/*******************************************************************************
Copyright (c) 2015, The OpenBLAS Project
All rights reserved.
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are
met:
1. Redistributions of source code must retain the above copyright
notice, this list of conditions and the following disclaimer.
2. Redistributions in binary form must reproduce the above copyright
notice, this list of conditions and the following disclaimer in
the documentation and/or other materials provided with the
distribution.
3. Neither the name of the OpenBLAS project nor the names of
its contributors may be used to endorse or promote products
derived from this software without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*******************************************************************************/
#define ASSEMBLER
#include "common.h"

/*                   X0          X1          X2          s0        X3        x4       x5           x6  */
/*int CNAME(BLASLONG bm,BLASLONG bn,BLASLONG bk,FLOAT alpha,FLOAT* ba,FLOAT* bb,FLOAT* C,BLASLONG ldc) */

#define origM		x0
#define origN		x1
#define origK		x2
#define origPA		x3
#define origPB		x4
#define pC		x5
#define LDC		x6
#define offset		x7
#define counterL	x8
#define counterI	x9
#define counterJ	x10
#define pB		x11
#define pCRow0		x12
#define pCRow1		x13
#define pCRow2		x14
#define pA		x15
#define temp		x16

#define alpha0		s10
#define alphaV0		v10.s[0]
#define alpha1		s11
#define alphaV1		v11.s[0]
#define alpha2		s14
#define alphaV2		v14.s[0]
#define alpha3		s15
#define alphaV3		v15.s[0]

#define A_PRE_SIZE	640
#define B_PRE_SIZE	224
#define C_PRE_SIZE	96

// 00 origM
// 01 origN
// 02 origK
// 03 origPA
// 04 origPB
// 05 pC
// 06 origLDC -> LDC
// 07 offset
// 08 counterL
// 09 counterI
// 10 counterJ
// 11 pB
// 12 pCRow0
// 13 pCRow1
// 14 pCRow2
// 15 pA
// 16 temp
// 17
// 18 must save
// 19 must save
// 20 must save pA0_2, pA0_3
// 21 must save pA0_6, pA0_7
// 22 must save pA1_2, pA1_3
// 23 must save pA1_6, pA1_7
// 24 must save pB0_2, pB0_3
// 25 must save pB0_6, pB0_7
// 26 must save pB1_2, pB1_3
// 27 must save pB1_6, pB1_7
// 28 must save
// 29 frame
// 30 link
// 31 sp

//v00 ALPHA -> pA0_0, pA0_1, pA0_2, pA0_3
//v01 pA0_4, pA0_5, pA0_6, pA0_7
//v02 pA1_0, pA1_1, pA1_2, pA1_3
//v03 pA1_4, pA1_5, pA1_6, pA1_7
//v04 pB0_0, pB0_1, pB0_2, pB0_3
//v05 pB0_4, pB0_5, pB0_6, pB0_7
//v06 pB1_0, pB1_1, pB1_2, pB1_3
//v07 pB1_4, pB1_5, pB1_6, pB1_7
//v08 must save
//v09 must save
//v10 must save ALPHA0
//v11 must save ALPHA1
//v12 must save
//v13 must save
//v14 must save ALPHA2
//v15 must save ALPHA3
//v16 must save C00, C01, C02, C03
//v17 must save C04, C05, C06, C07
//v18 C08, C09, C10, C11
//v19 C12, C13, C14, C15
//v20 C16, C17, C18, C19
//v21 C20, C21, C22, C23
//v22 C24, C25, C26, C27
//v23 C28, C29, C30, C31
//v24 C32, C33, C34, C35
//v25 C36, C37, C38, C39
//v26 C40, C41, C42, C43
//v27 C44, C45, C46, C47
//v28 C48, C49, C50, C51
//v29 C52, C53, C54, C55
//v30 C56, C57, C58, C59
//v31 C60, C61, C62, C63

/*******************************************************************************
* Macro definitions
*******************************************************************************/

.macro INIT8x8
	fmov		s16, wzr
	fmov		s17, wzr
	fmov		s18, s16
	fmov		s19, s17
	fmov		s20, wzr
	fmov		s21, s16
	fmov		s22, s17
	fmov		s23, s18
	fmov		s24, wzr
	fmov		s25, s16
	fmov		s26, s17
	fmov		s27, s18
	fmov		s28, wzr
	fmov		s29, s16
	fmov		s30, s17
	fmov		s31, s18
.endm

.macro KERNEL8x8_I
	ldp	q0, q1, [pA], #32
	ldp	q4, q5, [pB], #32

	ldr	d2, [pA], #8
	ldr	d6, [pB], #8
	ldr	d3, [pA, #8]
	ldr	d7, [pB, #8]
	ldr	x22, [pA], #16
	fmul	v16.4s, v0.4s, v4.s[0]
	ldr	x26, [pB], #16
	fmul	v17.4s, v1.4s, v4.s[0]
	ldr	x23, [pA], #8
	fmul	v18.4s, v0.4s, v4.s[1]
	ldr	x27, [pB], #8
	fmul	v19.4s, v1.4s, v4.s[1]
	fmul	v20.4s, v0.4s, v4.s[2]
	prfm	PLDL1KEEP, [pA, #A_PRE_SIZE]
	fmul	v21.4s, v1.4s, v4.s[2]
	prfm	PLDL1KEEP, [pB, #B_PRE_SIZE]
	fmul	v22.4s, v0.4s, v4.s[3]
	fmul	v23.4s, v1.4s, v4.s[3]
	fmul	v24.4s, v0.4s, v5.s[0]
	fmul	v25.4s, v1.4s, v5.s[0]
	fmul	v26.4s, v0.4s, v5.s[1]
	fmul	v27.4s, v1.4s, v5.s[1]
	fmul	v28.4s, v0.4s, v5.s[2]
	fmul	v29.4s, v1.4s, v5.s[2]
	fmul	v30.4s, v0.4s, v5.s[3]
	fmul	v31.4s, v1.4s, v5.s[3]
.endm

.macro KERNEL8x8_M1
	ldr	d2, [pA], #8
	fmov	v0.d[1], x20
	ldr	d6, [pB], #8
	fmov	v4.d[1], x24
	ldr	d3, [pA, #8]
	fmov	v1.d[1], x21
	ldr	d7, [pB, #8]
	fmov	v5.d[1], x25
	fmla	v16.4s, v0.4s, v4.s[0]
	ldr	x22, [pA], #16
	fmla	v17.4s, v1.4s, v4.s[0]
	ldr	x26, [pB], #16
	fmla	v18.4s, v0.4s, v4.s[1]
	ldr	x23, [pA], #8
	fmla	v19.4s, v1.4s, v4.s[1]
	ldr	x27, [pB], #8
	fmla	v20.4s, v0.4s, v4.s[2]
	prfm	PLDL1KEEP, [pA, #A_PRE_SIZE]
	fmla	v21.4s, v1.4s, v4.s[2]
	prfm	PLDL1KEEP, [pB, #B_PRE_SIZE]
	fmla	v22.4s, v0.4s, v4.s[3]
	fmla	v23.4s, v1.4s, v4.s[3]
	fmla	v24.4s, v0.4s, v5.s[0]
	fmla	v25.4s, v1.4s, v5.s[0]
	fmla	v26.4s, v0.4s, v5.s[1]
	fmla	v27.4s, v1.4s, v5.s[1]
	fmla	v28.4s, v0.4s, v5.s[2]
	fmla	v29.4s, v1.4s, v5.s[2]
	fmla	v30.4s, v0.4s, v5.s[3]
	fmla	v31.4s, v1.4s, v5.s[3]
.endm

.macro KERNEL8x8_M2
	ldr	d0, [pA], #8
	fmov	v2.d[1], x22
	ldr	d4, [pB], #8
	fmov	v6.d[1], x26
	ldr	d1, [pA, #8]
	fmov	v3.d[1], x23
	ldr	d5, [pB, #8]
	fmov	v7.d[1], x27
	fmla	v16.4s, v2.4s, v6.s[0]
	ldr	x20, [pA], #16
	fmla	v17.4s, v3.4s, v6.s[0]
	ldr	x24, [pB], #16
	fmla	v18.4s, v2.4s, v6.s[1]
	ldr	x21, [pA], #8
	fmla	v19.4s, v3.4s, v6.s[1]
	ldr	x25, [pB], #8
	fmla	v20.4s, v2.4s, v6.s[2]
	prfm	PLDL1KEEP, [pA, #A_PRE_SIZE]
	fmla	v21.4s, v3.4s, v6.s[2]
	prfm	PLDL1KEEP, [pB, #B_PRE_SIZE]
	fmla	v22.4s, v2.4s, v6.s[3]
	fmla	v23.4s, v3.4s, v6.s[3]
	fmla	v24.4s, v2.4s, v7.s[0]
	fmla	v25.4s, v3.4s, v7.s[0]
	fmla	v26.4s, v2.4s, v7.s[1]
	fmla	v27.4s, v3.4s, v7.s[1]
	fmla	v28.4s, v2.4s, v7.s[2]
	fmla	v29.4s, v3.4s, v7.s[2]
	fmla	v30.4s, v2.4s, v7.s[3]
	fmla	v31.4s, v3.4s, v7.s[3]
.endm

.macro KERNEL8x8_E
	fmov	v2.d[1], x22
	fmov	v6.d[1], x26
	fmov	v3.d[1], x23
	fmov	v7.d[1], x27
	fmla	v16.4s, v2.4s, v6.s[0]
	fmla	v17.4s, v3.4s, v6.s[0]
	fmla	v18.4s, v2.4s, v6.s[1]
	fmla	v19.4s, v3.4s, v6.s[1]
	fmla	v20.4s, v2.4s, v6.s[2]
	prfm	PLDL1KEEP, [pA, #A_PRE_SIZE]
	fmla	v21.4s, v3.4s, v6.s[2]
	prfm	PLDL1KEEP, [pB, #B_PRE_SIZE]
	fmla	v22.4s, v2.4s, v6.s[3]
	fmla	v23.4s, v3.4s, v6.s[3]
	fmla	v24.4s, v2.4s, v7.s[0]
	fmla	v25.4s, v3.4s, v7.s[0]
	fmla	v26.4s, v2.4s, v7.s[1]
	fmla	v27.4s, v3.4s, v7.s[1]
	fmla	v28.4s, v2.4s, v7.s[2]
	fmla	v29.4s, v3.4s, v7.s[2]
	fmla	v30.4s, v2.4s, v7.s[3]
	fmla	v31.4s, v3.4s, v7.s[3]
.endm

.macro KERNEL8x8_SUB
	ldp	q0, q1, [pA], #32
	ldp	q4, q5, [pB], #32

	fmla	v16.4s, v0.4s, v4.s[0]
	fmla	v17.4s, v1.4s, v4.s[0]
	fmla	v18.4s, v0.4s, v4.s[1]
	fmla	v19.4s, v1.4s, v4.s[1]
	fmla	v20.4s, v0.4s, v4.s[2]
	prfm	PLDL1KEEP, [pA, #A_PRE_SIZE]
	fmla	v21.4s, v1.4s, v4.s[2]
	prfm	PLDL1KEEP, [pB, #B_PRE_SIZE]
	fmla	v22.4s, v0.4s, v4.s[3]
	fmla	v23.4s, v1.4s, v4.s[3]
	fmla	v24.4s, v0.4s, v5.s[0]
	fmla	v25.4s, v1.4s, v5.s[0]
	fmla	v26.4s, v0.4s, v5.s[1]
	fmla	v27.4s, v1.4s, v5.s[1]
	fmla	v28.4s, v0.4s, v5.s[2]
	fmla	v29.4s, v1.4s, v5.s[2]
	fmla	v30.4s, v0.4s, v5.s[3]
	fmla	v31.4s, v1.4s, v5.s[3]
.endm

.macro SAVE8x8
	prfm	PLDL2KEEP, [pCRow0, #C_PRE_SIZE]
	add	pCRow1, pCRow0, LDC

	ldp	q0, q1, [pCRow0]
	fmla	v0.4s, v16.4s, alphaV0
	fmla	v1.4s, v17.4s, alphaV1
	stp	q0, q1, [pCRow0]

	prfm	PLDL2KEEP, [pCRow1, #C_PRE_SIZE]
	add	pCRow2, pCRow1, LDC

	ldp	q2, q3, [pCRow1]
	fmla	v2.4s, v18.4s, alphaV2
	fmla	v3.4s, v19.4s, alphaV3
	stp	q2, q3, [pCRow1]

	prfm	PLDL2KEEP, [pCRow2, #C_PRE_SIZE]
	add	pCRow1, pCRow2, LDC

	ldp	q4, q5, [pCRow2]
	fmla	v4.4s, v20.4s, alphaV0
	fmla	v5.4s, v21.4s, alphaV1
	stp	q4, q5, [pCRow2]

	prfm	PLDL2KEEP, [pCRow1, #C_PRE_SIZE]
	add	pCRow2, pCRow1, LDC

	ldp	q6, q7, [pCRow1]
	fmla	v6.4s, v22.4s, alphaV2
	fmla	v7.4s, v23.4s, alphaV3
	stp	q6, q7, [pCRow1]

	prfm	PLDL2KEEP, [pCRow2, #C_PRE_SIZE]
	add	pCRow1, pCRow2, LDC

	ldp	q0, q1, [pCRow2]
	fmla	v0.4s, v24.4s, alphaV0
	fmla	v1.4s, v25.4s, alphaV1
	stp	q0, q1, [pCRow2]

	prfm	PLDL2KEEP, [pCRow1, #C_PRE_SIZE]
	add	pCRow2, pCRow1, LDC

	ldp	q2, q3, [pCRow1]
	fmla	v2.4s, v26.4s, alphaV2
	fmla	v3.4s, v27.4s, alphaV3
	stp	q2, q3, [pCRow1]

	prfm	PLDL2KEEP, [pCRow2, #C_PRE_SIZE]
	add	pCRow1, pCRow2, LDC

	ldp	q4, q5, [pCRow2]
	fmla	v4.4s, v28.4s, alphaV0
	fmla	v5.4s, v29.4s, alphaV1
	stp	q4, q5, [pCRow2]

	prfm	PLDL2KEEP, [pCRow1, #C_PRE_SIZE]

	ldp	q6, q7, [pCRow1]
	fmla	v6.4s, v30.4s, alphaV2
	fmla	v7.4s, v31.4s, alphaV3
	stp	q6, q7, [pCRow1]

	add	pCRow0, pCRow0, #32
.endm

/******************************************************************************/

.macro INIT4x8
	fmov		s16, wzr
	fmov		s18, wzr
	fmov		s20, wzr
	fmov		s22, s16
	fmov		s24, wzr
	fmov		s26, s16
	fmov		s28, s18
	fmov		s30, s20
.endm

.macro KERNEL4x8_I
	ldr	q0, [pA], #16
	ldp	q4, q5, [pB], #32

	ldr	d2, [pA], #8
	ldr	d6, [pB], #8
	ldr	d7, [pB, #8]
	ldr	x22, [pA], #8
	fmul	v16.4s, v0.4s, v4.s[0]
	ldr	x26, [pB], #16
	fmul	v18.4s, v0.4s, v4.s[1]
	ldr	x27, [pB], #8
	fmul	v20.4s, v0.4s, v4.s[2]
	fmul	v22.4s, v0.4s, v4.s[3]
	prfm	PLDL1KEEP, [pB, #B_PRE_SIZE]
	fmul	v24.4s, v0.4s, v5.s[0]
	fmul	v26.4s, v0.4s, v5.s[1]
	fmul	v28.4s, v0.4s, v5.s[2]
	fmul	v30.4s, v0.4s, v5.s[3]
.endm

.macro KERNEL4x8_M1
	ldr	d2, [pA], #8
	fmov	v0.d[1], x20
	ldr	d6, [pB], #8
	fmov	v4.d[1], x24
	ldr	d7, [pB, #8]
	fmov	v5.d[1], x25
	ldr	x22, [pA], #8
	fmla	v16.4s, v0.4s, v4.s[0]
	ldr	x26, [pB], #16
	fmla	v18.4s, v0.4s, v4.s[1]
	ldr	x27, [pB], #8
	fmla	v20.4s, v0.4s, v4.s[2]
	fmla	v22.4s, v0.4s, v4.s[3]
	prfm	PLDL1KEEP, [pB, #B_PRE_SIZE]
	fmla	v24.4s, v0.4s, v5.s[0]
	fmla	v26.4s, v0.4s, v5.s[1]
	fmla	v28.4s, v0.4s, v5.s[2]
	fmla	v30.4s, v0.4s, v5.s[3]
.endm

.macro KERNEL4x8_M2
	ldr	d0, [pA], #8
	fmov	v2.d[1], x22
	ldr	d4, [pB], #8
	fmov	v6.d[1], x26
	ldr	d5, [pB, #8]
	fmov	v7.d[1], x27
	ldr	x20, [pA], #8
	fmla	v16.4s, v2.4s, v6.s[0]
	ldr	x24, [pB], #16
	fmla	v18.4s, v2.4s, v6.s[1]
	ldr	x25, [pB], #8
	fmla	v20.4s, v2.4s, v6.s[2]
	fmla	v22.4s, v2.4s, v6.s[3]
	prfm	PLDL1KEEP, [pB, #B_PRE_SIZE]
	fmla	v24.4s, v2.4s, v7.s[0]
	fmla	v26.4s, v2.4s, v7.s[1]
	fmla	v28.4s, v2.4s, v7.s[2]
	fmla	v30.4s, v2.4s, v7.s[3]
.endm

.macro KERNEL4x8_E
	fmov	v2.d[1], x22
	fmov	v6.d[1], x26
	fmov	v7.d[1], x27
	fmla	v16.4s, v2.4s, v6.s[0]
	fmla	v18.4s, v2.4s, v6.s[1]
	fmla	v20.4s, v2.4s, v6.s[2]
	fmla	v22.4s, v2.4s, v6.s[3]
	prfm	PLDL1KEEP, [pB, #B_PRE_SIZE]
	fmla	v24.4s, v2.4s, v7.s[0]
	fmla	v26.4s, v2.4s, v7.s[1]
	fmla	v28.4s, v2.4s, v7.s[2]
	fmla	v30.4s, v2.4s, v7.s[3]
.endm

.macro KERNEL4x8_SUB
	ldr	q0, [pA], #16
	ldp	q4, q5, [pB], #32

	fmla	v16.4s, v0.4s, v4.s[0]
	fmla	v18.4s, v0.4s, v4.s[1]
	fmla	v20.4s, v0.4s, v4.s[2]
	fmla	v22.4s, v0.4s, v4.s[3]
	prfm	PLDL1KEEP, [pB, #B_PRE_SIZE]
	fmla	v24.4s, v0.4s, v5.s[0]
	fmla	v26.4s, v0.4s, v5.s[1]
	fmla	v28.4s, v0.4s, v5.s[2]
	fmla	v30.4s, v0.4s, v5.s[3]
.endm

.macro SAVE4x8
	add	pCRow1, pCRow0, LDC

	ldr	q0, [pCRow0]
	fmla	v0.4s, v16.4s, alphaV0
	str	q0, [pCRow0]

	add	pCRow2, pCRow1, LDC

	ldr	q2, [pCRow1]
	fmla	v2.4s, v18.4s, alphaV2
	str	q2, [pCRow1]

	add	pCRow1, pCRow2, LDC

	ldr	q4, [pCRow2]
	fmla	v4.4s, v20.4s, alphaV0
	str	q4, [pCRow2]

	add	pCRow2, pCRow1, LDC

	ldr	q6, [pCRow1]
	fmla	v6.4s, v22.4s, alphaV2
	str	q6, [pCRow1]

	add	pCRow1, pCRow2, LDC

	ldr	q0, [pCRow2]
	fmla	v0.4s, v24.4s, alphaV0
	str	q0, [pCRow2]

	add	pCRow2, pCRow1, LDC

	ldr	q2, [pCRow1]
	fmla	v2.4s, v26.4s, alphaV2
	str	q2, [pCRow1]

	add	pCRow1, pCRow2, LDC

	ldr	q4, [pCRow2]
	fmla	v4.4s, v28.4s, alphaV0
	str	q4, [pCRow2]

	ldr	q6, [pCRow1]
	fmla	v6.4s, v30.4s, alphaV2
	str	q6, [pCRow1]

	add	pCRow0, pCRow0, #16
.endm

/******************************************************************************/

.macro INIT2x8
	fmov		s16, wzr
	fmov		s18, wzr
	fmov		s20, wzr
	fmov		s22, s16
	fmov		s24, wzr
	fmov		s26, s16
	fmov		s28, s18
	fmov		s30, s20
.endm

.macro KERNEL2x8_SUB
	ldr	d0, [pA], #8
	ldp	q4, q5, [pB], #32

	fmla	v16.2s, v0.2s, v4.s[0]
	fmla	v18.2s, v0.2s, v4.s[1]
	fmla	v20.2s, v0.2s, v4.s[2]
	fmla	v22.2s, v0.2s, v4.s[3]
	prfm	PLDL1KEEP, [pB, #B_PRE_SIZE]
	fmla	v24.2s, v0.2s, v5.s[0]
	fmla	v26.2s, v0.2s, v5.s[1]
	fmla	v28.2s, v0.2s, v5.s[2]
	fmla	v30.2s, v0.2s, v5.s[3]
.endm

.macro SAVE2x8
	add	pCRow1, pCRow0, LDC

	ldr	d0, [pCRow0]
	fmla	v0.2s, v16.2s, alphaV0
	str	d0, [pCRow0]

	add	pCRow2, pCRow1, LDC

	ldr	d2, [pCRow1]
	fmla	v2.2s, v18.2s, alphaV2
	str	d2, [pCRow1]

	add	pCRow1, pCRow2, LDC

	ldr	d4, [pCRow2]
	fmla	v4.2s, v20.2s, alphaV0
	str	d4, [pCRow2]

	add	pCRow2, pCRow1, LDC

	ldr	d6, [pCRow1]
	fmla	v6.2s, v22.2s, alphaV2
	str	d6, [pCRow1]

	add	pCRow1, pCRow2, LDC

	ldr	d0, [pCRow2]
	fmla	v0.2s, v24.2s, alphaV0
	str	d0, [pCRow2]

	add	pCRow2, pCRow1, LDC

	ldr	d2, [pCRow1]
	fmla	v2.2s, v26.2s, alphaV2
	str	d2, [pCRow1]

	add	pCRow1, pCRow2, LDC

	ldr	d4, [pCRow2]
	fmla	v4.2s, v28.2s, alphaV0
	str	d4, [pCRow2]

	ldr	d6, [pCRow1]
	fmla	v6.2s, v30.2s, alphaV2
	str	d6, [pCRow1]

	add	pCRow0, pCRow0, #8
.endm

/******************************************************************************/

.macro INIT1x8
	fmov		s16, wzr
	fmov		s18, wzr
	fmov		s20, wzr
	fmov		s22, s16
	fmov		s24, wzr
	fmov		s26, s16
	fmov		s28, s18
	fmov		s30, s20
.endm

.macro KERNEL1x8_SUB
	ldp	q4, q5, [pB], #32
	ldr	s0, [pA], #4

	fmla	s16, s0, v4.s[0]
	fmla	s18, s0, v4.s[1]
	fmla	s20, s0, v4.s[2]
	fmla	s22, s0, v4.s[3]
	prfm	PLDL1KEEP, [pB, #B_PRE_SIZE]
	fmla	s24, s0, v5.s[0]
	fmla	s26, s0, v5.s[1]
	fmla	s28, s0, v5.s[2]
	fmla	s30, s0, v5.s[3]
.endm

.macro SAVE1x8
	add	pCRow1, pCRow0, LDC

	ldr	s0, [pCRow0]
	fmla	s0, s16, alphaV0
	str	s0, [pCRow0]

	add	pCRow2, pCRow1, LDC

	ldr	s2, [pCRow1]
	fmla	s2, s18, alphaV2
	str	s2, [pCRow1]

	add	pCRow1, pCRow2, LDC

	ldr	s4, [pCRow2]
	fmla	s4, s20, alphaV0
	str	s4, [pCRow2]

	add	pCRow2, pCRow1, LDC

	ldr	s6, [pCRow1]
	fmla	s6, s22, alphaV2
	str	s6, [pCRow1]

	add	pCRow1, pCRow2, LDC

	ldr	s0, [pCRow2]
	fmla	s0, s24, alphaV0
	str	s0, [pCRow2]

	add	pCRow2, pCRow1, LDC

	ldr	s2, [pCRow1]
	fmla	s2, s26, alphaV2
	str	s2, [pCRow1]

	add	pCRow1, pCRow2, LDC

	ldr	s4, [pCRow2]
	fmla	s4, s28, alphaV0
	str	s4, [pCRow2]

	ldr	s6, [pCRow1]
	fmla	s6, s30, alphaV2
	str	s6, [pCRow1]

	add	pCRow0, pCRow0, #4
.endm

/******************************************************************************/

.macro INIT8x4
	fmov		s16, wzr
	fmov		s17, wzr
	fmov		s18, wzr
	fmov		s19, s16
	fmov		s20, wzr
	fmov		s21, s16
	fmov		s22, wzr
	fmov		s23, s16
.endm

.macro KERNEL8x4_I
	ldp	q0, q1, [pA], #32
	ldr	q4, [pB], #16

	ldr	d2, [pA], #8
	ldr	d6, [pB], #8
	ldr	d3, [pA, #8]
	fmul	v16.4s, v0.4s, v4.s[0]
	ldr	x22, [pA], #16
	fmul	v17.4s, v1.4s, v4.s[0]
	ldr	x26, [pB], #8
	fmul	v18.4s, v0.4s, v4.s[1]
	ldr	x23, [pA], #8
	fmul	v19.4s, v1.4s, v4.s[1]
	prfm	PLDL1KEEP, [pA, #A_PRE_SIZE]
	fmul	v20.4s, v0.4s, v4.s[2]
	fmul	v21.4s, v1.4s, v4.s[2]
	fmul	v22.4s, v0.4s, v4.s[3]
	fmul	v23.4s, v1.4s, v4.s[3]
.endm

.macro KERNEL8x4_M1
	ldr	d2, [pA], #8
	fmov	v0.d[1], x20
	ldr	d6, [pB], #8
	fmov	v4.d[1], x24
	ldr	d3, [pA, #8]
	fmov	v1.d[1], x21
	ldr	x22, [pA], #16
	fmla	v16.4s, v0.4s, v4.s[0]
	ldr	x26, [pB], #8
	fmla	v17.4s, v1.4s, v4.s[0]
	ldr	x23, [pA], #8
	fmla	v18.4s, v0.4s, v4.s[1]
	prfm	PLDL1KEEP, [pA, #A_PRE_SIZE]
	fmla	v19.4s, v1.4s, v4.s[1]
	fmla	v20.4s, v0.4s, v4.s[2]
	fmla	v21.4s, v1.4s, v4.s[2]
	fmla	v22.4s, v0.4s, v4.s[3]
	fmla	v23.4s, v1.4s, v4.s[3]
.endm

.macro KERNEL8x4_M2
	ldr	d0, [pA], #8
	fmov	v2.d[1], x22
	ldr	d4, [pB], #8
	fmov	v6.d[1], x26
	ldr	d1, [pA, #8]
	fmov	v3.d[1], x23
	ldr	x20, [pA], #16
	fmla	v16.4s, v2.4s, v6.s[0]
	ldr	x24, [pB], #8
	fmla	v17.4s, v3.4s, v6.s[0]
	ldr	x21, [pA], #8
	fmla	v18.4s, v2.4s, v6.s[1]
	prfm	PLDL1KEEP, [pA, #A_PRE_SIZE]
	fmla	v19.4s, v3.4s, v6.s[1]
	fmla	v20.4s, v2.4s, v6.s[2]
	fmla	v21.4s, v3.4s, v6.s[2]
	fmla	v22.4s, v2.4s, v6.s[3]
	fmla	v23.4s, v3.4s, v6.s[3]
.endm

.macro KERNEL8x4_E
	fmov	v2.d[1], x22
	fmov	v6.d[1], x26
	fmov	v3.d[1], x23
	fmla	v16.4s, v2.4s, v6.s[0]
	prfm	PLDL1KEEP, [pA, #A_PRE_SIZE]
	fmla	v17.4s, v3.4s, v6.s[0]
	fmla	v18.4s, v2.4s, v6.s[1]
	fmla	v19.4s, v3.4s, v6.s[1]
	fmla	v20.4s, v2.4s, v6.s[2]
	fmla	v21.4s, v3.4s, v6.s[2]
	fmla	v22.4s, v2.4s, v6.s[3]
	fmla	v23.4s, v3.4s, v6.s[3]
.endm

.macro KERNEL8x4_SUB
	ldp	q0, q1, [pA], #32
	ldr	q4, [pB], #16

	fmla	v16.4s, v0.4s, v4.s[0]
	fmla	v17.4s, v1.4s, v4.s[0]
	fmla	v18.4s, v0.4s, v4.s[1]
	fmla	v19.4s, v1.4s, v4.s[1]
	prfm	PLDL1KEEP, [pA, #A_PRE_SIZE]
	fmla	v20.4s, v0.4s, v4.s[2]
	fmla	v21.4s, v1.4s, v4.s[2]
	fmla	v22.4s, v0.4s, v4.s[3]
	fmla	v23.4s, v1.4s, v4.s[3]
.endm

.macro SAVE8x4
	prfm	PLDL2KEEP, [pCRow0, #C_PRE_SIZE]
	add	pCRow1, pCRow0, LDC

	ldp	q0, q1, [pCRow0]
	fmla	v0.4s, v16.4s, alphaV0
	fmla	v1.4s, v17.4s, alphaV1
	stp	q0, q1, [pCRow0]

	prfm	PLDL2KEEP, [pCRow1, #C_PRE_SIZE]
	add	pCRow2, pCRow1, LDC

	ldp	q4, q5, [pCRow1]
	fmla	v4.4s, v18.4s, alphaV0
	fmla	v5.4s, v19.4s, alphaV1
	stp	q4, q5, [pCRow1]

	prfm	PLDL2KEEP, [pCRow2, #C_PRE_SIZE]
	add	pCRow1, pCRow2, LDC

	ldp	q0, q1, [pCRow2]
	fmla	v0.4s, v20.4s, alphaV0
	fmla	v1.4s, v21.4s, alphaV1
	stp	q0, q1, [pCRow2]

	prfm	PLDL2KEEP, [pCRow1, #C_PRE_SIZE]

	ldp	q4, q5, [pCRow1]
	fmla	v4.4s, v22.4s, alphaV0
	fmla	v5.4s, v23.4s, alphaV1
	stp	q4, q5, [pCRow1]

	add	pCRow0, pCRow0, #32
.endm

/******************************************************************************/


.macro INIT4x4
	fmov		s16, wzr
	fmov		s18, wzr
	fmov		s20, wzr
	fmov		s22, wzr
.endm

.macro KERNEL4x4_I
	ldr	q0, [pA], #16
	ldr	q4, [pB], #16

	ldr	d2, [pA], #8
	ldr	d6, [pB], #8
	fmul	v16.4s, v0.4s, v4.s[0]
	ldr	x22, [pA], #8
	fmul	v18.4s, v0.4s, v4.s[1]
	ldr	x26, [pB], #8
	fmul	v20.4s, v0.4s, v4.s[2]
	fmul	v22.4s, v0.4s, v4.s[3]
.endm

.macro KERNEL4x4_M1
	ldr	d2, [pA], #8
	fmov	v0.d[1], x20
	ldr	d6, [pB], #8
	fmov	v4.d[1], x24
	ldr	x22, [pA], #8
	ldr	x26, [pB], #8
	fmla	v16.4s, v0.4s, v4.s[0]
	fmla	v18.4s, v0.4s, v4.s[1]
	fmla	v20.4s, v0.4s, v4.s[2]
	fmla	v22.4s, v0.4s, v4.s[3]
.endm

.macro KERNEL4x4_M2
	ldr	d0, [pA], #8
	fmov	v2.d[1], x22
	ldr	d4, [pB], #8
	fmov	v6.d[1], x26
	ldr	x20, [pA], #8
	ldr	x24, [pB], #8
	fmla	v16.4s, v2.4s, v6.s[0]
	fmla	v18.4s, v2.4s, v6.s[1]
	fmla	v20.4s, v2.4s, v6.s[2]
	fmla	v22.4s, v2.4s, v6.s[3]
.endm

.macro KERNEL4x4_E
	fmov	v2.d[1], x22
	fmov	v6.d[1], x26
	fmla	v16.4s, v2.4s, v6.s[0]
	fmla	v18.4s, v2.4s, v6.s[1]
	fmla	v20.4s, v2.4s, v6.s[2]
	fmla	v22.4s, v2.4s, v6.s[3]
.endm

.macro KERNEL4x4_SUB
	ldr	q0, [pA], #16
	ldr	q4, [pB], #16

	fmla	v16.4s, v0.4s, v4.s[0]
	fmla	v18.4s, v0.4s, v4.s[1]
	fmla	v20.4s, v0.4s, v4.s[2]
	fmla	v22.4s, v0.4s, v4.s[3]
.endm

.macro SAVE4x4
	ldr	q0, [pCRow0]
	fmla	v0.4s, v16.4s, alphaV0
	str	q0, [pCRow0]

	add	pCRow1, pCRow0, LDC
	ldr	q1, [pCRow1]
	fmla	v1.4s, v18.4s, alphaV2
	str	q1, [pCRow1]

	add	pCRow2, pCRow1, LDC
	ldr	q2, [pCRow2]
	fmla	v2.4s, v20.4s, alphaV0
	str	q2, [pCRow2]

	add	pCRow1, pCRow2, LDC
	ldr	q3, [pCRow1]
	fmla	v3.4s, v22.4s, alphaV2
	str	q3, [pCRow1]

	add	pCRow0, pCRow0, #16
.endm

/******************************************************************************/

.macro INIT2x4
	fmov		s16, wzr
	fmov		s18, wzr
	fmov		s20, wzr
	fmov		s22, s16
.endm

.macro KERNEL2x4_SUB
	ldr	d0, [pA], #8
	ldr	q4, [pB], #16

	fmla	v16.2s, v0.2s, v4.s[0]
	fmla	v18.2s, v0.2s, v4.s[1]
	fmla	v20.2s, v0.2s, v4.s[2]
	fmla	v22.2s, v0.2s, v4.s[3]
.endm

.macro SAVE2x4
	ldr	d8, [pCRow0]
	fmla	v8.2s, v16.2s, alphaV0
	str	d8, [pCRow0]

	add	pCRow1, pCRow0, LDC
	ldr	d12, [pCRow1]
	fmla	v12.2s, v18.2s, alphaV1
	str	d12, [pCRow1]

	add	pCRow2, pCRow1, LDC
	ldr	d8, [pCRow2]
	fmla	v8.2s, v20.2s, alphaV2
	str	d8, [pCRow2]

	add	pCRow1, pCRow2, LDC
	ldr	d12, [pCRow1]
	fmla	v12.2s, v22.2s, alphaV3
	str	d12, [pCRow1]

	add	pCRow0, pCRow0, #8
.endm

/******************************************************************************/

.macro INIT1x4
	fmov		s16, wzr
	fmov		s20, s16
.endm

.macro KERNEL1x4_SUB
	ldr	s0, [pA]
	add	pA, pA, #4

	ld1	{v8.2s, v9.2s}, [pB]
	add	pB, pB, #16

	fmla	v16.2s, v8.2s, v0.s[0]
	fmla	v20.2s, v9.2s, v0.s[0]
.endm

.macro SAVE1x4
	add	pCRow1, pCRow0, LDC
	ld1	{v8.s}[0], [pCRow0]
	ld1	{v8.s}[1], [pCRow1]
	fmla	v8.2s, v16.2s, alphaV0
	st1	{v8.s}[0], [pCRow0]
	st1	{v8.s}[1], [pCRow1]

	add	pCRow2, pCRow1, LDC
	add	pCRow1, pCRow2, LDC
	ld1	{v12.s}[0], [pCRow2]
	ld1	{v12.s}[1], [pCRow1]
	fmla	v12.2s, v20.2s, alphaV1
	st1	{v12.s}[0], [pCRow2]
	st1	{v12.s}[1], [pCRow1]

	add	pCRow0, pCRow0, #4
.endm

/******************************************************************************/

.macro INIT8x2
	fmov	s16, wzr
	fmov	s17, s16
	fmov	s18, s17
	fmov	s19, s16
.endm

.macro KERNEL8x2_SUB
	ldp	q0, q1, [pA], #32
	ldr	d4, [pB], #8

	fmla	v16.4s, v0.4s, v4.s[0]
	fmla	v17.4s, v1.4s, v4.s[0]
	prfm	PLDL1KEEP, [pA, #A_PRE_SIZE]
	fmla	v18.4s, v0.4s, v4.s[1]
	fmla	v19.4s, v1.4s, v4.s[1]
.endm

.macro SAVE8x2
	prfm	PLDL2KEEP, [pCRow0, #C_PRE_SIZE]
	add	pCRow1, pCRow0, LDC

	ldp	q0, q1, [pCRow0]
	fmla	v0.4s, v16.4s, alphaV0
	fmla	v1.4s, v17.4s, alphaV1
	stp	q0, q1, [pCRow0]

	prfm	PLDL2KEEP, [pCRow1, #C_PRE_SIZE]
	add	pCRow2, pCRow1, LDC

	ldp	q4, q5, [pCRow1]
	fmla	v4.4s, v18.4s, alphaV0
	fmla	v5.4s, v19.4s, alphaV1
	stp	q4, q5, [pCRow1]

	add	pCRow0, pCRow0, #32
.endm

/******************************************************************************/

.macro INIT4x2
	fmov	s16, wzr
	fmov	s17, s16
	fmov	s20, s17
	fmov	s21, s16
.endm

.macro KERNEL4x2_SUB
	ld1	{v8.2s}, [pB]
	add	pB, pB, #8
	ld1	{v0.2s, v1.2s}, [pA]
	add	pA, pA, #16

	fmla	v16.2s, v0.2s, v8.s[0]
	fmla	v17.2s, v1.2s, v8.s[0]
	fmla	v20.2s, v0.2s, v8.s[1]
	fmla	v21.2s, v1.2s, v8.s[1]
.endm

.macro SAVE4x2
	ld1	{v8.2s, v9.2s}, [pCRow0]
	fmla	v8.2s, v16.2s, alphaV0
	fmla	v9.2s, v17.2s, alphaV1
	st1	{v8.2s, v9.2s}, [pCRow0]

	add	pCRow1, pCRow0, LDC
	ld1	{v12.2s, v13.2s}, [pCRow1]
	fmla	v12.2s, v20.2s, alphaV2
	fmla	v13.2s, v21.2s, alphaV3
	st1	{v12.2s, v13.2s}, [pCRow1]

	add	pCRow0, pCRow0, #16
.endm

/******************************************************************************/

.macro INIT2x2
	fmov		s16, wzr
	fmov		s20, s16
.endm

.macro KERNEL2x2_SUB
	ld1	{v8.2s}, [pB]
	add	pB, pB, #8

	ld1	{v0.2s}, [pA]
	add	pA, pA, #8

	fmla	v16.2s, v0.2s, v8.s[0]
	fmla	v20.2s, v0.2s, v8.s[1]
.endm

.macro SAVE2x2
	ld1	{v8.2s}, [pCRow0]
	fmla	v8.2s, v16.2s, alphaV0
	st1	{v8.2s}, [pCRow0]

	add	pCRow1 , pCRow0, LDC
	ld1	{v12.2s}, [pCRow1]
	fmla	v12.2s, v20.2s, alphaV1
	st1	{v12.2s}, [pCRow1]

	add	pCRow0, pCRow0, #8
.endm

/******************************************************************************/

.macro INIT1x2
	fmov		s16, wzr
.endm

.macro KERNEL1x2_SUB
	ld1	{v8.2s} , [pB]
	add	pB , pB, #8

	ldr	s0 , [pA]
	add	pA, pA, #4

	fmla	v16.2s, v8.2s, v0.s[0]
.endm

.macro SAVE1x2
	add	pCRow1 , pCRow0, LDC
	ld1	{v8.s}[0], [pCRow0]
	ld1	{v8.s}[1], [pCRow1]
	fmla	v8.2s, v16.2s, alphaV0
	st1	{v8.s}[0], [pCRow0]
	st1	{v8.s}[1], [pCRow1]

	add	pCRow0, pCRow0, #4
.endm

/******************************************************************************/

.macro INIT8x1
	fmov	s16, wzr
	fmov	s17, wzr
.endm

.macro KERNEL8x1_SUB
	ldr	s4, [pB], #4
	ldp	q0, q1, [pA], #32

	fmla	v16.4s, v0.4s, v4.s[0]
	prfm	PLDL1KEEP, [pA, #A_PRE_SIZE]
	fmla	v17.4s, v1.4s, v4.s[0]
.endm

.macro SAVE8x1
	prfm	PLDL2KEEP, [pCRow0, #C_PRE_SIZE]

	ldp	q0, q1, [pCRow0]
	fmla	v0.4s, v16.4s, alphaV0
	fmla	v1.4s, v17.4s, alphaV1
	stp	q0, q1, [pCRow0]

	add	pCRow0, pCRow0, #32
.endm

/******************************************************************************/

.macro INIT4x1
	fmov	s16, wzr
	fmov	s17, s16
.endm

.macro KERNEL4x1_SUB
	ldr	s8, [pB]
	add	pB , pB, #4

	ld1	{v0.2s, v1.2s}, [pA]
	add	pA , pA, #16

	fmla	v16.2s, v0.2s, v8.s[0]
	fmla	v17.2s, v1.2s, v8.s[0]
.endm

.macro SAVE4x1
	ld1	{v8.2s, v9.2s}, [pCRow0]
	fmla	v8.2s, v16.2s, alphaV0
	fmla	v9.2s, v17.2s, alphaV1
	st1	{v8.2s, v9.2s}, [pCRow0]

	add	pCRow0, pCRow0, #16
.endm

/******************************************************************************/

.macro INIT2x1
	fmov		s16, wzr
.endm

.macro KERNEL2x1_SUB
	ldr	s8, [pB]
	add	pB , pB, #4

	ld1	{v0.2s}, [pA]
	add	pA , pA, #8

	fmla	v16.2s, v0.2s, v8.s[0]
.endm

.macro SAVE2x1
	ld1	{v8.2s}, [pCRow0]
	fmla	v8.2s, v16.2s, alphaV0
	st1	{v8.2s}, [pCRow0]

	add	pCRow0, pCRow0, #8
.endm

/******************************************************************************/

.macro INIT1x1
	fmov	s16, wzr
.endm

.macro KERNEL1x1_SUB
	ldr	s8, [pB]
	add	pB , pB, #4

	ldr	s0, [pA]
	add	pA , pA, #4

	fmadd	s16, s0, s8, s16
.endm

.macro SAVE1x1
	ldr	s8, [pCRow0]
	fmla	s8, s16, alphaV0
	str	s8, [pCRow0]

	add	pCRow0, pCRow0, #4
.endm

/*******************************************************************************
* End of macro definitions
*******************************************************************************/

	PROLOGUE

.Lsgemm_kernel_begin:

	.align 5
	add	sp, sp, #-(11 * 16)
	stp	d8, d9, [sp, #(0 * 16)]
	stp	d10, d11, [sp, #(1 * 16)]
	stp	d12, d13, [sp, #(2 * 16)]
	stp	d14, d15, [sp, #(3 * 16)]
	stp	d16, d17, [sp, #(4 * 16)]
	stp	x18, x19, [sp, #(5 * 16)]
	stp	x20, x21, [sp, #(6 * 16)]
	stp	x22, x23, [sp, #(7 * 16)]
	stp	x24, x25, [sp, #(8 * 16)]
	stp	x26, x27, [sp, #(9 * 16)]
	str	x28, [sp, #(10 * 16)]

	fmov	alpha0, s0
	fmov	alpha1, s0
	fmov	alpha2, s0
	fmov	alpha3, s0

	lsl	LDC, LDC, #2			// ldc = ldc * 4

	mov	pB, origPB

	mov	counterJ, origN
	asr	counterJ, counterJ, #3		// J = J / 8
	cmp	counterJ, #0
	ble	.Lsgemm_kernel_L4_BEGIN

/******************************************************************************/
/******************************************************************************/

.Lsgemm_kernel_L8_BEGIN:
	mov	pCRow0, pC			// pCRow0 = C
	add	pC, pC, LDC, lsl #3

	mov	pA, origPA			// pA = start of A array

/******************************************************************************/

.Lsgemm_kernel_L8_M8_BEGIN:

	mov	counterI, origM
	asr	counterI, counterI, #3		// counterI = counterI / 8
	cmp	counterI, #0
	ble	.Lsgemm_kernel_L8_M4_BEGIN

.Lsgemm_kernel_L8_M8_20:

	mov	pB, origPB

	asr	counterL , origK, #3		// L = K / 8
	cmp	counterL , #2			// is there at least 16 to do?
	blt	.Lsgemm_kernel_L8_M8_32

	KERNEL8x8_I				// do one in the K
	KERNEL8x8_M2				// do another in the K
	KERNEL8x8_M1
	KERNEL8x8_M2
	KERNEL8x8_M1
	KERNEL8x8_M2
	KERNEL8x8_M1
	KERNEL8x8_M2

	subs	counterL, counterL, #2
	ble	.Lsgemm_kernel_L8_M8_22a
	.align 5

.Lsgemm_kernel_L8_M8_22:

	KERNEL8x8_M1
	KERNEL8x8_M2
	KERNEL8x8_M1
	KERNEL8x8_M2
	KERNEL8x8_M1
	KERNEL8x8_M2
	KERNEL8x8_M1
	KERNEL8x8_M2

	subs	counterL, counterL, #1
	bgt	.Lsgemm_kernel_L8_M8_22

.Lsgemm_kernel_L8_M8_22a:

	KERNEL8x8_M1
	KERNEL8x8_M2
	KERNEL8x8_M1
	KERNEL8x8_M2
	KERNEL8x8_M1
	KERNEL8x8_M2
	KERNEL8x8_M1
	KERNEL8x8_E

	b	 .Lsgemm_kernel_L8_M8_44

.Lsgemm_kernel_L8_M8_32:

	tst	counterL, #1
	ble	.Lsgemm_kernel_L8_M8_40

	KERNEL8x8_I
	KERNEL8x8_M2
	KERNEL8x8_M1
	KERNEL8x8_M2
	KERNEL8x8_M1
	KERNEL8x8_M2
	KERNEL8x8_M1
	KERNEL8x8_E

	b	.Lsgemm_kernel_L8_M8_44

.Lsgemm_kernel_L8_M8_40:

	INIT8x8

.Lsgemm_kernel_L8_M8_44:

	ands	counterL , origK, #7
	ble	.Lsgemm_kernel_L8_M8_100

.Lsgemm_kernel_L8_M8_46:

	KERNEL8x8_SUB

	subs	counterL, counterL, 1
	bgt	.Lsgemm_kernel_L8_M8_46

.Lsgemm_kernel_L8_M8_100:

	SAVE8x8

.Lsgemm_kernel_L8_M8_END:
	subs	counterI, counterI, #1
	bne	.Lsgemm_kernel_L8_M8_20

/******************************************************************************/

.Lsgemm_kernel_L8_M4_BEGIN:

	mov	counterI, origM
	tst	counterI , #7
	ble	.Lsgemm_kernel_L8_END

	tst	counterI, #4
	ble	.Lsgemm_kernel_L8_M2_BEGIN

.Lsgemm_kernel_L8_M4_20:

	mov	pB, origPB

	asr	counterL , origK, #1		// L = K / 2
	cmp	counterL , #2			// is there at least 4 to do?
	blt	.Lsgemm_kernel_L8_M4_32

	KERNEL4x8_I				// do one in the K
	KERNEL4x8_M2				// do another in the K

	subs	counterL, counterL, #2
	ble	.Lsgemm_kernel_L8_M4_22a
	.align 5

.Lsgemm_kernel_L8_M4_22:

	KERNEL4x8_M1
	KERNEL4x8_M2

	subs	counterL, counterL, #1
	bgt	.Lsgemm_kernel_L8_M4_22

.Lsgemm_kernel_L8_M4_22a:

	KERNEL4x8_M1
	KERNEL4x8_E

	b	 .Lsgemm_kernel_L8_M4_44

.Lsgemm_kernel_L8_M4_32:

	tst	counterL, #1
	ble	.Lsgemm_kernel_L8_M4_40

	KERNEL4x8_I
	KERNEL4x8_E

	b	.Lsgemm_kernel_L8_M4_44

.Lsgemm_kernel_L8_M4_40:

	INIT4x8

.Lsgemm_kernel_L8_M4_44:

	ands	counterL , origK, #1
	ble	.Lsgemm_kernel_L8_M4_100

.Lsgemm_kernel_L8_M4_46:

	KERNEL4x8_SUB

.Lsgemm_kernel_L8_M4_100:

	SAVE4x8

.Lsgemm_kernel_L8_M4_END:

/******************************************************************************/

.Lsgemm_kernel_L8_M2_BEGIN:

	mov	counterI, origM
	tst	counterI , #3
	ble	.Lsgemm_kernel_L8_END

	tst	counterI, #2			// counterI = counterI / 2
	ble	.Lsgemm_kernel_L8_M1_BEGIN

.Lsgemm_kernel_L8_M2_20:

	INIT2x8

	mov	pB, origPB

	asr	counterL , origK, #3		// counterL = counterL / 8
	cmp	counterL , #0
	ble	.Lsgemm_kernel_L8_M2_40

.Lsgemm_kernel_L8_M2_22:

	KERNEL2x8_SUB
	KERNEL2x8_SUB
	KERNEL2x8_SUB
	KERNEL2x8_SUB

	KERNEL2x8_SUB
	KERNEL2x8_SUB
	KERNEL2x8_SUB
	KERNEL2x8_SUB

	subs	counterL, counterL, #1
	bgt	.Lsgemm_kernel_L8_M2_22


.Lsgemm_kernel_L8_M2_40:

	ands	counterL , origK, #7		// counterL = counterL % 8
	ble	.Lsgemm_kernel_L8_M2_100

.Lsgemm_kernel_L8_M2_42:

	KERNEL2x8_SUB

	subs	counterL, counterL, #1
	bgt	.Lsgemm_kernel_L8_M2_42

.Lsgemm_kernel_L8_M2_100:

	SAVE2x8

.Lsgemm_kernel_L8_M2_END:

/******************************************************************************/

.Lsgemm_kernel_L8_M1_BEGIN:

	tst	counterI, #1			// counterI = counterI % 2
	ble	.Lsgemm_kernel_L8_END

.Lsgemm_kernel_L8_M1_20:

	INIT1x8

	mov	pB, origPB

	asr	counterL , origK, #3		// counterL = counterL / 8
	cmp	counterL , #0
	ble	.Lsgemm_kernel_L8_M1_40

.Lsgemm_kernel_L8_M1_22:
	KERNEL1x8_SUB
	KERNEL1x8_SUB
	KERNEL1x8_SUB
	KERNEL1x8_SUB

	KERNEL1x8_SUB
	KERNEL1x8_SUB
	KERNEL1x8_SUB
	KERNEL1x8_SUB

	subs	counterL, counterL, #1
	bgt	.Lsgemm_kernel_L8_M1_22


.Lsgemm_kernel_L8_M1_40:

	ands	counterL , origK, #7		// counterL = counterL % 8
	ble	.Lsgemm_kernel_L8_M1_100

.Lsgemm_kernel_L8_M1_42:

	KERNEL1x8_SUB

	subs	counterL, counterL, #1
	bgt	.Lsgemm_kernel_L8_M1_42

.Lsgemm_kernel_L8_M1_100:

	SAVE1x8

.Lsgemm_kernel_L8_END:
	lsl	temp, origK, #5			// B = B + K * 4 * 8
	add	origPB, origPB, temp

	subs	counterJ, counterJ , #1		// j--
	bgt	.Lsgemm_kernel_L8_BEGIN

/******************************************************************************/
/******************************************************************************/

.Lsgemm_kernel_L4_BEGIN:

	mov	counterJ , origN
	tst	counterJ , #7
	ble	.Lsgemm_kernel_L999

	tst	counterJ , #4
	ble	.Lsgemm_kernel_L2_BEGIN

	mov	pCRow0, pC			// pCRow0 = pC

	add	pC,pC,LDC, lsl #2

	mov	pA, origPA			// pA = A

/******************************************************************************/

.Lsgemm_kernel_L4_M8_BEGIN:

	mov	counterI, origM
	asr	counterI, counterI, #3		// counterI = counterI / 8
	cmp	counterI, #0
	ble	.Lsgemm_kernel_L4_M4_BEGIN

.Lsgemm_kernel_L4_M8_20:

	mov	pB, origPB

	asr	counterL , origK, #1		// L = K / 2
	cmp	counterL , #2			// is there at least 4 to do?
	blt	.Lsgemm_kernel_L4_M8_32

	KERNEL8x4_I				// do one in the K
	KERNEL8x4_M2				// do another in the K

	subs	counterL, counterL, #2
	ble	.Lsgemm_kernel_L4_M8_22a
	.align 5

.Lsgemm_kernel_L4_M8_22:

	KERNEL8x4_M1
	KERNEL8x4_M2

	subs	counterL, counterL, #1
	bgt	.Lsgemm_kernel_L4_M8_22

.Lsgemm_kernel_L4_M8_22a:

	KERNEL8x4_M1
	KERNEL8x4_E

	b	 .Lsgemm_kernel_L4_M8_44

.Lsgemm_kernel_L4_M8_32:

	tst	counterL, #1
	ble	.Lsgemm_kernel_L4_M8_40

	KERNEL8x4_I
	KERNEL8x4_E

	b	.Lsgemm_kernel_L4_M8_44

.Lsgemm_kernel_L4_M8_40:

	INIT8x4

.Lsgemm_kernel_L4_M8_44:

	ands	counterL , origK, #1
	ble	.Lsgemm_kernel_L4_M8_100

.Lsgemm_kernel_L4_M8_46:

	KERNEL8x4_SUB

.Lsgemm_kernel_L4_M8_100:

