Add ARM SHA1 asm implementation from Cryptogams (GH #837, PR #838)

Add ARM SHA1 asm implementation from Cryptogams.

Cryptogams is Andy Polyakov's project used to create high speed crypto algorithms and share them with other developers. Cryptogams  has a dual license. First is the OpenSSL license because Andy contributes to OpenSSL. Second is a BSD license for those who want a more permissive license.

Andy's implementation runs about 30% faster than C/C++ code. Testing on a 1 GHz Cortex-A7 shows Cryptograms at 16 cpb, and C++ at 23 cpb.

The integration instructions are documented at [Cryptogams SHA](https://wiki.openssl.org/index.php/Cryptogams_SHA) on the OpenSSL wiki.

Filelist.txt

@@ -314,6 +314,8 @@ serpentp.h
 sha.cpp
 sha_simd.cpp
 sha.h
+sha1_armv4.h
+sha1_armv4.S
 sha3.cpp
 sha3.h
 shacal2.cpp

GNUmakefile

@@ -1057,9 +1057,9 @@ endif
 # Cryptogams AES for ARMv4 and above. We couple to ARMv7.
 # Avoid iOS. It cannot consume the assembly.
 ifeq ($(IS_ARM32),1)
-  CRYPTOGAMS_AES_FLAG = -march=armv7-a
+  CRYPTOGAMS_ARCH_FLAG = -march=armv7-a
-  CRYPTOGAMS_AES_FLAG += -Wa,--noexecstack
+  CRYPTOGAMS_ARCH_FLAG += -Wa,--noexecstack
-  SRCS += aes_armv4.S
+  SRCS += aes_armv4.S sha1_armv4.S
 endif
 
 # List cryptlib.cpp first, then cpu.cpp, then integer.cpp to tame C++ static initialization problems.
@@ -1423,7 +1423,7 @@ endif # Dependencies
 
 # Cryptogams ARM asm implementation.
 aes_armv4.o : aes_armv4.S
-	$(CC) $(strip $(CXXFLAGS) $(CRYPTOGAMS_AES_FLAG) -c) $<
+	$(CC) $(strip $(CXXFLAGS) $(CRYPTOGAMS_ARCH_FLAG) -c) $<
 
 # SSSE3 or NEON available
 aria_simd.o : aria_simd.cpp
@@ -1505,6 +1505,10 @@ rijndael_simd.o : rijndael_simd.cpp
 sha_simd.o : sha_simd.cpp
 	$(CXX) $(strip $(CXXFLAGS) $(SHA_FLAG) -c) $<
 
+# Cryptogams ARM asm implementation.
+sha1_armv4.o : sha1_armv4.S
+	$(CC) $(strip $(CXXFLAGS) $(CRYPTOGAMS_ARCH_FLAG) -c) $<
+
 sha3_simd.o : sha3_simd.cpp
 	$(CXX) $(strip $(CXXFLAGS) $(SHA3_FLAG) -c) $<
 

GNUmakefile-cross

@@ -592,9 +592,9 @@ INCL := $(filter-out resource.h,$(sort $(wildcard *.h)))
 # Avoid iOS. It cannot consume the assembly.
 ifeq ($(IS_ARM32),1)
   ifneq ($(IS_IOS),1)
-    CRYPTOGAMS_AES_FLAG = -march=armv7-a
+    CRYPTOGAMS_ARCH_FLAG = -march=armv7-a
-    CRYPTOGAMS_AES_FLAG += -Wa,--noexecstack
+    CRYPTOGAMS_ARCH_FLAG += -Wa,--noexecstack
-    SRCS += aes_armv4.S
+    SRCS += aes_armv4.S sha1_armv4.S
   endif
 endif
 
@@ -800,7 +800,7 @@ endif # Dependencies
 
 # Cryptogams ARM asm implementation.
 aes_armv4.o : aes_armv4.S
-	$(CXX) $(strip $(CXXFLAGS) -fpermissive $(CRYPTOGAMS_AES_FLAG) -c) $<
+	$(CXX) $(strip $(CXXFLAGS) -fpermissive $(CRYPTOGAMS_ARCH_FLAG) -c) $<
 
 cpu-features.o: cpu-features.h cpu-features.c
 	$(CXX) $(strip $(CXXFLAGS) -fpermissive -c) cpu-features.c

License.txt

@@ -29,8 +29,8 @@ Daniel J. Bernstein, Jack Lloyd - chacha.cpp, chacha_simd.cpp, chacha_avx.cpp
 Andrew Moon - ed25519, x25519, donna_32.cpp, donna_64.cpp, donna_sse.cpp
 
 The Crypto++ Library uses portions of Andy Polyakov's CRYPTOGAMS for Poly1305
-scalar multiplication and aes_armv4.S. CRYPTOGAMS is dual licensed with a
+scalar multiplication, aes_armv4.S and sha1_armv4.S. CRYPTOGAMS is dual licensed
-permissive BSD-style license. The CRYPTOGAMS license is reproduced below.
+with a permissive BSD-style license. The CRYPTOGAMS license is reproduced below.
 
