[Cherry-pick]Cherry pick 0d output

paddle/fluid/operators/matrix_rank_op.cc

@@ -27,7 +27,7 @@ namespace detail {
 static DDim CheckAndGetOutputDim(const DDim& dim_x) {
   auto x_vec = phi::vectorize(dim_x);
   if (x_vec.size() == 2) {
-    return phi::make_ddim({1});
+    return phi::make_ddim({});
   }
   x_vec.erase(x_vec.end() - 2, x_vec.end());
   return phi::make_ddim(x_vec);

paddle/phi/infermeta/binary.cc

@@ -72,7 +72,7 @@ static void BinarySameInputDimsCheck(const MetaTensor& x,
 static DDim CheckAndGetOutputDim(const DDim& dim_x) {
   auto x_vec = phi::vectorize(dim_x);
   if (x_vec.size() == 2) {
-    return phi::make_ddim({1});
+    return phi::make_ddim({});
   }
   x_vec.erase(x_vec.end() - 2, x_vec.end());
   return phi::make_ddim(x_vec);
@@ -990,7 +990,7 @@ void DistInferMeta(const MetaTensor& x,
                         "The Input(Y) has not been initialized properly. The "
                         "shape of Input(Y) = [%s].",
                         y_dims));
-  out->set_dims({1});
+  out->set_dims(phi::make_ddim({}));
   out->set_dtype(x.dtype());
 }
 

paddle/phi/infermeta/multiary.cc

@@ -2344,7 +2344,7 @@ void MultiDotInferMeta(const std::vector<const MetaTensor*>& x,
   // If the last tensor is 1D of size n view it as a column vector (n, 1)
   if (last_dim.size() == 1) {
     last_dim = phi::make_ddim({static_cast<int>(last_dim[0]), 1});
-    out_dim = is_vector ? phi::make_ddim({1}) : phi::make_ddim({first_dim[0]});
+    out_dim = is_vector ? phi::make_ddim({}) : phi::make_ddim({first_dim[0]});
   } else {
     out_dim = is_vector ? phi::make_ddim({last_dim[1]})
                         : phi::make_ddim({first_dim[0], last_dim[1]});

paddle/phi/infermeta/unary.cc

@@ -38,7 +38,7 @@ namespace detail {
 static DDim CheckAndGetOutputDim(const DDim& dim_x) {
   auto x_vec = phi::vectorize(dim_x);
   if (x_vec.size() == 2) {
-    return phi::make_ddim({1});
+    return phi::make_ddim({});
   }
   x_vec.erase(x_vec.end() - 2, x_vec.end());
   return phi::make_ddim(x_vec);
@@ -4405,7 +4405,6 @@ void TraceInferMeta(
   auto sizes = vectorize(x_dims);
   if (x_dims.size() == 2) {
     sizes.clear();
-    sizes.push_back(1);
   } else {
     sizes.erase(sizes.begin() + std::max(dim1_, dim2_));
     sizes.erase(sizes.begin() + std::min(dim1_, dim2_));

paddle/phi/kernels/gpu/trace_kernel.cu

@@ -32,7 +32,10 @@ void TraceKernel(const Context& ctx,
   auto diag = funcs::Diagonal<T, Context>(ctx, &x, offset, axis1, axis2);
   if (diag.numel() > 0) {
     std::vector<int> reduce_dims;
-    reduce_dims.push_back(out->dims().size());
+    // Adapt to 0D output
+    auto out_dim_size = out->dims().size();
+    if (out_dim_size == 0) out_dim_size = 1;
+    reduce_dims.push_back(out_dim_size);
     funcs::ReduceKernel<T, T, kps::AddFunctor, kps::IdentityFunctor<T>>(
         ctx, diag, out, kps::IdentityFunctor<T>(), reduce_dims);
   } else {

paddle/phi/kernels/impl/determinant_grad_kernel_impl.h

@@ -90,10 +90,10 @@ void DeterminantGradKernel(const Context& dev_ctx,
             " input tensor's, but here differ %d",
             input_dims_size - out_grad.dims().size()));
   } else if (input_dims_size == 2) {
-    // input dims size 2 and grad dims size 1 is possible
+    // input dims size 2 and grad dims size 0 is possible
     PADDLE_ENFORCE_EQ(
         out_grad.dims().size(),
-        1,
+        0,
         phi::errors::InvalidArgument(
             "The grad tensor of det dims size should be 2 less than"
             " input tensor's, but here differ %d",

paddle/phi/kernels/impl/determinant_kernel_impl.h

@@ -116,7 +116,7 @@ void DeterminantKernel(const Context& dev_ctx,
     out->Resize(output_dims);
   } else {
     // when input is a two-dimension matrix, The det value is a number.
-    out->Resize({1});
+    out->Resize(phi::make_ddim({}));
   }
   VLOG(10) << "output dim:" << out->dims();
 }

paddle/phi/kernels/impl/trace_grad_kernel_impl.h

@@ -91,7 +91,8 @@ void TraceGradKernel(const Context& ctx,
   auto input_dims = in_grad->dims();
   auto input_stride = phi::stride(input_dims);
   auto output_dims = out_grad.dims();
-  auto output_stride = phi::stride(output_dims);
+  auto output_stride = output_dims.size() == 0 ? phi::DDim(output_dims)
+                                               : phi::stride(output_dims);
 
   auto* out_data = out_grad.data<T>();
   T* x_data = ctx.template Alloc<T>(in_grad);