Skip to content

Conversation

@Yanxing-Shi
Copy link
Contributor

@Yanxing-Shi Yanxing-Shi commented Jan 16, 2026

Motivation

Add ck tile gemm a8w8 blockscale preshuffleB

Technical Details

Test Plan

Test Result

Submission Checklist

@Yanxing-Shi Yanxing-Shi requested a review from a team January 16, 2026 09:33
@Yanxing-Shi Yanxing-Shi marked this pull request as draft January 16, 2026 09:33
@Yanxing-Shi Yanxing-Shi force-pushed the yanxishi/tile_gemm_a8w8_blockscale_preshuffle branch from a41baa8 to 9a991af Compare January 16, 2026 09:41
@Yanxing-Shi Yanxing-Shi force-pushed the yanxishi/tile_gemm_a8w8_blockscale_preshuffle branch from 9a991af to e86ea16 Compare January 16, 2026 09:46
# with permute: {0, 2, 3, 1, 4}
#
# In our case: NW=BN, KW=BK, divisor=BK/K
x_ = x_.view(-1, x.shape[-2] // BN, x.shape[-1] // BK, BK // K, K)
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

why change this?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

why change this?

Modify this function to debug CK Tile tuning bugs,reverted to original version now.

@Yanxing-Shi Yanxing-Shi force-pushed the yanxishi/tile_gemm_a8w8_blockscale_preshuffle branch 2 times, most recently from ae2d1b0 to 4538488 Compare January 19, 2026 15:02
@Yanxing-Shi Yanxing-Shi marked this pull request as ready for review January 19, 2026 15:04
@Yanxing-Shi Yanxing-Shi force-pushed the yanxishi/tile_gemm_a8w8_blockscale_preshuffle branch 8 times, most recently from 5aa1f59 to 8490cf5 Compare January 20, 2026 02:32
@Yanxing-Shi Yanxing-Shi force-pushed the yanxishi/tile_gemm_a8w8_blockscale_preshuffle branch from 8490cf5 to 7b011c9 Compare January 20, 2026 02:35
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants