RFR: Load coops base shift from AOTRuntimeConstants in AOT code [v5]

[Andrew Dinn]

On Thu, 26 Sep 2024 10:52:02 GMT, Andrew Dinn <adinn at openjdk.org> wrote:

>> This PR modifies AOT compiled method code to load compressed oops base and shift constants via the AOTRuntimeConstants area rather than encode them as immediates. It also unlatches the currently forced setting of UseCompatibleCompressedOops, allowing the heap to be allocated wherever it will fit.
>
> Andrew Dinn has updated the pull request incrementally with one additional commit since the last revision:
> 
>   Use the force to wrangle register sets

Ok, so here are the proper results, first when the heap is defaulted:

Run 1:

    ============================== jvm1 ============================
    [1_xoff] Premain JDK (CDS disabled)                    235.10 ms
    [1_xon ] Premain JDK (CDS enabled)                     111.21 ms
    [1_td  ] Premain Prototype (CDS + Training Data)       109.45 ms
    [1_aot ] Premain Prototype (CDS + Training Data + AOT)  56.87 ms
    ============================== jvm2 ============================
    [2_xoff] Premain JDK (CDS disabled)                    233.21 ms
    [2_xon ] Premain JDK (CDS enabled)                     111.37 ms
    [2_td  ] Premain Prototype (CDS + Training Data)       107.96 ms
    [2_aot ] Premain Prototype (CDS + Training Data + AOT)  59.00 ms
    ================================================================

Run 2:

    ============================== jvm1 ============================
    [1_xoff] Premain JDK (CDS disabled)                    235.84 ms
    [1_xon ] Premain JDK (CDS enabled)                     112.28 ms
    [1_td  ] Premain Prototype (CDS + Training Data)       108.45 ms
    [1_aot ] Premain Prototype (CDS + Training Data + AOT)  57.76 ms
    ============================== jvm2 ============================
    [2_xoff] Premain JDK (CDS disabled)                    236.09 ms
    [2_xon ] Premain JDK (CDS enabled)                     111.91 ms
    [2_td  ] Premain Prototype (CDS + Training Data)       108.86 ms
    [2_aot ] Premain Prototype (CDS + Training Data + AOT)  59.80 ms
    ================================================================


Run 3:


    ============================== jvm1 ============================
    [1_xoff] Premain JDK (CDS disabled)                    231.32 ms
    [1_xon ] Premain JDK (CDS enabled)                     111.47 ms
    [1_td  ] Premain Prototype (CDS + Training Data)       108.92 ms
    [1_aot ] Premain Prototype (CDS + Training Data + AOT)  56.26 ms
    ============================== jvm2 ============================
    [2_xoff] Premain JDK (CDS disabled)                    233.96 ms
    [2_xon ] Premain JDK (CDS enabled)                     114.21 ms
    [2_td  ] Premain Prototype (CDS + Training Data)       107.92 ms
    [2_aot ] Premain Prototype (CDS + Training Data + AOT)  60.01 ms
    ================================================================

Now the results when heap size is set to 24Mb:

Run 1:

    ============================== jvm1 ============================
    [1_xoff] Premain JDK (CDS disabled)                    237.26 ms
    [1_xon ] Premain JDK (CDS enabled)                     114.10 ms
    [1_td  ] Premain Prototype (CDS + Training Data)       108.43 ms
    [1_aot ] Premain Prototype (CDS + Training Data + AOT)  56.43 ms
    ============================== jvm2 ============================
    [2_xoff] Premain JDK (CDS disabled)                    238.70 ms
    [2_xon ] Premain JDK (CDS enabled)                     115.12 ms
    [2_td  ] Premain Prototype (CDS + Training Data)       107.67 ms
    [2_aot ] Premain Prototype (CDS + Training Data + AOT)  58.27 ms
    ================================================================

Run 2:

    ============================== jvm1 ============================
    [1_xoff] Premain JDK (CDS disabled)                    237.81 ms
    [1_xon ] Premain JDK (CDS enabled)                     114.76 ms
    [1_td  ] Premain Prototype (CDS + Training Data)       107.24 ms
    [1_aot ] Premain Prototype (CDS + Training Data + AOT)  56.57 ms
    ============================== jvm2 ============================
    [2_xoff] Premain JDK (CDS disabled)                    237.65 ms
    [2_xon ] Premain JDK (CDS enabled)                     115.34 ms
    [2_td  ] Premain Prototype (CDS + Training Data)       107.66 ms
    [2_aot ] Premain Prototype (CDS + Training Data + AOT)  59.39 ms
    ================================================================

Run 3:

    ============================== jvm1 ============================
    [1_xoff] Premain JDK (CDS disabled)                    238.08 ms
    [1_xon ] Premain JDK (CDS enabled)                     114.87 ms
    [1_td  ] Premain Prototype (CDS + Training Data)       107.35 ms
    [1_aot ] Premain Prototype (CDS + Training Data + AOT)  56.20 ms
    ============================== jvm2 ============================
    [2_xoff] Premain JDK (CDS disabled)                    237.19 ms
    [2_xon ] Premain JDK (CDS enabled)                     114.43 ms
    [2_td  ] Premain Prototype (CDS + Training Data)       107.40 ms
    [2_aot ] Premain Prototype (CDS + Training Data + AOT)  58.24 ms
    ================================================================

So, there is stil some variability in the timings which makes it hard to gauge the full effect. However, it looks as if the patch is adding at least 3% overhead to the run times for the aot case with the default heap and also with the smaller heap. The ability to vary the coops model with a smaller heap does not appear to win anything back ineither the aot case or the non-aot case.

So, we definitely drop this patch.

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

PR Comment: https://git.openjdk.org/leyden/pull/20#issuecomment-2377067807