add v{gather,scatter}pf* vrange, vreduce, vpmadd, etc.
diff --git a/gen/gen_avx512.cpp b/gen/gen_avx512.cpp
index 0886e76..9035ecc 100644
--- a/gen/gen_avx512.cpp
+++ b/gen/gen_avx512.cpp
@@ -135,6 +135,8 @@
{ 0x26, "vptestnmw", T_F3 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW1, false },
{ 0x27, "vptestnmd", T_F3 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW0 | T_B32, false },
{ 0x27, "vptestnmq", T_F3 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW1 | T_B64, false },
+
+//QQQ { 0x66, "vfpclassps", T_66 | T_0F3A | T_MUST_EVEX | T_YMM |T_EW0 | T_B32, true },
};
for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) {
const Tbl *p = &tbl[i];
@@ -183,6 +185,14 @@
printf("void %s(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, %s, 0x%02X); }\n", p->name, type.c_str(), p->code);
}
puts("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); }");
+
+ puts("void vexp2pd(const Zmm& z, const Operand& op) { opAVX_X_XM_IMM(z, op, T_66 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW1 | T_B64 | T_SAE_Z, 0xC8); }");
+ puts("void vexp2ps(const Zmm& z, const Operand& op) { opAVX_X_XM_IMM(z, op, T_66 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW0 | T_B32 | T_SAE_Z, 0xC8); }");
+ puts("void vrcp28pd(const Zmm& z, const Operand& op) { opAVX_X_XM_IMM(z, op, T_66 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW1 | T_B64 | T_SAE_Z, 0xCA); }");
+ puts("void vrcp28ps(const Zmm& z, const Operand& op) { opAVX_X_XM_IMM(z, op, T_66 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW0 | T_B32 | T_SAE_Z, 0xCA); }");
+
+ puts("void vrsqrt28pd(const Zmm& z, const Operand& op) { opAVX_X_XM_IMM(z, op, T_66 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW1 | T_B64 | T_SAE_Z, 0xCC); }");
+ puts("void vrsqrt28ps(const Zmm& z, const Operand& op) { opAVX_X_XM_IMM(z, op, T_66 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW0 | T_B32 | T_SAE_Z, 0xCC); }");
}
void putM_X()
@@ -316,6 +326,22 @@
{ 0x14, "vprorvd", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_B32, false },
{ 0x14, "vprorvq", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1 | T_B64, false },
+ { 0xCB, "vrcp28sd", T_66 | T_0F38 | T_MUST_EVEX | T_EW1 | T_N8 | T_SAE_X, false },
+ { 0xCB, "vrcp28ss", T_66 | T_0F38 | T_MUST_EVEX | T_EW0 | T_N4 | T_SAE_X, false },
+
+ { 0xCD, "vrsqrt28sd", T_66 | T_0F38 | T_MUST_EVEX | T_EW1 | T_N8 | T_SAE_X, false },
+ { 0xCD, "vrsqrt28ss", T_66 | T_0F38 | T_MUST_EVEX | T_EW0 | T_N4 | T_SAE_X, false },
+
+ { 0x50, "vrangepd", T_66 | T_0F3A | T_YMM | T_MUST_EVEX | T_EW1 | T_B64 | T_SAE_Z, true },
+ { 0x50, "vrangeps", T_66 | T_0F3A | T_YMM | T_MUST_EVEX | T_EW0 | T_B32 | T_SAE_Z, true },
+ { 0x51, "vrangesd", T_66 | T_0F3A | T_MUST_EVEX | T_EW1 | T_SAE_X | T_N8, true },
+ { 0x51, "vrangess", T_66 | T_0F3A | T_MUST_EVEX | T_EW0 | T_SAE_X | T_N4, true },
+
