Merge #13386: SHA256 implementations based on Intel SHA Extensions

66b2cf1ccfad545a8ec3f2a854e23f647322bf30 Use immintrin.h everywhere for intrinsics (Pieter Wuille)
4c935e2eee456ff66cdfb908b0edffdd1e8a6c04 Add SHA256 implementation using using Intel SHA intrinsics (Pieter Wuille)
268400d3188200c9e3dcd3482c4853354388a721 [Refactor] CPU feature detection logic for SHA256 (Pieter Wuille)

Pull request description:

  Based on #13191.

  This adds SHA256 implementations that use Intel's SHA Extension instructions (using intrinsics). This needs GCC 4.9 or Clang 3.4.

  In addition to #13191, two extra implementations are provided:
  * (a) A variable-length SHA256 implementation using SHA extensions.
  * (b) A 2-way 64-byte input double-SHA256 implementation using SHA extensions.

  Benchmarks for 9001-element Merkle tree root computation on an AMD Ryzen 1800X system:
  * Using generic C++ code (pre-#10821): 6.1ms
  * Using SSE4 (master, #10821): 4.6ms
  * Using 4-way SSE4 specialized for 64-byte inputs (#13191): 2.8ms
  * Using 8-way AVX2 specialized for 64-byte inputs (#13191): 2.1ms
  * Using 2-way SHA-NI specialized for 64-byte inputs (this PR): 0.56ms

  Benchmarks for 32-byte SHA256 on the same system:
  * Using SSE4 (master, #10821): 190ns
  * Using SHA-NI (this PR): 53ns

  Benchmarks for 1000000-byte SHA256 on the same system:
  * Using SSE4 (master, #10821): 2.5ms
  * Using SHA-NI (this PR): 0.51ms

Tree-SHA512: 2b319e33b22579f815d91f9daf7994a5e1e799c4f73c13e15070dd54ba71f3f6438ccf77ae9cbd1ce76f972d9cbeb5f0edfea3d86f101bbc1055db70e42743b7

4 files changed, 433 insertions, 24 deletions

diff --git a/src/crypto/sha256.cpp b/src/crypto/sha256.cpp
index 3f02a05aab..fbdbef0bc6 100644
--- a/src/crypto/sha256.cpp
+++ b/src/crypto/sha256.cpp
@@ -29,6 +29,16 @@ namespace sha256d64_avx2
 void Transform_8way(unsigned char* out, const unsigned char* in);
 }
 
+namespace sha256d64_shani
+{
+void Transform_2way(unsigned char* out, const unsigned char* in);
+}
+
+namespace sha256_shani
+{
+void Transform(uint32_t* s, const unsigned char* chunk, size_t blocks);
+}
+
 // Internal implementation code.
 namespace
 {
@@ -448,6 +458,7 @@ void TransformD64Wrapper(unsigned char* out, const unsigned char* in)
 
 TransformType Transform = sha256::Transform;
 TransformD64Type TransformD64 = sha256::TransformD64;
+TransformD64Type TransformD64_2way = nullptr;
 TransformD64Type TransformD64_4way = nullptr;
 TransformD64Type TransformD64_8way = nullptr;
 
@@ -512,6 +523,13 @@ bool SelfTest() {
     TransformD64(out, data + 1);
     if (!std::equal(out, out + 32, result_d64)) return false;
 
