add vshuf*, vpternlog{d,q}

3 files changed, 63 insertions, 2 deletions

diff --git a/gen/gen_avx512.cpp b/gen/gen_avx512.cpp
index a912d33..d724edb 100644
--- a/gen/gen_avx512.cpp
+++ b/gen/gen_avx512.cpp
@@ -271,6 +271,15 @@ void putX_X_XM_IMM()
 		{ 0x7E, "vpermt2q", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1 | T_B64, false },
 		{ 0x7F, "vpermt2ps", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_B32, false },
 		{ 0x7F, "vpermt2pd", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1 | T_B64, false },
+
+		{ 0x75, "vpermi2w", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1, false },
+		{ 0x76, "vpermi2d", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_B32, false },
+		{ 0x76, "vpermi2q", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1 | T_B64, false },
+		{ 0x77, "vpermi2ps", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_B32, false },
+		{ 0x77, "vpermi2pd", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1 | T_B64, false },
+
+		{ 0x25, "vpternlogd", T_66 | T_0F3A | T_YMM | T_MUST_EVEX | T_EW0 | T_B32, true },
+		{ 0x25, "vpternlogq", T_66 | T_0F3A | T_YMM | T_MUST_EVEX | T_EW1 | T_B64, true },
 	};
 	for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) {
 		const Tbl *p = &tbl[i];
@@ -444,6 +453,14 @@ void putGather()
 	}
 }
 
