Stuff Michael Meeks is doing

• Opening O_WRONLY | O_NONBLOCK gives an inexplicable ENXIO error, odd
• Seems you need a pointless O_NONBLOCK reader opening the pipe before you can open O_NONBLOCK to write and poll on it - so wasting a file descriptor.
• Trying to close that after the open just gives endless EPIPE - so just leak it.
• Unfortunately now we have to read a lot of random numbers to fill the O_NONBLOCK pipe buffer which is probably not good for anyone; hmm.

// load and arrange 8x pixel chunks & a shifted version
vpermd %ymm0,%ymm1,%ymm5
vmovdqu 0x20(%rdi,%r10,4),%ymm0
vpand  %ymm5,%ymm2,%ymm5
vpermd %ymm0,%ymm3,%ymm6
vpand  %ymm4,%ymm6,%ymm6
vpor   %ymm5,%ymm6,%ymm5
// compare and turn that into a bitmask of duplicates
vpcmpeqd %ymm0,%ymm5,%ymm5
vmovmskps %ymm5,%eax
// store that bitmask (win was here)
mov    %al,(%rcx)
mov    %eax,%r11d
not    %r11b
shl    $0x5,%rax
// pack only new pixels into memory
vmovdqa (%rax,%r9,1),%ymm5
vpermd %ymm0,%ymm5,%ymm5
vmovdqu %ymm5,(%r8)
movzbl %r11b,%eax
popcnt %rax,%rax
lea    (%r8,%rax,4),%r8
add    $0x8,%r10
inc    %rcx
cmp    $0xf8,%r10
jb     // loop to top

// simplified inner loop for 64 pixel chunks
for (; bitToSet; ++x, bitToSet <<= 1)
{
    if (from[x] == lastPix)
        rleMask |= bitToSet;
    else
    {
        lastPix = from[x];
        scratch[outp++] = lastPix;
    }
}

"What is not covered[1]:
...
A. 19 asbestos risks or pollution.
...
A. 21 or contributed to by, resulting from or in connection with the Common Vulnerability and Exposure (CVE) reference CVE-2021-44228 listed in the National Vulnerability Database operated by the National Institute of Standards and Technology." (Log4j vulnerability)