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

Improve perf of vcmp cr6 update.

This commit is contained in:
Ben Vanik 2013-10-04 13:32:11 -07:00
parent e2842ea4b4
commit d3bf7813ea
1 changed files with 14 additions and 25 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

39
src/xenia/cpu/x64/x64_emit_altivec.cc
View File

@ -442,38 +442,27 @@ XEEMITTER(vcmpbfp128, VX128(6, 384), VX128_R)(X64Emitter& e, X86Compiler&
void InstrEmit_vcmp_cr6_(X64Emitter& e, X86Compiler& c, XmmVar& v) { void InstrEmit_vcmp_cr6_(X64Emitter& e, X86Compiler& c, XmmVar& v) {
// Testing for all 1's and all 0's. // Testing for all 1's and all 0's.
// if (Rc) CR6 = all_equal | 0 | none_equal | 0 // if (Rc) CR6 = all_equal | 0 | none_equal | 0
// Since none_equal and all_equal are mutually exclusive we optimize
// a bit here. This is still terrible.
GpVar lo(c.newGpVar());
GpVar hi(c.newGpVar());
c.pextrq(hi.m64(), v, imm(1));
c.movq(lo.m64(), v);
GpVar gt(c.newGpVar()); GpVar gt(c.newGpVar());
GpVar cr(c.newGpVar()); GpVar cr(c.newGpVar());
c.xor_(cr, cr); c.xor_(cr, cr);
Label skip(c.newLabel());
// cmp with 0xFF... and set all_equal // We do this fast by extracting the high bits (as all bits are the same)
c.mov(gt, lo); // and testing those.
c.and_(gt, hi); GpVar bmask(c.newGpVar());
c.test(gt, imm(0)); c.pmovmskb(bmask, v);
// !eq = all_equal
// all_equal= 0b1000
c.mov(gt, imm(0x8)); // 0b1000
c.cmovne(cr, gt);
c.jne(skip);
// cmp with 0 and set none_equal // zero = none_equal
c.mov(gt, lo); c.test(bmask, bmask);
c.or_(gt, hi); c.mov(gt, imm(0x2)); // none_equal=0b0010
c.test(gt, imm(0)); c.cmovz(cr, gt);
// eq = none_equal
// none_equal= 0b0010 // !zero = all_equal
c.mov(gt, imm(0x2)); // 0b0010 c.not_(bmask);
c.cmove(cr, gt); c.test(bmask, bmask);
c.mov(gt, imm(0x8)); // all_equal=0b1000
c.cmovz(cr, gt);
c.bind(skip);
e.update_cr_value(6, cr); e.update_cr_value(6, cr);
} }