add aeon support to backend cpu

- update auto suggestion default and hwloc
- extent hash function table to support aeon and xmr within one miner

3 files changed, 56 insertions, 36 deletions

diff --git a/xmrstak/backend/cpu/crypto/cryptonight.h b/xmrstak/backend/cpu/crypto/cryptonight.h
index 978c798..d07050e 100644
--- a/xmrstak/backend/cpu/crypto/cryptonight.h
+++ b/xmrstak/backend/cpu/crypto/cryptonight.h
@@ -7,8 +7,8 @@ extern "C" {
 
 #include <stddef.h>
 #include <inttypes.h>
+#include "../../cryptonight.hpp"
 
-#define MEMORY  2097152
 
 typedef struct {
 	uint8_t hash_state[224]; // Need only 200, explicit align
diff --git a/xmrstak/backend/cpu/crypto/cryptonight_aesni.h b/xmrstak/backend/cpu/crypto/cryptonight_aesni.h
index 942d511..2a6a769 100644
--- a/xmrstak/backend/cpu/crypto/cryptonight_aesni.h
+++ b/xmrstak/backend/cpu/crypto/cryptonight_aesni.h
@@ -287,7 +287,7 @@ void cn_implode_scratchpad(const __m128i* input, __m128i* output)
 	_mm_store_si128(output + 11, xout7);
 }
 
-template<size_t ITERATIONS, size_t MEM, bool SOFT_AES, bool PREFETCH>
+template<size_t MASK, size_t ITERATIONS, size_t MEM, bool SOFT_AES, bool PREFETCH>
 void cryptonight_hash(const void* input, size_t len, void* output, cryptonight_ctx* ctx0)
 {
 	keccak((const uint8_t *)input, len, ctx0->hash_state, 200);
@@ -308,36 +308,36 @@ void cryptonight_hash(const void* input, size_t len, void* output, cryptonight_c
 	for(size_t i = 0; i < ITERATIONS; i++)
 	{
 		__m128i cx;
-		cx = _mm_load_si128((__m128i *)&l0[idx0 & 0x1FFFF0]);
+		cx = _mm_load_si128((__m128i *)&l0[idx0 & MASK]);
 
 		if(SOFT_AES)
 			cx = soft_aesenc(cx, _mm_set_epi64x(ah0, al0));
 		else
 			cx = _mm_aesenc_si128(cx, _mm_set_epi64x(ah0, al0));
 
-		_mm_store_si128((__m128i *)&l0[idx0 & 0x1FFFF0], _mm_xor_si128(bx0, cx));
+		_mm_store_si128((__m128i *)&l0[idx0 & MASK], _mm_xor_si128(bx0, cx));
 		idx0 = _mm_cvtsi128_si64(cx);
 		bx0 = cx;
 
 		if(PREFETCH)
-			_mm_prefetch((const char*)&l0[idx0 & 0x1FFFF0], _MM_HINT_T0);
+			_mm_prefetch((const char*)&l0[idx0 & MASK], _MM_HINT_T0);
 
 		uint64_t hi, lo, cl, ch;
-		cl = ((uint64_t*)&l0[idx0 & 0x1FFFF0])[0];
-		ch = ((uint64_t*)&l0[idx0 & 0x1FFFF0])[1];
+		cl = ((uint64_t*)&l0[idx0 & MASK])[0];
+		ch = ((uint64_t*)&l0[idx0 & MASK])[1];
 
 		lo = _umul128(idx0, cl, &hi);
 
 		al0 += hi;
 		ah0 += lo;
-		((uint64_t*)&l0[idx0 & 0x1FFFF0])[0] = al0;
-		((uint64_t*)&l0[idx0 & 0x1FFFF0])[1] = ah0;
+		((uint64_t*)&l0[idx0 & MASK])[0] = al0;
+		((uint64_t*)&l0[idx0 & MASK])[1] = ah0;
 		ah0 ^= ch;
 		al0 ^= cl;
 		idx0 = al0;
 
 		if(PREFETCH)
-			_mm_prefetch((const char*)&l0[idx0 & 0x1FFFF0], _MM_HINT_T0);
+			_mm_prefetch((const char*)&l0[idx0 & MASK], _MM_HINT_T0);
 	}
 
 	// Optim - 90% time boundary
@@ -352,7 +352,7 @@ void cryptonight_hash(const void* input, size_t len, void* output, cryptonight_c
 // 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<size_t ITERATIONS, size_t MEM, bool SOFT_AES, bool PREFETCH>
+template<size_t MASK, size_t ITERATIONS, size_t MEM, bool SOFT_AES, bool PREFETCH>
 void cryptonight_double_hash(const void* input, size_t len, void* output, cryptonight_ctx* __restrict ctx0, cryptonight_ctx* __restrict ctx1)
 {
 	keccak((const uint8_t *)input, len, ctx0->hash_state, 200);
@@ -381,66 +381,66 @@ void cryptonight_double_hash(const void* input, size_t len, void* output, crypto
 	for (size_t i = 0; i < ITERATIONS; i++)
 	{
 		__m128i cx;
-		cx = _mm_load_si128((__m128i *)&l0[idx0 & 0x1FFFF0]);
+		cx = _mm_load_si128((__m128i *)&l0[idx0 & MASK]);
 
 		if(SOFT_AES)
 			cx = soft_aesenc(cx, _mm_set_epi64x(axh0, axl0));
 		else
 			cx = _mm_aesenc_si128(cx, _mm_set_epi64x(axh0, axl0));
 
-		_mm_store_si128((__m128i *)&l0[idx0 & 0x1FFFF0], _mm_xor_si128(bx0, cx));
+		_mm_store_si128((__m128i *)&l0[idx0 & MASK], _mm_xor_si128(bx0, cx));
 		idx0 = _mm_cvtsi128_si64(cx);
 		bx0 = cx;
 
 		if(PREFETCH)
-			_mm_prefetch((const char*)&l0[idx0 & 0x1FFFF0], _MM_HINT_T0);
+			_mm_prefetch((const char*)&l0[idx0 & MASK], _MM_HINT_T0);
 
-		cx = _mm_load_si128((__m128i *)&l1[idx1 & 0x1FFFF0]);
+		cx = _mm_load_si128((__m128i *)&l1[idx1 & MASK]);
 
 		if(SOFT_AES)
 			cx = soft_aesenc(cx, _mm_set_epi64x(axh1, axl1));
 		else
 			cx = _mm_aesenc_si128(cx, _mm_set_epi64x(axh1, axl1));
 
-		_mm_store_si128((__m128i *)&l1[idx1 & 0x1FFFF0], _mm_xor_si128(bx1, cx));
+		_mm_store_si128((__m128i *)&l1[idx1 & MASK], _mm_xor_si128(bx1, cx));
 		idx1 = _mm_cvtsi128_si64(cx);
 		bx1 = cx;
 
 		if(PREFETCH)
-			_mm_prefetch((const char*)&l1[idx1 & 0x1FFFF0], _MM_HINT_T0);
+			_mm_prefetch((const char*)&l1[idx1 & MASK], _MM_HINT_T0);
 
 		uint64_t hi, lo, cl, ch;
-		cl = ((uint64_t*)&l0[idx0 & 0x1FFFF0])[0];
-		ch = ((uint64_t*)&l0[idx0 & 0x1FFFF0])[1];
+		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 & 0x1FFFF0])[0] = axl0;
-		((uint64_t*)&l0[idx0 & 0x1FFFF0])[1] = axh0;
+		((uint64_t*)&l0[idx0 & MASK])[0] = axl0;
+		((uint64_t*)&l0[idx0 & MASK])[1] = axh0;
 		axh0 ^= ch;
 		axl0 ^= cl;
 		idx0 = axl0;
 
 		if(PREFETCH)
-			_mm_prefetch((const char*)&l0[idx0 & 0x1FFFF0], _MM_HINT_T0);
+			_mm_prefetch((const char*)&l0[idx0 & MASK], _MM_HINT_T0);
 
-		cl = ((uint64_t*)&l1[idx1 & 0x1FFFF0])[0];
-		ch = ((uint64_t*)&l1[idx1 & 0x1FFFF0])[1];
+		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 & 0x1FFFF0])[0] = axl1;
-		((uint64_t*)&l1[idx1 & 0x1FFFF0])[1] = axh1;
+		((uint64_t*)&l1[idx1 & MASK])[0] = axl1;
+		((uint64_t*)&l1[idx1 & MASK])[1] = axh1;
 		axh1 ^= ch;
 		axl1 ^= cl;
 		idx1 = axl1;
 
 		if(PREFETCH)
-			_mm_prefetch((const char*)&l1[idx1 & 0x1FFFF0], _MM_HINT_T0);
+			_mm_prefetch((const char*)&l1[idx1 & MASK], _MM_HINT_T0);
 	}
 
