eden-emu/xbyak
19
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

add vpexpand{b,w}

Browse source
This commit is contained in:
MITSUNARI Shigeo 2018-01-05 14:28:17 +09:00
parent e450f96513
commit 5af0ba3977
3 changed files with 44 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

3
gen/gen_avx512.cpp
View file

@ -653,6 +653,9 @@ void putX_XM_IMM()
{ 0x54, "vpopcntw", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1 | T_SAE_Z, false }, { 0x54, "vpopcntw", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1 | T_SAE_Z, false },
{ 0x55, "vpopcntd", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_SAE_Z | T_B32, false }, { 0x55, "vpopcntd", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_SAE_Z | T_B32, false },
{ 0x55, "vpopcntq", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1 | T_SAE_Z | T_B64, false }, { 0x55, "vpopcntq", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1 | T_SAE_Z | T_B64, false },
{ 0x62, "vpexpandb", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_SAE_Z | T_N1, false },
{ 0x62, "vpexpandw", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1 | T_SAE_Z | T_N2, false },
}; };
for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) { for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) {
const Tbl *p = &tbl[i]; const Tbl *p = &tbl[i];

39
test/misc.cpp
View file

@ -522,4 +522,43 @@ CYBOZU_TEST_AUTO(vpdpbus)
CYBOZU_TEST_EQUAL(c.getSize(), n); CYBOZU_TEST_EQUAL(c.getSize(), n);
CYBOZU_TEST_EQUAL_ARRAY(c.getCode(), tbl, n); CYBOZU_TEST_EQUAL_ARRAY(c.getCode(), tbl, n);
} }
CYBOZU_TEST_AUTO(vexpand)
{
struct Code : Xbyak::CodeGenerator {
Code()
{
vpexpandb(xmm5|k3|T_z, xmm30);
vpexpandb(ymm5|k3|T_z, ymm30);
vpexpandb(zmm5|k3|T_z, zmm30);
vpexpandb(xmm5|k3|T_z, ptr [rax + 0x40]);
vpexpandb(ymm5|k3|T_z, ptr [rax + 0x40]);
vpexpandb(zmm5|k3|T_z, ptr [rax + 0x40]);
vpexpandw(xmm5|k3|T_z, xmm30);
vpexpandw(ymm5|k3|T_z, ymm30);
vpexpandw(zmm5|k3|T_z, zmm30);
vpexpandw(xmm5|k3|T_z, ptr [rax + 0x40]);
vpexpandw(ymm5|k3|T_z, ptr [rax + 0x40]);
vpexpandw(zmm5|k3|T_z, ptr [rax + 0x40]);
}
} c;
const uint8_t tbl[] = {
0x62, 0x92, 0x7d, 0x8b, 0x62, 0xee,
0x62, 0x92, 0x7d, 0xab, 0x62, 0xee,
0x62, 0x92, 0x7d, 0xcb, 0x62, 0xee,
0x62, 0xf2, 0x7d, 0x8b, 0x62, 0x68, 0x40,
0x62, 0xf2, 0x7d, 0xab, 0x62, 0x68, 0x40,
0x62, 0xf2, 0x7d, 0xcb, 0x62, 0x68, 0x40,
0x62, 0x92, 0xfd, 0x8b, 0x62, 0xee,
0x62, 0x92, 0xfd, 0xab, 0x62, 0xee,
0x62, 0x92, 0xfd, 0xcb, 0x62, 0xee,
0x62, 0xf2, 0xfd, 0x8b, 0x62, 0x68, 0x20,
0x62, 0xf2, 0xfd, 0xab, 0x62, 0x68, 0x20,
0x62, 0xf2, 0xfd, 0xcb, 0x62, 0x68, 0x20,
};
const size_t n = sizeof(tbl) / sizeof(tbl[0]);
CYBOZU_TEST_EQUAL(c.getSize(), n);
CYBOZU_TEST_EQUAL_ARRAY(c.getCode(), tbl, n);
}
#endif #endif

2
xbyak/xbyak_mnemonic.h
View file

@ -1805,8 +1805,10 @@ void vpermt2ps(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x
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 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 vpermt2w(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, 0x7D); } void vpermt2w(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, 0x7D); }
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 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 vpexpandb(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_N1 | T_66 | T_0F38 | T_EW0 | T_YMM | T_SAE_Z | T_MUST_EVEX, 0x62); }
void vpexpandd(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_N4 | T_66 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX, 0x89); } void vpexpandd(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_N4 | T_66 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX, 0x89); }
void vpexpandq(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_N8 | T_66 | T_0F38 | T_EW1 | T_YMM | T_MUST_EVEX, 0x89); } void vpexpandq(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_N8 | T_66 | T_0F38 | T_EW1 | T_YMM | T_MUST_EVEX, 0x89); }
void vpexpandw(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_N2 | T_66 | T_0F38 | T_EW1 | T_YMM | T_SAE_Z | T_MUST_EVEX, 0x62); }
void vpgatherdd(const Xmm& x, const Address& addr) { opGather2(x, addr, T_N4 | T_66 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX, 0x90, 0); } void vpgatherdd(const Xmm& x, const Address& addr) { opGather2(x, addr, T_N4 | T_66 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX, 0x90, 0); }
void vpgatherdq(const Xmm& x, const Address& addr) { opGather2(x, addr, T_N8 | T_66 | T_0F38 | T_EW1 | T_YMM | T_MUST_EVEX, 0x90, 1); } void vpgatherdq(const Xmm& x, const Address& addr) { opGather2(x, addr, T_N8 | T_66 | T_0F38 | T_EW1 | T_YMM | T_MUST_EVEX, 0x90, 1); }
void vpgatherqd(const Xmm& x, const Address& addr) { opGather2(x, addr, T_N4 | T_66 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX, 0x91, 2); } void vpgatherqd(const Xmm& x, const Address& addr) { opGather2(x, addr, T_N4 | T_66 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX, 0x91, 2); }