[libre-riscv-dev] "simple" core
Luke Kenneth Casson Leighton
lkcl at lkcl.net
Mon Jun 8 00:31:21 BST 2020
the simple core is taking shape as a combination of all the pipelines,
connected to the (5) types of register files. all unit tests that
have been developed to test *individual* pipelines have been borrowed
through inheritance, to run against the register files, this time.
you can try it out as:
core.py really is *very* simple. its execution model is as follows:
* receive instruction
* decode instruction
* identify through bitmasking which pipeline can handle it
* enable *only* the MultiCompUnit managing that pipeline
* wait for it to indicate that it is not busy (this includes regfile writing)
* move on to the next instruction.
each MultiCompUnit, of which there are currently five connected up
(ALU, Logical, ShiftRot, CR, Branch), has a number of "operand in" and
"result out" ports. the register files have been *deliberately*
spec'd to match one-for-one the maximum number of ports of each
regfile type needed.
therefore, for example: whilst no other MCU needs INT 2W (2 write
ports), LDSTCompUnit *does* need 2 write ports and consequently the
INT regfile has been allocated 2W. another example: CR requires
*three* read ports *and* the "full" (32-bit) CR, and consequently the
CR regfile is 4R (1 full, 3 4-bit).
this dramatically simplifies the code needed to connect up the MCUs to
the Regfiles, because the port allocations (resource contention aside)
the majority of the code in core.py (at present) involves reorganising
"regspecs" into a dictionary-of-dictionaries-of-lists. the structure
* first dictionary key is the register file TYPE (INT, CR, SPRs)
* second dictionary key is the register port NAME (cr0, ra, rb, XER_so)
* list contains the *Function Unit* and operand/result read/write port
that list (on a per-file, per-regname basis) therefore contains *all*
Function Units that wish to contend for that register file port, and,
consequently, it is a simple matter of:
* A. creating a PriorityPicker to select one and *ONLY* one Function
Unit that is permitted to access that port at any one time
* B. creating a Broadcast Bus (fan-in in the case of write, fan-out in
the case of read) connecting regfile port to Function Unit port.
the next phases will involve:
* adding in LDSTCompUnit
* adding in minerva wishbone L1 I-Cache code (including bypass mode)
* including the pre-written scoreboard "Instruction Queue" code
* linking up NIA to the IQ, to fetch instructions and pass them to the decoder.
at that point we will have an actual core that is capable of executing
instructions on its own. after that, we have a basis for carrying out
further code-morphs, to add the precise-capable augmented scoreboard
and so on.
crowd-funded eco-conscious hardware: https://www.crowdsupply.com/eoma68
More information about the libre-riscv-dev