[Binding] ReduceOp should load element from buffer first

[chhzh123]

The current Python binding of ReduceOp returns the reduction variable (rv). However, since we actually create a size-1 buffer for the rv, it is not a register and should be read before other operations. Otherwise, it may cause the following error. (Example: sum(A[i, k] * B[k, j]) + 10)

error: 'arith.addi' op requires the same type for all operands and results
// Verification failed, printing generic form
#map0 = affine_map<(d0) -> (d0)>
#map1 = affine_map<(d0, d1) -> (d0, d1)>
#map2 = affine_map<() -> (0)>
#map3 = affine_map<() -> (32)>
"builtin.module"() ({
  "builtin.func"() ({
  ^bb0(%arg0: memref<32x32xi32>, %arg1: memref<32x32xi32>):
    %0 = "memref.alloc"() {name = "C", operand_segment_sizes = dense<0> : vector<2xi32>} : () -> memref<32x32xi32>
    %1 = "hcl.create_stage_handle"() {stage_name = "C"} : () -> !hcl.StageHandle
    %2 = "hcl.create_loop_handle"() {loop_name = "i"} : () -> !hcl.LoopHandle
    %3 = "hcl.create_loop_handle"() {loop_name = "j"} : () -> !hcl.LoopHandle
    %4 = "memref.alloc"() {name = "C", operand_segment_sizes = dense<0> : vector<2xi32>} : () -> memref<32x32xi32>
    "affine.for"() ({
    ^bb0(%arg2: index):
      "affine.for"() ({
      ^bb0(%arg3: index):
        %5 = "memref.alloc"() {name = "sum_rv", operand_segment_sizes = dense<0> : vector<2xi32>} : () -> memref<1xi32>
        %6 = "arith.constant"() {value = 0 : index} : () -> index
        %7 = "arith.constant"() {value = 0 : i32} : () -> i32
        "affine.store"(%7, %5, %6) {map = #map0, to = "sum_rv"} : (i32, memref<1xi32>, index) -> ()
        "affine.for"() ({
        ^bb0(%arg4: index):
          %10 = "affine.load"(%arg0, %arg2, %arg4) {from = "A", map = #map1} : (memref<32x32xi32>, index, index) -> i32
          %11 = "affine.load"(%arg1, %arg4, %arg3) {from = "B", map = #map1} : (memref<32x32xi32>, index, index) -> i32
          %12 = "arith.muli"(%10, %11) : (i32, i32) -> i32
          %13 = "affine.load"(%5, %6) {from = "sum_rv", map = #map0} : (memref<1xi32>, index) -> i32
          %14 = "arith.addi"(%12, %13) : (i32, i32) -> i32
          "affine.store"(%14, %5, %6) {map = #map0, to = "sum_rv"} : (i32, memref<1xi32>, index) -> ()
          "affine.yield"() : () -> ()
        }) {loop_name = "k", lower_bound = #map2, step = 1 : i32, upper_bound = #map3} : () -> ()
        %8 = "arith.constant"() {value = 10 : i32} : () -> i32
        %9 = "arith.addi"(%5, %8) : (memref<1xi32>, i32) -> memref<1xi32>
        "affine.store"(%9, %4, %arg2, %arg3) {map = #map1, to = "C"} : (memref<1xi32>, memref<32x32xi32>, index, index) -> ()
        "affine.yield"() : () -> ()
      }) {loop_name = "j", lower_bound = #map2, step = 1 : i32, upper_bound = #map3} : () -> ()
      "affine.yield"() : () -> ()
    }) {loop_name = "i", lower_bound = #map2, stage_name = "C", step = 1 : i32, upper_bound = #map3} : () -> ()
    "std.return"(%4) : (memref<32x32xi32>) -> ()
  }) {sym_name = "top", type = (memref<32x32xi32>, memref<32x32xi32>) -> memref<32x32xi32>} : () -> ()
}) : () -> ()

where %9 = "arith.addi"(%5, %8) : (memref<1xi32>, i32) -> memref<1xi32> tries to add the reduction variable but should load the element first.