+    // Test TransformD64_2way, if available.
+    if (TransformD64_2way) {
+        unsigned char out[64];
+        TransformD64_2way(out, data + 1);
+        if (!std::equal(out, out + 64, result_d64)) return false;
+    }
+
     // Test TransformD64_4way, if available.
     if (TransformD64_4way) {
         unsigned char out[128];
@@ -556,32 +574,64 @@ std::string SHA256AutoDetect()
 {
     std::string ret = "standard";
 #if defined(USE_ASM) && (defined(__x86_64__) || defined(__amd64__) || defined(__i386__))
-    (void)AVXEnabled; // Silence unused warning (in case ENABLE_AVX2 is not defined)
+    bool have_sse4 = false;
+    bool have_xsave = false;
+    bool have_avx = false;
+    bool have_avx2 = false;
+    bool have_shani = false;
+    bool enabled_avx = false;
+
+    (void)AVXEnabled;
+    (void)have_sse4;
+    (void)have_avx;
+    (void)have_xsave;
+    (void)have_avx2;
+    (void)have_shani;
+    (void)enabled_avx;
+
     uint32_t eax, ebx, ecx, edx;
     cpuid(1, 0, eax, ebx, ecx, edx);
-    if ((ecx >> 19) & 1) {
+    have_sse4 = (ecx >> 19) & 1;
+    have_xsave = (ecx >> 27) & 1;
+    have_avx = (ecx >> 28) & 1;
+    if (have_xsave && have_avx) {
+        enabled_avx = AVXEnabled();
+    }
+    if (have_sse4) {
+        cpuid(7, 0, eax, ebx, ecx, edx);
+        have_avx2 = (ebx >> 5) & 1;
+        have_shani = (ebx >> 29) & 1;
+    }
+
+#if defined(ENABLE_SHANI) && !defined(BUILD_BITCOIN_INTERNAL)
+    if (have_shani) {
+        Transform = sha256_shani::Transform;
+        TransformD64 = TransformD64Wrapper<sha256_shani::Transform>;
+        TransformD64_2way = sha256d64_shani::Transform_2way;
+        ret = "shani(1way,2way)";
+        have_sse4 = false; // Disable SSE4/AVX2;
+        have_avx2 = false;
+    }
+#endif
+
+    if (have_sse4) {
 #if defined(__x86_64__) || defined(__amd64__)
         Transform = sha256_sse4::Transform;
         TransformD64 = TransformD64Wrapper<sha256_sse4::Transform>;
+        ret = "sse4(1way)";
 #endif
 #if defined(ENABLE_SSE41) && !defined(BUILD_BITCOIN_INTERNAL)
         TransformD64_4way = sha256d64_sse41::Transform_4way;
-        ret = "sse4(1way+4way)";
-#if defined(ENABLE_AVX2) && !defined(BUILD_BITCOIN_INTERNAL)
-        if (((ecx >> 27) & 1) && ((ecx >> 28) & 1)) { // XSAVE and AVX
-            cpuid(7, 0, eax, ebx, ecx, edx);
-            if ((ebx >> 5) & 1) { // AVX2 flag
-                if (AVXEnabled()) { // OS has enabled AVX registers
-                    TransformD64_8way = sha256d64_avx2::Transform_8way;
-                    ret += ",avx2(8way)";
-                }
-            }
-        }
-#endif
-#else
-        ret = "sse4";
+        ret += ",sse41(4way)";
 #endif
     }
+
+#if defined(ENABLE_AVX2) && !defined(BUILD_BITCOIN_INTERNAL)
+    if (have_avx2 && have_avx && enabled_avx) {
+        TransformD64_8way = sha256d64_avx2::Transform_8way;
+        ret += ",avx2(8way)";
+    }
+#endif
 #endif
 
     assert(SelfTest());
@@ -663,6 +713,14 @@ void SHA256D64(unsigned char* out, const unsigned char* in, size_t blocks)
             blocks -= 4;
         }
     }
+    if (TransformD64_2way) {
+        while (blocks >= 2) {
+            TransformD64_2way(out, in);
+            out += 64;
+            in += 128;
+            blocks -= 2;
+        }
+    }
     while (blocks) {
         TransformD64(out, in);
         out += 32;
diff --git a/src/crypto/sha256_avx2.cpp b/src/crypto/sha256_avx2.cpp
index b338b06927..068e0e5ff6 100644
--- a/src/crypto/sha256_avx2.cpp
+++ b/src/crypto/sha256_avx2.cpp
@@ -1,11 +1,7 @@
 #ifdef ENABLE_AVX2
 
 #include <stdint.h>
-#if defined(_MSC_VER)
 #include <immintrin.h>
-#elif defined(__GNUC__)
-#include <x86intrin.h>
-#endif
 
