JIT: avoid saving the PC on every store #807
Conversation
|
@JMC47 can we get some estimates of impact on some games, like VC titles which are usually CPU bound? |
|
Also, LGTM, nice change. Almost 0 overhead for unpredicted RAM writes now. |
|
NICE BENCHMARK STATS |
| @@ -220,6 +222,14 @@ const u8 *Jitx86Base::BackPatch(u8 *codePtr, u32 emAddress, void *ctx_void) | |||
| else | |||
| { | |||
| // TODO: special case FIFO writes. Also, support 32-bit mode. | |||
| it = pcAtLoc.find(codePtr); | |||
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
|
It would be slightly faster to use one map containing both PC and registersInUse, rather than a map for each. |
|
I was thinking of that, but wasn't sure it was the best idea, because PCs are only needed for stores, not loads, I think. |
|
Looking through the code, I found only one place were the program counter is actually read during a write. It's a single debug message inside AudioInterface.when you write to the AI_INTERRUPT_TIMING io address. |
|
Isn't it read in the case of a DSI exception on write? |
|
Oh right. Someone did a |
|
Increases F-Zero X from 182fps to 184fps on my computer. Most other VC N64 games that I have aren't as fast as this one, and end up seeing a 1 fps increase or no noticeable increase. |
| @@ -110,6 +110,8 @@ const u8 *TrampolineCache::GetWriteTrampoline(const InstructionInfo &info, u32 r | |||
| // Don't treat FIFO writes specially for now because they require a burst | |||
| // check anyway. | |||
|
|
|||
| MOV(32, M(&PC), Imm32(pc)); // for exception handling | |||
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
|
Fix up those comments and it LGTM. |
Modify the backpatcher to store the PC in the trampolines. Should be ~3.5% faster overall (measured on POV-RAY benchmark).
|
@FioraAeterna: This comment grants you the permission to merge this pull request whenever you think it is ready. After addressing the remaining comments, click this link to merge. @dolphin-emu-bot allowmerge |
JIT: avoid saving the PC on every store
Modify the backpatcher to store the PC in the trampolines.
Should be ~3.5% faster overall (measured on POV-RAY benchmark).