Fold constant string null checks

[mikedn]

GT_CNS_STR can't ever produce a null value so we can constant fold "str" != null and similar expressions early, before GT_CNS_STR gets expanded into an IND(GT_CNS_INT) which is more problematic to recognize as non null.

Fixes#1368

[mikedn]

Example from #1368
Before:

G_M26703_IG01:56pushrsi 4883EC20 subrsp,32 ;; bbWeight=1 PerfScore 1.25G_M26703_IG02: 48B898314E43CE010000 movrax,0x1CE434E319848833800cmp gword ptr [rax],0 740E je SHORT G_M26703_IG06 ;; bbWeight=1 PerfScore 3.25G_M26703_IG03: 4885C9 testrcx,rcx7442je SHORT G_M26703_IG07 ;; bbWeight=0.50 PerfScore 0.63G_M26703_IG04: 488BC1 movrax,rcx ;; bbWeight=1 PerfScore 0.25G_M26703_IG05: 4883C420 addrsp,32 5E poprsi C3 ret ;; bbWeight=1 PerfScore 1.75G_M26703_IG06: 48B9C8B9D1A3FA7F0000 movrcx,0x7FFAA3D1B9C8 E81E178B5F call CORINFO_HELP_NEWSFAST…

After:

G_M26703_IG01:56pushrsi 4883EC20 subrsp,32 ;; bbWeight=1 PerfScore 1.25G_M26703_IG02: 4885C9 testrcx,rcx7409je SHORT G_M26703_IG05 ;; bbWeight=1 PerfScore 1.25G_M26703_IG03: 488BC1 movrax,rcx ;; bbWeight=1 PerfScore 0.25G_M26703_IG04: 4883C420 addrsp,32 5E poprsi C3 ret ;; bbWeight=1 PerfScore 1.75G_M26703_IG05: 48B9C8B9D2A3FA7F0000 movrcx,0x7FFAA3D2B9C8 E82E178A5F call CORINFO_HELP_NEWSFAST…

[mikedn]

Diff summary:

Total bytes of diff: -4786 (-0.02% of base)
diff is an improvement.
Top file improvements by size (bytes):
-2357 : Newtonsoft.Json.dasm (-0.38% of base)
-1622 : System.Private.CoreLib.dasm (-0.05% of base)
-594 : System.Private.Xml.dasm (-0.02% of base)
-68 : System.Reflection.Metadata.dasm (-0.02% of base)
-51 : System.Data.Common.dasm (-0.00% of base)
8 total files with size differences (8 improved, 0 regressed), 101 unchanged.
Top method regressions by size (bytes):
52 ( 1.03% of base) : System.Private.CoreLib.dasm - ManifestBuilder:CreateManifestString():String:this
1 ( 0.15% of base) : System.Private.CoreLib.dasm - MethodBase:AppendParameters(byref,ref,int)
Top method improvements by size (bytes):
-290 (-28.66% of base) : System.Private.CoreLib.dasm - LicenseInteropProxy:.ctor():this
-240 (-15.13% of base) : Newtonsoft.Json.dasm - JsonSerializerInternalReader:ReadMetadataPropertiesToken(JTokenReader,byref,byref,JsonProperty,JsonContainerContract,JsonProperty,Object,byref,byref):bool:this
-204 (-10.03% of base) : Newtonsoft.Json.dasm - JsonSerializerInternalReader:ResolvePropertyAndCreatorValues(JsonObjectContract,JsonProperty,JsonReader,Type):List`1:this
-160 (-5.91% of base) : Newtonsoft.Json.dasm - JsonTextReader:ReadStringIntoBuffer(ushort):this
-158 (-2.25% of base) : Newtonsoft.Json.dasm - <ExecuteFilter>d__4:MoveNext():bool:this (8 methods)
Top method regressions by size (percentage):
52 ( 1.03% of base) : System.Private.CoreLib.dasm - ManifestBuilder:CreateManifestString():String:this
1 ( 0.15% of base) : System.Private.CoreLib.dasm - MethodBase:AppendParameters(byref,ref,int)
Top method improvements by size (percentage):
-290 (-28.66% of base) : System.Private.CoreLib.dasm - LicenseInteropProxy:.ctor():this
-45 (-26.16% of base) : System.Private.DataContractSerialization.dasm - JsonReaderWriterFactory:CreateJsonWriter(Stream,Encoding,bool,bool):XmlDictionaryWriter
-68 (-22.59% of base) : System.Reflection.Metadata.dasm - ManagedPEBuilder:CreateSections():ImmutableArray`1:this
-49 (-20.16% of base) : System.Private.CoreLib.dasm - SynchronizationContext:GetWinRTSynchronizationContext(Object):SynchronizationContext
-96 (-18.05% of base) : System.Private.CoreLib.dasm - AppDomain:GetThreadPrincipal():IPrincipal:this
106 total methods with size differences (104 improved, 2 regressed), 146499 unchanged.

The CreateManifestString regression is due to additional CSEs. Same for AppendParameters

[EgorBo]

Heh, this PR along with #1378 should eliminate string.IsNullOrEmpty(...) completely for constant strings (it's value == null || 0u >= (uint)value.Length)

[EgorBo]

@mikedn May I ask why you closed this?
It looks quite useful for Spans of chars, e.g.:

ReadOnlySpan<char>span="hello";

currently produces a redundant nullcheck over ldstr.

[drieseng]

@EgorBo Mike decided to stop contributing. He maintains a fork where you can surely pick up great changes.

[EgorBo]

@EgorBo Mike decided to stop contributing. He maintains a fork where you can surely pick up great changes.

Ah, I see, that's sad, I learned a lot from his PRs and feedback 😢