mirrors/dynarmic
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

emit_x64_vector: Bugfix for EmitVectorLogicalShiftLeft8: shift_amount can be >= 8

Browse source
This commit is contained in:
Merry 2021-03-27 21:31:51 +00:00
parent dc37fe6e28
commit 91337788ee
1 changed files with 13 additions and 12 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

25
src/backend/x64/emit_x64_vector.cpp
View file

@ -1465,20 +1465,21 @@ void EmitX64::EmitVectorLogicalShiftLeft8(EmitContext& ctx, IR::Inst* inst) {
const Xbyak::Xmm result = ctx.reg_alloc.UseScratchXmm(args[0]);
const u8 shift_amount = args[1].GetImmediateU8();
if (shift_amount == 1) {
if (shift_amount == 0) {
// do nothing
} else if (shift_amount >= 8) {
code.pxor(result, result);
} else if (shift_amount == 1) {
code.paddb(result, result);
} else if (shift_amount > 0) {
if (code.HasAVX512_Icelake()) {
// Galois 8x8 identity matrix, bit-shifted by the shift-amount
const u64 shift_matrix = 0x0102040810204080 >> (shift_amount * 8);
code.vgf2p8affineqb(result, result, code.MConst(xword_b, shift_matrix), 0);
} else {
const u64 replicand = (0xFFULL << shift_amount) & 0xFF;
const u64 mask = Common::Replicate(replicand, Common::BitSize<u8>());
} else if (code.HasAVX512_Icelake()) {
const u64 shift_matrix = 0x0102040810204080 >> (shift_amount * 8);
code.vgf2p8affineqb(result, result, code.MConst(xword_b, shift_matrix), 0);
} else {
const u64 replicand = (0xFFULL << shift_amount) & 0xFF;
const u64 mask = Common::Replicate(replicand, Common::BitSize<u8>());
code.psllw(result, shift_amount);
code.pand(result, code.MConst(xword, mask, mask));
}
code.psllw(result, shift_amount);
code.pand(result, code.MConst(xword, mask, mask));
}
ctx.reg_alloc.DefineValue(inst, result);