 #include <crypto/sha256.h>
 #include <crypto/common.h>
diff --git a/src/crypto/sha256_shani.cpp b/src/crypto/sha256_shani.cpp
new file mode 100644
index 0000000000..e561da42c5
--- /dev/null
+++ b/src/crypto/sha256_shani.cpp
@@ -0,0 +1,359 @@
+// Copyright (c) 2018 The Bitcoin Core developers
+// Distributed under the MIT software license, see the accompanying
+// file COPYING or http://www.opensource.org/licenses/mit-license.php.
+//
+// Based on https://github.com/noloader/SHA-Intrinsics/blob/master/sha256-x86.c,
+// Written and placed in public domain by Jeffrey Walton.
+// Based on code from Intel, and by Sean Gulley for the miTLS project.
+
+#ifdef ENABLE_SHANI
+
+#include <stdint.h>
+#include <immintrin.h>
+
+#include <crypto/common.h>
+
+
+namespace {
+
+const __m128i MASK = _mm_set_epi64x(0x0c0d0e0f08090a0bULL, 0x0405060700010203ULL);
+const __m128i INIT0 = _mm_set_epi64x(0x6a09e667bb67ae85ull, 0x510e527f9b05688cull);
+const __m128i INIT1 = _mm_set_epi64x(0x3c6ef372a54ff53aull, 0x1f83d9ab5be0cd19ull);
+
+void inline  __attribute__((always_inline)) QuadRound(__m128i& state0, __m128i& state1, uint64_t k1, uint64_t k0)
+{
+    const __m128i msg = _mm_set_epi64x(k1, k0);
+    state1 = _mm_sha256rnds2_epu32(state1, state0, msg);
+    state0 = _mm_sha256rnds2_epu32(state0, state1, _mm_shuffle_epi32(msg, 0x0e));
+}
+
+void inline  __attribute__((always_inline)) QuadRound(__m128i& state0, __m128i& state1, __m128i m, uint64_t k1, uint64_t k0)
+{
+    const __m128i msg = _mm_add_epi32(m, _mm_set_epi64x(k1, k0));
+    state1 = _mm_sha256rnds2_epu32(state1, state0, msg);
+    state0 = _mm_sha256rnds2_epu32(state0, state1, _mm_shuffle_epi32(msg, 0x0e));
+}
+
+void inline  __attribute__((always_inline)) ShiftMessageA(__m128i& m0, __m128i m1)
+{
+    m0 = _mm_sha256msg1_epu32(m0, m1);
+}
+
+void inline  __attribute__((always_inline)) ShiftMessageC(__m128i& m0, __m128i m1, __m128i& m2)
+{
+    m2 = _mm_sha256msg2_epu32(_mm_add_epi32(m2, _mm_alignr_epi8(m1, m0, 4)), m1);
+}
+
+void inline __attribute__((always_inline)) ShiftMessageB(__m128i& m0, __m128i m1, __m128i& m2)
+{
+    ShiftMessageC(m0, m1, m2);
+    ShiftMessageA(m0, m1);
+}
+
+void inline __attribute__((always_inline)) Shuffle(__m128i& s0, __m128i& s1)
+{
+    const __m128i t1 = _mm_shuffle_epi32(s0, 0xB1);
+    const __m128i t2 = _mm_shuffle_epi32(s1, 0x1B);
+    s0 = _mm_alignr_epi8(t1, t2, 0x08);
+    s1 = _mm_blend_epi16(t2, t1, 0xF0);
+}
+
+void inline __attribute__((always_inline)) Unshuffle(__m128i& s0, __m128i& s1)
+{
+    const __m128i t1 = _mm_shuffle_epi32(s0, 0x1B);
+    const __m128i t2 = _mm_shuffle_epi32(s1, 0xB1);
+    s0 = _mm_blend_epi16(t1, t2, 0xF0);
+    s1 = _mm_alignr_epi8(t2, t1, 0x08);
+}
+
+__m128i inline  __attribute__((always_inline)) Load(const unsigned char* in)
+{
+    return _mm_shuffle_epi8(_mm_loadu_si128((const __m128i*)in), MASK);
+}
+
+void inline  __attribute__((always_inline)) Save(unsigned char* out, __m128i s)
+{
+    _mm_storeu_si128((__m128i*)out, _mm_shuffle_epi8(s, MASK));
+}
+}
+
+namespace sha256_shani {
+void Transform(uint32_t* s, const unsigned char* chunk, size_t blocks)
+{
+    __m128i m0, m1, m2, m3, s0, s1, so0, so1;
+
+    /* Load state */
+    s0 = _mm_loadu_si128((const __m128i*)s);
+    s1 = _mm_loadu_si128((const __m128i*)(s + 4));
+    Shuffle(s0, s1);
+
+    while (blocks--) {
+        /* Remember old state */
+        so0 = s0;
+        so1 = s1;
+
+        /* Load data and transform */
+        m0 = Load(chunk);
+        QuadRound(s0, s1, m0, 0xe9b5dba5b5c0fbcfull, 0x71374491428a2f98ull);
+        m1 = Load(chunk + 16);
+        QuadRound(s0, s1, m1, 0xab1c5ed5923f82a4ull, 0x59f111f13956c25bull);
+        ShiftMessageA(m0, m1);
+        m2 = Load(chunk + 32);
+        QuadRound(s0, s1, m2, 0x550c7dc3243185beull, 0x12835b01d807aa98ull);
+        ShiftMessageA(m1, m2);
+        m3 = Load(chunk + 48);
+        QuadRound(s0, s1, m3, 0xc19bf1749bdc06a7ull, 0x80deb1fe72be5d74ull);
+        ShiftMessageB(m2, m3, m0);
+        QuadRound(s0, s1, m0, 0x240ca1cc0fc19dc6ull, 0xefbe4786E49b69c1ull);
+        ShiftMessageB(m3, m0, m1);
+        QuadRound(s0, s1, m1, 0x76f988da5cb0a9dcull, 0x4a7484aa2de92c6full);
+        ShiftMessageB(m0, m1, m2);
+        QuadRound(s0, s1, m2, 0xbf597fc7b00327c8ull, 0xa831c66d983e5152ull);
+        ShiftMessageB(m1, m2, m3);
+        QuadRound(s0, s1, m3, 0x1429296706ca6351ull, 0xd5a79147c6e00bf3ull);
+        ShiftMessageB(m2, m3, m0);
+        QuadRound(s0, s1, m0, 0x53380d134d2c6dfcull, 0x2e1b213827b70a85ull);
+        ShiftMessageB(m3, m0, m1);
+        QuadRound(s0, s1, m1, 0x92722c8581c2c92eull, 0x766a0abb650a7354ull);
+        ShiftMessageB(m0, m1, m2);
+        QuadRound(s0, s1, m2, 0xc76c51A3c24b8b70ull, 0xa81a664ba2bfe8a1ull);
+        ShiftMessageB(m1, m2, m3);
+        QuadRound(s0, s1, m3, 0x106aa070f40e3585ull, 0xd6990624d192e819ull);
+        ShiftMessageB(m2, m3, m0);
+        QuadRound(s0, s1, m0, 0x34b0bcb52748774cull, 0x1e376c0819a4c116ull);
+        ShiftMessageB(m3, m0, m1);
+        QuadRound(s0, s1, m1, 0x682e6ff35b9cca4full, 0x4ed8aa4a391c0cb3ull);
+        ShiftMessageC(m0, m1, m2);
+        QuadRound(s0, s1, m2, 0x8cc7020884c87814ull, 0x78a5636f748f82eeull);
+        ShiftMessageC(m1, m2, m3);
+        QuadRound(s0, s1, m3, 0xc67178f2bef9A3f7ull, 0xa4506ceb90befffaull);
+
+        /* Combine with old state */
+        s0 = _mm_add_epi32(s0, so0);
+        s1 = _mm_add_epi32(s1, so1);
+
+        /* Advance */
+        chunk += 64;
+    }
+
+    Unshuffle(s0, s1);
+    _mm_storeu_si128((__m128i*)s, s0);
+    _mm_storeu_si128((__m128i*)(s + 4), s1);
+}
+}
+
+namespace sha256d64_shani {
+
+void Transform_2way(unsigned char* out, const unsigned char* in)
+{
+    __m128i am0, am1, am2, am3, as0, as1, aso0, aso1;
+    __m128i bm0, bm1, bm2, bm3, bs0, bs1, bso0, bso1;
+
+    /* Transform 1 */
+    bs0 = as0 = INIT0;
+    bs1 = as1 = INIT1;
+    am0 = Load(in);
+    bm0 = Load(in + 64);
+    QuadRound(as0, as1, am0, 0xe9b5dba5b5c0fbcfull, 0x71374491428a2f98ull);
+    QuadRound(bs0, bs1, bm0, 0xe9b5dba5b5c0fbcfull, 0x71374491428a2f98ull);
+    am1 = Load(in + 16);
+    bm1 = Load(in + 80);
+    QuadRound(as0, as1, am1, 0xab1c5ed5923f82a4ull, 0x59f111f13956c25bull);
+    QuadRound(bs0, bs1, bm1, 0xab1c5ed5923f82a4ull, 0x59f111f13956c25bull);
+    ShiftMessageA(am0, am1);
+    ShiftMessageA(bm0, bm1);
+    am2 = Load(in + 32);
+    bm2 = Load(in + 96);
+    QuadRound(as0, as1, am2, 0x550c7dc3243185beull, 0x12835b01d807aa98ull);
+    QuadRound(bs0, bs1, bm2, 0x550c7dc3243185beull, 0x12835b01d807aa98ull);
+    ShiftMessageA(am1, am2);
+    ShiftMessageA(bm1, bm2);
+    am3 = Load(in + 48);
+    bm3 = Load(in + 112);
+    QuadRound(as0, as1, am3, 0xc19bf1749bdc06a7ull, 0x80deb1fe72be5d74ull);
+    QuadRound(bs0, bs1, bm3, 0xc19bf1749bdc06a7ull, 0x80deb1fe72be5d74ull);
+    ShiftMessageB(am2, am3, am0);
+    ShiftMessageB(bm2, bm3, bm0);
+    QuadRound(as0, as1, am0, 0x240ca1cc0fc19dc6ull, 0xefbe4786E49b69c1ull);
+    QuadRound(bs0, bs1, bm0, 0x240ca1cc0fc19dc6ull, 0xefbe4786E49b69c1ull);
+    ShiftMessageB(am3, am0, am1);
+    ShiftMessageB(bm3, bm0, bm1);
+    QuadRound(as0, as1, am1, 0x76f988da5cb0a9dcull, 0x4a7484aa2de92c6full);
+    QuadRound(bs0, bs1, bm1, 0x76f988da5cb0a9dcull, 0x4a7484aa2de92c6full);
+    ShiftMessageB(am0, am1, am2);
+    ShiftMessageB(bm0, bm1, bm2);
+    QuadRound(as0, as1, am2, 0xbf597fc7b00327c8ull, 0xa831c66d983e5152ull);
+    QuadRound(bs0, bs1, bm2, 0xbf597fc7b00327c8ull, 0xa831c66d983e5152ull);
+    ShiftMessageB(am1, am2, am3);
+    ShiftMessageB(bm1, bm2, bm3);
+    QuadRound(as0, as1, am3, 0x1429296706ca6351ull, 0xd5a79147c6e00bf3ull);
+    QuadRound(bs0, bs1, bm3, 0x1429296706ca6351ull, 0xd5a79147c6e00bf3ull);
+    ShiftMessageB(am2, am3, am0);
+    ShiftMessageB(bm2, bm3, bm0);
+    QuadRound(as0, as1, am0, 0x53380d134d2c6dfcull, 0x2e1b213827b70a85ull);
+    QuadRound(bs0, bs1, bm0, 0x53380d134d2c6dfcull, 0x2e1b213827b70a85ull);
+    ShiftMessageB(am3, am0, am1);
+    ShiftMessageB(bm3, bm0, bm1);
+    QuadRound(as0, as1, am1, 0x92722c8581c2c92eull, 0x766a0abb650a7354ull);
+    QuadRound(bs0, bs1, bm1, 0x92722c8581c2c92eull, 0x766a0abb650a7354ull);
+    ShiftMessageB(am0, am1, am2);
+    ShiftMessageB(bm0, bm1, bm2);
+    QuadRound(as0, as1, am2, 0xc76c51A3c24b8b70ull, 0xa81a664ba2bfe8a1ull);
