d0s3t3ch/RandomWOW
1
0
Fork 0
mirror of https://git.wownero.com/wownero/RandomWOW.git synced 2024-08-15 00:23:14 +00:00
Code Issues Projects Releases Wiki Activity

Removed FPNEG instruction

Browse source 
Optimized instruction frequencies
Increased the range for A registers from [1,65536) to [1, 4294967296)
This commit is contained in:
tevador 2019-02-07 16:11:27 +01:00
parent ac4462ad42
commit a586751f6b
12 changed files with 738 additions and 759 deletions
Download patch file Download diff file Expand all files Collapse all files

10
src/AssemblyGeneratorX86.cpp
View file

@ -35,6 +35,8 @@ namespace RandomX {
static const char* regE[4] = { "xmm4", "xmm5", "xmm6", "xmm7" };
static const char* regA[4] = { "xmm8", "xmm9", "xmm10", "xmm11" };
static const char* fsumInstr[4] = { "paddb", "paddw", "paddd", "paddq" };
static const char* regA4 = "xmm12";
static const char* dblMin = "xmm13";
static const char* absMask = "xmm14";
@ -365,6 +367,7 @@ namespace RandomX {
instr.dst %= 4;
instr.src %= 4;
asmCode << "\taddpd " << regF[instr.dst] << ", " << regA[instr.src] << std::endl;
//asmCode << "\t" << fsumInstr[instr.mod % 4] << " " << signMask << ", " << regF[instr.dst] << std::endl;
}
//5 uOPs
@ -380,6 +383,7 @@ namespace RandomX {
instr.dst %= 4;
instr.src %= 4;
asmCode << "\tsubpd " << regF[instr.dst] << ", " << regA[instr.src] << std::endl;
//asmCode << "\t" << fsumInstr[instr.mod % 4] << " " << signMask << ", " << regF[instr.dst] << std::endl;
}
//5 uOPs
@ -391,9 +395,9 @@ namespace RandomX {
}
//1 uOP
void AssemblyGeneratorX86::h_FPNEG_R(Instruction& instr, int i) {
void AssemblyGeneratorX86::h_CFSUM_R(Instruction& instr, int i) {
instr.dst %= 4;
asmCode << "\txorps " << regF[instr.dst] << ", " << signMask << std::endl;
asmCode << "\t" << fsumInstr[instr.mod % 4] << " " << signMask << ", " << regF[instr.dst] << std::endl;
}
//1 uOPs
@ -538,7 +542,7 @@ namespace RandomX {
INST_HANDLE(FADD_M)
INST_HANDLE(FSUB_R)
INST_HANDLE(FSUB_M)
INST_HANDLE(FPNEG_R)
INST_HANDLE(CFSUM_R)
//Floating point group E
INST_HANDLE(FMUL_R)

2
src/AssemblyGeneratorX86.hpp
View file

@ -69,7 +69,7 @@ namespace RandomX {
void h_FADD_M(Instruction&, int);
void h_FSUB_R(Instruction&, int);
void h_FSUB_M(Instruction&, int);
void h_FPNEG_R(Instruction&, int);
void h_CFSUM_R(Instruction&, int);
void h_FMUL_R(Instruction&, int);
void h_FMUL_M(Instruction&, int);
void h_FDIV_R(Instruction&, int);

2
src/CompiledVirtualMachine.cpp
View file

@ -44,7 +44,7 @@ namespace RandomX {
}
static uint64_t getSmallPositiveFloatBits(uint64_t entropy) {
auto exponent = entropy >> 60; //0..15
auto exponent = entropy >> 59; //0..31
auto mantissa = entropy & mantissaMask;
exponent += exponentBias;
exponent &= exponentMask;

8
src/Instruction.cpp
View file

@ -247,9 +247,9 @@ namespace RandomX {
os << std::endl;
}
void Instruction::h_FPNEG_R(std::ostream& os) const {
void Instruction::h_CFSUM_R(std::ostream& os) const {
auto dstIndex = dst % 4;
os << "f" << dstIndex << std::endl;
os << "f" << dstIndex << ", " << (1 << ((mod % 4) + 3)) << std::endl;
}
void Instruction::h_FMUL_R(std::ostream& os) const {
@ -370,7 +370,7 @@ namespace RandomX {
INST_NAME(FADD_M)
INST_NAME(FSUB_R)
INST_NAME(FSUB_M)
INST_NAME(FPNEG_R)
INST_NAME(CFSUM_R)
//Floating point group E
INST_NAME(FMUL_R)
@ -421,7 +421,7 @@ namespace RandomX {
INST_HANDLE(FADD_M)
INST_HANDLE(FSUB_R)
INST_HANDLE(FSUB_M)
INST_HANDLE(FPNEG_R)
INST_HANDLE(CFSUM_R)
//Floating point group E
INST_HANDLE(FMUL_R)

