[QualcommQnn] Add stack、split、squeeze and concat op #9185
Conversation
shentanyue
requested review from
chenjiaoAngel,
hong19860320 and
liyancas
as code owners
|
Thanks for your contribution! |
| @@ -727,6 +797,70 @@ void NCHW2NHWCDataLayoutConverter::ConvertSplit(core::Operation* operation) { | |||
| } | |||
| } | |||
|
|
|||
| void NCHW2NHWCDataLayoutConverter::ConvertStack(core::Operation* operation) { | |||
There was a problem hiding this comment.
复用了ConvertConcat的逻辑，不同点在于stack会在axis维度上堆叠，所以insert了一个维度，并将原先的perm大于axis的维度值做了变化，保证transpose的次序不变。
for (auto& perm_data : reference_permutation) {
if (perm_data >= *axis) perm_data += 1;
}
reference_permutation.insert(reference_permutation.begin() + *axis, *axis);
| auto input_permutation = GetPermutation(input_operand); | ||
| std::vector<int32_t> nchw_layout = {0, 1, 2, 3}; | ||
| std::vector<int32_t> axes; | ||
| if (axes_operand && (axes_operand->length / sizeof(int32_t)) > 0) { |
There was a problem hiding this comment.
squeeze的axes有两种场景：
1、当axes_operand不为空，且数据不为空，则根据axes做降维操作。
2、当axes_operand为空或数据为空，则会将input所有为1的维度做降维度
| auto output_dimensions_count = output_operand->type.dimensions.count; | ||
| auto axes_operand = input_operands[1]; | ||
| // Recalculate the perm according to the dimorder vector of the input operand | ||
| auto input_permutation = GetPermutation(input_operand); |
There was a problem hiding this comment.
squeeze的实现逻辑可以理解如下：例如初始 input为(1,3,1,5)，初始axes为(0,2)，input_permutation为(0,2,3,1)，则经过input_permutation变化的axes为(0,1)
首先使用原始和经过input_permutation变化后的axes数据，分别将 Identity_permutation 和 input_permutation 对应的维度去掉，
即input_permutation从(0,2,3,1)变为(3,1)，Identity_permutation从(0,1,2,3)变为(1,3)。
其次再将 input_permutation 与 Identity_permutation 做对比，若一致不做数据transpose，若不一致，例如(3,1)和(1,3)不一致，则会根据相应的关系处理得到output_permutation为(1,0)。
No description provided.