 	// Optim - 90% time boundary
diff --git a/xmrstak/backend/cpu/crypto/cryptonight_common.cpp b/xmrstak/backend/cpu/crypto/cryptonight_common.cpp
index 0690415..70ad27c 100644
--- a/xmrstak/backend/cpu/crypto/cryptonight_common.cpp
+++ b/xmrstak/backend/cpu/crypto/cryptonight_common.cpp
@@ -30,6 +30,8 @@ extern "C"
 }
 #include "cryptonight.h"
 #include "cryptonight_aesni.h"
+#include "../../../jconf.hpp"
+#include "../../cryptonight.hpp"
 #include <stdio.h>
 #include <stdlib.h>
 
@@ -194,12 +196,21 @@ size_t cryptonight_init(size_t use_fast_mem, size_t use_mlock, alloc_msg* msg)
 
 cryptonight_ctx* cryptonight_alloc_ctx(size_t use_fast_mem, size_t use_mlock, alloc_msg* msg)
 {
+	size_t hashMemSize;
+	if(::jconf::inst()->IsCurrencyXMR())
+	{
+		hashMemSize = XMR_MEMORY;
+	}
+	else
+	{
+		hashMemSize = AEON_MEMORY;
+	}
 	cryptonight_ctx* ptr = (cryptonight_ctx*)_mm_malloc(sizeof(cryptonight_ctx), 4096);
 
 	if(use_fast_mem == 0)
 	{
 		// use 2MiB aligned memory
-		ptr->long_state = (uint8_t*)_mm_malloc(MEMORY, 2*1024*1024);
+		ptr->long_state = (uint8_t*)_mm_malloc(hashMemSize, hashMemSize);
 		ptr->ctx_info[0] = 0;
 		ptr->ctx_info[1] = 0;
 		return ptr;
@@ -208,7 +219,7 @@ cryptonight_ctx* cryptonight_alloc_ctx(size_t use_fast_mem, size_t use_mlock, al
 #ifdef _WIN32
 	SIZE_T iLargePageMin = GetLargePageMinimum();
 
-	if(MEMORY > iLargePageMin)
+	if(hashMemSize > iLargePageMin)
 		iLargePageMin *= 2;
 
 	ptr->long_state = (uint8_t*)VirtualAlloc(NULL, iLargePageMin,
@@ -231,13 +242,13 @@ cryptonight_ctx* cryptonight_alloc_ctx(size_t use_fast_mem, size_t use_mlock, al
 #else
 
 #if defined(__APPLE__)
-	ptr->long_state  = (uint8_t*)mmap(0, MEMORY, PROT_READ | PROT_WRITE,
+	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, MEMORY, PROT_READ | PROT_WRITE,
+	ptr->long_state = (uint8_t*)mmap(0, hashMemSize, PROT_READ | PROT_WRITE,
 		MAP_PRIVATE | MAP_ANONYMOUS | MAP_ALIGNED_SUPER | MAP_PREFAULT_READ, -1, 0);
 #else
-	ptr->long_state = (uint8_t*)mmap(0, MEMORY, PROT_READ | PROT_WRITE,
+	ptr->long_state = (uint8_t*)mmap(0, hashMemSize, PROT_READ | PROT_WRITE,
 		MAP_PRIVATE | MAP_ANONYMOUS | MAP_HUGETLB | MAP_POPULATE, 0, 0);
 #endif
 
@@ -250,11 +261,11 @@ cryptonight_ctx* cryptonight_alloc_ctx(size_t use_fast_mem, size_t use_mlock, al
 
 	ptr->ctx_info[0] = 1;
 
-	if(madvise(ptr->long_state, MEMORY, MADV_RANDOM|MADV_WILLNEED) != 0)
+	if(madvise(ptr->long_state, hashMemSize, MADV_RANDOM|MADV_WILLNEED) != 0)
 		msg->warning = "madvise failed";
 
 	ptr->ctx_info[1] = 0;
-	if(use_mlock != 0 && mlock(ptr->long_state, MEMORY) != 0)
+	if(use_mlock != 0 && mlock(ptr->long_state, hashMemSize) != 0)
 		msg->warning = "mlock failed";
 	else
 		ptr->ctx_info[1] = 1;
@@ -265,14 +276,23 @@ cryptonight_ctx* cryptonight_alloc_ctx(size_t use_fast_mem, size_t use_mlock, al
 
 void cryptonight_free_ctx(cryptonight_ctx* ctx)
 {
+	size_t hashMemSize;
+	if(::jconf::inst()->IsCurrencyXMR())
+	{
+		hashMemSize = XMR_MEMORY;
+	}
+	else
+	{
+		hashMemSize = AEON_MEMORY;
+	}
 	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, MEMORY);
-		munmap(ctx->long_state, MEMORY);
+			munlock(ctx->long_state, hashMemSize);
+		munmap(ctx->long_state, hashMemSize);
 #endif // _WIN32
 	}
 	else