+    QuadRound(bs0, bs1, bm2, 0xc76c51A3c24b8b70ull, 0xa81a664ba2bfe8a1ull);
+    ShiftMessageB(am1, am2, am3);
+    ShiftMessageB(bm1, bm2, bm3);
+    QuadRound(as0, as1, am3, 0x106aa070f40e3585ull, 0xd6990624d192e819ull);
+    QuadRound(bs0, bs1, bm3, 0x106aa070f40e3585ull, 0xd6990624d192e819ull);
+    ShiftMessageB(am2, am3, am0);
+    ShiftMessageB(bm2, bm3, bm0);
+    QuadRound(as0, as1, am0, 0x34b0bcb52748774cull, 0x1e376c0819a4c116ull);
+    QuadRound(bs0, bs1, bm0, 0x34b0bcb52748774cull, 0x1e376c0819a4c116ull);
+    ShiftMessageB(am3, am0, am1);
+    ShiftMessageB(bm3, bm0, bm1);
+    QuadRound(as0, as1, am1, 0x682e6ff35b9cca4full, 0x4ed8aa4a391c0cb3ull);
+    QuadRound(bs0, bs1, bm1, 0x682e6ff35b9cca4full, 0x4ed8aa4a391c0cb3ull);
+    ShiftMessageC(am0, am1, am2);
+    ShiftMessageC(bm0, bm1, bm2);
+    QuadRound(as0, as1, am2, 0x8cc7020884c87814ull, 0x78a5636f748f82eeull);
+    QuadRound(bs0, bs1, bm2, 0x8cc7020884c87814ull, 0x78a5636f748f82eeull);
+    ShiftMessageC(am1, am2, am3);
+    ShiftMessageC(bm1, bm2, bm3);
+    QuadRound(as0, as1, am3, 0xc67178f2bef9A3f7ull, 0xa4506ceb90befffaull);
+    QuadRound(bs0, bs1, bm3, 0xc67178f2bef9A3f7ull, 0xa4506ceb90befffaull);
+    as0 = _mm_add_epi32(as0, INIT0);
+    bs0 = _mm_add_epi32(bs0, INIT0);
+    as1 = _mm_add_epi32(as1, INIT1);
+    bs1 = _mm_add_epi32(bs1, INIT1);
+
+    /* Transform 2 */
+    aso0 = as0;
+    bso0 = bs0;
+    aso1 = as1;
+    bso1 = bs1;
+    QuadRound(as0, as1, 0xe9b5dba5b5c0fbcfull, 0x71374491c28a2f98ull);
+    QuadRound(bs0, bs1, 0xe9b5dba5b5c0fbcfull, 0x71374491c28a2f98ull);
+    QuadRound(as0, as1, 0xab1c5ed5923f82a4ull, 0x59f111f13956c25bull);
+    QuadRound(bs0, bs1, 0xab1c5ed5923f82a4ull, 0x59f111f13956c25bull);
+    QuadRound(as0, as1, 0x550c7dc3243185beull, 0x12835b01d807aa98ull);
+    QuadRound(bs0, bs1, 0x550c7dc3243185beull, 0x12835b01d807aa98ull);
+    QuadRound(as0, as1, 0xc19bf3749bdc06a7ull, 0x80deb1fe72be5d74ull);
+    QuadRound(bs0, bs1, 0xc19bf3749bdc06a7ull, 0x80deb1fe72be5d74ull);
+    QuadRound(as0, as1, 0x240cf2540fe1edc6ull, 0xf0fe4786649b69c1ull);
+    QuadRound(bs0, bs1, 0x240cf2540fe1edc6ull, 0xf0fe4786649b69c1ull);
+    QuadRound(as0, as1, 0x16f988fa61b9411eull, 0x6cc984be4fe9346full);
+    QuadRound(bs0, bs1, 0x16f988fa61b9411eull, 0x6cc984be4fe9346full);
+    QuadRound(as0, as1, 0xb9d99ec7b019fc65ull, 0xa88e5a6df2c65152ull);
+    QuadRound(bs0, bs1, 0xb9d99ec7b019fc65ull, 0xa88e5a6df2c65152ull);
+    QuadRound(as0, as1, 0xc7353eb0fdb1232bull, 0xe70eeaa09a1231c3ull);
+    QuadRound(bs0, bs1, 0xc7353eb0fdb1232bull, 0xe70eeaa09a1231c3ull);
+    QuadRound(as0, as1, 0xdc1eeefd5a0f118full, 0xcb976d5f3069bad5ull);
+    QuadRound(bs0, bs1, 0xdc1eeefd5a0f118full, 0xcb976d5f3069bad5ull);
+    QuadRound(as0, as1, 0xe15d5b1658f4ca9dull, 0xde0b7a040a35b689ull);
