Rocket core overview

The Rocket core is an in-order scalar processor that provides a 5-stage pipeline. It implements the RV64G variant of the RISC-V ISA. The Rocket core has one integer ALU and an optional FPU. An accelerator or co-processor interface, called RoCC, is also provided.

Further details of the RISC-V Rocket core pipeline can be found here. See p.13 of this document for a detailed diagram of Rocket’s microarchitecture. The Rocket core is sometimes described as a 6-stage pipeline with the addition of a ‘pcgen’ stage. While it is useful to layout the figure in this way, the stage is perhaps best considered as part of the other stages and is not a distinct pipeline stage in the traditional sense.

The pcgen and fetch stages are shown below. Instruction fetch is assisted by a gshare predictor, Return Address Stack (RAS) and Branch Target Buffer (BTB).

The remaining four pipeline stages are shown below:

The L1 data cache is shown below.

Note: The data cache’s metadata (tags etc.) and data arrays are shown in both stage 1 (for read operations) and stage 4 (for write operations).

Key:

• amoalu: Atomic memory operation ALU
• mshrs: Miss status handling registers
• prober: Handles incoming probe requests (i.e. tag lookups or requests to revoke permissions)
• code: placeholder for ECC support