Removed FPNEG instruction

Optimized instruction frequencies
Increased the range for A registers from [1,65536) to [1, 4294967296)

src/AssemblyGeneratorX86.cpp

@@ -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)

src/AssemblyGeneratorX86.hpp

@@ -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);

src/CompiledVirtualMachine.cpp

@@ -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;

src/Instruction.cpp

@@ -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)

src/Instruction.hpp

@@ -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;

src/JitCompilerX86.cpp

@@ -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)

src/JitCompilerX86.hpp

@@ -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&);

src/asm/program_loop_store.inc

@@ -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

src/asm/program_prologue_load.inc

@@ -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
 

src/executeProgram-win64.asm

@@ -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
 

src/instructionWeights.hpp

@@ -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,

src/program.inc

@@ -1,768 +1,731 @@
-	; IMUL_R r0, r7
-	imul r8, r15
-	; ISMULH_R r2, r1
-	mov rax, r10
-	imul r9
-	mov r10, rdx
-	; IMUL_R r2, r4
-	imul r10, r12
-	; IADD_R r7, r0
-	add r15, r8
-	; FPSQRT_R e0
-	sqrtpd xmm4, xmm4
-	; IMUL_R r3, r6
-	imul r11, r14
-	; FPMUL_R e3, a1
-	mulpd xmm7, xmm9
-	; IMULH_M r6, L1[r3]
+	; FMUL_R e0, a2
+	mulpd xmm4, xmm10
+	; IADD_RC r2, r5, -1621224194
+	lea r10, [r10+r13-1621224194]
+	; ISTORE L2[r2], r7
+	mov eax, r10d
+	and eax, 262136
+	mov qword ptr [rsi+rax], r15
+	; FSUB_M f2, L1[r2]
+	mov eax, r10d
+	and eax, 16376
+	cvtdq2pd xmm12, qword ptr [rsi+rax]
+	subpd xmm2, xmm12
+	; IMUL_9C r6, -1003503212
+	lea r14, [r14+r14*8-1003503212]
+	; FSUB_R f1, a0
+	subpd xmm1, xmm8
+	; IXOR_M r5, L2[r3]
+	mov eax, r11d
+	and eax, 262136
+	xor r13, qword ptr [rsi+rax]
+	; FSUB_M f2, L1[r4]
+	mov eax, r12d
+	and eax, 16376
+	cvtdq2pd xmm12, qword ptr [rsi+rax]
+	subpd xmm2, xmm12
+	; FSUB_R f3, a0
+	subpd xmm3, xmm8
+	; ISDIV_C r0, 1400272688
+	mov rax, 7072565507528518045
+	imul r8
+	xor eax, eax
+	sar rdx, 29
+	sets al
+	add rdx, rax
+	add r8, rdx
+	; IMUL_M r3, L1[r7]
+	mov eax, r15d
+	and eax, 16376
+	imul r11, qword ptr [rsi+rax]
+	; IROL_R r2, r3
 	mov ecx, r11d
-	and ecx, 16376
+	rol r10, cl
+	; IMULH_R r6, r0
 	mov rax, r14
-	mul qword ptr [rsi+rcx]
+	mul r8
 	mov r14, rdx
-	; IMUL_R r5, r1
-	imul r13, r9
-	; FPADD_M f0, L2[r6]
+	; FMUL_R e0, a2
+	mulpd xmm4, xmm10
+	; IADD_RC r3, r4, -52260428
+	lea r11, [r11+r12-52260428]
+	; IADD_R r7, -1138617760
+	add r15, -1138617760
+	; IXOR_M r2, L1[r6]
 	mov eax, r14d
-	and eax, 262136
-	cvtdq2pd xmm12, qword ptr [rsi+rax]
-	addpd xmm0, xmm12
-	; IROR_R r4, r3
-	mov ecx, r11d
-	ror r12, cl
-	; IXOR_M r4, L3[984888]
-	xor r12, qword ptr [rsi+984888]
-	; IROR_R r0, r3
-	mov ecx, r11d
-	ror r8, cl
-	; IROR_R r0, r4
-	mov ecx, r12d
-	ror r8, cl
-	; FPMUL_R e0, a1
-	mulpd xmm4, xmm9
-	; IMUL_R r0, r2
-	imul r8, r10
-	; ISUB_M r0, L1[r3]
-	mov eax, r11d
 	and eax, 16376
-	sub r8, qword ptr [rsi+rax]
-	; FPSUB_R f3, a1
-	subpd xmm3, xmm9
-	; ISWAP_R r7, r4
-	xchg r15, r12
-	; IDIV_C r1, 3690475308
-	mov rax, r9
-	shr rax, 2
-	mov rcx, 5367070356934653253
-	mul rcx
-	shr rdx, 28
-	add r9, rdx
-	; IROL_R r4, r2
-	mov ecx, r10d
-	rol r12, cl
-	; IMUL_M r5, L1[r4]
-	mov eax, r12d
+	xor r10, qword ptr [rsi+rax]
+	; FSUB_R f2, a1
+	subpd xmm2, xmm9
+	; IXOR_R r7, r1
+	xor r15, r9
+	; COND_R r2, lt(r7, -41618808)
+	xor ecx, ecx
+	cmp r15d, -41618808
+	setl cl
+	add r10, rcx
+	; FMUL_R e3, a0
