From be13cead34e566bdd561ad3ffc3f645b460e482e Mon Sep 17 00:00:00 2001 From: Michael Tyler Date: Tue, 17 Jan 2023 11:04:14 +0000 Subject: Revert "Update CPU kernels to remove x19" This reverts commit 3c59f01c209d2732a15d97d65565ead964787a8b. Resolves: COMPMID-5817 Change-Id: Ie2443a21854a95db1e3d0cafa2121c0187a5e237 Signed-off-by: Michael Tyler Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8974 Comments-Addressed: Arm Jenkins Tested-by: Arm Jenkins Reviewed-by: Gian Marco Iodice Benchmark: Arm Jenkins --- .../generic_direct.cpp | 1837 ++++++++++---------- 1 file changed, 919 insertions(+), 918 deletions(-) (limited to 'src/core/NEON/kernels/arm_conv/depthwise/kernels/a64_fp32_nhwc_3x3_s1_output4x4_mla_depthfirst/generic_direct.cpp') diff --git a/src/core/NEON/kernels/arm_conv/depthwise/kernels/a64_fp32_nhwc_3x3_s1_output4x4_mla_depthfirst/generic_direct.cpp b/src/core/NEON/kernels/arm_conv/depthwise/kernels/a64_fp32_nhwc_3x3_s1_output4x4_mla_depthfirst/generic_direct.cpp index 6d2b6ee998..6faacf144a 100644 --- a/src/core/NEON/kernels/arm_conv/depthwise/kernels/a64_fp32_nhwc_3x3_s1_output4x4_mla_depthfirst/generic_direct.cpp +++ b/src/core/NEON/kernels/arm_conv/depthwise/kernels/a64_fp32_nhwc_3x3_s1_output4x4_mla_depthfirst/generic_direct.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2021, 2023 Arm Limited. + * Copyright (c) 2021 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -87,1142 +87,1143 @@ void a64_fp32_nhwc_3x3_s1_output4x4_mla_depthfirst_direct_impl( ); __asm__ __volatile__( - "mov x27, #0x0\n" + "mov x4, #0x0\n" "mov x26, #0x0\n" "1:" // Tile loop - "str x27, [%x[params_struct], %[offsetof_args_tile_i]]\n" + "str x4, [%x[params_struct], %[offsetof_args_tile_i]]\n" "mov x25, #0x4\n" - "mov x23, #0x4\n" "str x26, [%x[params_struct], %[offsetof_args_tile_j]]\n" - "ldr x24, [%x[params_struct], %[offsetof_args_ld_input_row]]\n" - "ldr x22, [%x[params_struct], %[offsetof_args_ld_output_row]]\n" - "mul x21, x27, x24\n" // offset = tile_i * ld_input_row - "ldr x4, [%x[params_struct], %[offsetof_args_ld_input_col]]\n" - "ldr x5, [%x[params_struct], %[offsetof_args_ld_output_col]]\n" - "mul x20, x27, x22\n" // offset = tile_i * ld_output_row - "mov x6, #0x10\n" // cntb _, ALL, #1 - "madd x21, x26, x4, x21\n" // offset += tile_j * ld_input_col - "ldr x7, [%x[params_struct], %[offsetof_args_inptr]]\n" - "lsl x4, x4, #0x2\n" - "ldr x8, [%x[params_struct], %[offsetof_args_outptr]]\n" - "madd x20, x26, x5, x20\n" // offset += tile_j * ld_output_col - "lsl x5, x5, #0x2\n" - "add x17, x4, x4\n" - "ldr x16, [%x[params_struct], %[offsetof_args_params]]\n" - "mul x21, x21, x25\n" // offset *= kernel_stride * output_size - "add x7, x7, x21, LSL #2\n" // inptr[0] += offset * sizeof(float) - "add x15, x7, x24, LSL #2\n" - "mul x20, x20, x23\n" // offset *= output_tile_size - "add x14, x15, x24, LSL #2\n" - "add x8, x8, x20, LSL #2\n" // outptrs[0] += offset * sizeof(float) - "lsr x13, %x[n_channels], #0x2\n" - "add x12, x14, x24, LSL #2\n" - "add x11, x17, x4\n" - "add x10, x8, x22, LSL #2\n" - "add x9, x12, x24, LSL #2\n" - "add x28, x11, x4\n" - "add x27, x10, x22, LSL #2\n" - "add x23, x5, x5\n" - "add x20, %x[params_struct], %[offsetof_args_min]\n" - "ld1r { v15.4s }, [x20]\n" - "add x20, %x[params_struct], %[offsetof_args_max]\n" - "ld1r { v14.4s }, [x20]\n" - "add x26, x9, x24, LSL #2\n" - "add x25, x28, x4\n" - "add x24, x27, x22, LSL #2\n" - "add x22, x23, x5\n" - "mov x21, #0x0\n" - "sub x20, XZR, x6\n" - "cbz x13, 4f\n" - "ldr q13, [x16, #0x0]\n" - "ldr q0, [x16, #0x10]\n" - "cmp x6, x13, LSL #4\n" - "ldr q1, [x16, #0x20]\n" - "ldr q2, [x16, #0x30]\n" - "ldr q3, [x16, #0x40]\n" - "ldr q4, [x16, #0x50]\n" - "ldr q5, [x16, #0x60]\n" - "ldr q6, [x16, #0x70]\n" - "ldr q7, [x16, #0x80]\n" - "ldr q8, [x16, #0x90]\n" - "add x16, x16, #0xa0\n" - "ldr q9, [x14, x17]\n" - "ld1 { v10.4s }, [x7]\n" - "ldr q11, [x7, x25]\n" - "ldr q12, [x14, x11]\n" + "mov x24, #0x4\n" + "ldr x5, [%x[params_struct], %[offsetof_args_params]]\n" + "add x23, %x[params_struct], %[offsetof_args_min]\n" + "ldr x22, [%x[params_struct], %[offsetof_args_ld_input_row]]\n" + "add x21, %x[params_struct], %[offsetof_args_max]\n" + "ldr x6, [%x[params_struct], %[offsetof_args_ld_input_col]]\n" + "mov x7, #0x0\n" + "ldr x8, [%x[params_struct], %[offsetof_args_inptr]]\n" + "mul x19, x4, x22\n" // offset = tile_i * ld_input_row + "ldr x20, [%x[params_struct], %[offsetof_args_ld_output_row]]\n" + "madd x19, x26, x6, x19\n" // offset += tile_j * ld_input_col + "ldr x17, [%x[params_struct], %[offsetof_args_ld_output_col]]\n" + "mul x19, x19, x25\n" // offset *= kernel_stride * output_size + "ldr x16, [%x[params_struct], %[offsetof_args_outptr]]\n" + "add x8, x8, x19, LSL #2\n" // inptr[0] += offset * sizeof(float) + "ld1r { v15.4s }, [x23]\n" + "add x15, x8, x22, LSL #2\n" + "ld1r { v14.4s }, [x21]\n" + "add x14, x15, x22, LSL #2\n" + "lsl x6, x6, #0x2\n" + "add x13, x14, x22, LSL #2\n" + "add x12, x13, x22, LSL #2\n" + "add x11, x12, x22, LSL #2\n" + "add x10, x6, x6\n" + "add x9, x10, x6\n" + "add x28, x9, x6\n" + "add x27, x28, x6\n" + "mul x19, x4, x20\n" // offset = tile_i * ld_output_row + "madd x19, x26, x17, x19\n" // offset += tile_j * ld_output_col + "mul x19, x19, x24\n" // offset *= output_tile_size + "add x16, x16, x19, LSL #2\n" // outptrs[0] += offset * sizeof(float) + "add x26, x16, x20, LSL #2\n" + "add x25, x26, x20, LSL #2\n" + "add x24, x25, x20, LSL #2\n" + "lsl x17, x17, #0x2\n" + "add x23, x17, x17\n" + "add x22, x23, x17\n" + "mov x21, #0x10\n" // cntb _, ALL, #1 + "sub x20, XZR, x21\n" + "lsr x19, %x[n_channels], #0x2\n" + "cbz x19, 4f\n" + "ldr q13, [x5, #0x0]\n" + "ldr q0, [x5, #0x10]\n" + "cmp x21, x19, LSL #4\n" + "ldr q1, [x5, #0x20]\n" + "ldr q2, [x5, #0x30]\n" + "ldr q3, [x5, #0x40]\n" + "ldr q4, [x5, #0x50]\n" + "ldr q5, [x5, #0x60]\n" + "ldr q6, [x5, #0x70]\n" + "ldr q7, [x5, #0x80]\n" + "ldr q8, [x5, #0x90]\n" + "add x5, x5, #0xa0\n" + "ldr q9, [x14, x10]\n" + "ld1 { v10.4s }, [x8]\n" + "ldr q11, [x8, x27]\n" + "ldr q12, [x14, x9]\n" "bge 3f\n" "2:" // Tile loop: Channel loop - "mov v21.16b, v13.16b\n fmla v21.4s, v4.4s, v9.4s\n" - "mov v16.16b, v13.16b\n fmla v16.4s, v8.4s, v9.4s\n" - "add x6, x6, #0x10\n" - "cmp x6, x13, LSL #4\n" - "mov v22.16b, v13.16b\n fmla v22.4s, v3.4s, v9.4s\n" - "mov v25.16b, v13.16b\n fmla v25.4s, v1.4s, v9.4s\n" + "mov v31.16b, v13.16b\n fmla v31.4s, v8.4s, v9.4s\n" "add x20, x20, #0x10\n" + "mov v30.16b, v13.16b\n fmla v30.4s, v7.4s, v9.4s\n" + "add x7, x7, #0x10\n" + "mov v29.16b, v13.16b\n fmla v29.4s, v6.4s, v9.4s\n" "add x21, x21, #0x10\n" - "mov v26.16b, v13.16b\n fmla v26.4s, v0.4s, v9.4s\n" - "fmla v21.4s, v5.4s, v12.4s\n" - "mov v17.16b, v13.16b\n fmla v17.4s, v7.4s, v9.4s\n" - "mov v18.16b, v13.16b\n fmla v18.4s, v6.4s, v9.4s\n" - "mov v20.16b, v13.16b\n fmla v20.4s, v5.4s, v9.4s\n" - "mov v24.16b, v13.16b\n fmla v24.4s, v2.4s, v9.4s\n" - "ldr q9, [x12, x17]\n" - "fmla v16.4s, v0.4s, v10.4s\n" - "ld1 { v10.4s }, [x26]\n" - "mov v19.16b, v13.16b\n fmla v19.4s, v2.4s, v11.4s\n" - "ldr q11, [x26, x25]\n" - "fmla v22.4s, v4.4s, v12.4s\n" - "fmla v25.4s, v2.4s, v12.4s\n" - "fmla v26.4s, v1.4s, v12.4s\n" - "mov v28.16b, v13.16b\n fmla v28.4s, v6.4s, v10.4s\n" - "ldr q10, [x12, x11]\n" - "fmla v21.4s, v7.4s, v9.4s\n" - "fmla v17.4s, v8.4s, v12.4s\n" - "fmla v18.4s, v7.4s, v12.4s\n" - "fmla v19.4s, v6.4s, v12.4s\n" - "mov v23.16b, v13.16b\n fmla v23.4s, v3.4s, v12.4s\n" - "mov v27.16b, v13.16b\n fmla v27.4s, v0.4s, v12.4s\n" - "ldr q12, [x7, x4]\n" - "mov v31.16b, v13.16b\n fmla v31.4s, v8.4s, v11.4s\n" - "ldr q11, [x7, x28]\n" - "fmla v22.4s, v6.4s, v9.4s\n" - "fmla v25.4s, v4.4s, v9.4s\n" - "fmla v26.4s, v3.4s, v9.4s\n" - "fmla v20.4s, v8.4s, v9.4s\n" - "fmla v24.4s, v5.4s, v9.4s\n" - "fmla v28.4s, v2.4s, v9.4s\n" - "fmla v21.4s, v8.4s, v10.4s\n" - "fmla v16.4s, v1.4s, v12.4s\n" - "fmla v17.4s, v0.4s, v12.4s\n" - "ldr q12, [x15, x25]\n" - "fmla v18.4s, v2.4s, v11.4s\n" - "fmla v19.4s, v1.4s, v11.4s\n" - "ld1 { v11.4s }, [x9]\n" - "fmla v22.4s, v7.4s, v10.4s\n" - "fmla v23.4s, v6.4s, v10.4s\n" - "fmla v25.4s, v5.4s, v10.4s\n" - "fmla v26.4s, v4.4s, v10.4s\n" - "fmla v27.4s, v3.4s, v10.4s\n" + "mov v27.16b, v13.16b\n fmla v27.4s, v5.4s, v9.4s\n" + "cmp x21, x19, LSL #4\n" + "mov v26.16b, v13.16b\n fmla v26.4s, v4.4s, v9.4s\n" + "mov v25.16b, v13.16b\n fmla v25.4s, v3.4s, v9.4s\n" + "mov v23.16b, v13.16b\n fmla v23.4s, v2.4s, v9.4s\n" + "mov v22.16b, v13.16b\n fmla v22.4s, v1.4s, v9.4s\n" + "mov v21.16b, v13.16b\n fmla v21.4s, v0.4s, v9.4s\n" + "ldr q9, [x13, x10]\n" "fmla v31.4s, v0.4s, v10.4s\n" - "fmla v24.4s, v6.4s, v11.4s\n" - "fmla v28.4s, v3.4s, v11.4s\n" - "ldr q11, [x9, x25]\n" - "fmla v19.4s, v5.4s, v12.4s\n" - "fmla v23.4s, v2.4s, v12.4s\n" - "ldr q12, [x15, x11]\n" - "fmla