LZ4LLB Decoder (BJX2 ASM)

/*
Decoder for LZ4LLB (LZ4 variant with lengths limited to a single byte), as implemented in BJX2 ASM.
Relatively little bundling in this code, as for a lot of the cases where bundling could be used, interlock delays would have eaten any gains. Still, further improvements could be possible...
 */

byte *TKPE_UnpackL6(byte *ct, byte *ibuf, int isz);

__asm {
TKPE_UnpackL6:
	WEXMD	2
	ADD		R5, R6, R7
	.L0:

	/* Fetch Token Byte */
	MOVU.B	(R5, 0), R16
	MOVU.B	(R5, 1), R17
	ADD		1, R5
	SHLD	R16, -4, R18	|	AND		R16, 15, R19
	/* Fetch Extended Length */
	CMPEQ	15, R18
	ADD?T	R17, R18		|	ADD?T	1, R5
	ADD		R5, R18, R22
	MOV		R4,	R20			|	MOV		R5,	R21
	ADD		R18, R4			|	ADD		R18, R5
	/* Copy Literal Bytes */
	.L1:
	MOV.Q	(R21, 0), R16
	MOV.Q	(R21, 8), R17
	ADD		16, R21
	MOV.Q	R16, (R20, 0)
	MOV.Q	R17, (R20, 8)
	ADD		16, R20
	CMPGT	R21, R22
	BT		.L1

	MOVU.W	(R5, 0), R18
	MOVU.B	(R5, 2), R17
	ADD		R5, 1, R23
	ADD		2, R5

	/* Hit end of buffer yet? */
	CMPGT	R23, R7
	BF		.Done
	CMPEQ	15, R19
	ADD?T	1, R5		| ADD?T	R17, R19
	ADD		4, R19

	/* Zero-Distance Cases? */
	TEST	R18, R18
	BF		.Copy
	CMPEQ	5, R19
	BT		.Cont
	BRA		.Done

	.Cont:
	CMPGT	R5, R7
	BT		.L0

	.Done:
	MOV		R4, R2
	RTS

/* Match Copy Stuff */

	/* R4=Dest, R5=Source, R18=Dist, R19=Length */
	.Copy:
	ADD		R4, R19, R22	| SUB		R4, R18, R21
	MOV		R4,	R20
	ADD		R19, R4
	CMPGE	8, R18
	BF		.CopyRLE
	CMPGE	16, R18
	BF		.CopyL1S

	.CopyL1:
	MOV.Q	(R21, 0), R16
	MOV.Q	(R21, 8), R17
	ADD		16, R21
	MOV.Q	R16, (R20, 0)
	MOV.Q	R17, (R20, 8)
	ADD		16, R20
	CMPGT	R20, R22
	BT		.CopyL1
	BRA		.Cont

	.CopyL1S:
	MOV.Q	(R21, 0), R16
	MOV.Q	R16, (R20, 0)
	MOV.Q	(R21, 8), R17
	MOV.Q	R17, (R20, 8)
	ADD		16, R21
	ADD		16, R20
	CMPGT	R20, R22
	BT		.CopyL1S
	BRA		.Cont

	.CopyRLE:

	CMPEQ	1, R18
	BT		.CopyRLE1
	CMPEQ	2, R18
	BT		.CopyRLE2
	CMPEQ	4, R18
	BT		.CopyRLE4
	CMPEQ	3, R18
	BT		.CopyRLE3

	MOV.Q		(R21), R16
	.CopyRLE_L1:
	MOV.Q	R16, (R20, 0)
	ADD		R18, R20
	CMPGT	R20, R22
	BT		.CopyRLE_L1
	BRA		.Cont

	.CopyRLE1:
	MOVU.B		(R21), R16
	PSHUF.B		R16, 0x00, R16
	PSHUF.W		R16, 0x00, R16
	BRA			.CopyRLEN_L1
	.CopyRLE2:
	MOVU.W		(R21), R16
	PSHUF.W		R16, 0x00, R16
	BRA			.CopyRLEN_L1
	.CopyRLE4:
	MOVU.L		(R21), R16
	MOVLLD		R16, R16, R16
	BRA			.CopyRLEN_L1

	.CopyRLE3:
	MOVU.L		(R21), R16
	MOV			6, R18
	SHLD		R16, 8, R3		|	SHLD		R16, 24, R17
	SHLD		R3, -8, R16
	OR			R17, R16
	BRA			.CopyRLE_L1

	NOP

	.CopyRLEN_L1:
	MOV.Q	R16, (R20, 0)
	MOV.Q	R16, (R20, 8)
	ADD		16, R20
	CMPGT	R20, R22
	BT		.CopyRLEN_L1
	BRA		.Cont
};