	SAVE8x4

.Lsgemm_kernel_L4_M8_END:
	subs	counterI, counterI, #1
	bne	.Lsgemm_kernel_L4_M8_20

/******************************************************************************/

.Lsgemm_kernel_L4_M4_BEGIN:

	mov	counterI, origM
	tst	counterI , #7
	ble	.Lsgemm_kernel_L4_END

	tst	counterI, #4
	ble	.Lsgemm_kernel_L4_M2_BEGIN

.Lsgemm_kernel_L4_M4_20:

	mov	pB, origPB

	asr	counterL , origK, #1		// L = K / 2
	cmp	counterL , #2			// is there at least 4 to do?
	blt	.Lsgemm_kernel_L4_M4_32

	KERNEL4x4_I				// do one in the K
	KERNEL4x4_M2				// do another in the K

	subs	counterL, counterL, #2
	ble	.Lsgemm_kernel_L4_M4_22a
	.align 5

.Lsgemm_kernel_L4_M4_22:

	KERNEL4x4_M1
	KERNEL4x4_M2

	subs	counterL, counterL, #1
	bgt	.Lsgemm_kernel_L4_M4_22

.Lsgemm_kernel_L4_M4_22a:

	KERNEL4x4_M1
	KERNEL4x4_E

	b	 .Lsgemm_kernel_L4_M4_44

.Lsgemm_kernel_L4_M4_32:

	tst	counterL, #1
	ble	.Lsgemm_kernel_L4_M4_40

	KERNEL4x4_I
	KERNEL4x4_E

	b	.Lsgemm_kernel_L4_M4_44

.Lsgemm_kernel_L4_M4_40:

	INIT4x4

.Lsgemm_kernel_L4_M4_44:

	ands	counterL , origK, #1
	ble	.Lsgemm_kernel_L4_M4_100

.Lsgemm_kernel_L4_M4_46:

	KERNEL4x4_SUB

.Lsgemm_kernel_L4_M4_100:

	SAVE4x4

.Lsgemm_kernel_L4_M4_END:

/******************************************************************************/

.Lsgemm_kernel_L4_M2_BEGIN:

	mov	counterI, origM
	tst	counterI , #3
	ble	.Lsgemm_kernel_L4_END

	tst	counterI, #2			// counterI = counterI / 2
	ble	.Lsgemm_kernel_L4_M1_BEGIN

.Lsgemm_kernel_L4_M2_20:

	INIT2x4

	mov	pB, origPB

	asr	counterL , origK, #3		// counterL = counterL / 8
	cmp	counterL , #0
	ble	.Lsgemm_kernel_L4_M2_40

.Lsgemm_kernel_L4_M2_22:

	KERNEL2x4_SUB
	KERNEL2x4_SUB
	KERNEL2x4_SUB
	KERNEL2x4_SUB

	KERNEL2x4_SUB
	KERNEL2x4_SUB
	KERNEL2x4_SUB
	KERNEL2x4_SUB

	subs	counterL, counterL, #1
	bgt	.Lsgemm_kernel_L4_M2_22


.Lsgemm_kernel_L4_M2_40:

	ands	counterL , origK, #7		// counterL = counterL % 8
	ble	.Lsgemm_kernel_L4_M2_100

.Lsgemm_kernel_L4_M2_42:

	KERNEL2x4_SUB

	subs	counterL, counterL, #1
	bgt	.Lsgemm_kernel_L4_M2_42

.Lsgemm_kernel_L4_M2_100:

	SAVE2x4

.Lsgemm_kernel_L4_M2_END:

/******************************************************************************/

.Lsgemm_kernel_L4_M1_BEGIN:

	tst	counterI, #1			// counterI = counterI % 2
	ble	.Lsgemm_kernel_L4_END

.Lsgemm_kernel_L4_M1_20:

	INIT1x4

	mov	pB, origPB

	asr	counterL , origK, #3		// counterL = counterL / 8
	cmp	counterL , #0
	ble	.Lsgemm_kernel_L4_M1_40

.Lsgemm_kernel_L4_M1_22:
	KERNEL1x4_SUB
	KERNEL1x4_SUB
	KERNEL1x4_SUB
	KERNEL1x4_SUB

	KERNEL1x4_SUB
	KERNEL1x4_SUB
	KERNEL1x4_SUB
	KERNEL1x4_SUB

	subs	counterL, counterL, #1
	bgt	.Lsgemm_kernel_L4_M1_22


.Lsgemm_kernel_L4_M1_40:

	ands	counterL , origK, #7		// counterL = counterL % 8
	ble	.Lsgemm_kernel_L4_M1_100

.Lsgemm_kernel_L4_M1_42:

	KERNEL1x4_SUB

	subs	counterL, counterL, #1
	bgt	.Lsgemm_kernel_L4_M1_42

.Lsgemm_kernel_L4_M1_100:

	SAVE1x4

.Lsgemm_kernel_L4_END:
	add	origPB, origPB, origK, lsl #4	// B = B + K * 4 * 4

/******************************************************************************/
/******************************************************************************/

.Lsgemm_kernel_L2_BEGIN:   // less than 2 left in N direction

	mov	counterJ , origN
	tst	counterJ , #3
	ble	.Lsgemm_kernel_L999

	tst	counterJ , #2
	ble	.Lsgemm_kernel_L1_BEGIN

	mov	pCRow0, pC			// pCRow0 = pC

	add	pC,pC,LDC, lsl #1

	mov	pA, origPA			// pA = A

/******************************************************************************/

.Lsgemm_kernel_L2_M8_BEGIN:

	mov	counterI, origM
	asr	counterI, counterI, #3		// counterI = counterI / 8
	cmp	counterI,#0
	ble	.Lsgemm_kernel_L2_M4_BEGIN

.Lsgemm_kernel_L2_M8_20:

	INIT8x2

	mov	pB, origPB

	asr	counterL , origK, #3		// counterL = counterL / 8
	cmp	counterL,#0
	ble	.Lsgemm_kernel_L2_M8_40
	.align 5

.Lsgemm_kernel_L2_M8_22:
	KERNEL8x2_SUB
	KERNEL8x2_SUB
	KERNEL8x2_SUB
	KERNEL8x2_SUB

	KERNEL8x2_SUB
	KERNEL8x2_SUB
	KERNEL8x2_SUB
	KERNEL8x2_SUB

	subs	counterL, counterL, #1
	bgt	.Lsgemm_kernel_L2_M8_22


.Lsgemm_kernel_L2_M8_40:

	ands	counterL , origK, #7		// counterL = counterL % 8
	ble	.Lsgemm_kernel_L2_M8_100

.Lsgemm_kernel_L2_M8_42:

	KERNEL8x2_SUB

	subs	counterL, counterL, #1
	bgt	.Lsgemm_kernel_L2_M8_42

.Lsgemm_kernel_L2_M8_100:

	SAVE8x2

.Lsgemm_kernel_L2_M8_END:

	subs	counterI, counterI, #1
	bgt	.Lsgemm_kernel_L2_M8_20

/******************************************************************************/

.Lsgemm_kernel_L2_M4_BEGIN:

	mov	counterI, origM
	tst	counterI , #7
	ble	.Lsgemm_kernel_L2_END

	tst	counterI, #4
	ble	.Lsgemm_kernel_L2_M2_BEGIN

.Lsgemm_kernel_L2_M4_20:

	INIT4x2

	mov	pB, origPB

	asr	counterL , origK, #3		// counterL = counterL / 8
	cmp	counterL,#0
	ble	.Lsgemm_kernel_L2_M4_40
	.align 5

.Lsgemm_kernel_L2_M4_22:
	KERNEL4x2_SUB
	KERNEL4x2_SUB
	KERNEL4x2_SUB
	KERNEL4x2_SUB

	KERNEL4x2_SUB
	KERNEL4x2_SUB
	KERNEL4x2_SUB
	KERNEL4x2_SUB

	subs	counterL, counterL, #1
	bgt	.Lsgemm_kernel_L2_M4_22


.Lsgemm_kernel_L2_M4_40:

	ands	counterL , origK, #7		// counterL = counterL % 8
	ble	.Lsgemm_kernel_L2_M4_100

.Lsgemm_kernel_L2_M4_42:

	KERNEL4x2_SUB

	subs	counterL, counterL, #1
	bgt	.Lsgemm_kernel_L2_M4_42

.Lsgemm_kernel_L2_M4_100:

	SAVE4x2

.Lsgemm_kernel_L2_M4_END:

/******************************************************************************/

.Lsgemm_kernel_L2_M2_BEGIN:

	mov	counterI, origM
	tst	counterI , #3
	ble	.Lsgemm_kernel_L2_END

	tst	counterI, #2			// counterI = counterI / 2
	ble	.Lsgemm_kernel_L2_M1_BEGIN

.Lsgemm_kernel_L2_M2_20:

	INIT2x2

	mov	pB, origPB

	asr	counterL , origK, #3		// counterL = counterL / 8
        cmp	counterL,#0
	ble	.Lsgemm_kernel_L2_M2_40

.Lsgemm_kernel_L2_M2_22:

	KERNEL2x2_SUB
	KERNEL2x2_SUB
	KERNEL2x2_SUB
	KERNEL2x2_SUB

	KERNEL2x2_SUB
	KERNEL2x2_SUB
	KERNEL2x2_SUB
	KERNEL2x2_SUB

	subs	counterL, counterL, #1
	bgt	.Lsgemm_kernel_L2_M2_22


.Lsgemm_kernel_L2_M2_40:

	ands	counterL , origK, #7		// counterL = counterL % 8
	ble	.Lsgemm_kernel_L2_M2_100

.Lsgemm_kernel_L2_M2_42:

	KERNEL2x2_SUB

	subs	counterL, counterL, #1
	bgt	.Lsgemm_kernel_L2_M2_42

.Lsgemm_kernel_L2_M2_100:

	SAVE2x2

.Lsgemm_kernel_L2_M2_END:

/******************************************************************************/

.Lsgemm_kernel_L2_M1_BEGIN:

	tst	counterI, #1			// counterI = counterI % 2
	ble	.Lsgemm_kernel_L2_END

.Lsgemm_kernel_L2_M1_20:

	INIT1x2

	mov	pB, origPB

	asr	counterL , origK, #3		// counterL = counterL / 8
        cmp     counterL, #0
	ble	.Lsgemm_kernel_L2_M1_40

.Lsgemm_kernel_L2_M1_22:
	KERNEL1x2_SUB
	KERNEL1x2_SUB
	KERNEL1x2_SUB
	KERNEL1x2_SUB

	KERNEL1x2_SUB
	KERNEL1x2_SUB
	KERNEL1x2_SUB
	KERNEL1x2_SUB

	subs	counterL, counterL, #1
	bgt	.Lsgemm_kernel_L2_M1_22


.Lsgemm_kernel_L2_M1_40:

	ands	counterL , origK, #7		// counterL = counterL % 8
	ble	.Lsgemm_kernel_L2_M1_100

.Lsgemm_kernel_L2_M1_42:

	KERNEL1x2_SUB

	subs	counterL, counterL, #1
	bgt	.Lsgemm_kernel_L2_M1_42

.Lsgemm_kernel_L2_M1_100:

	SAVE1x2

.Lsgemm_kernel_L2_END:

	add	origPB, origPB, origK, lsl #3	// B = B + K * 2 * 4

/******************************************************************************/
/******************************************************************************/

.Lsgemm_kernel_L1_BEGIN:

	mov	counterJ , origN
	tst	counterJ , #1
	ble	.Lsgemm_kernel_L999 // done


	mov	pCRow0, pC			// pCRow0 = C
	add	pC , pC , LDC			// Update pC to point to next

	mov	pA, origPA			// pA = A

/******************************************************************************/

.Lsgemm_kernel_L1_M8_BEGIN:

	mov	counterI, origM
	asr	counterI, counterI, #3
	cmp	counterI, #0
	ble	.Lsgemm_kernel_L1_M4_BEGIN

.Lsgemm_kernel_L1_M8_20:

	INIT8x1

	mov	pB, origPB

	asr	counterL , origK, #3		// counterL = counterL / 8
	cmp	counterL , #0
	ble	.Lsgemm_kernel_L1_M8_40
	.align 5

.Lsgemm_kernel_L1_M8_22:
	KERNEL8x1_SUB
	KERNEL8x1_SUB
	KERNEL8x1_SUB
	KERNEL8x1_SUB

	KERNEL8x1_SUB
	KERNEL8x1_SUB
	KERNEL8x1_SUB
	KERNEL8x1_SUB

	subs	counterL, counterL, #1
	bgt	.Lsgemm_kernel_L1_M8_22


.Lsgemm_kernel_L1_M8_40:

	ands	counterL , origK, #7		// counterL = counterL % 8
	ble	.Lsgemm_kernel_L1_M8_100

.Lsgemm_kernel_L1_M8_42:

	KERNEL8x1_SUB

	subs	counterL, counterL, #1
	bgt	.Lsgemm_kernel_L1_M8_42

.Lsgemm_kernel_L1_M8_100:

	SAVE8x1

.Lsgemm_kernel_L1_M8_END:

	subs	counterI, counterI, #1
	bgt	.Lsgemm_kernel_L1_M8_20

/******************************************************************************/

.Lsgemm_kernel_L1_M4_BEGIN:

	mov	counterI, origM
	tst	counterI , #7
	ble	.Lsgemm_kernel_L1_END

	tst	counterI, #4
	ble	.Lsgemm_kernel_L1_M2_BEGIN

.Lsgemm_kernel_L1_M4_20:

	INIT4x1

	mov	pB, origPB

	asr	counterL , origK, #3		// counterL = counterL / 8
	cmp	counterL , #0
	ble	.Lsgemm_kernel_L1_M4_40
	.align 5

.Lsgemm_kernel_L1_M4_22:
	KERNEL4x1_SUB
	KERNEL4x1_SUB
	KERNEL4x1_SUB
	KERNEL4x1_SUB

	KERNEL4x1_SUB
	KERNEL4x1_SUB
	KERNEL4x1_SUB
	KERNEL4x1_SUB

	subs	counterL, counterL, #1
	bgt	.Lsgemm_kernel_L1_M4_22


.Lsgemm_kernel_L1_M4_40:

	ands	counterL , origK, #7		// counterL = counterL % 8
	ble	.Lsgemm_kernel_L1_M4_100

.Lsgemm_kernel_L1_M4_42:

	KERNEL4x1_SUB

	subs	counterL, counterL, #1
	bgt	.Lsgemm_kernel_L1_M4_42

.Lsgemm_kernel_L1_M4_100:

	SAVE4x1

.Lsgemm_kernel_L1_M4_END:

/******************************************************************************/

.Lsgemm_kernel_L1_M2_BEGIN:

	mov	counterI, origM
	tst	counterI , #3
	ble	.Lsgemm_kernel_L1_END

	tst	counterI, #2			// counterI = counterI / 2
	ble	.Lsgemm_kernel_L1_M1_BEGIN

.Lsgemm_kernel_L1_M2_20:

	INIT2x1

	mov	pB, origPB

	asr	counterL , origK, #3		// counterL = counterL / 8
	cmp	counterL , #0
	ble	.Lsgemm_kernel_L1_M2_40

.Lsgemm_kernel_L1_M2_22:

	KERNEL2x1_SUB
	KERNEL2x1_SUB
	KERNEL2x1_SUB
	KERNEL2x1_SUB

	KERNEL2x1_SUB
	KERNEL2x1_SUB
	KERNEL2x1_SUB
	KERNEL2x1_SUB

	subs	counterL, counterL, #1
	bgt	.Lsgemm_kernel_L1_M2_22


.Lsgemm_kernel_L1_M2_40:

	ands	counterL , origK, #7		// counterL = counterL % 8
	ble	.Lsgemm_kernel_L1_M2_100

.Lsgemm_kernel_L1_M2_42:

	KERNEL2x1_SUB

	subs	counterL, counterL, #1
	bgt	.Lsgemm_kernel_L1_M2_42

.Lsgemm_kernel_L1_M2_100:

	SAVE2x1

.Lsgemm_kernel_L1_M2_END:

/******************************************************************************/

.Lsgemm_kernel_L1_M1_BEGIN:

	tst	counterI, #1			// counterI = counterI % 2
	ble	.Lsgemm_kernel_L1_END

.Lsgemm_kernel_L1_M1_20:

	INIT1x1

	mov	pB, origPB

	asr	counterL , origK, #3		// counterL = counterL / 8
	cmp	counterL , #0
	ble	.Lsgemm_kernel_L1_M1_40

.Lsgemm_kernel_L1_M1_22:
	KERNEL1x1_SUB
	KERNEL1x1_SUB
	KERNEL1x1_SUB
	KERNEL1x1_SUB

	KERNEL1x1_SUB
	KERNEL1x1_SUB
	KERNEL1x1_SUB
	KERNEL1x1_SUB

	subs	counterL, counterL, #1
	bgt	.Lsgemm_kernel_L1_M1_22


.Lsgemm_kernel_L1_M1_40:

	ands	counterL , origK, #7		// counterL = counterL % 8
	ble	.Lsgemm_kernel_L1_M1_100

.Lsgemm_kernel_L1_M1_42:

	KERNEL1x1_SUB

	subs	counterL, counterL, #1
	bgt	.Lsgemm_kernel_L1_M1_42

.Lsgemm_kernel_L1_M1_100:

	SAVE1x1

.Lsgemm_kernel_L1_END:

/******************************************************************************/

.Lsgemm_kernel_L999:
	mov	x0, #0				// set return value
	ldp	d8, d9, [sp, #(0 * 16)]
	ldp	d10, d11, [sp, #(1 * 16)]
	ldp	d12, d13, [sp, #(2 * 16)]
	ldp	d14, d15, [sp, #(3 * 16)]
	ldp	d16, d17, [sp, #(4 * 16)]
	ldp	x18, x19, [sp, #(5 * 16)]
	ldp	x20, x21, [sp, #(6 * 16)]
	ldp	x22, x23, [sp, #(7 * 16)]
	ldp	x24, x25, [sp, #(8 * 16)]
	ldp	x26, x27, [sp, #(9 * 16)]
	ldr	x28, [sp, #(10 * 16)]
	add	sp, sp, #(11*16)
	ret

	EPILOGUE