[GPU] Support for converting zero-filled memory for i4 compressed weights #29937
+103
−4
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
ahnyoung-paul
added
category: GPU
isanghao
reviewed
Apr 4, 2025
| strides.push_back(1); | ||
| strides.push_back(padded_dims[1]); | ||
| dnnl::memory::data_type dt = convert_data_type(weights_layout.data_type); | ||
| weights_md = dnnl::memory::desc(dims, dt, strides); |
There was a problem hiding this comment.
- I think it would be better to be placed within
onednn::layout_to_memory_desc. - This function supports
padded_dims[1]only. Then could you add assert to checkpadded_dims[0] == 0? - Could you add test case for this fix?
| @@ -115,7 +189,17 @@ static void create_data(ProgramBuilder& p, const ov::Shape& const_shape, const s | |||
| auto buf = lock.data(); | |||
| auto bufSize = constLayout.bytes_count(); | |||
|
|
|||
| std::memcpy(&buf[0], &data[0], bufSize); | |||
| if (constLayout.data_padding) { | |||
There was a problem hiding this comment.
isn't it necessary to check also data_type == u4 || data_type == i4?
| std::memcpy(&padded_vec[padded_begin], &non_padded_vec[non_padded_begin], non_padded_row_size); | ||
| if (non_padded_end < non_padded_dims_size) { | ||
| uint8_t s0, s1; | ||
| unpack8to4(non_padded_vec[non_padded_end], s0, s1); |
There was a problem hiding this comment.
I'm afraid it will create wrong data if you convert char data into uint8_t. For example, when char contains negative value.
|
|
||
| // Function to convert a non-padded vector to a padded vector | ||
| static void copy_to_padded_vector(const std::vector<int> &non_padded_dims, const char* non_padded_vec, | ||
| const std::vector<int> &padded_dims, char* padded_vec) { |
There was a problem hiding this comment.
this function works only for 4bit data element. I think it would be better to clarify from the name.
| temp_padding_buffer[index] = s0; | ||
| ++index; | ||
| temp_padding_buffer[index] = s1; | ||
| } |
There was a problem hiding this comment.
How long does it take? I'm afraid this will be very slow. I think we need to find a better way.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Details:
Tickets: