[Flang] - Handle `BoxCharType` in `fir.box_offset` op #141713

bhandarkar-pranav · 2025-05-28T03:48:11Z

To map fir.boxchar types reliably onto an offload target, such as a GPU, the omp.map.info operation is used to map the underlying data pointer (fir.ref<fir.char<k, ?>>) wrapped by the fir.boxchar MLIR value. The omp.map.info operation needs a pointer to the underlying data pointer.
Given a reference to a descriptor (fir.box), the fir.box_offset is used to obtain the address of the underlying data pointer. This PR extends fir.box_offset to provide the same functionality for fir.boxchar as well.

PR Stack

…boxchar

llvmbot · 2025-05-28T03:48:46Z

@llvm/pr-subscribers-flang-codegen

@llvm/pr-subscribers-flang-fir-hlfir

Author: Pranav Bhandarkar (bhandarkar-pranav)

Changes

To map fir.boxchar types reliably onto an offload target, such as a GPU, the omp.map.info is used to map the underlying data pointer (fir.ref<fir.char<k, ?>) wrapped by the fir.boxchar MLIR value. The omp.map.info instruction needs a pointer to the underlying data pointer.
Given a reference to a descriptor (fir.box), the fir.box_offset is used to obtain the address of the underlying data pointer. This PR extends fir.box_offset to provide the same functionality for fir.boxchar as well.

Full diff: https://github.com/llvm/llvm-project/pull/141713.diff

8 Files Affected:

(modified) flang/include/flang/Optimizer/Dialect/FIROps.td (+6)
(modified) flang/include/flang/Optimizer/Dialect/FIRType.h (+3-2)
(modified) flang/lib/Optimizer/CodeGen/CodeGen.cpp (+19-6)
(modified) flang/lib/Optimizer/Dialect/FIROps.cpp (+13-3)
(modified) flang/lib/Optimizer/Dialect/FIRType.cpp (+1-1)
(modified) flang/test/Fir/box-offset-codegen.fir (+10)
(modified) flang/test/Fir/box-offset.fir (+5)
(modified) flang/test/Fir/invalid.fir (+9-1)

diff --git a/flang/include/flang/Optimizer/Dialect/FIROps.td b/flang/include/flang/Optimizer/Dialect/FIROps.td
index dc66885f776f0..160de05a33b41 100644
--- a/flang/include/flang/Optimizer/Dialect/FIROps.td
+++ b/flang/include/flang/Optimizer/Dialect/FIROps.td
@@ -3240,11 +3240,17 @@ def fir_BoxOffsetOp : fir_Op<"box_offset", [NoMemoryEffect]> {
     descriptor implementation must have, only the base_addr and derived_type
     descriptor fields can be addressed.
 
