LoadImageToLocal

Applicability

Product	Supported
Atlas 350 Accelerator Card	√
Atlas A3 training product/Atlas A3 inference product	√
Atlas A2 training product/Atlas A2 inference product	√
Atlas 200I/500 A2 inference product	√
Atlas inference product AI Core	√
Atlas inference product Vector Core	x
Atlas training product	x

Function Usage

Transfers image data from the global memory to the local memory. During data movement, you can preprocess images, including image flipping, image resizing (clipping, cropping, resizing, and stretching), color space conversion (CSC), and type conversion. The parameters related to image preprocessing are configured in SetAippFunctions.

Prototype

template <typename T>
__aicore__ inline void LoadImageToLocal(const LocalTensor<T>& dst, const LoadImageToLocalParams& loadDataParams)

Parameters

**Table 1** Parameters
Parameter	Input/Output	Description
dst	Output	Destination operand, which is of the LocalTensor type. The start address of LocalTensor must be 32-byte aligned. For the Atlas 350 Accelerator Card, the supported data types are uint8_t, int8_t, and half, and the supported TPosition values are VECIN, VECCALC, and VECOUT. For the Atlas A3 training product/Atlas A3 inference product, the supported data types are int8_t and half, and the supported TPosition values are A1 and B1. For the Atlas A2 training product/Atlas A2 inference product, the supported data types are int8_t and half, and the supported TPosition values are A1 and B1. For the Atlas 200I/500 A2 inference product, the supported data types are uint8_t, int8_t, and half, and the supported TPosition values are A1 and B1. For the Atlas inference product AI Core, the supported data types are uint8_t, int8_t, and half, and the supported TPosition values are A1 and B1.
loadDataParams	Input	LoadData parameter structure, of the LoadImageToLocalParams type. For details, see ${INSTALL_DIR}/include/ascendc/basic_api/interface/kernel_struct_mm.h. Replace ${INSTALL_DIR} with the actual path for storing files after the CANN software is installed. For details about the parameter description, see Table 2.

**Table 2** Parameters in the **LoadImageToLocalParams** structure
Parameter	Input/Output	Description
horizSize	Input	Horizontal width of the image loaded from the source image, in pixels. Value range: horizSize ∈ [2, 4095].
vertSize	Input	Vertical height of the image loaded from the source image, in pixels. Value range: vertSize ∈ [2, 4095].
horizStartPos	Input	Horizontal start address of the loaded image on the source image, in pixels. Value range: horizStartPos ∈ [0, 4095]. Default value: 0. Note: When the input image format is YUV420SP, XRGB8888, RGB888, or YUV400, the value of this parameter must be an even number.
vertStartPos	Input	Vertical start address of the loaded image on the source image, in pixels. Value range: vertStartPos ∈ [0, 4095]. Default value: 0. Note: If the input image is in YUV420SP format, the value of this parameter must be an even number.
srcHorizSize	Input	Horizontal width of the source image, in pixels. Value range: srcHorizSize ∈ [2, 4095]. Note: If the input image is in YUV420SP format, the value of this parameter must be an even number.
topPadSize	Input	Number of pixels padded on the top of the destination image. Value range: topPadSize ∈ [0, 32]. Default value: 0. This parameter is used for data padding. You need to call SetAippFunctions and use AippPaddingParams to configure the padding value, and then use topPadSize, botPadSize, leftPadSize, and rightPadSize to configure the padding size range.
botPadSize	Input	Number of pixels padded at the bottom of the destination image. Value range: botPadSize ∈ [0, 32]. Default value: 0.
leftPadSize	Input	Number of pixels padded to the left of the destination image. Value range: leftPadSize ∈ [0, 32]. Default value: 0.
rightPadSize	Input	Number of pixels padded to the right of the destination image. Value range: rightPadSize ∈ [0, 32]. Default value: 0.
sid	Input	Reserved parameter. This parameter is reserved for future functions. You can use the default value.

Restrictions

For details about the operand address alignment requirements, see General Address Alignment Restrictions.
The size of the image loaded to dst plus the padding size must be less than or equal to the size of the storage space.
When the padding mode is set to block padding or mirrored block padding by referring to SetAippFunctions, the left and right padding sizes (leftPadSize and rightPadSize) must be less than or equal to the horizontal size (horizSize) of the cropped image, and the top and bottom padding sizes (topPadSize and botPadSize) must be less than or equal to the vertical size (vertSize) of the cropped image because the padding data comes from the cropped image.

Returns

None

Examples

This calling example supports the Atlas inference product AI Core platform. The example image format is YUV420SP.

uint16_t horizSize = 32, vertSize = 32, horizStartPos = 0, vertStartPos = 0, srcHorizSize = 32, srcVertSize = 32, leftPadSize = 0, rightPadSize = 0;
uint32_t dstHorizSize = 32, dstVertSize = 32, cSize = 32;
uint8_t topPadSize = 0, botPadSize = 0;
uint32_t gmSrc0Size = 0, gmSrc1Size = 0, dstSize = 0;
AscendC::AippInputFormat inputFormat = AscendC::AippInputFormat::YUV420SP_U8;
uint32_t cPadMode = 0;
int8_t cPaddingValue = 0;

AscendC::TPipe pipe;
AscendC::TQue<AscendC::TPosition::A1, 1> inQueueA1;
AscendC::TQue<AscendC::TPosition::VECOUT, 1> outQueueUB;
AscendC::LocalTensor<int8_t> featureMapA1 = inQueueA1.AllocTensor<int8_t>();
uint64_t fm_addr = static_cast<uint64_t>(reinterpret_cast<uintptr_t>(fmGlobal.GetPhyAddr()));
// aipp config
AscendC::AippParams<int8_t> aippConfig;
aippConfig.cPaddingParams.cPaddingMode = cPadMode;
aippConfig.cPaddingParams.cPaddingValue = cPaddingValue;
        // fmGlobal is the entire input image. Set src1 to the start address of the UV dimension of the image.
AscendC::SetAippFunctions(fmGlobal, fmGlobal[gmSrc0Size], inputFormat, aippConfig);
AscendC::LoadImageToLocal(featureMapA1, { horizSize, vertSize, horizStartPos, vertStartPos, srcHorizSize, topPadSize, botPadSize, leftPadSize, rightPadSize });

Parent topic: Data Movement