add vcvtpd2dq, vcvtpd2udq

6 files changed, 34 insertions, 9 deletions

diff --git a/gen/gen_avx512.cpp b/gen/gen_avx512.cpp
index 94ba8f6..b7a34b4 100644
--- a/gen/gen_avx512.cpp
+++ b/gen/gen_avx512.cpp
@@ -138,6 +138,7 @@ void putX_XM()
 		{ 0x6F, "vmovdqu16", T_F2 | T_0F | T_MUST_EVEX | T_YMM | T_EW1 | T_ER_X | T_ER_Y | T_ER_Z  },
 		{ 0x6F, "vmovdqu32", T_F3 | T_0F | T_MUST_EVEX | T_YMM | T_EW0 | T_ER_X | T_ER_Y | T_ER_Z  },
 		{ 0x6F, "vmovdqu64", T_F3 | T_0F | T_MUST_EVEX | T_YMM | T_EW1 | T_ER_X | T_ER_Y | T_ER_Z  },
+		{ 0x7B, "vcvtpd2qq", T_66 | T_0F | T_MUST_EVEX | T_YMM | T_EW1 | T_B64 | T_ER_Z },
 	};
 	for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) {
 		const Tbl *p = &tbl[i];
@@ -311,6 +312,11 @@ void putBroadcast()
 	puts("void vbroadcasti64x4(const Zmm& z, const Operand& op) { opAVX_X_XM_IMM(z, op, T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1 | T_N32, 0x5B); }");
 }
 
+void putCvt()
+{
+	puts("void vcvtpd2udq(const Xmm& x, const Operand& op) { opCvt2(x, op, T_0F | T_YMM | T_MUST_EVEX | T_EW1 | T_B64 | T_ER_Z, 0x79); }");
+}
+
 int main()
 {
 	puts("#ifndef XBYAK_DISABLE_AVX512");
@@ -324,5 +330,6 @@ int main()
 #ifdef XBYAK64
 	putBroadcast();
 #endif
+	putCvt();
 	puts("#endif");
 }
diff --git a/gen/gen_code.cpp b/gen/gen_code.cpp
index 18d1bff..9881eb3 100644
--- a/gen/gen_code.cpp
+++ b/gen/gen_code.cpp
@@ -1588,12 +1588,8 @@ void put()
 		printf("void vcvtdq2pd(const Xmm& x, const Operand& op) { if (!(op.isMEM() || (x.is(Operand::XMM | Operand::YMM) && op.isXMM()) || (x.isZMM() && op.isYMM()))) throw Error(ERR_BAD_COMBINATION);"
 			"opAVX_X_X_XMcvt(x, false, cvtIdx0(x), op, !op.isMEM(), x.isXMM() ? Operand::XMM : Operand::YMM, T_0F | T_F3 | T_YMM | T_EVEX | T_EW0 | T_B32 | T_N8 | T_N_VL, 0xE6); }\n");
 
-		printf("void vcvtpd2ps(const Xmm& x, const Operand& op) { if (x.isYMM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XM(op.isYMM() ? Ymm(x.getIdx()) : x, cvtIdx0(op), op, T_0F | T_66 | T_YMM, 0x5A); }\n");
-
-		puts("void vcvtpd2dq(const Xmm& x, const Operand& op) {"
-			"if (!(op.isMEM() || (x.isXMM() && op.is(Operand::XMM | Operand::YMM)) || (x.isYMM() && op.isZMM()))) throw Error(ERR_BAD_COMBINATION);"
-			"Operand::Kind kind = x.isXMM() ? (op.isBit(256) ? Operand::YMM : Operand::XMM) : Operand::ZMM;"
-			"opVex(x.copyAndSetKind(kind), &xm0, op, T_0F | T_F2 | T_YMM | T_EVEX | T_EW1 | T_B64 | T_ER_Z, 0xE6);}");
+		puts("void vcvtpd2ps(const Xmm& x, const Operand& op) { opCvt2(x, op, T_0F | T_66 | T_YMM | T_EVEX | T_EW1 | T_B64 | T_ER_Z, 0x5A); }");
+		puts("void vcvtpd2dq(const Xmm& x, const Operand& op) { opCvt2(x, op, T_0F | T_F2 | T_YMM | T_EVEX | T_EW1 | T_B64 | T_ER_Z, 0xE6); }");
 
 		printf("void vcvttpd2dq(const Xmm& x, const Operand& op) { if (x.isYMM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XM(op.isYMM() ? Ymm(x.getIdx()) : x, cvtIdx0(op), op, T_0F | T_66 | T_YMM, 0xE6); }\n");
 
diff --git a/test/make_512.cpp b/test/make_512.cpp
index 46beab0..bba0afd 100644
--- a/test/make_512.cpp
+++ b/test/make_512.cpp
@@ -1415,6 +1415,18 @@ public:
 		put("vcvtpd2dq", _XMM | _XMM3, _XMM | M_xword | M_1to2);
 		put("vcvtpd2dq", _XMM | _XMM3, _YMM | M_yword | MY_1to4);
 		put("vcvtpd2dq", YMM | YMM_KZ, ZMM | _MEM | M_1to8);
+
+		put("vcvtpd2ps", _XMM | _XMM3, _XMM | M_xword | M_1to2);
+		put("vcvtpd2ps", _XMM | _XMM3, _YMM | M_yword | MY_1to4);
+		put("vcvtpd2ps", YMM | YMM_KZ, ZMM | _MEM | M_1to8);
+
+		put("vcvtpd2qq", XMM_KZ, _XMM | _MEM | M_1to2);
+		put("vcvtpd2qq", YMM_KZ, _YMM | _MEM | M_1to4);
+		put("vcvtpd2qq", ZMM_KZ, _ZMM | _MEM | M_1to8);
+
+		put("vcvtpd2udq", _XMM | _XMM3, _XMM | M_xword | M_1to2);
+		put("vcvtpd2udq", _XMM | _XMM3, _YMM | M_yword | MY_1to4);
+		put("vcvtpd2udq", YMM | YMM_KZ, ZMM | _MEM | M_1to8);
 #endif
 	}
 	void putMin()
@@ -1429,7 +1441,8 @@ public:
 	void putAVX512()
 	{
 #ifdef MIN_TEST
-		putMin();
+//		putMin();
+		put512_cvt();
 #else
 		putOpmask();
 		separateFunc();
diff --git a/xbyak/xbyak.h b/xbyak/xbyak.h
index 667aa51..99bf96c 100644
--- a/xbyak/xbyak.h
+++ b/xbyak/xbyak.h
@@ -1853,6 +1853,13 @@ private:
 		// use static_cast to avoid calling unintentional copy constructor on gcc
 		opAVX_X_X_XM(x, op1, cvt ? kind == Operand::XMM ? static_cast<const Operand&>(Xmm(op2.getIdx())) : static_cast<const Operand&>(Ymm(op2.getIdx())) : op2, type, code0, imm8);
 	}
+	void opCvt2(const Xmm& x, const Operand& op, int type, int code)
+	{
+		if (!(x.isXMM() && op.is(Operand::XMM | Operand::YMM | Operand::MEM)) && !(x.isYMM() && op.is(Operand::ZMM | Operand::MEM))) throw Error(ERR_BAD_COMBINATION);
+		Operand::Kind kind = x.isXMM() ? (op.isBit(256) ? Operand::YMM : Operand::XMM) : Operand::ZMM;
+		opVex(x.copyAndSetKind(kind), &xm0, op, type, code);
+	}
+
 	const Xmm& cvtIdx0(const Operand& x) const
 	{
 //		assert(!x.isMEM()); // QQQ
diff --git a/xbyak/xbyak_avx512.h b/xbyak/xbyak_avx512.h
index e91e1be..c9ecc9b 100644
--- a/xbyak/xbyak_avx512.h
+++ b/xbyak/xbyak_avx512.h
@@ -82,6 +82,7 @@ void vmovdqu8(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_F2 | T_
 void vmovdqu16(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_F2 | T_0F | T_EW1 | T_YMM | T_ER_X | T_ER_Y | T_ER_Z | T_MUST_EVEX, 0x6F); }
 void vmovdqu32(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_F3 | T_0F | T_EW0 | T_YMM | T_ER_X | T_ER_Y | T_ER_Z | T_MUST_EVEX, 0x6F); }
 void vmovdqu64(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_F3 | T_0F | T_EW1 | T_YMM | T_ER_X | T_ER_Y | T_ER_Z | T_MUST_EVEX, 0x6F); }
+void vcvtpd2qq(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_66 | T_0F | T_EW1 | T_YMM | T_ER_Z | T_MUST_EVEX | T_B64, 0x7B); }
 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); }
@@ -143,4 +144,5 @@ void vbroadcasti32x4(const Ymm& y, const Operand& op) { opAVX_X_XM_IMM(y, op, T_
 void vbroadcasti64x2(const Ymm& y, const Operand& op) { opAVX_X_XM_IMM(y, op, T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1 | T_N16, 0x5A); }
 void vbroadcasti32x8(const Zmm& z, const Operand& op) { opAVX_X_XM_IMM(z, op, T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_N32, 0x5B); }
 void vbroadcasti64x4(const Zmm& z, const Operand& op) { opAVX_X_XM_IMM(z, op, T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1 | T_N32, 0x5B); }
+void vcvtpd2udq(const Xmm& x, const Operand& op) { opCvt2(x, op, T_0F | T_YMM | T_MUST_EVEX | T_EW1 | T_B64 | T_ER_Z, 0x79); }
 #endif
diff --git a/xbyak/xbyak_mnemonic.h b/xbyak/xbyak_mnemonic.h
index 252e1c5..108769c 100644
--- a/xbyak/xbyak_mnemonic.h
+++ b/xbyak/xbyak_mnemonic.h
@@ -1482,8 +1482,8 @@ void vcvtsi2ss(const Xmm& x, const Operand& op1, const Operand& op2 = Operand())
 void vcvtsi2sd(const Xmm& x, const Operand& op1, const Operand& op2 = Operand()) { if (!op2.isNone() && !(op2.isREG(i32e) || op2.isMEM())) throw Error(ERR_BAD_COMBINATION); int type = T_0F | T_F2; if (!op1.isMEM() && !op2.isMEM()) type |= (op1.isREG(32) || op2.isREG(32)) ? T_W0 : T_W1; opAVX_X_X_XMcvt(x, false, op1, op2, op2.isREG(), Operand::XMM, type, 0x2A); }
 void vcvtps2pd(const Xmm& x, const Operand& op) { if (!op.isMEM() && !op.isXMM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x, false, cvtIdx0(x), op, !op.isMEM(), x.isXMM() ? Operand::XMM : Operand::YMM, T_0F | T_YMM, 0x5A); }
 void vcvtdq2pd(const Xmm& x, const Operand& op) { if (!(op.isMEM() || (x.is(Operand::XMM | Operand::YMM) && op.isXMM()) || (x.isZMM() && op.isYMM()))) throw Error(ERR_BAD_COMBINATION);opAVX_X_X_XMcvt(x, false, cvtIdx0(x), op, !op.isMEM(), x.isXMM() ? Operand::XMM : Operand::YMM, T_0F | T_F3 | T_YMM | T_EVEX | T_EW0 | T_B32 | T_N8 | T_N_VL, 0xE6); }
-void vcvtpd2ps(const Xmm& x, const Operand& op) { if (x.isYMM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XM(op.isYMM() ? Ymm(x.getIdx()) : x, cvtIdx0(op), op, T_0F | T_66 | T_YMM, 0x5A); }
-void vcvtpd2dq(const Xmm& x, const Operand& op) {if (!(op.isMEM() || (x.isXMM() && op.is(Operand::XMM | Operand::YMM)) || (x.isYMM() && op.isZMM()))) throw Error(ERR_BAD_COMBINATION);Operand::Kind kind = x.isXMM() ? (op.isBit(256) ? Operand::YMM : Operand::XMM) : Operand::ZMM;opVex(x.copyAndSetKind(kind), &xm0, op, T_0F | T_F2 | T_YMM | T_EVEX | T_EW1 | T_B64 | T_ER_Z, 0xE6);}
+void vcvtpd2ps(const Xmm& x, const Operand& op) { opCvt2(x, op, T_0F | T_66 | T_YMM | T_EVEX | T_EW1 | T_B64 | T_ER_Z, 0x5A); }
+void vcvtpd2dq(const Xmm& x, const Operand& op) { opCvt2(x, op, T_0F | T_F2 | T_YMM | T_EVEX | T_EW1 | T_B64 | T_ER_Z, 0xE6); }
 void vcvttpd2dq(const Xmm& x, const Operand& op) { if (x.isYMM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XM(op.isYMM() ? Ymm(x.getIdx()) : x, cvtIdx0(op), op, T_0F | T_66 | T_YMM, 0xE6); }
 void vcvtph2ps(const Xmm& x, const Operand& op) { if (!op.isMEM() && !op.isXMM()) throw Error(ERR_BAD_COMBINATION); opVex(x, 0, op, T_0F38 | T_66 | T_W0, 0x13); }
 void vcvtps2ph(const Operand& op, const Xmm& x, uint8 imm) { if (!op.isMEM() && !op.isXMM()) throw Error(ERR_BAD_COMBINATION); opVex(x, 0, op, T_0F3A | T_66 | T_W0, 0x1d, imm); }