Incorrect filtering of intrinsics

[mattgodbolt]

The following input:

bollo:~/d/c/asm-parser (main|✔) $ objdump -d /tmp/moo.o -l --insn-width=16 -r -C -M intel

/tmp/moo.o:     file format elf64-x86-64


Disassembly of section .text:

0000000000000000 <a(long long __vector(2))>:
_mm_bsrli_si128(long long __vector(2), int):
/opt/compiler-explorer/gcc-13.2.0/lib/gcc/x86_64-linux-gnu/13.2.0/include/emmintrin.h:1217
   0:	66 0f 73 d8 04                                  	psrldq xmm0,0x4
a(long long __vector(2)):
//tmp/moo.cc:6
   5:	c3                                              	ret

will ignore the psrldq instruction:

$ objdump -d /tmp/moo.o -l --insn-width=16 -r -C -M intel | ~/Downloads/asm-parser -plt -stdin -binary -unused_labels -directives -comment_only -library_functions
{"asm": [{"labels": [], "source": null, "text": "long __vector(2))::"},{"labels": [], "opcodes": ["c3"], "address": 5, "source": {"file": "//tmp/moo.cc", "mainsource": "false", "line": 6}, "text": " ret"}],"labelDefinitions": {"long __vector(2)):": 1}, "parsingTime": 2}

From compiler-explorer/compiler-explorer#6188

[mattgodbolt]

The line:

/opt/compiler-explorer/gcc-13.2.0/lib/gcc/x86_64-linux-gnu/13.2.0/include/emmintrin.h:1217

seems to be the thing that triggers the ignoring of the line after, even with just -plt -stdin -binary and no other filtering.

[Alcaro]

Yes, this definitely has something to do with that piece being inlined from immintrin.h. Considering that instruction to be library code is reasonable (everything in immintrin.h is tagged attribute((artificial)), so I suspect disassembly output shouldn't point to immintrin.h at all, but that's not a bug anywhere in CE).

Filtering library code when not asked to, less reasonable.

And deleting half of the line and printing long __vector(2)):: is not reasonable either. Looks like two different bugs to me; you may want to investigate the latter first, I suspect it may become hard to reproduce after fixing the library filtering thing.

[mattgodbolt]

The CE code correctly outputs this as:

{
  "asm": [
    {
      "labels": [],
      "source": null,
      "text": "a(long long __vector(2)):"
    },
    {
      "address": 0,
      "labels": [],
      "opcodes": [
        "66",
        "0f",
        "73",
        "d8",
        "04"
      ],
      "source": {
        "file": null,
        "line": 1217,
        "mainsource": true
      },
      "text": " psrldq xmm0,0x4"
    },
    {
      "address": 5,
      "labels": [],
      "opcodes": [
        "c3"
      ],
      "source": {
        "file": null,
        "line": 6,
        "mainsource": true
      },
      "text": " ret"
    }
  ],
  "labelDefinitions": {
    "a(long long __vector(2))": 1
  }
}