How srekrishnas05/soc works
This VHDL-2008 SoC runs bare-metal software on a five-stage ARMv7 CPU and includes a heterogeneous INT8 NPU cluster. The principal workflow is CPU job submission over MMIO, static tile planning, schedule storage, dispatch across clock domains, and systolic-array computation. The graph preserves the documented CPU features, memory/fabric roles, and NPU control and CDC stages. Source excerpts were unavailable, so relationships are limited to those stated in the README; implementation details beyond it are not inferred.
Main entry points: datapath.vhd CPU Datapath · npu_wrapper.vhd NPU Wrapper · controller_fsm.vhd NPU Sequencer · systolic_array.vhd Systolic Array
Components 24
ARMv7 CPU
- CPU Datapath
rtl/cpu/datapath.vhd - CPU Control
rtl/cpu/controlunit.vhd - Hazard Unit
rtl/cpu/hazardunit.vhd - ALU
rtl/cpu/alu.vhd - Wallace Multiplier
rtl/cpu/wallacemul.vhd - Branch Predictor
rtl/cpu/branchp.vhd
Fabric and Memory
- MMIO Bus
rtl/fabric/mmiobus.vhd - Data Cache
rtl/memory/dcache.vhd - Instruction Memory
rtl/memory/imem.vhd - Ping-Pong Buffers
rtl/memory/ping_pong_buffer.vhd
NPU Scheduling
- NPU Wrapper
rtl/npu/npu_wrapper.vhd - Tile Planner
rtl/npu/tile_planner.vhd - Schedule RAM
rtl/memory/sched_ram.vhd - Dispatch FSM
rtl/npu/dispatch_fsm.vhd - Command FIFO
rtl/fabric/async_fifo.vhd - NPU Sequencer
rtl/npu/controller_fsm.vhd - Done Synchronizer
rtl/fabric/cdc_pulse_sync.vhd
NPU Compute
- NPU Cluster
rtl/npu/npu_cluster_top.vhd - Accelerator
rtl/npu/accelerator_top.vhd - Skew Injector
rtl/npu/skew_injector.vhd - Systolic Array
rtl/npu/systolic_array.vhd - Processing Element
rtl/npu/pe.vhd
Other components
- Bare-metal Software
- SoC Top
top.vhd
Connections 24
- Bare-metal Software → CPU Datapath: submits job
README.md - SoC Top → CPU Datapath: integrates CPU
README.md - CPU Datapath → MMIO Bus: routes memory access
README.md - CPU Datapath → Data Cache: uses data cache
README.md - CPU Datapath → Instruction Memory: fetches instructions
- CPU Datapath → CPU Control: uses control
- CPU Datapath → Hazard Unit: checks hazards
README.md - CPU Datapath → ALU: executes operations
- CPU Datapath → Wallace Multiplier: requests multiply
README.md - CPU Datapath → Branch Predictor: uses prediction
README.md - MMIO Bus → NPU Wrapper: routes peripheral access
README.md - NPU Wrapper → Ping-Pong Buffers: sequences tile loads
README.md - NPU Wrapper → NPU Cluster: bridges NPU access
- NPU Cluster → Tile Planner: plans tiles
README.md - Tile Planner → Schedule RAM: writes schedule
README.md - Dispatch FSM → Schedule RAM: reads schedule
README.md - Dispatch FSM → Command FIFO: issues descriptors
README.md - Command FIFO → NPU Sequencer: delivers descriptors
README.md - NPU Sequencer → Accelerator: starts computation
README.md - Accelerator → Systolic Array: runs array
README.md - Systolic Array → Skew Injector: receives skewed operands
README.md - Systolic Array → Processing Element: uses processing elements
- NPU Sequencer → Done Synchronizer: synchronizes completion
README.md - Done Synchronizer → Dispatch FSM: signals completion