address logan's comments

src/runtime/micro/openocd_low_level_device.cc

@@ -207,8 +207,8 @@ class OpenOCDLowLevelDevice final : public LowLevelDevice {
 
   /*! \brief number of bytes in a word on the target device (64-bit) */
   static const constexpr ssize_t kWordSize = 8;
-  // NOTE: OpenOCD will call any request larger than this constant an "absurd
-  // request".
+  // NOTE: The OS pipe buffer must be able to handle a line long enough to
+  // print this transfer request.
   /*! \brief maximum number of bytes allowed in a single memory transfer */
   static const constexpr ssize_t kMemTransferLimit = 8000;
   /*! \brief number of milliseconds to wait for function execution to halt */

topi/python/topi/arm_cpu/conv2d_spatial_pack.py

@@ -173,7 +173,7 @@ def schedule_conv2d_spatial_pack_nchw(cfg, s, data_vec, kernel_vec,
                                  axis_lens=[cfg['tile_oh'].size[-1],
                                             cfg['tile_ow'].size[-1],
                                             cfg['tile_co'].size[-1]],
-                                 max_unroll=None,
+                                 max_unroll=16,
                                  cfg=cfg)
     s[conv].compute_at(s[last], ow)
 

topi/python/topi/arm_cpu/cortex_m7/micro_kernel/gemm.py

@@ -34,7 +34,7 @@ def intrin_gemm_MxKxN(M, K, N, in_dtype, out_dtype):
     # collisions in the generated source (e.g., if there are multiple operators
     # in the same module that use the same intrinsic)
     #
-    # TODO to cut down on memory usage, we should cache each intrinsic
+    # TODO(areusch): to cut down on memory usage, we should cache each intrinsic
     # instantiation and include it only once, eliminating the need for unique
     # IDs
     UNIQ_ID_LEN = 8
@@ -47,7 +47,7 @@ def intrin_gemm_MxKxN(M, K, N, in_dtype, out_dtype):
     if isinstance(N, tvm.tir.IntImm):
         N = N.value
     assert K % 4 == 0
-    # TODO support more dtypes?
+    # TODO(areusch): support more dtypes?
     assert in_dtype == 'int8'
     assert out_dtype == 'int32'
     A = te.placeholder((M, K), name='a', dtype=in_dtype)
@@ -124,7 +124,7 @@ def _body():
 
 def gemm_MxKxN_impl(M, K, N, uniq_id):
     """Emit C code for gemm impl."""
-    # TODO are there any SIMD tricks to zero out arrays quickly?
+    # TODO(areusch): are there any SIMD tricks to zero out arrays quickly?
     aa_pad_size = M * K
     bb_pad_size = N * K
     # code reference: CMSIS-NN paper (https://arxiv.org/abs/1801.06601)