+ { 0x57, "vreducesd", T_66 | T_0F3A | T_MUST_EVEX | T_EW1 | T_SAE_X | T_N8, true },
+ { 0x57, "vreducess", T_66 | T_0F3A | T_MUST_EVEX | T_EW0 | T_SAE_X | T_N4, true },
+
+ { 0xB4, "vpmadd52luq", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1 | T_B64, false },
+ { 0xB5, "vpmadd52huq", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1 | T_B64, false },
};
for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) {
const Tbl *p = &tbl[i];
@@ -601,6 +627,9 @@
{ 0x44, "vplzcntd", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_B32, false },
{ 0x44, "vplzcntq", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1 | T_B64, false },
+
+ { 0x56, "vreducepd", T_66 | T_0F3A | T_YMM | T_MUST_EVEX | T_EW1 | T_B64 | T_SAE_Z, true },
+ { 0x56, "vreduceps", T_66 | T_0F3A | T_YMM | T_MUST_EVEX | T_EW0 | T_B32 | T_SAE_Z, true },
};
for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) {
const Tbl *p = &tbl[i];
@@ -614,6 +643,26 @@
{
puts("void vpbroadcastmb2q(const Xmm& x, const Opmask& k) { opVex(x, 0, k, T_F3 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1, 0x2A); }");
puts("void vpbroadcastmw2d(const Xmm& x, const Opmask& k) { opVex(x, 0, k, T_F3 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0, 0x3A); }");
+
+ puts("void vgatherpf0dps(const Address& addr) { opGatherFetch(addr, zm1, T_66 | T_0F38 | T_EW0 | T_MUST_EVEX | T_N4, 0xC6, Operand::ZMM); }");
+ puts("void vgatherpf0qps(const Address& addr) { opGatherFetch(addr, zm1, T_66 | T_0F38 | T_EW0 | T_MUST_EVEX | T_N4, 0xC7, Operand::ZMM); }");
+ puts("void vgatherpf0dpd(const Address& addr) { opGatherFetch(addr, zm1, T_66 | T_0F38 | T_EW1 | T_MUST_EVEX | T_N8, 0xC6, Operand::YMM); }");
+ puts("void vgatherpf0qpd(const Address& addr) { opGatherFetch(addr, zm1, T_66 | T_0F38 | T_EW1 | T_MUST_EVEX | T_N8, 0xC7, Operand::ZMM); }");
+
+ puts("void vgatherpf1dps(const Address& addr) { opGatherFetch(addr, zm2, T_66 | T_0F38 | T_EW0 | T_MUST_EVEX | T_N4, 0xC6, Operand::ZMM); }");
+ puts("void vgatherpf1qps(const Address& addr) { opGatherFetch(addr, zm2, T_66 | T_0F38 | T_EW0 | T_MUST_EVEX | T_N4, 0xC7, Operand::ZMM); }");
+ puts("void vgatherpf1dpd(const Address& addr) { opGatherFetch(addr, zm2, T_66 | T_0F38 | T_EW1 | T_MUST_EVEX | T_N8, 0xC6, Operand::YMM); }");
+ puts("void vgatherpf1qpd(const Address& addr) { opGatherFetch(addr, zm2, T_66 | T_0F38 | T_EW1 | T_MUST_EVEX | T_N8, 0xC7, Operand::ZMM); }");
+
+ puts("void vscatterpf0dps(const Address& addr) { opGatherFetch(addr, zm5, T_66 | T_0F38 | T_EW0 | T_MUST_EVEX | T_N4, 0xC6, Operand::ZMM); }");
+ puts("void vscatterpf0qps(const Address& addr) { opGatherFetch(addr, zm5, T_66 | T_0F38 | T_EW0 | T_MUST_EVEX | T_N4, 0xC7, Operand::ZMM); }");
+ puts("void vscatterpf0dpd(const Address& addr) { opGatherFetch(addr, zm5, T_66 | T_0F38 | T_EW1 | T_MUST_EVEX | T_N8, 0xC6, Operand::YMM); }");
+ puts("void vscatterpf0qpd(const Address& addr) { opGatherFetch(addr, zm5, T_66 | T_0F38 | T_EW1 | T_MUST_EVEX | T_N8, 0xC7, Operand::ZMM); }");
+
+ puts("void vscatterpf1dps(const Address& addr) { opGatherFetch(addr, zm6, T_66 | T_0F38 | T_EW0 | T_MUST_EVEX | T_N4, 0xC6, Operand::ZMM); }");
+ puts("void vscatterpf1qps(const Address& addr) { opGatherFetch(addr, zm6, T_66 | T_0F38 | T_EW0 | T_MUST_EVEX | T_N4, 0xC7, Operand::ZMM); }");
+ puts("void vscatterpf1dpd(const Address& addr) { opGatherFetch(addr, zm6, T_66 | T_0F38 | T_EW1 | T_MUST_EVEX | T_N8, 0xC6, Operand::YMM); }");
+ puts("void vscatterpf1qpd(const Address& addr) { opGatherFetch(addr, zm6, T_66 | T_0F38 | T_EW1 | T_MUST_EVEX | T_N8, 0xC7, Operand::ZMM); }");
}
int main()
diff --git a/test/make_512.cpp b/test/make_512.cpp
index 9cee16a..cd53f39 100644
--- a/test/make_512.cpp
+++ b/test/make_512.cpp
@@ -1869,6 +1869,7 @@
}
void putMisc2()
{
+#ifdef XBYAK64
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);
@@ -1983,6 +1984,70 @@
put("vpbroadcastmb2q", _XMM | _YMM | _ZMM, K);
put("vpbroadcastmw2d", _XMM | _YMM | _ZMM, K);
+
+ put("vexp2pd", ZMM_KZ, _ZMM | _MEM | M_1to8 | ZMM_SAE);
+ put("vexp2ps", ZMM_KZ, _ZMM | _MEM | M_1to16 | ZMM_SAE);
+
+ put("vrcp28pd", ZMM_KZ, _ZMM | _MEM | M_1to8 | ZMM_SAE);
+ put("vrcp28ps", ZMM_KZ, _ZMM | _MEM | M_1to16 | ZMM_SAE);
+ put("vrcp28sd", XMM_KZ, _XMM, _XMM | _MEM | XMM_SAE);
+ put("vrcp28ss", XMM_KZ, _XMM, _XMM | _MEM | XMM_SAE);
+
+ put("vrsqrt28pd", ZMM_KZ, _ZMM | _MEM | M_1to8 | ZMM_SAE);
+ put("vrsqrt28ps", ZMM_KZ, _ZMM | _MEM | M_1to16 | ZMM_SAE);
+ put("vrsqrt28sd", XMM_KZ, _XMM, _XMM | _MEM | XMM_SAE);
+ put("vrsqrt28ss", XMM_KZ, _XMM, _XMM | _MEM | XMM_SAE);
+
+ put("vgatherpf0dps", VM32Z_K);
+ put("vgatherpf0qps", VM32Z_K);
+ put("vgatherpf0dpd", VM32Y_K);
+ put("vgatherpf0qpd", VM32Z_K);
+
+ put("vgatherpf1dps", VM32Z_K);
+ put("vgatherpf1qps", VM32Z_K);
+ put("vgatherpf1dpd", VM32Y_K);
+ put("vgatherpf1qpd", VM32Z_K);
+
+ put("vscatterpf0dps", VM32Z_K);
+ put("vscatterpf0qps", VM32Z_K);
+ put("vscatterpf0dpd", VM32Y_K);
+ put("vscatterpf0qpd", VM32Z_K);
+
+ put("vscatterpf1dps", VM32Z_K);
+ put("vscatterpf1qps", VM32Z_K);
+ put("vscatterpf1dpd", VM32Y_K);
+ put("vscatterpf1qpd", VM32Z_K);
+
+ put("vrangepd", XMM_KZ, _XMM, _XMM | _MEM | M_1to2, IMM8);
+ put("vrangepd", YMM_KZ, _YMM, _YMM | _MEM | M_1to4, IMM8);
+ put("vrangepd", ZMM_KZ, _ZMM, _ZMM | _MEM | M_1to8 | ZMM_SAE, IMM8);
+
+ put("vrangeps", XMM_KZ, _XMM, _XMM | _MEM | M_1to4, IMM8);
+ put("vrangeps", YMM_KZ, _YMM, _YMM | _MEM | M_1to8, IMM8);
+ put("vrangeps", ZMM_KZ, _ZMM, _ZMM | _MEM | M_1to16 | ZMM_SAE, IMM8);
+
+ put("vrangesd", XMM_KZ, _XMM, _XMM | _MEM | XMM_SAE, IMM8);
+ put("vrangess", XMM_KZ, _XMM, _XMM | _MEM | XMM_SAE, IMM8);
+
+ put("vreducepd", XMM_KZ, _XMM | _MEM | M_1to2, IMM8);
+ put("vreducepd", YMM_KZ, _YMM | _MEM | M_1to4, IMM8);
