Use AVX2 for SHA calculations

tbuktu · May 12, 2016 · 7d4d666 · 7d4d666
1 parent 6480853
commit 7d4d666
Show file tree

Hide file tree

Showing 15 changed files with 513 additions and 64 deletions.
diff --git a/Makefile.bsd b/Makefile.bsd
@@ -10,9 +10,24 @@ ifneq ($(SSE), no)
         SSE=yes
     endif
 endif
+AVX2_FLAG = $(shell /usr/bin/grep -o AVX2 /var/run/dmesg.boot | /usr/bin/head -1)
+ifneq ($(AVX2), no)
+    ifeq ($(AVX2_FLAG), AVX2)
+        AVX2=yes
+    endif
+endif
+ifeq ($(AVX2), yes)
+    SSE=yes
+endif
+ifeq ($(SSE), no)
+    AVX2=no
+endif
 ifeq ($(SSE), yes)
     CFLAGS+=-mssse3
 endif
+ifeq ($(AVX2), yes)
+    CFLAGS+=-mavx2
+endif
 
 # use -march=native if we're compiling for x86
 BENCH_ARCH_OPTION=
@@ -148,10 +163,10 @@ hybrid: static-lib
 $(SRCDIR)/%.o: $(SRCDIR)/%.c
 	$(CC) $(CFLAGS) $(CPPFLAGS) -c -fPIC $< -o $@
 
-$(SRCDIR)/sha1-mb-x86_64.s: $(SRCDIR)/sha1-mb-x86_64.pl; $(PERL) $(SRCDIR)/sha1-mb-x86_64.pl $(PERLASM_SCHEME) > $@
+$(SRCDIR)/sha1-mb-x86_64.s: $(SRCDIR)/sha1-mb-x86_64.pl; CC=$(CC) ASM="$(AS)" $(PERL) $(SRCDIR)/sha1-mb-x86_64.pl $(PERLASM_SCHEME) > $@
 $(SRCDIR)/sha1-mb-x86_64.o: $(SRCDIR)/sha1-mb-x86_64.s
 	$(AS) $(SRCDIR)/sha1-mb-x86_64.s -o $@
-$(SRCDIR)/sha256-mb-x86_64.s: $(SRCDIR)/sha256-mb-x86_64.pl; $(PERL) $(SRCDIR)/sha256-mb-x86_64.pl $(PERLASM_SCHEME) > $@
+$(SRCDIR)/sha256-mb-x86_64.s: $(SRCDIR)/sha256-mb-x86_64.pl; CC=$(CC) ASM="$(AS)" $(PERL) $(SRCDIR)/sha256-mb-x86_64.pl $(PERLASM_SCHEME) > $@
 $(SRCDIR)/sha256-mb-x86_64.o: $(SRCDIR)/sha256-mb-x86_64.s
 	$(AS) $(SRCDIR)/sha256-mb-x86_64.s -o $@
 

diff --git a/Makefile.linux b/Makefile.linux
@@ -10,9 +10,24 @@ ifneq ($(SSE), no)
         SSE=yes
     endif
 endif
+AVX2_FLAG = $(shell /bin/grep -m 1 -o avx2 /proc/cpuinfo)
+ifneq ($(AVX2), no)
+    ifeq ($(AVX2_FLAG), avx2)
+        AVX2=yes
+    endif
+endif
+ifeq ($(AVX2), yes)
+    SSE=yes
+endif
+ifeq ($(SSE), no)
+    AVX2=no
+endif
 ifeq ($(SSE), yes)
     CFLAGS+=-mssse3
 endif
+ifeq ($(AVX2), yes)
+    CFLAGS+=-mavx2
+endif
 
 # use -march=native if we're compiling for x86
 BENCH_ARCH_OPTION=
@@ -146,10 +161,10 @@ hybrid: static-lib
 $(SRCDIR)/%.o: $(SRCDIR)/%.c
 	$(CC) $(CFLAGS) $(CPPFLAGS) -c -fPIC $< -o $@
 
-$(SRCDIR)/sha1-mb-x86_64.s: $(SRCDIR)/sha1-mb-x86_64.pl; $(PERL) $(SRCDIR)/sha1-mb-x86_64.pl $(PERLASM_SCHEME) > $@
+$(SRCDIR)/sha1-mb-x86_64.s: $(SRCDIR)/sha1-mb-x86_64.pl; CC=$(CC) ASM="$(AS)" $(PERL) $(SRCDIR)/sha1-mb-x86_64.pl $(PERLASM_SCHEME) > $@
 $(SRCDIR)/sha1-mb-x86_64.o: $(SRCDIR)/sha1-mb-x86_64.s
 	$(AS) $(SRCDIR)/sha1-mb-x86_64.s -o $@
-$(SRCDIR)/sha256-mb-x86_64.s: $(SRCDIR)/sha256-mb-x86_64.pl; $(PERL) $(SRCDIR)/sha256-mb-x86_64.pl $(PERLASM_SCHEME) > $@
+$(SRCDIR)/sha256-mb-x86_64.s: $(SRCDIR)/sha256-mb-x86_64.pl; CC=$(CC) ASM="$(AS)" $(PERL) $(SRCDIR)/sha256-mb-x86_64.pl $(PERLASM_SCHEME) > $@
 $(SRCDIR)/sha256-mb-x86_64.o: $(SRCDIR)/sha256-mb-x86_64.s
 	$(AS) $(SRCDIR)/sha256-mb-x86_64.s -o $@
 

diff --git a/Makefile.os2 b/Makefile.os2
@@ -4,6 +4,10 @@ OPTFLAGS=-O2
 bench: OPTFLAGS=-O3 -march=native
 CFLAGS?=-g $(OPTFLAGS)
 CFLAGS+=-Wall -Wextra -Wno-unused-parameter
+ifeq ($(AVX2), yes)
+    CFLAGS+=-mavx2
+    SSE=yes
+endif
 ifeq ($(SSE), yes)
     CFLAGS+=-mssse3
 endif
@@ -106,10 +110,10 @@ hybrid: lib
 $(SRCDIR)/%.o: $(SRCDIR)/%.c
 	$(CC) $(CFLAGS) $(CPPFLAGS) -c $< -o $@
 
-$(SRCDIR)/sha1-mb-x86_64.s: $(SRCDIR)/sha1-mb-x86_64.pl; $(PERL) $(SRCDIR)/sha1-mb-x86_64.pl $(PERLASM_SCHEME) > $@
+$(SRCDIR)/sha1-mb-x86_64.s: $(SRCDIR)/sha1-mb-x86_64.pl; CC=$(CC) ASM="$(AS)" $(PERL) $(SRCDIR)/sha1-mb-x86_64.pl $(PERLASM_SCHEME) > $@
 $(SRCDIR)/sha1-mb-x86_64.o: $(SRCDIR)/sha1-mb-x86_64.s
 	$(AS) $(SRCDIR)/sha1-mb-x86_64.s -o $@
-$(SRCDIR)/sha256-mb-x86_64.s: $(SRCDIR)/sha256-mb-x86_64.pl; $(PERL) $(SRCDIR)/sha256-mb-x86_64.pl $(PERLASM_SCHEME) > $@
+$(SRCDIR)/sha256-mb-x86_64.s: $(SRCDIR)/sha256-mb-x86_64.pl; CC=$(CC) ASM="$(AS)" $(PERL) $(SRCDIR)/sha256-mb-x86_64.pl $(PERLASM_SCHEME) > $@
 $(SRCDIR)/sha256-mb-x86_64.o: $(SRCDIR)/sha256-mb-x86_64.s
 	$(AS) $(SRCDIR)/sha256-mb-x86_64.s -o $@
 