+    It also accepts the address of a fir.boxchar and returns
+    address of the data pointer encapsulated by the fir.boxchar.
+
     ```
         %addr = fir.box_offset %box base_addr : (!fir.ref<!fir.box<!fir.array<?xi32>>>) -> !fir.llvm_ptr<!fir.ref<!fir.array<?xi32>>>
         %tdesc = fir.box_offset %box derived_type : (!fir.ref<!fir.box<!fir.type<t>>>) -> !fir.llvm_ptr<!fir.tdesc<!fir.type<t>>>
 
+        %addr1 = fir.box_offset %boxchar base_addr : (!fir.ref<!fir.boxchar<1>>) -> !fir.llvm_ptr<!fir.ref<fir.char<1,?>>>
     ```
+
+    The derived_type field cannot be used when the input to this op is a reference to a fir.boxchar.
   }];
 
   let arguments = (ins
diff --git a/flang/include/flang/Optimizer/Dialect/FIRType.h b/flang/include/flang/Optimizer/Dialect/FIRType.h
index 52b14f15f89bd..01878aa41005c 100644
--- a/flang/include/flang/Optimizer/Dialect/FIRType.h
+++ b/flang/include/flang/Optimizer/Dialect/FIRType.h
@@ -278,8 +278,9 @@ inline mlir::Type unwrapRefType(mlir::Type t) {
 /// If `t` conforms with a pass-by-reference type (box, ref, ptr, etc.) then
 /// return the element type of `t`. Otherwise, return `t`.
 inline mlir::Type unwrapPassByRefType(mlir::Type t) {
-  if (auto eleTy = dyn_cast_ptrOrBoxEleTy(t))
-    return eleTy;
+  if (conformsWithPassByRef(t))
+    if (auto eleTy = dyn_cast_ptrOrBoxEleTy(t))
+      return eleTy;
   return t;
 }
 
diff --git a/flang/lib/Optimizer/CodeGen/CodeGen.cpp b/flang/lib/Optimizer/CodeGen/CodeGen.cpp
index 205807eab403a..e383c2e3e89ab 100644
--- a/flang/lib/Optimizer/CodeGen/CodeGen.cpp
+++ b/flang/lib/Optimizer/CodeGen/CodeGen.cpp
@@ -3930,12 +3930,25 @@ struct BoxOffsetOpConversion : public fir::FIROpConversion<fir::BoxOffsetOp> {
                   mlir::ConversionPatternRewriter &rewriter) const override {
 
     mlir::Type pty = ::getLlvmPtrType(boxOffset.getContext());
-    mlir::Type boxType = fir::unwrapRefType(boxOffset.getBoxRef().getType());
-    mlir::Type llvmBoxTy =
-        lowerTy().convertBoxTypeAsStruct(mlir::cast<fir::BaseBoxType>(boxType));
-    int fieldId = boxOffset.getField() == fir::BoxFieldAttr::derived_type
-                      ? getTypeDescFieldId(boxType)
-                      : kAddrPosInBox;
+    mlir::Type boxRefType = fir::unwrapRefType(boxOffset.getBoxRef().getType());
+
+    assert((mlir::isa<fir::BaseBoxType>(boxRefType) ||
+            mlir::isa<fir::BoxCharType>(boxRefType)) &&
+           "boxRef should be a reference to either fir.box or fir.boxchar");
+
+    mlir::Type llvmBoxTy;
+    int fieldId;
+    if (auto boxType = mlir::dyn_cast_or_null<fir::BaseBoxType>(boxRefType)) {
+      llvmBoxTy =
+          lowerTy().convertBoxTypeAsStruct(mlir::cast<fir::BaseBoxType>(boxType));
+      fieldId = boxOffset.getField() == fir::BoxFieldAttr::derived_type
+                        ? getTypeDescFieldId(boxType)
+                        : kAddrPosInBox;
+    } else {
+      auto boxCharType = mlir::cast<fir::BoxCharType>(boxRefType);
+      llvmBoxTy = lowerTy().convertType(boxCharType);
+      fieldId = kAddrPosInBox;
+    }
     rewriter.replaceOpWithNewOp<mlir::LLVM::GEPOp>(
         boxOffset, pty, llvmBoxTy, adaptor.getBoxRef(),
         llvm::ArrayRef<mlir::LLVM::GEPArg>{0, fieldId});
diff --git a/flang/lib/Optimizer/Dialect/FIROps.cpp b/flang/lib/Optimizer/Dialect/FIROps.cpp
index cbe93907265f6..6435886d73081 100644
--- a/flang/lib/Optimizer/Dialect/FIROps.cpp
+++ b/flang/lib/Optimizer/Dialect/FIROps.cpp
@@ -4484,15 +4484,25 @@ void fir::IfOp::resultToSourceOps(llvm::SmallVectorImpl<mlir::Value> &results,
 llvm::LogicalResult fir::BoxOffsetOp::verify() {
   auto boxType = mlir::dyn_cast_or_null<fir::BaseBoxType>(
       fir::dyn_cast_ptrEleTy(getBoxRef().getType()));
-  if (!boxType)
