[JIT] Recognize when length is never negative for `Span<T>::.ctor(void*, int32)` #83248

xtqqczze · 2023-03-10T01:23:41Z

I noticed this for the JIT intrinsic BitOperations.PopCount, for which the return value should be treated as never negative (since #64951).

unsafe Span<byte> M0(byte* p, byte b) => new(p, b);
unsafe Span<byte> M2(byte* p, int b) => new(p, (byte)b);
unsafe Span<byte> M3(byte* p, uint mask) => new(p, BitOperations.PopCount(mask));
unsafe Span<byte> M4(byte* p, uint mask) => new(p, (byte)BitOperations.PopCount(mask));

// crossgen2 8.0.0-preview.3.23159.99+1b2eb12a711410a358bdfa93af017a34a929d4cf

C:M0(ulong,ubyte):System.Span`1[ubyte]:this:
; Emitting BLENDED_CODE for X64 CPU with AVX - Unix
       push     rax
						;; size=1 bbWeight=1 PerfScore 1.00
       mov      rax, rsi
       movzx    rdx, dl
						;; size=6 bbWeight=1 PerfScore 0.50
       add      rsp, 8
       ret      
						;; size=5 bbWeight=1 PerfScore 1.25

C:M02(ulong,int):System.Span`1[ubyte]:this:
; Emitting BLENDED_CODE for X64 CPU with AVX - Unix
       push     rax
						;; size=1 bbWeight=1 PerfScore 1.00
       movzx    rdx, dl
       test     edx, edx
       jl       SHORT G_M62596_IG04
       mov      rax, rsi
						;; size=10 bbWeight=1 PerfScore 1.75
       add      rsp, 8
       ret      
						;; size=5 bbWeight=1 PerfScore 1.25
G_M62596_IG04:
       call     [System.ThrowHelper:ThrowArgumentOutOfRangeException()]
       int3     
						;; size=7 bbWeight=0 PerfScore 0.00

C:M1(ulong,uint):System.Span`1[ubyte]:this:
; Emitting BLENDED_CODE for X64 CPU with AVX - Unix
       push     rax
						;; size=1 bbWeight=1 PerfScore 1.00
       popcnt   edx, edx
       jl       SHORT G_M53218_IG04
       mov      rax, rsi
						;; size=9 bbWeight=1 PerfScore 3.25
       add      rsp, 8
       ret      
						;; size=5 bbWeight=1 PerfScore 1.25
G_M53218_IG04:
       call     [System.ThrowHelper:ThrowArgumentOutOfRangeException()]
       int3     
						;; size=7 bbWeight=0 PerfScore 0.00

C:M2(ulong,uint):System.Span`1[ubyte]:this:
; Emitting BLENDED_CODE for X64 CPU with AVX - Unix
       push     rax
						;; size=1 bbWeight=1 PerfScore 1.00
       popcnt   edx, edx
       jl       SHORT G_M40097_IG04
       mov      rax, rsi
						;; size=9 bbWeight=1 PerfScore 3.25
       add      rsp, 8
       ret      
						;; size=5 bbWeight=1 PerfScore 1.25
G_M40097_IG04:
       call     [System.ThrowHelper:ThrowArgumentOutOfRangeException()]
       int3     
						;; size=7 bbWeight=0 PerfScore 0.00
``

The text was updated successfully, but these errors were encountered:

ghost · 2023-03-10T01:23:50Z

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

Issue Details

I noticed this for BitOperations.PopCount which is a intrinsic where the return value is treated as never negative since #64951.

unsafe Span<byte> M0(byte* p, byte b) => new(p, b);
unsafe Span<byte> M2(byte* p, int b) => new(p, (byte)b);
unsafe Span<byte> M3(byte* p, uint mask) => new(p, BitOperations.PopCount(mask));
unsafe Span<byte> M4(byte* p, uint mask) => new(p, (byte)BitOperations.PopCount(mask));

// crossgen2 8.0.0-preview.3.23159.99+1b2eb12a711410a358bdfa93af017a34a929d4cf

C:M0(ulong,ubyte):System.Span`1[ubyte]:this:
; Emitting BLENDED_CODE for X64 CPU with AVX - Unix
       push     rax
						;; size=1 bbWeight=1 PerfScore 1.00
       mov      rax, rsi
       movzx    rdx, dl
						;; size=6 bbWeight=1 PerfScore 0.50
       add      rsp, 8
       ret      
						;; size=5 bbWeight=1 PerfScore 1.25

