[Driver, Util] change endianness reversal method for out unpack

src/finn/transformation/fpgadataflow/templates.py

@@ -86,7 +86,9 @@ assert ibuf_normal.shape == ishape_normal
 ibuf_folded = ibuf_normal.reshape(ishape_folded)
 
 # pack the input buffer, reversing both SIMD dim and endianness
-ibuf_packed = finnpy_to_packed_bytearray(ibuf_folded, idt, True, True)
+ibuf_packed = finnpy_to_packed_bytearray(
+    ibuf_folded, idt, reverse_endian=True, reverse_inner=True
+)
 # allocate a PYNQ buffer for the packed input buffer
 ibuf_packed_device = allocate(shape=ishape_packed, dtype=np.uint8)
 # copy the packed data into the PYNQ buffer
@@ -104,7 +106,7 @@ dma.recvchannel.wait()
 
 # unpack the packed output buffer from accelerator
 obuf_folded = packed_bytearray_to_finnpy(
-    obuf_packed, odt, oshape_folded, reverse_endian=True
+    obuf_packed, odt, oshape_folded, reverse_endian=True, reverse_inner=True
 )
 # convert to normal reshape and save
 obuf_normal = obuf_folded.reshape(oshape_normal)

src/finn/util/data_packing.py

@@ -258,7 +258,9 @@ def rtlsim_output_to_npy(
     return out_array
 
 
-def finnpy_to_packed_bytearray(ndarray, dtype, reverse_inner=False, reverse_endian=False):
+def finnpy_to_packed_bytearray(
+    ndarray, dtype, reverse_inner=False, reverse_endian=False
+):
     """Given a numpy ndarray with FINN DataType dtype, pack the innermost
     dimension and return the packed representation as an ndarray of uint8.
     The packed innermost dimension will be padded to the nearest multiple
@@ -316,14 +318,16 @@ def packed_bytearray_to_finnpy(
         assert packed_bits % target_bits == 0
         n_target_elems = packed_bits // target_bits
         output_shape = packed_bytearray.shape[:-1] + (n_target_elems,)
-    if reverse_endian and target_bits > 8:
-        # revse the endianness of each element
-        orig_shape = packed_bytearray.shape
-        assert target_bits % 8 == 0
-        target_bytes = target_bits // 8
-        new_shape = orig_shape[:-1] + (-1, target_bytes)
-        packed_bytearray = np.flip(packed_bytearray.reshape(new_shape), axis=-1)
-        packed_bytearray = packed_bytearray.reshape(orig_shape)
+    # if reverse_endian and target_bits > 8:
+    #     # revse the endianness of each element
+    #     orig_shape = packed_bytearray.shape
+    #     assert target_bits % 8 == 0
+    #     target_bytes = target_bits // 8
+    #     new_shape = orig_shape[:-1] + (-1, target_bytes)
+    #     packed_bytearray = np.flip(packed_bytearray.reshape(new_shape), axis=-1)
+    #     packed_bytearray = packed_bytearray.reshape(orig_shape)
+    if reverse_endian:
+        packed_bytearray = np.flip(packed_bytearray, axis=-1)
     # convert innermost dim of byte array to hex strings
     packed_hexstring = np.apply_along_axis(
         npbytearray2hexstring, packed_dim, packed_bytearray