IRInterpreter.cpp

#include <algorithm>
#include <cmath>

#include "ppsspp_config.h"
#include "Common/BitSet.h"
#include "Common/BitScan.h"
#include "Common/Common.h"
#include "Common/Data/Convert/SmallDataConvert.h"
#include "Common/Math/math_util.h"

#ifdef _M_SSE
#include <emmintrin.h>
#endif

#if PPSSPP_ARCH(ARM_NEON)
#if defined(_MSC_VER) && PPSSPP_ARCH(ARM64)
#include <arm64_neon.h>
#else
#include <arm_neon.h>
#endif
#endif

#include "Core/Core.h"
#include "Core/CoreTiming.h"
#include "Core/Debugger/Breakpoints.h"
#include "Core/HLE/HLE.h"
#include "Core/HLE/ReplaceTables.h"
#include "Core/MemMap.h"
#include "Core/MIPS/MIPS.h"
#include "Core/MIPS/MIPSTables.h"
#include "Core/MIPS/MIPSVFPUUtils.h"
#include "Core/MIPS/IR/IRInst.h"
#include "Core/MIPS/IR/IRInterpreter.h"
#include "Core/System.h"
#include "Core/MIPS/MIPSTracer.h"

#ifdef mips
// Why do MIPS compilers define something so generic?  Try to keep defined, at least...
#undef mips
#define mips mips
#endif

alignas(16) static const float vec4InitValues[8][4] = {
	{ 0.0f, 0.0f, 0.0f, 0.0f },
	{ 1.0f, 1.0f, 1.0f, 1.0f },
	{ -1.0f, -1.0f, -1.0f, -1.0f },
	{ 1.0f, 0.0f, 0.0f, 0.0f },
	{ 0.0f, 1.0f, 0.0f, 0.0f },
	{ 0.0f, 0.0f, 1.0f, 0.0f },
	{ 0.0f, 0.0f, 0.0f, 1.0f },
};

alignas(16) static const uint32_t signBits[4] = {
	0x80000000, 0x80000000, 0x80000000, 0x80000000,
};

alignas(16) static const uint32_t noSignMask[4] = {
	0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF,
};

alignas(16) static const uint32_t lowBytesMask[4] = {
	0x000000FF, 0x000000FF, 0x000000FF, 0x000000FF,
};

u32 IRRunBreakpoint(u32 pc) {
	// Should we skip this breakpoint?
	uint32_t skipFirst = CBreakPoints::CheckSkipFirst();
	if (skipFirst == pc || skipFirst == currentMIPS->pc)
		return 0;

	// Did we already hit one?
	if (coreState != CORE_RUNNING && coreState != CORE_NEXTFRAME)
		return 1;

	CBreakPoints::ExecBreakPoint(pc);
	return coreState != CORE_RUNNING ? 1 : 0;
}

u32 IRRunMemCheck(u32 pc, u32 addr) {
	// Should we skip this breakpoint?
	uint32_t skipFirst = CBreakPoints::CheckSkipFirst();
	if (skipFirst == pc || skipFirst == currentMIPS->pc)
		return 0;

	// Did we already hit one?
	if (coreState != CORE_RUNNING && coreState != CORE_NEXTFRAME)
		return 1;

	CBreakPoints::ExecOpMemCheck(addr, pc);
	return coreState != CORE_RUNNING ? 1 : 0;
}

void IRApplyRounding(MIPSState *mips) {
	u32 fcr1Bits = mips->fcr31 & 0x01000003;
	// If these are 0, we just leave things as they are.
	if (fcr1Bits) {
		int rmode = fcr1Bits & 3;
		bool ftz = (fcr1Bits & 0x01000000) != 0;
#if PPSSPP_ARCH(SSE2)
		u32 csr = _mm_getcsr() & ~0x6000;
		// Translate the rounding mode bits to X86, the same way as in Asm.cpp.
		if (rmode & 1) {
			rmode ^= 2;
		}
		csr |= rmode << 13;

		if (ftz) {
			// Flush to zero
			csr |= 0x8000;
		}
		_mm_setcsr(csr);
#elif PPSSPP_ARCH(ARM64) && !PPSSPP_PLATFORM(WINDOWS)
		// On ARM64 we need to use inline assembly for a portable solution.
		// Unfortunately we don't have this possibility on Windows with MSVC, so ifdeffed out above.
		// Note that in the JIT, for fcvts, we use specific conversions. We could use the FCVTS variants
		// directly through inline assembly.
		u64 fpcr;  // not really 64-bit, just to match the register size.
		asm volatile ("mrs %0, fpcr" : "=r" (fpcr));

		// Translate MIPS to ARM rounding mode
		static const u8 lookup[4] = {0, 3, 1, 2};

		fpcr &= ~(3 << 22);    // Clear bits [23:22]
		fpcr |= (lookup[rmode] << 22);

		if (ftz) {
			fpcr |= 1 << 24;
		}
		// Write back the modified FPCR
		asm volatile ("msr fpcr, %0" : : "r" (fpcr));
#endif
	}
}

void IRRestoreRounding() {
#if PPSSPP_ARCH(SSE2)
	// TODO: We should avoid this if we didn't apply rounding in the first place.
	// In the meantime, clear out FTZ and rounding mode bits.
	u32 csr = _mm_getcsr();
	csr &= ~(7 << 13);
	_mm_setcsr(csr);
#elif PPSSPP_ARCH(ARM64) && !PPSSPP_PLATFORM(WINDOWS)
	u64 fpcr;  // not really 64-bit, just to match the regsiter size.
	asm volatile ("mrs %0, fpcr" : "=r" (fpcr));
	fpcr &= ~(7 << 22);    // Clear bits [23:22] for rounding, 24 for FTZ
	// Write back the modified FPCR
	asm volatile ("msr fpcr, %0" : : "r" (fpcr));
#endif
}

// We cannot use NEON on ARM32 here until we make it a hard dependency. We can, however, on ARM64.
u32 IRInterpret(MIPSState *mips, const IRInst *inst) {
	while (true) {
		switch (inst->op) {
		case IROp::SetConst:
			mips->r[inst->dest] = inst->constant;
			break;
		case IROp::SetConstF:
			memcpy(&mips->f[inst->dest], &inst->constant, 4);
			break;
		case IROp::Add:
			mips->r[inst->dest] = mips->r[inst->src1] + mips->r[inst->src2];
			break;
		case IROp::Sub:
			mips->r[inst->dest] = mips->r[inst->src1] - mips->r[inst->src2];
			break;
		case IROp::And:
			mips->r[inst->dest] = mips->r[inst->src1] & mips->r[inst->src2];
			break;
		case IROp::Or:
			mips->r[inst->dest] = mips->r[inst->src1] | mips->r[inst->src2];
			break;
		case IROp::Xor:
			mips->r[inst->dest] = mips->r[inst->src1] ^ mips->r[inst->src2];
			break;
		case IROp::Mov:
			mips->r[inst->dest] = mips->r[inst->src1];
			break;
		case IROp::AddConst:
			mips->r[inst->dest] = mips->r[inst->src1] + inst->constant;
			break;
		case IROp::OptAddConst:  // For this one, it's worth having a "unary" variant of the above that only needs to read one register param.
			mips->r[inst->dest] += inst->constant;
			break;
		case IROp::SubConst:
			mips->r[inst->dest] = mips->r[inst->src1] - inst->constant;
			break;
		case IROp::AndConst:
			mips->r[inst->dest] = mips->r[inst->src1] & inst->constant;
			break;
		case IROp::OptAndConst:  // For this one, it's worth having a "unary" variant of the above that only needs to read one register param.
			mips->r[inst->dest] &= inst->constant;
			break;
		case IROp::OrConst:
			mips->r[inst->dest] = mips->r[inst->src1] | inst->constant;
			break;
		case IROp::OptOrConst:
			mips->r[inst->dest] |= inst->constant;
			break;
		case IROp::XorConst:
			mips->r[inst->dest] = mips->r[inst->src1] ^ inst->constant;
			break;
		case IROp::Neg:
			mips->r[inst->dest] = (u32)(-(s32)mips->r[inst->src1]);
			break;
		case IROp::Not:
			mips->r[inst->dest] = ~mips->r[inst->src1];
			break;
		case IROp::Ext8to32:
			mips->r[inst->dest] = SignExtend8ToU32(mips->r[inst->src1]);
			break;
		case IROp::Ext16to32:
			mips->r[inst->dest] = SignExtend16ToU32(mips->r[inst->src1]);
			break;
		case IROp::ReverseBits:
			mips->r[inst->dest] = ReverseBits32(mips->r[inst->src1]);
			break;

		case IROp::Load8:
			mips->r[inst->dest] = Memory::ReadUnchecked_U8(mips->r[inst->src1] + inst->constant);
			break;
		case IROp::Load8Ext:
			mips->r[inst->dest] = SignExtend8ToU32(Memory::ReadUnchecked_U8(mips->r[inst->src1] + inst->constant));
			break;
		case IROp::Load16:
			mips->r[inst->dest] = Memory::ReadUnchecked_U16(mips->r[inst->src1] + inst->constant);
			break;
		case IROp::Load16Ext:
			mips->r[inst->dest] = SignExtend16ToU32(Memory::ReadUnchecked_U16(mips->r[inst->src1] + inst->constant));
			break;
		case IROp::Load32:
			mips->r[inst->dest] = Memory::ReadUnchecked_U32(mips->r[inst->src1] + inst->constant);
			break;
		case IROp::Load32Left:
		{
			u32 addr = mips->r[inst->src1] + inst->constant;
			u32 shift = (addr & 3) * 8;
			u32 mem = Memory::ReadUnchecked_U32(addr & 0xfffffffc);
			u32 destMask = 0x00ffffff >> shift;
			mips->r[inst->dest] = (mips->r[inst->dest] & destMask) | (mem << (24 - shift));
			break;
		}
		case IROp::Load32Right:
		{
			u32 addr = mips->r[inst->src1] + inst->constant;
			u32 shift = (addr & 3) * 8;
			u32 mem = Memory::ReadUnchecked_U32(addr & 0xfffffffc);
			u32 destMask = 0xffffff00 << (24 - shift);
			mips->r[inst->dest] = (mips->r[inst->dest] & destMask) | (mem >> shift);
			break;
		}
		case IROp::Load32Linked:
			if (inst->dest != MIPS_REG_ZERO)
				mips->r[inst->dest] = Memory::ReadUnchecked_U32(mips->r[inst->src1] + inst->constant);
			mips->llBit = 1;
			break;
		case IROp::LoadFloat:
			mips->f[inst->dest] = Memory::ReadUnchecked_Float(mips->r[inst->src1] + inst->constant);
			break;

		case IROp::Store8:
			Memory::WriteUnchecked_U8(mips->r[inst->src3], mips->r[inst->src1] + inst->constant);
			break;
		case IROp::Store16:
			Memory::WriteUnchecked_U16(mips->r[inst->src3], mips->r[inst->src1] + inst->constant);
			break;
		case IROp::Store32:
			Memory::WriteUnchecked_U32(mips->r[inst->src3], mips->r[inst->src1] + inst->constant);
			break;
		case IROp::Store32Left:
		{
			u32 addr = mips->r[inst->src1] + inst->constant;
			u32 shift = (addr & 3) * 8;
			u32 mem = Memory::ReadUnchecked_U32(addr & 0xfffffffc);
			u32 memMask = 0xffffff00 << shift;
			u32 result = (mips->r[inst->src3] >> (24 - shift)) | (mem & memMask);
			Memory::WriteUnchecked_U32(result, addr & 0xfffffffc);
			break;
		}
		case IROp::Store32Right:
		{
			u32 addr = mips->r[inst->src1] + inst->constant;
			u32 shift = (addr & 3) * 8;
			u32 mem = Memory::ReadUnchecked_U32(addr & 0xfffffffc);
			u32 memMask = 0x00ffffff >> (24 - shift);
			u32 result = (mips->r[inst->src3] << shift) | (mem & memMask);
			Memory::WriteUnchecked_U32(result, addr & 0xfffffffc);
			break;
		}
		case IROp::Store32Conditional:
			if (mips->llBit) {
				Memory::WriteUnchecked_U32(mips->r[inst->src3], mips->r[inst->src1] + inst->constant);
				if (inst->dest != MIPS_REG_ZERO) {
					mips->r[inst->dest] = 1;
				}
			} else if (inst->dest != MIPS_REG_ZERO) {
				mips->r[inst->dest] = 0;
			}
			break;
		case IROp::StoreFloat:
			Memory::WriteUnchecked_Float(mips->f[inst->src3], mips->r[inst->src1] + inst->constant);
			break;

