emit_x64_vector: Simplify EmitVectorLogicalLeftShift8()

Similar to EmitVectorLogicalRightShift8(), we can determine a mask ahead
of time and just and the results of a halfword left shift.
This commit is contained in:
Lioncash 2018-08-31 11:14:51 -04:00 committed by MerryMage
parent 135107279d
commit 2cac6ad129

View file

@ -1055,12 +1055,17 @@ void EmitX64::EmitVectorInterleaveUpper64(EmitContext& ctx, IR::Inst* inst) {
void EmitX64::EmitVectorLogicalShiftLeft8(EmitContext& ctx, IR::Inst* inst) { void EmitX64::EmitVectorLogicalShiftLeft8(EmitContext& ctx, IR::Inst* inst) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst); auto args = ctx.reg_alloc.GetArgumentInfo(inst);
Xbyak::Xmm result = ctx.reg_alloc.UseScratchXmm(args[0]); const Xbyak::Xmm result = ctx.reg_alloc.UseScratchXmm(args[0]);
const u8 shift_amount = args[1].GetImmediateU8(); const u8 shift_amount = args[1].GetImmediateU8();
// TODO: Optimize if (shift_amount == 1) {
for (size_t i = 0; i < shift_amount; ++i) {
code.paddb(result, result); code.paddb(result, result);
} else if (shift_amount > 0) {
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));
} }
ctx.reg_alloc.DefineValue(inst, result); ctx.reg_alloc.DefineValue(inst, result);