added portable slow-hash, updated makefile targets, fixed readme

3 files changed, 166 insertions, 16 deletions

diff --git a/Makefile b/Makefile
index 4e3397230..232842a45 100644
--- a/Makefile
+++ b/Makefile
@@ -1,17 +1,17 @@
-all: all-release
+all: release-all
 
 cmake-debug:
 	mkdir -p build/debug
 	cd build/debug && cmake -D CMAKE_BUILD_TYPE=Debug ../..
 
-build-debug: cmake-debug
+debug: cmake-debug
 	cd build/debug && $(MAKE)
 
-test-debug: build-debug
+debug-test: debug
 	mkdir -p build/debug
 	cd build/debug && cmake -D BUILD_TESTS=ON -D CMAKE_BUILD_TYPE=Debug ../.. && $(MAKE) test
 
-all-debug:
+debug-all:
 	mkdir -p build/debug
 	cd build/debug && cmake -D BUILD_TESTS=ON -D CMAKE_BUILD_TYPE=Debug ../.. && $(MAKE)
 
@@ -19,14 +19,14 @@ cmake-release:
 	mkdir -p build/release
 	cd build/release && cmake -D CMAKE_BUILD_TYPE=Release ../..
 
-build-release: cmake-release
+release: cmake-release
 	cd build/release && $(MAKE)
 
-test-release: build-release
+release-test: release
 	mkdir -p build/release
 	cd build/release && cmake -D BUILD_TESTS=ON -D CMAKE_BUILD_TYPE=release ../.. && $(MAKE) test
 
-all-release:
+release-all:
 	mkdir -p build/release
 	cd build/release && cmake -D BUILD_TESTS=ON -D CMAKE_BUILD_TYPE=release ../.. && $(MAKE)
 
@@ -43,4 +43,4 @@ clean:
 tags:
 	ctags -R --sort=1 --c++-kinds=+p --fields=+iaS --extra=+q --language-force=C++ src contrib tests/gtest
 
-.PHONY: all cmake-debug build-debug test-debug all-debug cmake-release build-release test-release all-release clean tags
+.PHONY: all cmake-debug debug debug-test debug-all cmake-release release release-test release-all clean tags
diff --git a/README.md b/README.md
index bd0866002..2fb87ef5e 100644
--- a/README.md
+++ b/README.md
@@ -56,14 +56,14 @@ Static Build Additional Dependencies: ldns 1.6.17 or later, expat 1.1 or later,
 **Basic Process:**
 
 * To build, change to the root of the source code directory, and run `make`.
-* The resulting executables can be found in `build/release/src` or `build/debug/src`, depending on what you're building.
+* The resulting executables can be found in `build/release/bin` or `build/debug/bin`, depending on what you're building.
 
 **Advanced options:**
 
 * Parallel build: run `make -j<number of threads>` instead of `make`.
 * Statically linked release build: run `make release-static`.
-* Debug build: run `make build-debug`.
-* Test suite: run `make test-release` to run tests in addition to building. Running `make test-debug` will do the same to the debug version.
+* Debug build: run `make debug`.
+* Test suite: run `make release-test` to run tests in addition to building. Running `make debug-test` will do the same to the debug version.
 * Building with Clang: it may be possible to use Clang instead of GCC, but this may not work everywhere. To build, run `export CC=clang CXX=clang++` before running `make`.
 
 ### On OS X:
@@ -111,7 +111,7 @@ cmake -G "MSYS Makefiles" -D CMAKE_BUILD_TYPE=Release -D CMAKE_TOOLCHAIN_FILE=..
 cmake -G "MSYS Makefiles" -D CMAKE_BUILD_TYPE=Release -D CMAKE_TOOLCHAIN_FILE=../cmake/32-bit-toolchain.cmake -D MSYS2_FOLDER=c:/msys32 ..
 ```
 * You can now run `make` to have it build
-* The resulting executables can be found in `build/release/src` or `build/debug/src`, depending on what you're building.
+* The resulting executables can be found in `build/release/bin` or `build/debug/bin`, depending on what you're building.
 
 If you installed MSYS2 in a folder other than c:/msys64, make the appropriate substitution above.
 
@@ -119,8 +119,8 @@ If you installed MSYS2 in a folder other than c:/msys64, make the appropriate su
 
 * Parallel build: run `make -j<number of threads>` instead of `make`.
 * Statically linked release build: run `make release-static`.
-* Debug build: run `make build-debug`.
-* Test suite: run `make test-release` to run tests in addition to building. Running `make test-debug` will do the same to the debug version.
+* Debug build: run `make debug`.
+* Test suite: run `make release-test` to run tests in addition to building. Running `make debug-test` will do the same to the debug version.
 
 ### On FreeBSD:
 
diff --git a/src/crypto/slow-hash.c b/src/crypto/slow-hash.c
index 301aa3186..fda854ea5 100644
--- a/src/crypto/slow-hash.c
+++ b/src/crypto/slow-hash.c
@@ -37,9 +37,11 @@
 #include "hash-ops.h"
 #include "oaes_lib.h"
 
-#ifdef __SSE2__
+#if defined(__x86_64__) || defined(__i386)
+// Optimised code below, uses x86-specific intrinsics, SSE2, AES-NI
+// Fall back to more portable code is down at the bottom
+
 #include <emmintrin.h>
-#endif
 
 #if defined(_MSC_VER)
 #include <intrin.h>
@@ -621,3 +623,151 @@ void cn_slow_hash(const void *data, size_t length, char *hash)
     hash_permutation(&state.hs);
     extra_hashes[state.hs.b[0] & 3](&state, 200, hash);
 }
+
+#else
+// Portable implementation as a fallback
+
+static void (*const extra_hashes[4])(const void *, size_t, char *) = {
+  hash_extra_blake, hash_extra_groestl, hash_extra_jh, hash_extra_skein
+};
+
+#define MEMORY         (1 << 21) /* 2 MiB */
+#define ITER           (1 << 20)
+#define AES_BLOCK_SIZE  16
+#define AES_KEY_SIZE    32 /*16*/
+#define INIT_SIZE_BLK   8
+#define INIT_SIZE_BYTE (INIT_SIZE_BLK * AES_BLOCK_SIZE)
+
+extern int aesb_single_round(const uint8_t *in, uint8_t*out, const uint8_t *expandedKey);
+extern int aesb_pseudo_round(const uint8_t *in, uint8_t *out, const uint8_t *expandedKey);
+
+static size_t e2i(const uint8_t* a, size_t count) { return (*((uint64_t*)a) / AES_BLOCK_SIZE) & (count - 1); }
+
+static void mul(const uint8_t* a, const uint8_t* b, uint8_t* res) {
+  uint64_t a0, b0;
+  uint64_t hi, lo;
+
+  a0 = SWAP64LE(((uint64_t*)a)[0]);
+  b0 = SWAP64LE(((uint64_t*)b)[0]);
+  lo = mul128(a0, b0, &hi);
+  ((uint64_t*)res)[0] = SWAP64LE(hi);
+  ((uint64_t*)res)[1] = SWAP64LE(lo);
+}
+
+static void sum_half_blocks(uint8_t* a, const uint8_t* b) {
+  uint64_t a0, a1, b0, b1;
+
+  a0 = SWAP64LE(((uint64_t*)a)[0]);
+  a1 = SWAP64LE(((uint64_t*)a)[1]);
+  b0 = SWAP64LE(((uint64_t*)b)[0]);
+  b1 = SWAP64LE(((uint64_t*)b)[1]);
+  a0 += b0;
+  a1 += b1;
+  ((uint64_t*)a)[0] = SWAP64LE(a0);
