gcc-4.8: internal compiler error: in dwarf2out_frame_debug_adjust_cfa, at dwarf2cfi.c:1078

[Andy Whitcroft]

Seeing these when trying to compile linux-lts-vivid (v3.19 kernel) on the trusty compiler, also on the proposed SRU version:

  lib/raid6/neon4.c: In function 'raid6_neon4_gen_syndrome_real':
  lib/raid6/neon4.c:113:1: internal compiler error: in dwarf2out_frame_debug_adjust_cfa, at dwarf2cfi.c:1078
   }
   ^
   Please submit a full bug report,
  with preprocessed source if appropriate.
  See <file:///usr/share/doc/gcc-4.8/README.Bugs> for instructions.
  Preprocessed source stored into /tmp/ccg57WPc.out file, please attach this to your bugreport.

[Matthias Klose (doko)]

$ cat neon4.i
typedef uint8x16_t __attribute__((__vector_size__(16)));
typedef uint8x16_t unative_t;
raid6_neon4_gen_syndrome_real_d;
fn1() {
  uint8x16_t __trans_tmp_1;
  unative_t wq2;
  for (; raid6_neon4_gen_syndrome_real_d;) {
    fn2();
    fn2(wq2);
    wq2 = __trans_tmp_1;
  }
}

$ gcc-4.8 -mapcs -mabi=aapcs-linux -mno-thumb-interwork -marm -march=armv7-a -mfloat-abi=softfp -mfpu=neon -g -gdwarf-4 -O2 -std=gnu90 -fno-strict-aliasing -fno-common -fno-dwarf2-cfi-asm -funwind-tables -fno-delete-null-pointer-checks -fstack-protector -fno-omit-frame-pointer -fno-optimize-sibling-calls -fno-var-tracking-assignments -fno-inline-functions-called-once -fno-strict-overflow -fconserve-stack -ffreestanding neon4.i
neon4.i:3:1: warning: data definition has no type or storage class [enabled by default]
 raid6_neon4_gen_syndrome_real_d;
 ^
neon4.i: In function 'fn1':
neon4.i:12:1: internal compiler error: in dwarf2out_frame_debug_adjust_cfa, at dwarf2cfi.c:1078
 }
 ^
Please submit a full bug report,
with preprocessed source if appropriate.

not seen when building with -O1

[Matthias Klose (doko)]

seen with Linaro 4.8, works on the fsf 4.8 branch, can be avoided by using -O1

$ cat neon4.i
typedef uint8x16_t __attribute__((__vector_size__(16)));
typedef uint8x16_t unative_t;
raid6_neon4_gen_syndrome_real_d;
fn1() {
  uint8x16_t __trans_tmp_1;
  unative_t wq2;
  for (; raid6_neon4_gen_syndrome_real_d;) {
    fn2();
    fn2(wq2);
    wq2 = __trans_tmp_1;
  }
}

$ gcc-4.8 -mapcs -mabi=aapcs-linux -mno-thumb-interwork -marm -march=armv7-a -mfloat-abi=softfp -mfpu=neon -g -gdwarf-4 -O2 -std=gnu90 -fno-strict-aliasing -fno-common -fno-dwarf2-cfi-asm -funwind-tables -fno-delete-null-pointer-checks -fstack-protector -fno-omit-frame-pointer -fno-optimize-sibling-calls -fno-var-tracking-assignments -fno-inline-functions-called-once -fno-strict-overflow -fconserve-stack -ffreestanding neon4.i
neon4.i:3:1: warning: data definition has no type or storage class [enabled by default]
 raid6_neon4_gen_syndrome_real_d;
 ^
neon4.i: In function 'fn1':
neon4.i:12:1: internal compiler error: in dwarf2out_frame_debug_adjust_cfa, at dwarf2cfi.c:1078
 }
 ^
Please submit a full bug report,
with preprocessed source if appropriate.

[Matthias Klose (doko)]

Created attachment 289
preprocessed source

the unreduced test case ICEs with -O0 too.

[Christophe Lyon (christophe-lyon)]

How is your GCC configured?
I've just rebuilt our 4.8 branch for arm-none-linux-gnueabihf and I cannot reproduce the ICE you are seeing.

[Matthias Klose (doko)]

this seems to be fixed with current Linaro 4.8, was originally found in 4.8-2014.03. any hint on the specific fix?

[Yvan Roux (yvan-roux)]

It was fixed on our 4.8 branch at r209009 by backporting upstream fix for PR60264 (r208511):

    gcc/
    2014-04-02 Zhenqiang Chen <email address hidden>

        Backport from trunk r208511
        2014-03-12 Christian Bruel <email address hidden>

        PR target/60264
        * config/arm/arm.c (arm_emit_vfp_multi_reg_pop): Emit a
        REG_CFA_DEF_CFA note.
        (arm_expand_epilogue_apcs_frame): call arm_add_cfa_adjust_cfa_note.
        (arm_unwind_emit): Allow REG_CFA_DEF_CFA.

    gcc/testsuite/
    2014-04-02 Zhenqiang Chen <email address hidden>

        Backport from trunk r208511
        2014-03-12 Christian Bruel <email address hidden>

        PR target/60264
        * gcc.target/arm/pr60264.c

[Matthias Klose (doko)]

now fixed in trusty-proposed