+	mulpd xmm7, xmm8
+	; COND_R r4, sg(r1, -961190365)
+	xor ecx, ecx
+	cmp r9d, -961190365
+	sets cl
+	add r12, rcx
+	; FADD_R f2, a1
+	addpd xmm2, xmm9
+	; FSUB_R f0, a3
+	subpd xmm0, xmm11
+	; ISTORE L1[r6], r2
+	mov eax, r14d
 	and eax, 16376
-	imul r13, qword ptr [rsi+rax]
-	; IROL_R r4, r7
-	mov ecx, r15d
-	rol r12, cl
-	; ISUB_R r3, r1
-	sub r11, r9
-	; IADD_R r7, r0
-	add r15, r8
-	; IADD_M r1, L1[r3]
-	mov eax, r11d
+	mov qword ptr [rsi+rax], r10
+	; ISUB_R r6, r5
+	sub r14, r13
+	; IADD_M r0, L1[r4]
+	mov eax, r12d
 	and eax, 16376
-	add r9, qword ptr [rsi+rax]
-	; FPMUL_R e2, a2
-	mulpd xmm6, xmm10
-	; IADD_R r6, -1115286770
-	add r14, -1115286770
-	; FPDIV_R e2, a3
-	divpd xmm6, xmm11
-	maxpd xmm6, xmm13
-	; FPADD_R f1, a2
-	addpd xmm1, xmm10
-	; IXOR_R r3, r7
-	xor r11, r15
+	add r8, qword ptr [rsi+rax]
 	; ISTORE L1[r4], r3
 	mov eax, r12d
 	and eax, 16376
 	mov qword ptr [rsi+rax], r11
-	; IROR_R r3, r6
-	mov ecx, r14d
-	ror r11, cl
-	; ISMULH_R r0, r6
-	mov rax, r8
-	imul r14
-	mov r8, rdx
-	; IROR_R r6, r5
-	mov ecx, r13d
-	ror r14, cl
-	; IMULH_M r6, L2[r0]
-	mov ecx, r8d
-	and ecx, 262136
-	mov rax, r14
-	mul qword ptr [rsi+rcx]
-	mov r14, rdx
-	; ISUB_R r2, 1512125960
-	sub r10, 1512125960
+	; COND_M r6, sg(L1[r6], 1048782623)
+	xor ecx, ecx
+	mov eax, r14d
+	and eax, 16376
+	cmp dword ptr [rsi+rax], 1048782623
+	sets cl
+	add r14, rcx
+	; FSQRT_R e0
+	sqrtpd xmm4, xmm4
+	; INEG_R r2
+	neg r10
+	; FSQRT_R e1
+	sqrtpd xmm5, xmm5
+	; FMUL_R e1, a3
+	mulpd xmm5, xmm11
 	; IMUL_R r7, r6
 	imul r15, r14
-	; IMULH_R r6, r7
-	mov rax, r14
-	mul r15
-	mov r14, rdx
-	; ISUB_R r4, r1
-	sub r12, r9
-	; FPMUL_R e3, a2
+	; IMULH_R r0, r4
+	mov rax, r8
+	mul r12
+	mov r8, rdx
+	; IMUL_R r5, r3
+	imul r13, r11
+	; FSQRT_R e2
+	sqrtpd xmm6, xmm6
+	; FADD_R f3, a0
+	addpd xmm3, xmm8
+	; IADD_R r3, r2
+	add r11, r10
+	; FADD_R f1, a0
+	addpd xmm1, xmm8
+	; FMUL_R e3, a2
 	mulpd xmm7, xmm10
-	; FPSQRT_R e1
-	sqrtpd xmm5, xmm5
-	; IXOR_R r5, r2
-	xor r13, r10
-	; FPADD_M f2, L1[r0]
-	mov eax, r8d
-	and eax, 16376
-	cvtdq2pd xmm12, qword ptr [rsi+rax]
-	addpd xmm2, xmm12
-	; IMULH_R r6, r1
-	mov rax, r14
-	mul r9
-	mov r14, rdx
-	; ISUB_M r5, L1[r0]
-	mov eax, r8d
+	; FADD_R f0, a1
+	addpd xmm0, xmm9
+	; IMUL_R r5, r6
+	imul r13, r14
+	; IADD_RC r1, r2, -1263285243
+	lea r9, [r9+r10-1263285243]
+	; ISUB_M r4, L1[r6]
+	mov eax, r14d
 	and eax, 16376
-	sub r13, qword ptr [rsi+rax]
-	; FPMUL_R e2, a3
-	mulpd xmm6, xmm11
-	; IMUL_R r4, r6
-	imul r12, r14
-	; FPMUL_R e3, a2
-	mulpd xmm7, xmm10
-	; ISUB_R r3, r2
-	sub r11, r10
-	; FPMUL_R e3, a2
-	mulpd xmm7, xmm10
-	; IROL_R r7, r0
-	mov ecx, r8d
+	sub r12, qword ptr [rsi+rax]
+	; IROL_R r7, r2
+	mov ecx, r10d
 	rol r15, cl
-	; FPSUB_R f3, a2
-	subpd xmm3, xmm10
-	; IROL_R r3, r7
-	mov ecx, r15d
-	rol r11, cl
-	; ISWAP_R r5, r7
-	xchg r13, r15
-	; IDIV_C r5, 749951529
-	mov rax, 13205547200481862341
-	mul r13
-	shr rdx, 29
-	add r13, rdx
-	; FPADD_R f3, a0
-	addpd xmm3, xmm8
-	; IMUL_M r0, L1[r4]
+	; IMUL_R r0, r7
+	imul r8, r15
+	; IXOR_R r1, r6
+	xor r9, r14
+	; IXOR_M r2, L1[r4]
 	mov eax, r12d
 	and eax, 16376
-	imul r8, qword ptr [rsi+rax]
-	; FPADD_R f1, a1
-	addpd xmm1, xmm9
-	; IROR_R r2, 60
-	ror r10, 60
-	; IROR_R r5, r4
-	mov ecx, r12d
-	ror r13, cl
-	; FPADD_R f2, a0
-	addpd xmm2, xmm8
-	; IXOR_M r4, L1[r6]
-	mov eax, r14d
-	and eax, 16376
-	xor r12, qword ptr [rsi+rax]
-	; IXOR_R r2, r6
-	xor r10, r14
-	; FPADD_M f3, L1[r0]
+	xor r10, qword ptr [rsi+rax]
+	; FSUB_R f3, a1
+	subpd xmm3, xmm9
+	; ISTORE L1[r0], r5
 	mov eax, r8d
 	and eax, 16376
+	mov qword ptr [rsi+rax], r13
+	; FDIV_M e2, L2[r3]
