RuyJIT: Array.Length / cns can be converted to unsigned div (mod)

[EgorBo]

Array.Length is a signed integer but since it should not be negative we can optimize some arithmetic operations, e.g. division (which is faster for unsigned), e.g.:

static int TestDiv(int[] array) => array.Length / 4; static int TestMod(int[] array) => array.Length % 4;

Current codegen:

; Method Foo:TestDiv(System.Int32[]):int  8B4108 mov eax, dword ptr [rcx+8]  8BD0 mov edx, eax  C1FA1F sar edx, 31  83E203 and edx, 3  03C2 add eax, edx  C1F802 sar eax, 2  C3 ret  ; Total bytes of code: 17 ; Method Foo:TestMod(System.Int32[]):int  8B4108 mov eax, dword ptr [rcx+8]  8BD0 mov edx, eax  C1FA1F sar edx, 31  83E203 and edx, 3  03D0 add edx, eax  83E2FC and edx, -4  2BC2 sub eax, edx  C3 ret ; Total bytes of code: 19

New codegen:

; Method Foo:TestDiv(System.Int32[]):int  8B4108 mov eax, dword ptr [rcx+8]  C1E802 shr eax, 2  C3 ret  ; Total bytes of code: 7 ; Method Foo:TestMod(System.Int32[]):int  8B4108 mov eax, dword ptr [rcx+8]  83E003 and eax, 3  C3 ret  ; Total bytes of code: 7

(non-power of two constants are also optimized better than for signed div)

Jit-diff (-f -pmi)

Summary of Code Size diffs: (Lower is better) Total bytes of diff: -558 (-0.00% of base) diff is an improvement. Top file improvements (bytes): -98 : System.Collections.Concurrent.dasm (-0.03% of base) -84 : Microsoft.Diagnostics.Tracing.TraceEvent.dasm (-0.00% of base) -56 : Microsoft.CodeAnalysis.dasm (-0.00% of base) -49 : System.Private.Xml.Linq.dasm (-0.03% of base) -39 : System.Private.CoreLib.dasm (-0.00% of base) -32 : System.DirectoryServices.AccountManagement.dasm (-0.01% of base) -28 : System.Security.AccessControl.dasm (-0.03% of base) -28 : System.Security.Cryptography.Xml.dasm (-0.02% of base) -25 : Microsoft.CodeAnalysis.VisualBasic.dasm (-0.00% of base) -23 : Microsoft.CodeAnalysis.CSharp.dasm (-0.00% of base) -23 : System.Private.Xml.dasm (-0.00% of base) -17 : System.Text.Json.dasm (-0.00% of base) -11 : System.ComponentModel.Annotations.dasm (-0.03% of base) -10 : System.Private.DataContractSerialization.dasm (-0.00% of base) -7 : Microsoft.Diagnostics.FastSerialization.dasm (-0.01% of base) -7 : System.IO.Ports.dasm (-0.02% of base) -6 : System.Net.Http.dasm (-0.00% of base) -5 : System.Diagnostics.PerformanceCounter.dasm (-0.01% of base) -5 : System.Diagnostics.Process.dasm (-0.01% of base) -5 : System.Security.Cryptography.Pkcs.dasm (-0.00% of base) 20 total files with Code Size differences (20 improved, 0 regressed), 142 unchanged. 

Full jit-diff: https://gist.github.com/EgorBo/4e784cb90b1222bc928b1a27ccea7efa

[EgorBo]

@dotnet/jit-contrib Please take a look, the change is pretty harmless, isn't it?

[kunalspathak]

consider adding few cases for long.

[kunalspathak]

ArrayLengthDiv_cns0 [](start = 66, length = 19)

should we add cases where the operation could become CSE candidate?

 .. = array.Length / 3; ... .. = array.Length / 3; 

[sandreenko]

I do not see any potential issues here, because we don't change tree->gtType, so I can't construct a test where this transformation can lead to add new CSE candidates or reject an existing.
It would be nice to improve the test even more, but I would not block the PR because of that.

[sandreenko]

LGTM, thanks @EgorBo