eden-emu/xbyak
15
0
Fork 0
mirror of https://github.com/herumi/xbyak.git synced 2025-06-07 07:50:47 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

v(p)broadcast* supports disp8N

Browse source
This commit is contained in:
MITSUNARI Shigeo 2016-07-17 17:57:15 +09:00
parent e511e77e5c
commit 9019060b68
7 changed files with 57 additions and 49 deletions
Download patch file Download diff file Expand all files Collapse all files

9
gen/avx_type.hpp
View file

@ -28,8 +28,9 @@
T_N4 = 1 << 25,
T_N8 = 1 << 26,
T_N16 = 1 << 27,
T_N_VL = 1 << 28, // N * (1, 2, 4) for VL
T_DUP = 1 << 29, // N = (8, 32, 64)
T_N32 = 1 << 28,
T_N_VL = 1 << 29, // N * (1, 2, 4) for VL
T_DUP = 1 << 30, // N = (8, 32, 64)
T_XXX
};
@ -150,6 +151,10 @@ std::string type2String(int type)
if (!str.empty()) str += " | ";
str += "T_N16";
}
if (type & T_N32) {
if (!str.empty()) str += " | ";
str += "T_N32";
}
if (type & T_N_VL) {
if (!str.empty()) str += " | ";
str += "T_N_VL";

26
gen/gen_avx512.cpp
View file

@ -280,10 +280,10 @@ void putExtractInsert()
void putBroadcast()
{
puts("void vbroadcastf32x2(const Ymm& y, const Operand& op) { opAVX_X_XM_IMM(y, op, T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0, 0x19); }");
puts("void vbroadcastf32x4(const Ymm& y, const Address& addr) { opAVX_X_XM_IMM(y, addr, T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0, 0x1A); }");
puts("void vbroadcastf64x2(const Ymm& y, const Address& addr) { opAVX_X_XM_IMM(y, addr, T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1, 0x1A); }");
puts("void vbroadcastf64x4(const Zmm& y, const Address& addr) { opAVX_X_XM_IMM(y, addr, T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1, 0x1B); }");
puts("void vbroadcastf32x2(const Ymm& y, const Operand& op) { opAVX_X_XM_IMM(y, op, T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_N8, 0x19); }");
puts("void vbroadcastf32x4(const Ymm& y, const Address& addr) { opAVX_X_XM_IMM(y, addr, T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_N16, 0x1A); }");
puts("void vbroadcastf64x2(const Ymm& y, const Address& addr) { opAVX_X_XM_IMM(y, addr, T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1 | T_N16, 0x1A); }");
puts("void vbroadcastf64x4(const Zmm& y, const Address& addr) { opAVX_X_XM_IMM(y, addr, T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1 | T_N32, 0x1B); }");
{
const struct Tbl {
uint8 code;
@ -291,10 +291,10 @@ void putBroadcast()
int type;
int reg;
} tbl[] = {
{ 0x7A, "vpbroadcastb", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0, 8 },
{ 0x7B, "vpbroadcastw", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0, 16 },
{ 0x7C, "vpbroadcastd", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0, 32 },
{ 0x7C, "vpbroadcastq", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1, 64},
{ 0x7A, "vpbroadcastb", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_N4, 8 },
{ 0x7B, "vpbroadcastw", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_N4, 16 },
{ 0x7C, "vpbroadcastd", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_N4, 32 },
{ 0x7C, "vpbroadcastq", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1 | T_N8, 64},
};
for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) {
const Tbl& p = tbl[i];
@ -304,11 +304,11 @@ void putBroadcast()
if (p.reg == 64) puts("#endif");
}
}
puts("void vbroadcasti32x2(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0, 0x59); }");
puts("void vbroadcasti32x4(const Ymm& y, const Operand& op) { opAVX_X_XM_IMM(y, op, T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0, 0x5A); }");
puts("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, 0x5A); }");
puts("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, 0x5B); }");
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, 0x5B); }");
puts("void vbroadcasti32x2(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_N8, 0x59); }");
puts("void vbroadcasti32x4(const Ymm& y, const Operand& op) { opAVX_X_XM_IMM(y, op, T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_N16, 0x5A); }");
puts("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); }");
puts("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); }");
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); }");
}
int main()

