Implement Wormhole UBB support #592
Conversation
| @@ -13,8 +13,12 @@ class WormholeTTDevice : public TTDevice { | |||
| public: | |||
| WormholeTTDevice(std::unique_ptr<PCIDevice> pci_device); | |||
|
|
|||
| void wait_arc_core_start(const tt_xy_pair arc_core, const uint32_t timeout_ms = 1000) override; | |||
There was a problem hiding this comment.
Should this be in tt_device? Is this implemented for Blackhole?
Why pass position of the arc code?
| } | ||
| } | ||
| } else { | ||
| ubb_eth_connections(chips, desc); |
There was a problem hiding this comment.
I don't like that we do this here, but for blackhole we have whole implementation.
Can you create a separate function which encapsulates implementation for blackhole, and then Cluster::create_cluster_descriptor would remain as ~10 loc
Also this is a good signal that we should probably have a class TopologyDiscovery. Out of scope for this PR though. You can create an issue if you agree.
| const uint64_t niu_cfg_addr = 0x1000A0000 + 0x100; | ||
| read_from_device(&niu_cfg, dram_core, niu_cfg_addr, sizeof(uint32_t)); | ||
|
|
||
| bool noc_translation_enabled = (niu_cfg & (1 << 14)) != 0; |
There was a problem hiding this comment.
please put NOC_TRANSLATION_LOCATION from this line, and above niu_cfg_addr in some constants. wormhole::implementation probably
| } | ||
|
|
||
| void WormholeTTDevice::wait_arc_core_start(const tt_xy_pair arc_core, const uint32_t timeout_ms) {} |
There was a problem hiding this comment.
So you've added this without implementation? If its not used now, remove it. When we need it we will implement it.
pjanevskiTT
force-pushed
the
pjanevski/ubb
branch
from
37f3224 to
832ef83
Compare
Issue
#560
Description
Implement support for UBB galaxy board. ETH firmware is changing for UBB so topology discovery from scratch is needed. Also support for general recognition of UBB board type is needed.
This is a good step in implementing Wormhole topology discovery completely from scratch. Old WH boards still use legacy create-ethernet-map path.
List of the changes
Testing
CI + local testing. Hard to write UBB tests at the moment, we don't have this machines on CI
API Changes
Cluster descriptor API