Fp8 hipstream fix #3127
Fp8 hipstream fix #3127
Conversation
✅ Deploy Preview for pytorch-fbgemm-docs ready!
To edit notification comments on pull requests, go to your Netlify site configuration. |
acoskunses-AMD
force-pushed
the
fp8_hipstream_fix
branch
from
338a53a
to
0c9fc22
Compare
acoskunses-AMD
force-pushed
the
fp8_hipstream_fix
branch
from
0c9fc22
to
863f596
Compare
|
We use hipify script for this, is the change here needed? |
|
Rebase to resolve conflicts: also as Xiaodong mentioned, internally we don't need this (automatically hipify). Is this needed on OSS workflow? I guess hipify torch is not enabled? |
acoskunses-AMD
force-pushed
the
fp8_hipstream_fix
branch
from
c948fe8
to
e94d9e9
Compare
| @@ -12,10 +12,13 @@ | |||
| #include <numeric> | |||
|
|
|||
| #include <ATen/ATen.h> | |||
| #if !defined(USE_ROCM) | |||
There was a problem hiding this comment.
hipify script should hipify .h right?
There was a problem hiding this comment.
in any case, I think maybe we can just remove this block just to be clean.
| @@ -12,11 +12,14 @@ | |||
| #include <numeric> | |||
|
|
|||
| #include <ATen/ATen.h> | |||
| #include <c10/hip/HIPStream.h> | |||
| #if !defined(USE_ROCM) | |||
There was a problem hiding this comment.
is this block even needed? If USE_ROCM is not defined, it'll basically be an empty file. So we can just remove this block here.
There was a problem hiding this comment.
Let's just remove this include and move it down to the USE_ROCM. No need to include cuda header here.
| @@ -12,10 +12,13 @@ | |||
| #include <numeric> | |||
|
|
|||
| #include <ATen/ATen.h> | |||
| #if !defined(USE_ROCM) | |||
There was a problem hiding this comment.
in any case, I think maybe we can just remove this block just to be clean.
Pick hip stream where cuda is not defined