ShuriZma
/
xenia
mirror of https://github.com/xenia-project/xenia.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

SHL_V128.

This commit is contained in:
gibbed 2015-06-09 19:55:54 -05:00
parent f0c47e036b
commit 6d933c18f7
1 changed files with 28 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

29
src/xenia/cpu/backend/x64/x64_sequences.cc
View File

@ -4493,12 +4493,39 @@ EMITTER(SHL_I64, MATCH(I<OPCODE_SHL, I64<>, I64<>, I8<>>)) {
EmitShlXX<SHL_I64, Reg64>(e, i);
}
};
EMITTER(SHL_V128, MATCH(I<OPCODE_SHL, V128<>, V128<>, I8<>>)) {
static void Emit(X64Emitter& e, const EmitArgType& i) {
// TODO(benvanik): native version (with shift magic).
if (i.src2.is_constant) {
e.mov(e.r9, i.src2.constant());
} else {
e.mov(e.r9, i.src2);
}
e.lea(e.r8, e.StashXmm(0, i.src1));
e.CallNativeSafe(reinterpret_cast<void*>(EmulateShlV128));
e.vmovaps(i.dest, e.xmm0);
}
static __m128i EmulateShlV128(void*, __m128i src1, uint8_t src2) {
// Almost all instances are shamt = 1, but non-constant.
// shamt is [0,7]
uint8_t shamt = src2 & 0x7;
alignas(16) vec128_t value;
_mm_store_si128(reinterpret_cast<__m128i*>(&value), src1);
for (int i = 0; i < 15; ++i) {
value.u8[i ^ 0x3] = (value.u8[i ^ 0x3] << shamt) |
(value.u8[(i + 1) ^ 0x3] >> (8 - shamt));
}
value.u8[15 ^ 0x3] = value.u8[15 ^ 0x3] << shamt;
return _mm_load_si128(reinterpret_cast<__m128i*>(&value));
}
};
EMITTER_OPCODE_TABLE(
OPCODE_SHL,
SHL_I8,
SHL_I16,
SHL_I32,
SHL_I64);
SHL_I64,
SHL_V128);
// ============================================================================