Max/Min Reduction for floating point types.

[Vladimir Ivanov]

Looks good to me, Jatin.

src/hotspot/cpu/x86/x86.ad

-          // Float/Double intrinsics disabled till we fix the 
implementation to match Math.max/Math.min
-          if (bt == T_FLOAT || bt == T_DOUBLE)
+          // Float/Double intrinsics enabled for AVX family.
+          if (UseAVX == 0 && (bt == T_FLOAT || bt == T_DOUBLE))
+            ret_value = false;
+          if (UseAVX > 2 && (!VM_Version::supports_avx512dq() && 
size_in_bits == 512))
              ret_value = false;

I assume those are current implementation limitations. Please, leave a 
comment there which marks them as such.

Best regards,
Vladimir Ivanov

On 28/03/2019 06:38, Bhateja, Jatin wrote:
> 
> Hi All,
> 
> Submitting a patch for max/min reduction operation for floating point types at following link.
> 
> http://cr.openjdk.java.net/~sviswanathan/Jatin/vectorIntrinsics/FPMinMaxReduction/webrev.00/
> 
> Kindly review it.
> 
> Regards,
> Jatin
>