4
src/Instruction.hpp
View file

@ -54,7 +54,7 @@ namespace RandomX {
constexpr int FADD_M = 22;
constexpr int FSUB_R = 23;
constexpr int FSUB_M = 24;
constexpr int FPNEG_R = 25;
constexpr int CFSUM_R = 25;
constexpr int FMUL_R = 26;
constexpr int FMUL_M = 27;
constexpr int FDIV_R = 28;
@ -116,7 +116,7 @@ namespace RandomX {
void h_FADD_M(std::ostream&) const;
void h_FSUB_R(std::ostream&) const;
void h_FSUB_M(std::ostream&) const;
void h_FPNEG_R(std::ostream&) const;
void h_CFSUM_R(std::ostream&) const;
void h_FMUL_R(std::ostream&) const;
void h_FMUL_M(std::ostream&) const;
void h_FDIV_R(std::ostream&) const;

14
src/JitCompilerX86.cpp
View file

@ -87,7 +87,7 @@ namespace RandomX {
; xmm12 -> temporary
; xmm13 -> DBL_MIN
; xmm14 -> absolute value mask 0x7fffffffffffffff7fffffffffffffff
; xmm15 -> sign mask 0x80000000000000008000000000000000
; xmm15 -> unused
*/
@ -178,6 +178,8 @@ namespace RandomX {
static const uint8_t REX_XOR_RAX_R64[] = { 0x49, 0x33 };
static const uint8_t REX_XCHG[] = { 0x4d, 0x87 };
static const uint8_t REX_ANDPS_XMM12[] = { 0x41, 0x0f, 0x54, 0xe6 };
static const uint8_t REX_PADD[] = { 0x66, 0x44, 0x0f };
static const uint8_t PADD_OPCODES[] = { 0xfc, 0xfd, 0xfe, 0xd4 };
size_t JitCompilerX86::getCodeSize() {
return codePos - prologueSize;
@ -615,6 +617,9 @@ namespace RandomX {
instr.src %= 4;
emit(REX_ADDPD);
emitByte(0xc0 + instr.src + 8 * instr.dst);
//emit(REX_PADD);
//emitByte(PADD_OPCODES[instr.mod % 4]);
//emitByte(0xf8 + instr.dst);
}
void JitCompilerX86::h_FADD_M(Instruction& instr) {
@ -630,6 +635,9 @@ namespace RandomX {
instr.src %= 4;
emit(REX_SUBPD);
emitByte(0xc0 + instr.src + 8 * instr.dst);
//emit(REX_PADD);
//emitByte(PADD_OPCODES[instr.mod % 4]);
//emitByte(0xf8 + instr.dst);
}
void JitCompilerX86::h_FSUB_M(Instruction& instr) {
@ -640,7 +648,7 @@ namespace RandomX {
emitByte(0xc4 + 8 * instr.dst);
}
void JitCompilerX86::h_FPNEG_R(Instruction& instr) {
void JitCompilerX86::h_CFSUM_R(Instruction& instr) {
instr.dst %= 4;
emit(REX_XORPS);
emitByte(0xc7 + 8 * instr.dst);
@ -794,7 +802,7 @@ namespace RandomX {
INST_HANDLE(FADD_M)
INST_HANDLE(FSUB_R)
INST_HANDLE(FSUB_M)
INST_HANDLE(FPNEG_R)
INST_HANDLE(CFSUM_R)
INST_HANDLE(FMUL_R)
INST_HANDLE(FMUL_M)
INST_HANDLE(FDIV_R)

2
src/JitCompilerX86.hpp
View file

@ -115,7 +115,7 @@ namespace RandomX {
void h_FADD_M(Instruction&);
void h_FSUB_R(Instruction&);
void h_FSUB_M(Instruction&);
void h_FPNEG_R(Instruction&);
void h_CFSUM_R(Instruction&);
void h_FMUL_R(Instruction&);
void h_FMUL_M(Instruction&);
void h_FDIV_R(Instruction&);

