Pre-Review: improving Math.min/max on floats
Vladimir Kempik
vladimir.kempik at gmail.com
Wed Nov 23 08:49:22 UTC 2022
Hello
Got a results for new [1] version
it shows excelent perf improvements on thead and moderate on hifive ( and it’s better than both previous versions on hifive)
thead c910
before
Benchmark Mode Cnt Score Error Units
FpMinMaxIntrinsics.dMax avgt 25 53752.831 ± 97.198 ns/op
FpMinMaxIntrinsics.dMin avgt 25 53707.229 ± 177.559 ns/op
FpMinMaxIntrinsics.dMinReduce avgt 25 42805.985 ± 9.901 ns/op
FpMinMaxIntrinsics.fMax avgt 25 53449.568 ± 215.294 ns/op
FpMinMaxIntrinsics.fMin avgt 25 53504.106 ± 180.833 ns/op
FpMinMaxIntrinsics.fMinReduce avgt 25 42794.579 ± 7.013 ns/op
MaxMinOptimizeTest.dAdd avgt 25 381.138 ± 5.692 us/op
MaxMinOptimizeTest.dMax avgt 25 4575.094 ± 17.065 us/op
MaxMinOptimizeTest.dMin avgt 25 4584.648 ± 18.561 us/op
MaxMinOptimizeTest.dMul avgt 25 384.615 ± 7.751 us/op
MaxMinOptimizeTest.fAdd avgt 25 318.076 ± 3.308 us/op
MaxMinOptimizeTest.fMax avgt 25 4405.724 ± 20.353 us/op
MaxMinOptimizeTest.fMin avgt 25 4421.652 ± 18.029 us/op
MaxMinOptimizeTest.fMul avgt 25 305.462 ± 19.437 us/op
2fclass_new
Benchmark Mode Cnt Score Error Units
FpMinMaxIntrinsics.dMax avgt 25 10712.246 ± 5.607 ns/op
FpMinMaxIntrinsics.dMin avgt 25 10732.655 ± 41.894 ns/op
FpMinMaxIntrinsics.dMinReduce avgt 25 3248.106 ± 2.143 ns/op
FpMinMaxIntrinsics.fMax avgt 25 10707.084 ± 3.276 ns/op
FpMinMaxIntrinsics.fMin avgt 25 10719.771 ± 14.864 ns/op
FpMinMaxIntrinsics.fMinReduce avgt 25 3274.775 ± 0.996 ns/op
MaxMinOptimizeTest.dAdd avgt 25 383.720 ± 8.849 us/op
MaxMinOptimizeTest.dMax avgt 25 429.345 ± 11.160 us/op
MaxMinOptimizeTest.dMin avgt 25 439.980 ± 3.757 us/op
MaxMinOptimizeTest.dMul avgt 25 390.126 ± 10.258 us/op
MaxMinOptimizeTest.fAdd avgt 25 300.005 ± 18.206 us/op
MaxMinOptimizeTest.fMax avgt 25 370.467 ± 6.054 us/op
MaxMinOptimizeTest.fMin avgt 25 375.134 ± 4.568 us/op
MaxMinOptimizeTest.fMul avgt 25 305.344 ± 18.307 us/op
hifive
before
Benchmark Mode Cnt Score Error Units
FpMinMaxIntrinsics.dMax avgt 25 30234.224 ± 16.744 ns/op
FpMinMaxIntrinsics.dMin avgt 25 30227.686 ± 15.389 ns/op
FpMinMaxIntrinsics.dMinReduce avgt 25 15766.749 ± 3.724 ns/op
FpMinMaxIntrinsics.fMax avgt 25 30140.092 ± 10.243 ns/op
FpMinMaxIntrinsics.fMin avgt 25 30149.470 ± 34.041 ns/op
FpMinMaxIntrinsics.fMinReduce avgt 25 15760.770 ± 5.415 ns/op
MaxMinOptimizeTest.dAdd avgt 25 1155.234 ± 4.603 us/op
MaxMinOptimizeTest.dMax avgt 25 2597.897 ± 3.307 us/op
MaxMinOptimizeTest.dMin avgt 25 2599.183 ± 3.806 us/op
MaxMinOptimizeTest.dMul avgt 25 1155.281 ± 1.813 us/op
MaxMinOptimizeTest.fAdd avgt 25 750.967 ± 7.254 us/op
MaxMinOptimizeTest.fMax avgt 25 2305.085 ± 1.556 us/op
MaxMinOptimizeTest.fMin avgt 25 2305.306 ± 1.478 us/op
MaxMinOptimizeTest.fMul avgt 25 750.623 ± 7.357 us/op
2fclass_new
Benchmark Mode Cnt Score Error Units
FpMinMaxIntrinsics.dMax avgt 25 23599.547 ± 29.571 ns/op
FpMinMaxIntrinsics.dMin avgt 25 23593.236 ± 18.456 ns/op
FpMinMaxIntrinsics.dMinReduce avgt 25 8630.201 ± 1.353 ns/op
FpMinMaxIntrinsics.fMax avgt 25 23496.337 ± 18.340 ns/op
FpMinMaxIntrinsics.fMin avgt 25 23477.881 ± 8.545 ns/op
FpMinMaxIntrinsics.fMinReduce avgt 25 8629.135 ± 0.869 ns/op
MaxMinOptimizeTest.dAdd avgt 25 1155.479 ± 4.938 us/op
MaxMinOptimizeTest.dMax avgt 25 1560.323 ± 3.077 us/op
MaxMinOptimizeTest.dMin avgt 25 1558.668 ± 2.421 us/op
MaxMinOptimizeTest.dMul avgt 25 1154.919 ± 2.077 us/op
MaxMinOptimizeTest.fAdd avgt 25 751.325 ± 7.169 us/op
MaxMinOptimizeTest.fMax avgt 25 1306.131 ± 1.102 us/op
MaxMinOptimizeTest.fMin avgt 25 1306.134 ± 0.957 us/op
MaxMinOptimizeTest.fMul avgt 25 750.968 ± 7.334 us/op
Regards, Vladimir
[1] https://github.com/VladimirKempik/jdk/commit/fda44a8521f19b25d0fe155531d4bd1e3d7870a5
> 22 нояб. 2022 г., в 12:05, Vladimir Kempik <vladimir.kempik at gmail.com> написал(а):
>
> Hello Fei
>
> I think I can reduce the amount of opcodes for second version, but I need a second temp register for that ( to AND two results of fclass and check it just once for NaN)
> then it would look like:
>
> is_double ? fclass_d(t0, src1)
> : fclass_s(t0, src1);
> is_double ? fclass_d(t1, src2)
> : fclass_s(t1, src2);
> and(t0, t0, t1);
> andi(t0, t0, 0b1100000000); //if any of src is quiet or signaling NaN then return their sum
> beqz(t0, Compare);
> is_double ? fadd_d(dst, src1, src2)
> : fadd_s(dst, src1, src2);
> j(Done);
>
> bind(Compare);
>
> Any Hints on how to get a second temp register ?
>
> Regards, Vladimir
>
>> 22 нояб. 2022 г., в 11:28, Vladimir Kempik <vladimir.kempik at gmail.com> написал(а):
>>
>> Hello
>>
>> Found an issue with fadd+fclass version:
>>
>> jdk/incubator/vector/FloatMaxVectorTests.java
>>
>> test FloatMaxVectorTests.MAXReduceFloatMaxVectorTests(float[i * 5]): success
>> test FloatMaxVectorTests.MAXReduceFloatMaxVectorTests(float[i + 1]): success
>> test FloatMaxVectorTests.MAXReduceFloatMaxVectorTests(float[cornerCaseValue(i)]): failure
>> java.lang.AssertionError: at index #2 expected [Infinity] but found [NaN]
>> at org.testng.Assert.fail(Assert.java:99)
>> --
>> test FloatMaxVectorTests.MAXReduceFloatMaxVectorTestsMasked(float[i * 5], mask[i % 2]): success
>> test FloatMaxVectorTests.MAXReduceFloatMaxVectorTestsMasked(float[i + 1], mask[i % 2]): success
>> test FloatMaxVectorTests.MAXReduceFloatMaxVectorTestsMasked(float[cornerCaseValue(i)], mask[i % 2]): failure
>> java.lang.AssertionError: at index #10 expected [Infinity] but found [NaN]
>> at org.testng.Assert.fail(Assert.java:99)
>> --
>> test FloatMaxVectorTests.MAXReduceFloatMaxVectorTestsMasked(float[i * 5], mask[true]): success
>> test FloatMaxVectorTests.MAXReduceFloatMaxVectorTestsMasked(float[i + 1], mask[true]): success
>> test FloatMaxVectorTests.MAXReduceFloatMaxVectorTestsMasked(float[cornerCaseValue(i)], mask[true]): failure
>> java.lang.AssertionError: at index #2 expected [Infinity] but found [NaN]
>> at org.testng.Assert.fail(Assert.java:99)
>> --
>> test FloatMaxVectorTests.MINReduceFloatMaxVectorTests(float[i * 5]): success
>> test FloatMaxVectorTests.MINReduceFloatMaxVectorTests(float[i + 1]): success
