"null" in icc output

[sirzooro]

I do not know if it is problem with Compiler Explorer at https://godbolt.org/ or bug in icc. I tried to compile this code using x86_64 icc 19.0.1 with options -O3 -march=skylake-avx512:

#include <immintrin.h>

__mmask8 test(__mmask8 m)
{
    __mmask8 m2 = _kand_mask8(m, 3);
    return m2;
}

It generated this assembly code:

test(unsigned char):
        vkmovb    k0, edi                                       #5.19
        null      k1, 3                                         #5.19
        kandb     k2, k0, k1                                    #5.19
        vkmovb    eax, k2                                       #5.19
        ret                                                     #6.12

There is strange null instruction there. Please take a look on this.

[mattgodbolt]

https://godbolt.org/z/GKv7nb is an example demonstrating this. if we dump the binary (see the bottom window):

test(unsigned char):
 kmovd %edi,%k0
 roll $0x0,(%rbx)
 kandb %k1,%k0,%k2
 kmovd %k2,%eax
 retq 

[mattgodbolt]

Seems it's an icc bug:

ubuntu@admin-node ~> cat > /tmp/test.cc
#include <immintrin.h>

__mmask8 test(__mmask8 m)
{
    __mmask8 m2 = _kand_mask8(m, 3);
    return m2;
}
ubuntu@admin-node ~> /opt/compiler-explorer/intel-2019.1/bin/icc -S /tmp/test.cc -O3 -march=skylake-avx512 -o -
        .file "test.cc"
        .text
..TXTST0:
.L_2__routine_start__Z4testh_0:
# -- Begin  _Z4testh
        .text
       .align    16,0x90
        .globl _Z4testh
_Z4testh:
# parameter 1: %edi
..B1.1:
        .cfi_startproc
        .cfi_personality 0x3,__gxx_personality_v0
..___tag_value__Z4testh.1:
..L2:

        vkmovb    %edi, %k0
        null      $3, %k1
        kandb     %k1, %k0, %k2
        vkmovb    %k2, %eax
        ret       
        .align    16,0x90
        .cfi_endproc
        .type   _Z4testh,@function
        .size   _Z4testh,.-_Z4testh
..LN_Z4testh.0:
        .data
# -- End  _Z4testh
        .data
        .section .note.GNU-stack, ""
# End

[mattgodbolt]

I filed a bug with Intel: ref 03997020

[sirzooro]

Thanks for reporting it to Intel.

[mattgodbolt]

From Intel:

I have reproduced this bug and escalated (CMPLRIL0-30914). Should be corrected in the upcoming compiler updates.

I will notify you once it is fixed.

[mattgodbolt]

I've pinged Intel on this again, I've not heard anything.

[jeremy-rifkin]

Looks like this is fixed in newer versions of ICC. Do we want to leave this open?

[mattgodbolt]

I'm delighted to close anything where we can...please do!

[jeremy-rifkin]

Will go ahead and close for now