hwtLib.amba.axi_comp.lsu package

This package is dedicated to AXI Load/Store Units (LSU) a others like.

Other LSU implementations:

• https://github.com/riscv-boom/riscv-boom - has RISC-V LSU in chisel3, read kill, MSHRs

• https://github.com/rsd-devel/rsd - has N issue RISC-V LSU

• https://github.com/bluespec/Toooba - has RISC-V LSU in BlueSpecVerilog

• https://github.com/openhwgroup/cv32e40p - has RISC-V LSU for in order pipeline

• https://github.com/VerticalResearchGroup/miaow/

Submodules

hwtLib.amba.axi_comp.lsu.fifo_oooread module

class hwtLib.amba.axi_comp.lsu.fifo_oooread.FifoOutOfOrderRead(hdlName: str | None = None)

Bases: HwModule

Container of FIFO pointers and flags where the items can be discarded in out of order manner.

Item state control scheme:

• write_confirm: the item is now allocated in the FIFO and ready to be read

• read_execute: the item is locked for updates and is currently being read

• read_confirm: the item is entirely read and it is ready to be deallocated

HDL params:

• ITEMS - default value 4 of type int

• KEY_WIDTH - default value 0 of type int

• INIT_DATA - default value () of type tuple

HDL IO:

• clk - of type hwt.hwIOs.std.HwIOClk with dtype=<HBits, 1bit> - SLAVE

• rst_n - of type hwt.hwIOs.std.HwIORst_n with dtype=<HBits, 1bit, n> - SLAVE

• write_confirm - of type hwt.hwIOs.std.HwIORdVldSync - SLAVE

• read_execute - of type hwtLib.commonHwIO.index_key_hs.HwIOIndexKeyRdVld - MASTER

• read_confirm - of type hwt.hwIOs.std.HwIODataRdVld - SLAVE

schematic

__annotations__ = {}

class hwtLib.amba.axi_comp.lsu.fifo_oooread.FifoOutOfOrderReadFiltered(hdlName: str | None = None)

Bases: FifoOutOfOrderRead

FifoOutOfOrderRead with an additional cam to filter transactions by same key :attention: this component does not contains the item storage, it is just container of such a FIFO logic

Item state control scheme:

• write_execute: preallocate the item for writing (and add key to CAM for filtering)

• write_confirm: the item is now allocated in the fifo and ready to be read

• read_execute: the item is locked for updates and is currently being read

• read_confirm: the item is entirely readed and it is ready to be deallocated

HDL params:

• ITEMS - default value 4 of type int

• KEY_WIDTH - default value 8 of type int

• INIT_DATA - default value () of type tuple

• HAS_READ_LOOKUP - default value False of type bool

HDL IO:

• clk - of type hwt.hwIOs.std.HwIOClk with dtype=<HBits, 1bit> - SLAVE

• rst_n - of type hwt.hwIOs.std.HwIORst_n with dtype=<HBits, 1bit, n> - SLAVE

• write_confirm - of type hwt.hwIOs.std.HwIORdVldSync - SLAVE

• read_execute - of type hwtLib.commonHwIO.index_key_hs.HwIOIndexKeyRdVld - MASTER

• read_confirm - of type hwt.hwIOs.std.HwIODataRdVld - SLAVE

• write_pre_lookup - of type hwt.hwIOs.std.HwIODataRdVld - SLAVE

• write_pre_lookup_res - of type hwt.hwIOs.std.HwIODataRdVld - MASTER

• item_valid - of type hwt.hwIOs.std.HwIOSignal with dtype=<HBits, 4bits> - MASTER

• item_write_lock - of type hwt.hwIOs.std.HwIOSignal with dtype=<HBits, 4bits> - MASTER

• write_execute - of type hwtLib.commonHwIO.index_key_hs.HwIOIndexKeyInRdVld - MASTER

HDL components:

