Tiling Parameters and Buffer Descriptors - Tiling Parameters and Buffer Descriptors - 2026.1 English - UG1603

Buffer descriptors located in all three levels of memory manage DMA transfers:

• AI Engine-ML memory with co-located Buffer descriptors.
• Memory tile with co-located Buffer Descriptors.
• External memory with buffer descriptors located in the AI Engine-ML-PL interface.

These buffer descriptors handle 3D or 4D memory addressing, multiple iterations, lock ID and value change and buffer descriptor chaining.

struct tiling_parameters
{
  std::vector<uint32_t> buffer_dimension;
  std::vector<uint32_t> tiling_dimension;
  std::vector<int32_t> offset;
  std::vector<traversing_parameters> tile_traversal;
  int packet_port_id = -1;
  std::vector<uint32_t> boundary_dimension;
  int repetition = 1;
};

The members of this structure are as follows:

buffer_dimension

Buffer dimensions in the memory element type (for example, AI Engine-ML memory, memory tile, external memory). buffer_dimension[0] is contiguous in memory and has the fastest access. If you do not specify this member, the dimensions of the associated memory object are used. The AI Engine-ML memory can access data in the 1st, 2nd, and 3rd dimensions. Memory tile can access data in the 1st, 2nd, 3rd, and 4th dimensions. External memory can access data in the 1st, 2nd, and 3rd dimensions.

tiling_dimension

Tiling dimensions of the data transfer in buffer. The tiling dimension of AI Engine-ML memory can access data in the 1st, 2nd, and 3rd dimensions. Memory tile can access data in the 1st, 2nd, 3rd, and 4th dimensions. External memory can access data in the 1st, 2nd, and 3rd dimensions. If tiling dimensions exceed buffer dimensions, DMA transfers attempt to do zero padding.

Note: Zero padding is only supported on a memory tile. Only dimensions 0, 1 and 2 support zero padding. The maximum zero number padding before or after is 64 words for dimension 0, 32 words for dimension 1, and 16 words for dimension 2.

offset

Multidimensional offset with respect to the starting element in the buffer, assuming the buffer dimension is specified.

tile_traversal

Vector of traversing_parameters. tile_traversal[i] represents the i-th loop of inter-tile traversal, where i=0 represents most inner loop and i=N-1 represents most outer loop. tile_traversal structure is detailed the section below.

packet_port_id

Multiple connections can go through a single port that are previously merged through a pktmerge block or split afterward with a pktsplit block. This member represents the output port ID of the connected pktsplit or the input port ID of the connected pktmerge. If this member is set to a specific id, the data transfer only occurs if the incoming or outgoing data block ID matches it.

boundary_dimension

Real data boundary dimension for padding.

repetition

Sets the number of read repetitions for a buffer descriptor in dataflow designs to optimize data reading processes in memory tiles.

Note:

• The repetition feature allows data to be written one time and read multiple times, reducing redundant data writing cycles. That is, it allows multiple read accesses to a buffer without requiring a write access between each access.
• The semaphore locks manage repetition, so only single producer/consumer scenarios are supported.
• To form larger repetition values, multiple buffer descriptors with the same configuration are chained together. The AI Engine compiler throws an error if a configuration is not possible.

A key member of the tiling parameter is the tile_traversal vector which describes how the buffer is accessed. The structure traversing_parameters is defined as follows:

struct traversing_parameters
{
  uint32_t dimension;
  uint32_t stride;
  uint32_t wrap;
}; 

The members of this structure are as follows:

dimension

The buffer dimension on which this traversing loop applies. Depending on the type of memory element, it can be the 0, 1st, 2nd or 3rd dimension. The stride and wrap members of this structure are applied in the dimension specified.

stride

Represents the distance in terms of buffer element data type between consecutive inter-tile traversal in this dimension. When converting Tiling parameters into Buffer Descriptor parameters there can be overflow in the AI Engine Compiler. The stride is specified in the tiling parameters represents the number of samples counted in the associated dimension.

However, the Buffer Dimension stride is the number of samples counted in the linear addressing space, so the Tiling Parameter stride is multiplied by the product of the lengths of the smaller dimensions. For example, consider a 4D buffer space in a memory tile: { 64, 16, 32, 16 }. If the traversing parameter contains a stride of 8 on dimension 3, it means that the translated buffer descriptor stride is 64*16*32*8 = 256K (2^18) which cannot be encoded in the 17 bits dedicated to stride in the Buffer register of the Memory Tile.

The bitwidth of the register fields for the stepsize and wrap for the memory module, memory tile, and interface tile are as follows:

Table 1. Register Bitwidth

Module/Tile	Step Size	Wrap
Memory Module	13	8
Memory Tile	17	10
Interface Tile	20	10

wrap

Number of tiles to access in this dimension.

MM2S ports are responsible for zero-padding. If the offset contains negative values, the DMA automatically adds zero-padding to the stream to ensure the correct number of data is sent. The same behavior occurs when the tile_traversal parameters go beyond the end of the buffer.

When the stride value is lower than the tile size in one or more dimensions, the tiles overlap naturally in that dimension.