v27.4s, v8.4s, v11.4s\n" - "fmla v31.4s, v5.4s, v11.4s\n" - "mov v29.16b, v13.16b\n fmla v29.4s, v1.4s, v9.4s\n" - "mov v30.16b, v13.16b\n fmla v30.4s, v0.4s, v9.4s\n" + "ld1 { v10.4s }, [x11]\n" + "mov v28.16b, v13.16b\n fmla v28.4s, v2.4s, v11.4s\n" + "ldr q11, [x11, x27]\n" + "fmla v30.4s, v8.4s, v12.4s\n" + "fmla v29.4s, v7.4s, v12.4s\n" + "fmla v26.4s, v5.4s, v12.4s\n" + "fmla v28.4s, v6.4s, v12.4s\n" + "fmla v25.4s, v4.4s, v12.4s\n" + "mov v24.16b, v13.16b\n fmla v24.4s, v3.4s, v12.4s\n" + "fmla v22.4s, v2.4s, v12.4s\n" + "fmla v21.4s, v1.4s, v12.4s\n" + "mov v20.16b, v13.16b\n fmla v20.4s, v0.4s, v12.4s\n" + "ldr q12, [x8, x6]\n" + "mov v19.16b, v13.16b\n fmla v19.4s, v6.4s, v10.4s\n" + "ldr q10, [x13, x9]\n" + "mov v16.16b, v13.16b\n fmla v16.4s, v8.4s, v11.4s\n" + "ldr q11, [x8, x28]\n" + "fmla v27.4s, v8.4s, v9.4s\n" + "fmla v26.4s, v7.4s, v9.4s\n" + "fmla v25.4s, v6.4s, v9.4s\n" + "fmla v23.4s, v5.4s, v9.4s\n" + "fmla v22.4s, v4.4s, v9.4s\n" + "fmla v21.4s, v3.4s, v9.4s\n" + "fmla v19.4s, v2.4s, v9.4s\n" + "mov v18.16b, v13.16b\n fmla v18.4s, v1.4s, v9.4s\n" + "mov v17.16b, v13.16b\n fmla v17.4s, v0.4s, v9.4s\n" "ld1 { v9.4s }, [x15]\n" - "fmla v29.4s, v2.4s, v10.4s\n" - "fmla v30.4s, v1.4s, v10.4s\n" - "ldr q10, [x15, x17]\n" - "fmla v20.4s, v0.4s, v9.4s\n" - "fmla v21.4s, v1.4s, v10.4s\n" - "fmla v16.4s, v3.4s, v9.4s\n" - "ldr q11, [x26, x4]\n" - "fmla v17.4s, v4.4s, v10.4s\n" - "fmla v18.4s, v3.4s, v10.4s\n" - "fmla v22.4s, v0.4s, v10.4s\n" - "fmla v20.4s, v2.4s, v10.4s\n" - "fmla v21.4s, v2.4s, v12.4s\n" - "fmla v16.4s, v5.4s, v10.4s\n" - "ldr q10, [x14, x4]\n" - "fmla v17.4s, v5.4s, v12.4s\n" - "fmla v18.4s, v4.4s, v12.4s\n" - "fmla v19.4s, v3.4s, v12.4s\n" - "fmla v22.4s, v1.4s, v12.4s\n" - "fmla v23.4s, v0.4s, v12.4s\n" - "ldr q12, [x14, x28]\n" - "fmla v28.4s, v7.4s, v11.4s\n" - "fmla v29.4s, v6.4s, v11.4s\n" - "ldr q11, [x26, x28]\n" - "fmla v20.4s, v4.4s, v10.4s\n" - "fmla v21.4s, v3.4s, v10.4s\n" - "fmla v24.4s, v1.4s, v10.4s\n" - "fmla v25.4s, v0.4s, v10.4s\n" - "fmla v16.4s, v7.4s, v10.4s\n" - "fmla v17.4s, v6.4s, v10.4s\n" - "ldr q10, [x7, x17]\n" - "fmla v30.4s, v8.4s, v11.4s\n" - "fmla v31.4s, v7.4s, v11.4s\n" - "ldr q11, [x12, x4]\n" - "fmla v18.4s, v8.4s, v12.4s\n" - "fmla v19.4s, v7.4s, v12.4s\n" - "fmla v22.4s, v5.4s, v12.4s\n" - "fmla v23.4s, v4.4s, v12.4s\n" - "fmla v26.4s, v2.4s, v12.4s\n" - "fmla v27.4s, v1.4s, v12.4s\n" - "ldr q12, [x7, x11]\n" - "add x7, x7, #0x10\n" - "fmla v20.4s, v7.4s, v11.4s\n" - "fmla v21.4s, v6.4s, v11.4s\n" - "fmla v24.4s, v4.4s, v11.4s\n" - "fmla v25.4s, v3.4s, v11.4s\n" + "fmla v31.4s, v1.4s, v12.4s\n" + "ldr q13, [x5, #0x0]\n" + "fmla v30.4s, v0.4s, v12.4s\n" + "ldr q12, [x15, x27]\n" + "fmla v29.4s, v2.4s, v11.4s\n" "fmla v28.4s, v1.4s, v11.4s\n" - "fmla v29.4s, v0.4s, v11.4s\n" - "ldr q11, [x12, x28]\n" - "fmla v16.4s, v2.4s, v10.4s\n" - "fmla v17.4s, v1.4s, v10.4s\n" - "fmla v18.4s, v0.4s, v10.4s\n" - "ld1 { v10.4s }, [x14]\n" - "fmla v30.4s, v2.4s, v11.4s\n" - "fmla v19.4s, v0.4s, v12.4s\n" + "ld1 { v11.4s }, [x12]\n" + "fmla v26.4s, v8.4s, v10.4s\n" + "fmla v25.4s, v7.4s, v10.4s\n" + "fmla v24.4s, v6.4s, v10.4s\n" + "fmla v22.4s, v5.4s, v10.4s\n" + "fmla v21.4s, v4.4s, v10.4s\n" "fmla v20.4s, v3.4s, v10.4s\n" - "fmla v24.4s, v0.4s, v10.4s\n" - "fmla v22.4s, v8.4s, v11.4s\n" - "fmla v23.4s, v7.4s, v11.4s\n" - "fmla v26.4s, v5.4s, v11.4s\n" - "fmla v27.4s, v4.4s, v11.4s\n" - "fmla v31.4s, v1.4s, v11.4s\n" - "ldr q11, [x9, x17]\n" - "fmla v17.4s, v2.4s, v12.4s\n" - "fmla v18.4s, v1.4s, v12.4s\n" - "ldr q12, [x14, x25]\n" - "add x14, x14, #0x10\n" - "fmla v16.4s, v6.4s, v10.4s\n" - "ld1 { v10.4s }, [x12]\n" - "fmla v29.4s, v4.4s, v11.4s\n" - "fmla v30.4s, v3.4s, v11.4s\n" - "fmla v19.4s, v8.4s, v12.4s\n" - "fmla v23.4s, v5.4s, v12.4s\n" - "fmla v27.4s, v2.4s, v12.4s\n" - "ldr q12, [x12, x25]\n" - "add x12, x12, #0x10\n" - "fmla v20.4s, v6.4s, v10.4s\n" - "fmla v24.4s, v3.4s, v10.4s\n" - "fmla v28.4s, v0.4s, v10.4s\n" - "ldr q10, [x26, x17]\n" - "fmla v31.4s, v2.4s, v12.4s\n" - "fmla v29.4s, v7.4s, v10.4s\n" + "fmla v18.4s, v2.4s, v10.4s\n" + "fmla v17.4s, v1.4s, v10.4s\n" + "fmla v16.4s, v0.4s, v10.4s\n" + "ldr q10, [x15, x10]\n" + "fmla v31.4s, v3.4s, v9.4s\n" + "fmla v27.4s, v0.4s, v9.4s\n" + "fmla v28.4s, v5.4s, v12.4s\n" + "fmla v24.4s, v2.4s, v12.4s\n" + "ldr q12, [x15, x9]\n" + "fmla v23.4s, v6.4s, v11.4s\n" + "fmla v19.4s, v3.4s, v11.4s\n" + "ldr q11, [x12, x27]\n" + "fmla v31.4s, v5.4s, v10.4s\n" + "fmla v30.4s, v4.4s, v10.4s\n" + "fmla v29.4s, v3.4s, v10.4s\n" + "fmla v27.4s, v2.4s, v10.4s\n" + "fmla v26.4s, v1.4s, v10.4s\n" + "fmla v25.4s, v0.4s, v10.4s\n" + "ldr q10, [x14, x6]\n" + "fmla v20.4s, v8.4s, v11.4s\n" + "fmla v16.4s, v5.4s, v11.4s\n" + "ldr q11, [x11, x6]\n" + "fmla v30.4s, v5.4s, v12.4s\n" + "fmla v29.4s, v4.4s, v12.4s\n" + "fmla v28.4s, v3.4s, v12.4s\n" + "fmla v26.4s, v2.4s, v12.4s\n" + "fmla v25.4s, v1.4s, v12.4s\n" + "fmla v24.4s, v0.4s, v12.4s\n" + "ldr q12, [x14, x28]\n" + "fmla v19.4s, v7.4s, v11.4s\n" + "fmla v18.4s, v6.4s, v11.4s\n" + "ldr q11, [x11, x28]\n" + "fmla v31.4s, v7.4s, v10.4s\n" "fmla v30.4s, v6.4s, v10.4s\n" - "fmla v24.4s, v8.4s, v11.4s\n" - "fmla v25.4s, v7.4s, v11.4s\n" + "fmla v27.4s, v4.4s, v10.4s\n" + "fmla v26.4s, v3.4s, v10.4s\n" + "fmla v23.4s, v1.4s, v10.4s\n" + "fmla v22.4s, v0.4s, v10.4s\n" + "ldr q10, [x8, x10]\n" + "fmla v17.4s, v8.4s, v11.4s\n" + "fmla v16.4s, v7.4s, v11.4s\n" + "ldr q11, [x13, x6]\n" + "fmla v29.4s, v8.4s, v12.4s\n" + "fmla v28.4s, v7.4s, v12.4s\n" + "fmla v25.4s, v5.4s, v12.4s\n" + "fmla v24.4s, v4.4s, v12.4s\n" + "fmla v21.4s, v2.4s, v12.4s\n" + "fmla v20.4s, v1.4s, v12.4s\n" + "ldr q12, [x8, x9]\n" + "add x8, x8, #0x10\n" + "fmla v31.4s, v2.4s, v10.4s\n" + "fmla v30.4s, v1.4s, v10.4s\n" + "fmla v29.4s, v0.4s, v10.4s\n" + "ld1 { v10.4s }, [x14]\n" + "fmla v27.4s, v7.4s, v11.4s\n" "fmla v26.4s, v6.4s, v11.4s\n" - "fmla v28.4s, v5.4s, v11.4s\n" - "ldr q11, [x9, x11]\n" - "fmla v27.4s, v5.4s, v12.4s\n" - "fmla v29.4s, v5.4s, v11.4s\n" - "fmla v30.4s, v4.4s, v11.4s\n" - "fmla v31.4s, v3.4s, v11.4s\n" - "fmla v23.4s, v8.4s, v12.4s\n" - "ldr q12, [x26, x11]\n" - "fmla v28.4s, v8.4s, v10.4s\n" - "ldr q10, [x15, x4]\n" + "fmla v23.4s, v4.4s, v11.4s\n" + "fmla v22.4s, v3.4s, v11.4s\n" + "fmla v19.4s, v1.4s, v11.4s\n" + "fmla v18.4s, v0.4s, v11.4s\n" + "ldr q11, [x13, x28]\n" + "fmla v30.4s, v2.4s, v12.4s\n" + "fmla v29.4s, v1.4s, v12.4s\n" + "fmla v28.4s, v0.4s, v12.4s\n" + "ldr q12, [x14, x27]\n" + "add x14, x14, #0x10\n" + "fmla v31.4s, v6.4s, v10.4s\n" + "ldr q9, [x14, x10]\n" + "fmla v27.4s, v3.4s, v10.4s\n" + "fmla v23.4s, v0.4s, v10.4s\n" + "ld1 { v10.4s }, [x13]\n" "fmla v25.4s, v8.4s, v11.4s\n" - "fmla v26.4s, v7.4s, v11.4s\n" - "add x26, x26, #0x10\n" - "fmla v27.4s, v6.4s, v11.4s\n" + "fmla v24.4s, v7.4s, v11.4s\n" + "fmla v21.4s, v5.4s, v11.4s\n" + "fmla v20.4s, v4.4s, v11.4s\n" + "fmla v17.4s, v2.4s, v11.4s\n" + "fmla v16.4s, v1.4s, v11.4s\n" + "ldr q11, [x12, x10]\n" + "fmla v28.4s, v8.4s, v12.4s\n" + "fmla v24.4s, v5.4s, v12.4s\n" + "fmla v20.4s, v2.4s, v12.4s\n" + "ldr q12, [x13, x27]\n" + "add x13, x13, #0x10\n" + "fmla v27.4s, v6.4s, v10.4s\n" + "fmla v23.4s, v3.4s, v10.4s\n" + "fmla v19.4s, v0.4s, v10.4s\n" + "ldr q10, [x11, x10]\n" + "fmla v22.4s, v7.4s, v11.4s\n" + "fmla v21.4s, v6.4s, v11.4s\n" + "fmla v23.4s, v8.4s, v11.4s\n" + "fmla v19.4s, v5.4s, v11.4s\n" + "fmla v18.4s, v4.4s, v11.4s\n" + "fmla v17.4s, v3.4s, v11.4s\n" + "ldr q11, [x12, x9]\n" + "fmla v24.4s, v8.4s, v12.4s\n" + "fmla v20.4s, v5.4s, v12.4s\n" + "fmla v16.4s, v2.4s, v12.4s\n" + "ldr q12, [x11, x9]\n" + "add x11, x11, #0x10\n" + "fmla v19.4s, v8.4s, v10.4s\n" + "fmla v18.4s, v7.4s, v10.4s\n" + "fmla v17.4s, v6.4s, v10.4s\n" + "ldr q10, [x15, x6]\n" + "fmla v22.4s, v8.4s, v11.4s\n" + "fmla v21.4s, v7.4s, v11.4s\n" + "fmla v20.4s, v6.4s, v11.4s\n" + "fmla v18.4s, v5.4s, v11.4s\n" + "fmla v17.4s, v4.4s, v11.4s\n" + "fmla v16.4s, v3.4s, v11.4s\n" "ldr q11, [x15, x28]\n" - "fmla v29.4s, v8.4s, v12.4s\n" "add x15, x15, #0x10\n" - "fmla v30.4s, v7.4s, v12.4s\n" - "fmla v31.4s, v6.4s, v12.4s\n" - "ldr q12, [x9, x4]\n" - "fmla v16.4s, v4.4s, v10.4s\n" - "fmla v17.4s, v3.4s, v10.4s\n" - "fmax v16.4s, v16.4s, v15.4s\n" - "fmla v20.4s, v1.4s, v10.4s\n" - "fmla v21.4s, v0.4s, v10.4s\n" - "ldr q10, [x9, x28]\n" - "ldr q9, [x14, x17]\n" - "fmla v18.4s, v5.4s, v11.4s\n" - "fmla v19.4s, v4.4s, v11.4s\n" - "fmax v17.4s, v17.4s, v15.4s\n" - "add x9, x9, #0x10\n" - "fmla v22.4s, v2.4s, v11.4s\n" - "ldr q13, [x16, #0x0]\n" - "fmla v23.4s, v1.4s, v11.4s\n" - "ldr q11, [x7, x25]\n" - "ldr q0, [x16, #0x10]\n" - "fmla v24.4s, v7.4s, v12.4s\n" - "fmla v25.4s, v6.4s, v12.4s\n" - "ldr q1, [x16, #0x20]\n" - "fmla v28.4s, v4.4s, v12.4s\n" - "fmla v29.4s, v3.4s, v12.4s\n" - "ldr q12, [x14, x11]\n" - "ldr q2, [x16, #0x30]\n" - "fmla v26.4s, v8.4s, v10.4s\n" - "ldr q3, [x16, #0x40]\n" - "fmla v27.4s, v7.4s, v10.4s\n" - "ldr q6, [x16, #0x70]\n" - "fmla v30.4s, v5.4s, v10.4s\n" - "ldr q5, [x16, #0x60]\n" + "fmla v18.4s, v8.4s, v12.4s\n" "fmla v31.4s, v4.4s, v10.4s\n" - "ld1 { v10.4s }, [x7]\n" - "ldr q4, [x16, #0x50]\n" - "fmax v18.4s, v18.4s, v15.4s\n" - "fmax v19.4s, v19.4s, v15.4s\n" - "fmax v20.4s, v20.4s, v15.4s\n" - "fmax v21.4s, v21.4s, v15.4s\n" - "fmax v22.4s, v22.4s, v15.4s\n" - "fmax v23.4s, v23.4s, v15.4s\n" - "fmax v24.4s, v24.4s, v15.4s\n" - "fmax v25.4s, v25.4s, v15.4s\n" - "fmax v26.4s, v26.4s, v15.4s\n" - "fmax v27.4s, v27.4s, v15.4s\n" - "fmax v28.4s, v28.4s, v15.4s\n" - "fmax v29.4s, v29.4s, v15.4s\n" - "fmax v30.4s, v30.4s, v15.4s\n" + "fmla v17.4s, v7.4s, v12.4s\n" + "fmla v16.4s, v6.4s, v12.4s\n" + "ldr q12, [x12, x6]\n" + "fmla v30.4s, v3.4s, v10.4s\n" + "fmla v27.4s, v1.4s, v10.4s\n" + "fmla v26.4s, v0.4s, v10.4s\n" + "ldr q10, [x12, x28]\n" + "add x12, x12, #0x10\n" + "fmla v29.4s, v5.4s, v11.4s\n" + "ldr q0, [x5, #0x10]\n" + "fmla v28.4s, v4.4s, v11.4s\n" + "fmla v25.4s, v2.4s, v11.4s\n" + "ldr q2, [x5, #0x30]\n" + "fmla v24.4s, v1.4s, v11.4s\n" + "ldr q11, [x8, x27]\n" + "fmla v23.4s, v7.4s, v12.4s\n" + "ldr q1, [x5, #0x20]\n" + "fmla v22.4s, v6.4s, v12.4s\n" + "ldr q6, [x5, #0x70]\n" + "fmla v19.4s, v4.4s, v12.4s\n" + "fmla v18.4s, v3.4s, v12.4s\n" + "ldr q12, [x14, x9]\n" + "fmla v21.4s, v8.4s, v10.4s\n" + "ldr q3, [x5, #0x40]\n" + "fmla v20.4s, v7.4s, v10.4s\n" + "ldr q7, [x5, #0x80]\n" + "fmla v17.4s, v5.4s, v10.4s\n" + "ldr q5, [x5, #0x60]\n" + "fmla v16.4s, v4.4s, v10.4s\n" + "ld1 { v10.4s }, [x8]\n" "fmax v31.4s, v31.4s, v15.4s\n" - "fmin v16.4s, v16.4s, v14.4s\n" - "fmin v17.4s, v17.4s, v14.4s\n" - "st1 { v16.4s }, [x8]\n" - "ldr q7, [x16, #0x80]\n" - "fmin v18.4s, v18.4s, v14.4s\n" - "fmin v19.4s, v19.4s, v14.4s\n" - "str q17, [x8, x5]\n" - "ldr q8, [x16, #0x90]\n" - "fmin v20.4s, v20.4s, v14.4s\n" - "fmin v21.4s, v21.4s, v14.4s\n" - "str q18, [x8, x23]\n" - "add x16, x16, #0xa0\n" - "fmin v22.4s, v22.4s, v14.4s\n" - "fmin v23.4s, v23.4s, v14.4s\n" - "str q19, [x8, x22]\n" - "add x8, x8, #0x10\n" - "fmin v24.4s, v24.4s, v14.4s\n" - "fmin v25.4s, v25.4s, v14.4s\n" - "st1 { v20.4s }, [x10]\n" - "fmin v26.4s, v26.4s, v14.4s\n" - "fmin v27.4s, v27.4s, v14.4s\n" - "str q21, [x10, x5]\n" - "fmin v28.4s, v28.4s, v14.4s\n" - "fmin v29.4s, v29.4s, v14.4s\n" - "str q22, [x10, x23]\n" - "fmin v30.4s, v30.4s, v14.4s\n" + "ldr q4, [x5, #0x50]\n" + "fmax v30.4s, v30.4s, v15.4s\n" + "ldr q8, [x5, #0x90]\n" + "add x5, x5, #0xa0\n" "fmin v31.4s, v31.4s, v14.4s\n" - "str q23, [x10, x22]\n" - "add x10, x10, #0x10\n" - "st1 { v24.4s }, [x27]\n" - "str q25, [x27, x5]\n" - "str q26, [x27, x23]\n" - "str q27, [x27, x22]\n" - "add x27, x27, #0x10\n" - "st1 { v28.4s }, [x24]\n" - "str q29, [x24, x5]\n" - "str q30, [x24, x23]\n" - "str q31, [x24, x22]\n" + "st1 { v31.4s }, [x16]\n" + "fmin v30.4s, v30.4s, v14.4s\n" + "fmax v29.4s, v29.4s, v15.4s\n" + "str q30, [x16, x17]\n" + "fmin v29.4s, v29.4s, v14.4s\n" + "fmax v28.4s, v28.4s, v15.4s\n" + "str q29, [x16, x23]\n" + "fmin v28.4s, v28.4s, v14.4s\n" + "fmax v27.4s, v27.4s, v15.4s\n" + "str q28, [x16, x22]\n" + "fmin v27.4s, v27.4s, v14.4s\n" + "add x16, x16, #0x10\n" + "fmax v26.4s, v26.4s, v15.4s\n" + "st1 { v27.4s }, [x26]\n" + "fmax v25.4s, v25.4s, v15.4s\n" + "fmax v24.4s, v24.4s, v15.4s\n" + "fmin v26.4s, v26.4s, v14.4s\n" + "str q26, [x26, x17]\n" + "fmin v25.4s, v25.4s, v14.4s\n" + "fmin v24.4s, v24.4s, v14.4s\n" + "str q25, [x26, x23]\n" + "fmax v23.4s, v23.4s, v15.4s\n" + "fmax v22.4s, v22.4s, v15.4s\n" + "str q24, [x26, x22]\n" + "add x26, x26, #0x10\n" + "fmax v21.4s, v21.4s, v15.4s\n" + "fmax v20.4s, v20.4s, v15.4s\n" + "fmin v23.4s, v23.4s, v14.4s\n" + "st1 { v23.4s }, [x25]\n" + "fmin v22.4s, v22.4s, v14.4s\n" + "fmin v21.4s, v21.4s, v14.4s\n" + "str q22, [x25, x17]\n" + "fmin v20.4s, v20.4s, v14.4s\n" + "fmax v19.4s, v19.4s, v15.4s\n" + "str q21, [x25, x23]\n" + "fmax v18.4s, v18.4s, v15.4s\n" + "str q20, [x25, x22]\n" + "fmin v19.4s, v19.4s, v14.4s\n" + "add x25, x25, #0x10\n" + "fmin v18.4s, v18.4s, v14.4s\n" + "st1 { v19.4s }, [x24]\n" + "fmax v17.4s, v17.4s, v15.4s\n" + "fmax v16.4s, v16.4s, v15.4s\n" + "str q18, [x24, x17]\n" + "fmin v17.4s, v17.4s, v14.4s\n" + "str q17, [x24, x23]\n" + "fmin v16.4s, v16.4s, v14.4s\n" + "str q16, [x24, x22]\n" "add x24, x24, #0x10\n" "blt 2b\n" "3:" // Tile loop: Channel tail - "mov v21.16b, v13.16b\n fmla v21.4s, v4.4s, v9.4s\n" - "mov v16.16b, v13.16b\n fmla v16.4s, v8.4s, v9.4s\n" - "mov v22.16b, v13.16b\n fmla v22.4s, v3.4s, v9.4s\n" - "mov v25.16b, v13.16b\n fmla v25.4s, v1.4s, v9.4s\n" - "mov v26.16b, v13.16b\n fmla v26.4s, v0.4s, v9.4s\n" - "fmla v21.4s, v5.4s, v12.4s\n" - "mov v17.16b, v13.16b\n fmla v17.4s, v7.4s, v9.4s\n" - "mov v18.16b, v13.16b\n fmla v18.4s, v6.4s, v9.4s\n" - "mov v20.16b, v13.16b\n fmla v20.4s, v5.4s, v9.4s\n" - "mov v24.16b, v13.16b\n fmla v24.4s, v2.4s, v9.4s\n" - "ldr q9, [x12, x17]\n" - "fmla v16.4s, v0.4s, v10.4s\n" - "ld1 { v10.4s }, [x26]\n" - "mov v19.16b, v13.16b\n fmla v19.4s, v2.4s, v11.4s\n" - "ldr q11, [x26, x25]\n" - "fmla v22.4s, v4.4s, v12.4s\n" - "fmla v25.4s, v2.4s, v12.4s\n" - "fmla v26.4s, v1.4s, v12.4s\n" - "mov v28.16b, v13.16b\n fmla v28.4s, v6.4s, v10.4s\n" - "ldr q10, [x12, x11]\n" - "fmla v21.4s, v7.4s, v9.4s\n" - "fmla v17.4s, v8.4s, v12.4s\n" - "fmla v18.4s, v7.4s, v12.4s\n" - "fmla v19.4s, v6.4s, v12.4s\n" - "mov v23.16b, v13.16b\n fmla v23.4s, v3.4s, v12.4s\n" - "mov v27.16b, v13.16b\n fmla v27.4s, v0.4s, v12.4s\n" - "ldr q12, [x7, x4]\n" - "mov v31.16b, v13.16b\n fmla v31.4s, v8.4s, v11.4s\n" - "ldr q11, [x7, x28]\n" - "fmla v22.4s, v6.4s, v9.4s\n" - "fmla v25.4s, v4.4s, v9.4s\n" - "fmla v26.4s, v3.4s, v9.4s\n" - "fmla v20.4s, v8.4s, v9.4s\n" - "fmla v24.4s, v5.4s, v9.4s\n" - "fmla v28.4s, v2.4s, v9.4s\n" - "fmla v21.4s, v8.4s, v10.4s\n" - "fmla v16.4s, v1.4s, v12.4s\n" - "fmla v17.4s, v0.4s, v12.4s\n" - "ldr q12, [x15, x25]\n" - "fmla v18.4s, v2.4s, v11.4s\n" - "fmla v19.4s, v1.4s, v11.4s\n" - "ld1 { v11.4s }, [x9]\n" - "fmla v22.4s, v7.4s, v10.4s\n" - "fmla v23.4s, v6.4s, v10.4s\n" - "fmla v25.4s, v5.4s, v10.4s\n" - "fmla v26.4s, v4.4s, v10.4s\n" - "fmla v27.4s, v3.4s, v10.4s\n" + "mov v31.16b, v13.16b\n fmla v31.4s, v8.4s, v9.4s\n" + "mov v30.16b, v13.16b\n fmla v30.4s, v7.4s, v9.4s\n" + "mov v29.16b, v13.16b\n fmla v29.4s, v6.4s, v9.4s\n" + "mov v27.16b, v13.16b\n fmla v27.4s, v5.4s, v9.4s\n" + "mov v26.16b, v13.16b\n fmla v26.4s, v4.4s, v9.4s\n" + "mov v25.16b, v13.16b\n fmla v25.4s, v3.4s, v9.4s\n" + "mov v23.16b, v13.16b\n fmla v23.4s, v2.4s, v9.4s\n" + "mov v22.16b, v13.16b\n fmla v22.4s, v1.4s, v9.4s\n" + "mov v21.16b, v13.16b\n fmla v21.4s, v0.4s, v9.4s\n" + "ldr q9, [x13, x10]\n" "fmla v31.4s, v0.4s, v10.4s\n" - "fmla v24.4s, v6.4s, v11.4s\n" - "fmla v28.4s, v3.4s, v11.4s\n" - "ldr q11, [x9, x25]\n" - "fmla v19.4s, v5.4s, v12.4s\n" - "fmla v23.4s, v2.4s, v12.4s\n" - "ldr q12, [x15, x11]\n" - "fmla v27.4s, v8.4s, v11.4s\n" - "fmla v31.4s, v5.4s, v11.4s\n" - "mov v29.16b, v13.16b\n fmla v29.4s, v1.4s, v9.4s\n" - "mov v30.16b, v13.16b\n fmla v30.4s, v0.4s, v9.4s\n" + "ld1 { v10.4s }, [x11]\n" + "mov v28.16b, v13.16b\n fmla v28.4s, v2.4s, v11.4s\n" + "ldr q11, [x11, x27]\n" + "fmla v30.4s, v8.4s, v12.4s\n" + "fmla v29.4s, v7.4s, v12.4s\n" + "fmla v26.4s, v5.4s, v12.4s\n" + "fmla v28.4s, v6.4s, v12.4s\n" + "fmla v25.4s, v4.4s, v12.4s\n" + "mov v24.16b, v13.16b\n fmla v24.4s, v3.4s, v12.4s\n" + "fmla v22.4s, v2.4s, v12.4s\n" + "fmla v21.4s, v1.4s, v12.4s\n" + "mov v20.16b, v13.16b\n fmla v20.4s, v0.4s, v12.4s\n" + "ldr q12, [x8, x6]\n" + "mov v19.16b, v13.16b\n fmla v19.4s, v6.4s, v10.4s\n" + "ldr q10, [x13, x9]\n" + "mov v16.16b, v13.16b\n fmla v16.4s, v8.4s, v11.4s\n" + "ldr q11, [x8, x28]\n" + "fmla v27.4s, v8.4s, v9.4s\n" + "fmla v26.4s, v7.4s, v9.4s\n" + "fmla v25.4s, v6.4s, v9.4s\n" + "fmla v23.4s, v5.4s, v9.4s\n" + "fmla v22.4s, v4.4s, v9.4s\n" + "fmla v21.4s, v3.4s, v9.4s\n" + "fmla v19.4s, v2.4s, v9.4s\n" + "mov v18.16b, v13.16b\n fmla v18.4s, v1.4s, v9.4s\n" + "mov v17.16b, v13.16b\n fmla v17.4s, v0.4s, v9.4s\n" "ld1 { v9.4s }, [x15]\n" - "fmla v29.4s, v2.4s, v10.4s\n" - "fmla v30.4s, v1.4s, v10.4s\n" - "ldr q10, [x15, x17]\n" - "fmla v20.4s, v0.4s, v9.4s\n" - "fmla v21.4s, v1.4s, v10.4s\n" - "fmla v16.4s, v3.4s, v9.4s\n" - "ldr q11, [x26, x4]\n" - "fmla v17.4s, v4.4s, v10.4s\n" - "fmla v18.4s, v3.4s, v10.4s\n" - "fmla v22.4s, v0.4s, v10.4s\n" - "fmla v20.4s, v2.4s, v10.4s\n" - "fmla v21.4s, v2.4s, v12.4s\n" - "fmla v16.4s, v5.4s, v10.4s\n" - "ldr q10, [x14, x4]\n" - "fmla v17.4s, v5.4s, v12.4s\n" - "fmla v18.4s, v4.4s, v12.4s\n" - "fmla v19.4s, v3.4s, v12.4s\n" - "fmla v22.4s, v1.4s, v12.4s\n" - "fmla v23.4s, v0.4s, v12.4s\n" - "ldr q12, [x14, x28]\n" - "fmla v28.4s, v7.4s, v11.4s\n" - "fmla