cupvaTensorDataFlowParams_t

Defined in src/host/c_api/include/detail/workloads/cupva_host_dataflow.h

struct cupvaTensorDataFlowParams_t

Parameters structure for CupvaTensorDataFlowSetParams.

This structure is used to setup a TensorDataFlow with CupvaTensorDataFlowSetParams

Public Members

cupvaParameter_t *handler

cupvaParameter_t which has been initialized to refer to a TensorDataFlow handler

int32_t roiX

X offset in pixels into the surface where the ROI starts. Defaults to zero.

int32_t roiY

Y offset in pixels into the surface where the ROI starts. Defaults to zero.

int32_t roiWidth

ROI width in pixels. If zero, defaults to full tensor width.

int32_t roiHeight

ROI height in pixels. If zero, defaults to full tensor height.

int32_t tileWidth

Tile width in pixels.

int32_t tileHeight

Tile height in pixels.

int32_t tileDepth

Tile Depth in pixels.

int32_t bpp

Bytes per pixel.

int32_t haloX

X halo in pixels. Halo is the overlap between adjacent tiles.

int32_t haloY

Y halo in pixels. Halo is the overlap between adjacent tiles.

cupvaPadModeType_t padMode

Padding mode for out of bounds accesses. If a read tile would access out-of-bounds according to the tensor dimensions, DMA engine is used to generate padding instead.

int32_t padVal

Padding value to use for constant padding mode.

const void *ptrSrc

Source device pointer. Should be NULL if opSrc is specified.

cupvaOffsetPointer_t *opSrc

Source offset pointer. Should be NULL if ptrSrc is specified.

int32_t srcWidth

Valid width of the source tensor in pixels. Any accesses out-of-bounds on reads results in padding being applied to avoid out-of-bounds accesses.

int32_t srcHeight

Valid height of the source tensor in pixels. Any accesses out-of-bounds on reads results in padding being applied to avoid out-of-bounds accesses.

int32_t srcDepth

Valid depth of the source tensor in pixels. Any accesses out-of-bounds on reads results in padding being applied to avoid out-of-bounds accesses.

int32_t rowStrideSrc

Line pitch of the source tensor in pixels.

int32_t depthStrideSrc

Depth stride of the source tensor in pixels.

void *ptrDst

Destination device pointer. Should be NULL if opSrc is specified.

cupvaOffsetPointer_t *opDst

Destination offset pointer. Should be NULL if ptrSrc is specified.

int32_t dstWidth

Valid width of the destination tensor in pixels.

int32_t dstHeight

Valid height of the destination tensor in pixels.

int32_t dstDepth

Valid height of the destination tensor in pixels.

int32_t rowStrideDst

Line pitch of the destination tensor in pixels.

int32_t depthStrideDst

Depth stride of the destination tensor in pixels.

void *ptrTileBuffer

Device pointer to the VMEM tile buffer.