Unpin locals #70264

aromaa · 2022-06-05T18:27:20Z

This unpins pinned locals when the pinned address is pointing to the stack. This does not cover Spans created with stackalloc's as tracking them requires more extensive state keeping.

Diffs

Example:

 ; 0 inlinees with PGO data; 1 single block inlinees; 0 inlinees without PGO data
 ; Final local variable assignments
 ;
-;  V00 loc0         [V00    ] (  2,  2   )    long  ->  [rsp+30H]   do-not-enreg[X] addr-exposed ld-addr-op
+;  V00 loc0         [V00    ] (  2,  2   )    long  ->  [rsp+20H]   do-not-enreg[X] addr-exposed ld-addr-op
 ;  V01 OutArgs      [V01    ] (  1,  1   )  lclBlk (32) [rsp+00H]   "OutgoingArgSpace"
-;  V02 tmp1         [V02    ] (  4,  4   )   byref  ->  [rsp+28H]   pinned "Inline stloc first use temp"
+;  V02 tmp1         [V02,T01] (  2,  2   )   byref  ->  rcx         "Inline stloc first use temp"
 ;* V03 tmp2         [V03    ] (  0,  0   )    long  ->  zero-ref    "Inline stloc first use temp"
 ;  V04 tmp3         [V04,T00] (  2,  4   )    long  ->  rcx         "Cast away GC"
 ;
-; Lcl frame size = 56
+; Lcl frame size = 40

 G_M59712_IG01:              ;; offset=0000H
-       4883EC38             sub      rsp, 56
+       4883EC28             sub      rsp, 40
                                                 ;; size=4 bbWeight=1    PerfScore 0.25
 G_M59712_IG02:              ;; offset=0004H
-       488D4C2430           lea      rcx, bword ptr [rsp+30H]
-       48894C2428           mov      bword ptr [rsp+28H], rcx
-       488B4C2428           mov      rcx, bword ptr [rsp+28H]
+       488D4C2420           lea      rcx, bword ptr [rsp+20H]
        48B8105C861DFF7F0000 mov      rax, 0x7FFF1D865C10
-                                                ;; size=25 bbWeight=1    PerfScore 2.75
-G_M59712_IG03:              ;; offset=001DH
+                                                ;; size=15 bbWeight=1    PerfScore 0.75
+G_M59712_IG03:              ;; offset=0013H
        FFD0                 call     rax ; Program:QueryPerformanceCounter(long):int
-       33C0                 xor      eax, eax
-       4889442428           mov      bword ptr [rsp+28H], rax
-       4889442428           mov      bword ptr [rsp+28H], rax
-       833DC6E8FE5F00       cmp      dword ptr [(reloc 0x7ffe32111cf8)], 0
+       833DDCE8016000       cmp      dword ptr [(reloc 0x7ffe32111cf8)], 0
        750A                 jne      SHORT G_M59712_IG06
-                                                ;; size=23 bbWeight=1    PerfScore 9.25
-G_M59712_IG04:              ;; offset=0034H
-       488B442430           mov      rax, qword ptr [rsp+30H]
+                                                ;; size=11 bbWeight=1    PerfScore 7.00
+G_M59712_IG04:              ;; offset=001EH
+       488B442420           mov      rax, qword ptr [rsp+20H]
                                                 ;; size=5 bbWeight=1    PerfScore 1.00
-G_M59712_IG05:              ;; offset=0039H
-       4883C438             add      rsp, 56
+G_M59712_IG05:              ;; offset=0023H
+       4883C428             add      rsp, 40
        C3                   ret
                                                 ;; size=5 bbWeight=1    PerfScore 1.25
-G_M59712_IG06:              ;; offset=003EH
-       E88D77715E           call     CORINFO_HELP_POLL_GC
+G_M59712_IG06:              ;; offset=0028H
+       E8A377745E           call     CORINFO_HELP_POLL_GC
        EBEF                 jmp      SHORT G_M59712_IG04
                                                 ;; size=7 bbWeight=0    PerfScore 0.00

-; Total bytes of code 69, prolog size 4, PerfScore 21.40, instruction count 16, allocated bytes for code 69 (MethodHash=6ac616bf) for method Program:JitDisam():long
-; ============================================================
-
+; Total bytes of code 47, prolog size 4, PerfScore 14.95, instruction count 11, allocated bytes for code 47 (MethodHash=6ac616bf) for method Program:JitDisam():long

Contributes #40553

ghost · 2022-06-05T18:27:30Z

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

Issue Details

This unpins pinned locals when the pinned address is pointing to the stack. This does not cover Spans created with stackalloc's as tracking them requires more extensive state keeping.

