JIT: Faster vector == Vector128.Zero on arm64 #65632
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
dotnet-issue-labeler
bot
added
the
area-CodeGen-coreclr
ghost
assigned 
|
Tagging subscribers to this area: @JulieLeeMSFT Issue DetailsCloses #63829 static bool IsZero(Vector128<int> vec) => vec == Vector128<int>.Zero;Codegen diff: ; Assembly listing for method IsZero(System.Runtime.Intrinsics.Vector128`1[Int32]):bool
stp fp, lr, [sp,#-16]!
mov fp, sp
- cmeq v16.4s, v0.4s, #0
- uminv b16, v16.16b
+ umaxv b16, v0.16b
umov w0, v16.b[0]
cmp w0, #0
- cset x0, ne
+ cset x0, eq
ldp fp, lr, [sp],#16
ret lr
-; Total bytes of code 36
+; Total bytes of code 32This is needed for faster IndexOf from #63285. Also, #65288 relies on it. Perf_Regex_Industry_RustLang_Sherlock Benchmark:
|
|
PTAL @echesakovMSFT @TIHan |
|
cc @vargaz @fanyang-mono the Test I added in this PR crashes on |
|
This will fix it: |
EgorBo
requested review from
vargaz,
lambdageek,
SamMonoRT and
imhameed
as code owners
TIHan
reviewed
Feb 21, 2022
TIHan
reviewed
Feb 21, 2022
TIHan
approved these changes
Feb 21, 2022
|
I had created an issue to clean up the code for type checks of vector elements. Haven't get to it yet. (#65318) |
echesakov
reviewed
Feb 22, 2022
src/coreclr/jit/lowerarmarch.cpp
Outdated
| if (!varTypeIsFloating(simdBaseType) && (op != nullptr)) | ||
| { | ||
| GenTree* cmp = | ||
| comp->gtNewSimdHWIntrinsicNode(simdType, op, NI_AdvSimd_Arm64_MaxAcross, CORINFO_TYPE_UBYTE, simdSize); |
There was a problem hiding this comment.
According to Arm® Cortex®-A76 Software Optimization Guide:
UMAXV, 16B has Exec latency 6 and Execution throughput 1/2
while UMAXV, 4H/4S has Exec latency 3 and Execution throughput 1
Do we want CORINFO_TYPE_USHORT/CORINFO_TYPE_UINT as a base type instead?
There was a problem hiding this comment.
Sure, let's change, Although, I had this in mind when I was benchmarking it and saw zero difference, but I now I see why:
(Apple M1, Firestorm core)
echesakov
approved these changes
Feb 22, 2022
This was referenced Feb 23, 2022
|
Improvements in dotnet/perf-autofiling-issues#3833 and dotnet/perf-autofiling-issues#3829 |
|
wow, it's more than I expected |
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Closes #63829
Codegen diff:
; Assembly listing for method IsZero(System.Runtime.Intrinsics.Vector128`1[Int32]):bool stp fp, lr, [sp,#-16]! mov fp, sp - cmeq v16.4s, v0.4s, #0 - uminv b16, v16.16b - umov w0, v16.b[0] + umaxv b16, v0.16b + umov w0, v16.s[0] cmp w0, #0 - cset x0, ne + cset x0, eq ldp fp, lr, [sp],#16 ret lr -; Total bytes of code 36 +; Total bytes of code 32This is needed for faster IndexOf from #63285. Also, #65288 relies on it.
Perf_Regex_Industry_RustLang_Sherlock Benchmark:
Diffs