aboutsummaryrefslogtreecommitdiff
AgeCommit message (Collapse)AuthorFilesLines
2024-05-27sha3.c: permute_n_avx2(): theta: remove unnecessary assignmentPaul Duncan1-2/+1
2024-05-27sha3.c: permute_n_avx2(): replace some permutes with blends, minor cleanupsPaul Duncan1-19/+19
with these changes: - clang: avx2 comparable to scalar - gcc: avx2 still slower than scalar bench results ------------- gcc scalar: > make clean all BACKEND=1 CC=gcc && ./bench info: cpucycles: version=20240318 implementation=amd64-pmc persecond=4800000000 info: backend=scalar num_trials=2000 src_lens=64,256,1024,4096,16384 dst_lens=32 function,dst_len,64,256,1024,4096,16384 sha3_224,28,19.5,10.0,9.9,9.0,8.8 sha3_256,32,19.5,10.0,9.9,9.5,9.3 sha3_384,48,19.5,14.7,12.3,12.2,12.0 sha3_512,64,19.5,19.6,18.2,17.1,17.1 shake128,32,19.6,9.9,8.7,7.8,7.6 shake256,32,19.6,9.9,10.0,9.5,9.3 gcc avx2: > make clean all BACKEND=6 CC=gcc && ./bench info: cpucycles: version=20240318 implementation=amd64-pmc persecond=4800000000 info: backend=avx2 num_trials=2000 src_lens=64,256,1024,4096,16384 dst_lens=32 function,dst_len,64,256,1024,4096,16384 sha3_224,28,24.5,12.3,12.2,11.1,10.9 sha3_256,32,24.4,12.2,12.2,11.9,11.6 sha3_384,48,24.2,18.3,15.3,15.2,15.0 sha3_512,64,24.5,24.4,22.8,21.6,21.6 shake128,32,24.6,12.1,10.8,9.6,9.4 shake256,32,24.7,12.2,12.2,11.8,11.6 clang scalar: > make clean all BACKEND=1 CC=clang && ./bench info: cpucycles: version=20240318 implementation=amd64-pmc persecond=4800000000 info: backend=scalar num_trials=2000 src_lens=64,256,1024,4096,16384 dst_lens=32 function,dst_len,64,256,1024,4096,16384 sha3_224,28,21.8,9.9,9.7,8.8,8.7 sha3_256,32,21.1,9.9,9.8,9.4,9.2 sha3_384,48,21.1,14.6,12.1,12.0,11.8 sha3_512,64,21.2,19.2,17.9,16.9,16.9 shake128,32,21.0,9.9,8.6,7.7,7.5 shake256,32,20.9,9.9,9.8,9.5,9.2 clang avx2: > make clean all BACKEND=6 CC=clang && ./bench info: cpucycles: version=20240318 implementation=amd64-pmc persecond=4800000000 info: backend=avx2 num_trials=2000 src_lens=64,256,1024,4096,16384 dst_lens=32 function,dst_len,64,256,1024,4096,16384 sha3_224,28,19.9,10.0,9.9,9.0,8.9 sha3_256,32,19.9,10.0,9.9,9.6,9.4 sha3_384,48,20.1,14.9,12.4,12.3,12.2 sha3_512,64,19.9,19.6,18.4,17.4,17.4 shake128,32,19.9,10.0,8.8,7.9,7.7 shake256,32,20.0,10.0,9.9,9.6,9.4
2024-05-27sha3.c: remove extraneous tabsPaul Duncan1-88/+88
2024-05-27sha3.c: permute_n_avx2(): remove LOAD/STORE macros, minor comment fixesPaul Duncan1-36/+23
2024-05-27sha3.c: permute_n_avx2(): CHI(): remove tabsPaul Duncan1-5/+5
2024-05-27sha3.c: permute_n_avx2(): replace two instances of permute4x64_epi64() with ↵Paul Duncan1-4/+2
shuffle_epi32()
2024-05-27sha3.c: permute_n_avx2(): rho: use ROLI instead of ROLV for hiPaul Duncan1-11/+10
2024-05-27sha3.c: permute_n_avx2(): remove K64Paul Duncan1-1/+1
2024-05-27sha3.c: permute_n_avx2(): clean up commentsPaul Duncan1-17/+12
2024-05-27sha3.c: permute_n_avx2(): add CHI()Paul Duncan1-59/+25
2024-05-26sha3.c: avx2: s/loadu_epi64/loadu_si256Paul Duncan1-10/+10
2024-05-26sha3.c: permute_n_avx2(): chi: explicit andnotPaul Duncan1-5/+5
bench results ------------- before (gcc, avx2): > make clean all BACKEND=6 CC=gcc && ./bench info: cpucycles: version=20240318 implementation=amd64-pmc persecond=4800000000 info: backend=avx2 num_trials=2000 src_lens=64,256,1024,4096,16384 dst_lens=32 function,dst_len,64,256,1024,4096,16384 sha3_224,28,30.8,15.3,15.4,14.0,13.8 sha3_256,32,30.7,15.3,15.4,15.0,14.6 sha3_384,48,30.6,23.0,19.2,19.2,19.0 sha3_512,64,30.2,30.7,28.8,27.4,27.4 shake128,32,30.6,15.3,13.5,12.1,11.9 shake256,32,30.8,15.4,15.4,15.0,14.6 after (gcc, avx2): > make clean all BACKEND=6 CC=gcc && ./bench info: cpucycles: version=20240318 implementation=amd64-pmc persecond=4800000000 info: backend=avx2 num_trials=2000 src_lens=64,256,1024,4096,16384 dst_lens=32 function,dst_len,64,256,1024,4096,16384 sha3_224,28,29.4,14.8,14.8,13.5,13.2 sha3_256,32,29.4,14.8,14.8,14.4,14.0 sha3_384,48,29.3,22.1,18.5,18.5,18.3 sha3_512,64,29.4,29.4,27.6,26.3,26.3 shake128,32,29.4,14.8,13.0,11.6,11.4 shake256,32,29.5,14.8,14.9,14.4,14.1
2024-05-26sha3.c: permute_n_avx2(): chi: use blend instead of permute where possible, ↵Paul Duncan1-16/+19
remove unnecessary ands
2024-05-26sha3.c: permute_n_avx2(): refactor lane masksPaul Duncan1-44/+34
2024-05-26sha3.c: permute_n_avx2(): pi: remove gather (better, but still slower than ↵Paul Duncan1-29/+46
scalar) bench results ------------- scalar (gcc): > make clean all BACKEND=1 CC=gcc && ./bench info: cpucycles: version=20240318 implementation=amd64-pmc persecond=4800000000 info: backend=scalar num_trials=2000 src_lens=64,256,1024,4096,16384 dst_lens=32 function,dst_len,64,256,1024,4096,16384 sha3_224,28,19.7,10.1,10.0,9.0,8.8 sha3_256,32,20.0,10.1,9.9,9.6,9.3 sha3_384,48,20.0,14.9,12.3,12.2,12.0 sha3_512,64,20.0,19.5,18.3,17.2,17.1 shake128,32,20.2,10.1,8.7,7.8,7.6 shake256,32,20.3,10.1,10.0,9.6,9.3 scalar (clang): > make clean all BACKEND=1 CC=clang && ./bench info: cpucycles: version=20240318 implementation=amd64-pmc persecond=4800000000 info: backend=scalar num_trials=2000 src_lens=64,256,1024,4096,16384 dst_lens=32 function,dst_len,64,256,1024,4096,16384 sha3_224,28,19.7,9.9,9.8,8.8,8.7 sha3_256,32,19.6,9.8,9.8,9.4,9.2 sha3_384,48,19.8,14.6,12.1,12.0,11.8 sha3_512,64,19.6,19.3,17.9,16.9,16.9 shake128,32,19.7,9.9,8.6,7.7,7.5 shake256,32,19.7,9.9,9.8,9.4,9.2 avx2 (gcc): > make clean all BACKEND=6 CC=gcc && ./bench info: cpucycles: version=20240318 