Data Alignment Options - 1.0 English

Versal Adaptive SoC Integrated Block for PCI Express LogiCORE IP Product Guide (PG343)

Document ID
PG343
Release Date
2023-11-10
Version
1.0 English

A transaction layer packet (TLP) is transferred on each of the AXI4-Stream interfaces as a descriptor followed by payload data (when the TLP has a payload). The descriptor has a fixed size of 16 bytes on the request interfaces and 12 bytes on the completion interfaces. On its transmit side (towards the link), the integrated block assembles the TLP header from the parameters supplied by the user application in the descriptor. On its receive side (towards the user interface), the integrated block extracts parameters from the headers of received TLP and constructs the descriptors for delivering to the user application. Each TLP is transferred as a packet, as defined in the AXI4-Stream Interface protocol.

64/128/256-bit interface:

When a payload is present, there are two options for aligning the first byte of the payload with respect to the datapath.
  1. Dword-aligned mode: In this mode, the descriptor bytes are followed immediately by the payload bytes in the next Dword position, whenever a payload is present.
  2. Address-Aligned Mode: In this mode, the payload can begin at any byte position on the datapath. For data transferred from the integrated block to the user application, the position of the first byte is determined as
    n = A mod w

    where A is the memory or I/O address specified in the descriptor (for message and configuration requests, the address is taken as 0), and w is the configured width of the data bus in bytes. Any gap between the end of the descriptor and the start of the first byte of the payload is filled with null bytes.

For data transferred from the integrated block to the user application, the data alignment is determined based on the starting address where the data block is destined to in user memory. For data transferred from the user application to the integrated block, the user application must explicitly communicate the position of the first byte to the integrated block using the tuser sideband signals when the address-aligned mode is in use.

In the address-aligned mode, the payload and descriptor are not allowed to overlap. That is, the transmitter begins a new beat to start the transfer of the payload after it has transmitted the descriptor. The transmitter fills any gaps between the last byte of the descriptor and the first byte of the payload with null bytes.

512-bit interface:

When a payload is present, there are two options for aligning the first byte of the payload with respect to the datapath.
  1. Dword-aligned Mode: In this mode, the descriptor bytes are followed immediately by the payload bytes in the next Dword position, whenever a payload is present. If D is the size of the descriptor in bytes, the lane number corresponding to the first byte of the payload is determined as:
    n = (S + D + (A mod 4)) mod 64
    where S is the lane number where the first byte of the descriptor appears (which can be 0, 16, 32, or 48), D is the width of the descriptor (which can be 12 or 16 bytes), and A is the address of the first byte of the data block in user memory (for message and configuration requests, the address is taken as 0).
  2. 128b Address-aligned Mode: In this mode, the start of the payload on the 512-bit bus is aligned on a 128-bit boundary. The lane number corresponding to the first byte of the payload is determined as:
    n = (S + 16 + (A mod 16)) mod 64

    where S is the lane number where the first byte of the descriptor appears (which can be 0, 16, 32, or 48) and A is the memory or I/O address corresponding to the first byte of the payload (for message and configuration requests, the address is taken as 0). Any gap between the end of the descriptor and the start of the first byte of the payload is filled with null bytes.

    The source of address A used for alignment of the data varies among the four user interfaces, as described below:
    CQ Interface
    For data transferred from the core to the user application over the CQ interface, the address bits used for alignment are the lower address specified in the descriptor, which is the starting address of the data block in user memory.
    CC Interface
    For Completion data transferred from the user application to the core over the CC interface, the alignment is based on address bits supplied by the user in the descriptor.
    RQ Interface
    For memory requests transferred from the user application to the core over the RQ interface, the alignment is based on address bits supplied by the user alongside the request using sideband signals. You can specify any value for A, independent of the setting of the address field in the descriptor.
    RC Interface
    For Completion data transferred from the core to the user application over the RC interface, the alignment is based on address bits supplied by the user along with the request using sideband signals when it was issued on the RQ interface. The core saves the alignment information from the request and uses it to align the payload of the corresponding Completion when delivering the Completion payload over the RC interface.

    The 128b address-aligned mode divides the 512-bit AXI beat into four sub-beats of 128 bits each. The payload can begin only in the sub-beat following the descriptor. The payload and the descriptor are not allowed to overlap in the same sub-beat. The transmitter fills any gaps between the last byte of the descriptor and the first byte of the payload with null bytes.

The alignment mode can be selected independently for requester (RQ, RC) and completer (CQ, CC) interfaces by setting the IP customization GUI.

Note: If performance is a critical factor in the design, dword aligned mode should be used instead of address aligned mode.

The AMD Vivado™ IP catalog applies the data alignment option globally to all four interfaces. However, advanced users can select the alignment mode independently for each of the four AXI4-Stream interfaces. This is done by setting the corresponding alignment mode parameter. See 64/128/256-bit Completer Interface and 512-bit Completer Interface for more details on address alignment and example diagrams.