add vpcmp*, v(p)blend*
diff --git a/gen/gen_avx512.cpp b/gen/gen_avx512.cpp
index 07ee353..d4dadc0 100644
--- a/gen/gen_avx512.cpp
+++ b/gen/gen_avx512.cpp
@@ -115,6 +115,16 @@
{ 0x65, "vpcmpgtw", T_66 | T_0F | T_MUST_EVEX | T_YMM, false },
{ 0x66, "vpcmpgtd", T_66 | T_0F | T_MUST_EVEX | T_YMM | T_EW0 | T_B32, false },
{ 0x37, "vpcmpgtq", T_66 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW1 | T_B64, false },
+
+ { 0x3F, "vpcmpb", T_66 | T_0F3A | T_MUST_EVEX | T_YMM | T_EW0, true },
+ { 0x3E, "vpcmpub", T_66 | T_0F3A | T_MUST_EVEX | T_YMM | T_EW0, true },
+
+ { 0x3F, "vpcmpw", T_66 | T_0F3A | T_MUST_EVEX | T_YMM | T_EW1, true },
+ { 0x3E, "vpcmpuw", T_66 | T_0F3A | T_MUST_EVEX | T_YMM | T_EW1, true },
+ { 0x1F, "vpcmpd", T_66 | T_0F3A | T_MUST_EVEX | T_YMM | T_EW0 | T_B32, true },
+ { 0x1E, "vpcmpud", T_66 | T_0F3A | T_MUST_EVEX | T_YMM | T_EW0 | T_B32, true },
+ { 0x1F, "vpcmpq", T_66 | T_0F3A | T_MUST_EVEX | T_YMM | T_EW1 | T_B64, true },
+ { 0x1E, "vpcmpuq", T_66 | T_0F3A | T_MUST_EVEX | T_YMM | T_EW1 | T_B64, true },
};
for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) {
const Tbl *p = &tbl[i];
@@ -211,6 +221,13 @@
{ 0x8D, "vpermb", T_66 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW0, false },
{ 0x8D, "vpermw", T_66 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW1, false },
+
+ { 0x65, "vblendmpd", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1 | T_B64, false },
+ { 0x65, "vblendmps", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_B32, false },
+ { 0x66, "vpblendmb", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0, false },
+ { 0x66, "vpblendmw", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1, false },
+ { 0x64, "vpblendmd", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_B32, false },
+ { 0x64, "vpblendmq", 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];
diff --git a/test/make_512.cpp b/test/make_512.cpp
index c6999b8..3eab5cf 100644
--- a/test/make_512.cpp
+++ b/test/make_512.cpp
@@ -28,7 +28,7 @@
const uint64 MEM16 = 1ULL << 16;
const uint64 MEM32 = 1ULL << 17;
const uint64 VM32Z = 1ULL << 19;
-const uint64 CL = 1ULL << 20;
+const uint64 K_K = 1ULL << 20;
const uint64 MEM_ONLY_DISP = 1ULL << 21;
const uint64 NEG32 = 1ULL << 23;
const uint64 _YMM = 1ULL << 24;
@@ -335,8 +335,8 @@
return "ax";
case AL:
return "al";
- case CL:
- return "cl";
+ case K_K:
+ return isXbyak_ ? "k5 | k3" : "k5{k3}";
case IMM32:
return isXbyak_ ? "12345678" : "dword 12345678";
case IMM8:
@@ -1589,10 +1589,62 @@
}
#endif
}
+ void putBlend()
+ {
+ put("vblendmpd", XMM_KZ, _XMM, _XMM | _MEM | M_1to2);
+ put("vblendmpd", YMM_KZ, _YMM, _YMM | _MEM | M_1to4);
+ put("vblendmpd", ZMM_KZ, _ZMM, _ZMM | _MEM | M_1to8);
+
+ put("vblendmps", XMM_KZ, _XMM, _XMM | _MEM | M_1to4);
+ put("vblendmps", YMM_KZ, _YMM, _YMM | _MEM | M_1to8);
+ put("vblendmps", ZMM_KZ, _ZMM, _ZMM | _MEM | M_1to16);
+
+ put("vpblendmb", XMM_KZ, _XMM, _XMM | _MEM);
+ put("vpblendmb", YMM_KZ, _YMM, _YMM | _MEM);
+ put("vpblendmb", ZMM_KZ, _ZMM, _ZMM | _MEM);
+
+ put("vpblendmb", XMM_KZ, _XMM, _XMM | _MEM);
+ put("vpblendmb", YMM_KZ, _YMM, _YMM | _MEM);
+ put("vpblendmb", ZMM_KZ, _ZMM, _ZMM | _MEM);
+
+ put("vpblendmd", XMM_KZ, _XMM, _XMM | _MEM | M_1to4);
+ put("vpblendmd", YMM_KZ, _YMM, _YMM | _MEM | M_1to8);
+ put("vpblendmd", ZMM_KZ, _ZMM, _ZMM | _MEM | M_1to16);
+
+ put("vpblendmq", XMM_KZ, _XMM, _XMM | _MEM | M_1to2);
+ put("vpblendmq", YMM_KZ, _YMM, _YMM | _MEM | M_1to4);
+ put("vpblendmq", ZMM_KZ, _ZMM, _ZMM | _MEM | M_1to8);
+ }
+ void putVpcmp()
+ {
+ const uint64_t b0Tbl[] = { 0, 0, 0 };
+ const uint64_t b4Tbl[] = { M_1to4, M_1to8, M_1to16 };
+ const uint64_t b2Tbl[] = { M_1to2, M_1to4, M_1to8 };
+ const struct Tbl {
+ const char *name;
+ uint64_t b;
+ } tbl[] = {
+ { "vpcmpb", 0 },
+ { "vpcmpub", 0 },
+ { "vpcmpw", 0 },
+ { "vpcmpuw", 0 },
+ { "vpcmpd", M_1to4 },
+ { "vpcmpud", M_1to4 },
+ { "vpcmpq", M_1to2 },
+ { "vpcmpuq", M_1to2 },
+ };
+ for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) {
+ const Tbl& p = tbl[i];
+ const uint64_t *bTbl = p.b == 0 ? b0Tbl : p.b == M_1to4 ? b4Tbl : b2Tbl;
+ put(p.name, K_K, _XMM, _XMM | _MEM | bTbl[0], IMM8);
+ put(p.name, K_K, _YMM, _YMM | _MEM | bTbl[1], IMM8);
+ put(p.name, K_K, _ZMM, _ZMM | _MEM | bTbl[2], IMM8);
+ }
+ }
void putMin()
{
#ifdef XBYAK64
- putGather();
+ putVpcmp();
#endif
}
void putAVX512()
@@ -1631,6 +1683,10 @@
putMisc1();
separateFunc();
putGather();
+ separateFunc();
+ putBlend();
+ separateFunc();
+ putVpcmp();
#endif
}
};
diff --git a/xbyak/xbyak_avx512.h b/xbyak/xbyak_avx512.h
index afb8fce..5590638 100644
--- a/xbyak/xbyak_avx512.h
+++ b/xbyak/xbyak_avx512.h
@@ -76,6 +76,14 @@
void vpcmpgtw(const Opmask& k, const Xmm& x, const Operand& op) { opAVX_K_X_XM(k, x, op, T_66 | T_0F | T_YMM | T_MUST_EVEX, 0x65); }
void vpcmpgtd(const Opmask& k, const Xmm& x, const Operand& op) { opAVX_K_X_XM(k, x, op, T_66 | T_0F | T_EW0 | T_YMM | T_MUST_EVEX | T_B32, 0x66); }
void vpcmpgtq(const Opmask& k, const Xmm& x, const Operand& op) { opAVX_K_X_XM(k, x, op, T_66 | T_0F38 | T_EW1 | T_YMM | T_MUST_EVEX | T_B64, 0x37); }
+void vpcmpb(const Opmask& k, const Xmm& x, const Operand& op, uint8 imm) { opAVX_K_X_XM(k, x, op, T_66 | T_0F3A | T_EW0 | T_YMM | T_MUST_EVEX, 0x3F, imm); }
+void vpcmpub(const Opmask& k, const Xmm& x, const Operand& op, uint8 imm) { opAVX_K_X_XM(k, x, op, T_66 | T_0F3A | T_EW0 | T_YMM | T_MUST_EVEX, 0x3E, imm); }
+void vpcmpw(const Opmask& k, const Xmm& x, const Operand& op, uint8 imm) { opAVX_K_X_XM(k, x, op, T_66 | T_0F3A | T_EW1 | T_YMM | T_MUST_EVEX, 0x3F, imm); }
+void vpcmpuw(const Opmask& k, const Xmm& x, const Operand& op, uint8 imm) { opAVX_K_X_XM(k, x, op, T_66 | T_0F3A | T_EW1 | T_YMM | T_MUST_EVEX, 0x3E, imm); }
+void vpcmpd(const Opmask& k, const Xmm& x, const Operand& op, uint8 imm) { opAVX_K_X_XM(k, x, op, T_66 | T_0F3A | T_EW0 | T_YMM | T_MUST_EVEX | T_B32, 0x1F, imm); }
+void vpcmpud(const Opmask& k, const Xmm& x, const Operand& op, uint8 imm) { opAVX_K_X_XM(k, x, op, T_66 | T_0F3A | T_EW0 | T_YMM | T_MUST_EVEX | T_B32, 0x1E, imm); }
+void vpcmpq(const Opmask& k, const Xmm& x, const Operand& op, uint8 imm) { opAVX_K_X_XM(k, x, op, T_66 | T_0F3A | T_EW1 | T_YMM | T_MUST_EVEX | T_B64, 0x1F, imm); }
+void vpcmpuq(const Opmask& k, const Xmm& x, const Operand& op, uint8 imm) { opAVX_K_X_XM(k, x, op, T_66 | T_0F3A | T_EW1 | T_YMM | T_MUST_EVEX | T_B64, 0x1E, imm); }
void vmovdqa32(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_66 | T_0F | T_EW0 | T_YMM | T_ER_X | T_ER_Y | T_ER_Z | T_MUST_EVEX, 0x6F); }
void vmovdqa64(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_66 | T_0F | T_EW1 | T_YMM | T_ER_X | T_ER_Y | T_ER_Z | T_MUST_EVEX, 0x6F); }
void vmovdqu8(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_F2 | T_0F | T_EW0 | T_YMM | T_ER_X | T_ER_Y | T_ER_Z | T_MUST_EVEX, 0x6F); }
@@ -120,6 +128,12 @@
void vpmullq(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, 0x40); }
void vpermb(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, 0x8D); }
void vpermw(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, 0x8D); }
+void vblendmpd(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, 0x65); }
+void vblendmps(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, 0x65); }
+void vpblendmb(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, 0x66); }
+void vpblendmw(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, 0x66); }
+void vpblendmd(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, 0x64); }
+void vpblendmq(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, 0x64); }
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); }