>> test FloatMaxVectorTests.MINReduceFloatMaxVectorTests(float[cornerCaseValue(i)]): failure
>> java.lang.AssertionError: at index #2 expected [-Infinity] but found [NaN]
>> at org.testng.Assert.fail(Assert.java:99)
>> --
>> test FloatMaxVectorTests.MINReduceFloatMaxVectorTestsMasked(float[i * 5], mask[i % 2]): success
>> test FloatMaxVectorTests.MINReduceFloatMaxVectorTestsMasked(float[i + 1], mask[i % 2]): success
>> test FloatMaxVectorTests.MINReduceFloatMaxVectorTestsMasked(float[cornerCaseValue(i)], mask[i % 2]): failure
>> java.lang.AssertionError: at index #2 expected [-Infinity] but found [NaN]
>> at org.testng.Assert.fail(Assert.java:99)
>> --
>> test FloatMaxVectorTests.MINReduceFloatMaxVectorTestsMasked(float[i * 5], mask[true]): success
>> test FloatMaxVectorTests.MINReduceFloatMaxVectorTestsMasked(float[i + 1], mask[true]): success
>> test FloatMaxVectorTests.MINReduceFloatMaxVectorTestsMasked(float[cornerCaseValue(i)], mask[true]): failure
>> java.lang.AssertionError: at index #2 expected [-Infinity] but found [NaN]
>> at org.testng.Assert.fail(Assert.java:99)
>>
>>
>> And 2fclass version ( checking every src argument to be NaN) doesn’t have this issue.
>> So I think I’ll have to go v2 way.
>>
>> Regards, Vladimir.
>>
>>> 18 нояб. 2022 г., в 12:49, Vladimir Kempik <vladimir.kempik at gmail.com> написал(а):
>>>
>>> Hello
>>>
>>> Thanks for taking a look.
>>> I think v1 is better too.
>>>
>>> We have measured fadd latency with lmbench ( can’t remember which platform it was, hifive or thead) and it turned out to be just 4 cycles, that’s ok I think.
>>>
>>> I have also benched it on third platform - shallow OoO with dual-issue fpu, on fpga, and it showed gains similar to thead’s.
>>>
>>> I’ll run jtreg’s tiers and submit PR afterwards.
>>>
>>> Regards, Vladimir
>>>
>>>> 18 нояб. 2022 г., в 11:06, yangfei at iscas.ac.cn написал(а):
>>>>
>>>> Hi,
>>>>
>>>>
>>>> I went through both versions and looks like the resulting performance gain will depend on the micro-architecture implementations.
>>>>
>>>> Personally I prefer the first version in respect of instruction count (5 compared with 7 instructions when the inputs are not NaNs) and code readability.
>>>>
>>>> PS: I would suggest also carry out more conformance/compartibility test as needed for these changes.
>>>>
>>>>
>>>>
>>>> Thanks,
>>>>
>>>> Fei
>>>>
>>>>
>>>> -----Original Messages-----
>>>> From:"Vladimir Kempik" <vladimir.kempik at gmail.com>
>>>> Sent Time:2022-11-15 15:55:32 (Tuesday)
>>>> To: riscv-port-dev <riscv-port-dev at openjdk.org>
>>>> Cc:
>>>> Subject: Pre-Review: improving Math.min/max on floats
>>>>
>>>> Hello
>>>> Currently, in C2, Math.min/max is implemented in c2_MacroAssembler_riscv.cpp using
>>>>
>>>> void C2_MacroAssembler::minmax_FD(FloatRegister dst, FloatRegister src1, FloatRegister src2, bool is_double, bool is_min)
>>>>
>>>> The main issue there is Min/Max is required to return NaN if any of its arguments is NaN. In risc-v, fmin/fmax returns NaN only if both of src registers is NaN ( quiet NaN).
