diff options
| author | Freddie Liardet <frederick.liardet@arm.com> | 2021-08-04 11:57:37 +0100 |
|---|---|---|
| committer | Sheri Zhang <sheri.zhang@arm.com> | 2021-08-04 15:54:00 +0000 |
| commit | 6269f87f5caca3a83a4822e0556fbf284158615c (patch) | |
| tree | 0f814b2f6356aa073529a8d00c94913e006d1f6d /src/core/NEON/kernels/arm_conv/depthwise/kernels/a64_fp16_nhwc_3x3_s1_output3x3_mla_depthfirst | |
| parent | 38ce29d5dce9a09769b55b2717d105f984247e31 (diff) | |
| download | ComputeLibrary-6269f87f5caca3a83a4822e0556fbf284158615c.tar.gz | |
Fix depthwise convolution assembly kernels
Resolves: COMPMID-4710
Change-Id: I35b964731aeed0e6f4f873f59341bee48e4a41fd
Signed-off-by: Freddie Liardet <frederick.liardet@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/6039
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/core/NEON/kernels/arm_conv/depthwise/kernels/a64_fp16_nhwc_3x3_s1_output3x3_mla_depthfirst')
2 files changed, 2008 insertions, 1293 deletions
diff --git a/src/core/NEON/kernels/arm_conv/depthwise/kernels/a64_fp16_nhwc_3x3_s1_output3x3_mla_depthfirst/generic_direct.cpp b/src/core/NEON/kernels/arm_conv/depthwise/kernels/a64_fp16_nhwc_3x3_s1_output3x3_mla_depthfirst/generic_direct.cpp index 3bdd544a54..c2d67661f6 100644 --- a/src/core/NEON/kernels/arm_conv/depthwise/kernels/a64_fp16_nhwc_3x3_s1_output3x3_mla_depthfirst/generic_direct.cpp +++ b/src/core/NEON/kernels/arm_conv/depthwise/kernels/a64_fp16_nhwc_3x3_s1_output3x3_mla_depthfirst/generic_direct.cpp @@ -87,739 +87,1069 @@ void a64_fp16_nhwc_3x3_s1_output3x3_mla_depthfirst_direct_impl( ); __asm__ __volatile__( - "mov x7, #0x0\n" - "mov x8, #0x0\n" + "mov x23, #0x0\n" + "mov x22, #0x0\n" "1:" // Tile loop - "str x7, [%x[params_struct], %[offsetof_args_tile_i]]\n" + "str x23, [%x[params_struct], %[offsetof_args_tile_i]]\n" "mov x26, #0x3\n" - "str x8, [%x[params_struct], %[offsetof_args_tile_j]]\n" "mov x25, #0x3\n" - "ldr x17, [%x[params_struct], %[offsetof_args_params]]\n" - "add x24, %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 x16, [%x[params_struct], %[offsetof_args_ld_input_col]]\n" - "mov x23, #0x0\n" + "str x22, [%x[params_struct], %[offsetof_args_tile_j]]\n" + "ldr x24, [%x[params_struct], %[offsetof_args_ld_input_row]]\n" + "ldr x21, [%x[params_struct], %[offsetof_args_ld_output_row]]\n" + "mul x20, x23, x24\n" // offset = tile_i * ld_input_row + "ldr x17, [%x[params_struct], %[offsetof_args_ld_input_col]]\n" + "ldr x16, [%x[params_struct], %[offsetof_args_ld_output_col]]\n" + "mul x19, x23, x21\n" // offset = tile_i * ld_output_row + "mov x23, #0x10\n" // cntb _, ALL, #1 + "madd x20, x22, x17, x20\n" // offset += tile_j * ld_input_col "ldr x15, [%x[params_struct], %[offsetof_args_inptr]]\n" - "mul x19, x7, x22\n" // offset = tile_i * ld_input_row - "ldr x20, [%x[params_struct], %[offsetof_args_ld_output_row]]\n" - "madd x19, x8, x16, x19\n" // offset += tile_j * ld_input_col - "ldr x14, [%x[params_struct], %[offsetof_args_ld_output_col]]\n" - "mul x19, x19, x26\n" // offset *= kernel_stride * output_size - "ldr x13, [%x[params_struct], %[offsetof_args_outptr]]\n" - "add x15, x15, x19, LSL #1\n" // inptr[0] += offset * sizeof(__fp16) - "ld1r { v18.8h }, [x24]\n" - "add x12, x15, x22, LSL #1\n" - "ld1r { v17.8h }, [x21]\n" - "add x11, x12, x22, LSL #1\n" + "lsl x17, x17, #0x1\n" + "ldr x14, [%x[params_struct], %[offsetof_args_outptr]]\n" + "madd x19, x22, x16, x19\n" // offset += tile_j * ld_output_col "lsl x16, x16, #0x1\n" - "add x10, x11, x22, LSL #1\n" - "add x9, x10, x22, LSL #1\n" - "add x28, x16, x16\n" - "add x27, x28, x16\n" - "add x26, x27, x16\n" - "mul x19, x7, x20\n" // offset = tile_i * ld_output_row - "madd x19, x8, x14, x19\n" // offset += tile_j * ld_output_col - "mul x19, x19, x25\n" // offset *= output_tile_size - "add x13, x13, x19, LSL #1\n" // outptrs[0] += offset * sizeof(__fp16) - "add x25, x13, x20, LSL #1\n" - "add x24, x25, x20, LSL #1\n" - "lsl x14, x14, #0x1\n" - "add x22, x14, x14\n" - "mov x21, #0x10\n" // cntb _, ALL, #1 - "sub x20, XZR, x21\n" - "lsr x19, %x[n_channels], #0x3\n" - "cbz x19, 4f\n" - "ldr q16, [x17, #0x0]\n" - "ldr q0, [x17, #0x10]\n" - "cmp x21, x19, LSL #4\n" - "ldr q1, [x17, #0x20]\n" - "ldr q2, [x17, #0x30]\n" - "ldr q3, [x17, #0x40]\n" - "ldr q4, [x17, #0x50]\n" - "ldr q5, [x17, #0x60]\n" - "ldr q6, [x17, #0x70]\n" - "ldr q7, [x17, #0x80]\n" - "ldr q8, [x17, #0x90]\n" - "add x17, x17, #0xa0\n" - "ldr q9, [x11, x28]\n" + "lsr x22, %x[n_channels], #0x3\n" + "ldr x13, [%x[params_struct], %[offsetof_args_params]]\n" + "mul x20, x20, x26\n" // offset *= kernel_stride * output_size + "add x15, x15, x20, LSL #1\n" // inptr[0] += offset * sizeof(__fp16) + "add x12, x15, x24, LSL #1\n" + "mul x19, x19, x25\n" // offset *= output_tile_size + "add x11, x12, x24, LSL #1\n" + "add x10, x17, x17\n" + "add x14, x14, x19, LSL #1\n" // outptrs[0] += offset * sizeof(__fp16) + "add x9, x11, x24, LSL #1\n" + "add x28, x10, x17\n" + "add x27, x14, x21, LSL #1\n" + "add x20, %x[params_struct], %[offsetof_args_min]\n" + "add x19, %x[params_struct], %[offsetof_args_max]\n" + "ld1r { v18.8h }, [x20]\n" + "ld1r { v17.8h }, [x19]\n" + "add x26, x9, x24, LSL #1\n" + "add x25, x28, x17\n" + "add x24, x27, x21, LSL #1\n" + "add x21, x16, x16\n" + "mov x20, #0x0\n" + "sub x19, XZR, x23\n" + "cbz x22, 4f\n" + "ldr q16, [x13, #0x0]\n" + "cmp x23, x22, LSL #4\n" + "ldr q0, [x13, #0x10]\n" + "ldr q1, [x13, #0x20]\n" + "ldr q2, [x13, #0x30]\n" + "ldr q3, [x13, #0x40]\n" + "ldr q4, [x13, #0x50]\n" + "ldr q5, [x13, #0x60]\n" + "ldr q6, [x13, #0x70]\n" + "ldr q7, [x13, #0x80]\n" + "ldr q8, [x13, #0x90]\n" + "ldr q9, [x11, x10]\n" + "add x13, x13, #0xa0\n" "ld1 { v10.8h }, [x15]\n" - "ldr q11, [x15, x26]\n" - "ld1 { v12.8h }, [x9]\n" - "ldr q13, [x12, x28]\n" + "ldr q11, [x15, x25]\n" + "ld1 { v12.8h }, [x26]\n" + "ldr q13, [x12, x10]\n" "bge 3f\n" "2:" // Tile loop: Channel loop - "mov v31.16b, v16.16b\n fmla v31.8h, v8.8h, v9.8h\n" - "add x20, x20, #0x10\n" - "mov v30.16b, v16.16b\n fmla v30.8h, v7.8h, v9.8h\n" + "mov v24.16b, v16.16b\n fmla v24.8h, v7.8h, v9.8h\n" + "mov v23.16b, v16.16b\n fmla v23.8h, v8.8h, v9.8h\n" "add x23, x23, #0x10\n" - "mov v29.16b, v16.16b\n fmla v29.8h, v6.8h, v9.8h\n" - "add x21, x21, #0x10\n" - "mov v28.16b, v16.16b\n fmla v28.8h, v5.8h, v9.8h\n" - "cmp x21, x19, LSL #4\n" + "cmp x23, x22, LSL #4\n" + "mov v25.16b, v16.16b\n fmla v25.8h, v6.8h, v9.8h\n" + "fmla v24.8h, v4.8h, v13.8h\n" + "add x19, x19, #0x10\n" + "add x20, x20, #0x10\n" + "mov v26.16b, v16.16b\n fmla v26.8h, v5.8h, v9.8h\n" "mov v27.16b, v16.16b\n fmla v27.8h, v4.8h, v9.8h\n" - "mov v26.16b, v16.16b\n fmla v26.8h, v3.8h, v9.8h\n" - "mov v25.16b, v16.16b\n fmla v25.8h, v2.8h, v9.8h\n" - "mov v24.16b, v16.16b\n fmla v24.8h, v1.8h, v9.8h\n" - "mov v23.16b, v16.16b\n fmla v23.8h, v0.8h, v9.8h\n" - "ldr q16, [x17, #0x0]\n" - "fmla v31.8h, v0.8h, v10.8h\n" - "ldr q10, [x11, x27]\n" - "fmla v29.8h, v2.8h, v11.8h\n" - "ldr q11, [x11, x16]\n" - "fmla v25.8h, v6.8h, v12.8h\n" - "ldr q12, [x9, x26]\n" - "fmla