implementation=amd64-pmc persecond=4800000000 info: backend=avx2 num_trials=2000 src_lens=64,256,1024,4096,16384 dst_lens=32 function,dst_len,64,256,1024,4096,16384 sha3_224,28,29.7,14.9,15.0,13.6,13.4 sha3_256,32,30.0,14.9,15.0,14.6,14.2 sha3_384,48,29.8,22.4,18.7,18.7,18.5 sha3_512,64,29.8,29.8,28.0,26.6,26.6 shake128,32,29.9,15.0,13.2,11.8,11.6 shake256,32,30.0,14.9,15.0,14.6,14.2 avx2 (clang): > make clean all BACKEND=6 CC=clang && ./bench info: cpucycles: version=20240318 implementation=amd64-pmc persecond=4800000000 info: backend=avx2 num_trials=2000 src_lens=64,256,1024,4096,16384 dst_lens=32 function,dst_len,64,256,1024,4096,16384 sha3_224,28,23.8,11.6,11.3,10.3,10.1 sha3_256,32,23.7,11.6,11.3,10.9,10.7 sha3_384,48,23.8,17.2,14.1,14.1,13.9 sha3_512,64,23.8,22.8,21.1,20.0,20.0 shake128,32,22.8,11.2,9.8,8.8,8.6 shake256,32,22.9,11.2,11.1,10.8,10.5 chi appears to be the culprit now; look at replacing mask/and/or with blend
2024-05-26sha3.c: permute_n_avx2(): rename theta permute idsPaul Duncan1-3/+9
2024-05-26tests/avx2: use permutes for piPaul Duncan1-13/+50
2024-05-26sha3.c: add initial avx2 backend (slow)Paul Duncan1-0/+249
bench results (oof) ------------------- scalar (gcc): > make clean all BACKEND=1 && ./bench ... info: cpucycles: version=20240318 implementation=amd64-pmc persecond=4800000000 info: backend=scalar num_trials=2000 src_lens=64,256,1024,4096,16384 dst_lens=32 function,dst_len,64,256,1024,4096,16384 sha3_224,28,19.5,10.0,10.0,8.9,8.8 sha3_256,32,19.5,10.0,9.9,9.5,9.3 sha3_384,48,19.5,14.7,12.3,12.2,12.0 sha3_512,64,19.5,19.4,18.2,17.2,17.1 shake128,32,19.7,9.9,8.7,7.8,7.6 shake256,32,20.2,10.1,10.0,9.6,9.3 scalar (clang): > make clean all BACKEND=1 CC=clang && ./bench info: cpucycles: version=20240318 implementation=amd64-pmc persecond=4800000000 info: backend=scalar num_trials=2000 src_lens=64,256,1024,4096,16384 dst_lens=32 function,dst_len,64,256,1024,4096,16384 sha3_224,28,19.7,9.9,9.8,8.8,8.7 sha3_256,32,19.7,9.9,9.8,9.4,9.2 sha3_384,48,19.7,14.6,12.1,12.0,11.8 sha3_512,64,19.7,19.3,17.8,16.9,16.9 shake128,32,19.7,9.8,8.6,7.7,7.5 shake256,32,19.7,9.9,9.8,9.4,9.2 avx2 (gcc): > make clean all BACKEND=6 && ./bench info: cpucycles: version=20240318 implementation=amd64-pmc persecond=4800000000 info: backend=avx2 num_trials=2000 src_lens=64,256,1024,4096,16384 dst_lens=32 function,dst_len,64,256,1024,4096,16384 sha3_224,28,80.0,40.5,40.9,37.2,36.6 sha3_256,32,79.8,40.5,40.9,39.8,38.9 sha3_384,48,80.1,61.0,51.2,51.3,50.7 sha3_512,64,79.9,81.5,76.7,73.0,73.1 shake128,32,86.0,43.6,38.6,34.5,33.9 shake256,32,86.0,43.6,44.1,42.8,41.9 avx2 (clang): > make clean all BACKEND=6 CC=clang && ./bench info: cpucycles: version=20240318 implementation=amd64-pmc persecond=4800000000 info: backend=avx2 num_trials=2000 src_lens=64,256,1024,4096,16384 dst_lens=32 function,dst_len,64,256,1024,4096,16384 sha3_224,28,62.0,31.6,31.9,29.0,28.5 sha3_256,32,61.9,31.6,32.0,31.1,30.3 sha3_384,48,62.6,47.7,40.1,40.2,39.8 sha3_512,64,62.0,63.8,60.2,57.2,57.3 shake128,32,62.5,31.7,27.9,24.9,24.5 shake256,32,62.0,31.5,31.9,31.0,30.3 the culprit is using gather in the pi step for permutes, so those will need to be rewritten as permutes: > perf annotate -Mintel --stdio --quiet permute_n_avx2 | grep gatherqq 24.09 : 207e: vpgatherqq ymm0,QWORD PTR [rdi+ymm3*8],ymm6 14.07 : 20a0: vpgatherqq ymm4,QWORD PTR [rdi+ymm3*8],ymm6 13.36 : 20aa: vpgatherqq ymm8,QWORD PTR [rdi+ymm7*8],ymm6 13.50 : 20c4: vpgatherqq ymm3,QWORD PTR [rdi+ymm7*8],ymm6 13.61 : 20ce: vpgatherqq ymm7,QWORD PTR [rdi+ymm1*8],ymm6
2024-05-26add tests/avx2Paul Duncan5-0/+477
2024-05-26add tests/diagrams/avx512.svgPaul Duncan1-0/+61
2024-05-26add tests/diagramsPaul Duncan2-0/+26
2024-05-17sha3.c: permute_n_avx512(): simplify num_rounds logicPaul Duncan1-5/+2
bench results (samish across the board) --------------------------------------- x86-64-avx512-gcc before: > make clean all BACKEND=2 && ./bench 10000 ... info: cpucycles: version=20240318 implementation=amd64-pmc persecond=4800000000 info: backend=avx512 num_trials=10000 src_lens=64,256,1024,4096,16384 dst_lens=32 function,dst_len,64,256,1024,4096,16384 sha3_224,28,15.4,7.8,7.8,7.1,7.0 sha3_256,32,15.4,7.8,7.8,7.6,7.4 sha3_384,48,15.4,11.6,9.8,9.8,9.7 sha3_512,64,15.5,15.5,14.5,13.8,13.8 shake128,32,15.6,7.8,6.9,6.2,6.1 shake256,32,15.6,7.8,7.9,7.6,7.4 x86-64-avx512-clang before: > make clean all CC=clang BACKEND=2 && ./bench 10000 ... info: cpucycles: version=20240318 implementation=amd64-pmc persecond=4800000000 info: backend=avx512 num_trials=10000 src_lens=64,256,1024,4096,16384 dst_lens=32 function,dst_len,64,256,1024,4096,16384 sha3_224,28,21.6,10.8,10.7,9.6,9.5 sha3_256,32,21.6,10.8,10.7,10.3,10.1 sha3_384,48,21.5,16.0,13.3,13.2,13.1 sha3_512,64,21.5,21.2,19.8,18.8,18.7 shake128,32,21.8,10.8,9.4,8.4,8.2 shake256,32,21.6,10.8,10.7,10.3,10.1 x86-64-avx512-gcc after: > make clean all BACKEND=2 && ./bench 10000 ... info: cpucycles: version=20240318 implementation=amd64-pmc persecond=4800000000 info: backend=avx512 num_trials=10000 src_lens=64,256,1024,4096,16384 dst_lens=32 function,dst_len,64,256,1024,4096,16384 sha3_224,28,15.4,7.8,7.8,7.1,7.0 sha3_256,32,15.4,7.8,7.8,7.6,7.4 sha3_384,48,15.4,11.6,9.8,9.8,9.7 sha3_512,64,15.5,15.5,14.6,13.8,13.8 shake128,32,15.5,7.8,6.9,6.2,6.1 shake256,32,15.5,7.8,7.9,7.6,7.4 x86-64-avx512-clang after: > make clean all CC=clang BACKEND=2 && ./bench 10000 ... info: cpucycles: version=20240318 implementation=amd64-pmc persecond=4800000000 info: backend=avx512 num_trials=10000 src_lens=64,256,1024,4096,16384 dst_lens=32 function,dst_len,64,256,1024,4096,16384 sha3_224,28,21.5,10.8,10.7,9.6,9.5 sha3_256,32,21.5,10.8,10.7,10.3,10.0 sha3_384,48,21.6,16.0,13.3,13.2,13.1 sha3_512,64,21.5,21.2,19.8,18.8,18.7 shake128,32,21.7,10.8,9.3,8.3,8.2 shake256,32,21.7,10.8,10.7,10.3,10.1
