16 -> 8 chained programs

constant address loads are always from L3

src/AssemblyGeneratorX86.cpp

@@ -81,7 +81,7 @@ namespace RandomX {
 	}
 
 	int32_t AssemblyGeneratorX86::genAddressImm(Instruction& instr) {
-		return instr.imm32 & ((instr.mod % 4) ? ScratchpadL1Mask : ScratchpadL2Mask);
+		return instr.imm32 & ScratchpadL3Mask;
 	}
 
 	//1 uOP

src/Instruction.cpp

@@ -37,7 +37,7 @@ namespace RandomX {
 	}
 
 	void Instruction::genAddressImm(std::ostream& os) const {
-		os << ((mod % 4) ? "L1" : "L2") << "[" << (imm32 & ((mod % 4) ? ScratchpadL1Mask : ScratchpadL2Mask)) << "]";
+		os << "L3" << "[" << (imm32 & ScratchpadL3Mask) << "]";
 	}
 
 	void Instruction::h_IADD_R(std::ostream& os) const {

src/JitCompilerX86.cpp

@@ -262,7 +262,7 @@ namespace RandomX {
 	}
 
 	void JitCompilerX86::genAddressImm(Instruction& instr) {
-		emit32(instr.imm32 & ((instr.mod % 4) ? ScratchpadL1Mask : ScratchpadL2Mask));
+		emit32(instr.imm32 & ScratchpadL3Mask);
 	}
 
 	void JitCompilerX86::h_IADD_R(Instruction& instr) {

src/common.hpp

@@ -68,7 +68,7 @@ namespace RandomX {
 	};
 
 	constexpr int ProgramLength = 256;
-	constexpr uint32_t InstructionCount = 1024;
+	constexpr uint32_t InstructionCount = 2048;
 	constexpr uint32_t ScratchpadSize = 2 * 1024 * 1024;
 	constexpr uint32_t ScratchpadLength = ScratchpadSize / sizeof(convertible_t);
 	constexpr uint32_t ScratchpadL1 = ScratchpadSize / 128 / sizeof(convertible_t);
@@ -78,6 +78,7 @@ namespace RandomX {
 	constexpr int ScratchpadL2Mask = (ScratchpadL2 - 1) * 8;
 	constexpr int ScratchpadL1Mask16 = (ScratchpadL1 / 2 - 1) * 16;
 	constexpr int ScratchpadL2Mask16 = (ScratchpadL2 / 2 - 1) * 16;
+	constexpr int ScratchpadL3Mask = (ScratchpadLength - 1) * 8;
 	constexpr uint32_t TransformationCount = 90;
 	constexpr int RegistersCount = 8;
 

src/instructionWeights.hpp

@@ -25,7 +25,7 @@ along with RandomX.  If not, see<http://www.gnu.org/licenses/>.
 #define WT_IADD_RC 12
 #define WT_ISUB_R 12
 #define WT_ISUB_M 3
-#define WT_IMUL_9C 10
+#define WT_IMUL_9C 9
 #define WT_IMUL_R 16
 #define WT_IMUL_M 4
 #define WT_IMULH_R 4
@@ -36,7 +36,7 @@ along with RandomX.  If not, see<http://www.gnu.org/licenses/>.
 #define WT_ISDIV_C 4
 #define WT_INEG_R 2
 #define WT_IXOR_R 12
-#define WT_IXOR_M 3
+#define WT_IXOR_M 4
 #define WT_IROR_R 10
 #define WT_IROL_R 10
 #define WT_ISWAP_R 4

src/main.cpp

@@ -172,7 +172,7 @@ void mine(RandomX::VirtualMachine* vm, std::atomic<int>& atomicNonce, AtomicHash
 		//vm->initializeScratchpad(scratchpad, spIndex);
 		vm->setScratchpad(scratchpad);
 		//dump((char*)((RandomX::CompiledVirtualMachine*)vm)->getProgram(), RandomX::CodeSize, "code-1337-jmp.txt");
-		for (int chain = 0; chain < 16; ++chain) {
+		for (int chain = 0; chain < 8; ++chain) {
 			vm->initializeProgram(hash);
 			vm->execute();
 			vm->getResult(nullptr, 0, hash);

src/program.inc

@@ -1,381 +1,768 @@
-	; FPMUL_R e0, a2
-	mulpd xmm4, xmm10
-	; IADD_RC r2, r5, -1621224194
-	lea r10, [r10+r13-1621224194]
-	; ISTORE L2[r2], r7
-	mov eax, r10d
+	; 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]
+	mov ecx, r11d
+	and ecx, 16376
+	mov rax, r14
+	mul qword ptr [rsi+rcx]
+	mov r14, rdx
+	; IMUL_R r5, r1
+	imul r13, r9
+	; FPADD_M f0, L2[r6]
+	mov eax, r14d
 	and eax, 262136
-	mov qword ptr [rsi+rax], r15
+	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
+	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
+	and eax, 16376
+	add r9, qword ptr [rsi+rax]
 	; FPMUL_R e2, a2
 	mulpd xmm6, xmm10
-	; IMUL_R r6, r3
-	imul r14, r11
-	; FPSUB_M f1, L1[r4]
+	; 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
+	; ISTORE L1[r4], r3
 	mov eax, r12d
 	and eax, 16376
-	cvtdq2pd xmm12, qword ptr [rsi+rax]
-	subpd xmm1, xmm12
-	; IROL_R r5, r3
-	mov ecx, r11d
-	rol r13, cl
-	; FPMUL_R e2, a0
-	mulpd xmm6, xmm8
-	; FPSUB_R f3, a0
-	subpd xmm3, xmm8
-	; IXOR_R r0, r4
-	xor r8, r12
-	; ISMULH_M r3, L1[r7]
-	mov ecx, r15d
-	and ecx, 16376
-	mov rax, r11
-	imul qword ptr [rsi+rcx]
-	mov r11, rdx
-	; FPSWAP_R f2
-	shufpd xmm2, xmm2, 1
-	; IDIV_C r6, 1248528248
-	mov rax, 15864311168205210203
-	mul r14
-	shr rdx, 30
-	add r14, rdx
-	; FPMUL_R e0, a2
-	mulpd xmm4, xmm10
-	; IADD_RC r3, r4, -52260428
-	lea r11, [r11+r12-52260428]
-	; IADD_R r7, -1138617760
-	add r15, -1138617760
-	; IROL_R r2, r6
+	mov qword ptr [rsi+rax], r11
+	; IROR_R r3, r6
 	mov ecx, r14d
-	rol r10, cl
-	; FPNEG_R f2
-	xorps xmm2, xmm15
-	; IROR_R r7, r1
-	mov ecx, r9d
-	ror r15, cl
-	; COND_R r2, lt(r7, -41618808)
-	xor ecx, ecx
-	cmp r15d, -41618808
-	setl cl
-	add r10, rcx
-	; FPMUL_R e3, a0
-	mulpd xmm7, xmm8
-	; CFROUND r1, 43
-	mov rax, r9
-	rol rax, 34
-	and eax, 24576
-	or eax, 40896
-	mov dword ptr [rsp-8], eax
-	ldmxcsr dword ptr [rsp-8]
-	; FPADD_R f2, a1
-	addpd xmm2, xmm9
-	; FPSUB_M f0, L1[r7]
-	mov eax, r15d
+	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
+	; 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
+	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]
-	subpd xmm0, xmm12
-	; ISTORE L1[r6], r2
-	mov eax, r14d
-	and eax, 16376
-	mov qword ptr [rsi+rax], r10
-	; ISUB_R r6, r5
-	sub r14, r13
-	; IADD_M r0, L1[r4]
-	mov eax, r12d
+	addpd xmm2, xmm12
+	; IMULH_R r6, r1
+	mov rax, r14
+	mul r9
+	mov r14, rdx
+	; ISUB_M r5, L1[r0]
+	mov eax, r8d
 	and eax, 16376
-	add r8, qword ptr [rsi+rax]
-	; ISTORE L1[r4], r3
+	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
+	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]
 	mov eax, r12d
 	and eax, 16376
-	mov qword ptr [rsi+rax], r11
-	; ISTORE L1[r6], r6
+	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
-	mov qword ptr [rsi+rax], r14
-	; FPSQRT_R e0
-	sqrtpd xmm4, xmm4
-	; IXOR_R r2, r5
-	xor r10, r13
-	; FPSQRT_R e1
-	sqrtpd xmm5, xmm5
-	; FPMUL_R e1, a3
-	mulpd xmm5, xmm11
-	; IMULH_R r7, r6
-	mov rax, r15
-	mul r14
-	mov r15, rdx
-	; ISDIV_C r0, -1706892622
-	mov rax, -5802075764249827661
-	imul r8
+	xor r12, qword ptr [rsi+rax]
+	; IXOR_R r2, r6
+	xor r10, r14
+	; FPADD_M f3, L1[r0]
+	mov eax, r8d
+	and eax, 16376
+	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 r8, rdx
-	; IMUL_R r5, r3
-	imul r13, r11
-	; FPSQRT_R e2
-	sqrtpd xmm6, xmm6
-	; FPADD_M f3, L1[r4]
+	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]
 	mov eax, r12d
 	and eax, 16376
 	cvtdq2pd xmm12, qword ptr [rsi+rax]
-	addpd xmm3, xmm12
-	; IADD_R r3, r2
-	add r11, r10
-	; FPADD_R f1, a0
-	addpd xmm1, xmm8
-	; FPDIV_R e3, a2
-	divpd xmm7, xmm10
+	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
-	; FPSUB_R f0, a1
-	subpd xmm0, xmm9
-	; IMUL_M r5, L1[r6]
-	mov eax, r14d
+	; 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
-	imul r13, qword ptr [rsi+rax]
-	; IADD_RC r1, r2, -1263285243
-	lea r9, [r9+r10-1263285243]
-	; IMUL_9C r4, 1994773931
-	lea r12, [r12+r12*8+1994773931]
-	; FPSWAP_R e3
-	shufpd xmm7, xmm7, 1
-	; IMUL_M r0, L1[r7]
-	mov eax, r15d
+	xor r14, qword ptr [rsi+rax]
+	; ISUB_R r7, r4
+	sub r15, r12
+	; ISUB_M r6, L1[r1]
+	mov eax, r9d
 	and eax, 16376
-	imul r8, qword ptr [rsi+rax]
-	; IROR_R r1, r6
-	mov ecx, r14d
-	ror r9, cl
-	; IROL_R r2, r4
-	mov ecx, r12d
-	rol r10, cl
-	; FPSUB_R f3, a1
-	subpd xmm3, xmm9
-	; ISTORE L1[r0], r5
-	mov eax, r8d
+	sub r14, qword ptr [rsi+rax]
+	; ISTORE L1[r5], r3
+	mov eax, r13d
 	and eax, 16376
-	mov qword ptr [rsi+rax], r13
-	; FPDIV_M e2, L2[r3]
-	mov eax, r11d
+	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
+	xor eax, eax
+	sub rdx, r14
+	sar rdx, 25
+	sets al
+	add rdx, rax
+	add r14, rdx
+	; FPMUL_R e1, a2
+	mulpd xmm5, xmm10
+	; FPSUB_M f2, L2[r4]
+	mov eax, r12d
 	and eax, 262136
 	cvtdq2pd xmm12, qword ptr [rsi+rax]
-	divpd xmm6, xmm12
-	maxpd xmm6, xmm13
-	; FPSWAP_R f2
-	shufpd xmm2, xmm2, 1
-	; IADD_R r7, r5
-	add r15, r13
-	; FPDIV_M e0, L1[r4]
-	mov eax, r12d
+	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
+	maxpd xmm7, xmm13
+	; IXOR_M r5, L1[r1]
+	mov eax, r9d
 	and eax, 16376
-	cvtdq2pd xmm12, qword ptr [rsi+rax]
-	divpd xmm4, xmm12
+	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
+	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_M f3, L1[r5]
-	mov eax, r13d
+	; 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]
-	addpd xmm3, xmm12
-	; FPADD_R f0, a3
-	addpd xmm0, xmm11
-	; IADD_R r2, r0
-	add r10, r8
-	; ISTORE L1[r3], r6
-	mov eax, r11d
+	subpd xmm2, xmm12
+	; FPMUL_R e0, a0
+	mulpd xmm4, xmm8
+	; IXOR_M r4, L2[r7]
+	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
-	mov qword ptr [rsi+rax], r14
-	; IROR_R r1, r7
-	mov ecx, r15d
-	ror r9, cl
-	; IMUL_9C r5, 301671287
-	lea r13, [r13+r13*8+301671287]
-	; IXOR_R r7, 266992378
-	xor r15, 266992378
-	; FPSQRT_R e3
-	sqrtpd xmm7, xmm7
-	; IMUL_M r2, L2[r0]
-	mov eax, r8d
+	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
-	imul r10, qword ptr [rsi+rax]
+	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 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
+	mov ecx, r12d
+	ror r8, cl
 	; FPMUL_R e3, a2
 	mulpd xmm7, xmm10
-	; IMUL_R r0, r6
-	imul r8, r14
-	; ISTORE L1[r0], r7
-	mov eax, r8d
+	; 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
 	mov qword ptr [rsi+rax], r15
-	; FPNEG_R f0
-	xorps xmm0, xmm15
-	; FPADD_M f3, L1[r5]
+	; IMUL_9C r7, -1206742834
+	lea r15, [r15+r15*8-1206742834]
+	; ISMULH_R r0, r4
+	mov rax, r8
+	imul r12
+	mov r8, rdx
+	; FPADD_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
+	mov eax, r13d
+	and eax, 16376
+	cmp dword ptr [rsi+rax], -2056260506
+	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
 	and eax, 16376
 	cvtdq2pd xmm12, qword ptr [rsi+rax]
-	addpd xmm3, xmm12
-	; IROR_R r5, r4
-	mov ecx, r12d
-	ror r13, cl
-	; ISTORE L2[r7], r2
-	mov eax, r15d
+	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
-	mov qword ptr [rsi+rax], r10
-	; FPADD_R f2, a3
-	addpd xmm2, xmm11
-	; FPADD_M f3, L1[r2]
+	cvtdq2pd xmm12, qword ptr [rsi+rax]
+	mulpd xmm7, xmm12
+	maxpd xmm7, xmm13
+	; ISTORE L1[r2], r2
 	mov eax, r10d
 	and eax, 16376
-	cvtdq2pd xmm12, qword ptr [rsi+rax]
-	addpd xmm3, xmm12
-	; ISDIV_C r5, -2076168315
-	mov rax, -4770095103914078469
-	imul r13
-	xor eax, eax
-	sar rdx, 29
-	sets al
-	add rdx, rax
-	add r13, rdx
-	; IADD_RC r0, r4, -1321374359
-	lea r8, [r8+r12-1321374359]
-	; CFROUND r6, 28
-	mov rax, r14
-	rol rax, 49
-	and eax, 24576
-	or eax, 40896
-	mov dword ptr [rsp-8], eax
-	ldmxcsr dword ptr [rsp-8]
-	; FPADD_R f2, a2
-	addpd xmm2, xmm10
-	; IROL_R r7, r6
+	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
-	; ISUB_R r2, r4
-	sub r10, r12
-	; ISMULH_R r0, -1500893068
-	mov rax, -1500893068
-	imul r8
-	add r8, rdx
-	; IADD_R r2, r3
-	add r10, r11
-	; FPSQRT_R e2
-	sqrtpd xmm6, xmm6
-	; IROL_R r7, r4
-	mov ecx, r12d
-	rol r15, cl
-	; IMUL_R r4, r2
-	imul r12, r10
-	; ISUB_R r3, r7
-	sub r11, r15
-	; IADD_R r2, r7
-	add r10, r15
-	; FPDIV_R e3, a0
-	divpd xmm7, xmm8
-	maxpd xmm7, xmm13
-	; ISUB_R r6, 540663146
-	sub r14, 540663146
-	; IROL_R r5, 58
-	rol r13, 58
-	; FPADD_R f2, a1
-	addpd xmm2, xmm9
-	; FPADD_R f2, a2
-	addpd xmm2, xmm10
-	; FPDIV_R e1, a2
-	divpd xmm5, xmm10
-	maxpd xmm5, xmm13
+	; ISTORE L2[r1], r7
+	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)
+	xor ecx, ecx
+	cmp r14d, -444806705
+	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
+	and eax, 16376
+	mov qword ptr [rsi+rax], r8
 	; FPADD_R f1, a2
 	addpd xmm1, xmm10
-	; IADD_R r5, r3
-	add r13, r11
-	; IADD_R r7, -1780268176
-	add r15, -1780268176
-	; ISUB_R r7, r0
-	sub r15, r8
-	; ISTORE L2[r0], r7
-	mov eax, r8d
+	; ISUB_R r1, r4
+	sub r9, r12
+	; IXOR_R r5, r0
+	xor r13, r8
+	; IXOR_M r7, L2[r1]
+	mov eax, r9d
 	and eax, 262136
-	mov qword ptr [rsi+rax], r15
-	; INEG_R r2
-	neg r10
-	; FPNEG_R f0
-	xorps xmm0, xmm15
-	; INEG_R r2
-	neg r10
-	; IADD_R r0, r3
-	add r8, r11
-	; IMUL_9C r7, -2124093035
-	lea r15, [r15+r15*8-2124093035]
-	; FPADD_M f2, L1[r0]
-	mov eax, r8d
+	xor r15, qword ptr [rsi+rax]
+	; FPSUB_R f0, a0
+	subpd xmm0, xmm8
+	; IXOR_M r1, L1[r4]
+	mov eax, r12d
 	and eax, 16376
-	cvtdq2pd xmm12, qword ptr [rsi+rax]
-	addpd xmm2, xmm12
-	; FPMUL_M e0, L1[r6]
-	mov eax, r14d
+	xor r9, qword ptr [rsi+rax]
+	; FPMUL_R e3, a0
+	mulpd xmm7, xmm8
+	; ISDIV_C r1, 1473744194
+	mov rax, -5006799265644655925
+	imul r9
+	xor eax, eax
+	add rdx, r9
+	sar rdx, 30
+	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
 	and eax, 16376
-	cvtdq2pd xmm12, qword ptr [rsi+rax]
-	mulpd xmm4, xmm12
-	maxpd xmm4, xmm13
-	; FPSUB_R f2, a3
-	subpd xmm2, xmm11
-	; IMUL_R r1, r2
-	imul r9, r10
-	; IDIV_C r7, 3214009572
-	mov rax, 12325439725582798855
-	mul r15
-	shr rdx, 31
-	add r15, rdx
-	; IMULH_R r3, r2
-	mov rax, r11
+	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
-	; IROR_R r1, r0
-	mov ecx, r8d
-	ror r9, cl
-	; FPMUL_R e0, a1
-	mulpd xmm4, xmm9
-	; IADD_RC r4, r4, 1456841848
-	lea r12, [r12+r12+1456841848]
-	; IROR_R r3, r2
-	mov ecx, r10d
-	ror r11, cl
-	; COND_M r0, of(L1[r4], 1678513610)
+	; 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
 	and eax, 16376
-	cmp dword ptr [rsi+rax], 1678513610
-	seto cl
-	add r8, rcx
-	; INEG_R r4
-	neg r12
-	; IMUL_R r4, r1
-	imul r12, r9
-	; FPADD_R f1, a2
-	addpd xmm1, xmm10
-	; FPSUB_R f2, a0
-	subpd xmm2, xmm8
-	; FPMUL_R e1, a2
-	mulpd xmm5, xmm10
-	; FPSUB_R f0, a3
-	subpd xmm0, xmm11
+	cvtdq2pd xmm12, qword ptr [rsi+rax]
+	addpd xmm1, xmm12
+	; FPMUL_M e0, L1[r4]
+	mov eax, r12d
+	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
 	ror r8, cl
-	; ISTORE L2[r1], r4
-	mov eax, r9d
+	; 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
-	; IROL_R r7, r6
-	mov ecx, r14d
-	rol r15, cl
-	; IMUL_9C r2, 266593902
-	lea r10, [r10+r10*8+266593902]
-	; IMUL_R r4, r6
-	imul r12, r14
-	; FPSUB_R f2, a2
-	subpd xmm2, xmm10
-	; FPNEG_R f3
-	xorps xmm3, xmm15
-	; IROR_R r7, r2
-	mov ecx, r10d
+	; 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)
+	xor ecx, ecx
+	cmp r15d, -229570354
+	seta cl
+	add r9, rcx
+	; IROR_R r7, r3
+	mov ecx, r11d
 	ror r15, cl
-	; IROR_R r0, r5
-	mov ecx, r13d
-	ror r8, 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)
+	xor ecx, ecx
+	cmp r11d, 1932234501
+	seta cl
+	add r10, rcx
+	; FPMUL_R e1, a0
+	mulpd xmm5, xmm8
+	; FPSUB_M f1, 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)
+	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
+	xor eax, eax
+	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
+	subpd xmm0, xmm8
+	; IADD_R r0, r2
+	add r8, r10
+	; IADD_M r4, L1[r2]
+	mov eax, r10d
+	and eax, 16376
+	add r12, qword ptr [rsi+rax]
+	; ISTORE L1[r7], r2
+	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