Merge pull request bitcoin#128

b2c9681 Make {mul,sqr}_inner use the same argument order as {mul,sqr} (Pieter Wuille) 6793505 Convert YASM code into inline assembly (Pieter Wuille) f048615 Rewrite field assembly to match the C version (Pieter Wuille)
TheBlueMatt · Dec 7, 2014 · 376b28b · 376b28b
2 parents 1728806 + b2c9681
commit 376b28b
Show file tree

Hide file tree

Showing 9 changed files with 514 additions and 590 deletions.
diff --git a/Makefile.am b/Makefile.am
@@ -1,12 +1,6 @@
 ACLOCAL_AMFLAGS = -I build-aux/m4
 
 lib_LTLIBRARIES = libsecp256k1.la
-if USE_ASM
-COMMON_LIB = libsecp256k1_common.la
-else
-COMMON_LIB =
-endif
-noinst_LTLIBRARIES = $(COMMON_LIB)
 include_HEADERS = include/secp256k1.h
 noinst_HEADERS =
 noinst_HEADERS += src/scalar.h
@@ -47,13 +41,9 @@ noinst_HEADERS += src/field_impl.h
 pkgconfigdir = $(libdir)/pkgconfig
 pkgconfig_DATA = libsecp256k1.pc
 
-if USE_ASM
-libsecp256k1_common_la_SOURCES = src/field_5x52_asm.asm
-endif
-
 libsecp256k1_la_SOURCES = src/secp256k1.c
 libsecp256k1_la_CPPFLAGS = -I$(top_srcdir)/include $(SECP_INCLUDES)
-libsecp256k1_la_LIBADD = $(COMMON_LIB) $(SECP_LIBS)
+libsecp256k1_la_LIBADD = $(SECP_LIBS)
 
 
 noinst_PROGRAMS =
@@ -66,7 +56,7 @@ bench_sign_SOURCES = src/bench_sign.c
 bench_sign_LDADD = libsecp256k1.la $(SECP_LIBS)
 bench_sign_LDFLAGS = -static
 bench_inv_SOURCES = src/bench_inv.c
-bench_inv_LDADD = $(COMMON_LIB) $(SECP_LIBS)
+bench_inv_LDADD = $(SECP_LIBS)
 bench_inv_LDFLAGS = -static
 bench_inv_CPPFLAGS = $(SECP_INCLUDES)
 endif
@@ -75,15 +65,9 @@ if USE_TESTS
 noinst_PROGRAMS += tests
 tests_SOURCES = src/tests.c
 tests_CPPFLAGS = -DVERIFY $(SECP_INCLUDES) $(SECP_TEST_INCLUDES)
-tests_LDADD = $(COMMON_LIB) $(SECP_LIBS) $(SECP_TEST_LIBS)
+tests_LDADD = $(SECP_LIBS) $(SECP_TEST_LIBS)
 tests_LDFLAGS = -static
 TESTS = tests
 endif
 
-EXTRA_DIST = autogen.sh nasm_lt.sh
-
-#x86_64 only
-if USE_ASM
-.asm.lo:
-	$(LIBTOOL) --mode=compile --tag YASM $(srcdir)/nasm_lt.sh $(YASM) -f $(YASM_BINFMT) $(YAFLAGS) -I$(srcdir) -I. $< -o $@
-endif
+EXTRA_DIST = autogen.sh
diff --git a/build-aux/m4/bitcoin_secp.m4 b/build-aux/m4/bitcoin_secp.m4
@@ -11,38 +11,16 @@ fi
 
