diff options
| author | Patrik Gustavsson <patrik.gustavsson@arm.com> | 2021-02-16 12:57:03 +0100 |
|---|---|---|
| committer | Patrik Gustavsson <patrik.gustavsson@arm.com> | 2021-02-16 15:30:36 +0100 |
| commit | 56b6c711d8faaa6bcbc810e895efa650ddd97e73 (patch) | |
| tree | d6463e46e8b4147b6d03861f88999d954d5254ac | |
| parent | ab11bd1216cded591382ed01b6f5ce0f55a6de46 (diff) | |
| download | ethos-u-vela-56b6c711d8faaa6bcbc810e895efa650ddd97e73.tar.gz | |
MLBEDSW-4076 Fix read_offset at create_primary_op
Fix read_offset at create_primary_op
The read_offset need to be copied when adding
avg pool as primary operator.
Signed-off-by: Patrik Gustavsson <patrik.gustavsson@arm.com>
Change-Id: I6f168517a0e22543455b623b6b4f59237e8d530a
| -rw-r--r-- | ethosu/vela/pass_packing.py | 40 |
1 files changed, 15 insertions, 25 deletions
diff --git a/ethosu/vela/pass_packing.py b/ethosu/vela/pass_packing.py index 281c093..2a1903d 100644 --- a/ethosu/vela/pass_packing.py +++ b/ethosu/vela/pass_packing.py @@ -29,17 +29,16 @@ from .tensor import TensorPurpose class PassFlags(enum.Flag): Empty = 0 - Pre = 1 - Main = 2 - Post = 4 - Mac = 8 - Dma = 32 - ElementWise = 256 - Npu = 512 - Cpu = 1024 - StartupInit = 2048 - MemoryOnly = 4096 - PostFusingLimited = 8192 + Main = 1 + Post = 2 + Mac = 4 + Dma = 8 + ElementWise = 16 + Npu = 32 + Cpu = 64 + StartupInit = 128 + MemoryOnly = 256 + PostFusingLimited = 512 mac_main_ops = set( @@ -98,7 +97,7 @@ test_sequence = [ # ops_set npu_post_ops, # incompatible_pack_flags - PassFlags.Cpu | PassFlags.MemoryOnly | PassFlags.Pre | PassFlags.Main, + PassFlags.Cpu | PassFlags.MemoryOnly | PassFlags.Main, # flags_to_set PassFlags.Npu | PassFlags.Post, # flags_to_clear @@ -108,7 +107,7 @@ test_sequence = [ # ops_set npu_post_fuse_limited_ops, # incompatible_pack_flags - PassFlags.Cpu | PassFlags.MemoryOnly | PassFlags.Pre | PassFlags.Main, + PassFlags.Cpu | PassFlags.MemoryOnly | PassFlags.Main, # flags_to_set PassFlags.Npu | PassFlags.PostFusingLimited, # flags_to_clear @@ -118,12 +117,7 @@ test_sequence = [ # ops_set mac_main_ops, # incompatible_pack_flags - PassFlags.Cpu - | PassFlags.MemoryOnly - | PassFlags.ElementWise - | PassFlags.Pre - | PassFlags.Main - | PassFlags.PostFusingLimited, + PassFlags.Cpu | PassFlags.MemoryOnly | PassFlags.ElementWise | PassFlags.Main | PassFlags.PostFusingLimited, # flags_to_set PassFlags.Npu | PassFlags.Mac | PassFlags.Main, # flags_to_clear @@ -133,12 +127,7 @@ test_sequence = [ # ops_set elem_wise_main_ops, # incompatible_pack_flags - PassFlags.Cpu - | PassFlags.MemoryOnly - | PassFlags.Mac - | PassFlags.Pre - | PassFlags.Main - | PassFlags.PostFusingLimited, + PassFlags.Cpu | PassFlags.MemoryOnly | PassFlags.Mac | PassFlags.Main | PassFlags.PostFusingLimited, # flags_to_set PassFlags.Npu | PassFlags.ElementWise | PassFlags.Main, # flags_to_clear @@ -436,6 +425,7 @@ def pack_into_passes(nng, arch, verbose_packing=False): avgpool_op.set_output_tensor(avgpool_out) avgpool_op.ifm_shapes = op.ifm_shapes.copy() avgpool_op.ofm_shapes = op.ofm_shapes.copy() + avgpool_op.read_offsets = op.read_offsets.copy() op.inputs[0] = avgpool_out op_list.insert(0, avgpool_op) |