diff options
author | Andres Erbsen <andreser@mit.edu> | 2017-04-13 17:02:49 -0400 |
---|---|---|
committer | Andres Erbsen <andreser@mit.edu> | 2017-04-13 17:19:43 -0400 |
commit | 0d806d4a16820301926a6b511b3ccb65ee17b114 (patch) | |
tree | a34918c5591ad164e28d93cf242b91f9f8252dad /Makefile | |
parent | 03faa100b68a80281effa890d16c995f0d7d72a3 (diff) |
X25519: wrap synthesized code in donna-c64, run SUPERCOP benchmarks
21% slower than donna 36% slower than assembly with saturated limbs.
cycles impl compiler
593072 crypto_scalarmult/curve25519/amd64-64 gcc_-m64_-march=core-avx2_-O3_-fomit-frame-pointer
636660 crypto_scalarmult/curve25519/amd64-51 gcc_-funroll-loops_-fno-schedule-insns_-O2_-fomit-frame-pointer
660220 crypto_scalarmult/curve25519/donna_c64 gcc_-m64_-march=native_-mtune=native_-O3_-fomit-frame-pointer
804684 crypto_scalarmult/curve25519/fiat_c64 clang_-march=native_-O3_-fomit-frame-pointer_-fwrapv_-Qunused-arguments
815716 crypto_scalarmult/curve25519/fiat_c64 gcc_-m64_-march=core-avx2_-O3_-fomit-frame-pointer
Setup: Broadwell i7-5600U 2.60GHz in an X1C3. no HT, no TB, wall power.
Journalctl did not record any thermal throttling, idk if trustworthy.
Cc: @achlipala @JasonGross @jadephilipoom
Diffstat (limited to 'Makefile')
-rw-r--r-- | Makefile | 3 |
1 files changed, 3 insertions, 0 deletions
@@ -106,6 +106,9 @@ $(DISPLAY_JAVA_VO:.vo=.log) : %JavaDisplay.log : %.vo %JavaDisplay.v src/Compile $(SHOW)"COQC $*JavaDisplay > $@" $(HIDE)$(COQC) $(COQDEBUG) $(COQFLAGS) $*JavaDisplay.v > $@.tmp && mv -f $@.tmp $@ +src/Specific/x25519_c64.c: src/Specific/x25519_c64.c.sh src/Specific/IntegrationTestMulDisplay.log + bash src/Specific/x25519_c64.c.sh > src/Specific/x25519_c64.c + clean:: rm -f Makefile.coq |