14
gen/gen_code.cpp
View file

@ -1455,17 +1455,19 @@ void put()
const struct Tbl {
const char *name;
uint8 code;
int type;
bool ew1;
} tbl[] = {
{ "vbroadcastss", 0x18, false },
{ "vpbroadcastb", 0x78, false },
{ "vpbroadcastw", 0x79, false },
{ "vpbroadcastd", 0x58, false },
{ "vpbroadcastq", 0x59, true },
{ "vbroadcastss", 0x18, T_0F38 | T_66 | T_W0 | T_YMM | T_EVEX | T_N4 },
{ "vpbroadcastb", 0x78, T_0F38 | T_66 | T_W0 | T_YMM | T_EVEX | T_N4 },
{ "vpbroadcastw", 0x79, T_0F38 | T_66 | T_W0 | T_YMM | T_EVEX | T_N4 },
{ "vpbroadcastd", 0x58, T_0F38 | T_66 | T_W0 | T_YMM | T_EVEX | T_N4 },
{ "vpbroadcastq", 0x59, T_0F38 | T_66 | T_W0 | T_YMM | T_EVEX | T_EW1 | T_N8 },
};
for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) {
const Tbl& p = tbl[i];
printf("void %s(const Xmm& x, const Operand& op) { if (!(op.isXMM() || op.isMEM())) throw Error(ERR_BAD_COMBINATION); opAVX_X_XM_IMM(x, op, T_0F38 | T_66 | T_W0 | T_YMM | T_EVEX%s, 0x%02X); }\n", p.name, p.ew1 ? "|T_EW1" : "", p.code);
std::string type = type2String(p.type);
printf("void %s(const Xmm& x, const Operand& op) { if (!(op.isXMM() || op.isMEM())) throw Error(ERR_BAD_COMBINATION); opAVX_X_XM_IMM(x, op, %s, 0x%02X); }\n", p.name, type.c_str(), p.code);
}
printf("void vextractf128(const Operand& op, const Ymm& y, uint8 imm) { opAVX_X_X_XMcvt(y, false, cvtIdx0(y), op, op.isXMM(), Operand::YMM, T_0F3A | T_66 | T_W0 | T_YMM, 0x19, imm); }\n");

12
test/make_512.cpp
View file

@ -663,11 +663,11 @@ public:
putBroadcastSub(20, i);
#endif
}
put("vpbroadcastb", XMM_KZ | ZMM_KZ, REG8);
put("vpbroadcastw", XMM_KZ | ZMM_KZ, REG16);
put("vpbroadcastd", XMM_KZ | ZMM_KZ, REG32);
put("vpbroadcastb", XMM_KZ | ZMM_KZ, REG8 | _MEM);
put("vpbroadcastw", XMM_KZ | ZMM_KZ, REG16 | _MEM);
put("vpbroadcastd", XMM_KZ | ZMM_KZ, REG32 | _MEM);
#ifdef XBYAK64
put("vpbroadcastq", XMM_KZ | ZMM_KZ, REG64);
put("vpbroadcastq", XMM_KZ | ZMM_KZ, REG64 | _MEM);
#endif
{
const char *tbl[] = {
@ -1423,13 +1423,13 @@ public:
separateFunc();
putCombi();
separateFunc();
#endif
putCmpK();
#if 0
separateFunc();
putBroadcast();
separateFunc();
#endif
putAVX512_M_X();
#if 0
separateFunc();
put_vmov();
separateFunc();

9
xbyak/xbyak.h
View file

@ -1379,8 +1379,9 @@ private:
T_N4 = 1 << 25,
T_N8 = 1 << 26,
T_N16 = 1 << 27,
T_N_VL = 1 << 28, // N * (1, 2, 4) for VL
T_DUP = 1 << 29, // N = (8, 32, 64)
T_N32 = 1 << 28,
T_N_VL = 1 << 29, // N * (1, 2, 4) for VL
T_DUP = 1 << 30, // N = (8, 32, 64)
T_XXX
};
void vex(const Reg& reg, const Reg& base, const Operand *v, int type, int code, bool x = false)
@ -1447,8 +1448,8 @@ private:
LL = (VL == 512) ? 2 : (VL == 256) ? 1 : 0;
if (b) {
disp8N = (type & T_B32) ? 4 : 8;
} else if (type & (T_N2 | T_N4 | T_N8 | T_N16)) {
disp8N = (type & T_N2) ? 2 : (type & T_N4) ? 4 : (type & T_N8) ? 8 : 16;
} else if (type & (T_N2 | T_N4 | T_N8 | T_N16 | T_N32)) {
disp8N = (type & T_N2) ? 2 : (type & T_N4) ? 4 : (type & T_N8) ? 8 : (type & T_N16) ? 16 : 32;
if (type & T_N_VL) disp8N *= (VL == 512 ? 4 : VL == 256 ? 2 : 1);
} else if (type & T_DUP) {
disp8N = VL == 128 ? 8 : VL == 256 ? 32 : 64;

26
xbyak/xbyak_avx512.h
View file

@ -128,19 +128,19 @@ void vinserti32x4(const Ymm& r1, const Ymm& r2, const Operand& op, uint8 imm) {
void vinserti64x2(const Ymm& r1, const Ymm& r2, const Operand& op, uint8 imm) { opAVX_X_X_XMcvt(r1, false, r2, op, op.isXMM(), Operand::YMM, T_66 | T_0F3A | T_EW1 | T_YMM | T_MUST_EVEX, 0x38, imm); }
void vinserti32x8(const Zmm& r1, const Zmm& r2, const Operand& op, uint8 imm) { opAVX_X_X_XMcvt(r1, false, r2, op, op.isXMM(), Operand::YMM, T_66 | T_0F3A | T_EW0 | T_YMM | T_MUST_EVEX, 0x3A, imm); }
void vinserti64x4(const Zmm& r1, const Zmm& r2, const Operand& op, uint8 imm) { opAVX_X_X_XMcvt(r1, false, r2, op, op.isXMM(), Operand::YMM, T_66 | T_0F3A | T_EW1 | T_YMM | T_MUST_EVEX, 0x3A, imm); }
void vbroadcastf32x2(const Ymm& y, const Operand& op) { opAVX_X_XM_IMM(y, op, T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0, 0x19); }
void vbroadcastf32x4(const Ymm& y, const Address& addr) { opAVX_X_XM_IMM(y, addr, T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0, 0x1A); }
void vbroadcastf64x2(const Ymm& y, const Address& addr) { opAVX_X_XM_IMM(y, addr, T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1, 0x1A); }
void vbroadcastf64x4(const Zmm& y, const Address& addr) { opAVX_X_XM_IMM(y, addr, T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1, 0x1B); }
void vpbroadcastb(const Xmm& x, const Reg8& r) { opVex(x, &cvtIdx0(x), r, T_66 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX, 0x7A); }
void vpbroadcastw(const Xmm& x, const Reg16& r) { opVex(x, &cvtIdx0(x), r, T_66 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX, 0x7B); }
void vpbroadcastd(const Xmm& x, const Reg32& r) { opVex(x, &cvtIdx0(x), r, T_66 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX, 0x7C); }
void vbroadcastf32x2(const Ymm& y, const Operand& op) { opAVX_X_XM_IMM(y, op, T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_N8, 0x19); }
void vbroadcastf32x4(const Ymm& y, const Address& addr) { opAVX_X_XM_IMM(y, addr, T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_N16, 0x1A); }
void vbroadcastf64x2(const Ymm& y, const Address& addr) { opAVX_X_XM_IMM(y, addr, T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1 | T_N16, 0x1A); }
void vbroadcastf64x4(const Zmm& y, const Address& addr) { opAVX_X_XM_IMM(y, addr, T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1 | T_N32, 0x1B); }
void vpbroadcastb(const Xmm& x, const Reg8& r) { opVex(x, &cvtIdx0(x), r, T_66 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX | T_N4, 0x7A); }
void vpbroadcastw(const Xmm& x, const Reg16& r) { opVex(x, &cvtIdx0(x), r, T_66 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX | T_N4, 0x7B); }
void vpbroadcastd(const Xmm& x, const Reg32& r) { opVex(x, &cvtIdx0(x), r, T_66 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX | T_N4, 0x7C); }
#ifdef XBYAK64
void vpbroadcastq(const Xmm& x, const Reg64& r) { opVex(x, &cvtIdx0(x), r, T_66 | T_0F38 | T_EW1 | T_YMM | T_MUST_EVEX, 0x7C); }
void vpbroadcastq(const Xmm& x, const Reg64& r) { opVex(x, &cvtIdx0(x), r, T_66 | T_0F38 | T_EW1 | T_YMM | T_MUST_EVEX | T_N8, 0x7C); }
#endif
void vbroadcasti32x2(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0, 0x59); }
void vbroadcasti32x4(const Ymm& y, const Operand& op) { opAVX_X_XM_IMM(y, op, T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0, 0x5A); }
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, 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, 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, 0x5B); }
void vbroadcasti32x2(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_N8, 0x59); }
void vbroadcasti32x4(const Ymm& y, const Operand& op) { opAVX_X_XM_IMM(y, op, T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_N16, 0x5A); }
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); }
#endif

10
xbyak/xbyak_mnemonic.h
View file

@ -1403,11 +1403,11 @@ void vaesimc(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_0F38 | T
void vbroadcastf128(const Ymm& y, const Address& addr) { opAVX_X_XM_IMM(y, addr, T_0F38 | T_66 | T_W0 | T_YMM, 0x1A); }
void vbroadcasti128(const Ymm& y, const Address& addr) { opAVX_X_XM_IMM(y, addr, T_0F38 | T_66 | T_W0 | T_YMM, 0x5A); }
void vbroadcastsd(const Ymm& y, const Operand& op) { if (!op.isMEM() && !(y.isYMM() && op.isXMM()) && !(y.isZMM() && op.isXMM())) throw Error(ERR_BAD_COMBINATION); opAVX_X_XM_IMM(y, op, T_0F38 | T_66 | T_W0 | T_YMM | T_EVEX | T_EW1, 0x19); }
void vbroadcastss(const Xmm& x, const Operand& op) { if (!(op.isXMM() || op.isMEM())) throw Error(ERR_BAD_COMBINATION); opAVX_X_XM_IMM(x, op, T_0F38 | T_66 | T_W0 | T_YMM | T_EVEX, 0x18); }
void vpbroadcastb(const Xmm& x, const Operand& op) { if (!(op.isXMM() || op.isMEM())) throw Error(ERR_BAD_COMBINATION); opAVX_X_XM_IMM(x, op, T_0F38 | T_66 | T_W0 | T_YMM | T_EVEX, 0x78); }
void vpbroadcastw(const Xmm& x, const Operand& op) { if (!(op.isXMM() || op.isMEM())) throw Error(ERR_BAD_COMBINATION); opAVX_X_XM_IMM(x, op, T_0F38 | T_66 | T_W0 | T_YMM | T_EVEX, 0x79); }
void vpbroadcastd(const Xmm& x, const Operand& op) { if (!(op.isXMM() || op.isMEM())) throw Error(ERR_BAD_COMBINATION); opAVX_X_XM_IMM(x, op, T_0F38 | T_66 | T_W0 | T_YMM | T_EVEX, 0x58); }
void vpbroadcastq(const Xmm& x, const Operand& op) { if (!(op.isXMM() || op.isMEM())) throw Error(ERR_BAD_COMBINATION); opAVX_X_XM_IMM(x, op, T_0F38 | T_66 | T_W0 | T_YMM | T_EVEX|T_EW1, 0x59); }
void vbroadcastss(const Xmm& x, const Operand& op) { if (!(op.isXMM() || op.isMEM())) throw Error(ERR_BAD_COMBINATION); opAVX_X_XM_IMM(x, op, T_66 | T_0F38 | T_W0 | T_YMM | T_EVEX | T_N4, 0x18); }
void vpbroadcastb(const Xmm& x, const Operand& op) { if (!(op.isXMM() || op.isMEM())) throw Error(ERR_BAD_COMBINATION); opAVX_X_XM_IMM(x, op, T_66 | T_0F38 | T_W0 | T_YMM | T_EVEX | T_N4, 0x78); }
void vpbroadcastw(const Xmm& x, const Operand& op) { if (!(op.isXMM() || op.isMEM())) throw Error(ERR_BAD_COMBINATION); opAVX_X_XM_IMM(x, op, T_66 | T_0F38 | T_W0 | T_YMM | T_EVEX | T_N4, 0x79); }
void vpbroadcastd(const Xmm& x, const Operand& op) { if (!(op.isXMM() || op.isMEM())) throw Error(ERR_BAD_COMBINATION); opAVX_X_XM_IMM(x, op, T_66 | T_0F38 | T_W0 | T_YMM | T_EVEX | T_N4, 0x58); }
void vpbroadcastq(const Xmm& x, const Operand& op) { if (!(op.isXMM() || op.isMEM())) throw Error(ERR_BAD_COMBINATION); opAVX_X_XM_IMM(x, op, T_66 | T_0F38 | T_W0 | T_EW1 | T_YMM | T_EVEX | T_N8, 0x59); }
void vextractf128(const Operand& op, const Ymm& y, uint8 imm) { opAVX_X_X_XMcvt(y, false, cvtIdx0(y), op, op.isXMM(), Operand::YMM, T_0F3A | T_66 | T_W0 | T_YMM, 0x19, imm); }
void vextracti128(const Operand& op, const Ymm& y, uint8 imm) { opAVX_X_X_XMcvt(y, false, cvtIdx0(y), op, op.isXMM(), Operand::YMM, T_0F3A | T_66 | T_W0 | T_YMM, 0x39, imm); }
void vextractps(const Operand& op, const Xmm& x, uint8 imm) { if (!(op.isREG(32) || op.isMEM()) || x.isYMM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x, false, cvtIdx0(x), op, op.isREG(), Operand::XMM, T_0F3A | T_66 | T_W0 | T_EVEX, 0x17, imm); }