+	mov eax, r11d
+	and eax, 262136
 	cvtdq2pd xmm12, qword ptr [rsi+rax]
-	addpd xmm3, xmm12
-	; ISUB_R r7, r6
-	sub r15, r14
-	; IMUL_9C r2, -962375579
-	lea r10, [r10+r10*8-962375579]
-	; FPSUB_R f3, a2
-	subpd xmm3, xmm10
-	; FPSUB_R f3, a0
-	subpd xmm3, xmm8
-	; IMUL_R r1, r5
-	imul r9, r13
-	; IMUL_R r6, r4
-	imul r14, r12
-	; ISWAP_R r0, r2
-	xchg r8, r10
-	; ISUB_R r6, r5
-	sub r14, r13
-	; FPSUB_R f2, a1
-	subpd xmm2, xmm9
-	; ISDIV_C r6, 652931802
-	mov rax, -3278972671018643631
-	imul r14
-	xor eax, eax
-	add rdx, r14
-	sar rdx, 29
-	sets al
-	add rdx, rax
-	add r14, rdx
-	; IMUL_9C r5, -1142924545
-	lea r13, [r13+r13*8-1142924545]
-	; ISUB_R r7, 1085161834
-	sub r15, 1085161834
-	; IMUL_R r4, r6
-	imul r12, r14
-	; FPMUL_M e1, L1[r4]
+	andps xmm12, xmm14
+	divpd xmm6, xmm12
+	maxpd xmm6, xmm13
+	; IROL_R r2, r0
+	mov ecx, r8d
+	rol r10, cl
+	; IADD_R r7, r5
+	add r15, r13
+	; FDIV_M e0, L1[r4]
 	mov eax, r12d
 	and eax, 16376
 	cvtdq2pd xmm12, qword ptr [rsi+rax]
-	mulpd xmm5, xmm12
-	maxpd xmm5, xmm13
-	; FPMUL_M e3, L2[r1]
-	mov eax, r9d
-	and eax, 262136
-	cvtdq2pd xmm12, qword ptr [rsi+rax]
-	mulpd xmm7, xmm12
-	maxpd xmm7, xmm13
-	; COND_R r2, lt(r5, 1635027096)
-	xor ecx, ecx
-	cmp r13d, 1635027096
-	setl cl
-	add r10, rcx
-	; IMUL_R r5, -1219696062
-	imul r13, -1219696062
-	; IXOR_R r5, r0
-	xor r13, r8
-	; FPNEG_R f2
-	xorps xmm2, xmm15
-	; FPADD_R f3, a2
-	addpd xmm3, xmm10
-	; FPSUB_R f1, a3
-	subpd xmm1, xmm11
-	; FPADD_R f1, a2
-	addpd xmm1, xmm10
-	; FPDIV_R e1, a3
-	divpd xmm5, xmm11
-	maxpd xmm5, xmm13
-	; IXOR_M r6, L1[r0]
-	mov eax, r8d
-	and eax, 16376
-	xor r14, qword ptr [rsi+rax]
-	; ISUB_R r7, r4
-	sub r15, r12
-	; ISUB_M r6, L1[r1]
-	mov eax, r9d
-	and eax, 16376
-	sub r14, qword ptr [rsi+rax]
-	; ISTORE L1[r5], r3
-	mov eax, r13d
+	andps xmm12, xmm14
+	divpd xmm4, xmm12
+	maxpd xmm4, xmm13
+	; FADD_R f3, a1
+	addpd xmm3, xmm9
+	; FADD_R f0, a3
+	addpd xmm0, xmm11
+	; IADD_R r2, r0
+	add r10, r8
+	; ISTORE L1[r3], r6
+	mov eax, r11d
 	and eax, 16376
-	mov qword ptr [rsi+rax], r11
-	; IMUL_R r5, r1
-	imul r13, r9
-	; IROR_R r3, r2
-	mov ecx, r10d
-	ror r11, cl
-	; IMUL_R r4, r7
-	imul r12, r15
-	; ISDIV_C r6, -54134756
-	mov rax, 7012869325244995177
-	imul r14
+	mov qword ptr [rsi+rax], r14
+	; IXOR_R r1, r7
+	xor r9, r15
+	; ISUB_M r5, L2[r7]
+	mov eax, r15d
+	and eax, 262136
+	sub r13, qword ptr [rsi+rax]
+	; ISDIV_C r7, 266992378
+	mov rax, -9173520256920442565
+	imul r15
 	xor eax, eax
-	sub rdx, r14
-	sar rdx, 25
+	add rdx, r15
+	sar rdx, 27
 	sets al
 	add rdx, rax
-	add r14, rdx
-	; FPMUL_R e1, a2
-	mulpd xmm5, xmm10
-	; FPSUB_M f2, L2[r4]
+	add r15, rdx
+	; FDIV_M e3, L1[r4]
 	mov eax, r12d
-	and eax, 262136
+	and eax, 16376
 	cvtdq2pd xmm12, qword ptr [rsi+rax]
-	subpd xmm2, xmm12
-	; IMUL_R r0, r5
-	imul r8, r13
-	; FPMUL_R e3, a0
-	mulpd xmm7, xmm8
-	; COND_R r5, be(r4, 1545677311)
-	xor ecx, ecx
-	cmp r12d, 1545677311
-	setbe cl
-	add r13, rcx
-	; IMUL_R r6, r3
-	imul r14, r11
-	; IROL_R r6, r2
-	mov ecx, r10d
-	rol r14, cl
-	; FPDIV_R e3, a1
-	divpd xmm7, xmm9
+	andps xmm12, xmm14
+	divpd xmm7, xmm12
 	maxpd xmm7, xmm13
-	; IXOR_M r5, L1[r1]
-	mov eax, r9d
-	and eax, 16376
-	xor r13, qword ptr [rsi+rax]
-	; COND_R r3, ab(r2, 1734636060)
-	xor ecx, ecx
-	cmp r10d, 1734636060
-	seta cl
-	add r11, rcx
-	; ISTORE L1[r2], r7
-	mov eax, r10d
+	; IMUL_R r2, r0
+	imul r10, r8
+	; FMUL_R e3, a2
+	mulpd xmm7, xmm10
+	; IMUL_R r0, r6
+	imul r8, r14
+	; ISTORE L1[r0], r7
+	mov eax, r8d
 	and eax, 16376
 	mov qword ptr [rsi+rax], r15
-	; IADD_R r5, r6
-	add r13, r14
-	; FPSUB_R f1, a2
-	subpd xmm1, xmm10
-	; FPADD_R f2, a1
