Tool/software:
Hi. I had some problem with the test case prms,
// static MMALIB_CNN_fullyConnectedBias_ixX_ixX_oxX_testParams_t testParams[] =
// {
// STATIC, // test pattern
// staticRefInputX_test3000, // Input feature map
// MMALIB_INT8, // datatypeX
// staticRefInputH_test3000, // Pre-processed filter
// MMALIB_INT8, // datatypeH
// staticRefInputBias_test3000, // Bias vector
// MMALIB_INT32, // datatype for bias
// staticRefInputScale_test3000, // Scale vector/scalar
// MMALIB_INT8, // datatype scale
// staticRefInputShift_test3000, // Shift vector/scalar
// staticRefOutput_test3000, // output feature map reference
// MMALIB_INT8, // output datatype
// NULL, // DynamicRangeMin
// NULL, // DynamicRangeMax
// 64, // height of feature map
// 1280, // width of feature map
// 1600, // stride of feature map
// 1280, // height of kernel matrix
// 128, // width of kernel matrix
// 81920, // stride of kernel matrix
// 64, // height of output matrix
// 128, // width of output matrix
// 128, // stride of output matrix
// MMALIB_SATURATION, // activation type
// 0, // parameterized saturation min
// 0, // parameterized saturation max
// MMALIB_SUCCESS, // expectedStatus;
// MMALIB_DYNAMIC_RANGE_DISABLE, // enableDynamicRange
// MMALIB_DYNAMIC_RANGE_NOINIT, // initDynamicRange
// MMALIB_QUANT_PER_KERNEL, // quantization method
// MMALIB_TEST_OUTPUT_MSMC, // output data location
// 1, // num. of repetitions
// 0, // pre-process flag for kernel matrix
// 3000, // test ID
// };
About the test, i dont under stand about the stride of feature input matrix and stride of output matrix. Like the figure in MMALIB guideline of FC-layer, i could understand the stride of that weight matrix after pre-processing. But about the X^T and Y^T matrix (in and output matrices): it so ambiguous. So can you guys explain the single params above in details for me?
Tks,
About the test, i dont under stand about the stride of feature input matrix and stride of output matrix. Like the figure in MMALIB guideline of FC-layer, i could understand the stride of that weight matrix after pre-processing. But about the X^T and Y^T matrix (in and output matrices): it so ambiguous. So can you guys explain the single params above in details for me?
Tks,