		case IROp::LoadVec4:
		{
			u32 base = mips->r[inst->src1] + inst->constant;
			// This compiles to a nice SSE load/store on x86, and hopefully similar on ARM.
			memcpy(&mips->f[inst->dest], Memory::GetPointerUnchecked(base), 4 * 4);
			break;
		}
		case IROp::StoreVec4:
		{
			u32 base = mips->r[inst->src1] + inst->constant;
			memcpy((float *)Memory::GetPointerUnchecked(base), &mips->f[inst->dest], 4 * 4);
			break;
		}

		case IROp::Vec4Init:
		{
			memcpy(&mips->f[inst->dest], vec4InitValues[inst->src1], 4 * sizeof(float));
			break;
		}

		case IROp::Vec4Shuffle:
		{
			// Can't use the SSE shuffle here because it takes an immediate. pshufb with a table would work though,
			// or a big switch - there are only 256 shuffles possible (4^4)
			float temp[4];
			for (int i = 0; i < 4; i++)
				temp[i] = mips->f[inst->src1 + ((inst->src2 >> (i * 2)) & 3)];
			const int dest = inst->dest;
			for (int i = 0; i < 4; i++)
				mips->f[dest + i] = temp[i];
			break;
		}

		case IROp::Vec4Blend:
		{
			const int dest = inst->dest;
			const int src1 = inst->src1;
			const int src2 = inst->src2;
			const int constant = inst->constant;
			// 90% of calls to this is inst->constant == 7 or inst->constant == 8. Some are 1 and 4, others very rare.
			// Could use _mm_blendv_ps (SSE4+BMI), vbslq_f32 (ARM), __riscv_vmerge_vvm (RISC-V)
			for (int i = 0; i < 4; i++)
				mips->f[dest + i] = ((constant >> i) & 1) ? mips->f[src2 + i] : mips->f[src1 + i];
			break;
		}

		case IROp::Vec4Mov:
		{
#if defined(_M_SSE)
			_mm_store_ps(&mips->f[inst->dest], _mm_load_ps(&mips->f[inst->src1]));
#elif PPSSPP_ARCH(ARM_NEON)
			vst1q_f32(&mips->f[inst->dest], vld1q_f32(&mips->f[inst->src1]));
#else
			memcpy(&mips->f[inst->dest], &mips->f[inst->src1], 4 * sizeof(float));
#endif
			break;
		}

		case IROp::Vec4Add:
		{
#if defined(_M_SSE)
			_mm_store_ps(&mips->f[inst->dest], _mm_add_ps(_mm_load_ps(&mips->f[inst->src1]), _mm_load_ps(&mips->f[inst->src2])));
#elif PPSSPP_ARCH(ARM_NEON)
			vst1q_f32(&mips->f[inst->dest], vaddq_f32(vld1q_f32(&mips->f[inst->src1]), vld1q_f32(&mips->f[inst->src2])));
#else
			for (int i = 0; i < 4; i++)
				mips->f[inst->dest + i] = mips->f[inst->src1 + i] + mips->f[inst->src2 + i];
#endif
			break;
		}

		case IROp::Vec4Sub:
		{
#if defined(_M_SSE)
			_mm_store_ps(&mips->f[inst->dest], _mm_sub_ps(_mm_load_ps(&mips->f[inst->src1]), _mm_load_ps(&mips->f[inst->src2])));
#elif PPSSPP_ARCH(ARM_NEON)
			vst1q_f32(&mips->f[inst->dest], vsubq_f32(vld1q_f32(&mips->f[inst->src1]), vld1q_f32(&mips->f[inst->src2])));
#else
			for (int i = 0; i < 4; i++)
				mips->f[inst->dest + i] = mips->f[inst->src1 + i] - mips->f[inst->src2 + i];
#endif
			break;
		}

		case IROp::Vec4Mul:
		{
#if defined(_M_SSE)
			_mm_store_ps(&mips->f[inst->dest], _mm_mul_ps(_mm_load_ps(&mips->f[inst->src1]), _mm_load_ps(&mips->f[inst->src2])));
#elif PPSSPP_ARCH(ARM_NEON)
			vst1q_f32(&mips->f[inst->dest], vmulq_f32(vld1q_f32(&mips->f[inst->src1]), vld1q_f32(&mips->f[inst->src2])));
#else
			for (int i = 0; i < 4; i++)
				mips->f[inst->dest + i] = mips->f[inst->src1 + i] * mips->f[inst->src2 + i];
#endif
			break;
		}

		case IROp::Vec4Div:
		{
#if defined(_M_SSE)
			_mm_store_ps(&mips->f[inst->dest], _mm_div_ps(_mm_load_ps(&mips->f[inst->src1]), _mm_load_ps(&mips->f[inst->src2])));
#elif PPSSPP_ARCH(ARM64_NEON)
			vst1q_f32(&mips->f[inst->dest], vdivq_f32(vld1q_f32(&mips->f[inst->src1]), vld1q_f32(&mips->f[inst->src2])));
#else
			for (int i = 0; i < 4; i++)
				mips->f[inst->dest + i] = mips->f[inst->src1 + i] / mips->f[inst->src2 + i];
#endif
			break;
		}

		case IROp::Vec4Scale:
		{
#if defined(_M_SSE)
			_mm_store_ps(&mips->f[inst->dest], _mm_mul_ps(_mm_load_ps(&mips->f[inst->src1]), _mm_set1_ps(mips->f[inst->src2])));
#elif PPSSPP_ARCH(ARM_NEON)
			vst1q_f32(&mips->f[inst->dest], vmulq_lane_f32(vld1q_f32(&mips->f[inst->src1]), vdup_n_f32(mips->f[inst->src2]), 0));
#else
			const float factor = mips->f[inst->src2];
			for (int i = 0; i < 4; i++)
				mips->f[inst->dest + i] = mips->f[inst->src1 + i] * factor;
#endif
			break;
		}

		case IROp::Vec4Neg:
		{
#if defined(_M_SSE)
			_mm_store_ps(&mips->f[inst->dest], _mm_xor_ps(_mm_load_ps(&mips->f[inst->src1]), _mm_load_ps((const float *)signBits)));
#elif PPSSPP_ARCH(ARM_NEON)
			vst1q_f32(&mips->f[inst->dest], vnegq_f32(vld1q_f32(&mips->f[inst->src1])));
#else
			for (int i = 0; i < 4; i++)
				mips->f[inst->dest + i] = -mips->f[inst->src1 + i];
#endif
			break;
		}

