More flexible matmul test #476
Merged
+338
−200
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
maxtremblay
commented
Feb 11, 2025
Comment on lines
+379
to
+425
| // Perform matmul | ||
| for row in 0..m { | ||
| for col in 0..n { | ||
| for middle in 0..k { | ||
| let lhs_index = row * k + middle; | ||
| let rhs_index = middle * n + col; | ||
| let out_index = row * n + col; | ||
|
|
||
| let l = lhs[batch_lhs + lhs_index] as u16; | ||
| let r = rhs[batch_rhs + rhs_index] as u16; | ||
| let prod = l * r; | ||
|
|
||
| out[batch_out + out_index] += prod as i32; | ||
| } | ||
| } | ||
| } | ||
|
|
||
| // Substract rhs_zero_offset * sum_rows(lhs) | ||
| for row in 0..m { | ||
| let mut sum = 0; | ||
| for col in 0..k { | ||
| sum += lhs[batch_lhs + row * k + col] as i32; | ||
| } | ||
| sum *= rhs_zero_offset; | ||
| for col in 0..n { | ||
| out[batch_out + row * n + col] -= sum; | ||
| } | ||
| } | ||
|
|
||
| // Substract lhs_zero_offset * sum_cols(rhs) | ||
| for col in 0..n { | ||
| let mut sum = 0; | ||
| for row in 0..k { | ||
| sum += rhs[batch_lhs + row * n + col] as i32; | ||
| } | ||
| sum *= lhs_zero_offset; | ||
| for row in 0..m { | ||
| out[batch_out + row * n + col] -= sum; | ||
| } | ||
| } | ||
|
|
||
| // Add final constant term | ||
| for row in 0..m { | ||
| for col in 0..n { | ||
| out[batch_out + row * n + col] += (k as i32) * lhs_zero_offset * rhs_zero_offset; | ||
| } | ||
| } |
There was a problem hiding this comment.
Here you can see the flow for the quantized matmul.
There was a problem hiding this comment.
Another way to test it is to dequantize a tensor, perform matmul, requantize the tensor and see if the quantize matmul works.
nathanielsimard
approved these changes
Feb 11, 2025
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.
Prepare the matmul tests to be ready for quantization. However, this currently skip all tests related to quantization as it is not implemented yet. The strategy will be to partially enable quantization tests during development using this method:
https://github.com/tracel-ai/cubecl/blob/c1c333c56461aeea2bba6aac9f72c897ad725d01/crates/cubecl-linalg/src/matmul/tests/test_utils.rs#L148C1-L152C6