JIT: Allow experimenting with different CSE subsets #92918
Conversation
Introduce two config vars for experimenting with CSEs: * `JitCSEHash`: identifies a method for CSE experimentatin * `JitCSEMask`: specifies a bitmask of allowed CSEs (by "attempt") When the hash is nonzero, any method whose hash matches will perform only the subset of CSEs specified by the mask (up to 32 CSEs). Also introduce a config var to dump the total number of CSEs to either the assemby listing footer or the one-liner from the disassembly summary. * `JitMetrics` This can perhaps be generalized eventually to report more metrics and perhaps to report them back to SPMI when it is the jit host. Finally, note CSE lcl vars that represent hoisted trees and or are "multiple-def" CSEs in the local var table. Contributes to dotnet#92915.
ghost
assigned 
dotnet-issue-labeler
bot
added
the
area-CodeGen-coreclr
|
cc @dotnet/jit-contrib |
|
Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch Issue DetailsIntroduce two config vars for experimenting with CSEs:
When the hash is nonzero, any method whose hash matches will perform only the subset of CSEs specified by the mask (up to 32 CSEs). Also introduce a config var to dump the total number of CSEs to either the assemby listing footer or the one-liner from the disassembly summary.
This can perhaps be generalized eventually to report more metrics and perhaps to report them back to SPMI when it is the jit host. Finally, note CSE lcl vars that represent hoisted trees and or are "multiple-def" CSEs in the local var table. Contributes to #92915.
|
src/coreclr/jit/jitconfigvalues.h
Outdated
| // Allow fine-grained controls of CSEs done in a particular method | ||
| CONFIG_INTEGER(JitCSEHash, W("JitCSEHash"), 0) | ||
| CONFIG_INTEGER(JitCSEMask, W("JitCSEMask"), 0) | ||
|
|
||
| // Enable metric output in jit disasm & elsewhere | ||
| CONFIG_INTEGER(JitMetrics, W("JitMetrics"), 0) | ||
|
|
There was a problem hiding this comment.
These are available in Release. Makes sense for JitMetrics, but do you want that for the JitCSE* variables?
I suggest documenting the JitCSE* variables in more detail (how do you set JitCSEMask?)
There was a problem hiding this comment.
It doesn't look like JitMetrics actually works in release (even though it is made available there).
There was a problem hiding this comment.
Ah, let me move it under debug for now too.
src/coreclr/jit/compiler.cpp
Outdated
| const bool hasProf = fgHaveProfileData(); | ||
| printf("%4d: JIT compiled %s [%s%s%s%s, IL size=%u, code size=%u%s]\n", methodsCompiled, fullName, | ||
| printf("%4d: JIT compiled %s [%s%s%s%s, IL size=%u, code size=%u%s,%s]\n", methodsCompiled, fullName, |
There was a problem hiding this comment.
| printf("%4d: JIT compiled %s [%s%s%s%s, IL size=%u, code size=%u%s,%s]\n", methodsCompiled, fullName, | |
| printf("%4d: JIT compiled %s [%s%s%s%s, IL size=%u, code size=%u%s%s]\n", methodsCompiled, fullName, |
src/coreclr/jit/compiler.cpp
Outdated
| char metricPart[128] = {0}; | ||
| if (JitConfig.JitMetrics() > 0) | ||
| { | ||
| INDEBUG(sprintf_s(debugPart, 128, ", perfScore=%.2f, numCse=%u", info.compPerfScore, optCSEcount)); |
There was a problem hiding this comment.
| INDEBUG(sprintf_s(debugPart, 128, ", perfScore=%.2f, numCse=%u", info.compPerfScore, optCSEcount)); | |
| INDEBUG(sprintf_s(metricPart, 128, ", perfScore=%.2f, numCse=%u", info.compPerfScore, optCSEcount)); |
There was a problem hiding this comment.
Thanks, I added this but wasn't using it for anything, so didn't notice it was messed up.
src/coreclr/jit/optcse.cpp
Outdated
|
|
||
| if (m_pCompiler->info.compMethodHash() == (unsigned)JitConfig.JitCSEHash()) | ||
| { | ||
| doCSE = ((1ULL << attempt) & ((unsigned long long)JitConfig.JitCSEMask())) != 0; |
There was a problem hiding this comment.
Should you check that attempt is in range? i.e.,
if (attempt >= 32) { // or `sizeof(unsigned int) * BITS_PER_BYTE`
doCSE = false;
} else {
...
}
Presumably JitCSEMask is a 32-bit int.
|
@BruceForstall think I addressed your feedback, feel free to take another look. |
Add option to suppress dumping of OSR methods.
src/coreclr/jit/optcse.cpp
Outdated
| // We can only mask the first 32 CSE attempts, so suppress anything beyond that. | ||
| // Note methods with > 32 CSEs are currently quite rare. | ||
| // | ||
| if (attempt > 32) |
There was a problem hiding this comment.
| if (attempt > 32) | |
| if (attempt >= 32) |
otherwise if attempt == 32 you do (int) 1 << 32 which is undefined?
|
SPMI failures are "no space on device". No diffs expected and no diffs in the runs that succeeded. |
Introduce two config vars for experimenting with CSEs:
JitCSEHash: identifies a method for CSE experimentatinJitCSEMask: specifies a bitmask of allowed CSEs (by "attempt")When the hash is nonzero, any method whose hash matches will perform only the subset of CSEs specified by the mask (up to 32 CSEs).
Also introduce a config var to dump the total number of CSEs to either the assemby listing footer or the one-liner from the disassembly summary.
JitMetricsThis can perhaps be generalized eventually to report more metrics and perhaps to report them back to SPMI when it is the jit host.
Finally, note CSE lcl vars that represent hoisted trees and or are "multiple-def" CSEs in the local var table.
Contributes to #92915.