Debug self host: Atomic ops on non-global memory not allowed

kchristin22 · Oct 8, 2024 · 880a77b · 880a77b
1 parent 54c8954
commit 880a77b
Show file tree

Hide file tree

Showing 2 changed files with 3 additions and 5 deletions.
diff --git a/lib/Differentiator/ReverseModeVisitor.cpp b/lib/Differentiator/ReverseModeVisitor.cpp
@@ -105,9 +105,7 @@ Expr* getArraySizeExpr(const ArrayType* AT, ASTContext& context,
   }
 
   bool ReverseModeVisitor::shouldUseCudaAtomicOps() {
-    return m_DiffReq->hasAttr<clang::CUDAGlobalAttr>() ||
-           (m_DiffReq->hasAttr<clang::CUDADeviceAttr>() &&
-            !m_DiffReq->hasAttr<clang::CUDAHostAttr>());
+    return m_DiffReq->hasAttr<clang::CUDAGlobalAttr>();
   }
 
   clang::Expr* ReverseModeVisitor::BuildCallToCudaAtomicAdd(clang::Expr* LHS,

diff --git a/test/CUDA/GradientKernels.cu b/test/CUDA/GradientKernels.cu
@@ -391,8 +391,8 @@ double fn_memory(double *out, double *in) {
 
 // CHECK: __attribute__((device)) void device_fn_pullback(double in, double val, double _d_y, double *_d_in, double *_d_val) {
 //CHECK-NEXT:    {
-//CHECK-NEXT:        atomicAdd(_d_in, _d_y);
-//CHECK-NEXT:        atomicAdd(_d_val, _d_y);
+//CHECK-NEXT:                *_d_in += _d_y;
+//CHECK-NEXT:                *_d_val += _d_y;
 //CHECK-NEXT:    }
 //CHECK-NEXT:}