-    return emitOpError("box_ref operand must have !fir.ref<!fir.box<T>> type");
+  mlir::Type boxCharType;
+  bool isBoxChar = false;
+  if (!boxType) {
+    boxCharType = mlir::dyn_cast_or_null<fir::BoxCharType>(
+        fir::dyn_cast_ptrEleTy(getBoxRef().getType()));
+    if (!boxCharType)
+      return emitOpError("box_ref operand must have !fir.ref<!fir.box<T>> or !fir.ref<!fir.boxchar<k>> type");
+    isBoxChar = true;
+  }
   if (getField() != fir::BoxFieldAttr::base_addr &&
       getField() != fir::BoxFieldAttr::derived_type)
     return emitOpError("cannot address provided field");
-  if (getField() == fir::BoxFieldAttr::derived_type)
+  if (getField() == fir::BoxFieldAttr::derived_type) {
+    if (isBoxChar)
+      return emitOpError("cannot address derived_type field of a fir.boxchar");
     if (!fir::boxHasAddendum(boxType))
       return emitOpError("can only address derived_type field of derived type "
                          "or unlimited polymorphic fir.box");
+  }
   return mlir::success();
 }
 
diff --git a/flang/lib/Optimizer/Dialect/FIRType.cpp b/flang/lib/Optimizer/Dialect/FIRType.cpp
index 1e6e95393c2f7..da7aa17445404 100644
--- a/flang/lib/Optimizer/Dialect/FIRType.cpp
+++ b/flang/lib/Optimizer/Dialect/FIRType.cpp
@@ -255,7 +255,7 @@ mlir::Type dyn_cast_ptrOrBoxEleTy(mlir::Type t) {
   return llvm::TypeSwitch<mlir::Type, mlir::Type>(t)
       .Case<fir::ReferenceType, fir::PointerType, fir::HeapType,
             fir::LLVMPointerType>([](auto p) { return p.getEleTy(); })
-      .Case<fir::BaseBoxType>(
+      .Case<fir::BaseBoxType, fir::BoxCharType>(
           [](auto p) { return unwrapRefType(p.getEleTy()); })
       .Default([](mlir::Type) { return mlir::Type{}; });
 }
diff --git a/flang/test/Fir/box-offset-codegen.fir b/flang/test/Fir/box-offset-codegen.fir
index 15c9a11e5aefe..59cfda8523061 100644
--- a/flang/test/Fir/box-offset-codegen.fir
+++ b/flang/test/Fir/box-offset-codegen.fir
@@ -37,3 +37,13 @@ func.func @array_tdesc(%array : !fir.ref<!fir.class<!fir.ptr<!fir.array<?x!fir.t
 // CHECK-SAME: ptr captures(none) %[[BOX:.*]]){{.*}}{
 // CHECK:    %[[VAL_0:.*]] = getelementptr { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]], ptr, [1 x i64] }, ptr %[[BOX]], i32 0, i32 8
 // CHECK:    ret ptr %[[VAL_0]]
+
+func.func @boxchar_addr(%boxchar : !fir.ref<!fir.boxchar<1>>) -> !fir.llvm_ptr<!fir.ref<!fir.char<1,?>>> {
+  %addr = fir.box_offset %boxchar base_addr : (!fir.ref<!fir.boxchar<1>>) -> !fir.llvm_ptr<!fir.ref<!fir.char<1,?>>>
+  return %addr : !fir.llvm_ptr<!fir.ref<!fir.char<1,?>>>
+}
+
+// CHECK-LABEL: define ptr @boxchar_addr(
+// CHECK-SAME: ptr captures(none) %[[BOXCHAR:.*]]){{.*}} {
+// CHECK: %[[VAL_0:.*]] = getelementptr { ptr, i64 }, ptr %[[BOXCHAR]], i32 0, i32 0
+// CHECK: ret ptr %[[VAL_0]]
diff --git a/flang/test/Fir/box-offset.fir b/flang/test/Fir/box-offset.fir
index 98c2eaefb8d6b..181ad51a5dbe1 100644
--- a/flang/test/Fir/box-offset.fir
+++ b/flang/test/Fir/box-offset.fir
@@ -21,6 +21,9 @@ func.func @test_box_offset(%unlimited : !fir.ref<!fir.class<none>>, %type_star :
 
   %addr6 = fir.box_offset %type_star base_addr : (!fir.ref<!fir.box<!fir.array<?xnone>>>) -> !fir.llvm_ptr<!fir.ref<!fir.array<?xnone>>>
   %tdesc6 = fir.box_offset %type_star derived_type : (!fir.ref<!fir.box<!fir.array<?xnone>>>) -> !fir.llvm_ptr<!fir.tdesc<none>>
+
+  %boxchar = fir.alloca !fir.boxchar<1>
+  %addr7 = fir.box_offset %boxchar base_addr : (!fir.ref<!fir.boxchar<1>>) -> !fir.llvm_ptr<!fir.ref<!fir.char<1,?>>>
   return
 }
 // CHECK-LABEL:   func.func @test_box_offset(
@@ -40,3 +43,5 @@ func.func @test_box_offset(%unlimited : !fir.ref<!fir.class<none>>, %type_star :
 // CHECK:           %[[VAL_13:.*]] = fir.box_offset %[[VAL_0]] derived_type : (!fir.ref<!fir.class<none>>) -> !fir.llvm_ptr<!fir.tdesc<none>>
