-
Notifications
You must be signed in to change notification settings - Fork 11.4k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[flang][runtime] Fix another IsContiguous edge case #69199
Conversation
A recent PR addressed zero and one element edge cases but did not cover another case where the descriptors of arrays with more than two elements may have byte strides that are not perfect multiples, like when creating a descriptor for A(:, 1:1:2). In general, the byte stride in a dimension is only meaningful if that dimension has more than one element. Update IsContiguous and CFI_is_contiguous to reflect that.
@llvm/pr-subscribers-flang-semantics @llvm/pr-subscribers-flang-runtime Author: None (jeanPerier) ChangesA recent PR addressed zero and one element edge cases but did not cover another case where the descriptors of arrays with more than two elements may have byte strides that are not perfect multiples, like when creating a descriptor for A(:, 1:1:2). In general, the byte stride in a dimension is only meaningful if that dimension has more than one element. Update IsContiguous and CFI_is_contiguous to reflect that. Full diff: https://github.com/llvm/llvm-project/pull/69199.diff 3 Files Affected:
diff --git a/flang/include/flang/Runtime/descriptor.h b/flang/include/flang/Runtime/descriptor.h
index c69bb336dd29e7d..3eea3f30352947b 100644
--- a/flang/include/flang/Runtime/descriptor.h
+++ b/flang/include/flang/Runtime/descriptor.h
@@ -393,13 +393,17 @@ class Descriptor {
bool stridesAreContiguous{true};
for (int j{0}; j < leadingDimensions; ++j) {
const Dimension &dim{GetDimension(j)};
- stridesAreContiguous &= bytes == dim.ByteStride();
+ stridesAreContiguous &= bytes == dim.ByteStride() | dim.Extent() == 1;
bytes *= dim.Extent();
}
// One and zero element arrays are contiguous even if the descriptor
// byte strides are not perfect multiples.
- return stridesAreContiguous || bytes == 0 ||
- bytes == static_cast<SubscriptValue>(ElementBytes());
+ // Arrays with more than 2 elements may also be contiguous even if a
+ // byte stride in one dimension is not a perfect multiple, as long as
+ // this is the last dimension, or if the dimension has one extent and
+ // the following dimension have either one extents or contiguous byte
+ // strides.
+ return stridesAreContiguous || bytes == 0;
}
// Establishes a pointer to a section or element.
diff --git a/flang/runtime/ISO_Fortran_binding.cpp b/flang/runtime/ISO_Fortran_binding.cpp
index 103413cb7140aaa..c6320745ba17697 100644
--- a/flang/runtime/ISO_Fortran_binding.cpp
+++ b/flang/runtime/ISO_Fortran_binding.cpp
@@ -125,16 +125,15 @@ RT_API_ATTRS int CFI_establish(CFI_cdesc_t *descriptor, void *base_addr,
}
RT_API_ATTRS int CFI_is_contiguous(const CFI_cdesc_t *descriptor) {
+ // See Descriptor::IsContiguous for the rational.
bool stridesAreContiguous{true};
CFI_index_t bytes = descriptor->elem_len;
for (int j{0}; j < descriptor->rank; ++j) {
- stridesAreContiguous &= bytes == descriptor->dim[j].sm;
+ stridesAreContiguous &=
+ bytes == descriptor->dim[j].sm | descriptor->dim[j].extent == 1;
bytes *= descriptor->dim[j].extent;
}
- // One and zero element arrays are contiguous even if the descriptor
- // byte strides are not perfect multiples.
- if (stridesAreContiguous || bytes == 0 ||
- bytes == static_cast<CFI_index_t>(descriptor->elem_len)) {
+ if (stridesAreContiguous || bytes == 0) {
return 1;
}
return 0;
diff --git a/flang/unittests/Evaluate/ISO-Fortran-binding.cpp b/flang/unittests/Evaluate/ISO-Fortran-binding.cpp
index d1f0a31454056bf..3c98363f9004664 100644
--- a/flang/unittests/Evaluate/ISO-Fortran-binding.cpp
+++ b/flang/unittests/Evaluate/ISO-Fortran-binding.cpp
@@ -736,6 +736,38 @@ static void run_CFI_is_contiguous_tests() {
MATCH(true, retCode == CFI_SUCCESS);
MATCH(true, CFI_is_contiguous(section) == 0);
MATCH(false, sectionDesc->IsContiguous());
+
+ // Test section B = A(0:3:1,0:0:2) is contiguous.
+ lb[0] = 0;
+ lb[1] = 0;
+ ub[0] = 3;
+ ub[1] = 0;
+ strides[0] = 1;
+ strides[1] = 2;
+ retCode = CFI_section(section, dv, lb, ub, strides);
+ MATCH(true, retCode == CFI_SUCCESS);
+ MATCH(true, CFI_is_contiguous(section) == 1);
+ MATCH(true, sectionDesc->IsContiguous());
+
+ // INTEGER :: C(0:0, 0:3)
+ CFI_index_t c_extents[rank] = {1, 4};
+ CFI_CDESC_T(rank) c_dv_storage;
+ CFI_cdesc_t *cdv{&c_dv_storage};
+ retCode = CFI_establish(cdv, base_addr, CFI_attribute_other, CFI_type_int,
+ /*elem_len=*/0, rank, c_extents);
+ MATCH(retCode == CFI_SUCCESS, true);
+
+ // Test section B = C(0:0:2, 0:3:1) is contiguous.
+ lb[0] = 0;
+ lb[1] = 0;
+ ub[0] = 0;
+ ub[1] = 3;
+ strides[0] = 2;
+ strides[1] = 1;
+ retCode = CFI_section(section, cdv, lb, ub, strides);
+ MATCH(true, retCode == CFI_SUCCESS);
+ MATCH(true, CFI_is_contiguous(section) == 1);
+ MATCH(true, sectionDesc->IsContiguous());
}
int main() {
|
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Once I fixed all of the warnings, all built and tested successfully.
Thanks for fixing this bug!
Rational -> Rationale. Silence warning with extra parenthesis.
A recent PR addressed zero and one element edge cases but did not cover another case where the descriptors of arrays with more than two elements may have byte strides that are not perfect multiples, like when creating a descriptor for A(:, 1:1:2).
In general, the byte stride in a dimension is only meaningful if that dimension has more than one element. Update IsContiguous and CFI_is_contiguous to reflect that.