ppsspp/GPU/Common/VertexDecoderArm.cpp

// Copyright (c) 2013- PPSSPP Project.

// This program is free software: you can redistribute it and/or modify
// it under the terms of the GNU General Public License as published by
// the Free Software Foundation, version 2.0 or later versions.

// This program is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
// GNU General Public License 2.0 for more details.

// A copy of the GPL 2.0 should have been included with the program.
// If not, see http://www.gnu.org/licenses/

// Official git repository and contact information can be found at
// https://github.com/hrydgard/ppsspp and http://www.ppsspp.org/.

#include "ppsspp_config.h"

#if PPSSPP_ARCH(ARM)

// This allows highlighting to work.  Yay.
#ifdef __INTELLISENSE__
#define ARM
#endif

#include <stddef.h>

#include "Common/CPUDetect.h"
#include "Core/Config.h"
#include "Core/Reporting.h"
#include "GPU/GPUState.h"
#include "GPU/Common/VertexDecoderCommon.h"

extern void DisassembleArm(const u8 *data, int size);

alignas(16) static float bones[16 * 8];  // First two are kept in registers
alignas(16) static float boneMask[4] = {1.0f, 1.0f, 1.0f, 0.0f};

// NEON register allocation:
// Q0: Texture scaling parameters
// Q1: Temp storage
// Q2: Vector-by-matrix accumulator
// Q3: Unused (multiplier temp when morphing)
//
// When skinning, we'll use Q4-Q7 as the "matrix accumulator".
// First two matrices will be preloaded into Q8-Q11 and Q12-Q15 to reduce
// memory bandwidth requirements.
// The rest will be dumped to bones as on x86.
//
// When morphing, we never skin.  So we're free to use Q4+.
// Q4 is for color shift values, and Q5 is a secondary multipler inside the morph.
// TODO: Maybe load all morph weights to Q6+ to avoid memory access?

static const float by128 = 1.0f / 128.0f;
static const float by16384 = 1.0f / 16384.0f;
static const float by32768 = 1.0f / 32768.0f;

using namespace ArmGen;

// NOTE: Avoid R9, it's dangerous on iOS.
//
// r0-r3: parameters
// r4-r11: local vars. save, except R9.
// r12: interprocedure scratch
// r13: stack8

static const ARMReg tempReg1 = R3;
static const ARMReg tempReg2 = R4;
static const ARMReg tempReg3 = R5;
static const ARMReg scratchReg = R6;
static const ARMReg scratchReg2 = R7;
static const ARMReg scratchReg3 = R8;
static const ARMReg fullAlphaReg = R12;
static const ARMReg srcReg = R0;
static const ARMReg dstReg = R1;
static const ARMReg counterReg = R2;
static const ARMReg fpScratchReg = S4;
static const ARMReg fpScratchReg2 = S5;
static const ARMReg fpScratchReg3 = S6;
static const ARMReg fpScratchReg4 = S7;
static const ARMReg fpUscaleReg = S0;
static const ARMReg fpVscaleReg = S1;
static const ARMReg fpUoffsetReg = S2;
static const ARMReg fpVoffsetReg = S3;

// Simpler aliases for NEON. Overlaps with corresponding VFP regs.
static const ARMReg neonUVScaleReg = D0;
static const ARMReg neonUVOffsetReg = D1;
static const ARMReg neonScratchReg = D2;
static const ARMReg neonScratchReg2 = D3;
static const ARMReg neonScratchRegQ = Q1;  // Overlaps with all the scratch regs

// Everything above S6 is fair game for skinning

// S8-S15 are used during matrix generation

// These only live through the matrix multiplication
static const ARMReg src[3] = {S8, S9, S10};  // skin source
static const ARMReg acc[3] = {S11, S12, S13};  // skin accumulator

static const ARMReg srcNEON = Q2;
static const ARMReg accNEON = Q3;

static const JitLookup jitLookup[] = {
	{&VertexDecoder::Step_WeightsU8, &VertexDecoderJitCache::Jit_WeightsU8},
	{&VertexDecoder::Step_WeightsU16, &VertexDecoderJitCache::Jit_WeightsU16},
	{&VertexDecoder::Step_WeightsFloat, &VertexDecoderJitCache::Jit_WeightsFloat},
	{&VertexDecoder::Step_WeightsU8Skin, &VertexDecoderJitCache::Jit_WeightsU8Skin},
	{&VertexDecoder::Step_WeightsU16Skin, &VertexDecoderJitCache::Jit_WeightsU16Skin},
	{&VertexDecoder::Step_WeightsFloatSkin, &VertexDecoderJitCache::Jit_WeightsFloatSkin},

	{&VertexDecoder::Step_TcFloat, &VertexDecoderJitCache::Jit_TcFloat},
	{&VertexDecoder::Step_TcU8ToFloat, &VertexDecoderJitCache::Jit_TcU8ToFloat},
	{&VertexDecoder::Step_TcU16ToFloat, &VertexDecoderJitCache::Jit_TcU16ToFloat},

	{&VertexDecoder::Step_TcU8Prescale, &VertexDecoderJitCache::Jit_TcU8Prescale},
	{&VertexDecoder::Step_TcU16Prescale, &VertexDecoderJitCache::Jit_TcU16Prescale},
	{&VertexDecoder::Step_TcFloatPrescale, &VertexDecoderJitCache::Jit_TcFloatPrescale},

	{&VertexDecoder::Step_TcFloatThrough, &VertexDecoderJitCache::Jit_TcFloatThrough},
	{&VertexDecoder::Step_TcU16ThroughToFloat, &VertexDecoderJitCache::Jit_TcU16ThroughToFloat},

	{&VertexDecoder::Step_NormalS8, &VertexDecoderJitCache::Jit_NormalS8},
	{&VertexDecoder::Step_NormalS16, &VertexDecoderJitCache::Jit_NormalS16},
	{&VertexDecoder::Step_NormalFloat, &VertexDecoderJitCache::Jit_NormalFloat},

	{&VertexDecoder::Step_NormalS8Skin, &VertexDecoderJitCache::Jit_NormalS8Skin},
	{&VertexDecoder::Step_NormalS16Skin, &VertexDecoderJitCache::Jit_NormalS16Skin},
	{&VertexDecoder::Step_NormalFloatSkin, &VertexDecoderJitCache::Jit_NormalFloatSkin},

	{&VertexDecoder::Step_Color8888, &VertexDecoderJitCache::Jit_Color8888},
	{&VertexDecoder::Step_Color4444, &VertexDecoderJitCache::Jit_Color4444},
	{&VertexDecoder::Step_Color565, &VertexDecoderJitCache::Jit_Color565},
	{&VertexDecoder::Step_Color5551, &VertexDecoderJitCache::Jit_Color5551},

	{&VertexDecoder::Step_PosS8Through, &VertexDecoderJitCache::Jit_PosS8Through},
	{&VertexDecoder::Step_PosS16Through, &VertexDecoderJitCache::Jit_PosS16Through},
	{&VertexDecoder::Step_PosFloatThrough, &VertexDecoderJitCache::Jit_PosFloat},

	{&VertexDecoder::Step_PosS8, &VertexDecoderJitCache::Jit_PosS8},
	{&VertexDecoder::Step_PosS16, &VertexDecoderJitCache::Jit_PosS16},
	{&VertexDecoder::Step_PosFloat, &VertexDecoderJitCache::Jit_PosFloat},