Example:

 ; 0 inlinees with PGO data; 1 single block inlinees; 0 inlinees without PGO data
 ; Final local variable assignments
 ;
-;  V00 loc0         [V00    ] (  2,  2   )    long  ->  [rsp+30H]   do-not-enreg[X] addr-exposed ld-addr-op
+;  V00 loc0         [V00    ] (  2,  2   )    long  ->  [rsp+20H]   do-not-enreg[X] addr-exposed ld-addr-op
 ;  V01 OutArgs      [V01    ] (  1,  1   )  lclBlk (32) [rsp+00H]   "OutgoingArgSpace"
-;  V02 tmp1         [V02    ] (  4,  4   )   byref  ->  [rsp+28H]   pinned "Inline stloc first use temp"
+;  V02 tmp1         [V02,T01] (  2,  2   )   byref  ->  rcx         "Inline stloc first use temp"
 ;* V03 tmp2         [V03    ] (  0,  0   )    long  ->  zero-ref    "Inline stloc first use temp"
 ;  V04 tmp3         [V04,T00] (  2,  4   )    long  ->  rcx         "Cast away GC"
 ;
-; Lcl frame size = 56
+; Lcl frame size = 40

 G_M59712_IG01:              ;; offset=0000H
-       4883EC38             sub      rsp, 56
+       4883EC28             sub      rsp, 40
                                                 ;; size=4 bbWeight=1    PerfScore 0.25
 G_M59712_IG02:              ;; offset=0004H
-       488D4C2430           lea      rcx, bword ptr [rsp+30H]
-       48894C2428           mov      bword ptr [rsp+28H], rcx
-       488B4C2428           mov      rcx, bword ptr [rsp+28H]
+       488D4C2420           lea      rcx, bword ptr [rsp+20H]
        48B8105C861DFF7F0000 mov      rax, 0x7FFF1D865C10
-                                                ;; size=25 bbWeight=1    PerfScore 2.75
-G_M59712_IG03:              ;; offset=001DH
+                                                ;; size=15 bbWeight=1    PerfScore 0.75
+G_M59712_IG03:              ;; offset=0013H
        FFD0                 call     rax ; Program:QueryPerformanceCounter(long):int
-       33C0                 xor      eax, eax
-       4889442428           mov      bword ptr [rsp+28H], rax
-       4889442428           mov      bword ptr [rsp+28H], rax
-       833DC6E8FE5F00       cmp      dword ptr [(reloc 0x7ffe32111cf8)], 0
+       833DDCE8016000       cmp      dword ptr [(reloc 0x7ffe32111cf8)], 0
        750A                 jne      SHORT G_M59712_IG06
-                                                ;; size=23 bbWeight=1    PerfScore 9.25
-G_M59712_IG04:              ;; offset=0034H
-       488B442430           mov      rax, qword ptr [rsp+30H]
+                                                ;; size=11 bbWeight=1    PerfScore 7.00
+G_M59712_IG04:              ;; offset=001EH
+       488B442420           mov      rax, qword ptr [rsp+20H]
                                                 ;; size=5 bbWeight=1    PerfScore 1.00
-G_M59712_IG05:              ;; offset=0039H
-       4883C438             add      rsp, 56
+G_M59712_IG05:              ;; offset=0023H
+       4883C428             add      rsp, 40
        C3                   ret
                                                 ;; size=5 bbWeight=1    PerfScore 1.25
-G_M59712_IG06:              ;; offset=003EH
-       E88D77715E           call     CORINFO_HELP_POLL_GC
+G_M59712_IG06:              ;; offset=0028H
+       E8A377745E           call     CORINFO_HELP_POLL_GC
        EBEF                 jmp      SHORT G_M59712_IG04
                                                 ;; size=7 bbWeight=0    PerfScore 0.00

-; Total bytes of code 69, prolog size 4, PerfScore 21.40, instruction count 16, allocated bytes for code 69 (MethodHash=6ac616bf) for method Program:JitDisam():long
-; ============================================================
-
+; Total bytes of code 47, prolog size 4, PerfScore 14.95, instruction count 11, allocated bytes for code 47 (MethodHash=6ac616bf) for method Program:JitDisam():long

Contributes #40553

Author:	aromaa
Assignees:	-
Labels:	`area-CodeGen-coreclr`
Milestone:	-

AndyAyersMS

Looks good, and diffs look good too.

I've suggested a few minor changes.

Also not sure how I feel about IsNotGcDef ... let me think about it a bit more.

src/coreclr/jit/lclvars.cpp

src/coreclr/jit/compiler.h

Co-authored-by: Andy Ayers <[email protected]>

AndyAyersMS · 2022-06-10T20:26:38Z

Rerunning SPMI as we likely got caught in the middle of a jit GUID change.

AndyAyersMS

Thanks.

