NativeAOT: Inline tls access on linux/x64 #97413
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
|
Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch Issue DetailsBefore: G_M41298_IG02: ;; offset=0x0001
E800000000 call CORINFO_HELP_READYTORUN_THREADSTATIC_BASE
488D0D00000000 lea rcx, gword ptr [(reloc 0x40000000004210a8)] ; '"a"'
48894858 mov gword ptr [rax+0x58], rcx
;; size=16 bbWeight=1 PerfScore 2.50After: G_M41298_IG02: ;; offset=0x0008
66 data16
488D3D00000000 lea rdi, [(reloc 0x4000000000421230)]
66 data16
66 data16
48E800000000 call <unknown method>
488B18 mov rbx, gword ptr [rax]
4885DB test rbx, rbx
7412 je SHORT G_M41298_IG05
;; size=24 bbWeight=1 PerfScore 5.50
G_M41298_IG03: ;; offset=0x0020
488D0500000000 lea rax, gword ptr [(reloc 0x4000000000421278)] ; '"a"'
48894358 mov gword ptr [rbx+0x58], rax
;; size=11 bbWeight=1 PerfScore 1.50
G_M41298_IG04: ;; offset=0x002B
4883C408 add rsp, 8
5B pop rbx
5D pop rbp
C3 ret
;; size=7 bbWeight=1 PerfScore 2.25
G_M41298_IG05: ;; offset=0x0032
488BF8 mov rdi, rax
488D0500000000 lea rax, [(reloc 0x4000000000421238)]
FFD0 call rax
488BD8 mov rbx, rax
EBDD jmp SHORT G_M41298_IG03
;; size=17 bbWeight=0 PerfScore 0.00
; Total bytes of code 67, prolog size 8, PerfScore 12.75, instruction count 23, allocated bytes for code 67 (MethodHash=14305ead) for method Program:Test1() (FullOpts)The following sequence is patched by linker as seen below: 66 data16
488D3D00000000 lea rdi, [(reloc 0x4000000000421230)]
66 data16
66 data16
48E800000000 call <unknown method>
|
2 tasks
Diff results for #97413Throughput diffsThroughput diffs for linux/x64 ran on windows/x64Overall (+0.04% to +0.07%)
MinOpts (+0.06% to +0.27%)
FullOpts (+0.02% to +0.06%)
Throughput diffs for windows/x64 ran on windows/x64Overall (+0.02% to +0.05%)
MinOpts (+0.04% to +0.24%)
FullOpts (+0.02% to +0.05%)
Details here Throughput diffs for linux/x64 ran on linux/x64Overall (-0.00% to +0.01%)
MinOpts (-0.03% to +0.10%)
FullOpts (-0.01% to +0.01%)
Details here Throughput diffs for windows/x86 ran on windows/x86Overall (+0.02% to +0.04%)
MinOpts (+0.04% to +0.41%)
FullOpts (+0.01% to +0.02%)
Details here |
Diff results for #97413Throughput diffsThroughput diffs for linux/x64 ran on windows/x64Overall (+0.04% to +0.07%)
MinOpts (+0.06% to +0.27%)
FullOpts (+0.02% to +0.06%)
Throughput diffs for windows/x64 ran on windows/x64Overall (+0.02% to +0.05%)
MinOpts (+0.04% to +0.24%)
FullOpts (+0.02% to +0.05%)
Details here Throughput diffs for windows/x86 ran on windows/x86Overall (+0.02% to +0.04%)
MinOpts (+0.04% to +0.41%)
FullOpts (+0.01% to +0.02%)
Details here |
Diff results for #97413Throughput diffsThroughput diffs for linux/x64 ran on windows/x64Overall (+0.04% to +0.07%)
MinOpts (+0.06% to +0.27%)
FullOpts (+0.02% to +0.06%)
Throughput diffs for windows/x64 ran on windows/x64Overall (+0.02% to +0.05%)
MinOpts (+0.04% to +0.24%)
FullOpts (+0.02% to +0.05%)
Details here Throughput diffs for windows/x86 ran on windows/x86Overall (+0.02% to +0.04%)
MinOpts (+0.04% to +0.41%)
FullOpts (+0.01% to +0.02%)
Details here Throughput diffs for linux/x64 ran on linux/x64Overall (-0.00% to +0.01%)
MinOpts (-0.03% to +0.09%)
FullOpts (-0.01% to +0.01%)
Details here |
4 similar comments
Diff results for #97413Throughput diffsThroughput diffs for linux/x64 ran on windows/x64Overall (+0.04% to +0.07%)
MinOpts (+0.06% to +0.27%)
FullOpts (+0.02% to +0.06%)
Throughput diffs for windows/x64 ran on windows/x64Overall (+0.02% to +0.05%)
MinOpts (+0.04% to +0.24%)
FullOpts (+0.02% to +0.05%)
Details here Throughput diffs for windows/x86 ran on windows/x86Overall (+0.02% to +0.04%)
MinOpts (+0.04% to +0.41%)
FullOpts (+0.01% to +0.02%)
Details here Throughput diffs for linux/x64 ran on linux/x64Overall (-0.00% to +0.01%)
MinOpts (-0.03% to +0.09%)
FullOpts (-0.01% to +0.01%)
Details here |
Diff results for #97413Throughput diffsThroughput diffs for linux/x64 ran on windows/x64Overall (+0.04% to +0.07%)
MinOpts (+0.06% to +0.27%)
FullOpts (+0.02% to +0.06%)
Throughput diffs for windows/x64 ran on windows/x64Overall (+0.02% to +0.05%)
MinOpts (+0.04% to +0.24%)
FullOpts (+0.02% to +0.05%)
Details here Throughput diffs for windows/x86 ran on windows/x86Overall (+0.02% to +0.04%)
MinOpts (+0.04% to +0.41%)
FullOpts (+0.01% to +0.02%)
Details here Throughput diffs for linux/x64 ran on linux/x64Overall (-0.00% to +0.01%)
MinOpts (-0.03% to +0.09%)
FullOpts (-0.01% to +0.01%)
Details here |
Diff results for #97413Throughput diffsThroughput diffs for linux/x64 ran on windows/x64Overall (+0.04% to +0.07%)
MinOpts (+0.06% to +0.27%)
FullOpts (+0.02% to +0.06%)
Throughput diffs for windows/x64 ran on windows/x64Overall (+0.02% to +0.05%)
MinOpts (+0.04% to +0.24%)
FullOpts (+0.02% to +0.05%)
Details here Throughput diffs for windows/x86 ran on windows/x86Overall (+0.02% to +0.04%)
MinOpts (+0.04% to +0.41%)
FullOpts (+0.01% to +0.02%)
Details here Throughput diffs for linux/x64 ran on linux/x64Overall (-0.00% to +0.01%)
MinOpts (-0.03% to +0.09%)
FullOpts (-0.01% to +0.01%)
Details here |
Diff results for #97413Throughput diffsThroughput diffs for linux/x64 ran on windows/x64Overall (+0.04% to +0.07%)
MinOpts (+0.06% to +0.27%)
FullOpts (+0.02% to +0.06%)
Throughput diffs for windows/x64 ran on windows/x64Overall (+0.02% to +0.05%)
MinOpts (+0.04% to +0.24%)
FullOpts (+0.02% to +0.05%)
Details here Throughput diffs for windows/x86 ran on windows/x86Overall (+0.02% to +0.04%)
MinOpts (+0.04% to +0.41%)
FullOpts (+0.01% to +0.02%)
Details here Throughput diffs for linux/x64 ran on linux/x64Overall (-0.00% to +0.01%)
MinOpts (-0.03% to +0.09%)
FullOpts (-0.01% to +0.01%)
Details here |
Diff results for #97413Throughput diffsThroughput diffs for linux/x64 ran on windows/x64Overall (+0.04% to +0.07%)
MinOpts (+0.06% to +0.27%)
FullOpts (+0.02% to +0.06%)
Throughput diffs for windows/x64 ran on windows/x64Overall (+0.02% to +0.05%)
MinOpts (+0.04% to +0.24%)
FullOpts (+0.02% to +0.05%)
Details here Throughput diffs for windows/x86 ran on windows/x86Overall (+0.02% to +0.04%)
MinOpts (+0.04% to +0.41%)
FullOpts (+0.01% to +0.02%)
Details here Throughput diffs for linux/x64 ran on linux/x64Overall (-0.00% to +0.01%)
MinOpts (-0.03% to +0.09%)
FullOpts (-0.01% to +0.01%)
Details here |
Diff results for #97413Throughput diffsThroughput diffs for linux/x64 ran on windows/x64Overall (+0.04% to +0.07%)
MinOpts (+0.06% to +0.27%)
FullOpts (+0.02% to +0.06%)
Throughput diffs for windows/x64 ran on windows/x64Overall (+0.02% to +0.05%)
MinOpts (+0.04% to +0.24%)
FullOpts (+0.02% to +0.05%)
Details here Throughput diffs for windows/x86 ran on windows/x86Overall (+0.02% to +0.04%)
MinOpts (+0.04% to +0.41%)
FullOpts (+0.01% to +0.02%)
Details here Throughput diffs for linux/x64 ran on linux/x64Overall (-0.00% to +0.01%)
MinOpts (-0.03% to +0.09%)
FullOpts (-0.01% to +0.01%)
Details here |
Diff results for #97413Throughput diffsThroughput diffs for linux/x64 ran on windows/x64Overall (+0.04% to +0.07%)
MinOpts (+0.06% to +0.27%)
FullOpts (+0.02% to +0.06%)
Throughput diffs for windows/x64 ran on windows/x64Overall (+0.02% to +0.05%)
MinOpts (+0.04% to +0.24%)
FullOpts (+0.02% to +0.05%)
Details here Throughput diffs for windows/x86 ran on windows/x86Overall (+0.02% to +0.04%)
MinOpts (+0.04% to +0.41%)
FullOpts (+0.01% to +0.02%)
Details here Throughput diffs for linux/x64 ran on linux/x64Overall (-0.00% to +0.01%)
MinOpts (-0.03% to +0.09%)
FullOpts (-0.01% to +0.01%)
Details here |
|
@jkotas @VSadov @MichalStrehovsky @BruceForstall PTAL |
|
/azp run runtime-nativeaot-outerloop |
|
Azure Pipelines successfully started running 1 pipeline(s). |
|
@MihaZupan - is this way to trigger the jit-diff? |
|
Oh I see, the check I had was case-sensitive for the bot name. Fixed now |
|
Codegen looks nice. No more calls when acquiring a lock! System.Collections.Concurrent.Tests`System.Threading.Lock__EnterAndGetCurrentThreadId:
0x5555564cae70 <+0>: pushq %rbp
0x5555564cae71 <+1>: pushq %r15
0x5555564cae73 <+3>: pushq %rbx
0x5555564cae74 <+4>: subq $0x10, %rsp
0x5555564cae78 <+8>: leaq 0x20(%rsp), %rbp
0x5555564cae7d <+13>: movq %rdi, %rbx
// start fetching managed TLS
-> 0x5555564cae80 <+16>: movq %fs:0x0, %rax
0x5555564cae89 <+25>: leaq -0x108(%rax), %rax
0x5555564cae90 <+32>: movq %rax, %rdi
0x5555564cae93 <+35>: movq (%rdi), %r15 ; here r15 points to tlsRoot object
0x5555564cae96 <+38>: testq %r15, %r15 ; check if the root is initialized (or call helper to initialize)
0x5555564cae99 <+41>: je 0x5555564caee0 ; <+112> at Lock.cs:66
0x5555564cae9b <+43>: movl 0x88(%r15), %edx
// here edx has the managed thread ID - in 7 instructions!
// the rest of the Lock code follows . . .
// check if the states of the lock word and the thread ID permit fast path (typically it is the case)
0x5555564caea2 <+50>: testl %edx, %edx
0x5555564caea4 <+52>: je 0x5555564caec9 ; <+89> at Lock.cs:66
0x5555564caea6 <+54>: movl 0x14(%rbx), %eax
0x5555564caea9 <+57>: movl %eax, -0x14(%rbp)
0x5555564caeac <+60>: testb $0x3, %al
0x5555564caeae <+62>: jne 0x5555564caec9 ; <+89> at Lock.cs:66
0x5555564caeb0 <+64>: leal 0x1(%rax), %edi
0x5555564caeb3 <+67>: leaq 0x14(%rbx), %rsi
// acquire the lock
0x5555564caeb7 <+71>: lock
0x5555564caeb8 <+72>: cmpxchgl %edi, (%rsi)
0x5555564caebb <+75>: movl -0x14(%rbp), %edi
0x5555564caebe <+78>: cmpl %edi, %eax
0x5555564caec0 <+80>: jne 0x5555564caec9 ; <+89> at Lock.cs:66
0x5555564caec2 <+82>: movl %edx, 0x10(%rbx)
0x5555564caec5 <+85>: movl %edx, %eax
0x5555564caec7 <+87>: jmp 0x5555564caed6 ; <+102> at Lock.cs:69
0x5555564caec9 <+89>: movq %rbx, %rdi
0x5555564caecc <+92>: movl $0xffffffff, %esi ; imm = 0xFFFFFFFF
0x5555564caed1 <+97>: callq 0x5555564cb0a0 ; System.Threading.Lock__TryEnterSlow_0 at Lock.cs:297
0x5555564caed6 <+102>: nop
0x5555564caed7 <+103>: addq $0x10, %rsp
0x5555564caedb <+107>: popq %rbx
0x5555564caedc <+108>: popq %r15
0x5555564caede <+110>: popq %rbp
0x5555564caedf <+111>: retq
0x5555564caee0 <+112>: leaq 0x79959(%rip), %rax ; Internal.Runtime.ThreadStatics__GetInlinedThreadStaticBaseSlow at ThreadStatics.cs:40
0x5555564caee7 <+119>: callq *%rax
0x5555564caee9 <+121>: movq %rax, %r15
0x5555564caeec <+124>: jmp 0x5555564cae9b ; <+43> at Lock.cs:66 |
|
I've run a few libraries tests in a loop in various configurations (Checked/Release, ServerGC,...) and did not see any crashes or anomalies. Can't comment on the code change details, but the end result looks as expected! |
|
Tagging subscribers to this area: @agocke, @MichalStrehovsky, @jkotas Issue DetailsBefore: G_M41298_IG02: ;; offset=0x0001
E800000000 call CORINFO_HELP_READYTORUN_THREADSTATIC_BASE
488D0D00000000 lea rcx, gword ptr [(reloc 0x40000000004210a8)] ; '"a"'
48894858 mov gword ptr [rax+0x58], rcx
;; size=16 bbWeight=1 PerfScore 2.50After: G_M41298_IG02: ;; offset=0x0008
66 data16
488D3D00000000 lea rdi, [(reloc 0x4000000000421230)]
66 data16
66 data16
48E800000000 call <unknown method>
488B18 mov rbx, gword ptr [rax]
4885DB test rbx, rbx
7412 je SHORT G_M41298_IG05
;; size=24 bbWeight=1 PerfScore 5.50
G_M41298_IG03: ;; offset=0x0020
488D0500000000 lea rax, gword ptr [(reloc 0x4000000000421278)] ; '"a"'
48894358 mov gword ptr [rbx+0x58], rax
;; size=11 bbWeight=1 PerfScore 1.50
G_M41298_IG04: ;; offset=0x002B
4883C408 add rsp, 8
5B pop rbx
5D pop rbp
C3 ret
;; size=7 bbWeight=1 PerfScore 2.25
G_M41298_IG05: ;; offset=0x0032
488BF8 mov rdi, rax
488D0500000000 lea rax, [(reloc 0x4000000000421238)]
FFD0 call rax
488BD8 mov rbx, rax
EBDD jmp SHORT G_M41298_IG03
;; size=17 bbWeight=0 PerfScore 0.00
; Total bytes of code 67, prolog size 8, PerfScore 12.75, instruction count 23, allocated bytes for code 67 (MethodHash=14305ead) for method Program:Test1() (FullOpts)The following sequence is later patched by the linker as seen below: 66 data16
488D3D00000000 lea rdi, [(reloc 0x4000000000421230)]
66 data16
66 data16
48E800000000 call <unknown method>
The end result in this code:
|
jkotas
reviewed
Jan 25, 2024
BruceForstall
approved these changes
Jan 26, 2024
| gtNewIconHandleNode((size_t)threadStaticInfo.tlsGetAddrFtnPtr.handle, GTF_ICON_FTN_ADDR); | ||
| tls_get_addr_val->SetContained(); | ||
|
|
||
| // GenTreeCall* tlsRefCall = gtNewCallNode(CT_ tls_get_addr_val, TYP_I_IMPL); |
There was a problem hiding this comment.
I will do this in a follow-up PR
| // GenTreeCall* tlsRefCall = gtNewCallNode(CT_ tls_get_addr_val, TYP_I_IMPL); | ||
| GenTreeCall* tlsRefCall = gtNewIndCallNode(tls_get_addr_val, TYP_I_IMPL); | ||
| tlsRefCall->gtFlags |= GTF_TLS_GET_ADDR; | ||
| // // |
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.
Before:
After:
The following sequence is later patched by the linker as seen below:
The end result in this code: