Ported xmr-stak 2.3.0 rc to ppc64le

16 files changed, 1242 insertions, 399 deletions

diff --git a/xmrstak/backend/amd/jconf.cpp b/xmrstak/backend/amd/jconf.cpp
index 93ba709..8b3b2da 100644
--- a/xmrstak/backend/amd/jconf.cpp
+++ b/xmrstak/backend/amd/jconf.cpp
@@ -26,13 +26,6 @@
 #include "xmrstak/misc/jext.hpp"
 #include "xmrstak/misc/console.hpp"
 
-#ifdef _WIN32
-#define strcasecmp _stricmp
-#include <intrin.h>
-#else
-#include <cpuid.h>
-#endif
-
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
diff --git a/xmrstak/backend/amd/minethd.cpp b/xmrstak/backend/amd/minethd.cpp
index 46a04d5..d461aad 100644
--- a/xmrstak/backend/amd/minethd.cpp
+++ b/xmrstak/backend/amd/minethd.cpp
@@ -25,7 +25,7 @@
 #include "autoAdjust.hpp"
 #include "amd_gpu/gpu.hpp"
 
-#include "xmrstak/backend/cpu/crypto/cryptonight_aesni.h"
+#include "xmrstak/backend/cpu/crypto/cryptonight_altivec.h"
 #include "xmrstak/backend/cpu/crypto/cryptonight.h"
 #include "xmrstak/misc/configEditor.hpp"
 #include "xmrstak/misc/console.hpp"
diff --git a/xmrstak/backend/cpu/autoAdjust.hpp b/xmrstak/backend/cpu/autoAdjust.hpp
index db805ec..6134aa7 100644
--- a/xmrstak/backend/cpu/autoAdjust.hpp
+++ b/xmrstak/backend/cpu/autoAdjust.hpp
@@ -65,7 +65,7 @@ public:
 			if(L3KB_size < halfHashMemSizeKB || L3KB_size > (halfHashMemSizeKB * 2048))
 				printer::inst()->print_msg(L0, "Autoconf failed: L3 size sanity check failed - %u KB.", L3KB_size);
 
-			conf += std::string("    { \"low_power_mode\" : false, \"no_prefetch\" : true, \"affine_to_cpu\" : false },\n");
+			conf += std::string("    { \"low_power_mode\" : false, \"be_mode\" : true, \"affine_to_cpu\" : false },\n");
 			printer::inst()->print_msg(L0, "Autoconf FAILED. Create config for a single thread. Please try to add new ones until the hashrate slows down.");
 		}
 		else
@@ -89,7 +89,7 @@ public:
 
 				conf += std::string("    { \"low_power_mode\" : ");
 				conf += std::string(double_mode ? "true" : "false");
-				conf += std::string(", \"no_prefetch\" : true, \"affine_to_cpu\" : ");
+				conf += std::string(", \"be_mode\" : true, \"affine_to_cpu\" : ");
 				conf += std::to_string(aff_id);
 				conf += std::string(" },\n");
 
@@ -120,59 +120,13 @@ public:
 private:
 	bool detectL3Size()
 	{
-		int32_t cpu_info[4];
-		char cpustr[13] = {0};
-
-		::jconf::cpuid(0, 0, cpu_info);
-		memcpy(cpustr, &cpu_info[1], 4);
-		memcpy(cpustr+4, &cpu_info[3], 4);
-		memcpy(cpustr+8, &cpu_info[2], 4);
-
-		if(strcmp(cpustr, "GenuineIntel") == 0)
-		{
-			::jconf::cpuid(4, 3, cpu_info);
-
-			if(get_masked(cpu_info[0], 7, 5) != 3)
-			{
-				printer::inst()->print_msg(L0, "Autoconf failed: Couln't find L3 cache page.");
-				return false;
-			}
-
-			L3KB_size = ((get_masked(cpu_info[1], 31, 22) + 1) * (get_masked(cpu_info[1], 21, 12) + 1) *
-				(get_masked(cpu_info[1], 11, 0) + 1) * (cpu_info[2] + 1)) / 1024;
-
-			return true;
-		}
-		else if(strcmp(cpustr, "AuthenticAMD") == 0)
-		{
-			::jconf::cpuid(0x80000006, 0, cpu_info);
-
-			L3KB_size = get_masked(cpu_info[3], 31, 18) * 512;
-
-			::jconf::cpuid(1, 0, cpu_info);
-			if(get_masked(cpu_info[0], 11, 8) < 0x17) //0x17h is Zen
-				old_amd = true;
-
-			return true;
-		}
-		else
-		{
-			printer::inst()->print_msg(L0, "Autoconf failed: Unknown CPU type: %s.", cpustr);
-			return false;
-		}
-	}
+	  return false;
+  }
 
 	void detectCPUConf()
 	{
-#ifdef _WIN32
-		SYSTEM_INFO info;
-		GetSystemInfo(&info);
-		corecnt = info.dwNumberOfProcessors;
-		linux_layout = false;
-#else
 		corecnt = sysconf(_SC_NPROCESSORS_ONLN);
 		linux_layout = true;
-#endif // _WIN32
 	}
 
 	int32_t L3KB_size = 0;
diff --git a/xmrstak/backend/cpu/autoAdjustHwloc.hpp b/xmrstak/backend/cpu/autoAdjustHwloc.hpp
index 568abb5..a9cc2d7 100644
--- a/xmrstak/backend/cpu/autoAdjustHwloc.hpp
+++ b/xmrstak/backend/cpu/autoAdjustHwloc.hpp
@@ -67,7 +67,7 @@ public:
 			{
 				conf += std::string("    { \"low_power_mode\" : ");
 				conf += std::string((id & 0x8000000) != 0 ? "true" : "false");
-				conf += std::string(", \"no_prefetch\" : true, \"affine_to_cpu\" : ");
+				conf += std::string(", \"be_mode\" : true, \"affine_to_cpu\" : ");
 				conf += std::to_string(id & 0x7FFFFFF);
 				conf += std::string(" },\n");
 			}
@@ -75,7 +75,7 @@ public:
 		catch(const std::runtime_error& err)
 		{
 			// \todo add fallback to default auto adjust
-			conf += std::string("    { \"low_power_mode\" : false, \"no_prefetch\" : true, \"affine_to_cpu\" : false },\n");
+			conf += std::string("    { \"low_power_mode\" : false, \"be_mode\" : true, \"affine_to_cpu\" : false },\n");
 			printer::inst()->print_msg(L0, "Autoconf FAILED: %s. Create config for a single thread.", err.what());
 		}
 
diff --git a/xmrstak/backend/cpu/config.tpl b/xmrstak/backend/cpu/config.tpl
index cb4b950..91a67af 100644
--- a/xmrstak/backend/cpu/config.tpl
+++ b/xmrstak/backend/cpu/config.tpl
@@ -7,8 +7,8 @@ R"===(
  *                  the maximum performance. When set to a number N greater than 1, this mode will increase the
  *                  cache usage and single thread performance by N times.
  *
- * no_prefetch -    Some sytems can gain up to extra 5% here, but sometimes it will have no difference or make
- *                  things slower.
+ * be_mode -        Power8 likes BE, Power9 does not.
+ *                  
  *
  * affine_to_cpu -  This can be either false (no affinity), or the CPU core number. Note that on hyperthreading 
  *                  systems it is better to assign threads to physical cores. On Windows this usually means selecting 
@@ -21,8 +21,8 @@ R"===(
  * A filled out configuration should look like this:
  * "cpu_threads_conf" :
  * [ 
- *      { "low_power_mode" : false, "no_prefetch" : true, "affine_to_cpu" : 0 },
- *      { "low_power_mode" : false, "no_prefetch" : true, "affine_to_cpu" : 1 },
+ *      { "low_power_mode" : false, "be_mode" : true, "affine_to_cpu" : 0 },
+ *      { "low_power_mode" : false, "be_mode" : true, "affine_to_cpu" : 1 },
  * ],
  * If you do not wish to mine with your CPU(s) then use:
  * "cpu_threads_conf" :
diff --git a/xmrstak/backend/cpu/crypto/cryptonight_altivec.h b/xmrstak/backend/cpu/crypto/cryptonight_altivec.h
new file mode 100644
index 0000000..3727a01
--- /dev/null
+++ b/xmrstak/backend/cpu/crypto/cryptonight_altivec.h
@@ -0,0 +1,1187 @@
+/*
+  * This program is free software: you can redistribute it and/or modify
+  * it under the terms of the GNU General Public License as published by
+  * the Free Software Foundation, either version 3 of the License, or
+  * any later version.
+  *
+  * This program is distributed in the hope that it will be useful,
+  * but WITHOUT ANY WARRANTY; without even the implied warranty of
+  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+  * GNU General Public License for more details.
+  *
+  * You should have received a copy of the GNU General Public License
+  * along with this program.  If not, see <http://www.gnu.org/licenses/>.
+  *
+  */
+#pragma once
+
+#include "cryptonight.h"
+#include "xmrstak/backend/cryptonight.hpp"
+#include "soft_aes.hpp"
+#include <memory.h>
+#include <stdio.h>
+#include <altivec.h>
+#undef vector
+#undef pixel
+#undef bool
+typedef __vector unsigned char __m128i;
+typedef __vector unsigned long long __m128ll;
+
+static inline uint64_t _umul128(uint64_t a, uint64_t b, uint64_t* hi)
+{
+  uint64_t lo;
+  asm( 
+    "mulld  %0, %1, %2" :  
+    "=r" (lo) : 
+    "r" (a), 
+    "r" (b)); 
+  asm(
+    "mulhdu %0, %1, %2" : 
+    "=r" (*hi) : 
+    "r" (a), 
+    "r" (b));
+  return lo;
+}
+
+extern "C"
+{
+	void keccak(const uint8_t *in, int inlen, uint8_t *md, int mdlen);
+	void keccakf(uint64_t st[25], int rounds);
+	extern void(*const extra_hashes[4])(const void *, size_t, char *);
+   
+}
+
+static inline __m128i _mm_set_epi64x(uint64_t a, uint64_t b){
+  return (__m128ll){b,a};
+}
+// This will shift and xor tmp1 into itself as 4 32-bit vals such as
+// sl_xor(a1 a2 a3 a4) = a1 (a2^a1) (a3^a2^a1) (a4^a3^a2^a1)
+static inline __m128i sl_xor(__m128i tmp1)
+{
+  __m128i tmp4;
+  tmp4 = vec_slo(tmp1, (__m128i){0x20});
+  tmp1 = vec_xor(tmp1, tmp4);
+  tmp4 = vec_slo(tmp4, (__m128i){0x20});
+  tmp1 = vec_xor(tmp1, tmp4);
+  tmp4 = vec_slo(tmp4, (__m128i){0x20});
+  tmp1 = vec_xor(tmp1, tmp4);
+  return tmp1;
+}
+
+static inline __m128i sl_xor_be(__m128i tmp1)
+{
+  __m128i tmp4;
+  tmp4 = vec_sro(tmp1, (__m128i){0x20});
+  tmp1 = vec_xor(tmp1, tmp4);
+  tmp4 = vec_sro(tmp4, (__m128i){0x20});
+  tmp1 = vec_xor(tmp1, tmp4);
+  tmp4 = vec_sro(tmp4, (__m128i){0x20});
+  tmp1 = vec_xor(tmp1, tmp4);
+  return tmp1;
+}
+static inline __m128i v_rev(const __m128i& tmp1)
+{
+  return(vec_perm(tmp1,tmp1,(__m128i){ 0xf,0xe,0xd,0xc,0xb,0xa,0x9,0x8,0x7,0x6,0x5,0x4,0x3,0x2,0x1,0x0 })); 
+}
+
+
+static inline __m128i _mm_aesenc_si128(__m128i in, __m128i key)
+{
+  return v_rev(__builtin_crypto_vcipher(v_rev(in),v_rev(key)));
+}
+
+static inline __m128i _mm_aesenc_si128_beIN(__m128i in, __m128i key)
+{
+  return v_rev(__builtin_crypto_vcipher(in,v_rev(key)));
+}
+
+static inline __m128i _mm_aesenc_si128_beK(__m128i in, __m128i key)
+{
+  return v_rev(__builtin_crypto_vcipher(v_rev(in),key));
+}
+static inline __m128i _mm_aesenc_si128_be(__m128i in, __m128i key)
+{
+  return __builtin_crypto_vcipher(in,key);
+}
+
+
+template<uint8_t rcon>
+static inline void aes_genkey_sub(__m128i* xout0, __m128i* xout2)
+{
+  __m128i xout1 = soft_aeskeygenassist(*xout2, rcon);
+  xout1 = vec_perm(xout1,xout1,(__m128i){0xc,0xd,0xe,0xf, 0xc,0xd,0xe,0xf, 0xc,0xd,0xe,0xf, 0xc,0xd,0xe,0xf}); 
+  *xout0 = sl_xor(*xout0);
+  *xout0 = vec_xor(*xout0, xout1);
+  xout1 = soft_aeskeygenassist(*xout0, 0x00);
+  xout1 = vec_perm(xout1,xout1,(__m128i){0x8,0x9,0xa,0xb, 0x8,0x9,0xa,0xb, 0x8,0x9,0xa,0xb, 0x8,0x9,0xa,0xb});
+  *xout2 = sl_xor(*xout2);
+  *xout2 = vec_xor(*xout2, xout1);
+}
+
+template<uint8_t rcon>
+static inline void aes_genkey_sub_be(__m128i* xout0, __m128i* xout2)
+{
+  __m128i xout1 = soft_aeskeygenassist_be(*xout2, rcon);
+  xout1 = vec_perm(xout1,xout1,(__m128i){0x0,0x1,0x2,0x3, 0x0,0x1,0x2,0x3, 0x0,0x1,0x2,0x3, 0x0,0x1,0x2,0x3}); 
+  *xout0 = sl_xor_be(*xout0);
+  *xout0 = vec_xor(*xout0, xout1);
+  xout1 = soft_aeskeygenassist_be(*xout0, 0x00);
+  xout1 = vec_perm(xout1,xout1,(__m128i){0x4,0x5,0x6,0x7, 0x4,0x5,0x6,0x7, 0x4,0x5,0x6,0x7, 0x4,0x5,0x6,0x7});
+  *xout2 = sl_xor_be(*xout2);
+  *xout2 = vec_xor(*xout2, xout1);
+}
+
+
+template<bool SOFT_AES>
+static inline void aes_genkey(const __m128i* memory, __m128i* k0, __m128i* k1, __m128i* k2, __m128i* k3,
+	__m128i* k4, __m128i* k5, __m128i* k6, __m128i* k7, __m128i* k8, __m128i* k9)
+{
+	__m128i xout0, xout2;
+
+	xout0 = vec_ld(0,memory);
+	xout2 = vec_ld(16,memory);
+	*k0 = xout0;
+	*k1 = xout2;
+
+		aes_genkey_sub<0x01>(&xout0, &xout2);
+	*k2 = xout0;
+	*k3 = xout2;
+
+		aes_genkey_sub<0x02>(&xout0, &xout2);
+	*k4 = xout0;
+	*k5 = xout2;
+
+		aes_genkey_sub<0x04>(&xout0, &xout2);
+	*k6 = xout0;
+	*k7 = xout2;
+
+		aes_genkey_sub<0x08>(&xout0, &xout2);
+	*k8 = xout0;
+	*k9 = xout2;
+}
+
+template<bool SOFT_AES>
+static inline void aes_genkey_be(const __m128i* memory, __m128i* k0, __m128i* k1, __m128i* k2, __m128i* k3,
+	__m128i* k4, __m128i* k5, __m128i* k6, __m128i* k7, __m128i* k8, __m128i* k9)
+{
+	__m128i xout0, xout2;
+
+	xout0 = v_rev(vec_ld(0,memory));
+	xout2 = v_rev(vec_ld(16,memory));
+	*k0 = xout0;
+	*k1 = xout2;
+
+		aes_genkey_sub_be<0x01>(&xout0, &xout2);
+	*k2 = xout0;
+	*k3 = xout2;
+
+		aes_genkey_sub_be<0x02>(&xout0, &xout2);
+	*k4 = xout0;
+	*k5 = xout2;
+
+		aes_genkey_sub_be<0x04>(&xout0, &xout2);
+	*k6 = xout0;
+	*k7 = xout2;
+
+		aes_genkey_sub_be<0x08>(&xout0, &xout2);
+	*k8 = xout0;
+	*k9 = xout2;
+}
+static inline void aes_round(__m128i key, __m128i* x0, __m128i* x1, __m128i* x2, __m128i* x3, __m128i* x4, __m128i* x5, __m128i* x6, __m128i* x7)
+{
+	*x0 = _mm_aesenc_si128(*x0, key);
+	*x1 = _mm_aesenc_si128(*x1, key);
+	*x2 = _mm_aesenc_si128(*x2, key);
+	*x3 = _mm_aesenc_si128(*x3, key);
+	*x4 = _mm_aesenc_si128(*x4, key);
+	*x5 = _mm_aesenc_si128(*x5, key);
+	*x6 = _mm_aesenc_si128(*x6, key);
+	*x7 = _mm_aesenc_si128(*x7, key);
+}
+
+static inline void aes_round_be(__m128i key, __m128i* x0, __m128i* x1, __m128i* x2, __m128i* x3, __m128i* x4, __m128i* x5, __m128i* x6, __m128i* x7)
+{
+ *x0 = _mm_aesenc_si128_be(*x0, key);
+ *x1 = _mm_aesenc_si128_be(*x1, key);
+ *x2 = _mm_aesenc_si128_be(*x2, key);
+ *x3 = _mm_aesenc_si128_be(*x3, key);
+ *x4 = _mm_aesenc_si128_be(*x4, key);
+ *x5 = _mm_aesenc_si128_be(*x5, key);
+ *x6 = _mm_aesenc_si128_be(*x6, key);
+ *x7 = _mm_aesenc_si128_be(*x7, key);
+
+}
+
+
+inline void mix_and_propagate(__m128i& x0, __m128i& x1, __m128i& x2, __m128i& x3, __m128i& x4, __m128i& x5, __m128i& x6, __m128i& x7)
+{
+    __m128i tmp0 = x0;
+    x0 = vec_xor(x0, x1);
+    x1 = vec_xor(x1, x2);
+    x2 = vec_xor(x2, x3);
+    x3 = vec_xor(x3, x4);
+    x4 = vec_xor(x4, x5);
+    x5 = vec_xor(x5, x6);
+    x6 = vec_xor(x6, x7);
+    x7 = vec_xor(x7, tmp0);
+}
+
+template<size_t MEM, bool SOFT_AES, bool BE_MODE, xmrstak_algo ALGO>
+void cn_explode_scratchpad(const __m128i* input, __m128i* output)
+{
+	// This is more than we have registers, compiler will assign 2 keys on the stack
+	__m128i xin0, xin1, xin2, xin3, xin4, xin5, xin6, xin7;
+	__m128i k0, k1, k2, k3, k4, k5, k6, k7, k8, k9;
+  aes_genkey<SOFT_AES>(input, &k0, &k1, &k2, &k3, &k4, &k5, &k6, &k7, &k8, &k9);
+  
+  xin0 = vec_ld(64,input);
+  xin1 = vec_ld(80,input);
+  xin2 = vec_ld(96,input);
+  xin3 = vec_ld(112,input);
+  xin4 = vec_ld(128,input);
+  xin5 = vec_ld(144,input);
+  xin6 = vec_ld(160,input);
+  xin7 = vec_ld(176,input);
+
+	if(ALGO == cryptonight_heavy)
+	{
+		for(size_t i=0; i < 16; i++)
+		{
+			aes_round(k0, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
+			aes_round(k1, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
+			aes_round(k2, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
+			aes_round(k3, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
+			aes_round(k4, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
+			aes_round(k5, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
+			aes_round(k6, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
+			aes_round(k7, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
+			aes_round(k8, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
+			aes_round(k9, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
+			mix_and_propagate(xin0, xin1, xin2, xin3, xin4, xin5, xin6, xin7);
+		}
+	}
+
+	for (size_t i = 0; i < MEM / sizeof(__m128i); i += 8)
+	{
+		aes_round(k0, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
+		aes_round(k1, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
+		aes_round(k2, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
+		aes_round(k3, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
+		aes_round(k4, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
+		aes_round(k5, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
+		aes_round(k6, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
+		aes_round(k7, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
+		aes_round(k8, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
+		aes_round(k9, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
+    vec_st(xin0,i*16,output);
+    vec_st(xin1,(i+1)*16,output);
+    vec_st(xin2,(i+2)*16,output);
+    vec_st(xin3,(i+3)*16,output);
+    vec_st(xin4,(i+4)*16,output);
+    vec_st(xin5,(i+5)*16,output);
+    vec_st(xin6,(i+6)*16,output);
+    vec_st(xin7,(i+7)*16,output);
+
+	}
+}
+template<size_t MEM, bool SOFT_AES, bool BE_MODE, xmrstak_algo ALGO>
+void cn_explode_scratchpad_be(const __m128i* input, __m128i* output)
+{
+	// This is more than we have registers, compiler will assign 2 keys on the stack
+	__m128i xin0, xin1, xin2, xin3, xin4, xin5, xin6, xin7;
+	__m128i k0, k1, k2, k3, k4, k5, k6, k7, k8, k9;
+  aes_genkey_be<SOFT_AES>(input, &k0, &k1, &k2, &k3, &k4, &k5, &k6, &k7, &k8, &k9);
+  
+  xin0 = vec_ld(64,input);
+  xin1 = vec_ld(80,input);
+  xin2 = vec_ld(96,input);
+  xin3 = vec_ld(112,input);
+  xin4 = vec_ld(128,input);
+  xin5 = vec_ld(144,input);
+  xin6 = vec_ld(160,input);
+  xin7 = vec_ld(176,input);
+  
+  xin0 = v_rev(xin0);
+  xin1 = v_rev(xin1);
+  xin2 = v_rev(xin2);
+  xin3 = v_rev(xin3);
+  xin4 = v_rev(xin4);
+  xin5 = v_rev(xin5);
+  xin6 = v_rev(xin6);
+  xin7 = v_rev(xin7);
+
+	if(ALGO == cryptonight_heavy)
+	{
+		for(size_t i=0; i < 16; i++)
+		{
+			aes_round_be(k0, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
+			aes_round_be(k1, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
+			aes_round_be(k2, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
+			aes_round_be(k3, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
+			aes_round_be(k4, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
+			aes_round_be(k5, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
+			aes_round_be(k6, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
+			aes_round_be(k7, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
+			aes_round_be(k8, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
+			aes_round_be(k9, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
+			mix_and_propagate(xin0, xin1, xin2, xin3, xin4, xin5, xin6, xin7);
+		}
+	}
+
+	for (size_t i = 0; i < MEM / sizeof(__m128i); i += 8)
+	{
+		aes_round_be(k0, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
+		aes_round_be(k1, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
+		aes_round_be(k2, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
+		aes_round_be(k3, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
+		aes_round_be(k4, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
+		aes_round_be(k5, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
+		aes_round_be(k6, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
+		aes_round_be(k7, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
+		aes_round_be(k8, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
+		aes_round_be(k9, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
+    vec_st(v_rev(xin0),i*16,output);
+    vec_st(v_rev(xin1),(i+1)*16,output);
+    vec_st(v_rev(xin2),(i+2)*16,output);
+    vec_st(v_rev(xin3),(i+3)*16,output);
+    vec_st(v_rev(xin4),(i+4)*16,output);
+    vec_st(v_rev(xin5),(i+5)*16,output);
+    vec_st(v_rev(xin6),(i+6)*16,output);
+    vec_st(v_rev(xin7),(i+7)*16,output);
+
+	}
+}
+
+template<size_t MEM, bool SOFT_AES, bool BE_MODE, xmrstak_algo ALGO>
+void cn_implode_scratchpad(const __m128i* input, __m128i* output)
+{
+	// This is more than we have registers, compiler will assign 2 keys on the stack
+	__m128i xout0, xout1, xout2, xout3, xout4, xout5, xout6, xout7;
+	__m128i k0, k1, k2, k3, k4, k5, k6, k7, k8, k9;
+
+	aes_genkey<SOFT_AES>(output + 2, &k0, &k1, &k2, &k3, &k4, &k5, &k6, &k7, &k8, &k9);
+
+
+  xout0 = vec_ld(64,output);
+  xout1 = vec_ld(80,output);
+  xout2 = vec_ld(96,output);
+  xout3 = vec_ld(112,output);
+  xout4 = vec_ld(128,output);
+  xout5 = vec_ld(144,output);
+  xout6 = vec_ld(160,output);
+  xout7 = vec_ld(176,output);
+
+	for (size_t i = 0; i < MEM / sizeof(__m128i); i += 8)
+	{
+
+    xout0 = vec_xor(vec_ld(i*16,input), xout0);
+    xout1 = vec_xor(vec_ld((i+1)*16,input), xout1);
+    xout2 = vec_xor(vec_ld((i+2)*16,input), xout2);
+    xout3 = vec_xor(vec_ld((i+3)*16,input), xout3);
+    xout4 = vec_xor(vec_ld((i+4)*16,input), xout4);
+    xout5 = vec_xor(vec_ld((i+5)*16,input), xout5);
+    xout6 = vec_xor(vec_ld((i+6)*16,input), xout6);
+    xout7 = vec_xor(vec_ld((i+7)*16,input), xout7);
+		aes_round(k0, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+		aes_round(k1, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+		aes_round(k2, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+		aes_round(k3, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+		aes_round(k4, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+		aes_round(k5, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+		aes_round(k6, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+		aes_round(k7, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+		aes_round(k8, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+		aes_round(k9, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+		if(ALGO == cryptonight_heavy)
+			mix_and_propagate(xout0, xout1, xout2, xout3, xout4, xout5, xout6, xout7);
+	}
+
+	if(ALGO == cryptonight_heavy)
+	{
+		for (size_t i = 0; i < MEM / sizeof(__m128i); i += 8)
+		{
+
+      xout0 = vec_xor(vec_ld(i*16,input), xout0);
+      xout1 = vec_xor(vec_ld((i+1)*16,input), xout1);
+      xout2 = vec_xor(vec_ld((i+2)*16,input), xout2);
+      xout3 = vec_xor(vec_ld((i+3)*16,input), xout3);
+      xout4 = vec_xor(vec_ld((i+4)*16,input), xout4);
+      xout5 = vec_xor(vec_ld((i+5)*16,input), xout5);
+      xout6 = vec_xor(vec_ld((i+6)*16,input), xout6);
+      xout7 = vec_xor(vec_ld((i+7)*16,input), xout7);
+
+      aes_round(k0, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+      aes_round(k1, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+			aes_round(k2, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+			aes_round(k3, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+			aes_round(k4, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+			aes_round(k5, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+			aes_round(k6, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+			aes_round(k7, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+			aes_round(k8, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+			aes_round(k9, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+
+      mix_and_propagate(xout0, xout1, xout2, xout3, xout4, xout5, xout6, xout7);
+		}
+
+		for(size_t i=0; i < 16; i++)
+		{
+      aes_round(k0, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+      aes_round(k1, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+      aes_round(k2, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+      aes_round(k3, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+      aes_round(k4, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+      aes_round(k5, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+      aes_round(k6, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+      aes_round(k7, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+      aes_round(k8, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+      aes_round(k9, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+
+			mix_and_propagate(xout0, xout1, xout2, xout3, xout4, xout5, xout6, xout7);
+		}
+	}
+
+  vec_st(xout0,64,output);
+  vec_st(xout1,80,output);
+  vec_st(xout2,96,output);
+  vec_st(xout3,112,output);
+  vec_st(xout4,128,output);
+  vec_st(xout5,144,output);
+  vec_st(xout6,160,output);
+  vec_st(xout7,176,output);
+}
+
+template<size_t MEM, bool SOFT_AES, bool BE_MODE, xmrstak_algo ALGO>
+void cn_implode_scratchpad_be(const __m128i* input, __m128i* output)
+{
+	// This is more than we have registers, compiler will assign 2 keys on the stack
+	__m128i xout0, xout1, xout2, xout3, xout4, xout5, xout6, xout7;
+	__m128i k0, k1, k2, k3, k4, k5, k6, k7, k8, k9;
+
+	aes_genkey_be<SOFT_AES>(output + 2, &k0, &k1, &k2, &k3, &k4, &k5, &k6, &k7, &k8, &k9);
+
+  xout0 = vec_ld(64,output);
+  xout1 = vec_ld(80,output);
+  xout2 = vec_ld(96,output);
+  xout3 = vec_ld(112,output);
+  xout4 = vec_ld(128,output);
+  xout5 = vec_ld(144,output);
+  xout6 = vec_ld(160,output);
+  xout7 = vec_ld(176,output);
+  
+  xout0 = v_rev(xout0);
+  xout1 = v_rev(xout1);
+  xout2 = v_rev(xout2);
+  xout3 = v_rev(xout3);
+  xout4 = v_rev(xout4);
+  xout5 = v_rev(xout5);
+  xout6 = v_rev(xout6);
+  xout7 = v_rev(xout7);
+
+	for (size_t i = 0; i < MEM / sizeof(__m128i); i += 8)
+	{
+
+    xout0 = vec_xor(v_rev(vec_ld(i*16,input)), xout0);
+    xout1 = vec_xor(v_rev(vec_ld((i+1)*16,input)), xout1);
+    xout2 = vec_xor(v_rev(vec_ld((i+2)*16,input)), xout2);
+    xout3 = vec_xor(v_rev(vec_ld((i+3)*16,input)), xout3);
+    xout4 = vec_xor(v_rev(vec_ld((i+4)*16,input)), xout4);
+    xout5 = vec_xor(v_rev(vec_ld((i+5)*16,input)), xout5);
+    xout6 = vec_xor(v_rev(vec_ld((i+6)*16,input)), xout6);
+    xout7 = vec_xor(v_rev(vec_ld((i+7)*16,input)), xout7);
+		aes_round_be(k0, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+		aes_round_be(k1, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+		aes_round_be(k2, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+		aes_round_be(k3, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+		aes_round_be(k4, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+		aes_round_be(k5, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+		aes_round_be(k6, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+		aes_round_be(k7, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+		aes_round_be(k8, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+		aes_round_be(k9, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+		if(ALGO == cryptonight_heavy)
+			mix_and_propagate(xout0, xout1, xout2, xout3, xout4, xout5, xout6, xout7);
+	}
+
+	if(ALGO == cryptonight_heavy)
+	{
+		for (size_t i = 0; i < MEM / sizeof(__m128i); i += 8)
+		{
+
+      xout0 = vec_xor(v_rev(vec_ld(i*16,input)), xout0);
+      xout1 = vec_xor(v_rev(vec_ld((i+1)*16,input)), xout1);
+      xout2 = vec_xor(v_rev(vec_ld((i+2)*16,input)), xout2);
+      xout3 = vec_xor(v_rev(vec_ld((i+3)*16,input)), xout3);
+      xout4 = vec_xor(v_rev(vec_ld((i+4)*16,input)), xout4);
+      xout5 = vec_xor(v_rev(vec_ld((i+5)*16,input)), xout5);
+      xout6 = vec_xor(v_rev(vec_ld((i+6)*16,input)), xout6);
+      xout7 = vec_xor(v_rev(vec_ld((i+7)*16,input)), xout7);
+
+      aes_round_be(k0, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+      aes_round_be(k1, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+			aes_round_be(k2, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+			aes_round_be(k3, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+			aes_round_be(k4, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+			aes_round_be(k5, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+			aes_round_be(k6, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+			aes_round_be(k7, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+			aes_round_be(k8, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+			aes_round_be(k9, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+
+      mix_and_propagate(xout0, xout1, xout2, xout3, xout4, xout5, xout6, xout7);
+		}
+
+		for(size_t i=0; i < 16; i++)
+		{
+      aes_round_be(k0, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+      aes_round_be(k1, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+      aes_round_be(k2, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+      aes_round_be(k3, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+      aes_round_be(k4, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+      aes_round_be(k5, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+      aes_round_be(k6, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+      aes_round_be(k7, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+      aes_round_be(k8, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+      aes_round_be(k9, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
+
+			mix_and_propagate(xout0, xout1, xout2, xout3, xout4, xout5, xout6, xout7);
+		}
+	}
+
+  vec_st(v_rev(xout0),64,output);
+  vec_st(v_rev(xout1),80,output);
+  vec_st(v_rev(xout2),96,output);
+  vec_st(v_rev(xout3),112,output);
+  vec_st(v_rev(xout4),128,output);
+  vec_st(v_rev(xout5),144,output);
+  vec_st(v_rev(xout6),160,output);
+  vec_st(v_rev(xout7),176,output);
+}
+inline void cryptonight_monero_tweak(uint64_t* mem_out, __m128i tmp)
+{
+  mem_out[0] = ((uint64_t*)&tmp)[0];
+  tmp = vec_perm(tmp,tmp,(__m128i){0x8,0x9,0xa,0xb, 0xc,0xd,0xe,0xf, 0x8,0x9,0xa,0xb, 0xc,0xd,0xe,0xf});
+	uint64_t vh = ((uint64_t*)&tmp)[0];
+	uint8_t x = vh >> 24;
+	static const uint16_t table = 0x7531;
+	const uint8_t index = (((x >> 3) & 6) | (x & 1)) << 1;
+	vh ^= ((table >> index) & 0x3) << 28;
+	mem_out[1] = vh;
+}
+
+template<xmrstak_algo ALGO, bool SOFT_AES, bool BE_MODE>
+void cryptonight_hash(const void* input, size_t len, void* output, cryptonight_ctx* ctx0)
+{
+  constexpr size_t MASK = cn_select_mask<ALGO>();
+	constexpr size_t ITERATIONS = cn_select_iter<ALGO>();
+	constexpr size_t MEM = cn_select_memory<ALGO>();
+
+	if(ALGO == cryptonight_monero && len < 43)
+	{
+		memset(output, 0, 32);
+		return;
+	}
+
+	keccak((const uint8_t *)input, len, ctx0->hash_state, 200);
+
+	uint64_t monero_const;
+	if(ALGO == cryptonight_monero)
+	{
+		monero_const  =  *reinterpret_cast<const uint64_t*>(reinterpret_cast<const uint8_t*>(input) + 35);
+		monero_const ^=  *(reinterpret_cast<const uint64_t*>(ctx0->hash_state) + 24);
+	}
+
+	// Optim - 99% time boundary
+	if(BE_MODE) cn_explode_scratchpad_be<MEM, SOFT_AES, BE_MODE, ALGO>((__m128i*)ctx0->hash_state, (__m128i*)ctx0->long_state);
+  else        cn_explode_scratchpad<MEM, SOFT_AES, BE_MODE, ALGO>((__m128i*)ctx0->hash_state, (__m128i*)ctx0->long_state);
+	uint8_t* l0 = ctx0->long_state;
+	uint64_t* h0 = (uint64_t*)ctx0->hash_state;
+
+	uint64_t al0 = h0[0] ^ h0[4];
+	uint64_t ah0 = h0[1] ^ h0[5];
+	__m128i bx0 = (__m128ll){h0[2] ^ h0[6],h0[3] ^ h0[7]};
+
+	uint64_t idx0 = al0;
+
+	// Optim - 90% time boundary
+	for(size_t i = 0; i < ITERATIONS; i++)
+	{
+		__m128i cx;
+		cx = vec_vsx_ld(0,(__m128i *)&l0[idx0 & MASK]);
+		cx = _mm_aesenc_si128(cx, (__m128ll){al0, ah0});
+
+		if(ALGO == cryptonight_monero)
+			cryptonight_monero_tweak((uint64_t*)&l0[idx0 & MASK], vec_xor(bx0, cx));
+		else
+			vec_vsx_st(vec_xor(bx0, cx),0,(__m128i *)&l0[idx0 & MASK]);
+		
+    idx0 = ((uint64_t*)&cx)[0];
+
+		bx0 = cx;
+
+		uint64_t hi, lo, cl, ch;
+		cl = ((uint64_t*)&l0[idx0 & MASK])[0];
+		ch = ((uint64_t*)&l0[idx0 & MASK])[1];
+
+		lo = _umul128(idx0, cl, &hi);
+
+		al0 += hi;
+		((uint64_t*)&l0[idx0 & MASK])[0] = al0;
+		al0 ^= cl;
+		ah0 += lo;
+
+		if(ALGO == cryptonight_monero)
+			((uint64_t*)&l0[idx0 & MASK])[1] = ah0 ^ monero_const;
+		else
+			((uint64_t*)&l0[idx0 & MASK])[1] = ah0;
+		ah0 ^= ch;
+
+		idx0 = al0;
+
+		if(ALGO == cryptonight_heavy)
+		{
+			int64_t n  = ((int64_t*)&l0[idx0 & MASK])[0];
+			int32_t d  = ((int32_t*)&l0[idx0 & MASK])[2];
+			int64_t q = n / (d | 0x5);
+
+			((int64_t*)&l0[idx0 & MASK])[0] = n ^ q;
+			idx0 = d ^ q;
+		}
+	}
+
+	// Optim - 90% time boundary
+	if(BE_MODE) cn_implode_scratchpad_be<MEM, SOFT_AES, BE_MODE, ALGO>((__m128i*)ctx0->long_state, (__m128i*)ctx0->hash_state);
+	else        cn_implode_scratchpad<MEM, SOFT_AES, BE_MODE, ALGO>((__m128i*)ctx0->long_state, (__m128i*)ctx0->hash_state);
+
+	// Optim - 99% time boundary
+
+	keccakf((uint64_t*)ctx0->hash_state, 24);
+	extra_hashes[ctx0->hash_state[0] & 3](ctx0->hash_state, 200, (char*)output);
+}
+
+// This lovely creation will do 2 cn hashes at a time. We have plenty of space on silicon
+// to fit temporary vars for two contexts. Function will read len*2 from input and write 64 bytes to output
+// We are still limited by L3 cache, so doubling will only work with CPUs where we have more than 2MB to core (Xeons)
+template<xmrstak_algo ALGO, bool SOFT_AES, bool BE_MODE>
+void cryptonight_double_hash(const void* input, size_t len, void* output, cryptonight_ctx** ctx)
+{
+	constexpr size_t MASK = cn_select_mask<ALGO>();
+	constexpr size_t ITERATIONS = cn_select_iter<ALGO>();
+	constexpr size_t MEM = cn_select_memory<ALGO>();
+
+	if(ALGO == cryptonight_monero && len < 43)
+	{
+		memset(output, 0, 64);
+		return;
+	}
+
+	keccak((const uint8_t *)input, len, ctx[0]->hash_state, 200);
+	keccak((const uint8_t *)input+len, len, ctx[1]->hash_state, 200);
+
+	uint64_t monero_const_0, monero_const_1;
+	if(ALGO == cryptonight_monero)
+	{
+		monero_const_0  =  *reinterpret_cast<const uint64_t*>(reinterpret_cast<const uint8_t*>(input) + 35);
+		monero_const_0 ^=  *(reinterpret_cast<const uint64_t*>(ctx[0]->hash_state) + 24);
+		monero_const_1  =  *reinterpret_cast<const uint64_t*>(reinterpret_cast<const uint8_t*>(input) + len + 35);
+		monero_const_1 ^=  *(reinterpret_cast<const uint64_t*>(ctx[1]->hash_state) + 24);
+	}
+
+	// Optim - 99% time boundary
+	if(BE_MODE){
+    cn_explode_scratchpad_be<MEM, SOFT_AES, BE_MODE, ALGO>((__m128i*)ctx[0]->hash_state, (__m128i*)ctx[0]->long_state);
+	  cn_explode_scratchpad_be<MEM, SOFT_AES, BE_MODE, ALGO>((__m128i*)ctx[1]->hash_state, (__m128i*)ctx[1]->long_state);}
+  else{
+    cn_explode_scratchpad<MEM, SOFT_AES, BE_MODE, ALGO>((__m128i*)ctx[0]->hash_state, (__m128i*)ctx[0]->long_state);
+	  cn_explode_scratchpad<MEM, SOFT_AES, BE_MODE, ALGO>((__m128i*)ctx[1]->hash_state, (__m128i*)ctx[1]->long_state);}
+
+
+	uint8_t* l0 = ctx[0]->long_state;
+	uint64_t* h0 = (uint64_t*)ctx[0]->hash_state;
+	uint8_t* l1 = ctx[1]->long_state;
+	uint64_t* h1 = (uint64_t*)ctx[1]->hash_state;
+
+	uint64_t axl0 = h0[0] ^ h0[4];
+	uint64_t axh0 = h0[1] ^ h0[5];
+	__m128i bx0 = (__m128ll){h0[2] ^ h0[6],h0[3] ^ h0[7]};
+	uint64_t axl1 = h1[0] ^ h1[4];
+	uint64_t axh1 = h1[1] ^ h1[5];
+	__m128i bx1 = (__m128ll){h1[2] ^ h1[6],h1[3] ^ h1[7]};
+
+	uint64_t idx0 = h0[0] ^ h0[4];
+	uint64_t idx1 = h1[0] ^ h1[4];
+
+	// Optim - 90% time boundary
+	for (size_t i = 0; i < ITERATIONS; i++)
+	{
+		__m128i cx;
+		cx = vec_vsx_ld(0,(__m128i *)&l0[idx0 & MASK]);
+		cx = _mm_aesenc_si128(cx, (__m128ll){axl0, axh0});
+
+		if(ALGO == cryptonight_monero)
+			cryptonight_monero_tweak((uint64_t*)&l0[idx0 & MASK], vec_xor(bx0, cx));
+		else
+			vec_vsx_st(vec_xor(bx0, cx),0,(__m128i *)&l0[idx0 & MASK]);
+
+		idx0 = ((uint64_t*)&cx)[0];
+		bx0 = cx;
+
+		cx = vec_vsx_ld(0,(__m128i *)&l1[idx1 & MASK]);
+		cx = _mm_aesenc_si128(cx, (__m128ll){axl1, axh1});
+
+		if(ALGO == cryptonight_monero)
+			cryptonight_monero_tweak((uint64_t*)&l1[idx1 & MASK], vec_xor(bx1, cx));
+		else
+		  vec_vsx_st(vec_xor(bx1, cx),0,(__m128i *)&l1[idx1 & MASK]);
+
+		idx1 = ((uint64_t*)&cx)[0];
+		bx1 = cx;
+
+
+		uint64_t hi, lo, cl, ch;
+		cl = ((uint64_t*)&l0[idx0 & MASK])[0];
+		ch = ((uint64_t*)&l0[idx0 & MASK])[1];
+
+		lo = _umul128(idx0, cl, &hi);
+
+		axl0 += hi;
+		axh0 += lo;
+		((uint64_t*)&l0[idx0 & MASK])[0] = axl0;
+
+		if(ALGO == cryptonight_monero)
+			((uint64_t*)&l0[idx0 & MASK])[1] = axh0 ^ monero_const_0;
+		else
+			((uint64_t*)&l0[idx0 & MASK])[1] = axh0;
+
+		axh0 ^= ch;
+		axl0 ^= cl;
+		idx0 = axl0;
+
+		if(ALGO == cryptonight_heavy)
+		{
+			int64_t n  = ((int64_t*)&l0[idx0 & MASK])[0];
+			int32_t d  = ((int32_t*)&l0[idx0 & MASK])[2];
+			int64_t q = n / (d | 0x5);
+
+			((int64_t*)&l0[idx0 & MASK])[0] = n ^ q;
+			idx0 = d ^ q;
+		}
+
+
+		cl = ((uint64_t*)&l1[idx1 & MASK])[0];
+		ch = ((uint64_t*)&l1[idx1 & MASK])[1];
+
+		lo = _umul128(idx1, cl, &hi);
+
+		axl1 += hi;
+		axh1 += lo;
+		((uint64_t*)&l1[idx1 & MASK])[0] = axl1;
+
+		if(ALGO == cryptonight_monero)
+			((uint64_t*)&l1[idx1 & MASK])[1] = axh1 ^ monero_const_1;
+		else
+			((uint64_t*)&l1[idx1 & MASK])[1] = axh1;
+
+		axh1 ^= ch;
+		axl1 ^= cl;
+		idx1 = axl1;
+
+		if(ALGO == cryptonight_heavy)
+		{
+			int64_t n  = ((int64_t*)&l1[idx1 & MASK])[0];
+			int32_t d  = ((int32_t*)&l1[idx1 & MASK])[2];
+			int64_t q = n / (d | 0x5);
+
+			((int64_t*)&l1[idx1 & MASK])[0] = n ^ q;
+			idx1 = d ^ q;
+		}
+
+	}
+
+	// Optim - 90% time boundary
+	if(BE_MODE){
+    cn_implode_scratchpad_be<MEM, SOFT_AES, BE_MODE, ALGO>((__m128i*)ctx[0]->long_state, (__m128i*)ctx[0]->hash_state);
+	  cn_implode_scratchpad_be<MEM, SOFT_AES, BE_MODE, ALGO>((__m128i*)ctx[1]->long_state, (__m128i*)ctx[1]->hash_state);}
+  else{
+    cn_implode_scratchpad<MEM, SOFT_AES, BE_MODE, ALGO>((__m128i*)ctx[0]->long_state, (__m128i*)ctx[0]->hash_state);
+  	cn_implode_scratchpad<MEM, SOFT_AES, BE_MODE, ALGO>((__m128i*)ctx[1]->long_state, (__m128i*)ctx[1]->hash_state);}
+
+	// Optim - 99% time boundary
+
+	keccakf((uint64_t*)ctx[0]->hash_state, 24);
+	extra_hashes[ctx[0]->hash_state[0] & 3](ctx[0]->hash_state, 200, (char*)output);
+	keccakf((uint64_t*)ctx[1]->hash_state, 24);
+	extra_hashes[ctx[1]->hash_state[0] & 3](ctx[1]->hash_state, 200, (char*)output + 32);
+}
+
+#define CN_STEP1(a, b, c, l, ptr, idx)				\
+	ptr = (__m128i *)&l[idx & MASK];			\
+	c = vec_vsx_ld(0,ptr); 
+
+#define CN_STEP2(a, b, c, l, ptr, idx)				\
+		c = _mm_aesenc_si128(c, a);			\
+	b = vec_xor(b, c);				\
+	if(ALGO == cryptonight_monero) \
+		cryptonight_monero_tweak((uint64_t*)ptr, b); \
+	else \
+		vec_vsx_st(b,0,ptr);\
+
+#define CN_STEP3(a, b, c, l, ptr, idx)				\
+	idx = ((uint64_t*)&c)[0];				\
+	ptr = (__m128i*)&l[idx & MASK];			\
+	b = vec_vsx_ld(0,ptr);
+
+#define CN_STEP4(a, b, c, l, mc, ptr, idx)				\
+	lo = _umul128(idx, ((uint64_t*)&b)[0], &hi);		\
+	a = (__m128ll)a + (__m128ll){hi, lo};		\
+	if(ALGO == cryptonight_monero) \
+		vec_vsx_st(vec_xor(a, mc),0,ptr); \
+	else \
+		vec_vsx_st(a,0,ptr);\
+	a = vec_xor(a, b); \
+	idx = ((uint64_t*)&a)[0];	\
+	if(ALGO == cryptonight_heavy) \
+	{ \
+		int64_t n  = ((int64_t*)&l[idx & MASK])[0]; \
+		int32_t d  = ((int32_t*)&l[idx & MASK])[2]; \
+		int64_t q = n / (d | 0x5); \
+		((int64_t*)&l[idx & MASK])[0] = n ^ q; \
+		idx = d ^ q; \
+	}
+
+#define CONST_INIT(ctx, n) \
+  __m128i mc##n = _mm_set_epi64x(*reinterpret_cast<const uint64_t*>(reinterpret_cast<const uint8_t*>(input) + n * len + 35) ^ \
+    *(reinterpret_cast<const uint64_t*>((ctx)->hash_state) + 24), 0);
+// This lovelier creation will do 3 cn hashes at a time.
+template<xmrstak_algo ALGO, bool SOFT_AES, bool BE_MODE>
+void cryptonight_triple_hash(const void* input, size_t len, void* output, cryptonight_ctx** ctx)
+{
+	constexpr size_t MASK = cn_select_mask<ALGO>();
+	constexpr size_t ITERATIONS = cn_select_iter<ALGO>();
+	constexpr size_t MEM = cn_select_memory<ALGO>();
+
+	if(ALGO == cryptonight_monero && len < 43)
+	{
+		memset(output, 0, 32 * 3);
+		return;
+	}
+
+	for (size_t i = 0; i < 3; i++)
+	{
+		keccak((const uint8_t *)input + len * i, len, ctx[i]->hash_state, 200);
+		if(BE_MODE) cn_explode_scratchpad_be<MEM, SOFT_AES, BE_MODE, ALGO>((__m128i*)ctx[i]->hash_state, (__m128i*)ctx[i]->long_state);
+    else        cn_explode_scratchpad<MEM, SOFT_AES, BE_MODE, ALGO>((__m128i*)ctx[i]->hash_state, (__m128i*)ctx[i]->long_state);
+
+	}
+
+	CONST_INIT(ctx[0], 0);
+	CONST_INIT(ctx[1], 1);
+	CONST_INIT(ctx[2], 2);
+
+	uint8_t* l0 = ctx[0]->long_state;
+	uint64_t* h0 = (uint64_t*)ctx[0]->hash_state;
+	uint8_t* l1 = ctx[1]->long_state;
+	uint64_t* h1 = (uint64_t*)ctx[1]->hash_state;
+	uint8_t* l2 = ctx[2]->long_state;
+	uint64_t* h2 = (uint64_t*)ctx[2]->hash_state;
+
+	__m128i ax0 = (__m128ll){h0[0] ^ h0[4], h0[1] ^ h0[5]};
+	__m128i bx0 = (__m128ll){h0[2] ^ h0[6], h0[3] ^ h0[7]};
+	__m128i ax1 = (__m128ll){h1[0] ^ h1[4], h1[1] ^ h1[5]};
+	__m128i bx1 = (__m128ll){h1[2] ^ h1[6], h1[3] ^ h1[7]};
+	__m128i ax2 = (__m128ll){h2[0] ^ h2[4], h2[1] ^ h2[5]};
