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 | 1254 |
1 files changed, 627 insertions, 627 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 81ad8e5833..233b6bd61a 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) 2022-2023 Arm Limited. + * Copyright (c) 2022-2024 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -72,67 +72,67 @@ void sme2_fp32_planar_5x5_s2_4rows_mla_za_impl( "ldr x5, [%x[args], %[offsetof_Args_pad_bottom]]\n" "mov x20, #0xb\n" ".inst 0xd503477f // SMSTART ZA\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 { z2.s }, p2/Z, [%x[args], %[offsetof_Args_clamp_min]]\n" "ldr x7, [%x[args], %[offsetof_Args_n_channels]]\n" + "ld1rw { z17.s }, p2/Z, [%x[args], %[offsetof_Args_clamp_min]]\n" + "sub x20, x20, x5\n" + "ldr x17, [%x[args], %[offsetof_Args_current_channel]]\n" + "ld1rw { z16.s }, p2/Z, [%x[args], %[offsetof_Args_clamp_max]]\n" "whilelt p1.s, XZR, x7\n" "whilelt p9.s, XZR, x20\n" - "ld1rw { z3.s }, p2/Z, [%x[args], %[offsetof_Args_clamp_max]]\n" "whilelt p8.s, XZR, x6\n" "eor p8.b, p2/Z, p8.b, p9.b\n" - "ldr x17, [%x[args], %[offsetof_Args_current_channel]]\n" "1:" // Channel loop "ldr x20, [%x[args], %[offsetof_Args_bias]]\n" - "fmov z28.s, #0x0\n" + "fmov z20.s, #0x0\n" "cbz x20, 2f\n" - "ld1w { z28.s }, p1/Z, [x20, x17, LSL #2]\n" + "ld1w { z20.s }, p1/Z, [x20, x17, LSL #2]\n" "2:" // Load bias: Done "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 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" + "add x20, x6, x5\n" + "mov z21.d, z20.d\n" + "ldr x15, [%x[args], %[offsetof_Args_weights]]\n" + "lsl x21, %x[ld_in_row], #0x2\n" + "mov z22.d, z20.d\n" + "mov z23.d, z20.d\n" "ldr x14, [%x[args], %[offsetof_Args_inptr]]\n" - ".inst 0xa04149ea // ld1w { z10.s-z11.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" - "add x21, x6, x5\n" - "lsl x20, %x[ld_in_row], #0x2\n" - "ld1w { z9.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 0xa14049e7 // ld1w { z7.s, z15.s }, pn10.b/Z, [x15]\n" "mov x8, #0x0\n" + "sub x22, x22, x20\n" + "sub x20, x16, #0x1\n" "ldr x13, [%x[args], %[offsetof_Args_output_cols]]\n" - "lsl x23, x23, #0x2\n" + "orr x20, x20, %x[ld_in_col], LSL #18\n" + ".inst 0xa14049e7 // ld1w { z7.s, z15.s }, pn10.b/Z, [x15]\n" + ".inst 0xa14149e6 // ld1w { z6.s, z14.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "orr x20, x7, x20, LSL #20\n" + "madd x21, x21, x6, x14\n" + "ld1w { z8.s }, p2/Z, [x15, #4, MUL VL]\n" + "addvl x15, x15, #5\n" + "lsl x20, x20, #0x2\n" + ".inst 0xa14049e4 // ld1w { z4.s, z12.s }, pn10.b/Z, [x15]\n" ".inst 0xa04149e0 // ld1w { z0.s-z1.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" - "sub x22, x22, x21\n" - "madd x20, x20, x6, x14\n" - "ld1w { z6.s }, p2/Z, [x15, #4, MUL VL]\n" + "ld1w { z5.s }, p2/Z, [x15, #4, MUL VL]\n" "addvl x15, x15, #5\n" "3:" // Issue prefetches "subs x22, x22, #0x1\n" - ".inst 0xf8b74a9c // rprfm pldstrm, x23, [x20]\n" - "add x20, x20, %x[ld_in_col], LSL #2\n" + ".inst 0xf8b44abc // rprfm pldstrm, x20, [x21]\n" + "add x21, x21, %x[ld_in_col], LSL #2\n" "bgt 3b\n" "ldr x23, [%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 x20, [%x[args], %[offsetof_Args_ld_out_cols]]\n" - ".inst 0xc0040f81 // mova za.d[x8, #1], { z28.d-z31.d }\n" + "lsl x21, %x[ld_in_row], #0x2\n" + ".inst 0xc0040e80 // mova za.d[x8, #0], { z20.d-z23.d }\n" "mov x22, #0x4\n" + "ldr x20, [%x[args], %[offsetof_Args_ld_out_cols]]\n" + "msub x14, x6, x21, x14\n" + ".inst 0xc0040e81 // mova za.d[x8, #1], { z20.d-z23.d }\n" + "ldr x21, [%x[args], %[offsetof_Args_pad_left]]\n" + ".inst 0xc0040e82 // mova za.d[x8, #2], { z20.d-z23.d }\n" "ldp x11, x10, [x23], #0x10\n" - ".inst 0xc0040f82 // mova za.d[x8, #2], { z28.d-z31.d }\n" + ".inst 0xc0040e83 // mova za.d[x8, #3], { z20.d-z23.d }\n" "ldp x9, x28, [x20], #0x10\n" - ".inst 0xc0040f83 // mova za.d[x8, #3], { z28.d-z31.d }\n" - "ldr x21, [%x[args], %[offsetof_Args_pad_left]]\n" - ".inst 0xc0040f84 // mova za.d[x8, #4], { z28.d-z31.d }\n" + ".inst 0xc0040e84 // mova za.d[x8, #4], { z20.d-z23.d }\n" "ldp x27, x26, [x23], #0x10\n" "ldp x25, x24, [x20], #0x10\n" "cbz x21, 5f\n" @@ -141,21 +141,21 @@ void sme2_fp32_planar_5x5_s2_4rows_mla_za_impl( "sub x21, x21, x20\n" "sub x22, x22, x20\n" "cbz x21, 5f\n" - ".inst 0xc0060c10 // mova { z16.d-z19.d }, za.d[x8, #0]\n" + ".inst 0xc0060c18 // mova { z24.d-z27.d }, za.d[x8, #0]\n" "and x22, x21, #0x1\n" "add x21, x21, #0x1\n" - ".inst 0xc1a3c850 // fclamp { z16.s-z19.s }, z2.s, z3.s\n" "lsr x21, x21, #0x1\n" "sub x13, x13, x21\n" + ".inst 0xc1b0ca38 // fclamp { z24.s-z27.s }, z17.s, z16.s\n" "4:" // Left padding "subs x21, x21, #0x1\n" - "st1w { z16.s }, p1, [x11]\n" + "st1w { z24.s }, p1, [x11]\n" "add x11, x11, x9, LSL #2\n" - "st1w { z17.s }, p1, [x10]\n" + "st1w { z25.s }, p1, [x10]\n" "add x10, x10, x28, LSL #2\n" - "st1w { z18.s }, p1, [x27]\n" + "st1w { z26.s }, p1, [x27]\n" "add x27, x27, x25, LSL #2\n" - "st1w { z19.s }, p1, [x26]\n" + "st1w { z27.s }, p1, [x26]\n" "add x26, x26, x24, LSL #2\n" "bgt 4b\n" "5:" // Left padding: End @@ -171,331 +171,331 @@ void sme2_fp32_planar_5x5_s2_4rows_mla_za_impl( "beq 7f\n" "6:" // Unpadded: 4 priming loads "add x20, x14, %x[ld_in_row], LSL #2\n" - "ld1w { z9.s }, p1/Z, [x14]\n" + "ld1w { z24.s }, p1/Z, [x14]\n" "add x14, x14, %x[ld_in_col], LSL #2\n" - "ld1w { z20.s }, p1/Z, [x20]\n" + "ld1w { z9.s }, p1/Z, [x20]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z25.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" "ld1w { z10.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" - "ld1w { z21.s }, p1/Z, [x20]\n" + "ld1w { z26.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" "ld1w { z11.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" - "ld1w { z22.s }, p1/Z, [x20]\n" + "ld1w { z27.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" - ".inst 0xc1341920 // fmla za.s[x8, 0], { z9.s-z12.s }, z4.s\n" - "ld1w { z23.s }, p1/Z, [x20]\n" - "add x20, x20, %x[ld_in_row], LSL #2\n" - ".inst 0xc1371a80 // fmla za.s[x8, 0], { z20.s-z23.s }, z7.s\n" - "ld1w { z13.s }, p1/Z, [x20]\n" + "ld1w { z28.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" + ".inst 0xc1371b00 // fmla za.s[x8, 0], { z24.s-z27.s }, z7.s\n" + ".inst 0xa04049e2 // ld1w { z2.s-z3.s }, pn10.b/Z, [x15]\n" "addvl x15, x15, #5\n" - ".inst 0xc1341940 // fmla za.s[x8, 0], { z10.s-z13.s }, z4.s\n" - "ld1w { z24.s }, p1/Z, [x20]\n" + "ld1w { z13.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" - ".inst 0xa04049e0 // ld1w { z0.s-z1.s }, pn10.b/Z, [x15]\n" + "ld1w { z29.s }, p1/Z, [x20]\n" + ".inst 0xc1341920 // fmla za.s[x8, 0], { z9.s-z12.s }, z4.s\n" + ".inst 0xa14049e1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x15]\n" "addvl x15, x15, #5\n" - ".inst 0xc1301aa0 // fmla za.s[x8, 0], { z21.s-z24.s }, z0.s\n" - "ld1w { z14.s }, p1/Z, [x20]\n" + ".inst 0xc1321b20 // fmla za.s[x8, 0], { z25.s-z28.s }, z2.s\n" ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" "ldr x15, [%x[args], %[offsetof_Args_weights]]\n" - ".inst 0xc1341960 // fmla za.s[x8, 0], { z11.s-z14.s }, z4.s\n" - ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" + ".inst 0xc1311940 // fmla za.s[x8, 0], { z10.s-z13.s }, z1.s\n" + ".inst 0xc1341b40 // fmla za.s[x8, 0], { z26.s-z29.s }, z4.s\n" + ".inst 0xa04049ee // ld1w { z14.s-z15.s }, pn10.b/Z, [x15]\n" "addvl x15, x15, #5\n" - ".inst 0xa14049e7 // ld1w { z7.s, z15.s }, pn10.b/Z, [x15]\n" + ".inst 0xa14049e4 // ld1w { z4.s, z12.s }, pn10.b/Z, [x15]\n" "addvl x15, x15, #5\n" "7:" // Unpadded: 3 priming loads "add x20, x14, %x[ld_in_row], LSL #2\n" - "ld1w { z22.s }, p1/Z, [x14]\n" + "ld1w { z28.s }, p1/Z, [x14]\n" "add x14, x14, %x[ld_in_col], LSL #2\n" "ld1w { z7.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" - "ld1w { z23.s }, p1/Z, [x20]\n" + "ld1w { z29.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" "ld1w { z8.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" - "ld1w { z24.s }, p1/Z, [x20]\n" + "ld1w { z30.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" "ld1w { z9.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" - "ld1w { z25.s }, p1/Z, [x20]\n" + "ld1w { z31.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" - ".inst 0xc1351ac0 // fmla za.s[x8, 0], { z22.s-z25.s }, z5.s\n" "ld1w { z10.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" - ".inst 0xc13f18e0 // fmla za.s[x8, 0], { z7.s-z10.s }, z15.s\n" - "ld1w { z26.s }, p1/Z, [x20]\n" + "ld1w { z0.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" - ".inst 0xa04049ee // ld1w { z14.s-z15.s }, pn10.b/Z, [x15]\n" + ".inst 0xc13f1b80 // fmla za.s[x8, 0], { z28.s-z31.s }, z15.s\n" + ".inst 0xa04049e2 // ld1w { z2.s-z3.s }, pn10.b/Z, [x15]\n" "addvl x15, x15, #5\n" - ".inst 0xc13f1ae0 // fmla za.s[x8, 0], { z23.s-z26.s }, z15.s\n" "ld1w { z11.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" - ".inst 0xa04049e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15]\n" + "ld1w { z1.s }, p1/Z, [x20]\n" + ".inst 0xc13c18e0 // fmla za.s[x8, 0], { z7.s-z10.s }, z12.s\n" + ".inst 0xa14049e6 // ld1w { z6.s, z14.s }, pn10.b/Z, [x15]\n" "addvl x15, x15, #5\n" - ".inst 0xc1371900 // fmla za.s[x8, 0], { z8.s-z11.s }, z7.s\n" - "ld1w { z27.s }, p1/Z, [x20]\n" - ".inst 0xa04049ea // ld1w { z10.s-z11.s }, pn10.b/Z, [x15]\n" + ".inst 0xc1331ba0 // fmla za.s[x8, 0], { z29.s-z0.s }, z3.s\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" "ldr x15, [%x[args], %[offsetof_Args_weights]]\n" - ".inst 0xc13b1b00 // fmla za.s[x8, 0], { z24.s-z27.s }, z11.s\n" - ".inst 0xa14049e4 // ld1w { z4.s, z12.s }, pn10.b/Z, [x15]\n" - ".inst 0xa04149ea // ld1w { z10.s-z11.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" - "addvl x15, x15, #5\n" + ".inst 0xa04149e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + ".inst 0xc13e1900 // fmla za.s[x8, 0], { z8.s-z11.s }, z14.s\n" + ".inst 0xc1351bc0 // fmla za.s[x8, 0], { z30.s-z1.s }, z5.s\n" ".inst 0xa14049e7 // ld1w { z7.s, z15.s }, pn10.b/Z, [x15]\n" + "addvl x15, x15, #5\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" ".inst 0xa04149e0 // ld1w { z0.s-z1.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" "addvl x15, x15, #5\n" "8:" // Unpadded: 2 priming loads "add x20, x14, %x[ld_in_row], LSL #2\n" - "ld1w { z19.s }, p1/Z, [x14]\n" + "ld1w { z25.s }, p1/Z, [x14]\n" "add x14, x14, %x[ld_in_col], LSL #2\n" - "ld1w { z14.s }, p1/Z, [x20]\n" + "ld1w { z10.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" - "ld1w { z20.s }, p1/Z, [x20]\n" + "ld1w { z26.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" - "ld1w { z15.s }, p1/Z, [x20]\n" + "ld1w { z11.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" - "ld1w { z21.s }, p1/Z, [x20]\n" + "ld1w { z27.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" - "ld1w { z16.s }, p1/Z, [x20]\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" + "ld1w { z28.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" - ".inst 0xc13a1a60 // fmla za.s[x8, 0], { z19.s-z22.s }, z10.s\n" - ".inst 0xc1341a61 // fmla za.s[x8, 1], { z19.s-z22.s }, z4.s\n" - "ld1w { z17.s }, p1/Z, [x20]\n" + "ld1w { z13.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" - ".inst 0xc13019c0 // fmla za.s[x8, 0], { z14.s-z17.s }, z0.s\n" - "ld1w { z23.s }, p1/Z, [x20]\n" + "ld1w { z29.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" - ".inst 0xc13719c1 // fmla za.s[x8, 1], { z14.s-z17.s }, z7.s\n" - ".inst 0xa04049e8 // ld1w { z8.s-z9.s }, pn10.b/Z, [x15]\n" - ".inst 0xa04149ea // ld1w { z10.s-z11.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + ".inst 0xc1361b20 // fmla za.s[x8, 0], { z25.s-z28.s }, z6.s\n" + ".inst 0xa04149e2 // ld1w { z2.s-z3.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + ".inst 0xc1371b21 // fmla za.s[x8, 1], { z25.s-z28.s }, z7.s\n" + ".inst 0xa04049e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15]\n" "addvl x15, x15, #5\n" - ".inst 0xc13a1a80 // fmla za.s[x8, 0], { z20.s-z23.s }, z10.s\n" - ".inst 0xc1381a81 // fmla za.s[x8, 1], { z20.s-z23.s }, z8.s\n" - "ld1w { z18.s }, p1/Z, [x20]\n" + "ld1w { z14.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" - ".inst 0xa04049e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15]\n" - ".inst 0xc13619e1 // fmla za.s[x8, 1], { z15.s-z18.s }, z6.s\n" + "ld1w { z30.s }, p1/Z, [x20]\n" + ".inst 0xc1301940 // fmla za.s[x8, 0], { z10.s-z13.s }, z0.s\n" ".inst 0xa04149e8 // ld1w { z8.s-z9.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + ".inst 0xc1341941 // fmla za.s[x8, 1], { z10.s-z13.s }, z4.s\n" + ".inst 0xa14049e1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x15]\n" "addvl x15, x15, #5\n" - ".inst 0xc13819e0 // fmla za.s[x8, 0], { z15.s-z18.s }, z8.s\n" - "ld1w { z24.s }, p1/Z, [x20]\n" - ".inst 0xa04049ee // ld1w { z14.s-z15.s }, pn10.b/Z, [x15]\n" - ".inst 0xc13e1aa1 // fmla za.s[x8, 1], { z21.s-z24.s }, z14.s\n" - ".inst 0xa14149e7 // ld1w { z7.s, z15.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" - "ldr x15, [%x[args], %[offsetof_Args_weights]]\n" - ".inst 0xc1371aa0 // fmla za.s[x8, 0], { z21.s-z24.s }, z7.s\n" + ".inst 0xc1321b40 // fmla za.s[x8, 0], { z26.s-z29.s }, z2.s\n" + ".inst 0xa14149e2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + ".inst 0xc1361b41 // fmla za.s[x8, 1], { z26.s-z29.s }, z6.s\n" ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" - ".inst 0xa04149ea // ld1w { z10.s-z11.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "ldr x15, [%x[args], %[offsetof_Args_weights]]\n" + ".inst 0xc1381960 // fmla za.s[x8, 0], { z11.s-z14.s }, z8.s\n" + ".inst 0xc1311961 // fmla za.s[x8, 1], { z11.s-z14.s }, z1.s\n" + ".inst 0xc1321b60 // fmla za.s[x8, 0], { z27.s-z30.s }, z2.s\n" + ".inst 0xa14149e6 // ld1w { z6.s, z14.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + ".inst 0xc1341b61 // fmla za.s[x8, 1], { z27.s-z30.s }, z4.s\n" + ".inst 0xa14049e7 // ld1w { z7.s, z15.s }, pn10.b/Z, [x15]\n" "addvl x15, x15, #5\n" - ".inst 0xa04049ee // ld1w { z14.s-z15.s }, pn10.b/Z, [x15]\n" + ".inst 0xa14049e4 // ld1w { z4.s, z12.s }, pn10.b/Z, [x15]\n" ".inst 0xa04149e0 // ld1w { z0.s-z1.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" "addvl x15, x15, #5\n" "9:" // Unpadded: 1 priming loads "add x20, x14, %x[ld_in_row], LSL #2\n" - "ld1w { z7.s }, p1/Z, [x14]\n" + "ld1w { z24.s }, p1/Z, [x14]\n" "add x14, x14, %x[ld_in_col], LSL #2\n" - "ld1w { z18.s }, p1/Z, [x20]\n" + "ld1w { z3.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" - "ld1w { z8.s }, p1/Z, [x20]\n" + "ld1w { z25.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" - "ld1w { z19.s }, p1/Z, [x20]\n" + "ld1w { z4.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" - "ld1w { z9.s }, p1/Z, [x20]\n" + "ld1w { z26.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" - "ld1w { z20.s }, p1/Z, [x20]\n" + "ld1w { z5.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" - "ld1w { z10.s }, p1/Z, [x20]\n" + "ld1w { z27.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" - ".inst 0xc13b18e0 // fmla za.s[x8, 0], { z7.s-z10.s }, z11.s\n" - ".inst 0xc13518e1 // fmla za.s[x8, 1], { z7.s-z10.s }, z5.s\n" - "ld1w { z21.s }, p1/Z, [x20]\n" + "ld1w { z6.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" - ".inst 0xc1311a40 // fmla za.s[x8, 0], { z18.s-z21.s }, z1.s\n" - "ld1w { z11.s }, p1/Z, [x20]\n" + "ld1w { z28.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" - ".inst 0xc13f1a41 // fmla za.s[x8, 1], { z18.s-z21.s }, z15.s\n" - ".inst 0xa04049e0 // ld1w { z0.s-z1.s }, pn10.b/Z, [x15]\n" - ".inst 0xa04149ec // ld1w { z12.s-z13.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + ".inst 0xc13e1b00 // fmla za.s[x8, 0], { z24.s-z27.s }, z14.s\n" + ".inst 0xa04149e8 // ld1w { z8.s-z9.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + ".inst 0xc13f1b01 // fmla za.s[x8, 1], { z24.s-z27.s }, z15.s\n" + ".inst 0xa14049e2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x15]\n" "addvl x15, x15, #5\n" - ".inst 0xc13d1900 // fmla za.s[x8, 0], { z8.s-z11.s }, z13.s\n" - ".inst 0xc1311901 // fmla za.s[x8, 1], { z8.s-z11.s }, z1.s\n" - "ld1w { z22.s }, p1/Z, [x20]\n" + "ld1w { z7.s }, p1/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" - ".inst 0xa14049e6 // ld1w { z6.s, z14.s }, pn10.b/Z, [x15]\n" - ".inst 0xc13e1a61 // fmla za.s[x8, 1], { z19.s-z22.s }, z14.s\n" - ".inst 0xa14149e6 // ld1w { z6.s, z14.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" - "addvl x15, x15, #5\n" - ".inst 0xc13e1a60 // fmla za.s[x8, 0], { z19.s-z22.s }, z14.s\n" - "ld1w { z12.s }, p1/Z, [x20]\n" + "ld1w { z29.s }, p1/Z, [x20]\n" + ".inst 0xc1311860 // fmla za.s[x8, 0], { z3.s-z6.s }, z1.s\n" + ".inst 0xa04149e0 // ld1w { z0.s-z1.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + ".inst 0xc13c1861 // fmla za.s[x8, 1], { z3.s-z6.s }, z12.s\n" ".inst 0xa04049ee // ld1w { z14.s-z15.s }, pn10.b/Z, [x15]\n" - ".inst 0xc13f1921 // fmla za.s[x8, 1], { z9.s-z12.s }, z15.s\n" - ".inst 0xa04149ee // ld1w { z14.s-z15.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" - "ldr x15, [%x[args], %[offsetof_Args_weights]]\n" - ".inst 0xc13f1920 // fmla za.s[x8, 0], { z9.s-z12.s }, z15.s\n" - ".inst 0xa14049e4 // ld1w { z4.s, z12.s }, pn10.b/Z, [x15]\n" - ".inst 0xa04149ea // ld1w { z10.s-z11.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" - "ld1w { z9.s }, p2/Z, [x15, #4, MUL VL]\n" "addvl x15, x15, #5\n" + ".inst 0xc1391b20 // fmla za.s[x8, 0], { z25.s-z28.s }, z9.s\n" + ".inst 0xa04149e8 // ld1w { z8.s-z9.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + ".inst 0xc13a1b21 // fmla za.s[x8, 1], { z25.s-z28.s }, z10.s\n" + ".inst 0xa04049ec // ld1w { z12.s-z13.s }, pn10.b/Z, [x15]\n" + "ldr x15, [%x[args], %[offsetof_Args_weights]]\n" + ".inst 0xc1311880 // fmla za.s[x8, 0], { z4.s-z7.s }, z1.s\n" + "ld1w { z8.s }, p2/Z, [x15, #4, MUL VL]\n" + ".inst 0xc13f1881 // fmla za.s[x8, 1], { z4.s-z7.s }, z15.s\n" + ".inst 0xc1391b40 // fmla za.s[x8, 0], { z26.s-z29.s }, z9.s\n" + ".inst 0xa14149e6 // ld1w { z6.s, z14.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + ".inst 0xc13d1b41 // fmla za.s[x8, 1], { z26.s-z29.s }, z13.s\n" ".inst 0xa14049e7 // ld1w { z7.s, z15.s }, pn10.b/Z, [x15]\n" + "addvl x15, x15, #5\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" ".inst 0xa04149e0 // ld1w { z0.s-z1.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" - "ld1w { z6.s }, p2/Z, [x15, #4, MUL VL]\n" + "ld1w { z5.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 { z22.s }, p1/Z, [x14]\n" + "ld1w { z9.s }, p1/Z, [x14]\n" "sub x16, x16, #0x2\n" - "ld1w { z16.s }, p1/Z, [x21]\n" + "ld1w { z26.s }, p1/Z, [x21]\n" "add x21, x21, %x[ld_in_row], LSL #2\n" "sub x13, x13, #0x1\n" - "ld1w { z23.s }, p1/Z, [x21]\n" + "ld1w { z10.s }, p1/Z, [x21]\n" "add x21, x21, %x[ld_in_row], LSL #2\n" "lsr x20, x16, #0x1\n" - "ld1w { z17.s }, p1/Z, [x21]\n" + "ld1w { z27.s }, p1/Z, [x21]\n" "add x21, x21, %x[ld_in_row], LSL #2\n" "cmp x20, x13\n" - "ld1w { z24.s }, p1/Z, [x21]\n" + "ld1w { z11.s }, p1/Z, [x21]\n" "add x21, x21, %x[ld_in_row], LSL #2\n" "csel x23, x20, x13, LT\n" - "ld1w { z18.s }, p1/Z, [x21]\n" + "ld1w { z28.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 { z25.s }, p1/Z, [x21]\n" + "ld1w { z12.s }, p1/Z, [x21]\n" "add x21, x21, %x[ld_in_row], LSL #2\n" "and x16, x16, #0x1\n" - "ld1w { z19.s }, p1/Z, [x21]\n" + "ld1w { z29.s }, p1/Z, [x21]\n" "add x21, x21, %x[ld_in_row], LSL #2\n" "sub x13, x13, x23\n" - "ld1w { z26.s }, p1/Z, [x21]\n" + "ld1w { z13.s }, p1/Z, [x21]\n" "add x21, x21, %x[ld_in_row], LSL #2\n" - "ld1w { z20.s }, p1/Z, [x21]\n" + "ld1w { z30.s }, p1/Z, [x21]\n" "add x21, x21, %x[ld_in_row], LSL #2\n" - "ld1w { z27.s }, p1/Z, [x21]\n" + "ld1w { z14.s }, p1/Z, [x21]\n" "cbz x23, 19f\n" "11:" // Unpadded: Main loop - ".inst 0xc1391ac0 // fmla za.s[x8, 0], { z22.s-z25.s }, z9.s\n" - "ld1w { z13.s }, p2/Z, [x15, #4, MUL VL]\n" + ".inst 0xc1381920 // fmla za.s[x8, 0], { z9.s-z12.s }, z8.s\n" + "ld1w { z8.s }, p2/Z, [x15, #4, MUL VL]\n" "add x22, x14, %x[ld_in_row], LSL #2\n" "subs x23, x23, #0x1\n" - ".inst 0xc13a1ac1 // fmla za.s[x8, 1], { z22.s-z25.s }, z10.s\n" - ".inst 0xa14149e1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" - ".inst 0xc1341ac2 // fmla za.s[x8, 2], { z22.s-z25.s }, z4.s\n" - ".inst 0xa04049e8 // ld1w { z8.s-z9.s }, pn10.b/Z, [x15]\n" - "addvl x15, x15, #5\n" - ".inst 0xc1361a00 // fmla za.s[x8, 0], { z16.s-z19.s }, z6.s\n" - "ld1w { z11.s }, p2/Z, [x15, #4, MUL VL]\n" - ".inst 0xc1301a01 // fmla za.s[x8, 1], { z16.s-z19.s }, z0.s\n" - ".inst 0xa04149ee // ld1w { z14.s-z15.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" - ".inst 0xc1371a02 // fmla za.s[x8, 2], { z16.s-z19.s }, z7.s\n" + ".inst 0xc1361921 // fmla za.s[x8, 1], { z9.s-z12.s }, z6.s\n" + ".inst 0xa04149e2 // ld1w { z2.s-z3.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + ".inst 0xc1371922 // fmla za.s[x8, 2], { z9.s-z12.s }, z7.s\n" ".inst 0xa04049e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15]\n" "addvl x15, x15, #5\n" - ".inst 0xc13d1ae0 // fmla za.s[x8, 0], { z23.s-z26.s }, z13.s\n" - "ld1w { z4.s }, p2/Z, [x15, #4, MUL VL]\n" - ".inst 0xc1311ae1 // fmla za.s[x8, 1], { z23.s-z26.s }, z1.s\n" - ".inst 0xa04149e0 // ld1w { z0.s-z1.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" - ".inst 0xc1381ae2 // fmla za.s[x8, 2], { z23.s-z26.s }, z8.s\n" - ".inst 0xa04049ec // ld1w { z12.s-z13.s }, pn10.b/Z, [x15]\n" - "ldr x21, [%x[args], %[offsetof_Args_weights]]\n" - ".inst 0xc13b1a20 // fmla za.s[x8, 0], { z17.s-z20.s }, z11.s\n" - "ld1w { z15.s }, p1/Z, [x14]\n" + "ld1w { z31.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 0xc13e1a21 // fmla za.s[x8, 1], { z17.s-z20.s }, z14.s\n" - "ld1w { z22.s }, p1/Z, [x22]\n" + ".inst 0xc1351b40 // fmla za.s[x8, 0], { z26.s-z29.s }, z5.s\n" + "ld1w { z1.s }, p2/Z, [x15, #4, MUL VL]\n" + "add x21, x14, %x[ld_in_row], LSL #2\n" + ".inst 0xc1301b41 // fmla za.s[x8, 1], { z26.s-z29.s }, z0.s\n" + ".inst 0xa14149e7 // ld1w { z7.s, z15.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + ".inst 0xc1341b42 // fmla za.s[x8, 2], { z26.s-z29.s }, z4.s\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" + "addvl x15, x15, #5\n" + "ld1w { z26.s }, p1/Z, [x22]\n" "add x22, x22, %x[ld_in_row], LSL #2\n" - ".inst 0xc1361a22 // fmla za.s[x8, 2], { z17.s-z20.s }, z6.s\n" - "ld1w { z16.s }, p1/Z, [x22]\n" + ".inst 0xc1381940 // fmla za.s[x8, 0], { z10.s-z13.s }, z8.s\n" + "ld1w { z5.s }, p2/Z, [x15, #4, MUL VL]\n" + ".inst 0xc1321941 // fmla za.s[x8, 1], { z10.s-z13.s }, z2.s\n" + ".inst 0xa04149e2 // ld1w { z2.s-z3.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + ".inst 0xc1361942 // fmla za.s[x8, 2], { z10.s-z13.s }, z6.s\n" + ".inst 0xa04049e8 // ld1w { z8.s-z9.s }, pn10.b/Z, [x15]\n" + "ldr x20, [%x[args], %[offsetof_Args_weights]]\n" + "ld1w { z0.s }, p1/Z, [x22]\n" "add x22, x22, %x[ld_in_row], LSL #2\n" - "ld1w { z23.s }, p1/Z, [x22]\n" + ".inst 0xc1311b60 // fmla za.s[x8, 0], { z27.s-z30.s }, z1.s\n" + ".inst 0xc1371b61 // fmla za.s[x8, 1], { z27.s-z30.s }, z7.s\n" + ".inst 0xc1341b62 // fmla za.s[x8, 2], { z27.s-z30.s }, z4.s\n" + "ld1w { z27.s }, p1/Z, [x22]\n" "add x22, x22, %x[ld_in_row], LSL #2\n" - ".inst 0xc1341b00 // fmla za.s[x8, 0], { z24.s-z27.s }, z4.s\n" - ".inst 0xc1301b01 // fmla za.s[x8, 1], { z24.s-z27.s }, z0.s\n" - ".inst 0xa0414aa6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x21, #0x2, MUL VL]\n" - ".inst 0xc13c1b02 // fmla za.s[x8, 2], { z24.s-z27.s }, z12.s\n" - "ld1w { z17.s }, p1/Z, [x22]\n" + ".inst 0xc1351960 // fmla za.s[x8, 0], { z11.s-z14.s }, z5.s\n" + ".inst 0xc1321961 // fmla za.s[x8, 1], { z11.s-z14.s }, z2.s\n" + ".inst 0xa0414a82 // ld1w { z2.s-z3.s }, pn10.b/Z, [x20, #0x2, MUL VL]\n" + ".inst 0xc1381962 // fmla za.s[x8, 2], { z11.s-z14.s }, z8.s\n" + "ld1w { z1.s }, p1/Z, [x22]\n" "add x22, x22, %x[ld_in_row], LSL #2\n" - "ld1w { z24.s }, p1/Z, [x22]\n" + "ld1w { z28.s }, p1/Z, [x22]\n" "add x22, x22, %x[ld_in_row], LSL #2\n" - ".inst 0xc0060c08 // mova { z8.d-z11.d }, za.d[x8, #0]\n" + ".inst 0xa1404a82 // ld1w { z2.s, z10.s }, pn10.b/Z, [x20]\n" + ".inst 0xc0060c0c // mova { z12.d-z15.d }, za.d[x8, #0]\n" "add x8, x8, #0x1\n" - ".inst 0xa1404aa4 // ld1w { z4.s, z12.s }, pn10.b/Z, [x21]\n" - "addvl x21, x21, #5\n" - ".inst 0xc1a3c848 // fclamp { z8.s-z11.s }, z2.s, z3.s\n" - "st1w { z8.s }, p1, [x11]\n" - "ld1w { z18.s }, p1/Z, [x22]\n" + "addvl x20, x20, #5\n" + "ld1w { z2.s }, p1/Z, [x22]\n" + "add x22, x22, %x[ld_in_row], LSL #2\n" + ".inst 0xc0040e84 // mova za.d[x8, #4], { z20.d-z23.d }\n" + ".inst 0xa0404a84 // ld1w { z4.s-z5.s }, pn10.b/Z, [x20]\n" + ".inst 0xa0414a86 // ld1w { z6.s-z7.s }, pn10.b/Z, [x20, #0x2, MUL VL]\n" + "addvl x20, x20, #5\n" + ".inst 0xc1331be0 // fmla za.s[x8, 0], { z31.s-z2.s }, z3.s\n" + "ld1w { z29.s }, p1/Z, [x22]\n" + "add x22, x22, %x[ld_in_row], LSL #2\n" + ".inst 0xc1b0ca2c // fclamp { z12.s-z15.s }, z17.s, z16.s\n" + ".inst 0xc13a1be1 // fmla za.s[x8, 1], { z31.s-z2.s }, z10.s\n" + "ld1w { z3.s }, p1/Z, [x22]\n" + "add x22, x22, %x[ld_in_row], LSL #2\n" + ".inst 0xa0404a8a // ld1w { z10.s-z11.s }, pn10.b/Z, [x20]\n" + ".inst 0xa0414a88 // ld1w { z8.s-z9.s }, pn10.b/Z, [x20, #0x2, MUL VL]\n" + "addvl x20, x20, #5\n" + ".inst 0xc1371b40 // fmla za.s[x8, 0], { z26.s-z29.s }, z7.s\n" + "ld1w { z30.s }, p1/Z, [x22]\n" "add x22, x22, %x[ld_in_row], LSL #2\n" - ".inst 0xc13719e0 // fmla za.s[x8, 0], { z15.s-z18.s }, z7.s\n" + "st1w { z12.s }, p1, [x11]\n" + ".inst 0xc1351b41 // fmla za.s[x8, 1], { z26.s-z29.s }, z5.s\n" + ".inst 0xa1404a84 // ld1w { z4.s, z12.s }, pn10.b/Z, [x20]\n" "add x11, x11, x9, LSL #2\n" - ".inst 0xc13c19e1 // fmla za.s[x8, 1], { z15.s-z18.s }, z12.s\n" - ".inst 0xa1404aa7 // ld1w { z7.s, z15.s }, pn10.b/Z, [x21]\n" - "st1w { z9.s }, p1, [x10]\n" + "st1w { z13.s }, p1, [x10]\n" + ".inst 0xa0414a84 // ld1w { z4.s-z5.s }, pn10.b/Z, [x20, #0x2, MUL VL]\n" + "addvl x20, x20, #5\n" "add x10, x10, x28, LSL #2\n" - ".inst 0xa1414aa6 // ld1w { z6.s, z14.s }, pn10.b/Z, [x21, #0x2, MUL VL]\n" - "addvl x21, x21, #5\n" - "st1w { z10.s }, p1, [x27]\n" + "st1w { z14.s }, p1, [x27]\n" + "ld1w { z4.s }, p1/Z, [x22]\n" "add x27, x27, x25, LSL #2\n" - "ld1w { z25.s }, p1/Z, [x22]\n" - "add x22, x22, %x[ld_in_row], LSL #2\n" - ".inst 0xc13e1ac0 // fmla za.s[x8, 0], { z22.s-z25.s }, z14.s\n" - "st1w { z11.s }, p1, [x26]\n" - ".inst 0xc13f1ac1 // fmla za.s[x8, 1], { z22.s-z25.s }, z15.s\n" - "ld1w { z19.s }, p1/Z, [x22]\n" - "add x22, x22, %x[ld_in_row], LSL #2\n" + "st1w { z15.s }, p1, [x26]\n" "add x26, x26, x24, LSL #2\n" - ".inst 0xa0404aae // ld1w { z14.s-z15.s }, pn10.b/Z, [x21]\n" - ".inst 0xc13f1a01 // fmla za.s[x8, 1], { z16.s-z19.s }, z15.s\n" - ".inst 0xa1414aa4 // ld1w { z4.s, z12.s }, pn10.b/Z, [x21, #0x2, MUL VL]\n" - "addvl x21, x21, #5\n" - ".inst 0xc13c1a00 // fmla za.s[x8, 0], { z16.s-z19.s }, z12.s\n" - "ld1w { z26.s }, p1/Z, [x22]\n" - "add x22, x22, %x[ld_in_row], LSL #2\n" - ".inst 0xc0040f84 // mova za.d[x8, #4], { z28.d-z31.d }\n" - ".inst 0xa0404aac // ld1w { z12.s-z13.s }, pn10.b/Z, [x21]\n" - ".inst 0xc13d1ae1 // fmla za.s[x8, 1], { z23.s-z26.s }, z13.s\n" - ".inst 0xa1414aa4 // ld1w { z4.s, z12.s }, pn10.b/Z, [x21, #0x2, MUL VL]\n" - "addvl x21, x21, #5\n" - ".inst 0xc13c1ae0 // fmla za.s[x8, 0], { z23.s-z26.s }, z12.s\n" - "ld1w { z20.s }, p1/Z, [x22]\n" - ".inst 0xa1404aa7 // ld1w { z7.s, z15.s }, pn10.b/Z, [x21]\n" - ".inst 0xc13f1a21 // fmla za.s[x8, 1], { z17.s-z20.s }, z15.s\n" - ".inst 0xa0414aaa // ld1w { z10.s-z11.s }, pn10.b/Z, [x21, #0x2, MUL VL]\n" - ".inst 0xc13b1a20 // fmla za.s[x8, 0], { z17.s-z20.s }, z11.s\n" + ".inst 0xc1391800 // fmla za.s[x8, 0], { z0.s-z3.s }, z9.s\n" + ".inst 0xa0414a86 // ld1w { z6.s-z7.s }, pn10.b/Z, [x20, #0x2, MUL VL]\n" + ".inst 0xc13b1801 // fmla za.s[x8, 1], { z0.s-z3.s }, z11.s\n" + ".inst 0xa1404a86 // ld1w { z6.s, z14.s }, pn10.b/Z, [x20]\n" "ldr x15, [%x[args], %[offsetof_Args_weights]]\n" - "ld1w { z22.s }, p1/Z, [x14]\n" + "ld1w { z9.s }, p1/Z, [x14]\n" "add x14, x14, %x[ld_in_col], LSL #2\n" - "ld1w { z16.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 { z17.s }, p1/Z, [x20]\n" - "add x20, x20, %x[ld_in_row], LSL #2\n" - "ld1w { z24.s }, p1/Z, [x20]\n" - "add x20, x20, %x[ld_in_row], LSL #2\n" - "ld1w { z18.s }, p1/Z, [x20]\n" - "add x20, x20, %x[ld_in_row], LSL #2\n" - "ld1w { z25.s }, p1/Z, [x20]\n" - "add x20, x20, %x[ld_in_row], LSL #2\n" - "ld1w { z19.s }, p1/Z, [x20]\n" - "add x20, x20, %x[ld_in_row], LSL #2\n" - "ld1w { z26.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" - ".inst 0xa04149ea // ld1w { z10.s-z11.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" - "ld1w { z9.s }, p2/Z, [x15, #4, MUL VL]\n" - "addvl x15, x15, #5\n" - "ld1w { z20.s }, p1/Z, [x20]\n" - "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z26.s }, p1/Z, [x21]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" + ".inst 0xc1351b60 // fmla za.s[x8, 0], { z27.s-z30.s }, z5.s\n" + "ld1w { z10.s }, p1/Z, [x21]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" + ".inst 0xc13c1b61 // fmla za.s[x8, 1], { z27.s-z30.s }, z12.s\n" + "ld1w { z27.s }, p1/Z, [x21]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" + "ld1w { z8.s }, p2/Z, [x15, #4, MUL VL]\n" + ".inst 0xc1371820 // fmla za.s[x8, 0], { z1.s-z4.s }, z7.s\n" + ".inst 0xa04149e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + ".inst 0xc13e1821 // fmla za.s[x8, 1], { z1.s-z4.s }, z14.s\n" + "ld1w { z11.s }, p1/Z, [x21]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" + "ld1w { z28.s }, p1/Z, [x21]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" + "ld1w { z12.s }, p1/Z, [x21]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" + "ld1w { z29.s }, p1/Z, [x21]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" + "ld1w { z13.s }, p1/Z, [x21]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" ".inst 0xa14049e7 // ld1w { z7.s, z15.s }, pn10.b/Z, [x15]\n" + "addvl x15, x15, #5\n" + "ld1w { z30.s }, p1/Z, [x21]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" ".inst 0xa04149e0 // ld1w { z0.s-z1.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" - "ld1w { z6.s }, p2/Z, [x15, #4, MUL VL]\n" + "ld1w { z5.s }, p2/Z, [x15, #4, MUL VL]\n" "addvl x15, x15, #5\n" - "ld1w { z27.s }, p1/Z, [x20]\n" + "ld1w { z14.s }, p1/Z, [x21]\n" "bgt 11b\n" "b 19f\n" "12:" // Padded @@ -509,625 +509,625 @@ void sme2_fp32_planar_5x5_s2_4rows_mla_za_impl( "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 { z9.s }, p0/Z, [x14]\n" "add x20, x14, %x[ld_in_row], LSL #2\n" + ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" + "ld1w { z29.s }, p0/Z, [x14]\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" + "add x14, x14, %x[ld_in_col], LSL #2\n" + "ld1w { z8.s }, p0/Z, [x20]\n" ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" - "ld1w { z10.s }, p0/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z30.s }, p0/Z, [x20]\n" ".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n" - "ld1w { z24.s }, p0/Z, [x20]\n" - "mov x12, #0x4\n" "add x20, x20, %x[ld_in_row], LSL #2\n" + "mov x12, #0x4\n" + "ld1w { z9.s }, p0/Z, [x20]\n" ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" - "ld1w { z11.s }, p0/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z31.s }, p0/Z, [x20]\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" - "ld1w { z25.s }, p0/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z10.s }, p0/Z, [x20]\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 { z0.s }, p0/Z, [x20]\n" ".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n" - "mov x12, #0x8\n" - ".inst 0xc1341920 // fmla za.s[x8, 0], { z9.s-z12.s }, z4.s\n" - "ld1w { z26.s }, p0/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" + "mov x12, #0x8\n" + ".inst 0xc1371ba0 // fmla za.s[x8, 0], { z29.s-z0.s }, z7.s\n" + "ld1w { z11.s }, p0/Z, [x20]\n" ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" - ".inst 0xc1371ae0 // fmla za.s[x8, 0], { z23.s-z26.s }, z7.s\n" - "ld1w { z13.s }, p0/Z, [x20]\n" - ".inst 0xa04049e6 // ld1w { z6.s-z7.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 0xc1361940 // fmla za.s[x8, 0], { z10.s-z13.s }, z6.s\n" + ".inst 0xa04049e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15]\n" "addvl x15, x15, #5\n" - "ld1w { z27.s }, p0/Z, [x20]\n" + "ld1w { z1.s }, p0/Z, [x20]\n" + ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" - ".inst 0xa14049e6 // ld1w { z6.s, z14.s }, pn10.b/Z, [x15]\n" - ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" + ".inst 0xc1341900 // fmla za.s[x8, 0], { z8.s-z11.s }, z4.s\n" + ".inst 0xa14049e4 // ld1w { z4.s, z12.s }, pn10.b/Z, [x15]\n" "addvl x15, x15, #5\n" - ".inst 0xc1361b00 // fmla za.s[x8, 0], { z24.s-z27.s }, z6.s\n" - "ld1w { z14.s }, p0/Z, [x20]\n" - "add x14, x14, %x[ld_in_col], LSL #2\n" - ".inst 0xa04049e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15]\n" + "ld1w { z12.s }, p0/Z, [x20]\n" + ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" + ".inst 0xc1361bc0 // fmla za.s[x8, 0], { z30.s-z1.s }, z6.s\n" + ".inst 0xa04049ee // ld1w { z14.s-z15.s }, pn10.b/Z, [x15]\n" "ldr x15, [%x[args], %[offsetof_Args_weights]]\n" - ".inst 0xc1361960 // fmla za.s[x8, 0], { z11.s-z14.s }, z6.s\n" - ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" - "addvl x15, x15, #5\n" + "ld1w { z2.s }, p0/Z, [x20]\n" + ".inst 0xc1341920 // fmla za.s[x8, 0], { z9.s-z12.s }, z4.s\n" + ".inst 0xc13e1be0 // fmla za.s[x8, 0], { z31.s-z2.s }, z14.s\n" ".inst 0xa14049e7 // ld1w { z7.s, z15.s }, pn10.b/Z, [x15]\n" "addvl x15, x15, #5\n" + ".inst 0xa14049e4 // ld1w { z4.s, z12.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 { z22.s }, p0/Z, [x14]\n" "add x20, x14, %x[ld_in_row], LSL #2\n" + ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" + "ld1w { z30.s }, p0/Z, [x14]\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" - "ld1w { z9.s }, p0/Z, [x20]\n" - "add x20, x20, %x[ld_in_row], LSL #2\n" + "add x14, x14, %x[ld_in_col], LSL #2\n" + "ld1w { z26.s }, p0/Z, [x20]\n" ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" - "ld1w { z23.s }, p0/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z31.s }, p0/Z, [x20]\n" ".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n" - "ld1w { z10.s }, p0/Z, [x20]\n" - "mov x12, #0x4\n" "add x20, x20, %x[ld_in_row], LSL #2\n" + "mov x12, #0x4\n" + "ld1w { z27.s }, p0/Z, [x20]\n" ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" - "ld1w { z24.s }, p0/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z0.s }, p0/Z, [x20]\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" - "ld1w { z11.s }, p0/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z28.s }, p0/Z, [x20]\n" ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" - "ld1w { z25.s }, p0/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z1.s }, p0/Z, [x20]\n" ".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n" - "mov x12, #0x8\n" - ".inst 0xc1351ac0 // fmla za.s[x8, 0], { z22.s-z25.s }, z5.s\n" - "ld1w { z12.s }, p0/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" + "mov x12, #0x8\n" + ".inst 0xc13f1bc0 // fmla za.s[x8, 0], { z30.s-z1.s }, z15.s\n" + "ld1w { z29.s }, p0/Z, [x20]\n" ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" - ".inst 0xc13f1920 // fmla za.s[x8, 0], { z9.s-z12.s }, z15.s\n" - "ld1w { z26.s }, p0/Z, [x20]\n" - ".inst 0xa14049e0 // ld1w { z0.s, z8.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 0xc1381ae0 // fmla za.s[x8, 0], { z23.s-z26.s }, z8.s\n" + ".inst 0xa14049e6 // ld1w { z6.s, z14.s }, pn10.b/Z, [x15]\n" "addvl x15, x15, #5\n" - "ld1w { z13.s }, p0/Z, [x20]\n" + "ld1w { z2.s }, p0/Z, [x20]\n" + ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" - ".inst 0xa14049e7 // ld1w { z7.s, z15.s }, pn10.b/Z, [x15]\n" - ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" + ".inst 0xc13c1b40 // fmla za.s[x8, 0], { z26.s-z29.s }, z12.s\n" + ".inst 0xa04049ec // ld1w { z12.s-z13.s }, pn10.b/Z, [x15]\n" "addvl x15, x15, #5\n" - ".inst 0xc13f1940 // fmla za.s[x8, 0], { z10.s-z13.s }, z15.s\n" - "ld1w { z27.s }, p0/Z, [x20]\n" - "add x14, x14, %x[ld_in_col], LSL #2\n" - ".inst 0xa04049ee // ld1w { z14.s-z15.s }, pn10.b/Z, [x15]\n" + "ld1w { z30.s }, p0/Z, [x20]\n" + ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" + ".inst 0xc13e1be0 // fmla za.s[x8, 0], { z31.s-z2.s }, z14.s\n" + ".inst 0xa04049ea // ld1w { z10.s-z11.s }, pn10.b/Z, [x15]\n" "ldr x15, [%x[args], %[offsetof_Args_weights]]\n" - ".inst 0xc13f1b00 // fmla za.s[x8, 0], { z24.s-z27.s }, z15.s\n" - ".inst 0xa14049e4 // ld1w { z4.s, z12.s }, pn10.b/Z, [x15]\n" - ".inst 0xa04149ea // ld1w { z10.s-z11.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" - "addvl x15, x15, #5\n" + "ld1w { z3.s }, p0/Z, [x20]\n" + ".inst 0xa04149e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + ".inst 0xc13d1b60 // fmla za.s[x8, 0], { z27.s-z30.s }, z13.s\n" + ".inst 0xc13b1800 // fmla za.s[x8, 0], { z0.s-z3.s }, z11.s\n" ".inst 0xa14049e7 // ld1w { z7.s, z15.s }, pn10.b/Z, [x15]\n" + "addvl x15, x15, #5\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" ".inst 0xa04149e0 // ld1w { z0.s-z1.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 { z16.s }, p0/Z, [x14]\n" "add x20, x14, %x[ld_in_row], LSL #2\n" + ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" + "ld1w { z25.s }, p0/Z, [x14]\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" + "add x14, x14, %x[ld_in_col], LSL #2\n" + "ld1w { z11.s }, p0/Z, [x20]\n" ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" - "ld1w { z17.s }, p0/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z26.s }, p0/Z, [x20]\n" ".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n" - "ld1w { z24.s }, p0/Z, [x20]\n" - "mov x12, #0x4\n" "add x20, x20, %x[ld_in_row], LSL #2\n" + "mov x12, #0x4\n" + "ld1w { z12.s }, p0/Z, [x20]\n" ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" - "ld1w { z18.s }, p0/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z27.s }, p0/Z, [x20]\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" - "ld1w { z25.s }, p0/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z13.s }, p0/Z, [x20]\n" ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" - "ld1w { z19.s }, p0/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z28.s }, p0/Z, [x20]\n" ".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n" - "mov x12, #0x8\n" - ".inst 0xc13a1a00 // fmla za.s[x8, 0], { z16.s-z19.s }, z10.s\n" - "ld1w { z26.s }, p0/Z, [x20]\n" - ".inst 0xc1341a01 // fmla za.s[x8, 1], { z16.s-z19.s }, z4.s\n" "add x20, x20, %x[ld_in_row], LSL #2\n" + "mov x12, #0x8\n" + ".inst 0xc1361b20 // fmla za.s[x8, 0], { z25.s-z28.s }, z6.s\n" + "ld1w { z14.s }, p0/Z, [x20]\n" ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" - "ld1w { z20.s }, p0/Z, [x20]\n" - ".inst 0xc1301ae0 // fmla za.s[x8, 0], { z23.s-z26.s }, z0.s\n" - ".inst 0xa14049e0 // ld1w { z0.s, z8.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 0xc1371ae1 // fmla za.s[x8, 1], { z23.s-z26.s }, z7.s\n" - ".inst 0xa14149e7 // ld1w { z7.s, z15.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + ".inst 0xc1371b21 // fmla za.s[x8, 1], { z25.s-z28.s }, z7.s\n" + ".inst 0xa14049e2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x15]\n" + ".inst 0xa14149e1 // ld1w { z1.s, z9.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 0xc1371a20 // fmla za.s[x8, 0], { z17.s-z20.s }, z7.s\n" - "ld1w { z27.s }, p0/Z, [x20]\n" + "ld1w { z29.s }, p0/Z, [x20]\n" + ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\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 0xc1301a21 // fmla za.s[x8, 1], { z17.s-z20.s }, z0.s\n" - ".inst 0xa14049e5 // ld1w { z5.s, z13.s }, pn10.b/Z, [x15]\n" - ".inst 0xa04149ea // ld1w { z10.s-z11.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" - "addvl x15, x15, #5\n" - ".inst 0xc13a1b00 // fmla za.s[x8, 0], { z24.s-z27.s }, z10.s\n" - ".inst 0xc1351b01 // fmla za.s[x8, 1], { z24.s-z27.s }, z5.s\n" - "ld1w { z21.s }, p0/Z, [x20]\n" - ".inst 0xa14049e5 // ld1w { z5.s, z13.s }, pn10.b/Z, [x15]\n" - ".inst 0xc1351a41 // fmla za.s[x8, 1], { z18.s-z21.s }, z5.s\n" - ".inst 0xa04149e0 // ld1w { z0.s-z1.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" - "ldr x15, [%x[args], %[offsetof_Args_weights]]\n" - ".inst 0xc1301a40 // fmla za.s[x8, 0], { z18.s-z21.s }, z0.s\n" + ".inst 0xc1301960 // fmla za.s[x8, 0], { z11.s-z14.s }, z0.s\n" + ".inst 0xa04149e8 // ld1w { z8.s-z9.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + ".inst 0xc1341961 // fmla za.s[x8, 1], { z11.s-z14.s }, z4.s\n" ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" - ".inst 0xa04149ea // ld1w { z10.s-z11.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" "addvl x15, x15, #5\n" + "ld1w { z15.s }, p0/Z, [x20]\n" + ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" + ".inst 0xc1311b40 // fmla za.s[x8, 0], { z26.s-z29.s }, z1.s\n" + ".inst 0xa14149e3 // ld1w { z3.s, z11.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + ".inst 0xc1321b41 // fmla za.s[x8, 1], { z26.s-z29.s }, z2.s\n" + "ld1w { z30.s }, p0/Z, [x20]\n" + ".inst 0xa04049e0 // ld1w { z0.s-z1.s }, pn10.b/Z, [x15]\n" + "ldr x15, [%x[args], %[offsetof_Args_weights]]\n" + ".inst 0xc1381980 // fmla za.s[x8, 0], { z12.s-z15.s }, z8.s\n" + ".inst 0xc1341981 // fmla za.s[x8, 1], { z12.s-z15.s }, z4.s\n" + ".inst 0xc1331b60 // fmla za.s[x8, 0], { z27.s-z30.s }, z3.s\n" + ".inst 0xa14149e6 // ld1w { z6.s, z14.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + ".inst 0xc1301b61 // fmla za.s[x8, 1], { z27.s-z30.s }, z0.s\n" ".inst 0xa14049e7 // ld1w { z7.s, z15.s }, pn10.b/Z, [x15]\n" + "addvl x15, x15, #5\n" + ".inst 0xa14049e4 // ld1w { z4.s, z12.s }, pn10.b/Z, [x15]\n" ".inst 0xa04149e0 // ld1w { z0.s-z1.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 { z19.s }, p0/Z, [x14]\n" "add x20, x14, %x[ld_in_row], LSL #2\n" + ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" + "ld1w { z25.s }, p0/Z, [x14]\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" + "add x14, x14, %x[ld_in_col], LSL #2\n" "ld1w { z8.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 { z20.s }, p0/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z26.s }, p0/Z, [x20]\n" ".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n" - "ld1w { z9.s }, p0/Z, [x20]\n" - "mov x12, #0x4\n" "add x20, x20, %x[ld_in_row], LSL #2\n" + "mov x12, #0x4\n" + "ld1w { z9.s }, p0/Z, [x20]\n" ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" - "ld1w { z21.s }, p0/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z27.s }, p0/Z, [x20]\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" - "ld1w { z10.s }, p0/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z10.s }, p0/Z, [x20]\n" ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" - "ld1w { z22.s }, p0/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z28.s }, p0/Z, [x20]\n" ".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" "mov x12, #0x8\n" - ".inst 0xc13b1a60 // fmla za.s[x8, 0], { z19.s-z22.s }, z11.s\n" + ".inst 0xc13e1b20 // fmla za.s[x8, 0], { z25.s-z28.s }, z14.s\n" "ld1w { z11.s }, p0/Z, [x20]\n" - ".inst 0xc1351a61 // fmla za.s[x8, 1], { z19.s-z22.s }, z5.s\n" - "add x20, x20, %x[ld_in_row], LSL #2\n" ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" - "ld1w { z23.s }, p0/Z, [x20]\n" - ".inst 0xc1311900 // fmla za.s[x8, 0], { z8.s-z11.s }, z1.s\n" - ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" + ".inst 0xc13f1b21 // fmla za.s[x8, 1], { z25.s-z28.s }, z15.s\n" + ".inst 0xa14049e7 // ld1w { z7.s, z15.s }, pn10.b/Z, [x15]\n" + ".inst 0xa04149e2 // ld1w { z2.s-z3.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "addvl x15, x15, #5\n" + "ld1w { z29.s }, p0/Z, [x20]\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" - ".inst 0xc13f1901 // fmla za.s[x8, 1], { z8.s-z11.s }, z15.s\n" - ".inst 0xa14149e6 // ld1w { z6.s, z14.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" + ".inst 0xc1311900 // fmla za.s[x8, 0], { z8.s-z11.s }, z1.s\n" + ".inst 0xa14149e5 // ld1w { z5.s, z13.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + ".inst 0xc13c1901 // fmla za.s[x8, 1], { z8.s-z11.s }, z12.s\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" "addvl x15, x15, #5\n" - "add x14, x14, %x[ld_in_col], LSL #2\n" - ".inst 0xc13e1a80 // fmla za.s[x8, 0], { z20.s-z23.s }, z14.s\n" "ld1w { z12.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 0xc1351a81 // fmla za.s[x8, 1], { z20.s-z23.s }, z5.s\n" - ".inst 0xa04049e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15]\n" - ".inst 0xa14149e0 // ld1w { z0.s, z8.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" - "addvl x15, x15, #5\n" - ".inst 0xc1381920 // fmla za.s[x8, 0], { z9.s-z12.s }, z8.s\n" - ".inst 0xc1371921 // fmla za.s[x8, 1], { z9.s-z12.s }, z7.s\n" - "ld1w { z24.s }, p0/Z, [x20]\n" - ".inst 0xa14049e0 // ld1w { z0.s, z8.s }, pn10.b/Z, [x15]\n" - ".inst 0xc1381aa1 // fmla za.s[x8, 1], { z21.s-z24.s }, z8.s\n" - ".inst 0xa04149ec // ld1w { z12.s-z13.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" + ".inst 0xc1331b40 // fmla za.s[x8, 0], { z26.s-z29.s }, z3.s\n" + ".inst 0xa04149e0 // ld1w { z0.s-z1.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + ".inst 0xc13f1b41 // fmla za.s[x8, 1], { z26.s-z29.s }, z15.s\n" + "ld1w { z30.s }, p0/Z, [x20]\n" + ".inst 0xa04049e2 // ld1w { z2.s-z3.s }, pn10.b/Z, [x15]\n" "ldr x15, [%x[args], %[offsetof_Args_weights]]\n" - ".inst 0xc13d1aa0 // fmla za.s[x8, 0], { z21.s-z24.s }, z13.s\n" - ".inst 0xa14049e4 // ld1w { z4.s, z12.s }, pn10.b/Z, [x15]\n" - ".inst 0xa04149ea // ld1w { z10.s-z11.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" - "ld1w { z9.s }, p2/Z, [x15, #4, MUL VL]\n" - "addvl x15, x15, #5\n" + ".inst 0xc13d1920 // fmla za.s[x8, 0], { z9.s-z12.s }, z13.s\n" + ".inst 0xc1351921 // fmla za.s[x8, 1], { z9.s-z12.s }, z5.s\n" + "ld1w { z8.s }, p2/Z, [x15, #4, MUL VL]\n" + ".inst 0xc1311b60 // fmla za.s[x8, 0], { z27.s-z30.s }, z1.s\n" + ".inst 0xa04149e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + ".inst 0xc1331b61 // fmla za.s[x8, 1], { z27.s-z30.s }, z3.s\n" ".inst 0xa14049e7 // ld1w { z7.s, z15.s }, pn10.b/Z, [x15]\n" + "addvl x15, x15, #5\n" + ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" ".inst 0xa04149e0 // ld1w { z0.s-z1.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" - "ld1w { z6.s }, p2/Z, [x15, #4, MUL VL]\n" + "ld1w { z5.s }, p2/Z, [x15, #4, MUL VL]\n" "addvl x15, x15, #5\n" "17:" // Padded: 0 priming loads "cmp x16, #0x2\n" "blt 20f\n" "mov x12, #0x0\n" - ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" - "ld1w { z22.s }, p0/Z, [x14]\n" "add x21, x14, %x[ld_in_row], LSL #2\n" + ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" + "sub x16, x16, #0x2\n" + "sub x13, x13, #0x1\n" + "lsr x20, x16, #0x1\n" + "cmp x20, x13\n" + "and x16, x16, #0x1\n" + "ld1w { z9.s }, p0/Z, [x14]\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" - "ld1w { z16.s }, p0/Z, [x21]\n" - "add x21, x21, %x[ld_in_row], LSL #2\n" + "csel x23, x20, x13, LT\n" + "add x14, x14, %x[ld_in_col], LSL #2\n" + "sub x13, x13, x23\n" + "ld1w { z26.s }, p0/Z, [x21]\n" ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" - "ld1w { z23.s }, p0/Z, [x21]\n" "add x21, x21, %x[ld_in_row], LSL #2\n" + "ld1w { z10.s }, p0/Z, [x21]\n" ".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n" - "ld1w { z17.s }, p0/Z, [x21]\n" - "mov x12, #0x4\n" "add x21, x21, %x[ld_in_row], LSL #2\n" + "mov x12, #0x4\n" + "ld1w { z27.s }, p0/Z, [x21]\n" ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" - "ld1w { z24.s }, p0/Z, [x21]\n" "add x21, x21, %x[ld_in_row], LSL #2\n" + "ld1w { z11.s }, p0/Z, [x21]\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" - "ld1w { z18.s }, p0/Z, [x21]\n" "add x21, x21, %x[ld_in_row], LSL #2\n" + "ld1w { z28.s }, p0/Z, [x21]\n" ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" - "ld1w { z25.s }, p0/Z, [x21]\n" "add x21, x21, %x[ld_in_row], LSL #2\n" + "ld1w { z12.s }, p0/Z, [x21]\n" ".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n" - "mov x12, #0x8\n" - "ld1w { z19.s }, p0/Z, [x21]\n" - "sub x16, x16, #0x2\n" "add x21, x21, %x[ld_in_row], LSL #2\n" + "mov x12, #0x8\n" + "ld1w { z29.s }, p0/Z, [x21]\n" ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" - "sub x13, x13, #0x1\n" - "ld1w { z26.s }, p0/Z, [x21]\n" - "lsr x20, x16, #0x1\n" "add x21, x21, %x[ld_in_row], LSL #2\n" + "ld1w { z13.s }, p0/Z, [x21]\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" - "cmp x20, x13\n" - "ld1w { z20.s }, p0/Z, [x21]\n" "add x21, x21, %x[ld_in_row], LSL #2\n" + "ld1w { z30.s }, p0/Z, [x21]\n" ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" - "ld1w { z27.s }, p0/Z, [x21]\n" - "csel x23, x20, x13, LT\n" - "add x14, x14, %x[ld_in_col], LSL #2\n" - "and x16, x16, #0x1\n" - "sub x13, x13, x23\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" + "ld1w { z14.s }, p0/Z, [x21]\n" "cbz x23, 19f\n" "18:" // Padded: Main loop - ".inst 0xc1391ac0 // fmla za.s[x8, 0], { z22.s-z25.s }, z9.s\n" - "ld1w { z15.s }, p2/Z, [x15, #4, MUL VL]\n" + ".inst 0xc1381920 // fmla za.s[x8, 0], { z9.s-z12.s }, z8.s\n" "mov x12, #0x0\n" - ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" - ".inst 0xc13a1ac1 // fmla za.s[x8, 1], { z22.s-z25.s }, z10.s\n" - ".inst 0xa04149ea // ld1w { z10.s-z11.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "ld1w { z15.s }, p2/Z, [x15, #4, MUL VL]\n" "add x22, x14, %x[ld_in_row], LSL #2\n" + ".inst 0xc1361921 // fmla za.s[x8, 1], { z9.s-z12.s }, z6.s\n" + ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" + ".inst 0xa04149e2 // ld1w { z2.s-z3.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" "subs x23, x23, #0x1\n" - ".inst 0xc1341ac2 // fmla za.s[x8, 2], { z22.s-z25.s }, z4.s\n" - ".inst 0xa14049e5 // ld1w { z5.s, z13.s }, pn10.b/Z, [x15]\n" - "addvl x15, x15, #5\n" - ".inst 0xc1361a00 // fmla za.s[x8, 0], { z16.s-z19.s }, z6.s\n" - "ld1w { z1.s }, p2/Z, [x15, #4, MUL VL]\n" - ".inst 0xc1301a01 // fmla za.s[x8, 1], { z16.s-z19.s }, z0.s\n" - ".inst 0xa04149ec // ld1w { z12.s-z13.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" - ".inst 0xc1371a02 // fmla za.s[x8, 2], { z16.s-z19.s }, z7.s\n" - ".inst 0xa14049e6 // ld1w { z6.s, z14.s }, pn10.b/Z, [x15]\n" + ".inst 0xc1371922 // fmla za.s[x8, 2], { z9.s-z12.s }, z7.s\n" + ".inst 0xa04049e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15]\n" "addvl x15, x15, #5\n" - ".inst 0xc13f1ae0 // fmla za.s[x8, 0], { z23.s-z26.s }, z15.s\n" - "ld1w { z16.s }, p0/Z, [x14]\n" + ".inst 0xc1351b40 // fmla za.s[x8, 0], { z26.s-z29.s }, z5.s\n" + "ld1w { z8.s }, p0/Z, [x14]\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" "add x14, x14, %x[ld_in_col], LSL #2\n" - ".inst 0xc13a1ae1 // fmla za.s[x8, 1], { z23.s-z26.s }, z10.s\n" - ".inst 0xa04149ea // ld1w { z10.s-z11.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + ".inst 0xc1301b41 // fmla za.s[x8, 1], { z26.s-z29.s }, z0.s\n" + ".inst 0xa04149e0 // ld1w { z0.s-z1.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" "add x21, x14, %x[ld_in_row], LSL #2\n" - ".inst 0xc1351ae2 // fmla za.s[x8, 2], { z23.s-z26.s }, z5.s\n" - ".inst 0xa04049ee // ld1w { z14.s-z15.s }, pn10.b/Z, [x15]\n" - ".inst 0xc1311a20 // fmla za.s[x8, 0], { z17.s-z20.s }, z1.s\n" - "ld1w { z0.s }, p2/Z, [x15, #4, MUL VL]\n" - "ldr x20, [%x[args], %[offsetof_Args_weights]]\n" - ".inst 0xc13c1a21 // fmla za.s[x8, 1], { z17.s-z20.s }, z12.s\n" - "ld1w { z12.s }, p0/Z, [x22]\n" - "add x22, x22, %x[ld_in_row], LSL #2\n" + ".inst 0xc1341b42 // fmla za.s[x8, 2], { z26.s-z29.s }, z4.s\n" + ".inst 0xa14049e1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x15]\n" + "ld1w { z31.s }, p0/Z, [x22]\n" ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" - ".inst 0xc1361a22 // fmla za.s[x8, 2], { z17.s-z20.s }, z6.s\n" - "ld1w { z17.s }, p0/Z, [x22]\n" "add x22, x22, %x[ld_in_row], LSL #2\n" + ".inst 0xc13f1940 // fmla za.s[x8, 0], { z10.s-z13.s }, z15.s\n" + "ld1w { z15.s }, p2/Z, [x15, #4, MUL VL]\n" + "addvl x15, x15, #5\n" + ".inst 0xc1321941 // fmla za.s[x8, 1], { z10.s-z13.s }, z2.s\n" + ".inst 0xa04149e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + ".inst 0xc1361942 // fmla za.s[x8, 2], { z10.s-z13.s }, z6.s\n" + "ld1w { z9.s }, p0/Z, [x22]\n" ".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n" - "mov x12, #0x4\n" - "ld1w { z13.s }, p0/Z, [x22]\n" "add x22, x22, %x[ld_in_row], LSL #2\n" - ".inst 0xc1301b00 // fmla za.s[x8, 0], { z24.s-z27.s }, z0.s\n" + "mov x12, #0x4\n" + ".inst 0xa14049e2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x15]\n" + ".inst 0xc13f1b60 // fmla za.s[x8, 0], { z27.s-z30.s }, z15.s\n" + "ld1w { z5.s }, p2/Z, [x15, #4, MUL VL]\n" + "ldr x20, [%x[args], %[offsetof_Args_weights]]\n" + ".inst 0xc1301b61 // fmla za.s[x8, 1], { z27.s-z30.s }, z0.s\n" + ".inst 0xc1311b62 // fmla za.s[x8, 2], { z27.s-z30.s }, z1.s\n" + "ld1w { z0.s }, p0/Z, [x22]\n" ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" - ".inst 0xc13a1b01 // fmla za.s[x8, 1], { z24.s-z27.s }, z10.s\n" - ".inst 0xa1414a81 // ld1w { z1.s, z9.s }, pn10.b/Z, [x20, #0x2, MUL VL]\n" - ".inst 0xc13e1b02 // fmla za.s[x8, 2], { z24.s-z27.s }, z14.s\n" - "ld1w { z18.s }, p0/Z, [x22]\n" "add x22, x22, %x[ld_in_row], LSL #2\n" + ".inst 0xc1351960 // fmla za.s[x8, 0], { z11.s-z14.s }, z5.s\n" + ".inst 0xc1341961 // fmla za.s[x8, 1], { z11.s-z14.s }, z4.s\n" + ".inst 0xa0414a86 // ld1w { z6.s-z7.s }, pn10.b/Z, [x20, #0x2, MUL VL]\n" + ".inst 0xc1321962 // fmla za.s[x8, 2], { z11.s-z14.s }, z2.s\n" + "ld1w { z10.s }, p0/Z, [x22]\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" - "ld1w { z14.s }, p0/Z, [x22]\n" "add x22, x22, %x[ld_in_row], LSL #2\n" - ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" + ".inst 0xa0404a82 // ld1w { z2.s-z3.s }, pn10.b/Z, [x20]\n" + "addvl x20, x20, #5\n" ".inst 0xc0060c18 // mova { z24.d-z27.d }, za.d[x8, #0]\n" - ".inst 0xa0404a80 // ld1w { z0.s-z1.s }, pn10.b/Z, [x20]\n" "add x8, x8, #0x1\n" - "addvl x20, x20, #5\n" - ".inst 0xc1a3c858 // fclamp { z24.s-z27.s }, z2.s, z3.s\n" - "ld1w { z19.s }, p0/Z, [x22]\n" + ".inst 0xa0404a8e // ld1w { z14.s-z15.s }, pn10.b/Z, [x20]\n" + "ld1w { z1.s }, p0/Z, [x22]\n" + ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" "add x22, x22, %x[ld_in_row], LSL #2\n" + ".inst 0xc0040e84 // mova za.d[x8, #4], { z20.d-z23.d }\n" + ".inst 0xa1414a86 // ld1w { z6.s, z14.s }, pn10.b/Z, [x20, #0x2, MUL VL]\n" + "addvl x20, x20, #5\n" + "ld1w { z11.s }, p0/Z, [x22]\n" ".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n" - "st1w { z24.s }, p1, [x11]\n" + "add x22, x22, %x[ld_in_row], LSL #2\n" + ".inst 0xc1b0ca38 // fclamp { z24.s-z27.s }, z17.s, z16.s\n" "mov x12, #0x8\n" - ".inst 0xc1391a00 // fmla za.s[x8, 0], { z16.s-z19.s }, z9.s\n" - ".inst 0xa0404a88 // ld1w { z8.s-z9.s }, pn10.b/Z, [x20]\n" + "ld1w { z2.s }, p0/Z, [x22]\n" + ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" + "add x22, x22, %x[ld_in_row], LSL #2\n" + ".inst 0xc1371900 // fmla za.s[x8, 0], { z8.s-z11.s }, z7.s\n" + ".inst 0xc1331901 // fmla za.s[x8, 1], { z8.s-z11.s }, z3.s\n" + ".inst 0xa0404a84 // ld1w { z4.s-z5.s }, pn10.b/Z, [x20]\n" + "st1w { z24.s }, p1, [x11]\n" "add x11, x11, x9, LSL #2\n" - ".inst 0xc1311a01 // fmla za.s[x8, 1], { z16.s-z19.s }, z1.s\n" - ".inst 0xa0414a80 // ld1w { z0.s-z1.s }, pn10.b/Z, [x20, #0x2, MUL VL]\n" + ".inst 0xa0414a8c // ld1w { z12.s-z13.s }, pn10.b/Z, [x20, #0x2, MUL VL]\n" "addvl x20, x20, #5\n" "st1w { z25.s }, p1, [x10]\n" - "ld1w { z15.s }, p0/Z, [x22]\n" - "add x22, x22, %x[ld_in_row], LSL #2\n" - ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" - ".inst 0xc1311980 // fmla za.s[x8, 0], { z12.s-z15.s }, z1.s\n" - ".inst 0xc1391981 // fmla za.s[x8, 1], { z12.s-z15.s }, z9.s\n" - "ld1w { z20.s }, p0/Z, [x22]\n" - "add x22, x22, %x[ld_in_row], LSL #2\n" - ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" - ".inst 0xa0404a8a // ld1w { z10.s-z11.s }, pn10.b/Z, [x20]\n" - ".inst 0xc13b1a21 // fmla za.s[x8, 1], { z17.s-z20.s }, z11.s\n" "add x10, x10, x28, LSL #2\n" + "ld1w { z12.s }, p0/Z, [x22]\n" + ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" + "add x22, x22, %x[ld_in_row], LSL #2\n" "st1w { z26.s }, p1, [x27]\n" - ".inst 0xa1414a80 // ld1w { z0.s, z8.s }, pn10.b/Z, [x20, #0x2, MUL VL]\n" - "addvl x20, x20, #5\n" - ".inst 0xc1381a20 // fmla za.s[x8, 0], { z17.s-z20.s }, z8.s\n" + ".inst 0xc13e1be0 // fmla za.s[x8, 0], { z31.s-z2.s }, z14.s\n" + ".inst 0xa1414a86 // ld1w { z6.s, z14.s }, pn10.b/Z, [x20, #0x2, MUL VL]\n" "add x27, x27, x25, LSL #2\n" - "ld1w { z16.s }, p0/Z, [x22]\n" - "add x22, x22, %x[ld_in_row], LSL #2\n" - ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" "st1w { z27.s }, p1, [x26]\n" - ".inst 0xa0404a88 // ld1w { z8.s-z9.s }, pn10.b/Z, [x20]\n" - "mov x12, #0x0\n" - ".inst 0xc13919a1 // fmla za.s[x8, 1], { z13.s-z16.s }, z9.s\n" - "add x26, x26, x24, LSL #2\n" - ".inst 0xa1414a81 // ld1w { z1.s, z9.s }, pn10.b/Z, [x20, #0x2, MUL VL]\n" + ".inst 0xc13f1be1 // fmla za.s[x8, 1], { z31.s-z2.s }, z15.s\n" + ".inst 0xa1404a87 // ld1w { z7.s, z15.s }, pn10.b/Z, [x20]\n" "addvl x20, x20, #5\n" - ".inst 0xc13919a0 // fmla za.s[x8, 0], { z13.s-z16.s }, z9.s\n" - "ld1w { z21.s }, p0/Z, [x22]\n" + "add x26, x26, x24, LSL #2\n" + "ld1w { z3.s }, p0/Z, [x22]\n" + ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" + "add x22, x22, %x[ld_in_row], LSL #2\n" + "mov x12, #0x0\n" + ".inst 0xc13d1920 // fmla za.s[x8, 0], { z9.s-z12.s }, z13.s\n" + ".inst 0xa0414a86 // ld1w { z6.s-z7.s }, pn10.b/Z, [x20, #0x2, MUL VL]\n" + "ld1w { z13.s }, p0/Z, [x22]\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 { z22.s }, p0/Z, [x14]\n" + ".inst 0xc1351921 // fmla za.s[x8, 1], { z9.s-z12.s }, z5.s\n" + ".inst 0xa0404a84 // ld1w { z4.s-z5.s }, pn10.b/Z, [x20]\n" + "ldr x15, [%x[args], %[offsetof_Args_weights]]\n" + "ld1w { z9.s }, p0/Z, [x14]\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" + ".inst 0xc13e1800 // fmla za.s[x8, 0], { z0.s-z3.s }, z14.s\n" "add x14, x14, %x[ld_in_col], LSL #2\n" - ".inst 0xa0404a8e // ld1w { z14.s-z15.s }, pn10.b/Z, [x20]\n" - ".inst 0xc13f1a41 // fmla za.s[x8, 1], { z18.s-z21.s }, z15.s\n" - ".inst 0xa0414a80 // ld1w { z0.s-z1.s }, pn10.b/Z, [x20, #0x2, MUL VL]\n" - ".inst 0xc1311a40 // fmla za.s[x8, 0], { z18.s-z21.s }, z1.s\n" - "ldr x15, [%x[args], %[offsetof_Args_weights]]\n" - "ld1w { z16.s }, p0/Z, [x21]\n" - "add x21, x21, %x[ld_in_row], LSL #2\n" + ".inst 0xc13f1801 // fmla za.s[x8, 1], { z0.s-z3.s }, z15.s\n" + "ld1w { z8.s }, p2/Z, [x15, #4, MUL VL]\n" + "ld1w { z26.s }, p0/Z, [x21]\n" ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" - "ld1w { z23.s }, p0/Z, [x21]\n" "add x21, x21, %x[ld_in_row], LSL #2\n" + ".inst 0xc1371940 // fmla za.s[x8, 0], { z10.s-z13.s }, z7.s\n" + ".inst 0xa04149e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + ".inst 0xc1351941 // fmla za.s[x8, 1], { z10.s-z13.s }, z5.s\n" + ".inst 0xa14049e7 // ld1w { z7.s, z15.s }, pn10.b/Z, [x15]\n" + "addvl x15, x15, #5\n" + "ld1w { z10.s }, p0/Z, [x21]\n" ".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n" - "mov x12, #0x4\n" - "ld1w { z17.s }, p0/Z, [x21]\n" "add x21, x21, %x[ld_in_row], LSL #2\n" + "mov x12, #0x4\n" + ".inst 0xa14049e4 // ld1w { z4.s, z12.s }, pn10.b/Z, [x15]\n" + ".inst 0xa04149e0 // ld1w { z0.s-z1.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "ld1w { z27.s }, p0/Z, [x21]\n" ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" - "ld1w { z24.s }, p0/Z, [x21]\n" "add x21, x21, %x[ld_in_row], LSL #2\n" + "ld1w { z5.s }, p2/Z, [x15, #4, MUL VL]\n" + "addvl x15, x15, #5\n" + "ld1w { z11.s }, p0/Z, [x21]\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" - "ld1w { z18.s }, p0/Z, [x21]\n" "add x21, x21, %x[ld_in_row], LSL #2\n" + "ld1w { z28.s }, p0/Z, [x21]\n" ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" - "ld1w { z25.s }, p0/Z, [x21]\n" "add x21, x21, %x[ld_in_row], LSL #2\n" + "ld1w { z12.s }, p0/Z, [x21]\n" ".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n" - "mov x12, #0x8\n" - "ld1w { z19.s }, p0/Z, [x21]\n" "add x21, x21, %x[ld_in_row], LSL #2\n" + "mov x12, #0x8\n" + "ld1w { z29.s }, p0/Z, [x21]\n" ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" - "ld1w { z26.s }, p0/Z, [x21]\n" "add x21, x21, %x[ld_in_row], LSL #2\n" + "ld1w { z13.s }, p0/Z, [x21]\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" - ".inst 0xa04049e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15]\n" - ".inst 0xa04149ea // ld1w { z10.s-z11.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" - "ld1w { z9.s }, p2/Z, [x15, #4, MUL VL]\n" - "addvl x15, x15, #5\n" - "ld1w { z20.s }, p0/Z, [x21]\n" "add x21, x21, %x[ld_in_row], LSL #2\n" + "ld1w { z30.s }, p0/Z, [x21]\n" ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" - ".inst 0xa14049e7 // ld1w { z7.s, z15.s }, pn10.b/Z, [x15]\n" - ".inst 0xa04149e0 // ld1w { z0.s-z1.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" - "ld1w { z6.s }, p2/Z, [x15, #4, MUL VL]\n" - "addvl x15, x15, #5\n" - "ld1w { z27.s }, p0/Z, [x21]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" + "ld1w { z14.s }, p0/Z, [x21]\n" "bgt 18b\n" "19:" // Main loop tail - ".inst 0xc1391ac0 // fmla za.s[x8, 0], { z22.s-z25.s }, z9.s\n" - "ld1w { z8.s }, p2/Z, [x15, #4, MUL VL]\n" + ".inst 0xc1381920 // fmla za.s[x8, 0], { z9.s-z12.s }, z8.s\n" "mov x12, #0x0\n" - ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" - ".inst 0xc13a1ac1 // fmla za.s[x8, 1], { z22.s-z25.s }, z10.s\n" - ".inst 0xa04149ea // ld1w { z10.s-z11.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "ld1w { z8.s }, p2/Z, [x15, #4, MUL VL]\n" "add x21, x14, %x[ld_in_row], LSL #2\n" - ".inst 0xc1341ac2 // fmla za.s[x8, 2], { z22.s-z25.s }, z4.s\n" - ".inst 0xa14049e1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x15]\n" - "addvl x15, x15, #5\n" - ".inst 0xc1361a00 // fmla za.s[x8, 0], { z16.s-z19.s }, z6.s\n" - "ld1w { z9.s }, p2/Z, [x15, #4, MUL VL]\n" - ".inst 0xc1301a01 // fmla za.s[x8, 1], { z16.s-z19.s }, z0.s\n" - ".inst 0xa04149ee // ld1w { z14.s-z15.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" - ".inst 0xc1371a02 // fmla za.s[x8, 2], { z16.s-z19.s }, z7.s\n" - ".inst 0xa04049ec // ld1w { z12.s-z13.s }, pn10.b/Z, [x15]\n" + ".inst 0xc1361921 // fmla za.s[x8, 1], { z9.s-z12.s }, z6.s\n" + ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" + ".inst 0xa04149e2 // ld1w { z2.s-z3.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + ".inst 0xc1371922 // fmla za.s[x8, 2], { z9.s-z12.s }, z7.s\n" + ".inst 0xa04049e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15]\n" "addvl x15, x15, #5\n" - ".inst 0xc1381ae0 // fmla za.s[x8, 0], { z23.s-z26.s }, z8.s\n" - "ld1w { z16.s }, p0/Z, [x14]\n" + ".inst 0xc1351b40 // fmla za.s[x8, 0], { z26.s-z29.s }, z5.s\n" + "ld1w { z9.s }, p0/Z, [x14]\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" "add x14, x14, %x[ld_in_col], LSL #2\n" - ".inst 0xc13a1ae1 // fmla za.s[x8, 1], { z23.s-z26.s }, z10.s\n" - ".inst 0xa14149e5 // ld1w { z5.s, z13.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" - ".inst 0xc1311ae2 // fmla za.s[x8, 2], { z23.s-z26.s }, z1.s\n" + ".inst 0xc1301b41 // fmla za.s[x8, 1], { z26.s-z29.s }, z0.s\n" + ".inst 0xa04149e0 // ld1w { z0.s-z1.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + ".inst 0xc1341b42 // fmla za.s[x8, 2], { z26.s-z29.s }, z4.s\n" ".inst 0xa14049e7 // ld1w { z7.s, z15.s }, pn10.b/Z, [x15]\n" - ".inst 0xc1391a20 // fmla za.s[x8, 0], { z17.s-z20.s }, z9.s\n" - "ld1w { z1.s }, p2/Z, [x15, #4, MUL VL]\n" - "ldr x20, [%x[args], %[offsetof_Args_weights]]\n" - ".inst 0xc13e1a21 // fmla za.s[x8, 1], { z17.s-z20.s }, z14.s\n" - "ld1w { z22.s }, p0/Z, [x21]\n" - "add x21, x21, %x[ld_in_row], LSL #2\n" + "ld1w { z31.s }, p0/Z, [x21]\n" ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" - ".inst 0xc13c1a22 // fmla za.s[x8, 2], { z17.s-z20.s }, z12.s\n" - "ld1w { z17.s }, p0/Z, [x21]\n" "add x21, x21, %x[ld_in_row], LSL #2\n" + ".inst 0xc1381940 // fmla za.s[x8, 0], { z10.s-z13.s }, z8.s\n" + "ld1w { z15.s }, p2/Z, [x15, #4, MUL VL]\n" + "addvl x15, x15, #5\n" + ".inst 0xc1321941 // fmla za.s[x8, 1], { z10.s-z13.s }, z2.s\n" + ".inst 0xa04149e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + ".inst 0xc1361942 // fmla za.s[x8, 2], { z10.s-z13.s }, z6.s\n" + "ld1w { z10.s }, p0/Z, [x21]\n" ".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n" - "mov x12, #0x4\n" - "ld1w { z23.s }, p0/Z, [x21]\n" "add x21, x21, %x[ld_in_row], LSL #2\n" - ".inst 0xc1311b00 // fmla za.s[x8, 0], { z24.s-z27.s }, z1.s\n" + "mov x12, #0x4\n" + ".inst 0xa04049e2 // ld1w { z2.s-z3.s }, pn10.b/Z, [x15]\n" + ".inst 0xc13f1b60 // fmla za.s[x8, 0], { z27.s-z30.s }, z15.s\n" + "ld1w { z6.s }, p2/Z, [x15, #4, MUL VL]\n" + "ldr x20, [%x[args], %[offsetof_Args_weights]]\n" + ".inst 0xc1301b61 // fmla za.s[x8, 1], { z27.s-z30.s }, z0.s\n" + ".inst 0xc1371b62 // fmla za.s[x8, 2], { z27.s-z30.s }, z7.s\n" + "ld1w { z0.s }, p0/Z, [x21]\n" ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" - ".inst 0xc1351b01 // fmla za.s[x8, 1], { z24.s-z27.s }, z5.s\n" - ".inst 0xa0414a8e // ld1w { z14.s-z15.s }, pn10.b/Z, [x20, #0x2, MUL VL]\n" - ".inst 0xc1371b02 // fmla za.s[x8, 2], { z24.s-z27.s }, z7.s\n" - "ld1w { z18.s }, p0/Z, [x21]\n" "add x21, x21, %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" + ".inst 0xa0414a84 // ld1w { z4.s-z5.s }, pn10.b/Z, [x20, #0x2, MUL VL]\n" + ".inst 0xc1321962 // fmla za.s[x8, 2], { z11.s-z14.s }, z2.s\n" + "ld1w { z11.s }, p0/Z, [x21]\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" - "ld1w { z24.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 0xc0060c08 // mova { z8.d-z11.d }, za.d[x8, #0]\n" - ".inst 0xa0404a84 // ld1w { z4.s-z5.s }, pn10.b/Z, [x20]\n" - "add x8, x8, #0x1\n" + ".inst 0xa0404a86 // ld1w { z6.s-z7.s }, pn10.b/Z, [x20]\n" "addvl x20, x20, #5\n" - ".inst 0xc1a3c848 // fclamp { z8.s-z11.s }, z2.s, z3.s\n" - "ld1w { z19.s }, p0/Z, [x21]\n" + ".inst 0xc0060c18 // mova { z24.d-z27.d }, za.d[x8, #0]\n" + "add x8, x8, #0x1\n" + ".inst 0xa0404a82 // ld1w { z2.s-z3.s }, pn10.b/Z, [x20]\n" + "ld1w { z1.s }, p0/Z, [x21]\n" + ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\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 { z8.s }, p1, [x11]\n" - "mov x12, #0x8\n" - ".inst 0xc13f1a00 // fmla za.s[x8, 0], { z16.s-z19.s }, z15.s\n" - ".inst 0xa0404a80 // ld1w { z0.s-z1.s }, pn10.b/Z, [x20]\n" - "add x11, x11, x9, LSL #2\n" - ".inst 0xc1351a01 // fmla za.s[x8, 1], { z16.s-z19.s }, z5.s\n" - ".inst 0xa1414a80 // ld1w { z0.s, z8.s }, pn10.b/Z, [x20, #0x2, MUL VL]\n" + ".inst 0xc0040e84 // mova za.d[x8, #4], { z20.d-z23.d }\n" + ".inst 0xa0414a8e // ld1w { z14.s-z15.s }, pn10.b/Z, [x20, #0x2, MUL VL]\n" "addvl x20, x20, #5\n" - "st1w { z9.s }, p1, [x10]\n" - "ld1w { z25.s }, p0/Z, [x21]\n" + "ld1w { z12.s }, p0/Z, [x21]\n" + ".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n" "add x21, x21, %x[ld_in_row], LSL #2\n" + ".inst 0xc1b0ca38 // fclamp { z24.s-z27.s }, z17.s, z16.s\n" + "mov x12, #0x8\n" + ".inst 0xc1351920 // fmla za.s[x8, 0], { z9.s-z12.s }, z5.s\n" + "ld1w { z2.s }, p0/Z, [x21]\n" ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" - ".inst 0xc1381ac0 // fmla za.s[x8, 0], { z22.s-z25.s }, z8.s\n" - ".inst 0xc1311ac1 // fmla za.s[x8, 1], { z22.s-z25.s }, z1.s\n" - "ld1w { z20.s }, p0/Z, [x21]\n" "add x21, x21, %x[ld_in_row], LSL #2\n" + ".inst 0xc1371921 // fmla za.s[x8, 1], { z9.s-z12.s }, z7.s\n" + ".inst 0xa0404a88 // ld1w { z8.s-z9.s }, pn10.b/Z, [x20]\n" + "st1w { z24.s }, p1, [x11]\n" + "add x11, x11, x9, LSL #2\n" + ".inst 0xa1414a86 // ld1w { z6.s, z14.s }, pn10.b/Z, [x20, #0x2, MUL VL]\n" + "addvl x20, x20, #5\n" + "st1w { z25.s }, p1, [x10]\n" + "add x10, x10, x28, LSL #2\n" + "ld1w { z13.s }, p0/Z, [x21]\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" + "st1w { z26.s }, p1, [x27]\n" + ".inst 0xc13f1be0 // fmla za.s[x8, 0], { z31.s-z2.s }, z15.s\n" + ".inst 0xa0414a84 // ld1w { z4.s-z5.s }, pn10.b/Z, [x20, #0x2, MUL VL]\n" + "add x27, x27, x25, LSL #2\n" + "st1w { z27.s }, p1, [x26]\n" + ".inst 0xc1331be1 // fmla za.s[x8, 1], { z31.s-z2.s }, z3.s\n" ".inst 0xa0404a86 // ld1w { z6.s-z7.s }, pn10.b/Z, [x20]\n" - ".inst 0xc1371a21 // fmla za.s[x8, 1], { z17.s-z20.s }, z7.s\n" - "add x10, x10, x28, LSL #2\n" - "st1w { z10.s }, p1, [x27]\n" - ".inst 0xa1414a81 // ld1w { z1.s, z9.s }, pn10.b/Z, [x20, #0x2, MUL VL]\n" "addvl x20, x20, #5\n" - ".inst 0xc1391a20 // fmla za.s[x8, 0], { z17.s-z20.s }, z9.s\n" - "add x27, x27, x25, LSL #2\n" - "ld1w { z26.