Re: [PATCH v5 07/11] target/s390x: vxeh2: vector {load, store} byte reversed elements

[David Hildenbrand <david@redhat.com>]

> +static DisasJumpType op_vlbr(DisasContext *s, DisasOps *o)
> +{
> +    const uint8_t es = get_field(s, m3);
> +    TCGv_i64 t0, t1;
> +
> +    if (es < ES_16 || es > ES_128) {
> +        gen_program_exception(s, PGM_SPECIFICATION);
> +        return DISAS_NORETURN;
> +    }
> +
> +    t0 = tcg_temp_new_i64();
> +    t1 = tcg_temp_new_i64();
> +
> +
> +    if (es == ES_128) {
> +        tcg_gen_qemu_ld_i64(t1, o->addr1, get_mem_index(s), MO_LEUQ);
> +        gen_addi_and_wrap_i64(s, o->addr1, o->addr1, 8);
> +        tcg_gen_qemu_ld_i64(t0, o->addr1, get_mem_index(s), MO_LEUQ);
> +        goto write;
> +    }
> +
> +    /* Begin with byte reversed doublewords... */
> +    tcg_gen_qemu_ld_i64(t0, o->addr1, get_mem_index(s), MO_LEUQ);
> +    gen_addi_and_wrap_i64(s, o->addr1, o->addr1, 8);
> +    tcg_gen_qemu_ld_i64(t1, o->addr1, get_mem_index(s), MO_LEUQ);
> +
> +    /*
> +     * For 16 and 32-bit elements, the doubleword bswap also reversed
> +     * the order of the elements.  Perform a larger order swap to put
> +     * them back into place.  For the 128-bit "element", finish the
> +     * bswap by swapping the doublewords.

The last sentence still needs to go.

[...]

> +static DisasJumpType op_vstbr(DisasContext *s, DisasOps *o)
> +{
> +    const uint8_t es = get_field(s, m3);
> +    TCGv_i64 t0, t1;
> +
> +    if (es < ES_16 || es > ES_128) {
> +        gen_program_exception(s, PGM_SPECIFICATION);
> +        return DISAS_NORETURN;
> +    }
> +
> +    /* Probe write access before actually modifying memory */
> +    gen_helper_probe_write_access(cpu_env, o->addr1, tcg_constant_i64(16));
> +
> +    t0 = tcg_temp_new_i64();
> +    t1 = tcg_temp_new_i64();
> +
> +
> +    if (es == ES_128) {
> +        read_vec_element_i64(t1, get_field(s, v1), 0, ES_64);
> +        read_vec_element_i64(t0, get_field(s, v1), 1, ES_64);
> +        goto write;
> +    }
> +
> +    read_vec_element_i64(t0, get_field(s, v1), 0, ES_64);
> +    read_vec_element_i64(t1, get_field(s, v1), 1, ES_64);
> +
> +    /*
> +     * For 16 and 32-bit elements, the doubleword bswap below will
> +     * reverse the order of the elements.  Perform a larger order
> +     * swap to put them back into place.  For the 128-bit "element",
> +     * finish the bswap by swapping the doublewords.

Dito.

I assume Thomas can fixup when applying.

-- 
Thanks,

David / dhildenb