2024-05-17sha3.c: permute_n_scalar(): simplify num_rounds logicPaul Duncan1-2/+2
bench results (samish across the board) ------------- x86-64-scalar-gcc before: > make clean all BACKEND=1 && perf record ./bench 10000 ... info: cpucycles: version=20240318 implementation=amd64-pmc persecond=4800000000 info: backend=scalar num_trials=10000 src_lens=64,256,1024,4096,16384 dst_lens=32 function,dst_len,64,256,1024,4096,16384 sha3_224,28,19.5,10.0,10.0,9.0,8.8 sha3_256,32,19.5,10.0,9.9,9.5,9.3 sha3_384,48,20.1,14.7,12.3,12.2,12.0 sha3_512,64,19.5,19.5,18.2,17.2,17.1 shake128,32,19.6,10.0,8.7,7.8,7.6 shake256,32,19.6,10.1,10.0,9.6,9.3 x86-64-scalar-clang before: > make clean all CC=clang BACKEND=1 && perf record ./bench 10000 ... info: cpucycles: version=20240318 implementation=amd64-pmc persecond=4800000000 info: backend=scalar num_trials=10000 src_lens=64,256,1024,4096,16384 dst_lens=32 function,dst_len,64,256,1024,4096,16384 sha3_224,28,19.7,9.9,9.7,8.8,8.6 sha3_256,32,19.7,9.8,9.7,9.4,9.1 sha3_384,48,19.7,14.5,12.0,11.9,11.8 sha3_512,64,19.7,19.1,17.7,16.8,16.8 shake128,32,20.4,10.0,8.7,7.7,7.5 shake256,32,20.3,10.0,9.8,9.4,9.1 x86-64-scalar-gcc after: > make clean all BACKEND=1 && perf record ./bench 10000 ... info: cpucycles: version=20240318 implementation=amd64-pmc persecond=4800000000 info: backend=scalar num_trials=10000 src_lens=64,256,1024,4096,16384 dst_lens=32 function,dst_len,64,256,1024,4096,16384 sha3_224,28,19.5,10.0,9.9,9.0,8.8 sha3_256,32,19.5,10.0,9.9,9.5,9.3 sha3_384,48,19.5,14.7,12.3,12.2,12.0 sha3_512,64,19.7,19.5,18.2,17.2,17.1 shake128,32,19.6,10.0,8.7,7.8,7.6 shake256,32,19.6,10.1,10.0,9.6,9.3 x86-64-scalar-gcc after: > make clean all CC=clang BACKEND=1 && perf record ./bench 10000 ... info: cpucycles: version=20240318 implementation=amd64-pmc persecond=4800000000 info: backend=scalar num_trials=10000 src_lens=64,256,1024,4096,16384 dst_lens=32 function,dst_len,64,256,1024,4096,16384 sha3_224,28,19.7,9.9,9.8,8.8,8.7 sha3_256,32,19.6,9.9,9.8,9.4,9.2 sha3_384,48,19.7,14.6,12.1,12.0,11.9 sha3_512,64,19.7,19.3,17.9,16.9,16.9 shake128,32,19.7,9.9,8.7,7.7,7.5 shake256,32,19.7,9.9,9.8,9.5,9.2 a76-scalar-gcc before: > make clean all BACKEND=1 && ./bench 5000 info: cpucycles: version=20240318 implementation=arm64-vct persecond=2400000000 info: backend=scalar num_trials=5000 src_lens=64,256,1024,4096,16384 