Target,
@@ -1066,10 +1067,11 @@ genTargetOp(Fortran::lower::AbstractConverter &converter,
   cp.processNowait(nowaitAttr);
   cp.processMap(currentLocation, directive, stmtCtx, mapOperands, &mapSymTypes,
                 &mapSymLocs, &mapSymbols);
-  cp.processIsDevicePtr(devicePtrOperands, useDeviceTypes, useDeviceLocs,
-                        useDeviceSymbols);
-  cp.processHasDeviceAddr(deviceAddrOperands, useDeviceTypes, useDeviceLocs,
-                          useDeviceSymbols);
+  cp.processIsDevicePtr(devicePtrOperands, devicePtrTypes, devicePtrLocs,
+                        devicePtrSymbols);
+  cp.processHasDeviceAddr(deviceAddrOperands, deviceAddrTypes, deviceAddrLocs,
+                          deviceAddrSymbols);
+
   cp.processTODO<Fortran::parser::OmpClause::Private,
                  Fortran::parser::OmpClause::Firstprivate,
                  Fortran::parser::OmpClause::Reduction,
diff --git a/flang/test/Lower/OpenMP/FIR/target.f90 b/flang/test/Lower/OpenMP/FIR/target.f90
index 2e03e8064d80ce..fcda1ecddfc0b3 100644
--- a/flang/test/Lower/OpenMP/FIR/target.f90
+++ b/flang/test/Lower/OpenMP/FIR/target.f90
@@ -515,15 +515,15 @@ end subroutine omp_target_parallel_do
 !CHECK-LABEL: func.func @_QPomp_target_is_device_ptr() {
 subroutine omp_target_is_device_ptr
    use iso_c_binding, only : c_ptr, c_loc
-   !CHECK: %[[DEV_PTR:.*]] = fir.alloca !fir.type<_QM__fortran_builtinsT__builtin_c_ptr{__address:i64}> {bindc_name = "a", uniq_name = "_QFomp_target_is_device_ptrEa"}
+   !CHECK: %[[VAL_0:.*]] = fir.alloca !fir.type<_QM__fortran_builtinsT__builtin_c_ptr{__address:i64}> {bindc_name = "a", uniq_name = "_QFomp_target_is_device_ptrEa"}
    type(c_ptr) :: a
-   !CHECK %[[VAL_0:.*]] = fir.alloca i32 {bindc_name = "b", fir.target, uniq_name = "_QFomp_target_is_device_ptrEb"}
+   !CHECK: %[[VAL_1:.*]] = fir.alloca i32 {bindc_name = "b", fir.target, uniq_name = "_QFomp_target_is_device_ptrEb"}
    integer, target :: b
-   !CHECK: %[[MAP_0:.*]] = omp.map_info var_ptr(%[[DEV_PTR:.*]] : !fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_c_ptr{__address:i64}>>, !fir.type<_QM__fortran_builtinsT__builtin_c_ptr{__address:i64}>) map_clauses(tofrom) capture(ByRef) -> !fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_c_ptr{__address:i64}>> {name = "a"}
-   !CHECK: %[[MAP_1:.*]] = omp.map_info var_ptr(%[[VAL_0:.*]] : !fir.ref<i32>, i32)   map_clauses(tofrom) capture(ByRef) -> !fir.ref<i32> {name = "b"}
-   !CHECK: omp.target is_device_ptr(%[[DEV_PTR:.*]] : !fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_c_ptr{__address:i64}>>) map_entries(%[[MAP_0:.*]], %[[MAP_1:.*]] : !fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_c_ptr{__address:i64}>>, !fir.ref<i32>, !fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_c_ptr{__address:i64}>>) {
+   !CHECK: %[[VAL_2:.*]] = omp.map_info var_ptr(%[[VAL_0:.*]] : !fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_c_ptr{__address:i64}>>, !fir.type<_QM__fortran_builtinsT__builtin_c_ptr{__address:i64}>) map_clauses(tofrom) capture(ByRef) -> !fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_c_ptr{__address:i64}>> {name = "a"}
+   !CHECK: %[[VAL_3:.*]] = omp.map_info var_ptr(%[[VAL_1:.*]] : !fir.ref<i32>, i32) map_clauses(tofrom) capture(ByRef) -> !fir.ref<i32> {name = "b"}
+   !CHECK: omp.target is_device_ptr(%[[VAL_0:.*]] : !fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_c_ptr{__address:i64}>>) map_entries(%[[VAL_2:.*]] -> {{.*}}, %[[VAL_0:.*]] -> {{.*}}, {{.*}} -> {{.*}} : !fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_c_ptr{__address:i64}>>, !fir.ref<i32>, !fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_c_ptr{__address:i64}>>) {
    !$omp target map(tofrom: a,b) is_device_ptr(a)
-      !CHECK: {{.*}} = fir.coordinate_of %[[DEV_PTR:.*]], {{.*}} : (!fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_c_ptr{__address:i64}>>, !fir.field) -> !fir.ref<i64>
+      !CHECK: {{.*}} = fir.coordinate_of %[[VAL_0:.*]], {{.*}} : (!fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_c_ptr{__address:i64}>>, !fir.field) -> !fir.ref<i64>
       a = c_loc(b)
    !CHECK: omp.terminator
    !$omp end target
@@ -536,8 +536,8 @@ end subroutine omp_target_is_device_ptr
 
