Add s390x/zSeries implementation of Poly1305
* cipher/Makefile.am: Add 'poly1305-s390x.S' and 'asm-poly1305-s390x.h'. * cipher/asm-poly1305-s390x.h: New * cipher/chacha20-s390x.S (_gcry_chacha20_poly1305_s390x_vx_blocks8) (_gcry_chacha20_poly1305_s390x_vx_blocks4_2_1): New, stitched chacha20-poly1305 implementation. * cipher/chacha20.c (USE_S390X_VX_POLY1305): New. (_gcry_chacha20_poly1305_s390x_vx_blocks8) (_gcry_chacha20_poly1305_s390x_vx_blocks4_2_1): New prototypes. (_gcry_chacha20_poly1305_encrypt, _gcry_chacha20_poly1305_decrypt): Add s390x/VX stitched chacha20-poly1305 code-path. * cipher/poly1305-s390x.S: New. * cipher/poly1305.c (USE_S390X_ASM, HAVE_ASM_POLY1305_BLOCKS): New. [USE_S390X_ASM] (_gcry_poly1305_s390x_blocks1, poly1305_blocks): New. * configure.ac (gcry_cv_gcc_inline_asm_s390x): Check for 'risbgn' and 'algrk' instructions. * tests/basic.c (_check_poly1305_cipher): Add large chacha20-poly1305 test vector.
Patch adds Poly1305 and stitched ChaCha20-Poly1305 implementation
for zSeries. Stitched implementation interleaves ChaCha20 and Poly1305
processing for higher instruction level parallelism and better
utilization of execution units.
Benchmark on z15 (4504 Mhz):
Before:
CHACHA20 | nanosecs/byte mebibytes/sec cycles/byte
POLY1305 enc | 1.16 ns/B 823.2 MiB/s 5.22 c/B POLY1305 dec | 1.16 ns/B 823.2 MiB/s 5.22 c/B POLY1305 auth | 0.736 ns/B 1295 MiB/s 3.32 c/B
After (chacha20-poly1305 ~71% faster, poly1305 ~29% faster):
CHACHA20 | nanosecs/byte mebibytes/sec cycles/byte
POLY1305 enc | 0.677 ns/B 1409 MiB/s 3.05 c/B POLY1305 dec | 0.655 ns/B 1456 MiB/s 2.95 c/B POLY1305 auth | 0.569 ns/B 1675 MiB/s 2.56 c/B
- GnuPG-bug-id: T5202
- Signed-off-by: Jussi Kivilinna <jussi.kivilinna@iki.fi>