Compiling a Cilk/C++ program with Cilkscale instrumentation prints LLVM IR to stderr #129
Description
Describe the bug
When I compile a Cilk/C++ program with Cilkscale instrumentation (-fcilktool=cilkscale), LLVM IR for my Cilk function is printed to stderr. Compilation seems to proceed OK otherwise: the binary runs and passes correctness tests, and Cilkscale outputs its work-span analysis report as expected.
The same is true when compiling with Cilkscale-benchmark instrumentation (-fcilktool=cilkscale-benchmark).
I do not see this behavior with Cilk/C code or if I use OpenCilk 1.0.
Expected behavior
No printing of LLVM IR when compiling a Cilk/C++ program with Cilkscale instrumentation.
OpenCilk version
- Release version: 2.0
- Distribution method: built from source (tags
opencilk/v2.0for all repos)
System information
- OS: Ubuntu 20.04 (via WSL2 on Windows 10)
- CPU: Intel Core i7-10875H
Steps to reproduce (include relevant output)
Minimal working example, hello.cpp:
#include <iostream>
#include <cilk/cilk.h>
void hello() {
std::cout << "Hello world!\n";
}
int main() {
cilk_scope {
cilk_spawn hello();
hello();
}
return 0;
}Compile with Cilkscale instrumentation:
$ /opt/opencilk-2/bin/clang++ hello.cpp -fopencilk -fcilktool=cilkscale -o hello
Sync sync within %syncreg, label %sync.continuehas unwind
; Function Attrs: mustprogress noinline norecurse optnone uwtable
define dso_local noundef i32 @main() #5 personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) {
entry:
%retval = alloca i32, align 4
%syncreg = call token @llvm.syncregion.start()
%exn.slot10 = alloca i8*, align 8
%ehselector.slot11 = alloca i32, align 4
store i32 0, i32* %retval, align 4
%0 = call token @llvm.tapir.runtime.start()
br label %entry.split
entry.split: ; preds = %entry
%1 = call token @llvm.taskframe.create()
%exn.slot3 = alloca i8*, align 8
%ehselector.slot4 = alloca i32, align 4
detach within %syncreg, label %det.achd, label %det.cont unwind label %lpad2
det.achd: ; preds = %entry.split
%exn.slot = alloca i8*, align 8
%ehselector.slot = alloca i32, align 4
call void @llvm.taskframe.use(token %1)
invoke void @_Z5hellov()
to label %invoke.cont unwind label %lpad
invoke.cont: ; preds = %det.achd
reattach within %syncreg, label %det.cont
det.cont: ; preds = %entry.split, %invoke.cont
invoke void @_Z5hellov()
to label %invoke.cont12 unwind label %lpad9.csi-split.csi-split-lp
invoke.cont12: ; preds = %det.cont
sync within %syncreg, label %sync.continue
sync.continue: ; preds = %invoke.cont12
invoke void @llvm.sync.unwind(token %syncreg)
to label %invoke.cont13 unwind label %lpad9.csi-split-lp
invoke.cont13: ; preds = %sync.continue
call void @llvm.tapir.runtime.end(token %0)
ret i32 0
lpad: ; preds = %det.achd
%2 = landingpad { i8*, i32 }
cleanup
%3 = extractvalue { i8*, i32 } %2, 0
store i8* %3, i8** %exn.slot, align 8
%4 = extractvalue { i8*, i32 } %2, 1
store i32 %4, i32* %ehselector.slot, align 4
%exn = load i8*, i8** %exn.slot, align 8
%sel = load i32, i32* %ehselector.slot, align 4
%lpad.val = insertvalue { i8*, i32 } undef, i8* %exn, 0
%lpad.val1 = insertvalue { i8*, i32 } %lpad.val, i32 %sel, 1
invoke void @llvm.detached.rethrow.sl_p0i8i32s(token %syncreg, { i8*, i32 } %lpad.val1)
to label %unreachable unwind label %lpad2
lpad2: ; preds = %entry.split, %lpad
%5 = landingpad { i8*, i32 }
cleanup
%6 = extractvalue { i8*, i32 } %5, 0
store i8* %6, i8** %exn.slot3, align 8
%7 = extractvalue { i8*, i32 } %5, 1
store i32 %7, i32* %ehselector.slot4, align 4
br label %ehcleanup
ehcleanup: ; preds = %lpad2
%exn5 = load i8*, i8** %exn.slot3, align 8
%sel6 = load i32, i32* %ehselector.slot4, align 4
%lpad.val7 = insertvalue { i8*, i32 } undef, i8* %exn5, 0
%lpad.val8 = insertvalue { i8*, i32 } %lpad.val7, i32 %sel6, 1
invoke void @llvm.taskframe.resume.sl_p0i8i32s(token %1, { i8*, i32 } %lpad.val8)
to label %unreachable unwind label %lpad9.csi-split.csi-split
lpad9.csi-split-lp: ; preds = %sync.continue
%lpad.csi-split-lp = landingpad { i8*, i32 }
cleanup
br label %lpad9
lpad9.csi-split.csi-split-lp: ; preds = %det.cont
%lpad.csi-split-lp21 = landingpad { i8*, i32 }
cleanup
br label %lpad9.csi-split
lpad9.csi-split.csi-split: ; preds = %ehcleanup
%lpad.csi-split22 = landingpad { i8*, i32 }
cleanup
br label %lpad9.csi-split
lpad9.csi-split: ; preds = %lpad9.csi-split.csi-split, %lpad9.csi-split.csi-split-lp
%lpad.phi23 = phi { i8*, i32 } [ %lpad.csi-split-lp21, %lpad9.csi-split.csi-split-lp ], [ %lpad.csi-split22, %lpad9.csi-split.csi-split ]
br label %lpad9
lpad9: ; preds = %lpad9.csi-split, %lpad9.csi-split-lp
%lpad.phi = phi { i8*, i32 } [ %lpad.csi-split-lp, %lpad9.csi-split-lp ], [ %lpad.phi23, %lpad9.csi-split ]
%8 = extractvalue { i8*, i32 } %lpad.phi, 0
store i8* %8, i8** %exn.slot10, align 8
%9 = extractvalue { i8*, i32 } %lpad.phi, 1
store i32 %9, i32* %ehselector.slot11, align 4
call void @llvm.tapir.runtime.end(token %0)
br label %eh.resume
eh.resume: ; preds = %lpad9
%exn15 = load i8*, i8** %exn.slot10, align 8
%sel16 = load i32, i32* %ehselector.slot11, align 4
%lpad.val17 = insertvalue { i8*, i32 } undef, i8* %exn15, 0
%lpad.val18 = insertvalue { i8*, i32 } %lpad.val17, i32 %sel16, 1
resume { i8*, i32 } %lpad.val18
unreachable: ; preds = %ehcleanup, %lpad
unreachable
}
Run:
$ ./hello
Hello world!
Hello world!
tag,work (seconds),span (seconds),parallelism,burdened_span (seconds),burdened_parallelism
,0.0030673,0.0030097,1.01914,0.00301595,1.01703