 The Crypto++ Library uses portions of Jack Lloyd's Botan for ChaCha SSE2 and
 AVX. Botan placed the code in public domain for Crypto++ to use.

config_asm.h

@@ -354,15 +354,18 @@
 # undef CRYPTOPP_ARM_SHA2_AVAILABLE
 #endif
 
-// Cryptogams offers an ARM asm AES implementation. Crypto++ does
+// Cryptogams offers an ARM asm implementations for AES and SHA. Crypto++ does
-// not provide an asm implementation. The Cryptogams implementation
+// not provide an asm implementation. The Cryptogams AES implementation is
-// is about 2x faster than C/C++. Define this to use the Cryptogams
+// about 50% faster than C/C++, and SHA implementation is about 30% faster
-// AES implementation on GNU Linux systems. When defined, Crypto++
+// than C/C++. Define this to use the Cryptogams AES and SHA implementations
-// will use aes_armv4.S. LLVM miscompiles aes_armv4.S so disable
+// on GNU Linux systems. When defined, Crypto++ will use aes_armv4.S and
-// under Clang. See https://bugs.llvm.org/show_bug.cgi?id=38133.
+// sha1_armv4.S. LLVM miscompiles aes_armv4.S so disable under Clang. Also see
+// https://bugs.llvm.org/show_bug.cgi?id=38133.
 #if !defined(CRYPTOPP_DISABLE_ASM) && defined(__arm__)
 # if defined(__GNUC__) && !defined(__clang__)
 #  define CRYPTOGAMS_ARM_AES      1
+#  define CRYPTOGAMS_ARM_SHA1     1
+#  define CRYPTOGAMS_ARM_SHA256   1
 # endif
 #endif
 

sha.cpp

@@ -56,6 +56,11 @@ extern void SHA1_HashMultipleBlocks_SHANI(word32 *state, const word32 *data, siz
 extern void SHA256_HashMultipleBlocks_SHANI(word32 *state, const word32 *data, size_t length, ByteOrder order);
 #endif
 
+#if (CRYPTOGAMS_ARM_SHA1)
+extern "C" unsigned int CRYPTOGAMS_armcaps;
+extern "C" int sha1_block_data_order(word32* state, const word32 *data, size_t blocks);
+#endif
+
 #if CRYPTOPP_ARM_SHA1_AVAILABLE
 extern void SHA1_HashMultipleBlocks_ARMV8(word32 *state, const word32 *data, size_t length, ByteOrder order);
 #endif
@@ -232,6 +237,12 @@ std::string SHA1::AlgorithmProvider() const
     if (HasSSE2())
         return "SSE2";
 #endif
+#if CRYPTOGAMS_ARM_SHA1
+    if (HasNEON())
+        return "NEON";
+    if (HasARMv7())
+        return "ARMv7";
+#endif
 #if CRYPTOPP_ARM_SHA1_AVAILABLE
     if (HasSHA1())
         return "ARMv8";
@@ -260,6 +271,14 @@ void SHA1::Transform(word32 *state, const word32 *data)
         return;
     }
 #endif
+#if CRYPTOGAMS_ARM_SHA1 && 0
+    // TODO: convert LE to BE and use Cryptogams code
+    if (HasARMv7())
+    {
+        sha1_block_data_order(state, data, 1);
+        return;
+    }
+#endif
 #if CRYPTOPP_ARM_SHA1_AVAILABLE
     if (HasSHA1())
     {
@@ -283,6 +302,19 @@ size_t SHA1::HashMultipleBlocks(const word32 *input, size_t length)
         return length & (SHA1::BLOCKSIZE - 1);
     }
 #endif
+#if CRYPTOGAMS_ARM_SHA1
+    if (HasARMv7())
+    {
+        // The Cryptogams code uses a global variable named CRYPTOGAMS_armcaps
+        // for capabilities like ARMv7 and NEON. Storage is allocated in the
+        // module. We still need to set CRYPTOGAMS_armcaps accordingly.
+        // The Cryptogams code defines NEON as 1<<0; see ARMV7_NEON.
+        static unsigned int unused = CRYPTOGAMS_armcaps = HasNEON() ? (1<<0) : 0;
+
+        sha1_block_data_order(m_state, input, length / SHA1::BLOCKSIZE);
+        return length & (SHA1::BLOCKSIZE - 1);
+    }
+#endif
 #if CRYPTOPP_ARM_SHA1_AVAILABLE
     if (HasSHA1())
     {

sha1_armv4.h

@@ -0,0 +1,21 @@
+/* Header file for use with Cryptogam's ARMv4 SHA1.    */
+/* Also see http://www.openssl.org/~appro/cryptogams/  */
+/* https://wiki.openssl.org/index.php/Cryptogams_SHA.  */
+
+#ifndef CRYPTOGAMS_SHA1_ARMV4_H
+#define CRYPTOGAMS_SHA1_ARMV4_H
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+void sha1_block_data_order(void *state, const void *data, size_t blocks);
+
+/* Cryptogams arm caps */
+#define ARMV7_NEON (1<<0)
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif  /* CRYPTOGAMS_SHA1_ARMV4_H */