add vcvtsi2sh

5 files changed, 24 insertions, 2 deletions

diff --git a/gen/avx_type.hpp b/gen/avx_type.hpp
index 083f646..4652d98 100644
--- a/gen/avx_type.hpp
+++ b/gen/avx_type.hpp
@@ -15,7 +15,7 @@
 		T_66 = 1 << 5, // pp = 1
 		T_F3 = 1 << 6, // pp = 2
 		T_F2 = T_66 | T_F3, // pp = 3
-		// 1 << 7, not used
+		T_ER_R = 1 << 7, // reg{er}
 		T_0F = 1 << 8,
 		T_0F38 = 1 << 9,
 		T_0F3A = 1 << 10,
@@ -143,6 +143,10 @@ std::string type2String(int type)
 		if (!str.empty()) str += " | ";
 		str += "T_ER_Z";
 	}
+	if (type & T_ER_R) {
+		if (!str.empty()) str += " | ";
+		str += "T_ER_R";
+	}
 	if (type & T_SAE_X) {
 		if (!str.empty()) str += " | ";
 		str += "T_SAE_X";
diff --git a/gen/gen_avx512.cpp b/gen/gen_avx512.cpp
index 7e067a1..99a5f8d 100644
--- a/gen/gen_avx512.cpp
+++ b/gen/gen_avx512.cpp
@@ -580,6 +580,8 @@ void putCvt()
 		{ 0x5A, "vcvtpd2ph", T_66 | T_MAP5 | T_MUST_EVEX | T_EW1 | T_B64 | T_ER_Z | T_N16 | T_N_VL, 5 },
 		{ 0x5B, "vcvtqq2ph", T_MAP5 | T_MUST_EVEX | T_EW1 | T_B64 | T_ER_Z | T_N16 | T_N_VL, 5 },
 		{ 0x7A, "vcvtuqq2ph", T_F2 | T_MAP5 | T_MUST_EVEX | T_EW1 | T_B64 | T_ER_Z | T_N16 | T_N_VL, 5 },
+
+		{ 0x2A, "vcvtsi2sh", T_F3 | T_MAP5 | T_MUST_EVEX | T_ER_R | T_M_K, 6 },
 	};
 	for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) {
 		const Tbl& p = tbl[i];
@@ -603,6 +605,9 @@ void putCvt()
 		case 5:
 			printf("void %s(const Xmm& x, const Operand& op) { opCvt5(x, op, %s, 0x%02X); }\n", p.name, type.c_str(), p.code);
 			break;
+		case 6:
+			printf("void %s(const Xmm& x1, const Xmm& x2, const Operand& op) { if (!(x1.isXMM() && x2.isXMM() && op.isBit(32|64))) XBYAK_THROW(ERR_BAD_COMBINATION) int type = (%s) | (op.isBit(32) ? (T_EW0 | T_N4) : (T_EW1 | T_N8)); opVex(x1, &x2, op, type, 0x%02X); }\n", p.name, type.c_str(), p.code);
+			break;
 		}
 	}
 	puts("void vcvtusi2sd(const Xmm& x1, const Xmm& x2, const Operand& op) { opCvt3(x1, x2, op, T_F2 | T_0F | T_MUST_EVEX, T_W1 | T_EW1 | T_ER_X | T_N8, T_W0 | T_EW0 | T_N4, 0x7B); }");
diff --git a/test/misc.cpp b/test/misc.cpp
index 411afa3..1f920a8 100644
--- a/test/misc.cpp
+++ b/test/misc.cpp
@@ -1316,6 +1316,11 @@ CYBOZU_TEST_AUTO(vaddph)
 			vcvttph2qq(zmm1|k5|T_z|T_sae, xmm3);
 			vcvttph2qq(zmm1|k5|T_z, ptr [rax+0x40]);
 			vcvttph2qq(zmm1|k5|T_z, ptr_b [rax+0x40]);
+
+			vcvtsi2sh(xmm1|T_rd_sae, xmm2, eax);
+			vcvtsi2sh(xmm1, xmm2, dword [rax+0x40]);
+			vcvtsi2sh(xmm1|T_rd_sae, xmm2, r9);
+			vcvtsi2sh(xmm1, xmm2, qword [rax+0x40]);
 		}
 	} c;
 	const uint8_t tbl[] = {
@@ -1835,6 +1840,12 @@ CYBOZU_TEST_AUTO(vaddph)
 		0x62, 0xf5, 0x7d, 0x9d, 0x7a, 0xcb,
 		0x62, 0xf5, 0x7d, 0xcd, 0x7a, 0x48, 0x04,
 		0x62, 0xf5, 0x7d, 0xdd, 0x7a, 0x48, 0x20,
+
+		// vcvtsi2sh
+		0x62, 0xf5, 0x6e, 0x38, 0x2a, 0xc8,
+		0x62, 0xf5, 0x6e, 0x08, 0x2a, 0x48, 0x10,
+		0x62, 0xd5, 0xee, 0x38, 0x2a, 0xc9,
+		0x62, 0xf5, 0xee, 0x08, 0x2a, 0x48, 0x08,
 	};
 	const size_t n = sizeof(tbl) / sizeof(tbl[0]);
 	CYBOZU_TEST_EQUAL(c.getSize(), n);
diff --git a/xbyak/xbyak.h b/xbyak/xbyak.h
index 42ce5ce..461a0b6 100644
--- a/xbyak/xbyak.h
+++ b/xbyak/xbyak.h
@@ -1655,7 +1655,7 @@ private:
 		T_66 = 1 << 5, // pp = 1
 		T_F3 = 1 << 6, // pp = 2
 		T_F2 = T_66 | T_F3, // pp = 3
-		// 1 << 7, not used
+		T_ER_R = 1 << 7, // reg{er}
 		T_0F = 1 << 8,
 		T_0F38 = 1 << 9,
 		T_0F3A = 1 << 10,
@@ -1712,6 +1712,7 @@ private:
 	}
 	void verifyER(const Reg& r, int type) const
 	{
+		if ((type & T_ER_R) && r.isREG(32|64)) return;
 		if (((type & T_ER_X) && r.isXMM()) || ((type & T_ER_Y) && r.isYMM()) || ((type & T_ER_Z) && r.isZMM())) return;
 		XBYAK_THROW(ERR_ER_IS_INVALID)
 	}
diff --git a/xbyak/xbyak_mnemonic.h b/xbyak/xbyak_mnemonic.h
index b21986e..8c67a32 100644
--- a/xbyak/xbyak_mnemonic.h
+++ b/xbyak/xbyak_mnemonic.h
@@ -1924,6 +1924,7 @@ void vcvtsh2sd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x
 void vcvtsh2si(const Reg32e& r, const Operand& op) { int type = (T_N2 | T_F3 | T_MAP5 | T_ER_X | T_MUST_EVEX) | (r.isREG(64) ? T_EW1 : T_EW0); opVex(r, &xm0, op, type, 0x2D); }
 void vcvtsh2ss(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_N2 | T_MAP6 | T_EW0 | T_SAE_X | T_MUST_EVEX, 0x13); }
 void vcvtsh2usi(const Reg32e& r, const Operand& op) { int type = (T_N2 | T_F3 | T_MAP5 | T_ER_X | T_MUST_EVEX) | (r.isREG(64) ? T_EW1 : T_EW0); opVex(r, &xm0, op, type, 0x79); }
+void vcvtsi2sh(const Xmm& x1, const Xmm& x2, const Operand& op) { if (!(x1.isXMM() && x2.isXMM() && op.isBit(32|64))) XBYAK_THROW(ERR_BAD_COMBINATION) int type = (T_F3 | T_MAP5 | T_ER_R | T_MUST_EVEX | T_M_K) | (op.isBit(32) ? (T_EW0 | T_N4) : (T_EW1 | T_N8)); opVex(x1, &x2, op, type, 0x2A); }
 void vcvtss2sh(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_N4 | T_MAP5 | T_EW0 | T_ER_X | T_MUST_EVEX, 0x1D); }
 void vcvtss2usi(const Reg32e& r, const Operand& op) { int type = (T_N4 | T_F3 | T_0F | T_ER_X | T_MUST_EVEX) | (r.isREG(64) ? T_EW1 : T_EW0); opVex(r, &xm0, op, type, 0x79); }
 void vcvttpd2qq(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_66 | T_0F | T_EW1 | T_YMM | T_SAE_Z | T_MUST_EVEX | T_B64, 0x7A); }