fabi/Xenia-Canary
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[JIT] Full vctuxs support

Browse Source
This commit is contained in:
DrChat
2018-03-01 23:49:44 -06:00
parent 4766a5ec24
commit 0577b6c9be
5 changed files with 113 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
src/xenia/cpu/ppc/ppc_emit_altivec.cc
View File

@@ -554,7 +554,7 @@ int InstrEmit_vctsxs_(PPCHIRBuilder& f, uint32_t vd, uint32_t vb,
float fuimm = static_cast<float>(std::exp2(uimm));
Value* v =
f.Mul(f.LoadVR(vb), f.Splat(f.LoadConstantFloat32(fuimm), VEC128_TYPE));
v = f.VectorConvertF2I(v, ARITHMETIC_SATURATE);
v = f.VectorConvertF2I(v);
f.StoreSAT(f.DidSaturate(v));
f.StoreVR(vd, v);
return 0;
@@ -572,7 +572,7 @@ int InstrEmit_vctuxs_(PPCHIRBuilder& f, uint32_t vd, uint32_t vb,
float fuimm = static_cast<float>(std::exp2(uimm));
Value* v =
f.Mul(f.LoadVR(vb), f.Splat(f.LoadConstantFloat32(fuimm), VEC128_TYPE));
v = f.VectorConvertF2I(v, ARITHMETIC_UNSIGNED | ARITHMETIC_SATURATE);
v = f.VectorConvertF2I(v, ARITHMETIC_UNSIGNED);
f.StoreSAT(f.DidSaturate(v));
f.StoreVR(vd, v);
return 0;

79
src/xenia/cpu/ppc/testing/instr_vctuxs.s Normal file
View File

@@ -0,0 +1,79 @@
# 0 * 2^31
test_vctuxs_1:
#_ REGISTER_IN v0 [00000000, 00000000, 00000000, 00000000]
vctuxs v3, v0, 31
blr
#_ REGISTER_OUT v0 [00000000, 00000000, 00000000, 00000000]
#_ REGISTER_OUT v3 [00000000, 00000000, 00000000, 00000000]
# -0 ^ 2^31
test_vctuxs_2:
#_ REGISTER_IN v0 [80000000, 80000000, 80000000, 80000000]
vctuxs v3, v0, 31
blr
#_ REGISTER_OUT v0 [80000000, 80000000, 80000000, 80000000]
#_ REGISTER_OUT v3 [00000000, 00000000, 00000000, 00000000]
# smallest positive subnormal * 2^31
test_vctuxs_3:
#_ REGISTER_IN v0 [00000001, 00000001, 00000001, 00000001]
vctuxs v3, v0, 31
blr
#_ REGISTER_OUT v0 [00000001, 00000001, 00000001, 00000001]
#_ REGISTER_OUT v3 [00000000, 00000000, 00000000, 00000000]
# largest subnormal * 2^31
test_vctuxs_4:
#_ REGISTER_IN v0 [007FFFFF, 007FFFFF, 007FFFFF, 007FFFFF]
vctuxs v3, v0, 31
blr
#_ REGISTER_OUT v0 [007FFFFF, 007FFFFF, 007FFFFF, 007FFFFF]
#_ REGISTER_OUT v3 [00000000, 00000000, 00000000, 00000000]
# +1 * 2^0
test_vctuxs_5:
#_ REGISTER_IN v0 [3F800000, 3F800000, 3F800000, 3F800000]
vctuxs v3, v0, 0
blr
#_ REGISTER_OUT v0 [3F800000, 3F800000, 3F800000, 3F800000]
#_ REGISTER_OUT v3 [00000001, 00000001, 00000001, 00000001]
# -1 * 2^0
test_vctuxs_6:
#_ REGISTER_IN v0 [BF800000, BF800000, BF800000, BF800000]
vctuxs v3, v0, 0
blr
#_ REGISTER_OUT v0 [BF800000, BF800000, BF800000, BF800000]
#_ REGISTER_OUT v3 [00000000, 00000000, 00000000, 00000000]
# 2^31 * 2^0
test_vctuxs_7:
#_ REGISTER_IN v0 [4F000000, 4F000000, 4F000000, 4F000000]
vctuxs v3, v0, 0
blr
#_ REGISTER_OUT v0 [4F000000, 4F000000, 4F000000, 4F000000]
#_ REGISTER_OUT v3 [80000000, 80000000, 80000000, 80000000]
# 2^32 * 2^0
test_vctuxs_8:
#_ REGISTER_IN v0 [4F800000, 4F800000, 4F800000, 4F800000]
vctuxs v3, v0, 0
blr
#_ REGISTER_OUT v0 [4F800000, 4F800000, 4F800000, 4F800000]
#_ REGISTER_OUT v3 [FFFFFFFF, FFFFFFFF, FFFFFFFF, FFFFFFFF]
# +infinity * 2^0
test_vctuxs_9:
#_ REGISTER_IN v0 [7F800000, 7F800000, 7F800000, 7F800000]
vctuxs v3, v0, 0
blr
#_ REGISTER_OUT v0 [7F800000, 7F800000, 7F800000, 7F800000]
#_ REGISTER_OUT v3 [FFFFFFFF, FFFFFFFF, FFFFFFFF, FFFFFFFF]
# -infinity * 2^0
test_vctuxs_10:
#_ REGISTER_IN v0 [FF800000, FF800000, FF800000, FF800000]
vctuxs v3, v0, 0
blr
#_ REGISTER_OUT v0 [FF800000, FF800000, FF800000, FF800000]
#_ REGISTER_OUT v3 [00000000, 00000000, 00000000, 00000000]