tweak vmovd

3 files changed, 22 insertions, 37 deletions

diff --git a/gen/gen_code.cpp b/gen/gen_code.cpp
index e72df50..df4e5b9 100644
--- a/gen/gen_code.cpp
+++ b/gen/gen_code.cpp
@@ -1734,8 +1734,8 @@ void put()
 	}
 	// mov
 	{
-		printf("void vmovd(const Xmm& x, const Operand& op) { if (!op.isREG(32) && !op.isMEM()) XBYAK_THROW(ERR_BAD_COMBINATION) opAVX_X_X_XM(x, xm0, op, T_0F | T_66 | T_W0 | T_EVEX | T_N4, 0x6E); }\n");
-		printf("void vmovd(const Operand& op, const Xmm& x) { if (!op.isREG(32) && !op.isMEM()) XBYAK_THROW(ERR_BAD_COMBINATION) opAVX_X_X_XM(x, xm0, op, T_0F | T_66 | T_W0 | T_EVEX | T_N4, 0x7E); }\n");
+		puts("void vmovd(const Xmm& x, const Operand& op, PreferredEncoding encoding = DefaultEncoding) { opVmovd(x, op, false, encoding); }");
+		puts("void vmovd(const Operand& op, const Xmm& x, PreferredEncoding encoding = DefaultEncoding) { opVmovd(x, op, true, encoding); }");
 
 		printf("void vmovq(const Xmm& x, const Address& addr) { uint64_t type; uint8_t code; if (x.getIdx() < 16) { type = T_0F | T_F3; code = 0x7E; } else { type = T_0F | T_66 | T_EVEX | T_EW1 | T_N8; code = 0x6E; } opAVX_X_X_XM(x, xm0, addr, type, code); }\n");
 		printf("void vmovq(const Address& addr, const Xmm& x) { opAVX_X_X_XM(x, xm0, addr, T_0F | T_66 | T_EVEX | T_EW1 | T_N8, x.getIdx() < 16 ? 0xD6 : 0x7E); }\n");
@@ -1900,36 +1900,6 @@ void put()
 			printf("void %s(const Xmm& x1, const Xmm& x2, const Operand& op, PreferredEncoding encoding = DefaultEncoding) { opEncoding(x1, x2, op, %s, 0x%02X, encoding); }\n", p->name, s.c_str(), p->code);
 		}
 	}
-	// avx-vnni-int8
-	// avx-vnni-int16
-#if 0
-	{
-		const struct Tbl {
-			uint8_t code;
-			const char *name;
-			uint64_t type;
-		} tbl[] = {
-//			{ 0x50, "vpdpbssd", T_F2 | T_0F38 | T_W0 | T_YMM },
-//			{ 0x51, "vpdpbssds", T_F2 | T_0F38 | T_W0 | T_YMM },
-//			{ 0x50, "vpdpbsud", T_F3 | T_0F38 | T_W0 | T_YMM },
-//			{ 0x51, "vpdpbsuds", T_F3 | T_0F38 | T_W0 | T_YMM },
-//			{ 0x50, "vpdpbuud", T_0F38 | T_W0 | T_YMM },
-//			{ 0x51, "vpdpbuuds", T_0F38 | T_W0 | T_YMM },
-
-//			{ 0xD2, "vpdpwsud", T_F3 | T_0F38 | T_W0 | T_YMM },
-//			{ 0xD3, "vpdpwsuds", T_F3 | T_0F38 | T_W0 | T_YMM },
-//			{ 0xD2, "vpdpwusd", T_66 | T_0F38 | T_W0 | T_YMM },
-//			{ 0xD3, "vpdpwusds", T_66 | T_0F38 | T_W0 | T_YMM },
-//			{ 0xD2, "vpdpwuud", T_0F38 | T_W0 | T_YMM },
-//			{ 0xD3, "vpdpwuuds", T_0F38 | T_W0 | T_YMM },
-		};
-		for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) {
-			const Tbl *p = &tbl[i];
-			std::string s = type2String(p->type);
-			printf("void %s(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, %s, 0x%02X); }\n", p->name, s.c_str(), p->code);
-		}
-	}
-#endif
 }
 
 void put32()
diff --git a/xbyak/xbyak.h b/xbyak/xbyak.h
index 17a9597..ed5c361 100644
--- a/xbyak/xbyak.h
+++ b/xbyak/xbyak.h
@@ -2665,7 +2665,8 @@ private:
 	{
 		opAVX_X_X_XM(x1, x2, op, type | orEvexIf(encoding, typeVex, typeEvex, sel), code, imm);
 	}
-	uint64_t orEvexIf(PreferredEncoding encoding, uint64_t typeVex, uint64_t typeEvex, int sel) {
+	bool isVexEncoding(PreferredEncoding encoding, int sel) const
+	{
 		if (encoding == DefaultEncoding) {
 			encoding = defaultEncoding_[sel];
 		}
@@ -2673,9 +2674,13 @@ private:
 #ifdef XBYAK_DISABLE_AVX512
 			XBYAK_THROW(ERR_EVEX_IS_INVALID)
 #endif
-			return T_MUST_EVEX | typeEvex;
+			return false;
 		}
-		return typeVex;
+		return true;
+	}
+	uint64_t orEvexIf(PreferredEncoding encoding, uint64_t typeVex, uint64_t typeEvex, int sel) {
+		bool isVex = isVexEncoding(encoding, sel);
+		return isVex ?  typeVex : T_MUST_EVEX | typeEvex;
 	}
 	void opInOut(const Reg& a, const Reg& d, uint8_t code)
 	{
@@ -3188,6 +3193,16 @@ public:
 		}
 		db(0xC8 + (idx & 7));
 	}
+	void opVmovd(const Xmm& x, const Operand& op, bool rev, PreferredEncoding encoding)
+	{
+		if (isVexEncoding(encoding, 1)) {
+			if (!op.isREG(32) && !op.isMEM()) XBYAK_THROW(ERR_BAD_COMBINATION)
+			uint64_t type = T_0F | T_66 | T_W0 | T_EVEX | T_N4;
+			int code = rev ? 0x7E : 0x6E;
+			opAVX_X_X_XM(x, xm0, op, type, code);
+		} else {
+		}
+	}
 	/*
 		use single byte nop if useMultiByteNop = false
 	*/
diff --git a/xbyak/xbyak_mnemonic.h b/xbyak/xbyak_mnemonic.h
index 07ef43e..efd207a 100644
--- a/xbyak/xbyak_mnemonic.h
+++ b/xbyak/xbyak_mnemonic.h
@@ -1332,8 +1332,8 @@ void vmovapd(const Address& addr, const Xmm& xmm) { opAVX_X_XM_IMM(xmm, addr, T_
 void vmovapd(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_66|T_0F|T_EW1|T_YMM|T_EVEX, 0x28); }
 void vmovaps(const Address& addr, const Xmm& xmm) { opAVX_X_XM_IMM(xmm, addr, T_0F|T_EW0|T_YMM|T_EVEX|T_M_K, 0x29); }
 void vmovaps(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F|T_EW0|T_YMM|T_EVEX, 0x28); }
-void vmovd(const Operand& op, const Xmm& x) { if (!op.isREG(32) && !op.isMEM()) XBYAK_THROW(ERR_BAD_COMBINATION) opAVX_X_X_XM(x, xm0, op, T_0F | T_66 | T_W0 | T_EVEX | T_N4, 0x7E); }
-void vmovd(const Xmm& x, const Operand& op) { if (!op.isREG(32) && !op.isMEM()) XBYAK_THROW(ERR_BAD_COMBINATION) opAVX_X_X_XM(x, xm0, op, T_0F | T_66 | T_W0 | T_EVEX | T_N4, 0x6E); }
+void vmovd(const Operand& op, const Xmm& x, PreferredEncoding encoding = DefaultEncoding) { opVmovd(x, op, true, encoding); }
+void vmovd(const Xmm& x, const Operand& op, PreferredEncoding encoding = DefaultEncoding) { opVmovd(x, op, false, encoding); }
 void vmovddup(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_DUP|T_F2|T_0F|T_EW1|T_YMM|T_EVEX|T_ER_X|T_ER_Y|T_ER_Z, 0x12); }
 void vmovdqa(const Address& addr, const Xmm& xmm) { opAVX_X_XM_IMM(xmm, addr, T_66|T_0F|T_YMM, 0x7F); }
 void vmovdqa(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_66|T_0F|T_YMM, 0x6F); }