		case IROp::Vec4Abs:
		{
#if defined(_M_SSE)
			_mm_store_ps(&mips->f[inst->dest], _mm_and_ps(_mm_load_ps(&mips->f[inst->src1]), _mm_load_ps((const float *)noSignMask)));
#elif PPSSPP_ARCH(ARM_NEON)
			vst1q_f32(&mips->f[inst->dest], vabsq_f32(vld1q_f32(&mips->f[inst->src1])));
#else
			for (int i = 0; i < 4; i++)
				mips->f[inst->dest + i] = fabsf(mips->f[inst->src1 + i]);
#endif
			break;
		}

		case IROp::Vec2Unpack16To31:
		{
			const int dest = inst->dest;
			const int src1 = inst->src1;
			mips->fi[dest] = (mips->fi[src1] << 16) >> 1;
			mips->fi[dest + 1] = (mips->fi[src1] & 0xFFFF0000) >> 1;
			break;
		}

		case IROp::Vec2Unpack16To32:
		{
			const int dest = inst->dest;
			const int src1 = inst->src1;
			mips->fi[dest] = (mips->fi[src1] << 16);
			mips->fi[dest + 1] = (mips->fi[src1] & 0xFFFF0000);
			break;
		}

		case IROp::Vec4Unpack8To32:
		{
#if defined(_M_SSE)
			__m128i src = _mm_cvtsi32_si128(mips->fi[inst->src1]);
			src = _mm_unpacklo_epi8(src, _mm_setzero_si128());
			src = _mm_unpacklo_epi16(src, _mm_setzero_si128());
			_mm_store_si128((__m128i *)&mips->fi[inst->dest], _mm_slli_epi32(src, 24));
#elif PPSSPP_ARCH(ARM_NEON) && 0 // Untested
			const uint8x8_t value = (uint8x8_t)vdup_n_u32(mips->fi[inst->src1]);
			const uint16x8_t value16 = vmovl_u8(value);
			const uint32x4_t value32 = vshll_n_u16(vget_low_u16(value16), 24);
			vst1q_u32(&mips->fi[inst->dest], value32);
#else
			mips->fi[inst->dest] = (mips->fi[inst->src1] << 24);
			mips->fi[inst->dest + 1] = (mips->fi[inst->src1] << 16) & 0xFF000000;
			mips->fi[inst->dest + 2] = (mips->fi[inst->src1] << 8) & 0xFF000000;
			mips->fi[inst->dest + 3] = (mips->fi[inst->src1]) & 0xFF000000;
#endif
			break;
		}

		case IROp::Vec2Pack32To16:
		{
			u32 val = mips->fi[inst->src1] >> 16;
			mips->fi[inst->dest] = (mips->fi[inst->src1 + 1] & 0xFFFF0000) | val;
			break;
		}

		case IROp::Vec2Pack31To16:
		{
			// Used in Tekken 6

			u32 val = (mips->fi[inst->src1] >> 15) & 0xFFFF;
			val |= (mips->fi[inst->src1 + 1] << 1) & 0xFFFF0000;
			mips->fi[inst->dest] = val;
			break;
		}

		case IROp::Vec4Pack32To8:
		{
			// Removed previous SSE code due to the need for unsigned 16-bit pack, which I'm too lazy to work around the lack of in SSE2.
			// pshufb or SSE4 instructions can be used instead.
			u32 val = mips->fi[inst->src1] >> 24;
			val |= (mips->fi[inst->src1 + 1] >> 16) & 0xFF00;
			val |= (mips->fi[inst->src1 + 2] >> 8) & 0xFF0000;
			val |= (mips->fi[inst->src1 + 3]) & 0xFF000000;
			mips->fi[inst->dest] = val;
			break;
		}

		case IROp::Vec4Pack31To8:
		{
			// Used in Tekken 6

			// Removed previous SSE code due to the need for unsigned 16-bit pack, which I'm too lazy to work around the lack of in SSE2.
			// pshufb or SSE4 instructions can be used instead.
#if PPSSPP_ARCH(ARM_NEON) && 0
			// Untested
			uint32x4_t value = vld1q_u32(&mips->fi[inst->src1]);
			value = vshlq_n_u32(value, 1);
			uint32x2_t halved = vshrn_n_u32(value, 8);
			uint32x2_t halvedAgain = vshrn_n_u32(vcombine_u32(halved, vdup_n_u32(0)), 8);
			mips->fi[inst->dest] = vget_lane_u32(halvedAgain, 0);
#else
			u32 val = (mips->fi[inst->src1] >> 23) & 0xFF;
			val |= (mips->fi[inst->src1 + 1] >> 15) & 0xFF00;
			val |= (mips->fi[inst->src1 + 2] >> 7) & 0xFF0000;
			val |= (mips->fi[inst->src1 + 3] << 1) & 0xFF000000;
			mips->fi[inst->dest] = val;
#endif
			break;
		}

		case IROp::Vec2ClampToZero:
		{
			for (int i = 0; i < 2; i++) {
				u32 val = mips->fi[inst->src1 + i];
				mips->fi[inst->dest + i] = (int)val >= 0 ? val : 0;
			}
			break;
		}

		case IROp::Vec4ClampToZero:
		{
#if defined(_M_SSE)
			// Trickery: Expand the sign bit, and use andnot to zero negative values.
			__m128i val = _mm_load_si128((const __m128i *)&mips->fi[inst->src1]);
			__m128i mask = _mm_srai_epi32(val, 31);
			val = _mm_andnot_si128(mask, val);
			_mm_store_si128((__m128i *)&mips->fi[inst->dest], val);
#else
			const int src1 = inst->src1;
			const int dest = inst->dest;
			for (int i = 0; i < 4; i++) {
				u32 val = mips->fi[src1 + i];
				mips->fi[dest + i] = (int)val >= 0 ? val : 0;
			}
#endif
			break;
		}

		case IROp::Vec4DuplicateUpperBitsAndShift1:  // For vuc2i, the weird one.
		{
			const int src1 = inst->src1;
			const int dest = inst->dest;
			for (int i = 0; i < 4; i++) {
				u32 val = mips->fi[src1 + i];
				val = val | (val >> 8);
				val = val | (val >> 16);
				val >>= 1;
				mips->fi[dest + i] = val;
			}
			break;
		}