	{&VertexDecoder::Step_PosS8Skin, &VertexDecoderJitCache::Jit_PosS8Skin},
	{&VertexDecoder::Step_PosS16Skin, &VertexDecoderJitCache::Jit_PosS16Skin},
	{&VertexDecoder::Step_PosFloatSkin, &VertexDecoderJitCache::Jit_PosFloatSkin},

	{&VertexDecoder::Step_NormalS8Morph, &VertexDecoderJitCache::Jit_NormalS8Morph},
	{&VertexDecoder::Step_NormalS16Morph, &VertexDecoderJitCache::Jit_NormalS16Morph},
	{&VertexDecoder::Step_NormalFloatMorph, &VertexDecoderJitCache::Jit_NormalFloatMorph},

	{&VertexDecoder::Step_PosS8Morph, &VertexDecoderJitCache::Jit_PosS8Morph},
	{&VertexDecoder::Step_PosS16Morph, &VertexDecoderJitCache::Jit_PosS16Morph},
	{&VertexDecoder::Step_PosFloatMorph, &VertexDecoderJitCache::Jit_PosFloatMorph},

	{&VertexDecoder::Step_Color8888Morph, &VertexDecoderJitCache::Jit_Color8888Morph},
	{&VertexDecoder::Step_Color4444Morph, &VertexDecoderJitCache::Jit_Color4444Morph},
	{&VertexDecoder::Step_Color565Morph, &VertexDecoderJitCache::Jit_Color565Morph},
	{&VertexDecoder::Step_Color5551Morph, &VertexDecoderJitCache::Jit_Color5551Morph},
};

JittedVertexDecoder VertexDecoderJitCache::Compile(const VertexDecoder &dec, int32_t *jittedSize) {
	dec_ = &dec;
	BeginWrite();
	const u8 *start = AlignCode16();

	bool prescaleStep = false;
	bool skinning = false;

	// Look for prescaled texcoord steps
	for (int i = 0; i < dec.numSteps_; i++) {
		if (dec.steps_[i] == &VertexDecoder::Step_TcU8Prescale ||
			dec.steps_[i] == &VertexDecoder::Step_TcU16Prescale ||
			dec.steps_[i] == &VertexDecoder::Step_TcFloatPrescale) {
				prescaleStep = true;
		}
		if (dec.steps_[i] == &VertexDecoder::Step_WeightsU8Skin ||
			dec.steps_[i] == &VertexDecoder::Step_WeightsU16Skin ||
			dec.steps_[i] == &VertexDecoder::Step_WeightsFloatSkin) {
				skinning = true;
		}
	}

	// Not used below, but useful for logging.
	(void)skinning;

	SetCC(CC_AL);

	PUSH(8, R4, R5, R6, R7, R8, R10, R11, R_LR);
	VPUSH(D8, 8);

	// Keep the scale/offset in a few fp registers if we need it.
	if (prescaleStep) {
		MOVP2R(R3, &gstate_c.uv);
		VLD1(F_32, neonUVScaleReg, R3, 2, ALIGN_NONE);
		if ((dec.VertexType() & GE_VTYPE_TC_MASK) == GE_VTYPE_TC_8BIT) {
			VMOV_neon(F_32, neonScratchReg, by128);
			VMUL(F_32, neonUVScaleReg, neonUVScaleReg, neonScratchReg);
		} else if ((dec.VertexType() & GE_VTYPE_TC_MASK) == GE_VTYPE_TC_16BIT) {
			VMOV_neon(F_32, neonScratchReg, by32768);
			VMUL(F_32, neonUVScaleReg, neonUVScaleReg, neonScratchReg);
		}
	}

	// Add code to convert matrices to 4x4.
	// Later we might want to do this when the matrices are loaded instead.
	if (dec.weighttype && g_Config.bSoftwareSkinning) {
		// Copying from R3 to R4
		MOVP2R(R3, gstate.boneMatrix);
		MOVP2R(R4, bones);
		MOVP2R(R5, boneMask);
		VLD1(F_32, Q3, R5, 2, ALIGN_128);
		for (int i = 0; i < dec.nweights; i++) {
			VLD1(F_32, Q4, R3, 2);  // Load 128 bits even though we just want 96
			VMUL(F_32, Q4, Q4, Q3);
			ADD(R3, R3, 12);
			VLD1(F_32, Q5, R3, 2);
			VMUL(F_32, Q5, Q5, Q3);
			ADD(R3, R3, 12);
			VLD1(F_32, Q6, R3, 2);
			VMUL(F_32, Q6, Q6, Q3);
			ADD(R3, R3, 12);
			VLD1(F_32, Q7, R3, 2);
			VMUL(F_32, Q7, Q7, Q3);
			ADD(R3, R3, 12);
			// First two matrices are in registers.
			if (i == 0) {
				VMOV(Q8, Q4);
				VMOV(Q9, Q5);
				VMOV(Q10, Q6);
				VMOV(Q11, Q7);
				ADD(R4, R4, 16 * 4);
			} else if (i == 1) {
				VMOV(Q12, Q4);
				VMOV(Q13, Q5);
				VMOV(Q14, Q6);
				VMOV(Q15, Q7);
				ADD(R4, R4, 16 * 4);
			} else {
				VST1(F_32, Q4, R4, 2, ALIGN_128, REG_UPDATE);
				VST1(F_32, Q5, R4, 2, ALIGN_128, REG_UPDATE);
				VST1(F_32, Q6, R4, 2, ALIGN_128, REG_UPDATE);
				VST1(F_32, Q7, R4, 2, ALIGN_128, REG_UPDATE);
			}
		}
	}

	if (dec.col) {
		// Or LDB and skip the conditional?  This is probably cheaper.
		MOV(fullAlphaReg, 0xFF);
	}

	JumpTarget loopStart = GetCodePtr();
	// Preload data cache ahead of reading. This offset seems pretty good.
	PLD(srcReg, 64);
	for (int i = 0; i < dec.numSteps_; i++) {
		if (!CompileStep(dec, i)) {
			EndWrite();
			// Reset the code ptr and return zero to indicate that we failed.
			ResetCodePtr(GetOffset(start));
			char temp[1024] = {0};
			dec.ToString(temp);
			INFO_LOG(G3D, "Could not compile vertex decoder: %s", temp);
			return 0;
		}
	}

	ADDI2R(srcReg, srcReg, dec.VertexSize(), scratchReg);
	ADDI2R(dstReg, dstReg, dec.decFmt.stride, scratchReg);
	SUBS(counterReg, counterReg, 1);
	B_CC(CC_NEQ, loopStart);

	if (dec.col) {
		MOVP2R(tempReg1, &gstate_c.vertexFullAlpha);
		CMP(fullAlphaReg, 0);
		SetCC(CC_EQ);
		STRB(fullAlphaReg, tempReg1, 0);
		SetCC(CC_AL);
	}

	VPOP(D8, 8);
	POP(8, R4, R5, R6, R7, R8, R10, R11, R_PC);

	FlushLitPool();
	FlushIcache();

	/*
	DisassembleArm(start, GetCodePtr() - start);
	char temp[1024] = {0};
	dec.ToString(temp);
	INFO_LOG(G3D, "%s", temp);
	*/

	*jittedSize = GetCodePtr() - start;
	EndWrite();
	return (JittedVertexDecoder)start;
}

void VertexDecoderJitCache::Jit_WeightsU8() {
	// Basic implementation - a byte at a time. TODO: Optimize
	int j;
	for (j = 0; j < dec_->nweights; j++) {
		LDRB(tempReg1, srcReg, dec_->weightoff + j);
		STRB(tempReg1, dstReg, dec_->decFmt.w0off + j);
	}
	if (j & 3) {
		// Create a zero register. Might want to make a fixed one.
		EOR(scratchReg, scratchReg, scratchReg);
	}
	while (j & 3) {
		STRB(scratchReg, dstReg, dec_->decFmt.w0off + j);
		j++;
	}
}

void VertexDecoderJitCache::Jit_WeightsU16() {
	// Basic implementation - a short at a time. TODO: Optimize
	int j;
	for (j = 0; j < dec_->nweights; j++) {
		LDRH(tempReg1, srcReg, dec_->weightoff + j * 2);
		STRH(tempReg1, dstReg, dec_->decFmt.w0off + j * 2);
	}
	if (j & 3) {
		// Create a zero register. Might want to make a fixed one.
		EOR(scratchReg, scratchReg, scratchReg);
	}
	while (j & 3) {
		STRH(scratchReg, dstReg, dec_->decFmt.w0off + j * 2);
		j++;
	}
}

void VertexDecoderJitCache::Jit_WeightsFloat() {
	int j;
	for (j = 0; j < dec_->nweights; j++) {
		LDR(tempReg1, srcReg, dec_->weightoff + j * 4);
		STR(tempReg1, dstReg, dec_->decFmt.w0off + j * 4);
	}
	if (j & 3) {
		EOR(tempReg1, tempReg1, tempReg1);
	}
	while (j & 3) {  // Zero additional weights rounding up to 4.
		STR(tempReg1, dstReg, dec_->decFmt.w0off + j * 4);
		j++;
	}
}

static const ARMReg weightRegs[8] = { S8, S9, S10, S11, S12, S13, S14, S15 };
static const ARMReg neonWeightRegsD[4] = { D4, D5, D6, D7 };
static const ARMReg neonWeightRegsQ[2] = { Q2, Q3 };

void VertexDecoderJitCache::Jit_ApplyWeights() {
	// We construct a matrix in Q4-Q7
	// We can use Q1 as temp.
	if (dec_->nweights >= 2) {
		MOVP2R(scratchReg, bones + 16 * 2);
	}
	for (int i = 0; i < dec_->nweights; i++) {
		switch (i) {
		case 0:
			VMUL_scalar(F_32, Q4, Q8, QScalar(neonWeightRegsQ[0], 0));
			VMUL_scalar(F_32, Q5, Q9, QScalar(neonWeightRegsQ[0], 0));
			VMUL_scalar(F_32, Q6, Q10, QScalar(neonWeightRegsQ[0], 0));
			VMUL_scalar(F_32, Q7, Q11, QScalar(neonWeightRegsQ[0], 0));
			break;
		case 1:
			// Krait likes VDUP + VFMA better than VMLA, and it's easy to do here.
			if (cpu_info.bVFPv4) {
				VDUP(F_32, Q1, neonWeightRegsQ[i >> 2], i & 1);
				VFMA(F_32, Q4, Q12, Q1);
				VFMA(F_32, Q5, Q13, Q1);
				VFMA(F_32, Q6, Q14, Q1);
				VFMA(F_32, Q7, Q15, Q1);
			} else {
				VMLA_scalar(F_32, Q4, Q12, QScalar(neonWeightRegsQ[0], 1));
				VMLA_scalar(F_32, Q5, Q13, QScalar(neonWeightRegsQ[0], 1));
				VMLA_scalar(F_32, Q6, Q14, QScalar(neonWeightRegsQ[0], 1));
				VMLA_scalar(F_32, Q7, Q15, QScalar(neonWeightRegsQ[0], 1));
			}
			break;
		default:
			// Matrices 2+ need to be loaded from memory.
			// Wonder if we can free up one more register so we could get some parallelism.
			// Actually Q3 is free if there are fewer than 5 weights...
			if (dec_->nweights <= 4) {
				VLD1(F_32, Q1, scratchReg, 2, ALIGN_128, REG_UPDATE);
				VLD1(F_32, Q3, scratchReg, 2, ALIGN_128, REG_UPDATE);
				VMLA_scalar(F_32, Q4, Q1, QScalar(neonWeightRegsQ[i >> 2], i & 3));
				VMLA_scalar(F_32, Q5, Q3, QScalar(neonWeightRegsQ[i >> 2], i & 3));
				VLD1(F_32, Q1, scratchReg, 2, ALIGN_128, REG_UPDATE);
				VLD1(F_32, Q3, scratchReg, 2, ALIGN_128, REG_UPDATE);
				VMLA_scalar(F_32, Q6, Q1, QScalar(neonWeightRegsQ[i >> 2], i & 3));
				VMLA_scalar(F_32, Q7, Q3, QScalar(neonWeightRegsQ[i >> 2], i & 3));
			} else {
				VLD1(F_32, Q1, scratchReg, 2, ALIGN_128, REG_UPDATE);
				VMLA_scalar(F_32, Q4, Q1, QScalar(neonWeightRegsQ[i >> 2], i & 3));
				VLD1(F_32, Q1, scratchReg, 2, ALIGN_128, REG_UPDATE);
				VMLA_scalar(F_32, Q5, Q1, QScalar(neonWeightRegsQ[i >> 2], i & 3));
				VLD1(F_32, Q1, scratchReg, 2, ALIGN_128, REG_UPDATE);
				VMLA_scalar(F_32, Q6, Q1, QScalar(neonWeightRegsQ[i >> 2], i & 3));
				VLD1(F_32, Q1, scratchReg, 2, ALIGN_128, REG_UPDATE);
				VMLA_scalar(F_32, Q7, Q1, QScalar(neonWeightRegsQ[i >> 2], i & 3));
			}
			break;
		}
	}
}

void VertexDecoderJitCache::Jit_WeightsU8Skin() {
	// Weight is first so srcReg is correct.
	switch (dec_->nweights) {
	case 1: VLD1_lane(I_8, neonScratchReg, srcReg, 0, false); break;
	case 2: VLD1_lane(I_16, neonScratchReg, srcReg, 0, false); break;
	default:
		// For 3, we over read, for over 4, we read more later.
		VLD1_lane(I_32, neonScratchReg, srcReg, 0, false);
		break;
	}
	// This can be represented as a constant.
	VMOV_neon(F_32, Q3, by128);
	VMOVL(I_8 | I_UNSIGNED, neonScratchRegQ, neonScratchReg);
	VMOVL(I_16 | I_UNSIGNED, neonScratchRegQ, neonScratchReg);
	VCVT(F_32 | I_UNSIGNED, neonScratchRegQ, neonScratchRegQ);
	VMUL(F_32, neonWeightRegsQ[0], neonScratchRegQ, Q3);

	if (dec_->nweights > 4) {
		ADD(tempReg1, srcReg, 4 * sizeof(u8));
		switch (dec_->nweights) {
		case 5: VLD1_lane(I_8, neonScratchReg, tempReg1, 0, false); break;
		case 6: VLD1_lane(I_16, neonScratchReg, tempReg1, 0, false); break;
		case 7:
		case 8:
			VLD1_lane(I_32, neonScratchReg, tempReg1, 0, false);
			break;
		}
		VMOVL(I_8 | I_UNSIGNED, neonScratchRegQ, neonScratchReg);
		VMOVL(I_16 | I_UNSIGNED, neonScratchRegQ, neonScratchReg);
		VCVT(F_32 | I_UNSIGNED, neonScratchRegQ, neonScratchRegQ);
		VMUL(F_32, neonWeightRegsQ[1], neonScratchRegQ, Q3);
	}
	Jit_ApplyWeights();
}

void VertexDecoderJitCache::Jit_WeightsU16Skin() {
	switch (dec_->nweights) {
	case 1: VLD1_lane(I_16, neonScratchReg, srcReg, 0, true); break;
	case 2: VLD1_lane(I_32, neonScratchReg, srcReg, 0, false); break;
	default:
		// For 3, we over read, for over 4, we read more later.
		VLD1(I_32, neonScratchReg, srcReg, 1, ALIGN_NONE);
		break;
	}
	// This can be represented as a constant.
	VMOV_neon(F_32, Q3, by32768);
	VMOVL(I_16 | I_UNSIGNED, neonScratchRegQ, neonScratchReg);
	VCVT(F_32 | I_UNSIGNED, neonScratchRegQ, neonScratchRegQ);
	VMUL(F_32, neonWeightRegsQ[0], neonScratchRegQ, Q3);

	if (dec_->nweights > 4) {
		ADD(tempReg1, srcReg, 4 * sizeof(u16));
		switch (dec_->nweights) {
		case 5: VLD1_lane(I_16, neonScratchReg, tempReg1, 0, true); break;
		case 6: VLD1_lane(I_32, neonScratchReg, tempReg1, 0, false); break;
		case 7:
		case 8:
			VLD1(I_32, neonScratchReg, tempReg1, 1, ALIGN_NONE);
			break;
		}
		VMOVL(I_16 | I_UNSIGNED, neonScratchRegQ, neonScratchReg);
		VCVT(F_32 | I_UNSIGNED, neonScratchRegQ, neonScratchRegQ);
		VMUL(F_32, neonWeightRegsQ[1], neonScratchRegQ, Q3);
	}
	Jit_ApplyWeights();
}

void VertexDecoderJitCache::Jit_WeightsFloatSkin() {
	for (int i = 1; i < dec_->nweights; ++i) {
		_dbg_assert_msg_(weightRegs[i - 1] + 1 == weightRegs[i], "VertexDecoder weightRegs must be in order.");
	}

	// Weights are always first, so we can use srcReg directly.
	if (dec_->nweights == 1) {
		VLD1_lane(F_32, neonWeightRegsD[0], srcReg, 0, true);
	} else {
		// We may over-read by one float but this is not a tragedy.
		VLD1(F_32, neonWeightRegsD[0], srcReg, (dec_->nweights + 1) / 2);
	}
	Jit_ApplyWeights();
}

void VertexDecoderJitCache::Jit_TcFloat() {
	LDR(tempReg1, srcReg, dec_->tcoff);
	LDR(tempReg2, srcReg, dec_->tcoff + 4);
	STR(tempReg1, dstReg, dec_->decFmt.uvoff);
	STR(tempReg2, dstReg, dec_->decFmt.uvoff + 4);
}

void VertexDecoderJitCache::Jit_TcU16ThroughToFloat() {
	LDRH(tempReg1, srcReg, dec_->tcoff);
	LDRH(tempReg2, srcReg, dec_->tcoff + 2);

	MOVP2R(scratchReg, &gstate_c.vertBounds.minU);

	auto updateSide = [&](ARMReg r, CCFlags cc, u32 off) {
		LDRH(tempReg3, scratchReg, off);
		CMP(r, tempReg3);
		SetCC(cc);
		STRH(r, scratchReg, off);
		SetCC(CC_AL);
	};

	// TODO: Can this actually be fast?  Hmm, floats aren't better.
	updateSide(tempReg1, CC_LT, offsetof(KnownVertexBounds, minU));
	updateSide(tempReg1, CC_GT, offsetof(KnownVertexBounds, maxU));
	updateSide(tempReg2, CC_LT, offsetof(KnownVertexBounds, minV));
	updateSide(tempReg2, CC_GT, offsetof(KnownVertexBounds, maxV));

	ADD(scratchReg, srcReg, dec_->tcoff);
	VLD1_lane(I_32, neonScratchReg, scratchReg, 0, false);
	VMOVL(I_16 | I_UNSIGNED, neonScratchRegQ, neonScratchReg);  // Widen to 32-bit
	VCVT(F_32 | I_UNSIGNED, neonScratchRegQ, neonScratchRegQ);
	ADD(scratchReg2, dstReg, dec_->decFmt.uvoff);
	VST1(F_32, neonScratchReg, scratchReg2, 1, ALIGN_NONE);
}

void VertexDecoderJitCache::Jit_TcFloatThrough() {
	LDR(tempReg1, srcReg, dec_->tcoff);
	LDR(tempReg2, srcReg, dec_->tcoff + 4);
	STR(tempReg1, dstReg, dec_->decFmt.uvoff);
	STR(tempReg2, dstReg, dec_->decFmt.uvoff + 4);
}

void VertexDecoderJitCache::Jit_TcU8Prescale() {
	ADD(scratchReg, srcReg, dec_->tcoff);
	VLD1_lane(I_16, neonScratchReg, scratchReg, 0, false);
	VMOVL(I_8 | I_UNSIGNED, neonScratchRegQ, neonScratchReg);  // Widen to 16-bit
	VMOVL(I_16 | I_UNSIGNED, neonScratchRegQ, neonScratchReg);  // Widen to 32-bit
	VCVT(F_32 | I_UNSIGNED, neonScratchRegQ, neonScratchRegQ);
	ADD(scratchReg2, dstReg, dec_->decFmt.uvoff);
	VMUL(F_32, neonScratchReg, neonScratchReg, neonUVScaleReg);
	VADD(F_32, neonScratchReg, neonScratchReg, neonUVOffsetReg);
	VST1(F_32, neonScratchReg, scratchReg2, 1, ALIGN_NONE);
}

void VertexDecoderJitCache::Jit_TcU8ToFloat() {
	ADD(scratchReg, srcReg, dec_->tcoff);
	VLD1_lane(I_16, neonScratchReg, scratchReg, 0, false);
	VMOVL(I_8 | I_UNSIGNED, neonScratchRegQ, neonScratchReg);  // Widen to 16-bit
	VMOVL(I_16 | I_UNSIGNED, neonScratchRegQ, neonScratchReg);  // Widen to 32-bit
	VCVT(F_32 | I_UNSIGNED, neonScratchRegQ, neonScratchRegQ);
	VMOV_neon(F_32, neonScratchReg2, by128);
	VMUL(F_32, neonScratchReg, neonScratchReg, neonScratchReg2);
	ADD(scratchReg2, dstReg, dec_->decFmt.uvoff);
	VST1(F_32, neonScratchReg, scratchReg2, 1, ALIGN_NONE);
}

void VertexDecoderJitCache::Jit_TcU16Prescale() {
	ADD(scratchReg, srcReg, dec_->tcoff);
	VLD1_lane(I_32, neonScratchReg, scratchReg, 0, false);
	VMOVL(I_16 | I_UNSIGNED, neonScratchRegQ, neonScratchReg);  // Widen to 32-bit
	VCVT(F_32 | I_UNSIGNED, neonScratchRegQ, neonScratchRegQ);
	ADD(scratchReg2, dstReg, dec_->decFmt.uvoff);
	VMUL(F_32, neonScratchReg, neonScratchReg, neonUVScaleReg);
	VADD(F_32, neonScratchReg, neonScratchReg, neonUVOffsetReg);
	VST1(F_32, neonScratchReg, scratchReg2, 1, ALIGN_NONE);
}

void VertexDecoderJitCache::Jit_TcU16ToFloat() {
	ADD(scratchReg, srcReg, dec_->tcoff);
	VLD1_lane(I_32, neonScratchReg, scratchReg, 0, false);
	VMOVL(I_16 | I_UNSIGNED, neonScratchRegQ, neonScratchReg);  // Widen to 32-bit
	VCVT(F_32 | I_UNSIGNED, neonScratchRegQ, neonScratchRegQ);
	ADD(scratchReg2, dstReg, dec_->decFmt.uvoff);
	VMOV_neon(F_32, neonScratchReg2, by32768);
	VMUL(F_32, neonScratchReg, neonScratchReg, neonScratchReg2);
	VST1(F_32, neonScratchReg, scratchReg2, 1, ALIGN_NONE);
}

void VertexDecoderJitCache::Jit_TcFloatPrescale() {
	ADD(scratchReg, srcReg, dec_->tcoff);
	VLD1(F_32, neonScratchReg, scratchReg, 1, ALIGN_NONE);
	ADD(scratchReg2, dstReg, dec_->decFmt.uvoff);
	VMUL(F_32, neonScratchReg, neonScratchReg, neonUVScaleReg);
	VADD(F_32, neonScratchReg, neonScratchReg, neonUVOffsetReg);
	VST1(F_32, neonScratchReg, scratchReg2, 1, ALIGN_NONE);
}

void VertexDecoderJitCache::Jit_Color8888() {
	LDR(tempReg1, srcReg, dec_->coloff);
	// Set flags to determine if alpha != 0xFF.
	MVNS(tempReg2, Operand2(tempReg1, ST_ASR, 24));
	STR(tempReg1, dstReg, dec_->decFmt.c0off);
	SetCC(CC_NEQ);
	MOV(fullAlphaReg, 0);
	SetCC(CC_AL);
}

void VertexDecoderJitCache::Jit_Color4444() {
	LDRH(tempReg1, srcReg, dec_->coloff);

	// Spread out the components.
	ANDI2R(tempReg2, tempReg1, 0x000F, scratchReg);
	ANDI2R(tempReg3, tempReg1, 0x00F0, scratchReg);
	ORR(tempReg2, tempReg2, Operand2(tempReg3, ST_LSL, 4));
	ANDI2R(tempReg3, tempReg1, 0x0F00, scratchReg);
	ORR(tempReg2, tempReg2, Operand2(tempReg3, ST_LSL, 8));
	ANDI2R(tempReg3, tempReg1, 0xF000, scratchReg);
	ORR(tempReg2, tempReg2, Operand2(tempReg3, ST_LSL, 12));

	// And expand to 8 bits.
	ORR(tempReg1, tempReg2, Operand2(tempReg2, ST_LSL, 4));

	STR(tempReg1, dstReg, dec_->decFmt.c0off);

	// Set flags to determine if alpha != 0xFF.
	MVNS(tempReg2, Operand2(tempReg1, ST_ASR, 24));
	SetCC(CC_NEQ);
	MOV(fullAlphaReg, 0);
	SetCC(CC_AL);
}

void VertexDecoderJitCache::Jit_Color565() {
	LDRH(tempReg1, srcReg, dec_->coloff);

	// Spread out R and B first.  This puts them in 0x001F001F.
	ANDI2R(tempReg2, tempReg1, 0x001F, scratchReg);
	ANDI2R(tempReg3, tempReg1, 0xF800, scratchReg);
	ORR(tempReg2, tempReg2, Operand2(tempReg3, ST_LSL, 5));

	// Expand 5 -> 8.
	LSL(tempReg3, tempReg2, 3);
	ORR(tempReg2, tempReg3, Operand2(tempReg2, ST_LSR, 2));
	ANDI2R(tempReg2, tempReg2, 0xFFFF00FF, scratchReg);

	// Now finally G.  We start by shoving it into a wall.
	LSR(tempReg1, tempReg1, 5);
	ANDI2R(tempReg1, tempReg1, 0x003F, scratchReg);
	LSL(tempReg3, tempReg1, 2);
	// Don't worry, shifts into a wall.
	ORR(tempReg3, tempReg3, Operand2(tempReg1, ST_LSR, 4));
	ORR(tempReg2, tempReg2, Operand2(tempReg3, ST_LSL, 8));

	// Add in full alpha.  No need to update fullAlphaReg.
	ORI2R(tempReg1, tempReg2, 0xFF000000, scratchReg);

	STR(tempReg1, dstReg, dec_->decFmt.c0off);
}

void VertexDecoderJitCache::Jit_Color5551() {
	LDRSH(tempReg1, srcReg, dec_->coloff);

	ANDI2R(tempReg2, tempReg1, 0x001F, scratchReg);
	ANDI2R(tempReg3, tempReg1, 0x03E0, scratchReg);
	ORR(tempReg2, tempReg2, Operand2(tempReg3, ST_LSL, 3));
	ANDI2R(tempReg3, tempReg1, 0x7C00, scratchReg);
	ORR(tempReg2, tempReg2, Operand2(tempReg3, ST_LSL, 6));

	// Expand 5 -> 8.
	LSR(tempReg3, tempReg2, 2);
	// Clean up the bits that were shifted right.
	BIC(tempReg3, tempReg3, AssumeMakeOperand2(0x000000F8));
	BIC(tempReg3, tempReg3, AssumeMakeOperand2(0x0000F800));
	ORR(tempReg2, tempReg3, Operand2(tempReg2, ST_LSL, 3));

	// Now we just need alpha.  Since we loaded as signed, it'll be extended.
	ANDI2R(tempReg1, tempReg1, 0xFF000000, scratchReg);
	ORR(tempReg2, tempReg2, tempReg1);
	
	// Set flags to determine if alpha != 0xFF.
	MVNS(tempReg3, Operand2(tempReg1, ST_ASR, 24));
	STR(tempReg2, dstReg, dec_->decFmt.c0off);
	SetCC(CC_NEQ);
	MOV(fullAlphaReg, 0);
	SetCC(CC_AL);
}

void VertexDecoderJitCache::Jit_Color8888Morph() {
	ADDI2R(tempReg1, srcReg, dec_->coloff, scratchReg);
	MOVP2R(tempReg2, &gstate_c.morphWeights[0]);

	bool first = true;
	for (int n = 0; n < dec_->morphcount; ++n) {
		VLD1_lane(I_32, neonScratchReg, tempReg1, 0, true);
		VLD1_all_lanes(F_32, Q3, tempReg2, true, REG_UPDATE);

		ADDI2R(tempReg1, tempReg1, dec_->onesize_, scratchReg);
		VMOVL(I_8 | I_UNSIGNED, neonScratchRegQ, neonScratchReg);
		VMOVL(I_16 | I_UNSIGNED, neonScratchRegQ, neonScratchReg);
		VCVT(F_32 | I_UNSIGNED, neonScratchRegQ, neonScratchRegQ);

		if (first) {
			first = false;
			VMUL(F_32, Q2, neonScratchRegQ, Q3);
		} else if (cpu_info.bVFPv4) {
			VFMA(F_32, Q2, neonScratchRegQ, Q3);
		} else {
			VMLA(F_32, Q2, neonScratchRegQ, Q3);
		}
	}