-	addpd xmm2, xmm9
-	; FPSWAP_R f1
-	shufpd xmm1, xmm1, 1
-	; IROL_R r2, r6
-	mov ecx, r14d
-	rol r10, cl
-	; IMUL_R r0, r4
-	imul r8, r12
-	; FPSUB_R f0, a2
-	subpd xmm0, xmm10
-	; ISUB_R r6, r7
-	sub r14, r15
-	; IROL_R r4, r7
-	mov ecx, r15d
-	rol r12, cl
-	; FPMUL_R e2, a0
-	mulpd xmm6, xmm8
-	; ISUB_R r1, r3
-	sub r9, r11
-	; FPDIV_R e0, a1
-	divpd xmm4, xmm9
-	maxpd xmm4, xmm13
-	; FPADD_R f0, a1
-	addpd xmm0, xmm9
-	; FPMUL_R e0, a2
-	mulpd xmm4, xmm10
-	; FPSUB_R f2, a2
-	subpd xmm2, xmm10
-	; FPSUB_M f2, L1[r6]
-	mov eax, r14d
-	and eax, 16376
-	cvtdq2pd xmm12, qword ptr [rsi+rax]
-	subpd xmm2, xmm12
-	; FPMUL_R e0, a0
-	mulpd xmm4, xmm8
-	; IXOR_M r4, L2[r7]
+	; FSUB_R f0, a1
+	subpd xmm0, xmm9
+	; FADD_R f3, a1
+	addpd xmm3, xmm9
+	; IXOR_R r5, r4
+	xor r13, r12
+	; ISTORE L2[r7], r2
 	mov eax, r15d
 	and eax, 262136
-	xor r12, qword ptr [rsi+rax]
-	; FPSUB_R f3, a3
-	subpd xmm3, xmm11
-	; ISMULH_R r1, r6
-	mov rax, r9
-	imul r14
-	mov r9, rdx
-	; COND_R r4, be(r7, 224524971)
-	xor ecx, ecx
-	cmp r15d, 224524971
-	setbe cl
-	add r12, rcx
-	; FPADD_M f2, L1[r1]
-	mov eax, r9d
-	and eax, 16376
-	cvtdq2pd xmm12, qword ptr [rsi+rax]
-	addpd xmm2, xmm12
-	; IMUL_R r5, r4
-	imul r13, r12
-	; IADD_RC r1, r5, 370966979
-	lea r9, [r9+r13+370966979]
-	; IADD_RC r7, r3, -1762209698
-	lea r15, [r15+r11-1762209698]
-	; FPMUL_M e3, L2[r2]
-	mov eax, r10d
-	and eax, 262136
-	cvtdq2pd xmm12, qword ptr [rsi+rax]
-	mulpd xmm7, xmm12
-	maxpd xmm7, xmm13
-	; ISUB_R r2, r7
-	sub r10, r15
-	; IMUL_9C r3, 171157280
-	lea r11, [r11+r11*8+171157280]
-	; ISUB_R r3, r5
-	sub r11, r13
-	; FPNEG_R f3
-	xorps xmm3, xmm15
-	; FPNEG_R f2
-	xorps xmm2, xmm15
-	; ISTORE L1[r4], r1
+	mov qword ptr [rsi+rax], r10
+	; ISWAP_R r6, r7
+	xchg r14, r15
+	; FADD_R f3, a2
+	addpd xmm3, xmm10
+	; ISMULH_R r5, r0
+	mov rax, r13
+	imul r8
+	mov r13, rdx
+	; IADD_M r0, L1[r4]
 	mov eax, r12d
 	and eax, 16376
-	mov qword ptr [rsi+rax], r9
-	; IADD_R r0, r2
-	add r8, r10
-	; IXOR_R r7, r6
-	xor r15, r14
-	; IROR_R r0, r4
+	add r8, qword ptr [rsi+rax]
+	; COND_R r7, ge(r6, -1972898485)
+	xor ecx, ecx
+	cmp r14d, -1972898485
+	setge cl
+	add r15, rcx
+	; FADD_R f2, a2
+	addpd xmm2, xmm10
+	; IROR_R r7, r6
+	mov ecx, r14d
+	ror r15, cl
+	; IADD_RC r2, r4, -117457973
+	lea r10, [r10+r12-117457973]
+	; IMUL_R r0, -1500893068
+	imul r8, -1500893068
+	; IADD_R r2, r3
+	add r10, r11
+	; FSQRT_R e2
+	sqrtpd xmm6, xmm6
+	; IROR_R r7, r4
 	mov ecx, r12d
-	ror r8, cl
-	; FPMUL_R e3, a2
-	mulpd xmm7, xmm10
-	; IXOR_M r4, L1[r7]
-	mov eax, r15d
-	and eax, 16376
-	xor r12, qword ptr [rsi+rax]
-	; ISTORE L1[r5], r7
-	mov eax, r13d
-	and eax, 16376
+	ror r15, cl
+	; IMUL_9C r4, 381194890
+	lea r12, [r12+r12*8+381194890]
+	; IADD_RC r3, r7, 1050899263
+	lea r11, [r11+r15+1050899263]
+	; IADD_R r2, r7
+	add r10, r15
+	; FMUL_R e3, a0
+	mulpd xmm7, xmm8
+	; IADD_RC r6, r6, 540663146
+	lea r14, [r14+r14+540663146]
+	; IROR_R r5, 58
+	ror r13, 58
+	; FSWAP_R f2
+	shufpd xmm2, xmm2, 1
+	; FSWAP_R f2
+	shufpd xmm2, xmm2, 1
+	; FMUL_R e1, a2
+	mulpd xmm5, xmm10
+	; ISWAP_R r5, r6
+	xchg r13, r14
+	; IADD_R r5, r3
+	add r13, r11
+	; IADD_R r7, -1780268176
+	add r15, -1780268176
+	; IADD_RC r7, r0, -1497756854
+	lea r15, [r15+r8-1497756854]
+	; ISTORE L2[r0], r7
+	mov eax, r8d
+	and eax, 262136
 	mov qword ptr [rsi+rax], r15
-	; IMUL_9C r7, -1206742834
-	lea r15, [r15+r15*8-1206742834]
-	; ISMULH_R r0, r4
-	mov rax, r8
+	; ISMULH_R r2, r4
+	mov rax, r10
 	imul r12
-	mov r8, rdx
-	; FPADD_R f2, a0
+	mov r10, rdx
+	; FSUB_R f0, a2
+	subpd xmm0, xmm10
+	; ISMULH_R r2, r3
+	mov rax, r10
+	imul r11
+	mov r10, rdx