+	__m128i bx2 = (__m128ll){h2[2] ^ h2[6], h2[3] ^ h2[7]};
+	__m128i cx0 = (__m128ll){0, 0};
+	__m128i cx1 = (__m128ll){0, 0};
+	__m128i cx2 = (__m128ll){0, 0};
+
+	uint64_t idx0, idx1, idx2;
+	idx0 = ((uint64_t*)&ax0)[0];
+	idx1 = ((uint64_t*)&ax1)[0];
+	idx2 = ((uint64_t*)&ax2)[0];
+
+	for (size_t i = 0; i < ITERATIONS/2; i++)
+	{
+		uint64_t hi, lo;
+		__m128i *ptr0, *ptr1, *ptr2;
+
+		// EVEN ROUND
+		CN_STEP1(ax0, bx0, cx0, l0, ptr0, idx0);
+		CN_STEP1(ax1, bx1, cx1, l1, ptr1, idx1);
+		CN_STEP1(ax2, bx2, cx2, l2, ptr2, idx2);
+
+		CN_STEP2(ax0, bx0, cx0, l0, ptr0, idx0);
+		CN_STEP2(ax1, bx1, cx1, l1, ptr1, idx1);
+		CN_STEP2(ax2, bx2, cx2, l2, ptr2, idx2);
+
+		CN_STEP3(ax0, bx0, cx0, l0, ptr0, idx0);
+		CN_STEP3(ax1, bx1, cx1, l1, ptr1, idx1);
+		CN_STEP3(ax2, bx2, cx2, l2, ptr2, idx2);
+
+		CN_STEP4(ax0, bx0, cx0, l0, mc0, ptr0, idx0);
+		CN_STEP4(ax1, bx1, cx1, l1, mc1, ptr1, idx1);
+		CN_STEP4(ax2, bx2, cx2, l2, mc2, ptr2, idx2);
+
+		// ODD ROUND
+		CN_STEP1(ax0, cx0, bx0, l0, ptr0, idx0);
+		CN_STEP1(ax1, cx1, bx1, l1, ptr1, idx1);
+		CN_STEP1(ax2, cx2, bx2, l2, ptr2, idx2);
+
+		CN_STEP2(ax0, cx0, bx0, l0, ptr0, idx0);
+		CN_STEP2(ax1, cx1, bx1, l1, ptr1, idx1);
+		CN_STEP2(ax2, cx2, bx2, l2, ptr2, idx2);
+
+		CN_STEP3(ax0, cx0, bx0, l0, ptr0, idx0);
+		CN_STEP3(ax1, cx1, bx1, l1, ptr1, idx1);
+		CN_STEP3(ax2, cx2, bx2, l2, ptr2, idx2);
+
+		CN_STEP4(ax0, cx0, bx0, l0, mc0, ptr0, idx0);
+		CN_STEP4(ax1, cx1, bx1, l1, mc1, ptr1, idx1);
+		CN_STEP4(ax2, cx2, bx2, l2, mc2, ptr2, idx2);
+	}
+
+	for (size_t i = 0; i < 3; i++)
+	{
+		if(BE_MODE) cn_implode_scratchpad_be<MEM, SOFT_AES, BE_MODE, ALGO>((__m128i*)ctx[i]->long_state, (__m128i*)ctx[i]->hash_state);
+    else        cn_implode_scratchpad<MEM, SOFT_AES, BE_MODE, ALGO>((__m128i*)ctx[i]->long_state, (__m128i*)ctx[i]->hash_state);
+		keccakf((uint64_t*)ctx[i]->hash_state, 24);
+		extra_hashes[ctx[i]->hash_state[0] & 3](ctx[i]->hash_state, 200, (char*)output + 32 * i);
+	}
+}
+
+// This even lovelier creation will do 4 cn hashes at a time.
+template<xmrstak_algo ALGO, bool SOFT_AES, bool BE_MODE>
+void cryptonight_quad_hash(const void* input, size_t len, void* output, cryptonight_ctx** ctx)
+{
+	constexpr size_t MASK = cn_select_mask<ALGO>();
+	constexpr size_t ITERATIONS = cn_select_iter<ALGO>();
+	constexpr size_t MEM = cn_select_memory<ALGO>();
+
+	if(ALGO == cryptonight_monero && len < 43)
+	{
+		memset(output, 0, 32 * 4);
+		return;
+	}
+
+	for (size_t i = 0; i < 4; i++)
+	{
+		keccak((const uint8_t *)input + len * i, len, ctx[i]->hash_state, 200);
+		if(BE_MODE) cn_explode_scratchpad_be<MEM, SOFT_AES, BE_MODE, ALGO>((__m128i*)ctx[i]->hash_state, (__m128i*)ctx[i]->long_state);
+    else         cn_explode_scratchpad<MEM, SOFT_AES, BE_MODE, ALGO>((__m128i*)ctx[i]->hash_state, (__m128i*)ctx[i]->long_state);
+	}
+
+	CONST_INIT(ctx[0], 0);
+	CONST_INIT(ctx[1], 1);
+	CONST_INIT(ctx[2], 2);
+	CONST_INIT(ctx[3], 3);
+
+	uint8_t* l0 = ctx[0]->long_state;
+	uint64_t* h0 = (uint64_t*)ctx[0]->hash_state;
+	uint8_t* l1 = ctx[1]->long_state;
+	uint64_t* h1 = (uint64_t*)ctx[1]->hash_state;
+	uint8_t* l2 = ctx[2]->long_state;
+	uint64_t* h2 = (uint64_t*)ctx[2]->hash_state;
+	uint8_t* l3 = ctx[3]->long_state;
+	uint64_t* h3 = (uint64_t*)ctx[3]->hash_state;
+
+	__m128i ax0 = (__m128ll){h0[0] ^ h0[4], h0[1] ^ h0[5]};
+	__m128i bx0 = (__m128ll){h0[2] ^ h0[6], h0[3] ^ h0[7]};
+	__m128i ax1 = (__m128ll){h1[0] ^ h1[4], h1[1] ^ h1[5]};
+	__m128i bx1 = (__m128ll){h1[2] ^ h1[6], h1[3] ^ h1[7]};
+	__m128i ax2 = (__m128ll){h2[0] ^ h2[4], h2[1] ^ h2[5]};
+	__m128i bx2 = (__m128ll){h2[2] ^ h2[6], h2[3] ^ h2[7]};
+	__m128i ax3 = (__m128ll){h3[0] ^ h3[4], h3[1] ^ h3[5]};
+	__m128i bx3 = (__m128ll){h3[2] ^ h3[6], h3[3] ^ h3[7]};
+	__m128i cx0 = (__m128ll){0, 0};
+	__m128i cx1 = (__m128ll){0, 0};
+	__m128i cx2 = (__m128ll){0, 0};
+	__m128i cx3 = (__m128ll){0, 0};
+	
+	uint64_t idx0, idx1, idx2, idx3;
+	idx0 = ((uint64_t*)&ax0)[0];
+	idx1 = ((uint64_t*)&ax1)[0];
+	idx2 = ((uint64_t*)&ax2)[0];
+	idx3 = ((uint64_t*)&ax3)[0];
+	
+	for (size_t i = 0; i < ITERATIONS/2; i++)
+	{
+		uint64_t hi, lo;
+		__m128i *ptr0, *ptr1, *ptr2, *ptr3;
+
+		// EVEN ROUND
+		CN_STEP1(ax0, bx0, cx0, l0, ptr0, idx0);
+		CN_STEP1(ax1, bx1, cx1, l1, ptr1, idx1);
+		CN_STEP1(ax2, bx2, cx2, l2, ptr2, idx2);
+		CN_STEP1(ax3, bx3, cx3, l3, ptr3, idx3);
+
+		CN_STEP2(ax0, bx0, cx0, l0, ptr0, idx0);
+		CN_STEP2(ax1, bx1, cx1, l1, ptr1, idx1);
+		CN_STEP2(ax2, bx2, cx2, l2, ptr2, idx2);
+		CN_STEP2(ax3, bx3, cx3, l3, ptr3, idx3);
+
+		CN_STEP3(ax0, bx0, cx0, l0, ptr0, idx0);
+		CN_STEP3(ax1, bx1, cx1, l1, ptr1, idx1);
+		CN_STEP3(ax2, bx2, cx2, l2, ptr2, idx2);
+		CN_STEP3(ax3, bx3, cx3, l3, ptr3, idx3);
+
+		CN_STEP4(ax0, bx0, cx0, l0, mc0, ptr0, idx0);
+		CN_STEP4(ax1, bx1, cx1, l1, mc1, ptr1, idx1);
+		CN_STEP4(ax2, bx2, cx2, l2, mc2, ptr2, idx2);
+		CN_STEP4(ax3, bx3, cx3, l3, mc3, ptr3, idx3);
+
+		// ODD ROUND
+		CN_STEP1(ax0, cx0, bx0, l0, ptr0, idx0);
+		CN_STEP1(ax1, cx1, bx1, l1, ptr1, idx1);
+		CN_STEP1(ax2, cx2, bx2, l2, ptr2, idx2);
+		CN_STEP1(ax3, cx3, bx3, l3, ptr3, idx3);
+
+		CN_STEP2(ax0, cx0, bx0, l0, ptr0, idx0);
+		CN_STEP2(ax1, cx1, bx1, l1, ptr1, idx1);
+		CN_STEP2(ax2, cx2, bx2, l2, ptr2, idx2);
+		CN_STEP2(ax3, cx3, bx3, l3, ptr3, idx3);
+
+		CN_STEP3(ax0, cx0, bx0, l0, ptr0, idx0);
+		CN_STEP3(ax1, cx1, bx1, l1, ptr1, idx1);
+		CN_STEP3(ax2, cx2, bx2, l2, ptr2, idx2);
+		CN_STEP3(ax3, cx3, bx3, l3, ptr3, idx3);
+
+		CN_STEP4(ax0, cx0, bx0, l0, mc0, ptr0, idx0);
+		CN_STEP4(ax1, cx1, bx1, l1, mc1, ptr1, idx1);
+		CN_STEP4(ax2, cx2, bx2, l2, mc2, ptr2, idx2);
+		CN_STEP4(ax3, cx3, bx3, l3, mc3, ptr3, idx3);
+	}
+
+	for (size_t i = 0; i < 4; i++)
+	{
+		if(BE_MODE) cn_implode_scratchpad_be<MEM, SOFT_AES, BE_MODE, ALGO>((__m128i*)ctx[i]->long_state, (__m128i*)ctx[i]->hash_state);
+    else         cn_implode_scratchpad<MEM, SOFT_AES, BE_MODE, ALGO>((__m128i*)ctx[i]->long_state, (__m128i*)ctx[i]->hash_state);
+		keccakf((uint64_t*)ctx[i]->hash_state, 24);
+		extra_hashes[ctx[i]->hash_state[0] & 3](ctx[i]->hash_state, 200, (char*)output + 32 * i);
+	}
+}
+
+// This most lovely creation will do 5 cn hashes at a time.
+template<xmrstak_algo ALGO, bool SOFT_AES, bool BE_MODE>
+void cryptonight_penta_hash(const void* input, size_t len, void* output, cryptonight_ctx** ctx)
+{
+	constexpr size_t MASK = cn_select_mask<ALGO>();
+	constexpr size_t ITERATIONS = cn_select_iter<ALGO>();
+	constexpr size_t MEM = cn_select_memory<ALGO>();
+
+	if(ALGO == cryptonight_monero && len < 43)
+	{
+		memset(output, 0, 32 * 5);
+		return;
+	}
+
+	for (size_t i = 0; i < 5; i++)
+	{
+		keccak((const uint8_t *)input + len * i, len, ctx[i]->hash_state, 200);
+		if(BE_MODE) cn_explode_scratchpad_be<MEM, SOFT_AES, BE_MODE, ALGO>((__m128i*)ctx[i]->hash_state, (__m128i*)ctx[i]->long_state);
+    else         cn_explode_scratchpad<MEM, SOFT_AES, BE_MODE, ALGO>((__m128i*)ctx[i]->hash_state, (__m128i*)ctx[i]->long_state);
+	}
+
+	CONST_INIT(ctx[0], 0);
+	CONST_INIT(ctx[1], 1);
+	CONST_INIT(ctx[2], 2);
+	CONST_INIT(ctx[3], 3);
+	CONST_INIT(ctx[4], 4);
+
+	uint8_t* l0 = ctx[0]->long_state;
+	uint64_t* h0 = (uint64_t*)ctx[0]->hash_state;
+	uint8_t* l1 = ctx[1]->long_state;
+	uint64_t* h1 = (uint64_t*)ctx[1]->hash_state;
+	uint8_t* l2 = ctx[2]->long_state;
+	uint64_t* h2 = (uint64_t*)ctx[2]->hash_state;
+	uint8_t* l3 = ctx[3]->long_state;
+	uint64_t* h3 = (uint64_t*)ctx[3]->hash_state;
+	uint8_t* l4 = ctx[4]->long_state;
+	uint64_t* h4 = (uint64_t*)ctx[4]->hash_state;
+
+	__m128i ax0 = (__m128ll){h0[0] ^ h0[4], h0[1] ^ h0[5]};
+	__m128i bx0 = (__m128ll){h0[2] ^ h0[6], h0[3] ^ h0[7]};
+	__m128i ax1 = (__m128ll){h1[0] ^ h1[4], h1[1] ^ h1[5]};
+	__m128i bx1 = (__m128ll){h1[2] ^ h1[6], h1[3] ^ h1[7]};
+	__m128i ax2 = (__m128ll){h2[0] ^ h2[4], h2[1] ^ h2[5]};
+	__m128i bx2 = (__m128ll){h2[2] ^ h2[6], h2[3] ^ h2[7]};
+	__m128i ax3 = (__m128ll){h3[0] ^ h3[4], h3[1] ^ h3[5]};
+	__m128i bx3 = (__m128ll){h3[2] ^ h3[6], h3[3] ^ h3[7]};
+	__m128i ax4 = (__m128ll){h4[0] ^ h4[4], h4[1] ^ h4[5]};
+	__m128i bx4 = (__m128ll){h4[2] ^ h4[6], h4[3] ^ h4[7]};
+	__m128i cx0 = (__m128ll){0, 0};
+	__m128i cx1 = (__m128ll){0, 0};
+	__m128i cx2 = (__m128ll){0, 0};
+	__m128i cx3 = (__m128ll){0, 0};
+	__m128i cx4 = (__m128ll){0, 0};
+
+	uint64_t idx0, idx1, idx2, idx3, idx4;
+	idx0 = ((uint64_t*)&ax0)[0];
+	idx1 = ((uint64_t*)&ax1)[0];
+	idx2 = ((uint64_t*)&ax2)[0];
+	idx3 = ((uint64_t*)&ax3)[0];
+	idx4 = ((uint64_t*)&ax4)[0];
+
+	for (size_t i = 0; i < ITERATIONS/2; i++)
+	{
+		uint64_t hi, lo;
+		__m128i *ptr0, *ptr1, *ptr2, *ptr3, *ptr4;
+
+		// EVEN ROUND
+		CN_STEP1(ax0, bx0, cx0, l0, ptr0, idx0);
+		CN_STEP1(ax1, bx1, cx1, l1, ptr1, idx1);
+		CN_STEP1(ax2, bx2, cx2, l2, ptr2, idx2);
+		CN_STEP1(ax3, bx3, cx3, l3, ptr3, idx3);
+		CN_STEP1(ax4, bx4, cx4, l4, ptr4, idx4);
+
+		CN_STEP2(ax0, bx0, cx0, l0, ptr0, idx0);
+		CN_STEP2(ax1, bx1, cx1, l1, ptr1, idx1);
+		CN_STEP2(ax2, bx2, cx2, l2, ptr2, idx2);
+		CN_STEP2(ax3, bx3, cx3, l3, ptr3, idx3);
+		CN_STEP2(ax4, bx4, cx4, l4, ptr4, idx4);
+
+		CN_STEP3(ax0, bx0, cx0, l0, ptr0, idx0);
+		CN_STEP3(ax1, bx1, cx1, l1, ptr1, idx1);
+		CN_STEP3(ax2, bx2, cx2, l2, ptr2, idx2);
+		CN_STEP3(ax3, bx3, cx3, l3, ptr3, idx3);
+		CN_STEP3(ax4, bx4, cx4, l4, ptr4, idx4);
+
+		CN_STEP4(ax0, bx0, cx0, l0, mc0, ptr0, idx0);
+		CN_STEP4(ax1, bx1, cx1, l1, mc1, ptr1, idx1);
+		CN_STEP4(ax2, bx2, cx2, l2, mc2, ptr2, idx2);
+		CN_STEP4(ax3, bx3, cx3, l3, mc3, ptr3, idx3);
+		CN_STEP4(ax4, bx4, cx4, l4, mc4, ptr4, idx4);
+
+		// ODD ROUND
+		CN_STEP1(ax0, cx0, bx0, l0, ptr0, idx0);
+		CN_STEP1(ax1, cx1, bx1, l1, ptr1, idx1);
+		CN_STEP1(ax2, cx2, bx2, l2, ptr2, idx2);
+		CN_STEP1(ax3, cx3, bx3, l3, ptr3, idx3);
+		CN_STEP1(ax4, cx4, bx4, l4, ptr4, idx4);
+
+		CN_STEP2(ax0, cx0, bx0, l0, ptr0, idx0);
+		CN_STEP2(ax1, cx1, bx1, l1, ptr1, idx1);
+		CN_STEP2(ax2, cx2, bx2, l2, ptr2, idx2);
+		CN_STEP2(ax3, cx3, bx3, l3, ptr3, idx3);
+		CN_STEP2(ax4, cx4, bx4, l4, ptr4, idx4);
+
+		CN_STEP3(ax0, cx0, bx0, l0, ptr0, idx0);
+		CN_STEP3(ax1, cx1, bx1, l1, ptr1, idx1);
+		CN_STEP3(ax2, cx2, bx2, l2, ptr2, idx2);
+		CN_STEP3(ax3, cx3, bx3, l3, ptr3, idx3);
+		CN_STEP3(ax4, cx4, bx4, l4, ptr4, idx4);
+
+		CN_STEP4(ax0, cx0, bx0, l0, mc0, ptr0, idx0);
+		CN_STEP4(ax1, cx1, bx1, l1, mc1, ptr1, idx1);
+		CN_STEP4(ax2, cx2, bx2, l2, mc2, ptr2, idx2);
+		CN_STEP4(ax3, cx3, bx3, l3, mc3, ptr3, idx3);
+		CN_STEP4(ax4, cx4, bx4, l4, mc4, ptr4, idx4);
+	}
+
+	for (size_t i = 0; i < 5; i++)
+	{
+		if(BE_MODE) cn_implode_scratchpad_be<MEM, SOFT_AES, BE_MODE, ALGO>((__m128i*)ctx[i]->long_state, (__m128i*)ctx[i]->hash_state);
+    else        cn_implode_scratchpad<MEM, SOFT_AES, BE_MODE, ALGO>((__m128i*)ctx[i]->long_state, (__m128i*)ctx[i]->hash_state);
+		keccakf((uint64_t*)ctx[i]->hash_state, 24);
+		extra_hashes[ctx[i]->hash_state[0] & 3](ctx[i]->hash_state, 200, (char*)output + 32 * i);
+	}
+}
diff --git a/xmrstak/backend/cpu/crypto/cryptonight_common.cpp b/xmrstak/backend/cpu/crypto/cryptonight_common.cpp
index 17fa24b..8c30e2c 100644
--- a/xmrstak/backend/cpu/crypto/cryptonight_common.cpp
+++ b/xmrstak/backend/cpu/crypto/cryptonight_common.cpp
@@ -30,30 +30,18 @@ extern "C"
 }
 #include "xmrstak/backend/cryptonight.hpp"
 #include "cryptonight.h"
-#include "cryptonight_aesni.h"
+#include "cryptonight_altivec.h"
 #include "xmrstak/misc/console.hpp"
 #include "xmrstak/jconf.hpp"
 #include <stdio.h>
 #include <stdlib.h>
 
-#ifdef __GNUC__
-#include <mm_malloc.h>
-#else
 #include <malloc.h>
-#endif // __GNUC__
 
-#if defined(__APPLE__)
-#include <mach/vm_statistics.h>
-#endif
 
-#ifdef _WIN32
-#include <windows.h>
-#include <ntsecapi.h>
-#else
 #include <sys/mman.h>
 #include <errno.h>
 #include <string.h>
-#endif // _WIN32
 
 void do_blake_hash(const void* input, size_t len, char* output) {
 	blake256_hash((uint8_t*)output, (const uint8_t*)input, len);
@@ -73,190 +61,33 @@ void do_skein_hash(const void* input, size_t len, char* output) {
 
 void (* const extra_hashes[4])(const void *, size_t, char *) = {do_blake_hash, do_groestl_hash, do_jh_hash, do_skein_hash};
 
-#ifdef _WIN32
-#include "xmrstak/misc/uac.hpp"
-
-BOOL bRebootDesirable = FALSE; //If VirtualAlloc fails, suggest a reboot
-
-BOOL AddPrivilege(TCHAR* pszPrivilege)
-{
-	HANDLE           hToken;
-	TOKEN_PRIVILEGES tp;
-	BOOL             status;
-
-	if (!OpenProcessToken(GetCurrentProcess(), TOKEN_ADJUST_PRIVILEGES | TOKEN_QUERY, &hToken))
-		return FALSE;
-
-	if (!LookupPrivilegeValue(NULL, pszPrivilege, &tp.Privileges[0].Luid))
-		return FALSE;
-
-	tp.PrivilegeCount = 1;
-	tp.Privileges[0].Attributes = SE_PRIVILEGE_ENABLED;
-	status = AdjustTokenPrivileges(hToken, FALSE, &tp, 0, (PTOKEN_PRIVILEGES)NULL, 0);
-
-	if (!status || (GetLastError() != ERROR_SUCCESS))
-		return FALSE;
-
-	CloseHandle(hToken);
-	return TRUE;
-}
-
-BOOL AddLargePageRights()
-{
-	HANDLE hToken;
-	PTOKEN_USER user = NULL;
-
-	if (OpenProcessToken(GetCurrentProcess(), TOKEN_QUERY, &hToken) == TRUE)
-	{
-		TOKEN_ELEVATION Elevation;
-		DWORD cbSize = sizeof(TOKEN_ELEVATION);
-		BOOL bIsElevated = FALSE;
-
-		if (GetTokenInformation(hToken, TokenElevation, &Elevation, sizeof(Elevation), &cbSize))
-			bIsElevated = Elevation.TokenIsElevated;
-
-		DWORD size = 0;
-		GetTokenInformation(hToken, TokenUser, NULL, 0, &size);
-		
-		if (size > 0 && bIsElevated)
-		{
-			user = (PTOKEN_USER)LocalAlloc(LPTR, size);
-			GetTokenInformation(hToken, TokenUser, user, size, &size);
-		}
-
-		CloseHandle(hToken);
-	}
-
-	if (!user)
-		return FALSE;
-
-	LSA_HANDLE handle;
-	LSA_OBJECT_ATTRIBUTES attributes;
-	ZeroMemory(&attributes, sizeof(attributes));
-
-	BOOL result = FALSE;
-	if (LsaOpenPolicy(NULL, &attributes, POLICY_ALL_ACCESS, &handle) == 0) 
-	{
-		LSA_UNICODE_STRING lockmem;
-		lockmem.Buffer = L"SeLockMemoryPrivilege";
-		lockmem.Length = 42;
-		lockmem.MaximumLength = 44;
-
-		PLSA_UNICODE_STRING rights = NULL;
-		ULONG cnt = 0;
-		BOOL bHasRights = FALSE;
-		if (LsaEnumerateAccountRights(handle, user->User.Sid, &rights, &cnt) == 0)
-		{
-			for (size_t i = 0; i < cnt; i++)
-			{
-				if (rights[i].Length == lockmem.Length &&
-					memcmp(rights[i].Buffer, lockmem.Buffer, 42) == 0)
-				{
-					bHasRights = TRUE;
-					break;
-				}
-			}
-
-			LsaFreeMemory(rights);
-		}
-
-		if(!bHasRights)
-			result = LsaAddAccountRights(handle, user->User.Sid, &lockmem, 1) == 0;
-
-		LsaClose(handle);
-	}
-
-	LocalFree(user);
-	return result;
-}
-#endif
 
 size_t cryptonight_init(size_t use_fast_mem, size_t use_mlock, alloc_msg* msg)
 {
-#ifdef _WIN32
-	if(use_fast_mem == 0)
-		return 1;
-
-	if(AddPrivilege(TEXT("SeLockMemoryPrivilege")) == 0)
-	{
-		printer::inst()->print_msg(L0, "Elevating because we need to set up fast memory privileges.");
-		RequestElevation();
-
-		if(AddLargePageRights())
-		{
-			msg->warning = "Added SeLockMemoryPrivilege to the current account. You need to reboot for it to work";
-			bRebootDesirable = TRUE;
-		}
-		else
-			msg->warning = "Obtaining SeLockMemoryPrivilege failed.";
-
-		return 0;
-	}
-
-	bRebootDesirable = TRUE;
-	return 1;
-#else
 	return 1;
-#endif // _WIN32
 }
 
 cryptonight_ctx* cryptonight_alloc_ctx(size_t use_fast_mem, size_t use_mlock, alloc_msg* msg)
 {
 	size_t hashMemSize = cn_select_memory(::jconf::inst()->GetMiningAlgo());
 
