Rieman fit in CA #169
Merged
Rieman fit in CA #169
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
Validation summaryReference release CMSSW_10_2_4 at 6e8f631
|
|
@rovere what changes do we expect running the usual workflows ? |
|
none, since the products still live only on the GPU. |
|
Looks like this PR introduces a GPU memory leak: Can you have a look ? |
|
I will |
cudaFree(helix_fit_resultsGPU_) together with the other buffers
Validation summaryReference release CMSSW_10_2_4 at 6e8f631
|
Reported by cuda-memcheck --tool memcheck --leak-check full --report-api-errors all:
CUDA-MEMCHECK
Leaked 3040000 bytes at 0x75ab80000
Saved host backtrace up to driver entry point at cudaMalloc time
Host Frame:/lib64/libcuda.so.1 (cuMemAlloc_v2 + 0x17f) [0x22bdaf]
Host Frame:/data/cmssw/slc7_amd64_gcc700/cms/cmssw/CMSSW_10_2_4_Patatrack/external/slc7_amd64_gcc700/lib/libcudart.so.9.2 [0x2fdc3]
Host Frame:/data/cmssw/slc7_amd64_gcc700/cms/cmssw/CMSSW_10_2_4_Patatrack/external/slc7_amd64_gcc700/lib/libcudart.so.9.2 [0x1196b]
Host Frame:/data/cmssw/slc7_amd64_gcc700/cms/cmssw/CMSSW_10_2_4_Patatrack/external/slc7_amd64_gcc700/lib/libcudart.so.9.2 (cudaMalloc + 0x16f) [0x3fb2f]
Host Frame:/data/user/fwyzard/patatrack/validation/run.ho9ucTUSKo/testing/lib/slc7_amd64_gcc700/pluginRecoPixelVertexingPixelTripletsPlugins.so (CAHitQuadrupletGeneratorGPU::allocateOnGPU() + 0x341) [0x27301]
Host Frame:/data/cmssw/slc7_amd64_gcc700/cms/cmssw/CMSSW_10_2_4_Patatrack/lib/slc7_amd64_gcc700/libHeterogeneousCoreCUDACore.so (heterogeneous::GPUCuda::call_beginStreamGPUCuda(edm::StreamID) + 0x945) [0x6785]
Host Frame:/data/user/fwyzard/patatrack/validation/run.ho9ucTUSKo/testing/lib/slc7_amd64_gcc700/pluginRecoPixelVertexingPixelTripletsPlugins.so [0x2ada6]
Host Frame:/data/cmssw/slc7_amd64_gcc700/cms/cmssw/CMSSW_10_2_4_Patatrack/lib/slc7_amd64_gcc700/libFWCoreFramework.so (edm::Worker::beginStream(edm::StreamID, edm::StreamContext&) + 0x110) [0x23e900]
Host Frame:/data/cmssw/slc7_amd64_gcc700/cms/cmssw/CMSSW_10_2_4_Patatrack/lib/slc7_amd64_gcc700/libFWCoreFramework.so (edm::WorkerManager::beginStream(edm::StreamID, edm::StreamContext&) + 0x32) [0x1e6b32]
Host Frame:/data/cmssw/slc7_amd64_gcc700/cms/cmssw/CMSSW_10_2_4_Patatrack/lib/slc7_amd64_gcc700/libFWCoreFramework.so (edm::EventProcessor::beginJob() + 0x1df) [0x20599f]
Host Frame:cmsRun [0xead4]
Host Frame:cmsRun (main + 0x162) [0xd2c2]
Host Frame:/lib64/libc.so.6 (__libc_start_main + 0xf5) [0x22445]
Host Frame:cmsRun [0xd598]
LEAK SUMMARY: 3040000 bytes leaked in 1 allocations
ERROR SUMMARY: 1 error
Validation summaryReference release CMSSW_10_2_4 at 6e8f631
|
|
No changes to the standard workflows (as expected). For the 10824.8 TTbar workflow it adds to the profile: the new kernels: |
|
@rovere @felicepantaleo do we want to have a flag to turn on the Riemann fit, or run it unconditionally ? |
|
@fwyzard Could you elaborate in what ways does running the Riemann fit unconditionally make benchmarking much less useful? The motivation for unconditional execution is to replace the CPU-side poor-man fit (that is used for filtering quadruplets) with the real fit. On the other hand, the circle fit becoming now the second heaviest kernel, maybe we have/want to reconsider that. |
|
