Refactor build so we can build arch-independent

[kkwong10]

Current mechanism is reliant upon ARCH_NAME specified in the build. Which is a compile-time flag that we have to use for all the different locations which we need to include for defines etc.

In general we want to move away from this in the long term or figure out a plan for this. Tagging @tt-rkim @DrJessop as well,

There are two considerations to this from a higher leve, the way to do the build should stem from this:

1. How do we do internal maintain code-base when we have two arches
2. How do we want to release to customers.

Since we have no plans to support heterogeneous systems, I think we can do internal testing with a compile-time flag(as it currently stands) or we can change to something more define based. But it does take more effort if things are split as they are.

For customers, we should think about this, @tt-rkim for input.

[tt-rkim]

• we need to think about compile-time arch name during library build, arch name during compile time kernel build, arch name during runtime
• might be better to query device if we have set list of archs that we know work and will never go

[kkwong10]

@tt-rkim @davorchap We want make build to not need ARCH_NAME=<>

• we use ARCH_NAME=wormhole_b0 in everything right now.
• host API + harvesting changes are all runtime based off the ARCH_NAME
• tt_device/silicon driver includes arch specific defines — This is in make build unfortunately

I think if we solve 3. We can definitely make build without ARCH_NAME.

Kernels and fw are JIT compiled so we can ignore as we already solve this with a ARCH passed into compile_program

[tt-rkim]

@kkwong10 and I have had on off discussions concerning this.

From what I understood, the key guiding principle will be to select, at runtime, a namespaced version of the current variables which are separated by arch into different header files.

It's been a while since we discussed this issue, so @kkwong10 may have been some considerable progress already. Curious to see what he decided is a good way forward

The following modules in tt_metal will need to have ARCH_NAME torn out (just to enumerate):

• tt_metal/device includes build-time ARCH_NAME-dependent files from src
• tt_metal/device includes files within itself that are ARCH_NAME-dependent
• common for some address stuff, similar to device
• tt_cluster.cpp in llrt as that's where runtime assert occurs to ensure build arch matches runtime arch
• tt_gdb and memset tools need uplifting to other archs @DrJessop

[TT-billteng]

We should consider raising the priority of this as blackhole is coming soon

[tt-rkim]

cc: @jliangTT - a part of the larger BH story

[tt-rkim]

@TT-billteng @davorchap @rtawfik01 @pgkeller This won't be immediately contribute to features for BH or javelin, but I think this will repay us in droves once BH actually comes in.

I urge we move this up in priority.

[tt-rkim]

@TT-billteng @teijo

This came up in a recent weekly meeting.

There's more interest in moving this up in priority. One underlying issue is that ti's more so a UMD refactor issue

[blozano-tt]

Enumeration of Required Changes

• [x] hw/toolchain depends on it
• [ ] Remove Runtime Checks of compiled time define in tt_cluster.cpp Pending in PR: https://github.com/tenstorrent/tt-metal/pull/13354 https://github.com/tenstorrent/tt-metal/blob/ce99dc80acdbd8d38fb251930fc00791ae42541a/tt_metal/llrt/tt_cluster.cpp#L74-L91
• [ ] Remove Architecture Specific Include Paths in CMakeLists.txt https://github.com/tenstorrent/tt-metal/blob/ce99dc80acdbd8d38fb251930fc00791ae42541a/CMakeLists.txt#L221-L234
• [ ] HAL Enhancements to give host access to dev_mem_map independent of compile time define https://github.com/tenstorrent/tt-metal/blob/ce99dc80acdbd8d38fb251930fc00791ae42541a/tt_metal/llrt/grayskull/gs_hal.cpp#L9-L16
• [x] UMD Remove Architecture Specific Include Paths in CMakeLists.txt Pending in PR: https://github.com/tenstorrent/tt-umd/pull/87 https://github.com/tenstorrent/tt-umd/blob/f7b1ce0f6ed3101bdd4f1367145378a59996f07b/CMakeLists.txt#L81-L101
• [x] UMD remove ISSUE FIX ifdef https://github.com/tenstorrent/tt-umd/issues/51#issuecomment-2372389730

  File List

• [ ] CMakeLists.txt
• [ ] tt_metal/llrt/grayskull/gs_hal.cpp
• [ ] tt_metal/llrt/wormhole/wh_hal_idle_eth.cpp
• [ ] tt_metal/llrt/wormhole/wh_hal_tensix.cpp
• [ ] tt_metal/llrt/wormhole/wh_hal_active_eth.cpp
• [ ] tt_metal/llrt/wormhole/wh_hal.cpp
• [ ] tt_metal/llrt/blackhole/bh_hal.cpp
• [ ] tt_metal/llrt/blackhole/bh_hal_active_eth.cpp
• [ ] tt_metal/llrt/blackhole/bh_hal_idle_eth.cpp
• [ ] tt_metal/llrt/blackhole/bh_hal_tensix.cpp
• [ ] tt_metal/llrt/tt_cluster.cpp
• [ ] tt_metal/third_party/umd/CMakeLists.txt

[afuller-TT]

Interest in this from others in the community: https://discord.com/channels/863154240319258674/1177107133167321158/1293915781305864284