Skip to content
This repository has been archived by the owner on Jul 10, 2023. It is now read-only.

sh4: Fix FTRC (ipr, canonical, x86) by using correct positive cutoff value #1428

Merged
merged 1 commit into from Oct 2, 2018
Merged

sh4: Fix FTRC (ipr, canonical, x86) by using correct positive cutoff value #1428

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Jump to
Jump to file
Failed to load files.
Diff view
Diff view
9 changes: 5 additions & 4 deletions core/hw/sh4/dyna/shil_canonical.h
View file
Open in desktop
Expand Up @@ -677,9 +677,10 @@ shil_opc(cvt_f2i_t)
shil_canonical
(
u32,f1,(f32 f1),
if (f1 > 0x7FFFFFBF)
f1 = 0x7FFFFFBF;
return (s32)f1;
if (f1 > 2147483520.0f) // IEEE 754: 0x4effffff
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We should propably use a const for this instead of some magic number...

return 0x7fffffff;
else
return (s32)f1;
)

shil_compile
Expand Down Expand Up @@ -1044,4 +1045,4 @@ SHIL_END
#undef UN_OP_I
#undef UN_OP_F
#undef BIN_OP_FU
#undef shil_recimp
#undef shil_recimp
15 changes: 9 additions & 6 deletions core/hw/sh4/interpr/sh4_fpu.cpp
View file
Open in desktop
Expand Up @@ -628,22 +628,25 @@ sh4op(i1111_nnnn_0011_1101)
if (fpscr.PR == 0)
{
u32 n = GetN(op);
fpul = (u32)(s32)min(fr[n],(float)0x7FFFFFBF);
fpul = (u32)(s32)min(fr[n], 2147483520.0f); // IEEE 754: 0x4effffff
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

...that way it would be much easier...


if (fpul==0x80000000) //this is actually a x86-specific fix. I think ARM saturates
// Intel CPUs convert out of range float numbers to 0x80000000. Manually set the correct sign
if (fpul == 0x80000000)
{
if (fr[n]>0)
if (*(int *)&fr[n] > 0) // Using integer math to avoid issues with Inf and NaN
fpul--;
}
}
else
{
u32 n = (op >> 9) & 0x07;
fpul = (u32)(s32)GetDR(n);
f64 f = GetDR(n);
fpul = (u32)(s32)f;

if (fpul==0x80000000) //this is actually a x86-specific fix. I think ARM saturates
// Intel CPUs convert out of range float numbers to 0x80000000. Manually set the correct sign
if (fpul == 0x80000000)
{
if (GetDR(n)>0)
if (*(s64 *)&f > 0) // Using integer math to avoid issues with Inf and NaN
fpul--;
}
}
Expand Down
4 changes: 2 additions & 2 deletions core/rec-x86/rec_x86_il.cpp
View file
Open in desktop
Expand Up @@ -1453,7 +1453,7 @@ void ngen_opcode(RuntimeBlockInfo* block, shil_opcode* op,x86_block* x86e, bool
verify(op->rs1.is_r32f());
verify(reg.IsAllocg(op->rd));
verify(reg.IsAllocf(op->rs1));
static f32 sse_ftrc_saturate = 0x7FFFFFBF;//1.11111111111111111111111 << 31
static f32 sse_ftrc_saturate = 2147483520.0f; // IEEE 754: 0x4effffff
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

...to see that all these are the same.

x86e->Emit(op_movaps, XMM0, reg.mapf(op->rs1));
x86e->Emit(op_minss, XMM0, &sse_ftrc_saturate);
x86e->Emit(op_cvttss2si, reg.mapg(op->rd), XMM0);
Expand Down Expand Up @@ -1536,4 +1536,4 @@ void ngen_opcode(RuntimeBlockInfo* block, shil_opcode* op,x86_block* x86e, bool
}
}

#endif
#endif