Optimize lsra for MinOpts #96386

kunalspathak · 2024-01-01T18:37:00Z

Created separate methods *Minimal() and moved the enregisterLocalVars=false specific code in it
Only run REG_ORDER and REG_NUM heuristics when enregisterLocalVars=false

ghost · 2024-01-01T18:37:11Z

Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch
See info in area-owners.md if you want to be subscribed.

Issue Details

null

Author:	kunalspathak
Assignees:	kunalspathak
Labels:	`area-CodeGen-coreclr`
Milestone:	-

kunalspathak · 2024-01-02T00:50:15Z

Need to find out what is going on with arm64, but initial results are very motivating!

kunalspathak · 2024-01-02T18:59:59Z

I tried to investigate the windows-arm64 TP regression

It turns out that the majority of slowness is attributed to 2 method contexts of benchmarks.run.windows.arm64.checked.mch : 2999 and 15658 and the reason being the majority of gains from register allocation are compensated by the equal work that emitter has to do to produce more instructions. Below is the pin difference for just 2 method contexts.

Base: 414060, Diff: 433023, +4.5798%

??$allocateReg@$0A@@LinearScan@@AEAA?AW4_regNumber_enum@@PEAVInterval@@PEAVRefPosition@@@Z : 24193  : +942.46% : 13.83% : +5.8429%
??$allocateRegistersForMinOpt@$0A@@LinearScan@@QEAAXXZ                                     : 22858  : NA       : 13.07% : +5.5205%
?emitOutputInstr@emitter@@IEAA_KPEAUinsGroup@@PEAUinstrDesc@1@PEAPEAE@Z                    : 8062   : +61.79%  : 4.61%  : +1.9471%
?lvaFrameAddress@Compiler@@QEAAHHPEA_N@Z                                                   : 6491   : +575.44% : 3.71%  : +1.5676%
?emitAllocAnyInstr@emitter@@AEAAPEAX_KW4emitAttr@@@Z                                       : 2664   : +45.73%  : 1.52%  : +0.6434%
?emitIns_S_R@emitter@@QEAAXW4instruction@@W4emitAttr@@W4_regNumber_enum@@HH@Z              : 2568   : +143.87% : 1.47%  : +0.6202%
?rsSpillTree@RegSet@@AEAAXW4_regNumber_enum@@PEAUGenTree@@I@Z                              : 2544   : NA       : 1.45%  : +0.6144%
?emitIns_R_S@emitter@@QEAAXW4instruction@@W4emitAttr@@W4_regNumber_enum@@HH@Z              : 2391   : +127.05% : 1.37%  : +0.5775%
?IsMultiRegNode@GenTree@@QEBA_NXZ                                                          : 1988   : +177.03% : 1.14%  : +0.4801%
?emitInsCode@emitter@@AEAAIW4instruction@@W4insFormat@1@@Z                                 : 1496   : +65.99%  : 0.86%  : +0.3613%
?idCodeSize@instrDesc@emitter@@QEBAIXZ                                                     : 1290   : +45.94%  : 0.74%  : +0.3115%
?updateMaxSpill@LinearScan@@QEAAXPEAVRefPosition@@@Z                                       : 1267   : +54.61%  : 0.72%  : +0.3060%
?resetAllRegistersState@LinearScan@@AEAAXXZ                                                : 1255   : NA       : 0.72%  : +0.3031%
?unassignPhysReg@LinearScan@@AEAAXPEAVRegRecord@@PEAVRefPosition@@@Z                       : 1232   : +47.26%  : 0.70%  : +0.2975%
memset                                                                                     : 1218   : +44.71%  : 0.70%  : +0.2942%
?initLclVarAddr@emitLclVarAddr@@QEAAXHI@Z                                                  : 1075   : +139.97% : 0.61%  : +0.2596%
?emitSizeOfInsDsc@emitter@@AEBA_KPEAUinstrDesc@1@@Z                                        : 1056   : +106.88% : 0.60%  : +0.2550%
?genUnspillRegIfNeeded@CodeGen@@IEAAXPEAUGenTree@@@Z                                       : 996    : +101.01% : 0.57%  : +0.2405%
?insertCopyOrReload@LinearScan@@QEAAXPEAUBasicBlock@@PEAUGenTree@@IPEAVRefPosition@@@Z     : 935    : NA       : 0.53%  : +0.2258%
?emitEndCodeGen@emitter@@QEAAIPEAVCompiler@@_N11IPEAI2PEAPEAX33@Z                          : 903    : +26.23%  : 0.52%  : +0.2181%
?updateAssignedInterval@LinearScan@@AEAAXPEAVRegRecord@@PEAVInterval@@@Z                   : 777    : +35.41%  : 0.44%  : +0.1877%
?appendToCurIG@emitter@@IEAAXPEAUinstrDesc@1@@Z                                            : 731    : +52.18%  : 0.42%  : +0.1765%
?rsUnspillInPlace@RegSet@@AEAAPEAVTempDsc@@PEAUGenTree@@W4_regNumber_enum@@I@Z             : 711    : NA       : 0.41%  : +0.1717%
?assignPhysReg@LinearScan@@AEAAXPEAVRegRecord@@PEAVInterval@@@Z                            : 693    : +36.21%  : 0.40%  : +0.1674%
?emitFindOffset@emitter@@IEBAIPEBUinsGroup@@I@Z                                            : 672    : +72.41%  : 0.38%  : +0.1623%
?insEncodeDatasizeLS@emitter@@CAIIW4emitAttr@@@Z                                           : 598    : +120.56% : 0.34%  : +0.1444%
?tmpRlsTemp@RegSet@@QEAAXPEAVTempDsc@@@Z                                                   : 588    : NA       : 0.34%  : +0.1420%
?TryGetUse@Range@LIR@@QEAA_NPEAUGenTree@@PEAVUse@2@@Z                                      : 539    : NA       : 0.31%  : +0.1302%
?genCodeForTreeNode@CodeGen@@IEAAXPEAUGenTree@@@Z                                          : 451    : +9.59%   : 0.26%  : +0.1089%
?emitGCregDeadUpd@emitter@@QEAAXW4_regNumber_enum@@PEAE@Z                                  : 402    : +33.67%  : 0.23%  : +0.0971%
??$resolveRegisters@$0A@@LinearScan@@QEAAXXZ                                               : 396    : +4.76%   : 0.23%  : +0.0956%
?allocateMemory@ArenaAllocator@@QEAAPEAX_K@Z                                               : 368    : +3.28%   : 0.21%  : +0.0889%
?emitNewInstrCns@emitter@@AEAAPEAUinstrDesc@1@W4emitAttr@@_J@Z                             : 357    : +87.50%  : 0.20%  : +0.0862%
?tmpSlot@RegSet@@CAII@Z                                                                    : 286    : NA       : 0.16%  : +0.0691%
?setIntervalAsSpilled@LinearScan@@AEAAXPEAVInterval@@@Z                                    : 252    : NA       : 0.14%  : +0.0609%
?gcMakeRegPtrTable@GCInfo@@QEAAXPEAVGcInfoEncoder@@IIW4MakeRegPtrMode@1@PEAI@Z             : 244    : +10.95%  : 0.14%  : +0.0589%
?ins_Load@CodeGenInterface@@QEAA?AW4instruction@@W4var_types@@_N@Z                         : 237    : +179.55% : 0.14%  : +0.0572%
?lvaAllocateTemps@Compiler@@QEAAHH_N@Z                                                     : 221    : +153.47% : 0.13%  : +0.0534%
?TryGetUse@GenTree@@QEAA_NPEAU1@PEAPEAPEAU1@@Z                                             : 212    : NA       : 0.12%  : +0.0512%
?getWeight@LinearScan@@AEAANPEAVRefPosition@@@Z                                            : 189    : +10.41%  : 0.11%  : +0.0456%
?recordMaxSpill@LinearScan@@QEAAXXZ                                                        : 182    : +55.83%  : 0.10%  : +0.0440%
?UpdateLifeVar@?$TreeLifeUpdater@$00@@AEAAXPEAUGenTree@@PEAUGenTreeLclVarCommon@@@Z        : -175   : -21.74%  : 0.10%  : -0.0423%
?getMatchingConstants@LinearScan@@AEAA_K_KPEAVInterval@@PEAVRefPosition@@@Z                : -180   : -100.00% : 0.10%  : -0.0435%
?processBlockEndAllocation@LinearScan@@AEAAXPEAUBasicBlock@@@Z                             : -1896  : -100.00% : 1.08%  : -0.4579%
??$allocateRegisters@$0A@@LinearScan@@QEAAXXZ                                              : -30030 : -100.00% : 17.17% : -7.2526%
??$select@$0A@@RegisterSelection@LinearScan@@QEAA_KPEAVInterval@@PEAVRefPosition@@@Z       : -45417 : -100.00% : 25.97% : -10.9687%

If I exclude those 2 contexts from the benchmarks.run.windows collection, then locally I do see TP gains that matches for other platforms:

2999 diff and 15658 diff

kunalspathak · 2024-01-04T06:00:25Z

There was a bug where I was almost always assigning busy register. That was having us generate lot of spilling and emit spilling code increase the TP. I fixed that in 7f29c4e. The new numbers are even better, especially on Arm64, we see almost 13% TP improvements. Diffs:

kunalspathak · 2024-01-19T19:54:19Z

Yes, so here is the breakdown:

FullOpts:

Blue - All x64 platforms is showing around 17K+ regression
Green - All arm64 platforms is showing around 32K+ regression (acceptable because of arm64 larger code size)

MinOpts:

Yellow - All arm64 platforms MinOpts is showing around 145K+ regression
Purple - Linux/x64 platform MinOpts is showing 464K+ improvement

Looking in more detail, I checked what % of contents are affected in each of the configuration. For x64, there are around 28% (both windows/linux) and on arm64, there are around 15% (both windows/linux), so we at least see that the contexts affected are same for a particular architecture.

OS	arch	total context	diff context	% diff
windows	x64	2384370	666648	28%
linux	x64	2546536	716232	28%
windows	arm64	2329306	366025	16%
linux	arm64	2495514	385512	15%
windows	x86	`2244188`	505186	23%
linux	arm	2239583	671422	30%

Linux/x64 analysis

Values are mostly in rax (because of REG_ORDER), I see that instructions are more compact with it, Here is an example from benchmarks.run.linux collection

-  400FB67F4C          movzx    rdi, byte  ptr [rdi+0x4C]
+ 0FB6404C             movzx    rax, byte  ptr [rax+0x4C]

Some other places, since value is mostly in rax, I see cwde instruction is used instead of movsx which is smaller in size:

That could justify the improvements on linux, but if you want I can dig further.

jakobbotsch · 2024-01-19T20:12:08Z

IMO, since the change is about MinOpts, it should be no-diff for FullOpts. How does the change end up affecting codegen for FullOpts in the cases where we have no tracking variables? Isn't that unexpected? And if it's expected, can the new logic be made to run only if OptimizationDisabled() && !enregisterLocalVar or something?

Edit: It makes sense to me that we'd still want the optimized heuristics for LIR temps, even when we don't have any locals to enregister, so I think we should base the use of the fast and less optimal heuristics around OptimizationDisabled() instead.

kunalspathak · 2024-01-20T00:13:50Z

IMO, since the change is about MinOpts, it should be no-diff for FullOpts.

This change is not just for MinOpts, but when we decide that enregistering of local vars is not needed. This is fairly common in MinOpts/Tier0/MethodImplOptions.NoOptimization but also happen in FullOpts, when there are no tracking variables.

How does the change end up affecting codegen for FullOpts in the cases where we have no tracking variables? Isn't that unexpected?

We turn off enregisterLocalVars if we don't have any tracking variable and hence this change is affecting exactly and only those scenarios for FullOpts.

runtime/src/coreclr/jit/lsra.cpp

Lines 1353 to 1358 in 62d33ee

    
           // but we don't want to spend time on the lclVar parts of LinearScan 
        
           // if we have no tracked locals. 
        
           if (enregisterLocalVars && (compiler->lvaTrackedCount == 0)) 
        
           { 
        
               enregisterLocalVars = false; 
        
           }

And if it's expected, can the new logic be made to run only if OptimizationDisabled() && !enregisterLocalVar or something? Edit: It makes sense to me that we'd still want the optimized heuristics for LIR temps, even when we don't have any locals to enregister, so I think we should base the use of the fast and less optimal heuristics around OptimizationDisabled() instead.

I am not sure I follow what you are suggesting.

jakobbotsch · 2024-01-20T00:32:11Z

We turn off enregisterLocalVars if we don't have any tracking variable and hence this change is affecting exactly and only those scenarios for FullOpts.