v30.8h, v4.8h, v13.8h\n" - "fmla v31.8h, v5.8h, v13.8h\n" - "fmla v29.8h, v3.8h, v13.8h\n" - "fmla v28.8h, v2.8h, v13.8h\n" + "mov v28.16b, v16.16b\n fmla v28.8h, v3.8h, v9.8h\n" + "fmla v23.8h, v0.8h, v10.8h\n" + "ldr q10, [x11, x28]\n" + "fmla v25.8h, v2.8h, v11.8h\n" + "ldr q11, [x11, x17]\n" + "mov v29.16b, v16.16b\n fmla v29.8h, v2.8h, v9.8h\n" + "fmla v24.8h, v6.8h, v11.8h\n" + "mov v31.16b, v16.16b\n fmla v31.8h, v0.8h, v9.8h\n" + "fmla v23.8h, v5.8h, v13.8h\n" + "fmla v25.8h, v3.8h, v13.8h\n" + "fmla v26.8h, v2.8h, v13.8h\n" "fmla v27.8h, v1.8h, v13.8h\n" - "fmla v26.8h, v0.8h, v13.8h\n" - "ldr q13, [x15, x16]\n" - "fmla v23.8h, v8.8h, v12.8h\n" - "ldr q12, [x15, x27]\n" - "fmla v31.8h, v7.8h, v11.8h\n" - "fmla v30.8h, v6.8h, v11.8h\n" - "fmla v28.8h, v4.8h, v11.8h\n" + "fmla v28.8h, v0.8h, v13.8h\n" + "ldr q13, [x15, x17]\n" + "fmla v29.8h, v6.8h, v12.8h\n" + "ldr q12, [x26, x25]\n" + "mov v30.16b, v16.16b\n fmla v30.8h, v1.8h, v9.8h\n" + "fmla v24.8h, v0.8h, v13.8h\n" + "ldr q16, [x13, #0x0]\n" + "fmla v31.8h, v8.8h, v12.8h\n" + "ldr q12, [x15, x28]\n" + "fmla v23.8h, v7.8h, v11.8h\n" + "fmla v30.8h, v0.8h, v11.8h\n" + "fmla v26.8h, v4.8h, v11.8h\n" "fmla v27.8h, v3.8h, v11.8h\n" - "fmla v25.8h, v1.8h, v11.8h\n" - "fmla v24.8h, v0.8h, v11.8h\n" + "fmla v29.8h, v1.8h, v11.8h\n" "ld1 { v11.8h }, [x12]\n" - "fmla v31.8h, v1.8h, v13.8h\n" - "fmla v30.8h, v0.8h, v13.8h\n" - "ldr q13, [x12, x26]\n" - "fmla v29.8h, v1.8h, v12.8h\n" + "fmla v24.8h, v2.8h, v12.8h\n" + "fmla v25.8h, v1.8h, v12.8h\n" + "ld1 { v12.8h }, [x9]\n" + "fmla v28.8h, v4.8h, v10.8h\n" + "fmla v23.8h, v1.8h, v13.8h\n" + "ldr q13, [x12, x25]\n" + "fmla v30.8h, v2.8h, v10.8h\n" + "fmla v31.8h, v1.8h, v10.8h\n" + "fmla v24.8h, v8.8h, v10.8h\n" + "fmla v25.8h, v7.8h, v10.8h\n" "fmla v27.8h, v5.8h, v10.8h\n" - "fmla v26.8h, v4.8h, v10.8h\n" - "fmla v30.8h, v2.8h, v12.8h\n" - "ld1 { v12.8h }, [x10]\n" - "fmla v29.8h, v7.8h, v10.8h\n" - "fmla v24.8h, v2.8h, v10.8h\n" - "fmla v23.8h, v1.8h, v10.8h\n" - "fmla v30.8h, v8.8h, v10.8h\n" - "ldr q10, [x10, x28]\n" - "fmla v31.8h, v3.8h, v11.8h\n" - "fmla v28.8h, v0.8h, v11.8h\n" - "ldr q11, [x10, x26]\n" - "fmla v29.8h, v5.8h, v13.8h\n" - "fmla v26.8h, v2.8h, v13.8h\n" - "ldr q13, [x9, x16]\n" - "fmla v25.8h, v3.8h, v12.8h\n" - "fmla v28.8h, v6.8h, v12.8h\n" - "ldr q12, [x12, x16]\n" + "fmla v26.8h, v0.8h, v11.8h\n" + "ldr q10, [x9, x10]\n" + "fmla v29.8h, v3.8h, v12.8h\n" + "fmla v28.8h, v2.8h, v13.8h\n" + "fmla v30.8h, v4.8h, v10.8h\n" + "fmla v31.8h, v3.8h, v10.8h\n" + "fmla v23.8h, v3.8h, v11.8h\n" + "fmla v25.8h, v5.8h, v13.8h\n" + "ldr q11, [x9, x25]\n" + "ldr q13, [x26, x17]\n" + "fmla v26.8h, v6.8h, v12.8h\n" "fmla v27.8h, v7.8h, v10.8h\n" - "fmla v26.8h, v6.8h, v10.8h\n" - "fmla v25.8h, v5.8h, v10.8h\n" - "fmla v28.8h, v8.8h, v10.8h\n" - "fmla v24.8h, v4.8h, v10.8h\n" - "fmla v23.8h, v3.8h, v10.8h\n" - "fmla v26.8h, v8.8h, v11.8h\n" - "fmla v25.8h, v7.8h, v13.8h\n" - "fmla v24.8h, v6.8h, v13.8h\n" - "ldr q13, [x9, x27]\n" - "fmla v23.8h, v5.8h, v11.8h\n" - "ldr q11, [x12, x27]\n" - "add x12, x12, #0x10\n" - "fmla v31.8h, v4.8h, v12.8h\n" - "fmla v30.8h, v3.8h, v12.8h\n" - "fmla v28.8h, v1.8h, v12.8h\n" + "ldr q12, [x12, x17]\n" + "fmla v29.8h, v5.8h, v10.8h\n" + "fmla v28.8h, v6.8h, v10.8h\n" + "fmla v31.8h, v5.8h, v11.8h\n" + "fmla v30.8h, v6.8h, v13.8h\n" + "fmla v26.8h, v8.8h, v10.8h\n" + "fmla v29.8h, v7.8h, v13.8h\n" + "ldr q13, [x26, x28]\n" + "fmla v24.8h, v3.8h, v12.8h\n" "fmla v27.8h, v0.8h, v12.8h\n" - "ldr q12, [x10, x16]\n" - "fmla v29.8h, v4.8h, v11.8h\n" - "fmla v30.8h, v5.8h, v11.8h\n" - "fmla v26.8h, v1.8h, v11.8h\n" + "fmla v28.8h, v8.8h, v11.8h\n" + "ldr q11, [x12, x28]\n" + "fmla v30.8h, v8.8h, v13.8h\n" + "add x12, x12, #0x10\n" + "fmla v31.8h, v7.8h, v13.8h\n" + "fmla v23.8h, v4.8h, v12.8h\n" + "ldr q13, [x9, x28]\n" + "fmla v26.8h, v1.8h, v12.8h\n" + "fmla v24.8h, v5.8h, v11.8h\n" + "ldr q12, [x9, x17]\n" + "add x9, x9, #0x10\n" + "fmla v25.8h, v4.8h, v11.8h\n" "fmla v27.8h, v2.8h, v11.8h\n" - "ldr q11, [x15, x28]\n" + "fmla v28.8h, v1.8h, v11.8h\n" + "ldr q11, [x15, x10]\n" + "fmla v29.8h, v4.8h, v12.8h\n" "add x15, x15, #0x10\n" - "fmla v24.8h, v8.8h, v13.8h\n" + "fmla v30.8h, v3.8h, v12.8h\n" + "fmla v31.8h, v4.8h, v13.8h\n" "ld1 { v10.8h }, [x15]\n" - "fmla v23.8h, v7.8h, v13.8h\n" - "ldr q13, [x10, x27]\n" - "add x10, x10, #0x10\n" - "fmla v28.8h, v7.8h, v12.8h\n" + "ldr q4, [x13, #0x50]\n" + "fmla v26.8h, v7.8h, v12.8h\n" "fmla v27.8h, v6.8h, v12.8h\n" - "fmla v25.8h, v4.8h, v12.8h\n" - "fmla v24.8h, v3.8h, v12.8h\n" "ld1 { v12.8h }, [x11]\n" - "fmla v31.8h, v2.8h, v11.8h\n" - "fmla v30.8h, v1.8h, v11.8h\n" - "ldr q1, [x17, #0x20]\n" - "fmla v29.8h, v0.8h, v11.8h\n" - "ldr q11, [x11, x26]\n" + "fmla v23.8h, v2.8h, v11.8h\n" + "fmla v24.8h, v1.8h, v11.8h\n" + "fmax v24.8h, v24.8h, v18.8h\n" + "ldr q1, [x13, #0x20]\n" + "fmla v25.8h, v0.8h, v11.8h\n" + "ldr q11, [x11, x25]\n" + "fmla v28.8h, v7.8h, v13.8h\n" "add x11, x11, #0x10\n" + "fmla v30.8h, v5.8h, v13.8h\n" + "fmla v29.8h, v0.8h, v12.8h\n" + "fmin v24.8h, v24.8h, v17.8h\n" + "ldr q9, [x11, x10]\n" + "fmla v31.8h, v2.8h, v11.8h\n" "fmla v27.8h, v8.8h, v13.8h\n" - "ldr q9, [x11, x28]\n" - "fmla v26.8h, v7.8h, v13.8h\n" - "fmla v24.8h, v5.8h, v13.8h\n" - "fmla v23.8h, v4.8h, v13.8h\n" - "ldr q13, [x9, x28]\n" - "add x9, x9, #0x10\n" - "fmla v31.8h, v6.8h, v12.8h\n" - "ldr q4, [x17, #0x50]\n" - "fmla v28.8h, v3.8h, v12.8h\n" - "ldr q3, [x17, #0x40]\n" - "fmla v25.8h, v0.8h, v12.8h\n" - "ld1 { v12.8h }, [x9]\n" - "fmla v29.8h, v8.8h, v11.8h\n" - "ldr q0, [x17, #0x10]\n" - "fmla v26.8h, v5.8h, v11.8h\n" - "ldr q5, [x17, #0x60]\n" - "fmla v23.8h, v2.8h, v11.8h\n" - "ldr q11, [x15, x26]\n" - "fmla v25.8h, v8.8h, v13.8h\n" - "ldr q2, [x17, #0x30]\n" - "fmla v24.8h, v7.8h, v13.8h\n" - "ldr q7, [x17, #0x80]\n" - "fmax v31.8h, v31.8h, v18.8h\n" - "ldr q8, [x17, #0x90]\n" - "fmla v23.8h, v6.8h, v13.8h\n" - "ldr q13, [x12, x28]\n" - "fmax v30.8h, v30.8h, v18.8h\n" - "ldr q6, [x17, #0x70]\n" - "add x17, x17, #0xa0\n" - "fmin v31.8h, v31.8h, v17.8h\n" - "st1 { v31.8h }, [x13]\n" - "fmin v30.8h, v30.8h, v17.8h\n" - "fmax v29.8h, v29.8h, v18.8h\n" - "str q30, [x13, x14]\n" - "fmin v29.8h, v29.8h, v17.8h\n" - "fmax v28.8h, v28.8h, v18.8h\n" - "str q29, [x13, x22]\n" - "fmin v28.8h, v28.8h, v17.8h\n" - "add x13, x13, #0x10\n" + "ldr q13, [x26, x10]\n" "fmax v27.8h, v27.8h, v18.8h\n" - "st1 { v28.8h }, [x25]\n" - "fmax v26.8h, v26.8h, v18.8h\n" - "fmax v25.8h, v25.8h, v18.8h\n" - "fmin v27.8h, v27.8h, v17.8h\n" - "str q27, [x25, x14]\n" - "fmin v26.8h, v26.8h, v17.8h\n" - "fmin v25.8h, v25.8h, v17.8h\n" - "str q26, [x25, x22]\n" - "fmax v24.8h, v24.8h, v18.8h\n" - "add x25, x25, #0x10\n" + "fmla v23.8h, v6.8h, v12.8h\n" + "fmla v26.8h, v3.8h, v12.8h\n" "fmax v23.8h, v23.8h, v18.8h\n" - "st1 { v25.8h }, [x24]\n" - "fmin v24.8h, v24.8h, v17.8h\n" - "str q24, [x24, x14]\n" + "add x26, x26, #0x10\n" + "fmla v25.8h, v8.8h, v11.8h\n" + "fmla v28.8h, v5.8h, v11.8h\n" + "fmax v25.8h, v25.8h, v18.8h\n" + "ldr q11, [x15, x25]\n" + "fmla