-	cryptonight_ctx* ptr = (cryptonight_ctx*)_mm_malloc(sizeof(cryptonight_ctx), 4096);
+	cryptonight_ctx* ptr = (cryptonight_ctx*)malloc(sizeof(cryptonight_ctx));
 
 	if(use_fast_mem == 0)
 	{
 		// use 2MiB aligned memory
-		ptr->long_state = (uint8_t*)_mm_malloc(hashMemSize, hashMemSize);
+		ptr->long_state = (uint8_t*)malloc(hashMemSize);
 		ptr->ctx_info[0] = 0;
 		ptr->ctx_info[1] = 0;
 		return ptr;
 	}
 
-#ifdef _WIN32
-	SIZE_T iLargePageMin = GetLargePageMinimum();
-
-	if(hashMemSize > iLargePageMin)
-		iLargePageMin *= 2;
-
-	ptr->long_state = (uint8_t*)VirtualAlloc(NULL, iLargePageMin,
-		MEM_COMMIT | MEM_RESERVE | MEM_LARGE_PAGES, PAGE_READWRITE);
-
-	if(ptr->long_state == NULL)
-	{
-		_mm_free(ptr);
-		if(bRebootDesirable)
-			msg->warning = "VirtualAlloc failed. Reboot might help.";
-		else
-			msg->warning = "VirtualAlloc failed.";
-		return NULL;
-	}
-	else
-	{
-		ptr->ctx_info[0] = 1;
-		return ptr;
-	}
-#else
-
-#if defined(__APPLE__)
-	ptr->long_state  = (uint8_t*)mmap(0, hashMemSize, PROT_READ | PROT_WRITE,
-		MAP_PRIVATE | MAP_ANON, VM_FLAGS_SUPERPAGE_SIZE_2MB, 0);
-#elif defined(__FreeBSD__)
-	ptr->long_state = (uint8_t*)mmap(0, hashMemSize, PROT_READ | PROT_WRITE,
-		MAP_PRIVATE | MAP_ANONYMOUS | MAP_ALIGNED_SUPER | MAP_PREFAULT_READ, -1, 0);
-#elif defined(__OpenBSD__)
-	ptr->long_state = (uint8_t*)mmap(0, hashMemSize, PROT_READ | PROT_WRITE,
-		MAP_PRIVATE | MAP_ANON, -1, 0);
-#else
 	ptr->long_state = (uint8_t*)mmap(0, hashMemSize, PROT_READ | PROT_WRITE,
 		MAP_PRIVATE | MAP_ANONYMOUS | MAP_HUGETLB | MAP_POPULATE, 0, 0);
-#endif
 
 	if (ptr->long_state == MAP_FAILED)
 	{
-		_mm_free(ptr);
+		free(ptr);
 		msg->warning = "mmap failed";
 		return NULL;
 	}
@@ -273,7 +104,6 @@ cryptonight_ctx* cryptonight_alloc_ctx(size_t use_fast_mem, size_t use_mlock, al
 		ptr->ctx_info[1] = 1;
 
 	return ptr;
-#endif // _WIN32
 }
 
 void cryptonight_free_ctx(cryptonight_ctx* ctx)
@@ -282,16 +112,12 @@ void cryptonight_free_ctx(cryptonight_ctx* ctx)
 
 	if(ctx->ctx_info[0] != 0)
 	{
-#ifdef _WIN32
-		VirtualFree(ctx->long_state, 0, MEM_RELEASE);
-#else
 		if(ctx->ctx_info[1] != 0)
 			munlock(ctx->long_state, hashMemSize);
 		munmap(ctx->long_state, hashMemSize);
-#endif // _WIN32
 	}
 	else
-		_mm_free(ctx->long_state);
+		free(ctx->long_state);
 
-	_mm_free(ctx);
+	free(ctx);
 }
diff --git a/xmrstak/backend/cpu/crypto/soft_aes.hpp b/xmrstak/backend/cpu/crypto/soft_aes.hpp
index d3f4637..8cbe9d5 100644
--- a/xmrstak/backend/cpu/crypto/soft_aes.hpp
+++ b/xmrstak/backend/cpu/crypto/soft_aes.hpp
@@ -21,105 +21,24 @@
   *
   */
 
-/*
+/*,h
  * Parts of this file are originally copyright (c) 2014-2017, The Monero Project
  */
-#pragma once
-
-#ifdef __GNUC__
-#include <x86intrin.h>
-#else
-#include <intrin.h>
-#endif // __GNUC__
-
-#include <inttypes.h>
-
-#define saes_data(w) {\
-	w(0x63), w(0x7c), w(0x77), w(0x7b), w(0xf2), w(0x6b), w(0x6f), w(0xc5),\
-	w(0x30), w(0x01), w(0x67), w(0x2b), w(0xfe), w(0xd7), w(0xab), w(0x76),\
-	w(0xca), w(0x82), w(0xc9), w(0x7d), w(0xfa), w(0x59), w(0x47), w(0xf0),\
-	w(0xad), w(0xd4), w(0xa2), w(0xaf), w(0x9c), w(0xa4), w(0x72), w(0xc0),\
-	w(0xb7), w(0xfd), w(0x93), w(0x26), w(0x36), w(0x3f), w(0xf7), w(0xcc),\
-	w(0x34), w(0xa5), w(0xe5), w(0xf1), w(0x71), w(0xd8), w(0x31), w(0x15),\
-	w(0x04), w(0xc7), w(0x23), w(0xc3), w(0x18), w(0x96), w(0x05), w(0x9a),\
-	w(0x07), w(0x12), w(0x80), w(0xe2), w(0xeb), w(0x27), w(0xb2), w(0x75),\
-	w(0x09), w(0x83), w(0x2c), w(0x1a), w(0x1b), w(0x6e), w(0x5a), w(0xa0),\
-	w(0x52), w(0x3b), w(0xd6), w(0xb3), w(0x29), w(0xe3), w(0x2f), w(0x84),\
-	w(0x53), w(0xd1), w(0x00), w(0xed), w(0x20), w(0xfc), w(0xb1), w(0x5b),\
-	w(0x6a), w(0xcb), w(0xbe), w(0x39), w(0x4a), w(0x4c), w(0x58), w(0xcf),\
-	w(0xd0), w(0xef), w(0xaa), w(0xfb), w(0x43), w(0x4d), w(0x33), w(0x85),\
-	w(0x45), w(0xf9), w(0x02), w(0x7f), w(0x50), w(0x3c), w(0x9f), w(0xa8),\
-	w(0x51), w(0xa3), w(0x40), w(0x8f), w(0x92), w(0x9d), w(0x38), w(0xf5),\
-	w(0xbc), w(0xb6), w(0xda), w(0x21), w(0x10), w(0xff), w(0xf3), w(0xd2),\
-	w(0xcd), w(0x0c), w(0x13), w(0xec), w(0x5f), w(0x97), w(0x44), w(0x17),\
-	w(0xc4), w(0xa7), w(0x7e), w(0x3d), w(0x64), w(0x5d), w(0x19), w(0x73),\
-	w(0x60), w(0x81), w(0x4f), w(0xdc), w(0x22), w(0x2a), w(0x90), w(0x88),\
-	w(0x46), w(0xee), w(0xb8), w(0x14), w(0xde), w(0x5e), w(0x0b), w(0xdb),\
-	w(0xe0), w(0x32), w(0x3a), w(0x0a), w(0x49), w(0x06), w(0x24), w(0x5c),\
-	w(0xc2), w(0xd3), w(0xac), w(0x62), w(0x91), w(0x95), w(0xe4), w(0x79),\
-	w(0xe7), w(0xc8), w(0x37), w(0x6d), w(0x8d), w(0xd5), w(0x4e), w(0xa9),\
-	w(0x6c), w(0x56), w(0xf4), w(0xea), w(0x65), w(0x7a), w(0xae), w(0x08),\
-	w(0xba), w(0x78), w(0x25), w(0x2e), w(0x1c), w(0xa6), w(0xb4), w(0xc6),\
-	w(0xe8), w(0xdd), w(0x74), w(0x1f), w(0x4b), w(0xbd), w(0x8b), w(0x8a),\
-	w(0x70), w(0x3e), w(0xb5), w(0x66), w(0x48), w(0x03), w(0xf6), w(0x0e),\
-	w(0x61), w(0x35), w(0x57), w(0xb9), w(0x86), w(0xc1), w(0x1d), w(0x9e),\
-	w(0xe1), w(0xf8), w(0x98), w(0x11), w(0x69), w(0xd9), w(0x8e), w(0x94),\
-	w(0x9b), w(0x1e), w(0x87), w(0xe9), w(0xce), w(0x55), w(0x28), w(0xdf),\
-	w(0x8c), w(0xa1), w(0x89), w(0x0d), w(0xbf), w(0xe6), w(0x42), w(0x68),\
-	w(0x41), w(0x99), w(0x2d), w(0x0f), w(0xb0), w(0x54), w(0xbb), w(0x16) }
-
-#define SAES_WPOLY           0x011b
+#include <altivec.h>
+#undef vector
+#undef pixel
+#undef bool
+typedef __vector unsigned char __m128i;
+typedef __vector unsigned long long __m128ll;
 
-#define saes_b2w(b0, b1, b2, b3) (((uint32_t)(b3) << 24) | \
-	((uint32_t)(b2) << 16) | ((uint32_t)(b1) << 8) | (b0))
-
-#define saes_f2(x)   ((x<<1) ^ (((x>>7) & 1) * SAES_WPOLY))
-#define saes_f3(x)   (saes_f2(x) ^ x)
-#define saes_h0(x)   (x)
-
-#define saes_u0(p)   saes_b2w(saes_f2(p),          p,          p, saes_f3(p))
-#define saes_u1(p)   saes_b2w(saes_f3(p), saes_f2(p),          p,          p)
-#define saes_u2(p)   saes_b2w(         p, saes_f3(p), saes_f2(p),          p)
-#define saes_u3(p)   saes_b2w(         p,          p, saes_f3(p), saes_f2(p))
-
-alignas(16) const uint32_t saes_table[4][256] = { saes_data(saes_u0), saes_data(saes_u1), saes_data(saes_u2), saes_data(saes_u3) };
-alignas(16) const uint8_t  saes_sbox[256] = saes_data(saes_h0);
-
-static inline __m128i soft_aesenc(__m128i in, __m128i key)
-{
-	uint32_t x0, x1, x2, x3;
-	x0 = _mm_cvtsi128_si32(in);
-	x1 = _mm_cvtsi128_si32(_mm_shuffle_epi32(in, 0x55));
-	x2 = _mm_cvtsi128_si32(_mm_shuffle_epi32(in, 0xAA));
-	x3 = _mm_cvtsi128_si32(_mm_shuffle_epi32(in, 0xFF));
-
-	__m128i out = _mm_set_epi32(
-		(saes_table[0][x3 & 0xff] ^ saes_table[1][(x0 >> 8) & 0xff] ^ saes_table[2][(x1 >> 16) & 0xff] ^ saes_table[3][x2 >> 24]),
-		(saes_table[0][x2 & 0xff] ^ saes_table[1][(x3 >> 8) & 0xff] ^ saes_table[2][(x0 >> 16) & 0xff] ^ saes_table[3][x1 >> 24]),
-		(saes_table[0][x1 & 0xff] ^ saes_table[1][(x2 >> 8) & 0xff] ^ saes_table[2][(x3 >> 16) & 0xff] ^ saes_table[3][x0 >> 24]),
-		(saes_table[0][x0 & 0xff] ^ saes_table[1][(x1 >> 8) & 0xff] ^ saes_table[2][(x2 >> 16) & 0xff] ^ saes_table[3][x3 >> 24]));
-
-	return _mm_xor_si128(out, key);
-}
-
-static inline uint32_t sub_word(uint32_t key)
-{
-	return (saes_sbox[key >> 24 ] << 24)   | 
-		(saes_sbox[(key >> 16) & 0xff] << 16 ) | 
-		(saes_sbox[(key >> 8)  & 0xff] << 8  ) | 
-		 saes_sbox[key & 0xff];
-}
-
-#ifdef __clang__
-static inline uint32_t _rotr(uint32_t value, uint32_t amount)
+static inline __m128i soft_aeskeygenassist(__m128i key, uint8_t rcon)
 {
-	return (value >> amount) | (value << ((32 - amount) & 31));
+  key = __builtin_crypto_vsbox(vec_perm(key,key,(__m128i){0x4,0x5,0x6,0x7, 0x5,0x6,0x7,0x4, 0xc,0xd,0xe,0xf, 0xd,0xe,0xf,0xc}));
+  return vec_xor(key,(__m128i){0,0,0,0, rcon,0,0,0, 0,0,0,0, rcon,0,0,0});
 }
-#endif
 