C:M02(ulong,int):System.Span`1[ubyte]:this:
; Emitting BLENDED_CODE for X64 CPU with AVX - Unix
       push     rax
						;; size=1 bbWeight=1 PerfScore 1.00
       movzx    rdx, dl
       test     edx, edx
       jl       SHORT G_M62596_IG04
       mov      rax, rsi
						;; size=10 bbWeight=1 PerfScore 1.75
       add      rsp, 8
       ret      
						;; size=5 bbWeight=1 PerfScore 1.25
G_M62596_IG04:
       call     [System.ThrowHelper:ThrowArgumentOutOfRangeException()]
       int3     
						;; size=7 bbWeight=0 PerfScore 0.00

C:M1(ulong,uint):System.Span`1[ubyte]:this:
; Emitting BLENDED_CODE for X64 CPU with AVX - Unix
       push     rax
						;; size=1 bbWeight=1 PerfScore 1.00
       popcnt   edx, edx
       jl       SHORT G_M53218_IG04
       mov      rax, rsi
						;; size=9 bbWeight=1 PerfScore 3.25
       add      rsp, 8
       ret      
						;; size=5 bbWeight=1 PerfScore 1.25
G_M53218_IG04:
       call     [System.ThrowHelper:ThrowArgumentOutOfRangeException()]
       int3     
						;; size=7 bbWeight=0 PerfScore 0.00

C:M2(ulong,uint):System.Span`1[ubyte]:this:
; Emitting BLENDED_CODE for X64 CPU with AVX - Unix
       push     rax
						;; size=1 bbWeight=1 PerfScore 1.00
       popcnt   edx, edx
       jl       SHORT G_M40097_IG04
       mov      rax, rsi
						;; size=9 bbWeight=1 PerfScore 3.25
       add      rsp, 8
       ret      
						;; size=5 bbWeight=1 PerfScore 1.25
G_M40097_IG04:
       call     [System.ThrowHelper:ThrowArgumentOutOfRangeException()]
       int3     
						;; size=7 bbWeight=0 PerfScore 0.00
``

<table>
  <tr>
    <th align="left">Author:</th>
    <td>xtqqczze</td>
  </tr>
  <tr>
    <th align="left">Assignees:</th>
    <td>-</td>
  </tr>
  <tr>
    <th align="left">Labels:</th>
    <td>

`tenet-performance`, `area-CodeGen-coreclr`, `untriaged`

</td>
  </tr>
  <tr>
    <th align="left">Milestone:</th>
    <td>-</td>
  </tr>
</table>
</details>

xtqqczze · 2023-03-10T01:28:26Z

Reproduces for XARCH and ARM64.

JulieLeeMSFT · 2023-03-14T11:34:18Z

@TIHan PTAL.

TIHan · 2023-03-20T18:29:21Z

Did a quick investigation, IsNeverNegative does handle the PopCount intrinsics.

The problem may have something to do with value numbering with assertion prop or redundant branch opt. Will look into it further.

xtqqczze · 2023-04-06T13:41:05Z

Confirmed fixed for BitOperations.PopCount on X64.

// crossgen2 8.0.0-preview.4.23206.99+18e2c5fd9e2239a8b06fe49dbb6492d40f5e5e19

C:M0(ulong,ubyte):System.Span`1[ubyte]:this:
; Emitting BLENDED_CODE for X64 CPU with SSE2 - Unix
       push     rax
						;; size=1 bbWeight=1 PerfScore 1.00
       mov      rax, rsi
       movzx    rdx, dl
						;; size=6 bbWeight=1 PerfScore 0.50
       add      rsp, 8
       ret      
						;; size=5 bbWeight=1 PerfScore 1.25

C:M2(ulong,int):System.Span`1[ubyte]:this:
; Emitting BLENDED_CODE for X64 CPU with SSE2 - Unix
       push     rax
						;; size=1 bbWeight=1 PerfScore 1.00
       movzx    rdx, dl
       test     edx, edx
       jl       SHORT G_M26260_IG04
       mov      rax, rsi
						;; size=10 bbWeight=1 PerfScore 1.75
       add      rsp, 8
       ret      
						;; size=5 bbWeight=1 PerfScore 1.25
G_M26260_IG04:
       call     [System.ThrowHelper:ThrowArgumentOutOfRangeException()]
       int3     
						;; size=7 bbWeight=0 PerfScore 0.00

C:M3(ulong,uint):System.Span`1[ubyte]:this:
; Emitting BLENDED_CODE for X64 CPU with SSE2 - Unix
       push     rax
						;; size=1 bbWeight=1 PerfScore 1.00
       popcnt   edx, edx
       mov      rax, rsi
						;; size=7 bbWeight=1 PerfScore 2.25
       add      rsp, 8
       ret      
						;; size=5 bbWeight=1 PerfScore 1.25

C:M4(ulong,uint):System.Span`1[ubyte]:this:
; Emitting BLENDED_CODE for X64 CPU with SSE2 - Unix
       push     rax
						;; size=1 bbWeight=1 PerfScore 1.00
       popcnt   edx, edx
       mov      rax, rsi
						;; size=7 bbWeight=1 PerfScore 2.25
       add      rsp, 8
       ret      
						;; size=5 bbWeight=1 PerfScore 1.25

C:.ctor():this:
; Emitting BLENDED_CODE for X64 CPU with SSE2 - Unix
						;; size=0 bbWeight=1 PerfScore 0.00
       ret      
						;; size=1 bbWeight=1 PerfScore 1.00

https://csharp.godbolt.org/z/9Px9W3T5v

xtqqczze · 2023-04-06T13:43:59Z

However, there is still a range check for:

unsafe Span<byte> M2(byte* p, int b) => new(p, (byte)b);

xtqqczze added the tenet-performance Performance related issue label Mar 10, 2023

ghost added the untriaged New issue has not been triaged by the area owner label Mar 10, 2023

dotnet-issue-labeler bot added the area-CodeGen-coreclr CLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI label Mar 10, 2023

JulieLeeMSFT assigned TIHan Mar 14, 2023

JulieLeeMSFT removed the untriaged New issue has not been triaged by the area owner label Mar 14, 2023

JulieLeeMSFT added this to the 8.0.0 milestone Mar 14, 2023

xtqqczze changed the title ~~JIT fails to recognize when length is never negative for Span<T>::.ctor(void*, int32)~~ [JIT] Recognize when length is never negative for Span<T>::.ctor(void*, int32) Mar 14, 2023

TIHan mentioned this issue Mar 21, 2023

[JIT] ARM64/X64 - Add IsVNNeverNegative #83694

Merged

TIHan closed this as completed in #83694 Apr 4, 2023

ghost locked as resolved and limited conversation to collaborators May 6, 2023

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[JIT] Recognize when length is never negative for `Span<T>::.ctor(void*, int32)` #83248

[JIT] Recognize when length is never negative for `Span<T>::.ctor(void*, int32)` #83248

xtqqczze commented Mar 10, 2023 •

edited

Loading

ghost commented Mar 10, 2023

xtqqczze commented Mar 10, 2023 •

edited

Loading

JulieLeeMSFT commented Mar 14, 2023

TIHan commented Mar 20, 2023

xtqqczze commented Apr 6, 2023

xtqqczze commented Apr 6, 2023

[JIT] Recognize when length is never negative for Span<T>::.ctor(void*, int32) #83248

[JIT] Recognize when length is never negative for Span<T>::.ctor(void*, int32) #83248

Comments

xtqqczze commented Mar 10, 2023 • edited Loading

ghost commented Mar 10, 2023

xtqqczze commented Mar 10, 2023 • edited Loading

JulieLeeMSFT commented Mar 14, 2023

TIHan commented Mar 20, 2023

xtqqczze commented Apr 6, 2023

xtqqczze commented Apr 6, 2023

[JIT] Recognize when length is never negative for `Span<T>::.ctor(void*, int32)` #83248

[JIT] Recognize when length is never negative for `Span<T>::.ctor(void*, int32)` #83248

xtqqczze commented Mar 10, 2023 •

edited

Loading

xtqqczze commented Mar 10, 2023 •

edited

Loading