The end result seems to be that we are regressing CQ in FullOpts for a throughput optimization. That doesn't seem right.

enregisterLocalVars does not seem like the right check to me. Sure if there are no local vars then there are also no local vars to enregister, even in FullOpts, but that doesn't mean there are no other intervals that should get good register allocation.

I am not sure I follow what you are suggesting.

Call allocateRegistersMinimal() only if opts.OptimizationDisabled(). Optionally assert that enregisterLocalVars is false in that case. Keep allocating all intervals with the full set of heuristics in FullOpts.

BruceForstall · 2024-01-20T00:35:20Z

If @jakobbotsch suggestion restricts the change the MinOpts, then this change could be only about MinOpts CQ + TP. @kunalspathak if you think a version of this change applies to FullOpts, then perhaps that could be made as a follow-up, to isolate the conversation about the FullOpts impact to that PR.

kunalspathak · 2024-01-20T00:54:20Z

Call allocateRegistersMinimal() only if opts.OptimizationDisabled(). Optionally assert that enregisterLocalVars is false in that case. Keep allocating all intervals with the full set of heuristics in FullOpts.

ah i see. Yes, I was trying out similar change locally to eliminate the full opt impact.

kunalspathak · 2024-01-20T06:21:28Z

superpmi failures are: OSError: [Errno 28] No space left on device

/azp run runtime-coreclr superpmi-replay

kunalspathak · 2024-01-20T06:21:49Z

/azp run runtime-coreclr superpmi-replay

azure-pipelines · 2024-01-20T06:22:04Z

Azure Pipelines successfully started running 1 pipeline(s).

kunalspathak · 2024-01-22T05:24:49Z

kunalspathak · 2024-01-22T19:27:13Z

Assembly diffs

Assembly diffs for linux/arm64 ran on windows/x64

Diffs are based on ${\color{orange}2,491,707}$ contexts (${\color{orange}1,001,189}$ MinOpts, ${\color{orange}1,490,518}$ FullOpts).

${\color{blue}MISSED}$ contexts: ${\color{blue}3,807 (0.15\%)}$

Overall (${\color{red}+148,204}$ bytes)

Collection	Base size (bytes)	Diff size (bytes)
benchmarks.run.linux.arm64.checked.mch	14,270,080	+0
benchmarks.run_pgo.linux.arm64.checked.mch	83,058,256	${\color{red}+1,980}$
benchmarks.run_tiered.linux.arm64.checked.mch	21,983,044	${\color{red}+588}$
coreclr_tests.run.linux.arm64.checked.mch	510,170,492	${\color{red}+144,156}$
libraries.crossgen2.linux.arm64.checked.mch	55,689,604	+0
libraries.pmi.linux.arm64.checked.mch	75,931,904	+0
libraries_tests.run.linux.arm64.Release.mch	379,668,404	${\color{red}+1,372}$
libraries_tests_no_tiered_compilation.run.linux.arm64.Release.mch	162,138,568	${\color{red}+108}$
realworld.run.linux.arm64.checked.mch	15,784,696	+0

MinOpts (${\color{red}+148,204}$ bytes)

Collection	Base size (bytes)	Diff size (bytes)
benchmarks.run.linux.arm64.checked.mch	218,508	+0
benchmarks.run_pgo.linux.arm64.checked.mch	26,552,600	${\color{red}+1,980}$
benchmarks.run_tiered.linux.arm64.checked.mch	17,400,224	${\color{red}+588}$
coreclr_tests.run.linux.arm64.checked.mch	349,264,408	${\color{red}+144,156}$
libraries.crossgen2.linux.arm64.checked.mch	1,636	+0
libraries.pmi.linux.arm64.checked.mch	119,984	+0
libraries_tests.run.linux.arm64.Release.mch	214,980,496	${\color{red}+1,372}$
libraries_tests_no_tiered_compilation.run.linux.arm64.Release.mch	13,480,568	${\color{red}+108}$
realworld.run.linux.arm64.checked.mch	575,816	+0

Assembly diffs for linux/x64 ran on windows/x64

Diffs are based on ${\color{orange}2,542,496}$ contexts (${\color{orange}1,013,312}$ MinOpts, ${\color{orange}1,529,184}$ FullOpts).

${\color{blue}MISSED}$ contexts: ${\color{blue}4,042 (0.16\%)}$

Overall (${\color{green}-464,740}$ bytes)

Collection	Base size (bytes)	Diff size (bytes)
benchmarks.run.linux.x64.checked.mch	13,063,008	${\color{green}-5}$
benchmarks.run_pgo.linux.x64.checked.mch	64,981,453	${\color{green}-15,309}$
benchmarks.run_tiered.linux.x64.checked.mch	17,221,305	${\color{green}-11,296}$
coreclr_tests.run.linux.x64.checked.mch	439,635,202	${\color{green}-246,288}$
libraries.crossgen2.linux.x64.checked.mch	38,636,779	${\color{green}-4}$
libraries.pmi.linux.x64.checked.mch	59,927,414	${\color{green}-13}$
libraries_tests.run.linux.x64.Release.mch	328,845,908	${\color{green}-136,481}$
libraries_tests_no_tiered_compilation.run.linux.x64.Release.mch	129,732,524	${\color{green}-54,318}$
realworld.run.linux.x64.checked.mch	13,174,701	${\color{green}-1,026}$
smoke_tests.nativeaot.linux.x64.checked.mch	4,191,403	+0

MinOpts (${\color{green}-464,740}$ bytes)

Collection	Base size (bytes)	Diff size (bytes)
benchmarks.run.linux.x64.checked.mch	230,051	${\color{green}-5}$
benchmarks.run_pgo.linux.x64.checked.mch	18,706,882	${\color{green}-15,309}$
benchmarks.run_tiered.linux.x64.checked.mch	13,644,734	${\color{green}-11,296}$
coreclr_tests.run.linux.x64.checked.mch	310,584,524	${\color{green}-246,288}$
libraries.crossgen2.linux.x64.checked.mch	1,202	${\color{green}-4}$
libraries.pmi.linux.x64.checked.mch	112,870	${\color{green}-13}$
libraries_tests.run.linux.x64.Release.mch	183,459,665	${\color{green}-136,481}$
libraries_tests_no_tiered_compilation.run.linux.x64.Release.mch	10,657,089	${\color{green}-54,318}$
realworld.run.linux.x64.checked.mch	387,256	${\color{green}-1,026}$
smoke_tests.nativeaot.linux.x64.checked.mch	911	+0

Assembly diffs for osx/arm64 ran on windows/x64

Diffs are based on ${\color{orange}2,257,766}$ contexts (${\color{orange}928,286}$ MinOpts, ${\color{orange}1,329,480}$ FullOpts).

${\color{blue}MISSED}$ contexts: ${\color{blue}2,387 (0.11\%)}$

Overall (${\color{red}+150,496}$ bytes)

Collection	Base size (bytes)	Diff size (bytes)
benchmarks.run_pgo.osx.arm64.checked.mch	34,138,564	${\color{red}+1,480}$
benchmarks.run_tiered.osx.arm64.checked.mch	15,442,448	${\color{red}+316}$
coreclr_tests.run.osx.arm64.checked.mch	486,102,776	${\color{red}+148,116}$
libraries.crossgen2.osx.arm64.checked.mch	55,570,716	+0
libraries.pmi.osx.arm64.checked.mch	79,907,992	+0
libraries_tests.run.osx.arm64.Release.mch	314,762,740	${\color{red}+476}$
libraries_tests_no_tiered_compilation.run.osx.arm64.Release.mch	159,436,088	${\color{red}+108}$
realworld.run.osx.arm64.checked.mch	15,065,616	+0

MinOpts (${\color{red}+150,496}$ bytes)

Collection	Base size (bytes)	Diff size (bytes)
benchmarks.run_pgo.osx.arm64.checked.mch	16,484,956	${\color{red}+1,480}$
benchmarks.run_tiered.osx.arm64.checked.mch	11,499,424	${\color{red}+316}$
coreclr_tests.run.osx.arm64.checked.mch	332,741,972	${\color{red}+148,116}$
libraries.crossgen2.osx.arm64.checked.mch	1,628	+0
libraries.pmi.osx.arm64.checked.mch	121,128	+0
libraries_tests.run.osx.arm64.Release.mch	203,635,640	${\color{red}+476}$
libraries_tests_no_tiered_compilation.run.osx.arm64.Release.mch	13,136,596	${\color{red}+108}$
realworld.run.osx.arm64.checked.mch	568,396	+0

Assembly diffs for windows/arm64 ran on windows/x64

Diffs are based on ${\color{orange}2,326,811}$ contexts (${\color{orange}934,686}$ MinOpts, ${\color{orange}1,392,125}$ FullOpts).

${\color{blue}MISSED}$ contexts: ${\color{blue}2,495 (0.11\%)}$

Overall (${\color{red}+148,732}$ bytes)

Collection	Base size (bytes)	Diff size (bytes)
benchmarks.run_pgo.windows.arm64.checked.mch	45,619,340	${\color{red}+1,376}$
benchmarks.run_tiered.windows.arm64.checked.mch	15,264,644	${\color{red}+368}$
coreclr_tests.run.windows.arm64.checked.mch	496,117,708	${\color{red}+144,188}$
libraries.crossgen2.windows.arm64.checked.mch	58,913,600	+0
libraries.pmi.windows.arm64.checked.mch	79,525,232	+0
libraries_tests.run.windows.arm64.Release.mch	319,612,456	${\color{red}+2,688}$
libraries_tests_no_tiered_compilation.run.windows.arm64.Release.mch	169,012,984	${\color{red}+112}$
realworld.run.windows.arm64.checked.mch	15,917,176	+0

MinOpts (${\color{red}+148,732}$ bytes)

Collection	Base size (bytes)	Diff size (bytes)
benchmarks.run_pgo.windows.arm64.checked.mch	16,205,148	${\color{red}+1,376}$
benchmarks.run_tiered.windows.arm64.checked.mch	11,172,336	${\color{red}+368}$
coreclr_tests.run.windows.arm64.checked.mch	339,719,232	${\color{red}+144,188}$
libraries.crossgen2.windows.arm64.checked.mch	1,636	+0
libraries.pmi.windows.arm64.checked.mch	119,984	+0
libraries_tests.run.windows.arm64.Release.mch	203,904,856	${\color{red}+2,688}$
libraries_tests_no_tiered_compilation.run.windows.arm64.Release.mch	13,136,180	${\color{red}+112}$
realworld.run.windows.arm64.checked.mch	568,424	+0

Assembly diffs for windows/x64 ran on windows/x64

Diffs are based on ${\color{orange}2,495,240}$ contexts (${\color{orange}977,958}$ MinOpts, ${\color{orange}1,517,282}$ FullOpts).

${\color{blue}MISSED}$ contexts: ${\color{blue}3,546 (0.14\%)}$

Overall (${\color{green}-4,578}$ bytes)

Collection	Base size (bytes)	Diff size (bytes)
aspnet.run.windows.x64.checked.mch	42,070,891	${\color{green}-3,161}$
benchmarks.run.windows.x64.checked.mch	8,666,396	${\color{red}+2}$
benchmarks.run_pgo.windows.x64.checked.mch	35,349,995	${\color{green}-8,059}$
benchmarks.run_tiered.windows.x64.checked.mch	12,686,993	${\color{green}-10,190}$
coreclr_tests.run.windows.x64.checked.mch	393,136,663	${\color{red}+155,469}$
libraries.crossgen2.windows.x64.checked.mch	39,411,196	${\color{green}-2}$
libraries.pmi.windows.x64.checked.mch	61,131,115	${\color{red}+2}$
libraries_tests.run.windows.x64.Release.mch	279,572,690	${\color{green}-126,077}$
libraries_tests_no_tiered_compilation.run.windows.x64.Release.mch	133,740,839	${\color{green}-12,568}$
realworld.run.windows.x64.checked.mch	14,188,290	${\color{red}+6}$
smoke_tests.nativeaot.windows.x64.checked.mch	5,085,035	+0

MinOpts (${\color{green}-4,578}$ bytes)

Collection	Base size (bytes)	Diff size (bytes)
aspnet.run.windows.x64.checked.mch	14,658,725	${\color{green}-3,161}$
benchmarks.run.windows.x64.checked.mch	361	${\color{red}+2}$
benchmarks.run_pgo.windows.x64.checked.mch	14,238,248	${\color{green}-8,059}$
benchmarks.run_tiered.windows.x64.checked.mch	9,174,699	${\color{green}-10,190}$
coreclr_tests.run.windows.x64.checked.mch	273,514,441	${\color{red}+155,469}$
libraries.crossgen2.windows.x64.checked.mch	1,189	${\color{green}-2}$
libraries.pmi.windows.x64.checked.mch	113,519	${\color{red}+2}$
libraries_tests.run.windows.x64.Release.mch	175,492,237	${\color{green}-126,077}$
libraries_tests_no_tiered_compilation.run.windows.x64.Release.mch	10,810,850	${\color{green}-12,568}$
realworld.run.windows.x64.checked.mch	389,705	${\color{red}+6}$
smoke_tests.nativeaot.windows.x64.checked.mch	909	+0

Details here

Assembly diffs for linux/arm ran on windows/x86

Diffs are based on ${\color{orange}2,232,466}$ contexts (${\color{orange}824,262}$ MinOpts, ${\color{orange}1,408,204}$ FullOpts).

