Re: [RFC PATCH] target/i386: avoid copying junk to extended ZMMReg fields

[Paolo Bonzini]

On 4/11/22 17:18, Peter Maydell wrote:
> Looking a bit more closely, this won't work on big-endian
> hosts, because there we want to copy across the last 16
> bytes of the struct, not the first 16. So I think we need
> some more macro magic:
>
> #if SHIFT == 0
> #define COPY_REG(DEST, SRC) (DEST) = (SRC)
> #else
> #define COPY_REG(DEST, SRC) do { \
>      (DEST).Q(0) = (SRC).Q(0);    \
>      (DEST).Q(1) = (SRC).Q(1);    \
>    } while (0)
> #endif
>
> and then use COPY_REG(*d, r);

Right, I have written something similar after seeing your response to
the bug.

> We could probably try to write endian-specific flavours of
> memcpy() invocation, but "do two 64-bit word copies" is what
> the compiler would hopefully turn the memcpy into anyway :-)

Yeah, I actually wrote the memcpy() invocation because I was going to
look at AVX later this year, which of course you couldn't know. :)
What I came up after stealing parts of your nice comment is the
following:

/*
 * Copy the relevant parts of a Reg value around. In the case where
 * sizeof(Reg) > SIZE, these helpers operate only on the lower bytes of
 * a 64 byte ZMMReg, so we must copy only those and keep the top bytes
 * untouched in the guest-visible destination destination register.
 * Note that the "lower bytes" are placed last in memory on big-endian
 * hosts, which store the vector backwards in memory.  In that case the
 * copy *starts* at B(SIZE - 1) and ends at B(0), the opposite of
 * the little-endian case.
 */
#ifdef HOST_WORDS_BIGENDIAN
#define MOVE(d, r) memcpy(&((d).B(SIZE - 1)), &(d).B(SIZE - 1), SIZE)
#else
#define MOVE(d, r) memcpy(&(d).B(0), &(r).B(0), SIZE)
#endif

I'll still your nice comment and submit a patch later when 7.1 opens.

Paolo