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 { z11.s }, p1, [x26]\n" - ".inst 0xa1404a84 // ld1w { z4.s, z12.s }, pn10.b/Z, [x20]\n" - ".inst 0xc13c1ae1 // fmla za.s[x8, 1], { z23.s-z26.s }, z12.s\n" "add x26, x26, x24, LSL #2\n" - ".inst 0xa1414a84 // ld1w { z4.s, z12.s }, pn10.b/Z, [x20, #0x2, MUL VL]\n" - "addvl x20, x20, #5\n" - ".inst 0xc13c1ae0 // fmla za.s[x8, 0], { z23.s-z26.s }, z12.s\n" - "ld1w { z21.s }, p0/Z, [x21]\n" - ".inst 0xc0040f84 // mova za.d[x8, #4], { z28.d-z31.d }\n" - ".inst 0xa0404a80 // ld1w { z0.s-z1.s }, pn10.b/Z, [x20]\n" - ".inst 0xc1311a41 // fmla za.s[x8, 1], { z18.s-z21.s }, z1.s\n" - ".inst 0xa0414a80 // ld1w { z0.s-z1.s }, pn10.b/Z, [x20, #0x2, MUL VL]\n" + "ld1w { z3.s }, p0/Z, [x21]\n" + ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" + "add x21, x21, %x[ld_in_row], LSL #2\n" + ".inst 0xc13e1940 // fmla za.s[x8, 0], { z10.s-z13.s }, z14.s\n" + ".inst 0xa0414a8e // ld1w { z14.s-z15.s }, pn10.b/Z, [x20, #0x2, MUL VL]\n" + ".inst 0xc1391941 // fmla za.s[x8, 1], { z10.s-z13.s }, z9.s\n" + "ld1w { z14.s }, p0/Z, [x21]\n" + ".inst 0xa0404a88 // ld1w { z8.s-z9.s }, pn10.b/Z, [x20]\n" "ldr x15, [%x[args], %[offsetof_Args_weights]]\n" - ".inst 0xc1311a40 // fmla za.s[x8, 0], { z18.s-z21.s }, z1.s\n" - ".inst 0xa14049e4 // ld1w { z4.s, z12.s }, pn10.b/Z, [x15]\n" - ".inst 0xa04149ea // ld1w { z10.s-z11.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" - "ld1w { z9.s }, p2/Z, [x15, #4, MUL VL]\n" - "addvl x15, x15, #5\n" + ".inst 0xc1351800 // fmla za.s[x8, 0], { z0.s-z3.s }, z5.s\n" + ".inst 0xc1371801 // fmla za.s[x8, 1], { z0.s-z3.s }, z7.s\n" + "ld1w { z8.s }, p2/Z, [x15, #4, MUL VL]\n" + ".inst 0xc13f1960 // fmla za.s[x8, 0], { z11.s-z14.s }, z15.s\n" + ".inst 0xa04149e6 // ld1w { z6.s-z7.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + ".inst 0xc1391961 // fmla za.s[x8, 1], { z11.s-z14.s }, z9.s\n" ".inst 0xa14049e7 // ld1w { z7.s, z15.s }, pn10.b/Z, [x15]\n" + "addvl x15, x15, #5\n" + ".inst 0xa14049e4 // ld1w { z4.s, z12.s }, pn10.b/Z, [x15]\n" ".inst 0xa04149e0 // ld1w { z0.s-z1.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" - "ld1w { z6.s }, p2/Z, [x15, #4, MUL VL]\n" + "ld1w { z5.s }, p2/Z, [x15, #4, MUL VL]\n" "addvl x15, x15, #5\n" "20:" // Main loop skip tail "cbz x16, 21f\n" // Skip remainder inputs "mov x12, #0x0\n" - ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" - "ld1w { z16.s }, p0/Z, [x14]\n" "add x20, x14, %x[ld_in_row], LSL #2\n" + ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" + "sub x13, x13, #0x1\n" + "ld1w { z25.s }, p0/Z, [x14]\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 { z11.s }, p0/Z, [x20]\n" ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" - "ld1w { z17.s }, p0/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z26.s }, p0/Z, [x20]\n" ".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n" - "ld1w { z24.s }, p0/Z, [x20]\n" - "mov x12, #0x4\n" "add x20, x20, %x[ld_in_row], LSL #2\n" + "mov x12, #0x4\n" + "ld1w { z12.s }, p0/Z, [x20]\n" ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" - "ld1w { z18.s }, p0/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z27.s }, p0/Z, [x20]\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" - "ld1w { z25.s }, p0/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z13.s }, p0/Z, [x20]\n" ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" - "ld1w { z19.s }, p0/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z28.s }, p0/Z, [x20]\n" ".inst 0x25f04500 // psel p0.s, p1.s/Z, p8.s[w12, #3]\n" - "mov x12, #0x8\n" - ".inst 0xc1391a00 // fmla za.s[x8, 0], { z16.s-z19.s }, z9.s\n" - "ld1w { z26.s }, p0/Z, [x20]\n" "add x20, x20, %x[ld_in_row], LSL #2\n" + "mov x12, #0x8\n" + ".inst 0xc1381b20 // fmla za.s[x8, 0], { z25.s-z28.s }, z8.s\n" + "ld1w { z14.s }, p0/Z, [x20]\n" ".inst 0x25304500 // psel p0.s, p1.s/Z, p8.s[w12]\n" - ".inst 0xc1361ae0 // fmla za.s[x8, 0], { z23.s-z26.s }, z6.s\n" - "ld1w { z20.s }, p0/Z, [x20]\n" - "ld1w { z8.s }, p2/Z, [x15, #4, MUL VL]\n" - ".inst 0xc13a1a01 // fmla za.s[x8, 1], { z16.s-z19.s }, z10.s\n" "add x20, x20, %x[ld_in_row], LSL #2\n" + "ld1w { z8.s }, p2/Z, [x15, #4, MUL VL]\n" + ".inst 0xc1361b21 // fmla za.s[x8, 1], { z25.s-z28.s }, z6.s\n" + ".inst 0xc1371b22 // fmla za.s[x8, 2], { z25.s-z28.s }, z7.s\n" + ".inst 0xa14049e1 // ld1w { z1.s, z9.s }, pn10.b/Z, [x15]\n" + "ld1w { z29.s }, p0/Z, [x20]\n" ".inst 0x25704500 // psel p0.s, p1.s/Z, p8.s[w12, #1]\n" - ".inst 0xc1341a02 // fmla za.s[x8, 2], { z16.s-z19.s }, z4.s\n" - ".inst 0xa04049ea // ld1w { z10.s-z11.s }, pn10.b/Z, [x15]\n" - "sub x13, x13, #0x1\n" - ".inst 0xa04149ee // ld1w { z14.s-z15.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" - "addvl x15, x15, #5\n" - ".inst 0xc1381a20 // fmla za.s[x8, 0], { z17.s-z20.s }, z8.s\n" - "ld1w { z27.s }, p0/Z, [x20]\n" - ".inst 0xc1301ae1 // fmla za.s[x8, 1], { z23.s-z26.s }, z0.s\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" + ".inst 0xa14149e2 // ld1w { z2.s, z10.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + "addvl x15, x15, #5\n" + "ld1w { z10.s }, p2/Z, [x15, #4, MUL VL]\n" + ".inst 0xc1301961 // fmla za.s[x8, 1], { z11.s-z14.s }, z0.s\n" + "ld1w { z15.s }, p0/Z, [x20]\n" ".inst 0x25b04500 // psel p0.s, p1.s/Z, p8.s[w12, #2]\n" - "ld1w { z9.s }, p2/Z, [x15, #4, MUL VL]\n" - ".inst 0xc1371ae2 // fmla za.s[x8, 2], { z23.s-z26.s }, z7.s\n" - ".inst 0xa04049e0 // ld1w { z0.s-z1.s }, pn10.b/Z, [x15]\n" - ".inst 0xc1391b00 // fmla za.s[x8, 0], { z24.s-z27.s }, z9.s\n" - ".inst 0xa14149e5 // ld1w { z5.s, z13.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + ".inst 0xc1341962 // fmla za.s[x8, 2], { z11.s-z14.s }, z4.s\n" + "add x20, x20, %x[ld_in_row], LSL #2\n" + ".inst 0xa14049e3 // ld1w { z3.s, z11.s }, pn10.b/Z, [x15]\n" + ".inst 0xc1381b40 // fmla za.s[x8, 0], { z26.s-z29.s }, z8.s\n" + ".inst 0xa14149e0 // ld1w { z0.s, z8.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" "addvl x15, x15, #5\n" - ".inst 0xc13e1a21 // fmla za.s[x8, 1], { z17.s-z20.s }, z14.s\n" - "ld1w { z21.s }, p0/Z, [x20]\n" - ".inst 0xc13a1a22 // fmla za.s[x8, 2], { z17.s-z20.s }, z10.s\n" - "ld1w { z8.s }, p2/Z, [x15, #4, MUL VL]\n" - ".inst 0xc1381a40 // fmla za.s[x8, 0], { z18.s-z21.s }, z8.s\n" - ".inst 0xc1351b01 // fmla za.s[x8, 1], { z24.s-z27.s }, z5.s\n" + "ld1w { z30.s }, p0/Z, [x20]\n" + ".inst 0xc1321b41 // fmla za.s[x8, 1], { z26.s-z29.s }, z2.s\n" + "ld1w { z6.s }, p2/Z, [x15, #4, MUL VL]\n" + ".inst 0xc1311b42 // fmla za.s[x8, 2], { z26.s-z29.s }, z1.s\n" ".inst 0xa04049e8 // ld1w { z8.s-z9.s }, pn10.b/Z, [x15]\n" - ".inst 0xc1301b02 // fmla za.s[x8, 2], { z24.s-z27.s }, z0.s\n" - ".inst 0xa04149e0 // ld1w { z0.s-z1.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" - ".inst 0xc0060c18 // mova { z24.d-z27.d }, za.d[x8, #0]\n" - ".inst 0xc1a3c858 // fclamp { z24.s-z27.s }, z2.s, z3.s\n" - "st1w { z24.s }, p1, [x11]\n" + ".inst 0xc13a1980 // fmla za.s[x8, 0], { z12.s-z15.s }, z10.s\n" + ".inst 0xa04149e4 // ld1w { z4.s-z5.s }, pn10.b/Z, [x15, #0x2, MUL VL]\n" + ".inst 0xc1301981 // fmla za.s[x8, 1], { z12.s-z15.s }, z0.s\n" + ".inst 0xc1331982 // fmla za.s[x8, 2], { z12.s-z15.s }, z3.s\n" + ".inst 0xc1361b60 // fmla za.s[x8, 0], { z27.s-z30.s }, z6.s\n" + ".inst 0xc1341b61 // fmla za.s[x8, 1], { z27.s-z30.s }, z4.s\n" + ".inst 0xc1381b62 // fmla za.s[x8, 2], { z27.s-z30.s }, z8.s\n" + ".inst 0xc0060c04 // mova { z4.d-z7.d }, za.d[x8, #0]\n" + "add x8, x8, #0x1\n" + ".inst 0xc0040e84 // mova za.d[x8, #4], { z20.d-z23.d }\n" + ".inst 0xc1b0ca24 // fclamp { z4.s-z7.s }, z17.s, z16.s\n" + "st1w { z4.s }, p1, [x11]\n" "add x11, x11, x9, LSL #2\n" - ".inst 0xc1301a41 // fmla za.s[x8, 1], { z18.s-z21.s }, z0.s\n" - "st1w { z25.s }, p1, [x10]\n" + "st1w { z5.s }, p1, [x10]\n" "add x10, x10, x28, LSL #2\n" - ".inst 0xc1381a42 // fmla za.s[x8, 2], { z18.s-z21.s }, z8.s\n" - "add x8, x8, #0x1\n" - "st1w { z26.s }, p1, [x27]\n" + "st1w { z6.s }, p1, [x27]\n" "add x27, x27, x25, LSL #2\n" - "st1w { z27.s }, p1, [x26]\n" + "st1w { z7.s }, p1, [x26]\n" "add x26, x26, x24, LSL #2\n" - ".inst 0xc0040f84 // mova za.d[x8, #4], { z28.d-z31.d }\n" "21:" // Tail input: End "cbz x13, 23f\n" "22:" // Right padding loop ".inst 0xc0060c08 // mova { z8.d-z11.d }, za.d[x8, #0]\n" "add x8, x8, #0x1\n" "subs x13, x13, #0x1\n" - ".inst 0xc1a3c848 // fclamp { z8.s-z11.s }, z2.s, z3.s\n" + ".inst 0xc0040e84 // mova za.d[x8, #4], { z20.d-z23.d }\n" + ".inst 0xc1b0ca28 // fclamp { z8.s-z11.s }, z17.s, z16.s\n" "st1w { z8.s }, p1, [x11]\n" "add x11, x11, x9, LSL #2\n" - ".inst 0xc0040f84 // mova za.d[x8, #4], { z28.d-z31.d }\n" "st1w { z9.s }, p1, [x10]\n" "add x10, x10, x28, LSL #2\n" "st1w { z10.s }, p1, [x27]\n" @@ -1137,12 +1137,12 @@ void sme2_fp32_planar_5x5_s2_4rows_mla_za_impl( "bgt 22b\n" "23:" // End "ldr x20, [%x[args], %[offsetof_Args_weights]]\n" + "incw x17\n" + "whilelt p1.s, x17, x7\n" "incb x20, ALL, MUL #16\n" "incb x20, ALL, MUL #9\n" "str x20, [%x[args], %[offsetof_Args_weights]]\n" "ldr x21, [%x[args], %[offsetof_Args_ld_in_vl]]\n" - "incw x17\n" - "whilelt p1.s, x17, x7\n" "ldr x20, [%x[args], %[offsetof_Args_inptr]]\n" "add x20, x20, x21, LSL #2\n" "str x20, [%x[args], %[offsetof_Args_inptr]]\n" |