Add KUL Cluster SW regression test

[xiaoling-yi]

In this PR, we add the software test for the KUL cluster. More specifically, we add:

• sw/tests/kulClusterOffload.c which is the main test to offload the test function kul_cluster_sw_test() to KUL cluster

• sw/kul-cluster-test/src/snax-kul-cluster-TO-test.c which contains the entry of the test function kul_cluster_sw_test()

• sw/kul-cluster-test/data/data.h which contains the test-related data

• sw/kul-cluster-test/include/* contains header files that need to be added in the include path for compiling the kul_cluster_sw_test()

• sw/kul-cluster-test/lib/* contains two libraries (snax-gemmx-lib and snax-data-reshuffler-lib) that need to be compiled and likned before compiling kul_cluster_sw_test(). To compile these two libraries, include/snrt_TO.h is needed.

[Scheremo]

Hi Xiaoling,
Thanks for your contribution! I have a few questions before going further in-depth with reviewing.

1. I see you use the DMA in some circumstances. Would it be possible to instead preload all your data in L1? This reduces turn around time by a lot.
2. I didn't see you setting the stack pointer; does this test work on your end?

Best,
Moritz

[xiaoling-yi]

Hi Moritz,

I see you use the DMA in some circumstances. Would it be possible to instead preload all your data in L1? This reduces turn around time by a lot.

What do you mean by preload data?
Is it putting all the DMA functions together at the beginning of the test function kul_cluster_sw_test?

I didn't see you setting the stack pointer; does this test work on your end?

I set the stack pointer just now to the end of the KUL cluster TCDM (size = 128KB) address. Is it the right way to set it before the offloadToCluster call?
Yes, I tested it on the snax cluster level but not the Chimera level yet.

Thanks!

Best regards,
Xiaoling

[xiaoling-yi]

Hi Moritz!

The two snitch core is waked up through the host. This is the only way that I managed it to work.
Also the return value at the host level is a combination of the two workload test. As the two workloads both turn 0 from the snitch core and turned out 1 from the view of host (the bootrom did something: uint32_t retVal = ret | 0x000000001;). So the final return value is 10001.

Please free feel to modify them as you need.

Best regards,
Xiaoling