${\color{blue}MISSED}$ contexts: ${\color{blue}77,883 (3.37\%)}$

Overall (${\color{green}-852,700}$ bytes)

Collection	Base size (bytes)	Diff size (bytes)
benchmarks.run.linux.arm.checked.mch	14,172,892	${\color{green}-42}$
benchmarks.run_pgo.linux.arm.checked.mch	58,450,272	${\color{green}-48,252}$
benchmarks.run_tiered.linux.arm.checked.mch	17,364,030	${\color{green}-32,012}$
coreclr_tests.run.linux.arm.checked.mch	321,914,498	${\color{green}-199,178}$
libraries.crossgen2.linux.arm.checked.mch	37,768,318	+0
libraries.pmi.linux.arm.checked.mch	49,500,180	${\color{green}-280}$
libraries_tests.run.linux.arm.Release.mch	240,715,714	${\color{green}-533,380}$
libraries_tests_no_tiered_compilation.run.linux.arm.Release.mch	92,863,410	${\color{green}-24,910}$
realworld.run.linux.arm.checked.mch	13,608,080	${\color{green}-14,646}$

MinOpts (${\color{green}-852,700}$ bytes)

Collection	Base size (bytes)	Diff size (bytes)
benchmarks.run.linux.arm.checked.mch	334,862	${\color{green}-42}$
benchmarks.run_pgo.linux.arm.checked.mch	11,222,492	${\color{green}-48,252}$
benchmarks.run_tiered.linux.arm.checked.mch	7,194,320	${\color{green}-32,012}$
coreclr_tests.run.linux.arm.checked.mch	212,840,102	${\color{green}-199,178}$
libraries.crossgen2.linux.arm.checked.mch	1,230	+0
libraries.pmi.linux.arm.checked.mch	106,504	${\color{green}-280}$
libraries_tests.run.linux.arm.Release.mch	122,210,132	${\color{green}-533,380}$
libraries_tests_no_tiered_compilation.run.linux.arm.Release.mch	10,081,872	${\color{green}-24,910}$
realworld.run.linux.arm.checked.mch	449,296	${\color{green}-14,646}$

Assembly diffs for windows/x86 ran on windows/x86

Diffs are based on ${\color{orange}2,240,380}$ contexts (${\color{orange}789,389}$ MinOpts, ${\color{orange}1,450,991}$ FullOpts).

${\color{blue}MISSED}$ contexts: base: ${\color{blue}3,847 (0.17\%)}$, diff: ${\color{blue}49,158 (2.15\%)}$

Overall (${\color{red}+199,265}$ bytes)

Collection	Base size (bytes)	Diff size (bytes)
benchmarks.run.windows.x86.checked.mch	7,037,944	${\color{red}+2}$
benchmarks.run_pgo.windows.x86.checked.mch	43,076,375	${\color{red}+1,814}$
benchmarks.run_tiered.windows.x86.checked.mch	8,947,008	${\color{red}+247}$
coreclr_tests.run.windows.x86.checked.mch	305,219,044	${\color{red}+201,221}$
libraries.crossgen2.windows.x86.checked.mch	31,623,120	${\color{red}+3}$
libraries.pmi.windows.x86.checked.mch	48,807,890	${\color{green}-81}$
libraries_tests.run.windows.x86.Release.mch	162,974,883	${\color{green}-7,140}$
libraries_tests_no_tiered_compilation.run.windows.x86.Release.mch	100,565,179	${\color{red}+3,185}$
realworld.run.windows.x86.checked.mch	11,349,106	${\color{red}+14}$

MinOpts (${\color{red}+199,265}$ bytes)

Collection	Base size (bytes)	Diff size (bytes)
benchmarks.run.windows.x86.checked.mch	279	${\color{red}+2}$
benchmarks.run_pgo.windows.x86.checked.mch	5,966,425	${\color{red}+1,814}$
benchmarks.run_tiered.windows.x86.checked.mch	3,848,434	${\color{red}+247}$
coreclr_tests.run.windows.x86.checked.mch	198,156,671	${\color{red}+201,221}$
libraries.crossgen2.windows.x86.checked.mch	1,057	${\color{red}+3}$
libraries.pmi.windows.x86.checked.mch	95,314	${\color{green}-81}$
libraries_tests.run.windows.x86.Release.mch	77,656,212	${\color{green}-7,140}$
libraries_tests_no_tiered_compilation.run.windows.x86.Release.mch	7,273,446	${\color{red}+3,185}$
realworld.run.windows.x86.checked.mch	295,700	${\color{red}+14}$

Details here

Throughput diffs

Throughput diffs for linux/arm64 ran on windows/x64

Overall (${\color{green}-4.82\%}$ to ${\color{green}-0.04\%}$)

Collection	PDIFF
benchmarks.run.linux.arm64.checked.mch	${\color{green}-0.09\%}$
benchmarks.run_pgo.linux.arm64.checked.mch	${\color{green}-1.12\%}$
benchmarks.run_tiered.linux.arm64.checked.mch	${\color{green}-4.82\%}$
coreclr_tests.run.linux.arm64.checked.mch	${\color{green}-4.00\%}$
libraries.crossgen2.linux.arm64.checked.mch	${\color{green}-0.07\%}$
libraries.pmi.linux.arm64.checked.mch	${\color{green}-0.05\%}$
libraries_tests.run.linux.arm64.Release.mch	${\color{green}-2.41\%}$
libraries_tests_no_tiered_compilation.run.linux.arm64.Release.mch	${\color{green}-0.30\%}$
realworld.run.linux.arm64.checked.mch	${\color{green}-0.13\%}$
smoke_tests.nativeaot.linux.arm64.checked.mch	${\color{green}-0.04\%}$

MinOpts (${\color{green}-10.51\%}$ to ${\color{green}-6.50\%}$)

Collection	PDIFF
benchmarks.run.linux.arm64.checked.mch	${\color{green}-8.44\%}$
benchmarks.run_pgo.linux.arm64.checked.mch	${\color{green}-9.02\%}$
benchmarks.run_tiered.linux.arm64.checked.mch	${\color{green}-9.30\%}$
coreclr_tests.run.linux.arm64.checked.mch	${\color{green}-9.24\%}$
libraries.crossgen2.linux.arm64.checked.mch	${\color{green}-8.09\%}$
libraries.pmi.linux.arm64.checked.mch	${\color{green}-9.87\%}$
libraries_tests.run.linux.arm64.Release.mch	${\color{green}-9.06\%}$
libraries_tests_no_tiered_compilation.run.linux.arm64.Release.mch	${\color{green}-8.69\%}$
realworld.run.linux.arm64.checked.mch	${\color{green}-10.51\%}$
smoke_tests.nativeaot.linux.arm64.checked.mch	${\color{green}-6.50\%}$

FullOpts (${\color{green}-0.07\%}$ to ${\color{green}-0.03\%}$)

Collection	PDIFF
benchmarks.run.linux.arm64.checked.mch	${\color{green}-0.05\%}$
benchmarks.run_pgo.linux.arm64.checked.mch	${\color{green}-0.04\%}$
benchmarks.run_tiered.linux.arm64.checked.mch	${\color{green}-0.03\%}$
coreclr_tests.run.linux.arm64.checked.mch	${\color{green}-0.05\%}$
libraries.crossgen2.linux.arm64.checked.mch	${\color{green}-0.07\%}$
libraries.pmi.linux.arm64.checked.mch	${\color{green}-0.04\%}$
libraries_tests.run.linux.arm64.Release.mch	${\color{green}-0.04\%}$
libraries_tests_no_tiered_compilation.run.linux.arm64.Release.mch	${\color{green}-0.05\%}$
realworld.run.linux.arm64.checked.mch	${\color{green}-0.04\%}$
smoke_tests.nativeaot.linux.arm64.checked.mch	${\color{green}-0.04\%}$

Throughput diffs for linux/x64 ran on windows/x64

Overall (${\color{green}-2.72\%}$ to ${\color{green}-0.05\%}$)

Collection	PDIFF
benchmarks.run.linux.x64.checked.mch	${\color{green}-0.09\%}$
benchmarks.run_pgo.linux.x64.checked.mch	${\color{green}-0.56\%}$
benchmarks.run_tiered.linux.x64.checked.mch	${\color{green}-2.72\%}$
coreclr_tests.run.linux.x64.checked.mch	${\color{green}-2.70\%}$
libraries.crossgen2.linux.x64.checked.mch	${\color{green}-0.08\%}$
libraries.pmi.linux.x64.checked.mch	${\color{green}-0.06\%}$
libraries_tests.run.linux.x64.Release.mch	${\color{green}-1.34\%}$
libraries_tests_no_tiered_compilation.run.linux.x64.Release.mch	${\color{green}-0.21\%}$
realworld.run.linux.x64.checked.mch	${\color{green}-0.10\%}$
smoke_tests.nativeaot.linux.x64.checked.mch	${\color{green}-0.05\%}$

MinOpts (${\color{green}-6.56\%}$ to ${\color{green}-4.49\%}$)

Collection	PDIFF
benchmarks.run.linux.x64.checked.mch	${\color{green}-5.03\%}$
benchmarks.run_pgo.linux.x64.checked.mch	${\color{green}-5.63\%}$
benchmarks.run_tiered.linux.x64.checked.mch	${\color{green}-5.71\%}$
coreclr_tests.run.linux.x64.checked.mch	${\color{green}-6.24\%}$
libraries.crossgen2.linux.x64.checked.mch	${\color{green}-4.49\%}$
libraries.pmi.linux.x64.checked.mch	${\color{green}-5.88\%}$
libraries_tests.run.linux.x64.Release.mch	${\color{green}-5.66\%}$
libraries_tests_no_tiered_compilation.run.linux.x64.Release.mch	${\color{green}-5.15\%}$
realworld.run.linux.x64.checked.mch	${\color{green}-6.56\%}$
smoke_tests.nativeaot.linux.x64.checked.mch	${\color{green}-4.77\%}$

FullOpts (${\color{green}-0.08\%}$ to ${\color{green}-0.04\%}$)

Collection	PDIFF
benchmarks.run.linux.x64.checked.mch	${\color{green}-0.06\%}$
benchmarks.run_pgo.linux.x64.checked.mch	${\color{green}-0.04\%}$
benchmarks.run_tiered.linux.x64.checked.mch	${\color{green}-0.04\%}$
coreclr_tests.run.linux.x64.checked.mch	${\color{green}-0.06\%}$
libraries.crossgen2.linux.x64.checked.mch	${\color{green}-0.08\%}$
libraries.pmi.linux.x64.checked.mch	${\color{green}-0.06\%}$
libraries_tests.run.linux.x64.Release.mch	${\color{green}-0.05\%}$
libraries_tests_no_tiered_compilation.run.linux.x64.Release.mch	${\color{green}-0.07\%}$
realworld.run.linux.x64.checked.mch	${\color{green}-0.06\%}$
smoke_tests.nativeaot.linux.x64.checked.mch	${\color{green}-0.05\%}$

Throughput diffs for osx/arm64 ran on windows/x64

Overall (${\color{green}-4.32\%}$ to ${\color{green}-0.04\%}$)

Collection	PDIFF
benchmarks.run.osx.arm64.checked.mch	${\color{green}-0.04\%}$
benchmarks.run_pgo.osx.arm64.checked.mch	${\color{green}-1.94\%}$
benchmarks.run_tiered.osx.arm64.checked.mch	${\color{green}-4.32\%}$
coreclr_tests.run.osx.arm64.checked.mch	${\color{green}-3.96\%}$
libraries.crossgen2.osx.arm64.checked.mch	${\color{green}-0.07\%}$
libraries.pmi.osx.arm64.checked.mch	${\color{green}-0.05\%}$
libraries_tests.run.osx.arm64.Release.mch	${\color{green}-3.05\%}$
libraries_tests_no_tiered_compilation.run.osx.arm64.Release.mch	${\color{green}-0.30\%}$
realworld.run.osx.arm64.checked.mch	${\color{green}-0.13\%}$

MinOpts (${\color{green}-10.55\%}$ to ${\color{green}-7.99\%}$)

