github-wownero/RandomWOW
1
0
Fork 0
mirror of https://github.com/wownero/RandomWOW.git synced 2026-10-17 23:41:36 +02:00
Code Issues Packages Projects Releases Wiki Activity

Interpreter with bytecode

Browse Source 
Fixed some undefined behavior with signed types
Fixed different results on big endian systems
Removed unused code files
Restored FNEG_R instructions
Updated documentation
This commit is contained in:
tevador
2019-02-09 15:45:26 +01:00
parent a586751f6b
commit 32d827d0a6
41 changed files with 1517 additions and 3621 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/instructionsPortable.cpp
+49 -130
View File
@@ -18,12 +18,14 @@ along with RandomX. If not, see<http://www.gnu.org/licenses/>.
*/
//#define DEBUG
#include "intrinPortable.h"
#include "blake2/endian.h"
#pragma STDC FENV_ACCESS on
#include <cfenv>
#include <cmath>
#ifdef DEBUG
#include <iostream>
#endif
#include "common.hpp"
#if defined(__SIZEOF_INT128__)
typedef unsigned __int128 uint128_t;
@@ -136,18 +138,18 @@ static inline int32_t safeSub(int32_t a, int32_t b) {
#if defined(__has_builtin)
#if __has_builtin(__builtin_sub_overflow)
static inline bool subOverflow__(int32_t a, int32_t b) {
static inline bool subOverflow__(uint32_t a, uint32_t b) {
int32_t temp;
return __builtin_sub_overflow(a, b, &temp);
return __builtin_sub_overflow(unsigned32ToSigned2sCompl(a), unsigned32ToSigned2sCompl(b), &temp);
}
#define HAVE_SUB_OVERFLOW
#endif
#endif
#ifndef HAVE_SUB_OVERFLOW
static inline bool subOverflow__(int32_t a, int32_t b) {
auto c = safeSub(a, b);
return (c < a) != (b > 0);
static inline bool subOverflow__(uint32_t a, uint32_t b) {
auto c = unsigned32ToSigned2sCompl(a - b);
return (c < unsigned32ToSigned2sCompl(a)) != (unsigned32ToSigned2sCompl(b) > 0);
}
#define HAVE_SUB_OVERFLOW
#endif
@@ -166,40 +168,44 @@ static inline double FlushNaN(double x) {
void setRoundMode(uint32_t rcflag) {
switch (rcflag & 3) {
case RoundDown:
setRoundMode__(FE_DOWNWARD);
break;
case RoundUp:
setRoundMode__(FE_UPWARD);
break;
case RoundToZero:
setRoundMode__(FE_TOWARDZERO);
break;
default:
setRoundMode__(FE_TONEAREST);
break;
case RoundDown:
setRoundMode__(FE_DOWNWARD);
break;
case RoundUp:
setRoundMode__(FE_UPWARD);
break;
case RoundToZero:
setRoundMode__(FE_TOWARDZERO);
break;
case RoundToNearest:
setRoundMode__(FE_TONEAREST);
break;
default:
UNREACHABLE;
}
}
bool condition(uint32_t type, int32_t value, int32_t imm32) {
bool condition(uint32_t type, uint32_t value, uint32_t imm32) {
switch (type & 7)
{
case 0:
return (uint32_t)value <= (uint32_t)imm32;
case 1:
return (uint32_t)value > (uint32_t)imm32;
case 2:
return safeSub(value, imm32) < 0;
case 3:
return safeSub(value, imm32) >= 0;
case 4:
return subOverflow__(value, imm32);
case 5:
return !subOverflow__(value, imm32);
case 6:
return value < imm32;
case 7:
return value >= imm32;
case 0:
return value <= imm32;
case 1:
return value > imm32;
case 2:
return unsigned32ToSigned2sCompl(value - imm32) < 0;
case 3:
return unsigned32ToSigned2sCompl(value - imm32) >= 0;
