Revert "[NPU] split to aclnn. (PaddlePaddle#1218)" (PaddlePaddle#1295)

Author: tianhaodongbd

backends/npu/kernels/split_kernel.cc

@@ -18,11 +18,11 @@
 namespace custom_kernel {
 
 template <typename T, typename Context>
-void AclopSplitKernel(const Context& dev_ctx,
-                      const phi::DenseTensor& x,
-                      const phi::IntArray& num_or_sections,
-                      const phi::Scalar& axis_scalar,
-                      std::vector<phi::DenseTensor*> outs) {
+void SplitKernel(const Context& dev_ctx,
+                 const phi::DenseTensor& x,
+                 const phi::IntArray& num_or_sections,
+                 const phi::Scalar& axis_scalar,
+                 std::vector<phi::DenseTensor*> outs) {
   // need to infershape output
   auto sections = num_or_sections.GetData();
   int axis = axis_scalar.to<int>();
@@ -77,58 +77,6 @@ void AclopSplitKernel(const Context& dev_ctx,
   }
 }
 
-template <typename T, typename Context>
-void SplitKernel(const Context& dev_ctx,
-                 const phi::DenseTensor& x,
-                 const phi::IntArray& num_or_sections,
-                 const phi::Scalar& axis_scalar,
-                 std::vector<phi::DenseTensor*> outs) {
-  auto sections = num_or_sections.GetData();
-  int64_t axis = axis_scalar.to<int64_t>();
-  if (axis < 0) {
-    axis = x.dims().size() + axis;
-  }
-
-  std::vector<phi::DenseTensor*> outputs;
-  for (size_t j = 0; j < outs.size(); ++j) {
-    dev_ctx.template Alloc<T>(outs[j]);
-    outputs.push_back(outs[j]);
-  }
-
-  if (!num_or_sections.FromTensor() && !axis_scalar.FromTensor() &&
-      sections.size() == 1 && outs.size() == sections[0]) {
-    DO_COMPATIBILITY(aclnnSplitTensor,
-                     (custom_kernel::AclopSplitKernel<T, Context>(
-                         dev_ctx, x, num_or_sections, axis_scalar, outs)));
-
-    uint64_t splitSections = x.dims()[axis] / sections[0];
-    EXEC_NPU_CMD(aclnnSplitTensor, dev_ctx, x, splitSections, axis, outputs);
-  } else {
-    DO_COMPATIBILITY(aclnnSplitWithSize,
-                     (custom_kernel::AclopSplitKernel<T, Context>(
-                         dev_ctx, x, num_or_sections, axis_scalar, outs)));
-
-    std::vector<int64_t> sections_;
-    int sum = 0;
-    int minusOneIndex = -1;
-    int all = x.dims()[axis];
-
-    for (int i = 0; i < sections.size(); ++i) {
-      sections_.push_back(sections[i]);
-      if (sections_[i] == -1) {
-        minusOneIndex = i;
-      } else {
-        sum += sections_[i];
-      }
-    }
-    if (minusOneIndex != -1) {
-      sections_[minusOneIndex] = all - sum;
-    }
-
-    EXEC_NPU_CMD(aclnnSplitWithSize, dev_ctx, x, sections_, axis, outputs);
-  }
-}
-
 template <typename T, typename Context>
 void SplitWithNumKernel(const Context& dev_ctx,
                         const phi::DenseTensor& x,