+ put("vreducepd", ZMM_KZ, _ZMM | _MEM | M_1to8 | ZMM_SAE, IMM8);
+
+ put("vreduceps", XMM_KZ, _XMM | _MEM | M_1to4, IMM8);
+ put("vreduceps", YMM_KZ, _YMM | _MEM | M_1to8, IMM8);
+ put("vreduceps", ZMM_KZ, _ZMM | _MEM | M_1to16 | ZMM_SAE, IMM8);
+
+ put("vreducesd", XMM_KZ, _XMM, _XMM | _MEM | XMM_SAE, IMM8);
+ put("vreducess", XMM_KZ, _XMM, _XMM | _MEM | XMM_SAE, IMM8);
+
+ put("vpmadd52luq", XMM_KZ, _XMM, _XMM | _MEM | M_1to2);
+ put("vpmadd52luq", YMM_KZ, _YMM, _YMM | _MEM | M_1to4);
+ put("vpmadd52luq", ZMM_KZ, _ZMM, _ZMM | _MEM | M_1to8);
+
+ put("vpmadd52huq", XMM_KZ, _XMM, _XMM | _MEM | M_1to2);
+ put("vpmadd52huq", YMM_KZ, _YMM, _YMM | _MEM | M_1to4);
+ put("vpmadd52huq", ZMM_KZ, _ZMM, _ZMM | _MEM | M_1to8);
+#endif
}
void putMin()
{
diff --git a/xbyak/xbyak.h b/xbyak/xbyak.h
index 4dd14a2..a7baf5a 100644
--- a/xbyak/xbyak.h
+++ b/xbyak/xbyak.h
@@ -1948,6 +1948,13 @@
}
opVex(x, 0, op, type, code);
}
+ void opGatherFetch(const Address& addr, const Xmm& x, int type, uint8 code, Operand::Kind kind)
+ {
+ if (addr.hasZero()) throw Error(ERR_INVALID_ZERO);
+ if (addr.getRegExp().getIndex().getKind() != kind) throw Error(ERR_BAD_VSIB_ADDRESSING);
+ addr.permitVsib();
+ opVex(x, 0, addr, type, code);
+ }
public:
unsigned int getVersion() const { return VERSION; }
using CodeArray::db;
diff --git a/xbyak/xbyak_avx512.h b/xbyak/xbyak_avx512.h
index 188aa9d..8c01fe2 100644
--- a/xbyak/xbyak_avx512.h
+++ b/xbyak/xbyak_avx512.h
@@ -114,6 +114,12 @@
void vgetexppd(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_66 | T_0F38 | T_EW1 | T_YMM | T_SAE_Z | T_MUST_EVEX | T_B64, 0x42); }
void vgetexpps(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_66 | T_0F38 | T_EW0 | T_YMM | T_SAE_Z | T_MUST_EVEX | T_B32, 0x42); }
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 vexp2pd(const Zmm& z, const Operand& op) { opAVX_X_XM_IMM(z, op, T_66 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW1 | T_B64 | T_SAE_Z, 0xC8); }
+void vexp2ps(const Zmm& z, const Operand& op) { opAVX_X_XM_IMM(z, op, T_66 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW0 | T_B32 | T_SAE_Z, 0xC8); }
+void vrcp28pd(const Zmm& z, const Operand& op) { opAVX_X_XM_IMM(z, op, T_66 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW1 | T_B64 | T_SAE_Z, 0xCA); }
+void vrcp28ps(const Zmm& z, const Operand& op) { opAVX_X_XM_IMM(z, op, T_66 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW0 | T_B32 | T_SAE_Z, 0xCA); }
+void vrsqrt28pd(const Zmm& z, const Operand& op) { opAVX_X_XM_IMM(z, op, T_66 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW1 | T_B64 | T_SAE_Z, 0xCC); }
+void vrsqrt28ps(const Zmm& z, const Operand& op) { opAVX_X_XM_IMM(z, op, T_66 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW0 | T_B32 | T_SAE_Z, 0xCC); }
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); }
