[BUG] Modifying the block/warptile shapes and the output datatype in the unit test causes the tests to fail.

Describe the bug
I modified the block/warptile shapes and the output datatype in https://github.com/NVIDIA/cutlass/blob/main/test/unit/gemm/device/gemm_s8t_s8n_s32t_tensor_op_s32_sm80.cu, and found some shapes cause the tests to fail.
I modified the ElementOutput to cutlass::half_t and tested various block/warptile shapes. While some shapes passed, others failed.

Passed block/warptile shapes, for example, are
<32,128,64>/<16,64,64>,
<32,128,64>/<32,32,64>,
<32,128,64>/<32,64,64>,
<32,256,64>/<32,64,64>,
<64,32,64>/<32,32,64>,
<64,32,64>/<64,32,64>, etc.

Failed block/warptile shapes, I found, are
<16,16,64>/<16,16,64>,
<16,128,64>/<16,128,64>,
<16,256,64>/<16,128,64>,
<32,16,64>/<32,16,64>,
<32,128,64>/<16,128,64>,
<32,128,64>/<32,128,64>,
<32,256,64>/<16,128,64>,
<32,256,64>/<32,128,64>,
<64,16,64>/<64,16,64>,
<64,128,64>/<16,128,64>,
<64,128,64>/<32,128,64>,
<64,128,64>/<64,128,64>,
<64,256,64>/<16,128,64>,
<64,256,64>/<32,128,64>,
<64,256,64>/<64,128,64>,
<128,16,64>/<128,16,64>,
<128,32,64>/<128,32,64>,
<128,64,64>/<128,64,64>,
<128,128,64>/<16,128,64>,
<128,128,64>/<32,128,64>,
<128,128,64>/<64,128,64>,
<128,256,64>/<128,128,64>,
<256,32,64>/<128,32,64>,
<256,64,64>/<128,64,64>,
<256,128,64>/<32,128,64>,
<256,128,64>/<64,128,64>,
<256,128,64>/<128,128,64>.

Steps/Code to reproduce bug

CUTLASS_TEST_L1(SM80_Device_Gemm_s8t_s8n_s32t_tensor_op_s32, 64x128x64_32x128x64, {
  using ElementOutput = cutlass::half_t;
  using ElementAccumulator = int32_t;
  using ElementCompute = int32_t;

  using Gemm = cutlass::gemm::device::Gemm<
      int8_t, cutlass::layout::RowMajor, int8_t,
      cutlass::layout::ColumnMajor, ElementOutput, cutlass::layout::RowMajor,
      ElementAccumulator, cutlass::arch::OpClassTensorOp, cutlass::arch::Sm80,
      cutlass::gemm::GemmShape<64, 128, 64>,
      cutlass::gemm::GemmShape<32, 128, 64>, cutlass::gemm::GemmShape<16, 8, 32>,
      cutlass::epilogue::thread::LinearCombination<
          ElementOutput, 128 / cutlass::sizeof_bits<ElementOutput>::value,
          ElementAccumulator, ElementCompute>,
      cutlass::gemm::threadblock::GemmIdentityThreadblockSwizzle<>, 3>;

  EXPECT_TRUE(test::gemm::device::TestAllGemm<Gemm>());
} )

Start with test/unit/gemm/device/gemm_s8t_s8n_s32t_tensor_op_s32_sm80.cu and the shown SM80_Device_Gemm_s8t_s8n_s32t_tensor_op_s32 test case. Reproduce the listed passing and failing block/warptile shapes after changing ElementOutput to cutlass::half_t, then compare their results and failures. Done means the reported failing configurations pass without regressing the listed passing cases.