dst_lens=32 function,dst_len,64,256,1024,4096,16384 sha3_224,28,20.2,10.1,10.2,9.3,9.2 sha3_256,32,20.2,10.3,10.3,10.0,9.7 sha3_384,48,20.9,15.3,12.8,12.7,12.5 sha3_512,64,20.9,20.3,18.9,18.0,17.9 shake128,32,20.2,10.3,9.0,8.1,7.9 shake256,32,20.2,10.1,10.3,9.9,9.7 a76-scalar-clang before: > make clean all CC=clang BACKEND=1 && ./bench 5000 info: cpucycles: version=20240318 implementation=arm64-vct persecond=2400000000 info: backend=scalar num_trials=5000 src_lens=64,256,1024,4096,16384 dst_lens=32 function,dst_len,64,256,1024,4096,16384 sha3_224,28,18.8,9.6,9.5,8.7,8.5 sha3_256,32,18.8,9.6,9.5,9.2,9.0 sha3_384,48,18.8,14.1,11.8,11.8,11.6 sha3_512,64,18.8,18.6,17.5,16.6,16.6 shake128,32,18.8,9.6,8.4,7.5,7.4 shake256,32,18.8,9.6,9.6,9.2,9.0 a76-scalar-gcc after: > make clean all BACKEND=1 && ./bench 5000 info: cpucycles: version=20240318 implementation=arm64-vct persecond=2400000000 info: backend=scalar num_trials=5000 src_lens=64,256,1024,4096,16384 dst_lens=32 function,dst_len,64,256,1024,4096,16384 sha3_224,28,20.2,10.1,10.3,9.3,9.2 sha3_256,32,20.2,10.3,10.3,10.0,9.7 sha3_384,48,20.9,15.3,12.8,12.7,12.5 sha3_512,64,20.9,20.3,18.9,18.0,17.9 shake128,32,20.2,10.3,9.0,8.1,7.9 shake256,32,20.2,10.3,10.3,9.9,9.7 a76-scalar-clang after: > make clean all CC=clang BACKEND=1 && ./bench 5000 info: cpucycles: version=20240318 implementation=arm64-vct persecond=2400000000 info: backend=scalar num_trials=5000 src_lens=64,256,1024,4096,16384 dst_lens=32 function,dst_len,64,256,1024,4096,16384 sha3_224,28,18.8,9.4,9.3,8.5,8.3 sha3_256,32,18.8,9.4,9.3,9.0,8.8 sha3_384,48,18.8,13.9,11.6,11.6,11.4 sha3_512,64,18.8,18.3,17.1,16.3,16.3 shake128,32,18.8,9.4,8.3,7.4,7.2 shake256,32,18.8,9.4,9.4,9.1,8.9
2024-05-09examples/*/Makefile, tests/*/Makefile: add BACKEND support, mark clean ↵v0.8Paul Duncan9-16/+40
target as phony
2024-05-09sha3.c: document how to add a new backendPaul Duncan1-3/+30
2024-05-09Makefile: remove -fsanitize=leak (not supported in macos)Paul Duncan1-1/+10
2024-05-09sha3.[hc]: add sha3_xof12_t, refactor xor12 and turboshake functions to use itPaul Duncan2-12/+28
2024-05-09sha3.h: remove extraneous newlinePaul Duncan1-1/+0
2024-05-09sha3.c: add internal documentation, s/absorb12/absorb_12/Paul Duncan1-39/+227
2024-05-08sha3.c: update internal documentationPaul Duncan1-16/+50
2024-05-08rand-bytes.h: use getentropy() instead of getrandom() to support macosPaul Duncan1-8/+24
2024-05-08sha3.c, Makefile: s/SHA3_TEST/TEST_SHA3/Paul Duncan2-5/+5
2024-05-08.gitignore: add tests/neon/neonPaul Duncan1-0/+1
2024-05-08s/permute/permute_24/, s/permute12/permute_12/, misc comment cleanupPaul Duncan1-99/+89
2024-05-08sha3.c: rename hybrid-neon to hybrid, fix commentsPaul Duncan1-30/+26