Collection	PDIFF
benchmarks.run.osx.arm64.checked.mch	${\color{green}-9.05\%}$
benchmarks.run_pgo.osx.arm64.checked.mch	${\color{green}-9.34\%}$
benchmarks.run_tiered.osx.arm64.checked.mch	${\color{green}-9.62\%}$
coreclr_tests.run.osx.arm64.checked.mch	${\color{green}-9.13\%}$
libraries.crossgen2.osx.arm64.checked.mch	${\color{green}-7.99\%}$
libraries.pmi.osx.arm64.checked.mch	${\color{green}-9.86\%}$
libraries_tests.run.osx.arm64.Release.mch	${\color{green}-9.09\%}$
libraries_tests_no_tiered_compilation.run.osx.arm64.Release.mch	${\color{green}-8.68\%}$
realworld.run.osx.arm64.checked.mch	${\color{green}-10.55\%}$

FullOpts (${\color{green}-0.07\%}$ to ${\color{green}-0.03\%}$)

Collection	PDIFF
benchmarks.run.osx.arm64.checked.mch	${\color{green}-0.04\%}$
benchmarks.run_pgo.osx.arm64.checked.mch	${\color{green}-0.03\%}$
benchmarks.run_tiered.osx.arm64.checked.mch	${\color{green}-0.03\%}$
coreclr_tests.run.osx.arm64.checked.mch	${\color{green}-0.06\%}$
libraries.crossgen2.osx.arm64.checked.mch	${\color{green}-0.07\%}$
libraries.pmi.osx.arm64.checked.mch	${\color{green}-0.04\%}$
libraries_tests.run.osx.arm64.Release.mch	${\color{green}-0.04\%}$
libraries_tests_no_tiered_compilation.run.osx.arm64.Release.mch	${\color{green}-0.05\%}$
realworld.run.osx.arm64.checked.mch	${\color{green}-0.04\%}$

Throughput diffs for windows/arm64 ran on windows/x64

Overall (${\color{green}-4.16\%}$ to ${\color{green}-0.04\%}$)

Collection	PDIFF
benchmarks.run.windows.arm64.checked.mch	${\color{green}-0.04\%}$
benchmarks.run_pgo.windows.arm64.checked.mch	${\color{green}-1.27\%}$
benchmarks.run_tiered.windows.arm64.checked.mch	${\color{green}-4.16\%}$
coreclr_tests.run.windows.arm64.checked.mch	${\color{green}-4.01\%}$
libraries.crossgen2.windows.arm64.checked.mch	${\color{green}-0.07\%}$
libraries.pmi.windows.arm64.checked.mch	${\color{green}-0.05\%}$
libraries_tests.run.windows.arm64.Release.mch	${\color{green}-2.98\%}$
libraries_tests_no_tiered_compilation.run.windows.arm64.Release.mch	${\color{green}-0.28\%}$
realworld.run.windows.arm64.checked.mch	${\color{green}-0.13\%}$
smoke_tests.nativeaot.windows.arm64.checked.mch	${\color{green}-0.04\%}$

MinOpts (${\color{green}-10.55\%}$ to ${\color{green}-6.32\%}$)

Collection	PDIFF
benchmarks.run.windows.arm64.checked.mch	${\color{green}-9.05\%}$
benchmarks.run_pgo.windows.arm64.checked.mch	${\color{green}-9.25\%}$
benchmarks.run_tiered.windows.arm64.checked.mch	${\color{green}-9.57\%}$
coreclr_tests.run.windows.arm64.checked.mch	${\color{green}-9.25\%}$
libraries.crossgen2.windows.arm64.checked.mch	${\color{green}-8.09\%}$
libraries.pmi.windows.arm64.checked.mch	${\color{green}-9.86\%}$
libraries_tests.run.windows.arm64.Release.mch	${\color{green}-9.07\%}$
libraries_tests_no_tiered_compilation.run.windows.arm64.Release.mch	${\color{green}-8.68\%}$
realworld.run.windows.arm64.checked.mch	${\color{green}-10.55\%}$
smoke_tests.nativeaot.windows.arm64.checked.mch	${\color{green}-6.32\%}$

FullOpts (${\color{green}-0.07\%}$ to ${\color{green}-0.03\%}$)

Collection	PDIFF
benchmarks.run.windows.arm64.checked.mch	${\color{green}-0.04\%}$
benchmarks.run_pgo.windows.arm64.checked.mch	${\color{green}-0.03\%}$
benchmarks.run_tiered.windows.arm64.checked.mch	${\color{green}-0.03\%}$
coreclr_tests.run.windows.arm64.checked.mch	${\color{green}-0.06\%}$
libraries.crossgen2.windows.arm64.checked.mch	${\color{green}-0.07\%}$
libraries.pmi.windows.arm64.checked.mch	${\color{green}-0.04\%}$
libraries_tests.run.windows.arm64.Release.mch	${\color{green}-0.04\%}$
libraries_tests_no_tiered_compilation.run.windows.arm64.Release.mch	${\color{green}-0.05\%}$
realworld.run.windows.arm64.checked.mch	${\color{green}-0.04\%}$
smoke_tests.nativeaot.windows.arm64.checked.mch	${\color{green}-0.04\%}$

Throughput diffs for windows/x64 ran on windows/x64

Overall (${\color{green}-2.54\%}$ to ${\color{green}-0.08\%}$)

Collection	PDIFF
aspnet.run.windows.x64.checked.mch	${\color{green}-0.66\%}$
benchmarks.run.windows.x64.checked.mch	${\color{green}-0.08\%}$
benchmarks.run_pgo.windows.x64.checked.mch	${\color{green}-0.83\%}$
benchmarks.run_tiered.windows.x64.checked.mch	${\color{green}-2.22\%}$
coreclr_tests.run.windows.x64.checked.mch	${\color{green}-2.54\%}$
libraries.crossgen2.windows.x64.checked.mch	${\color{green}-0.12\%}$
libraries.pmi.windows.x64.checked.mch	${\color{green}-0.10\%}$
libraries_tests.run.windows.x64.Release.mch	${\color{green}-1.64\%}$
libraries_tests_no_tiered_compilation.run.windows.x64.Release.mch	${\color{green}-0.25\%}$
realworld.run.windows.x64.checked.mch	${\color{green}-0.13\%}$
smoke_tests.nativeaot.windows.x64.checked.mch	${\color{green}-0.08\%}$

MinOpts (${\color{green}-6.49\%}$ to ${\color{green}-4.45\%}$)

Collection	PDIFF
aspnet.run.windows.x64.checked.mch	${\color{green}-5.50\%}$
benchmarks.run.windows.x64.checked.mch	${\color{green}-4.89\%}$
benchmarks.run_pgo.windows.x64.checked.mch	${\color{green}-5.70\%}$
benchmarks.run_tiered.windows.x64.checked.mch	${\color{green}-5.72\%}$
coreclr_tests.run.windows.x64.checked.mch	${\color{green}-5.95\%}$
libraries.crossgen2.windows.x64.checked.mch	${\color{green}-4.45\%}$
libraries.pmi.windows.x64.checked.mch	${\color{green}-5.72\%}$
libraries_tests.run.windows.x64.Release.mch	${\color{green}-5.74\%}$
libraries_tests_no_tiered_compilation.run.windows.x64.Release.mch	${\color{green}-5.52\%}$
realworld.run.windows.x64.checked.mch	${\color{green}-6.49\%}$
smoke_tests.nativeaot.windows.x64.checked.mch	${\color{green}-4.70\%}$

FullOpts (${\color{green}-0.12\%}$ to ${\color{green}-0.07\%}$)

Collection	PDIFF
aspnet.run.windows.x64.checked.mch	${\color{green}-0.07\%}$
benchmarks.run.windows.x64.checked.mch	${\color{green}-0.08\%}$
benchmarks.run_pgo.windows.x64.checked.mch	${\color{green}-0.07\%}$
benchmarks.run_tiered.windows.x64.checked.mch	${\color{green}-0.07\%}$
coreclr_tests.run.windows.x64.checked.mch	${\color{green}-0.11\%}$
libraries.crossgen2.windows.x64.checked.mch	${\color{green}-0.12\%}$
libraries.pmi.windows.x64.checked.mch	${\color{green}-0.09\%}$
libraries_tests.run.windows.x64.Release.mch	${\color{green}-0.08\%}$
libraries_tests_no_tiered_compilation.run.windows.x64.Release.mch	${\color{green}-0.11\%}$
realworld.run.windows.x64.checked.mch	${\color{green}-0.10\%}$
smoke_tests.nativeaot.windows.x64.checked.mch	${\color{green}-0.08\%}$

Details here

Throughput diffs for linux/arm64 ran on linux/x64

Overall (${\color{green}-5.32\%}$ to ${\color{red}+0.04\%}$)

Collection	PDIFF
libraries.crossgen2.linux.arm64.checked.mch	${\color{red}+0.04\%}$
libraries_tests.run.linux.arm64.Release.mch	${\color{green}-2.61\%}$
realworld.run.linux.arm64.checked.mch	${\color{green}-0.07\%}$
coreclr_tests.run.linux.arm64.checked.mch	${\color{green}-4.33\%}$
benchmarks.run_pgo.linux.arm64.checked.mch	${\color{green}-1.17\%}$
libraries.pmi.linux.arm64.checked.mch	${\color{red}+0.03\%}$
benchmarks.run_tiered.linux.arm64.checked.mch	${\color{green}-5.32\%}$
libraries_tests_no_tiered_compilation.run.linux.arm64.Release.mch	${\color{green}-0.24\%}$
smoke_tests.nativeaot.linux.arm64.checked.mch	${\color{red}+0.04\%}$

MinOpts (${\color{green}-11.63\%}$ to ${\color{green}-6.46\%}$)

Collection	PDIFF
libraries.crossgen2.linux.arm64.checked.mch	${\color{green}-9.09\%}$
libraries_tests.run.linux.arm64.Release.mch	${\color{green}-9.98\%}$
realworld.run.linux.arm64.checked.mch	${\color{green}-11.63\%}$
coreclr_tests.run.linux.arm64.checked.mch	${\color{green}-9.71\%}$
benchmarks.run_pgo.linux.arm64.checked.mch	${\color{green}-9.97\%}$
libraries.pmi.linux.arm64.checked.mch	${\color{green}-11.20\%}$
benchmarks.run.linux.arm64.checked.mch	${\color{green}-9.20\%}$
benchmarks.run_tiered.linux.arm64.checked.mch	${\color{green}-10.35\%}$
libraries_tests_no_tiered_compilation.run.linux.arm64.Release.mch	${\color{green}-10.06\%}$
smoke_tests.nativeaot.linux.arm64.checked.mch	${\color{green}-6.46\%}$

FullOpts (${\color{red}+0.02\%}$ to ${\color{red}+0.05\%}$)

Collection	PDIFF
libraries.crossgen2.linux.arm64.checked.mch	${\color{red}+0.04\%}$
libraries_tests.run.linux.arm64.Release.mch	${\color{red}+0.03\%}$
realworld.run.linux.arm64.checked.mch	${\color{red}+0.03\%}$
coreclr_tests.run.linux.arm64.checked.mch	${\color{red}+0.05\%}$
benchmarks.run_pgo.linux.arm64.checked.mch	${\color{red}+0.03\%}$
libraries.pmi.linux.arm64.checked.mch	${\color{red}+0.03\%}$
benchmarks.run.linux.arm64.checked.mch	${\color{red}+0.05\%}$
benchmarks.run_tiered.linux.arm64.checked.mch	${\color{red}+0.02\%}$
libraries_tests_no_tiered_compilation.run.linux.arm64.Release.mch	${\color{red}+0.05\%}$
smoke_tests.nativeaot.linux.arm64.checked.mch	${\color{red}+0.04\%}$

Throughput diffs for linux/x64 ran on linux/x64

Overall (${\color{green}-2.87\%}$ to ${\color{green}-0.15\%}$)

Collection	PDIFF
benchmarks.run_pgo.linux.x64.checked.mch	${\color{green}-0.66\%}$
benchmarks.run_tiered.linux.x64.checked.mch	${\color{green}-2.87\%}$
libraries.crossgen2.linux.x64.checked.mch	${\color{green}-0.22\%}$
libraries.pmi.linux.x64.checked.mch	${\color{green}-0.17\%}$
libraries_tests_no_tiered_compilation.run.linux.x64.Release.mch	${\color{green}-0.34\%}$
realworld.run.linux.x64.checked.mch	${\color{green}-0.20\%}$
smoke_tests.nativeaot.linux.x64.checked.mch	${\color{green}-0.15\%}$
benchmarks.run.linux.x64.checked.mch	${\color{green}-0.20\%}$
libraries_tests.run.linux.x64.Release.mch	${\color{green}-1.45\%}$
coreclr_tests.run.linux.x64.checked.mch	${\color{green}-2.80\%}$

