[flang][cuda] Add support for character type in cuf.alloc and cuf.data_transfer

Author: clementval

flang/lib/Optimizer/Transforms/CUFOpConversion.cpp

@@ -268,24 +268,23 @@ static bool inDeviceContext(mlir::Operation *op) {
 static int computeWidth(mlir::Location loc, mlir::Type type,
                         fir::KindMapping &kindMap) {
   auto eleTy = fir::unwrapSequenceType(type);
-  int width = 0;
-  if (auto t{mlir::dyn_cast<mlir::IntegerType>(eleTy)}) {
-    width = t.getWidth() / 8;
-  } else if (auto t{mlir::dyn_cast<mlir::FloatType>(eleTy)}) {
-    width = t.getWidth() / 8;
-  } else if (eleTy.isInteger(1)) {
-    width = 1;
-  } else if (auto t{mlir::dyn_cast<fir::LogicalType>(eleTy)}) {
-    int kind = t.getFKind();
-    width = kindMap.getLogicalBitsize(kind) / 8;
-  } else if (auto t{mlir::dyn_cast<mlir::ComplexType>(eleTy)}) {
+  if (auto t{mlir::dyn_cast<mlir::IntegerType>(eleTy)})
+    return t.getWidth() / 8;
+  if (auto t{mlir::dyn_cast<mlir::FloatType>(eleTy)})
+    return t.getWidth() / 8;
+  if (eleTy.isInteger(1))
+    return 1;
+  if (auto t{mlir::dyn_cast<fir::LogicalType>(eleTy)})
+    return kindMap.getLogicalBitsize(t.getFKind()) / 8;
+  if (auto t{mlir::dyn_cast<mlir::ComplexType>(eleTy)}) {
     int elemSize =
         mlir::cast<mlir::FloatType>(t.getElementType()).getWidth() / 8;
-    width = 2 * elemSize;
-  } else {
-    mlir::emitError(loc, "unsupported type");
+    return 2 * elemSize;
   }
-  return width;
+  if (auto t{mlir::dyn_cast_or_null<fir::CharacterType>(eleTy)})
+    return kindMap.getCharacterBitsize(t.getFKind()) / 8;
+  mlir::emitError(loc, "unsupported type");
+  return 0;
 }
 
 struct CUFAllocOpConversion : public mlir::OpRewritePattern<cuf::AllocOp> {

flang/test/Fir/CUDA/cuda-alloc-free.fir

@@ -83,4 +83,15 @@ gpu.module @cuda_device_mod [#nvvm.target] {
 // CHECK-LABEL: gpu.func @_QMalloc() kernel
 // CHECK: fir.alloca !fir.box<!fir.heap<!fir.array<?xf32>>> {bindc_name = "a", uniq_name = "_QMallocEa"}
 
+func.func @_QQalloc_char() attributes {fir.bindc_name = "alloc_char"} {
+  %c1 = arith.constant 1 : index
+  %0 = cuf.alloc !fir.array<10x!fir.char<1>>(%c1 : index) {bindc_name = "a", data_attr = #cuf.cuda<device>, uniq_name = "_QFEa"} -> !fir.ref<!fir.array<10x!fir.char<1>>>
+  return
+}
+
+// CHECK-LABEL: func.func @_QQalloc_char()
+// CHECK: %[[BYTES:.*]] = arith.muli %c10{{.*}}, %c1{{.*}} : index
+// CHECK: %[[BYTES_CONV:.*]] = fir.convert %[[BYTES]] : (index) -> i64
+// CHECK: fir.call @_FortranACUFMemAlloc(%[[BYTES_CONV]], %c0{{.*}}, %{{.*}}, %{{.*}}) : (i64, i32, !fir.ref<i8>, i32) -> !fir.llvm_ptr<i8>
+
 } // end module

flang/test/Fir/CUDA/cuda-data-transfer.fir

@@ -385,4 +385,22 @@ func.func @_QPdevice_addr_conv() {
 // CHECK: fir.embox %[[DEV_ADDR_CONV]](%{{.*}}) : (!fir.ref<!fir.array<4xf32>>, !fir.shape<1>) -> !fir.box<!fir.array<4xf32>>
 // CHECK: fir.call @_FortranACUFDataTransferDescDescNoRealloc
 
+func.func @_QQchar_transfer() attributes {fir.bindc_name = "char_transfer"} {
+  %c1 = arith.constant 1 : index
+  %c10 = arith.constant 10 : index
+  %0 = cuf.alloc !fir.array<10x!fir.char<1>>(%c1 : index) {bindc_name = "a", data_attr = #cuf.cuda<device>, uniq_name = "_QFEa"} -> !fir.ref<!fir.array<10x!fir.char<1>>>
+  %1 = fir.shape %c10 : (index) -> !fir.shape<1>
+  %2 = fir.declare %0(%1) typeparams %c1 {data_attr = #cuf.cuda<device>, uniq_name = "_QFEa"} : (!fir.ref<!fir.array<10x!fir.char<1>>>, !fir.shape<1>, index) -> !fir.ref<!fir.array<10x!fir.char<1>>>
+  %3 = fir.alloca !fir.array<10x!fir.char<1>> {bindc_name = "b", uniq_name = "_QFEb"}
+  %4 = fir.declare %3(%1) typeparams %c1 {uniq_name = "_QFEb"} : (!fir.ref<!fir.array<10x!fir.char<1>>>, !fir.shape<1>, index) -> !fir.ref<!fir.array<10x!fir.char<1>>>
+  cuf.data_transfer %4 to %2 {transfer_kind = #cuf.cuda_transfer<host_device>} : !fir.ref<!fir.array<10x!fir.char<1>>>, !fir.ref<!fir.array<10x!fir.char<1>>>
+  cuf.free %2 : !fir.ref<!fir.array<10x!fir.char<1>>> {data_attr = #cuf.cuda<device>}
+  return
+}
+
+// CHECK-LABEL:  func.func @_QQchar_transfer()
+// CHECK: fir.call @_FortranACUFMemAlloc
+// CHECK: %[[BYTES:.*]] = arith.muli %c10{{.*}}, %c1{{.*}} : i64
+// CHECK: fir.call @_FortranACUFDataTransferPtrPtr(%{{.*}}, %{{.*}}, %[[BYTES]], %c0{{.*}}, %{{.*}}, %{{.*}}) : (!fir.llvm_ptr<i8>, !fir.llvm_ptr<i8>, i64, i32, !fir.ref<i8>, i32) -> none
+
 } // end of module