aboutsummaryrefslogtreecommitdiff
path: root/ethosu/vela/register_command_stream_generator.py
diff options
context:
space:
mode:
authorTim Hall <tim.hall@arm.com>2021-06-17 17:03:49 +0100
committerTim Hall <tim.hall@arm.com>2021-06-17 17:03:49 +0100
commit3016157e5099a50075d1a8b54d1b2cac2ee3899e (patch)
treee51fc057e07362720d6082bbff7ff20957b49bd7 /ethosu/vela/register_command_stream_generator.py
parent789e6f3acd1a377dfba80aa18d513579fd33fc93 (diff)
downloadethos-u-vela-3016157e5099a50075d1a8b54d1b2cac2ee3899e.tar.gz
Block config optimisation for 256/512 configurations
- 256 and 512 configuration variants execute 1D convolutions in an optimised manner compared to their 2x2 microblock dimensions. This commit takes this into account to improve Conv1D throughput on these configurations. Signed-off-by: Tim Hall <tim.hall@arm.com> Change-Id: I6ecdf6e4a219e356327b22f8393f50ee8817af23
Diffstat (limited to 'ethosu/vela/register_command_stream_generator.py')
-rw-r--r--ethosu/vela/register_command_stream_generator.py1
1 files changed, 1 insertions, 0 deletions
diff --git a/ethosu/vela/register_command_stream_generator.py b/ethosu/vela/register_command_stream_generator.py
index 20431273..b4a633e9 100644
--- a/ethosu/vela/register_command_stream_generator.py
+++ b/ethosu/vela/register_command_stream_generator.py
@@ -583,6 +583,7 @@ def get_arch_block_config(
block_config,
arch,
block_type,
+ npu_op.ofm.shape,
ifm_shape,
ifm2_shape,
uses_scalar,