 // CHECK:           %[[VAL_14:.*]] = fir.box_offset %[[VAL_1]] base_addr : (!fir.ref<!fir.box<!fir.array<?xnone>>>) -> !fir.llvm_ptr<!fir.ref<!fir.array<?xnone>>>
 // CHECK:           %[[VAL_15:.*]] = fir.box_offset %[[VAL_1]] derived_type : (!fir.ref<!fir.box<!fir.array<?xnone>>>) -> !fir.llvm_ptr<!fir.tdesc<none>>
+// CHECK:           %[[VAL_16:.*]] = fir.alloca !fir.boxchar<1>
+// CHECK:           %[[VAL_17:.*]] = fir.box_offset %[[VAL_16]] base_addr : (!fir.ref<!fir.boxchar<1>>) -> !fir.llvm_ptr<!fir.ref<!fir.char<1,?>>>
diff --git a/flang/test/Fir/invalid.fir b/flang/test/Fir/invalid.fir
index fd607fd9066f7..45cae1f82cb8e 100644
--- a/flang/test/Fir/invalid.fir
+++ b/flang/test/Fir/invalid.fir
@@ -972,13 +972,21 @@ func.func @rec_to_rec(%arg0: !fir.type<t1{i:i32, f:f32}>) -> !fir.type<t2{f:f32,
 // -----
 
 func.func @bad_box_offset(%not_a_box : !fir.ref<i32>) {
-  // expected-error@+1{{'fir.box_offset' op box_ref operand must have !fir.ref<!fir.box<T>> type}}
+  // expected-error@+1{{'fir.box_offset' op box_ref operand must have !fir.ref<!fir.box<T>> or !fir.ref<!fir.boxchar<k>> type}}
   %addr1 = fir.box_offset %not_a_box base_addr : (!fir.ref<i32>) -> !fir.llvm_ptr<!fir.ref<i32>>
   return
 }
 
 // -----
 
+func.func @bad_box_offset(%boxchar : !fir.ref<!fir.boxchar<1>>) {
+  // expected-error@+1{{'fir.box_offset' op cannot address derived_type field of a fir.boxchar}}
+  %addr1 = fir.box_offset %boxchar derived_type : (!fir.ref<!fir.boxchar<1>>) -> !fir.llvm_ptr<!fir.ref<!fir.char<1,?>>>
+  return
+}
+
+// -----
+
 func.func @bad_box_offset(%no_addendum : !fir.ref<!fir.box<i32>>) {
   // expected-error@+1{{'fir.box_offset' op can only address derived_type field of derived type or unlimited polymorphic fir.box}}
   %addr1 = fir.box_offset %no_addendum derived_type : (!fir.ref<!fir.box<i32>>) -> !fir.llvm_ptr<!fir.tdesc<!fir.type<none>>>

github-actions · 2025-05-28T03:50:38Z

✅ With the latest revision this PR passed the C/C++ code formatter.

tblah

LGTM with one minor comment

flang/include/flang/Optimizer/Dialect/FIRType.h

jeanPerier

LGTM outside of the unwrapPassByRefType change.

flang/include/flang/Optimizer/Dialect/FIRType.h

flang/lib/Optimizer/Dialect/FIROps.cpp

agozillon

LGTM based on my understanding, but I'll leave final approval up to the other reviewers!

jeanPerier · 2025-06-04T08:26:39Z

flang/test/Fir/box-offset-codegen.fir

@@ -37,3 +37,13 @@ func.func @array_tdesc(%array : !fir.ref<!fir.class<!fir.ptr<!fir.array<?x!fir.t
 // CHECK-SAME: ptr captures(none) %[[BOX:.*]]){{.*}}{
 // CHECK:    %[[VAL_0:.*]] = getelementptr { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]], ptr, [1 x i64] }, ptr %[[BOX]], i32 0, i32 8
 // CHECK:    ret ptr %[[VAL_0]]
+


Beware the github CI indicates this test is failing.

Thank you for you review. I noticed. I'll take care of it.

bhandarkar-pranav · 2025-06-06T15:47:33Z

Merging this because failing testcase (Flang.lower/namelist.f90) is a testcase problem fixed by #142363.

llvm-ci · 2025-06-06T15:49:15Z

LLVM Buildbot has detected a new failure on builder amdgpu-offload-rhel-8-cmake-build-only running on rocm-docker-rhel-8 while building flang at step 3 "checkout".

Full details are available at: https://lab.llvm.org/buildbot/#/builders/204/builds/11557

Here is the relevant piece of the build log for the reference

Step 3 (checkout) failure: update (failure)
git version 2.43.5
fatal: unable to access 'https://github.com/llvm/llvm-project.git/': Could not resolve host: github.com
fatal: unable to access 'https://github.com/llvm/llvm-project.git/': Could not resolve host: github.com