	Jit_WriteMorphColor(dec_->decFmt.c0off);
}

// First is the left shift, second is the right shift (against walls, to get the RGBA values.)
alignas(16) static const s16 color4444Shift[2][4] = {{12, 8, 4, 0}, {-12, -12, -12, -12}};

void VertexDecoderJitCache::Jit_Color4444Morph() {
	ADDI2R(tempReg1, srcReg, dec_->coloff, scratchReg);
	MOVP2R(tempReg2, &gstate_c.morphWeights[0]);

	MOVP2R(scratchReg, color4444Shift);
	MOVI2FR(scratchReg2, 255.0f / 15.0f);
	VDUP(I_32, Q5, scratchReg2);
	VLD1(I_16, D8, scratchReg, 2, ALIGN_128);

	bool first = true;
	for (int n = 0; n < dec_->morphcount; ++n) {
		VLD1_all_lanes(I_16, neonScratchReg, tempReg1, true);
		VLD1_all_lanes(F_32, Q3, tempReg2, true, REG_UPDATE);

		// Shift against walls and then back to get R, G, B, A in each 16-bit lane.
		VSHL(I_16 | I_UNSIGNED, neonScratchReg, neonScratchReg, D8);
		VSHL(I_16 | I_UNSIGNED, neonScratchReg, neonScratchReg, D9);
		ADDI2R(tempReg1, tempReg1, dec_->onesize_, scratchReg);
		VMOVL(I_16 | I_UNSIGNED, neonScratchRegQ, neonScratchReg);
		VCVT(F_32 | I_UNSIGNED, neonScratchRegQ, neonScratchRegQ);

		VMUL(F_32, Q3, Q3, Q5);

		if (first) {
			first = false;
			VMUL(F_32, Q2, neonScratchRegQ, Q3);
		} else if (cpu_info.bVFPv4) {
			VFMA(F_32, Q2, neonScratchRegQ, Q3);
		} else {
			VMLA(F_32, Q2, neonScratchRegQ, Q3);
		}
	}

	Jit_WriteMorphColor(dec_->decFmt.c0off);
}

// First is the left shift, second is the right shift (against walls, to get the RGBA values.)
alignas(16) static const s16 color565Shift[2][4] = {{11, 5, 0, 0}, {-11, -10, -11, 0}};
alignas(16) static const float byColor565[4] = {255.0f / 31.0f, 255.0f / 63.0f, 255.0f / 31.0f, 0.0f};

void VertexDecoderJitCache::Jit_Color565Morph() {
	ADDI2R(tempReg1, srcReg, dec_->coloff, scratchReg);
	MOVP2R(tempReg2, &gstate_c.morphWeights[0]);
	MOVI2FR(tempReg3, 255.0f);

	MOVP2R(scratchReg, color565Shift);
	MOVP2R(scratchReg2, byColor565);
	VLD1(I_16, D8, scratchReg, 2, ALIGN_128);
	VLD1(F_32, D10, scratchReg2, 2, ALIGN_128);

	bool first = true;
	for (int n = 0; n < dec_->morphcount; ++n) {
		VLD1_all_lanes(I_16, neonScratchReg, tempReg1, true);
		VLD1_all_lanes(F_32, Q3, tempReg2, true, REG_UPDATE);

		VSHL(I_16 | I_UNSIGNED, neonScratchReg, neonScratchReg, D8);
		VSHL(I_16 | I_UNSIGNED, neonScratchReg, neonScratchReg, D9);
		ADDI2R(tempReg1, tempReg1, dec_->onesize_, scratchReg);
		VMOVL(I_16 | I_UNSIGNED, neonScratchRegQ, neonScratchReg);
		VCVT(F_32 | I_UNSIGNED, neonScratchRegQ, neonScratchRegQ);

		VMUL(F_32, Q3, Q3, Q5);

		if (first) {
			first = false;
			VMUL(F_32, Q2, neonScratchRegQ, Q3);
		} else if (cpu_info.bVFPv4) {
			VFMA(F_32, Q2, neonScratchRegQ, Q3);
		} else {
			VMLA(F_32, Q2, neonScratchRegQ, Q3);
		}
	}

	// Overwrite A with 255.0f.
	VMOV_neon(F_32, D5, tempReg3, 1);

	Jit_WriteMorphColor(dec_->decFmt.c0off, false);
}

// First is the left shift, second is the right shift (against walls, to get the RGBA values.)
alignas(16) static const s16 color5551Shift[2][4] = {{11, 6, 1, 0}, {-11, -11, -11, -15}};
alignas(16) static const float byColor5551[4] = {255.0f / 31.0f, 255.0f / 31.0f, 255.0f / 31.0f, 255.0f / 1.0f};

void VertexDecoderJitCache::Jit_Color5551Morph() {
	ADDI2R(tempReg1, srcReg, dec_->coloff, scratchReg);
	MOVP2R(tempReg2, &gstate_c.morphWeights[0]);

	MOVP2R(scratchReg, color5551Shift);
	MOVP2R(scratchReg2, byColor5551);
	VLD1(I_16, D8, scratchReg, 2, ALIGN_128);
	VLD1(F_32, D10, scratchReg2, 2, ALIGN_128);

	bool first = true;
	for (int n = 0; n < dec_->morphcount; ++n) {
		VLD1_all_lanes(I_16, neonScratchReg, tempReg1, true);
		VLD1_all_lanes(F_32, Q3, tempReg2, true, REG_UPDATE);

		VSHL(I_16 | I_UNSIGNED, neonScratchReg, neonScratchReg, D8);
		VSHL(I_16 | I_UNSIGNED, neonScratchReg, neonScratchReg, D9);
		ADDI2R(tempReg1, tempReg1, dec_->onesize_, scratchReg);
		VMOVL(I_16 | I_UNSIGNED, neonScratchRegQ, neonScratchReg);
		VCVT(F_32 | I_UNSIGNED, neonScratchRegQ, neonScratchRegQ);

		VMUL(F_32, Q3, Q3, Q5);

		if (first) {
			first = false;
			VMUL(F_32, Q2, neonScratchRegQ, Q3);
		} else if (cpu_info.bVFPv4) {
			VFMA(F_32, Q2, neonScratchRegQ, Q3);
		} else {
			VMLA(F_32, Q2, neonScratchRegQ, Q3);
		}
	}

	Jit_WriteMorphColor(dec_->decFmt.c0off);
}

// Expects RGBA color in S8 - S11, which is Q2.
void VertexDecoderJitCache::Jit_WriteMorphColor(int outOff, bool checkAlpha) {
	ADDI2R(tempReg1, dstReg, outOff, scratchReg);
	VCVT(I_32 | I_UNSIGNED, Q2, Q2);
	VQMOVN(I_32 | I_UNSIGNED, D4, Q2);
	VQMOVN(I_16 | I_UNSIGNED, D4, Q2);
	VST1_lane(I_32, D4, tempReg1, 0, true);
	if (checkAlpha) {
		VMOV_neon(I_32, scratchReg, D4, 0);
	}
	// Set flags to determine if alpha != 0xFF.
	if (checkAlpha) {
		MVNS(tempReg2, Operand2(scratchReg, ST_ASR, 24));
		SetCC(CC_NEQ);
		MOV(fullAlphaReg, 0);
		SetCC(CC_AL);
	}
}

void VertexDecoderJitCache::Jit_NormalS8() {
	LDRB(tempReg1, srcReg, dec_->nrmoff);
	LDRB(tempReg2, srcReg, dec_->nrmoff + 1);
	LDRB(tempReg3, srcReg, dec_->nrmoff + 2);
	ORR(tempReg1, tempReg1, Operand2(tempReg2, ST_LSL, 8));
	ORR(tempReg1, tempReg1, Operand2(tempReg3, ST_LSL, 16));
	STR(tempReg1, dstReg, dec_->decFmt.nrmoff);

