emit_x64_vector: Simplify "position == 0" case for EmitVectorExtractLower()

In the event position == 0, we can just treat it as a simple movq,
clearing the upper half of the XMM register. This also makes that case
use only one register.
This commit is contained in:
Lioncash 2018-09-08 01:51:50 -04:00 committed by MerryMage
parent f5fb496e7e
commit 87372917f9

View file

@ -780,13 +780,16 @@ void EmitX64::EmitVectorExtractLower(EmitContext& ctx, IR::Inst* inst) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst); auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const Xbyak::Xmm xmm_a = ctx.reg_alloc.UseScratchXmm(args[0]); const Xbyak::Xmm xmm_a = ctx.reg_alloc.UseScratchXmm(args[0]);
const Xbyak::Xmm xmm_b = ctx.reg_alloc.UseXmm(args[1]);
const u8 position = args[2].GetImmediateU8(); const u8 position = args[2].GetImmediateU8();
ASSERT(position % 8 == 0); ASSERT(position % 8 == 0);
if (position != 0) {
const Xbyak::Xmm xmm_b = ctx.reg_alloc.UseXmm(args[1]);
code.punpcklqdq(xmm_a, xmm_b); code.punpcklqdq(xmm_a, xmm_b);
code.psrldq(xmm_a, position / 8); code.psrldq(xmm_a, position / 8);
}
code.movq(xmm_a, xmm_a); code.movq(xmm_a, xmm_a);
ctx.reg_alloc.DefineValue(inst, xmm_a); ctx.reg_alloc.DefineValue(inst, xmm_a);