JitArm64: Implement HLE function hooking #9218
Conversation
| @@ -209,20 +209,15 @@ void JitArm64::FallBackToInterpreter(UGeckoInstruction inst) | |||
| } | |||
| } | |||
|
|
|||
| void JitArm64::HLEFunction(UGeckoInstruction inst) | |||
| void JitArm64::HLEFunction(u32 hook_index) | |||
| { | |||
| gpr.Flush(FlushMode::FLUSH_ALL); | |||
| fpr.Flush(FlushMode::FLUSH_ALL); | |||
There was a problem hiding this comment.
Is there a chance that the first instruction is an integer instruction? If so, "FlushCarry();" might need to be called as well.
| return false; | ||
|
|
||
| LDR(INDEX_UNSIGNED, DISPATCHER_PC, PPC_REG, PPCSTATE_OFF(npc)); | ||
| js.downcountAmount += js.st.numCycles; |
There was a problem hiding this comment.
Honestly, I do not get the point why we add the cycles of this single instruction. I assume te HLE functions shall decrease downCount on their own, so this single instruction likely has no effect. However it is the same in the x64 JIT, so I'm fine with it.
There was a problem hiding this comment.
js.st.numCycles is for the whole block, not the current instruction. Though, implementing it that way also seems weird, considering that we might hit the HLE case halfway through a block after we've already added some cycles to js.downcountAmount the normal way...
JosJuice
force-pushed
the
aarch64-hle-hooks
branch
from
66732da
to
aaf4e70
Compare
Closely based on the x86-64 implementation. Fixes https://bugs.dolphin-emu.org/issues/10965.
JosJuice
force-pushed
the
aarch64-hle-hooks
branch
from
aaf4e70
to
fe986b6
Compare
Closely based on the x86-64 implementation. Fixes https://bugs.dolphin-emu.org/issues/10965.