MinOpts (${\color{green}-6.69\%}$ to ${\color{green}-4.65\%}$)

Collection	PDIFF
benchmarks.run_pgo.linux.x64.checked.mch	${\color{green}-5.91\%}$
benchmarks.run_tiered.linux.x64.checked.mch	${\color{green}-5.96\%}$
libraries.crossgen2.linux.x64.checked.mch	${\color{green}-4.86\%}$
libraries.pmi.linux.x64.checked.mch	${\color{green}-6.32\%}$
libraries_tests_no_tiered_compilation.run.linux.x64.Release.mch	${\color{green}-5.60\%}$
realworld.run.linux.x64.checked.mch	${\color{green}-6.69\%}$
smoke_tests.nativeaot.linux.x64.checked.mch	${\color{green}-4.65\%}$
benchmarks.run.linux.x64.checked.mch	${\color{green}-5.12\%}$
libraries_tests.run.linux.x64.Release.mch	${\color{green}-5.81\%}$
coreclr_tests.run.linux.x64.checked.mch	${\color{green}-6.04\%}$

FullOpts (${\color{green}-0.22\%}$ to ${\color{green}-0.12\%}$)

Collection	PDIFF
benchmarks.run_pgo.linux.x64.checked.mch	${\color{green}-0.13\%}$
benchmarks.run_tiered.linux.x64.checked.mch	${\color{green}-0.12\%}$
libraries.crossgen2.linux.x64.checked.mch	${\color{green}-0.22\%}$
libraries.pmi.linux.x64.checked.mch	${\color{green}-0.16\%}$
libraries_tests_no_tiered_compilation.run.linux.x64.Release.mch	${\color{green}-0.19\%}$
realworld.run.linux.x64.checked.mch	${\color{green}-0.16\%}$
smoke_tests.nativeaot.linux.x64.checked.mch	${\color{green}-0.15\%}$
benchmarks.run.linux.x64.checked.mch	${\color{green}-0.17\%}$
libraries_tests.run.linux.x64.Release.mch	${\color{green}-0.14\%}$
coreclr_tests.run.linux.x64.checked.mch	${\color{green}-0.18\%}$

Details here

Throughput diffs for linux/arm ran on windows/x86

Overall (${\color{green}-4.15\%}$ to ${\color{green}-0.08\%}$)

Collection	PDIFF
benchmarks.run.linux.arm.checked.mch	${\color{green}-0.18\%}$
benchmarks.run_pgo.linux.arm.checked.mch	${\color{green}-0.72\%}$
benchmarks.run_tiered.linux.arm.checked.mch	${\color{green}-2.03\%}$
coreclr_tests.run.linux.arm.checked.mch	${\color{green}-4.15\%}$
libraries.crossgen2.linux.arm.checked.mch	${\color{green}-0.15\%}$
libraries.pmi.linux.arm.checked.mch	${\color{green}-0.08\%}$
libraries_tests.run.linux.arm.Release.mch	${\color{green}-2.51\%}$
libraries_tests_no_tiered_compilation.run.linux.arm.Release.mch	${\color{green}-0.47\%}$
realworld.run.linux.arm.checked.mch	${\color{green}-0.15\%}$

MinOpts (${\color{green}-11.56\%}$ to ${\color{green}-9.30\%}$)

Collection	PDIFF
benchmarks.run.linux.arm.checked.mch	${\color{green}-9.64\%}$
benchmarks.run_pgo.linux.arm.checked.mch	${\color{green}-10.51\%}$
benchmarks.run_tiered.linux.arm.checked.mch	${\color{green}-10.52\%}$
coreclr_tests.run.linux.arm.checked.mch	${\color{green}-9.88\%}$
libraries.crossgen2.linux.arm.checked.mch	${\color{green}-9.83\%}$
libraries.pmi.linux.arm.checked.mch	${\color{green}-9.30\%}$
libraries_tests.run.linux.arm.Release.mch	${\color{green}-10.61\%}$
libraries_tests_no_tiered_compilation.run.linux.arm.Release.mch	${\color{green}-9.92\%}$
realworld.run.linux.arm.checked.mch	${\color{green}-11.56\%}$

FullOpts (${\color{green}-0.15\%}$ to ${\color{green}-0.05\%}$)

Collection	PDIFF
benchmarks.run.linux.arm.checked.mch	${\color{green}-0.09\%}$
benchmarks.run_pgo.linux.arm.checked.mch	${\color{green}-0.05\%}$
benchmarks.run_tiered.linux.arm.checked.mch	${\color{green}-0.09\%}$
coreclr_tests.run.linux.arm.checked.mch	${\color{green}-0.11\%}$
libraries.crossgen2.linux.arm.checked.mch	${\color{green}-0.15\%}$
libraries.pmi.linux.arm.checked.mch	${\color{green}-0.07\%}$
libraries_tests.run.linux.arm.Release.mch	${\color{green}-0.05\%}$
libraries_tests_no_tiered_compilation.run.linux.arm.Release.mch	${\color{green}-0.09\%}$
realworld.run.linux.arm.checked.mch	${\color{green}-0.06\%}$

Throughput diffs for windows/x86 ran on windows/x86

Overall (${\color{green}-1.68\%}$ to ${\color{red}+0.02\%}$)

Collection	PDIFF
benchmarks.run.windows.x86.checked.mch	${\color{red}+0.02\%}$
benchmarks.run_pgo.windows.x86.checked.mch	${\color{green}-0.12\%}$
benchmarks.run_tiered.windows.x86.checked.mch	${\color{green}-0.59\%}$
coreclr_tests.run.windows.x86.checked.mch	${\color{green}-1.68\%}$
libraries.crossgen2.windows.x86.checked.mch	${\color{red}+0.01\%}$
libraries.pmi.windows.x86.checked.mch	${\color{red}+0.02\%}$
libraries_tests.run.windows.x86.Release.mch	${\color{green}-0.74\%}$
libraries_tests_no_tiered_compilation.run.windows.x86.Release.mch	${\color{green}-0.07\%}$
realworld.run.windows.x86.checked.mch	${\color{green}-0.01\%}$

MinOpts (${\color{green}-5.03\%}$ to ${\color{green}-3.05\%}$)

Collection	PDIFF
benchmarks.run.windows.x86.checked.mch	${\color{green}-3.15\%}$
benchmarks.run_pgo.windows.x86.checked.mch	${\color{green}-3.99\%}$
benchmarks.run_tiered.windows.x86.checked.mch	${\color{green}-3.90\%}$
coreclr_tests.run.windows.x86.checked.mch	${\color{green}-5.03\%}$
libraries.crossgen2.windows.x86.checked.mch	${\color{green}-3.05\%}$
libraries.pmi.windows.x86.checked.mch	${\color{green}-4.50\%}$
libraries_tests.run.windows.x86.Release.mch	${\color{green}-4.24\%}$
libraries_tests_no_tiered_compilation.run.windows.x86.Release.mch	${\color{green}-3.69\%}$
realworld.run.windows.x86.checked.mch	${\color{green}-4.85\%}$

FullOpts (${\color{red}+0.01\%}$ to ${\color{red}+0.02\%}$)

Collection	PDIFF
benchmarks.run.windows.x86.checked.mch	${\color{red}+0.02\%}$
benchmarks.run_pgo.windows.x86.checked.mch	${\color{red}+0.02\%}$
benchmarks.run_tiered.windows.x86.checked.mch	${\color{red}+0.02\%}$
coreclr_tests.run.windows.x86.checked.mch	${\color{red}+0.01\%}$
libraries.crossgen2.windows.x86.checked.mch	${\color{red}+0.01\%}$
libraries.pmi.windows.x86.checked.mch	${\color{red}+0.02\%}$
libraries_tests.run.windows.x86.Release.mch	${\color{red}+0.02\%}$
libraries_tests_no_tiered_compilation.run.windows.x86.Release.mch	${\color{red}+0.01\%}$
realworld.run.windows.x86.checked.mch	${\color{red}+0.02\%}$

Details here

kunalspathak · 2024-01-22T22:59:12Z

@BruceForstall @jakobbotsch - can you please take another look? Thanks!

src/coreclr/jit/lsra.cpp

jakobbotsch · 2024-01-23T12:05:06Z

src/coreclr/jit/lsra.cpp

+#ifdef DEBUG
+                // Validate the current state just after we've freed the registers. This ensures that any pending
+                // freed registers will have had their state updated to reflect the intervals they were holding.
+                for (regNumber reg = REG_FIRST; reg < AVAILABLE_REG_COUNT; reg = REG_NEXT(reg))


Is it possible/straightforward to factor this validation into a method and use it from both allocateRegisters and allocateRegistersMinimal?

A related question -- what are the fundamental difference between the two, except the reduced set of heuristics used in the minimal version?

jakobbotsch

I've skimmed the code. I'm assuming allocateRegistersMinimal was copied mostly from allocateRegisters. I had a question on what the differences between them is, apart from the use of a reduced set of heuristics -- I'm assuming it wouldn't be straightforward to unify them to avoid code duplication? (For example, by the use of another template parameter)

The trade offs in CQ vs TP look just fine to me. We should definitely be taking this kind of tradeoff in MinOpts, and the TP improvements are amazing.

kunalspathak · 2024-01-25T22:43:21Z

I'm assuming allocateRegistersMinimal was copied mostly from allocateRegisters.

Pretty much. To easily spot the difference, I extracted out here:

I'm assuming it wouldn't be straightforward to unify them to avoid code duplication? (For example, by the use of another template parameter)

Correct, that was the direction I started with, but I quickly noticed that it scatters the templated parameter everywhere and very hard to see what gets executed for MinOpts. In future, it will be good to focus on *minimal() APIs to further simplify the algorithm.

apart from the use of a reduced set of heuristics

Yes, that's the main change, and then removing the code for things that never happen for MinOpts, like we don't see RefTypeExpUse or RefTypeDummyDef.

ryujit-bot · 2024-01-26T06:29:12Z

Diff results for #96386

Assembly diffs

Assembly diffs for linux/arm64 ran on windows/x64

Diffs are based on 2,498,771 contexts (1,011,240 MinOpts, 1,487,531 FullOpts).

MISSED contexts: 6,580 (0.26%)

Overall (+147,632 bytes)

Collection	Base size (bytes)	Diff size (bytes)
benchmarks.run.linux.arm64.checked.mch	15,908,516	+0
benchmarks.run_pgo.linux.arm64.checked.mch	78,672,796	+1,492
benchmarks.run_tiered.linux.arm64.checked.mch	29,592,944	+620
coreclr_tests.run.linux.arm64.checked.mch	508,895,360	+143,476
libraries.crossgen2.linux.arm64.checked.mch	55,781,064	+0
libraries.pmi.linux.arm64.checked.mch	74,608,176	+0
libraries_tests.run.linux.arm64.Release.mch	383,838,152	+1,936
libraries_tests_no_tiered_compilation.run.linux.arm64.Release.mch	163,129,488	+108
realworld.run.linux.arm64.checked.mch	15,368,676	+0

MinOpts (+147,632 bytes)

Collection	Base size (bytes)	Diff size (bytes)
benchmarks.run.linux.arm64.checked.mch	346,256	+0
benchmarks.run_pgo.linux.arm64.checked.mch	23,982,040	+1,492
benchmarks.run_tiered.linux.arm64.checked.mch	24,621,220	+620
coreclr_tests.run.linux.arm64.checked.mch	348,829,192	+143,476
libraries.crossgen2.linux.arm64.checked.mch	1,636	+0
libraries.pmi.linux.arm64.checked.mch	119,984	+0
libraries_tests.run.linux.arm64.Release.mch	215,421,676	+1,936
libraries_tests_no_tiered_compilation.run.linux.arm64.Release.mch	13,479,996	+108
realworld.run.linux.arm64.checked.mch	574,152	+0

Assembly diffs for linux/x64 ran on windows/x64

Diffs are based on 2,505,340 contexts (977,766 MinOpts, 1,527,574 FullOpts).

MISSED contexts: 6,922 (0.28%)

Overall (-485,333 bytes)

Collection	Base size (bytes)	Diff size (bytes)
benchmarks.run.linux.x64.checked.mch	16,148,228	-5
benchmarks.run_pgo.linux.x64.checked.mch	70,126,965	-18,248
benchmarks.run_tiered.linux.x64.checked.mch	15,918,176	-11,382
coreclr_tests.run.linux.x64.checked.mch	402,653,853	-237,129
libraries.crossgen2.linux.x64.checked.mch	38,689,357	-4
libraries.pmi.linux.x64.checked.mch	58,919,195	-13
libraries_tests.run.linux.x64.Release.mch	339,819,743	-163,282
libraries_tests_no_tiered_compilation.run.linux.x64.Release.mch	130,995,704	-54,244
realworld.run.linux.x64.checked.mch	12,767,093	-1,026
smoke_tests.nativeaot.linux.x64.checked.mch	4,192,959	+0

MinOpts (-485,333 bytes)

Collection	Base size (bytes)	Diff size (bytes)
benchmarks.run.linux.x64.checked.mch	264,173	-5
benchmarks.run_pgo.linux.x64.checked.mch	21,310,577	-18,248
benchmarks.run_tiered.linux.x64.checked.mch	12,281,125	-11,382
coreclr_tests.run.linux.x64.checked.mch	279,538,531	-237,129
libraries.crossgen2.linux.x64.checked.mch	1,202	-4
libraries.pmi.linux.x64.checked.mch	112,870	-13
libraries_tests.run.linux.x64.Release.mch	183,511,120	-163,282
libraries_tests_no_tiered_compilation.run.linux.x64.Release.mch	10,658,448	-54,244
realworld.run.linux.x64.checked.mch	386,910	-1,026
smoke_tests.nativeaot.linux.x64.checked.mch	949	+0

Assembly diffs for osx/arm64 ran on windows/x64

Diffs are based on 2,229,922 contexts (927,360 MinOpts, 1,302,562 FullOpts).

MISSED contexts: 6,095 (0.27%)

Overall (+150,252 bytes)

Collection	Base size (bytes)	Diff size (bytes)
benchmarks.run_pgo.osx.arm64.checked.mch	34,339,712	+952
benchmarks.run_tiered.osx.arm64.checked.mch	15,547,160	+328
coreclr_tests.run.osx.arm64.checked.mch	483,275,704	+147,732
libraries.crossgen2.osx.arm64.checked.mch	55,660,852	+0
libraries.pmi.osx.arm64.checked.mch	78,453,800	+0
libraries_tests.run.osx.arm64.Release.mch	311,227,792	+1,132
libraries_tests_no_tiered_compilation.run.osx.arm64.Release.mch	161,150,176	+108
realworld.run.osx.arm64.checked.mch	14,659,812	+0

MinOpts (+150,252 bytes)

Collection	Base size (bytes)	Diff size (bytes)
benchmarks.run_pgo.osx.arm64.checked.mch	16,407,348	+952
benchmarks.run_tiered.osx.arm64.checked.mch	11,499,248	+328
coreclr_tests.run.osx.arm64.checked.mch	331,292,952	+147,732
libraries.crossgen2.osx.arm64.checked.mch	1,628	+0
libraries.pmi.osx.arm64.checked.mch	121,128	+0
libraries_tests.run.osx.arm64.Release.mch	201,319,184	+1,132
libraries_tests_no_tiered_compilation.run.osx.arm64.Release.mch	13,136,524	+108
realworld.run.osx.arm64.checked.mch	563,956	+0

Assembly diffs for windows/arm64 ran on windows/x64

Diffs are based on 2,308,445 contexts (929,692 MinOpts, 1,378,753 FullOpts).

MISSED contexts: 6,353 (0.27%)

Overall (+148,112 bytes)

Collection	Base size (bytes)	Diff size (bytes)
benchmarks.run_pgo.windows.arm64.checked.mch	46,475,100	+988
benchmarks.run_tiered.windows.arm64.checked.mch	15,251,468	+384
coreclr_tests.run.windows.arm64.checked.mch	494,449,448	+143,628
libraries.crossgen2.windows.arm64.checked.mch	59,008,088	+0
libraries.pmi.windows.arm64.checked.mch	78,059,240	+0
libraries_tests.run.windows.arm64.Release.mch	308,519,220	+3,000
libraries_tests_no_tiered_compilation.run.windows.arm64.Release.mch	169,413,408	+112
realworld.run.windows.arm64.checked.mch	15,463,108	+0

MinOpts (+148,112 bytes)

Collection	Base size (bytes)	Diff size (bytes)
benchmarks.run_pgo.windows.arm64.checked.mch	16,058,416	+988
benchmarks.run_tiered.windows.arm64.checked.mch	11,172,728	+384
coreclr_tests.run.windows.arm64.checked.mch	338,587,172	+143,628
libraries.crossgen2.windows.arm64.checked.mch	1,636	+0
libraries.pmi.windows.arm64.checked.mch	119,984	+0
libraries_tests.run.windows.arm64.Release.mch	201,005,952	+3,000
libraries_tests_no_tiered_compilation.run.windows.arm64.Release.mch	13,136,540	+112
realworld.run.windows.arm64.checked.mch	563,980	+0

Assembly diffs for windows/x64 ran on windows/x64

Diffs are based on 2,366,385 contexts (928,740 MinOpts, 1,437,645 FullOpts).

MISSED contexts: 6,816 (0.29%)

Overall (-62,154 bytes)

Collection	Base size (bytes)	Diff size (bytes)
benchmarks.run.windows.x64.checked.mch	8,536,477	+2
benchmarks.run_pgo.windows.x64.checked.mch	35,464,350	-13,500
benchmarks.run_tiered.windows.x64.checked.mch	12,551,598	-12,300
coreclr_tests.run.windows.x64.checked.mch	392,239,828	+139,386
libraries.crossgen2.windows.x64.checked.mch	39,446,742	-2
libraries.pmi.windows.x64.checked.mch	60,117,403	+2
libraries_tests.run.windows.x64.Release.mch	276,943,385	-163,110
libraries_tests_no_tiered_compilation.run.windows.x64.Release.mch	135,762,558	-12,638
realworld.run.windows.x64.checked.mch	13,729,153	+6
smoke_tests.nativeaot.windows.x64.checked.mch	5,087,013	+0

MinOpts (-62,154 bytes)

Collection	Base size (bytes)	Diff size (bytes)
benchmarks.run.windows.x64.checked.mch	361	+2
benchmarks.run_pgo.windows.x64.checked.mch	13,997,605	-13,500
benchmarks.run_tiered.windows.x64.checked.mch	9,105,849	-12,300
coreclr_tests.run.windows.x64.checked.mch	272,531,567	+139,386
libraries.crossgen2.windows.x64.checked.mch	1,189	-2
libraries.pmi.windows.x64.checked.mch	113,519	+2
libraries_tests.run.windows.x64.Release.mch	173,720,676	-163,110
libraries_tests_no_tiered_compilation.run.windows.x64.Release.mch	10,806,556	-12,638
realworld.run.windows.x64.checked.mch	386,603	+6
smoke_tests.nativeaot.windows.x64.checked.mch	947	+0

Details here

Assembly diffs for linux/arm ran on windows/x86

Diffs are based on 2,230,528 contexts (825,130 MinOpts, 1,405,398 FullOpts).

MISSED contexts: 77,529 (3.36%)

Overall (-868,478 bytes)

Collection	Base size (bytes)	Diff size (bytes)
benchmarks.run.linux.arm.checked.mch	15,056,962	-54
benchmarks.run_pgo.linux.arm.checked.mch	63,327,810	-49,248
benchmarks.run_tiered.linux.arm.checked.mch	17,395,270	-32,052
coreclr_tests.run.linux.arm.checked.mch	321,199,404	-200,016
libraries.crossgen2.linux.arm.checked.mch	36,758,116	+0
libraries.pmi.linux.arm.checked.mch	48,738,124	-280
libraries_tests.run.linux.arm.Release.mch	244,888,660	-547,278
libraries_tests_no_tiered_compilation.run.linux.arm.Release.mch	93,389,302	-24,904
realworld.run.linux.arm.checked.mch	13,304,600	-14,646

MinOpts (-868,478 bytes)

Collection	Base size (bytes)	Diff size (bytes)
benchmarks.run.linux.arm.checked.mch	383,466	-54
benchmarks.run_pgo.linux.arm.checked.mch	11,641,744	-49,248
benchmarks.run_tiered.linux.arm.checked.mch	7,192,228	-32,052
coreclr_tests.run.linux.arm.checked.mch	212,641,656	-200,016
libraries.crossgen2.linux.arm.checked.mch	1,230	+0
libraries.pmi.linux.arm.checked.mch	106,504	-280
libraries_tests.run.linux.arm.Release.mch	122,290,786	-547,278
libraries_tests_no_tiered_compilation.run.linux.arm.Release.mch	10,083,802	-24,904
realworld.run.linux.arm.checked.mch	449,686	-14,646

Assembly diffs for windows/x86 ran on windows/x86

Diffs are based on 2,246,531 contexts (794,865 MinOpts, 1,451,666 FullOpts).

MISSED contexts: base: 7,010 (0.30%), diff: 52,597 (2.29%)

Overall (+190,955 bytes)

Collection	Base size (bytes)	Diff size (bytes)
benchmarks.run.windows.x86.checked.mch	6,960,783	+2
benchmarks.run_pgo.windows.x86.checked.mch	44,966,304	+504
benchmarks.run_tiered.windows.x86.checked.mch	8,898,260	+195
coreclr_tests.run.windows.x86.checked.mch	305,383,476	+201,068
libraries.crossgen2.windows.x86.checked.mch	31,679,639	+3
libraries.pmi.windows.x86.checked.mch	48,041,556	-81
libraries_tests.run.windows.x86.Release.mch	167,902,249	-13,935
libraries_tests_no_tiered_compilation.run.windows.x86.Release.mch	101,249,179	+3,185
realworld.run.windows.x86.checked.mch	11,011,414	+14

MinOpts (+190,955 bytes)

Collection	Base size (bytes)	Diff size (bytes)
benchmarks.run.windows.x86.checked.mch	279	+2
benchmarks.run_pgo.windows.x86.checked.mch	5,930,678	+504
benchmarks.run_tiered.windows.x86.checked.mch	3,850,704	+195
coreclr_tests.run.windows.x86.checked.mch	198,326,162	+201,068
libraries.crossgen2.windows.x86.checked.mch	1,057	+3
libraries.pmi.windows.x86.checked.mch	95,314	-81
libraries_tests.run.windows.x86.Release.mch	78,712,933	-13,935
libraries_tests_no_tiered_compilation.run.windows.x86.Release.mch	7,275,039	+3,185
realworld.run.windows.x86.checked.mch	295,700	+14

Details here

Throughput diffs

Throughput diffs for linux/arm64 ran on windows/x64

Overall (-5.32% to -0.04%)

Collection	PDIFF
benchmarks.run.linux.arm64.checked.mch	-0.11%
benchmarks.run_pgo.linux.arm64.checked.mch	-1.07%
benchmarks.run_tiered.linux.arm64.checked.mch	-5.32%
coreclr_tests.run.linux.arm64.checked.mch	-3.98%
libraries.crossgen2.linux.arm64.checked.mch	-0.07%
libraries.pmi.linux.arm64.checked.mch	-0.05%
libraries_tests.run.linux.arm64.Release.mch	-2.35%
libraries_tests_no_tiered_compilation.run.linux.arm64.Release.mch	-0.29%
realworld.run.linux.arm64.checked.mch	-0.13%
smoke_tests.nativeaot.linux.arm64.checked.mch	-0.04%

MinOpts (-10.39% to -6.49%)

Collection	PDIFF
benchmarks.run.linux.arm64.checked.mch	-8.36%
benchmarks.run_pgo.linux.arm64.checked.mch	-8.98%
benchmarks.run_tiered.linux.arm64.checked.mch	-9.07%
coreclr_tests.run.linux.arm64.checked.mch	-9.16%
libraries.crossgen2.linux.arm64.checked.mch	-8.01%
libraries.pmi.linux.arm64.checked.mch	-9.80%
libraries_tests.run.linux.arm64.Release.mch	-8.97%
libraries_tests_no_tiered_compilation.run.linux.arm64.Release.mch	-8.60%
realworld.run.linux.arm64.checked.mch	-10.39%
smoke_tests.nativeaot.linux.arm64.checked.mch	-6.49%

FullOpts (-0.07% to -0.03%)

Collection	PDIFF
benchmarks.run.linux.arm64.checked.mch	-0.05%
benchmarks.run_pgo.linux.arm64.checked.mch	-0.04%
benchmarks.run_tiered.linux.arm64.checked.mch	-0.03%
coreclr_tests.run.linux.arm64.checked.mch	-0.05%
libraries.crossgen2.linux.arm64.checked.mch	-0.07%
libraries.pmi.linux.arm64.checked.mch	-0.04%
libraries_tests.run.linux.arm64.Release.mch	-0.04%
libraries_tests_no_tiered_compilation.run.linux.arm64.Release.mch	-0.05%
realworld.run.linux.arm64.checked.mch	-0.04%
smoke_tests.nativeaot.linux.arm64.checked.mch	-0.04%

