8345298: RISC-V: Add riscv backend for Float16 operations - scalar

[Hamlin-Li]

Hi,
Can you help to review this patch?
It's an implementation of #22754 on riscv.

Performance

data

Benchmark	(vectorDim)	Mode	Cnt	Score -master	Error	Score - patch	Error	Units	Improvement (master/patch)
Float16OperationsBenchmark.absBenchmark	256	avgt	10	219.564	0.076	219.597	0.081	ns/op	1
Float16OperationsBenchmark.absBenchmark	512	avgt	10	358.873	0.575	355.011	0.07	ns/op	1.011
Float16OperationsBenchmark.absBenchmark	1024	avgt	10	582.361	0.189	581.832	0.006	ns/op	1.001
Float16OperationsBenchmark.absBenchmark	2048	avgt	10	1035.633	0.239	1034.854	0.284	ns/op	1.001
Float16OperationsBenchmark.addBenchmark	256	avgt	10	4951.702	0.194	2593.835	0.066	ns/op	1.909
Float16OperationsBenchmark.addBenchmark	512	avgt	10	9867.909	0.314	5167.568	0.162	ns/op	1.91
Float16OperationsBenchmark.addBenchmark	1024	avgt	10	21324.318	1.651	10016.456	1.07	ns/op	2.129
Float16OperationsBenchmark.addBenchmark	2048	avgt	10	42618.969	3.877	19985.662	1.233	ns/op	2.132
Float16OperationsBenchmark.cosineSimilarityDequantizedFP16	256	avgt	10	2811.45	0.441	2701.419	140.699	ns/op	1.041
Float16OperationsBenchmark.cosineSimilarityDequantizedFP16	512	avgt	10	5568.561	0.654	5577.598	1.123	ns/op	0.998
Float16OperationsBenchmark.cosineSimilarityDequantizedFP16	1024	avgt	10	11109.108	1.7	11095.644	0.644	ns/op	1.001
Float16OperationsBenchmark.cosineSimilarityDequantizedFP16	2048	avgt	10	20017.095	0.778	21560.165	0.515	ns/op	0.928
Float16OperationsBenchmark.cosineSimilarityDoubleRoundingFP16	256	avgt	10	20864.303	23.768	1345.192	0.274	ns/op	15.51
Float16OperationsBenchmark.cosineSimilarityDoubleRoundingFP16	512	avgt	10	43596.262	102.075	2580.035	0.397	ns/op	16.898
Float16OperationsBenchmark.cosineSimilarityDoubleRoundingFP16	1024	avgt	10	91565.818	250.761	5191.12	64.598	ns/op	17.639
Float16OperationsBenchmark.cosineSimilarityDoubleRoundingFP16	2048	avgt	10	185679.064	617.369	10343	20.025	ns/op	17.952
Float16OperationsBenchmark.cosineSimilaritySingleRoundingFP16	256	avgt	10	23473.486	0.498	1856.59	0.344	ns/op	12.643
Float16OperationsBenchmark.cosineSimilaritySingleRoundingFP16	512	avgt	10	47056.983	1.335	3833.72	2.487	ns/op	12.274
Float16OperationsBenchmark.cosineSimilaritySingleRoundingFP16	1024	avgt	10	94193.06	2.748	7054.642	1.243	ns/op	13.352
Float16OperationsBenchmark.cosineSimilaritySingleRoundingFP16	2048	avgt	10	188399.012	376.598	13982.963	1.533	ns/op	13.473
Float16OperationsBenchmark.divBenchmark	256	avgt	10	6865.998	1.289	4029.496	1.093	ns/op	1.704
Float16OperationsBenchmark.divBenchmark	512	avgt	10	13908.829	32.138	8034.805	0.133	ns/op	1.731
Float16OperationsBenchmark.divBenchmark	1024	avgt	10	28975.886	0.595	15674.755	5.269	ns/op	1.849
Float16OperationsBenchmark.divBenchmark	2048	avgt	10	57924.129	4.542	31420.095	96.01	ns/op	1.844
Float16OperationsBenchmark.euclideanDistanceDequantizedFP16	256	avgt	10	3301.349	1.798	3150.946	0.722	ns/op	1.048
Float16OperationsBenchmark.euclideanDistanceDequantizedFP16	512	avgt	10	6325.897	3.003	6252.479	0.15	ns/op	1.012
Float16OperationsBenchmark.euclideanDistanceDequantizedFP16	1024	avgt	10	12691.852	3.678	12439.328	0.78	ns/op	1.02
Float16OperationsBenchmark.euclideanDistanceDequantizedFP16	2048	avgt	10	25219.846	3.786	24731.585	151.057	ns/op	1.02
Float16OperationsBenchmark.euclideanDistanceFP16	256	avgt	10	11596.839	8.701	1119.235	0.033	ns/op	10.361
Float16OperationsBenchmark.euclideanDistanceFP16	512	avgt	10	23342.733	4.513	2131.066	0.231	ns/op	10.954
Float16OperationsBenchmark.euclideanDistanceFP16	1024	avgt	10	46112.17	44.496	4152.953	0.146	ns/op	11.103
Float16OperationsBenchmark.euclideanDistanceFP16	2048	avgt	10	93860.737	107.312	8198.33	1.291	ns/op	11.449
Float16OperationsBenchmark.fmaBenchmark	256	avgt	10	75400.936	60.886	2681.222	0.111	ns/op	28.122
Float16OperationsBenchmark.fmaBenchmark	512	avgt	10	139251.977	619.017	3703.767	0.471	ns/op	37.597
Float16OperationsBenchmark.fmaBenchmark	1024	avgt	10	267586.32	789.884	7286.745	0.152	ns/op	36.722
Float16OperationsBenchmark.fmaBenchmark	2048	avgt	10	509900.498	656.287	14470.574	2.257	ns/op	35.237
Float16OperationsBenchmark.getExponentBenchmark	256	avgt	10	684.578	0.486	683.204	0.101	ns/op	1.002
Float16OperationsBenchmark.getExponentBenchmark	512	avgt	10	1325.275	1.96	1323.645	0.92	ns/op	1.001
Float16OperationsBenchmark.getExponentBenchmark	1024	avgt	10	2641.872	15.528	2605.375	0.261	ns/op	1.014
Float16OperationsBenchmark.getExponentBenchmark	2048	avgt	10	5028.168	1.879	5023.716	2.348	ns/op	1.001
Float16OperationsBenchmark.isFiniteBenchmark	256	avgt	10	612.872	0.68	613.996	1.142	ns/op	0.998
Float16OperationsBenchmark.isFiniteBenchmark	512	avgt	10	1205.306	2.812	1195.51	4.283	ns/op	1.008
Float16OperationsBenchmark.isFiniteBenchmark	1024	avgt	10	2335.283	17.772	2332.876	0.349	ns/op	1.001
Float16OperationsBenchmark.isFiniteBenchmark	2048	avgt	10	4697.091	10.683	5051.694	39.746	ns/op	0.93
Float16OperationsBenchmark.isFiniteCMovBenchmark	256	avgt	10	914.27	0.704	914.195	0.202	ns/op	1
Float16OperationsBenchmark.isFiniteCMovBenchmark	512	avgt	10	1844.587	6.638	1814.631	5.204	ns/op	1.017
Float16OperationsBenchmark.isFiniteCMovBenchmark	1024	avgt	10	3595.557	0.453	3597.302	0.594	ns/op	1
Float16OperationsBenchmark.isFiniteCMovBenchmark	2048	avgt	10	7157.07	0.673	7155.312	0.37	ns/op	1
Float16OperationsBenchmark.isFiniteStoreBenchmark	256	avgt	10	677.109	0.214	677.146	0.249	ns/op	1
Float16OperationsBenchmark.isFiniteStoreBenchmark	512	avgt	10	1336.623	1.292	1336.019	0.067	ns/op	1
Float16OperationsBenchmark.isFiniteStoreBenchmark	1024	avgt	10	2640.22	10.33	2637.142	0.464	ns/op	1.001
Float16OperationsBenchmark.isFiniteStoreBenchmark	2048	avgt	10	5249.69	1.726	5255.893	23.545	ns/op	0.999
Float16OperationsBenchmark.isInfiniteBenchmark	256	avgt	10	728.603	1.596	693.678	1.057	ns/op	1.05
Float16OperationsBenchmark.isInfiniteBenchmark	512	avgt	10	1348.349	0.21	1297.384	1.692	ns/op	1.039
Float16OperationsBenchmark.isInfiniteBenchmark	1024	avgt	10	2610.138	0.492	2518.27	5.889	ns/op	1.036
Float16OperationsBenchmark.isInfiniteBenchmark	2048	avgt	10	5132.662	0.833	4908.692	3.115	ns/op	1.046
Float16OperationsBenchmark.isInfiniteCMovBenchmark	256	avgt	10	915.357	0.228	915.45	0.464	ns/op	1
Float16OperationsBenchmark.isInfiniteCMovBenchmark	512	avgt	10	1769.662	0.53	1767.386	0.119	ns/op	1.001
Float16OperationsBenchmark.isInfiniteCMovBenchmark	1024	avgt	10	3470.309	0.284	3472.84	1.01	ns/op	0.999
Float16OperationsBenchmark.isInfiniteCMovBenchmark	2048	avgt	10	6932.564	3.019	6932.776	2.647	ns/op	1
Float16OperationsBenchmark.isInfiniteStoreBenchmark	256	avgt	10	730.842	0.085	730.841	0.308	ns/op	1
Float16OperationsBenchmark.isInfiniteStoreBenchmark	512	avgt	10	1434.546	0.066	1434.594	0.344	ns/op	1
Float16OperationsBenchmark.isInfiniteStoreBenchmark	1024	avgt	10	2321.078	0.048	2320.688	1.318	ns/op	1
Float16OperationsBenchmark.isInfiniteStoreBenchmark	2048	avgt	10	4465.404	3.87	4463.767	1.083	ns/op	1
Float16OperationsBenchmark.isNaNBenchmark	256	avgt	10	636.582	0.118	637.523	0.012	ns/op	0.999
Float16OperationsBenchmark.isNaNBenchmark	512	avgt	10	1252.001	0.199	1251.084	0.227	ns/op	1.001
Float16OperationsBenchmark.isNaNBenchmark	1024	avgt	10	2477.057	0.761	2476.489	0.378	ns/op	1
Float16OperationsBenchmark.isNaNBenchmark	2048	avgt	10	4923.221	0.192	4920.501	0.247	ns/op	1.001
Float16OperationsBenchmark.isNaNCMovBenchmark	256	avgt	10	689.366	2.118	680.408	0.073	ns/op	1.013
Float16OperationsBenchmark.isNaNCMovBenchmark	512	avgt	10	1340.416	0.228	1341.398	0.411	ns/op	0.999
Float16OperationsBenchmark.isNaNCMovBenchmark	1024	avgt	10	2651.651	0.392	2649.589	0.475	ns/op	1.001
Float16OperationsBenchmark.isNaNCMovBenchmark	2048	avgt	10	5261.672	0.906	5312.504	7.238	ns/op	0.99
Float16OperationsBenchmark.isNaNStoreBenchmark	256	avgt	10	642.548	0.288	640.636	0.128	ns/op	1.003
Float16OperationsBenchmark.isNaNStoreBenchmark	512	avgt	10	1258.312	0.368	1258.208	1.662	ns/op	1
Float16OperationsBenchmark.isNaNStoreBenchmark	1024	avgt	10	2477.097	0.654	2476.883	0.623	ns/op	1
Float16OperationsBenchmark.isNaNStoreBenchmark	2048	avgt	10	4932.261	2.213	4933.748	6.211	ns/op	1
Float16OperationsBenchmark.maxBenchmark	256	avgt	10	6262.318	0.184	3737.469	0.442	ns/op	1.676
Float16OperationsBenchmark.maxBenchmark	512	avgt	10	13346.887	9.165	7427.994	0.388	ns/op	1.797
Float16OperationsBenchmark.maxBenchmark	1024	avgt	10	26453.036	4.278	14793.968	2.125	ns/op	1.788
Float16OperationsBenchmark.maxBenchmark	2048	avgt	10	52887.135	1.596	29524.416	5.099	ns/op	1.791
Float16OperationsBenchmark.minBenchmark	256	avgt	10	6294.667	24.257	3764.081	2.646	ns/op	1.672
Float16OperationsBenchmark.minBenchmark	512	avgt	10	13246.531	2.183	7426.716	0.868	ns/op	1.784
Float16OperationsBenchmark.minBenchmark	1024	avgt	10	26772.888	457.212	14792.145	0.623	ns/op	1.81
Float16OperationsBenchmark.minBenchmark	2048	avgt	10	53933.3	820.653	29521.248	0.755	ns/op	1.827
Float16OperationsBenchmark.mulBenchmark	256	avgt	10	4953.512	1.465	2596.331	0.089	ns/op	1.908
Float16OperationsBenchmark.mulBenchmark	512	avgt	10	9905.903	51.296	5168.052	0.773	ns/op	1.917
Float16OperationsBenchmark.mulBenchmark	1024	avgt	10	21513.646	31.823	10015.796	0.14	ns/op	2.148
Float16OperationsBenchmark.mulBenchmark	2048	avgt	10	42641.223	1.12	19990.315	0.805	ns/op	2.133
Float16OperationsBenchmark.negateBenchmark	256	avgt	10	218.717	0.13	218.341	0.006	ns/op	1.002
Float16OperationsBenchmark.negateBenchmark	512	avgt	10	352.992	0.189	356.34	0.384	ns/op	0.991
Float16OperationsBenchmark.negateBenchmark	1024	avgt	10	581.243	3.774	581.086	0.08	ns/op	1
Float16OperationsBenchmark.negateBenchmark	2048	avgt	10	1034.852	0.629	1034.441	0.462	ns/op	1
Float16OperationsBenchmark.sqrtBenchmark	256	avgt	10	49892.244	115.282	3807.597	0.089	ns/op	13.103
Float16OperationsBenchmark.sqrtBenchmark	512	avgt	10	30012.543	14.094	7678.139	1.214	ns/op	3.909
Float16OperationsBenchmark.sqrtBenchmark	1024	avgt	10	59690.622	23.593	15406.483	0.357	ns/op	3.874
Float16OperationsBenchmark.sqrtBenchmark	2048	avgt	10	118895.955	19.41	30877.922	0.35	ns/op	3.851
Float16OperationsBenchmark.subBenchmark	256	avgt	10	4991.926	6.009	2593.842	0.047	ns/op	1.925
Float16OperationsBenchmark.subBenchmark	512	avgt	10	9871.741	4.777	5165.983	0.325	ns/op	1.911
Float16OperationsBenchmark.subBenchmark	1024	avgt	10	21324.947	2.31	10015.511	0.23	ns/op	2.129
Float16OperationsBenchmark.subBenchmark	2048	avgt	10	42604.606	2.182	19992.064	0.818	ns/op	2.131

---

Progress

• Change must be properly reviewed (1 review required, with at least 1 Reviewer)
• Change must not contain extraneous whitespace
• Commit message must refer to an issue

Issue

• JDK-8345298: RISC-V: Add riscv backend for Float16 operations - scalar (Enhancement - P4)

Reviewers

• Fei Yang (@RealFYang - Reviewer) 🔄 Re-review required (review applies to e63061d2)
• Robbin Ehn (@robehn - Reviewer) 🔄 Re-review required (review applies to e63061d2)

Reviewing

Using git

Checkout this PR locally:
$ git fetch https://git.openjdk.org/jdk.git pull/23844/head:pull/23844
$ git checkout pull/23844

Update a local copy of the PR:
$ git checkout pull/23844
$ git pull https://git.openjdk.org/jdk.git pull/23844/head

Using Skara CLI tools

Checkout this PR locally:
$ git pr checkout 23844

View PR using the GUI difftool:
$ git pr show -t 23844

Using diff file

Download this PR as a diff file:
https://git.openjdk.org/jdk/pull/23844.diff

Using Webrev

Link to Webrev Comment

[bridgekeeper]

👋 Welcome back mli! A progress list of the required criteria for merging this PR into master will be added to the body of your pull request. There are additional pull request commands available for use with this pull request.

