diff options
| author | Patrik Gustavsson <patrik.gustavsson@arm.com> | 2020-12-01 16:02:29 +0100 |
|---|---|---|
| committer | Patrik Gustavsson <patrik.gustavsson@arm.com> | 2020-12-18 16:33:32 +0100 |
| commit | 2349d429d926e258e9a61d34c7fd97660ab9fb98 (patch) | |
| tree | b5151d0f12428e47d64b1fb2ce4f2f8c19304a0d /ethosu/vela/npu_performance.py | |
| parent | 528a56df829b65f7a2c61953650b123c461095f7 (diff) | |
| download | ethos-u-vela-2349d429d926e258e9a61d34c7fd97660ab9fb98.tar.gz | |
MLBEDSW-3654 Add/use op ifm/ofm shapes
Add ifm/ofm shapes to op
Changed to rely on these shapes
Signed-off-by: Patrik Gustavsson <patrik.gustavsson@arm.com>
Change-Id: I571535a1dcadc2bdb04a3c727a8e1c49703b174d
Diffstat (limited to 'ethosu/vela/npu_performance.py')
| -rw-r--r-- | ethosu/vela/npu_performance.py | 10 |
1 files changed, 5 insertions, 5 deletions
diff --git a/ethosu/vela/npu_performance.py b/ethosu/vela/npu_performance.py index 9d83f6fb..c2ec4424 100644 --- a/ethosu/vela/npu_performance.py +++ b/ethosu/vela/npu_performance.py @@ -48,7 +48,7 @@ def rolling_buffer_dims_from_passes(arch, ps1, block_config_ps1, ps2, block_conf if ps2.npu_block_type in (NpuBlockType.ConvolutionMxN, NpuBlockType.VectorProduct): op = ps2.primary_op - ifm_block_depth = arch.calc_ifm_block_depth(op.ifm.shape[-1], op.ifm.dtype.size_in_bits()) + ifm_block_depth = arch.calc_ifm_block_depth(op.ifm_shapes[0][-1], op.ifm.dtype.size_in_bits()) else: ifm_block_depth = block_config_ps2[-1] @@ -224,8 +224,8 @@ def estimate_conv_pooling_cycles( scale_tensor=None, ): ofm_ublock = Block(arch.config.ofm_ublock.width, arch.config.ofm_ublock.height, arch.config.ofm_ublock.depth) - ifm_tens_shape = numeric_util.full_shape(4, ifm_tensor.shape, 1) - ofm_tens_shape = numeric_util.full_shape(4, ofm_tensor.shape, 1) + ifm_tens_shape = primary_op.ifm_shapes[0] + ofm_tens_shape = primary_op.ofm_shapes[0] if ( arch.config.ofm_ublock.height == 2 @@ -420,8 +420,8 @@ def performance_metrics_for_pass(arch, ps, block_config=None, rewrite_list=None, npu_block_type = primary_op.type.npu_block_type ifm_tensor, _, weight_tensor, ofm_tensor = ps.get_primary_op_ifm_ifm2_weights_ofm() - ifm_tensor_shape = numeric_util.full_shape(4, ifm_tensor.shape, 1) - ofm_tensor_shape = numeric_util.full_shape(4, ofm_tensor.shape, 1) + ifm_tensor_shape = list(ps.primary_op.ifm_shapes[0]) + ofm_tensor_shape = list(ps.primary_op.ofm_shapes[0]) if npu_block_type == NpuBlockType.ReduceSum: block_traversal = TensorBlockTraversal.DepthFirst |