[VectorCombine] foldShuffleToIdentity can't handle BITCAST nodes #96884
Comments
|
Ah interesting. I left bitcasts out as they can make tracking the lanes more difficult (and I honestly didn't think they can up as often as they apparently do from intrinsics). |
RKSimon
added a commit
to RKSimon/llvm-project
that referenced
this issue
Jul 4, 2024
…ement counts Basic initial patch for llvm#96884 that just handles case where we bitcast between float/intgers of the same element width
RKSimon
added a commit
to RKSimon/llvm-project
that referenced
this issue
Jul 4, 2024
…ement counts Basic initial patch for llvm#96884 that just handles case where we bitcast between float/intgers of the same element width
kbluck
pushed a commit
to kbluck/llvm-project
that referenced
this issue
Jul 6, 2024
…ement counts (llvm#97731) Basic initial patch for llvm#96884 that just handles case where we bitcast between float/integers of the same element width
RKSimon
added a commit
to RKSimon/llvm-project
that referenced
this issue
Jul 10, 2024
…if they come from the same value to form identity sequence Workaround until I can get llvm#96884 fixed properly - when trying to find identity sequences, peek through any bitcasts to see if the values all came from the same source. We don't run CSE frequently enough to merge all the bitcasts that we end up with.
RKSimon
added a commit
to RKSimon/llvm-project
that referenced
this issue
Jul 15, 2024
…if they come from the same value to form identity sequence Workaround until I can get llvm#96884 fixed properly - when trying to find identity sequences, peek through any bitcasts to see if the values all came from the same source. We don't run CSE frequently enough to merge all the bitcasts that we end up with.
RKSimon
added a commit
that referenced
this issue
Jul 15, 2024
…if they come from the same value to form identity sequence (#98334) Workaround until I can get #96884 fixed properly - when trying to find identity sequences, peek through any bitcasts to see if the values all came from the same source. We don't run CSE frequently enough to merge all the bitcasts that we end up with.
yuxuanchen1997
pushed a commit
that referenced
this issue
Jul 25, 2024
…if they come from the same value to form identity sequence (#98334) Summary: Workaround until I can get #96884 fixed properly - when trying to find identity sequences, peek through any bitcasts to see if the values all came from the same source. We don't run CSE frequently enough to merge all the bitcasts that we end up with. Test Plan: Reviewers: Subscribers: Tasks: Tags: Differential Revision: https://phabricator.intern.facebook.com/D60251606
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Noticed on #96882 - SSE/AVX code tends to result in a lot of bitcasts as the
__m128i / __m256i / __m512itypes are always treated asvXi64.CC @davemgreen
The text was updated successfully, but these errors were encountered: