--- /dev/null
+/*******************************************************************************
+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 alpha0,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 temp x7
+#define counterL x8
+#define counterI x9
+#define counterJ x10
+#define pB x11
+#define pCRow0 x12
+#define pCRow1 x13
+#define pCRow2 x14
+#define pCRow3 x15
+#define pA x16
+#define alpha x17
+
+#define alpha0 d10
+#define alphaZ z10.d
+#define alphaV0 v10.d[0]
+
+#define A_PRE_SIZE 2560
+#define B_PRE_SIZE 448
+#define C_PRE_SIZE 128
+
+// 00 origM
+// 01 origN
+// 02 origK
+// 03 origPA
+// 04 origPB
+// 05 pC
+// 06 origLDC -> LDC
+// 07 temp
+// 08 counterL
+// 09 counterI
+// 10 counterJ
+// 11 pB
+// 12 pCRow0
+// 13 pCRow1
+// 14 pCRow2
+// 15 pCRow3
+// 16 pA
+// 17
+// 18 must save
+// 19 must save
+// 20 must save
+// 21 must save
+// 22 must save
+// 23 must save
+// 24 must save
+// 25 must save
+// 26 must save
+// 27 must save
+// 28 must save
+// 29 frame
+// 30 link
+// 31 sp
+
+//v00 ALPHA -> pA0_0
+//v01 pA0_1
+//v02 pA0_2
+//v03 pA0_3
+//v04 pA0_4
+//v05 pA0_5
+//v06 pA0_6
+//v07 pA0_7
+//v08 must save pB0_0
+//v09 must save pB0_1
+//v10 must save pB0_2 --> ALPHA0
+//v11 must save pB0_3
+//v12 must save pB1_0
+//v13 must save pB1_1
+//v14 must save pB1_2
+//v15 must save pB1_3
+//v16 must save C0
+//v17 must save C1
+//v18 must save C2
+//v19 must save C3
+//v20 must save C4
+//v21 must save C5
+//v22 must save C6
+//v23 must save C7
+
+/*******************************************************************************
+* Macro definitions
+*******************************************************************************/
+
+.macro INITv1x8
+ dup z16.d, #0
+ dup z17.d, #0
+ dup z18.d, #0
+ dup z19.d, #0
+ dup z20.d, #0
+ dup z21.d, #0
+ dup z22.d, #0
+ dup z23.d, #0
+.endm
+
+.macro KERNELv1x8_I
+ ld1d z0.d, p1/z, [pA]
+ ld1d z1.d, p1/z, [pA, x18, lsl #3] // next one
+ //incb pA, all, mul #2
+ add pA, pA, x18, lsl #4 // pA = pA + cnt_active * 2 * 8
+
+ ld1rd z8.d, p0/z, [pB]
+ ld1rd z9.d, p0/z, [pB, 8]
+ ld1rd z10.d, p0/z, [pB, 16]
+ ld1rd z11.d, p0/z, [pB, 24]
+ ld1rd z12.d, p0/z, [pB, 32]
+ ld1rd z13.d, p0/z, [pB, 40]
+ ld1rd z14.d, p0/z, [pB, 48]
+ ld1rd z15.d, p0/z, [pB, 56]
+
+ add pB, pB, 64
+
+ fmla z16.d, p1/m, z0.d, z8.d
+ ld1rd z8.d, p0/z, [pB]
+ fmla z17.d, p1/m, z0.d, z9.d
+ ld1rd z9.d, p0/z, [pB, 8]
+ fmla z18.d, p1/m, z0.d, z10.d
+ ld1rd z10.d, p0/z, [pB, 16]
+ fmla z19.d, p1/m, z0.d, z11.d
+ ld1rd z11.d, p0/z, [pB, 24]
+ fmla z20.d, p1/m, z0.d, z12.d
+ ld1rd z12.d, p0/z, [pB, 32]
+ fmla z21.d, p1/m, z0.d, z13.d
+ ld1rd z13.d, p0/z, [pB, 40]
+ fmla z22.d, p1/m, z0.d, z14.d
+ ld1rd z14.d, p0/z, [pB, 48]
+ fmla z23.d, p1/m, z0.d, z15.d
+ ld1rd z15.d, p0/z, [pB, 56]
+
+ add pB, pB, 64
+.endm
+
+.macro KERNELv1x8_M1
+ ld1d z1.d, p1/z, [pA]
+ add pA, pA, x18, lsl #3 // pA = pA + cnt_active * 8
+
+ fmla z16.d, p1/m, z0.d, z8.d
+ ld1rd z8.d, p0/z, [pB]
+ fmla z17.d, p1/m, z0.d, z9.d
+ ld1rd z9.d, p0/z, [pB, 8]
+ fmla z18.d, p1/m, z0.d, z10.d
+ ld1rd z10.d, p0/z, [pB, 16]
+ fmla z19.d, p1/m, z0.d, z11.d
+ ld1rd z11.d, p0/z, [pB, 24]
+ fmla z20.d, p1/m, z0.d, z12.d
+ ld1rd z12.d, p0/z, [pB, 32]
+ fmla z21.d, p1/m, z0.d, z13.d
+ ld1rd z13.d, p0/z, [pB, 40]
+ fmla z22.d, p1/m, z0.d, z14.d
+ ld1rd z14.d, p0/z, [pB, 48]
+ fmla z23.d, p1/m, z0.d, z15.d
+ ld1rd z15.d, p0/z, [pB, 56]
+
+ add pB, pB, 64
+.endm
+
+.macro KERNELv1x8_M2
+ ld1d z0.d, p1/z, [pA]
+ add pA, pA, x18, lsl #3 // pA = pA + cnt_active * 8
+
+ fmla z16.d, p1/m, z1.d, z8.d
+ ld1rd z8.d, p0/z, [pB]
+ fmla z17.d, p1/m, z1.d, z9.d
+ ld1rd z9.d, p0/z, [pB, 8]
+ fmla z18.d, p1/m, z1.d, z10.d
+ ld1rd z10.d, p0/z, [pB, 16]
+ fmla z19.d, p1/m, z1.d, z11.d
+ ld1rd z11.d, p0/z, [pB, 24]
+ fmla z20.d, p1/m, z1.d, z12.d
+ ld1rd z12.d, p0/z, [pB, 32]
+ fmla z21.d, p1/m, z1.d, z13.d
+ ld1rd z13.d, p0/z, [pB, 40]
+ fmla z22.d, p1/m, z1.d, z14.d
+ ld1rd z14.d, p0/z, [pB, 48]
+ fmla z23.d, p1/m, z1.d, z15.d
+ ld1rd z15.d, p0/z, [pB, 56]
+
+ add pB, pB, 64
+.endm
+
+.macro KERNELv1x8_E
+ fmla z16.d, p1/m, z1.d, z8.d
+ fmla z17.d, p1/m, z1.d, z9.d
+ fmla z18.d, p1/m, z1.d, z10.d
+ fmla z19.d, p1/m, z1.d, z11.d
+ fmla z20.d, p1/m, z1.d, z12.d
+ fmla z21.d, p1/m, z1.d, z13.d
+ fmla z22.d, p1/m, z1.d, z14.d
+ fmla z23.d, p1/m, z1.d, z15.d
+.endm
+
+.macro KERNELv1x8_SUB
+ ld1d z0.d, p1/z, [pA]
+ add pA, pA, x18, lsl #3 // pA = pA + cnt_active * 8
+
+ ld1rd z8.d, p0/z, [pB]
+ ld1rd z9.d, p0/z, [pB, 8]
+ ld1rd z10.d, p0/z, [pB, 16]
+ ld1rd z11.d, p0/z, [pB, 24]
+ ld1rd z12.d, p0/z, [pB, 32]
+ ld1rd z13.d, p0/z, [pB, 40]
+ ld1rd z14.d, p0/z, [pB, 48]
+ ld1rd z15.d, p0/z, [pB, 56]
+
+ add pB, pB, 64
+
+ fmla z16.d, p1/m, z0.d, z8.d
+ fmla z17.d, p1/m, z0.d, z9.d
+ fmla z18.d, p1/m, z0.d, z10.d
+ fmla z19.d, p1/m, z0.d, z11.d
+ fmla z20.d, p1/m, z0.d, z12.d
+ fmla z21.d, p1/m, z0.d, z13.d
+ fmla z22.d, p1/m, z0.d, z14.d
+ fmla z23.d, p1/m, z0.d, z15.d
+
+.endm
+
+.macro SAVEv1x8
+ dup alphaZ, alpha
+
+ prfm PLDL2KEEP, [pCRow0, #C_PRE_SIZE]
+
+ add pCRow1, pCRow0, LDC
+ ld1d z24.d, p1/z, [pCRow0]
+ fmla z24.d, p1/m, z16.d, alphaZ
+ st1d z24.d, p1, [pCRow0]
+
+ add pCRow2, pCRow1, LDC
+ ld1d z25.d, p1/z, [pCRow1]
+ fmla z25.d, p1/m, z17.d, alphaZ
+ st1d z25.d, p1, [pCRow1]
+
+ add pCRow1, pCRow2, LDC
+ ld1d z26.d, p1/z, [pCRow2]
+ fmla z26.d, p1/m, z18.d, alphaZ
+ st1d z26.d, p1, [pCRow2]
+
+ add pCRow2, pCRow1, LDC
+ ld1d z27.d, p1/z, [pCRow1]
+ fmla z27.d, p1/m, z19.d, alphaZ
+ st1d z27.d, p1, [pCRow1]
+
+ add pCRow1, pCRow2, LDC
+ ld1d z28.d, p1/z, [pCRow2]
+ fmla z28.d, p1/m, z20.d, alphaZ
+ st1d z28.d, p1, [pCRow2]
+
+ add pCRow2, pCRow1, LDC
+ ld1d z29.d, p1/z, [pCRow1]
+ fmla z29.d, p1/m, z21.d, alphaZ
+ st1d z29.d, p1, [pCRow1]
+
+ add pCRow1, pCRow2, LDC
+ ld1d z30.d, p1/z, [pCRow2]
+ fmla z30.d, p1/m, z22.d, alphaZ
+ st1d z30.d, p1, [pCRow2]
+
+ add pCRow2, pCRow1, LDC
+ ld1d z31.d, p1/z, [pCRow1]
+ fmla z31.d, p1/m, z23.d, alphaZ
+ st1d z31.d, p1, [pCRow1]
+
+ add pCRow0, pCRow0, x18, lsl #3 // pC = pC + cnt_active * 8
+
+.endm
+
+/******************************************************************************/
+
+.macro INITv1x4
+ dup z16.d, #0
+ dup z17.d, #0
+ dup z18.d, #0
+ dup z19.d, #0
+.endm
+
+.macro KERNELv1x4_SUB
+ ld1d z0.d, p1/z, [pA]
+ add pA, pA, x18, lsl #3 // pA = pA + cnt_active * 8
+
+ ld1rd z8.d, p0/z, [pB]
+ ld1rd z9.d, p0/z, [pB, 8]
+ ld1rd z10.d, p0/z, [pB, 16]
+ ld1rd z11.d, p0/z, [pB, 24]
+
+ add pB, pB, 32
+
+ fmla z16.d, p1/m, z0.d, z8.d
+ fmla z17.d, p1/m, z0.d, z9.d
+ fmla z18.d, p1/m, z0.d, z10.d
+ fmla z19.d, p1/m, z0.d, z11.d
+
+.endm
+
+.macro SAVEv1x4
+ dup alphaZ, alpha
+
+ prfm PLDL2KEEP, [pCRow0, #C_PRE_SIZE]
+
+ add pCRow1, pCRow0, LDC
+ ld1d z24.d, p1/z, [pCRow0]
+ fmla z24.d, p1/m, z16.d, alphaZ
+ st1d z24.d, p1, [pCRow0]
+
+ add pCRow2, pCRow1, LDC
+ ld1d z25.d, p1/z, [pCRow1]
+ fmla z25.d, p1/m, z17.d, alphaZ
+ st1d z25.d, p1, [pCRow1]
+
+ add pCRow1, pCRow2, LDC
+ ld1d z26.d, p1/z, [pCRow2]
+ fmla z26.d, p1/m, z18.d, alphaZ
+ st1d z26.d, p1, [pCRow2]
+
+ add pCRow2, pCRow1, LDC
+ ld1d z27.d, p1/z, [pCRow1]
+ fmla z27.d, p1/m, z19.d, alphaZ
+ st1d z27.d, p1, [pCRow1]
+
+ add pCRow0, pCRow0, x18, lsl #3 // pC = pC + cnt_active * 8
+
+.endm
+
+/******************************************************************************/
+
+.macro INITv1x2
+ dup z16.d, #0
+ dup z17.d, #0
+.endm
+
+.macro KERNELv1x2_SUB
+ ld1d z0.d, p1/z, [pA]
+ add pA, pA, x18, lsl #3 // pA = pA + cnt_active * 8
+
+ ld1rd z8.d, p0/z, [pB]
+ ld1rd z9.d, p0/z, [pB, 8]
+
+ add pB, pB, 16
+
+ fmla z16.d, p1/m, z0.d, z8.d
+ fmla z17.d, p1/m, z0.d, z9.d
+
+.endm
+
+.macro SAVEv1x2
+ dup alphaZ, alpha
+
+ prfm PLDL2KEEP, [pCRow0, #C_PRE_SIZE]
+
+ add pCRow1, pCRow0, LDC
+ ld1d z24.d, p1/z, [pCRow0]
+ fmla z24.d, p1/m, z16.d, alphaZ
+ st1d z24.d, p1, [pCRow0]
+
+ add pCRow2, pCRow1, LDC
+ ld1d z25.d, p1/z, [pCRow1]
+ fmla z25.d, p1/m, z17.d, alphaZ
+ st1d z25.d, p1, [pCRow1]
+
+ add pCRow0, pCRow0, x18, lsl #3 // pC = pC + cnt_active * 8
+
+.endm
+
+/******************************************************************************/
+
+.macro INITv1x1
+ dup z16.d, #0
+.endm
+
+.macro KERNELv1x1_SUB
+ ld1d z0.d, p1/z, [pA]
+ add pA, pA, x18, lsl #3 // pA = pA + cnt_active * 8
+
+ ld1rd z8.d, p0/z, [pB]
+
+ add pB, pB, 8
+
+ fmla z16.d, p1/m, z0.d, z8.d
+
+.endm
+
+.macro SAVEv1x1
+ dup alphaZ, alpha
+
+ prfm PLDL2KEEP, [pCRow0, #C_PRE_SIZE]
+
+ add pCRow1, pCRow0, LDC
+ ld1d z24.d, p1/z, [pCRow0]
+ fmla z24.d, p1/m, z16.d, alphaZ
+ st1d z24.d, p1, [pCRow0]
+
+
+ add pCRow0, pCRow0, x18, lsl #3 // pC = pC + cnt_active * 8
+
+.endm
+
+
+/*******************************************************************************
+* End of macro definitions
+*******************************************************************************/
+
+ PROLOGUE
+
+ .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)]
+
+ prfm PLDL1KEEP, [origPB]
+ prfm PLDL1KEEP, [origPA]
+
+ fmov alpha, d0
+
+ lsl LDC, LDC, #3 // ldc = ldc * 8
+ ptrue p0.d // create true predicate
+
+ mov pB, origPB
+
+ mov counterJ, origN
+ asr counterJ, counterJ, #3 // J = J / 8
+ cmp counterJ, #0
+ ble .Ldgemm_kernel_L4_BEGIN
+
+/******************************************************************************/
+
+ .align 5
+.Ldgemm_kernel_L8_BEGIN:
+ mov pCRow0, pC
+
+ add pC, pCRow0, LDC, lsl #3 // add 8 x LDC
+
+ mov pA, origPA // pA = start of A array
+
+.Ldgemm_kernel_L8_Mv1_BEGIN:
+
+ mov counterI, #0
+ whilelt p1.d, counterI, origM //SVE instruction
+ cntp x18, p0, p1.d
+ /* mov counterI, origM */
+ /* asr counterI, counterI, #3 // counterI = counterI / 8 */
+ /* cmp counterI, #0 */
+ /* ble .Ldgemm_kernel_L4_M4_BEGIN */
+
+ .align 5
+.Ldgemm_kernel_L8_Mv1_20:
+
+ mov pB, origPB
+ INITv1x8 // fill with zeros
+
+ asr counterL , origK, #3 // L = K / 8
+ cmp counterL , #2 // is there at least 4 to do?
+ blt .Ldgemm_kernel_L8_Mv1_32
+
+ KERNELv1x8_I
+ KERNELv1x8_M2
+ KERNELv1x8_M1
+ KERNELv1x8_M2
+ KERNELv1x8_M1
+ KERNELv1x8_M2
+ KERNELv1x8_M1
+ KERNELv1x8_M2
+
+ subs counterL, counterL, #2 // subtract 2
+ ble .Ldgemm_kernel_L8_Mv1_22a
+
+ .align 5
+.Ldgemm_kernel_L8_Mv1_22:
+
+ KERNELv1x8_M1
+ KERNELv1x8_M2
+ KERNELv1x8_M1
+ KERNELv1x8_M2
+ KERNELv1x8_M1
+ KERNELv1x8_M2
+ KERNELv1x8_M1
+ KERNELv1x8_M2
+
+ subs counterL, counterL, #1
+ bgt .Ldgemm_kernel_L8_Mv1_22
+
+ .align 5
+.Ldgemm_kernel_L8_Mv1_22a:
+
+ KERNELv1x8_M1
+ KERNELv1x8_M2
+ KERNELv1x8_M1
+ KERNELv1x8_M2
+ KERNELv1x8_M1
+ KERNELv1x8_M2
+ KERNELv1x8_M1
+ KERNELv1x8_E
+
+ b .Ldgemm_kernel_L8_Mv1_44
+
+ .align 5
+.Ldgemm_kernel_L8_Mv1_32:
+
+ tst counterL, #1
+ ble .Ldgemm_kernel_L8_Mv1_40
+
+ KERNELv1x8_I
+ KERNELv1x8_M2
+ KERNELv1x8_M1
+ KERNELv1x8_M2
+ KERNELv1x8_M1
+ KERNELv1x8_M2
+ KERNELv1x8_M1
+ KERNELv1x8_E
+
+
+ b .Ldgemm_kernel_L8_Mv1_44
+
+.Ldgemm_kernel_L8_Mv1_40:
+
+ INITv1x8
+
+.Ldgemm_kernel_L8_Mv1_44:
+
+ ands counterL , origK, #7
+ ble .Ldgemm_kernel_L8_Mv1_100
+
+ .align 5
+.Ldgemm_kernel_L8_Mv1_46:
+
+ KERNELv1x8_SUB
+
+ subs counterL, counterL, #1
+ bne .Ldgemm_kernel_L8_Mv1_46
+
+.Ldgemm_kernel_L8_Mv1_100:
+ prfm PLDL1KEEP, [pA]
+ prfm PLDL1KEEP, [pA, #64]
+ prfm PLDL1KEEP, [origPB]
+
+ SAVEv1x8
+
+.Ldgemm_kernel_L8_Mv1_END:
+
+ incd counterI
+ whilelt p1.d, counterI, origM //SVE instruction
+ cntp x18, p0, p1.d
+ b.any .Ldgemm_kernel_L8_Mv1_20
+
+.Ldgemm_kernel_L8_END:
+
+ lsl temp, origK, #6
+ add origPB, origPB, temp // B = B + K * 8 * 8
+
+ subs counterJ, counterJ , #1 // j--
+ bgt .Ldgemm_kernel_L8_BEGIN
+
+/******************************************************************************/
+/******************************************************************************/
+
+ .align 5
+.Ldgemm_kernel_L4_BEGIN:
+
+ mov counterJ , origN
+ tst counterJ , #4
+ ble .Ldgemm_kernel_L2_BEGIN
+
+
+ mov pCRow0, pC
+
+ add pC, pCRow0, LDC, lsl #2 // add 4 x LDC
+
+ mov pA, origPA // pA = start of A array
+
+.Ldgemm_kernel_L4_Mv1_BEGIN:
+
+ mov counterI, #0
+ whilelt p1.d, counterI, origM //SVE instruction
+ cntp x18, p0, p1.d
+
+ .align 5
+.Ldgemm_kernel_L4_Mv1_20:
+
+ mov pB, origPB
+ INITv1x4 // fill with zeros
+
+ asr counterL , origK, #3 // L = K / 8
+ cmp counterL , #0 // is there at least 4 to do?
+ blt .Ldgemm_kernel_L4_Mv1_44
+
+ .align 5
+.Ldgemm_kernel_L4_Mv1_22:
+
+ KERNELv1x4_SUB
+ KERNELv1x4_SUB
+ KERNELv1x4_SUB
+ KERNELv1x4_SUB
+ KERNELv1x4_SUB
+ KERNELv1x4_SUB
+ KERNELv1x4_SUB
+ KERNELv1x4_SUB
+
+ subs counterL, counterL, #1
+ bgt .Ldgemm_kernel_L4_Mv1_22
+
+.Ldgemm_kernel_L4_Mv1_44:
+
+ ands counterL , origK, #7
+ ble .Ldgemm_kernel_L4_Mv1_100
+
+ .align 5
+.Ldgemm_kernel_L4_Mv1_46:
+
+ KERNELv1x4_SUB
+
+ subs counterL, counterL, #1
+ bne .Ldgemm_kernel_L4_Mv1_46
+
+.Ldgemm_kernel_L4_Mv1_100:
+ prfm PLDL1KEEP, [pA]
+ prfm PLDL1KEEP, [pA, #64]
+ prfm PLDL1KEEP, [origPB]
+
+ SAVEv1x4
+
+.Ldgemm_kernel_L4_Mv1_END:
+
+ incd counterI
+ whilelt p1.d, counterI, origM //SVE instruction
+ cntp x18, p0, p1.d
+ b.any .Ldgemm_kernel_L4_Mv1_20
+
+
+.Ldgemm_kernel_L4_END:
+ add origPB, origPB, origK, lsl #5 // B = B + K * 4 * 8
+
+/******************************************************************************/
+/******************************************************************************/
+
+ .align 5
+.Ldgemm_kernel_L2_BEGIN:
+
+ mov counterJ , origN
+ tst counterJ , #2
+ ble .Ldgemm_kernel_L1_BEGIN
+
+ mov pCRow0, pC
+
+ add pC, pCRow0, LDC, lsl #1 // add 2 x LDC
+
+ mov pA, origPA // pA = start of A array
+
+.Ldgemm_kernel_L2_Mv1_BEGIN:
+
+ mov counterI, #0
+ whilelt p1.d, counterI, origM //SVE instruction
+ cntp x18, p0, p1.d
+
+ .align 5
+.Ldgemm_kernel_L2_Mv1_20:
+
+ mov pB, origPB
+ INITv1x2 // fill with zeros
+
+ asr counterL , origK, #3 // L = K / 8
+ cmp counterL , #0 // is there at least 4 to do?
+ blt .Ldgemm_kernel_L2_Mv1_44
+
+ .align 5
+.Ldgemm_kernel_L2_Mv1_22:
+
+ KERNELv1x2_SUB
+ KERNELv1x2_SUB
+ KERNELv1x2_SUB
+ KERNELv1x2_SUB
+ KERNELv1x2_SUB
+ KERNELv1x2_SUB
+ KERNELv1x2_SUB
+ KERNELv1x2_SUB
+
+ subs counterL, counterL, #1
+ bgt .Ldgemm_kernel_L2_Mv1_22
+
+.Ldgemm_kernel_L2_Mv1_44:
+
+ ands counterL , origK, #7
+ ble .Ldgemm_kernel_L2_Mv1_100
+
+ .align 5
+.Ldgemm_kernel_L2_Mv1_46:
+
+ KERNELv1x2_SUB
+
+ subs counterL, counterL, #1
+ bne .Ldgemm_kernel_L2_Mv1_46
+
+.Ldgemm_kernel_L2_Mv1_100:
+ prfm PLDL1KEEP, [pA]
+ prfm PLDL1KEEP, [pA, #64]
+ prfm PLDL1KEEP, [origPB]
+
+ SAVEv1x2
+
+.Ldgemm_kernel_L2_Mv1_END:
+
+ incd counterI
+ whilelt p1.d, counterI, origM //SVE instruction
+ cntp x18, p0, p1.d
+ b.any .Ldgemm_kernel_L2_Mv1_20
+
+
+.Ldgemm_kernel_L2_END:
+ add origPB, origPB, origK, lsl #4 // B = B + K * 2 * 8
+
+/******************************************************************************/
+/******************************************************************************/
+
+ .align 5
+.Ldgemm_kernel_L1_BEGIN:
+
+ mov counterJ , origN
+ tst counterJ , #1
+ ble .Ldgemm_kernel_L999 // done
+
+ mov pCRow0, pC
+
+ add pC, pCRow0, LDC, lsl #1 // add 2 x LDC
+
+ mov pA, origPA // pA = start of A array
+
+.Ldgemm_kernel_L1_Mv1_BEGIN:
+
+ mov counterI, #0
+ whilelt p1.d, counterI, origM //SVE instruction
+ cntp x18, p0, p1.d
+
+ .align 5
+.Ldgemm_kernel_L1_Mv1_20:
+
+ mov pB, origPB
+ INITv1x1 // fill with zeros
+
+ asr counterL , origK, #3 // L = K / 8
+ cmp counterL , #0 // is there at least 4 to do?
+ blt .Ldgemm_kernel_L1_Mv1_44
+
+ .align 5
+.Ldgemm_kernel_L1_Mv1_22:
+
+ KERNELv1x1_SUB
+ KERNELv1x1_SUB
+ KERNELv1x1_SUB
+ KERNELv1x1_SUB
+ KERNELv1x1_SUB
+ KERNELv1x1_SUB
+ KERNELv1x1_SUB
+ KERNELv1x1_SUB
+
+ subs counterL, counterL, #1
+ bgt .Ldgemm_kernel_L1_Mv1_22
+
+.Ldgemm_kernel_L1_Mv1_44:
+
+ ands counterL , origK, #7
+ ble .Ldgemm_kernel_L1_Mv1_100
+
+ .align 5
+.Ldgemm_kernel_L1_Mv1_46:
+
+ KERNELv1x1_SUB
+
+ subs counterL, counterL, #1
+ bne .Ldgemm_kernel_L1_Mv1_46
+
+.Ldgemm_kernel_L1_Mv1_100:
+ prfm PLDL1KEEP, [pA]
+ prfm PLDL1KEEP, [pA, #64]
+ prfm PLDL1KEEP, [origPB]
+
+ SAVEv1x1
+
+.Ldgemm_kernel_L1_Mv1_END:
+
+ incd counterI
+ whilelt p1.d, counterI, origM //SVE instruction
+ cntp x18, p0, p1.d
+ b.any .Ldgemm_kernel_L1_Mv1_20
+
+
+.Ldgemm_kernel_L1_END:
+
+/******************************************************************************/
+
+.Ldgemm_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
+