diff --git a/Makefile.osx b/Makefile.osx
@@ -9,12 +9,27 @@ ifneq ($(SSE), no)
         SSE=yes
     endif
 endif
+AVX2_FLAG = $(shell /usr/sbin/sysctl machdep.cpu.features | grep -m 1 -ow AVX2)
+ifneq ($(AVX2), no)
+    ifeq ($(AVX2_FLAG), AVX2)
+        AVX2=yes
+    endif
+endif
+ifeq ($(AVX2), yes)
+    SSE=yes
+endif
+ifeq ($(SSE), no)
+    AVX2=no
+endif
 ifeq ($(SSE), yes)
     CFLAGS+=-mssse3
 endif
 ifeq ($(SSE), no)
     CFLAGS+=-march=x86-64
 endif
+ifeq ($(AVX2), yes)
+    CFLAGS+=-mavx2
+endif
 SRCDIR=src
 TESTDIR=tests
 LIB_OBJS=bitstring.o encparams.o hash.o idxgen.o key.o mgf.o ntru.o poly.o rand.o arith.o sha1.o sha2.o nist_ctr_drbg.o rijndael.o rijndael-alg-fst.o
@@ -104,10 +119,10 @@ hybrid: lib
 $(SRCDIR)/%.o: $(SRCDIR)/%.c
 	$(CC) $(CFLAGS) $(CPPFLAGS) -c -fPIC $< -o $@
 
-$(SRCDIR)/sha1-mb-x86_64.s: $(SRCDIR)/sha1-mb-x86_64.pl; $(PERL) $(SRCDIR)/sha1-mb-x86_64.pl $(PERLASM_SCHEME) > $@
+$(SRCDIR)/sha1-mb-x86_64.s: $(SRCDIR)/sha1-mb-x86_64.pl; CC=$(CC) ASM="$(AS)" $(PERL) $(SRCDIR)/sha1-mb-x86_64.pl $(PERLASM_SCHEME) > $@
 $(SRCDIR)/sha1-mb-x86_64.o: $(SRCDIR)/sha1-mb-x86_64.s
 	$(AS) $(SRCDIR)/sha1-mb-x86_64.s -o $@
-$(SRCDIR)/sha256-mb-x86_64.s: $(SRCDIR)/sha256-mb-x86_64.pl; $(PERL) $(SRCDIR)/sha256-mb-x86_64.pl $(PERLASM_SCHEME) > $@
+$(SRCDIR)/sha256-mb-x86_64.s: $(SRCDIR)/sha256-mb-x86_64.pl; CC=$(CC) ASM="$(AS)" $(PERL) $(SRCDIR)/sha256-mb-x86_64.pl $(PERLASM_SCHEME) > $@
 $(SRCDIR)/sha256-mb-x86_64.o: $(SRCDIR)/sha256-mb-x86_64.s
 	$(AS) $(SRCDIR)/sha256-mb-x86_64.s -o $@
 

diff --git a/Makefile.win b/Makefile.win
@@ -4,6 +4,10 @@ OPTFLAGS=-O2
 bench: OPTFLAGS=-O3 -march=native
 CFLAGS?=-g $(OPTFLAGS)
 CFLAGS+=-Wall -Wextra -Wno-unused-parameter
+ifeq ($(AVX2), yes)
+    CFLAGS+=-mavx2
+    SSE=yes
+endif
 ifeq ($(SSE), yes)
     CFLAGS+=-mssse3
 endif
@@ -108,10 +112,10 @@ hybrid: lib
 $(SRCDIR)/%.o: $(SRCDIR)/%.c
 	$(CC) $(CFLAGS) $(CPPFLAGS) -c $< -o $@
 
