Make ToDictionary() selectors parallel.

[lateapexearlyspeed]

Fix #96262

[jozkee]

As stated in #96262 (comment), this will likely regress cases with a trivial key selector. @lateapexearlyspeed could you please provide a benchmark to see how bad it hits it so we can do an informed decision?

[lateapexearlyspeed]

@jozkee following is benchmark result of trivial key (and element) selector case. Toolchain in "runtime-main" is library before change and toolchain in "runtime" is library after change.

// * Summary *

BenchmarkDotNet=v0.13.1.1603-nightly, OS=Windows 10.0.22631
12th Gen Intel Core i7-12800H, 1 CPU, 20 logical and 14 physical cores
.NET SDK=8.0.202
[Host] : .NET 6.0.31 (6.0.3124.26714), X64 RyuJIT
Job-XVKCDM : .NET 9.0.0 (42.42.42.42424), X64 RyuJIT
Job-ODFYHP : .NET 9.0.0 (42.42.42.42424), X64 RyuJIT

PowerPlanMode=00000000-0000-0000-0000-000000000000 Arguments=/p:DebugType=portable,-bl:benchmarkdotnet.binlog IterationTime=250.0000 ms
MaxIterationCount=20 MinIterationCount=15 WarmupCount=1

Method	Job	Toolchain	ItemCount	Mean	Error	StdDev	Median	Min	Max	Ratio	RatioSD	Gen 0	Gen 1	Gen 2	Allocated
TestParallelToDictionary	Job-XVKCDM	\runtime-main\artifacts\bin\testhost\net9.0-windows-Release-x64\shared\Microsoft.NETCore.App\9.0.0\CoreRun.exe	1	42.00 ns	9.363 ns	10.782 ns	35.46 ns	32.76 ns	62.01 ns	1.00	0.00	0.0298	-	-	376 B
TestParallelToDictionary	Job-ODFYHP	\runtime\artifacts\bin\testhost\net9.0-windows-Release-x64\shared\Microsoft.NETCore.App\9.0.0\CoreRun.exe	1	7,778.30 ns	149.193 ns	139.555 ns	7,773.13 ns	7,457.77 ns	8,025.18 ns	214.43	27.42	0.6916	-	-	8,896 B
TestParallelToDictionary	Job-XVKCDM	\runtime-main\artifacts\bin\testhost\net9.0-windows-Release-x64\shared\Microsoft.NETCore.App\9.0.0\CoreRun.exe	10	120.94 ns	2.399 ns	2.356 ns	121.89 ns	116.36 ns	124.39 ns	1.00	0.00	0.0765	-	-	960 B
TestParallelToDictionary	Job-ODFYHP	\runtime\artifacts\bin\testhost\net9.0-windows-Release-x64\shared\Microsoft.NETCore.App\9.0.0\CoreRun.exe	10	8,034.68 ns	190.599 ns	219.495 ns	8,038.56 ns	7,658.90 ns	8,423.41 ns	66.69	2.11	0.7844	-	-	9,984 B
TestParallelToDictionary	Job-XVKCDM	\runtime-main\artifacts\bin\testhost\net9.0-windows-Release-x64\shared\Microsoft.NETCore.App\9.0.0\CoreRun.exe	100	1,054.74 ns	94.978 ns	109.377 ns	1,025.83 ns	905.00 ns	1,335.60 ns	1.00	0.00	0.6014	0.0073	-	7,576 B
TestParallelToDictionary	Job-ODFYHP	\runtime\artifacts\bin\testhost\net9.0-windows-Release-x64\shared\Microsoft.NETCore.App\9.0.0\CoreRun.exe	100	10,700.76 ns	295.269 ns	328.191 ns	10,768.11 ns	10,152.59 ns	11,369.07 ns	10.20	1.00	1.5175	-	-	18,920 B
TestParallelToDictionary	Job-XVKCDM	\runtime-main\artifacts\bin\testhost\net9.0-windows-Release-x64\shared\Microsoft.NETCore.App\9.0.0\CoreRun.exe	1000	8,536.74 ns	206.045 ns	237.282 ns	8,593.82 ns	8,068.56 ns	8,911.05 ns	1.00	0.00	5.8294	0.6955	-	73,352 B
TestParallelToDictionary	Job-ODFYHP	\runtime\artifacts\bin\testhost\net9.0-windows-Release-x64\shared\Microsoft.NETCore.App\9.0.0\CoreRun.exe	1000	45,359.53 ns	1,780.828 ns	2,050.804 ns	44,628.87 ns	42,440.57 ns	50,316.83 ns	5.31	0.22	8.3220	1.3587	-	104,118 B
TestParallelToDictionary	Job-XVKCDM	\runtime-main\artifacts\bin\testhost\net9.0-windows-Release-x64\shared\Microsoft.NETCore.App\9.0.0\CoreRun.exe	10000	175,304.48 ns	5,273.794 ns	6,073.309 ns	174,695.64 ns	166,454.45 ns	186,023.27 ns	1.00	0.00	124.3351	124.3351	124.3351	673,290 B
TestParallelToDictionary	Job-ODFYHP	\runtime\artifacts\bin\testhost\net9.0-windows-Release-x64\shared\Microsoft.NETCore.App\9.0.0\CoreRun.exe	10000	425,559.34 ns	21,829.962 ns	25,139.417 ns	422,713.75 ns	382,274.84 ns	461,048.75 ns	2.43	0.20	131.2500	131.2500	123.4375	848,901 B

---

        [Benchmark]
        public void TestParallelToDictionary()
        {
            IEnumerable<int> collection = Enumerable.Range(0, ItemCount);
            Dictionary<int, int> _ = collection.AsParallel().ToDictionary(i => i, i => i);
        }

        [Params(1, 10, 100, 1000, 10000)]
        public int ItemCount { get; set; }

[stephentoub]

@EgorBot -amd

using System.Linq;
using BenchmarkDotNet.Attributes;
using BenchmarkDotNet.Running;

BenchmarkRunner.Run<Tests>(args: args);

public class Tests
{
        [Benchmark]
        public void TestParallelToDictionary()
        {
            IEnumerable<int> collection = Enumerable.Range(0, ItemCount);
            Dictionary<int, int> _ = collection.AsParallel().ToDictionary(i => i, i => i);
        }

        [Benchmark]
        public void TestSelectParallelToDictionary()
        {
            IEnumerable<int> collection = Enumerable.Range(0, ItemCount);
            Dictionary<int, int> _ = collection.AsParallel().Select(i => i).ToDictionary(i => i, i => i);
        }

        [Benchmark]
        public void TestWhereSelectParallelToDictionary()
        {
            IEnumerable<int> collection = Enumerable.Range(0, ItemCount);
            Dictionary<int, int> _ = collection.AsParallel().Where(i => true).Select(i => i).ToDictionary(i => i, i => i);
        }

        [Params(1, 10, 100, 1000, 10000)]
        public int ItemCount { get; set; }
}

[stephentoub]

@EgorBo, what's the invocation syntax for your bot?

[EgorBo]

@EgorBo, what's the invocation syntax for your bot?

I typically use -intel -arm64 -profiler [the rest of the args are directly passed to BDN app e.g. --disasm]
it means "run for Intel (Xeon) and Arm64 (Ampere), with profiler and with the given BDN args".
There is also -amd if you need AMD Epyc (Milano) specific results. All args are optional (-intel is used by default when no target specified, mostly, becuase it's the fastest VM instance I have)

Example: #102805 (comment)

-profiler is not useful when you have multiple [Benchmark]s.

[stephentoub]

@EgorBot -intel -arm64

using System.Linq;
using BenchmarkDotNet.Attributes;
using BenchmarkDotNet.Running;

BenchmarkRunner.Run<Tests>(args: args);

public class Tests
{
        [Benchmark]
        public void TestParallelToDictionary()
        {
            IEnumerable<int> collection = Enumerable.Range(0, ItemCount);
            Dictionary<int, int> _ = collection.AsParallel().ToDictionary(i => i, i => i);
        }

        [Benchmark]
        public void TestSelectParallelToDictionary()
        {
            IEnumerable<int> collection = Enumerable.Range(0, ItemCount);
            Dictionary<int, int> _ = collection.AsParallel().Select(i => i).ToDictionary(i => i, i => i);
        }

        [Benchmark]
        public void TestWhereSelectParallelToDictionary()
        {
            IEnumerable<int> collection = Enumerable.Range(0, ItemCount);
            Dictionary<int, int> _ = collection.AsParallel().Where(i => true).Select(i => i).ToDictionary(i => i, i => i);
        }

        [Params(1, 10, 100, 1000, 10000)]
        public int ItemCount { get; set; }
}

[stephentoub]

@EgorBo, what's the invocation syntax for your bot?

I typically use -intel -arm64 -profiler [the rest of the args are directly passed to BDN app e.g. --disasm] it means "run for Intel (Xeon) and Arm64 (Ampere), with profiler and with the given BDN args". There is also -amd if you need AMD Epyc (Milano) specific results. All args are optional (-intel is used by default when no target specified, mostly, becuase it's the fastest VM instance I have)

Example: #102805 (comment)

-profiler is not useful when you have multiple [Benchmark]s.

Thanks. What am I doing wrong that it's not kicking in?

[EgorBo]

@EgorBo, what's the invocation syntax for your bot?

I typically use -intel -arm64 -profiler [the rest of the args are directly passed to BDN app e.g. --disasm] it means "run for Intel (Xeon) and Arm64 (Ampere), with profiler and with the given BDN args". There is also -amd if you need AMD Epyc (Milano) specific results. All args are optional (-intel is used by default when no target specified, mostly, becuase it's the fastest VM instance I have)
Example: #102805 (comment)
-profiler is not useful when you have multiple [Benchmark]s.

Thanks. What am I doing wrong that it's not kicking in?

It's correct as is (I already see 3 16-core VMs allocated for these requests😆) - it's just that currently there is no feedback that the jobs have started yet

[stephentoub]

it's just that currently there is no feedback that the jobs have started yet

Ah! Ok. I'm used to the other bots that post a comment indicating that they've received the request. Thanks.

[EgorBot]

BenchmarkDotNet v0.13.12, Ubuntu 22.04.4 LTS (Jammy Jellyfish)
Unknown processor
  Job-UDSLNT : .NET 9.0.0 (42.42.42.42424), Arm64 RyuJIT AdvSIMD
  Job-JPHOCC : .NET 9.0.0 (42.42.42.42424), Arm64 RyuJIT AdvSIMD

Method	Toolchain	ItemCount	Mean	Error	Ratio
TestParallelToDictionary	Main	1	105.8 ns	0.90 ns	1.00
TestParallelToDictionary	PR	1	8,586.0 ns	47.85 ns	81.20
TestSelectParallelToDictionary	Main	1	8,611.2 ns	37.32 ns	1.00
TestSelectParallelToDictionary	PR	1	8,805.7 ns	72.94 ns	1.02
TestWhereSelectParallelToDictionary	Main	1	10,038.1 ns	45.88 ns	1.00
TestWhereSelectParallelToDictionary	PR	1	10,348.0 ns	60.22 ns	1.03
TestParallelToDictionary	Main	10	390.3 ns	3.63 ns	1.00
TestParallelToDictionary	PR	10	9,635.3 ns	64.86 ns	24.69
TestSelectParallelToDictionary	Main	10	9,300.1 ns	86.65 ns	1.00
TestSelectParallelToDictionary	PR	10	9,630.1 ns	163.45 ns	1.03
TestWhereSelectParallelToDictionary	Main	10	10,586.7 ns	67.26 ns	1.00
TestWhereSelectParallelToDictionary	PR	10	11,616.0 ns	96.22 ns	1.10
TestParallelToDictionary	Main	100	2,414.5 ns	5.49 ns	1.00
TestParallelToDictionary	PR	100	14,295.5 ns	110.44 ns	5.92
TestSelectParallelToDictionary	Main	100	13,759.9 ns	110.77 ns	1.00
TestSelectParallelToDictionary	PR	100	14,346.2 ns	179.50 ns	1.04
TestWhereSelectParallelToDictionary	Main	100	15,354.5 ns	148.65 ns	1.00
TestWhereSelectParallelToDictionary	PR	100	16,238.8 ns	137.54 ns	1.06
TestParallelToDictionary	Main	1000	22,040.0 ns	404.10 ns	1.00
TestParallelToDictionary	PR	1000	49,659.3 ns	992.25 ns	2.33
TestSelectParallelToDictionary	Main	1000	50,769.5 ns	992.20 ns	1.00
TestSelectParallelToDictionary	PR	1000	52,010.0 ns	896.11 ns	1.02
TestWhereSelectParallelToDictionary	Main	1000	52,453.1 ns	496.28 ns	1.00
TestWhereSelectParallelToDictionary	PR	1000	51,200.5 ns	1,087.75 ns	1.05
TestParallelToDictionary	Main	10000	317,249.3 ns	4,007.57 ns	1.00
TestParallelToDictionary	PR	10000	598,908.9 ns	3,963.41 ns	1.89
TestSelectParallelToDictionary	Main	10000	552,960.2 ns	9,053.46 ns	1.00
TestSelectParallelToDictionary	PR	10000	599,659.8 ns	3,316.79 ns	1.08
TestWhereSelectParallelToDictionary	Main	10000	560,577.1 ns	5,482.61 ns	1.00
TestWhereSelectParallelToDictionary	PR	10000	611,083.7 ns	6,500.18 ns	1.09

