add vexpand*, vcomp*

gen/gen_avx512.cpp

@@ -168,6 +168,12 @@ void putX_XM()
 		{ 0x78, "vcvttps2udq", T_0F | T_MUST_EVEX | T_YMM | T_EW0 | T_B32 | T_SAE_Z },
 		{ 0x7A, "vcvtudq2ps", T_F2 | T_0F | T_YMM | T_MUST_EVEX | T_EW0 | T_B32 | T_ER_Z },
 		{ 0x7A, "vcvtuqq2pd", T_F3 | T_0F | T_YMM | T_MUST_EVEX | T_EW1 | T_B64 | T_ER_Z },
+
+		{ 0x88, "vexpandpd", T_66 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW1 | T_N8 },
+		{ 0x88, "vexpandps", T_66 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW0 | T_N4 },
+
+		{ 0x89, "vpexpandd", T_66 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW0 | T_N4 },
+		{ 0x89, "vpexpandq", T_66 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW1 | T_N8 },
 	};
 	for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) {
 		const Tbl *p = &tbl[i];
@@ -198,6 +204,26 @@ void putM_X()
 	}
 }
 
+void putXM_X()
+{
+	const struct Tbl {
+		uint8 code;
+		const char *name;
+		int type;
+	} tbl[] = {
+		{ 0x8A, "vcompresspd", T_66 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW1 | T_N8 },
+		{ 0x8A, "vcompressps", T_66 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW0 | T_N4 },
+
+		{ 0x8B, "vpcompressd", T_66 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW0 | T_N4 },
+		{ 0x8B, "vpcompressq", T_66 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW1 | T_N8 },
+	};
+	for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) {
+		const Tbl *p = &tbl[i];
+		std::string type = type2String(p->type);
+		printf("void %s(const Operand& op, const Xmm& x) { opAVX_X_XM_IMM(x, op, %s, 0x%02X); }\n", p->name, type.c_str(), p->code);
+	}
+}
+
 void putX_X_XM_IMM()
 {
 	const struct Tbl {
@@ -418,6 +444,7 @@ int main()
 	putVcmp();
 	putX_XM();
 	putM_X();
+	putXM_X();
 	putX_X_XM_IMM();
 	putShift();
 	putExtractInsert();

test/make_512.cpp

@@ -1668,10 +1668,41 @@ public:
 			put(p.name, K_K, _ZMM, _ZMM | _MEM | bTbl[2]);
 		}
 	}
+	void putCompExp()
+	{
+		{
+			const char *tbl[] = {
+				"vcompresspd",
+				"vcompressps",
+				"vpcompressd",
+				"vpcompressq",
+			};
+			for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) {
+				const char *name = tbl[i];
+				put(name, XMM_KZ | _MEM, _XMM);
+				put(name, YMM_KZ | _MEM, _YMM);
+				put(name, ZMM_KZ | _MEM, _ZMM);
+			}
+		}
+		{
+			const char *tbl[] = {
+				"vexpandpd",
+				"vexpandps",
+				"vpexpandd",
+				"vpexpandq",
+			};
+			for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) {
+				const char *name = tbl[i];
+				put(name, XMM_KZ, _XMM | _MEM);
+				put(name, YMM_KZ, _YMM | _MEM);
+				put(name, ZMM_KZ, _ZMM | _MEM);
+			}
+		}
+	}
 	void putMin()
 	{
 #ifdef XBYAK64
-		putVtest();
+		putCompExp();
 #endif
 	}
 	void putAVX512()
@@ -1716,6 +1747,8 @@ public:
 		putVpcmp();
 		separateFunc();
 		putVtest();
+		separateFunc();
+		putCompExp();
 #endif
 	}
 };

xbyak/xbyak_avx512.h

@@ -107,6 +107,10 @@ void vcvttpd2uqq(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_66 |
 void vcvttps2udq(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_0F | T_EW0 | T_YMM | T_SAE_Z | T_MUST_EVEX | T_B32, 0x78); }
 void vcvtudq2ps(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_F2 | T_0F | T_EW0 | T_YMM | T_ER_Z | T_MUST_EVEX | T_B32, 0x7A); }
 void vcvtuqq2pd(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_F3 | T_0F | T_EW1 | T_YMM | T_ER_Z | T_MUST_EVEX | T_B64, 0x7A); }
+void vexpandpd(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_66 | T_0F38 | T_EW1 | T_YMM | T_MUST_EVEX | T_N8, 0x88); }
+void vexpandps(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_66 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX | T_N4, 0x88); }
+void vpexpandd(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_66 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX | T_N4, 0x89); }
+void vpexpandq(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_66 | T_0F38 | T_EW1 | T_YMM | T_MUST_EVEX | T_N8, 0x89); }
 void vpabsq(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_66 | T_0F38 | T_MUST_EVEX | T_EW1 | T_B64 | T_YMM, 0x1F); }
 void vmovdqa32(const Address& addr, const Xmm& x) { opAVX_X_XM_IMM(x, addr, T_66 | T_0F | T_EW0 | T_YMM | T_ER_X | T_ER_Y | T_ER_Z | T_MUST_EVEX, 0x7F); }
 void vmovdqa64(const Address& addr, const Xmm& x) { opAVX_X_XM_IMM(x, addr, T_66 | T_0F | T_EW1 | T_YMM | T_ER_X | T_ER_Y | T_ER_Z | T_MUST_EVEX, 0x7F); }
@@ -114,6 +118,10 @@ void vmovdqu8(const Address& addr, const Xmm& x) { opAVX_X_XM_IMM(x, addr, T_F2
 void vmovdqu16(const Address& addr, const Xmm& x) { opAVX_X_XM_IMM(x, addr, T_F2 | T_0F | T_EW1 | T_YMM | T_ER_X | T_ER_Y | T_ER_Z | T_MUST_EVEX, 0x7F); }
 void vmovdqu32(const Address& addr, const Xmm& x) { opAVX_X_XM_IMM(x, addr, T_F3 | T_0F | T_EW0 | T_YMM | T_ER_X | T_ER_Y | T_ER_Z | T_MUST_EVEX, 0x7F); }
 void vmovdqu64(const Address& addr, const Xmm& x) { opAVX_X_XM_IMM(x, addr, T_F3 | T_0F | T_EW1 | T_YMM | T_ER_X | T_ER_Y | T_ER_Z | T_MUST_EVEX, 0x7F); }
+void vcompresspd(const Operand& op, const Xmm& x) { opAVX_X_XM_IMM(x, op, T_66 | T_0F38 | T_EW1 | T_YMM | T_MUST_EVEX | T_N8, 0x8A); }
+void vcompressps(const Operand& op, const Xmm& x) { opAVX_X_XM_IMM(x, op, T_66 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX | T_N4, 0x8A); }
+void vpcompressd(const Operand& op, const Xmm& x) { opAVX_X_XM_IMM(x, op, T_66 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX | T_N4, 0x8B); }
+void vpcompressq(const Operand& op, const Xmm& x) { opAVX_X_XM_IMM(x, op, T_66 | T_0F38 | T_EW1 | T_YMM | T_MUST_EVEX | T_N8, 0x8B); }
 void valignd(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F3A | T_EW0 | T_YMM | T_MUST_EVEX, 0x03, imm); }
 void valignq(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F3A | T_EW1 | T_YMM | T_MUST_EVEX, 0x03, imm); }
 void vpandd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F | T_EW0 | T_YMM | T_MUST_EVEX | T_B32, 0xDB); }