qemu-devel
[Top][All Lists]
Advanced

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

Re: [Qemu-devel] [PATCH v3 1/6] tcg: Add types and operations for host v


From: Richard Henderson
Subject: Re: [Qemu-devel] [PATCH v3 1/6] tcg: Add types and operations for host vectors
Date: Wed, 27 Sep 2017 09:18:50 -0700
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:52.0) Gecko/20100101 Thunderbird/52.3.0

On 09/26/2017 12:28 PM, Alex Bennée wrote:
>>      * TCGv_ptr : a host pointer type
>> +    * TCGv_vec : a host vector type; the exact size is not exposed
>> +                 to the CPU front-end code.
> 
> Isn't this a guest vector type (which is pointed to by a host pointer)?

No, it's a host vector, which we have created in response to expanding a guest
vector operation.

> A one line comment wouldn't go amiss here. This looks like we are
> allocating a new temp of the same type as an existing temp?
> 
>> +TCGv_vec tcg_temp_new_vec_matching(TCGv_vec match)

Yes.

>> +All of the vector ops have a final constant argument that specifies the
>> +length of the vector operation LEN as 64 << LEN bits.
> 
> That doesn't scan well. So would a 4 lane operation be encoded as 64 <<
> 4? Is this because we are using the bottom bits for something?

64 << 0 = 64
64 << 1 = 128
64 << 2 = 256.

I've fixed up the wording a bit.

>> +  Copy C across the entire vector.
>> +  At present the only supported values for C are 0 and -1.
> 
> I guess this is why the size in unimportant? This is for clearing or
> setting the whole of the vector? What does len mean in this case?

Yes.  Len still means the length of the whole vector.

Elsewhere there's a comment about maybe using dupi{8,16,32,64}_vec instead.
However I wanted to put that off until we do some more conversions and see
exactly what's going to be needed.


>> +* and_vec     v0, v1, v2, len
>> +* or_vec      v0, v1, v2, len
>> +* xor_vec     v0, v1, v2, len
>> +* andc_vec    v0, v1, v2, len
>> +* orc_vec     v0, v1, v2, len
>> +* not_vec     v0, v1, len
>> +
>> +  Similarly, logical operations.
> 
> Similarly, logical operations with and without compliment?

Sure.


r~



reply via email to

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