		case IROp::FCmpVfpuBit:
		{
			const int op = inst->dest & 0xF;
			const int bit = inst->dest >> 4;
			int result = 0;
			switch (op) {
			case VC_EQ: result = mips->f[inst->src1] == mips->f[inst->src2]; break;
			case VC_NE: result = mips->f[inst->src1] != mips->f[inst->src2]; break;
			case VC_LT: result = mips->f[inst->src1] < mips->f[inst->src2]; break;
			case VC_LE: result = mips->f[inst->src1] <= mips->f[inst->src2]; break;
			case VC_GT: result = mips->f[inst->src1] > mips->f[inst->src2]; break;
			case VC_GE: result = mips->f[inst->src1] >= mips->f[inst->src2]; break;
			case VC_EZ: result = mips->f[inst->src1] == 0.0f; break;
			case VC_NZ: result = mips->f[inst->src1] != 0.0f; break;
			case VC_EN: result = my_isnan(mips->f[inst->src1]); break;
			case VC_NN: result = !my_isnan(mips->f[inst->src1]); break;
			case VC_EI: result = my_isinf(mips->f[inst->src1]); break;
			case VC_NI: result = !my_isinf(mips->f[inst->src1]); break;
			case VC_ES: result = my_isnanorinf(mips->f[inst->src1]); break;
			case VC_NS: result = !my_isnanorinf(mips->f[inst->src1]); break;
			case VC_TR: result = 1; break;
			case VC_FL: result = 0; break;
			default:
				result = 0;
			}
			if (result != 0) {
				mips->vfpuCtrl[VFPU_CTRL_CC] |= (1 << bit);
			} else {
				mips->vfpuCtrl[VFPU_CTRL_CC] &= ~(1 << bit);
			}
			break;
		}

		case IROp::FCmpVfpuAggregate:
		{
			const u32 mask = inst->dest;
			const u32 cc = mips->vfpuCtrl[VFPU_CTRL_CC];
			int anyBit = (cc & mask) ? 0x10 : 0x00;
			int allBit = (cc & mask) == mask ? 0x20 : 0x00;
			mips->vfpuCtrl[VFPU_CTRL_CC] = (cc & ~0x30) | anyBit | allBit;
			break;
		}

		case IROp::FCmovVfpuCC:
			if (((mips->vfpuCtrl[VFPU_CTRL_CC] >> (inst->src2 & 0xf)) & 1) == ((u32)inst->src2 >> 7)) {
				mips->f[inst->dest] = mips->f[inst->src1];
			}
			break;

		case IROp::Vec4Dot:
		{
			// Not quickly implementable on all platforms, unfortunately.
			// Though, this is still pretty fast compared to one split into multiple IR instructions.
			// This might be good though: https://gist.github.com/rikusalminen/3040241
			float dot = mips->f[inst->src1] * mips->f[inst->src2];
			for (int i = 1; i < 4; i++)
				dot += mips->f[inst->src1 + i] * mips->f[inst->src2 + i];
			mips->f[inst->dest] = dot;
			break;
		}

		case IROp::FSin:
			mips->f[inst->dest] = vfpu_sin(mips->f[inst->src1]);
			break;
		case IROp::FCos:
			mips->f[inst->dest] = vfpu_cos(mips->f[inst->src1]);
			break;
		case IROp::FRSqrt:
			mips->f[inst->dest] = 1.0f / sqrtf(mips->f[inst->src1]);
			break;
		case IROp::FRecip:
			mips->f[inst->dest] = 1.0f / mips->f[inst->src1];
			break;
		case IROp::FAsin:
			mips->f[inst->dest] = vfpu_asin(mips->f[inst->src1]);
			break;

		case IROp::ShlImm:
			mips->r[inst->dest] = mips->r[inst->src1] << (int)inst->src2;
			break;
		case IROp::ShrImm:
			mips->r[inst->dest] = mips->r[inst->src1] >> (int)inst->src2;
			break;
		case IROp::SarImm:
			mips->r[inst->dest] = (s32)mips->r[inst->src1] >> (int)inst->src2;
			break;
		case IROp::RorImm:
		{
			u32 x = mips->r[inst->src1];
			int sa = inst->src2;
			mips->r[inst->dest] = (x >> sa) | (x << (32 - sa));
		}
		break;

		case IROp::Shl:
			mips->r[inst->dest] = mips->r[inst->src1] << (mips->r[inst->src2] & 31);
			break;
		case IROp::Shr:
			mips->r[inst->dest] = mips->r[inst->src1] >> (mips->r[inst->src2] & 31);
			break;
		case IROp::Sar:
			mips->r[inst->dest] = (s32)mips->r[inst->src1] >> (mips->r[inst->src2] & 31);
			break;
		case IROp::Ror:
		{
			u32 x = mips->r[inst->src1];
			int sa = mips->r[inst->src2] & 31;
			mips->r[inst->dest] = (x >> sa) | (x << (32 - sa));
			break;
		}

		case IROp::Clz:
		{
			mips->r[inst->dest] = clz32(mips->r[inst->src1]);
			break;
		}

		case IROp::Slt:
			mips->r[inst->dest] = (s32)mips->r[inst->src1] < (s32)mips->r[inst->src2];
			break;

		case IROp::SltU:
			mips->r[inst->dest] = mips->r[inst->src1] < mips->r[inst->src2];
			break;

		case IROp::SltConst:
			mips->r[inst->dest] = (s32)mips->r[inst->src1] < (s32)inst->constant;
			break;

		case IROp::SltUConst:
			mips->r[inst->dest] = mips->r[inst->src1] < inst->constant;
			break;

		case IROp::MovZ:
			if (mips->r[inst->src1] == 0)
				mips->r[inst->dest] = mips->r[inst->src2];
			break;
		case IROp::MovNZ:
			if (mips->r[inst->src1] != 0)
				mips->r[inst->dest] = mips->r[inst->src2];
			break;

		case IROp::Max:
			mips->r[inst->dest] = (s32)mips->r[inst->src1] > (s32)mips->r[inst->src2] ? mips->r[inst->src1] : mips->r[inst->src2];
			break;
		case IROp::Min:
			mips->r[inst->dest] = (s32)mips->r[inst->src1] < (s32)mips->r[inst->src2] ? mips->r[inst->src1] : mips->r[inst->src2];
			break;