+    QuadRound(bs0, bs1, 0xe15d5b1658f4ca9dull, 0xde0b7a040a35b689ull);
+    QuadRound(as0, as1, 0x6fab9537a507ea32ull, 0x37088980007f3e86ull);
+    QuadRound(bs0, bs1, 0x6fab9537a507ea32ull, 0x37088980007f3e86ull);
+    QuadRound(as0, as1, 0xc0bbbe37cdaa3b6dull, 0x0d8cd6f117406110ull);
+    QuadRound(bs0, bs1, 0xc0bbbe37cdaa3b6dull, 0x0d8cd6f117406110ull);
+    QuadRound(as0, as1, 0x6fd15ca70b02e931ull, 0xdb48a36383613bdaull);
+    QuadRound(bs0, bs1, 0x6fd15ca70b02e931ull, 0xdb48a36383613bdaull);
+    QuadRound(as0, as1, 0x6d4378906ed41a95ull, 0x31338431521afacaull);
+    QuadRound(bs0, bs1, 0x6d4378906ed41a95ull, 0x31338431521afacaull);
+    QuadRound(as0, as1, 0x532fb63cb5c9a0e6ull, 0x9eccabbdc39c91f2ull);
+    QuadRound(bs0, bs1, 0x532fb63cb5c9a0e6ull, 0x9eccabbdc39c91f2ull);
+    QuadRound(as0, as1, 0x4c191d76a4954b68ull, 0x07237ea3d2c741c6ull);
+    QuadRound(bs0, bs1, 0x4c191d76a4954b68ull, 0x07237ea3d2c741c6ull);
+    as0 = _mm_add_epi32(as0, aso0);
+    bs0 = _mm_add_epi32(bs0, bso0);
+    as1 = _mm_add_epi32(as1, aso1);
+    bs1 = _mm_add_epi32(bs1, bso1);
+
+    /* Extract hash */
+    Unshuffle(as0, as1);
+    Unshuffle(bs0, bs1);
+    am0 = as0;
+    bm0 = bs0;
+    am1 = as1;
+    bm1 = bs1;
+
+    /* Transform 3 */
+    bs0 = as0 = INIT0;
+    bs1 = as1 = INIT1;
+    QuadRound(as0, as1, am0, 0xe9b5dba5B5c0fbcfull, 0x71374491428a2f98ull);
+    QuadRound(bs0, bs1, bm0, 0xe9b5dba5B5c0fbcfull, 0x71374491428a2f98ull);
+    QuadRound(as0, as1, am1, 0xab1c5ed5923f82a4ull, 0x59f111f13956c25bull);
+    QuadRound(bs0, bs1, bm1, 0xab1c5ed5923f82a4ull, 0x59f111f13956c25bull);
+    ShiftMessageA(am0, am1);
+    ShiftMessageA(bm0, bm1);
+    bm2 = am2 = _mm_set_epi64x(0x0ull, 0x80000000ull);
+    QuadRound(as0, as1, 0x550c7dc3243185beull, 0x12835b015807aa98ull);
+    QuadRound(bs0, bs1, 0x550c7dc3243185beull, 0x12835b015807aa98ull);
+    ShiftMessageA(am1, am2);
+    ShiftMessageA(bm1, bm2);
+    bm3 = am3 = _mm_set_epi64x(0x10000000000ull, 0x0ull);
+    QuadRound(as0, as1, 0xc19bf2749bdc06a7ull, 0x80deb1fe72be5d74ull);
+    QuadRound(bs0, bs1, 0xc19bf2749bdc06a7ull, 0x80deb1fe72be5d74ull);
+    ShiftMessageB(am2, am3, am0);
+    ShiftMessageB(bm2, bm3, bm0);
+    QuadRound(as0, as1, am0, 0x240ca1cc0fc19dc6ull, 0xefbe4786e49b69c1ull);
+    QuadRound(bs0, bs1, bm0, 0x240ca1cc0fc19dc6ull, 0xefbe4786e49b69c1ull);
+    ShiftMessageB(am3, am0, am1);
+    ShiftMessageB(bm3, bm0, bm1);
+    QuadRound(as0, as1, am1, 0x76f988da5cb0a9dcull, 0x4a7484aa2de92c6full);
+    QuadRound(bs0, bs1, bm1, 0x76f988da5cb0a9dcull, 0x4a7484aa2de92c6full);
+    ShiftMessageB(am0, am1, am2);
+    ShiftMessageB(bm0, bm1, bm2);
+    QuadRound(as0, as1, am2, 0xbf597fc7b00327c8ull, 0xa831c66d983e5152ull);
+    QuadRound(bs0, bs1, bm2, 0xbf597fc7b00327c8ull, 0xa831c66d983e5152ull);