+	; IADD_R r0, r3
+	add r8, r11
+	; ISUB_R r7, r2
+	sub r15, r10
+	; FADD_R f2, a0
 	addpd xmm2, xmm8
-	; FPSUB_R f1, a0
-	subpd xmm1, xmm8
-	; INEG_R r7
-	neg r15
-	; COND_M r0, of(L1[r5], -2056260506)
-	xor ecx, ecx
+	; FMUL_R e0, a2
+	mulpd xmm4, xmm10
+	; FADD_R f2, a3
+	addpd xmm2, xmm11
+	; IMUL_R r1, r2
+	imul r9, r10
+	; IMUL_M r7, L1[r5]
+	mov eax, r13d
+	and eax, 16376
+	imul r15, qword ptr [rsi+rax]
+	; IMUL_R r3, r2
+	imul r11, r10
+	; IXOR_R r1, r0
+	xor r9, r8
+	; FSUB_M f0, L1[r5]
 	mov eax, r13d
 	and eax, 16376
-	cmp dword ptr [rsi+rax], -2056260506
+	cvtdq2pd xmm12, qword ptr [rsi+rax]
+	subpd xmm0, xmm12
+	; IADD_RC r4, r4, 1456841848
+	lea r12, [r12+r12+1456841848]
+	; IXOR_R r3, r2
+	xor r11, r10
+	; COND_R r0, of(r4, 1678513610)
+	xor ecx, ecx
+	cmp r12d, 1678513610
 	seto cl
 	add r8, rcx
-	; FPSQRT_R e2
-	sqrtpd xmm6, xmm6
-	; IMUL_R r3, r4
-	imul r11, r12
-	; FPNEG_R f1
-	xorps xmm1, xmm15
-	; FPADD_M f2, L1[r5]
-	mov eax, r13d
+	; ISMULH_R r4, -1620573087
+	mov rax, -1620573087
+	imul r12
+	add r12, rdx
+	; IMUL_R r4, r1
+	imul r12, r9
+	; FSWAP_R e1
+	shufpd xmm5, xmm5, 1
+	; FADD_M f2, L1[r0]
+	mov eax, r8d
 	and eax, 16376
 	cvtdq2pd xmm12, qword ptr [rsi+rax]
 	addpd xmm2, xmm12
-	; FPSUB_R f3, a0
-	subpd xmm3, xmm8
-	; FPNEG_R f3
-	xorps xmm3, xmm15
-	; FPMUL_M e3, L2[r5]
-	mov eax, r13d
-	and eax, 262136
-	cvtdq2pd xmm12, qword ptr [rsi+rax]
-	mulpd xmm7, xmm12
-	maxpd xmm7, xmm13
-	; ISTORE L1[r2], r2
-	mov eax, r10d
-	and eax, 16376
-	mov qword ptr [rsi+rax], r10
-	; IMUL_M r3, L2[r4]
-	mov eax, r12d
-	and eax, 262136
-	imul r11, qword ptr [rsi+rax]
-	; IROL_R r5, r6
-	mov ecx, r14d
-	rol r13, cl
-	; IADD_RC r4, r3, -904431293
-	lea r12, [r12+r11-904431293]
-	; FPSUB_R f1, a1
-	subpd xmm1, xmm9
-	; IROL_R r7, r0
-	mov ecx, r8d
-	rol r15, cl
-	; ISTORE L2[r1], r7
+	; FMUL_R e1, a2
+	mulpd xmm5, xmm10
+	; FSUB_R f0, a3
+	subpd xmm0, xmm11
+	; IXOR_R r0, r7
+	xor r8, r15
+	; ISTORE L2[r1], r4
 	mov eax, r9d
 	and eax, 262136
-	mov qword ptr [rsi+rax], r15
-	; IROL_R r4, r3
-	mov ecx, r11d
-	rol r12, cl
-	; IADD_R r5, r2
-	add r13, r10
-	; COND_R r3, ge(r6, -444806705)
+	mov qword ptr [rsi+rax], r12
+	; IXOR_M r7, L1[r6]
+	mov eax, r14d
+	and eax, 16376
+	xor r15, qword ptr [rsi+rax]
+	; ISUB_R r2, r4
+	sub r10, r12
+	; ISUB_M r4, L1[r6]
+	mov eax, r14d
+	and eax, 16376
+	sub r12, qword ptr [rsi+rax]
+	; FADD_R f2, a2
+	addpd xmm2, xmm10
+	; FSUB_R f3, a0
+	subpd xmm3, xmm8
+	; IXOR_R r7, r2
+	xor r15, r10
+	; IXOR_R r0, r5
+	xor r8, r13
+	; FSWAP_R f1
+	shufpd xmm1, xmm1, 1
+	; FMUL_R e3, a2
+	mulpd xmm7, xmm10
+	; ISWAP_R r7, r1
+	xchg r15, r9
+	; ISWAP_R r1, r4
+	xchg r9, r12
+	; COND_R r2, ge(r2, -226330940)
 	xor ecx, ecx
-	cmp r14d, -444806705
+	cmp r10d, -226330940
 	setge cl
-	add r11, rcx
-	; FPADD_R f0, a1
-	addpd xmm0, xmm9
-	; IROL_R r0, 57
-	rol r8, 57
-	; IADD_R r0, r2
-	add r8, r10
-	; IADD_R r7, r4
-	add r15, r12
-	; IROL_R r1, r7
-	mov ecx, r15d
-	rol r9, cl
-	; IXOR_M r7, L2[r5]
-	mov eax, r13d
-	and eax, 262136
-	xor r15, qword ptr [rsi+rax]
-	; ISTORE L1[r2], r0
-	mov eax, r10d
+	add r10, rcx
+	; FMUL_R e2, a3
+	mulpd xmm6, xmm11
+	; FSUB_R f2, a1
+	subpd xmm2, xmm9
+	; FADD_R f1, a0
+	addpd xmm1, xmm8
+	; ISUB_R r7, r5
+	sub r15, r13
+	; ISUB_M r0, L1[r1]
+	mov eax, r9d
 	and eax, 16376
-	mov qword ptr [rsi+rax], r8
-	; FPADD_R f1, a2
-	addpd xmm1, xmm10
-	; ISUB_R r1, r4
-	sub r9, r12
-	; IXOR_R r5, r0
-	xor r13, r8
-	; IXOR_M r7, L2[r1]
+	sub r8, qword ptr [rsi+rax]
+	; FSUB_R f3, a1
+	subpd xmm3, xmm9
+	; IROL_R r3, r5
+	mov ecx, r13d
+	rol r11, cl
+	; IADD_RC r5, r2, 795784298
+	lea r13, [r13+r10+795784298]
