Files
FEX-Emu--FEX/FEXCore/Source/Interface/Core/CPUBackend.h
T
Ryan Houdek dcfbc2f20e FEXCore/JIT: Encode the JITRIPReconstructionEntries using variable length integer
When #2722 implemented this initially and #4271 switched over to signed
int16_t there was assumptions made that int16_t was a reasonable
trade-off in encoding size versus needing to deal with 8-bit values
being too small in some cases.

In the common case we are almost always encoding 8-bit values because
instructions are typically linear (and less than 15-bytes in size), but
16-bit was chosen because optimizing JIT and multiple instructions that
don't cause exceptions can add up to larger than 8-bit.

Instead of hardcoding 16-bit values, implement a variable length integer
class where ~96.8% of values are 8-bit encoded, and the remaining 3.19% are encoded using 16-bit.
Due to some constraints that #4271 put in place, we can basically
guarantee currently that branch targets are within 16-bit. The VL class
does support 32-bit and 64-bit as well so if we change behaviour then
nothing needs to change.

Some stats when running Sonic Mania with multiblock enabled.
Encoded integers: 3,504,907
Encoded 8-bit:    3,393,095 (96.8%)
Encoded 16-bit:     111,812 (3.19%)
Encoded 32/64-bit:        0

Encoded Size:       3,615,181 bytes (3.44MiB)
Fixed encoded size: 7,007,604 bytes (6.68MiB)

Definitely worth using and saves the headache of large RIP/PC offsets
causing problems.
2025-02-03 11:54:52 -08:00

176 lines
5.4 KiB
C++

// SPDX-License-Identifier: MIT
/*
$info$
category: backend ~ IR to host code generation
tags: backend|shared
$end_info$
*/
#pragma once
#include <FEXCore/Utils/CompilerDefs.h>
#include <FEXCore/fextl/string.h>
#include <FEXCore/fextl/vector.h>
#include <cstdint>
#include <memory>
namespace FEXCore {
namespace IR {
class IRListView;
class RegisterAllocationData;
} // namespace IR
namespace Core {
struct DebugData;
struct ThreadState;
struct CpuStateFrame;
struct InternalThreadState;
} // namespace Core
namespace CodeSerialize {
struct CodeObjectFileSection;
}
namespace CPU {
class CPUBackend {
public:
struct CodeBuffer {
uint8_t* Ptr;
size_t Size;
};
/**
* @param InitialCodeSize - Initial size for the code buffers
* @param MaxCodeSize - Max size for the code buffers
*/
CPUBackend(FEXCore::Core::InternalThreadState* ThreadState, size_t InitialCodeSize, size_t MaxCodeSize);
virtual ~CPUBackend();
struct CompiledCode {
// Where this code block begins.
uint8_t* BlockBegin;
/**
* The function entrypoint to this codeblock.
*
* This may or may not equal `BlockBegin` above. Depending on the CPU backend, it may stick data
* prior to the BlockEntry.
*
* Is actually a function pointer of type `void (FEXCore::Core::ThreadState *Thread)`
*/
uint8_t* BlockEntry;
// The total size of the codeblock from [BlockBegin, BlockBegin+Size).
size_t Size;
};
// Header that can live at the start of a JIT block.
// We want the header to be quite small, with most data living in the tail object.
struct JITCodeHeader {
// Offset from the start of this header to where the tail lives.
// Only 32-bit since the tail block won't ever be more than 4GB away.
uint32_t OffsetToBlockTail;
};
// Header that can live at the end of the JIT block.
// For any state reconstruction or other data, this is where it should live.
// Any data that is explicitly tied to the JIT code and needs to be cached with it
// should end up in this data structure.
struct JITCodeTail {
// The total size of the codeblock from [BlockBegin, BlockBegin+Size).
size_t Size;
// RIP that the block's entry comes from.
uint64_t RIP;
// The length of the guest code for this block.
size_t GuestSize;
// If this block represents a single guest instruction.
bool SingleInst;
// Number of RIP entries for this JIT Code section.
uint32_t NumberOfRIPEntries;
// Offset after this block to the start of the RIP entries.
uint32_t OffsetToRIPEntries;
// Shared-code modification spin-loop futex.
uint32_t SpinLockFutex;
uint32_t _Pad;
};
/**
* @brief Tells this CPUBackend to compile code for the provided IR and DebugData
*
* The returned pointer needs to be long lived and be executable in the host environment
* FEXCore's frontend will store this pointer in to a cache for the current RIP when this was executed
*
* This is a thread specific compilation unit since there is one CPUBackend per guest thread
*
* @param Size - The byte size of the guest code for this block
* @param SingleInst - If this block represents a single guest instruction
* @param IR - IR that maps to the IR for this RIP
* @param DebugData - Debug data that is available for this IR indirectly
* @param CheckTF - If EFLAGS.TF checks should be emitted at the start of the block
*
* @return Information about the compiled code block.
*/
[[nodiscard]]
virtual CompiledCode CompileCode(uint64_t Entry, uint64_t Size, bool SingleInst, const FEXCore::IR::IRListView* IR,
FEXCore::Core::DebugData* DebugData, const FEXCore::IR::RegisterAllocationData* RAData, bool CheckTF) = 0;
/**
* @brief Relocates a block of code from the JIT code object cache
*
* @param Entry - RIP of the entry
* @param SerializationData - Serialization data referring to the object cache for `Entry`
*
* @return An executable function pointer relocated from the cache object
*/
[[nodiscard]]
virtual void* RelocateJITObjectCode(uint64_t Entry, const CodeSerialize::CodeObjectFileSection* SerializationData) {
return nullptr;
}
virtual void ClearCache() {}
/**
* @brief Clear any relocations after JIT compiling
*/
virtual void ClearRelocations() {}
bool IsAddressInCodeBuffer(uintptr_t Address) const;
protected:
// Max spill slot size in bytes. We need at most 32 bytes
// to be able to handle a 256-bit vector store to a slot.
constexpr static uint32_t MaxSpillSlotSize = 32;
FEXCore::Core::InternalThreadState* ThreadState;
size_t InitialCodeSize, MaxCodeSize;
[[nodiscard]]
CodeBuffer* GetEmptyCodeBuffer();
// This is the current code buffer that we are tracking
CodeBuffer* CurrentCodeBuffer {};
private:
CodeBuffer AllocateNewCodeBuffer(size_t Size);
void FreeCodeBuffer(CodeBuffer Buffer);
void EmplaceNewCodeBuffer(CodeBuffer Buffer) {
CurrentCodeBuffer = &CodeBuffers.emplace_back(Buffer);
}
// This is the array of code buffers. Unless signals force us to keep more than
// buffer, there will be only one entry here
fextl::vector<CodeBuffer> CodeBuffers {};
};
} // namespace CPU
} // namespace FEXCore