// license that can be found in the LICENSE file.

//go:build gc

#include "textflag.h"

// try to run "vmov.f64 d0, d0" instruction
TEXT ·useVFPv1(SB),NOSPLIT,$0
	WORD $0xeeb00b40	// vmov.f64 d0, d0
	RET

// try to run VFPv3-only "vmov.f64 d0, #112" instruction
TEXT ·useVFPv3(SB),NOSPLIT,$0
	WORD $0xeeb70b00	// vmov.f64 d0, #112
	RET

// try to run ARMv6K (or above) "ldrexd" instruction
TEXT ·useARMv6K(SB),NOSPLIT,$32

	MOVD.W	-8(srcPtr), H0
	VMOV	H0, T0.D[0]
	VMOV	H1, T3.D[0]
ld4:
	TBZ	$2, srcPtrLen, ld2
	MOVW.W	-4(srcPtr), H0
	VEXT	$12, T0.B16, ZERO.B16, T0.B16
	VEXT	$12, T3.B16, ZERO.B16, T3.B16
	VMOV	H0, T0.S[0]
	VMOV	H1, T3.S[0]
ld2:
	TBZ	$1, srcPtrLen, ld1
	MOVH.W	-2(srcPtr), H0
	VEXT	$14, T0.B16, ZERO.B16, T0.B16
	VEXT	$14, T3.B16, ZERO.B16, T3.B16
	VMOV	H0, T0.H[0]
	VMOV	H1, T3.H[0]
ld1:

	MOVD	dec+24(FP), R11
	LDP	rotInvSRows<>(SB), (R0, R1)
	VMOV	R0, V3.D[0]
	VMOV	R1, V3.D[1]
	VEOR	V0.B16, V0.B16, V0.B16 // All zeroes
	MOVW	$1, R13
	TBZ	$1, R8, ks192
	TBNZ	$2, R8, ks256
	LDPW	(R9), (R4, R5)
	LDPW	8(R9), (R6, R7)
	STPW.P	(R4, R5), 8(R10)
	STPW.P	(R6, R7), 8(R10)
	MOVW	$0x1b, R14
ks128Loop:
		VMOV	R7, V2.S[0]
		WORD	$0x4E030042       // TBL V3.B16, [V2.B16], V2.B16

// testvmovs() (r1, r2 uint64)
TEXT ·testvmovs(SB), NOSPLIT, $0-16
	VMOVS   $0x80402010, V1
	VMOV    V1.D[0], R0
	VMOV    V1.D[1], R1
	MOVD    R0, r1+0(FP)
	MOVD    R1, r2+8(FP)
	RET

// testvmovd() (r1, r2 uint64)
TEXT ·testvmovd(SB), NOSPLIT, $0-16
	VMOVD   $0x7040201008040201, V1
	VMOV    V1.D[0], R0
	VMOV    V1.D[1], R1
	MOVD    R0, r1+0(FP)
	MOVD    R1, r2+8(FP)
	RET

// testvmovq() (r1, r2 uint64)

	VEOR V14.B16, V14.B16, V14.B16
	VEOR V15.B16, V15.B16, V15.B16

	// Save seed state for adding back later.
	VMOV V4.B16, V20.B16
	VMOV V5.B16, V21.B16
	VMOV V6.B16, V22.B16
	VMOV V7.B16, V23.B16
	VMOV V8.B16, V24.B16
	VMOV V9.B16, V25.B16
	VMOV V10.B16, V26.B16
	VMOV V11.B16, V27.B16

	// 4 iterations. Each iteration is 8 quarter-rounds.
	MOVD $4, R0
loop:
	QR(V0, V4, V8, V12)

//go:build !purego

#include "textflag.h"

#define HASHUPDATECHOOSE \
	SHA1C	V16.S4, V1, V2 \
	SHA1H	V3, V1 \
	VMOV	V2.B16, V3.B16

#define HASHUPDATEPARITY \
	SHA1P	V16.S4, V1, V2 \
	SHA1H	V3, V1 \
	VMOV	V2.B16, V3.B16

#define HASHUPDATEMAJ \
	SHA1M	V16.S4, V1, V2 \
	SHA1H	V3, V1 \
	VMOV	V2.B16, V3.B16

// func sha1block(h []uint32, p []byte, k []uint32)
TEXT ·sha1block(SB),NOSPLIT,$0

	// which lane matches the requested byte.
	// 0x40100401 = ((1<<0) + (4<<8) + (16<<16) + (64<<24))
	// Different bytes have different bit masks (i.e: 1, 4, 16, 64)
	MOVD	$0x40100401, R5
	VMOV	R1, V0.B16
	// Work with aligned 32-byte chunks
	BIC	$0x1f, R0, R3
	VMOV	R5, V5.S4
	ANDS	$0x1f, R0, R9
	AND	$0x1f, R2, R10
	BEQ	loop

	// Input string is not 32-byte aligned. We calculate the
	// syndrome value for the aligned 32 bytes block containing

	VLD1.P	16(R1), [V6.B16]                            // load 16bytes message
	VLD1.P	16(R1), [V7.B16]                            // load 16bytes message
	VMOV	V0.B16, V2.B16                              // backup: VO h(dcba)
	VMOV	V1.B16, V3.B16                              // backup: V1 h(hgfe)
	VMOV	V2.B16, V8.B16
	VREV32	V4.B16, V4.B16                              // prepare for using message in Byte format
	VREV32	V5.B16, V5.B16
	VREV32	V6.B16, V6.B16

chunk:
	BIC	$0x1f, R2, R9
	// The first chunk can also be the last
	CBZ	R9, tail
	// R3 = end of 32-byte chunks
	ADD	R0, R9, R3
	MOVD	$1, R5
	VMOV	R5, V5.B16
	// R2 = length of tail
	SUB	R9, R2, R2
	// Duplicate R1 (byte to search) to 16 1-byte elements of V0
	VMOV	R1, V0.B16
	// Clear the low 64-bit element of V7 and V8
	VEOR	V7.B8, V7.B8, V7.B8
	VEOR	V8.B8, V8.B8, V8.B8
	PCALIGN $16

	VMOV	V8.D[0], V12.S[1]                                // ERROR "operand mismatch"
	VMOV	V8.D[0], V12.H[1]                                // ERROR "operand mismatch"
	VMOV	V8.D[0], V12.B[1]                                // ERROR "operand mismatch"
	VMOV	V8.S[0], V12.H[1]                                // ERROR "operand mismatch"
	VMOV	V8.S[0], V12.B[1]                                // ERROR "operand mismatch"