		case IROp::MtLo:
			mips->lo = mips->r[inst->src1];
			break;
		case IROp::MtHi:
			mips->hi = mips->r[inst->src1];
			break;
		case IROp::MfLo:
			mips->r[inst->dest] = mips->lo;
			break;
		case IROp::MfHi:
			mips->r[inst->dest] = mips->hi;
			break;

		case IROp::Mult:
		{
			s64 result = (s64)(s32)mips->r[inst->src1] * (s64)(s32)mips->r[inst->src2];
			memcpy(&mips->lo, &result, 8);
			break;
		}
		case IROp::MultU:
		{
			u64 result = (u64)mips->r[inst->src1] * (u64)mips->r[inst->src2];
			memcpy(&mips->lo, &result, 8);
			break;
		}
		case IROp::Madd:
		{
			s64 result;
			memcpy(&result, &mips->lo, 8);
			result += (s64)(s32)mips->r[inst->src1] * (s64)(s32)mips->r[inst->src2];
			memcpy(&mips->lo, &result, 8);
			break;
		}
		case IROp::MaddU:
		{
			s64 result;
			memcpy(&result, &mips->lo, 8);
			result += (u64)mips->r[inst->src1] * (u64)mips->r[inst->src2];
			memcpy(&mips->lo, &result, 8);
			break;
		}
		case IROp::Msub:
		{
			s64 result;
			memcpy(&result, &mips->lo, 8);
			result -= (s64)(s32)mips->r[inst->src1] * (s64)(s32)mips->r[inst->src2];
			memcpy(&mips->lo, &result, 8);
			break;
		}
		case IROp::MsubU:
		{
			s64 result;
			memcpy(&result, &mips->lo, 8);
			result -= (u64)mips->r[inst->src1] * (u64)mips->r[inst->src2];
			memcpy(&mips->lo, &result, 8);
			break;
		}

		case IROp::Div:
		{
			s32 numerator = (s32)mips->r[inst->src1];
			s32 denominator = (s32)mips->r[inst->src2];
			if (numerator == (s32)0x80000000 && denominator == -1) {
				mips->lo = 0x80000000;
				mips->hi = -1;
			} else if (denominator != 0) {
				mips->lo = (u32)(numerator / denominator);
				mips->hi = (u32)(numerator % denominator);
			} else {
				mips->lo = numerator < 0 ? 1 : -1;
				mips->hi = numerator;
			}
			break;
		}
		case IROp::DivU:
		{
			u32 numerator = mips->r[inst->src1];
			u32 denominator = mips->r[inst->src2];
			if (denominator != 0) {
				mips->lo = numerator / denominator;
				mips->hi = numerator % denominator;
			} else {
				mips->lo = numerator <= 0xFFFF ? 0xFFFF : -1;
				mips->hi = numerator;
			}
			break;
		}

		case IROp::BSwap16:
		{
			u32 x = mips->r[inst->src1];
			// Don't think we can beat this with intrinsics.
			mips->r[inst->dest] = ((x & 0xFF00FF00) >> 8) | ((x & 0x00FF00FF) << 8);
			break;
		}
		case IROp::BSwap32:
		{
			mips->r[inst->dest] = swap32(mips->r[inst->src1]);
			break;
		}

		case IROp::FAdd:
			mips->f[inst->dest] = mips->f[inst->src1] + mips->f[inst->src2];
			break;
		case IROp::FSub:
			mips->f[inst->dest] = mips->f[inst->src1] - mips->f[inst->src2];
			break;
		case IROp::FMul:
#if 1
		{
			float a = mips->f[inst->src1];
			float b = mips->f[inst->src2];
			if ((b == 0.0f && my_isinf(a)) || (a == 0.0f && my_isinf(b))) {
				mips->fi[inst->dest] = 0x7fc00000;
			} else {
				mips->f[inst->dest] = a * b;
			}
		}
			break;
#else
			// Not sure if faster since it needs to load the operands twice? But the code is simpler.
			{
				// Takes care of negative zero by masking away the top bit, which also makes the inf check shorter.
				u32 a = mips->fi[inst->src1] & 0x7FFFFFFF;
				u32 b = mips->fi[inst->src2] & 0x7FFFFFFF;
				if ((a == 0 && b == 0x7F800000) || (b == 0 && a == 0x7F800000)) {
					mips->fi[inst->dest] = 0x7fc00000;
				} else {
					mips->f[inst->dest] = mips->f[inst->src1] * mips->f[inst->src2];
				}
				break;
			}
#endif
		case IROp::FDiv:
			mips->f[inst->dest] = mips->f[inst->src1] / mips->f[inst->src2];
			break;
		case IROp::FMin:
			if (my_isnan(mips->f[inst->src1]) || my_isnan(mips->f[inst->src2])) {
				// See interpreter for this logic: this is for vmin, we're comparing mantissa+exp.
				if (mips->fs[inst->src1] < 0 && mips->fs[inst->src2] < 0) {
					mips->fs[inst->dest] = std::max(mips->fs[inst->src1], mips->fs[inst->src2]);
				} else {
					mips->fs[inst->dest] = std::min(mips->fs[inst->src1], mips->fs[inst->src2]);
				}
			} else {
				mips->f[inst->dest] = std::min(mips->f[inst->src1], mips->f[inst->src2]);
			}
			break;
		case IROp::FMax:
			if (my_isnan(mips->f[inst->src1]) || my_isnan(mips->f[inst->src2])) {
				// See interpreter for this logic: this is for vmax, we're comparing mantissa+exp.
				if (mips->fs[inst->src1] < 0 && mips->fs[inst->src2] < 0) {
					mips->fs[inst->dest] = std::min(mips->fs[inst->src1], mips->fs[inst->src2]);
				} else {
					mips->fs[inst->dest] = std::max(mips->fs[inst->src1], mips->fs[inst->src2]);
				}
			} else {
				mips->f[inst->dest] = std::max(mips->f[inst->src1], mips->f[inst->src2]);
			}
			break;