+	; IADD_RC r0, r4, -2050178553
+	lea r8, [r8+r12-2050178553]
+	; IMUL_9C r5, 1062534001
+	lea r13, [r13+r13*8+1062534001]
+	; FADD_R f0, a2
+	addpd xmm0, xmm10
+	; FMUL_R e3, a1
+	mulpd xmm7, xmm9
+	; IDIV_C r3, 1662492575
+	mov rax, 11914062610815620875
+	mul r11
+	shr rdx, 30
+	add r11, rdx
+	; IMUL_M r5, L1[r0]
+	mov eax, r8d
+	and eax, 16376
+	imul r13, qword ptr [rsi+rax]
+	; IDIV_C r4, 1963597892
+	mov rax, r12
+	shr rax, 2
+	mov rcx, 1260889558222626443
+	mul rcx
+	shr rdx, 25
+	add r12, rdx
+	; IMUL_9C r7, 1820045218
+	lea r15, [r15+r15*8+1820045218]
+	; IMUL_M r0, L1[r3]
+	mov eax, r11d
+	and eax, 16376
+	imul r8, qword ptr [rsi+rax]
+	; IXOR_R r3, r7
+	xor r11, r15
+	; ISMULH_R r4, r2
+	mov rax, r12
+	imul r10
+	mov r12, rdx
+	; IROL_R r3, r0
+	mov ecx, r8d
+	rol r11, cl
+	; IXOR_R r2, r0
+	xor r10, r8
+	; IXOR_M r0, L2[r1]
 	mov eax, r9d
 	and eax, 262136
-	xor r15, qword ptr [rsi+rax]
-	; FPSUB_R f0, a0
-	subpd xmm0, xmm8
-	; IXOR_M r1, L1[r4]
-	mov eax, r12d
-	and eax, 16376
-	xor r9, qword ptr [rsi+rax]
-	; FPMUL_R e3, a0
-	mulpd xmm7, xmm8
-	; ISDIV_C r1, 1473744194
-	mov rax, -5006799265644655925
-	imul r9
+	xor r8, qword ptr [rsi+rax]
+	; ISDIV_C r7, -935446980
+	mov rax, 7859804860668271393
+	imul r15
 	xor eax, eax
-	add rdx, r9
-	sar rdx, 30
+	sub rdx, r15
+	sar rdx, 29
 	sets al
 	add rdx, rax
-	add r9, rdx
-	; IMUL_9C r1, 1626151459
-	lea r9, [r9+r9*8+1626151459]
-	; IXOR_M r6, L1[r4]
-	mov eax, r12d
+	add r15, rdx
+	; IMUL_M r6, L1[r2]
+	mov eax, r10d
 	and eax, 16376
-	xor r14, qword ptr [rsi+rax]
-	; FPADD_R f0, a0
-	addpd xmm0, xmm8
-	; FPADD_R f3, a2
-	addpd xmm3, xmm10
-	; ISUB_R r6, r7
-	sub r14, r15
-	; IADD_RC r1, r5, 2075955307
-	lea r9, [r9+r13+2075955307]
-	; IROL_R r6, r3
-	mov ecx, r11d
-	rol r14, cl
-	; IMULH_R r2, -1135671124
-	mov eax, -1135671124
-	mul r10
-	add r10, rdx
-	; ISUB_R r5, r2
-	sub r13, r10
-	; IMULH_R r3, r5
-	mov rax, r11
-	mul r13
-	mov r11, rdx
-	; IADD_M r4, L3[386040]
-	add r12, qword ptr [rsi+386040]
-	; COND_R r6, ge(r4, 1518758207)
-	xor ecx, ecx
-	cmp r12d, 1518758207
-	setge cl
-	add r14, rcx
-	; FPDIV_R e3, a1
-	divpd xmm7, xmm9
-	maxpd xmm7, xmm13
-	; FPNEG_R f2
-	xorps xmm2, xmm15
-	; FPADD_M f1, L1[r4]
-	mov eax, r12d
+	imul r14, qword ptr [rsi+rax]
+	; FSUB_M f3, L1[r6]
+	mov eax, r14d
 	and eax, 16376
 	cvtdq2pd xmm12, qword ptr [rsi+rax]
-	addpd xmm1, xmm12
-	; FPMUL_M e0, L1[r4]
-	mov eax, r12d
+	subpd xmm3, xmm12
+	; IADD_RC r4, r2, 1704868083
+	lea r12, [r12+r10+1704868083]
+	; FADD_R f2, a0
+	addpd xmm2, xmm8
+	; ISTORE L1[r0], r0
+	mov eax, r8d
 	and eax, 16376
-	cvtdq2pd xmm12, qword ptr [rsi+rax]
-	mulpd xmm4, xmm12
-	maxpd xmm4, xmm13
-	; FPSQRT_R e2
-	sqrtpd xmm6, xmm6
-	; IROL_R r5, r1
-	mov ecx, r9d
-	rol r13, cl
-	; FPADD_R f3, a0
-	addpd xmm3, xmm8
-	; IROL_R r3, r0
-	mov ecx, r8d
-	rol r11, cl
-	; FPMUL_R e3, a1
-	mulpd xmm7, xmm9
-	; IROR_R r0, r7
-	mov ecx, r15d
+	mov qword ptr [rsi+rax], r8
+	; FADD_R f0, a3
+	addpd xmm0, xmm11
+	; FMUL_R e0, a3
+	mulpd xmm4, xmm11
+	; FSUB_R f3, a2
+	subpd xmm3, xmm10
+	; IADD_RC r7, r7, 1302457878
+	lea r15, [r15+r15+1302457878]
+	; ISUB_R r1, 1330165941
+	sub r9, 1330165941
+	; FSUB_R f1, a3
+	subpd xmm1, xmm11
+	; IROR_R r0, r4
+	mov ecx, r12d
 	ror r8, cl
-	; FPADD_R f2, a2
-	addpd xmm2, xmm10
-	; IXOR_R r7, r0
-	xor r15, r8
-	; ISTORE L1[r4], r1
-	mov eax, r12d
-	and eax, 16376
-	mov qword ptr [rsi+rax], r9
-	; ISTORE L2[r0], r4
-	mov eax, r8d
-	and eax, 262136
-	mov qword ptr [rsi+rax], r12
-	; FPDIV_R e3, a3
-	divpd xmm7, xmm11
-	maxpd xmm7, xmm13
-	; ISTORE L2[r4], r6
-	mov eax, r12d
-	and eax, 262136
-	mov qword ptr [rsi+rax], r14
-	; IMUL_R r3, r1
-	imul r11, r9
-	; IXOR_R r2, r4
-	xor r10, r12
-	; ISTORE L2[r3], r5
-	mov eax, r11d
-	and eax, 262136
-	mov qword ptr [rsi+rax], r13
-	; FPMUL_M e2, L2[r4]
-	mov eax, r12d
-	and eax, 262136
-	cvtdq2pd xmm12, qword ptr [rsi+rax]
-	mulpd xmm6, xmm12
-	maxpd xmm6, xmm13
-	; FPSUB_R f3, a0
-	subpd xmm3, xmm8
-	; COND_R r1, ab(r7, -229570354)
+	; FSUB_R f1, a0
+	subpd xmm1, xmm8
+	; IROR_R r5, r6
+	mov ecx, r14d
+	ror r13, cl
+	; COND_R r0, ab(r1, -310933871)
 	xor ecx, ecx
-	cmp r15d, -229570354
+	cmp r9d, -310933871
 	seta cl
-	add r9, rcx
-	; IROR_R r7, r3
-	mov ecx, r11d
-	ror r15, cl
-	; FPDIV_R e2, a0
-	divpd xmm6, xmm8
-	maxpd xmm6, xmm13
-	; IADD_R r2, r5
-	add r10, r13
-	; FPDIV_R e1, a3
-	divpd xmm5, xmm11
-	maxpd xmm5, xmm13
-	; FPSQRT_R e2
-	sqrtpd xmm6, xmm6
-	; ISUB_R r3, r7
-	sub r11, r15
-	; FPADD_R f0, a0
-	addpd xmm0, xmm8
-	; IMUL_M r0, L3[98136]
-	imul r8, qword ptr [rsi+98136]
-	; IMUL_9C r5, -895487055
-	lea r13, [r13+r13*8-895487055]
-	; IMULH_R r2, r7
-	mov rax, r10
-	mul r15
-	mov r10, rdx
-	; IADD_R r4, r1
-	add r12, r9
-	; ISDIV_C r0, 494395999
-	mov rax, 5007888582388710937
-	imul r8
-	xor eax, eax
-	sar rdx, 27
-	sets al
-	add rdx, rax
-	add r8, rdx
-	; FPSWAP_R e0
-	shufpd xmm4, xmm4, 1
-	; IXOR_R r1, r5
-	xor r9, r13
-	; COND_R r2, ab(r3, 1932234501)
+	add r8, rcx
+	; COND_R r4, ab(r7, 757929676)
 	xor ecx, ecx
-	cmp r11d, 1932234501
+	cmp r15d, 757929676
 	seta cl
-	add r10, rcx
-	; FPMUL_R e1, a0
-	mulpd xmm5, xmm8
-	; FPSUB_M f1, L1[r1]
+	add r12, rcx
+	; FMUL_R e0, a1
+	mulpd xmm4, xmm9
+	; IMUL_R r1, r3
+	imul r9, r11
+	; ISUB_R r3, r2
+	sub r11, r10
+	; FSUB_R f3, a2
+	subpd xmm3, xmm10
+	; FDIV_M e1, L1[r4]
+	mov eax, r12d
+	and eax, 16376
+	cvtdq2pd xmm12, qword ptr [rsi+rax]
+	andps xmm12, xmm14
+	divpd xmm5, xmm12
+	maxpd xmm5, xmm13
+	; IROL_R r1, 5
+	rol r9, 5
+	; IADD_R r7, -1421188024
+	add r15, -1421188024
+	; FSUB_R f3, a2
+	subpd xmm3, xmm10
+	; FSUB_R f2, a3
+	subpd xmm2, xmm11
+	; FADD_M f3, L1[r1]
 	mov eax, r9d
 	and eax, 16376
 	cvtdq2pd xmm12, qword ptr [rsi+rax]
-	subpd xmm1, xmm12
-	; FPSUB_R f0, a0
-	subpd xmm0, xmm8
-	; IROL_R r1, r7
-	mov ecx, r15d
-	rol r9, cl
-	; IADD_RC r0, r5, -2051588680
-	lea r8, [r8+r13-2051588680]
-	; COND_R r6, of(r5, -795593984)
+	addpd xmm3, xmm12
+	; FMUL_R e1, a3
+	mulpd xmm5, xmm11
+	; IADD_RC r2, r4, -317832028
+	lea r10, [r10+r12-317832028]
+	; IMUL_M r4, L1[r5]
+	mov eax, r13d
+	and eax, 16376
+	imul r12, qword ptr [rsi+rax]
+	; FDIV_M e1, L1[r7]
+	mov eax, r15d
+	and eax, 16376
+	cvtdq2pd xmm12, qword ptr [rsi+rax]
+	andps xmm12, xmm14
+	divpd xmm5, xmm12
+	maxpd xmm5, xmm13
+	; IADD_R r5, r2
+	add r13, r10
+	; ISUB_R r4, 401020510
+	sub r12, 401020510
+	; IROR_R r3, r0
+	mov ecx, r8d
+	ror r11, cl
+	; ISTORE L1[r7], r0
+	mov eax, r15d
+	and eax, 16376
+	mov qword ptr [rsi+rax], r8
+	; FSUB_R f2, a1
+	subpd xmm2, xmm9
+	; FMUL_R e3, a1
+	mulpd xmm7, xmm9
+	; IMUL_9C r3, 720965215
+	lea r11, [r11+r11*8+720965215]
+	; IMUL_9C r6, 74948046
+	lea r14, [r14+r14*8+74948046]
+	; ISTORE L1[r7], r3
+	mov eax, r15d
+	and eax, 16376
+	mov qword ptr [rsi+rax], r11
+	; IXOR_R r2, r6
+	xor r10, r14
+	; FMUL_R e3, a1
+	mulpd xmm7, xmm9
+	; ISUB_R r4, r1
+	sub r12, r9
+	; ISUB_R r3, r0
+	sub r11, r8
+	; IROL_R r7, r5
+	mov ecx, r13d
+	rol r15, cl
+	; IMUL_R r2, r6
+	imul r10, r14
+	; COND_R r2, ge(r2, -1892157506)
 	xor ecx, ecx
-	cmp r13d, -795593984
-	seto cl
-	add r14, rcx
-	; FPADD_R f1, a0
-	addpd xmm1, xmm8
-	; IMULH_R r7, r3
-	mov rax, r15
-	mul r11
-	mov r15, rdx
-	; ISUB_R r7, r4
-	sub r15, r12
-	; IROL_R r0, r6
-	mov ecx, r14d