	// Copy 3 bytes and then a zero. Might as well copy four.
	// LDR(tempReg1, srcReg, dec_->nrmoff);
	// ANDI2R(tempReg1, tempReg1, 0x00FFFFFF, scratchReg);
	// STR(tempReg1, dstReg, dec_->decFmt.nrmoff);
}

// Copy 6 bytes and then 2 zeroes.
void VertexDecoderJitCache::Jit_NormalS16() {
	LDRH(tempReg1, srcReg, dec_->nrmoff);
	LDRH(tempReg2, srcReg, dec_->nrmoff + 2);
	LDRH(tempReg3, srcReg, dec_->nrmoff + 4);
	ORR(tempReg1, tempReg1, Operand2(tempReg2, ST_LSL, 16));
	STR(tempReg1, dstReg, dec_->decFmt.nrmoff);
	STR(tempReg3, dstReg, dec_->decFmt.nrmoff + 4);
}

void VertexDecoderJitCache::Jit_NormalFloat() {
	ADD(scratchReg, srcReg, dec_->nrmoff);
	LDMIA(scratchReg, false, 3, tempReg1, tempReg2, tempReg3);
	ADD(scratchReg, dstReg, dec_->decFmt.nrmoff);
	STMIA(scratchReg, false, 3, tempReg1, tempReg2, tempReg3);
}

// Through expands into floats, always. Might want to look at changing this.
void VertexDecoderJitCache::Jit_PosS8Through() {
	DEBUG_LOG_REPORT_ONCE(vertexS8Through, G3D, "Using S8 positions in throughmode");
	_dbg_assert_msg_(fpScratchReg + 1 == fpScratchReg2, "VertexDecoder fpScratchRegs must be in order.");
	_dbg_assert_msg_(fpScratchReg2 + 1 == fpScratchReg3, "VertexDecoder fpScratchRegs must be in order.");

	// TODO: SIMD
	LDRSB(tempReg1, srcReg, dec_->posoff);
	LDRSB(tempReg2, srcReg, dec_->posoff + 1);
	LDRSB(tempReg3, srcReg, dec_->posoff + 2);  // signed?
	static const ARMReg tr[3] = { tempReg1, tempReg2, tempReg3 };
	static const ARMReg fr[3] = { fpScratchReg, fpScratchReg2, fpScratchReg3 };
	ADD(scratchReg, dstReg, dec_->decFmt.posoff);
	VMOV(neonScratchReg, tempReg1, tempReg2);
	VMOV(neonScratchReg2, tempReg3, tempReg3);
	VCVT(F_32 | I_SIGNED, neonScratchRegQ, neonScratchRegQ);
	VST1(F_32, neonScratchReg, scratchReg, 2, ALIGN_NONE);
}

// Through expands into floats, always. Might want to look at changing this.
void VertexDecoderJitCache::Jit_PosS16Through() {
	_dbg_assert_msg_(fpScratchReg + 1 == fpScratchReg2, "VertexDecoder fpScratchRegs must be in order.");
	_dbg_assert_msg_(fpScratchReg2 + 1 == fpScratchReg3, "VertexDecoder fpScratchRegs must be in order.");

	LDRSH(tempReg1, srcReg, dec_->posoff);
	LDRSH(tempReg2, srcReg, dec_->posoff + 2);
	LDRH(tempReg3, srcReg, dec_->posoff + 4);
	static const ARMReg tr[3] = { tempReg1, tempReg2, tempReg3 };
	static const ARMReg fr[3] = { fpScratchReg, fpScratchReg2, fpScratchReg3 };
	ADD(scratchReg, dstReg, dec_->decFmt.posoff);
	VMOV(neonScratchReg, tempReg1, tempReg2);
	VMOV(neonScratchReg2, tempReg3, tempReg3);
	VCVT(F_32 | I_SIGNED, neonScratchRegQ, neonScratchRegQ);
	VST1(F_32, neonScratchReg, scratchReg, 2, ALIGN_NONE);
}

void VertexDecoderJitCache::Jit_PosS8() {
	Jit_AnyS8ToFloat(dec_->posoff);

	ADD(scratchReg, dstReg, dec_->decFmt.posoff);
	VST1(F_32, srcNEON, scratchReg, 2);
}

void VertexDecoderJitCache::Jit_PosS16() {
	Jit_AnyS16ToFloat(dec_->posoff);

	ADD(scratchReg, dstReg, dec_->decFmt.posoff);
	VST1(F_32, srcNEON, scratchReg, 2);
}

// Just copy 12 bytes.
void VertexDecoderJitCache::Jit_PosFloat() {
	ADD(scratchReg, srcReg, dec_->posoff);
	LDMIA(scratchReg, false, 3, tempReg1, tempReg2, tempReg3);
	ADD(scratchReg, dstReg, dec_->decFmt.posoff);
	STMIA(scratchReg, false, 3, tempReg1, tempReg2, tempReg3);
}

void VertexDecoderJitCache::Jit_NormalS8Skin() {
	Jit_AnyS8ToFloat(dec_->nrmoff);
	Jit_WriteMatrixMul(dec_->decFmt.nrmoff, false);
}

void VertexDecoderJitCache::Jit_NormalS16Skin() {
	Jit_AnyS16ToFloat(dec_->nrmoff);
	Jit_WriteMatrixMul(dec_->decFmt.nrmoff, false);
}

void VertexDecoderJitCache::Jit_NormalFloatSkin() {
	for (int i = 1; i < 3; ++i) {
		_dbg_assert_msg_(src[i - 1] + 1 == src[i], "VertexDecoder src regs must be in order.");
	}

	ADD(tempReg1, srcReg, dec_->nrmoff);
	VLD1(F_32, srcNEON, tempReg1, 2, ALIGN_NONE);
	Jit_WriteMatrixMul(dec_->decFmt.nrmoff, false);
}

void VertexDecoderJitCache::Jit_WriteMatrixMul(int outOff, bool pos) {
	// Multiply with the matrix sitting in Q4-Q7.
	ADD(scratchReg, dstReg, outOff);
	VMUL_scalar(F_32, accNEON, Q4, QScalar(srcNEON, 0));
	VMLA_scalar(F_32, accNEON, Q5, QScalar(srcNEON, 1));
	VMLA_scalar(F_32, accNEON, Q6, QScalar(srcNEON, 2));
	if (pos) {
		VADD(F_32, accNEON, accNEON, Q7);
	}
	VST1(F_32, accNEON, scratchReg, 2);
}

void VertexDecoderJitCache::Jit_PosS8Skin() {
	Jit_AnyS8ToFloat(dec_->posoff);
	Jit_WriteMatrixMul(dec_->decFmt.posoff, true);
}

void VertexDecoderJitCache::Jit_PosS16Skin() {
	Jit_AnyS16ToFloat(dec_->posoff);
	Jit_WriteMatrixMul(dec_->decFmt.posoff, true);
}

void VertexDecoderJitCache::Jit_PosFloatSkin() {
	for (int i = 1; i < 3; ++i) {
		_dbg_assert_msg_(src[i - 1] + 1 == src[i], "VertexDecoder src regs must be in order.");
	}

