This is an archive of the discontinued LLVM Phabricator instance.

Differential D73837

Add minimal test for the ThinLtoJIT example
AbandonedPublic

Authored by sgraenitz on Feb 1 2020, 12:52 PM.

Details

Reviewers
lhames
aprantl
Summary

A lit test for the ThinLtoJIT example from D72486. It will only run in builds configured with -DLLVM_BUILD_EXAMPLES=ON (just like the tests for the Kaleidoscope examples).

The ThinLtoJIT example implements "per-module speculation" for lazy loading and compilation of bitcode. Thus, the test is not self-contained in a single module, but comes with a number .ll files in the Inputs subfolder. These files are excluded from test discovery via the local lit config file.

The test uses a workload() function to simulate CPU load. It's used to give the speculator a chance to submit modules, that will likely be executed soon, asynchronously, before execution reaches them. It's supposed to maximize code coverage and NOT a requirement for the test to succeed. I wanted to use a sleep system function first, but I didn't know how to implement it cross-platform in LLVM IR. So, I chose a function that finds the largest prime-factor of the Nth Fibonacci number. It has exponential complexity and can easily be made dependent on the input argc.

The test's call graph looks like this:

             sub1() - workload(3)
           /
--> main() -- workload(40)
           \
            \           sub21() - workload(4)
             \        /
               sub2() - sub22() - workload(5)
                      \
                        sub23() - workload(6)

All functions are defined in separate .ll files. The module that contains the main entry point is always loaded synchronously (not via the speculator). The same will happen for sub1, workload and sub2, because execution reaches their caller too soon. It will then be blocked for a bit in workload(40) and the speculator will likely be faster to discover the sub2x modules.

The ThinLtoJIT executable is invoked with -print-stats, which dumps the number of synchronously/asynchronously loaded modules on shutdown. Getting these two lines of output is what the test currently checks. The JITed code itself doesn't print anything.

Diff Detail

Event Timeline

sgraenitz created this revision.Feb 1 2020, 12:52 PM
Herald added a project: Restricted Project. · View Herald TranscriptFeb 1 2020, 12:52 PM
Herald added subscribers: dexonsmith, steven_wu, hiraditya and 2 others. · View Herald Transcript
Harbormaster completed remote builds in B45521: Diff 241896.Feb 1 2020, 1:02 PM
sgraenitz marked 3 inline comments as done.Feb 1 2020, 1:19 PM

I wrote this in C, compiled to LLVM IR (on macOS with a TOT clang) and removed a lot of bulk manually. Would you recommend to commit the C sources too? For now I put them here: https://github.com/weliveindetail/ThinLtoJitTests

llvm/test/Examples/ThinLtoJIT/minimal.test
7

The input files must be assembled into .bc files, because the ThinLTO bitcode reader is not designed for human-readable .ll files.

9

Passing the ThinLTO module index file as input (instead of the actual module files) is a feature.

12

@aprantl Did you show me the regex matchers for lit check lines last year? I cannot find the documentation for it anymore. It would be nice to grab the actual numbers here and check that their sum is 6. Is that possible? (In 99% of cases it will be 3 and 3, but in the end it's a timing issue and I don't want to rely on it.)

aprantl added inline comments.Feb 1 2020, 8:16 PM
llvm/test/Examples/ThinLtoJIT/minimal.test
12

The general documentation is here: https://www.llvm.org/docs/CommandGuide/FileCheck.html

I remember there was discussion in https://reviews.llvm.org/D49084 to add a computation mechanism to FileCheck and apparently this went in:

https://www.llvm.org/docs/CommandGuide/FileCheck.html#filecheck-numeric-substitution-blocks

!

sgraenitz updated this revision to Diff 241933.Feb 2 2020, 11:13 AM
sgraenitz marked 2 inline comments as done.

Use numeric substitution blocks to verify the sum of submitted modules is 6

llvm/test/Examples/ThinLtoJIT/minimal.test
12

Thank you! That's exactly what I was looking for.

Harbormaster completed remote builds in B45533: Diff 241933.Feb 2 2020, 11:20 AM
sgraenitz abandoned this revision.Aug 14 2020, 7:59 AM

Revision Contents

PathSize
llvm/
test/
Examples/
ThinLtoJIT/
Inputs/
56 lines
13 lines
25 lines
13 lines
13 lines
13 lines
78 lines
12 lines
5 lines