Throughput diffs for linux/x64 ran on windows/x64

Overall (-2.61% to -0.05%)

Collection	PDIFF
benchmarks.run.linux.x64.checked.mch	-0.09%
benchmarks.run_pgo.linux.x64.checked.mch	-0.61%
benchmarks.run_tiered.linux.x64.checked.mch	-2.55%
coreclr_tests.run.linux.x64.checked.mch	-2.61%
libraries.crossgen2.linux.x64.checked.mch	-0.08%
libraries.pmi.linux.x64.checked.mch	-0.06%
libraries_tests.run.linux.x64.Release.mch	-1.28%
libraries_tests_no_tiered_compilation.run.linux.x64.Release.mch	-0.20%
realworld.run.linux.x64.checked.mch	-0.10%
smoke_tests.nativeaot.linux.x64.checked.mch	-0.05%

MinOpts (-6.57% to -4.49%)

Collection	PDIFF
benchmarks.run.linux.x64.checked.mch	-5.03%
benchmarks.run_pgo.linux.x64.checked.mch	-5.61%
benchmarks.run_tiered.linux.x64.checked.mch	-5.72%
coreclr_tests.run.linux.x64.checked.mch	-6.14%
libraries.crossgen2.linux.x64.checked.mch	-4.49%
libraries.pmi.linux.x64.checked.mch	-5.89%
libraries_tests.run.linux.x64.Release.mch	-5.66%
libraries_tests_no_tiered_compilation.run.linux.x64.Release.mch	-5.15%
realworld.run.linux.x64.checked.mch	-6.57%
smoke_tests.nativeaot.linux.x64.checked.mch	-4.76%

FullOpts (-0.08% to -0.04%)

Collection	PDIFF
benchmarks.run.linux.x64.checked.mch	-0.06%
benchmarks.run_pgo.linux.x64.checked.mch	-0.04%
benchmarks.run_tiered.linux.x64.checked.mch	-0.04%
coreclr_tests.run.linux.x64.checked.mch	-0.06%
libraries.crossgen2.linux.x64.checked.mch	-0.08%
libraries.pmi.linux.x64.checked.mch	-0.06%
libraries_tests.run.linux.x64.Release.mch	-0.05%
libraries_tests_no_tiered_compilation.run.linux.x64.Release.mch	-0.07%
realworld.run.linux.x64.checked.mch	-0.06%
smoke_tests.nativeaot.linux.x64.checked.mch	-0.05%

Throughput diffs for osx/arm64 ran on windows/x64

Overall (-4.23% to -0.05%)

Collection	PDIFF
benchmarks.run_pgo.osx.arm64.checked.mch	-1.91%
benchmarks.run_tiered.osx.arm64.checked.mch	-4.23%
coreclr_tests.run.osx.arm64.checked.mch	-3.94%
libraries.crossgen2.osx.arm64.checked.mch	-0.07%
libraries.pmi.osx.arm64.checked.mch	-0.05%
libraries_tests.run.osx.arm64.Release.mch	-3.02%
libraries_tests_no_tiered_compilation.run.osx.arm64.Release.mch	-0.29%
realworld.run.osx.arm64.checked.mch	-0.14%

MinOpts (-10.43% to -7.92%)

Collection	PDIFF
benchmarks.run_pgo.osx.arm64.checked.mch	-9.25%
benchmarks.run_tiered.osx.arm64.checked.mch	-9.54%
coreclr_tests.run.osx.arm64.checked.mch	-9.03%
libraries.crossgen2.osx.arm64.checked.mch	-7.92%
libraries.pmi.osx.arm64.checked.mch	-9.79%
libraries_tests.run.osx.arm64.Release.mch	-9.00%
libraries_tests_no_tiered_compilation.run.osx.arm64.Release.mch	-8.60%
realworld.run.osx.arm64.checked.mch	-10.43%

FullOpts (-0.07% to -0.03%)

Collection	PDIFF
benchmarks.run_pgo.osx.arm64.checked.mch	-0.03%
benchmarks.run_tiered.osx.arm64.checked.mch	-0.03%
coreclr_tests.run.osx.arm64.checked.mch	-0.06%
libraries.crossgen2.osx.arm64.checked.mch	-0.07%
libraries.pmi.osx.arm64.checked.mch	-0.04%
libraries_tests.run.osx.arm64.Release.mch	-0.04%
libraries_tests_no_tiered_compilation.run.osx.arm64.Release.mch	-0.05%
realworld.run.osx.arm64.checked.mch	-0.04%

Throughput diffs for windows/arm64 ran on windows/x64

Overall (-4.15% to -0.04%)

Collection	PDIFF
benchmarks.run.windows.arm64.checked.mch	-0.04%
benchmarks.run_pgo.windows.arm64.checked.mch	-1.23%
benchmarks.run_tiered.windows.arm64.checked.mch	-4.15%
coreclr_tests.run.windows.arm64.checked.mch	-3.98%
libraries.crossgen2.windows.arm64.checked.mch	-0.07%
libraries.pmi.windows.arm64.checked.mch	-0.05%
libraries_tests.run.windows.arm64.Release.mch	-3.07%
libraries_tests_no_tiered_compilation.run.windows.arm64.Release.mch	-0.28%
realworld.run.windows.arm64.checked.mch	-0.13%
smoke_tests.nativeaot.windows.arm64.checked.mch	-0.04%

MinOpts (-10.42% to -6.32%)

Collection	PDIFF
benchmarks.run.windows.arm64.checked.mch	-8.96%
benchmarks.run_pgo.windows.arm64.checked.mch	-9.17%
benchmarks.run_tiered.windows.arm64.checked.mch	-9.48%
coreclr_tests.run.windows.arm64.checked.mch	-9.16%
libraries.crossgen2.windows.arm64.checked.mch	-8.01%
libraries.pmi.windows.arm64.checked.mch	-9.80%
libraries_tests.run.windows.arm64.Release.mch	-8.99%
libraries_tests_no_tiered_compilation.run.windows.arm64.Release.mch	-8.60%
realworld.run.windows.arm64.checked.mch	-10.42%
smoke_tests.nativeaot.windows.arm64.checked.mch	-6.32%

FullOpts (-0.07% to -0.03%)

Collection	PDIFF
benchmarks.run.windows.arm64.checked.mch	-0.04%
benchmarks.run_pgo.windows.arm64.checked.mch	-0.03%
benchmarks.run_tiered.windows.arm64.checked.mch	-0.03%
coreclr_tests.run.windows.arm64.checked.mch	-0.05%
libraries.crossgen2.windows.arm64.checked.mch	-0.07%
libraries.pmi.windows.arm64.checked.mch	-0.04%
libraries_tests.run.windows.arm64.Release.mch	-0.04%
libraries_tests_no_tiered_compilation.run.windows.arm64.Release.mch	-0.05%
realworld.run.windows.arm64.checked.mch	-0.04%
smoke_tests.nativeaot.windows.arm64.checked.mch	-0.04%

Throughput diffs for windows/x64 ran on windows/x64

Overall (-2.54% to -0.08%)

Collection	PDIFF
benchmarks.run.windows.x64.checked.mch	-0.08%
benchmarks.run_pgo.windows.x64.checked.mch	-0.83%
benchmarks.run_tiered.windows.x64.checked.mch	-2.24%
coreclr_tests.run.windows.x64.checked.mch	-2.54%
libraries.crossgen2.windows.x64.checked.mch	-0.12%
libraries.pmi.windows.x64.checked.mch	-0.10%
libraries_tests.run.windows.x64.Release.mch	-1.66%
libraries_tests_no_tiered_compilation.run.windows.x64.Release.mch	-0.24%
realworld.run.windows.x64.checked.mch	-0.14%
smoke_tests.nativeaot.windows.x64.checked.mch	-0.08%

MinOpts (-6.48% to -4.45%)

Collection	PDIFF
benchmarks.run.windows.x64.checked.mch	-4.89%
benchmarks.run_pgo.windows.x64.checked.mch	-5.70%
benchmarks.run_tiered.windows.x64.checked.mch	-5.72%
coreclr_tests.run.windows.x64.checked.mch	-5.96%
libraries.crossgen2.windows.x64.checked.mch	-4.45%
libraries.pmi.windows.x64.checked.mch	-5.72%
libraries_tests.run.windows.x64.Release.mch	-5.73%
libraries_tests_no_tiered_compilation.run.windows.x64.Release.mch	-5.56%
realworld.run.windows.x64.checked.mch	-6.48%
smoke_tests.nativeaot.windows.x64.checked.mch	-4.69%

FullOpts (-0.12% to -0.07%)

Collection	PDIFF
benchmarks.run.windows.x64.checked.mch	-0.08%
benchmarks.run_pgo.windows.x64.checked.mch	-0.07%
benchmarks.run_tiered.windows.x64.checked.mch	-0.07%
coreclr_tests.run.windows.x64.checked.mch	-0.11%
libraries.crossgen2.windows.x64.checked.mch	-0.12%
libraries.pmi.windows.x64.checked.mch	-0.09%
libraries_tests.run.windows.x64.Release.mch	-0.08%
libraries_tests_no_tiered_compilation.run.windows.x64.Release.mch	-0.11%
realworld.run.windows.x64.checked.mch	-0.10%
smoke_tests.nativeaot.windows.x64.checked.mch	-0.08%

Details here

Throughput diffs for linux/arm ran on windows/x86

Overall (-4.16% to -0.08%)

Collection	PDIFF
benchmarks.run.linux.arm.checked.mch	-0.20%
benchmarks.run_pgo.linux.arm.checked.mch	-0.71%
benchmarks.run_tiered.linux.arm.checked.mch	-2.03%
coreclr_tests.run.linux.arm.checked.mch	-4.16%
libraries.crossgen2.linux.arm.checked.mch	-0.15%
libraries.pmi.linux.arm.checked.mch	-0.08%
libraries_tests.run.linux.arm.Release.mch	-2.45%
libraries_tests_no_tiered_compilation.run.linux.arm.Release.mch	-0.46%
realworld.run.linux.arm.checked.mch	-0.15%

MinOpts (-11.57% to -9.31%)

Collection	PDIFF
benchmarks.run.linux.arm.checked.mch	-9.63%
benchmarks.run_pgo.linux.arm.checked.mch	-10.51%
benchmarks.run_tiered.linux.arm.checked.mch	-10.53%
coreclr_tests.run.linux.arm.checked.mch	-9.89%
libraries.crossgen2.linux.arm.checked.mch	-9.84%
libraries.pmi.linux.arm.checked.mch	-9.31%
libraries_tests.run.linux.arm.Release.mch	-10.61%
libraries_tests_no_tiered_compilation.run.linux.arm.Release.mch	-9.93%
realworld.run.linux.arm.checked.mch	-11.57%

FullOpts (-0.15% to -0.05%)

Collection	PDIFF
benchmarks.run.linux.arm.checked.mch	-0.10%
benchmarks.run_pgo.linux.arm.checked.mch	-0.06%
benchmarks.run_tiered.linux.arm.checked.mch	-0.09%
coreclr_tests.run.linux.arm.checked.mch	-0.11%
libraries.crossgen2.linux.arm.checked.mch	-0.15%
libraries.pmi.linux.arm.checked.mch	-0.07%
libraries_tests.run.linux.arm.Release.mch	-0.05%
libraries_tests_no_tiered_compilation.run.linux.arm.Release.mch	-0.09%
realworld.run.linux.arm.checked.mch	-0.06%

Throughput diffs for windows/x86 ran on windows/x86

Overall (-1.69% to +0.02%)

Collection	PDIFF
benchmarks.run.windows.x86.checked.mch	+0.02%
benchmarks.run_pgo.windows.x86.checked.mch	-0.12%
benchmarks.run_tiered.windows.x86.checked.mch	-0.60%
coreclr_tests.run.windows.x86.checked.mch	-1.69%
libraries.crossgen2.windows.x86.checked.mch	+0.01%
libraries.pmi.windows.x86.checked.mch	+0.02%
libraries_tests.run.windows.x86.Release.mch	-0.73%
libraries_tests_no_tiered_compilation.run.windows.x86.Release.mch	-0.06%
realworld.run.windows.x86.checked.mch	-0.01%

MinOpts (-5.04% to -3.05%)

Collection	PDIFF
benchmarks.run.windows.x86.checked.mch	-3.15%
benchmarks.run_pgo.windows.x86.checked.mch	-3.99%
benchmarks.run_tiered.windows.x86.checked.mch	-3.90%
coreclr_tests.run.windows.x86.checked.mch	-5.04%
libraries.crossgen2.windows.x86.checked.mch	-3.05%
libraries.pmi.windows.x86.checked.mch	-4.50%
libraries_tests.run.windows.x86.Release.mch	-4.25%
libraries_tests_no_tiered_compilation.run.windows.x86.Release.mch	-3.69%
realworld.run.windows.x86.checked.mch	-4.85%

