RFR: 8320448: Accelerate IndexOf using AVX2 [v33]

[Daniel Jeliński]

On Thu, 23 May 2024 17:25:34 GMT, Scott Gibbons <sgibbons at openjdk.org> wrote:

>> Re-write the IndexOf code without the use of the pcmpestri instruction, only using AVX2 instructions.  This change accelerates String.IndexOf on average 1.3x for AVX2.  The benchmark numbers:
>> 
>> 
>> Benchmark	                                               Score		Latest		
>> StringIndexOf.advancedWithMediumSub   343.573		317.934		0.925375393x
>> StringIndexOf.advancedWithShortSub1	  1039.081		1053.96		1.014319384x
>> StringIndexOf.advancedWithShortSub2	      55.828		110.541		1.980027943x
>> StringIndexOf.constantPattern	                9.361		11.906		1.271872663x
>> StringIndexOf.searchCharLongSuccess	        4.216		4.218		1.000474383x
>> StringIndexOf.searchCharMediumSuccess	3.133		3.216		1.02649218x
>> StringIndexOf.searchCharShortSuccess	3.76		        3.761		1.000265957x
>> StringIndexOf.success	                                9.186		9.713		1.057369911x
>> StringIndexOf.successBig	                      14.341		46.343		3.231504079x
>> StringIndexOfChar.latin1_AVX2_String	  6220.918		12154.52		1.953814533x
>> StringIndexOfChar.latin1_AVX2_char	  5503.556		5540.044		1.006629895x
>> StringIndexOfChar.latin1_SSE4_String	  6978.854		6818.689		0.977049957x
>> StringIndexOfChar.latin1_SSE4_char	  5657.499		5474.624		0.967675646x
>> StringIndexOfChar.latin1_Short_String	  7132.541		6863.359		0.962260014x
>> StringIndexOfChar.latin1_Short_char	16013.389	      16162.437		1.009307711x
>> StringIndexOfChar.latin1_mixed_String	  7386.123	      14771.622		1.999915517x
>> StringIndexOfChar.latin1_mixed_char	  9901.671		9782.245		0.987938803
>
> Scott Gibbons has updated the pull request incrementally with one additional commit since the last revision:
> 
>   Fix for IndexOf.java on mac

src/hotspot/cpu/x86/stubGenerator_x86_64_string.cpp line 268:

> 266:   __ cmpq(needle_len_p, 0);
> 267:   __ jg_b(L_nextCheck);
> 268:   __ xorq(rax, rax);

out of curiosity, is there any advantage to using `xorq` instead of `xorl` here?

https://stackoverflow.com/a/33668295/7707617 suggests that `xorl` might be better, but it's a bit dated now.

src/hotspot/cpu/x86/stubGenerator_x86_64_string.cpp line 449:

> 447:   __ cmpq(r13, NUMBER_OF_CASES - 1);
> 448:   __ ja(L_smallCaseDefault);
> 449:   __ mov64(r15, (int64_t)small_jump_table);

would it make sense to use `lea` here?

src/hotspot/cpu/x86/stubGenerator_x86_64_string.cpp line 803:

> 801:     __ movq(index, needle_len);
> 802:     __ andq(index, 0xf);  // nLen % 16
> 803:     __ movq(offset, 0x10);

`movl` or `movptr` would produce a shorter encoding

src/hotspot/cpu/x86/stubGenerator_x86_64_string.cpp line 1544:

> 1542:   }
> 1543: 
> 1544:   __ align(8);

why `8` and not `OptoLoopAlignment` ?

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

PR Review Comment: https://git.openjdk.org/jdk/pull/16753#discussion_r1612178285
PR Review Comment: https://git.openjdk.org/jdk/pull/16753#discussion_r1612179069
PR Review Comment: https://git.openjdk.org/jdk/pull/16753#discussion_r1612180163
PR Review Comment: https://git.openjdk.org/jdk/pull/16753#discussion_r1612183311