SPMI should avoid DASM roundtrip for analysis #76178
Labels
area-CodeGen-coreclr
CLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI
Milestone
Comments
dotnet-issue-labeler
bot
added
the
area-CodeGen-coreclr
|
Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch Issue DetailsFor large diffs it is prohibitively slow to produce disassembly for all method contexts with diffs, and most of us are probably familiar with superpmi-diffs timing out for a jit change with a large amount of diffs. We are already returning a list of contexts with diffs when using As a final step we'll probably want to produce .dasm files for the most important contexts identified by jit-analyze-spmi.
|
jakobbotsch
removed
the
untriaged
jakobbotsch
added a commit
to jakobbotsch/runtime
that referenced
this issue
Sep 27, 2022
This starts communicating more information about diffs back from superpmi and starts using it in the driver. The current information is the base size, diff size, base instructions, diff instructions and context size. The driver uses the base size/diff size to pick a small number of interesting contexts and only creates disassembly for these contexts. jit-analyze is then also invoked on only these contexts, but the contexts are picked so that they overlap with what jit-analyze would display. Additionally, we also pick the top 20 smallest contexts (in terms of context size) -- I frequently use the smallest contexts with diffs to investigate my change. The new behavior is only enabled when no custom metrics are specified. If custom metrics are specified we fall back to producing all disassembly files and leave it up to jit-analyze to extract and analyze the metrics specified. The net result is that we can now get SPMI diffs for changes with even hundreds of thousands of diffs in the same time as it takes to get diffs for a small change. Fix dotnet#76178
ghost
added
the
in-pr
jakobbotsch
added a commit
to jakobbotsch/runtime
that referenced
this issue
Sep 28, 2022
This starts communicating more information about diffs back from superpmi and starts using it in the driver. The current information is the base size, diff size, base instructions, diff instructions and context size. The driver uses the base size/diff size to pick a small number of interesting contexts and only creates disassembly for these contexts. jit-analyze is then also invoked on only these contexts, but the contexts are picked so that they overlap with what jit-analyze would display. Additionally, we also pick the top 20 smallest contexts (in terms of context size) -- I frequently use the smallest contexts with diffs to investigate my change. The new behavior is only enabled when no custom metrics are specified. If custom metrics are specified we fall back to producing all disassembly files and leave it up to jit-analyze to extract and analyze the metrics specified. Also remove --retainOnlyTopFiles. This is no longer necessary since we avoid creating these disassembly files in the first place. Another benefit is that all contexts mentioned by jit-analyze output will now be part of the artifacts. The net result is that we can now get SPMI diffs for changes with even hundreds of thousands of diffs in the same time as it takes to get diffs for a small change. Fix dotnet#76178
jakobbotsch
added a commit
to jakobbotsch/runtime
that referenced
this issue
Sep 29, 2022
This starts communicating more information about diffs back from superpmi and starts using it in the driver. The current information is the base size, diff size, base instructions, diff instructions and context size. The driver uses the base size/diff size to pick a small number of interesting contexts and only creates disassembly for these contexts. jit-analyze is then also invoked on only these contexts, but the contexts are picked so that they overlap with what jit-analyze would display. Additionally, we also pick the top 20 smallest contexts (in terms of context size) -- I frequently use the smallest contexts with diffs to investigate my change. The new behavior is only enabled when no custom metrics are specified. If custom metrics are specified we fall back to producing all disassembly files and leave it up to jit-analyze to extract and analyze the metrics specified. Also remove --retainOnlyTopFiles. This is no longer necessary since we avoid creating these disassembly files in the first place. Another benefit is that all contexts mentioned by jit-analyze output will now be part of the artifacts. The net result is that we can now get SPMI diffs for changes with even hundreds of thousands of diffs in the same time as it takes to get diffs for a small change. Fix dotnet#76178
|
Related: #52877 |
jakobbotsch
added a commit
that referenced
this issue
Sep 29, 2022
This starts communicating more information about diffs back from superpmi and starts using it in the driver. The current information is the base size, diff size, base instructions, diff instructions and context size. The driver uses the base size/diff size to pick a small number of interesting contexts and only creates disassembly for these contexts. jit-analyze is then also invoked on only these contexts, but the contexts are picked so that they overlap with what jit-analyze would display. Additionally, we also pick the top 20 smallest contexts (in terms of context size) -- I frequently use the smallest contexts with diffs to investigate my change. The new behavior is only enabled when no custom metrics are specified. If custom metrics are specified we fall back to producing all disassembly files and leave it up to jit-analyze to extract and analyze the metrics specified. Also, the retainTopFilesOnly option is no longer necessary since our CI pipeline will at most produce 100 .dasm files now. Another benefit is that this means that all contexts mentioned in the jit-analyze output will now be part of the artifacts. The net result is that we can now get SPMI diffs for changes with even hundreds of thousands of diffs in the same time as it takes to get diffs for a small change. Fix #76178
ghost
removed
the
in-pr
ghost
locked as resolved and limited conversation to collaborators
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
For large diffs it is prohibitively slow to produce disassembly for all method contexts with diffs, and most of us are probably familiar with superpmi-diffs timing out for a jit change with a large amount of diffs.
We are already returning a list of contexts with diffs when using
-diffMCList. This is how we figure out which contexts to produce .dasm files for. It should not be that much extra work to produce a comma-separated file instead that also includes metrics like diff size for each of the contexts. Then we can create jit-analyze-spmi that can analyze this, or enhance jit-analyze with support for this, to produce output similar to jit-analyze directly from the .csv file, without having to go through intermediate disassembly.As a final step we'll probably want to produce .dasm files for the most important contexts identified by jit-analyze-spmi.
The text was updated successfully, but these errors were encountered: