add adcx, adox with APX

herumi · Nov 1, 2023 · 6b19515 · 6b19515
1 parent 77d6ace
commit 6b19515
Show file tree

Hide file tree

Showing 4 changed files with 53 additions and 5 deletions.
diff --git a/gen/gen_code.cpp b/gen/gen_code.cpp
@@ -1054,6 +1054,20 @@ void put()
 			printf("void %s(const Reg& reg, const Operand& op) { opMovxx(reg, op, 0x%02X); }\n", p->name, p->code);
 		}
 	}
+	{
+		const struct Tbl {
+			uint8_t code;
+			const char *name;
+		} tbl[] = {
+			{ 0x66, "adcx" },
+			{ 0xF3, "adox" },
+		};
+		for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) {
+			const Tbl *p = &tbl[i];
+			printf("void %s(const Reg32e& reg, const Operand& op) { if (opROO(Reg(), op, reg, T_66, 0x%02X)) return; opGen(reg, op, 0xF6, 0x%02X, isREG32_REG32orMEM, NONE, 0x38); }\n", p->name, p->code, p->code);
+			printf("void %s(const Reg32e& d, const Reg32e& reg, const Operand& op) { opROO(d, op, reg, T_66, 0x%02X); }\n", p->name, p->code);
+		}
+	}
 	{ // in/out
 		puts("void in_(const Reg& a, uint8_t v) { opInOut(a, 0xE4, v); }");
 		puts("void in_(const Reg& a, const Reg& d) { opInOut(a, d, 0xEC); }");
@@ -1084,8 +1098,6 @@ void put()
 		puts("void movbe(const Address& addr, const Reg& reg) { opModM(addr, reg, 0x0F, 0x38, 0xF1); }");
 		puts("void movdiri(const Address& addr, const Reg32e& reg) { opModM(addr, reg, 0x0F, 0x38, 0xF9); }");
 		puts("void movdir64b(const Reg& reg, const Address& addr) { db(0x66); opModM(addr, reg.cvt32(), 0x0F, 0x38, 0xF8); }");
-		puts("void adcx(const Reg32e& reg, const Operand& op) { opGen(reg, op, 0xF6, 0x66, isREG32_REG32orMEM, NONE, 0x38); }");
-		puts("void adox(const Reg32e& reg, const Operand& op) { opGen(reg, op, 0xF6, 0xF3, isREG32_REG32orMEM, NONE, 0x38); }");
 		puts("void cmpxchg8b(const Address& addr) { opModM(addr, Reg32(1), 0x0F, 0xC7); }");
 
 		puts("void pextrw(const Operand& op, const Mmx& xmm, uint8_t imm) { opExt(op, xmm, 0x15, imm, true); }");

diff --git a/test/apx.cpp b/test/apx.cpp
@@ -341,3 +341,27 @@ CYBOZU_TEST_AUTO(rm3_2)
 	CYBOZU_TEST_EQUAL(c.getSize(), n);
 	CYBOZU_TEST_EQUAL_ARRAY(c.getCode(), tbl, n);
 }
+
+CYBOZU_TEST_AUTO(adcx)
+{
+	struct Code : Xbyak::CodeGenerator {
+		Code()
+		{
+			adcx(rax, r30);
+			adcx(ecx, r20d);
+			adcx(ecx, ptr [r31+r29*4]);
+			adcx(r20d, ptr [rax]);
+			adcx(r16, ptr [r31+r29*4]);
+			adcx(r17, ptr [rax]);
+			adcx(rax, rcx, rdx);
+		}
+	}  c;
+	const uint8_t tbl[] = {
+		0x62, 0xdc, 0xfd, 0x08, 0x66, 0xc6, 0x62, 0xfc, 0x7d, 0x08, 0x66, 0xcc, 0x62, 0x9c, 0x79, 0x08,
+		0x66, 0x0c, 0xaf, 0x62, 0xe4, 0x7d, 0x08, 0x66, 0x20, 0x62, 0x8c, 0xf9, 0x08, 0x66, 0x04, 0xaf,
+		0x62, 0xe4, 0xfd, 0x08, 0x66, 0x08, 0x62, 0xf4, 0xfd, 0x18, 0x66, 0xca,
+	};
+	const size_t n = sizeof(tbl);
+	CYBOZU_TEST_EQUAL(c.getSize(), n);
+	CYBOZU_TEST_EQUAL_ARRAY(c.getCode(), tbl, n);
+}
diff --git a/xbyak/xbyak.h b/xbyak/xbyak.h
@@ -2853,8 +2853,9 @@ class CodeGenerator : public CodeArray {
 	}
 
 	// (r, r, m) or (r, m, r)
-	void opROO(const Reg& d, const Operand& op1, const Operand& op2, int type, int code0, int code1 = NONE, int code2 = NONE, int immSize = 0)
+	bool opROO(const Reg& d, const Operand& op1, const Operand& op2, int type, int code0, int code1 = NONE, int code2 = NONE, int immSize = 0)
 	{
+//		if (type == 0 && !(d.hasRex2() || op1.hasRex2() || op2.hasRex2())) return false;
 		const Operand *p1 = &op1, *p2 = &op2;
 		if (p1->isMEM()) { std::swap(p1, p2); } else { if (p2->isMEM()) code0 |= 2; }
 		if (p1->isMEM()) XBYAK_THROW(ERR_BAD_COMBINATION)
@@ -2870,6 +2871,7 @@ class CodeGenerator : public CodeArray {
 			writeCode(type, d, code0, code1, code2);
 			setModRM(3, op2.getIdx(), op1.getIdx());
 		}
+		return true;
 	}
 #endif
 
@@ -3010,6 +3012,14 @@ class CodeGenerator : public CodeArray {
 			size -= len;
 		}
 	}
+	void adcx2(const Reg32e& r1, const Operand& op)
+	{
+		opROO(Reg(), op, r1, T_66, 0x66);
+	}
+	void adcx2(const Reg32e& d, const Reg32e& r1, const Operand& op)
+	{
+		opROO(d, op, r1, T_66, 0x66);
+	}
 #ifndef XBYAK_DONT_READ_LIST
 #include "xbyak_mnemonic.h"
 	/*

diff --git a/xbyak/xbyak_mnemonic.h b/xbyak/xbyak_mnemonic.h
@@ -5,7 +5,8 @@ void adc(const Operand& op, uint32_t imm) { opRM_I(op, imm, 0x10, 2); }
 void adc(const Operand& op1, const Operand& op2) { opRM_RM(op1, op2, 0x10); }
 void adc(const Reg& d, const Operand& op, uint32_t imm) { opROI(d, op, imm, 0, 2); }
 void adc(const Reg& d, const Operand& op1, const Operand& op2) { opROO(d, op1, op2, 0, 0x10); }
-void adcx(const Reg32e& reg, const Operand& op) { opGen(reg, op, 0xF6, 0x66, isREG32_REG32orMEM, NONE, 0x38); }
+void adcx(const Reg32e& d, const Reg32e& reg, const Operand& op) { opROO(d, op, reg, T_66, 0x66); }
+void adcx(const Reg32e& reg, const Operand& op) { if (opROO(Reg(), op, reg, T_66, 0x66)) return; opGen(reg, op, 0xF6, 0x66, isREG32_REG32orMEM, NONE, 0x38); }
 void add(const Operand& op, uint32_t imm) { opRM_I(op, imm, 0x00, 0); }
 void add(const Operand& op1, const Operand& op2) { opRM_RM(op1, op2, 0x00); }
 void add(const Reg& d, const Operand& op, uint32_t imm) { opROI(d, op, imm, 0, 0); }
@@ -16,7 +17,8 @@ void addsd(const Xmm& xmm, const Operand& op) { opGen(xmm, op, 0x58, 0xF2, isXMM
 void addss(const Xmm& xmm, const Operand& op) { opGen(xmm, op, 0x58, 0xF3, isXMM_XMMorMEM); }
 void addsubpd(const Xmm& xmm, const Operand& op) { opGen(xmm, op, 0xD0, 0x66, isXMM_XMMorMEM); }
 void addsubps(const Xmm& xmm, const Operand& op) { opGen(xmm, op, 0xD0, 0xF2, isXMM_XMMorMEM); }
-void adox(const Reg32e& reg, const Operand& op) { opGen(reg, op, 0xF6, 0xF3, isREG32_REG32orMEM, NONE, 0x38); }
+void adox(const Reg32e& d, const Reg32e& reg, const Operand& op) { opROO(d, op, reg, T_66, 0xF3); }
+void adox(const Reg32e& reg, const Operand& op) { if (opROO(Reg(), op, reg, T_66, 0xF3)) return; opGen(reg, op, 0xF6, 0xF3, isREG32_REG32orMEM, NONE, 0x38); }
 void aesdec(const Xmm& xmm, const Operand& op) { opGen(xmm, op, 0xDE, 0x66, isXMM_XMMorMEM, NONE, 0x38); }
 void aesdeclast(const Xmm& xmm, const Operand& op) { opGen(xmm, op, 0xDF, 0x66, isXMM_XMMorMEM, NONE, 0x38); }
 void aesenc(const Xmm& xmm, const Operand& op) { opGen(xmm, op, 0xDC, 0x66, isXMM_XMMorMEM, NONE, 0x38); }