  !CHECK-LABEL: func.func @_QPomp_target_has_device_addr() {
  subroutine omp_target_has_device_addr
-   integer, pointer :: a
    !CHECK: %[[VAL_0:.*]] = fir.alloca !fir.box<!fir.ptr<i32>> {bindc_name = "a", uniq_name = "_QFomp_target_has_device_addrEa"}
+   integer, pointer :: a
    !CHECK: omp.target has_device_addr(%[[VAL_0:.*]] : !fir.ref<!fir.box<!fir.ptr<i32>>>) map_entries({{.*}} -> {{.*}}, {{.*}} -> {{.*}} : !fir.llvm_ptr<!fir.ref<i32>>, !fir.ref<!fir.box<!fir.ptr<i32>>>) {
    !$omp target has_device_addr(a)
    !CHECK: {{.*}} = fir.load %[[VAL_0:.*]] : !fir.ref<!fir.box<!fir.ptr<i32>>>

>From b2693f08a9938da50a4f37739e3566be37af9a26 Mon Sep 17 00:00:00 2001
From: Raghu Maddhipatla <Raghu.Maddhipatla at amd.com>
Date: Fri, 16 Feb 2024 13:59:54 -0600
Subject: [PATCH 3/4] clang-format changes

---
 flang/lib/Lower/OpenMP/OpenMP.cpp | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/flang/lib/Lower/OpenMP/OpenMP.cpp b/flang/lib/Lower/OpenMP/OpenMP.cpp
index 986e418cf5432d..9d8856bdcc2611 100644
--- a/flang/lib/Lower/OpenMP/OpenMP.cpp
+++ b/flang/lib/Lower/OpenMP/OpenMP.cpp
@@ -1056,7 +1056,7 @@ genTargetOp(Fortran::lower::AbstractConverter &converter,
   llvm::SmallVector<mlir::Type> devicePtrTypes, deviceAddrTypes;
   llvm::SmallVector<mlir::Location> devicePtrLocs, deviceAddrLocs;
   llvm::SmallVector<const Fortran::semantics::Symbol *> devicePtrSymbols,
-                                                        deviceAddrSymbols;
+      deviceAddrSymbols;
 
   ClauseProcessor cp(converter, semaCtx, clauseList);
   cp.processIf(Fortran::parser::OmpIfClause::DirectiveNameModifier::Target,

>From 2b487499f5691e7b66e125e551cabe69c6d47b67 Mon Sep 17 00:00:00 2001
From: Raghu Maddhipatla <Raghu.Maddhipatla at amd.com>
Date: Thu, 22 Feb 2024 19:25:23 -0600
Subject: [PATCH 4/4] Rebasing patch with Lower/OpenMP/OpenMP.cpp changes.

---
 flang/lib/Lower/OpenMP/ClauseProcessor.h | 4 ++--
 flang/lib/Lower/OpenMP/OpenMP.cpp        | 4 ++--
 2 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/flang/lib/Lower/OpenMP/ClauseProcessor.h b/flang/lib/Lower/OpenMP/ClauseProcessor.h
index 7e36c84c4662fe..6687206f811882 100644
--- a/flang/lib/Lower/OpenMP/ClauseProcessor.h
+++ b/flang/lib/Lower/OpenMP/ClauseProcessor.h
@@ -148,14 +148,14 @@ class ClauseProcessor {
                      llvm::SmallVectorImpl<mlir::Type> &isDeviceTypes,
                      llvm::SmallVectorImpl<mlir::Location> &isDeviceLocs,
                      llvm::SmallVectorImpl<const Fortran::semantics::Symbol *>
-                     &isDeviceSymbols) const;
+                         &isDeviceSymbols) const;
 
   bool
   processHasDeviceAddr(llvm::SmallVectorImpl<mlir::Value> &operands,
                        llvm::SmallVectorImpl<mlir::Type> &isDeviceTypes,
                        llvm::SmallVectorImpl<mlir::Location> &isDeviceLocs,
                        llvm::SmallVectorImpl<const Fortran::semantics::Symbol *>
-                       &isDeviceSymbols) const;
+                           &isDeviceSymbols) const;
 
   template <typename T>
   bool processMotionClauses(Fortran::lower::StatementContext &stmtCtx,
diff --git a/flang/lib/Lower/OpenMP/OpenMP.cpp b/flang/lib/Lower/OpenMP/OpenMP.cpp
index 9d8856bdcc2611..ae44b4daa6700b 100644
--- a/flang/lib/Lower/OpenMP/OpenMP.cpp
+++ b/flang/lib/Lower/OpenMP/OpenMP.cpp
@@ -1157,8 +1157,8 @@ genTargetOp(Fortran::lower::AbstractConverter &converter,
           ? nullptr
           : mlir::ArrayAttr::get(converter.getFirOpBuilder().getContext(),
                                  dependTypeOperands),
-      dependOperands, nowaitAttr, devicePtrOperands,
-      deviceAddrOperands, mapOperands);
+      dependOperands, nowaitAttr, devicePtrOperands, deviceAddrOperands,
+      mapOperands);
 
   genBodyOfTargetOp(converter, semaCtx, eval, genNested, targetOp, mapSymTypes,
                     mapSymLocs, mapSymbols, currentLocation);



More information about the flang-commits mailing list