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

[JIT] Kill a couple instructions in VECTOR_CONVERT_F2I

This commit is contained in:
DrChat 2018-03-02 11:41:04 -06:00
parent 04d20e33a5
commit 131fcd1320
1 changed files with 3 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
src/xenia/cpu/backend/x64/x64_sequences.cc
View File

@ -1610,12 +1610,10 @@ struct VECTOR_CONVERT_F2I
// scale any values >= (unsigned)INT_MIN back to [0, ...] // scale any values >= (unsigned)INT_MIN back to [0, ...]
e.vsubps(e.xmm2, e.xmm0, e.GetXmmConstPtr(XMMPosIntMinPS)); e.vsubps(e.xmm2, e.xmm0, e.GetXmmConstPtr(XMMPosIntMinPS));
e.vandps(e.xmm2, e.xmm1, e.xmm2); // 0 if < (unsigned)INT_MIN e.vblendvps(e.xmm0, e.xmm0, e.xmm2, e.xmm1);
e.vandnps(e.xmm0, e.xmm1, e.xmm0); // 0 if >= (unsigned)INT_MIN
// xmm0 = [0, INT_MAX] // xmm0 = [0, INT_MAX]
// this may still contain values > INT_MAX (if src has vals > UINT_MAX) // this may still contain values > INT_MAX (if src has vals > UINT_MAX)
e.vorps(e.xmm0, e.xmm0, e.xmm2);
e.vcvttps2dq(i.dest, e.xmm0); e.vcvttps2dq(i.dest, e.xmm0);
// xmm0 = mask of values that need saturation // xmm0 = mask of values that need saturation
@ -3816,12 +3814,12 @@ struct VECTOR_ADD
// Set any negative overflowed elements of src1 to INT_MIN // Set any negative overflowed elements of src1 to INT_MIN
e.vpand(e.xmm2, src1, e.xmm1); e.vpand(e.xmm2, src1, e.xmm1);
e.vblendvps(dest, dest, e.GetXmmConstPtr(XMMSignMaskI32), e.vblendvps(dest, dest, e.GetXmmConstPtr(XMMSignMaskI32),
e.xmm2); e.xmm2);
// Set any positive overflowed elements of src1 to INT_MAX // Set any positive overflowed elements of src1 to INT_MAX
e.vpandn(e.xmm2, src1, e.xmm1); e.vpandn(e.xmm2, src1, e.xmm1);
e.vblendvps(dest, dest, e.GetXmmConstPtr(XMMAbsMaskPS), e.vblendvps(dest, dest, e.GetXmmConstPtr(XMMAbsMaskPS),
e.xmm2); e.xmm2);
} }
} else { } else {
e.vpaddd(dest, src1, src2); e.vpaddd(dest, src1, src2);