Comments (7)
3.x is now almost 4 year old release.
Can you upgrade it to 5.6+ and check.
Also can you provide a bit more data here like: your gpu name/architecture, a sample code that reproduces this issue, compiler command you used to compile.
from hip.
3.x is now almost 4 year old release.
Can you upgrade it to 5.6+ and check.
Also can you provide a bit more data here like: your gpu name/architecture, a sample code that reproduces this issue, compiler command you used to compile.
Sorry, only version 3.3 and 2.9 are available on the hpc I'm using.
If I comment out hip_impl::hip_init() in the hipLaunchKernelGGL of functional_grid_launch.hpp, then ps.get_kernargs_size_align(reinterpret_caststd::uintptr_t(kernel)) in make_kernarg() will report an error. If I don't comment it out, everything is fine. But now, I have to comment it out.
template<typename... Args, typename F = void (*)(Args...)>
inline void hipLaunchKernelGGL(F kernel, const dim3 &numBlocks, const dim3 &dimBlocks, std::uint32_t sharedMemBytes, hipStream_t stream,
Args... args) {
// hip_impl::hip_init();
auto kernarg = hip_impl::make_kernarg(kernel, std::tuple<Args...>{std::move(args)...});
std::size_t kernarg_size = kernarg.size();
...
}
After my analysis, the problem is in the read_kernarg_metadata() in program_state.inl. In my opinion, hip_init should be unnecessary to read the arguments to the kernel function. But In hip 2.9, things work well when commenting out the hip_init(). Can you help me to analyse why the above happens.
from hip.
in 3.x hip_impl::hip_init()
was internal hip init (not to be confused with hipInit()
). It is required before you query kernel arguments because it initializes visible devices which is required by program_state.
You can not comment that line out, it is required.
from hip.
in 3.x
hip_impl::hip_init()
was internal hip init (not to be confused withhipInit()
). It is required before you query kernel arguments because it initializes visible devices which is required by program_state. You can not comment that line out, it is required.
Well, that's a real pity. I wanna be able to run get_kernargs_size_align without calling hipimpl::hip_init(). Looks like there's no way to implement my features.
from hip.
Hi @yulingao can you comment on your features? kernarg is not a concept of HIP, its an implementation detail. So it's hard to understand why you care about kernarg if you are writing HIP programs.
from hip.
@b-sumner Apologize for the delay in my response. What I would like to do is intercept the hiplaunchkernelGGL function in order to achieve remote execution of the HIP program.
from hip.
Hi @yulingao thanks for the information. I can think of a few ways to do that, but only with some pretty significant restrictions. For example, many HIP (and Cuda) programs allow device functions to directly access host memory or the memory of other GPUs. Are your applications of interest not using such features?
from hip.
Related Issues (20)
- rocm 5.7: hipcc crashes if ranges is included nd g++-12 is installed HOT 5
- __match functions HOT 2
- hipHostMalloc: hipHostMallocNumaUser flag is ignored HOT 3
- Multi-GPU, P2P access, Memcpy2D, SIGSEGV HOT 18
- HIP 5.7 Makes blender freeze HOT 19
- How to estimate the accuracy of converting HIP to CUDA or converting CUDA to HIP? HOT 4
- Number of VGPRs and SGPRs vs. hipDeviceProp_t::regsPerBlock HOT 4
- Different behaviour between clang hip (clang from llvm), /opt/rocm/llvm/bin/clang and hipcc HOT 1
- `hipEventRecord` time depends heavily on number of streams, synchronization, and HIP version HOT 7
- `hipStreamDestroy` hangs HOT 10
- Device assert broken on gfx1030 with "Bus error", or hanging after synchronize HOT 27
- `hipFreeAsync` hangs HOT 11
- graph conditional handles
- Kernels produced by `hipcc` vs `hiprtc` run the GPU in different compute modes (gfx1100). HOT 3
- `operator*` overloads for complex types are broken in 5.7 HOT 2
- Please provide pkg-config file
- Memleak when using ROCm 5.7 with Pytorch 2.2.0-dev HOT 2
- hip-6.0.0 clang/ld.lld compilation fails with: ld.lld: error: version script assignment of 'hip_4.2' to symbol 'hipExtModuleLaunchKernel' failed: symbol not defined HOT 3
- The documentation of `hipEventQuery` is incorrect HOT 4
- [Issue]: hipMalloc fails even though there is enough memory (I just freed it) HOT 5
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from hip.