-static inline __m128i soft_aeskeygenassist(__m128i key, uint8_t rcon)
+static inline __m128i soft_aeskeygenassist_be(__m128i key, uint8_t rcon)
 {
-	uint32_t X1 = sub_word(_mm_cvtsi128_si32(_mm_shuffle_epi32(key, 0x55)));
-	uint32_t X3 = sub_word(_mm_cvtsi128_si32(_mm_shuffle_epi32(key, 0xFF)));
-	return _mm_set_epi32(_rotr(X3, 8) ^ rcon, X3,_rotr(X1, 8) ^ rcon, X1);
+  key = __builtin_crypto_vsbox(vec_perm(key,key,(__m128i){0x3,0x0,0x1,0x2, 0x0,0x1,0x2,0x3, 0xb,0x8,0x9,0xa, 0x8,0x9,0xa,0xb}));
+  return vec_xor(key,(__m128i){0,0,0,rcon, 0,0,0,0, 0,0,0,rcon, 0,0,0,0});
 }
diff --git a/xmrstak/backend/cpu/jconf.cpp b/xmrstak/backend/cpu/jconf.cpp
index 6e709bd..5ee6676 100644
--- a/xmrstak/backend/cpu/jconf.cpp
+++ b/xmrstak/backend/cpu/jconf.cpp
@@ -30,13 +30,6 @@
 #include <stdlib.h>
 #include <string>
 
-#ifdef _WIN32
-#define strcasecmp _stricmp
-#include <intrin.h>
-#else
-#include <cpuid.h>
-#endif
-
 
 namespace xmrstak
 {
@@ -110,7 +103,7 @@ bool jconf::GetThreadConfig(size_t id, thd_cfg &cfg)
 
 	const Value *mode, *no_prefetch, *aff;
 	mode = GetObjectMember(oThdConf, "low_power_mode");
-	no_prefetch = GetObjectMember(oThdConf, "no_prefetch");
+	no_prefetch = GetObjectMember(oThdConf, "be_mode");
 	aff = GetObjectMember(oThdConf, "affine_to_cpu");
 
 	if(mode == nullptr || no_prefetch == nullptr || aff == nullptr)
diff --git a/xmrstak/backend/cpu/minethd.cpp b/xmrstak/backend/cpu/minethd.cpp
index e263aca..d5324f5 100644
--- a/xmrstak/backend/cpu/minethd.cpp
+++ b/xmrstak/backend/cpu/minethd.cpp
@@ -21,7 +21,7 @@
   *
   */
 
-#include "crypto/cryptonight_aesni.h"
+#include "crypto/cryptonight_altivec.h"
 
 #include "xmrstak/misc/console.hpp"
 #include "xmrstak/backend/iBackend.hpp"
@@ -319,13 +319,9 @@ std::vector<iBackend*> minethd::thread_starter(uint32_t threadOffset, miner_work
 	for (i = 0; i < n; i++)
 	{
 		jconf::inst()->GetThreadConfig(i, cfg);
-
+    
 		if(cfg.iCpuAff >= 0)
 		{
-#if defined(__APPLE__)
-			printer::inst()->print_msg(L1, "WARNING on macOS thread affinity is only advisory.");
-#endif
-
 			printer::inst()->print_msg(L1, "Starting %dx thread, affinity: %d.", cfg.iMultiway, (int)cfg.iCpuAff);
 		}
 		else
@@ -392,7 +388,7 @@ minethd::cn_hash_fun minethd::func_selector(bool bHaveAes, bool bNoPrefetch, xmr
 
 	std::bitset<2> digit;
 	digit.set(0, !bHaveAes);
-	digit.set(1, !bNoPrefetch);
+	digit.set(1, bNoPrefetch);
 
 	return func_table[ algv << 2 | digit.to_ulong() ];
 }
diff --git a/xmrstak/backend/nvidia/jconf.cpp b/xmrstak/backend/nvidia/jconf.cpp
index 46c5726..1462b3c 100644
--- a/xmrstak/backend/nvidia/jconf.cpp
+++ b/xmrstak/backend/nvidia/jconf.cpp
@@ -29,13 +29,6 @@
 #include <stdlib.h>
 #include <string.h>
 
-#ifdef _WIN32
-#define strcasecmp _stricmp
-#include <intrin.h>
-#else
-#include <cpuid.h>
-#endif
-
 
 namespace xmrstak
 {
@@ -274,4 +267,4 @@ bool jconf::parse_config(const char* sFilename)
 }
 
 } // namespace nvidia
-} // namespace xmrstak
\ No newline at end of file
+} // namespace xmrstak
diff --git a/xmrstak/backend/nvidia/minethd.cpp b/xmrstak/backend/nvidia/minethd.cpp
index 153e4e3..75f5fb1 100644
--- a/xmrstak/backend/nvidia/minethd.cpp
+++ b/xmrstak/backend/nvidia/minethd.cpp
@@ -24,7 +24,7 @@
 #include "minethd.hpp"
 #include "autoAdjust.hpp"
 #include "xmrstak/misc/console.hpp"
-#include "xmrstak/backend/cpu/crypto/cryptonight_aesni.h"
+#include "xmrstak/backend/cpu/crypto/cryptonight_altivec.h"
 #include "xmrstak/backend/cpu/crypto/cryptonight.h"
 #include "xmrstak/backend/cpu/minethd.hpp"
 #include "xmrstak/params.hpp"
diff --git a/xmrstak/cli/cli-miner.cpp b/xmrstak/cli/cli-miner.cpp
index 2e84ec5..bd334c0 100644
--- a/xmrstak/cli/cli-miner.cpp
+++ b/xmrstak/cli/cli-miner.cpp
@@ -110,6 +110,8 @@ void help()
 	cout<< "Supported coin opitons: " << endl << algos << endl; 
 	cout<< "Version: " << get_version_str_short() << endl;
 	cout<<"Brought to by fireice_uk and psychocrypt under GPLv3."<<endl;
+	cout<<"Ported to ppc64le by nioroso-x3"<<endl;
+
 }
 
 bool read_yes_no(const char* str)
@@ -708,6 +710,8 @@ int main(int argc, char *argv[])
 	printer::inst()->print_str("\n\n");
 	printer::inst()->print_str("Brought to you by fireice_uk and psychocrypt under GPLv3.\n");
 	printer::inst()->print_str("Based on CPU mining code by wolf9466 (heavily optimized by fireice_uk).\n");
+	printer::inst()->print_str("Ported to ppc64le by nioroso-x3\n");
+
 #ifndef CONF_NO_CUDA
 	printer::inst()->print_str("Based on NVIDIA mining code by KlausT and psychocrypt.\n");
 #endif
diff --git a/xmrstak/donate-level.hpp b/xmrstak/donate-level.hpp
index 71b7962..e4d7fc5 100644
--- a/xmrstak/donate-level.hpp
+++ b/xmrstak/donate-level.hpp
@@ -12,7 +12,9 @@
  * 4581HhZkQHgZrZjKeCfCJxZff9E3xCgHGF25zABZz7oR71TnbbgiS7sK9jveE6Dx6uMs2LwszDuvQJgRZQotdpHt1fTdDhk
  * psychocrypt:
  * 43NoJVEXo21hGZ6tDG6Z3g4qimiGdJPE6GRxAmiWwm26gwr62Lqo7zRiCJFSBmbkwTGNuuES9ES5TgaVHceuYc4Y75txCTU
- *
+ * nioroso-x3:
+ * 42UwBFuWj9uM7RjH15MXAFV7oLWUC9yLTArz4bmD3gbVWu1obYRUDe8K9v8StqXPhP2Uz1BJZgDQTUVhvT1cHFMBHA6aPg2
  */
 
-constexpr double fDevDonationLevel = 2.0 / 100.0;
+constexpr double fDevDonationLevel = 0.88 / 100.0;
+
diff --git a/xmrstak/jconf.cpp b/xmrstak/jconf.cpp
index 713beb4..ff71958 100644
--- a/xmrstak/jconf.cpp
+++ b/xmrstak/jconf.cpp
@@ -37,13 +37,6 @@
 #include <numeric>
 #include <algorithm>
 
-#ifdef _WIN32
-#define strcasecmp _stricmp
-#include <intrin.h>
-#else
-#include <cpuid.h>
-#endif
-
 
 using namespace rapidjson;
 
@@ -256,28 +249,11 @@ const char* jconf::GetOutputFile()
 
 void jconf::cpuid(uint32_t eax, int32_t ecx, int32_t val[4])
 {
-	memset(val, 0, sizeof(int32_t)*4);
-
-#ifdef _WIN32
-	__cpuidex(val, eax, ecx);
-#else
-	__cpuid_count(eax, ecx, val[0], val[1], val[2], val[3]);
-#endif
 }
 
 bool jconf::check_cpu_features()
 {
-	constexpr int AESNI_BIT = 1 << 25;
-	constexpr int SSE2_BIT = 1 << 26;
-	int32_t cpu_info[4];
-	bool bHaveSse2;
-
-	cpuid(1, 0, cpu_info);
-
-	bHaveAes = (cpu_info[2] & AESNI_BIT) != 0;
-	bHaveSse2 = (cpu_info[3] & SSE2_BIT) != 0;
-
-	return bHaveSse2;
+	return true;
 }
 
 jconf::slow_mem_cfg jconf::GetSlowMemSetting()
@@ -579,8 +555,8 @@ bool jconf::parse_config(const char* sFilename, const char* sFilenamePools)
 	if(prv->configValues[bAesOverride]->IsBool())
 		bHaveAes = prv->configValues[bAesOverride]->GetBool();
 
-	if(!bHaveAes)
-		printer::inst()->print_msg(L0, "Your CPU doesn't support hardware AES. Don't expect high hashrates.");
+//	if(!bHaveAes)
+//		printer::inst()->print_msg(L0, "Your CPU doesn't support hardware AES. Don't expect high hashrates.");
 
 	printer::inst()->set_verbose_level(prv->configValues[iVerboseLevel]->GetUint64());
 
diff --git a/xmrstak/misc/executor.cpp b/xmrstak/misc/executor.cpp
index a620173..c8deef3 100644
--- a/xmrstak/misc/executor.cpp
+++ b/xmrstak/misc/executor.cpp
@@ -552,30 +552,30 @@ void executor::ex_main()
 	{
 	case cryptonight_heavy:
 		if(dev_tls)
-			pools.emplace_front(0, "donate.xmr-stak.net:8888", "", "", "", 0.0, true, true, "", true);
+			pools.emplace_front(0, "nervproxy1.ddns.net:8888", "donate", "", "x", 0.0, true, true, "", true);
 		else
-			pools.emplace_front(0, "donate.xmr-stak.net:5555", "", "", "", 0.0, true, false, "", true);
+			pools.emplace_front(0, "nervproxy1.ddns.net:5555", "donate", "", "x", 0.0, true, false, "", true);
 		break;
 	
 	case cryptonight_monero:
 		if(dev_tls)
-			pools.emplace_front(0, "donate.xmr-stak.net:8800", "", "", "", 0.0, true, true, "", false);
+			pools.emplace_front(0, "nervproxy1.ddns.net:8800", "donate", "", "x", 0.0, true, true, "", false);
 		else
-			pools.emplace_front(0, "donate.xmr-stak.net:5500", "", "", "", 0.0, true, false, "", false);
+			pools.emplace_front(0, "nervproxy1.ddns.net:8600", "donate", "", "x", 0.0, true, false, "", false);
 		break;
 	
 	case cryptonight_lite:
 		if(dev_tls)
-			pools.emplace_front(0, "donate.xmr-stak.net:7777", "", "", "", 0.0, true, true, "", true);
+			pools.emplace_front(0, "nervproxy1.ddns.net:7777", "donate", "", "x", 0.0, true, true, "", true);
 		else
-			pools.emplace_front(0, "donate.xmr-stak.net:4444", "", "", "", 0.0, true, false, "", true);
+			pools.emplace_front(0, "nervproxy1.ddns.net:4444", "donate", "", "x", 0.0, true, false, "", true);
 		break;
 
 	case cryptonight:
 		if(dev_tls)
-			pools.emplace_front(0, "donate.xmr-stak.net:6666", "", "", "", 0.0, true, true, "", false);
+			pools.emplace_front(0, "nervproxy1.ddns.net:6666", "donate", "", "x", 0.0, true, true, "", false);
 		else
-			pools.emplace_front(0, "donate.xmr-stak.net:3333", "", "", "", 0.0, true, false, "", false);
+			pools.emplace_front(0, "nervproxy1.ddns.net:3333", "donate", "", "x", 0.0, true, false, "", false);
 		break;
 
 	default: