qemu-devel
[Top][All Lists]
Advanced

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

Re: [Qemu-devel] [PATCH v1 06/23] s390x/tcg: Implement VECTOR FP COMPARE


From: David Hildenbrand
Subject: Re: [Qemu-devel] [PATCH v1 06/23] s390x/tcg: Implement VECTOR FP COMPARE (EQUAL|HIGH|HIGH OR EQUAL)
Date: Fri, 31 May 2019 19:18:34 +0200
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:60.0) Gecko/20100101 Thunderbird/60.6.1

On 31.05.19 18:53, Richard Henderson wrote:
> On 5/31/19 5:44 AM, David Hildenbrand wrote:
>> +static int vfc64(S390Vector *v1, const S390Vector *v2, const S390Vector *v3,
>> +                 CPUS390XState *env, bool s, bool test_equal, bool 
>> test_high,
>> +                 uintptr_t retaddr)
>> +{
>> +    uint8_t vxc, vec_exc = 0;
>> +    S390Vector tmp = {};
>> +    int match = 0;
>> +    int i;
>> +
>> +    for (i = 0; i < 2; i++) {
>> +        const float64 a = make_float64(s390_vec_read_element64(v2, i));
>> +        const float64 b = make_float64(s390_vec_read_element64(v3, i));
>> +        const int cmp = float64_compare_quiet(a, b, &env->fpu_status);
>> +
>> +        if ((cmp == float_relation_equal && test_equal) ||
>> +            (cmp == float_relation_greater && test_high)) {
> 
> It might be easier to pass in the comparison function instead of test_equal 
> and
> test_high (float64_eq_quiet, float64_lt_quiet) and swap the arguments to turn
> lt into gt (not affecting eq).
> 
> This will let you pass float64_eq and float64_lt when it comes time to support
> the SQ bit for the vector-enhancment-1 facility.
> 
> Otherwise you'll have 3 bools passed in and a bit of a mess here.

Very good idea!

> 
> 
>> +    if (match == i + 1) {
>> +        return 0;
> 
> This doesn't look right.  How can match == 3,
> with i == 2 when not exiting the loop early.
> 
> The vxc case is handled via longjmp, I think,
> which leaves the S case to handle here.
> 
> Perhaps better as
> 
>       if (match) {
>           return s || match == 2 ? 0 : 1;
>       }
>       return 3;

Yes indeed, thanks for catching this.

> 
> 
> r~
> 


-- 

Thanks,

David / dhildenb



reply via email to

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