[flang][cuda] Fix offset with multiple assumed size shared array (#154844)

When multiple assumed size variable are used in a kernel with dynamic
shared memory, each variable use the 0 offset. Update the pass to
account for that.

```
attributes(global) subroutine testany( a )
    real(4), shared :: smasks(*)
    real(8), shared :: dmasks(*)
end subroutine
```

Author: clementval

flang/lib/Optimizer/Transforms/CUFComputeSharedMemoryOffsetsAndSize.cpp

@@ -38,6 +38,15 @@ using namespace Fortran::runtime::cuda;
 
 namespace {
 
+static bool isAssumedSize(mlir::ValueRange shape) {
+  if (shape.size() != 1)
+    return false;
+  std::optional<std::int64_t> val = fir::getIntIfConstant(shape[0]);
+  if (val && *val == -1)
+    return true;
+  return false;
+}
+
 struct CUFComputeSharedMemoryOffsetsAndSize
     : public fir::impl::CUFComputeSharedMemoryOffsetsAndSizeBase<
           CUFComputeSharedMemoryOffsetsAndSize> {
@@ -82,12 +91,12 @@ struct CUFComputeSharedMemoryOffsetsAndSize
           alignment = std::max(alignment, align);
           uint64_t tySize = dl->getTypeSize(ty);
           ++nbDynamicSharedVariables;
-          if (crtDynOffset) {
-            sharedOp.getOffsetMutable().assign(
-                builder.createConvert(loc, i32Ty, crtDynOffset));
-          } else {
+          if (isAssumedSize(sharedOp.getShape()) || !crtDynOffset) {
             mlir::Value zero = builder.createIntegerConstant(loc, i32Ty, 0);
             sharedOp.getOffsetMutable().assign(zero);
+          } else {
+            sharedOp.getOffsetMutable().assign(
+                builder.createConvert(loc, i32Ty, crtDynOffset));
           }
 
           mlir::Value dynSize =

flang/test/Fir/CUDA/cuda-shared-offset.mlir

@@ -121,4 +121,40 @@ module attributes {dlti.dl_spec = #dlti.dl_spec<#dlti.dl_entry<!llvm.ptr, dense<
 // CHECK-LABEL: gpu.func @_QPnoshared()
 // CHECK-NOT: fir.global internal @_QPnoshared__shared_mem
 