		case IROp::FMov:
			mips->f[inst->dest] = mips->f[inst->src1];
			break;
		case IROp::FAbs:
			mips->f[inst->dest] = fabsf(mips->f[inst->src1]);
			break;
		case IROp::FSqrt:
			mips->f[inst->dest] = sqrtf(mips->f[inst->src1]);
			break;
		case IROp::FNeg:
			mips->f[inst->dest] = -mips->f[inst->src1];
			break;
		case IROp::FSat0_1:
			// We have to do this carefully to handle NAN and -0.0f.
			mips->f[inst->dest] = vfpu_clamp(mips->f[inst->src1], 0.0f, 1.0f);
			break;
		case IROp::FSatMinus1_1:
			mips->f[inst->dest] = vfpu_clamp(mips->f[inst->src1], -1.0f, 1.0f);
			break;

		case IROp::FSign:
		{
			// Bitwise trickery
			u32 val;
			memcpy(&val, &mips->f[inst->src1], sizeof(u32));
			if (val == 0 || val == 0x80000000)
				mips->f[inst->dest] = 0.0f;
			else if ((val >> 31) == 0)
				mips->f[inst->dest] = 1.0f;
			else
				mips->f[inst->dest] = -1.0f;
			break;
		}

		case IROp::FpCondFromReg:
			mips->fpcond = mips->r[inst->dest];
			break;
		case IROp::FpCondToReg:
			mips->r[inst->dest] = mips->fpcond;
			break;
		case IROp::FpCtrlFromReg:
			mips->fcr31 = mips->r[inst->src1] & 0x0181FFFF;
			// Extract the new fpcond value.
			// TODO: Is it really helping us to keep it separate?
			mips->fpcond = (mips->fcr31 >> 23) & 1;
			break;
		case IROp::FpCtrlToReg:
			// Update the fpcond bit first.
			mips->fcr31 = (mips->fcr31 & ~(1 << 23)) | ((mips->fpcond & 1) << 23);
			mips->r[inst->dest] = mips->fcr31;
			break;
		case IROp::VfpuCtrlToReg:
			mips->r[inst->dest] = mips->vfpuCtrl[inst->src1];
			break;
		case IROp::FRound:
		{
			float value = mips->f[inst->src1];
			if (my_isnanorinf(value)) {
				mips->fi[inst->dest] = my_isinf(value) && value < 0.0f ? -2147483648LL : 2147483647LL;
				break;
			} else {
				mips->fs[inst->dest] = (int)round_ieee_754(value);
			}
			break;
		}
		case IROp::FTrunc:
		{
			float value = mips->f[inst->src1];
			if (my_isnanorinf(value)) {
				mips->fi[inst->dest] = my_isinf(value) && value < 0.0f ? -2147483648LL : 2147483647LL;
				break;
			} else {
				if (value >= 0.0f) {
					mips->fs[inst->dest] = (int)floorf(value);
					// Overflow, but it was positive.
					if (mips->fs[inst->dest] == -2147483648LL) {
						mips->fs[inst->dest] = 2147483647LL;
					}
				} else {
					// Overflow happens to be the right value anyway.
					mips->fs[inst->dest] = (int)ceilf(value);
				}
				break;
			}
		}
		case IROp::FCeil:
		{
			float value = mips->f[inst->src1];
			if (my_isnanorinf(value)) {
				mips->fi[inst->dest] = my_isinf(value) && value < 0.0f ? -2147483648LL : 2147483647LL;
				break;
			} else {
				mips->fs[inst->dest] = (int)ceilf(value);
			}
			break;
		}
		case IROp::FFloor:
		{
			float value = mips->f[inst->src1];
			if (my_isnanorinf(value)) {
				mips->fi[inst->dest] = my_isinf(value) && value < 0.0f ? -2147483648LL : 2147483647LL;
				break;
			} else {
				mips->fs[inst->dest] = (int)floorf(value);
			}
			break;
		}
		case IROp::FCmp:
			switch (inst->dest) {
			case IRFpCompareMode::False:
				mips->fpcond = 0;
				break;
			case IRFpCompareMode::EitherUnordered:
			{
				float a = mips->f[inst->src1];
				float b = mips->f[inst->src2];
				mips->fpcond = !(a > b || a < b || a == b);
				break;
			}
			case IRFpCompareMode::EqualOrdered:
				mips->fpcond = mips->f[inst->src1] == mips->f[inst->src2];
				break;
			case IRFpCompareMode::EqualUnordered:
				mips->fpcond = mips->f[inst->src1] == mips->f[inst->src2] || my_isnan(mips->f[inst->src1]) || my_isnan(mips->f[inst->src2]);
				break;
			case IRFpCompareMode::LessEqualOrdered:
				mips->fpcond = mips->f[inst->src1] <= mips->f[inst->src2];
				break;
			case IRFpCompareMode::LessEqualUnordered:
				mips->fpcond = !(mips->f[inst->src1] > mips->f[inst->src2]);
				break;
			case IRFpCompareMode::LessOrdered:
				mips->fpcond = mips->f[inst->src1] < mips->f[inst->src2];
				break;
			case IRFpCompareMode::LessUnordered:
				mips->fpcond = !(mips->f[inst->src1] >= mips->f[inst->src2]);
				break;
			}
			break;

		case IROp::FCvtSW:
			mips->f[inst->dest] = (float)mips->fs[inst->src1];
			break;
		case IROp::FCvtWS:
		{
			float src = mips->f[inst->src1];
			if (my_isnanorinf(src)) {
				mips->fs[inst->dest] = my_isinf(src) && src < 0.0f ? -2147483648LL : 2147483647LL;
				break;
			}
			// TODO: Inline assembly to use here would be better.
			switch (IRRoundMode(mips->fcr31 & 3)) {
			case IRRoundMode::RINT_0: mips->fs[inst->dest] = (int)round_ieee_754(src); break;
			case IRRoundMode::CAST_1: mips->fs[inst->dest] = (int)src; break;
			case IRRoundMode::CEIL_2: mips->fs[inst->dest] = (int)ceilf(src); break;
			case IRRoundMode::FLOOR_3: mips->fs[inst->dest] = (int)floorf(src); break;
			}
			break; //cvt.w.s
		}
		case IROp::FCvtScaledSW:
			mips->f[inst->dest] = (float)mips->fs[inst->src1] * (1.0f / (1UL << (inst->src2 & 0x1F)));
			break;
		case IROp::FCvtScaledWS:
		{
			float src = mips->f[inst->src1];
			if (my_isnan(src)) {
				// TODO: True for negatives too?
				mips->fs[inst->dest] = 2147483647L;
				break;
			}

