On 10/24/21 19:26, Richard Henderson wrote:
> On 10/24/21 9:44 AM, Philippe Mathieu-Daudé wrote:
>> On 10/24/21 07:04, Richard Henderson wrote:
>>> On 10/23/21 2:47 PM, Philippe Mathieu-Daudé wrote:
>>>> Convert the FILL opcode (Vector Fill from GPR) to decodetree.
>>>>
>>>> Signed-off-by: Philippe Mathieu-Daudé <f4...@amsat.org>
>>>> ---
>>>> target/mips/tcg/msa.decode | 2 ++
>>>> target/mips/tcg/msa_translate.c | 40
>>>> +++++++++++++++++++++++----------
>>>> 2 files changed, 30 insertions(+), 12 deletions(-)
>>
>>>> +static bool trans_FILL(DisasContext *ctx, arg_msa_r *a)
>>>> +{
>>>> + TCGv_i32 twd;
>>>> + TCGv_i32 tws;
>>>> + TCGv_i32 tdf;
>>>> +
>>>> + if (!check_msa_access(ctx)) {
>>>> + return false;
>>>> + }
>>>> +
>>>> + if (TARGET_LONG_BITS != 64 && a->df == DF_DOUBLE) {
>>>> + /* Double format valid only for MIPS64 */
>>>> + gen_reserved_instruction(ctx);
>>>> + return true;
>>>> + }
>>>
>>> I expect this reserved check should be above the MSA is disabled check,
>>> within check_msa_access.
>>>
>>>> + twd = tcg_const_i32(a->wd);
>>>> + tws = tcg_const_i32(a->ws);
>>>
>>> tcg_constant_i32.
>>
>> Hmm I am confused here, only 'df' is constant, 'ws' is GPR[$rs].
>> Since it is limited in [0,32[ and used read-only by the helper,
>> we can also pass it as constant?
>
> What? You're passing the constant ws to the helper, not the contents of
> the gpr -- that's what the helper is expecting.
OK got it now, thanks :)