2024-05-08sha3.c: hybrid: do not mix neon and non-neon instructions (still slow)Paul Duncan1-15/+26
2024-05-08s/SHA3_BACKEND/BACKEND/gPaul Duncan5-66/+66
2024-05-08sha3.c: s/call permute_n_.*(/call permute_n(/ in test commentsPaul Duncan1-4/+4
2024-05-08sha3.c: add hybrid-neon backend (slow)Paul Duncan1-1/+202
2024-05-08sha3.c: diet-neon: misc fixes. still too slowPaul Duncan1-46/+36
2024-05-08sha3.c: neon, diet-neon: use replace vorrq(vshlq, vshrq) and vorr(vshl,vshr) ↵Paul Duncan1-3/+4
with vsri(vshlq) and vsri(vshl), respectively
2024-05-08sha3.c: move INTERNAL before includesPaul Duncan1-2/+2
2024-05-08sha3.h: fix typo in sha3_backend() documentationPaul Duncan1-1/+1
2024-05-08README.md: add median linkPaul Duncan1-0/+2
2024-05-08sha3.c: prefer scalar backend to neon for nowv0.7Paul Duncan1-2/+2
2024-05-08README.md: add Backends and Benchmarks sectionsPaul Duncan1-1/+71
2024-05-08tests/bench/README.md: fix cpb links, update default trial count, add pi5 ↵Paul Duncan1-21/+103
results, show results for multiple backends on all test systems
2024-05-08tests/bench/bench.c: reduce defailt number of trails from 100k to 2kPaul Duncan1-1/+1
2024-05-08sha3.c: neon: refactor, add documentationPaul Duncan1-115/+157
- switch row_eor() from macro to static inline function - compress rho rotate values into from 15 128-bit registers to two to reduce register pressure (still spilling, though) - remove PERMUTE macro - switch from unrolled loop with macro in body of permute_n_neon() to regular loop - add documentation for register/lane layout and for compressed rho rotations with these changes the neon backend is still uses ~50% more cycles than the scalar backend, so i will probably leave it disabled for the initial release. scalar (pi5): > ./bench 2000 info: cpucycles: version=20240318 implementation=arm64-vct persecond=2400000000 info: backend=scalar num_trials=2000 src_lens=64,256,1024,4096,16384 dst_lens=32 function,dst_len,64,256,1024,4096,16384 sha3_224,28,20.2,10.3,10.3,9.3,9.2 sha3_256,32,20.2,10.3,10.3,9.9,9.7 sha3_384,48,20.9,15.3,12.8,12.7,12.5 sha3_512,64,20.2,20.2,18.9,25.3,17.9 shake128,32,20.2,10.1,9.0,8.1,7.9 shake256,32,20.2,10.3,10.3,9.9,9.7 neon backend bench results (pi5): > ./bench 2000 info: cpucycles: version=20240318 implementation=arm64-vct persecond=2400000000 info: backend=neon num_trials=2000 src_lens=64,256,1024,4096,16384 dst_lens=32 function,dst_len,64,256,1024,4096,16384 sha3_224,28,32.7,16.3,16.4,14.9,14.6 sha3_256,32,32.0,16.2,16.4,15.9,15.5 sha3_384,48,32.7,24.2,20.4,20.2,20.0 sha3_512,64,32.0,32.2,30.1,28.6,28.5 shake128,32,32.7,16.2,14.2,12.8,12.5 shake256,32,32.7,16.2,16.3,15.7,15.4
2024-05-08tests/neon/neon.c: port changes back from sha3.cPaul Duncan1-302/+405