[LLVMbugs] [Bug 23414] New: [AArch64]Clang crush on an assertion when calling a NEON intrinsic with incorrect type in CPP file

bugzilla-daemon at llvm.org bugzilla-daemon at llvm.org
Mon May 4 19:49:46 PDT 2015


            Bug ID: 23414
           Summary: [AArch64]Clang crush on an assertion when calling a
                    NEON intrinsic with incorrect type in CPP file
           Product: clang
           Version: trunk
          Hardware: PC
                OS: Linux
            Status: NEW
          Severity: normal
          Priority: P
         Component: C++
          Assignee: unassignedclangbugs at nondot.org
          Reporter: Hao.Liu at arm.com
                CC: dgregor at apple.com, llvmbugs at cs.uiuc.edu
    Classification: Unclassified

The test case is test.cpp
#include <arm_neon.h>

int8x8_t test_vadd_s8(int8x16_t v1, int8x16_t v2) {
//  int8x8_t vadd_s8(int8x8_t a, int8x8_t b)
  return vadd_s8(v1, v2);

int8x8x2_t test_vld2_s8(float *a) {
//  int8x8x2_t vld2_s8(int8_t const * ptr)
  return vld2_s8(a);

To reproduce the crush:
$clang -S -O3 -target aarch64 test.cpp 
unexpected Neon polynomial vector element type
UNREACHABLE executed at
#0 0x26620c5 llvm::sys::PrintStackTrace(llvm::raw_ostream&)
#1 0x26623da PrintStackTraceSignalHandler(void*)

But the C file test.c with the same code as test.cpp can report error/warning
$clang -S -O3 -target aarch64 test.c
test.c:5:18: error: passing 'int8x16_t' (vector of 16 'int8_t' values) to
parameter of incompatible type
      'int8x8_t' (vector of 8 'int8_t' values)
  return vadd_s8(v1, v2);
note: passing argument to
      parameter '__p0' here
__ai int8x8_t vadd_s8(int8x8_t __p0, int8x8_t __p1) {
test.c:10:18: warning: incompatible pointer types assigning to 'const signed
char *' from 'float *'
  return vld2_s8(a);
note: expanded from macro
  __builtin_neon_vld2_v(&__ret, __p0, 0); \
1 warning and 1 error generated.

Also, GCC can report errors (I used the linaro toolchain):
$aarch64-linux-gnu-g++ -O3 -S test.cpp 
test.cpp: In function ‘int8x8_t test_vadd_s8(int8x16_t, int8x16_t)’:
test.cpp:5:24: error: cannot convert ‘int8x16_t {aka __vector(16)
__builtin_aarch64_simd_qi}’ to ‘int8x8_t {aka __vector(8)
__builtin_aarch64_simd_qi}’ for argument ‘1’ to ‘int8x8_t vadd_s8(int8x8_t,
   return vadd_s8(v1, v2);
test.cpp: In function ‘int8x8x2_t test_vld2_s8(float*)’:
test.cpp:10:19: error: cannot convert ‘float*’ to ‘const int8_t* {aka const
signed char*}’ for argument ‘1’ to ‘int8x8x2_t vld2_s8(const int8_t*)’
   return vld2_s8(a);

You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20150505/9cc356d3/attachment.html>

More information about the llvm-bugs mailing list