v29.8h, v8.8h, v13.8h\n" + "fmla v30.8h, v7.8h, v13.8h\n" + "fmax v26.8h, v26.8h, v18.8h\n" + "ld1 { v12.8h }, [x26]\n" + "fmla v31.8h, v6.8h, v13.8h\n" + "fmax v28.8h, v28.8h, v18.8h\n" + "ldr q13, [x12, x10]\n" + "ldr q0, [x13, #0x10]\n" + "fmax v29.8h, v29.8h, v18.8h\n" + "fmax v30.8h, v30.8h, v18.8h\n" + "ldr q2, [x13, #0x30]\n" + "ldr q3, [x13, #0x40]\n" + "fmax v31.8h, v31.8h, v18.8h\n" "fmin v23.8h, v23.8h, v17.8h\n" - "str q23, [x24, x22]\n" + "st1 { v23.8h }, [x14]\n" + "ldr q5, [x13, #0x60]\n" + "fmin v25.8h, v25.8h, v17.8h\n" + "fmin v26.8h, v26.8h, v17.8h\n" + "str q24, [x14, x16]\n" + "ldr q6, [x13, #0x70]\n" + "fmin v27.8h, v27.8h, v17.8h\n" + "fmin v28.8h, v28.8h, v17.8h\n" + "str q25, [x14, x21]\n" + "add x14, x14, #0x10\n" + "fmin v29.8h, v29.8h, v17.8h\n" + "fmin v30.8h, v30.8h, v17.8h\n" + "st1 { v26.8h }, [x27]\n" + "ldr q7, [x13, #0x80]\n" + "fmin v31.8h, v31.8h, v17.8h\n" + "str q27, [x27, x16]\n" + "ldr q8, [x13, #0x90]\n" + "add x13, x13, #0xa0\n" + "str q28, [x27, x21]\n" + "add x27, x27, #0x10\n" + "st1 { v29.8h }, [x24]\n" + "str q30, [x24, x16]\n" + "str q31, [x24, x21]\n" "add x24, x24, #0x10\n" "blt 2b\n" "3:" // Tile loop: Channel tail - "mov v31.16b, v16.16b\n fmla v31.8h, v8.8h, v9.8h\n" - "mov v30.16b, v16.16b\n fmla v30.8h, v7.8h, v9.8h\n" - "mov v29.16b, v16.16b\n fmla v29.8h, v6.8h, v9.8h\n" - "mov v28.16b, v16.16b\n fmla v28.8h, v5.8h, v9.8h\n" + "mov v24.16b, v16.16b\n fmla v24.8h, v7.8h, v9.8h\n" + "mov v23.16b, v16.16b\n fmla v23.8h, v8.8h, v9.8h\n" + "mov v25.16b, v16.16b\n fmla v25.8h, v6.8h, v9.8h\n" + "fmla v24.8h, v4.8h, v13.8h\n" + "mov v26.16b, v16.16b\n fmla v26.8h, v5.8h, v9.8h\n" "mov v27.16b, v16.16b\n fmla v27.8h, v4.8h, v9.8h\n" - "mov v26.16b, v16.16b\n fmla v26.8h, v3.8h, v9.8h\n" - "mov v25.16b, v16.16b\n fmla v25.8h, v2.8h, v9.8h\n" - "mov v24.16b, v16.16b\n fmla v24.8h, v1.8h, v9.8h\n" - "mov v23.16b, v16.16b\n fmla v23.8h, v0.8h, v9.8h\n" - "fmla v31.8h, v0.8h, v10.8h\n" - "ldr q10, [x11, x27]\n" - "fmla v29.8h, v2.8h, v11.8h\n" - "ldr q11, [x11, x16]\n" - "fmla v25.8h, v6.8h, v12.8h\n" - "ldr q12, [x9, x26]\n" - "fmla v30.8h, v4.8h, v13.8h\n" - "fmla v31.8h, v5.8h, v13.8h\n" - "fmla v29.8h, v3.8h, v13.8h\n" - "fmla v28.8h, v2.8h, v13.8h\n" + "mov v28.16b, v16.16b\n fmla v28.8h, v3.8h, v9.8h\n" + "fmla v23.8h, v0.8h, v10.8h\n" + "ldr q10, [x11, x28]\n" + "fmla v25.8h, v2.8h, v11.8h\n" + "ldr q11, [x11, x17]\n" + "mov v29.16b, v16.16b\n fmla v29.8h, v2.8h, v9.8h\n" + "fmla v24.8h, v6.8h, v11.8h\n" + "mov v31.16b, v16.16b\n fmla v31.8h, v0.8h, v9.8h\n" + "fmla v23.8h, v5.8h, v13.8h\n" + "fmla v25.8h, v3.8h, v13.8h\n" + "fmla v26.8h, v2.8h, v13.8h\n" "fmla v27.8h, v1.8h, v13.8h\n" - "fmla v26.8h, v0.8h, v13.8h\n" - "ldr q13, [x15, x16]\n" - "fmla v23.8h, v8.8h, v12.8h\n" - "ldr q12, [x15, x27]\n" - "fmla v31.8h, v7.8h, v11.8h\n" - "fmla v30.8h, v6.8h, v11.8h\n" - "fmla v28.8h, v4.8h, v11.8h\n" + "fmla v28.8h, v0.8h, v13.8h\n" + "ldr q13, [x15, x17]\n" + "fmla v29.8h, v6.8h, v12.8h\n" + "ldr q12, [x26, x25]\n" + "mov v30.16b, v16.16b\n fmla v30.8h, v1.8h, v9.8h\n" + "fmla v24.8h, v0.8h, v13.8h\n" + "fmla v31.8h, v8.8h, v12.8h\n" + "ldr q12, [x15, x28]\n" + "fmla v23.8h, v7.8h, v11.8h\n" + "fmla v30.8h, v0.8h, v11.8h\n" + "fmla v26.8h, v4.8h, v11.8h\n" "fmla v27.8h, v3.8h, v11.8h\n" - "fmla v25.8h, v1.8h, v11.8h\n" - "fmla v24.8h, v0.8h, v11.8h\n" + "fmla v29.8h, v1.8h, v11.8h\n" "ld1 { v11.8h }, [x12]\n" - "fmla v31.8h, v1.8h, v13.8h\n" - "fmla v30.8h, v0.8h, v13.8h\n" - "ldr q13, [x12, x26]\n" - "fmla v29.8h, v1.8h, v12.8h\n" + "fmla v24.8h, v2.8h, v12.8h\n" + "fmla v25.8h, v1.8h, v12.8h\n" + "ld1 { v12.8h }, [x9]\n" + "fmla v28.8h, v4.8h, v10.8h\n" + "fmla v23.8h, v1.8h, v13.8h\n" + "ldr q13, [x12, x25]\n" + "fmla v30.8h, v2.8h, v10.8h\n" + "fmla v31.8h, v1.8h, v10.8h\n" + "fmla v24.8h, v8.8h, v10.8h\n" + "fmla v25.8h, v7.8h, v10.8h\n" "fmla v27.8h, v5.8h, v10.8h\n" - "fmla v26.8h, v4.8h, v10.8h\n" - "fmla v30.8h, v2.8h, v12.8h\n" - "ld1 { v12.8h }, [x10]\n" - "fmla v29.8h, v7.8h, v10.8h\n" - "fmla v24.8h, v2.8h, v10.8h\n" - "fmla v23.8h, v1.8h, v10.8h\n" - "fmla v30.8h, v8.8h, v10.8h\n" - "ldr q10, [x10, x28]\n" - "fmla v31.8h, v3.8h, v11.8h\n" - "fmla v28.8h, v0.8h, v11.8h\n" - "ldr q11, [x10, x26]\n" - "fmla v29.8h, v5.8h, v13.8h\n" - "fmla v26.8h, v2.8h, v13.8h\n" - "ldr q13, [x9, x16]\n" - "fmla v25.8h, v3.8h, v12.8h\n" - "fmla v28.8h, v6.8h, v12.8h\n" - "ldr q12, [x12, x16]\n" + "fmla v26.8h, v0.8h, v11.8h\n" + "ldr q10, [x9, x10]\n" + "fmla v29.8h, v3.8h, v12.8h\n" + "fmla v28.8h, v2.8h, v13.8h\n" + "fmla v30.8h, v4.8h, v10.8h\n" + "fmla v31.8h, v3.8h, v10.8h\n" + "fmla v23.8h, v3.8h, v11.8h\n" + "fmla v25.8h, v5.8h, v13.8h\n" + "ldr q11, [x9, x25]\n" + "ldr q13, [x26, x17]\n" + "fmla v26.8h, v6.8h, v12.8h\n" "fmla v27.8h, v7.8h, v10.8h\n" - "fmla v26.8h, v6.8h, v10.8h\n" - "fmla v25.8h, v5.8h, v10.8h\n" - "fmla v28.8h, v8.8h, v10.8h\n" - "fmla v24.8h, v4.8h, v10.8h\n" - "fmla v23.8h, v3.8h, v10.8h\n" - "fmla v26.8h, v8.8h, v11.8h\n" - "fmla v25.8h, v7.8h, v13.8h\n" - "fmla v24.8h, v6.8h, v13.8h\n" - "ldr q13, [x9, x27]\n" - "fmla v23.8h, v5.8h, v11.8h\n" - "ldr q11, [x12, x27]\n" - "add x12, x12, #0x10\n" - "fmla v31.8h, v4.8h, v12.8h\n" - "fmla v30.8h, v3.8h, v12.8h\n" - "fmla v28.8h, v1.8h, v12.8h\n" + "ldr q12, [x12, x17]\n" + "fmla v29.8h, v5.8h, v10.8h\n" + "fmla v28.8h, v6.8h, v10.8h\n" + "fmla v31.8h, v5.8h, v11.8h\n" + "fmla v30.8h, v6.8h, v13.8h\n" + "fmla v26.8h, v8.8h, v10.8h\n" + "fmla v29.8h, v7.8h, v13.8h\n" + "ldr q13, [x26, x28]\n" + "fmla v24.8h, v3.8h, v12.8h\n" "fmla v27.8h, v0.8h, v12.8h\n" - "ldr q12, [x10, x16]\n" - "fmla v29.8h, v4.8h, v11.8h\n" - "fmla v30.8h, v5.8h, v11.8h\n" - "fmla v26.8h, v1.8h, v11.8h\n" + "fmla v28.8h, v8.8h, v11.8h\n" + "ldr q11, [x12, x28]\n" + "fmla v30.8h, v8.8h, v13.8h\n" + "add x12, x12, #0x10\n" + "fmla v31.8h, v7.8h, v13.8h\n" + "fmla v23.8h, v4.8h, v12.8h\n" + "ldr q13, [x9, x28]\n" + "fmla v26.8h, v1.8h, v12.8h\n" + "fmla v24.8h, v5.8h, v11.8h\n" + "ldr q12, [x9, x17]\n" + "add x9, x9, #0x10\n" + "fmla v25.8h, v4.8h, v11.8h\n" "fmla v27.8h, v2.8h, v11.8h\n" - "ldr q11, [x15, x28]\n" + "fmla v28.8h, v1.8h, v11.8h\n" + "ldr q11, [x15, x10]\n" + "fmla v29.8h, v4.8h, v12.8h\n" "add x15, x15, #0x10\n" - "fmla v24.8h, v8.8h, v13.8h\n" - "fmla v23.8h, v7.8h, v13.8h\n" - "ldr q13, [x10, x27]\n" - "add x10, x10, #0x10\n" - "fmla v28.8h, v7.8h, v12.8h\n" + "fmla v30.8h, v3.8h, v12.8h\n" + "fmla v31.8h, v4.8h, v13.8h\n" + "fmla v26.8h, v7.8h, v12.8h\n" "fmla v27.8h, v6.8h, v12.8h\n" - "fmla v25.8h, v4.8h, v12.8h\n" - "fmla v24.8h, v3.8h, v12.8h\n" "ld1 { v12.8h }, [x11]\n" - "fmla v31.8h, v2.8h, v11.8h\n" - "fmla v30.8h, v1.8h, v11.8h\n" - "fmla v29.8h, v0.8h, v11.8h\n" - "ldr q11, [x11, x26]\n" + "fmla v23.8h, v2.8h, v11.8h\n" + "fmla v24.8h, v1.8h, v11.8h\n" + "fmax v24.8h, v24.8h, v18.8h\n" + "fmla v25.8h, v0.8h, v11.8h\n" + "ldr q11, [x11, x25]\n" + "fmla v28.8h, v7.8h, v13.8h\n" + "fmin v24.8h, v24.8h, v17.8h\n" + "fmla v30.8h, v5.8h, v13.8h\n" + "fmla v29.8h, v0.8h, v12.8h\n" "add x11, x11, #0x10\n" + "fmla v31.8h, v2.8h, v11.8h\n" "fmla v27.8h, v8.8h, v13.8h\n" - "fmla v26.8h, v7.8h, v13.8h\n" - "fmla v24.8h, v5.8h, v13.8h\n" - "fmla v23.8h, v4.8h, v13.8h\n" - "ldr q13, [x9, x28]\n" - "add x9, x9, #0x10\n" - "fmla v31.8h, v6.8h, v12.8h\n" - "fmla v28.8h, v3.8h, v12.8h\n" - "fmla v25.8h, v0.8h, v12.8h\n" - "fmla v29.8h, v8.8h, v11.8h\n" - "fmla v26.8h, v5.8h, v11.8h\n" - "fmla v23.8h, v2.8h, v11.8h\n" - "fmla v25.8h, v8.8h, v13.8h\n" - "fmla v24.8h, v7.8h, v13.8h\n" - "fmax v31.8h, v31.8h, v18.8h\n" - "fmla v23.8h, v6.8h, v13.8h\n" - "fmax v30.8h, v30.8h, v18.8h\n" - "fmin v31.8h, v31.8h, v17.8h\n" - "st1 { v31.8h }, [x13]\n" - "fmin v30.8h, v30.8h, v17.8h\n" - "fmax v29.8h, v29.8h, v18.8h\n" - "str q30, [x13, x14]\n" - "fmin v29.8h, v29.8h, v17.8h\n" - "fmax v28.8h, v28.8h, v18.8h\n" - "str q29, [x13, x22]\n" - "fmin v28.8h, v28.8h, v17.8h\n" - "add x13, x13, #0x10\n" + "ldr q13, [x26, x10]\n" "fmax v27.8h, v27.8h, v18.8h\n" - "st1 { v28.8h }, [x25]\n" - "fmax v26.8h, v26.8h, v18.8h\n" - "fmax v25.8h, v25.8h, v18.8h\n" - "fmin v27.8h, v27.8h, v17.8h\n" - "str q27, [x25, x14]\n" - "fmin v26.8h, v26.8h, v17.8h\n" - "fmin v25.8h, v25.8h, v17.8h\n" - "str q26, [x25, x22]\n" - "fmax v24.8h, v24.8h, v18.8h\n" - "add x25, x25, #0x10\n" + "fmla v23.8h, v6.8h, v12.8h\n" + "fmla v26.8h, v3.8h, v12.8h\n" "fmax v23.8h, v23.8h, v18.8h\n" - "st1 { v25.8h }, [x24]\n" - "fmin v24.8h, v24.8h, v17.8h\n" - "str q24, [x24, x14]\n" + "add x26, x26, #0x10\n" + "fmla v25.8h, v8.8h, v11.8h\n" + "fmla v28.8h, v5.8h, v11.8h\n" + "fmax v25.8h, v25.8h, v18.8h\n" + "fmla v29.8h, v8.8h, v13.8h\n" + "fmla v30.8h, v7.8h, v13.8h\n" + "fmax v26.8h, v26.8h, v18.8h\n" + "fmla v31.8h, v6.8h, v13.8h\n" + "fmax v28.8h, v28.8h, v18.8h\n" + "fmax v29.8h, v29.8h, v18.8h\n" + "fmax v30.8h, v30.8h, v18.8h\n" + "fmax v31.8h, v31.8h, v18.8h\n" "fmin v23.8h, v23.8h, v17.8h\n" - "str q23, [x24, x22]\n" + "st1 { v23.8h }, [x14]\n" + "fmin v25.8h, v25.8h, v17.8h\n" + "fmin v26.8h, v26.8h, v17.8h\n" + "str q24, [x14, x16]\n" + "fmin v27.8h, v27.8h, v17.8h\n" + "fmin v28.8h, v28.8h, v17.8h\n" + "str q25, [x14, x21]\n" + "add x14, x14, #0x10\n" + "fmin v29.8h, v29.8h, v17.8h\n" + "fmin v30.8h, v30.8h, v17.8h\n" + "st1 { v26.8h }, [x27]\n" + "fmin v31.8h, v31.8h, v17.8h\n" + "str q27, [x27, x16]\n" + "str q28, [x27, x21]\n" + "add x27, x27, #0x10\n" + "st1 { v29.8h }, [x24]\n" + "str q30, [x24, x16]\n" + "str q31, [x24, x21]\n" "add x24, x24, #0x10\n" "4:" // Tile loop: Oddments - "tst %x[n_channels], #0x1\n" - "beq 49f\n" - "ldr q16, [x17, #0x0]\n" - "ldr q0, [x17, #0x10]\n" - "add x23, x11, x28\n" - "ldr q1, [x17, #0x20]\n" + "tst %x[n_channels], #0x7\n" + "beq 93f\n" + "ldr q16, [x13, #0x0]\n" + "ldr q0, [x13, #0x10]\n" + "ldr q1, [x13, #0x20]\n" + "ldr q2, [x13, #0x30]\n" + "add x23, x11, x10\n" "add x22, x15, XZR\n" - "ldr q2, [x17, #0x30]\n" - "add x21, x15, x26\n" - "ldr q3, [x17, #0x40]\n" - "add x20, x9, XZR\n" - "ldr q4, [x17, #0x50]\n" - "add x19, x12, x28\n" - "ldr q5, [x17, #0x60]\n" - "ldr q6, [x17, #0x70]\n" - "ldr q7, [x17, #0x80]\n" - "ldr q8, [x17, #0x90]\n" + "ldr q3, [x13, #0x40]\n" + "ldr q4, [x13, #0x50]\n" + "add x21, x15, x25\n" + "add x20, x26, XZR\n" + "ldr q5, [x13, #0x60]\n" + "ldr q6, [x13, #0x70]\n" + "add x19, x12, x10\n" + "ldr q7, [x13, #0x80]\n" + "ldr q8, [x13, #0x90]\n" + "tbz %x[n_channels], #2, 6f\n" + "ldr d9, [x23], #0x8\n" + "ldr d10, [x22], #0x8\n" + "ldr d11, [x21], #0x8\n" + "ldr d12, [x20], #0x8\n" + "ldr d13, [x19], #0x8\n" "tbz %x[n_channels], #1, 5f\n" + "ld1 { v9.s }[2], [x23], #0x4\n" + "ld1 { v10.s }[2], [x22], #0x4\n" + "ld1 { v11.s }[2], [x21], #0x4\n" + "ld1 { v12.s }[2], [x20], #0x4\n" + "ld1 { v13.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 8f\n" + "ld1 { v9.h }[6], [x23]\n" + "ld1 { v10.h }[6], [x22]\n" + "ld1 { v11.h }[6], [x21]\n" + "ld1 { v12.h }[6], [x20]\n" + "ld1 { v13.h }[6], [x19]\n" + "b 8f\n" + "5:" // Tile loop: Oddments: Load inputs: (2, 2), (0, 0), (0, 4), (4, 0), (1, 2): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 8f\n" + "ld1 { v9.h }[4], [x23]\n" + "ld1 { v10.h }[4], [x22]\n" + "ld1 { v11.h }[4], [x21]\n" + "ld1 { v12.h }[4], [x20]\n" + "ld1 { v13.h }[4], [x19]\n" + "b 8f\n" + "6:" // Tile loop: Oddments: Load inputs: (2, 2), (0, 0), (0, 4), (4, 0), (1, 2): Bit 2: Unset + "tbz %x[n_channels], #1, 7f\n" "ldr s9, [x23], #0x4\n" "ldr s10, [x22], #0x4\n" "ldr s11, [x21], #0x4\n" "ldr s12, [x20], #0x4\n" "ldr s13, [x19], #0x4\n" - "tbz %x[n_channels], #0, 6f\n" + "tbz %x[n_channels], #0, 8f\n" "ld1 { v9.h }[2], [x23]\n" "ld1 { v10.h }[2], [x22]\n" "ld1 { v11.h }[2], [x21]\n" "ld1 { v12.h }[2], [x20]\n" "ld1 { v13.h }[2], [x19]\n" - "b 6f\n" - "5:" // Tile loop: Oddments: Load inputs: (2, 2), (0, 0), (0, 4), (4, 0), (1, 2): Bit 1: Unset + "b 8f\n" + "7:" // Tile loop: Oddments: Load inputs: (2, 2), (0, 0), (0, 4), (4, 0), (1, 2): Bit 2: Unset: Bit 1: Unset "ldr h9, [x23, #0x0]\n" "ldr h10, [x22, #0x0]\n" "ldr h11, [x21, #0x0]\n" "ldr h12, [x20, #0x0]\n" "ldr h13, [x19, #0x0]\n" - "6:" // Tile loop: Oddments: Load inputs: (2, 2), (0, 0), (0, 4), (4, 0), (1, 2): Bit 1: End - "mov v31.16b, v16.16b\n fmla v31.8h, v8.8h, v9.8h\n" - "add x19, x9, x26\n" - "mov v30.16b, v16.16b\n fmla v30.8h, v7.8h, v9.8h\n" - "mov v29.16b, v16.16b\n fmla v29.8h, v6.8h, v9.8h\n" - "mov v28.16b, v16.16b\n fmla v28.8h, v5.8h, v9.8h\n" + "8:" // Tile loop: Oddments: Load inputs: (2, 2), (0, 0), (0, 4), (4, 0), (1, 2): Bit 2: End + "mov v23.16b, v16.16b\n fmla v23.8h, v8.8h, v9.8h\n" + "mov v25.16b, v16.16b\n fmla v25.8h, v6.8h, v9.8h\n" + "add x19, x26, x25\n" + "mov v24.16b, v16.16b\n fmla v24.8h, v7.8h, v9.8h\n" + "mov v26.16b, v16.16b\n fmla v26.8h, v5.8h, v9.8h\n" "mov v27.16b, v16.16b\n fmla v27.8h, v4.8h, v9.8h\n" - "mov v26.16b, v16.16b\n fmla v26.8h, v3.8h, v9.8h\n" - "mov v25.16b, v16.16b\n fmla v25.8h, v2.8h, v9.8h\n" - "mov v24.16b, v16.16b\n fmla v24.8h, v1.8h, v9.8h\n" - "mov v23.16b, v16.16b\n fmla v23.8h, v0.8h, v9.8h\n" - "fmla v31.8h, v0.8h, v10.8h\n" - "fmla v29.8h, v2.8h, v11.8h\n" - "fmla v25.8h, v6.8h, v12.8h\n" - "fmla v30.8h, v4.8h, v13.8h\n" - "fmla v31.8h, v5.8h, v13.8h\n" - "fmla v29.8h, v3.8h, v13.8h\n" - "fmla v28.8h, v2.8h, v13.8h\n" + "mov v28.16b, v16.16b\n fmla v28.8h, v3.8h, v9.8h\n" + "mov v29.16b, v16.16b\n fmla v29.8h, v2.8h, v9.8h\n" + "fmla v23.8h, v0.8h, v10.8h\n" + "fmla v25.8h, v2.8h, v11.8h\n" + "mov v30.16b, v16.16b\n fmla v30.8h, v1.8h, v9.8h\n" + "mov v31.16b, v16.16b\n fmla v31.8h, v0.8h, v9.8h\n" + "fmla v29.8h, v6.8h, v12.8h\n" + "fmla v23.8h, v5.8h, v13.8h\n" + "fmla v24.8h, v4.8h, v13.8h\n" + "fmla v25.8h, v3.8h, v13.8h\n" + "fmla v26.8h, v2.8h, v13.8h\n" "fmla v27.8h, v1.8h, v13.8h\n" - "fmla v26.8h, v0.8h, v13.8h\n" - "tbz %x[n_channels], #1, 7f\n" + "fmla v28.8h, v0.8h, v13.8h\n" + "tbz %x[n_channels], #2, 10f\n" + "ldr d12, [x19], #0x8\n" + "tbz %x[n_channels], #1, 9f\n" + "ld1 { v12.