RFR: 8261671: X86 I2L conversion can be skipped for certain masked positive values [v3]

Marcus G K Williams github.com+168222+mgkwill at openjdk.java.net
Thu Mar 4 22:42:40 UTC 2021


On Thu, 4 Mar 2021 22:26:47 GMT, Vladimir Kozlov <kvn at openjdk.org> wrote:

> Note, the test only failed on machines which do not support BMI2: AMD EPYC and Aarch64.

AH! Thanks. I couldn't reproduce here. I was writing a long response about that.

I'll take a look at the predicate and the test to see if I can exclude systems that cannot run BMI2. 

The predicate already had a check for BMI2: 
`predicate(VM_Version::supports_bmi2());`

and the bzhiq assembler method had a check:
`assert(VM_Version::supports_bmi2(), "bit manipulation instructions not supported");`



Here is a warning in Test, though this doesn't preclude running on non-BMI2 supporting processors.
if (!CPUInfo.hasFeature("bmi2")) {
            System.out.println("INFO: CPU does not support bmi2 feature.");
}

That i7 failure is concerning however. I'll look into that also. Thanks @vnkozlov

-------------

PR: https://git.openjdk.java.net/jdk/pull/2590


More information about the hotspot-compiler-dev mailing list