yggdrasil/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

[OpenMP] Fix calculation of dependencies for multi-dimensional iteration space (#99347)

Browse Source 
The expectation for multiple iterators used in a single depend clause
(`depend(iterator(i=0:5,j=0:5), in:x[i][j])`) is that the iterator space
is the product of the iteration vectors (25 in that case). The current
codeGen only works correctly, if `numIterators() = 1`. For more
iterators, the execution results in runtime assertions or segfaults.
The modified codeGen first calculates the iteration space, then
multiplies to the number of dependencies in the depend clause and
finally adds to the total number of iterator dependencies.
This commit is contained in:
Joachim 2024-07-18 07:41:41 +02:00 committed by GitHub
parent 27ee33d136
commit 4782a4ab0a
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
3 changed files with 39 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
clang/lib/CodeGen/CGOpenMPRuntime.cpp
View File

@ -4259,14 +4259,18 @@ std::pair<llvm::Value *, Address> CGOpenMPRuntime::emitDependClause(
// Include number of iterations, if any.
if (const auto *IE = cast_or_null<OMPIteratorExpr>(D.IteratorExpr)) {
llvm::Value *ClauseIteratorSpace =
llvm::ConstantInt::get(CGF.IntPtrTy, 1);
for (unsigned I = 0, E = IE->numOfIterators(); I < E; ++I) {
llvm::Value *Sz = CGF.EmitScalarExpr(IE->getHelper(I).Upper);
Sz = CGF.Builder.CreateIntCast(Sz, CGF.IntPtrTy, /*isSigned=*/false);
ClauseIteratorSpace = CGF.Builder.CreateNUWMul(Sz, ClauseIteratorSpace);
}
llvm::Value *NumClauseDeps = CGF.Builder.CreateNUWMul(
Sz, llvm::ConstantInt::get(CGF.IntPtrTy, D.DepExprs.size()));
ClauseIteratorSpace,
llvm::ConstantInt::get(CGF.IntPtrTy, D.DepExprs.size()));
NumOfRegularWithIterators =
CGF.Builder.CreateNUWAdd(NumOfRegularWithIterators, NumClauseDeps);
}
HasRegularWithIterators = true;
continue;
}

29
clang/test/OpenMP/depend_iterator_bug.c
View File

@ -5,6 +5,7 @@
int x[100];
int y[100];
int z[100][100];
// CHECK-LABEL: @many_iterators_single_clause(
// CHECK: [[VLA:%.*]] = alloca [[STRUCT_KMP_DEPEND_INFO:%.*]], i64 10, align 16
@ -24,3 +25,31 @@ void many_iterators_many_clauses(void) {
{
}
}
// CHECK-LABEL: @multidim_iterators_clause1(
// CHECK: [[VLA:%.*]] = alloca [[STRUCT_KMP_DEPEND_INFO:%.*]], i64 1, align 16
// CHECK: = call i32 @__kmpc_omp_task_with_deps(ptr {{.*}}, i32 {{.*}}, ptr {{.*}}, i32 1, ptr {{.*}}, i32 0, ptr null)
void multidim_iterators_clause1(void) {
#pragma omp task depend(iterator(i=0:1, j=0:1), in: z[i][j])
{
}
}
// CHECK-LABEL: @multidim_iterators_offset_clause(
// CHECK: [[VLA:%.*]] = alloca [[STRUCT_KMP_DEPEND_INFO:%.*]], i64 1, align 16
// CHECK: = call i32 @__kmpc_omp_task_with_deps(ptr {{.*}}, i32 {{.*}}, ptr {{.*}}, i32 1, ptr {{.*}}, i32 0, ptr null)
void multidim_iterators_offset_clause(void) {
#pragma omp task depend(iterator(i=5:6, j=10:11), in: z[i][j])
{
}
}
// CHECK-LABEL: @multidim_iterators_clause25(
// CHECK: [[VLA:%.*]] = alloca [[STRUCT_KMP_DEPEND_INFO:%.*]], i64 25, align 16
// CHECK: = call i32 @__kmpc_omp_task_with_deps(ptr {{.*}}, i32 {{.*}}, ptr {{.*}}, i32 25, ptr {{.*}}, i32 0, ptr null)
void multidim_iterators_clause25(void) {
#pragma omp task depend(iterator(i=0:5, j=0:5), in: z[i][j])
{
}
}

3
clang/test/OpenMP/task_codegen.c
View File

@ -139,7 +139,8 @@ for (int i = 0; i < 10; ++i)
// CHECK: [[EB_SUB_2_ADD_1_SUB:%.+]] = sub i32 [[EB_SUB_2_ADD]], 1
// CHECK: [[EB_SUB_2_ADD_1_SUB_2_DIV:%.+]] = udiv i32 [[EB_SUB_2_ADD_1_SUB]], 2
// CHECK: [[ELEMS:%.+]] = zext i32 [[EB_SUB_2_ADD_1_SUB_2_DIV]] to i64
// CHECK: [[NELEMS:%.+]] = mul nuw i64 [[ELEMS]], 1
// CHECK: [[ELEMS2:%.+]] = mul nuw i64 [[ELEMS]], 1
// CHECK: [[NELEMS:%.+]] = mul nuw i64 [[ELEMS2]], 1
// ITERATOR_TOTAL = NELEMS + 0;
// CHECK: [[ITERATOR_TOTAL:%.+]] = add nuw i64 0, [[NELEMS]]