v29.4s, v6.4s, v11.4s\n" - "ldr q11, [x26, x28]\n" - "fmla v20.4s, v4.4s, v10.4s\n" - "fmla v21.4s, v3.4s, v10.4s\n" - "fmla v24.4s, v1.4s, v10.4s\n" - "fmla v25.4s, v0.4s, v10.4s\n" - "fmla v16.4s, v7.4s, v10.4s\n" - "fmla v17.4s, v6.4s, v10.4s\n" - "ldr q10, [x7, x17]\n" - "fmla v30.4s, v8.4s, v11.4s\n" - "fmla v31.4s, v7.4s, v11.4s\n" - "ldr q11, [x12, x4]\n" - "fmla v18.4s, v8.4s, v12.4s\n" - "fmla v19.4s, v7.4s, v12.4s\n" - "fmla v22.4s, v5.4s, v12.4s\n" - "fmla v23.4s, v4.4s, v12.4s\n" - "fmla v26.4s, v2.4s, v12.4s\n" - "fmla v27.4s, v1.4s, v12.4s\n" - "ldr q12, [x7, x11]\n" - "add x7, x7, #0x10\n" - "fmla v20.4s, v7.4s, v11.4s\n" - "fmla v21.4s, v6.4s, v11.4s\n" - "fmla v24.4s, v4.4s, v11.4s\n" - "fmla v25.4s, v3.4s, v11.4s\n" + "fmla v31.4s, v1.4s, v12.4s\n" + "fmla v30.4s, v0.4s, v12.4s\n" + "ldr q12, [x15, x27]\n" + "fmla v29.4s, v2.4s, v11.4s\n" "fmla v28.4s, v1.4s, v11.4s\n" - "fmla v29.4s, v0.4s, v11.4s\n" - "ldr q11, [x12, x28]\n" - "fmla v16.4s, v2.4s, v10.4s\n" - "fmla v17.4s, v1.4s, v10.4s\n" - "fmla v18.4s, v0.4s, v10.4s\n" - "ld1 { v10.4s }, [x14]\n" - "fmla v30.4s, v2.4s, v11.4s\n" - "fmla v19.4s, v0.4s, v12.4s\n" + "ld1 { v11.4s }, [x12]\n" + "fmla v26.4s, v8.4s, v10.4s\n" + "fmla v25.4s, v7.4s, v10.4s\n" + "fmla v24.4s, v6.4s, v10.4s\n" + "fmla v22.4s, v5.4s, v10.4s\n" + "fmla v21.4s, v4.4s, v10.4s\n" "fmla v20.4s, v3.4s, v10.4s\n" - "fmla v24.4s, v0.4s, v10.4s\n" - "fmla v22.4s, v8.4s, v11.4s\n" - "fmla v23.4s, v7.4s, v11.4s\n" - "fmla v26.4s, v5.4s, v11.4s\n" - "fmla v27.4s, v4.4s, v11.4s\n" - "fmla v31.4s, v1.4s, v11.4s\n" - "ldr q11, [x9, x17]\n" - "fmla v17.4s, v2.4s, v12.4s\n" - "fmla v18.4s, v1.4s, v12.4s\n" - "ldr q12, [x14, x25]\n" - "add x14, x14, #0x10\n" - "fmla v16.4s, v6.4s, v10.4s\n" - "ld1 { v10.4s }, [x12]\n" - "fmla v29.4s, v4.4s, v11.4s\n" - "fmla v30.4s, v3.4s, v11.4s\n" - "fmla v19.4s, v8.4s, v12.4s\n" - "fmla v23.4s, v5.4s, v12.4s\n" - "fmla v27.4s, v2.4s, v12.4s\n" - "ldr q12, [x12, x25]\n" - "add x12, x12, #0x10\n" - "fmla v20.4s, v6.4s, v10.4s\n" - "fmla v24.4s, v3.4s, v10.4s\n" - "fmla v28.4s, v0.4s, v10.4s\n" - "ldr q10, [x26, x17]\n" - "fmla v31.4s, v2.4s, v12.4s\n" - "fmla v29.4s, v7.4s, v10.4s\n" + "fmla v18.4s, v2.4s, v10.4s\n" + "fmla v17.4s, v1.4s, v10.4s\n" + "fmla v16.4s, v0.4s, v10.4s\n" + "ldr q10, [x15, x10]\n" + "fmla v31.4s, v3.4s, v9.4s\n" + "fmla v27.4s, v0.4s, v9.4s\n" + "fmla v28.4s, v5.4s, v12.4s\n" + "fmla v24.4s, v2.4s, v12.4s\n" + "ldr q12, [x15, x9]\n" + "fmla v23.4s, v6.4s, v11.4s\n" + "fmla v19.4s, v3.4s, v11.4s\n" + "ldr q11, [x12, x27]\n" + "fmla v31.4s, v5.4s, v10.4s\n" + "fmla v30.4s, v4.4s, v10.4s\n" + "fmla v29.4s, v3.4s, v10.4s\n" + "fmla v27.4s, v2.4s, v10.4s\n" + "fmla v26.4s, v1.4s, v10.4s\n" + "fmla v25.4s, v0.4s, v10.4s\n" + "ldr q10, [x14, x6]\n" + "fmla v20.4s, v8.4s, v11.4s\n" + "fmla v16.4s, v5.4s, v11.4s\n" + "ldr q11, [x11, x6]\n" + "fmla v30.4s, v5.4s, v12.4s\n" + "fmla v29.4s, v4.4s, v12.4s\n" + "fmla v28.4s, v3.4s, v12.4s\n" + "fmla v26.4s, v2.4s, v12.4s\n" + "fmla v25.4s, v1.4s, v12.4s\n" + "fmla v24.4s, v0.4s, v12.4s\n" + "ldr q12, [x14, x28]\n" + "fmla v19.4s, v7.4s, v11.4s\n" + "fmla v18.4s, v6.4s, v11.4s\n" + "ldr q11, [x11, x28]\n" + "fmla v31.4s, v7.4s, v10.4s\n" "fmla v30.4s, v6.4s, v10.4s\n" - "fmla v24.4s, v8.4s, v11.4s\n" - "fmla v25.4s, v7.4s, v11.4s\n" + "fmla v27.4s, v4.4s, v10.4s\n" + "fmla v26.4s, v3.4s, v10.4s\n" + "fmla v23.4s, v1.4s, v10.4s\n" + "fmla v22.4s, v0.4s, v10.4s\n" + "ldr q10, [x8, x10]\n" + "fmla v17.4s, v8.4s, v11.4s\n" + "fmla v16.4s, v7.4s, v11.4s\n" + "ldr q11, [x13, x6]\n" + "fmla v29.4s, v8.4s, v12.4s\n" + "fmla v28.4s, v7.4s, v12.4s\n" + "fmla v25.4s, v5.4s, v12.4s\n" + "fmla v24.4s, v4.4s, v12.4s\n" + "fmla v21.4s, v2.4s, v12.4s\n" + "fmla v20.4s, v1.4s, v12.4s\n" + "ldr q12, [x8, x9]\n" + "add x8, x8, #0x10\n" + "fmla v31.4s, v2.4s, v10.4s\n" + "fmla v30.4s, v1.4s, v10.4s\n" + "fmla v29.4s, v0.4s, v10.4s\n" + "ld1 { v10.4s }, [x14]\n" + "fmla v27.4s, v7.4s, v11.4s\n" "fmla v26.4s, v6.4s, v11.4s\n" - "fmla v28.4s, v5.4s, v11.4s\n" - "ldr q11, [x9, x11]\n" - "fmla v27.4s, v5.4s, v12.4s\n" - "fmla v29.4s, v5.4s, v11.4s\n" - "fmla v30.4s, v4.4s, v11.4s\n" - "fmla v31.4s, v3.4s, v11.4s\n" - "fmla v23.4s, v8.4s, v12.4s\n" - "ldr q12, [x26, x11]\n" - "fmla v28.4s, v8.4s, v10.4s\n" - "ldr q10, [x15, x4]\n" + "fmla v23.4s, v4.4s, v11.4s\n" + "fmla v22.4s, v3.4s, v11.4s\n" + "fmla v19.4s, v1.4s, v11.4s\n" + "fmla v18.4s, v0.4s, v11.4s\n" + "ldr q11, [x13, x28]\n" + "fmla v30.4s, v2.4s, v12.4s\n" + "fmla v29.4s, v1.4s, v12.4s\n" + "fmla v28.4s, v0.4s, v12.4s\n" + "ldr q12, [x14, x27]\n" + "add x14, x14, #0x10\n" + "fmla v31.4s, v6.4s, v10.4s\n" + "fmla v27.4s, v3.4s, v10.4s\n" + "fmla v23.4s, v0.4s, v10.4s\n" + "ld1 { v10.4s }, [x13]\n" "fmla v25.4s, v8.4s, v11.4s\n" - "fmla v26.4s, v7.4s, v11.4s\n" - "add x26, x26, #0x10\n" - "fmla v27.4s, v6.4s, v11.4s\n" + "fmla v24.4s, v7.4s, v11.4s\n" + "fmla v21.4s, v5.4s, v11.4s\n" + "fmla v20.4s, v4.4s, v11.4s\n" + "fmla v17.4s, v2.4s, v11.4s\n" + "fmla v16.4s, v1.4s, v11.4s\n" + "ldr q11, [x12, x10]\n" + "fmla v28.4s, v8.4s, v12.4s\n" + "fmla v24.4s, v5.4s, v12.4s\n" + "fmla v20.4s, v2.4s, v12.4s\n" + "ldr q12, [x13, x27]\n" + "add x13, x13, #0x10\n" + "fmla v27.4s, v6.4s, v10.4s\n" + "fmla v23.4s, v3.4s, v10.4s\n" + "fmla v19.4s, v0.4s, v10.4s\n" + "ldr q10, [x11, x10]\n" + "fmla v22.4s, v7.4s, v11.4s\n" + "fmla v21.4s, v6.4s, v11.4s\n" + "fmla v23.4s, v8.4s, v11.4s\n" + "fmla v19.4s, v5.4s, v11.4s\n" + "fmla v18.4s, v4.4s, v11.4s\n" + "fmla v17.4s, v3.4s, v11.4s\n" + "ldr q11, [x12, x9]\n" + "fmla v24.4s, v8.4s, v12.4s\n" + "fmla v20.4s, v5.4s, v12.4s\n" + "fmla v16.4s, v2.4s, v12.4s\n" + "ldr q12, [x11, x9]\n" + "add x11, x11, #0x10\n" + "fmla v19.4s, v8.4s, v10.4s\n" + "fmla v18.4s, v7.4s, v10.4s\n" + "fmla v17.4s, v6.4s, v10.4s\n" + "ldr q10, [x15, x6]\n" + "fmla v22.4s, v8.4s, v11.4s\n" + "fmla v21.4s, v7.4s, v11.4s\n" + "fmla v20.4s, v6.4s, v11.4s\n" + "fmla v18.4s, v5.4s, v11.4s\n" + "fmla v17.4s, v4.4s, v11.4s\n" + "fmla v16.4s, v3.4s, v11.4s\n" "ldr q11, [x15, x28]\n" - "fmla v29.4s, v8.4s, v12.4s\n" "add x15, x15, #0x10\n" - "fmla v30.4s, v7.4s, v12.4s\n" - "fmla v31.4s, v6.4s, v12.4s\n" - "ldr q12, [x9, x4]\n" - "fmla v16.4s, v4.4s, v10.4s\n" - "fmla v17.4s, v3.4s, v10.4s\n" - "fmax v16.4s, v16.4s, v15.4s\n" - "fmla v20.4s, v1.4s, v10.4s\n" - "fmla v21.4s, v0.4s, v10.4s\n" - "ldr q10, [x9, x28]\n" - "fmax v17.4s, v17.4s, v15.4s\n" - "fmla v18.4s, v5.4s, v11.4s\n" - "fmla v19.4s, v4.4s, v11.4s\n" - "fmax v18.4s, v18.4s, v15.4s\n" - "add x9, x9, #0x10\n" - "fmla v22.4s, v2.4s, v11.4s\n" - "fmla v23.4s, v1.4s, v11.4s\n" - "fmax v19.4s, v19.4s, v15.4s\n" - "fmla v24.4s, v7.4s, v12.4s\n" - "fmla v25.4s, v6.4s, v12.4s\n" - "fmax v20.4s, v20.4s, v15.4s\n" - "fmla v28.4s, v4.4s, v12.4s\n" - "fmla v29.4s, v3.4s, v12.4s\n" - "fmax v21.4s, v21.4s, v15.4s\n" - "fmla v26.4s, v8.4s, v10.4s\n" - "fmla v27.4s, v7.4s, v10.4s\n" - "fmax v22.4s, v22.4s, v15.4s\n" - "fmla v30.4s, v5.4s, v10.4s\n" + "fmla v18.4s, v8.4s, v12.4s\n" "fmla v31.4s, v4.4s, v10.4s\n" - "fmax v23.4s, v23.4s, v15.4s\n" - "fmax v24.4s, v24.4s, v15.4s\n" - "fmax v25.4s, v25.4s, v15.4s\n" - "fmax v26.4s, v26.4s, v15.4s\n" - "fmax v27.4s, v27.4s, v15.4s\n" - "fmax v28.4s, v28.4s, v15.4s\n" - "fmax v29.4s, v29.4s, v15.4s\n" - "fmax v30.4s, v30.4s, v15.4s\n" + "fmla v17.4s, v7.4s, v12.4s\n" + "fmla v16.4s, v6.4s, v12.4s\n" + "ldr q12, [x12, x6]\n" + "fmla v30.4s, v3.4s, v10.4s\n" + "fmla v27.4s, v1.4s, v10.4s\n" + "fmla v26.4s, v0.4s, v10.4s\n" + "ldr q10, [x12, x28]\n" + "add x12, x12, #0x10\n" + "fmla v29.4s, v5.4s, v11.4s\n" + "fmla v28.4s, v4.4s, v11.4s\n" + "fmla v25.4s, v2.4s, v11.4s\n" + "fmla v24.4s, v1.4s, v11.4s\n" + "fmla v23.4s, v7.4s, v12.4s\n" + "fmla v22.4s, v6.4s, v12.4s\n" + "fmla v19.4s, v4.4s, v12.4s\n" + "fmla v18.4s, v3.4s, v12.4s\n" + "fmla v21.4s, v8.4s, v10.4s\n" + "fmla v20.4s, v7.4s, v10.4s\n" + "fmla v17.4s, v5.4s, v10.4s\n" + "fmla v16.4s, v4.4s, v10.4s\n" "fmax v31.4s, v31.4s, v15.4s\n" - "fmin v16.4s, v16.4s, v14.4s\n" - "fmin v17.4s, v17.4s, v14.4s\n" - "st1 { v16.4s }, [x8]\n" - "fmin v18.4s, v18.4s, v14.4s\n" - "fmin v19.4s, v19.4s, v14.4s\n" - "str q17, [x8, x5]\n" - "fmin v20.4s, v20.4s, v14.4s\n" - "fmin v21.4s, v21.4s, v14.4s\n" - "str q18, [x8, x23]\n" - "fmin v22.4s, v22.4s, v14.4s\n" - "fmin v23.4s, v23.4s, v14.4s\n" - "str q19, [x8, x22]\n" - "add x8, x8, #0x10\n" - "fmin v24.4s, v24.4s, v14.4s\n" - "fmin v25.4s, v25.4s, v14.4s\n" - "st1 { v20.4s }, [x10]\n" - "fmin v26.4s, v26.4s, v14.4s\n" - "fmin v27.4s, v27.4s, v14.4s\n" - "str q21, [x10, x5]\n" - "fmin v28.4s, v28.4s, v14.4s\n" - "fmin v29.4s, v29.4s, v14.4s\n" - "str q22, [x10, x23]\n" - "fmin v30.4s, v30.4s, v14.4s\n" + "fmax v30.4s, v30.4s, v15.4s\n" + "fmax v29.4s, v29.4s, v15.4s\n" "fmin v31.4s, v31.4s, v14.4s\n" - "str q23, [x10, x22]\n" - "add x10, x10, #0x10\n" - "st1 { v24.4s }, [x27]\n" - "str q25, [x27, x5]\n" - "str q26, [x27, x23]\n" - "str q27, [x27, x22]\n" - "add x27, x27, #0x10\n" - "st1 { v28.4s }, [x24]\n" - "str q29, [x24, x5]\n" - "str q30, [x24, x23]\n" - "str q31, [x24, x22]\n" + "st1 { v31.4s }, [x16]\n" + "fmin v30.4s, v30.4s, v14.4s\n" + "fmin v29.4s, v29.4s, v14.4s\n" + "str q30, [x16, x17]\n" + "fmax v28.4s, v28.4s, v15.4s\n" + "fmax v27.4s, v27.4s, v15.4s\n" + "str q29, [x16, x23]\n" + "fmax v26.4s, v26.4s, v15.4s\n" + "fmax v25.4s, v25.4s, v15.4s\n" + "fmin v28.4s, v28.4s, v14.4s\n" + "str q28, [x16, x22]\n" + "fmin v27.4s, v27.4s, v14.4s\n" + "add x16, x16, #0x10\n" + "fmin v26.4s, v26.4s, v14.4s\n" + "st1 { v27.4s }, [x26]\n" + "fmin v25.4s, v25.4s, v14.4s\n" + "fmax v24.4s, v24.4s, v15.4s\n" + "str q26, [x26, x17]\n" + "fmax v23.4s, v23.4s, v15.4s\n" + "str q25, [x26, x23]\n" + "fmin v24.4s, v24.4s, v14.4s\n" + "fmax v22.4s, v22.4s, v15.4s\n" + "str q24, [x26, x22]\n" + "fmin v23.4s, v23.4s, v14.4s\n" + "add x26, x26, #0x10\n" + "fmin v22.4s, v22.4s, v14.4s\n" + "st1 { v23.4s }, [x25]\n" + "fmax v21.4s, v21.4s, v15.4s\n" + "fmax v20.4s, v20.4s, v15.4s\n" + "str q22, [x25, x17]\n" + "fmax v19.4s, v19.4s, v15.4s\n" + "fmax v18.4s, v18.4s, v15.4s\n" + "fmin v21.4s, v21.4s, v14.4s\n" + "str q21, [x25, x23]\n" + "fmin v20.4s, v20.4s, v14.4s\n" + "fmin v19.4s, v19.4s, v14.4s\n" + "str q20, [x25, x22]\n" + "fmin v18.4s, v18.4s, v14.4s\n" + "add x25, x25, #0x10\n" + "fmax v17.4s, v17.4s, v15.4s\n" + "st1 { v19.4s }, [x24]\n" + "fmax v16.4s, v16.4s, v15.4s\n" + "str q18, [x24, x17]\n" + "fmin v17.4s, v17.4s, v14.4s\n" + "str q17, [x24, x23]\n" + "fmin v16.4s, v16.4s, v14.4s\n" + "str q16, [x24, x22]\n" "add x24, x24, #0x10\n" "4:" // Tile loop: Oddments "tst %x[n_channels], #0x3\n" "beq 73f\n" - "ldr q13, [x16, #0x0]\n" - "ldr q0, [x16, #0x10]\n" - "add x23, x14, x17\n" - "add x22, x7, XZR\n" - "ldr q1, [x16, #0x20]\n" - "ldr q2, [x16, #0x30]\n" - "add x21, x7, x25\n" - "add x20, x14, x11\n" - "ldr q3, [x16, #0x40]\n" - "ldr q4, [x16, #0x50]\n" - "ldr q5, [x16, #0x60]\n" - "ldr q6, [x16, #0x70]\n" - "ldr q7, [x16, #0x80]\n" - "ldr q8, [x16, #0x90]\n" + "ldr q13, [x5, #0x0]\n" + "ldr q0, [x5, #0x10]\n" + "add x22, x14, x10\n" + "ldr q1, [x5, #0x20]\n" + "add x21, x8, XZR\n" + "ldr q2, [x5, #0x30]\n" + "add x20, x8, x27\n" + "ldr q3, [x5, #0x40]\n" + "add x19, x14, x9\n" + "ldr q4, [x5, #0x50]\n" + "ldr q5, [x5, #0x60]\n" + "ldr q6, [x5, #0x70]\n" + "ldr q7, [x5, #0x80]\n" + "ldr q8, [x5, #0x90]\n" "tbz %x[n_channels], #1, 5f\n" - "ldr d9, [x23], #0x8\n" - "ldr d10, [x22], #0x8\n" - "ldr d11, [x21], #0x8\n" - "ldr d12, [x20], #0x8\n" + "ldr d9, [x22], #0x8\n" + "ldr d10, [x21], #0x8\n" + "ldr d11, [x20], #0x8\n" + "ldr d12, [x19], #0x8\n" "tbz %x[n_channels], #0, 6f\n" - "ld1 { v9.s }[2], [x23]\n" - "ld1 { v10.s }[2], [x22]\n" - "ld1 { v11.s }[2], [x21]\n" - "ld1 { v12.s }[2], [x20]\n" + "ld1 { v9.s }[2], [x22]\n" + "ld1 { v10.s }[2], [x21]\n" + "ld1 { v11.s }[2], [x20]\n" + "ld1 { v12.s }[2], [x19]\n" "b 6f\n" "5:" // Tile loop: Oddments: Load inputs: (2, 2), (0, 0), (0, 5), (2, 3): Bit 1: Unset - "ldr s9, [x23, #0x0]\n" - "ldr s10, [x22, #0x0]\n" - "ldr s11, [x21, #0x0]\n" - "ldr s12, [x20, #0x0]\n" + "ldr s9, [x22, #0x0]\n" + "ldr s10, [x21, #0x0]\n" + "ldr s11, [x20, #0x0]\n" + "ldr s12, [x19, #0x0]\n" "6:" // Tile loop: Oddments: Load inputs: (2, 2), (0, 0), (0, 5), (2, 3): Bit 1: End - "mov v16.16b, v13.16b\n fmla v16.4s, v8.4s, v9.4s\n" - "mov v17.16b, v13.16b\n fmla v17.4s, v7.4s, v9.4s\n" - "add x20, x26, XZR\n" - "mov v18.16b, v13.16b\n fmla v18.4s, v6.4s, v9.4s\n" - "mov v21.16b, v13.16b\n fmla v21.4s, v4.4s, v9.4s\n" - "mov v22.16b, v13.16b\n fmla v22.4s, v3.4s, v9.4s\n" - "mov v25.16b, v13.16b\n fmla v25.4s, v1.4s, v9.4s\n" - "mov v26.16b, v13.16b\n fmla v26.4s, v0.4s, v9.4s\n" - "mov v19.16b, v13.16b\n fmla v19.4s, v2.4s, v11.4s\n" - "mov v20.16b, v13.16b\n fmla v20.4s, v5.4s, v9.4s\n" - "mov v24.16b, v13.16b\n fmla v24.4s, v2.4s, v9.4s\n" - "fmla v16.4s, v0.4s, v10.4s\n" - "fmla v17.4s, v8.4s, v12.4s\n" - "fmla v18.4s, v7.4s, v12.4s\n" - "fmla v19.4s, v6.4s, v12.4s\n" - "fmla v21.4s, v5.4s, v12.4s\n" - "fmla v22.4s, v4.4s, v12.4s\n" - "mov v23.16b, v13.16b\n fmla v23.4s, v3.4s, v12.4s\n" - "fmla v25.4s, v2.4s, v12.4s\n" - "fmla v26.4s, v1.4s, v12.4s\n" - "mov v27.16b, v13.16b\n fmla v27.4s, v0.4s, v12.4s\n" + "mov v31.16b, v13.16b\n fmla v31.4s, v8.4s, v9.4s\n" + "add x19, x11, XZR\n" + "mov v30.16b, v13.16b\n fmla v30.4s, v7.4s, v9.4s\n" + "mov v29.16b, v13.16b\n fmla v29.4s, v6.4s, v9.4s\n" + "mov v27.16b, v13.16b\n fmla v27.4s, v5.4s, v9.4s\n" + "mov v26.16b, v13.16b\n fmla v26.4s, v4.4s, v9.4s\n" + "mov v25.16b, v13.16b\n fmla v25.4s, v3.4s, v9.4s\n" + "mov v23.16b, v13.16b\n fmla v23.4s, v2.4s, v9.4s\n" + "mov v22.16b, v13.16b\n fmla v22.4s, v1.4s, v9.4s\n" + "mov v21.16b, v13.16b\n fmla v21.4s, v0.4s, v9.4s\n" + "fmla v31.4s, v0.4s, v10.4s\n" + "mov v28.16b, v13.16b\n fmla v28.4s, v2.4s, v11.4s\n" + "fmla v30.4s, v8.4s, v12.4s\n" + "fmla v29.4s, v7.4s, v12.4s\n" + "fmla v26.4s, v5.4s, v12.4s\n" + "fmla v28.4s, v6.4s, v12.4s\n" + "fmla v25.4s, v4.4s, v12.4s\n" + "mov v24.16b, v13.16b\n fmla v24.4s, v3.4s, v12.4s\n" + "fmla v22.4s, v2.4s, v12.4s\n" + "fmla v21.4s, v1.4s, v12.4s\n" + "mov v20.16b, v13.16b\n fmla v20.4s, v0.4s, v12.4s\n" "tbz %x[n_channels], #1, 7f\n" - "ldr d10, [x20], #0x8\n" + "ldr d10, [x19], #0x8\n" "tbz %x[n_channels], #0, 8f\n" - "ld1 { v10.s }[2], [x20]\n" + "ld1 { v10.s }[2], [x19]\n" "b 8f\n" "7:" // Tile loop: Oddments: Load inputs: (5, 0): Bit 1: Unset - "ldr s10, [x20, #0x0]\n" + "ldr s10, [x19, #0x0]\n" "8:" // Tile loop: Oddments: Load inputs: (5, 0): Bit 1: End - "mov v28.16b, v13.16b\n fmla v28.4s, v6.4s, v10.4s\n" - "add x20, x26, x25\n" + "mov v19.16b, v13.16b\n fmla v19.4s, v6.4s, v10.4s\n" + "add x19, x11, x27\n" "tbz %x[n_channels], #1, 9f\n" - "ldr d11, [x20], #0x8\n" + "ldr d11, [x19], #0x8\n" "tbz %x[n_channels], #0, 10f\n" - "ld1 { v11.s }[2], [x20]\n" + "ld1 { v11.s }[2], [x19]\n" "b 10f\n" "9:" // Tile loop: Oddments: Load inputs: (5, 5): Bit 1: Unset - "ldr s11, [x20, #0x0]\n" + "ldr s11, [x19, #0x0]\n" "10:" // Tile loop: Oddments: Load inputs: (5, 5): Bit 1: End - "mov v31.16b, v13.16b\n fmla v31.4s, v8.4s, v11.4s\n" - "add x20, x12, x17\n" + "mov v16.16b, v13.16b\n fmla v16.4s, v8.4s, v11.4s\n" + "add x19, x13, x10\n" "tbz %x[n_channels], #1, 11f\n" - "ldr d9, [x20], #0x8\n" + "ldr d9, [x19], #0x8\n" "tbz %x[n_channels], #0, 12f\n" - "ld1 { v9.s }[2], [x20]\n" + "ld1 { v9.s }[2], [x19]\n" "b 12f\n" "11:" // Tile loop: Oddments: Load inputs: (3, 2): Bit 1: Unset - "ldr s9, [x20, #0x0]\n" + "ldr s9, [x19, #0x0]\n" "12:" // Tile loop: Oddments: Load inputs: (3, 2): Bit 1: End - "fmla v20.4s, v8.4s, v9.4s\n" - "fmla v21.4s, v7.4s, v9.4s\n" - "add x20, x7, x4\n" - "fmla v22.4s, v6.4s, v9.4s\n" - "fmla v24.4s, v5.4s, v9.4s\n" - "fmla v25.4s, v4.4s, v9.4s\n" - "fmla v26.4s, v3.4s, v9.4s\n" - "fmla v28.4s, v2.4s, v9.4s\n" - "mov v29.16b, v13.16b\n fmla v29.4s, v1.4s, v9.4s\n" - "mov v30.16b, v13.16b\n fmla v30.4s, v0.4s, v9.4s\n" + "fmla v27.4s, v8.4s, v9.4s\n" + "add x19, x8, x6\n" + "fmla v26.4s, v7.4s, v9.4s\n" + "fmla v25.4s, v6.4s, v9.4s\n" + "fmla v23.4s, v5.4s, v9.4s\n" + "fmla v22.4s, v4.4s, v9.4s\n" + "fmla v21.4s, v3.4s, v9.4s\n" + "fmla v19.4s, v2.4s, v9.4s\n" + "mov v18.16b, v13.16b\n fmla v18.4s, v1.4s, v9.4s\n" + "mov v17.16b, v13.16b\n fmla v17.4s, v0.4s, v9.4s\n" "tbz %x[n_channels], #1, 13f\n" - "ldr d12, [x20], #0x8\n" + "ldr d12, [x19], #0x8\n" "tbz %x[n_channels], #0, 14f\n" - "ld1 { v12.s }[2], [x20]\n" + "ld1 { v12.s }[2], [x19]\n" "b 14f\n" "13:" // Tile loop: Oddments: Load inputs: (0, 1): Bit 1: Unset - "ldr s12, [x20, #0x0]\n" + "ldr s12, [x19, #0x0]\n" "14:" // Tile loop: Oddments: Load inputs: (0, 1): Bit 1: End - "fmla v16.4s, v1.4s, v12.4s\n" - "fmla v17.4s, v0.4s, v12.4s\n" - "add x20, x7, x28\n" + "fmla v31.4s, v1.4s, v12.4s\n" + "add x19, x8, x28\n" + "fmla v30.4s, v0.4s, v12.4s\n" "tbz %x[n_channels], #1, 15f\n" - "ldr d11, [x20], #0x8\n" + "ldr d11, [x19], #0x8\n" "tbz %x[n_channels], #0, 16f\n" - "ld1 { v11.s }[2], [x20]\n" + "ld1 { v11.s }[2], [x19]\n" "b 16f\n" "15:" // Tile loop: Oddments: Load inputs: (0, 4): Bit 1: Unset - "ldr s11, [x20, #0x0]\n" + "ldr s11, [x19, #0x0]\n" "16:" // Tile loop: Oddments: Load inputs: (0, 4): Bit 1: End - "fmla v18.4s, v2.4s, v11.4s\n" - "fmla v19.4s, v1.4s, v11.4s\n" - "add x20, x12, x11\n" + "fmla v29.4s, v2.4s, v11.4s\n" + "add x19, x13, x9\n" + "fmla v28.4s, v1.4s, v11.4s\n" "tbz %x[n_channels], #1, 17f\n" - "ldr d10, [x20], #0x8\n" + "ldr d10, [x19], #0x8\n" "tbz %x[n_channels], #0, 18f\n" - "ld1 { v10.s }[2], [x20]\n" + "ld1 { v10.s }[2], [x19]\n" "b 18f\n" "17:" // Tile loop: Oddments: Load inputs: (3, 3): Bit 1: Unset - "ldr s10, [x20, #0x0]\n" + "ldr s10, [x19, #0x0]\n" "18:" // Tile loop: Oddments: Load inputs: (3, 3): Bit 1: End - "fmla v21.4s, v8.4s, v10.4s\n" - "fmla v22.4s, v7.4s, v10.4s\n" - "add x20, x15, XZR\n" - "fmla v23.4s, v6.4s, v10.4s\n" - "fmla v25.4s, v5.4s, v10.4s\n" - "fmla v26.4s, v4.4s, v10.4s\n" - "fmla v27.4s, v3.4s, v10.4s\n" - "fmla v29.4s, v2.4s, v10.4s\n" - "fmla v30.4s, v1.4s, v10.4s\n" - "fmla v31.4s, v0.4s, v10.4s\n" + "fmla v26.4s, v8.4s, v10.4s\n" + "add x19, x15, XZR\n" + "fmla v25.4s, v7.4s, v10.4s\n" + "fmla v24.4s, v6.4s, v10.4s\n" + "fmla v22.4s, v5.4s, v10.4s\n" + "fmla v21.4s, v4.4s, v10.4s\n" + "fmla v20.4s, v3.4s, v10.4s\n" + "fmla v18.4s, v2.4s, v10.4s\n" + "fmla v17.4s, v1.4s, v10.4s\n" + "fmla v16.4s, v0.4s, v10.4s\n" "tbz %x[n_channels], #1, 19f\n" - "ldr d9, [x20], #0x8\n" + "ldr d9, [x19], #0x8\n" "tbz %x[n_channels], #0, 20f\n" - "ld1 { v9.s }[2], [x20]\n" + "ld1 { v9.s }[2], [x19]\n" "b 20f\n" "19:" // Tile loop: Oddments: Load inputs: (1, 0): Bit 1: Unset - "ldr s9, [x20, #0x0]\n" + "ldr s9, [x19, #0x0]\n" "20:" // Tile loop: Oddments: Load inputs: (1, 0): Bit 1: End - "fmla v16.4s, v3.4s, v9.4s\n" - "fmla v20.4s, v0.4s, v9.4s\n" - "add x20, x15, x25\n" + "fmla v31.4s, v3.4s, v9.4s\n" + "add x19, x15, x27\n" + "fmla v27.4s, v0.4s, v9.4s\n" "tbz %x[n_channels], #1, 21f\n" - "ldr d12, [x20], #0x8\n" + "ldr d12, [x19], #0x8\n" "tbz %x[n_channels], #0, 22f\n" - "ld1 { v12.s }[2], [x20]\n" + "ld1 { v12.s }[2], [x19]\n" "b 22f\n" "21:" // Tile loop: Oddments: Load inputs: (1, 5): Bit 1: Unset - "ldr s12, [x20, #0x0]\n" + "ldr s12, [x19, #0x0]\n" "22:" // Tile loop: Oddments: Load inputs: (1, 5): Bit 1: End - "fmla v19.4s, v5.4s, v12.4s\n" - "fmla v23.4s, v2.4s, v12.4s\n" - "add x20, x9, XZR\n" + "fmla v28.4s, v5.4s, v12.4s\n" + "add x19, x12, XZR\n" + "fmla v24.4s, v2.4s, v12.4s\n" "tbz %x[n_channels], #1, 23f\n" - "ldr d11, [x20], #0x8\n" + "ldr d11, [x19], #0x8\n" "tbz %x[n_channels], #0, 24f\n" - "ld1 { v11.s }[2], [x20]\n" + "ld1 { v11.s }[2], [x19]\n" "b 24f\n" "23:" // Tile loop: Oddments: Load inputs: (4, 0): Bit 1: Unset - "ldr s11, [x20, #0x0]\n" + "ldr s11, [x19, #0x0]\n" "24:" // Tile loop: Oddments: Load inputs: (4, 0): Bit 1: End - "fmla v24.4s, v6.4s, v11.4s\n" - "fmla v28.4s, v3.4s, v11.4s\n" - "add x20, x15, x17\n" + "fmla v23.4s, v6.4s, v11.4s\n" + "add x19, x15, x10\n" + "fmla v19.4s, v3.4s, v11.4s\n" "tbz %x[n_channels], #1, 25f\n" - "ldr d10, [x20], #0x8\n" + "ldr d10, [x19], #0x8\n" "tbz %x[n_channels], #0, 26f\n" - "ld1 { v10.s }[2], [x20]\n" + "ld1 { v10.s }[2], [x19]\n" "b 26f\n" "25:" // Tile loop: Oddments: Load inputs: (1, 2): Bit 1: Unset - "ldr s10, [x20, #0x0]\n" + "ldr s10, [x19, #0x0]\n" "26:" // Tile loop: Oddments: Load inputs: (1, 2): Bit 1: End - "fmla v16.4s, v5.4s, v10.4s\n" - "fmla v17.4s, v4.4s, v10.4s\n" - "add x20, x9, x25\n" - "fmla v18.4s, v3.4s, v10.4s\n" - "fmla v20.4s, v2.4s, v10.4s\n" - "fmla v21.4s, v1.4s, v10.4s\n" - "fmla v22.4s, v0.4s, v10.4s\n" + "fmla v31.4s, v5.4s, v10.4s\n" + "add x19, x12, x27\n" + "fmla v30.4s, v4.4s, v10.4s\n" + "fmla v29.4s, v3.4s, v10.4s\n" + "fmla v27.4s, v2.4s, v10.4s\n" + "fmla v26.4s, v1.4s, v10.4s\n" + "fmla v25.4s, v0.4s, v10.4s\n" "tbz %x[n_channels], #1, 27f\n" - "ldr d11, [x20], #0x8\n" + "ldr d11, [x19], #0x8\n" "tbz %x[n_channels], #0, 28f\n" - "ld1 { v11.s }[2], [x20]\n" + "ld1 { v11.s }[2], [x19]\n" "b 28f\n" "27:" // Tile loop: Oddments: Load inputs: (4, 5): Bit 1: Unset - "ldr s11, [x20, #0x0]\n" + "ldr s11, [x19, #0x0]\n" "28:" // Tile loop: Oddments: Load inputs: (4, 5): Bit 1: End - "fmla v27.4s, v8.4s, v11.4s\n" - "fmla v31.4s, v5.4s, v11.4s\n" - "add x20, x15, x11\n" + "fmla v20.4s, v8.4s, v11.4s\n" + "add x19, x15, x9\n" + "fmla v16.4s, v5.4s, v11.4s\n" "tbz %x[n_channels], #1, 29f\n" - "ldr d12, [x20], #0x8\n" + "ldr d12, [x19], #0x8\n" "tbz %x[n_channels], #0, 30f\n" - "ld1 { v12.s }[2], [x20]\n" + "ld1 { v12.s }[2], [x19]\n" "b 30f\n" "29:" // Tile loop: Oddments: Load inputs: (1, 3): Bit 1: Unset - "ldr s12, [x20, #0x0]\n" + "ldr s12, [x19, #0x0]\n" "30:" // Tile loop: Oddments: Load inputs: (1, 3): Bit 1: End - "fmla v17.4s, v5.4s, v12.4s\n" - "fmla v18.4s, v4.4s, v12.4s\n" - "add x20, x26, x4\n" - "fmla v19.4s, v3.4s, v12.4s\n" - "fmla v21.4s, v2.4s, v12.4s\n" - "fmla v22.4s, v1.4s, v12.4s\n" - "fmla v23.4s, v0.4s, v12.4s\n" + "fmla v30.4s, v5.4s, v12.4s\n" + "add x19, x11, x6\n" + "fmla v29.4s, v4.4s, v12.4s\n" + "fmla v28.4s, v3.4s, v12.4s\n" + "fmla v26.4s, v2.4s, v12.4s\n" + "fmla v25.4s, v1.4s, v12.4s\n" + "fmla v24.4s, v0.4s, v12.4s\n" "tbz %x[n_channels], #1, 31f\n" - "ldr d11, [x20], #0x8\n" + "ldr d11, [x19], #0x8\n" "tbz %x[n_channels], #0, 32f\n" - "ld1 { v11.s }[2], [x20]\n" + "ld1 { v11.s }[2], [x19]\n" "b 32f\n" "31:" // Tile loop: Oddments: Load inputs: (5, 1): Bit 1: Unset - "ldr s11, [x20, #0x0]\n" + "ldr s11, [x19, #0x0]\n" "32:" // Tile loop: Oddments: Load inputs: (5, 1): Bit 1: End - "fmla v28.4s, v7.4s, v11.4s\n" - "fmla v29.4s, v6.4s, v11.4s\n" - "add x20, x14, x4\n" + "fmla v19.4s, v7.4s, v11.4s\n" + "add x19, x14, x6\n" + "fmla v18.4s, v6.4s, v11.4s\n" "tbz %x[n_channels], #1, 33f\n" - "ldr d10, [x20], #0x8\n" + "ldr d10, [x19], #0x8\n" "tbz %x[n_channels], #0, 34f\n" - "ld1 { v10.s }[2], [x20]\n" + "ld1 { v10.s }[2], [x19]\n" "b 34f\n" "33:" // Tile loop: Oddments: Load inputs: (2, 1): Bit 1: Unset - "ldr s10, [x20, #0x0]\n" + "ldr s10, [x19, #0x0]\n" "34:" // Tile loop: Oddments: Load inputs: (2, 1): Bit 1: End - "fmla v16.4s, v7.4s, v10.4s\n" - "fmla v17.4s, v6.4s, v10.4s\n" - "add x20, x26, x28\n" - "fmla v20.4s, v4.4s, v10.4s\n" - "fmla v21.4s, v3.4s, v10.4s\n" - "fmla v24.4s, v1.4s, v10.4s\n" - "fmla v25.4s, v0.4s, v10.4s\n" + "fmla v31.4s, v7.4s, v10.4s\n" + "add x19, x11, x28\n" + "fmla v30.4s, v6.4s, v10.4s\n" + "fmla v27.4s, v4.4s, v10.4s\n" + "fmla v26.4s, v3.4s, v10.4s\n" + "fmla v23.4s, v1.4s, v10.4s\n" + "fmla v22.4s, v0.4s, v10.4s\n" "tbz %x[n_channels], #1, 35f\n" - "ldr d11, [x20], #0x8\n" + "ldr d11, [x19], #0x8\n" "tbz %x[n_channels], #0, 36f\n" - "ld1 { v11.s }[2], [x20]\n" + "ld1 { v11.s }[2], [x19]\n" "b 36f\n" "35:" // Tile loop: Oddments: Load inputs: (5, 4): Bit 1: Unset - "ldr s11, [x20, #0x0]\n" + "ldr s11, [x19, #0x0]\n" "36:" // Tile loop: Oddments: Load inputs: (5, 4): Bit 1: End - "fmla v30.4s, v8.4s, v11.4s\n" - "fmla v31.4s, v7.4s, v11.4s\n" - "add x20, x14, x28\n" + "fmla v17.4s, v8.4s, v11.4s\n" + "add x19, x14, x28\n" + "fmla v16.4s, v7.4s, v11.4s\n" "tbz %x[n_channels], #1, 37f\n" - "ldr d12, [x20], #0x8\n" + "ldr d12, [x19], #0x8\n" "tbz %x[n_channels], #0, 38f\n" - "ld1 { v12.s }[2], [x20]\n" + "ld1 { v12.s }[2], [x19]\n" "b 38f\n" "37:" // Tile loop: Oddments: Load inputs: (2, 4): Bit 1: Unset - "ldr s12, [x20, #0x0]\n" + "ldr s12, [x19, #0x0]\n" "38:" // Tile loop: Oddments: Load inputs: (2, 4): Bit 1: End - "fmla v18.4s, v8.4s, v12.4s\n" - "fmla v19.4s, v7.4s, v12.4s\n" - "add x20, x7, x17\n" - "fmla v22.4s, v5.4s, v12.4s\n" - "fmla v23.4s, v4.4s, v12.4s\n" - "fmla v26.4s, v2.4s, v12.4s\n" - "fmla v27.4s, v1.4s, v12.4s\n" + "fmla v29.4s, v8.4s, v12.4s\n" + "add x19, x8, x10\n" + "fmla v28.4s, v7.4s, v12.4s\n" + "fmla v25.4s, v5.4s, v12.4s\n" + "fmla v24.4s, v4.4s, v12.4s\n" + "fmla v21.4s, v2.4s, v12.4s\n" + "fmla v20.4s, v1.4s, v12.4s\n" "tbz %x[n_channels], #1, 39f\n" - "ldr d10, [x20], #0x8\n" + "ldr d10, [x19], #0x8\n" "tbz %x[n_channels], #0, 40f\n" - "ld1 { v10.s }[2], [x20]\n" + "ld1 { v10.s }[2], [x19]\n" "b 40f\n" "39:" // Tile loop: Oddments: Load inputs: (0, 2): Bit 1: Unset - "ldr s10, [x20, #0x0]\n" + "ldr s10, [x19, #0x0]\n" "40:" // Tile loop: Oddments: Load inputs: (0, 2): Bit 1: End - "fmla v16.4s, v2.4s, v10.4s\n" - "fmla v17.4s, v1.4s, v10.4s\n" - "add x20, x12, x4\n" - "fmla v18.4s, v0.4s, v10.4s\n" + "fmla v31.4s, v2.4s, v10.4s\n" + "add x19, x13, x6\n" + "fmla v30.4s, v1.4s, v10.4s\n" + "fmla v29.4s, v0.4s, v10.4s\n" "tbz %x[n_channels], #1, 41f\n" - "ldr d11, [x20], #0x8\n" + "ldr d11, [x19], #0x8\n" "tbz %x[n_channels], #0, 42f\n" - "ld1 { v11.s }[2], [x20]\n" + "ld1 { v11.s }[2], [x19]\n" "b 42f\n" "41:" // Tile loop: Oddments: Load inputs: (3, 1): Bit 1: Unset - "ldr s11, [x20, #0x0]\n" + "ldr s11, [x19, #0x0]\n" "42:" // Tile loop: Oddments: Load inputs: (3, 1): Bit 1: End - "fmla v20.4s, v7.4s, v11.4s\n" - "fmla v21.4s, v6.4s, v11.4s\n" - "add x20, x7, x11\n" - "fmla v24.4s, v4.4s, v11.4s\n" - "fmla v25.4s, v3.4s, v11.4s\n" - "fmla v28.4s, v1.4s, v11.4s\n" - "fmla v29.4s, v0.4s, v11.4s\n" + "fmla v27.4s, v7.4s, v11.4s\n" + "add x19, x8, x9\n" + "fmla v26.4s, v6.4s, v11.4s\n" + "fmla v23.4s, v4.4s, v11.4s\n" + "fmla v22.4s, v3.4s, v11.4s\n" + "fmla v19.4s, v1.4s, v11.4s\n" + "fmla v18.4s, v0.4s, v11.4s\n" "tbz %x[n_channels], #1, 43f\n" - "ldr d12, [x20], #0x8\n" + "ldr