Hi Matti, let me try to summarise the chat with Felice, and explain my point of view here. There are (at least) two things going on here:
The goal of the former is to improve the physics performance, at a negligible (or at least affordable) increase in computing cost. Making the Riemann fit the default workflow goes in this direction. The goal of the latter is to show that a given algorithm can be implemented and offloaded to GPUs in a way that significantly reduces the overall cost at a fixed performance. Being able to measure the event processing throughput with the same inputs and outputs is a requirement here. Felice claims we should do former first, and use that as a baseline for the latter. My point is that I do not care much about the former, but I am quite invested in the latter, so I would rather not have the Patatrack code base in a situation where it cannot be measured. |
|
Thanks Andrea for the explanation. Allow me to ask one further clarification: in
with "it" do you refer to physics performance, technical performance, or both? |
|
Sorry for the confusion - I mean in a state where we cannot compare the
throughput on GPU vs CPU for a similar enough workflow.
|
|
Ok, thanks for the clarification. I'm fine with adding configuration flags, we could e.g. keep all the Riemann fit work in the .9 workflow, and leave .8 as it is with this respect. |
fwyzard
reviewed
Sep 20, 2018
| // ge[] ==> [(0,0), (0,1), (0,2), (1,1), (1,2), (2,2)] | ||
| // So: | ||
| // (i=0,j=0,l=1) ==> (x_0,y_0) | ||
| // (i=1,j=2,l=0) ==> (z_1,x_1) |
fwyzard
reviewed
Sep 20, 2018
| // So: | ||
| // (i=0,j=0,l=1) ==> (x_0,y_0) | ||
| // (i=1,j=2,l=0) ==> (z_1,x_1) | ||
| auto ge_idx = j + l; |
There was a problem hiding this comment.
wouldn't it be simpler to rewrite this as
auto ge_idx = j + l + (j > 0 and l > 0);?
Validation summaryReference release CMSSW_10_2_4 at 6e8f631
|
fwyzard
pushed a commit
that referenced
this pull request
Oct 8, 2020
Also, add back the stand-alone GPU fit test.
fwyzard
pushed a commit
that referenced
this pull request
Oct 20, 2020
Also, add back the stand-alone GPU fit test.
fwyzard
pushed a commit
that referenced
this pull request
Oct 20, 2020
Also, add back the stand-alone GPU fit test.
fwyzard
pushed a commit
that referenced
this pull request
Oct 23, 2020
Also, add back the stand-alone GPU fit test.
fwyzard
pushed a commit
that referenced
this pull request
Nov 6, 2020
Also, add back the stand-alone GPU fit test.
fwyzard
pushed a commit
that referenced
this pull request
Nov 6, 2020
Also, add back the stand-alone GPU fit test.
fwyzard
pushed a commit
that referenced
this pull request
Nov 16, 2020
Also, add back the stand-alone GPU fit test.
fwyzard
added a commit
that referenced
this pull request
Nov 27, 2020
Also, add back the stand-alone GPU fit test.
fwyzard
pushed a commit
that referenced
this pull request
Dec 26, 2020
Also, add back the stand-alone GPU fit test.
fwyzard
pushed a commit
that referenced
this pull request
Jan 15, 2021
Also, add back the stand-alone GPU fit test.
fwyzard
pushed a commit
that referenced
this pull request
Mar 23, 2021
Also, add back the stand-alone GPU fit test.
fwyzard
pushed a commit
that referenced
this pull request
Apr 1, 2021
Also, add back the stand-alone GPU fit test.
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.
This supersedes #162