	ADD(tempReg1, srcReg, dec_->posoff);
	VLD1(F_32, srcNEON, tempReg1, 2, ALIGN_NONE);
	Jit_WriteMatrixMul(dec_->decFmt.posoff, true);
}

void VertexDecoderJitCache::Jit_AnyS8ToFloat(int srcoff) {
	ADD(scratchReg, srcReg, srcoff);
	VMOV_neon(F_32, Q3, by128);
	VLD1_lane(I_32, neonScratchReg, scratchReg, 0, false);
	VMOVL(I_8 | I_SIGNED, neonScratchRegQ, neonScratchReg);  // Widen to 16-bit
	VMOVL(I_16 | I_SIGNED, neonScratchRegQ, neonScratchReg);  // Widen to 32-bit
	VCVT(F_32 | I_SIGNED, neonScratchRegQ, neonScratchRegQ);
	VMUL(F_32, srcNEON, neonScratchReg, Q3);
}

void VertexDecoderJitCache::Jit_AnyS16ToFloat(int srcoff) {
	ADD(scratchReg, srcReg, srcoff);
	VMOV_neon(F_32, Q3, by32768);
	VLD1(I_32, neonScratchReg, scratchReg, 1, ALIGN_NONE);
	VMOVL(I_16 | I_SIGNED, neonScratchRegQ, neonScratchReg);  // Widen to 32-bit
	VCVT(F_32 | I_SIGNED, neonScratchRegQ, neonScratchRegQ);
	VMUL(F_32, srcNEON, neonScratchReg, Q3);
}

void VertexDecoderJitCache::Jit_AnyS8Morph(int srcoff, int dstoff) {
	ADDI2R(tempReg1, srcReg, srcoff, scratchReg);
	MOVP2R(tempReg2, &gstate_c.morphWeights[0]);

	MOVI2FR(scratchReg2, by128);
	VDUP(I_32, Q5, scratchReg2);

	bool first = true;
	for (int n = 0; n < dec_->morphcount; ++n) {
		VLD1_lane(I_32, neonScratchReg, tempReg1, 0, false);
		VLD1_all_lanes(F_32, Q3, tempReg2, true, REG_UPDATE);

		ADDI2R(tempReg1, tempReg1, dec_->onesize_, scratchReg);
		VMOVL(I_8 | I_SIGNED, neonScratchRegQ, neonScratchReg);
		VMOVL(I_16 | I_SIGNED, neonScratchRegQ, neonScratchReg);
		VCVT(F_32 | I_SIGNED, neonScratchRegQ, neonScratchRegQ);

		VMUL(F_32, Q3, Q3, Q5);

		if (first) {
			first = false;
			VMUL(F_32, Q2, neonScratchRegQ, Q3);
		} else if (cpu_info.bVFPv4) {
			VFMA(F_32, Q2, neonScratchRegQ, Q3);
		} else {
			VMLA(F_32, Q2, neonScratchRegQ, Q3);
		}
	}

	ADDI2R(tempReg1, dstReg, dstoff, scratchReg);
	// TODO: Is it okay that we're over-writing by 4 bytes?  Probably...
	VSTMIA(tempReg1, false, D4, 2);
}

void VertexDecoderJitCache::Jit_AnyS16Morph(int srcoff, int dstoff) {
	ADDI2R(tempReg1, srcReg, srcoff, scratchReg);
	MOVP2R(tempReg2, &gstate_c.morphWeights[0]);

	MOVI2FR(scratchReg, by32768);
	VDUP(I_32, Q5, scratchReg);

	bool first = true;
	for (int n = 0; n < dec_->morphcount; ++n) {
		VLD1(I_32, neonScratchReg, tempReg1, 1, ALIGN_NONE);
		VLD1_all_lanes(F_32, Q3, tempReg2, true, REG_UPDATE);

		ADDI2R(tempReg1, tempReg1, dec_->onesize_, scratchReg);
		VMOVL(I_16 | I_SIGNED, neonScratchRegQ, neonScratchReg);
		VCVT(F_32 | I_SIGNED, neonScratchRegQ, neonScratchRegQ);

		VMUL(F_32, Q3, Q3, Q5);

		if (first) {
			first = false;
			VMUL(F_32, Q2, neonScratchRegQ, Q3);
		} else if (cpu_info.bVFPv4) {
			VFMA(F_32, Q2, neonScratchRegQ, Q3);
		} else {
			VMLA(F_32, Q2, neonScratchRegQ, Q3);
		}
	}

	ADDI2R(tempReg1, dstReg, dstoff, scratchReg);
	// TODO: Is it okay that we're over-writing by 4 bytes?  Probably...
	VSTMIA(tempReg1, false, D4, 2);
}

void VertexDecoderJitCache::Jit_AnyFloatMorph(int srcoff, int dstoff) {
	ADDI2R(tempReg1, srcReg, srcoff, scratchReg);
	MOVP2R(tempReg2, &gstate_c.morphWeights[0]);

	bool first = true;
	for (int n = 0; n < dec_->morphcount; ++n) {
		// Load an extra float to stay in NEON mode.
		VLD1(F_32, neonScratchRegQ, tempReg1, 2, ALIGN_NONE);
		VLD1_all_lanes(F_32, Q3, tempReg2, true, REG_UPDATE);
		ADDI2R(tempReg1, tempReg1, dec_->onesize_, scratchReg);

		if (first) {
			first = false;
			VMUL(F_32, Q2, neonScratchRegQ, Q3);
		} else if (cpu_info.bVFPv4) {
			VFMA(F_32, Q2, neonScratchRegQ, Q3);
		} else {
			VMLA(F_32, Q2, neonScratchRegQ, Q3);
		}
	}

	ADDI2R(tempReg1, dstReg, dstoff, scratchReg);
	// TODO: Is it okay that we're over-writing by 4 bytes?  Probably...
	VSTMIA(tempReg1, false, D4, 2);
}

void VertexDecoderJitCache::Jit_PosS8Morph() {
	Jit_AnyS8Morph(dec_->posoff, dec_->decFmt.posoff);
}

void VertexDecoderJitCache::Jit_PosS16Morph() {
	Jit_AnyS16Morph(dec_->posoff, dec_->decFmt.posoff);
}

void VertexDecoderJitCache::Jit_PosFloatMorph() {
	Jit_AnyFloatMorph(dec_->posoff, dec_->decFmt.posoff);
}

void VertexDecoderJitCache::Jit_NormalS8Morph() {
	Jit_AnyS8Morph(dec_->nrmoff, dec_->decFmt.nrmoff);
}

void VertexDecoderJitCache::Jit_NormalS16Morph() {
	Jit_AnyS16Morph(dec_->nrmoff, dec_->decFmt.nrmoff);
}

void VertexDecoderJitCache::Jit_NormalFloatMorph() {
	Jit_AnyFloatMorph(dec_->nrmoff, dec_->decFmt.nrmoff);
}

bool VertexDecoderJitCache::CompileStep(const VertexDecoder &dec, int step) {
	// See if we find a matching JIT function
	for (size_t i = 0; i < ARRAY_SIZE(jitLookup); i++) {
		if (dec.steps_[step] == jitLookup[i].func) {
			((*this).*jitLookup[i].jitFunc)();
			return true;
		}
	}
	return false;
}

#endif // PPSSPP_ARCH(ARM)