Re: [PATCH v7 3/4] tests/tcg/s390x: Tests for Miscellaneous-Instruction-Extensions Facility 3

[Thomas Huth]

On 28/02/2022 19.31, David Miller wrote:
> Had it on my TODO list for this morning, thank you.

Thanks! Please send it as additional patch on top of my s390x-next, since I 
already sent a pull request for the other patches yesterday:

 https://gitlab.com/thuth/qemu/-/commits/s390x-next/

> On Mon, Feb 28, 2022 at 12:59 PM Richard Henderson 
> <richard.henderson@linaro.org <mailto:richard.henderson@linaro.org>> wrote:
>
>     On 2/28/22 00:14, Thomas Huth wrote:
>      > Full patch can be seen here:
>      >
>      > https://gitlab.com/thuth/qemu/-/commit/38af118ea2fef0c473
>     <https://gitlab.com/thuth/qemu/-/commit/38af118ea2fef0c473>
>
>
>      > static inline void mvcrl_8(const char *dst, const char *src)
>      > {
>      >     asm volatile (
>      >     "llill %%r0, 8\n"
>      >     ".insn sse, 0xE50A00000000, 0(%[dst]), 0(%[src])"
>      >     : : [dst] "d" (dst), [src] "d" (src)
>      >     : "memory");
>      > }
>
>     Need clobber of r0 here.

Right. This test fails with Clang, indeed, as I discovered today, since 
Clang uses r0 more often than GCC, as it seems. I've already sent some 
patches for some other tests today, so there'll be another s390x pull 
request next week for TCG tests fixups :-)

 Thomas


>      > #define Fi3(S, ASM) uint64_t S(uint64_t a, uint64_t b, uint64_t c) \
>      > {                            \
>      >     uint64_t res = 0;        \
>      >     asm (                    \
>      >          "lg %%r2, %[a]\n"   \
>      >          "lg %%r3, %[b]\n"   \
>      >          "lg %%r0, %[c]\n"   \
>      >          "ltgr %%r0, %%r0\n" \
>      >          ASM                 \
>      >          "stg %%r0, %[res] " \
>      >          : [res] "=m" (res)  \
>      >          : [a] "m" (a),      \
>      >            [b] "m" (b),      \
>      >            [c] "m" (c)       \
>      >          : "r0", "r2",       \
>      >            "r3", "r4"        \
>      >     );                       \
>      >     return res;              \
>      > }
>      >
>      > Fi3 (_selre,     ".insn rrf, 0xB9F00000, %%r0, %%r3, %%r2, 8\n")
>      > Fi3 (_selgrz,    ".insn rrf, 0xB9E30000, %%r0, %%r3, %%r2, 8\n")
>      > Fi3 (_selfhrnz,  ".insn rrf, 0xB9C00000, %%r0, %%r3, %%r2, 7\n")
>
>     This isn't actively broken, but could use the same treatment as NCRK et al:
>
>     #define Fi3(S, ASM) uint64_t S(uint64_t a, uint64_t b, uint64_t c) \
>     {                            \
>           uint64_t res;            \
>           asm("ltgr %[c], %[c]\n\t" ASM
>               : [res] "=&r" (res)
>               : [a] "r" (a), [b] "r" (b), [c] "r" (c)
>               : "cc");
>           return res;
>     }
>
>     Fi3(_selre,   ".insn rrf, 0xB9F00000, %[res], %[a], %[b], 8")
>
>     etc.
>
>
>     r~