4
src/asm/program_loop_store.inc
View file

@ -12,6 +12,10 @@
mulpd xmm1, xmm5
mulpd xmm2, xmm6
mulpd xmm3, xmm7
;# xorpd xmm0, xmm15
;# xorpd xmm1, xmm15
;# xorpd xmm2, xmm15
;# xorpd xmm3, xmm15
movapd xmmword ptr [rcx+0], xmm0
movapd xmmword ptr [rcx+16], xmm1
movapd xmmword ptr [rcx+32], xmm2

2
src/asm/program_prologue_load.inc
View file

@ -18,5 +18,5 @@
movapd xmm11, xmmword ptr [rcx+120]
movapd xmm13, xmmword ptr [minDbl]
movapd xmm14, xmmword ptr [absMask]
movapd xmm15, xmmword ptr [signMask]
;# xorpd xmm15, xmm15

4
src/executeProgram-win64.asm
View file

@ -54,7 +54,7 @@ executeProgram PROC
; xmm12 -> temporary
; xmm13 -> DBL_MIN
; xmm14 -> absolute value mask
; xmm15 -> sign mask
; xmm15 -> unused
; store callee-saved registers
push rbx
@ -104,7 +104,7 @@ executeProgram PROC
movapd xmm11, xmmword ptr [rcx+120]
movapd xmm13, xmmword ptr [minDbl]
movapd xmm14, xmmword ptr [absMask]
movapd xmm15, xmmword ptr [signMask]
;# xorps xmm15, xmm15
jmp program_begin

26
src/instructionWeights.hpp
View file

@ -21,10 +21,10 @@ along with RandomX. If not, see<http://www.gnu.org/licenses/>.
//Integer
#define WT_IADD_R 12
#define WT_IADD_M 3
#define WT_IADD_RC 12
#define WT_IADD_M 7
#define WT_IADD_RC 16
#define WT_ISUB_R 12
#define WT_ISUB_M 3
#define WT_ISUB_M 7
#define WT_IMUL_9C 9
#define WT_IMUL_R 16
#define WT_IMUL_M 4
@ -35,10 +35,10 @@ along with RandomX. If not, see<http://www.gnu.org/licenses/>.
#define WT_IDIV_C 4
#define WT_ISDIV_C 4
#define WT_INEG_R 2
#define WT_IXOR_R 12
#define WT_IXOR_R 16
#define WT_IXOR_M 4
#define WT_IROR_R 10
#define WT_IROL_R 10
#define WT_IROR_R 8
#define WT_IROL_R 8
#define WT_ISWAP_R 4
//Common floating point
@ -49,22 +49,22 @@ along with RandomX. If not, see<http://www.gnu.org/licenses/>.
#define WT_FADD_M 5
#define WT_FSUB_R 20
#define WT_FSUB_M 5
#define WT_FPNEG_R 6
//Floating point group E
#define WT_FMUL_R 16
#define WT_FMUL_M 4
#define WT_FDIV_R 7
#define WT_FDIV_M 1
#define WT_FMUL_R 20
#define WT_FMUL_M 0
#define WT_FDIV_R 0
#define WT_FDIV_M 4
#define WT_FSQRT_R 6
//Control
#define WT_COND_R 7
#define WT_COND_M 1
#define WT_CFROUND 1
#define WT_CFSUM_R 0
//Store
#define WT_ISTORE 18
#define WT_ISTORE 16
#define WT_FSTORE 0
#define WT_NOP 0
@ -74,7 +74,7 @@ WT_ISUB_M + WT_IMUL_9C + WT_IMUL_R + WT_IMUL_M + WT_IMULH_R + \
WT_IMULH_M + WT_ISMULH_R + WT_ISMULH_M + WT_IDIV_C + WT_ISDIV_C + \
WT_INEG_R + WT_IXOR_R + WT_IXOR_M + WT_IROR_R + WT_IROL_R + \
WT_ISWAP_R + WT_FSWAP_R + WT_FADD_R + WT_FADD_M + WT_FSUB_R + WT_FSUB_M + \
WT_FPNEG_R + WT_FMUL_R + WT_FMUL_M + WT_FDIV_R + WT_FDIV_M + \
WT_CFSUM_R + WT_FMUL_R + WT_FMUL_M + WT_FDIV_R + WT_FDIV_M + \
WT_FSQRT_R + WT_COND_R + WT_COND_M + WT_CFROUND + WT_ISTORE + WT_FSTORE + WT_NOP;
static_assert(wtSum == 256,

1419
src/program.inc
View file

File diff suppressed because it is too large Load diff