However, in tt_metal/hw/inc/debug/sanitize_noc.h, it still uses the hardcoded cmd buffers, resulting in it reading from the wrong command buffer, ex for DEBUG_SANITIZE_NOC_WRITE_TRANSACTION_WITH_ADDR_AND_SIZE_STATE when using DYNAMIC_NOC. It reads 0 for the size when running ./build/test/tt_metal/perf_microbenchmark/11_remote_cb_sync_matmul_single_core/test_remote_cb_sync_matmul --m 32 --k 2048 --n 128 --num-blocks 8 --cb-num-blocks 10 --cb-padding 256 --num-tests 1 --data-type 1 --num-receivers 2 --num-layers 15 because it is reading from the wrong command buffer.
In
dataflow_api.h
, there are now aliases for the specific command buffers used when in the original/DEDICATED_NOC mode, and DYNAMIC_NOC mode.However, in
tt_metal/hw/inc/debug/sanitize_noc.h
, it still uses the hardcoded cmd buffers, resulting in it reading from the wrong command buffer, ex forDEBUG_SANITIZE_NOC_WRITE_TRANSACTION_WITH_ADDR_AND_SIZE_STATE
when using DYNAMIC_NOC. It reads 0 for the size when running./build/test/tt_metal/perf_microbenchmark/11_remote_cb_sync_matmul_single_core/test_remote_cb_sync_matmul --m 32 --k 2048 --n 128 --num-blocks 8 --cb-num-blocks 10 --cb-padding 256 --num-tests 1 --data-type 1 --num-receivers 2 --num-layers 15
because it is reading from the wrong command buffer.