Diff 241896

llvm/test/Examples/ThinLtoJIT/Inputs/main.ll

  • This file was added.
; ModuleID = 'main.s'
source_filename = "main.c"
define i32 @main(i32 %0, i8** nocapture readonly %1) {
%3 = alloca i8*, align 8
%4 = icmp sgt i32 %0, 1
br i1 %4, label %5, label %14
5:
%6 = bitcast i8** %3 to i8*
call void @llvm.lifetime.start.p0i8(i64 8, i8* nonnull %6)
%7 = getelementptr inbounds i8*, i8** %1, i64 1
%8 = load i8*, i8** %7, align 8
%9 = call i64 @strtol(i8* %8, i8** nonnull %3, i32 10)
%10 = load i8*, i8** %3, align 8
%11 = load i8, i8* %10, align 1
%12 = icmp eq i8 %11, 0
%13 = xor i1 %12, true
call void @llvm.lifetime.end.p0i8(i64 8, i8* nonnull %6)
br i1 %12, label %14, label %24
14:
%15 = phi i64 [ %9, %5 ], [ 40, %2 ]
%16 = call i32 @sub1(i32 3)
%17 = call i32 @max(i32 0, i32 %16)
%18 = trunc i64 %15 to i32
%19 = call i32 @workload(i32 %18)
%20 = call i32 @max(i32 %17, i32 %19)
%21 = call i32 @sub2(i32 4)
%22 = call i32 @max(i32 %20, i32 %21)
%23 = icmp slt i32 %22, 1
br label %24
24:
%25 = phi i1 [ %23, %14 ], [ %13, %5 ]
%26 = zext i1 %25 to i32
ret i32 %26
}
declare void @llvm.lifetime.start.p0i8(i64 immarg, i8* nocapture)
declare i64 @strtol(i8* readonly, i8** nocapture, i32)
declare void @llvm.lifetime.end.p0i8(i64 immarg, i8* nocapture)
declare i32 @max(i32, i32)
declare i32 @sub1(i32)
declare i32 @workload(i32)
declare i32 @sub2(i32)
^0 = module: (path: "main.s", hash: (527116715, 3332069392, 338859151, 688235912, 2536759033))
^1 = gv: (name: "workload") ; guid = 2904160244401423009
^2 = gv: (name: "llvm.lifetime.start.p0i8") ; guid = 3657761528566682672
^3 = gv: (name: "sub1") ; guid = 6667511020869341723
^4 = gv: (name: "max") ; guid = 8185957744030711343
^5 = gv: (name: "strtol") ; guid = 10216030853587034628
^6 = gv: (name: "llvm.lifetime.end.p0i8") ; guid = 14311549039910520616
^7 = gv: (name: "main", summaries: (function: (module: ^0, flags: (linkage: external, notEligibleToImport: 0, live: 0, dsoLocal: 0, canAutoHide: 0), insts: 27, calls: ((callee: ^5), (callee: ^3), (callee: ^4), (callee: ^1), (callee: ^8))))) ; guid = 15822663052811949562
^8 = gv: (name: "sub2") ; guid = 15968543173600788063

llvm/test/Examples/ThinLtoJIT/Inputs/sub1.ll

  • This file was added.
; ModuleID = 'sub1.s'
source_filename = "sub1.c"
define i32 @sub1(i32 %0) {
%2 = call i32 @workload(i32 %0)
ret i32 %2
}
declare i32 @workload(i32)
^0 = module: (path: "sub1.s", hash: (716800708, 1140043336, 1672117802, 4083982944, 497481660))
^1 = gv: (name: "workload") ; guid = 2904160244401423009
^2 = gv: (name: "sub1", summaries: (function: (module: ^0, flags: (linkage: external, notEligibleToImport: 0, live: 0, dsoLocal: 0, canAutoHide: 0), insts: 2, calls: ((callee: ^1))))) ; guid = 6667511020869341723

llvm/test/Examples/ThinLtoJIT/Inputs/sub2.ll

  • This file was added.
; ModuleID = 'sub2.s'
source_filename = "sub2.c"
define i32 @sub2(i32 %0) {
%2 = call i32 @sub21(i32 %0)
%3 = call i32 @max(i32 0, i32 %2)
%4 = add nsw i32 %0, 1
%5 = call i32 @sub22(i32 %4)
%6 = call i32 @max(i32 %3, i32 %5)
%7 = add nsw i32 %0, 2
%8 = call i32 @sub23(i32 %7)
%9 = call i32 @max(i32 %6, i32 %8)
ret i32 %9
}
declare i32 @max(i32, i32)
declare i32 @sub21(i32)
declare i32 @sub22(i32)
declare i32 @sub23(i32)
^0 = module: (path: "sub2.s", hash: (1967678436, 1954842929, 2113549141, 2651172409, 3315993195))
^1 = gv: (name: "sub23") ; guid = 1031837926433398383
^2 = gv: (name: "max") ; guid = 8185957744030711343
^3 = gv: (name: "sub22") ; guid = 8601446283390290348
^4 = gv: (name: "sub21") ; guid = 12793086811415836454
^5 = gv: (name: "sub2", summaries: (function: (module: ^0, flags: (linkage: external, notEligibleToImport: 0, live: 0, dsoLocal: 0, canAutoHide: 0), insts: 9, calls: ((callee: ^4), (callee: ^2), (callee: ^3), (callee: ^1))))) ; guid = 15968543173600788063

llvm/test/Examples/ThinLtoJIT/Inputs/sub21.ll

  • This file was added.
; ModuleID = 'sub21.s'
source_filename = "sub21.c"
define i32 @sub21(i32 %0) {
%2 = call i32 @workload(i32 %0)
ret i32 %2
}
declare i32 @workload(i32)
^0 = module: (path: "sub21.s", hash: (2324013215, 2786549134, 2291471355, 584671689, 2308799624))
^1 = gv: (name: "workload") ; guid = 2904160244401423009
^2 = gv: (name: "sub21", summaries: (function: (module: ^0, flags: (linkage: external, notEligibleToImport: 0, live: 0, dsoLocal: 0, canAutoHide: 0), insts: 2, calls: ((callee: ^1))))) ; guid = 12793086811415836454

llvm/test/Examples/ThinLtoJIT/Inputs/sub22.ll

  • This file was added.
; ModuleID = 'sub22.s'
source_filename = "sub22.c"
define i32 @sub22(i32 %0) {
%2 = call i32 @workload(i32 %0)
ret i32 %2
}
declare i32 @workload(i32)
^0 = module: (path: "sub22.s", hash: (4057258347, 3191877851, 3120355814, 3202334300, 790339932))
^1 = gv: (name: "workload") ; guid = 2904160244401423009
^2 = gv: (name: "sub22", summaries: (function: (module: ^0, flags: (linkage: external, notEligibleToImport: 0, live: 0, dsoLocal: 0, canAutoHide: 0), insts: 2, calls: ((callee: ^1))))) ; guid = 8601446283390290348

llvm/test/Examples/ThinLtoJIT/Inputs/sub23.ll

  • This file was added.
; ModuleID = 'sub23.s'
source_filename = "sub23.c"
define i32 @sub23(i32 %0) {
%2 = call i32 @workload(i32 %0)
ret i32 %2
}
declare i32 @workload(i32)
^0 = module: (path: "sub23.s", hash: (1187217899, 2239201331, 3030240015, 3752221195, 912650771))
^1 = gv: (name: "sub23", summaries: (function: (module: ^0, flags: (linkage: external, notEligibleToImport: 0, live: 0, dsoLocal: 0, canAutoHide: 0), insts: 2, calls: ((callee: ^2))))) ; guid = 1031837926433398383
^2 = gv: (name: "workload") ; guid = 2904160244401423009

llvm/test/Examples/ThinLtoJIT/Inputs/workload.ll

  • This file was added.
; ModuleID = 'workload.s'
source_filename = "workload.c"
define i32 @max(i32 %0, i32 %1) {
%3 = icmp sgt i32 %0, %1
%4 = select i1 %3, i32 %0, i32 %1
ret i32 %4
}
define i32 @workload(i32 %0) {
%2 = icmp slt i32 %0, 64
%3 = select i1 %2, i32 %0, i32 64
%4 = icmp sgt i32 %3, 1
br i1 %4, label %5, label %12
5:
%6 = phi i32 [ %10, %5 ], [ 1, %1 ]
%7 = phi i64 [ %9, %5 ], [ 1, %1 ]
%8 = phi i64 [ %7, %5 ], [ 0, %1 ]
%9 = add i64 %7, %8
%10 = add nuw nsw i32 %6, 1
%11 = icmp slt i32 %10, %3
br i1 %11, label %5, label %12
12:
%13 = phi i64 [ 1, %1 ], [ %9, %5 ]
%14 = call fastcc i32 @fact(i64 %13)
ret i32 %14
}
define internal fastcc i32 @fact(i64 %0) {
%2 = icmp ult i64 %0, 2
br i1 %2, label %26, label %3
3:
%4 = phi i32 [ %23, %22 ], [ 0, %1 ]
%5 = phi i64 [ %24, %22 ], [ 2, %1 ]
%6 = urem i64 %0, %5
%7 = icmp eq i64 %6, 0
br i1 %7, label %8, label %22
8:
%9 = lshr i64 %5, 1
%10 = icmp ult i64 %5, 4
br i1 %10, label %18, label %14
11:
%12 = add nuw i64 %15, 1
%13 = icmp ult i64 %15, %9
br i1 %13, label %14, label %18
14:
%15 = phi i64 [ %12, %11 ], [ 2, %8 ]
%16 = urem i64 %5, %15
%17 = icmp eq i64 %16, 0
br i1 %17, label %18, label %11
18:
%19 = phi i1 [ true, %8 ], [ true, %11 ], [ false, %14 ]
%20 = trunc i64 %5 to i32
%21 = select i1 %19, i32 %20, i32 %4
br label %22
22:
%23 = phi i32 [ %21, %18 ], [ %4, %3 ]
%24 = add i64 %5, 1
%25 = icmp ugt i64 %24, %0
br i1 %25, label %26, label %3
26:
%27 = phi i32 [ 0, %1 ], [ %23, %22 ]
ret i32 %27
}
^0 = module: (path: "workload.s", hash: (3100996161, 646883963, 3338709300, 3710134105, 1603428348))
^1 = gv: (name: "workload", summaries: (function: (module: ^0, flags: (linkage: external, notEligibleToImport: 0, live: 0, dsoLocal: 0, canAutoHide: 0), insts: 14, funcFlags: (readNone: 1, readOnly: 0, noRecurse: 1, returnDoesNotAlias: 0, noInline: 0, alwaysInline: 0), calls: ((callee: ^3))))) ; guid = 2904160244401423009
^2 = gv: (name: "max", summaries: (function: (module: ^0, flags: (linkage: external, notEligibleToImport: 0, live: 0, dsoLocal: 0, canAutoHide: 0), insts: 3, funcFlags: (readNone: 1, readOnly: 0, noRecurse: 1, returnDoesNotAlias: 0, noInline: 0, alwaysInline: 0)))) ; guid = 8185957744030711343
^3 = gv: (name: "fact", summaries: (function: (module: ^0, flags: (linkage: internal, notEligibleToImport: 0, live: 0, dsoLocal: 1, canAutoHide: 0), insts: 27, funcFlags: (readNone: 1, readOnly: 0, noRecurse: 1, returnDoesNotAlias: 0, noInline: 0, alwaysInline: 0)))) ; guid = 17650377645318605574

llvm/test/Examples/ThinLtoJIT/minimal.test

  • This file was added.
# RUN: llvm-as %S/Inputs/main.ll -o main.bc
# RUN: llvm-as %S/Inputs/sub1.ll -o sub1.bc
# RUN: llvm-as %S/Inputs/sub2.ll -o sub2.bc
# RUN: llvm-as %S/Inputs/sub21.ll -o sub21.bc
# RUN: llvm-as %S/Inputs/sub22.ll -o sub22.bc
# RUN: llvm-as %S/Inputs/sub23.ll -o sub23.bc
# RUN: llvm-as %S/Inputs/workload.ll -o workload.bc
sgraenitzAuthorUnsubmitted
Done
ReplyInline Actions

The input files must be assembled into .bc files, because the ThinLTO bitcode reader is not designed for human-readable .ll files.

sgraenitz: The input files must be assembled into .bc files, because the ThinLTO bitcode reader is not…
# RUN: llvm-lto --thinlto -o index main.bc sub1.bc sub2.bc sub21.bc sub22.bc sub23.bc workload.bc
# RUN: ThinLtoJIT -print-stats index.thinlto.bc 2>&1 | FileCheck %s
sgraenitzAuthorUnsubmitted
Done
ReplyInline Actions

Passing the ThinLTO module index file as input (instead of the actual module files) is a feature.

sgraenitz: Passing the ThinLTO module index file as input (instead of the actual module files) is a…
# CHECK: Modules submitted asynchronously: {{[0-6]}}
# CHECK: Modules submitted synchronously: {{[0-6]}}
sgraenitzAuthorUnsubmitted
Done
ReplyInline Actions

@aprantl Did you show me the regex matchers for lit check lines last year? I cannot find the documentation for it anymore. It would be nice to grab the actual numbers here and check that their sum is 6. Is that possible? (In 99% of cases it will be 3 and 3, but in the end it's a timing issue and I don't want to rely on it.)

sgraenitz: @aprantl Did you show me the regex matchers for lit check lines last year? I cannot find the…
aprantlUnsubmitted
Done
ReplyInline Actions

The general documentation is here: https://www.llvm.org/docs/CommandGuide/FileCheck.html

I remember there was discussion in https://reviews.llvm.org/D49084 to add a computation mechanism to FileCheck and apparently this went in:

https://www.llvm.org/docs/CommandGuide/FileCheck.html#filecheck-numeric-substitution-blocks

!

aprantl: The general documentation is here: https://www.llvm.org/docs/CommandGuide/FileCheck.html I…
sgraenitzAuthorUnsubmitted
Done
ReplyInline Actions

Thank you! That's exactly what I was looking for.

sgraenitz: Thank you! That's exactly what I was looking for.

llvm/test/Examples/lit.local.cfg

if not config.build_examples or sys.platform in ['win32']: if not config.build_examples or sys.platform in ['win32']:
config.unsupported = True config.unsupported = True
No newline at end of file
# Test discovery should ignore subdirectories that contain auxiliary inputs.
config.excludes = ['Inputs']