void vmovdqu8(const Address& addr, const Xmm& x) { opAVX_X_XM_IMM(x, addr, T_F2 | T_0F | T_EW0 | T_YMM | T_ER_X | T_ER_Y | T_ER_Z | T_MUST_EVEX, 0x7F); }
@@ -189,6 +195,18 @@
void vprolvq(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, 0x15); }
void vprorvd(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, 0x14); }
void vprorvq(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, 0x14); }
+void vrcp28sd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_EW1 | T_SAE_X | T_MUST_EVEX | T_N8, 0xCB); }
+void vrcp28ss(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_EW0 | T_SAE_X | T_MUST_EVEX | T_N4, 0xCB); }
+void vrsqrt28sd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_EW1 | T_SAE_X | T_MUST_EVEX | T_N8, 0xCD); }
+void vrsqrt28ss(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_EW0 | T_SAE_X | T_MUST_EVEX | T_N4, 0xCD); }
+void vrangepd(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_SAE_Z | T_MUST_EVEX | T_B64, 0x50, imm); }
+void vrangeps(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_SAE_Z | T_MUST_EVEX | T_B32, 0x50, imm); }
+void vrangesd(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_SAE_X | T_MUST_EVEX | T_N8, 0x51, imm); }
+void vrangess(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_SAE_X | T_MUST_EVEX | T_N4, 0x51, imm); }
+void vreducesd(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_SAE_X | T_MUST_EVEX | T_N8, 0x57, imm); }
+void vreducess(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_SAE_X | T_MUST_EVEX | T_N4, 0x57, imm); }
+void vpmadd52luq(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, 0xB4); }
+void vpmadd52huq(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, 0xB5); }
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 vprold(const Xmm& x, const Operand& op, uint8 imm) { opAVX_X_X_XM(Xmm(x.getKind(), 1), x, op, T_66 | T_0F | T_EW0 | T_YMM | T_MUST_EVEX | T_B32, 0x72, imm); }
void vprolq(const Xmm& x, const Operand& op, uint8 imm) { opAVX_X_X_XM(Xmm(x.getKind(), 1), x, op, T_66 | T_0F | T_EW1 | T_YMM | T_MUST_EVEX | T_B64, 0x72, imm); }
@@ -297,8 +315,26 @@
void vpconflictq(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_66 | T_0F38 | T_EW1 | T_YMM | T_MUST_EVEX | T_B64, 0xC4); }
void vplzcntd(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_66 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX | T_B32, 0x44); }
void vplzcntq(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_66 | T_0F38 | T_EW1 | T_YMM | T_MUST_EVEX | T_B64, 0x44); }
+void vreducepd(const Xmm& x, const Operand& op, uint8 imm) { opAVX_X_XM_IMM(x, op, T_66 | T_0F3A | T_EW1 | T_YMM | T_SAE_Z | T_MUST_EVEX | T_B64, 0x56, imm); }
+void vreduceps(const Xmm& x, const Operand& op, uint8 imm) { opAVX_X_XM_IMM(x, op, T_66 | T_0F3A | T_EW0 | T_YMM | T_SAE_Z | T_MUST_EVEX | T_B32, 0x56, imm); }
void vpbroadcastmb2q(const Xmm& x, const Opmask& k) { opVex(x, 0, k, T_F3 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1, 0x2A); }
void vpbroadcastmw2d(const Xmm& x, const Opmask& k) { opVex(x, 0, k, T_F3 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0, 0x3A); }
+void vgatherpf0dps(const Address& addr) { opGatherFetch(addr, zm1, T_66 | T_0F38 | T_EW0 | T_MUST_EVEX | T_N4, 0xC6, Operand::ZMM); }
+void vgatherpf0qps(const Address& addr) { opGatherFetch(addr, zm1, T_66 | T_0F38 | T_EW0 | T_MUST_EVEX | T_N4, 0xC7, Operand::ZMM); }
+void vgatherpf0dpd(const Address& addr) { opGatherFetch(addr, zm1, T_66 | T_0F38 | T_EW1 | T_MUST_EVEX | T_N8, 0xC6, Operand::YMM); }
+void vgatherpf0qpd(const Address& addr) { opGatherFetch(addr, zm1, T_66 | T_0F38 | T_EW1 | T_MUST_EVEX | T_N8, 0xC7, Operand::ZMM); }
+void vgatherpf1dps(const Address& addr) { opGatherFetch(addr, zm2, T_66 | T_0F38 | T_EW0 | T_MUST_EVEX | T_N4, 0xC6, Operand::ZMM); }
+void vgatherpf1qps(const Address& addr) { opGatherFetch(addr, zm2, T_66 | T_0F38 | T_EW0 | T_MUST_EVEX | T_N4, 0xC7, Operand::ZMM); }
+void vgatherpf1dpd(const Address& addr) { opGatherFetch(addr, zm2, T_66 | T_0F38 | T_EW1 | T_MUST_EVEX | T_N8, 0xC6, Operand::YMM); }
+void vgatherpf1qpd(const Address& addr) { opGatherFetch(addr, zm2, T_66 | T_0F38 | T_EW1 | T_MUST_EVEX | T_N8, 0xC7, Operand::ZMM); }
+void vscatterpf0dps(const Address& addr) { opGatherFetch(addr, zm5, T_66 | T_0F38 | T_EW0 | T_MUST_EVEX | T_N4, 0xC6, Operand::ZMM); }
+void vscatterpf0qps(const Address& addr) { opGatherFetch(addr, zm5, T_66 | T_0F38 | T_EW0 | T_MUST_EVEX | T_N4, 0xC7, Operand::ZMM); }
+void vscatterpf0dpd(const Address& addr) { opGatherFetch(addr, zm5, T_66 | T_0F38 | T_EW1 | T_MUST_EVEX | T_N8, 0xC6, Operand::YMM); }
+void vscatterpf0qpd(const Address& addr) { opGatherFetch(addr, zm5, T_66 | T_0F38 | T_EW1 | T_MUST_EVEX | T_N8, 0xC7, Operand::ZMM); }
+void vscatterpf1dps(const Address& addr) { opGatherFetch(addr, zm6, T_66 | T_0F38 | T_EW0 | T_MUST_EVEX | T_N4, 0xC6, Operand::ZMM); }
+void vscatterpf1qps(const Address& addr) { opGatherFetch(addr, zm6, T_66 | T_0F38 | T_EW0 | T_MUST_EVEX | T_N4, 0xC7, Operand::ZMM); }
+void vscatterpf1dpd(const Address& addr) { opGatherFetch(addr, zm6, T_66 | T_0F38 | T_EW1 | T_MUST_EVEX | T_N8, 0xC6, Operand::YMM); }
+void vscatterpf1qpd(const Address& addr) { opGatherFetch(addr, zm6, T_66 | T_0F38 | T_EW1 | T_MUST_EVEX | T_N8, 0xC7, Operand::ZMM); }
void vpscatterdd(const Address& addr, const Xmm& x) { opGather2(x, addr, T_66 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX | T_M_K | T_N4, 0xA0, 0); }
void vpscatterdq(const Address& addr, const Xmm& x) { opGather2(x, addr, T_66 | T_0F38 | T_EW1 | T_YMM | T_MUST_EVEX | T_M_K | T_N8, 0xA0, 1); }
void vpscatterqd(const Address& addr, const Xmm& x) { opGather2(x, addr, T_66 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX | T_M_K | T_N4, 0xA1, 2); }
