[flang] Improve alias analysis to be precise for box and box.base_addr #80335
Conversation
After PR#68727 the source for both the fir.box_addr and a box became the same. Thus the detection that only one of the sources was direct and the special logic around it was being skipped. As a result, the test included would show a "MayAlias" result instead of a "NoAlias" result.
llvmbot
added
flang
|
@llvm/pr-subscribers-flang-fir-hlfir Author: Razvan Lupusoru (razvanlupusoru) ChangesAfter PR#68727 the source for both the fir.box_addr and a box became the same. Thus the detection that only one of the sources was direct and the special logic around it was being skipped. As a result, the test included would show a "MayAlias" result instead of a "NoAlias" result. Full diff: https://github.com/llvm/llvm-project/pull/80335.diff 2 Files Affected:
diff --git a/flang/lib/Optimizer/Analysis/AliasAnalysis.cpp b/flang/lib/Optimizer/Analysis/AliasAnalysis.cpp
index 90072eea323be..ea2064e32a44d 100644
--- a/flang/lib/Optimizer/Analysis/AliasAnalysis.cpp
+++ b/flang/lib/Optimizer/Analysis/AliasAnalysis.cpp
@@ -106,7 +106,7 @@ AliasResult AliasAnalysis::alias(Value lhs, Value rhs) {
// Though nothing is known about them, they would only alias with targets or
// pointers
bool directSourceToNonTargetOrPointer = false;
- if (lhsSrc.u != rhsSrc.u) {
+ if (lhsSrc.u != rhsSrc.u || lhsSrc.kind != rhsSrc.kind) {
if ((lhsSrc.kind == SourceKind::Direct && !rhsSrc.isTargetOrPointer()) ||
(rhsSrc.kind == SourceKind::Direct && !lhsSrc.isTargetOrPointer()))
directSourceToNonTargetOrPointer = true;
diff --git a/flang/test/Analysis/AliasAnalysis/alias-analysis-8.fir b/flang/test/Analysis/AliasAnalysis/alias-analysis-8.fir
new file mode 100644
index 0000000000000..a04c9523f472a
--- /dev/null
+++ b/flang/test/Analysis/AliasAnalysis/alias-analysis-8.fir
@@ -0,0 +1,40 @@
+// RUN: fir-opt %s -pass-pipeline='builtin.module(func.func(test-fir-alias-analysis))' 2>&1 | FileCheck %s
+
+// Exercise that a box reference does not alias with the address loaded from the box
+// module mm
+// real, allocatable :: arr(:)
+// contains
+// subroutine sub
+// arr(1) = ubound(arr, 1)
+// end subroutine
+// end module
+
+// CHECK: box#0 <-> boxaddr#0: NoAlias
+
+fir.global @_QMmmEarr : !fir.box<!fir.heap<!fir.array<?xf32>>> {
+ %c0 = arith.constant 0 : index
+ %0 = fir.zero_bits !fir.heap<!fir.array<?xf32>>
+ %1 = fir.shape %c0 : (index) -> !fir.shape<1>
+ %2 = fir.embox %0(%1) : (!fir.heap<!fir.array<?xf32>>, !fir.shape<1>) -> !fir.box<!fir.heap<!fir.array<?xf32>>>
+ fir.has_value %2 : !fir.box<!fir.heap<!fir.array<?xf32>>>
+}
+func.func @_QMmmPsub() {
+ %c1 = arith.constant 1 : index
+ %c1_i64 = arith.constant 1 : i64
+ %c0 = arith.constant 0 : index
+ %0 = fir.address_of(@_QMmmEarr) : !fir.ref<!fir.box<!fir.heap<!fir.array<?xf32>>>>
+ %1 = fir.declare %0 {fortran_attrs = #fir.var_attrs<allocatable>, uniq_name = "_QMmmEarr", test.ptr = "box"} : (!fir.ref<!fir.box<!fir.heap<!fir.array<?xf32>>>>) -> !fir.ref<!fir.box<!fir.heap<!fir.array<?xf32>>>>
+ %2 = fir.load %1 : !fir.ref<!fir.box<!fir.heap<!fir.array<?xf32>>>>
+ %3:3 = fir.box_dims %2, %c0 : (!fir.box<!fir.heap<!fir.array<?xf32>>>, index) -> (index, index, index)
+ %4 = fir.convert %3#1 : (index) -> i64
+ %5 = fir.convert %3#0 : (index) -> i64
+ %6 = arith.addi %4, %5 : i64
+ %7 = arith.subi %6, %c1_i64 : i64
+ %8 = fir.convert %7 : (i64) -> i32
+ %9 = fir.convert %8 : (i32) -> f32
+ %10 = fir.box_addr %2 {test.ptr = "boxaddr"} : (!fir.box<!fir.heap<!fir.array<?xf32>>>) -> !fir.heap<!fir.array<?xf32>>
+ %11 = fir.shape_shift %3#0, %3#1 : (index, index) -> !fir.shapeshift<1>
+ %12 = fir.array_coor %10(%11) %c1 : (!fir.heap<!fir.array<?xf32>>, !fir.shapeshift<1>, index) -> !fir.ref<f32>
+ fir.store %9 to %12 : !fir.ref<f32>
+ return
+}
|
There was a problem hiding this comment.
This makes sense to me, but please wait for approval from at least one of the other reviewers.
Out of interest, are you able to share what you are using this for? I believe the existing alias analysis infrastructure should already be able to distinguish between box and non-box accesses.
I was trying it out on some OpenACC. I am interested in creating an alias analysis framework (at MLIR level) for it as well.
Are you referring to the TBAA implementation? As far as I understood this was mostly for passing information to LLVM, not for alias analysis in MLIR passes. |
Yes that was what I meant. Yes it is not for MLIR passes. |
llvm#80335) After PR#68727 the source for both the fir.box_addr and a box became the same. Thus the detection that only one of the sources was direct and the special logic around it was being skipped. As a result, the test included would show a "MayAlias" result instead of a "NoAlias" result.
After PR#68727 the source for both the fir.box_addr and a box became the same. Thus the detection that only one of the sources was direct and the special logic around it was being skipped. As a result, the test included would show a "MayAlias" result instead of a "NoAlias" result.