+  ((uint64_t*)a)[1] = SWAP64LE(a1);
+}
+#define U64(x) ((uint64_t *) (x))
+
+static void copy_block(uint8_t* dst, const uint8_t* src) {
+  memcpy(dst, src, AES_BLOCK_SIZE);
+}
+
+static void swap_blocks(uint8_t *a, uint8_t *b){
+	uint64_t t[2];
+	U64(t)[0] = U64(a)[0];
+	U64(t)[1] = U64(a)[1];
+	U64(a)[0] = U64(b)[0];
+	U64(a)[1] = U64(b)[1];
+	U64(b)[0] = U64(t)[0];
+	U64(b)[1] = U64(t)[1];
+}
+
+static void xor_blocks(uint8_t* a, const uint8_t* b) {
+  size_t i;
+  for (i = 0; i < AES_BLOCK_SIZE; i++) {
+    a[i] ^= b[i];
+  }
+}
+
+#pragma pack(push, 1)
+union cn_slow_hash_state {
+  union hash_state hs;
+  struct {
+    uint8_t k[64];
+    uint8_t init[INIT_SIZE_BYTE];
+  };
+};
+#pragma pack(pop)
+
+void cn_slow_hash(const void *data, size_t length, char *hash) {
+  uint8_t long_state[MEMORY];
+  union cn_slow_hash_state state;
+  uint8_t text[INIT_SIZE_BYTE];
+  uint8_t a[AES_BLOCK_SIZE];
+  uint8_t b[AES_BLOCK_SIZE];
+  uint8_t c[AES_BLOCK_SIZE];
+  uint8_t d[AES_BLOCK_SIZE];
+  size_t i, j;
+  uint8_t aes_key[AES_KEY_SIZE];
+  oaes_ctx *aes_ctx;
+
+  hash_process(&state.hs, data, length);
+  memcpy(text, state.init, INIT_SIZE_BYTE);
+  memcpy(aes_key, state.hs.b, AES_KEY_SIZE);
+  aes_ctx = (oaes_ctx *) oaes_alloc();
+  
+  oaes_key_import_data(aes_ctx, aes_key, AES_KEY_SIZE);
+  for (i = 0; i < MEMORY / INIT_SIZE_BYTE; i++) {
+    for (j = 0; j < INIT_SIZE_BLK; j++) {    
+		  aesb_pseudo_round(&text[AES_BLOCK_SIZE * j], &text[AES_BLOCK_SIZE * j], aes_ctx->key->exp_data);
+    }
+    memcpy(&long_state[i * INIT_SIZE_BYTE], text, INIT_SIZE_BYTE);
+  }
+
+  for (i = 0; i < 16; i++) {
+    a[i] = state.k[     i] ^ state.k[32 + i];
+    b[i] = state.k[16 + i] ^ state.k[48 + i];
+  }
+
+  for (i = 0; i < ITER / 2; i++) {
+    /* Dependency chain: address -> read value ------+
+     * written value <-+ hard function (AES or MUL) <+
+     * next address  <-+
+     */
+    /* Iteration 1 */
+    j = e2i(a, MEMORY / AES_BLOCK_SIZE);
+    copy_block(c, &long_state[j * AES_BLOCK_SIZE]);
+	  aesb_single_round(c, c, a);
+    xor_blocks(b, c);
+    swap_blocks(b, c);
+    copy_block(&long_state[j * AES_BLOCK_SIZE], c);
+    assert(j == e2i(a, MEMORY / AES_BLOCK_SIZE));
+    swap_blocks(a, b);
+    /* Iteration 2 */
+    j = e2i(a, MEMORY / AES_BLOCK_SIZE);
+    copy_block(c, &long_state[j * AES_BLOCK_SIZE]);
+    mul(a, c, d);
+    sum_half_blocks(b, d);
+    swap_blocks(b, c);
+    xor_blocks(b, c);
+    copy_block(&long_state[j * AES_BLOCK_SIZE], c);
+    assert(j == e2i(a, MEMORY / AES_BLOCK_SIZE));
+    swap_blocks(a, b);
+  }
+
+  memcpy(text, state.init, INIT_SIZE_BYTE);
+  oaes_key_import_data(aes_ctx, &state.hs.b[32], AES_KEY_SIZE);
+  for (i = 0; i < MEMORY / INIT_SIZE_BYTE; i++) {
+    for (j = 0; j < INIT_SIZE_BLK; j++) {
+      xor_blocks(&text[j * AES_BLOCK_SIZE], &long_state[i * INIT_SIZE_BYTE + j * AES_BLOCK_SIZE]);
+	    aesb_pseudo_round(&text[AES_BLOCK_SIZE * j], &text[AES_BLOCK_SIZE * j], aes_ctx->key->exp_data);
+    }
+  }
+  memcpy(state.init, text, INIT_SIZE_BYTE);
+  hash_permutation(&state.hs);
+  /*memcpy(hash, &state, 32);*/
+  extra_hashes[state.hs.b[0] & 3](&state, 200, hash);
+  oaes_free((OAES_CTX **) &aes_ctx);
+}
+
+#endif
\ No newline at end of file