[openjdk]

@Hamlin-Li This change is no longer ready for integration - check the PR body for details.

[openjdk]

@Hamlin-Li this pull request can not be integrated into master due to one or more merge conflicts. To resolve these merge conflicts and update this pull request you can run the following commands in the local repository for your personal fork:

git checkout float16-scalar
git fetch https://git.openjdk.org/jdk.git master
git merge FETCH_HEAD
# resolve conflicts and follow the instructions given by git merge
git commit -m "Merge master"
git push

[openjdk]

@Hamlin-Li The following label will be automatically applied to this pull request:

• hotspot

When this pull request is ready to be reviewed, an "RFR" email will be sent to the corresponding mailing list. If you would like to change these labels, use the /label pull request command.

[mlbridge]

Webrevs

• 06: Full - Incremental (3957bbca)
• 05: Full - Incremental (e63061d2)
• 04: Full - Incremental (143869ff)
• 03: Full - Incremental (a6d36051)
• 02: Full - Incremental (83b3bc76)
• 01: Full (54bf239a)
• 00: Full (d7ebbe69)

[RealFYang]

Hi, Thanks for working on this part. Some initial comments after a cursory look.

[Hamlin-Li]

BTW, I just removed TestFloat16VectorConvChain test for riscv temporariely, as with this patch the ConvF2HF(AddF(ConvHF2F, ConvHF2F)) will be replaced by a ReinterpretHF2S(AddHF(ReinterpretS2HF, ReinterpretS2HF)).

In the short future, when we support vectorization of float16 operations, we can enable the test, but for riscv it will still be different form of Nodes, could be something like VectorReinterpretHF2S(VectorAddHF(VectorReinterpretS2HF, VectorReinterpretS2HF)).

[RealFYang]

Several more comments after a closer look.

[RealFYang]

BTW, I just removed TestFloat16VectorConvChain test for riscv temporariely, as with this patch the ConvF2HF(AddF(ConvHF2F, ConvHF2F)) will be replaced by a ReinterpretHF2S(AddHF(ReinterpretS2HF, ReinterpretS2HF)).

Maybe we should also update the @requires of the test at the same time?
Currently, it says | (os.arch == "riscv64" & vm.cpu.features ~= ".*zvfh.*").
Maybe we change zvfh into zfh?

[Hamlin-Li]

Maybe we should also update the @requires of the test at the same time? Currently, it says | (os.arch == "riscv64" & vm.cpu.features ~= ".*zvfh.*"). Maybe we change zvfh into zfh?

No, as this test is for "vector conversion chain", only support of zfh should not trigger the test. BTW, scalar tests are in other test files.
In the future, when we support vectorization the IR verification test should be enabled again, but it still depends on zvfh rather than zfh.
Hope this answer your question?

[RealFYang]

Maybe we should also update the @requires of the test at the same time? Currently, it says | (os.arch == "riscv64" & vm.cpu.features ~= ".*zvfh.*"). Maybe we change zvfh into zfh?

No, as this test is for "vector conversion chain", only support of zfh should not trigger the test. BTW, scalar tests are in other test files. In the future, when we support vectorization the IR verification test should be enabled again, but it still depends on zvfh rather than zfh. Hope this answer your question?

That makes sense to me. Thanks.

[RealFYang]

Thanks for the update. There minor comments remain. Seems fine to me otherwise.

[RealFYang]

Overall LGTM. Thanks for the updates.

[RealFYang]

Still good to me.

[Hamlin-Li]

Still good to me.

Thank you!

[robehn]

Seems alright, thanks!