			float mult = (float)(1UL << (inst->src2 & 0x1F));
			double sv = src * mult; // (float)0x7fffffff == (float)0x80000000
			// Cap/floor it to 0x7fffffff / 0x80000000
			if (sv > (double)0x7fffffff) {
				mips->fs[inst->dest] = 0x7fffffff;
			} else if (sv <= (double)(int)0x80000000) {
				mips->fs[inst->dest] = 0x80000000;
			} else {
				switch (IRRoundMode(inst->src2 >> 6)) {
				case IRRoundMode::RINT_0: mips->fs[inst->dest] = (int)round_ieee_754(sv); break;
				case IRRoundMode::CAST_1: mips->fs[inst->dest] = src >= 0 ? (int)floor(sv) : (int)ceil(sv); break;
				case IRRoundMode::CEIL_2: mips->fs[inst->dest] = (int)ceil(sv); break;
				case IRRoundMode::FLOOR_3: mips->fs[inst->dest] = (int)floor(sv); break;
				}
			}
			break;
		}

		case IROp::FMovFromGPR:
			memcpy(&mips->f[inst->dest], &mips->r[inst->src1], 4);
			break;
		case IROp::OptFCvtSWFromGPR:
			mips->f[inst->dest] = (float)(int)mips->r[inst->src1];
			break;
		case IROp::FMovToGPR:
			memcpy(&mips->r[inst->dest], &mips->f[inst->src1], 4);
			break;
		case IROp::OptFMovToGPRShr8:
		{
			u32 temp;
			memcpy(&temp, &mips->f[inst->src1], 4);
			mips->r[inst->dest] = temp >> 8;
			break;
		}

		case IROp::ExitToConst:
			return inst->constant;

		case IROp::ExitToReg:
			return mips->r[inst->src1];

		case IROp::ExitToConstIfEq:
			if (mips->r[inst->src1] == mips->r[inst->src2])
				return inst->constant;
			break;
		case IROp::ExitToConstIfNeq:
			if (mips->r[inst->src1] != mips->r[inst->src2])
				return inst->constant;
			break;
		case IROp::ExitToConstIfGtZ:
			if ((s32)mips->r[inst->src1] > 0)
				return inst->constant;
			break;
		case IROp::ExitToConstIfGeZ:
			if ((s32)mips->r[inst->src1] >= 0)
				return inst->constant;
			break;
		case IROp::ExitToConstIfLtZ:
			if ((s32)mips->r[inst->src1] < 0)
				return inst->constant;
			break;
		case IROp::ExitToConstIfLeZ:
			if ((s32)mips->r[inst->src1] <= 0)
				return inst->constant;
			break;

		case IROp::Downcount:
			mips->downcount -= (int)inst->constant;
			break;

		case IROp::SetPC:
			mips->pc = mips->r[inst->src1];
			break;

		case IROp::SetPCConst:
			mips->pc = inst->constant;
			break;

		case IROp::Syscall:
			// IROp::SetPC was (hopefully) executed before.
		{
			MIPSOpcode op(inst->constant);
			CallSyscall(op);
			if (coreState != CORE_RUNNING)
				CoreTiming::ForceCheck();
			break;
		}

		case IROp::ExitToPC:
			return mips->pc;

		case IROp::Interpret:  // SLOW fallback. Can be made faster. Ideally should be removed but may be useful for debugging.
		{
			MIPSOpcode op(inst->constant);
			MIPSInterpret(op);
			break;
		}

		case IROp::CallReplacement:
		{
			int funcIndex = inst->constant;
			const ReplacementTableEntry *f = GetReplacementFunc(funcIndex);
			int cycles = f->replaceFunc();
			mips->r[inst->dest] = cycles < 0 ? -1 : 0;
			mips->downcount -= cycles < 0 ? -cycles : cycles;
			break;
		}

		case IROp::SetCtrlVFPU:
			mips->vfpuCtrl[inst->dest] = inst->constant;
			break;

		case IROp::SetCtrlVFPUReg:
			mips->vfpuCtrl[inst->dest] = mips->r[inst->src1];
			break;

		case IROp::SetCtrlVFPUFReg:
			memcpy(&mips->vfpuCtrl[inst->dest], &mips->f[inst->src1], 4);
			break;

		case IROp::ApplyRoundingMode:
			IRApplyRounding(mips);
			break;
		case IROp::RestoreRoundingMode:
			IRRestoreRounding();
			break;
		case IROp::UpdateRoundingMode:
			// TODO: Implement
			break;

		case IROp::Break:
			Core_Break(mips->pc);
			return mips->pc + 4;

		case IROp::Breakpoint:
			if (IRRunBreakpoint(inst->constant)) {
				CoreTiming::ForceCheck();
				return mips->pc;
			}
			break;

		case IROp::MemoryCheck:
			if (IRRunMemCheck(mips->pc + inst->dest, mips->r[inst->src1] + inst->constant)) {
				CoreTiming::ForceCheck();
				return mips->pc;
			}
			break;

		case IROp::ValidateAddress8:
			if (RunValidateAddress<1>(mips->pc, mips->r[inst->src1] + inst->constant, inst->src2)) {
				CoreTiming::ForceCheck();
				return mips->pc;
			}
			break;
		case IROp::ValidateAddress16:
			if (RunValidateAddress<2>(mips->pc, mips->r[inst->src1] + inst->constant, inst->src2)) {
				CoreTiming::ForceCheck();
				return mips->pc;
			}
			break;
		case IROp::ValidateAddress32:
			if (RunValidateAddress<4>(mips->pc, mips->r[inst->src1] + inst->constant, inst->src2)) {
				CoreTiming::ForceCheck();
				return mips->pc;
			}
			break;
		case IROp::ValidateAddress128:
			if (RunValidateAddress<16>(mips->pc, mips->r[inst->src1] + inst->constant, inst->src2)) {
				CoreTiming::ForceCheck();
				return mips->pc;
			}
			break;
		case IROp::LogIRBlock:
			if (mipsTracer.tracing_enabled) {
				mipsTracer.executed_blocks.push_back(inst->constant);
			}
			break;

		case IROp::Nop: // TODO: This shouldn't crash, but for now we should not emit nops, so...
		case IROp::Bad:
		default:
			Crash();
			break;
			// Unimplemented IR op. Bad.
		}

#ifdef _DEBUG
		if (mips->r[0] != 0)
			Crash();
#endif
		inst++;
	}

	// We should not reach here anymore.
	return 0;
}