 dnl 
 AC_DEFUN([SECP_64BIT_ASM_CHECK],[
-if test x"$host_cpu" == x"x86_64"; then
-  AC_CHECK_PROG(YASM, yasm, yasm)
-else
-  if test x"$set_field" = x"64bit_asm"; then
-    AC_MSG_ERROR([$set_field field support explicitly requested but is not compatible with this host])
-  fi
-fi
-if test x$YASM = x; then
-  if test x"$set_field" = x"64bit_asm"; then
-    AC_MSG_ERROR([$set_field field support explicitly requested but yasm was not found])
-  fi
-  has_64bit_asm=no
-else
-  case x"$host_os" in
-  xdarwin*)
-    YASM_BINFMT=macho64
-    ;;
-  x*-gnux32)
-    YASM_BINFMT=elfx32
-    ;;
-  *)
-    YASM_BINFMT=elf64
-    ;;
-  esac
-  if $YASM -f help | grep -q $YASM_BINFMT; then
-    has_64bit_asm=yes
-  else
-    if test x"$set_field" = x"64bit_asm"; then
-      AC_MSG_ERROR([$set_field field support explicitly requested but yasm doesn't support $YASM_BINFMT format])
-    fi
-    AC_MSG_WARN([yasm too old for $YASM_BINFMT format])
-    has_64bit_asm=no
+AC_MSG_CHECKING(for x86_64 assembly availability)
+AC_COMPILE_IFELSE([AC_LANG_PROGRAM([[
+  #include <stdint.h>]],[[
+  uint64_t a = 11, tmp;
+  __asm__ __volatile__("movq $0x100000000,%1; mulq %%rsi" : "+a"(a) : "S"(tmp) : "cc", "%rdx");
+  ]])],[has_64bit_asm=yes],[has_64bit_asm=no])
+AC_MSG_RESULT([$has_64bit_asm])
+if test x"$set_field" == x"64bit_asm"; then
+  if test x"$has_64bit_asm" == x"no"; then
+    AC_MSG_ERROR([$set_field field support explicitly requested but no x86_64 assembly available])
   fi
 fi
 ])

diff --git a/configure.ac b/configure.ac
@@ -283,7 +283,6 @@ AC_SUBST(SECP_INCLUDES)
 AC_SUBST(SECP_LIBS)
 AC_SUBST(SECP_TEST_LIBS)
 AC_SUBST(SECP_TEST_INCLUDES)
-AC_SUBST(YASM_BINFMT)
 AM_CONDITIONAL([USE_ASM], [test x"$set_field" == x"64bit_asm"])
 AM_CONDITIONAL([USE_TESTS], [test x"$use_tests" != x"no"])
 AM_CONDITIONAL([USE_BENCHMARK], [test x"$use_benchmark" != x"no"])

diff --git a/nasm_lt.sh b/nasm_lt.sh
diff --git a/src/field_10x26_impl.h b/src/field_10x26_impl.h
@@ -271,7 +271,7 @@ SECP256K1_INLINE static void secp256k1_fe_add(secp256k1_fe_t *r, const secp256k1
 #define VERIFY_BITS(x, n) do { } while(0)
 #endif
 
-SECP256K1_INLINE static void secp256k1_fe_mul_inner(const uint32_t *a, const uint32_t * SECP256K1_RESTRICT b, uint32_t *r) {
+SECP256K1_INLINE static void secp256k1_fe_mul_inner(uint32_t *r, const uint32_t *a, const uint32_t * SECP256K1_RESTRICT b) {
     VERIFY_BITS(a[0], 30);
     VERIFY_BITS(a[1], 30);
     VERIFY_BITS(a[2], 30);
@@ -598,7 +598,7 @@ SECP256K1_INLINE static void secp256k1_fe_mul_inner(const uint32_t *a, const uin
     /* [r9 r8 r7 r6 r5 r4 r3 r2 r1 r0] = [p18 p17 p16 p15 p14 p13 p12 p11 p10 p9 p8 p7 p6 p5 p4 p3 p2 p1 p0] */
 }
 
-SECP256K1_INLINE static void secp256k1_fe_sqr_inner(const uint32_t *a, uint32_t *r) {
+SECP256K1_INLINE static void secp256k1_fe_sqr_inner(uint32_t *r, const uint32_t *a) {
     VERIFY_BITS(a[0], 30);
     VERIFY_BITS(a[1], 30);
     VERIFY_BITS(a[2], 30);
@@ -879,7 +879,7 @@ static void secp256k1_fe_mul(secp256k1_fe_t *r, const secp256k1_fe_t *a, const s
     secp256k1_fe_verify(b);
     VERIFY_CHECK(r != b);
 #endif
-    secp256k1_fe_mul_inner(a->n, b->n, r->n);
+    secp256k1_fe_mul_inner(r->n, a->n, b->n);
 #ifdef VERIFY
     r->magnitude = 1;
     r->normalized = 0;
@@ -892,7 +892,7 @@ static void secp256k1_fe_sqr(secp256k1_fe_t *r, const secp256k1_fe_t *a) {
     VERIFY_CHECK(a->magnitude <= 8);
     secp256k1_fe_verify(a);
 #endif
-    secp256k1_fe_sqr_inner(a->n, r->n);
+    secp256k1_fe_sqr_inner(r->n, a->n);
 #ifdef VERIFY
     r->magnitude = 1;
     r->normalized = 0;