diff options
| author | Patrik Gustavsson <patrik.gustavsson@arm.com> | 2020-05-05 17:49:35 +0200 |
|---|---|---|
| committer | Tim Hall <tim.hall@arm.com> | 2020-06-18 17:53:52 +0100 |
| commit | 2213e90570af328418d4f4a0d54269ed21dc40bc (patch) | |
| tree | a40f53e04e1c9200eec5101ee7fc884a3e5feed0 | |
| parent | cf72890e5dd89dada3189816a61d174b984086bd (diff) | |
| download | ethos-u-vela-2213e90570af328418d4f4a0d54269ed21dc40bc.tar.gz | |
MLBEDSW-2241: Fix for NHCWB16 with int16
Changes in strides and rounding for int16 and NHCWB16
Change-Id: I195890215b55ee7a4eab2e6ce4da95fb41587acb
Signed-off-by: Patrik Gustavsson <patrik.gustavsson@arm.com>
| -rw-r--r-- | ethosu/vela/tensor.py | 11 |
1 files changed, 3 insertions, 8 deletions
diff --git a/ethosu/vela/tensor.py b/ethosu/vela/tensor.py index 46040a46..5d0206cc 100644 --- a/ethosu/vela/tensor.py +++ b/ethosu/vela/tensor.py @@ -331,10 +331,6 @@ class Tensor: self.storage_rounding_quantum = arch.storage_rounding_quantums[self.format] self.storage_rounding_quantum = self.storage_rounding_quantum[-shape_len:] - if self.format == TensorFormat.NHCWB16: - self.storage_rounding_quantum = self.storage_rounding_quantum[:-1] + ( - int(self.storage_rounding_quantum[-1] / self.dtype.size_in_bytes()), - ) self.brick_size = arch.brick_sizes[self.format] self.brick_size = self.brick_size[-shape_len:] if self.shape is None: @@ -491,7 +487,7 @@ class Tensor: stride_order = [4, 1, 3, 2, 0] elif self.format == TensorFormat.NHCWB16: - channel_divisor = int(16 / self.element_size()) + channel_divisor = 16 augmented_shape = augmented_shape[0:4] + [1] augmented_coord = ( [augmented_coord[0], augmented_coord[3] // channel_divisor] @@ -515,11 +511,10 @@ class Tensor: stride *= augmented_shape[i] else: assert len(strides) == 5 - channel_divisor = int(16 / self.element_size()) strides[4] = stride - strides[3] = channel_divisor # STRIDE_X + strides[3] = 16 * stride # STRIDE_X strides[1] = strides[3] * augmented_shape[2] # STRIDE_C - strides[2] = augmented_shape[2] * augmented_shape[3] # STRIDE_Y + strides[2] = augmented_shape[2] * augmented_shape[3] * stride # STRIDE_Y strides[0] = strides[2] * augmented_shape[1] # STRIDE_N return strides, augmented_coord |