Still trying to get a clean test run. Running into CI issues, nothing specific with your changes.

AndyAyersMS · 2022-06-10T22:04:45Z

Going to bounce this as SPMI seems confused.

AndyAyersMS · 2022-06-11T01:10:47Z

Installer build & test failure AppHost.Bundle.Tests_net7.0_x64.html seems unique.
Native AOT failure on arm64 seems less so (though no open issue for it).

Am going to see if these repro.

AndyAyersMS · 2022-06-11T16:00:58Z

Native AOT timed out (#70549)

Running assembly:System.Runtime.Tests, Version=7.0.0.0, Culture=neutral, PublicKeyToken=cc7b13ffcd2ddd51
[SKIP] System.ComponentModel.Tests.DefaultValueAttributeTests.Ctor_TypeDescriptorNotFound_ExceptionFallback
[SKIP] System.Tests.ArraySegment_Tests_string.IGenericSharedAPI_SerializeDeserialize
[SKIP] System.Text.Tests.EncodingTests.GetEncoding_FromProvider_ByCodePage_WithDisallowedEncoding_Throws
[SKIP] System.Text.Tests.EncodingTests.GetEncoding_FromProvider_ByEncodingName_WithDisallowedEncoding_Throws
[SKIP] System.Text.Tests.EncodingTests.GetEncodings_FromProvider_DoesNotContainDisallowedEncodings
[SKIP] System.Text.Tests.EncodingTests.RegisterProvider_EncodingsAreUsable

...
[EXECUTION TIMED OUT]
Exit Code:-3Executor timed out after 2700 seconds and was killed
['System.Runtime.Tests' END OF WORK ITEM LOG: Command timed out, and was killed]

This reverts commit b1c2275.

Contributes to dotnet#40553

Unpin locals

0dfecac

dotnet-issue-labeler bot added the area-CodeGen-coreclr CLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI label Jun 5, 2022

ghost added the community-contribution Indicates that the PR has been added by a community member label Jun 5, 2022

AndyAyersMS reviewed Jun 6, 2022

View reviewed changes

src/coreclr/jit/lclvars.cpp Outdated Show resolved Hide resolved

src/coreclr/jit/compiler.h Outdated Show resolved Hide resolved

aromaa and others added 2 commits June 7, 2022 18:14

Apply suggestions from code review

20ed170

Co-authored-by: Andy Ayers <[email protected]>

Fix compile

816bd0d

JulieLeeMSFT assigned aromaa and AndyAyersMS Jun 10, 2022

JulieLeeMSFT added this to the 7.0.0 milestone Jun 10, 2022

AndyAyersMS approved these changes Jun 10, 2022

View reviewed changes

AndyAyersMS closed this Jun 10, 2022

AndyAyersMS reopened this Jun 10, 2022

AndyAyersMS merged commit b1c2275 into dotnet:main Jun 11, 2022

aromaa deleted the opts/unpin-locals branch June 11, 2022 16:25

jkotas added a commit to jkotas/runtime that referenced this pull request Jun 12, 2022

Revert "Unpin locals (dotnet#70264)"

ddc0b20

This reverts commit b1c2275.

aromaa mentioned this pull request Jun 12, 2022

Revert "Unpin locals (dotnet#70264)" #70627

Merged

jkotas added a commit that referenced this pull request Jun 13, 2022

Revert "Unpin locals (#70264)" (#70627)

4f81257

This reverts commit b1c2275.

aromaa added a commit to aromaa/runtime that referenced this pull request Jun 13, 2022

Unpin locals (dotnet#70264)

6cc1125

Contributes to dotnet#40553

aromaa mentioned this pull request Jun 13, 2022

Unpin locals, attempt 2 #70655

Merged

ghost locked as resolved and limited conversation to collaborators Jul 11, 2022

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Unpin locals #70264

Unpin locals #70264

aromaa commented Jun 5, 2022 •

edited

Loading

ghost commented Jun 5, 2022

AndyAyersMS left a comment

AndyAyersMS commented Jun 10, 2022

AndyAyersMS left a comment

AndyAyersMS commented Jun 10, 2022

AndyAyersMS commented Jun 11, 2022

AndyAyersMS commented Jun 11, 2022

Unpin locals #70264

Unpin locals #70264

Conversation

aromaa commented Jun 5, 2022 • edited Loading

ghost commented Jun 5, 2022

AndyAyersMS left a comment

Choose a reason for hiding this comment

AndyAyersMS commented Jun 10, 2022

AndyAyersMS left a comment

Choose a reason for hiding this comment

AndyAyersMS commented Jun 10, 2022

AndyAyersMS commented Jun 11, 2022

AndyAyersMS commented Jun 11, 2022

aromaa commented Jun 5, 2022 •

edited

Loading