• tag_cam - of type hwtLib.amba.axi_comp.cache.utils.CamWithReadPort

schematic

__annotations__ = {}

hwtLib.amba.axi_comp.lsu.hIOs module

class hwtLib.amba.axi_comp.lsu.hIOs.HwIOAxiWriteAggregatorRead(masterDir=DIRECTION.OUT, hdlName: str | dict[str, str] | None = None, loadConfig=True)

Bases: HwIO

An interface which is used to speculatively read data from AxiWriteAggregator

HDL params:

• ADDR_WIDTH - default value 32 of type int

• DATA_WIDTH - default value 64 of type int

HDL IO:

• a - of type hwtLib.amba.axi3Lite.Axi3Lite_addr - UNKNOWN

• r_data_available - of type hwt.hwIOs.std.HwIODataRdVld - UNKNOWN

• r - of type hwtLib.amba.axi3Lite.Axi3Lite_r - UNKNOWN

__annotations__ = {}

_initSimAgent(sim: HdlSimulator)

class hwtLib.amba.axi_comp.lsu.hIOs.HwIOAxiWriteAggregatorWriteTmp(masterDir=DIRECTION.OUT, hdlName: str | dict[str, str] | None = None, loadConfig=True)

Bases: Axi4Stream

HwIO for tmp input register on store buffer write input

HDL params:

• IS_BIGENDIAN - default value False of type bool

• USE_STRB - default value True of type bool

• USE_KEEP - default value False of type bool

• ID_WIDTH - default value 0 of type int

• DEST_WIDTH - default value 0 of type int

• DATA_WIDTH - default value 64 of type int

• USER_WIDTH - default value 0 of type int

• ADDR_WIDTH - default value 32 of type int

• ITEMS - default value 64 of type int

HDL IO:

• addr - of type hwt.hwIOs.std.HwIOSignal with dtype=<HBits, 32bits> - UNKNOWN

• data - of type hwt.hwIOs.std.HwIOSignal with dtype=<HBits, 64bits> - UNKNOWN

• strb - of type hwt.hwIOs.std.HwIOSignal with dtype=<HBits, 8bits> - UNKNOWN

• last - of type hwt.hwIOs.std.HwIOSignal with dtype=<HBits, 1bit> - UNKNOWN

• ready - of type hwt.hwIOs.std.HwIOSignal with dtype=<HBits, 1bit> - UNKNOWN (Master=IN)

• valid - of type hwt.hwIOs.std.HwIOSignal with dtype=<HBits, 1bit> - UNKNOWN

• colides_with_last_addr - of type hwt.hwIOs.std.HwIOSignal with dtype=<HBits, 1bit> - UNKNOWN

• cam_lookup - of type hwt.hwIOs.std.HwIOSignal with dtype=<HBits, 64bits> - UNKNOWN HwIOVectSignal with value of lookup from item cam (1 if cacheline present in store buffer)

• mask_byte_unaligned - of type hwt.hwIOs.std.HwIOSignal with dtype=<HBits, 1bit> - UNKNOWN HwIOSignal 1 if any byte of mask is 0 or all 1

__annotations__ = {}

_initSimAgent(sim: HdlSimulator)

hwtLib.amba.axi_comp.lsu.read_aggregator module

class hwtLib.amba.axi_comp.lsu.read_aggregator.AxiReadAggregator(hdlName: str | None = None)

Bases: HwModule

This is a component which reduces reads from same address.

This component has several slots for read transactions, Each slot has it’s own address record in CAM which is used to detect reads from same address, if the read is from same address which is currently being loaded. The read thread is put to sleep until data for previous read is received. After data is received it is copied as a response also for this transaction.

