[amdgpu] Part2 add runtime #6482
Merged
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
✅ Deploy Preview for docsite-preview ready!
To edit notification comments on pull requests, go to your Netlify site settings. |
for more information, see https://pre-commit.ci
for more information, see https://pre-commit.ci
galeselee
requested review from
ailzhang,
jim19930609,
strongoier and
turbo0628
for more information, see https://pre-commit.ci
jim19930609
reviewed
Jan 11, 2023
There was a problem hiding this comment.
I remember we removed all the Runtime::register_impl() in #6830, so you don't actually need this file: taichi/runtime/amdgpu/runtime.cpp( btw, it's not even compiled )
Thanks! I forgot to delete it when I submitted :-) |
for more information, see https://pre-commit.ci
jim19930609
approved these changes
Jan 11, 2023
quadpixels
pushed a commit
to quadpixels/taichi
that referenced
this pull request
May 13, 2023
Issue: taichi-dev#6434 ### Brief Summary 1. This is the third part of adding the backend of amdgpu: adding the runtime part of the implementation. The main code for runtime is llvm ir generating gcn-isa/object and hsaco (which is a file format that can be accepted by the module launch api provided by hip) 2. After calling the relevant api to generate the gcn isa/obj, the linker of llvm (ld.lld) needs to be called to generate the hsaco file format, so there is a command line call `ld.lld -shared xxx.o -o xxx.hsaco` in the code, and the temporarily generated file is stored in the `/tmp/taichi_hsaco/` folder 3. To deal with the problem of multiple `hsaco` files being generated at the same time, a random number is used to name the related generated files, as follows: in `JITSessionAMDGPU` there is a `random_num_` and `tmp_dir_` which are assigned when the `JITSessionAMDGPU` instance is created. Each `ti.kernel` will be devided into offload-tasks which is compiled into a separate `hsaco` file. A random number bound to the `hsaco` file is obtained when the `hsaco` file is generated. Here is an example of the file after running the `ti example mpm128`: ``` taichi_hsaco/ └── 4858208420434830779 ├── taichi_amdgcn_10476395765980093855.hsaco ├── taichi_amdgcn_10476395765980093855.o ├── taichi_amdgcn_11369096326162657620.hsaco ├── taichi_amdgcn_11369096326162657620.o ├── taichi_amdgcn_11700031850871498261.hsaco ├── taichi_amdgcn_11700031850871498261.o ├── taichi_amdgcn_14803499569653867868.hsaco ├── taichi_amdgcn_14803499569653867868.o ├── taichi_amdgcn_14949458395707884954.hsaco ├── taichi_amdgcn_14949458395707884954.o ├── taichi_amdgcn_15955762247261446379.hsaco ├── taichi_amdgcn_15955762247261446379.o ├── taichi_amdgcn_16891452471041191610.hsaco ├── taichi_amdgcn_16891452471041191610.o ├── taichi_amdgcn_17615766226135707772.hsaco ├── taichi_amdgcn_17615766226135707772.o ├── taichi_amdgcn_18033844193337069056.hsaco ├── taichi_amdgcn_18033844193337069056.o ├── taichi_amdgcn_5951151729973841331.hsaco ├── taichi_amdgcn_5951151729973841331.o ├── taichi_amdgcn_6012043323411824926.hsaco ├── taichi_amdgcn_6012043323411824926.o ├── taichi_amdgcn_6796840558965541322.hsaco ├── taichi_amdgcn_6796840558965541322.o ├── taichi_amdgcn_6835984424286808860.hsaco ├── taichi_amdgcn_6835984424286808860.o ├── taichi_amdgcn_7872622170129629907.hsaco ├── taichi_amdgcn_7872622170129629907.o ├── taichi_amdgcn_8760441738982760858.hsaco ├── taichi_amdgcn_8760441738982760858.o ├── taichi_amdgcn_9006625347419529255.hsaco └── taichi_amdgcn_9006625347419529255.o ``` Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
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.
Issue: #6434
Brief Summary
This is the third part of adding the backend of amdgpu: adding the runtime part of the implementation. The main code for runtime is llvm ir generating gcn-isa/object and hsaco (which is a file format that can be accepted by the module launch api provided by hip)
After calling the relevant api to generate the gcn isa/obj, the linker of llvm (ld.lld) needs to be called to generate the hsaco file format, so there is a command line call
ld.lld -shared xxx.o -o xxx.hsacoin the code, and the temporarily generated file is stored in the/tmp/taichi_hsaco/folderTo deal with the problem of multiple
hsacofiles being generated at the same time, a random number is used to name the related generated files, as follows: inJITSessionAMDGPUthere is arandom_num_andtmp_dir_which are assigned when theJITSessionAMDGPUinstance is created. Eachti.kernelwill be devided into offload-tasks which is compiled into a separatehsacofile. A random number bound to thehsacofile is obtained when thehsacofile is generated. Here is an example of the file after running theti example mpm128: