diff options
Diffstat (limited to 'src/core/NEON/kernels/arm_conv/depthwise/kernels/sme2_fp32_planar_5x5_s2_4rows_mla_za/generic.cpp')
| -rw-r--r-- | src/core/NEON/kernels/arm_conv/depthwise/kernels/sme2_fp32_planar_5x5_s2_4rows_mla_za/generic.cpp | 1450 |
1 files changed, 725 insertions, 725 deletions
diff --git a/src/core/NEON/kernels/arm_conv/depthwise/kernels/sme2_fp32_planar_5x5_s2_4rows_mla_za/generic.cpp b/src/core/NEON/kernels/arm_conv/depthwise/kernels/sme2_fp32_planar_5x5_s2_4rows_mla_za/generic.cpp index 8920b3b749..755265835d 100644 --- a/src/core/NEON/kernels/arm_conv/depthwise/kernels/sme2_fp32_planar_5x5_s2_4rows_mla_za/generic.cpp +++ b/src/core/NEON/kernels/arm_conv/depthwise/kernels/sme2_fp32_planar_5x5_s2_4rows_mla_za/generic.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2022 Arm Limited. + * Copyright (c) 2022-2023 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -69,1100 +69,1100 @@ void sme2_fp32_planar_5x5_s2_4rows_mla_za_impl( Args args = { inptr, ld_in_vl, pad_top, 11u - std::min(11u, pad_top + valid_input_rows), pad_left, weights, bias, valid_input_cols, output_cols, outptrs, outlds, outvllds, start_channel, valid_channels, act_min, act_max }; __asm__ __volatile__( - "ldr x6, [%x[args], %[offsetof_Args_pad_bottom]]\n" - "mov x19, #0xb\n" + "ldr x5, [%x[args], %[offsetof_Args_pad_bottom]]\n" + "mov x20, #0xb\n" ".inst 0xd503477f // SMSTART ZA\n" - "sub x19, x19, x6\n" - "ldr x7, [%x[args], %[offsetof_Args_pad_top]]\n" + "sub x20, x20, x5\n" + "ldr x6, [%x[args], %[offsetof_Args_pad_top]]\n" "ptrue p2.b\n" ".inst 0x25207812 // ptrue pn10.b\n" "ld1rw { z0.s }, p2/Z, [%x[args], %[offsetof_Args_clamp_min]]\n" - "ldr x17, [%x[args], %[offsetof_Args_n_channels]]\n" - "whilelt p1.s, XZR, x17\n" - "whilelt p9.s, XZR, x19\n" + "ldr x7, [%x[args], %[offsetof_Args_n_channels]]\n" + "whilelt p1.s, XZR, x7\n" + "whilelt p9.s, XZR, x20\n" "ld1rw { z17.s }, p2/Z, [%x[args], %[offsetof_Args_clamp_max]]\n" - "whilelt p8.s, XZR, x7\n" + "whilelt p8.s, XZR, x6\n" "eor p8.b, p2/Z, p8.b, p9.b\n" - "ldr x16, [%x[args], %[offsetof_Args_current_channel]]\n" + "ldr x17, [%x[args], %[offsetof_Args_current_channel]]\n" "1:" // Channel loop - "ldr x19, [%x[args], %[offsetof_Args_bias]]\n" + "ldr x20, [%x[args], %[offsetof_Args_bias]]\n" "fmov z28.s, #0x0\n" - "cbz x19, 2f\n" - "ld1w { z28.s }, p1/Z, [x19, x16, LSL #2]\n" + "cbz x20, 2f\n" + "ld1w { z28.s }, p1/Z, [x20, x17, LSL #2]\n" "2:" // Load bias: Done - "ldr x15, [%x[args], %[offsetof_Args_input_cols]]\n" - "sub x19, x15, #0x1\n" - "orr x22, x19, %x[ld_in_col], LSL #18\n" + "ldr x16, [%x[args], %[offsetof_Args_input_cols]]\n" + "sub x20, x16, #0x1\n" + "orr x23, x20, %x[ld_in_col], LSL #18\n" "mov z29.d, z28.d\n" - "ldr x14, [%x[args], %[offsetof_Args_weights]]\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" - "orr x22, x17, x22, LSL #20\n" - "mov x21, #0xb\n" - "ldr x13, [%x[args], %[offsetof_Args_inptr]]\n" - ".inst 0xa04149c6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "add x20, x7, x6\n" - "lsl x19, %x[ld_in_row], #0x2\n" - "ld1w { z8.s }, p2/Z, [x14, #4, MUL VL]\n" - "addvl x14, x14, #5\n" + "ldr x15, [%x[args], %[offsetof_Args_weights]]\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" + "orr x23, x7, x23, LSL #20\n" + "mov x22, #0xb\n" + "ldr x14, [%x[args], %[offsetof_Args_inptr]]\n" + ".inst 0xa04149e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "add x21, x6, x5\n" + "lsl x20, %x[ld_in_row], #0x2\n" + "ld1w { z8.s }, p2/Z, [x15, #4, MUL VL]\n" + "addvl x15, x15, #5\n" "mov z30.d, z28.d\n" "mov z31.d, z28.d\n" - ".inst 0xa14049c1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x14]\n" + ".inst 0xa14049e1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x15]\n" "mov x8, #0x0\n" - "ldr x11, [%x[args], %[offsetof_Args_output_cols]]\n" - "lsl x22, x22, #0x2\n" - ".inst 0xa14149c2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "sub x21, x21, x20\n" - "madd x19, x19, x7, x13\n" - "ld1w { z3.s }, p2/Z, [x14, #4, MUL VL]\n" - "addvl x14, x14, #5\n" + "ldr x13, [%x[args], %[offsetof_Args_output_cols]]\n" + "lsl x23, x23, #0x2\n" + ".inst 0xa14149e2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "sub x22, x22, x21\n" + "madd x20, x20, x6, x14\n" + "ld1w { z3.s }, p2/Z, [x15, #4, MUL VL]\n" + "addvl x15, x15, #5\n" "3:" // Issue prefetches - "subs x21, x21, #0x1\n" - ".inst 0xf8b64a7c // rprfm pldstrm, x22, [x19]\n" - "add x19, x19, %x[ld_in_col], LSL #2\n" + "subs x22, x22, #0x1\n" + ".inst 0xf8b74a9c // rprfm pldstrm, x23, [x20]\n" + "add x20, x20, %x[ld_in_col], LSL #2\n" "bgt 3b\n" - "ldr x10, [%x[args], %[offsetof_Args_outptrs]]\n" - "lsl x19, %x[ld_in_row], #0x2\n" - "msub x13, x7, x19, x13\n" + "ldr x11, [%x[args], %[offsetof_Args_outptrs]]\n" + "lsl x20, %x[ld_in_row], #0x2\n" + "msub x14, x6, x20, x14\n" ".inst 0xc0040f80 // mova za.d[x8, #0], { z28.d-z31.d }\n" - "ldr x19, [%x[args], %[offsetof_Args_ld_out_cols]]\n" + "ldr x20, [%x[args], %[offsetof_Args_ld_out_cols]]\n" ".inst 0xc0040f81 // mova za.d[x8, #1], { z28.d-z31.d }\n" - "mov x21, #0x4\n" - "ldp x9, x28, [x10], #0x10\n" + "mov x22, #0x4\n" + "ldp x10, x9, [x11], #0x10\n" ".inst 0xc0040f82 // mova za.d[x8, #2], { z28.d-z31.d }\n" - "ldp x27, x26, [x19], #0x10\n" + "ldp x28, x27, [x20], #0x10\n" ".inst 0xc0040f83 // mova za.d[x8, #3], { z28.d-z31.d }\n" - "ldr x20, [%x[args], %[offsetof_Args_pad_left]]\n" + "ldr x21, [%x[args], %[offsetof_Args_pad_left]]\n" ".inst 0xc0040f84 // mova za.d[x8, #4], { z28.d-z31.d }\n" - "ldp x25, x24, [x10], #0x10\n" - "ldp x23, x22, [x19], #0x10\n" - "cbz x20, 5f\n" - "cmp x20, x21\n" - "csel x19, x20, x21, LT\n" - "sub x20, x20, x19\n" - "sub x21, x21, x19\n" - "cbz x20, 5f\n" + "ldp x26, x25, [x11], #0x10\n" + "ldp x24, x23, [x20], #0x10\n" + "cbz x21, 5f\n" + "cmp x21, x22\n" + "csel x20, x21, x22, LT\n" + "sub x21, x21, x20\n" + "sub x22, x22, x20\n" + "cbz x21, 5f\n" ".inst 0xc0060c18 // mova { z24.d-z27.d }, za.d[x8, #0]\n" - "and x21, x20, #0x1\n" - "add x20, x20, #0x1\n" + "and x22, x21, #0x1\n" + "add x21, x21, #0x1\n" ".inst 0xc1b1c818 // fclamp { z24.s-z27.s }, z0.s, z17.s\n" - "lsr x20, x20, #0x1\n" - "sub x11, x11, x20\n" + "lsr x21, x21, #0x1\n" + "sub x13, x13, x21\n" "4:" // Left padding - "subs x20, x20, #0x1\n" - "st1w { z24.s }, p1, [x9]\n" + "subs x21, x21, #0x1\n" + "st1w { z24.s }, p1, [x10]\n" + "add x10, x10, x28, LSL #2\n" + "st1w { z25.s }, p1, [x9]\n" "add x9, x9, x27, LSL #2\n" - "st1w { z25.s }, p1, [x28]\n" - "add x28, x28, x26, LSL #2\n" - "st1w { z26.s }, p1, [x25]\n" + "st1w { z26.s }, p1, [x26]\n" + "add x26, x26, x24, LSL #2\n" + "st1w { z27.s }, p1, [x25]\n" "add x25, x25, x23, LSL #2\n" - "st1w { z27.s }, p1, [x24]\n" - "add x24, x24, x22, LSL #2\n" "bgt 4b\n" "5:" // Left padding: End - "adds XZR, x7, x6\n" + "adds XZR, x6, x5\n" "bne 12f\n" - "cbz x21, 10f\n" - "cmp x21, #0x1\n" - "sub x15, x15, x21\n" + "cbz x22, 10f\n" + "cmp x22, #0x1\n" + "sub x16, x16, x22\n" "beq 9f\n" - "cmp x21, #0x2\n" + "cmp x22, #0x2\n" "beq 8f\n" - "cmp x21, #0x3\n" + "cmp x22, #0x3\n" "beq 7f\n" "6:" // Unpadded: 4 priming loads - "add x19, x13, %x[ld_in_row], LSL #2\n" - "ld1w { z11.s }, p1/Z, [x13]\n" - "add x13, x13, %x[ld_in_col], LSL #2\n" - "ld1w { z21.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" - "ld1w { z12.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" - "ld1w { z22.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" - "ld1w { z13.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" - "ld1w { z23.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" - "ld1w { z14.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "add x20, x14, %x[ld_in_row], LSL #2\n" + "ld1w { z11.s }, p1/Z, [x14]\n" + "add x14, x14, %x[ld_in_col], LSL #2\n" + "ld1w { z21.s }, p1/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z12.s }, p1/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z22.s }, p1/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z13.s }, p1/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z23.s }, p1/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z14.s }, p1/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0xc1341960 // fmla za.s[x8, 0], { z11.s-z14.s }, z4.s\n" - "ld1w { z24.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z24.s }, p1/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0xc1311aa0 // fmla za.s[x8, 0], { z21.s-z24.s }, z1.s\n" - "ld1w { z15.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" - "addvl x14, x14, #5\n" + "ld1w { z15.s }, p1/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" + "addvl x15, x15, #5\n" ".inst 0xc1341980 // fmla za.s[x8, 0], { z12.s-z15.s }, z4.s\n" - "ld1w { z25.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" - ".inst 0xa14049c1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x14]\n" - "addvl x14, x14, #5\n" + "ld1w { z25.s }, p1/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" + ".inst 0xa14049e1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x15]\n" + "addvl x15, x15, #5\n" ".inst 0xc1311ac0 // fmla za.s[x8, 0], { z22.s-z25.s }, z1.s\n" - "ld1w { z16.s }, p1/Z, [x19]\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" - "ldr x14, [%x[args], %[offsetof_Args_weights]]\n" + "ld1w { z16.s }, p1/Z, [x20]\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" + "ldr x15, [%x[args], %[offsetof_Args_weights]]\n" ".inst 0xc13419a0 // fmla za.s[x8, 0], { z13.s-z16.s }, z4.s\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" - "addvl x14, x14, #5\n" - ".inst 0xa14049c1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x14]\n" - "addvl x14, x14, #5\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" + "addvl x15, x15, #5\n" + ".inst 0xa14049e1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x15]\n" + "addvl x15, x15, #5\n" "7:" // Unpadded: 3 priming loads - "add x19, x13, %x[ld_in_row], LSL #2\n" - "ld1w { z11.s }, p1/Z, [x13]\n" - "add x13, x13, %x[ld_in_col], LSL #2\n" - "ld1w { z21.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" - "ld1w { z12.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" - "ld1w { z22.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" - "ld1w { z13.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" - "ld1w { z23.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" - "ld1w { z14.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "add x20, x14, %x[ld_in_row], LSL #2\n" + "ld1w { z11.s }, p1/Z, [x14]\n" + "add x14, x14, %x[ld_in_col], LSL #2\n" + "ld1w { z21.s }, p1/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z12.s }, p1/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z22.s }, p1/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z13.s }, p1/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z23.s }, p1/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z14.s }, p1/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0xc1351960 // fmla za.s[x8, 0], { z11.s-z14.s }, z5.s\n" - "ld1w { z24.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z24.s }, p1/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0xc1391aa0 // fmla za.s[x8, 0], { z21.s-z24.s }, z9.s\n" - "ld1w { z15.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" - "addvl x14, x14, #5\n" + "ld1w { z15.s }, p1/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" + "addvl x15, x15, #5\n" ".inst 0xc1351980 // fmla za.s[x8, 0], { z12.s-z15.s }, z5.s\n" - "ld1w { z25.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" - ".inst 0xa14049c1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x14]\n" - "addvl x14, x14, #5\n" + "ld1w { z25.s }, p1/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" + ".inst 0xa14049e1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x15]\n" + "addvl x15, x15, #5\n" ".inst 0xc1391ac0 // fmla za.s[x8, 0], { z22.s-z25.s }, z9.s\n" - "ld1w { z16.s }, p1/Z, [x19]\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" - "ldr x14, [%x[args], %[offsetof_Args_weights]]\n" + "ld1w { z16.s }, p1/Z, [x20]\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" + "ldr x15, [%x[args], %[offsetof_Args_weights]]\n" ".inst 0xc13519a0 // fmla za.s[x8, 0], { z13.s-z16.s }, z5.s\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" - ".inst 0xa04149c6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "addvl x14, x14, #5\n" - ".inst 0xa14049c1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x14]\n" - ".inst 0xa14149c2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "addvl x14, x14, #5\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" + ".inst 0xa04149e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "addvl x15, x15, #5\n" + ".inst 0xa14049e1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x15]\n" + ".inst 0xa14149e2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "addvl x15, x15, #5\n" "8:" // Unpadded: 2 priming loads - "add x19, x13, %x[ld_in_row], LSL #2\n" - "ld1w { z11.s }, p1/Z, [x13]\n" - "add x13, x13, %x[ld_in_col], LSL #2\n" - "ld1w { z21.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" - "ld1w { z12.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" - "ld1w { z22.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" - "ld1w { z13.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" - "ld1w { z23.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" - "ld1w { z14.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "add x20, x14, %x[ld_in_row], LSL #2\n" + "ld1w { z11.s }, p1/Z, [x14]\n" + "add x14, x14, %x[ld_in_col], LSL #2\n" + "ld1w { z21.s }, p1/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z12.s }, p1/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z22.s }, p1/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z13.s }, p1/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z23.s }, p1/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z14.s }, p1/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0xc1361960 // fmla za.s[x8, 0], { z11.s-z14.s }, z6.s\n" ".inst 0xc1341961 // fmla za.s[x8, 1], { z11.s-z14.s }, z4.s\n" - "ld1w { z24.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z24.s }, p1/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0xc1321aa0 // fmla za.s[x8, 0], { z21.s-z24.s }, z2.s\n" - "ld1w { z15.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z15.s }, p1/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0xc1311aa1 // fmla za.s[x8, 1], { z21.s-z24.s }, z1.s\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" - ".inst 0xa04149c6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "addvl x14, x14, #5\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" + ".inst 0xa04149e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "addvl x15, x15, #5\n" ".inst 0xc1361980 // fmla za.s[x8, 0], { z12.s-z15.s }, z6.s\n" ".inst 0xc1341981 // fmla za.s[x8, 1], { z12.s-z15.s }, z4.s\n" - "ld1w { z25.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" - ".inst 0xa14049c1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x14]\n" + "ld1w { z25.s }, p1/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" + ".inst 0xa14049e1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x15]\n" ".inst 0xc1311ac1 // fmla za.s[x8, 1], { z22.s-z25.s }, z1.s\n" - ".inst 0xa14149c2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "addvl x14, x14, #5\n" + ".inst 0xa14149e2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "addvl x15, x15, #5\n" ".inst 0xc1321ac0 // fmla za.s[x8, 0], { z22.s-z25.s }, z2.s\n" - "ld1w { z16.s }, p1/Z, [x19]\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" + "ld1w { z16.s }, p1/Z, [x20]\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" ".inst 0xc13419a1 // fmla za.s[x8, 1], { z13.s-z16.s }, z4.s\n" - ".inst 0xa04149c6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "ldr x14, [%x[args], %[offsetof_Args_weights]]\n" + ".inst 0xa04149e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "ldr x15, [%x[args], %[offsetof_Args_weights]]\n" ".inst 0xc13619a0 // fmla za.s[x8, 0], { z13.s-z16.s }, z6.s\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" - ".inst 0xa04149c6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "addvl x14, x14, #5\n" - ".inst 0xa14049c1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x14]\n" - ".inst 0xa14149c2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "addvl x14, x14, #5\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" + ".inst 0xa04149e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "addvl x15, x15, #5\n" + ".inst 0xa14049e1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x15]\n" + ".inst 0xa14149e2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "addvl x15, x15, #5\n" "9:" // Unpadded: 1 priming loads - "add x19, x13, %x[ld_in_row], LSL #2\n" - "ld1w { z11.s }, p1/Z, [x13]\n" - "add x13, x13, %x[ld_in_col], LSL #2\n" - "ld1w { z21.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" - "ld1w { z12.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" - "ld1w { z22.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" - "ld1w { z13.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" - "ld1w { z23.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" - "ld1w { z14.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" - ".inst 0xc1371960 // fmla za.s[x8, 0], { z11.s-z14.s }, z7.s\n" - ".inst 0xc1351961 // fmla za.s[x8, 1], { z11.s-z14.s }, z5.s\n" - "ld1w { z24.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" - ".inst 0xc13a1aa0 // fmla za.s[x8, 0], { z21.s-z24.s }, z10.s\n" - "ld1w { z15.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" - ".inst 0xc1391aa1 // fmla za.s[x8, 1], { z21.s-z24.s }, z9.s\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" - ".inst 0xa04149c6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "addvl x14, x14, #5\n" - ".inst 0xc1371980 // fmla za.s[x8, 0], { z12.s-z15.s }, z7.s\n" - ".inst 0xc1351981 // fmla za.s[x8, 1], { z12.s-z15.s }, z5.s\n" - "ld1w { z25.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" - ".inst 0xa14049c1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x14]\n" - ".inst 0xc1391ac1 // fmla za.s[x8, 1], { z22.s-z25.s }, z9.s\n" - ".inst 0xa14149c2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "addvl x14, x14, #5\n" - ".inst 0xc13a1ac0 // fmla za.s[x8, 0], { z22.s-z25.s }, z10.s\n" - "ld1w { z16.s }, p1/Z, [x19]\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" - ".inst 0xc13519a1 // fmla za.s[x8, 1], { z13.s-z16.s }, z5.s\n" - ".inst 0xa04149c6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "ldr x14, [%x[args], %[offsetof_Args_weights]]\n" - ".inst 0xc13719a0 // fmla za.s[x8, 0], { z13.s-z16.s }, z7.s\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" - ".inst 0xa04149c6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "ld1w { z8.s }, p2/Z, [x14, #4, MUL VL]\n" - "addvl x14, x14, #5\n" - ".inst 0xa14049c1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x14]\n" - ".inst 0xa14149c2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "ld1w { z3.s }, p2/Z, [x14, #4, MUL VL]\n" - "addvl x14, x14, #5\n" - "10:" // Unpadded: 0 priming loads - "cmp x15, #0x2\n" - "blt 20f\n" - "add x20, x13, %x[ld_in_row], LSL #2\n" - "ld1w { z11.s }, p1/Z, [x13]\n" - "sub x15, x15, #0x2\n" + "add x20, x14, %x[ld_in_row], LSL #2\n" + "ld1w { z11.s }, p1/Z, [x14]\n" + "add x14, x14, %x[ld_in_col], LSL #2\n" "ld1w { z21.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" - "sub x11, x11, #0x1\n" "ld1w { z12.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" - "lsr x19, x15, #0x1\n" "ld1w { z22.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" - "cmp x19, x11\n" "ld1w { z13.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" - "csel x21, x19, x11, LT\n" "ld1w { z23.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" - "add x13, x13, %x[ld_in_col], LSL #2\n" "ld1w { z14.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" - "and x15, x15, #0x1\n" + ".inst 0xc1371960 // fmla za.s[x8, 0], { z11.s-z14.s }, z7.s\n" + ".inst 0xc1351961 // fmla za.s[x8, 1], { z11.s-z14.s }, z5.s\n" "ld1w { z24.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" - "sub x11, x11, x21\n" + ".inst 0xc13a1aa0 // fmla za.s[x8, 0], { z21.s-z24.s }, z10.s\n" "ld1w { z15.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" + ".inst 0xc1391aa1 // fmla za.s[x8, 1], { z21.s-z24.s }, z9.s\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" + ".inst 0xa04149e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "addvl x15, x15, #5\n" + ".inst 0xc1371980 // fmla za.s[x8, 0], { z12.s-z15.s }, z7.s\n" + ".inst 0xc1351981 // fmla za.s[x8, 1], { z12.s-z15.s }, z5.s\n" "ld1w { z25.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" + ".inst 0xa14049e1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x15]\n" + ".inst 0xc1391ac1 // fmla za.s[x8, 1], { z22.s-z25.s }, z9.s\n" + ".inst 0xa14149e2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "addvl x15, x15, #5\n" + ".inst 0xc13a1ac0 // fmla za.s[x8, 0], { z22.s-z25.s }, z10.s\n" "ld1w { z16.s }, p1/Z, [x20]\n" - "cbz x21, 19f\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" + ".inst 0xc13519a1 // fmla za.s[x8, 1], { z13.s-z16.s }, z5.s\n" + ".inst 0xa04149e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "ldr x15, [%x[args], %[offsetof_Args_weights]]\n" + ".inst 0xc13719a0 // fmla za.s[x8, 0], { z13.s-z16.s }, z7.s\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" + ".inst 0xa04149e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "ld1w { z8.s }, p2/Z, [x15, #4, MUL VL]\n" + "addvl x15, x15, #5\n" + ".inst 0xa14049e1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x15]\n" + ".inst 0xa14149e2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "ld1w { z3.s }, p2/Z, [x15, #4, MUL VL]\n" + "addvl x15, x15, #5\n" + "10:" // Unpadded: 0 priming loads + "cmp x16, #0x2\n" + "blt 20f\n" + "add x21, x14, %x[ld_in_row], LSL #2\n" + "ld1w { z11.s }, p1/Z, [x14]\n" + "sub x16, x16, #0x2\n" + "ld1w { z21.s }, p1/Z, [x21]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" + "sub x13, x13, #0x1\n" + "ld1w { z12.s }, p1/Z, [x21]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" + "lsr x20, x16, #0x1\n" + "ld1w { z22.s }, p1/Z, [x21]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" + "cmp x20, x13\n" + "ld1w { z13.s }, p1/Z, [x21]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" + "csel x22, x20, x13, LT\n" + "ld1w { z23.s }, p1/Z, [x21]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" + "add x14, x14, %x[ld_in_col], LSL #2\n" + "ld1w { z14.s }, p1/Z, [x21]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" + "and x16, x16, #0x1\n" + "ld1w { z24.s }, p1/Z, [x21]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" + "sub x13, x13, x22\n" + "ld1w { z15.s }, p1/Z, [x21]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" + "ld1w { z25.s }, p1/Z, [x21]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" + "ld1w { z16.s }, p1/Z, [x21]\n" + "cbz x22, 19f\n" "11:" // Unpadded: Main loop ".inst 0xc1381960 // fmla za.s[x8, 0], { z11.s-z14.s }, z8.s\n" - "ld1w { z8.s }, p2/Z, [x14, #4, MUL VL]\n" - "add x20, x13, %x[ld_in_row], LSL #2\n" - "subs x21, x21, #0x1\n" + "ld1w { z8.s }, p2/Z, [x15, #4, MUL VL]\n" + "add x21, x14, %x[ld_in_row], LSL #2\n" + "subs x22, x22, #0x1\n" ".inst 0xc1361961 // fmla za.s[x8, 1], { z11.s-z14.s }, z6.s\n" - ".inst 0xa04149c6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" + ".inst 0xa04149e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" ".inst 0xc1341962 // fmla za.s[x8, 2], { z11.s-z14.s }, z4.s\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" - "addvl x14, x14, #5\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" + "addvl x15, x15, #5\n" ".inst 0xc1331aa0 // fmla za.s[x8, 0], { z21.s-z24.s }, z3.s\n" - "ld1w { z3.s }, p2/Z, [x14, #4, MUL VL]\n" + "ld1w { z3.s }, p2/Z, [x15, #4, MUL VL]\n" ".inst 0xc1321aa1 // fmla za.s[x8, 1], { z21.s-z24.s }, z2.s\n" - ".inst 0xa14149c2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" + ".inst 0xa14149e2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" ".inst 0xc1311aa2 // fmla za.s[x8, 2], { z21.s-z24.s }, z1.s\n" - ".inst 0xa14049c1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x14]\n" - "addvl x14, x14, #5\n" + ".inst 0xa14049e1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x15]\n" + "addvl x15, x15, #5\n" ".inst 0xc1381980 // fmla za.s[x8, 0], { z12.s-z15.s }, z8.s\n" - "ld1w { z8.s }, p2/Z, [x14, #4, MUL VL]\n" + "ld1w { z8.s }, p2/Z, [x15, #4, MUL VL]\n" ".inst 0xc1361981 // fmla za.s[x8, 1], { z12.s-z15.s }, z6.s\n" - ".inst 0xa04149c6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" + ".inst 0xa04149e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" ".inst 0xc1341982 // fmla za.s[x8, 2], { z12.s-z15.s }, z4.s\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" - "ldr x14, [%x[args], %[offsetof_Args_weights]]\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" + "ldr x15, [%x[args], %[offsetof_Args_weights]]\n" ".inst 0xc1331ac0 // fmla za.s[x8, 0], { z22.s-z25.s }, z3.s\n" - "ld1w { z11.s }, p1/Z, [x13]\n" - "add x13, x13, %x[ld_in_col], LSL #2\n" - "add x19, x13, %x[ld_in_row], LSL #2\n" + "ld1w { z11.s }, p1/Z, [x14]\n" + "add x14, x14, %x[ld_in_col], LSL #2\n" + "add x20, x14, %x[ld_in_row], LSL #2\n" ".inst 0xc1321ac1 // fmla za.s[x8, 1], { z22.s-z25.s }, z2.s\n" + "ld1w { z21.s }, p1/Z, [x21]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" + ".inst 0xc1311ac2 // fmla za.s[x8, 2], { z22.s-z25.s }, z1.s\n" + "ld1w { z12.s }, p1/Z, [x21]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" + "ld1w { z22.s }, p1/Z, [x21]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" + ".inst 0xc13819a0 // fmla za.s[x8, 0], { z13.s-z16.s }, z8.s\n" + ".inst 0xc13619a1 // fmla za.s[x8, 1], { z13.s-z16.s }, z6.s\n" + ".inst 0xa04149e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + ".inst 0xc13419a2 // fmla za.s[x8, 2], { z13.s-z16.s }, z4.s\n" + "ld1w { z13.s }, p1/Z, [x21]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" + "ld1w { z23.s }, p1/Z, [x21]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" + ".inst 0xc0060c18 // mova { z24.d-z27.d }, za.d[x8, #0]\n" + "add x8, x8, #0x1\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" + "addvl x15, x15, #5\n" + ".inst 0xc1b1c818 // fclamp { z24.s-z27.s }, z0.s, z17.s\n" + "st1w { z24.s }, p1, [x10]\n" + "ld1w { z14.s }, p1/Z, [x21]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" + ".inst 0xc1371960 // fmla za.s[x8, 0], { z11.s-z14.s }, z7.s\n" + "add x10, x10, x28, LSL #2\n" + ".inst 0xc1351961 // fmla za.s[x8, 1], { z11.s-z14.s }, z5.s\n" + ".inst 0xa14049e1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x15]\n" + "st1w { z25.s }, p1, [x9]\n" + "add x9, x9, x27, LSL #2\n" + ".inst 0xa14149e2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "addvl x15, x15, #5\n" + "st1w { z26.s }, p1, [x26]\n" + "add x26, x26, x24, LSL #2\n" + "ld1w { z24.s }, p1/Z, [x21]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" + ".inst 0xc13a1aa0 // fmla za.s[x8, 0], { z21.s-z24.s }, z10.s\n" + "st1w { z27.s }, p1, [x25]\n" + ".inst 0xc1391aa1 // fmla za.s[x8, 1], { z21.s-z24.s }, z9.s\n" + "ld1w { z15.s }, p1/Z, [x21]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" + "add x25, x25, x23, LSL #2\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" + ".inst 0xc1351981 // fmla za.s[x8, 1], { z12.s-z15.s }, z5.s\n" + ".inst 0xa04149e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "addvl x15, x15, #5\n" + ".inst 0xc1371980 // fmla za.s[x8, 0], { z12.s-z15.s }, z7.s\n" + "ld1w { z25.s }, p1/Z, [x21]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" + ".inst 0xc0040f84 // mova za.d[x8, #4], { z28.d-z31.d }\n" + ".inst 0xa14049e1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x15]\n" + ".inst 0xc1391ac1 // fmla za.s[x8, 1], { z22.s-z25.s }, z9.s\n" + ".inst 0xa14149e2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "addvl x15, x15, #5\n" + ".inst 0xc13a1ac0 // fmla za.s[x8, 0], { z22.s-z25.s }, z10.s\n" + "ld1w { z16.s }, p1/Z, [x21]\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" + ".inst 0xc13519a1 // fmla za.s[x8, 1], { z13.s-z16.s }, z5.s\n" + ".inst 0xa04149e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + ".inst 0xc13719a0 // fmla za.s[x8, 0], { z13.s-z16.s }, z7.s\n" + "ldr x15, [%x[args], %[offsetof_Args_weights]]\n" + "ld1w { z11.s }, p1/Z, [x14]\n" + "add x14, x14, %x[ld_in_col], LSL #2\n" "ld1w { z21.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" - ".inst 0xc1311ac2 // fmla za.s[x8, 2], { z22.s-z25.s }, z1.s\n" "ld1w { z12.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" "ld1w { z22.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" - ".inst 0xc13819a0 // fmla za.s[x8, 0], { z13.s-z16.s }, z8.s\n" - ".inst 0xc13619a1 // fmla za.s[x8, 1], { z13.s-z16.s }, z6.s\n" - ".inst 0xa04149c6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - ".inst 0xc13419a2 // fmla za.s[x8, 2], { z13.s-z16.s }, z4.s\n" "ld1w { z13.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" "ld1w { z23.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" - ".inst 0xc0060c18 // mova { z24.d-z27.d }, za.d[x8, #0]\n" - "add x8, x8, #0x1\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" - "addvl x14, x14, #5\n" - ".inst 0xc1b1c818 // fclamp { z24.s-z27.s }, z0.s, z17.s\n" - "st1w { z24.s }, p1, [x9]\n" "ld1w { z14.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" - ".inst 0xc1371960 // fmla za.s[x8, 0], { z11.s-z14.s }, z7.s\n" - "add x9, x9, x27, LSL #2\n" - ".inst 0xc1351961 // fmla za.s[x8, 1], { z11.s-z14.s }, z5.s\n" - ".inst 0xa14049c1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x14]\n" - "st1w { z25.s }, p1, [x28]\n" - "add x28, x28, x26, LSL #2\n" - ".inst 0xa14149c2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "addvl x14, x14, #5\n" - "st1w { z26.s }, p1, [x25]\n" - "add x25, x25, x23, LSL #2\n" "ld1w { z24.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" - ".inst 0xc13a1aa0 // fmla za.s[x8, 0], { z21.s-z24.s }, z10.s\n" - "st1w { z27.s }, p1, [x24]\n" - ".inst 0xc1391aa1 // fmla za.s[x8, 1], { z21.s-z24.s }, z9.s\n" "ld1w { z15.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" - "add x24, x24, x22, LSL #2\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" - ".inst 0xc1351981 // fmla za.s[x8, 1], { z12.s-z15.s }, z5.s\n" - ".inst 0xa04149c6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "addvl x14, x14, #5\n" - ".inst 0xc1371980 // fmla za.s[x8, 0], { z12.s-z15.s }, z7.s\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" + ".inst 0xa04149e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "ld1w { z8.s }, p2/Z, [x15, #4, MUL VL]\n" + "addvl x15, x15, #5\n" "ld1w { z25.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" - ".inst 0xc0040f84 // mova za.d[x8, #4], { z28.d-z31.d }\n" - ".inst 0xa14049c1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x14]\n" - ".inst 0xc1391ac1 // fmla za.s[x8, 1], { z22.s-z25.s }, z9.s\n" - ".inst 0xa14149c2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "addvl x14, x14, #5\n" - ".inst 0xc13a1ac0 // fmla za.s[x8, 0], { z22.s-z25.s }, z10.s\n" + ".inst 0xa14049e1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x15]\n" + ".inst 0xa14149e2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "ld1w { z3.s }, p2/Z, [x15, #4, MUL VL]\n" + "addvl x15, x15, #5\n" "ld1w { z16.s }, p1/Z, [x20]\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" - ".inst 0xc13519a1 // fmla za.s[x8, 1], { z13.s-z16.s }, z5.s\n" - ".inst 0xa04149c6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - ".inst 0xc13719a0 // fmla za.s[x8, 0], { z13.s-z16.s }, z7.s\n" - "ldr x14, [%x[args], %[offsetof_Args_weights]]\n" - "ld1w { z11.s }, p1/Z, [x13]\n" - "add x13, x13, %x[ld_in_col], LSL #2\n" - "ld1w { z21.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" - "ld1w { z12.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" - "ld1w { z22.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" - "ld1w { z13.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" - "ld1w { z23.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" - "ld1w { z14.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" - "ld1w { z24.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" - "ld1w { z15.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" - ".inst 0xa04149c6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "ld1w { z8.s }, p2/Z, [x14, #4, MUL VL]\n" - "addvl x14, x14, #5\n" - "ld1w { z25.s }, p1/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" - ".inst 0xa14049c1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x14]\n" - ".inst 0xa14149c2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "ld1w { z3.s }, p2/Z, [x14, #4, MUL VL]\n" - "addvl x14, x14, #5\n" - "ld1w { z16.s }, p1/Z, [x19]\n" "bgt 11b\n" "b 19f\n" "12:" // Padded - "cbz x21, 17f\n" - "cmp x21, #0x1\n" - "sub x15, x15, x21\n" + "cbz x22, 17f\n" + "cmp x22, #0x1\n" + "sub x16, x16, x22\n" "beq 16f\n" - "cmp x21, #0x2\n" + "cmp x22, #0x2\n" "beq 15f\n" - "cmp x21, #0x3\n" + "cmp x22, #0x3\n" "beq 14f\n" "13:" // Padded: 4 priming loads "mov x12, #0x0\n" ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" - "ld1w { z11.s }, p0/Z, [x13]\n" - "add x19, x13, %x[ld_in_row], LSL #2\n" + "ld1w { z11.s }, p0/Z, [x14]\n" + "add x20, x14, %x[ld_in_row], LSL #2\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" - "ld1w { z21.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z21.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" - "ld1w { z12.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z12.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n" - "ld1w { z22.s }, p0/Z, [x19]\n" + "ld1w { z22.s }, p0/Z, [x20]\n" "mov x12, #0x4\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" - "ld1w { z13.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z13.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" - "ld1w { z23.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z23.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" - "ld1w { z14.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z14.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n" "mov x12, #0x8\n" ".inst 0xc1341960 // fmla za.s[x8, 0], { z11.s-z14.s }, z4.s\n" - "ld1w { z24.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z24.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" ".inst 0xc1311aa0 // fmla za.s[x8, 0], { z21.s-z24.s }, z1.s\n" - "ld1w { z15.s }, p0/Z, [x19]\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z15.s }, p0/Z, [x20]\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" ".inst 0xc1341980 // fmla za.s[x8, 0], { z12.s-z15.s }, z4.s\n" - "addvl x14, x14, #5\n" - "ld1w { z25.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" - ".inst 0xa14049c1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x14]\n" + "addvl x15, x15, #5\n" + "ld1w { z25.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" + ".inst 0xa14049e1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x15]\n" ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" - "addvl x14, x14, #5\n" + "addvl x15, x15, #5\n" ".inst 0xc1311ac0 // fmla za.s[x8, 0], { z22.s-z25.s }, z1.s\n" - "ld1w { z16.s }, p0/Z, [x19]\n" - "add x13, x13, %x[ld_in_col], LSL #2\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" - "ldr x14, [%x[args], %[offsetof_Args_weights]]\n" + "ld1w { z16.s }, p0/Z, [x20]\n" + "add x14, x14, %x[ld_in_col], LSL #2\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" + "ldr x15, [%x[args], %[offsetof_Args_weights]]\n" ".inst 0xc13419a0 // fmla za.s[x8, 0], { z13.s-z16.s }, z4.s\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" - "addvl x14, x14, #5\n" - ".inst 0xa14049c1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x14]\n" - "addvl x14, x14, #5\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" + "addvl x15, x15, #5\n" + ".inst 0xa14049e1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x15]\n" + "addvl x15, x15, #5\n" "14:" // Padded: 3 priming loads "mov x12, #0x0\n" ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" - "ld1w { z11.s }, p0/Z, [x13]\n" - "add x19, x13, %x[ld_in_row], LSL #2\n" + "ld1w { z11.s }, p0/Z, [x14]\n" + "add x20, x14, %x[ld_in_row], LSL #2\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" - "ld1w { z21.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z21.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" - "ld1w { z12.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z12.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n" - "ld1w { z22.s }, p0/Z, [x19]\n" + "ld1w { z22.s }, p0/Z, [x20]\n" "mov x12, #0x4\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" - "ld1w { z13.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z13.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" - "ld1w { z23.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z23.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" - "ld1w { z14.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z14.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n" "mov x12, #0x8\n" ".inst 0xc1351960 // fmla za.s[x8, 0], { z11.s-z14.s }, z5.s\n" - "ld1w { z24.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z24.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" ".inst 0xc1391aa0 // fmla za.s[x8, 0], { z21.s-z24.s }, z9.s\n" - "ld1w { z15.s }, p0/Z, [x19]\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z15.s }, p0/Z, [x20]\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" ".inst 0xc1351980 // fmla za.s[x8, 0], { z12.s-z15.s }, z5.s\n" - "addvl x14, x14, #5\n" - "ld1w { z25.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" - ".inst 0xa14049c1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x14]\n" + "addvl x15, x15, #5\n" + "ld1w { z25.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" + ".inst 0xa14049e1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x15]\n" ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" - "addvl x14, x14, #5\n" + "addvl x15, x15, #5\n" ".inst 0xc1391ac0 // fmla za.s[x8, 0], { z22.s-z25.s }, z9.s\n" - "ld1w { z16.s }, p0/Z, [x19]\n" - "add x13, x13, %x[ld_in_col], LSL #2\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" - "ldr x14, [%x[args], %[offsetof_Args_weights]]\n" + "ld1w { z16.s }, p0/Z, [x20]\n" + "add x14, x14, %x[ld_in_col], LSL #2\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" + "ldr x15, [%x[args], %[offsetof_Args_weights]]\n" ".inst 0xc13519a0 // fmla za.s[x8, 0], { z13.s-z16.s }, z5.s\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" - ".inst 0xa04149c6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "addvl x14, x14, #5\n" - ".inst 0xa14049c1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x14]\n" - ".inst 0xa14149c2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "addvl x14, x14, #5\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" + ".inst 0xa04149e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "addvl x15, x15, #5\n" + ".inst 0xa14049e1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x15]\n" + ".inst 0xa14149e2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "addvl x15, x15, #5\n" "15:" // Padded: 2 priming loads "mov x12, #0x0\n" ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" - "ld1w { z11.s }, p0/Z, [x13]\n" - "add x19, x13, %x[ld_in_row], LSL #2\n" + "ld1w { z11.s }, p0/Z, [x14]\n" + "add x20, x14, %x[ld_in_row], LSL #2\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" - "ld1w { z21.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z21.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" - "ld1w { z12.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z12.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n" - "ld1w { z22.s }, p0/Z, [x19]\n" + "ld1w { z22.s }, p0/Z, [x20]\n" "mov x12, #0x4\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" - "ld1w { z13.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z13.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" - "ld1w { z23.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z23.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" - "ld1w { z14.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z14.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n" "mov x12, #0x8\n" ".inst 0xc1361960 // fmla za.s[x8, 0], { z11.s-z14.s }, z6.s\n" - "ld1w { z24.s }, p0/Z, [x19]\n" + "ld1w { z24.s }, p0/Z, [x20]\n" ".inst 0xc1341961 // fmla za.s[x8, 1], { z11.s-z14.s }, z4.s\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" - "ld1w { z15.s }, p0/Z, [x19]\n" + "ld1w { z15.s }, p0/Z, [x20]\n" ".inst 0xc1321aa0 // fmla za.s[x8, 0], { z21.s-z24.s }, z2.s\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" ".inst 0xc1311aa1 // fmla za.s[x8, 1], { z21.s-z24.s }, z1.s\n" - ".inst 0xa04149c6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "addvl x14, x14, #5\n" - "add x13, x13, %x[ld_in_col], LSL #2\n" + ".inst 0xa04149e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "addvl x15, x15, #5\n" + "add x14, x14, %x[ld_in_col], LSL #2\n" ".inst 0xc1361980 // fmla za.s[x8, 0], { z12.s-z15.s }, z6.s\n" - "ld1w { z25.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z25.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" ".inst 0xc1341981 // fmla za.s[x8, 1], { z12.s-z15.s }, z4.s\n" - ".inst 0xa14049c1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x14]\n" - ".inst 0xa14149c2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "addvl x14, x14, #5\n" + ".inst 0xa14049e1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x15]\n" + ".inst 0xa14149e2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "addvl x15, x15, #5\n" ".inst 0xc1321ac0 // fmla za.s[x8, 0], { z22.s-z25.s }, z2.s\n" ".inst 0xc1311ac1 // fmla za.s[x8, 1], { z22.s-z25.s }, z1.s\n" - "ld1w { z16.s }, p0/Z, [x19]\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" + "ld1w { z16.s }, p0/Z, [x20]\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" ".inst 0xc13419a1 // fmla za.s[x8, 1], { z13.s-z16.s }, z4.s\n" - ".inst 0xa04149c6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "ldr x14, [%x[args], %[offsetof_Args_weights]]\n" + ".inst 0xa04149e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "ldr x15, [%x[args], %[offsetof_Args_weights]]\n" ".inst 0xc13619a0 // fmla za.s[x8, 0], { z13.s-z16.s }, z6.s\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" - ".inst 0xa04149c6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "addvl x14, x14, #5\n" - ".inst 0xa14049c1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x14]\n" - ".inst 0xa14149c2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "addvl x14, x14, #5\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" + ".inst 0xa04149e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "addvl x15, x15, #5\n" + ".inst 0xa14049e1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x15]\n" + ".inst 0xa14149e2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "addvl x15, x15, #5\n" "16:" // Padded: 1 priming loads "mov x12, #0x0\n" ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" - "ld1w { z11.s }, p0/Z, [x13]\n" - "add x19, x13, %x[ld_in_row], LSL #2\n" + "ld1w { z11.s }, p0/Z, [x14]\n" + "add x20, x14, %x[ld_in_row], LSL #2\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" - "ld1w { z21.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z21.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" - "ld1w { z12.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z12.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n" - "ld1w { z22.s }, p0/Z, [x19]\n" + "ld1w { z22.s }, p0/Z, [x20]\n" "mov x12, #0x4\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" - "ld1w { z13.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z13.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" - "ld1w { z23.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z23.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" - "ld1w { z14.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z14.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n" "mov x12, #0x8\n" ".inst 0xc1371960 // fmla za.s[x8, 0], { z11.s-z14.s }, z7.s\n" - "ld1w { z24.s }, p0/Z, [x19]\n" + "ld1w { z24.s }, p0/Z, [x20]\n" ".inst 0xc1351961 // fmla za.s[x8, 1], { z11.s-z14.s }, z5.s\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" - "ld1w { z15.s }, p0/Z, [x19]\n" + "ld1w { z15.s }, p0/Z, [x20]\n" ".inst 0xc13a1aa0 // fmla za.s[x8, 0], { z21.s-z24.s }, z10.s\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" ".inst 0xc1391aa1 // fmla za.s[x8, 1], { z21.s-z24.s }, z9.s\n" - ".inst 0xa04149c6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "addvl x14, x14, #5\n" - "add x13, x13, %x[ld_in_col], LSL #2\n" + ".inst 0xa04149e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "addvl x15, x15, #5\n" + "add x14, x14, %x[ld_in_col], LSL #2\n" ".inst 0xc1371980 // fmla za.s[x8, 0], { z12.s-z15.s }, z7.s\n" - "ld1w { z25.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z25.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" ".inst 0xc1351981 // fmla za.s[x8, 1], { z12.s-z15.s }, z5.s\n" - ".inst 0xa14049c1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x14]\n" - ".inst 0xa14149c2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "addvl x14, x14, #5\n" + ".inst 0xa14049e1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x15]\n" + ".inst 0xa14149e2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "addvl x15, x15, #5\n" ".inst 0xc13a1ac0 // fmla za.s[x8, 0], { z22.s-z25.s }, z10.s\n" ".inst 0xc1391ac1 // fmla za.s[x8, 1], { z22.s-z25.s }, z9.s\n" - "ld1w { z16.s }, p0/Z, [x19]\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" + "ld1w { z16.s }, p0/Z, [x20]\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" ".inst 0xc13519a1 // fmla za.s[x8, 1], { z13.s-z16.s }, z5.s\n" - ".inst 0xa04149c6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "ldr x14, [%x[args], %[offsetof_Args_weights]]\n" + ".inst 0xa04149e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "ldr x15, [%x[args], %[offsetof_Args_weights]]\n" ".inst 0xc13719a0 // fmla za.s[x8, 0], { z13.s-z16.s }, z7.s\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" - ".inst 0xa04149c6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "ld1w { z8.s }, p2/Z, [x14, #4, MUL VL]\n" - "addvl x14, x14, #5\n" - ".inst 0xa14049c1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x14]\n" - ".inst 0xa14149c2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "ld1w { z3.s }, p2/Z, [x14, #4, MUL VL]\n" - "addvl x14, x14, #5\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" + ".inst 0xa04149e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "ld1w { z8.s }, p2/Z, [x15, #4, MUL VL]\n" + "addvl x15, x15, #5\n" + ".inst 0xa14049e1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x15]\n" + ".inst 0xa14149e2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "ld1w { z3.s }, p2/Z, [x15, #4, MUL VL]\n" + "addvl x15, x15, #5\n" "17:" // Padded: 0 priming loads - "cmp x15, #0x2\n" + "cmp x16, #0x2\n" "blt 20f\n" "mov x12, #0x0\n" ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" - "ld1w { z11.s }, p0/Z, [x13]\n" - "add x20, x13, %x[ld_in_row], LSL #2\n" + "ld1w { z11.s }, p0/Z, [x14]\n" + "add x21, x14, %x[ld_in_row], LSL #2\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" - "ld1w { z21.s }, p0/Z, [x20]\n" - "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z21.s }, p0/Z, [x21]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" - "ld1w { z12.s }, p0/Z, [x20]\n" - "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z12.s }, p0/Z, [x21]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" ".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n" - "ld1w { z22.s }, p0/Z, [x20]\n" + "ld1w { z22.s }, p0/Z, [x21]\n" "mov x12, #0x4\n" - "add x20, x20, %x[ld_in_row], LSL #2\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" - "ld1w { z13.s }, p0/Z, [x20]\n" - "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z13.s }, p0/Z, [x21]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" - "ld1w { z23.s }, p0/Z, [x20]\n" - "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z23.s }, p0/Z, [x21]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" - "ld1w { z14.s }, p0/Z, [x20]\n" - "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z14.s }, p0/Z, [x21]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" ".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n" "mov x12, #0x8\n" - "ld1w { z24.s }, p0/Z, [x20]\n" - "sub x15, x15, #0x2\n" - "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z24.s }, p0/Z, [x21]\n" + "sub x16, x16, #0x2\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" - "sub x11, x11, #0x1\n" - "ld1w { z15.s }, p0/Z, [x20]\n" - "lsr x19, x15, #0x1\n" - "add x20, x20, %x[ld_in_row], LSL #2\n" + "sub x13, x13, #0x1\n" + "ld1w { z15.s }, p0/Z, [x21]\n" + "lsr x20, x16, #0x1\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" - "cmp x19, x11\n" - "ld1w { z25.s }, p0/Z, [x20]\n" - "add x20, x20, %x[ld_in_row], LSL #2\n" + "cmp x20, x13\n" + "ld1w { z25.s }, p0/Z, [x21]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" - "ld1w { z16.s }, p0/Z, [x20]\n" - "csel x21, x19, x11, LT\n" - "add x13, x13, %x[ld_in_col], LSL #2\n" - "and x15, x15, #0x1\n" - "sub x11, x11, x21\n" - "cbz x21, 19f\n" + "ld1w { z16.s }, p0/Z, [x21]\n" + "csel x22, x20, x13, LT\n" + "add x14, x14, %x[ld_in_col], LSL #2\n" + "and x16, x16, #0x1\n" + "sub x13, x13, x22\n" + "cbz x22, 19f\n" "18:" // Padded: Main loop ".inst 0xc1381960 // fmla za.s[x8, 0], { z11.s-z14.s }, z8.s\n" - "ld1w { z8.s }, p2/Z, [x14, #4, MUL VL]\n" + "ld1w { z8.s }, p2/Z, [x15, #4, MUL VL]\n" "mov x12, #0x0\n" ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" ".inst 0xc1361961 // fmla za.s[x8, 1], { z11.s-z14.s }, z6.s\n" - ".inst 0xa04149c6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "add x20, x13, %x[ld_in_row], LSL #2\n" - "subs x21, x21, #0x1\n" + ".inst 0xa04149e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "add x21, x14, %x[ld_in_row], LSL #2\n" + "subs x22, x22, #0x1\n" ".inst 0xc1341962 // fmla za.s[x8, 2], { z11.s-z14.s }, z4.s\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" - "addvl x14, x14, #5\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" + "addvl x15, x15, #5\n" ".inst 0xc1331aa0 // fmla za.s[x8, 0], { z21.s-z24.s }, z3.s\n" - "ld1w { z3.s }, p2/Z, [x14, #4, MUL VL]\n" + "ld1w { z3.s }, p2/Z, [x15, #4, MUL VL]\n" ".inst 0xc1321aa1 // fmla za.s[x8, 1], { z21.s-z24.s }, z2.s\n" - ".inst 0xa14149c2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" + ".inst 0xa14149e2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" ".inst 0xc1311aa2 // fmla za.s[x8, 2], { z21.s-z24.s }, z1.s\n" - ".inst 0xa14049c1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x14]\n" - "addvl x14, x14, #5\n" + ".inst 0xa14049e1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x15]\n" + "addvl x15, x15, #5\n" ".inst 0xc1381980 // fmla za.s[x8, 0], { z12.s-z15.s }, z8.s\n" - "ld1w { z11.s }, p0/Z, [x13]\n" + "ld1w { z11.s }, p0/Z, [x14]\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" - "add x13, x13, %x[ld_in_col], LSL #2\n" + "add x14, x14, %x[ld_in_col], LSL #2\n" ".inst 0xc1361981 // fmla za.s[x8, 1], { z12.s-z15.s }, z6.s\n" - ".inst 0xa04149c6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "add x19, x13, %x[ld_in_row], LSL #2\n" + ".inst 0xa04149e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "add x20, x14, %x[ld_in_row], LSL #2\n" ".inst 0xc1341982 // fmla za.s[x8, 2], { z12.s-z15.s }, z4.s\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" ".inst 0xc1331ac0 // fmla za.s[x8, 0], { z22.s-z25.s }, z3.s\n" - "ld1w { z8.s }, p2/Z, [x14, #4, MUL VL]\n" - "ldr x14, [%x[args], %[offsetof_Args_weights]]\n" + "ld1w { z8.s }, p2/Z, [x15, #4, MUL VL]\n" + "ldr x15, [%x[args], %[offsetof_Args_weights]]\n" ".inst 0xc1321ac1 // fmla za.s[x8, 1], { z22.s-z25.s }, z2.s\n" - "ld1w { z21.s }, p0/Z, [x20]\n" - "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z21.s }, p0/Z, [x21]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" ".inst 0xc1311ac2 // fmla za.s[x8, 2], { z22.s-z25.s }, z1.s\n" - "ld1w { z12.s }, p0/Z, [x20]\n" - "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z12.s }, p0/Z, [x21]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" ".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n" "mov x12, #0x4\n" - "ld1w { z22.s }, p0/Z, [x20]\n" - "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z22.s }, p0/Z, [x21]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" ".inst 0xc13819a0 // fmla za.s[x8, 0], { z13.s-z16.s }, z8.s\n" ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" ".inst 0xc13619a1 // fmla za.s[x8, 1], { z13.s-z16.s }, z6.s\n" - ".inst 0xa04149c6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" + ".inst 0xa04149e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" ".inst 0xc13419a2 // fmla za.s[x8, 2], { z13.s-z16.s }, z4.s\n" - "ld1w { z13.s }, p0/Z, [x20]\n" - "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z13.s }, p0/Z, [x21]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" - "ld1w { z23.s }, p0/Z, [x20]\n" - "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z23.s }, p0/Z, [x21]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" ".inst 0xc0060c18 // mova { z24.d-z27.d }, za.d[x8, #0]\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" "add x8, x8, #0x1\n" - "addvl x14, x14, #5\n" + "addvl x15, x15, #5\n" ".inst 0xc1b1c818 // fclamp { z24.s-z27.s }, z0.s, z17.s\n" - "ld1w { z14.s }, p0/Z, [x20]\n" - "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z14.s }, p0/Z, [x21]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" ".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n" - "st1w { z24.s }, p1, [x9]\n" + "st1w { z24.s }, p1, [x10]\n" "mov x12, #0x8\n" ".inst 0xc1371960 // fmla za.s[x8, 0], { z11.s-z14.s }, z7.s\n" - ".inst 0xa14049c1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x14]\n" - "add x9, x9, x27, LSL #2\n" + ".inst 0xa14049e1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x15]\n" + "add x10, x10, x28, LSL #2\n" ".inst 0xc1351961 // fmla za.s[x8, 1], { z11.s-z14.s }, z5.s\n" - ".inst 0xa14149c2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "addvl x14, x14, #5\n" - "st1w { z25.s }, p1, [x28]\n" - "ld1w { z24.s }, p0/Z, [x20]\n" - "add x20, x20, %x[ld_in_row], LSL #2\n" + ".inst 0xa14149e2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "addvl x15, x15, #5\n" + "st1w { z25.s }, p1, [x9]\n" + "ld1w { z24.s }, p0/Z, [x21]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" ".inst 0xc13a1aa0 // fmla za.s[x8, 0], { z21.s-z24.s }, z10.s\n" ".inst 0xc1391aa1 // fmla za.s[x8, 1], { z21.s-z24.s }, z9.s\n" - "ld1w { z15.s }, p0/Z, [x20]\n" - "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z15.s }, p0/Z, [x21]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" ".inst 0xc1351981 // fmla za.s[x8, 1], { z12.s-z15.s }, z5.s\n" - "add x28, x28, x26, LSL #2\n" - "st1w { z26.s }, p1, [x25]\n" - ".inst 0xa04149c6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "addvl x14, x14, #5\n" + "add x9, x9, x27, LSL #2\n" + "st1w { z26.s }, p1, [x26]\n" + ".inst 0xa04149e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "addvl x15, x15, #5\n" ".inst 0xc1371980 // fmla za.s[x8, 0], { z12.s-z15.s }, z7.s\n" - "add x25, x25, x23, LSL #2\n" - "ld1w { z25.s }, p0/Z, [x20]\n" - "add x20, x20, %x[ld_in_row], LSL #2\n" + "add x26, x26, x24, LSL #2\n" + "ld1w { z25.s }, p0/Z, [x21]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" - "st1w { z27.s }, p1, [x24]\n" - ".inst 0xa14049c1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x14]\n" + "st1w { z27.s }, p1, [x25]\n" + ".inst 0xa14049e1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x15]\n" "mov x12, #0x0\n" ".inst 0xc1391ac1 // fmla za.s[x8, 1], { z22.s-z25.s }, z9.s\n" - "add x24, x24, x22, LSL #2\n" - ".inst 0xa14149c2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "addvl x14, x14, #5\n" + "add x25, x25, x23, LSL #2\n" + ".inst 0xa14149e2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "addvl x15, x15, #5\n" ".inst 0xc13a1ac0 // fmla za.s[x8, 0], { z22.s-z25.s }, z10.s\n" - "ld1w { z16.s }, p0/Z, [x20]\n" + "ld1w { z16.s }, p0/Z, [x21]\n" ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" ".inst 0xc0040f84 // mova za.d[x8, #4], { z28.d-z31.d }\n" - "ld1w { z11.s }, p0/Z, [x13]\n" + "ld1w { z11.s }, p0/Z, [x14]\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" - "add x13, x13, %x[ld_in_col], LSL #2\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" + "add x14, x14, %x[ld_in_col], LSL #2\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" ".inst 0xc13519a1 // fmla za.s[x8, 1], { z13.s-z16.s }, z5.s\n" - ".inst 0xa04149c6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" + ".inst 0xa04149e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" ".inst 0xc13719a0 // fmla za.s[x8, 0], { z13.s-z16.s }, z7.s\n" - "ldr x14, [%x[args], %[offsetof_Args_weights]]\n" - "ld1w { z21.