Re: [PATCH v2 02/14] util: Add RISC-V vector extension probe in cpuinfo

[Richard Henderson]

On 8/30/24 16:15, LIU Zhiwei wrote:
> +extern unsigned riscv_vlen;

Do you really want to store vlen and not vlenb?
It seems that would simplify some of your computation in the tcg backend.


> @@ -49,6 +50,9 @@ unsigned __attribute__((constructor)) cpuinfo_init(void)
>   #endif
>   #if defined(__riscv_arch_test) && defined(__riscv_zicond)
>       info |= CPUINFO_ZICOND;
> +#endif
> +#if defined(__riscv_arch_test) && defined(__riscv_zve64x)
> +    info |= CPUINFO_ZVE64X;
>   #endif
>       left &= ~info;
>   
> @@ -64,7 +68,8 @@ unsigned __attribute__((constructor)) cpuinfo_init(void)
>               && pair.key >= 0) {
>               info |= pair.value & RISCV_HWPROBE_EXT_ZBA ? CPUINFO_ZBA : 0;
>               info |= pair.value & RISCV_HWPROBE_EXT_ZBB ? CPUINFO_ZBB : 0;
> -            left &= ~(CPUINFO_ZBA | CPUINFO_ZBB);
> +            info |= pair.value & RISCV_HWPROBE_IMA_V ? CPUINFO_ZVE64X : 0;
> +            left &= ~(CPUINFO_ZBA | CPUINFO_ZBB | CPUINFO_ZVE64X);
>   #ifdef RISCV_HWPROBE_EXT_ZICOND
>               info |= pair.value & RISCV_HWPROBE_EXT_ZICOND ? CPUINFO_ZICOND : 
> 0;
>               left &= ~CPUINFO_ZICOND;
> @@ -112,6 +117,23 @@ unsigned __attribute__((constructor)) cpuinfo_init(void)
>           assert(left == 0);
>       }
>   
> +    if (info & CPUINFO_ZVE64X) {
> +        /*
> +         * Get vlen for Vector.
> +         * VLMAX = LMUL * VLEN / SEW.
> +         * The "vsetvli rd, x0, e64" means "LMUL = 1, SEW = 64, rd = VLMAX",
> +         * so "vlen = VLMAX * 64".
> +         */
> +        unsigned long vlmax = 0;
> +        asm("vsetvli %0, x0, e64" : "=r"(vlmax));

This doesn't compile, surely.

s/x0/zero/ is surely the minimum change, but this still won't work unless V is enabled at 
compile-time.

You need to use the .insn form, like we do for the Zba probe above:

        .insn i 0x57, 7, %0, zero, 3 << 3

I have verified that RISCV_HWPROBE_IMA_V went into the linux kernel at the same time as 
vector support, so this probing is complete and sufficient.


r~