qemu-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [PATCH v5 06/60] target/riscv: add vector index load and store instr


From: Richard Henderson
Subject: Re: [PATCH v5 06/60] target/riscv: add vector index load and store instructions
Date: Fri, 13 Mar 2020 18:49:38 -0700
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:68.0) Gecko/20100101 Thunderbird/68.4.1

On 3/12/20 7:58 AM, LIU Zhiwei wrote:
> +static inline void vext_ldst_index(void *vd, void *v0, target_ulong base,
> +        void *vs2, CPURISCVState *env, uint32_t desc,
> +        vext_get_index_addr get_index_addr,
> +        vext_ldst_elem_fn ldst_elem,
> +        vext_ld_clear_elem clear_elem,
> +        uint32_t esz, uint32_t msz, uintptr_t ra,
> +        MMUAccessType access_type)
> +{
> +    uint32_t i, k;
> +    uint32_t nf = vext_nf(desc);
> +    uint32_t vm = vext_vm(desc);
> +    uint32_t mlen = vext_mlen(desc);
> +    uint32_t vlmax = vext_maxsz(desc) / esz;
> +
> +    if (env->vl == 0) {
> +        return;
> +    }
> +    /* probe every access*/
> +    for (i = 0; i < env->vl; i++) {
> +        if (!vm && !vext_elem_mask(v0, mlen, i)) {
> +            continue;
> +        }
> +        probe_pages(env, get_index_addr(base, i, vs2), nf * msz, ra,
> +                access_type);

Indentation.

> +    /* load bytes from guest memory */
> +    for (i = 0; i < env->vl; i++) {
> +        k = 0;
> +        if (!vm && !vext_elem_mask(v0, mlen, i)) {
> +            continue;
> +        }
> +        while (k < nf) {
> +            abi_ptr addr = get_index_addr(base, i, vs2) + k * msz;
> +            ldst_elem(env, addr, i + k * vlmax, vd, ra);
> +            k++;
> +        }

Why the odd formulation with k?

> +        for (k = 0; k < nf; k++) {
> +            clear_elem(vd, env->vl + k * vlmax, env->vl * esz, vlmax * esz);
> +        }

Using a for is certainly a bit clearer.

Which does bring to mind an optimization -- letting the compiler know that
these loops always go at least once.

We can do that either by writing all of them as do { } while.

Or by encoding NF in desc like the instruction does:

static inline uint32_t vext_nf(uint32_t desc)
{
    return FIELD_EX32(simd_data(desc), VDATA, NF) + 1;
}

which will let the compiler know that NF >= 1.

But that's minor, and we can look at these sorts of things later.

Reviewed-by: Richard Henderson <address@hidden>


r~



reply via email to

[Prev in Thread] Current Thread [Next in Thread]