case 4:
return subOverflow__(value, imm32);
case 5:
return !subOverflow__(value, imm32);
case 6:
return unsigned32ToSigned2sCompl(value) < unsigned32ToSigned2sCompl(imm32);
case 7:
return unsigned32ToSigned2sCompl(value) >= unsigned32ToSigned2sCompl(imm32);
default:
UNREACHABLE;
}
}
@@ -211,100 +217,13 @@ void initFpu() {
#endif
}
namespace RandomX {
union double_ser_t {
double f;
uint64_t i;
};
extern "C" {
/*void DIV_64(convertible_t& a, convertible_t& b, convertible_t& c) {
c.u64 = a.u64 / (b.u32 != 0 ? b.u32 : 1U);
}
void IDIV_64(convertible_t& a, convertible_t& b, convertible_t& c) {
if (a.i64 == INT64_MIN && b.i32 == -1)
c.i64 = INT64_MIN;
else
c.i64 = a.i64 / (b.i32 != 0 ? b.i32 : 1);
}
void FADD(convertible_t& a, fpu_reg_t& b, fpu_reg_t& c) {
#ifdef __SSE2__
__m128i ai = _mm_loadl_epi64((const __m128i*)&a);
__m128d ad = _mm_cvtepi32_pd(ai);
__m128d bd = _mm_load_pd(&b.lo.f64);
__m128d cd = _mm_add_pd(ad, bd);
_mm_store_pd(&c.lo.f64, cd);
#else
double alo = (double)a.i32lo;
double ahi = (double)a.i32hi;
c.lo.f64 = alo + b.lo.f64;
c.hi.f64 = ahi + b.hi.f64;
#endif
}
void FSUB(convertible_t& a, fpu_reg_t& b, fpu_reg_t& c) {
#ifdef __SSE2__
__m128i ai = _mm_loadl_epi64((const __m128i*)&a);
__m128d ad = _mm_cvtepi32_pd(ai);
__m128d bd = _mm_load_pd(&b.lo.f64);
__m128d cd = _mm_sub_pd(ad, bd);
_mm_store_pd(&c.lo.f64, cd);
#else
double alo = (double)a.i32lo;
double ahi = (double)a.i32hi;
c.lo.f64 = alo - b.lo.f64;
c.hi.f64 = ahi - b.hi.f64;
#endif
}
void FMUL(convertible_t& a, fpu_reg_t& b, fpu_reg_t& c) {
#ifdef __SSE2__
__m128i ai = _mm_loadl_epi64((const __m128i*)&a);
__m128d ad = _mm_cvtepi32_pd(ai);
__m128d bd = _mm_load_pd(&b.lo.f64);
__m128d cd = _mm_mul_pd(ad, bd);
__m128d mask = _mm_cmpeq_pd(cd, cd);
cd = _mm_and_pd(cd, mask);
_mm_store_pd(&c.lo.f64, cd);
#else
double alo = (double)a.i32lo;
double ahi = (double)a.i32hi;
c.lo.f64 = FlushNaN(alo * b.lo.f64);
c.hi.f64 = FlushNaN(ahi * b.hi.f64);
#endif
}
void FDIV(convertible_t& a, fpu_reg_t& b, fpu_reg_t& c) {
#ifdef __SSE2__
__m128i ai = _mm_loadl_epi64((const __m128i*)&a);
__m128d ad = _mm_cvtepi32_pd(ai);
__m128d bd = _mm_load_pd(&b.lo.f64);
__m128d cd = _mm_div_pd(ad, bd);
__m128d mask = _mm_cmpeq_pd(cd, cd);
cd = _mm_and_pd(cd, mask);
_mm_store_pd(&c.lo.f64, cd);
#else
double alo = (double)a.i32lo;
double ahi = (double)a.i32hi;
c.lo.f64 = FlushDenormalNaN(alo / b.lo.f64);
c.hi.f64 = FlushDenormalNaN(ahi / b.hi.f64);
#endif
}
void FSQRT(convertible_t& a, fpu_reg_t& b, fpu_reg_t& c) {
#ifdef __SSE2__
__m128i ai = _mm_loadl_epi64((const __m128i*)&a);
__m128d ad = _mm_cvtepi32_pd(ai);
const __m128d absmask = _mm_castsi128_pd(_mm_set1_epi64x(~(1LL << 63)));
ad = _mm_and_pd(ad, absmask);
__m128d cd = _mm_sqrt_pd(ad);
_mm_store_pd(&c.lo.f64, cd);
#else
double alo = (double)a.i32lo;
double ahi = (double)a.i32hi;
c.lo.f64 = sqrt(std::abs(alo));
c.hi.f64 = sqrt(std::abs(ahi));
#endif
}*/
}
}
double loadDoublePortable(const void* addr) {
double_ser_t ds;
ds.i = load64(addr);
return ds.f;
}