>>>> That requires additional logic to handle the case where only of of src is NaN.
>>>> Currently it’s done this way ( i’ve reduced is_double and is_min case for readability)
>>>>
>>>>
>>>> fmax_s(dst, src1, src2);
>>>> // Checking NaNs
>>>> flt_s(zr, src1, src2);
>>>>
>>>> frflags(t0);
>>>> beqz(t0, Done);
>>>>
>>>> // In case of NaNs
>>>> fadd_s(dst, src1, src2);
>>>>
>>>> bind(Done);
>>>>
>>>>
>>>> here we always do two float comparisons ( one in fmax, one in flt), perf shows they are taking equal time ( checking on thead c910)
>>>>
>>>> I think that’s suboptimal and can be improved: first, move the check before fmin/fmax and if check fails return NaN without doing fmax
>>>> second thing:
>>>>
>>>> I have prepared two version, first one [1] sums src1 and src2, if result is NaN - return it, result is checked with fclass, checking for quiet NaN and signaling NaN.
>>>> if result of sum is not NaN - do fmax and return result.
>>>>
>>>> second version [2] checks both src1 and src2 for being NaN with fclass, without doing any FP arithmetics. if any of them is NaN - return NaN, otherwise do the fmax.
>>>>
>>>> I have built both versions and compared results to unpatched JDK on hifive unmatched and thead c910.
>>>> While on hifive the perf win is moderate ( ~10%), on thead I’m getting up to 10x better results sometimes.
>>>>
>>>> MicroBenches fAdd/fMul/dAdd/dMul doesn’t show any difference, I think that happens because these
>>>> private double dAddBench(double a, double b) { return Math.max(a, b) + Math.min(a, b); } private double dMulBench(double a, double b) { return Math.max(a, b) * Math.min(a, b); }
>>>> may get reduces to just a + b and a*b respectively.
>>>>
>>>> Looking for opinions, which way is better.
>>>>
>>>> The results, thead c910:
>>>>
>>>> before
>>>>
>>>> Benchmark Mode Cnt Score Error Units
>>>> FpMinMaxIntrinsics.dMax avgt 25 54023.827 ± 268.645 ns/op
>>>> FpMinMaxIntrinsics.dMin avgt 25 54309.850 ± 323.551 ns/op
>>>> FpMinMaxIntrinsics.dMinReduce avgt 25 42192.140 ± 12.114 ns/op
>>>> FpMinMaxIntrinsics.fMax avgt 25 53797.657 ± 15.816 ns/op
>>>> FpMinMaxIntrinsics.fMin avgt 25 54135.710 ± 313.185 ns/op
>>>> FpMinMaxIntrinsics.fMinReduce avgt 25 42196.156 ± 13.424 ns/op
>>>> MaxMinOptimizeTest.dAdd avgt 25 650.810 ± 169.998 us/op
>>>> MaxMinOptimizeTest.dMax avgt 25 4561.967 ± 40.367 us/op
>>>> MaxMinOptimizeTest.dMin avgt 25 4589.100 ± 75.854 us/op
>>>> MaxMinOptimizeTest.dMul avgt 25 759.821 ± 240.092 us/op
>>>> MaxMinOptimizeTest.fAdd avgt 25 300.137 ± 13.495 us/op
>>>> MaxMinOptimizeTest.fMax avgt 25 4348.885 ± 20.061 us/op
>>>> MaxMinOptimizeTest.fMin avgt 25 4372.799 ± 27.296 us/op
>>>> MaxMinOptimizeTest.fMul avgt 25 304.024 ± 12.120 us/op
>>>>
>>>> fadd+fclass
>>>>
>>>> Benchmark Mode Cnt Score Error Units
>>>> FpMinMaxIntrinsics.dMax avgt 25 10545.196 ± 140.137 ns/op
>>>> FpMinMaxIntrinsics.dMin avgt 25 10454.525 ± 9.972 ns/op
>>>> FpMinMaxIntrinsics.dMinReduce avgt 25 3104.703 ± 0.892 ns/op
>>>> FpMinMaxIntrinsics.fMax avgt 25 10449.709 ± 7.284 ns/op
>>>> FpMinMaxIntrinsics.fMin avgt 25 10445.261 ± 7.206 ns/op
>>>> FpMinMaxIntrinsics.fMinReduce avgt 25 3104.769 ± 0.951 ns/op
>>>> MaxMinOptimizeTest.dAdd avgt 25 487.769 ± 170.711 us/op
>>>> MaxMinOptimizeTest.dMax avgt 25 929.394 ± 158.697 us/op
>>>> MaxMinOptimizeTest.dMin avgt 25 864.230 ± 284.794 us/op
>>>> MaxMinOptimizeTest.dMul avgt 25 894.116 ± 342.550 us/op
>>>> MaxMinOptimizeTest.fAdd avgt 25 284.664 ± 1.446 us/op
>>>> MaxMinOptimizeTest.fMax avgt 25 384.388 ± 15.004 us/op
>>>> MaxMinOptimizeTest.fMin avgt 25 371.952 ± 15.295 us/op
>>>> MaxMinOptimizeTest.fMul avgt 25 305.226 ± 12.467 us/op
>>>>
>>>>
>>>> 2fclass
>>>> Benchmark Mode Cnt Score Error Units
>>>> FpMinMaxIntrinsics.dMax avgt 25 11415.817 ± 403.757 ns/op
>>>> FpMinMaxIntrinsics.dMin avgt 25 11835.521 ± 329.380 ns/op
>>>> FpMinMaxIntrinsics.dMinReduce avgt 25 5188.436 ± 3.723 ns/op
>>>> FpMinMaxIntrinsics.fMax avgt 25 11667.456 ± 426.731 ns/op
>>>> FpMinMaxIntrinsics.fMin avgt 25 11646.682 ± 416.883 ns/op
>>>> FpMinMaxIntrinsics.fMinReduce avgt 25 5190.395 ± 3.628 ns/op
>>>> MaxMinOptimizeTest.dAdd avgt 25 745.417 ± 209.376 us/op
>>>> MaxMinOptimizeTest.dMax avgt 25 581.580 ± 38.046 us/op
>>>> MaxMinOptimizeTest.dMin avgt 25 533.442 ± 41.184 us/op
>>>> MaxMinOptimizeTest.dMul avgt 25 654.667 ± 267.537 us/op
>>>> MaxMinOptimizeTest.fAdd avgt 25 294.606 ± 11.712 us/op
>>>> MaxMinOptimizeTest.fMax avgt 25 433.842 ± 3.935 us/op
>>>> MaxMinOptimizeTest.fMin avgt 25 434.727 ± 1.894 us/op
>>>> MaxMinOptimizeTest.fMul avgt 25 305.385 ± 12.980 us/op
>>>>
>>>>
>>>> hifive:
>>>>
>>>> before
>>>> Benchmark Mode Cnt Score Error Units
>>>> FpMinMaxIntrinsics.dMax avgt 25 30219.666 ± 12.878 ns/op
>>>> FpMinMaxIntrinsics.dMin avgt 25 30242.249 ± 31.374 ns/op
>>>> FpMinMaxIntrinsics.dMinReduce avgt 25 15394.622 ± 2.803 ns/op
>>>> FpMinMaxIntrinsics.fMax avgt 25 30150.114 ± 22.421 ns/op
>>>> FpMinMaxIntrinsics.fMin avgt 25 30149.752 ± 20.813 ns/op
>>>> FpMinMaxIntrinsics.fMinReduce avgt 25 15396.402 ± 4.251 ns/op
>>>> MaxMinOptimizeTest.dAdd avgt 25 1143.582 ± 4.444 us/op
>>>> MaxMinOptimizeTest.dMax avgt 25 2556.317 ± 3.795 us/op
>>>> MaxMinOptimizeTest.dMin avgt 25 2556.569 ± 2.274 us/op
>>>> MaxMinOptimizeTest.dMul avgt 25 1142.769 ± 1.593 us/op
>>>> MaxMinOptimizeTest.fAdd avgt 25 748.688 ± 7.342 us/op
>>>> MaxMinOptimizeTest.fMax avgt 25 2280.381 ± 1.535 us/op
>>>> MaxMinOptimizeTest.fMin avgt 25 2280.760 ± 1.532 us/op
>>>> MaxMinOptimizeTest.fMul avgt 25 748.991 ± 7.261 us/op
>>>>
>>>> fadd+fclass
>>>>
>>>> Benchmark Mode Cnt Score Error Units
>>>> FpMinMaxIntrinsics.dMax avgt 25 27723.791 ± 22.784 ns/op
>>>> FpMinMaxIntrinsics.dMin avgt 25 27760.799 ± 45.411 ns/op
>>>> FpMinMaxIntrinsics.dMinReduce avgt 25 12875.949 ± 2.829 ns/op
>>>> FpMinMaxIntrinsics.fMax avgt 25 25992.753 ± 23.788 ns/op
>>>> FpMinMaxIntrinsics.fMin avgt 25 25994.554 ± 32.060 ns/op
>>>> FpMinMaxIntrinsics.fMinReduce avgt 25 11200.737 ± 2.169 ns/op
>>>> MaxMinOptimizeTest.dAdd avgt 25 1144.128 ± 4.371 us/op
>>>> MaxMinOptimizeTest.dMax avgt 25 1968.145 ± 2.346 us/op
>>>> MaxMinOptimizeTest.dMin avgt 25 1970.249 ± 4.712 us/op
>>>> MaxMinOptimizeTest.dMul avgt 25 1143.356 ± 2.203 us/op
>>>> MaxMinOptimizeTest.fAdd avgt 25 748.634 ± 7.229 us/op
>>>> MaxMinOptimizeTest.fMax avgt 25 1523.719 ± 0.570 us/op
>>>> MaxMinOptimizeTest.fMin avgt 25 1524.534 ± 1.109 us/op
>>>> MaxMinOptimizeTest.fMul avgt 25 748.643 ± 7.291 us/op
>>>>
>>>> 2fclass
>>>>
>>>> Benchmark Mode Cnt Score Error Units
>>>> FpMinMaxIntrinsics.dMax avgt 25 26890.963 ± 13.928 ns/op
>>>> FpMinMaxIntrinsics.dMin avgt 25 26919.595 ± 23.140 ns/op
>>>> FpMinMaxIntrinsics.dMinReduce avgt 25 11928.938 ± 1.985 ns/op
>>>> FpMinMaxIntrinsics.fMax avgt 25 26843.782 ± 27.956 ns/op
>>>> FpMinMaxIntrinsics.fMin avgt 25 26825.124 ± 24.104 ns/op
>>>> FpMinMaxIntrinsics.fMinReduce avgt 25 11927.765 ± 1.238 ns/op
>>>> MaxMinOptimizeTest.dAdd avgt 25 1144.860 ± 3.467 us/op
>>>> MaxMinOptimizeTest.dMax avgt 25 1881.809 ± 1.986 us/op
>>>> MaxMinOptimizeTest.dMin avgt 25 1882.623 ± 2.225 us/op
>>>> MaxMinOptimizeTest.dMul avgt 25 1142.860 ± 1.755 us/op
>>>> MaxMinOptimizeTest.fAdd avgt 25 752.557 ± 8.708 us/op
>>>> MaxMinOptimizeTest.fMax avgt 25 1587.139 ± 0.903 us/op
>>>> MaxMinOptimizeTest.fMin avgt 25 1587.140 ± 1.067 us/op
>>>> MaxMinOptimizeTest.fMul avgt 25 748.653 ± 7.278 us/op
>>>>
>>>> Regards, Vladimir
>>>>
>>>> P.S. for some reason I can’t use mv opcode on two FloatRegisters ( I think it was possible before) and had to use fmv_s/fmv_d which might be not exactly what I want.
>>>>
>>>> [1] https://github.com/VladimirKempik/jdk/commit/b6752492f7efd82e248e49e136dc9f5929cc19a2
>>>> [2] https://github.com/VladimirKempik/jdk/commit/384efc3ca59c2e301ec43f8d716f142828d2ac6a
>>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.openjdk.org/pipermail/riscv-port-dev/attachments/20221123/7de4f525/attachment-0001.htm>
More information about the riscv-port-dev
mailing list