+    ShiftMessageB(am1, am2, am3);
+    ShiftMessageB(bm1, bm2, bm3);
+    QuadRound(as0, as1, am3, 0x1429296706ca6351ull, 0xd5a79147c6e00bf3ull);
+    QuadRound(bs0, bs1, bm3, 0x1429296706ca6351ull, 0xd5a79147c6e00bf3ull);
+    ShiftMessageB(am2, am3, am0);
+    ShiftMessageB(bm2, bm3, bm0);
+    QuadRound(as0, as1, am0, 0x53380d134d2c6dfcull, 0x2e1b213827b70a85ull);
+    QuadRound(bs0, bs1, bm0, 0x53380d134d2c6dfcull, 0x2e1b213827b70a85ull);
+    ShiftMessageB(am3, am0, am1);
+    ShiftMessageB(bm3, bm0, bm1);
+    QuadRound(as0, as1, am1, 0x92722c8581c2c92eull, 0x766a0abb650a7354ull);
+    QuadRound(bs0, bs1, bm1, 0x92722c8581c2c92eull, 0x766a0abb650a7354ull);
+    ShiftMessageB(am0, am1, am2);
+    ShiftMessageB(bm0, bm1, bm2);
+    QuadRound(as0, as1, am2, 0xc76c51a3c24b8b70ull, 0xa81a664ba2bfe8A1ull);
+    QuadRound(bs0, bs1, bm2, 0xc76c51a3c24b8b70ull, 0xa81a664ba2bfe8A1ull);
+    ShiftMessageB(am1, am2, am3);
+    ShiftMessageB(bm1, bm2, bm3);
+    QuadRound(as0, as1, am3, 0x106aa070f40e3585ull, 0xd6990624d192e819ull);
+    QuadRound(bs0, bs1, bm3, 0x106aa070f40e3585ull, 0xd6990624d192e819ull);
+    ShiftMessageB(am2, am3, am0);
+    ShiftMessageB(bm2, bm3, bm0);
+    QuadRound(as0, as1, am0, 0x34b0bcb52748774cull, 0x1e376c0819a4c116ull);
+    QuadRound(bs0, bs1, bm0, 0x34b0bcb52748774cull, 0x1e376c0819a4c116ull);
+    ShiftMessageB(am3, am0, am1);
+    ShiftMessageB(bm3, bm0, bm1);
+    QuadRound(as0, as1, am1, 0x682e6ff35b9cca4full, 0x4ed8aa4a391c0cb3ull);
+    QuadRound(bs0, bs1, bm1, 0x682e6ff35b9cca4full, 0x4ed8aa4a391c0cb3ull);
+    ShiftMessageC(am0, am1, am2);
+    ShiftMessageC(bm0, bm1, bm2);
+    QuadRound(as0, as1, am2, 0x8cc7020884c87814ull, 0x78a5636f748f82eeull);
+    QuadRound(bs0, bs1, bm2, 0x8cc7020884c87814ull, 0x78a5636f748f82eeull);
+    ShiftMessageC(am1, am2, am3);
+    ShiftMessageC(bm1, bm2, bm3);
+    QuadRound(as0, as1, am3, 0xc67178f2bef9a3f7ull, 0xa4506ceb90befffaull);
+    QuadRound(bs0, bs1, bm3, 0xc67178f2bef9a3f7ull, 0xa4506ceb90befffaull);
+    as0 = _mm_add_epi32(as0, INIT0);
+    bs0 = _mm_add_epi32(bs0, INIT0);
+    as1 = _mm_add_epi32(as1, INIT1);
+    bs1 = _mm_add_epi32(bs1, INIT1);
+
+    /* Extract hash into out */
+    Unshuffle(as0, as1);
+    Unshuffle(bs0, bs1);
+    Save(out, as0);
+    Save(out + 16, as1);
+    Save(out + 32, bs0);
+    Save(out + 48, bs1);
+}
+
+}
+
+#endif
diff --git a/src/crypto/sha256_sse41.cpp b/src/crypto/sha256_sse41.cpp
index be71dd8fb8..adca870e2d 100644
--- a/src/crypto/sha256_sse41.cpp
+++ b/src/crypto/sha256_sse41.cpp
@@ -1,11 +1,7 @@
 #ifdef ENABLE_SSE41
 
 #include <stdint.h>
-#if defined(_MSC_VER)
 #include <immintrin.h>
-#elif defined(__GNUC__)
-#include <x86intrin.h>
-#endif
 
 #include <crypto/sha256.h>
 #include <crypto/common.h>