FullOpts (+0.01% to +0.02%)

Collection	PDIFF
benchmarks.run.windows.x86.checked.mch	+0.02%
benchmarks.run_pgo.windows.x86.checked.mch	+0.02%
benchmarks.run_tiered.windows.x86.checked.mch	+0.02%
coreclr_tests.run.windows.x86.checked.mch	+0.01%
libraries.crossgen2.windows.x86.checked.mch	+0.01%
libraries.pmi.windows.x86.checked.mch	+0.02%
libraries_tests.run.windows.x86.Release.mch	+0.01%
libraries_tests_no_tiered_compilation.run.windows.x86.Release.mch	+0.01%
realworld.run.windows.x86.checked.mch	+0.02%

Details here

Throughput diffs for linux/arm64 ran on linux/x64

Overall (-5.82% to +0.04%)

Collection	PDIFF
smoke_tests.nativeaot.linux.arm64.checked.mch	+0.04%
benchmarks.run_tiered.linux.arm64.checked.mch	-5.82%
benchmarks.run_pgo.linux.arm64.checked.mch	-1.12%
libraries.crossgen2.linux.arm64.checked.mch	+0.04%
libraries.pmi.linux.arm64.checked.mch	+0.03%
realworld.run.linux.arm64.checked.mch	-0.07%
libraries_tests.run.linux.arm64.Release.mch	-2.55%
benchmarks.run.linux.arm64.checked.mch	-0.01%
coreclr_tests.run.linux.arm64.checked.mch	-4.32%
libraries_tests_no_tiered_compilation.run.linux.arm64.Release.mch	-0.24%

MinOpts (-11.62% to -6.54%)

Collection	PDIFF
smoke_tests.nativeaot.linux.arm64.checked.mch	-6.54%
benchmarks.run_tiered.linux.arm64.checked.mch	-10.03%
benchmarks.run_pgo.linux.arm64.checked.mch	-10.07%
libraries.crossgen2.linux.arm64.checked.mch	-9.09%
libraries.pmi.linux.arm64.checked.mch	-11.21%
realworld.run.linux.arm64.checked.mch	-11.62%
libraries_tests.run.linux.arm64.Release.mch	-9.97%
benchmarks.run.linux.arm64.checked.mch	-9.19%
coreclr_tests.run.linux.arm64.checked.mch	-9.71%
libraries_tests_no_tiered_compilation.run.linux.arm64.Release.mch	-10.06%

FullOpts (+0.02% to +0.06%)

Collection	PDIFF
smoke_tests.nativeaot.linux.arm64.checked.mch	+0.04%
benchmarks.run_tiered.linux.arm64.checked.mch	+0.02%
benchmarks.run_pgo.linux.arm64.checked.mch	+0.03%
libraries.crossgen2.linux.arm64.checked.mch	+0.04%
libraries.pmi.linux.arm64.checked.mch	+0.03%
realworld.run.linux.arm64.checked.mch	+0.03%
libraries_tests.run.linux.arm64.Release.mch	+0.03%
benchmarks.run.linux.arm64.checked.mch	+0.06%
coreclr_tests.run.linux.arm64.checked.mch	+0.05%
libraries_tests_no_tiered_compilation.run.linux.arm64.Release.mch	+0.04%

Throughput diffs for linux/x64 ran on linux/x64

Overall (-2.72% to -0.15%)

Collection	PDIFF
benchmarks.run.linux.x64.checked.mch	-0.20%
libraries.pmi.linux.x64.checked.mch	-0.17%
coreclr_tests.run.linux.x64.checked.mch	-2.70%
libraries_tests.run.linux.x64.Release.mch	-1.39%
benchmarks.run_pgo.linux.x64.checked.mch	-0.71%
libraries.crossgen2.linux.x64.checked.mch	-0.22%
libraries_tests_no_tiered_compilation.run.linux.x64.Release.mch	-0.33%
realworld.run.linux.x64.checked.mch	-0.20%
smoke_tests.nativeaot.linux.x64.checked.mch	-0.15%
benchmarks.run_tiered.linux.x64.checked.mch	-2.72%

MinOpts (-6.70% to -4.66%)

Collection	PDIFF
benchmarks.run.linux.x64.checked.mch	-5.13%
libraries.pmi.linux.x64.checked.mch	-6.33%
coreclr_tests.run.linux.x64.checked.mch	-5.89%
libraries_tests.run.linux.x64.Release.mch	-5.81%
benchmarks.run_pgo.linux.x64.checked.mch	-5.87%
libraries.crossgen2.linux.x64.checked.mch	-4.86%
libraries_tests_no_tiered_compilation.run.linux.x64.Release.mch	-5.60%
realworld.run.linux.x64.checked.mch	-6.70%
smoke_tests.nativeaot.linux.x64.checked.mch	-4.66%
benchmarks.run_tiered.linux.x64.checked.mch	-6.00%

FullOpts (-0.22% to -0.12%)

Collection	PDIFF
benchmarks.run.linux.x64.checked.mch	-0.17%
libraries.pmi.linux.x64.checked.mch	-0.16%
coreclr_tests.run.linux.x64.checked.mch	-0.19%
libraries_tests.run.linux.x64.Release.mch	-0.13%
benchmarks.run_pgo.linux.x64.checked.mch	-0.13%
libraries.crossgen2.linux.x64.checked.mch	-0.22%
libraries_tests_no_tiered_compilation.run.linux.x64.Release.mch	-0.19%
realworld.run.linux.x64.checked.mch	-0.16%
smoke_tests.nativeaot.linux.x64.checked.mch	-0.15%
benchmarks.run_tiered.linux.x64.checked.mch	-0.12%

Details here

kunalspathak added 13 commits December 28, 2023 07:57

Optimize resetAllRegistersState()

6deeff4

Add assert about refType for minopts

7ff5cd3

Templatize processBlockEndAllocation()

33852e4

Add resetMinOpts()

d42e5da

Add selectMinOpts()

0d0f27f

Add calculateUnassignedSet()

bc00b94

Call spillCostMinOpts() for minopts

62af8b4

Add enregisterLocalVars check in setIntervalAsSpilled()

df18666

optimize few paths in resolveRegisters()

f24ef56

optimize few paths in allocateRegisters()

edd0815

Adjust RegisterSelection creation

a55144b

Add allocateRegistersForMinOpts()

b2bea01

jit format

8ce699b

ghost assigned kunalspathak Jan 1, 2024

dotnet-issue-labeler bot added the area-CodeGen-coreclr CLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI label Jan 1, 2024

build-analysis bot mentioned this pull request Jan 1, 2024

An error occurred trying to start process 'llvm-symbolizer' #91975

Open

Update the assert for processBlockStartLocations()

4ae62ed

Fix bug by using freeCandidates

7f29c4e

This was referenced Jan 3, 2024

Tracking issue for CI build timeouts #76454

Closed

[libraries-jitstress] Failures in System.Numerics.Tensors.Tests.*Generic* #96464

Closed

kunalspathak added 6 commits January 3, 2024 23:02

Merge remote-tracking branch 'origin/main' into minopts-lsra

b8180f1

Set candidates only if freeCandidates are NONE

755b9aa

Remove unnecessary setting of foundRegBit

a0222da

Remove unnecessary checks for freeCandidates == RBM_NONE

01b29b0

Remove unnecessary checks for freeCandidates == RBM_NONE --amend

f6d8aa7

wip

1cbd0b7

Quick LSRA only if optimization is disabled

39d3e9d

kunalspathak changed the title ~~Experiment: Optimize lsra when enregisterLocalVars is false~~ Optimize lsra for MinOpts and enregisterLocalVars is false Jan 20, 2024

build-analysis bot mentioned this pull request Jan 20, 2024

Test failing: System.Globalization.Tests.CompareInfoHashCodeTests.CheckHashingOfSkippedChars #97241

Closed

build-analysis bot mentioned this pull request Jan 20, 2024

System.Diagnostics.Tracing.Tests BasicEventSourceTests.TestsManifestGeneration test fails on windows x86 leg #97255

Closed

kunalspathak changed the title ~~Optimize lsra for MinOpts and enregisterLocalVars is false~~ Optimize lsra for MinOpts Jan 22, 2024

jakobbotsch reviewed Jan 23, 2024

View reviewed changes

src/coreclr/jit/lsra.cpp Outdated Show resolved Hide resolved

jakobbotsch reviewed Jan 23, 2024

View reviewed changes

jakobbotsch approved these changes Jan 23, 2024

View reviewed changes

kunalspathak added 2 commits January 25, 2024 15:07

Add verifyFreeRegisters()

55f5b4b

Merge remote-tracking branch 'origin/main' into minopts-lsra

9e767eb

kunalspathak merged commit 468447b into dotnet:main Jan 26, 2024
129 checks passed

kunalspathak deleted the minopts-lsra branch January 26, 2024 15:39

jakobbotsch mentioned this pull request Jan 29, 2024

JIT: checked/release diffs #97638

Closed

kunalspathak mentioned this pull request Jan 29, 2024

For Debug, make sure we use right register selection heuristics to match with Release #97657

Merged

tomeksowi mentioned this pull request Feb 1, 2024

[RISC-V] Call missing genProduceReg in genCodeForCompare #97805

Merged

github-actions bot locked and limited conversation to collaborators Feb 26, 2024

Optimize lsra for MinOpts #96386

Optimize lsra for MinOpts #96386

Conversation

kunalspathak commented Jan 1, 2024 • edited Loading

ghost commented Jan 1, 2024

kunalspathak commented Jan 2, 2024

kunalspathak commented Jan 2, 2024

kunalspathak commented Jan 4, 2024

kunalspathak commented Jan 19, 2024

Linux/x64 analysis

jakobbotsch commented Jan 19, 2024 • edited Loading

kunalspathak commented Jan 20, 2024 • edited Loading

jakobbotsch commented Jan 20, 2024

BruceForstall commented Jan 20, 2024

kunalspathak commented Jan 20, 2024

kunalspathak commented Jan 20, 2024

kunalspathak commented Jan 20, 2024

azure-pipelines bot commented Jan 20, 2024

kunalspathak commented Jan 22, 2024

kunalspathak commented Jan 22, 2024

Assembly diffs

Assembly diffs for linux/arm64 ran on windows/x64

Assembly diffs for linux/x64 ran on windows/x64

Assembly diffs for osx/arm64 ran on windows/x64

Assembly diffs for windows/arm64 ran on windows/x64

Assembly diffs for windows/x64 ran on windows/x64

Assembly diffs for linux/arm ran on windows/x86

Assembly diffs for windows/x86 ran on windows/x86

Throughput diffs

Throughput diffs for linux/arm64 ran on windows/x64

Throughput diffs for linux/x64 ran on windows/x64

Throughput diffs for osx/arm64 ran on windows/x64

Throughput diffs for windows/arm64 ran on windows/x64

Throughput diffs for windows/x64 ran on windows/x64

Throughput diffs for linux/arm64 ran on linux/x64

Throughput diffs for linux/x64 ran on linux/x64

Throughput diffs for linux/arm ran on windows/x86

Throughput diffs for windows/x86 ran on windows/x86

kunalspathak commented Jan 22, 2024

jakobbotsch Jan 23, 2024

Choose a reason for hiding this comment

jakobbotsch left a comment

Choose a reason for hiding this comment

kunalspathak commented Jan 25, 2024 • edited Loading

ryujit-bot commented Jan 26, 2024

Assembly diffs

Assembly diffs for linux/arm64 ran on windows/x64

Assembly diffs for linux/x64 ran on windows/x64

Assembly diffs for osx/arm64 ran on windows/x64

Assembly diffs for windows/arm64 ran on windows/x64

Assembly diffs for windows/x64 ran on windows/x64

Assembly diffs for linux/arm ran on windows/x86

Assembly diffs for windows/x86 ran on windows/x86

Throughput diffs

Throughput diffs for linux/arm64 ran on windows/x64

Throughput diffs for linux/x64 ran on windows/x64

Throughput diffs for osx/arm64 ran on windows/x64

Throughput diffs for windows/arm64 ran on windows/x64

Throughput diffs for windows/x64 ran on windows/x64

Throughput diffs for linux/arm ran on windows/x86

Throughput diffs for windows/x86 ran on windows/x86

Throughput diffs for linux/arm64 ran on linux/x64

Throughput diffs for linux/x64 ran on linux/x64

kunalspathak commented Jan 1, 2024 •

edited

Loading

jakobbotsch commented Jan 19, 2024 •

edited

Loading

kunalspathak commented Jan 20, 2024 •

edited

Loading

kunalspathak commented Jan 25, 2024 •

edited

Loading