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ldr x15, [%x[args], %[offsetof_Args_weights]]\n" + "ld1w { z21.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" - "ld1w { z12.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z12.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n" "mov x12, #0x4\n" - "ld1w { z22.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z22.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" - "ld1w { z13.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z13.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" - "ld1w { z23.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z23.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" - "ld1w { z14.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z14.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n" "mov x12, #0x8\n" - "ld1w { z24.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z24.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" - "ld1w { z15.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z15.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" - ".inst 0xa04149c6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "ld1w { z8.s }, p2/Z, [x14, #4, MUL VL]\n" - "addvl x14, x14, #5\n" - "ld1w { z25.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" + ".inst 0xa04149e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "ld1w { z8.s }, p2/Z, [x15, #4, MUL VL]\n" + "addvl x15, x15, #5\n" + "ld1w { z25.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" - ".inst 0xa14049c1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x14]\n" - ".inst 0xa14149c2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "ld1w { z3.s }, p2/Z, [x14, #4, MUL VL]\n" - "addvl x14, x14, #5\n" - "ld1w { z16.s }, p0/Z, [x19]\n" + ".inst 0xa14049e1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x15]\n" + ".inst 0xa14149e2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "ld1w { z3.s }, p2/Z, [x15, #4, MUL VL]\n" + "addvl x15, x15, #5\n" + "ld1w { z16.s }, p0/Z, [x20]\n" "bgt 18b\n" "19:" // Main loop tail ".inst 0xc1381960 // fmla za.s[x8, 0], { z11.s-z14.s }, z8.s\n" - "ld1w { z8.s }, p2/Z, [x14, #4, MUL VL]\n" + "ld1w { z8.s }, p2/Z, [x15, #4, MUL VL]\n" "mov x12, #0x0\n" ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" ".inst 0xc1361961 // fmla za.s[x8, 1], { z11.s-z14.s }, z6.s\n" - ".inst 0xa04149c6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "add x19, x13, %x[ld_in_row], LSL #2\n" + ".inst 0xa04149e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "add x20, x14, %x[ld_in_row], LSL #2\n" ".inst 0xc1341962 // fmla za.s[x8, 2], { z11.s-z14.s }, z4.s\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" - "addvl x14, x14, #5\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" + "addvl x15, x15, #5\n" ".inst 0xc1331aa0 // fmla za.s[x8, 0], { z21.s-z24.s }, z3.s\n" - "ld1w { z3.s }, p2/Z, [x14, #4, MUL VL]\n" + "ld1w { z3.s }, p2/Z, [x15, #4, MUL VL]\n" ".inst 0xc1321aa1 // fmla za.s[x8, 1], { z21.s-z24.s }, z2.s\n" - ".inst 0xa14149c2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" + ".inst 0xa14149e2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" ".inst 0xc1311aa2 // fmla za.s[x8, 2], { z21.s-z24.s }, z1.s\n" - ".inst 0xa14049c1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x14]\n" - "addvl x14, x14, #5\n" + ".inst 0xa14049e1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x15]\n" + "addvl x15, x15, #5\n" ".inst 0xc1381980 // fmla za.s[x8, 0], { z12.s-z15.s }, z8.s\n" - "ld1w { z11.s }, p0/Z, [x13]\n" + "ld1w { z11.s }, p0/Z, [x14]\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" - "add x13, x13, %x[ld_in_col], LSL #2\n" + "add x14, x14, %x[ld_in_col], LSL #2\n" ".inst 0xc1361981 // fmla za.s[x8, 1], { z12.s-z15.s }, z6.s\n" - ".inst 0xa04149c6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" + ".inst 0xa04149e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" ".inst 0xc1341982 // fmla za.s[x8, 2], { z12.s-z15.s }, z4.s\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" ".inst 0xc1331ac0 // fmla za.s[x8, 0], { z22.s-z25.s }, z3.s\n" - "ld1w { z8.s }, p2/Z, [x14, #4, MUL VL]\n" - "ldr x14, [%x[args], %[offsetof_Args_weights]]\n" + "ld1w { z8.s }, p2/Z, [x15, #4, MUL VL]\n" + "ldr x15, [%x[args], %[offsetof_Args_weights]]\n" ".inst 0xc1321ac1 // fmla za.s[x8, 1], { z22.s-z25.s }, z2.s\n" - "ld1w { z21.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z21.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" ".inst 0xc1311ac2 // fmla za.s[x8, 2], { z22.s-z25.s }, z1.s\n" - "ld1w { z12.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z12.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n" "mov x12, #0x4\n" - "ld1w { z22.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z22.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0xc13819a0 // fmla za.s[x8, 0], { z13.s-z16.s }, z8.s\n" ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" ".inst 0xc13619a1 // fmla za.s[x8, 1], { z13.s-z16.s }, z6.s\n" - ".inst 0xa04149c6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" + ".inst 0xa04149e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" ".inst 0xc13419a2 // fmla za.s[x8, 2], { z13.s-z16.s }, z4.s\n" - "ld1w { z13.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z13.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" - "ld1w { z23.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z23.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" ".inst 0xc0060c18 // mova { z24.d-z27.d }, za.d[x8, #0]\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" "add x8, x8, #0x1\n" - "addvl x14, x14, #5\n" + "addvl x15, x15, #5\n" ".inst 0xc1b1c818 // fclamp { z24.s-z27.s }, z0.s, z17.s\n" - "ld1w { z14.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z14.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n" - "st1w { z24.s }, p1, [x9]\n" + "st1w { z24.s }, p1, [x10]\n" "mov x12, #0x8\n" ".inst 0xc1371960 // fmla za.s[x8, 0], { z11.s-z14.s }, z7.s\n" - ".inst 0xa14049c1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x14]\n" - "add x9, x9, x27, LSL #2\n" + ".inst 0xa14049e1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x15]\n" + "add x10, x10, x28, LSL #2\n" ".inst 0xc1351961 // fmla za.s[x8, 1], { z11.s-z14.s }, z5.s\n" - ".inst 0xa14149c2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "addvl x14, x14, #5\n" - "st1w { z25.s }, p1, [x28]\n" - "ld1w { z24.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + ".inst 0xa14149e2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "addvl x15, x15, #5\n" + "st1w { z25.s }, p1, [x9]\n" + "ld1w { z24.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" ".inst 0xc13a1aa0 // fmla za.s[x8, 0], { z21.s-z24.s }, z10.s\n" ".inst 0xc1391aa1 // fmla za.s[x8, 1], { z21.s-z24.s }, z9.s\n" - "ld1w { z15.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z15.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" ".inst 0xc1351981 // fmla za.s[x8, 1], { z12.s-z15.s }, z5.s\n" - "add x28, x28, x26, LSL #2\n" - "st1w { z26.s }, p1, [x25]\n" - ".inst 0xa04149c6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "addvl x14, x14, #5\n" + "add x9, x9, x27, LSL #2\n" + "st1w { z26.s }, p1, [x26]\n" + ".inst 0xa04149e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "addvl x15, x15, #5\n" ".inst 0xc1371980 // fmla za.s[x8, 0], { z12.s-z15.s }, z7.s\n" - "add x25, x25, x23, LSL #2\n" - "ld1w { z25.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "add x26, x26, x24, LSL #2\n" + "ld1w { z25.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" - "st1w { z27.s }, p1, [x24]\n" - ".inst 0xa14049c1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x14]\n" + "st1w { z27.s }, p1, [x25]\n" + ".inst 0xa14049e1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x15]\n" ".inst 0xc1391ac1 // fmla za.s[x8, 1], { z22.s-z25.s }, z9.s\n" - "add x24, x24, x22, LSL #2\n" - ".inst 0xa14149c2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "addvl x14, x14, #5\n" + "add x25, x25, x23, LSL #2\n" + ".inst 0xa14149e2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "addvl x15, x15, #5\n" ".inst 0xc13a1ac0 // fmla za.s[x8, 0], { z22.s-z25.s }, z10.s\n" - "ld1w { z16.s }, p0/Z, [x19]\n" + "ld1w { z16.s }, p0/Z, [x20]\n" ".inst 0xc0040f84 // mova za.d[x8, #4], { z28.d-z31.d }\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" ".inst 0xc13519a1 // fmla za.s[x8, 1], { z13.s-z16.s }, z5.s\n" - ".inst 0xa04149c6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "ldr x14, [%x[args], %[offsetof_Args_weights]]\n" + ".inst 0xa04149e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "ldr x15, [%x[args], %[offsetof_Args_weights]]\n" ".inst 0xc13719a0 // fmla za.s[x8, 0], { z13.s-z16.s }, z7.s\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" - ".inst 0xa04149c6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "ld1w { z8.s }, p2/Z, [x14, #4, MUL VL]\n" - "addvl x14, x14, #5\n" - ".inst 0xa14049c1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x14]\n" - ".inst 0xa14149c2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "ld1w { z3.s }, p2/Z, [x14, #4, MUL VL]\n" - "addvl x14, x14, #5\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" + ".inst 0xa04149e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "ld1w { z8.s }, p2/Z, [x15, #4, MUL VL]\n" + "addvl x15, x15, #5\n" + ".inst 0xa14049e1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x15]\n" + ".inst 0xa14149e2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "ld1w { z3.s }, p2/Z, [x15, #4, MUL VL]\n" + "addvl x15, x15, #5\n" "20:" // Main loop skip tail - "cbz x15, 21f\n" // Skip remainder inputs + "cbz x16, 21f\n" // Skip remainder inputs "mov x12, #0x0\n" ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" - "ld1w { z11.s }, p0/Z, [x13]\n" - "add x19, x13, %x[ld_in_row], LSL #2\n" + "ld1w { z11.s }, p0/Z, [x14]\n" + "add x20, x14, %x[ld_in_row], LSL #2\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" - "ld1w { z21.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z21.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" - "ld1w { z12.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z12.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n" - "ld1w { z22.s }, p0/Z, [x19]\n" + "ld1w { z22.s }, p0/Z, [x20]\n" "mov x12, #0x4\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" - "ld1w { z13.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z13.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" - "ld1w { z23.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z23.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" - "ld1w { z14.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z14.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n" "mov x12, #0x8\n" ".inst 0xc1381960 // fmla za.s[x8, 0], { z11.s-z14.s }, z8.s\n" - "ld1w { z24.s }, p0/Z, [x19]\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "ld1w { z24.s }, p0/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" ".inst 0xc1331aa0 // fmla za.s[x8, 0], { z21.s-z24.s }, z3.s\n" - "ld1w { z15.s }, p0/Z, [x19]\n" - "ld1w { z8.s }, p2/Z, [x14, #4, MUL VL]\n" + "ld1w { z15.s }, p0/Z, [x20]\n" + "ld1w { z8.s }, p2/Z, [x15, #4, MUL VL]\n" ".inst 0xc1361961 // fmla za.s[x8, 1], { z11.s-z14.s }, z6.s\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" ".inst 0xc1341962 // fmla za.s[x8, 2], { z11.s-z14.s }, z4.s\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" - "sub x11, x11, #0x1\n" - ".inst 0xa04149c6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "addvl x14, x14, #5\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" + "sub x13, x13, #0x1\n" + ".inst 0xa04149e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "addvl x15, x15, #5\n" ".inst 0xc1381980 // fmla za.s[x8, 0], { z12.s-z15.s }, z8.s\n" - "ld1w { z25.s }, p0/Z, [x19]\n" + "ld1w { z25.s }, p0/Z, [x20]\n" ".inst 0xc1321aa1 // fmla za.s[x8, 1], { z21.s-z24.s }, z2.s\n" - "add x19, x19, %x[ld_in_row], LSL #2\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" - "ld1w { z3.s }, p2/Z, [x14, #4, MUL VL]\n" + "ld1w { z3.s }, p2/Z, [x15, #4, MUL VL]\n" ".inst 0xc1311aa2 // fmla za.s[x8, 2], { z21.s-z24.s }, z1.s\n" - ".inst 0xa14049c1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x14]\n" + ".inst 0xa14049e1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x15]\n" ".inst 0xc1331ac0 // fmla za.s[x8, 0], { z22.s-z25.s }, z3.s\n" - ".inst 0xa14149c2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" - "addvl x14, x14, #5\n" + ".inst 0xa14149e2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "addvl x15, x15, #5\n" ".inst 0xc1361981 // fmla za.s[x8, 1], { z12.s-z15.s }, z6.s\n" - "ld1w { z16.s }, p0/Z, [x19]\n" + "ld1w { z16.s }, p0/Z, [x20]\n" ".inst 0xc1341982 // fmla za.s[x8, 2], { z12.s-z15.s }, z4.s\n" - "ld1w { z8.s }, p2/Z, [x14, #4, MUL VL]\n" + "ld1w { z8.s }, p2/Z, [x15, #4, MUL VL]\n" ".inst 0xc13819a0 // fmla za.s[x8, 0], { z13.s-z16.s }, z8.s\n" ".inst 0xc1321ac1 // fmla za.s[x8, 1], { z22.s-z25.s }, z2.s\n" - ".inst 0xa04049c4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x14]\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" ".inst 0xc1311ac2 // fmla za.s[x8, 2], { z22.s-z25.s }, z1.s\n" - ".inst 0xa04149c6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x14, #0x2, MUL VL]\n" + ".inst 0xa04149e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" ".inst 0xc0060c18 // mova { z24.d-z27.d }, za.d[x8, #0]\n" ".inst 0xc1b1c818 // fclamp { z24.s-z27.s }, z0.s, z17.s\n" - "st1w { z24.s }, p1, [x9]\n" - "add x9, x9, x27, LSL #2\n" + "st1w { z24.s }, p1, [x10]\n" + "add x10, x10, x28, LSL #2\n" ".inst 0xc13619a1 // fmla za.s[x8, 1], { z13.s-z16.s }, z6.s\n" - "st1w { z25.s }, p1, [x28]\n" - "add x28, x28, x26, LSL #2\n" + "st1w { z25.s }, p1, [x9]\n" + "add x9, x9, x27, LSL #2\n" ".inst 0xc13419a2 // fmla za.s[x8, 2], { z13.s-z16.s }, z4.s\n" "add x8, x8, #0x1\n" - "st1w { z26.s }, p1, [x25]\n" + "st1w { z26.s }, p1, [x26]\n" + "add x26, x26, x24, LSL #2\n" + "st1w { z27.s }, p1, [x25]\n" "add x25, x25, x23, LSL #2\n" - "st1w { z27.s }, p1, [x24]\n" - "add x24, x24, x22, LSL #2\n" ".inst 0xc0040f84 // mova za.d[x8, #4], { z28.d-z31.d }\n" "21:" // Tail input: End - "cbz x11, 23f\n" + "cbz x13, 23f\n" "22:" // Right padding loop ".inst 0xc0060c18 // mova { z24.d-z27.d }, za.d[x8, #0]\n" "add x8, x8, #0x1\n" - "subs x11, x11, #0x1\n" + "subs x13, x13, #0x1\n" ".inst 0xc1b1c818 // fclamp { z24.s-z27.s }, z0.s, z17.s\n" - "st1w { z24.s }, p1, [x9]\n" - "add x9, x9, x27, LSL #2\n" + "st1w { z24.s }, p1, [x10]\n" + "add x10, x10, x28, LSL #2\n" ".inst 0xc0040f84 // mova za.d[x8, #4], { z28.d-z31.d }\n" - "st1w { z25.s }, p1, [x28]\n" - "add x28, x28, x26, LSL #2\n" - "st1w { z26.s }, p1, [x25]\n" + "st1w { z25.s }, p1, [x9]\n" + "add x9, x9, x27, LSL #2\n" + "st1w { z26.s }, p1, [x26]\n" + "add x26, x26, x24, LSL #2\n" + "st1w { z27.s }, p1, [x25]\n" "add x25, x25, x23, LSL #2\n" - "st1w { z27.s }, p1, [x24]\n" - "add x24, x24, x22, LSL #2\n" "bgt 22b\n" "23:" // End - "ldr x14, [%x[args], %[offsetof_Args_weights]]\n" - "incb x14, ALL, MUL #16\n" - "incb x14, ALL, MUL #9\n" - "str x14, [%x[args], %[offsetof_Args_weights]]\n" - "ldr x19, [%x[args], %[offsetof_Args_ld_in_vl]]\n" - "incw x16\n" - "whilelt p1.s, x16, x17\n" - "ldr x13, [%x[args], %[offsetof_Args_inptr]]\n" - "add x13, x13, x19, LSL #2\n" - "str x13, [%x[args], %[offsetof_Args_inptr]]\n" - "ldr x10, [%x[args], %[offsetof_Args_outptrs]]\n" - "ldr x23, [%x[args], %[offsetof_Args_ld_out_vls]]\n" - "ldp x22, x21, [x10, #0x0]\n" - "ldp x20, x19, [x23, #0x0]\n" + "ldr x15, [%x[args], %[offsetof_Args_weights]]\n" + "incb x15, ALL, MUL #16\n" + "incb x15, ALL, MUL #9\n" + "str x15, [%x[args], %[offsetof_Args_weights]]\n" + "ldr x20, [%x[args], %[offsetof_Args_ld_in_vl]]\n" + "incw x17\n" + "whilelt p1.s, x17, x7\n" + "ldr x14, [%x[args], %[offsetof_Args_inptr]]\n" + "add x14, x14, x20, LSL #2\n" + "str x14, [%x[args], %[offsetof_Args_inptr]]\n" + "ldr x11, [%x[args], %[offsetof_Args_outptrs]]\n" + "ldr x24, [%x[args], %[offsetof_Args_ld_out_vls]]\n" + "ldp x23, x22, [x11, #0x0]\n" + "ldp x21, x20, [x24, #0x0]\n" + "add x23, x23, x21, LSL #2\n" "add x22, x22, x20, LSL #2\n" - "add x21, x21, x19, LSL #2\n" - "stp x22, x21, [x10, #0x0]\n" - "ldp x22, x21, [x10, #0x10]\n" - "ldp x20, x19, [x23, #0x10]\n" + "stp x23, x22, [x11, #0x0]\n" + "ldp x23, x22, [x11, #0x10]\n" + "ldp x21, x20, [x24, #0x10]\n" + "add x23, x23, x21, LSL #2\n" "add x22, x22, x20, LSL #2\n" - "add x21, x21, x19, LSL #2\n" - "stp x22, x21, [x10, #0x10]\n" + "stp x23, x22, [x11, #0x10]\n" "b.any 1b\n" ".inst 0xd503467f // SMSTOP\n" : : [args] "r" (&args), [ld_in_col] "r" (ld_in_col), [ld_in_row] "r" (ld_in_row), [offsetof_Args_bias] "I" (offsetof(Args, bias)), [offsetof_Args_clamp_max] "I" (offsetof(Args, clamp_max)), [offsetof_Args_clamp_min] "I" (offsetof(Args, clamp_min)), [offsetof_Args_current_channel] "I" (offsetof(Args, current_channel)), [offsetof_Args_inptr] "I" (offsetof(Args, inptr)), [offsetof_Args_input_cols] "I" (offsetof(Args, input_cols)), [offsetof_Args_ld_in_vl] "I" (offsetof(Args, ld_in_vl)), [offsetof_Args_ld_out_cols] "I" (offsetof(Args, ld_out_cols)), [offsetof_Args_ld_out_vls] "I" (offsetof(Args, ld_out_vls)), [offsetof_Args_n_channels] "I" (offsetof(Args, n_channels)), [offsetof_Args_outptrs] "I" (offsetof(Args, outptrs)), [offsetof_Args_output_cols] "I" (offsetof(Args, output_cols)), [offsetof_Args_pad_bottom] "I" (offsetof(Args, pad_bottom)), [offsetof_Args_pad_left] "I" (offsetof(Args, pad_left)), [offsetof_Args_pad_top] "I" (offsetof(Args, pad_top)), [offsetof_Args_weights] "I" (offsetof(Args, weights)) - : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x6", "x7", "x8", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x16", "x17", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31" + : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x5", "x6", "x7", "x8", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x16", "x17", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31" ); } |