+void putShuff()
+{
+	puts("void vshuff32x4(const Ymm& y1, const Ymm& y2, const Operand& op, uint8 imm) { opAVX_X_X_XM(y1, y2, op, T_66 | T_0F3A | T_YMM | T_MUST_EVEX | T_EW0 | T_B32, 0x23, imm); }");
+	puts("void vshuff64x2(const Ymm& y1, const Ymm& y2, const Operand& op, uint8 imm) { opAVX_X_X_XM(y1, y2, op, T_66 | T_0F3A | T_YMM | T_MUST_EVEX | T_EW1 | T_B64, 0x23, imm); }");
+	puts("void vshufi32x4(const Ymm& y1, const Ymm& y2, const Operand& op, uint8 imm) { opAVX_X_X_XM(y1, y2, op, T_66 | T_0F3A | T_YMM | T_MUST_EVEX | T_EW0 | T_B32, 0x43, imm); }");
+	puts("void vshufi64x2(const Ymm& y1, const Ymm& y2, const Operand& op, uint8 imm) { opAVX_X_X_XM(y1, y2, op, T_66 | T_0F3A | T_YMM | T_MUST_EVEX | T_EW1 | T_B64, 0x43, imm); }");
+}
+
 int main()
 {
 	puts("#ifndef XBYAK_DISABLE_AVX512");
@@ -460,5 +477,6 @@ int main()
 #endif
 	putCvt();
 	putGather();
+	putShuff();
 	puts("#endif");
 }
diff --git a/test/make_512.cpp b/test/make_512.cpp
index 3f65659..9a40304 100644
--- a/test/make_512.cpp
+++ b/test/make_512.cpp
@@ -30,7 +30,7 @@ const uint64 MEM32 = 1ULL << 17;
 const uint64 VM32Z = 1ULL << 19;
 const uint64 K_K = 1ULL << 20;
 const uint64 MEM_ONLY_DISP = 1ULL << 21;
-const uint64 NEG32 = 1ULL << 23;
+//const uint64 QQQ = 1ULL << 23;
 const uint64 _YMM = 1ULL << 24;
 const uint64 VM32X_32 = 1ULL << 39;
 const uint64 VM32X_64 = 1ULL << 40;
@@ -1714,6 +1714,11 @@ public:
 			{ "vpermt2q", M_1to2 },
 			{ "vpermt2ps", M_1to4 },
 			{ "vpermt2pd", M_1to2 },
+
+			{ "vpermi2w", 0 },
+			{ "vpermi2d", M_1to4 },
+			{ "vpermi2q", M_1to2 },
+			{ "vpermi2ps", M_1to4 },
 		};
 		for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) {
 			const Tbl& p = tbl[i];
@@ -1723,10 +1728,33 @@ public:
 			put(p.name, ZMM_KZ, _ZMM, _ZMM | _MEM | bTbl[2]);
 		}
 	}
+	void putShuff()
+	{
+		put("vshuff32x4", YMM_KZ, _YMM, _YMM | _MEM | M_1to8, IMM8);
+		put("vshuff32x4", ZMM_KZ, _ZMM, _ZMM | _MEM | M_1to16, IMM8);
+
+		put("vshuff64x2", YMM_KZ, _YMM, _YMM | _MEM | M_1to4, IMM8);
+		put("vshuff64x2", ZMM_KZ, _ZMM, _ZMM | _MEM | M_1to8, IMM8);
+
+		put("vshufi32x4", YMM_KZ, _YMM, _YMM | _MEM | M_1to8, IMM8);
+		put("vshufi32x4", ZMM_KZ, _ZMM, _ZMM | _MEM | M_1to16, IMM8);
+
+		put("vshufi64x2", YMM_KZ, _YMM, _YMM | _MEM | M_1to4, IMM8);
+		put("vshufi64x2", ZMM_KZ, _ZMM, _ZMM | _MEM | M_1to8, IMM8);
+	}
+	void putMisc2()
+	{
+		put("vpternlogd", XMM_KZ, _XMM, _XMM | _MEM | M_1to4, IMM8);
+		put("vpternlogd", YMM_KZ, _YMM, _YMM | _MEM | M_1to8, IMM8);
+		put("vpternlogd", ZMM_KZ, _ZMM, _ZMM | _MEM | M_1to16, IMM8);
+
+		put("vpternlogq", XMM_KZ, _XMM, _XMM | _MEM | M_1to2, IMM8);
+		put("vpternlogq", YMM_KZ, _YMM, _YMM | _MEM | M_1to4, IMM8);
+		put("vpternlogq", ZMM_KZ, _ZMM, _ZMM | _MEM | M_1to8, IMM8);
+	}
 	void putMin()
 	{
 #ifdef XBYAK64
-		putPerm();
 #endif
 	}
 	void putAVX512()
@@ -1775,6 +1803,10 @@ public:
 		putCompExp();
 		separateFunc();
 		putPerm();
+		separateFunc();
+		putShuff();
+		separateFunc();
+		putMisc2();
 #endif
 	}
 };
diff --git a/xbyak/xbyak_avx512.h b/xbyak/xbyak_avx512.h
index 48c5ec9..e22bd05 100644
--- a/xbyak/xbyak_avx512.h
+++ b/xbyak/xbyak_avx512.h
@@ -156,6 +156,13 @@ void vpermt2d(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1
 void vpermt2q(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_EW1 | T_YMM | T_MUST_EVEX | T_B64, 0x7E); }
 void vpermt2ps(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX | T_B32, 0x7F); }
 void vpermt2pd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_EW1 | T_YMM | T_MUST_EVEX | T_B64, 0x7F); }
+void vpermi2w(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_EW1 | T_YMM | T_MUST_EVEX, 0x75); }
+void vpermi2d(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX | T_B32, 0x76); }
+void vpermi2q(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_EW1 | T_YMM | T_MUST_EVEX | T_B64, 0x76); }
+void vpermi2ps(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX | T_B32, 0x77); }
+void vpermi2pd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_EW1 | T_YMM | T_MUST_EVEX | T_B64, 0x77); }
+void vpternlogd(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 | T_B32, 0x25, imm); }
+void vpternlogq(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 | T_B64, 0x25, imm); }
 void vpsraq(const Xmm& x, const Operand& op, uint8 imm) { opAVX_X_X_XM(Xmm(x.getKind(), 4), x, op, T_66 | T_0F | T_EW1 | T_YMM | T_MUST_EVEX | T_B64, 0x72, imm); }
 void vextractf32x4(const Operand& op, const Ymm& r, uint8 imm) { opAVX_X_X_XMcvt(r, true, cvtIdx0(r), op, op.isXMM(), Operand::YMM, T_66 | T_0F3A | T_EW0 | T_YMM | T_MUST_EVEX | T_N16, 0x19, imm); }
 void vextractf64x2(const Operand& op, const Ymm& r, uint8 imm) { opAVX_X_X_XMcvt(r, true, cvtIdx0(r), op, op.isXMM(), Operand::YMM, T_66 | T_0F3A | T_EW1 | T_YMM | T_MUST_EVEX | T_N16, 0x19, imm); }
@@ -217,4 +224,8 @@ void vgatherdps(const Xmm& x, const Address& addr) { opGather2(x, addr, T_66 | T
 void vgatherdpd(const Xmm& x, const Address& addr) { opGather2(x, addr, T_66 | T_0F38 | T_EW1 | T_YMM | T_MUST_EVEX | T_N8, 0x92, 1); }
 void vgatherqps(const Xmm& x, const Address& addr) { opGather2(x, addr, T_66 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX | T_N4, 0x93, 2); }
 void vgatherqpd(const Xmm& x, const Address& addr) { opGather2(x, addr, T_66 | T_0F38 | T_EW1 | T_YMM | T_MUST_EVEX | T_N8, 0x93, 0); }
+void vshuff32x4(const Ymm& y1, const Ymm& y2, const Operand& op, uint8 imm) { opAVX_X_X_XM(y1, y2, op, T_66 | T_0F3A | T_YMM | T_MUST_EVEX | T_EW0 | T_B32, 0x23, imm); }
+void vshuff64x2(const Ymm& y1, const Ymm& y2, const Operand& op, uint8 imm) { opAVX_X_X_XM(y1, y2, op, T_66 | T_0F3A | T_YMM | T_MUST_EVEX | T_EW1 | T_B64, 0x23, imm); }
+void vshufi32x4(const Ymm& y1, const Ymm& y2, const Operand& op, uint8 imm) { opAVX_X_X_XM(y1, y2, op, T_66 | T_0F3A | T_YMM | T_MUST_EVEX | T_EW0 | T_B32, 0x43, imm); }
+void vshufi64x2(const Ymm& y1, const Ymm& y2, const Operand& op, uint8 imm) { opAVX_X_X_XM(y1, y2, op, T_66 | T_0F3A | T_YMM | T_MUST_EVEX | T_EW1 | T_B64, 0x43, imm); }
 #endif