+// -----
+
+module attributes {dlti.dl_spec = #dlti.dl_spec<#dlti.dl_entry<!llvm.ptr, dense<64> : vector<4xi64>>, #dlti.dl_entry<!llvm.ptr<271>, dense<32> : vector<4xi64>>, #dlti.dl_entry<!llvm.ptr<270>, dense<32> : vector<4xi64>>, #dlti.dl_entry<f128, dense<128> : vector<2xi64>>, #dlti.dl_entry<f64, dense<64> : vector<2xi64>>, #dlti.dl_entry<f80, dense<128> : vector<2xi64>>, #dlti.dl_entry<f16, dense<16> : vector<2xi64>>, #dlti.dl_entry<i32, dense<32> : vector<2xi64>>, #dlti.dl_entry<i16, dense<16> : vector<2xi64>>, #dlti.dl_entry<i128, dense<128> : vector<2xi64>>, #dlti.dl_entry<i8, dense<8> : vector<2xi64>>, #dlti.dl_entry<!llvm.ptr<272>, dense<64> : vector<4xi64>>, #dlti.dl_entry<i64, dense<64> : vector<2xi64>>, #dlti.dl_entry<i1, dense<8> : vector<2xi64>>, #dlti.dl_entry<"dlti.endianness", "little">, #dlti.dl_entry<"dlti.stack_alignment", 128 : i64>>, fir.defaultkind = "a1c4d8i4l4r4", fir.kindmap = "", gpu.container_module, llvm.data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", llvm.ident = "flang version 20.0.0 (https://github.com/llvm/llvm-project.git cae351f3453a0a26ec8eb2ddaf773c24a29d929e)", llvm.target_triple = "x86_64-unknown-linux-gnu"} {
+  gpu.module @cuda_device_mod {
+    gpu.func @_QMmtestsPtestany(%arg0: !fir.ref<!fir.array<?xf32>> {cuf.data_attr = #cuf.cuda<device>, fir.bindc_name = "a"}) attributes {cuf.proc_attr = #cuf.cuda_proc<global>} {
+      %0 = fir.dummy_scope : !fir.dscope
+      %c-1 = arith.constant -1 : index
+      %1 = fir.shape %c-1 : (index) -> !fir.shape<1>
+      %2:2 = hlfir.declare %arg0(%1) dummy_scope %0 {data_attr = #cuf.cuda<device>, uniq_name = "_QMmtestsFtestanyEa"} : (!fir.ref<!fir.array<?xf32>>, !fir.shape<1>, !fir.dscope) -> (!fir.box<!fir.array<?xf32>>, !fir.ref<!fir.array<?xf32>>)
+      %3 = fir.address_of(@_QM__fortran_builtinsE__builtin_blockdim) : !fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_dim3{x:i32,y:i32,z:i32}>>
+      %4:2 = hlfir.declare %3 {uniq_name = "_QM__fortran_builtinsE__builtin_blockdim"} : (!fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_dim3{x:i32,y:i32,z:i32}>>) -> (!fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_dim3{x:i32,y:i32,z:i32}>>, !fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_dim3{x:i32,y:i32,z:i32}>>)
+      %5 = fir.address_of(@_QM__fortran_builtinsE__builtin_blockidx) : !fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_dim3{x:i32,y:i32,z:i32}>>
+      %6:2 = hlfir.declare %5 {uniq_name = "_QM__fortran_builtinsE__builtin_blockidx"} : (!fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_dim3{x:i32,y:i32,z:i32}>>) -> (!fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_dim3{x:i32,y:i32,z:i32}>>, !fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_dim3{x:i32,y:i32,z:i32}>>)
+      %c-1_0 = arith.constant -1 : index
+      %7 = cuf.shared_memory !fir.array<?xf64>, %c-1_0 : index {bindc_name = "dmasks", uniq_name = "_QMmtestsFtestanyEdmasks"} -> !fir.ref<!fir.array<?xf64>>
+      %8 = fir.shape %c-1_0 : (index) -> !fir.shape<1>
+      %9:2 = hlfir.declare %7(%8) {data_attr = #cuf.cuda<shared>, uniq_name = "_QMmtestsFtestanyEdmasks"} : (!fir.ref<!fir.array<?xf64>>, !fir.shape<1>) -> (!fir.box<!fir.array<?xf64>>, !fir.ref<!fir.array<?xf64>>)
+      %10 = fir.address_of(@_QM__fortran_builtinsE__builtin_griddim) : !fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_dim3{x:i32,y:i32,z:i32}>>
+      %11:2 = hlfir.declare %10 {uniq_name = "_QM__fortran_builtinsE__builtin_griddim"} : (!fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_dim3{x:i32,y:i32,z:i32}>>) -> (!fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_dim3{x:i32,y:i32,z:i32}>>, !fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_dim3{x:i32,y:i32,z:i32}>>)
+      %12 = fir.alloca i32 {bindc_name = "i", uniq_name = "_QMmtestsFtestanyEi"}
+      %13:2 = hlfir.declare %12 {uniq_name = "_QMmtestsFtestanyEi"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
+      %14 = fir.alloca i32 {bindc_name = "iam", uniq_name = "_QMmtestsFtestanyEiam"}
+      %15:2 = hlfir.declare %14 {uniq_name = "_QMmtestsFtestanyEiam"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
+      %16 = fir.alloca i32 {bindc_name = "j", uniq_name = "_QMmtestsFtestanyEj"}
+      %17:2 = hlfir.declare %16 {uniq_name = "_QMmtestsFtestanyEj"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
+      %c-1_1 = arith.constant -1 : index
+      %18 = cuf.shared_memory !fir.array<?xf32>, %c-1_1 : index {bindc_name = "smasks", uniq_name = "_QMmtestsFtestanyEsmasks"} -> !fir.ref<!fir.array<?xf32>>
+      %19 = fir.shape %c-1_1 : (index) -> !fir.shape<1>
+      %20:2 = hlfir.declare %18(%19) {data_attr = #cuf.cuda<shared>, uniq_name = "_QMmtestsFtestanyEsmasks"} : (!fir.ref<!fir.array<?xf32>>, !fir.shape<1>) -> (!fir.box<!fir.array<?xf32>>, !fir.ref<!fir.array<?xf32>>)
+      gpu.return
+    }
+  }
+}
 
+// CHECK-LABEL: gpu.func @_QMmtestsPtestany
+// CHECK: %{{.*}} = cuf.shared_memory[%c0{{.*}} : i32] !fir.array<?xf64>, %c-1{{.*}} : index {bindc_name = "dmasks", uniq_name = "_QMmtestsFtestanyEdmasks"} -> !fir.ref<!fir.array<?xf64>>
+// CHECK: %{{.*}} = cuf.shared_memory[%c0{{.*}} : i32] !fir.array<?xf32>, %c-1{{.*}} : index {bindc_name = "smasks", uniq_name = "_QMmtestsFtestanyEsmasks"} -> !fir.ref<!fir.array<?xf32>>