To map `fir.boxchar` types reliably onto an offload target, such as a GPU, the `omp.map.info` operation is used to map the underlying data pointer (`fir.ref<fir.char<k, ?>>`) wrapped by the `fir.boxchar` MLIR value. The `omp.map.info` operation needs a pointer to the underlying data pointer. Given a reference to a descriptor (`fir.box`), the `fir.box_offset` is used to obtain the address of the underlying data pointer. This PR extends `fir.box_offset` to provide the same functionality for `fir.boxchar` as well.

Add the ability to the fir.box_offset op to handle references to fir.…

271272f

…boxchar

bhandarkar-pranav requested review from ergawy, skatrak, jsjodin, tblah, agozillon, jeanPerier, TIFitis and kparzysz May 28, 2025 03:48

llvmbot added flang Flang issues not falling into any other category flang:fir-hlfir flang:codegen labels May 28, 2025

bhandarkar-pranav mentioned this pull request May 28, 2025

[Flang][OpenMP] - When mapping a fir.boxchar, map the underlying data pointer as a member #141715

Merged

tblah approved these changes May 28, 2025

View reviewed changes

flang/include/flang/Optimizer/Dialect/FIRType.h Outdated Show resolved Hide resolved

Fix clang formatt issues

ada55c9

jeanPerier reviewed Jun 2, 2025

View reviewed changes

flang/include/flang/Optimizer/Dialect/FIRType.h Outdated Show resolved Hide resolved

flang/lib/Optimizer/Dialect/FIROps.cpp Outdated Show resolved Hide resolved

bhandarkar-pranav added 2 commits June 3, 2025 10:54

Make changes based on review comments

f3e8cbe

Fix testcase

3300531

agozillon approved these changes Jun 3, 2025

View reviewed changes

jeanPerier approved these changes Jun 4, 2025

View reviewed changes

jeanPerier reviewed Jun 4, 2025

View reviewed changes

fix testcase again

3bde346

bhandarkar-pranav merged commit 8395912 into main Jun 6, 2025
10 of 12 checks passed

bhandarkar-pranav deleted the users/bhandarkar-pranav/boxchar_firstprivate_fix0 branch June 6, 2025 15:48

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

[Flang] - Handle `BoxCharType` in `fir.box_offset` op #141713

[Flang] - Handle `BoxCharType` in `fir.box_offset` op #141713

Uh oh!

bhandarkar-pranav commented May 28, 2025 •

edited

Loading

Uh oh!

llvmbot commented May 28, 2025 •

edited

Loading

Uh oh!

github-actions bot commented May 28, 2025 •

edited

Loading

Uh oh!

tblah left a comment

Uh oh!

Uh oh!

jeanPerier left a comment

Uh oh!

Uh oh!

Uh oh!

agozillon left a comment

Uh oh!

jeanPerier Jun 4, 2025

Uh oh!

bhandarkar-pranav Jun 4, 2025

Uh oh!

bhandarkar-pranav commented Jun 6, 2025

Uh oh!

Uh oh!

llvm-ci commented Jun 6, 2025

Uh oh!

Uh oh!

[Flang] - Handle BoxCharType in fir.box_offset op #141713

[Flang] - Handle BoxCharType in fir.box_offset op #141713

Uh oh!

Conversation

bhandarkar-pranav commented May 28, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

llvmbot commented May 28, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions bot commented May 28, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

tblah left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

jeanPerier left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

agozillon left a comment

Choose a reason for hiding this comment

Uh oh!

jeanPerier Jun 4, 2025

Choose a reason for hiding this comment

Uh oh!

bhandarkar-pranav Jun 4, 2025

Choose a reason for hiding this comment

Uh oh!

bhandarkar-pranav commented Jun 6, 2025

Uh oh!

Uh oh!

llvm-ci commented Jun 6, 2025

Uh oh!

Uh oh!

[Flang] - Handle `BoxCharType` in `fir.box_offset` op #141713

[Flang] - Handle `BoxCharType` in `fir.box_offset` op #141713

bhandarkar-pranav commented May 28, 2025 •

edited

Loading

llvmbot commented May 28, 2025 •

edited

Loading

github-actions bot commented May 28, 2025 •

edited

Loading