On Sat, 12 Mar 2022 19:58:37 GMT, Jatin Bhateja <jbhat...@openjdk.org> wrote:
>> Summary of changes:
>> - Intrinsify Math.round(float) and Math.round(double) APIs.
>> - Extend auto-vectorizer to infer vector operations on encountering scalar
>> IR nodes for above intrinsics.
>> - Test creation using new IR testing framework.
>>
>> Following are the performance number of a JMH micro included with the patch
>>
>> Test System: Intel(R) Xeon(R) Platinum 8380 CPU @ 2.30GHz (Icelake Server)
>>
>>
>> Benchmark | TESTSIZE | Baseline AVX3 (ops/ms) | Withopt AVX3 (ops/ms) | Gain
>> ratio | Baseline AVX2 (ops/ms) | Withopt AVX2 (ops/ms) | Gain ratio
>> -- | -- | -- | -- | -- | -- | -- | --
>> FpRoundingBenchmark.test_round_double | 1024.00 | 504.15 | 2209.54 | 4.38 |
>> 510.36 | 548.39 | 1.07
>> FpRoundingBenchmark.test_round_double | 2048.00 | 293.64 | 1271.98 | 4.33 |
>> 293.48 | 274.01 | 0.93
>> FpRoundingBenchmark.test_round_float | 1024.00 | 825.99 | 4754.66 | 5.76 |
>> 751.83 | 2274.13 | 3.02
>> FpRoundingBenchmark.test_round_float | 2048.00 | 412.22 | 2490.09 | 6.04 |
>> 388.52 | 1334.18 | 3.43
>>
>>
>> Kindly review and share your feedback.
>>
>> Best Regards,
>> Jatin
>
> Jatin Bhateja has updated the pull request incrementally with one additional
> commit since the last revision:
>
> 8279508: Creating separate test for round double under feature check.
src/hotspot/cpu/x86/assembler_x86.hpp line 1159:
> 1157: void cvttsd2siq(Register dst, Address src);
> 1158: void cvttsd2siq(Register dst, XMMRegister src);
> 1159: void cvtsd2siq(Register dst, XMMRegister src);
Hi, some small suggestions only, the instructions are sorted alphabetically so
the `cvtsd2si` should come before `scttsd2si`, the same for the below
instructions.
src/hotspot/cpu/x86/c2_MacroAssembler_x86.cpp line 4024:
> 4022: * the result is equal to the value of Integer.MAX_VALUE.
> 4023: */
> 4024: void C2_MacroAssembler::vector_cast_float_special_cases_avx(XMMRegister
> dst, XMMRegister src, XMMRegister xtmp1,
This special handling is really large, could we use a stub routine for it?
src/hotspot/cpu/x86/c2_MacroAssembler_x86.cpp line 4161:
> 4159: movl(scratch, 1056964608);
> 4160: movq(xtmp1, scratch);
> 4161: vbroadcastss(xtmp1, xtmp1, vec_enc);
An `evpbroadcastd` would reduce this by one instruction I guess?
src/hotspot/cpu/x86/c2_MacroAssembler_x86.cpp line 4178:
> 4176: movl(scratch, 1056964608);
> 4177: movq(xtmp1, scratch);
> 4178: vbroadcastss(xtmp1, xtmp1, vec_enc);
You could put the constant in the constant table and use `vbroadcastss` here
also.
Thank you very much.
src/hotspot/cpu/x86/x86.ad line 7297:
> 7295: ins_encode %{
> 7296: int vlen_enc = vector_length_encoding(this);
> 7297: InternalAddress new_mxcsr = $constantaddress(0x3F80L);
`ldmxcsr` takes a `m32` argument so this constant can be an `int` instead.
Also, I would suggest putting the `mxcst_std` in the constant table also.
src/hotspot/cpu/x86/x86_64.ad line 10699:
> 10697: match(Set dst (ConvD2L src));
> 10698: effect(KILL cr);
> 10699: format %{ "round_or_convert_d2l $dst,$src"%}
You could revert the changes for `ConvD2L` and `ConvF2I` here
-------------
PR: https://git.openjdk.java.net/jdk/pull/7094