BDN_Artifacts.zip

[EgorBot]

BenchmarkDotNet v0.13.12, Ubuntu 22.04.4 LTS (Jammy Jellyfish)
Intel Xeon Platinum 8370C CPU 2.80GHz, 1 CPU, 16 logical and 8 physical cores
  Job-DMAJRO : .NET 9.0.0 (42.42.42.42424), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI
  Job-CCUEBP : .NET 9.0.0 (42.42.42.42424), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI

Method	Toolchain	ItemCount	Mean	Error	Ratio
TestParallelToDictionary	Main	1	68.87 ns	0.774 ns	1.00
TestParallelToDictionary	PR	1	9,353.72 ns	182.088 ns	135.81
TestSelectParallelToDictionary	Main	1	9,613.30 ns	187.263 ns	1.00
TestSelectParallelToDictionary	PR	1	9,438.78 ns	185.942 ns	0.98
TestWhereSelectParallelToDictionary	Main	1	10,173.87 ns	202.386 ns	1.00
TestWhereSelectParallelToDictionary	PR	1	11,065.01 ns	217.562 ns	1.10
TestParallelToDictionary	Main	10	235.89 ns	2.557 ns	1.00
TestParallelToDictionary	PR	10	9,453.04 ns	182.520 ns	40.04
TestSelectParallelToDictionary	Main	10	9,254.72 ns	157.309 ns	1.00
TestSelectParallelToDictionary	PR	10	10,072.27 ns	199.426 ns	1.11
TestWhereSelectParallelToDictionary	Main	10	10,812.37 ns	264.585 ns	1.00
TestWhereSelectParallelToDictionary	PR	10	11,822.02 ns	200.344 ns	1.11
TestParallelToDictionary	Main	100	1,676.41 ns	26.051 ns	1.00
TestParallelToDictionary	PR	100	13,394.86 ns	256.131 ns	7.99
TestSelectParallelToDictionary	Main	100	12,813.12 ns	255.738 ns	1.00
TestSelectParallelToDictionary	PR	100	13,425.19 ns	240.527 ns	1.05
TestWhereSelectParallelToDictionary	Main	100	13,975.55 ns	271.767 ns	1.00
TestWhereSelectParallelToDictionary	PR	100	14,356.32 ns	286.720 ns	1.03
TestParallelToDictionary	Main	1000	14,119.17 ns	175.553 ns	1.00
TestParallelToDictionary	PR	1000	33,632.66 ns	655.722 ns	2.39
TestSelectParallelToDictionary	Main	1000	31,916.77 ns	507.908 ns	1.00
TestSelectParallelToDictionary	PR	1000	32,717.81 ns	633.848 ns	1.02
TestWhereSelectParallelToDictionary	Main	1000	34,647.44 ns	684.606 ns	1.00
TestWhereSelectParallelToDictionary	PR	1000	42,381.94 ns	2,569.036 ns	1.28
TestParallelToDictionary	Main	10000	216,167.31 ns	341.834 ns	1.00
TestParallelToDictionary	PR	10000	383,979.18 ns	6,465.315 ns	1.78
TestSelectParallelToDictionary	Main	10000	375,850.28 ns	4,573.328 ns	1.00
TestSelectParallelToDictionary	PR	10000	396,121.31 ns	5,709.440 ns	1.05
TestWhereSelectParallelToDictionary	Main	10000	376,878.77 ns	7,402.242 ns	1.00
TestWhereSelectParallelToDictionary	PR	10000	391,383.93 ns	7,662.109 ns	1.03

BDN_Artifacts.zip

[ericstj]

It looks like @EgorBo's test results show some significant regressions. Am I reading that correctly @jozkee @stephentoub?

[stephentoub]

Yes