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 12f\n" + "ld1 { v12.h }[6], [x19]\n" + "b 12f\n" + "9:" // Tile loop: Oddments: Load inputs: (4, 4): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 12f\n" + "ld1 { v12.h }[4], [x19]\n" + "b 12f\n" + "10:" // Tile loop: Oddments: Load inputs: (4, 4): Bit 2: Unset + "tbz %x[n_channels], #1, 11f\n" "ldr s12, [x19], #0x4\n" - "tbz %x[n_channels], #0, 8f\n" + "tbz %x[n_channels], #0, 12f\n" "ld1 { v12.h }[2], [x19]\n" - "b 8f\n" - "7:" // Tile loop: Oddments: Load inputs: (4, 4): Bit 1: Unset + "b 12f\n" + "11:" // Tile loop: Oddments: Load inputs: (4, 4): Bit 2: Unset: Bit 1: Unset "ldr h12, [x19, #0x0]\n" - "8:" // Tile loop: Oddments: Load inputs: (4, 4): Bit 1: End - "fmla v23.8h, v8.8h, v12.8h\n" - "add x19, x11, x16\n" - "tbz %x[n_channels], #1, 9f\n" + "12:" // Tile loop: Oddments: Load inputs: (4, 4): Bit 2: End + "fmla v31.8h, v8.8h, v12.8h\n" + "add x19, x11, x17\n" + "tbz %x[n_channels], #2, 14f\n" + "ldr d11, [x19], #0x8\n" + "tbz %x[n_channels], #1, 13f\n" + "ld1 { v11.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 16f\n" + "ld1 { v11.h }[6], [x19]\n" + "b 16f\n" + "13:" // Tile loop: Oddments: Load inputs: (2, 1): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 16f\n" + "ld1 { v11.h }[4], [x19]\n" + "b 16f\n" + "14:" // Tile loop: Oddments: Load inputs: (2, 1): Bit 2: Unset + "tbz %x[n_channels], #1, 15f\n" "ldr s11, [x19], #0x4\n" - "tbz %x[n_channels], #0, 10f\n" + "tbz %x[n_channels], #0, 16f\n" "ld1 { v11.h }[2], [x19]\n" - "b 10f\n" - "9:" // Tile loop: Oddments: Load inputs: (2, 1): Bit 1: Unset + "b 16f\n" + "15:" // Tile loop: Oddments: Load inputs: (2, 1): Bit 2: Unset: Bit 1: Unset "ldr h11, [x19, #0x0]\n" - "10:" // Tile loop: Oddments: Load inputs: (2, 1): Bit 1: End - "fmla v31.8h, v7.8h, v11.8h\n" - "add x19, x15, x16\n" - "fmla v30.8h, v6.8h, v11.8h\n" - "fmla v28.8h, v4.8h, v11.8h\n" + "16:" // Tile loop: Oddments: Load inputs: (2, 1): Bit 2: End + "fmla v23.8h, v7.8h, v11.8h\n" + "fmla v24.8h, v6.8h, v11.8h\n" + "add x19, x15, x17\n" + "fmla v26.8h, v4.8h, v11.8h\n" "fmla v27.8h, v3.8h, v11.8h\n" - "fmla v25.8h, v1.8h, v11.8h\n" - "fmla v24.8h, v0.8h, v11.8h\n" - "tbz %x[n_channels], #1, 11f\n" + "fmla v29.8h, v1.8h, v11.8h\n" + "fmla v30.8h, v0.8h, v11.8h\n" + "tbz %x[n_channels], #2, 18f\n" + "ldr d13, [x19], #0x8\n" + "tbz %x[n_channels], #1, 17f\n" + "ld1 { v13.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 20f\n" + "ld1 { v13.h }[6], [x19]\n" + "b 20f\n" + "17:" // Tile loop: Oddments: Load inputs: (0, 1): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 20f\n" + "ld1 { v13.h }[4], [x19]\n" + "b 20f\n" + "18:" // Tile loop: Oddments: Load inputs: (0, 1): Bit 2: Unset + "tbz %x[n_channels], #1, 19f\n" "ldr s13, [x19], #0x4\n" - "tbz %x[n_channels], #0, 12f\n" + "tbz %x[n_channels], #0, 20f\n" "ld1 { v13.h }[2], [x19]\n" - "b 12f\n" - "11:" // Tile loop: Oddments: Load inputs: (0, 1): Bit 1: Unset + "b 20f\n" + "19:" // Tile loop: Oddments: Load inputs: (0, 1): Bit 2: Unset: Bit 1: Unset "ldr h13, [x19, #0x0]\n" - "12:" // Tile loop: Oddments: Load inputs: (0, 1): Bit 1: End - "fmla v31.8h, v1.8h, v13.8h\n" - "add x19, x15, x27\n" - "fmla v30.8h, v0.8h, v13.8h\n" - "tbz %x[n_channels], #1, 13f\n" + "20:" // Tile loop: Oddments: Load inputs: (0, 1): Bit 2: End + "fmla v23.8h, v1.8h, v13.8h\n" + "fmla v24.8h, v0.8h, v13.8h\n" + "add x19, x15, x28\n" + "tbz %x[n_channels], #2, 22f\n" + "ldr d12, [x19], #0x8\n" + "tbz %x[n_channels], #1, 21f\n" + "ld1 { v12.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 24f\n" + "ld1 { v12.h }[6], [x19]\n" + "b 24f\n" + "21:" // Tile loop: Oddments: Load inputs: (0, 3): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 24f\n" + "ld1 { v12.h }[4], [x19]\n" + "b 24f\n" + "22:" // Tile loop: Oddments: Load inputs: (0, 3): Bit 2: Unset + "tbz %x[n_channels], #1, 23f\n" "ldr s12, [x19], #0x4\n" - "tbz %x[n_channels], #0, 14f\n" + "tbz %x[n_channels], #0, 24f\n" "ld1 { v12.h }[2], [x19]\n" - "b 14f\n" - "13:" // Tile loop: Oddments: Load inputs: (0, 3): Bit 1: Unset + "b 24f\n" + "23:" // Tile loop: Oddments: Load inputs: (0, 3): Bit 2: Unset: Bit 1: Unset "ldr h12, [x19, #0x0]\n" - "14:" // Tile loop: Oddments: Load inputs: (0, 3): Bit 1: End - "fmla v30.8h, v2.8h, v12.8h\n" - "add x19, x11, x27\n" - "fmla v29.8h, v1.8h, v12.8h\n" - "tbz %x[n_channels], #1, 15f\n" + "24:" // Tile loop: Oddments: Load inputs: (0, 3): Bit 2: End + "fmla v24.8h, v2.8h, v12.8h\n" + "fmla v25.8h, v1.8h, v12.8h\n" + "add x19, x11, x28\n" + "tbz %x[n_channels], #2, 26f\n" + "ldr d10, [x19], #0x8\n" + "tbz %x[n_channels], #1, 25f\n" + "ld1 { v10.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 28f\n" + "ld1 { v10.h }[6], [x19]\n" + "b 28f\n" + "25:" // Tile loop: Oddments: Load inputs: (2, 3): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 28f\n" + "ld1 { v10.h }[4], [x19]\n" + "b 28f\n" + "26:" // Tile loop: Oddments: Load inputs: (2, 3): Bit 2: Unset + "tbz %x[n_channels], #1, 27f\n" "ldr s10, [x19], #0x4\n" - "tbz %x[n_channels], #0, 16f\n" + "tbz %x[n_channels], #0, 28f\n" "ld1 { v10.h }[2], [x19]\n" - "b 16f\n" - "15:" // Tile loop: Oddments: Load inputs: (2, 3): Bit 1: Unset + "b 28f\n" + "27:" // Tile loop: Oddments: Load inputs: (2, 3): Bit 2: Unset: Bit 1: Unset "ldr h10, [x19, #0x0]\n" - "16:" // Tile loop: Oddments: Load inputs: (2, 3): Bit 1: End - "fmla v30.8h, v8.8h, v10.8h\n" + "28:" // Tile loop: Oddments: Load inputs: (2, 3): Bit 2: End + "fmla v24.8h, v8.8h, v10.8h\n" + "fmla v25.8h, v7.8h, v10.8h\n" "add x19, x12, XZR\n" - "fmla v29.8h, v7.8h, v10.8h\n" "fmla v27.8h, v5.8h, v10.8h\n" - "fmla v26.8h, v4.8h, v10.8h\n" - "fmla v24.8h, v2.8h, v10.8h\n" - "fmla v23.8h, v1.8h, v10.8h\n" - "tbz %x[n_channels], #1, 17f\n" + "fmla v28.8h, v4.8h, v10.8h\n" + "fmla v30.8h, v2.8h, v10.8h\n" + "fmla v31.8h, v1.8h, v10.8h\n" + "tbz %x[n_channels], #2, 30f\n" + "ldr d11, [x19], #0x8\n" + "tbz %x[n_channels], #1, 29f\n" + "ld1 { v11.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 32f\n" + "ld1 { v11.h }[6], [x19]\n" + "b 32f\n" + "29:" // Tile loop: Oddments: Load inputs: (1, 0): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 32f\n" + "ld1 { v11.h }[4], [x19]\n" + "b 32f\n" + "30:" // Tile loop: Oddments: Load inputs: (1, 0): Bit 2: Unset + "tbz %x[n_channels], #1, 31f\n" "ldr s11, [x19], #0x4\n" - "tbz %x[n_channels], #0, 18f\n" + "tbz %x[n_channels], #0, 32f\n" "ld1 { v11.h }[2], [x19]\n" - "b 18f\n" - "17:" // Tile loop: Oddments: Load inputs: (1, 0): Bit 1: Unset + "b 32f\n" + "31:" // Tile loop: Oddments: Load inputs: (1, 0): Bit 2: Unset: Bit 1: Unset "ldr h11, [x19, #0x0]\n" - "18:" // Tile loop: Oddments: Load inputs: (1, 0): Bit 1: End - "fmla v31.8h, v3.8h, v11.8h\n" - "add x19, x12, x26\n" - "fmla v28.8h, v0.8h, v11.8h\n" - "tbz %x[n_channels], #1, 19f\n" + "32:" // Tile loop: Oddments: Load inputs: (1, 0): Bit 2: End + "fmla v23.8h, v3.8h, v11.8h\n" + "fmla v26.8h, v0.8h, v11.8h\n" + "add x19, x12, x25\n" + "tbz %x[n_channels], #2, 34f\n" + "ldr d13, [x19], #0x8\n" + "tbz %x[n_channels], #1, 33f\n" + "ld1 { v13.