llvm dev - Oct 2017 - alloca + strd issue on arm freebsd

Hi All,

Below code is compiled with clang on raspberry pi2(bcm2836) + freebsd11.0.
But it doesn't work well if address from alloca is not 32-bit aligned. I am
trying to put down a scenario encountered on a large build system, but I am
unable to generate strd with a simple c example so using asm here. This
problem is encountered only on strd instruction and not on str. If strd
instruction has a alignment requirement why is alloca returning an
unaligned memory address(Not telling in the context of this example)?
Please help me understand what is the best way to handle this issue.

$ clang -v
FreeBSD clang version 3.8.0 (tags/RELEASE_380/final 262564) (based on LLVM
3.8.0)
Target: armv6--freebsd11.0-gnueabihf
Thread model: posix
InstalledDir: /usr/bin

$ ./a.out
dst addr = 0xbfbfecaf
Bus error (core dumped)

But, if the address is 32-bit aligned
$ ./a.out
dst addr = 0xbfbfecb0
value = 123456789123456

Regards,
Karan

--------- CODE -----------
#include <stdio.h>
#include <stdlib.h>

void test_strd(unsigned long long* addr)
{
  unsigned long long value = 0x123456789123456;
  asm("ldr     r4, [sp, #20];"
      "ldr     r3, [sp, #12];"
      "ldr     r2, [sp, #8];"
      "strd    r2, r3, [r4];");
}

/* void test_strd(unsigned long long* addr) */
/* { */
/*   unsigned long long value = 0x123456789123456; */
/*   asm("ldr     r4, [sp, #20];" */
/*       "ldr     r3, [sp, #12];" */
/*       "ldr     r2, [sp, #8];" */
/*       "str     r3, [r4, #4];" */
/*       "str     r2, [r4];"); */
/* } */

int main()
{
  alloca(1);
  /* alloca(1); */
  /* alloca(1); */
  /* alloca(1); */
  unsigned long long *dst = (unsigned long long *)alloca(sizeof(unsigned
long long));
  printf("dst addr = %p\n", (void *)dst);
  test_strd(dst);
  printf("value = %llx\n", *dst);

  return 0;
}
-------------------------------

$ cat strd_test.ll
; ModuleID = 'strd_test.c'
target datalayout = "e-m:e-p:32:32-i64:64-v128:64:128-a:0:32-n32-S64"
target triple = "armv6kz--freebsd11.0-gnueabihf"

@.str = private unnamed_addr constant [15 x i8] c"dst addr =
%p\0A\00",
align 1
@.str.1 = private unnamed_addr constant [14 x i8] c"value =
%llx\0A\00",
align 1

; Function Attrs: nounwind
define void @test_strd(i64* %addr) #0 {
  %1 = alloca i64*, align 4
  %value = alloca i64, align 8
  store i64* %addr, i64** %1, align 4
  store i64 81985529206420566, i64* %value, align 8
  call void asm sideeffect "ldr     r4, [sp, #20];ldr     r3, [sp,
#12];ldr     r2, [sp, #8];strd    r2, r3, [r4];", ""() #2,
!srcloc !3
  ret void
}

; Function Attrs: nounwind
define i32 @main() #0 {
  %1 = alloca i32, align 4
  %dst = alloca i64*, align 4
  store i32 0, i32* %1, align 4
  %2 = alloca i8
  %3 = alloca i8, i32 8
  %4 = bitcast i8* %3 to i64*
  store i64* %4, i64** %dst, align 4
  %5 = load i64*, i64** %dst, align 4
  %6 = bitcast i64* %5 to i8*
  %7 = call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([15 x i8],
[15 x i8]* @.str, i32 0, i32 0), i8* %6)
  %8 = load i64*, i64** %dst, align 4
  call void @test_strd(i64* %8)
  %9 = load i64*, i64** %dst, align 4
  %10 = load i64, i64* %9, align 8
  %11 = call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([14 x i8],
[14 x i8]* @.str.1, i32 0, i32 0), i64 %10)
  ret i32 0
}

declare i32 @printf(i8*, ...) #1

attributes #0 = { nounwind "disable-tail-calls"="false"
"less-precise-fpmad"="false"
"no-frame-pointer-elim"="true"
"no-frame-pointer-elim-non-leaf"
"no-infs-fp-math"="false"
"no-nans-fp-math"="false"
"stack-protector-buffer-size"="8"
"target-cpu"="arm1176jzf-s"
"target-features"="+dsp,+strict-align,+vfp2"
"unsafe-fp-math"="false"
"use-soft-float"="false" }
attributes #1 = { "disable-tail-calls"="false"
"less-precise-fpmad"="false"
"no-frame-pointer-elim"="true"
"no-frame-pointer-elim-non-leaf"
"no-infs-fp-math"="false"
"no-nans-fp-math"="false"
"stack-protector-buffer-size"="8"
"target-cpu"="arm1176jzf-s"
"target-features"="+dsp,+strict-align,+vfp2"
"unsafe-fp-math"="false"
"use-soft-float"="false" }
attributes #2 = { nounwind }

!llvm.module.flags = !{!0, !1}
!llvm.ident = !{!2}

!0 = !{i32 1, !"wchar_size", i32 4}
!1 = !{i32 1, !"min_enum_size", i32 4}
!2 = !{!"FreeBSD clang version 3.8.0 (tags/RELEASE_380/final 262564) (based
on LLVM 3.8.0)"}
!3 = !{i32 139}
-----------------------------
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20171011/a485de60/attachment.html>

Hi Karnajit,

On 10 October 2017 at 21:16, karnajit wangkhem via llvm-dev
<llvm-dev at lists.llvm.org> wrote:
> $ clang -v
> FreeBSD clang version 3.8.0 (tags/RELEASE_380/final 262564) (based on LLVM
> 3.8.0)
> Target: armv6--freebsd11.0-gnueabihf
> Thread model: posix
> InstalledDir: /usr/bin
[...]
>   %2 = alloca i8
>   %3 = alloca i8, i32 8
This looks like a bug in how the alloca function is handled in Clang
3.8. Those instructions should be tagged with some ABI-sensible
alignment, and they are in newer versions of Clang but obviously not
back then.

So you probably either need to upgrade your compiler or work around
the bug by allocating more than you need and manually aligning the
result.

Cheers.

Tim.