[all-commits] [llvm/llvm-project] a1c736: [Flang] Cray pointer Lowering

[madanial0 via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: a1c736ec08f25e83552b20c94a5b2afdcd021a40
      https://github.com/llvm/llvm-project/commit/a1c736ec08f25e83552b20c94a5b2afdcd021a40
  Author: Mark Danial <mark.danial at ibm.com>
  Date:   2023-08-22 (Tue, 22 Aug 2023)

  Changed paths:
    M flang/include/flang/Lower/ConvertExpr.h
    M flang/lib/Lower/Bridge.cpp
    M flang/lib/Lower/ConvertExpr.cpp
    A flang/test/Lower/cray-pointer.f90

  Log Message:
  -----------
  [Flang] Cray pointer Lowering

This patch is to add cray pointer (aka integer pointer) support to flang. Syntax and semantic checking were already available in flang.
Cray pointers reference (https://gcc.gnu.org/onlinedocs/gfortran/Cray-pointers.html)

In order to implement the feature we create the following sequence for a simple scalar load and store:

```
integer pte, i
pointer(ptr, pte)
i = pte
```

```
    %1 = fir.alloca i32 {bindc_name = "i", uniq_name = "_QFEi"}
    %2 = fir.alloca i32 {bindc_name = "pte", uniq_name = "_QFEpte"}
    %3 = fir.alloca i64 {bindc_name = "ptr", uniq_name = "_QFEptr"}
    ...
    %7 = fir.embox %3 : (!fir.ref<i64>) -> !fir.box<i64>
    %8 = fir.box_addr %7 : (!fir.box<i64>) -> !fir.ref<i64>
    %9 = fir.convert %8 : (!fir.ref<i64>) -> !fir.ref<!fir.ptr<i32>>
    %10 = fir.load %9 : !fir.ref<!fir.ptr<i32>>
    %11 = fir.load %10 : !fir.ptr<i32>
    fir.store %11 to %1 : !fir.ref<i32>
```

```
  integer pte, i
  pointer(ptr, pte)
  pte = i
```

```
    %1 = fir.alloca i32 {bindc_name = "i", uniq_name = "_QFEi"}
    %2 = fir.alloca i32 {bindc_name = "pte", uniq_name = "_QFEpte"}
    %3 = fir.alloca i64 {bindc_name = "ptr", uniq_name = "_QFEptr"}

    %7 = fir.load %1 : !fir.ref<i32>
    %8 = fir.embox %3 : (!fir.ref<i64>) -> !fir.box<i64>
    %9 = fir.box_addr %8 : (!fir.box<i64>) -> !fir.ref<i64>
    %10 = fir.convert %9 : (!fir.ref<i64>) -> !fir.ref<!fir.ptr<i32>>
    %11 = fir.load %10 : !fir.ref<!fir.ptr<i32>>
    fir.store %7 to %11 : !fir.ptr<i32>
```
The sequence is very similar for array element cases with the addition of fir.coordinate_of for the specific element.
The whole array case is slightly different but uses the same sequence before the fir.array_load and fir.array_merge_store.

Reviewed By: kkwli0

Differential Revision: https://reviews.llvm.org/D151478