From 9f1753cc4f54dfe22dad97a5b45a9b69027d19b7 Mon Sep 17 00:00:00 2001
From: SChernykh <sergey.v.chernykh@gmail.com>
Date: Wed, 22 Jan 2020 20:11:00 +0100
Subject: [PATCH] Optimized CFROUND

---
 .../randomx/asm/program_epilogue_store.inc    |  2 +-
 src/crypto/randomx/asm/program_loop_load.inc  |  4 ++--
 src/crypto/randomx/asm/program_loop_store.inc |  4 ++--
 src/crypto/randomx/jit_compiler_x86.cpp       | 23 +++++++++++--------
 .../randomx/jit_compiler_x86_static.asm       |  8 +++++--
 5 files changed, 24 insertions(+), 17 deletions(-)

diff --git a/src/crypto/randomx/asm/program_epilogue_store.inc b/src/crypto/randomx/asm/program_epilogue_store.inc
index 6fa1a1fcf..82067d191 100644
--- a/src/crypto/randomx/asm/program_epilogue_store.inc
+++ b/src/crypto/randomx/asm/program_epilogue_store.inc
@@ -1,5 +1,5 @@
 	;# save VM register values
-	add rsp, 24
+	add rsp, 40
 	pop rcx
 	mov qword ptr [rcx+0], r8
 	mov qword ptr [rcx+8], r9
diff --git a/src/crypto/randomx/asm/program_loop_load.inc b/src/crypto/randomx/asm/program_loop_load.inc
index f5b23a931..5d8a84918 100644
--- a/src/crypto/randomx/asm/program_loop_load.inc
+++ b/src/crypto/randomx/asm/program_loop_load.inc
@@ -1,5 +1,5 @@
 	lea rcx, [rsi+rax]
-	mov [rsp+8], rcx
+	mov [rsp+16], rcx
 	xor r8,  qword ptr [rcx+0]
 	xor r9,  qword ptr [rcx+8]
 	xor r10, qword ptr [rcx+16]
@@ -9,7 +9,7 @@
 	xor r14, qword ptr [rcx+48]
 	xor r15, qword ptr [rcx+56]
 	lea rcx, [rsi+rdx]
-	mov [rsp+16], rcx
+	mov [rsp+24], rcx
 	cvtdq2pd xmm0, qword ptr [rcx+0]
 	cvtdq2pd xmm1, qword ptr [rcx+8]
 	cvtdq2pd xmm2, qword ptr [rcx+16]
diff --git a/src/crypto/randomx/asm/program_loop_store.inc b/src/crypto/randomx/asm/program_loop_store.inc
index f778f134f..f579bb0c5 100644
--- a/src/crypto/randomx/asm/program_loop_store.inc
+++ b/src/crypto/randomx/asm/program_loop_store.inc
@@ -1,4 +1,4 @@
-	mov rcx, [rsp+16]
+	mov rcx, [rsp+24]
 	mov qword ptr [rcx+0], r8
 	mov qword ptr [rcx+8], r9
 	mov qword ptr [rcx+16], r10
@@ -7,7 +7,7 @@
 	mov qword ptr [rcx+40], r13
 	mov qword ptr [rcx+48], r14
 	mov qword ptr [rcx+56], r15
-	mov rcx, [rsp+8]
+	mov rcx, [rsp+16]
 	xorpd xmm0, xmm4
 	xorpd xmm1, xmm5
 	xorpd xmm2, xmm6
diff --git a/src/crypto/randomx/jit_compiler_x86.cpp b/src/crypto/randomx/jit_compiler_x86.cpp
index b31355263..34f98cb96 100644
--- a/src/crypto/randomx/jit_compiler_x86.cpp
+++ b/src/crypto/randomx/jit_compiler_x86.cpp
@@ -400,7 +400,7 @@ namespace randomx {
 		*(uint32_t*)(code + codePos + 10) = RandomX_CurrentConfig.ScratchpadL3Mask64_Calculated;
 		*(uint32_t*)(code + codePos + 20) = RandomX_CurrentConfig.ScratchpadL3Mask64_Calculated;
 		if (hasAVX) {
-			uint32_t* p = (uint32_t*)(code + codePos + 32);
+			uint32_t* p = (uint32_t*)(code + codePos + 67);
 			*p = (*p & 0xFF000000U) | 0x0077F8C5U;
 		}
 
@@ -1072,18 +1072,21 @@ namespace randomx {
 		uint8_t* const p = code;
 		int pos = codePos;
 
-		emit(REX_MOV_RR64, p, pos);
-		emitByte(0xc0 + instr.src, p, pos);
-		int rotate = (13 - (instr.getImm32() & 63)) & 63;
-		if (rotate != 0) {
-			emit(ROL_RAX, p, pos);
-			emitByte(rotate, p, pos);
-		}
+		const uint32_t src = instr.src;
+
+		*(uint32_t*)(p + pos) = 0x00C08B49 + (src << 16);
+		const int rotate = (static_cast<int>(instr.getImm32() & 63) - 2) & 63;
+		*(uint32_t*)(p + pos + 3) = 0x00C8C148 + (rotate << 24);
+
 		if (vm_flags & RANDOMX_FLAG_AMD) {
-			emit(AND_OR_MOV_LDMXCSR_RYZEN, p, pos);
+			*(uint64_t*)(p + pos + 7) = 0x742024443B0CE083ULL;
+			*(uint8_t*)(p + pos + 15) = 8;
+			*(uint64_t*)(p + pos + 16) = 0x202444890414AE0FULL;
+			pos += 24;
 		}
 		else {
-			emit(AND_OR_MOV_LDMXCSR, p, pos);
+			*(uint64_t*)(p + pos + 7) = 0x0414AE0F0CE083ULL;
+			pos += 14;
 		}
 
 		codePos = pos;
diff --git a/src/crypto/randomx/jit_compiler_x86_static.asm b/src/crypto/randomx/jit_compiler_x86_static.asm
index 8e7714002..e36e5aafa 100644
--- a/src/crypto/randomx/jit_compiler_x86_static.asm
+++ b/src/crypto/randomx/jit_compiler_x86_static.asm
@@ -82,8 +82,12 @@ randomx_program_prologue_first_load PROC
 	and eax, RANDOMX_SCRATCHPAD_MASK
 	ror rdx, 32
 	and edx, RANDOMX_SCRATCHPAD_MASK
-	sub rsp, 24
-	stmxcsr dword ptr [rsp]
+	sub rsp, 40
+	mov dword ptr [rsp], 9FC0h
+	mov dword ptr [rsp+4], 0BFC0h
+	mov dword ptr [rsp+8], 0DFC0h
+	mov dword ptr [rsp+12], 0FFC0h
+	mov dword ptr [rsp+32], -1
 	nop
 	nop
 	nop