-$(SRCDIR)/sha1-mb-x86_64.s: $(SRCDIR)/sha1-mb-x86_64.pl; $(PERL) $(SRCDIR)/sha1-mb-x86_64.pl $(PERLASM_SCHEME) > $@
+$(SRCDIR)/sha1-mb-x86_64.s: $(SRCDIR)/sha1-mb-x86_64.pl; CC=$(CC) ASM="$(AS)" $(PERL) $(SRCDIR)/sha1-mb-x86_64.pl $(PERLASM_SCHEME) > $@
 $(SRCDIR)/sha1-mb-x86_64.o: $(SRCDIR)/sha1-mb-x86_64.s
 	$(AS) $(SRCDIR)/sha1-mb-x86_64.s -o $@
-$(SRCDIR)/sha256-mb-x86_64.s: $(SRCDIR)/sha256-mb-x86_64.pl; $(PERL) $(SRCDIR)/sha256-mb-x86_64.pl $(PERLASM_SCHEME) > $@
+$(SRCDIR)/sha256-mb-x86_64.s: $(SRCDIR)/sha256-mb-x86_64.pl; CC=$(CC) ASM="$(AS)" $(PERL) $(SRCDIR)/sha256-mb-x86_64.pl $(PERLASM_SCHEME) > $@
 $(SRCDIR)/sha256-mb-x86_64.o: $(SRCDIR)/sha256-mb-x86_64.s
 	$(AS) $(SRCDIR)/sha256-mb-x86_64.s -o $@
 