HDL params:

• ADDR_WIDTH - default value 32 of type int

• DATA_WIDTH - default value 64 of type int

• ID_WIDTH - default value 2 of type int

• ADDR_USER_WIDTH - default value 0 of type int

• CACHE_LINE_SIZE - default value 64 of type int

HDL IO:

• clk - of type hwt.hwIOs.std.HwIOClk with dtype=<HBits, 1bit> - SLAVE

• rst_n - of type hwt.hwIOs.std.HwIORst_n with dtype=<HBits, 1bit, n> - SLAVE

• s - of type hwtLib.amba.axi4.Axi4 - SLAVE

• m - of type hwtLib.amba.axi4.Axi4 - MASTER

HDL components:

• frame_buff - of type hwtLib.amba.axis_comp.fifoCopy.Axi4SFifoCopy

• addr_cam - of type hwtLib.mem.cam.CamMultiPort

• addr_cam_out_reg - of type hwtLib.handshaked.reg.HandshakedReg

• s_ar_tmp - of type hwtLib.amba.axis_comp.reg.Axi4SReg

schematic

__annotations__ = {}

add_addr_cam_out_reg(item_vld: RtlSignal)

read_data_section(read_ack: RtlSignal, waiting_transaction_id: RtlSignal, waiting_transaction_vld: RtlSignal, data_copy_override: HwIODataVld)

read_request_section(read_ack: RtlSignal, item_vld: RtlSignal, waiting_transaction_id: RtlSignal, waiting_transaction_vld: RtlSignal, data_copy_override: HwIODataVld)

hwtLib.amba.axi_comp.lsu.read_aggregator._example_AxiReadAggregator()

hwtLib.amba.axi_comp.lsu.store_queue_write_propagating module

class hwtLib.amba.axi_comp.lsu.store_queue_write_propagating.Axi4StoreQueueWritePropagating(hdlName: str | None = None)

Bases: AxiWriteAggregator

An extension of hwtLib.amba.axi_comp.lsu.write_aggregator.AxiWriteAggregator with an IO for a communication with an hwtLib.amba.axi_comp.lsu.load_queue.AxiLoadQueue Does the same thing and allows hwtLib.amba.axi_comp.lsu.load_queue_write_propagating.AxiLoadQueueWritePropagating to speculatively read the data and listen for write transactions. The component allows for write to bypass read, which makes it suitable for cumulative operations, but more complicated for a generic use.

Variables:

• speculative_read_addr – port used for load buffer to speculatively read the data from this component. If data is not present the speculative_read_data returns RESP_EXOKAY error. The data may also be flushed during the read this is marked with RESP_SLVERR error.

• speculative_read_data – Read data for speculative read.

__annotations__ = {}

speculative_read_handler()

Connect the speculative_read port to internal storages of the AxiWriteAggregator

We need to handle several cases:

1. the data is currently stuck in tmp register (we need to wait)

2. the data was in tmp register and now is in data_ram

3. the data was and is in data_ram

4. the data was in data_ram and now it is in main memory

5. the data was not found in this component (is in main memory)

Handling of speculative read has following stages:

1. search input register and main address CAM for data

2. optionaly load the data from ram

3. send data to speculative_read_data and set resp to error if was not found it may also happen that the data was flushed in the mean time we can not blok other channel and we need to have a buffer for 1 transaction to prevent data loose where we would need to ask main memory

Note:

speculative read never block the write channel and thus data may be invalid if the speculative read data is stalled. This should be handled in master of speculative read port (Other component which is using this component).

class hwtLib.amba.axi_comp.lsu.store_queue_write_propagating.HwIOAxiWriteAggregatorWriteTmp(masterDir=DIRECTION.OUT, hdlName: str | dict[str, str] | None = None, loadConfig=True)

Bases: HwIODataRdVld

__annotations__ = {}

hwtLib.amba.axi_comp.lsu.write_aggregator module

class hwtLib.amba.axi_comp.lsu.write_aggregator.AxiWriteAggregator(hdlName: str | None = None)

Bases: HwModule

A buffer which is used for write data from cache.

