diff options
| author | Tim Hall <tim.hall@arm.com> | 2020-06-25 16:55:02 +0100 |
|---|---|---|
| committer | Tim Hall <tim.hall@arm.com> | 2020-06-25 16:55:02 +0100 |
| commit | 6231676d1751caa353a51f9bdfc50642930f4802 (patch) | |
| tree | 60952c5d4593aa640e676f7d408a8ef7bf8d3521 /ethosu/vela/weight_compressor.py | |
| parent | f7e810a695c1426799d945d61671126543efc123 (diff) | |
| download | ethos-u-vela-6231676d1751caa353a51f9bdfc50642930f4802.tar.gz | |
MLBEDSW-2556: Odd core/block depth weight interleaving update
- If blockdepth or core count resulted in empty or non-existent substreams, the
command generator generated an error. This commit changes the command stream
generator to only program cores that have streams and are enabled for the
configuration.
Change-Id: I4e724b19de14d3a12e886ec6b17d0038593dfb59
Signed-off-by: Tim Hall <tim.hall@arm.com>
Diffstat (limited to 'ethosu/vela/weight_compressor.py')
| -rw-r--r-- | ethosu/vela/weight_compressor.py | 9 |
1 files changed, 7 insertions, 2 deletions
diff --git a/ethosu/vela/weight_compressor.py b/ethosu/vela/weight_compressor.py index fe8f04b9..d3562891 100644 --- a/ethosu/vela/weight_compressor.py +++ b/ethosu/vela/weight_compressor.py @@ -246,7 +246,6 @@ def compress_weights(arch, nng, tens, npu_block_type, ofm_block_depth, ofm_depth # Slice weight stream up depth-ways into bricks and compress full_ofm_depth = quant_buf.shape[-1] - ofm_block_depth = ofm_block_depth // arch.ncores for idx in range(0, full_ofm_depth, ofm_depth_step): # Get the weights necessary for this brick count = min(full_ofm_depth - idx, ofm_depth_step) @@ -260,7 +259,13 @@ def compress_weights(arch, nng, tens, npu_block_type, ofm_block_depth, ofm_depth # and generate separate compressed streams. for core in range(0, min(arch.ncores, full_ofm_depth)): core_weights = core_deinterleave(brick_weights, core, arch.ncores) - raw_stream = generate_brick(arch, core_weights, ofm_block_depth, tens.block_traversal, ifm_bitdepth, dilation) + + block_depth = (ofm_block_depth + arch.ncores - 1 - core) // arch.ncores + if block_depth != 0: + raw_stream = generate_brick(arch, core_weights, block_depth, tens.block_traversal, ifm_bitdepth, dilation) + else: + raw_stream = [] + raw_size += len( raw_stream ) encoded_substream = encode( raw_stream ) encoded_stream.extend( encoded_substream ) |