diff --git a/src/encparams.c b/src/encparams.c
@@ -19,6 +19,7 @@ const NtruEncParams EES401EP1 = {\
     {0, 2, 4},     /* oid */\
     ntru_sha1,     /* hash */\
     ntru_sha1_4way, /* hash_4way */\
+    ntru_sha1_8way, /* hash_8way */\
     20,            /* hlen */\
     114            /* pklen */\
 };
@@ -41,6 +42,7 @@ const NtruEncParams EES449EP1 = {\
     {0, 3, 3},     /* oid */\
     ntru_sha1,     /* hash */\
     ntru_sha1_4way, /* hash_4way */\
+    ntru_sha1_8way, /* hash_8way */\
     20,            /* hlen */\
     128            /* pklen */\
 };
@@ -63,6 +65,7 @@ const NtruEncParams EES677EP1 = {\
     {0, 5, 3},     /* oid */\
     ntru_sha256,   /* hash */\
     ntru_sha256_4way, /* hash_4way */\
+    ntru_sha256_8way, /* hash_8way */\
     32,            /* hlen */\
     192            /* pklen */\
 };
@@ -85,6 +88,7 @@ const NtruEncParams EES1087EP2 = {\
     {0, 6, 3},     /* oid */\
     ntru_sha256,   /* hash */\
     ntru_sha256_4way, /* hash_4way */\
+    ntru_sha256_8way, /* hash_8way */\
     32,            /* hlen */\
     256            /* pklen */\
 };
@@ -107,6 +111,7 @@ const NtruEncParams EES541EP1 = {\
     {0, 2, 5},     /* oid */\
     ntru_sha1,     /* hash */\
     ntru_sha1_4way, /* hash_4way */\
+    ntru_sha1_8way, /* hash_8way */\
     20,            /* hlen */\
     112            /* pklen */\
 };
@@ -129,6 +134,7 @@ const NtruEncParams EES613EP1 = {\
     {0, 3, 4},     /* oid */\
     ntru_sha1,     /* hash */\
     ntru_sha1_4way, /* hash_4way */\
+    ntru_sha1_8way, /* hash_8way */\
     20,            /* hlen */\
     128            /* pklen */\
 };
@@ -151,6 +157,7 @@ const NtruEncParams EES887EP1 = {\
     {0, 5, 4},     /* oid */\
     ntru_sha256,   /* hash */\
     ntru_sha256_4way, /* hash_4way */\
+    ntru_sha256_8way, /* hash_8way */\
     32,            /* hlen */\
     192            /* pklen */\
 };
@@ -173,6 +180,7 @@ const NtruEncParams EES1171EP1 = {\
     {0, 6, 4},     /* oid */\
     ntru_sha256,   /* hash */\
     ntru_sha256_4way, /* hash_4way */\
+    ntru_sha256_8way, /* hash_8way */\
     32,            /* hlen */\
     256            /* pklen */\
 };
@@ -195,6 +203,7 @@ const NtruEncParams EES659EP1 = {\
     {0, 2, 6},     /* oid */\
     ntru_sha1,     /* hash */\
     ntru_sha1_4way, /* hash_4way */\
+    ntru_sha1_8way, /* hash_8way */\
     20,            /* hlen */\
     112            /* pklen */\
 };
@@ -217,6 +226,7 @@ const NtruEncParams EES761EP1 = {\
     {0, 3, 5},     /* oid */\
     ntru_sha1,     /* hash */\
     ntru_sha1_4way, /* hash_4way */\
+    ntru_sha1_8way, /* hash_8way */\
     20,            /* hlen */\
     128            /* pklen */\
 };
@@ -239,6 +249,7 @@ const NtruEncParams EES1087EP1 = {\
     {0, 5, 5},     /* oid */\
     ntru_sha256,   /* hash */\
     ntru_sha256_4way, /* hash_4way */\
+    ntru_sha256_8way, /* hash_8way */\
     32,            /* hlen */\
     192            /* pklen */\
 };
@@ -261,6 +272,7 @@ const NtruEncParams EES1499EP1 = {\
     {0, 6, 5},     /* oid */\
     ntru_sha256,   /* hash */\
     ntru_sha256_4way, /* hash_4way */\
+    ntru_sha256_8way, /* hash_8way */\
     32,            /* hlen */\
     256            /* pklen */\
 };
@@ -285,6 +297,7 @@ const NtruEncParams EES401EP2 = {\
     {0, 2, 16},    /* oid */\
     ntru_sha1,     /* hash */\
     ntru_sha1_4way, /* hash_4way */\
+    ntru_sha1_8way, /* hash_8way */\
     20,            /* hlen */\
     112            /* pklen */\
 };
@@ -307,6 +320,7 @@ const NtruEncParams EES439EP1 = {\
     {0, 3, 16},    /* oid */\
     ntru_sha1,     /* hash */\
     ntru_sha1_4way, /* hash_4way */\
+    ntru_sha1_8way, /* hash_8way */\
     20,            /* hlen */\
     128            /* pklen */\
 };
@@ -329,6 +343,7 @@ const NtruEncParams EES443EP1 = {\
     {0, 3, 17},    /* oid */\
     ntru_sha256,   /* hash */\
     ntru_sha256_4way, /* hash_4way */\
+    ntru_sha256_8way, /* hash_8way */\
     32,            /* hlen */\
     128            /* pklen */\
 };
@@ -351,6 +366,7 @@ const NtruEncParams EES593EP1 = {\
     {0, 5, 16},    /* oid */\
     ntru_sha256,   /* hash */\
     ntru_sha256_4way, /* hash_4way */\
+    ntru_sha256_8way, /* hash_8way */\
     32,            /* hlen */\
     192            /* pklen */\
 };
@@ -373,6 +389,7 @@ const NtruEncParams EES587EP1 = {\
     {0, 5, 17},    /* oid */\
     ntru_sha256,   /* hash */\
     ntru_sha256_4way, /* hash_4way */\
+    ntru_sha256_8way, /* hash_8way */\
     32,            /* hlen */\
     192            /* pklen */\
 };
@@ -395,6 +412,7 @@ const NtruEncParams EES743EP1 = {\
     {0, 6, 16},    /* oid */\
     ntru_sha256,   /* hash */\
     ntru_sha256_4way, /* hash_4way */\
+    ntru_sha256_8way, /* hash_8way */\
     32,            /* hlen */\
     256            /* pklen */\
 };

diff --git a/src/encparams.h b/src/encparams.h
@@ -80,6 +80,9 @@ typedef struct NtruEncParams {
     /* hash function for 4 inputs, e.g. ntru_sha256_4way */
     void (*hash_4way)(uint8_t *[4], uint16_t, uint8_t *[4]);
 
+    /* hash function for 8 inputs, e.g. ntru_sha256_8way */
+    void (*hash_8way)(uint8_t *[8], uint16_t, uint8_t *[8]);
+
     /* output length of the hash function */
     uint16_t hlen;