diff options
| author | Johan Alfvén <johan.alfven@arm.com> | 2022-12-07 12:40:55 +0100 |
|---|---|---|
| committer | Fredrik Svedberg <fredrik.svedberg@arm.com> | 2022-12-15 12:43:07 +0000 |
| commit | 2f87617ef1ed682a5a0ac2138fc9fd551c93fd74 (patch) | |
| tree | df040056097a07e1507e0e71760fc45e050a3b2c | |
| parent | faa4b7861317385ecd2357b7af9b90c6a5fbdd08 (diff) | |
| download | ethos-u-vela-2f87617ef1ed682a5a0ac2138fc9fd551c93fd74.tar.gz | |
MLBEDSW-7173: MLCE: NPU performance for reversed operands
- When introducing the support for reversed operands the npu performance
was not updated. The result is larger numbers (degrade) from the performance
estimater compared to the previous release. In reality there is no degrade
and the real performance is the same.
- Updated npu performance to reflect the behavior implemented by the
reversed operands attribute.
Change-Id: I1b37a07f25def8f7a8adbdaadcf931bfe49165cb
Signed-off-by: Johan Alfven <johan.alfven@arm.com>
| -rw-r--r-- | ethosu/vela/npu_performance.py | 5 |
1 files changed, 3 insertions, 2 deletions
diff --git a/ethosu/vela/npu_performance.py b/ethosu/vela/npu_performance.py index cfe13d03..967a7ac0 100644 --- a/ethosu/vela/npu_performance.py +++ b/ethosu/vela/npu_performance.py @@ -705,14 +705,15 @@ def estimate_full_op_performance( ) # IFM read - ifm = op.parent_op.ifm + ifm = op.parent_op.ifm2 if op.reversed_operands else op.parent_op.ifm bw = access.ifm_read[0] * ifm.element_size() bws[ifm.mem_area][ifm.purpose][BandwidthDirection.Read] += bw scaled_bws[ifm.mem_area][ifm.purpose][BandwidthDirection.Read] += _estimate_memory_transfer_efficiency( arch, True, query.ifm_memory_area, ifm.format, query.ifm_bits, query.config.ifm_block, query.ifm_shape, bw ) + if query.ifm2_shape: - ifm2 = op.parent_op.ifm2 + ifm2 = op.parent_op.ifm if op.reversed_operands else op.parent_op.ifm2 bw = access.ifm_read[1] * ifm2.element_size() bws[ifm2.mem_area][ifm2.purpose][BandwidthDirection.Read] += bw scaled_bws[ifm2.mem_area][ifm2.purpose][BandwidthDirection.Read] += _estimate_memory_transfer_efficiency( |