[llvm-bugs] [Bug 121135] ARM32: Missing gprof support
Issue
121135
Summary
ARM32: Missing gprof support
Labels
backend:ARM
Assignees
Reporter
wzssyqa
```
int fib(int n) {return n <=1 ? 1 : fib(n-1) + fib(n-2); }
int main() {fib(10); return 0;}
```
How to test native:
```
bin/clang -pg test.c && ./a.out && gprof -b a.out gmon.out
```
How to test cross with qemu-user
```
apt install gcc-arm-linux-gnueabi gcc-arm-linux-gnueabihf
bin/clang --target=arm-linux-gnueabi -pg test.c -static && ./a.out && gprof -b a.out gmon.out
bin/clang --target=arm-linux-gnueabihf -pg test.c -static && ./a.out && gprof -b a.out gmon.out
```
Known problem:
```
$ bin/clang --target=arm-linux-gnueabi -pg test.c -static && ./a.out && gprof -b a.out gmon.out
Flat profile:
Each sample counts as 0.01 seconds.
no time accumulated
% cumulative self self total
time seconds secondscalls Ts/call Ts/call name
Call graph
granularity: each sample hit covers 2 byte(s) no time propagated
index % timeself childrencalled name
1 main [130]
[130]0.00.00 0.00 0+1 main [130]
1 main [130]
---
Index by function name
```
```
$ bin/clang --target=arm-linux-gnueabihf -pg test.c -static && ./a.out && gprof -b a.out gmon.out
/usr/bin/arm-linux-gnueabihf-ld: /tmp/n-9e8be5.o: in function `fib':
n.c:(.text+0xc): undefined reference to `mcount'
/usr/bin/arm-linux-gnueabihf-ld: /tmp/n-9e8be5.o: in function `main':
n.c:(.text+0x4c): undefined reference to `mcount'
clang: error: linker command failed with exit code 1 (use -v to see invocation)
```
___
llvm-bugs mailing list
llvm-bugs@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
[llvm-bugs] [Bug 121125] [Flang] Performance issues in assignments
Issue 121125 Summary [Flang] Performance issues in assignments Labels performance, flang Assignees Reporter yus3710-fj [The measurement](https://discourse.llvm.org/t/optimization-of-complex-number-division/83468/1) of the performance of 627.cam4_s shows that assignments for non-contiguous arrays spends time. I checked the performance of some kinds of assignments and found issues. This issue intends to track the status of those issues. Any help is welcome. ## Issue list - [ ] [Flang] Assignments for local non-contiguous allocatable arrays are slower than GFortran - [ ] [Flang] Assignments for arrays of a derived type are slower than GFortran - [ ] [Flang] Assignments for overlapping arrays of a derived type with FORALL/WHERE are slower than GFortran - [ ] [Flang] Assignments for integer arrays with FORALL/WHERE are slower than GFortran - [ ] [Flang] Defined assignments are slower than GFortran - [ ] [Flang] Assignments for complex variables are slower than GFortran - [ ] [Flang] Assignments for variables of a derived type are slower than GFortran - [ ] [Flang] Pointer assignments for scalar pointers are slower than GFortran - [ ] [Flang] Pointer assignments for array pointers are slower than GFortran ___ llvm-bugs mailing list llvm-bugs@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
[llvm-bugs] [Bug 121126] [Flang] Assignments for local non-contiguous allocatable arrays are slower than GFortran
Issue 121126 Summary [Flang] Assignments for local non-contiguous allocatable arrays are slower than GFortran Labels flang Assignees Reporter yus3710-fj ___ llvm-bugs mailing list llvm-bugs@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
[llvm-bugs] [Bug 121136] Missing gprof support on some target
Issue
121136
Summary
Missing gprof support on some target
Labels
new issue
Assignees
Reporter
wzssyqa
Test code
```
; bin/clang -pg test.c && ./a.out && gprof -b a.out gmon.out
int fib(int n) {return n <=1 ? 1 : fib(n-1) + fib(n-2); }
int main() {fib(10); return 0;}
```
Known workable platforms
```
X86-64
X86-32
MIPS32
MIPS64
SPARC64
PowerPC64LE
```
Known unworkable platforms
```
AARCH64 https://github.com/llvm/llvm-project/pull/121107
LoongArch64 https://github.com/llvm/llvm-project/pull/121107
RISC-V64 https://github.com/llvm/llvm-project/pull/121107
ARM32: https://github.com/llvm/llvm-project/issues/121135
S390X/SystemZ
```
___
llvm-bugs mailing list
llvm-bugs@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
[llvm-bugs] [Bug 121137] SystemZ: Missing gprof support
Issue
121137
Summary
SystemZ: Missing gprof support
Labels
new issue
Assignees
Reporter
wzssyqa
```
int fib(int n) {return n <=1 ? 1 : fib(n-1) + fib(n-2); }
int main() {fib(10); return 0;}
```
How to test native:
```
bin/clang -pg test.c && ./a.out && gprof -b a.out gmon.out
```
How to test cross with qemu-user
```
apt install gcc-s390x-linux-gnu qemu-user-static
bin/clang --target=s390x-linux-gnu -pg test.c -static && ./a.out && gprof -b a.out gmon.out
```
Known problem:
```
$ bin/clang --target=s390x-linux-gnu -pg n.c -O2 -static && ./a.out && gprof -b ./a.out gmon.out
Flat profile:
Each sample counts as 0.01 seconds.
no time accumulated
% cumulative self self total
time seconds secondscalls Ts/call Ts/call name
```
___
llvm-bugs mailing list
llvm-bugs@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
[llvm-bugs] [Bug 121142] Can't specify a physical register in inline assembly issue
Issue
121142
Summary
Can't specify a physical register in inline assembly issue
Labels
new issue
Assignees
Reporter
LuoYuanke
It seems below inline assembly fails to be compiled with clang, but can be successful compiled with GCC. It can be duplicated in https://godbolt.org/z/xhrq6YEvT
```
void foo() {
int src = ""
int dst, dst2;
asm("movl %1, %0\n\t"
"addl $1, %0"
: "=eax"(dst)
: "i"(1));
}
```
___
llvm-bugs mailing list
llvm-bugs@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
[llvm-bugs] [Bug 121108] [Clang] Missing flag to allow using `[[gnu::init_priority]]` in source files
Issue 121108 Summary [Clang] Missing flag to allow using `[[gnu::init_priority]]` in source files Labels clang Assignees Reporter philnik777 Currently it's not possible to use `[[gnu::init_priority(100)]]` in source files except by using line control markers. It would be way nicer if there was a diagnostic group to disable this with the diagnostic being an error by default. GCC currently accepts this with a warning without a group instead, which is also annoying, but not as bad as making it only possible with a hack. ___ llvm-bugs mailing list llvm-bugs@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
[llvm-bugs] [Bug 121105] [LLVM] Inlining breaks __builtin_expect branch weight metadata
Issue
121105
Summary
[LLVM] Inlining breaks __builtin_expect branch weight metadata
Labels
new issue
Assignees
Reporter
sillycross
Minimal repro:
```
#define likely(expr) __builtin_expect((expr) != 0, 1)
bool f(int** p) {
return *p != nullptr && **p == 1;
}
void h();
void i();
void g(int** p) {
if (likely(f(p))) h(); else i();
}
```
Compile: `clang++ test.cpp -O3 -emit-llvm -S` gives following result ([https://godbolt.org/z/81rrxbhab](https://godbolt.org/z/81rrxbhab)):
```
... omitted ...
%6 = icmp eq i32 %5, 1# <- **p == 1 check
br i1 %6, label %7, label %8, !prof !49 # <- metadata broken
... omitted ...
# -2147483648 means invalid. Should be "i32 2000, i32 1" (due to __builtin_expect)
!49 = !{!"branch_weights", !"expected", i32 -2147483648, i32 0}
```
After bisecting, this bug is introduced by commit [#a1b78fb](https://github.com/llvm/llvm-project/commit/a1b78fb929fccf96acaa0212cf68fee82298e747), authored by Evgeniy Brevnov (ybrev...@azul.com) and reviewed by @xortator (Max Kazantsev). Reverting the commit fixes the issue.
Ironically, this commit is already breaking existing tests by producing invalid metadata (see [#a1b78fb](https://github.com/llvm/llvm-project/commit/a1b78fb929fccf96acaa0212cf68fee82298e747) at the end). Instead of fixing the issues, the author changed the expected test output to the invalid metadata.
___
llvm-bugs mailing list
llvm-bugs@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
[llvm-bugs] [Bug 121103] [AArch64][RISCV][Loongarch] return address should be passed into `_mcount`
Issue
121103
Summary
[AArch64][RISCV][Loongarch] return address should be passed into `_mcount`
Labels
backend:AArch64,
backend:RISC-V,
miscompilation,
backend:loongarch
Assignees
dtcxzyw
Reporter
dtcxzyw
Reproducer:
```
; bin/clang -pg test.c && ./a.out && gprof -b a.out gmon.out
int fib(int n) {return n <=1 ? 1 : fib(n-1) + fib(n-2); }
int main() {fib(10); return 0;}
```
The call graph is missing if it is compiled with clang:
```
Flat profile:
Each sample counts as 0.01 seconds.
no time accumulated
% cumulative self self total
time seconds secondscalls Ts/call Ts/call name
```
With gcc, it works as expected:
```
Flat profile:
Each sample counts as 0.01 seconds.
no time accumulated
% cumulative self self total
time seconds secondscalls Ts/call Ts/call name
0.00 0.00 0.001 0.00 0.00 fib
Call graph
granularity: each sample hit covers 2 byte(s) no time propagated
index % timeself childrencalled name
176 fib [1]
0.000.00 1/1 main [6]
[1] 0.00.000.00 1+176 fib [1]
176 fib [1]
---
Index by function name
[1] fib
```
This issue can be reproduced on AArch64/RISCV/Loongarch platforms.
After investing this issue, I found that we should pass the return address into `_mcount`:
https://github.com/gcc-mirror/gcc/blob/c6b7d0343f2c0eb3102341c9326fe314a047bee6/gcc/config/riscv/riscv.h#L466-L473
https://github.com/bminor/glibc/blob/757ac24f8f2344e5f8afd2465c808d0f6adcc948/sysdeps/riscv/machine-gmon.h
This issue was originally reported by @wzssyqa.
___
llvm-bugs mailing list
llvm-bugs@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
[llvm-bugs] [Bug 121110] [clang] Miscompile at -Os
Issue
121110
Summary
[clang] Miscompile at -Os
Labels
Assignees
Reporter
cardigan1008
This code prints 9 at `-Os` and 0 at `-O0/1/2/3`:
```c
int printf(const char *, ...);
char a = 9, c, d;
int b, e;
unsigned short f;
char *g = &a;
void h(int i) {
for (; b; b++)
c &= 0 <= i;
}
static short j(unsigned long i) {
int k;
for (; e + d + 4 > 0;) {
k = i + 49;
h(k + i - 52 + i);
*g = 0;
return 0;
}
return 0;
}
int main() {
j(6 < (unsigned short)(f - 7) + f);
printf("%d\n", a);
}
```
Compiler Explorer: https://godbolt.org/z/Mo5TEKh8r
Bisected to https://github.com/llvm/llvm-project/commit/5287299f8809ae927a0acafb179c4b37ce9ff21d, which was committed by @RKSimon
___
llvm-bugs mailing list
llvm-bugs@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
[llvm-bugs] [Bug 121119] MCAsmStreamer: Omit initial ".text" causing lot of breakages
Issue 121119 Summary MCAsmStreamer: Omit initial ".text" causing lot of breakages Labels new issue Assignees Reporter pranavk @MaskRay https://github.com/llvm/llvm-project/commit/7b23f413d1f76532825e470b523e971818d453ca#diff-c1cd0d8700618033d5c6439791e7203361e531ee49f9a53865ca158302a79adc is causing us a lot of problems. I know you have changed all the tests that were dependent on it in the LLVM repo but it's a disruptive change in general. I am trying to fix-forward but failing that, could we revert this and reland it later after the holidays? ___ llvm-bugs mailing list llvm-bugs@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
[llvm-bugs] [Bug 121122] [amdgpu] Using fp128 fails with unsupported libcalls
Issue
121122
Summary
[amdgpu] Using fp128 fails with unsupported libcalls
Labels
backend:AMDGPU
Assignees
Reporter
Flakebi
`fp128` seems to work fine in arguments and return values, but trying to do computations like `fmul` or `fdiv` fails:
```llvm
target triple = "amdgcn-amd-amdhsa"
define fp128 @fp128(fp128 %a, fp128 %b) {
%res = fmul fp128 %a, %b
ret fp128 %res
}
```
results in
```
LLVM ERROR: unsupported libcall legalization
Stack dump:
…
#10 llvm::SITargetLowering::LowerCall(llvm::TargetLowering::CallLoweringInfo&, llvm::SmallVectorImpl&) const llvm-project/llvm/lib/Target/AMDGPU/SIISelLowering.cpp:3877:34
```
I assume this is expected.
I encountered this while trying to compile Rust code, see also https://github.com/rust-lang/compiler-builtins/pull/737.
___
llvm-bugs mailing list
llvm-bugs@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
[llvm-bugs] [Bug 121131] [Flang] Assignments for complex variables are slower than GFortran
Issue 121131 Summary [Flang] Assignments for complex variables are slower than GFortran Labels flang Assignees Reporter yus3710-fj ___ llvm-bugs mailing list llvm-bugs@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
[llvm-bugs] [Bug 121128] [Flang] Assignments for overlapping arrays of a derived type with FORALL/WHERE are slower than GFortran
Issue 121128 Summary [Flang] Assignments for overlapping arrays of a derived type with FORALL/WHERE are slower than GFortran Labels flang Assignees Reporter yus3710-fj ___ llvm-bugs mailing list llvm-bugs@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
[llvm-bugs] [Bug 121132] [Flang] Assignments for variables of a derived type are slower than GFortran
Issue 121132 Summary [Flang] Assignments for variables of a derived type are slower than GFortran Labels flang Assignees Reporter yus3710-fj ___ llvm-bugs mailing list llvm-bugs@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
[llvm-bugs] [Bug 121134] [Flang] Pointer assignments for array pointers are slower than GFortran
Issue 121134 Summary [Flang] Pointer assignments for array pointers are slower than GFortran Labels flang Assignees Reporter yus3710-fj ___ llvm-bugs mailing list llvm-bugs@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
[llvm-bugs] [Bug 121129] [Flang] Assignments for integer arrays with FORALL/WHERE are slower than GFortran
Issue 121129 Summary [Flang] Assignments for integer arrays with FORALL/WHERE are slower than GFortran Labels flang Assignees Reporter yus3710-fj ___ llvm-bugs mailing list llvm-bugs@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
[llvm-bugs] [Bug 121127] [Flang] Assignments for arrays of a derived type are slower than GFortran
Issue 121127 Summary [Flang] Assignments for arrays of a derived type are slower than GFortran Labels flang Assignees Reporter yus3710-fj ___ llvm-bugs mailing list llvm-bugs@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
[llvm-bugs] [Bug 121130] [Flang] Defined assignments are slower than GFortran
Issue 121130 Summary [Flang] Defined assignments are slower than GFortran Labels flang Assignees Reporter yus3710-fj ___ llvm-bugs mailing list llvm-bugs@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
[llvm-bugs] [Bug 121133] [Flang] Pointer assignments for scalar pointers are slower than GFortran
Issue 121133 Summary [Flang] Pointer assignments for scalar pointers are slower than GFortran Labels flang Assignees Reporter yus3710-fj ___ llvm-bugs mailing list llvm-bugs@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