d12, [x19], #0x8\n" "tbz %x[n_channels], #0, 44f\n" - "ld1 { v12.s }[2], [x20]\n" + "ld1 { v12.s }[2], [x19]\n" "b 44f\n" "43:" // Tile loop: Oddments: Load inputs: (0, 3): Bit 1: Unset - "ldr s12, [x20, #0x0]\n" + "ldr s12, [x19, #0x0]\n" "44:" // Tile loop: Oddments: Load inputs: (0, 3): Bit 1: End - "fmla v17.4s, v2.4s, v12.4s\n" - "fmla v18.4s, v1.4s, v12.4s\n" - "add x20, x14, XZR\n" - "fmla v19.4s, v0.4s, v12.4s\n" + "fmla v30.4s, v2.4s, v12.4s\n" + "add x19, x14, XZR\n" + "fmla v29.4s, v1.4s, v12.4s\n" + "fmla v28.4s, v0.4s, v12.4s\n" "tbz %x[n_channels], #1, 45f\n" - "ldr d10, [x20], #0x8\n" + "ldr d10, [x19], #0x8\n" "tbz %x[n_channels], #0, 46f\n" - "ld1 { v10.s }[2], [x20]\n" + "ld1 { v10.s }[2], [x19]\n" "b 46f\n" "45:" // Tile loop: Oddments: Load inputs: (2, 0): Bit 1: Unset - "ldr s10, [x20, #0x0]\n" + "ldr s10, [x19, #0x0]\n" "46:" // Tile loop: Oddments: Load inputs: (2, 0): Bit 1: End - "fmla v16.4s, v6.4s, v10.4s\n" - "fmla v20.4s, v3.4s, v10.4s\n" - "add x20, x12, x28\n" - "fmla v24.4s, v0.4s, v10.4s\n" + "fmla v31.4s, v6.4s, v10.4s\n" + "add x19, x13, x28\n" + "fmla v27.4s, v3.4s, v10.4s\n" + "fmla v23.4s, v0.4s, v10.4s\n" "tbz %x[n_channels], #1, 47f\n" - "ldr d11, [x20], #0x8\n" + "ldr d11, [x19], #0x8\n" "tbz %x[n_channels], #0, 48f\n" - "ld1 { v11.s }[2], [x20]\n" + "ld1 { v11.s }[2], [x19]\n" "b 48f\n" "47:" // Tile loop: Oddments: Load inputs: (3, 4): Bit 1: Unset - "ldr s11, [x20, #0x0]\n" + "ldr s11, [x19, #0x0]\n" "48:" // Tile loop: Oddments: Load inputs: (3, 4): Bit 1: End - "fmla v22.4s, v8.4s, v11.4s\n" - "fmla v23.4s, v7.4s, v11.4s\n" - "add x20, x14, x25\n" - "fmla v26.4s, v5.4s, v11.4s\n" - "fmla v27.4s, v4.4s, v11.4s\n" - "fmla v30.4s, v2.4s, v11.4s\n" - "fmla v31.4s, v1.4s, v11.4s\n" + "fmla v25.4s, v8.4s, v11.4s\n" + "add x19, x14, x27\n" + "fmla v24.4s, v7.4s, v11.4s\n" + "fmla v21.4s, v5.4s, v11.4s\n" + "fmla v20.4s, v4.4s, v11.4s\n" + "fmla v17.4s, v2.4s, v11.4s\n" + "fmla v16.4s, v1.4s, v11.4s\n" "tbz %x[n_channels], #1, 49f\n" - "ldr d12, [x20], #0x8\n" + "ldr d12, [x19], #0x8\n" "tbz %x[n_channels], #0, 50f\n" - "ld1 { v12.s }[2], [x20]\n" + "ld1 { v12.s }[2], [x19]\n" "b 50f\n" "49:" // Tile loop: Oddments: Load inputs: (2, 5): Bit 1: Unset - "ldr s12, [x20, #0x0]\n" + "ldr s12, [x19, #0x0]\n" "50:" // Tile loop: Oddments: Load inputs: (2, 5): Bit 1: End - "fmla v19.4s, v8.4s, v12.4s\n" - "fmla v23.4s, v5.4s, v12.4s\n" - "add x20, x12, XZR\n" - "fmla v27.4s, v2.4s, v12.4s\n" + "fmla v28.4s, v8.4s, v12.4s\n" + "add x19, x13, XZR\n" + "fmla v24.4s, v5.4s, v12.4s\n" + "fmla v20.4s, v2.4s, v12.4s\n" "tbz %x[n_channels], #1, 51f\n" - "ldr d10, [x20], #0x8\n" + "ldr d10, [x19], #0x8\n" "tbz %x[n_channels], #0, 52f\n" - "ld1 { v10.s }[2], [x20]\n" + "ld1 { v10.s }[2], [x19]\n" "b 52f\n" "51:" // Tile loop: Oddments: Load inputs: (3, 0): Bit 1: Unset - "ldr s10, [x20, #0x0]\n" + "ldr s10, [x19, #0x0]\n" "52:" // Tile loop: Oddments: Load inputs: (3, 0): Bit 1: End - "fmla v20.4s, v6.4s, v10.4s\n" - "fmla v24.4s, v3.4s, v10.4s\n" - "add x20, x9, x17\n" - "fmla v28.4s, v0.4s, v10.4s\n" + "fmla v27.4s, v6.4s, v10.4s\n" + "add x19, x12, x10\n" + "fmla v23.4s, v3.4s, v10.4s\n" + "fmla v19.4s, v0.4s, v10.4s\n" "tbz %x[n_channels], #1, 53f\n" - "ldr d11, [x20], #0x8\n" + "ldr d11, [x19], #0x8\n" "tbz %x[n_channels], #0, 54f\n" - "ld1 { v11.s }[2], [x20]\n" + "ld1 { v11.s }[2], [x19]\n" "b 54f\n" "53:" // Tile loop: Oddments: Load inputs: (4, 2): Bit 1: Unset - "ldr s11, [x20, #0x0]\n" + "ldr s11, [x19, #0x0]\n" "54:" // Tile loop: Oddments: Load inputs: (4, 2): Bit 1: End - "fmla v24.4s, v8.4s, v11.4s\n" - "fmla v25.4s, v7.4s, v11.4s\n" - "add x20, x12, x25\n" - "fmla v26.4s, v6.4s, v11.4s\n" - "fmla v28.4s, v5.4s, v11.4s\n" - "fmla v29.4s, v4.4s, v11.4s\n" - "fmla v30.4s, v3.4s, v11.4s\n" + "fmla v23.4s, v8.4s, v11.4s\n" + "add x19, x13, x27\n" + "fmla v22.4s, v7.4s, v11.4s\n" + "fmla v21.4s, v6.4s, v11.4s\n" + "fmla v19.4s, v5.4s, v11.4s\n" + "fmla v18.4s, v4.4s, v11.4s\n" + "fmla v17.4s, v3.4s, v11.4s\n" "tbz %x[n_channels], #1, 55f\n" - "ldr d12, [x20], #0x8\n" + "ldr d12, [x19], #0x8\n" "tbz %x[n_channels], #0, 56f\n" - "ld1 { v12.s }[2], [x20]\n" + "ld1 { v12.s }[2], [x19]\n" "b 56f\n" "55:" // Tile loop: Oddments: Load inputs: (3, 5): Bit 1: Unset - "ldr s12, [x20, #0x0]\n" + "ldr s12, [x19, #0x0]\n" "56:" // Tile loop: Oddments: Load inputs: (3, 5): Bit 1: End - "fmla v23.4s, v8.4s, v12.4s\n" - "fmla v27.4s, v5.4s, v12.4s\n" - "add x20, x26, x17\n" - "fmla v31.4s, v2.4s, v12.4s\n" + "fmla v24.4s, v8.4s, v12.4s\n" + "add x19, x11, x10\n" + "fmla v20.4s, v5.4s, v12.4s\n" + "fmla v16.4s, v2.4s, v12.4s\n" "tbz %x[n_channels], #1, 57f\n" - "ldr d10, [x20], #0x8\n" + "ldr d10, [x19], #0x8\n" "tbz %x[n_channels], #0, 58f\n" - "ld1 { v10.s }[2], [x20]\n" + "ld1 { v10.s }[2], [x19]\n" "b 58f\n" "57:" // Tile loop: Oddments: Load inputs: (5, 2): Bit 1: Unset - "ldr s10, [x20, #0x0]\n" + "ldr s10, [x19, #0x0]\n" "58:" // Tile loop: Oddments: Load inputs: (5, 2): Bit 1: End - "fmla v28.4s, v8.4s, v10.4s\n" - "fmla v29.4s, v7.4s, v10.4s\n" - "add x20, x9, x11\n" - "fmla v30.4s, v6.4s, v10.4s\n" + "fmla v19.4s, v8.4s, v10.4s\n" + "add x19, x12, x9\n" + "fmla v18.4s, v7.4s, v10.4s\n" + "fmla v17.4s, v6.4s, v10.4s\n" "tbz %x[n_channels], #1, 59f\n" - "ldr d11, [x20], #0x8\n" + "ldr d11, [x19], #0x8\n" "tbz %x[n_channels], #0, 60f\n" - "ld1 { v11.s }[2], [x20]\n" + "ld1 { v11.s }[2], [x19]\n" "b 60f\n" "59:" // Tile loop: Oddments: Load inputs: (4, 3): Bit 1: Unset - "ldr s11, [x20, #0x0]\n" + "ldr s11, [x19, #0x0]\n" "60:" // Tile loop: Oddments: Load inputs: (4, 3): Bit 1: End - "fmla v25.4s, v8.4s, v11.4s\n" - "fmla v26.4s, v7.4s, v11.4s\n" - "add x20, x26, x11\n" - "fmla v27.4s, v6.4s, v11.4s\n" - "fmla v29.4s, v5.4s, v11.4s\n" - "fmla v30.4s, v4.4s, v11.4s\n" - "fmla v31.4s, v3.4s, v11.4s\n" + "fmla v22.4s, v8.4s, v11.4s\n" + "add x19, x11, x9\n" + "fmla v21.4s, v7.4s, v11.4s\n" + "fmla v20.4s, v6.4s, v11.4s\n" + "fmla v18.4s, v5.4s, v11.4s\n" + "fmla v17.4s, v4.4s, v11.4s\n" + "fmla v16.4s, v3.4s, v11.4s\n" "tbz %x[n_channels], #1, 61f\n" - "ldr d12, [x20], #0x8\n" + "ldr d12, [x19], #0x8\n" "tbz %x[n_channels], #0, 62f\n" - "ld1 { v12.s }[2], [x20]\n" + "ld1 { v12.s }[2], [x19]\n" "b 62f\n" "61:" // Tile loop: Oddments: Load inputs: (5, 3): Bit 1: Unset - "ldr s12, [x20, #0x0]\n" + "ldr s12, [x19, #0x0]\n" "62:" // Tile loop: Oddments: Load inputs: (5, 3): Bit 1: End - "fmla v29.4s, v8.4s, v12.4s\n" - "fmla v30.4s, v7.4s, v12.4s\n" - "add x20, x15, x4\n" - "fmla v31.4s, v6.4s, v12.4s\n" + "fmla v18.4s, v8.4s, v12.4s\n" + "add x19, x15, x6\n" + "fmla v17.4s, v7.4s, v12.4s\n" + "fmla v16.4s, v6.4s, v12.4s\n" "tbz %x[n_channels], #1, 63f\n" - "ldr d10, [x20], #0x8\n" + "ldr d10, [x19], #0x8\n" "tbz %x[n_channels], #0, 64f\n" - "ld1 { v10.s }[2], [x20]\n" + "ld1 { v10.s }[2], [x19]\n" "b 64f\n" "63:" // Tile loop: Oddments: Load inputs: (1, 1): Bit 1: Unset - "ldr s10, [x20, #0x0]\n" + "ldr s10, [x19, #0x0]\n" "64:" // Tile loop: Oddments: Load inputs: (1, 1): Bit 1: End - "fmla v16.4s, v4.4s, v10.4s\n" - "fmla v17.4s, v3.4s, v10.4s\n" - "add x20, x15, x28\n" - "fmla v20.4s, v1.4s, v10.4s\n" - "fmla v21.4s, v0.4s, v10.4s\n" + "fmla v31.4s, v4.4s, v10.4s\n" + "add x19, x15, x28\n" + "fmla v30.4s, v3.4s, v10.4s\n" + "fmla v27.4s, v1.4s, v10.4s\n" + "fmla v26.4s, v0.4s, v10.4s\n" "tbz %x[n_channels], #1, 65f\n" - "ldr d11, [x20], #0x8\n" + "ldr d11, [x19], #0x8\n" "tbz %x[n_channels], #0, 66f\n" - "ld1 { v11.s }[2], [x20]\n" + "ld1 { v11.s }[2], [x19]\n" "b 66f\n" "65:" // Tile loop: Oddments: Load inputs: (1, 4): Bit 1: Unset - "ldr s11, [x20, #0x0]\n" + "ldr s11, [x19, #0x0]\n" "66:" // Tile loop: Oddments: Load inputs: (1, 4): Bit 1: End - "fmla v18.4s, v5.4s, v11.4s\n" - "fmla v19.4s, v4.4s, v11.4s\n" - "add x20, x9, x4\n" - "fmla v22.4s, v2.4s, v11.4s\n" - "fmla v23.4s, v1.4s, v11.4s\n" + "fmla v29.4s, v5.4s, v11.4s\n" + "add x19, x12, x6\n" + "fmla v28.4s, v4.4s, v11.4s\n" + "fmla v25.4s, v2.4s, v11.4s\n" + "fmla v24.4s, v1.4s, v11.4s\n" "tbz %x[n_channels], #1, 67f\n" - "ldr d12, [x20], #0x8\n" + "ldr d12, [x19], #0x8\n" "tbz %x[n_channels], #0, 68f\n" - "ld1 { v12.s }[2], [x20]\n" + "ld1 { v12.s }[2], [x19]\n" "b 68f\n" "67:" // Tile loop: Oddments: Load inputs: (4, 1): Bit 1: Unset - "ldr s12, [x20, #0x0]\n" + "ldr s12, [x19, #0x0]\n" "68:" // Tile loop: Oddments: Load inputs: (4, 1): Bit 1: End - "fmla v24.4s, v7.4s, v12.4s\n" - "fmla v25.4s, v6.4s, v12.4s\n" - "add x20, x9, x28\n" - "fmla v28.4s, v4.4s, v12.4s\n" - "fmla v29.4s, v3.4s, v12.4s\n" + "fmla v23.4s, v7.4s, v12.4s\n" + "add x19, x12, x28\n" + "fmla v22.4s, v6.4s, v12.4s\n" + "fmla v19.4s, v4.4s, v12.4s\n" + "fmla v18.4s, v3.4s, v12.4s\n" "tbz %x[n_channels], #1, 69f\n" - "ldr d10, [x20], #0x8\n" + "ldr d10, [x19], #0x8\n" "tbz %x[n_channels], #0, 70f\n" - "ld1 { v10.s }[2], [x20]\n" + "ld1 { v10.s }[2], [x19]\n" "b 70f\n" "69:" // Tile loop: Oddments: Load inputs: (4, 4): Bit 1: Unset - "ldr s10, [x20, #0x0]\n" + "ldr s10, [x19, #0x0]\n" "70:" // Tile loop: Oddments: Load inputs: (4, 4): Bit 1: End - "fmla v26.4s, v8.4s, v10.4s\n" - "fmla v27.4s, v7.4s, v10.4s\n" - "fmax v16.4s, v16.4s, v15.4s\n" - "fmla v30.4s, v5.4s, v10.4s\n" - "fmla v31.4s, v4.4s, v10.4s\n" - "fmax v17.4s, v17.4s, v15.4s\n" - "fmax v18.4s, v18.4s, v15.4s\n" - "fmax v19.4s, v19.4s, v15.4s\n" - "fmax v20.4s, v20.4s, v15.4s\n" - "fmax v21.4s, v21.4s, v15.4s\n" - "fmax v22.4s, v22.4s, v15.4s\n" - "fmax v23.4s, v23.4s, v15.4s\n" - "fmax v24.4s, v24.4s, v15.4s\n" - "fmax v25.4s, v25.4s, v15.4s\n" - "fmax v26.4s, v26.4s, v15.4s\n" - "fmax v27.4s, v27.4s, v15.4s\n" - "fmax v28.4s, v28.4s, v15.4s\n" - "fmax v29.4s, v29.4s, v15.4s\n" - "fmax v30.4s, v30.4s, v15.4s\n" + "fmla v21.4s, v8.4s, v10.4s\n" + "fmla v20.4s, v7.4s, v10.4s\n" + "fmla v17.4s, v5.4s, v10.4s\n" + "fmla v16.4s, v4.4s, v10.4s\n" "fmax v31.4s, v31.4s, v15.4s\n" - "fmin v16.4s, v16.4s, v14.4s\n" - "fmin v17.4s, v17.4s, v14.4s\n" - "fmin v18.4s, v18.4s, v14.4s\n" - "fmin v19.4s, v19.4s, v14.4s\n" - "fmin v20.4s, v20.4s, v14.4s\n" - "fmin v21.4s, v21.4s, v14.4s\n" - "fmin v22.4s, v22.4s, v14.4s\n" - "fmin v23.4s, v23.4s, v14.4s\n" - "fmin v24.4s, v24.4s, v14.4s\n" - "fmin v25.4s, v25.4s, v14.4s\n" - "fmin v26.4s, v26.4s, v14.4s\n" - "fmin v27.4s, v27.4s, v14.4s\n" - "fmin v28.4s, v28.4s, v14.4s\n" - "fmin v29.4s, v29.4s, v14.4s\n" - "fmin v30.4s, v30.4s, v14.4s\n" + "fmax v30.4s, v30.4s, v15.4s\n" + "fmax v29.4s, v29.4s, v15.4s\n" "fmin v31.4s, v31.4s, v14.4s\n" + "fmin v30.4s, v30.4s, v14.4s\n" + "fmin v29.4s, v29.4s, v14.4s\n" + "fmax v28.4s, v28.4s, v15.4s\n" + "fmax v27.4s, v27.4s, v15.4s\n" + "fmax v26.4s, v26.4s, v15.4s\n" + "fmin v28.4s, v28.4s, v14.4s\n" + "fmin v27.4s, v27.4s, v14.4s\n" + "fmin v26.4s, v26.4s, v14.4s\n" + "fmax v25.4s, v25.4s, v15.4s\n" + "fmax v24.4s, v24.4s, v15.4s\n" + "fmax v23.4s, v23.4s, v15.4s\n" + "fmin v25.4s, v25.4s, v14.4s\n" + "fmin v24.4s, v24.4s, v14.4s\n" + "fmin v23.4s, v23.4s, v14.4s\n" + "fmax v22.4s, v22.4s, v15.4s\n" + "fmax v21.4s, v21.4s, v15.4s\n" + "fmax v20.4s, v20.4s, v15.4s\n" + "fmin v22.4s, v22.4s, v14.4s\n" + "fmin v21.4s, v21.4s, v14.4s\n" + "fmin v20.4s, v20.4s, v14.4s\n" + "fmax v19.4s, v19.4s, v15.4s\n" + "fmax v18.4s, v18.4s, v15.4s\n" + "fmax v17.4s, v17.4s, v15.4s\n" + "fmin v19.4s, v19.4s, v14.4s\n" + "fmin v18.4s, v18.4s, v14.4s\n" + "fmin v17.4s, v17.4s, v14.4s\n" + "fmax v16.4s, v16.4s, v15.4s\n" + "fmin v16.4s, v16.4s, v14.4s\n" "tbz %x[n_channels], #1, 71f\n" - "mov x23, x8\n" - "mov x22, x10\n" - "st1 { v16.d }[0], [x23], x5\n" - "mov x21, x27\n" - "mov x20, x24\n" - "st1 { v20.d }[0], [x22], x5\n" - "st1 { v24.d }[0], [x21], x5\n" - "add x8, x8, #0x8\n" - "add x10, x10, #0x8\n" - "st1 { v28.d }[0], [x20], x5\n" - "add x27, x27, #0x8\n" + "mov x19, x16\n" + "st1 { v31.d }[0], [x19], x17\n" + "add x16, x16, #0x8\n" + "st1 { v30.d }[0], [x19], x17\n" + "mov x21, x26\n" + "st1 { v29.d }[0], [x19], x17\n" + "st1 { v27.d }[0], [x21], x17\n" + "add x26, x26, #0x8\n" + "st1 { v28.d }[0], [x19]\n" + "mov x20, x25\n" + "st1 { v26.d }[0], [x21], x17\n" + "add x25, x25, #0x8\n" + "st1 { v25.d }[0], [x21], x17\n" + "mov x19, x24\n" + "st1 { v24.d }[0], [x21]\n" "add x24, x24, #0x8\n" - "st1 { v17.d }[0], [x23], x5\n" - "st1 { v21.d }[0], [x22], x5\n" - "st1 { v25.d }[0], [x21], x5\n" - "st1 { v29.d }[0], [x20], x5\n" - "st1 { v18.d }[0], [x23], x5\n" - "st1 { v22.d }[0], [x22], x5\n" - "st1 { v26.d }[0], [x21], x5\n" - "st1 { v30.d }[0], [x20], x5\n" - "st1 { v19.d }[0], [x23]\n" - "st1 { v23.d }[0], [x22]\n" - "st1 { v27.d }[0], [x21]\n" - "st1 { v31.d }[0], [x20]\n" + "st1 { v23.d }[0], [x20], x17\n" + "st1 { v22.d }[0], [x20], x17\n" + "st1 { v21.d }[0], [x20], x17\n" + "st1 { v20.d }[0], [x20]\n" + "st1 { v19.d }[0], [x19], x17\n" + "st1 { v18.d }[0], [x19], x17\n" + "st1 { v17.d }[0], [x19], x17\n" + "st1 { v16.d }[0], [x19]\n" "tbz %x[n_channels], #0, 72f\n" - "mov x23, x8\n" - "mov x22, x10\n" - "st1 { v16.s }[2], [x23], x5\n" - "mov x21, x27\n" - "mov x20, x24\n" - "st1 { v20.s }[2], [x22], x5\n" - "st1 { v24.s }[2], [x21], x5\n" - "st1 { v28.s }[2], [x20], x5\n" - "st1 { v17.s }[2], [x23], x5\n" - "st1 { v21.s }[2], [x22], x5\n" - "st1 { v25.s }[2], [x21], x5\n" - "st1 { v29.s }[2], [x20], x5\n" - "st1 { v18.s }[2], [x23], x5\n" - "st1 { v22.s }[2], [x22], x5\n" - "st1 { v26.s }[2], [x21], x5\n" - "st1 { v30.s }[2], [x20], x5\n" - "st1 { v19.s }[2], [x23]\n" - "st1 { v23.s }[2], [x22]\n" - "st1 { v27.s }[2], [x21]\n" - "st1 { v31.s }[2], [x20]\n" + "mov x22, x16\n" + "st1 { v31.s }[2], [x22], x17\n" + "mov x21, x26\n" + "st1 { v30.s }[2], [x22], x17\n" + "st1 { v27.s }[2], [x21], x17\n" + "mov x20, x25\n" + "st1 { v29.s }[2], [x22], x17\n" + "mov x19, x24\n" + "st1 { v28.s }[2], [x22]\n" + "st1 { v26.s }[2], [x21], x17\n" + "st1 { v25.s }[2], [x21], x17\n" + "st1 { v24.s }[2], [x21]\n" + "st1 { v23.s }[2], [x20], x17\n" + "st1 { v22.s }[2], [x20], x17\n" + "st1 { v21.s }[2], [x20], x17\n" + "st1 { v20.s }[2], [x20]\n" + "st1 { v19.s }[2], [x19], x17\n" + "st1 { v18.s }[2], [x19], x17\n" + "st1 { v17.s }[2], [x19], x17\n" + "st1 { v16.s }[2], [x19]\n" "b 72f\n" "71:" // Tile loop: Oddments: Store: Bit 1: Unset - "mov x23, x8\n" - "mov x22, x10\n" - "st1 { v16.s }[0], [x23], x5\n" - "mov x21, x27\n" - "mov x20, x24\n" - "st1 { v20.s }[0], [x22], x5\n" - "st1 { v24.s }[0], [x21], x5\n" - "st1 { v28.s }[0], [x20], x5\n" - "st1 { v17.s }[0], [x23], x5\n" - "st1 { v21.s }[0], [x22], x5\n" - "st1 { v25.s }[0], [x21], x5\n" - "st1 { v29.s }[0], [x20], x5\n" - "st1 { v18.s }[0], [x23], x5\n" - "st1 { v22.s }[0], [x22], x5\n" - "st1 { v26.s }[0], [x21], x5\n" - "st1 { v30.s }[0], [x20], x5\n" - "st1 { v19.s }[0], [x23]\n" - "st1 { v23.s }[0], [x22]\n" - "st1 { v27.s }[0], [x21]\n" - "st1 { v31.s }[0], [x20]\n" + "mov x22, x16\n" + "st1 { v31.s }[0], [x22], x17\n" + "mov x21, x26\n" + "mov x20, x25\n" + "st1 { v30.s }[0], [x22], x17\n" + "st1 { v27.s }[0], [x21], x17\n" + "mov x19, x24\n" + "st1 { v29.s }[0], [x22], x17\n" + "st1 { v28.s }[0], [x22]\n" + "st1 { v26.s }[0], [x21], x17\n" + "st1 { v25.s }[0], [x21], x17\n" + "st1 { v24.s }[0], [x21]\n" + "st1 { v23.s }[0], [x20], x17\n" + "st1 { v22.s }[0], [x20], x17\n" + "st1 { v21.s }[0], [x20], x17\n" + "st1 { v20.s }[0], [x20]\n" + "st1 { v19.s }[0], [x19], x17\n" + "st1 { v18.s }[0], [x19], x17\n" + "st1 { v17.s }[0], [x19], x17\n" + "st1 { v16.s }[0], [x19]\n" "72:" // Tile loop: Oddments: Store: Bit 1: End + "73:" // Tile loop: End + "ldr x4, [%x[params_struct], %[offsetof_args_tile_i]]\n" + "add x21, x4, #0x1\n" "ldr x26, [%x[params_struct], %[offsetof_args_tile_j]]\n" - "ldr x27, [%x[params_struct], %[offsetof_args_tile_i]]\n" - "add x26, x26, #0x1\n" - "add x21, x27, #0x1\n" - "ldr x20, [%x[params_struct], %[offsetof_args_n_tile_cols]]\n" - "cmp x26, x20\n" "ldr x20, [%x[params_struct], %[offsetof_args_n_tile_rows]]\n" - "csel x27, x27, x21, LT\n" + "add x26, x26, #0x1\n" + "ldr x19, [%x[params_struct], %[offsetof_args_n_tile_cols]]\n" + "cmp x26, x19\n" "csel x26, x26, XZR, LT\n" - "cmp x27, x20\n" + "csel x4, x4, x21, LT\n" + "cmp x4, x20\n" "blt 1b\n" : : [n_channels] "r" ((unsigned long) n_channels), [offsetof_args_inptr] "I" (offsetof(Args, inptr)), [offsetof_args_ld_input_col] "I" (offsetof(Args, ld_input_col)), [offsetof_args_ld_input_row] "I" (offsetof(Args, ld_input_row)), [offsetof_args_ld_output_col] "I" (offsetof(Args, ld_output_col)), [offsetof_args_ld_output_row] "I" (offsetof(Args, ld_output_row)), [offsetof_args_max] "I" (offsetof(Args, max)), [offsetof_args_min] "I" (offsetof(Args, min)), [offsetof_args_n_tile_cols] "I" (offsetof(Args, n_tile_cols)), [offsetof_args_n_tile_rows] "I" (offsetof(Args, n_tile_rows)), [offsetof_args_outptr] "I" (offsetof(Args, outptr)), [offsetof_args_params] "I" (offsetof(Args, params)), [offsetof_args_tile_i] "I" (offsetof(Args, tile_i)), [offsetof_args_tile_j] "I" (offsetof(Args, tile_j)), [params_struct] "r" (¶ms_struct) - : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v8", "v9", "v10", "v11", "v12", "v13", "v14", "v15", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31", "x4", "x5", "x6", "x7", "x8", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x16", "x17", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28" + : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v8", "v9", "v10", "v11", "v12", "v13", "v14", "v15", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31", "x4", "x5", "x6", "x7", "x8", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x16", "x17", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28" ); } -- cgit v1.2.1