It manages:

• out of order write acknowledge

• write transaction merging

HDL params:

• ADDR_WIDTH - default value 32 of type int

• DATA_WIDTH - default value 32 of type int

• ID_WIDTH - default value 2 of type inta parameter which specifies width of axi id signal, it also specifies the number of items in this buffer (2**ID_WIDTH)

• CACHE_LINE_SIZE - default value 4 of type int

• MAX_BLOCK_DATA_WIDTH - default value 8 of type intspecifies maximum data width of RAM (used to prevent synthesis problems for tools which can not handle too wide memories with byte enable)

HDL IO:

• clk - of type hwt.hwIOs.std.HwIOClk with dtype=<HBits, 1bit> - SLAVE

• rst_n - of type hwt.hwIOs.std.HwIORst_n with dtype=<HBits, 1bit, n> - SLAVE

• s - of type hwtLib.amba.axi4.Axi4 - SLAVE

• m - of type hwtLib.amba.axi4.Axi4 - MASTER

HDL components:

• write_dispatch - of type hwtLib.amba.axi_comp.lsu.write_aggregator_write_dispatcher.AxiWriteAggregatorWriteDispatcher

• ooo_fifo - of type hwtLib.amba.axi_comp.lsu.fifo_oooread.FifoOutOfOrderReadFiltered

• data_ram - of type hwtLib.mem.ramCumulativeMask.RamCumulativeMask

• gen_b_reg_0 - of type hwtLib.amba.axis_comp.reg.Axi4SReg

• w_in_reg - of type hwtLib.amba.axis_comp.reg.Axi4SReg

schematic

__annotations__ = {}

_addr_to_index(addr: RtlSignal)

_declr_data_ram()

data_insert(items: BramPort_withReadMask_withoutClk)

• if it is possible to update data in data_ram of this buffer

• else allocate new data (insert to address CAM of ooo_fifo) and store data to w_in_reg

resolve_cam_index(w_tmp_out: HwIOAxiWriteAggregatorWriteTmp)

w_in_tmp_reg_load() → HwIOAxiWriteAggregatorWriteTmp

• check if this address is already present in address CAM or w_in_reg

hwtLib.amba.axi_comp.lsu.write_aggregator._example_AxiWriteAggregator()

hwtLib.amba.axi_comp.lsu.write_aggregator_write_dispatcher module

class hwtLib.amba.axi_comp.lsu.write_aggregator_write_dispatcher.AxiWriteAggregatorWriteDispatcher(hdlName: str | None = None)

Bases: HwModule

Use hwtLib.amba.axi_comp.lsu.fifo_oooread.FifoOutOfOrderReadFiltered read ports to query an AXI transaction info and copy paste this transaction from BRAM to AXI.

HDL params:

• ADDR_WIDTH - default value 32 of type int

• DATA_WIDTH - default value 64 of type int

• ID_WIDTH - default value 6 of type int

• CACHE_LINE_SIZE - default value 64 of type int

HDL IO:

• clk - of type hwt.hwIOs.std.HwIOClk with dtype=<HBits, 1bit> - SLAVE

• rst_n - of type hwt.hwIOs.std.HwIORst_n with dtype=<HBits, 1bit, n> - SLAVE

• data - of type hwtLib.mem.ramCumulativeMask.BramPort_withReadMask_withoutClk - MASTER

• read_execute - of type hwtLib.commonHwIO.index_key_hs.HwIOIndexKeyRdVld - SLAVE

• read_confirm - of type hwt.hwIOs.std.HwIODataRdVld - MASTER

• m - of type hwtLib.amba.axi4.Axi4 - MASTER

schematic

__annotations__ = {}

data_ram_read_to_bus_w(items: BramPort_withReadMask_withoutClk, item_last: RtlSignal, w_out: Axi4_w)

Read write data from data_ram

Parameters:

item_last – the signal with last flag for data (notes a last beat in burst) sampled in the same time as an read address

dispatch_addr(id_to_use: RtlSignal, addr: RtlSignal, a: Axi4_addr)

• if there is a valid item in buffer dispatch read request

dispatch_data(aw_ld: RtlSignal, pop_ptr: RtlSignal, pop_req: RtlSignal, pop_wait: RtlSignal, items: BramPort_withReadMask_withoutClk, aw_out_ready: RtlSignal, w_out: Axi4_w)

• if there is a valid item in buffer dispatch write request and write data

• handshaked FIFO read logic

precompute_constants()

receive_write_confirm(b: Axi4_b, read_confirm: HwIODataRdVld)