-	rol r8, cl
-	; ISDIV_C r1, -675825513
-	mov rax, -7326980207007250257
-	imul r9
+	cmp r10d, -1892157506
+	setge cl
+	add r10, rcx
+	; FADD_R f1, a3
+	addpd xmm1, xmm11
+	; IADD_R r7, r0
+	add r15, r8
+	; IDIV_C r1, 624867857
+	mov rax, 15848983434401622933
+	mul r9
+	shr rdx, 29
+	add r9, rdx
+	; FADD_R f0, a1
+	addpd xmm0, xmm9
+	; IADD_RC r5, r7, -477591118
+	lea r13, [r13+r15-477591118]
+	; FSUB_R f0, a3
+	subpd xmm0, xmm11
+	; ISUB_M r6, L1[r2]
+	mov eax, r10d
+	and eax, 16376
+	sub r14, qword ptr [rsi+rax]
+	; FMUL_R e3, a1
+	mulpd xmm7, xmm9
+	; IADD_R r0, r4
+	add r8, r12
+	; FSUB_R f3, a1
+	subpd xmm3, xmm9
+	; FSUB_R f2, a0
+	subpd xmm2, xmm8
+	; ISDIV_C r2, -396711688
+	mov rax, 5964731804029407733
+	imul r10
 	xor eax, eax
+	sub rdx, r10
 	sar rdx, 28
 	sets al
 	add rdx, rax
-	add r9, rdx
-	; ISTORE L1[r6], r3
-	mov eax, r14d
-	and eax, 16376
-	mov qword ptr [rsi+rax], r11
-	; IROR_R r4, r3
-	mov ecx, r11d
-	ror r12, cl
-	; IDIV_C r4, 3919226376
-	mov rax, r12
-	shr rax, 3
-	mov rcx, 2526906936258851663
-	mul rcx
-	shr rdx, 26
-	add r12, rdx
-	; FPSUB_R f1, a1
-	subpd xmm1, xmm9
-	; FPSUB_R f0, a0
+	add r10, rdx
+	; FSUB_R f2, a2
+	subpd xmm2, xmm10
+	; FSUB_R f3, a2
+	subpd xmm3, xmm10
+	; FADD_R f1, a3
+	addpd xmm1, xmm11
+	; IMUL_R r3, r2
+	imul r11, r10
+	; FADD_R f0, a3
+	addpd xmm0, xmm11
+	; ISMULH_R r5, r2
+	mov rax, r13
+	imul r10
+	mov r13, rdx
+	; IMULH_R r6, r2
+	mov rax, r14
+	mul r10
+	mov r14, rdx
+	; FADD_R f3, a3
+	addpd xmm3, xmm11
+	; IMUL_R r6, r7
+	imul r14, r15
+	; FSUB_R f0, a0
 	subpd xmm0, xmm8
-	; IADD_R r0, r2
-	add r8, r10
-	; IADD_M r4, L1[r2]
-	mov eax, r10d
+	; FSUB_R f2, a0
+	subpd xmm2, xmm8
+	; ISUB_R r6, r4
+	sub r14, r12
+	; FSWAP_R f1
+	shufpd xmm1, xmm1, 1
+	; IXOR_R r0, r5
+	xor r8, r13
+	; FADD_R f2, a1
+	addpd xmm2, xmm9
+	; IROL_R r7, r5
+	mov ecx, r13d
+	rol r15, cl
+	; FMUL_R e3, a2
+	mulpd xmm7, xmm10
+	; IADD_RC r3, r6, -1317630728
+	lea r11, [r11+r14-1317630728]
+	; IMUL_R r2, r3
+	imul r10, r11
+	; IADD_RC r1, r4, 894105694
+	lea r9, [r9+r12+894105694]
+	; IMUL_9C r7, 504293473
+	lea r15, [r15+r15*8+504293473]
+	; FADD_M f1, L2[r0]
+	mov eax, r8d
+	and eax, 262136
+	cvtdq2pd xmm12, qword ptr [rsi+rax]
+	addpd xmm1, xmm12
+	; IMUL_R r7, r1
+	imul r15, r9
+	; IXOR_R r2, r4
+	xor r10, r12
+	; IADD_RC r0, r1, 392362094
+	lea r8, [r8+r9+392362094]
+	; IDIV_C r4, 1645771433
+	mov rax, 376097195048767223
+	mul r12
+	shr rdx, 25
+	add r12, rdx
+	; ISUB_R r4, r3
+	sub r12, r11
+	; ISUB_M r7, L1[r4]
+	mov eax, r12d
 	and eax, 16376
-	add r12, qword ptr [rsi+rax]
-	; ISTORE L1[r7], r2
+	sub r15, qword ptr [rsi+rax]
+	; IMUL_M r5, L1[r7]
 	mov eax, r15d
 	and eax, 16376
-	mov qword ptr [rsi+rax], r10
-	; FPSQRT_R e1
-	sqrtpd xmm5, xmm5
-	; IADD_R r5, r4
-	add r13, r12
-	; IXOR_R r6, r7
-	xor r14, r15
-	; ISMULH_R r4, r7
-	mov rax, r12
-	imul r15
-	mov r12, rdx
-	; FPSQRT_R e1
-	sqrtpd xmm5, xmm5
+	imul r13, qword ptr [rsi+rax]
+	; IROR_R r1, r7
+	mov ecx, r15d
+	ror r9, cl
+	; INEG_R r4
+	neg r12
+	; IMUL_R r3, 1863959234
+	imul r11, 1863959234
+	; IROR_R r4, 59
+	ror r12, 59
+	; IMUL_M r1, L3[363256]
+	imul r9, qword ptr [rsi+363256]
+	; ISTORE L2[r6], r7
+	mov eax, r14d
+	and eax, 262136
+	mov qword ptr [rsi+rax], r15
+	; ISTORE L1[r1], r5
+	mov eax, r9d
+	and eax, 16376
+	mov qword ptr [rsi+rax], r13
+	; FSUB_M f0, L1[r1]
+	mov eax, r9d
+	and eax, 16376
+	cvtdq2pd xmm12, qword ptr [rsi+rax]
+	subpd xmm0, xmm12
+	; FSQRT_R e2
+	sqrtpd xmm6, xmm6
+	; FMUL_R e0, a3
+	mulpd xmm4, xmm11
+	; FMUL_R e3, a2
+	mulpd xmm7, xmm10
+	; IROL_R r5, r2
+	mov ecx, r10d
+	rol r13, cl
+	; IADD_R r0, r4
+	add r8, r12