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 36f\n" + "ld1 { v13.h }[6], [x19]\n" + "b 36f\n" + "33:" // Tile loop: Oddments: Load inputs: (1, 4): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 36f\n" + "ld1 { v13.h }[4], [x19]\n" + "b 36f\n" + "34:" // Tile loop: Oddments: Load inputs: (1, 4): Bit 2: Unset + "tbz %x[n_channels], #1, 35f\n" "ldr s13, [x19], #0x4\n" - "tbz %x[n_channels], #0, 20f\n" + "tbz %x[n_channels], #0, 36f\n" "ld1 { v13.h }[2], [x19]\n" - "b 20f\n" - "19:" // Tile loop: Oddments: Load inputs: (1, 4): Bit 1: Unset + "b 36f\n" + "35:" // Tile loop: Oddments: Load inputs: (1, 4): Bit 2: Unset: Bit 1: Unset "ldr h13, [x19, #0x0]\n" - "20:" // Tile loop: Oddments: Load inputs: (1, 4): Bit 1: End - "fmla v29.8h, v5.8h, v13.8h\n" - "add x19, x10, XZR\n" - "fmla v26.8h, v2.8h, v13.8h\n" - "tbz %x[n_channels], #1, 21f\n" + "36:" // Tile loop: Oddments: Load inputs: (1, 4): Bit 2: End + "fmla v25.8h, v5.8h, v13.8h\n" + "fmla v28.8h, v2.8h, v13.8h\n" + "add x19, x9, XZR\n" + "tbz %x[n_channels], #2, 38f\n" + "ldr d12, [x19], #0x8\n" + "tbz %x[n_channels], #1, 37f\n" + "ld1 { v12.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 40f\n" + "ld1 { v12.h }[6], [x19]\n" + "b 40f\n" + "37:" // Tile loop: Oddments: Load inputs: (3, 0): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 40f\n" + "ld1 { v12.h }[4], [x19]\n" + "b 40f\n" + "38:" // Tile loop: Oddments: Load inputs: (3, 0): Bit 2: Unset + "tbz %x[n_channels], #1, 39f\n" "ldr s12, [x19], #0x4\n" - "tbz %x[n_channels], #0, 22f\n" + "tbz %x[n_channels], #0, 40f\n" "ld1 { v12.h }[2], [x19]\n" - "b 22f\n" - "21:" // Tile loop: Oddments: Load inputs: (3, 0): Bit 1: Unset + "b 40f\n" + "39:" // Tile loop: Oddments: Load inputs: (3, 0): Bit 2: Unset: Bit 1: Unset "ldr h12, [x19, #0x0]\n" - "22:" // Tile loop: Oddments: Load inputs: (3, 0): Bit 1: End - "fmla v28.8h, v6.8h, v12.8h\n" - "add x19, x10, x28\n" - "fmla v25.8h, v3.8h, v12.8h\n" - "tbz %x[n_channels], #1, 23f\n" + "40:" // Tile loop: Oddments: Load inputs: (3, 0): Bit 2: End + "fmla v26.8h, v6.8h, v12.8h\n" + "fmla v29.8h, v3.8h, v12.8h\n" + "add x19, x9, x10\n" + "tbz %x[n_channels], #2, 42f\n" + "ldr d10, [x19], #0x8\n" + "tbz %x[n_channels], #1, 41f\n" + "ld1 { v10.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 44f\n" + "ld1 { v10.h }[6], [x19]\n" + "b 44f\n" + "41:" // Tile loop: Oddments: Load inputs: (3, 2): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 44f\n" + "ld1 { v10.h }[4], [x19]\n" + "b 44f\n" + "42:" // Tile loop: Oddments: Load inputs: (3, 2): Bit 2: Unset + "tbz %x[n_channels], #1, 43f\n" "ldr s10, [x19], #0x4\n" - "tbz %x[n_channels], #0, 24f\n" + "tbz %x[n_channels], #0, 44f\n" "ld1 { v10.h }[2], [x19]\n" - "b 24f\n" - "23:" // Tile loop: Oddments: Load inputs: (3, 2): Bit 1: Unset + "b 44f\n" + "43:" // Tile loop: Oddments: Load inputs: (3, 2): Bit 2: Unset: Bit 1: Unset "ldr h10, [x19, #0x0]\n" - "24:" // Tile loop: Oddments: Load inputs: (3, 2): Bit 1: End - "fmla v28.8h, v8.8h, v10.8h\n" - "add x19, x10, x26\n" + "44:" // Tile loop: Oddments: Load inputs: (3, 2): Bit 2: End + "fmla v26.8h, v8.8h, v10.8h\n" "fmla v27.8h, v7.8h, v10.8h\n" - "fmla v26.8h, v6.8h, v10.8h\n" - "fmla v25.8h, v5.8h, v10.8h\n" - "fmla v24.8h, v4.8h, v10.8h\n" - "fmla v23.8h, v3.8h, v10.8h\n" - "tbz %x[n_channels], #1, 25f\n" + "add x19, x9, x25\n" + "fmla v28.8h, v6.8h, v10.8h\n" + "fmla v29.8h, v5.8h, v10.8h\n" + "fmla v30.8h, v4.8h, v10.8h\n" + "fmla v31.8h, v3.8h, v10.8h\n" + "tbz %x[n_channels], #2, 46f\n" + "ldr d11, [x19], #0x8\n" + "tbz %x[n_channels], #1, 45f\n" + "ld1 { v11.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 48f\n" + "ld1 { v11.h }[6], [x19]\n" + "b 48f\n" + "45:" // Tile loop: Oddments: Load inputs: (3, 4): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 48f\n" + "ld1 { v11.h }[4], [x19]\n" + "b 48f\n" + "46:" // Tile loop: Oddments: Load inputs: (3, 4): Bit 2: Unset + "tbz %x[n_channels], #1, 47f\n" "ldr s11, [x19], #0x4\n" - "tbz %x[n_channels], #0, 26f\n" + "tbz %x[n_channels], #0, 48f\n" "ld1 { v11.h }[2], [x19]\n" - "b 26f\n" - "25:" // Tile loop: Oddments: Load inputs: (3, 4): Bit 1: Unset + "b 48f\n" + "47:" // Tile loop: Oddments: Load inputs: (3, 4): Bit 2: Unset: Bit 1: Unset "ldr h11, [x19, #0x0]\n" - "26:" // Tile loop: Oddments: Load inputs: (3, 4): Bit 1: End - "fmla v26.8h, v8.8h, v11.8h\n" - "add x19, x9, x16\n" - "fmla v23.8h, v5.8h, v11.8h\n" - "tbz %x[n_channels], #1, 27f\n" + "48:" // Tile loop: Oddments: Load inputs: (3, 4): Bit 2: End + "fmla v28.8h, v8.8h, v11.8h\n" + "fmla v31.8h, v5.8h, v11.8h\n" + "add x19, x26, x17\n" + "tbz %x[n_channels], #2, 50f\n" + "ldr d13, [x19], #0x8\n" + "tbz %x[n_channels], #1, 49f\n" + "ld1 { v13.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 52f\n" + "ld1 { v13.h }[6], [x19]\n" + "b 52f\n" + "49:" // Tile loop: Oddments: Load inputs: (4, 1): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 52f\n" + "ld1 { v13.h }[4], [x19]\n" + "b 52f\n" + "50:" // Tile loop: Oddments: Load inputs: (4, 1): Bit 2: Unset + "tbz %x[n_channels], #1, 51f\n" "ldr s13, [x19], #0x4\n" - "tbz %x[n_channels], #0, 28f\n" + "tbz %x[n_channels], #0, 52f\n" "ld1 { v13.h }[2], [x19]\n" - "b 28f\n" - "27:" // Tile loop: Oddments: Load inputs: (4, 1): Bit 1: Unset + "b 52f\n" + "51:" // Tile loop: Oddments: Load inputs: (4, 1): Bit 2: Unset: Bit 1: Unset "ldr h13, [x19, #0x0]\n" - "28:" // Tile loop: Oddments: Load inputs: (4, 1): Bit 1: End - "fmla v25.8h, v7.8h, v13.8h\n" - "add x19, x12, x16\n" - "fmla v24.8h, v6.8h, v13.8h\n" - "tbz %x[n_channels], #1, 29f\n" + "52:" // Tile loop: Oddments: Load inputs: (4, 1): Bit 2: End + "fmla v29.8h, v7.8h, v13.8h\n" + "fmla v30.8h, v6.8h, v13.8h\n" + "add x19, x12, x17\n" + "tbz %x[n_channels], #2, 54f\n" + "ldr d12, [x19], #0x8\n" + "tbz %x[n_channels], #1, 53f\n" + "ld1 { v12.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 56f\n" + "ld1 { v12.h }[6], [x19]\n" + "b 56f\n" + "53:" // Tile loop: Oddments: Load inputs: (1, 1): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 56f\n" + "ld1 { v12.h }[4], [x19]\n" + "b 56f\n" + "54:" // Tile loop: Oddments: Load inputs: (1, 1): Bit 2: Unset + "tbz %x[n_channels], #1, 55f\n" "ldr s12, [x19], #0x4\n" - "tbz %x[n_channels], #0, 30f\n" + "tbz %x[n_channels], #0, 56f\n" "ld1 { v12.h }[2], [x19]\n" - "b 30f\n" - "29:" // Tile loop: Oddments: Load inputs: (1, 1): Bit 1: Unset + "b 56f\n" + "55:" // Tile loop: Oddments: Load inputs: (1, 1): Bit 2: Unset: Bit 1: Unset "ldr h12, [x19, #0x0]\n" - "30:" // Tile loop: Oddments: Load inputs: (1, 1): Bit 1: End - "fmla v31.8h, v4.8h, v12.8h\n" - "add x19, x12, x27\n" - "fmla v30.8h, v3.8h, v12.8h\n" - "fmla v28.8h, v1.8h, v12.8h\n" + "56:" // Tile loop: Oddments: Load inputs: (1, 1): Bit 2: End + "fmla v23.8h, v4.8h, v12.8h\n" + "fmla v24.8h, v3.8h, v12.8h\n" + "add x19, x12, x28\n" + "fmla v26.8h, v1.8h, v12.8h\n" "fmla v27.8h, v0.8h, v12.8h\n" - "tbz %x[n_channels], #1, 31f\n" + "tbz %x[n_channels], #2, 58f\n" + "ldr d11, [x19], #0x8\n" + "tbz %x[n_channels], #1, 57f\n" + "ld1 { v11.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 60f\n" + "ld1 { v11.h }[6], [x19]\n" + "b 60f\n" + "57:" // Tile loop: Oddments: Load inputs: (1, 3): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 60f\n" + "ld1 { v11.h }[4], [x19]\n" + "b 60f\n" + "58:" // Tile loop: Oddments: Load inputs: (1, 3): Bit 2: Unset + "tbz %x[n_channels], #1, 59f\n" "ldr s11, [x19], #0x4\n" - "tbz %x[n_channels], #0, 32f\n" + "tbz %x[n_channels], #0, 60f\n" "ld1 { v11.h }[2], [x19]\n" - "b 32f\n" - "31:" // Tile loop: Oddments: Load inputs: (1, 3): Bit 1: Unset + "b 60f\n" + "59:" // Tile loop: Oddments: Load inputs: (1, 3): Bit 2: Unset: Bit 1: Unset "ldr h11, [x19, #0x0]\n" - "32:" // Tile loop: Oddments: Load inputs: (1, 3): Bit 1: End - "fmla v30.8h, v5.8h, v11.8h\n" - "add x19, x9, x27\n" - "fmla v29.8h, v4.8h, v11.8h\n" + "60:" // Tile loop: Oddments: Load inputs: (1, 3): Bit 2: End + "fmla v24.8h, v5.8h, v11.8h\n" + "fmla v25.8h, v4.8h, v11.8h\n" + "add x19, x26, x28\n" "fmla v27.8h, v2.8h, v11.8h\n" - "fmla v26.8h, v1.8h, v11.8h\n" - "tbz %x[n_channels], #1, 33f\n" + "fmla v28.8h, v1.8h, v11.8h\n" + "tbz %x[n_channels], #2, 62f\n" + "ldr d13, [x19], #0x8\n" + "tbz %x[n_channels], #1, 61f\n" + "ld1 { v13.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 64f\n" + "ld1 { v13.h }[6], [x19]\n" + "b 64f\n" + "61:" // Tile loop: Oddments: Load inputs: (4, 3): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 64f\n" + "ld1 { v13.h }[4], [x19]\n" + "b 64f\n" + "62:" // Tile loop: Oddments: Load inputs: (4, 3): Bit 2: Unset + "tbz %x[n_channels], #1, 63f\n" "ldr s13, [x19], #0x4\n" - "tbz %x[n_channels], #0, 34f\n" + "tbz %x[n_channels], #0, 64f\n" "ld1 { v13.h }[2], [x19]\n" - "b 34f\n" - "33:" // Tile loop: Oddments: Load inputs: (4, 3): Bit 1: Unset + "b 64f\n" + "63:" // Tile loop: Oddments: Load inputs: (4, 3): Bit 2: Unset: Bit 1: Unset "ldr h13, [x19, #0x0]\n" - "34:" // Tile loop: Oddments: Load inputs: (4, 3): Bit 1: End - "fmla v24.8h, v8.8h, v13.8h\n" - "add x19, x10, x16\n" - "fmla v23.8h, v7.8h, v13.8h\n" - "tbz %x[n_channels], #1, 35f\n" + "64:" // Tile loop: Oddments: Load inputs: (4, 3): Bit 2: End + "fmla v30.8h, v8.8h, v13.8h\n" + "fmla v31.8h, v7.8h, v13.8h\n" + "add x19, x9, x17\n" + "tbz %x[n_channels], #2, 66f\n" + "ldr d12, [x19], #0x8\n" + "tbz %x[n_channels], #1, 65f\n" + "ld1 { v12.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 68f\n" + "ld1 { v12.h }[6], [x19]\n" + "b 68f\n" + "65:" // Tile loop: Oddments: Load inputs: (3, 1): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 68f\n" + "ld1 { v12.h }[4], [x19]\n" + "b 68f\n" + "66:" // Tile loop: Oddments: Load inputs: (3, 1): Bit 2: Unset + "tbz %x[n_channels], #1, 67f\n" "ldr s12, [x19], #0x4\n" - "tbz %x[n_channels], #0, 36f\n" + "tbz %x[n_channels], #0, 68f\n" "ld1 { v12.h }[2], [x19]\n" - "b 36f\n" - "35:" // Tile loop: Oddments: Load inputs: (3, 1): Bit 1: Unset + "b 68f\n" + "67:" // Tile loop: Oddments: Load inputs: (3, 1): Bit 2: Unset: Bit 1: Unset "ldr h12, [x19, #0x0]\n" - "36:" // Tile loop: Oddments: Load inputs: (3, 1): Bit 1: End - "fmla v28.8h, v7.8h, v12.8h\n" - "add x19, x15, x28\n" + "68:" // Tile loop: Oddments: Load inputs: (3, 1): Bit 2: End + "fmla v26.8h, v7.8h, v12.8h\n" "fmla v27.8h, v6.8h, v12.8h\n" - "fmla v25.8h, v4.8h, v12.8h\n" - "fmla v24.8h, v3.8h, v12.8h\n" - "tbz %x[n_channels], #1, 37f\n" + "add x19, x15, x10\n" + "fmla v29.8h, v4.8h, v12.8h\n" + "fmla v30.8h, v3.8h, v12.8h\n" + "tbz %x[n_channels], #2, 70f\n" + "ldr d11, [x19], #0x8\n" + "tbz %x[n_channels], #1, 69f\n" + "ld1 { v11.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 72f\n" + "ld1 { v11.h }[6], [x19]\n" + "b 72f\n" + "69:" // Tile loop: Oddments: Load inputs: (0, 2): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 72f\n" + "ld1 { v11.h }[4], [x19]\n" + "b 72f\n" + "70:" // Tile loop: Oddments: Load inputs: (0, 2): Bit 2: Unset + "tbz %x[n_channels], #1, 71f\n" "ldr s11, [x19], #0x4\n" - "tbz %x[n_channels], #0, 38f\n" + "tbz %x[n_channels], #0, 72f\n" "ld1 { v11.h }[2], [x19]\n" - "b 38f\n" - "37:" // Tile loop: Oddments: Load inputs: (0, 2): Bit 1: Unset + "b 72f\n" + "71:" // Tile loop: Oddments: Load inputs: (0, 2): Bit 2: Unset: Bit 1: Unset "ldr h11, [x19, #0x0]\n" - "38:" // Tile loop: Oddments: Load inputs: (0, 2): Bit 1: End - "fmla v31.8h, v2.8h, v11.8h\n" - "add x19, x10, x27\n" - "fmla v30.8h, v1.8h, v11.8h\n" - "fmla v29.8h, v0.8h, v11.8h\n" - "tbz %x[n_channels], #1, 39f\n" + "72:" // Tile loop: Oddments: Load inputs: (0, 2): Bit 2: End + "fmla v23.8h, v2.8h, v11.8h\n" + "fmla v24.8h, v1.8h, v11.8h\n" + "add x19, x9, x28\n" + "fmla v25.8h, v0.8h, v11.8h\n" + "tbz %x[n_channels], #2, 74f\n" + "ldr d13, [x19], #0x8\n" + "tbz %x[n_channels], #1, 73f\n" + "ld1 { v13.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 76f\n" + "ld1 { v13.h }[6], [x19]\n" + "b 76f\n" + "73:" // Tile loop: Oddments: Load inputs: (3, 3): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 76f\n" + "ld1 { v13.h }[4], [x19]\n" + "b 76f\n" + "74:" // Tile loop: Oddments: Load inputs: (3, 3): Bit 2: Unset + "tbz %x[n_channels], #1, 75f\n" "ldr s13, [x19], #0x4\n" - "tbz %x[n_channels], #0, 40f\n" + "tbz %x[n_channels], #0, 76f\n" "ld1 { v13.h }[2], [x19]\n" - "b 40f\n" - "39:" // Tile loop: Oddments: Load inputs: (3, 3): Bit 1: Unset + "b 76f\n" + "75:" // Tile loop: Oddments: Load inputs: (3, 3): Bit 2: Unset: Bit 1: Unset "ldr h13, [x19, #0x0]\n" - "40:" // Tile loop: Oddments: Load inputs: (3, 3): Bit 1: End + "76:" // Tile loop: Oddments: Load inputs: (3, 3): Bit 2: End "fmla v27.8h, v8.8h, v13.8h\n" + "fmla v28.8h, v7.8h, v13.8h\n" "add x19, x11, XZR\n" - "fmla v26.8h, v7.8h, v13.8h\n" - "fmla v24.8h, v5.8h, v13.8h\n" - "fmla v23.8h, v4.8h, v13.8h\n" - "tbz %x[n_channels], #1, 41f\n" + "fmla v30.8h, v5.8h, v13.8h\n" + "fmla v31.8h, v4.8h, v13.8h\n" + "tbz %x[n_channels], #2, 78f\n" + "ldr d12, [x19], #0x8\n" + "tbz %x[n_channels], #1, 77f\n" + "ld1 { v12.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 80f\n" + "ld1 { v12.h }[6], [x19]\n" + "b 80f\n" + "77:" // Tile loop: Oddments: Load inputs: (2, 0): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 80f\n" + "ld1 { v12.h }[4], [x19]\n" + "b 80f\n" + "78:" // Tile loop: Oddments: Load inputs: (2, 0): Bit 2: Unset + "tbz %x[n_channels], #1, 79f\n" "ldr s12, [x19], #0x4\n" - "tbz %x[n_channels], #0, 42f\n" + "tbz %x[n_channels], #0, 80f\n" "ld1 { v12.h }[2], [x19]\n" - "b 42f\n" - "41:" // Tile loop: Oddments: Load inputs: (2, 0): Bit 1: Unset + "b 80f\n" + "79:" // Tile loop: Oddments: Load inputs: (2, 0): Bit 2: Unset: Bit 1: Unset "ldr h12, [x19, #0x0]\n" - "42:" // Tile loop: Oddments: Load inputs: (2, 0): Bit 1: End - "fmla v31.8h, v6.8h, v12.8h\n" - "add x19, x11, x26\n" - "fmla v28.8h, v3.8h, v12.8h\n" - "fmla v25.8h, v0.8h, v12.8h\n" - "tbz %x[n_channels], #1, 43f\n" + "80:" // Tile loop: Oddments: Load inputs: (2, 0): Bit 2: End + "fmla v23.8h, v6.8h, v12.8h\n" + "fmla v26.8h, v3.8h, v12.8h\n" + "add x19, x11, x25\n" + "fmla v29.8h, v0.8h, v12.8h\n" + "tbz %x[n_channels], #2, 82f\n" + "ldr d11, [x19], #0x8\n" + "tbz %x[n_channels], #1, 81f\n" + "ld1 { v11.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 84f\n" + "ld1 { v11.h }[6], [x19]\n" + "b 84f\n" + "81:" // Tile loop: Oddments: Load inputs: (2, 4): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 84f\n" + "ld1 { v11.h }[4], [x19]\n" + "b 84f\n" + "82:" // Tile loop: Oddments: Load inputs: (2, 4): Bit 2: Unset + "tbz %x[n_channels], #1, 83f\n" "ldr s11, [x19], #0x4\n" - "tbz %x[n_channels], #0, 44f\n" + "tbz %x[n_channels], #0, 84f\n" "ld1 { v11.h }[2], [x19]\n" - "b 44f\n" - "43:" // Tile loop: Oddments: Load inputs: (2, 4): Bit 1: Unset + "b 84f\n" + "83:" // Tile loop: Oddments: Load inputs: (2, 4): Bit 2: Unset: Bit 1: Unset "ldr h11, [x19, #0x0]\n" - "44:" // Tile loop: Oddments: Load inputs: (2, 4): Bit 1: End - "fmla v29.8h, v8.8h, v11.8h\n" - "add x19, x9, x28\n" - "fmla v26.8h, v5.8h, v11.8h\n" - "fmla v23.8h, v2.8h, v11.8h\n" - "tbz %x[n_channels], #1, 45f\n" + "84:" // Tile loop: Oddments: Load inputs: (2, 4): Bit 2: End + "fmla v25.8h, v8.8h, v11.8h\n" + "fmla v28.8h, v5.8h, v11.8h\n" + "add x19, x26, x10\n" + "fmla v31.8h, v2.8h, v11.8h\n" + "tbz %x[n_channels], #2, 86f\n" + "ldr d13, [x19], #0x8\n" + "tbz %x[n_channels], #1, 85f\n" + "ld1 { v13.s }[2], [x19], #0x4\n" + "tbz %x[n_channels], #0, 88f\n" + "ld1 { v13.h }[6], [x19]\n" + "b 88f\n" + "85:" // Tile loop: Oddments: Load inputs: (4, 2): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 88f\n" + "ld1 { v13.h }[4], [x19]\n" + "b 88f\n" + "86:" // Tile loop: Oddments: Load inputs: (4, 2): Bit 2: Unset + "tbz %x[n_channels], #1, 87f\n" "ldr s13, [x19], #0x4\n" - "tbz %x[n_channels], #0, 46f\n" + "tbz %x[n_channels], #0, 88f\n" "ld1 { v13.h }[2], [x19]\n" - "b 46f\n" - "45:" // Tile loop: Oddments: Load inputs: (4, 2): Bit 1: Unset + "b 88f\n" + "87:" // Tile loop: Oddments: Load inputs: (4, 2): Bit 2: Unset: Bit 1: Unset "ldr h13, [x19, #0x0]\n" - "46:" // Tile loop: Oddments: Load inputs: (4, 2): Bit 1: End - "fmla v25.8h, v8.8h, v13.8h\n" - "fmla v24.8h, v7.8h, v13.8h\n" - "fmla v23.8h, v6.8h, v13.8h\n" - "fmax v31.8h, v31.8h, v18.8h\n" - "fmax v30.8h, v30.8h, v18.8h\n" - "fmax v29.8h, v29.8h, v18.8h\n" - "fmin v31.8h, v31.8h, v17.8h\n" - "fmin v30.8h, v30.8h, v17.8h\n" - "fmin v29.8h, v29.8h, v17.8h\n" - "fmax v28.8h, v28.8h, v18.8h\n" - "fmax v27.8h, v27.8h, v18.8h\n" - "fmax v26.8h, v26.8h, v18.8h\n" - "fmin v28.8h, v28.8h, v17.8h\n" - "fmin v27.8h, v27.8h, v17.8h\n" - "fmin v26.8h, v26.8h, v17.8h\n" - "fmax v25.8h, v25.8h, v18.8h\n" - "fmax v24.8h, v24.8h, v18.8h\n" + "88:" // Tile loop: Oddments: Load inputs: (4, 2): Bit 2: End + "fmla v29.8h, v8.8h, v13.8h\n" + "fmla v30.8h, v7.8h, v13.8h\n" "fmax v23.8h, v23.8h, v18.8h\n" - "fmin v25.8h, v25.8h, v17.8h\n" - "fmin v24.8h, v24.8h, v17.8h\n" + "fmla v31.8h, v6.8h, v13.8h\n" + "fmax v24.8h, v24.8h, v18.8h\n" + "fmax v25.8h, v25.8h, v18.8h\n" + "fmax v26.8h, v26.8h, v18.8h\n" + "fmax v27.8h, v27.8h, v18.8h\n" + "fmax v28.8h, v28.8h, v18.8h\n" + "fmax v29.8h, v29.8h, v18.8h\n" + "fmax v30.8h, v30.8h, v18.8h\n" + "fmax v31.8h, v31.8h, v18.8h\n" "fmin v23.8h, v23.8h, v17.8h\n" - "tbz %x[n_channels], #1, 47f\n" - "mov x19, x13\n" - "st1 { v31.s }[0], [x19], x14\n" - "add x13, x13, #0x4\n" - "st1 { v30.s }[0], [x19], x14\n" - "mov x20, x25\n" - "st1 { v29.s }[0], [x19]\n" - "st1 { v28.s }[0], [x20], x14\n" - "add x25, x25, #0x4\n" - "st1 { v27.s }[0], [x20], x14\n" + "fmin v24.8h, v24.8h, v17.8h\n" + "fmin v25.8h, v25.8h, v17.8h\n" + "fmin v26.8h, v26.8h, v17.8h\n" + "fmin v27.8h, v27.8h, v17.8h\n" + "fmin v28.8h, v28.8h, v17.8h\n" + "fmin v29.8h, v29.8h, v17.8h\n" + "fmin v30.8h, v30.8h, v17.8h\n" + "fmin v31.8h, v31.8h, v17.8h\n" + "tbz %x[n_channels], #2, 90f\n" + "mov x21, x14\n" + "mov x20, x27\n" + "mov x19, x24\n" + "st1 { v23.d }[0], [x21], x16\n" + "st1 { v26.d }[0], [x20], x16\n" + "add x14, x14, #0x8\n" + "add x27, x27, #0x8\n" + "st1 { v29.d }[0], [x19], x16\n" + "add x24, x24, #0x8\n" + "st1 { v24.d }[0], [x21], x16\n" + "st1 { v27.d }[0], [x20], x16\n" + "st1 { v30.d }[0], [x19], x16\n" + "st1 { v25.d }[0], [x21]\n" + "st1 { v28.d }[0], [x20]\n" + "st1 { v31.d }[0], [x19]\n" + "tbz %x[n_channels], #1, 89f\n" + "mov x21, x14\n" + "mov x20, x27\n" "mov x19, x24\n" - "st1 { v26.s }[0], [x20]\n" + "st1 { v23.s }[2], [x21], x16\n" + "add x14, x14, #0x4\n" + "st1 { v26.s }[2], [x20], x16\n" + "add x27, x27, #0x4\n" "add x24, x24, #0x4\n" - "st1 { v25.s }[0], [x19], x14\n" - "st1 { v24.s }[0], [x19], x14\n" - "st1 { v23.s }[0], [x19]\n" - "tbz %x[n_channels], #0, 48f\n" - "mov x21, x13\n" - "st1 { v31.h }[2], [x21], x14\n" - "mov x20, x25\n" - "st1 { v30.h }[2], [x21], x14\n" - "st1 { v28.h }[2], [x20], x14\n" + "st1 { v29.s }[2], [x19], x16\n" + "st1 { v24.s }[2], [x21], x16\n" + "st1 { v27.s }[2], [x20], x16\n" + "st1 { v30.s }[2], [x19], x16\n" + "st1 { v25.s }[2], [x21]\n" + "st1 { v28.s }[2], [x20]\n" + "st1 { v31.s }[2], [x19]\n" + "tbz %x[n_channels], #0, 92f\n" + "mov x21, x14\n" + "mov x20, x27\n" "mov x19, x24\n" - "st1 { v29.h }[2], [x21]\n" - "st1 { v27.h }[2], [x20], x14\n" - "st1 { v26.h }[2], [x20]\n" - "st1 { v25.h }[2], [x19], x14\n" - "st1 { v24.h }[2], [x19], x14\n" - "st1 { v23.h }[2], [x19]\n" - "b 48f\n" - "47:" // Tile loop: Oddments: Store: Bit 1: Unset - "mov x21, x13\n" - "st1 { v31.h }[0], [x21], x14\n" - "mov x20, x25\n" + "st1 { v23.h }[6], [x21], x16\n" + "st1 { v26.h }[6], [x20], x16\n" + "st1 { v29.h }[6], [x19], x16\n" + "st1 { v24.h }[6], [x21], x16\n" + "st1 { v27.h }[6], [x20], x16\n" + "st1 { v30.h }[6], [x19], x16\n" + "st1 { v25.h }[6], [x21]\n" + "st1 { v28.h }[6], [x20]\n" + "st1 { v31.h }[6], [x19]\n" + "b 92f\n" + "89:" // Tile loop: Oddments: Store: Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 92f\n" + "mov x21, x14\n" + "mov x20, x27\n" + "st1 { v23.h }[4], [x21], x16\n" + "mov x19, x24\n" + "st1 { v26.h }[4], [x20], x16\n" + "st1 { v29.h }[4], [x19], x16\n" + "st1 { v24.h }[4], [x21], x16\n" + "st1 { v27.h }[4], [x20], x16\n" + "st1 { v30.h }[4], [x19], x16\n" + "st1 { v25.h }[4], [x21]\n" + "st1 { v28.h }[4], [x20]\n" + "st1 { v31.h }[4], [x19]\n" + "b 92f\n" + "90:" // Tile loop: Oddments: Store: Bit 2: Unset + "tbz %x[n_channels], #1, 91f\n" + "mov x21, x14\n" + "mov x20, x27\n" + "st1 { v23.s }[0], [x21], x16\n" + "mov x19, x24\n" + "st1 { v26.s }[0], [x20], x16\n" + "add x14, x14, #0x4\n" + "st1 { v29.s }[0], [x19], x16\n" + "add x27, x27, #0x4\n" + "add x24, x24, #0x4\n" + "st1 { v24.s }[0], [x21], x16\n" + "st1 { v27.s }[0], [x20], x16\n" + "st1 { v30.s }[0], [x19], x16\n" + "st1 { v25.s }[0], [x21]\n" + "st1 { v28.s }[0], [x20]\n" + "st1 { v31.s }[0], [x19]\n" + "tbz %x[n_channels], #0, 92f\n" + "mov x21, x14\n" + "mov x20, x27\n" + "mov x19, x24\n" + "st1 { v23.h }[2], [x21], x16\n" + "st1 { v26.h }[2], [x20], x16\n" + "st1 { v29.h }[2], [x19], x16\n" + "st1 { v24.h }[2], [x21], x16\n" + "st1 { v27.h }[2], [x20], x16\n" + "st1 { v30.h }[2], [x19], x16\n" + "st1 { v25.h }[2], [x21]\n" + "st1 { v28.h }[2], [x20]\n" + "st1 { v31.h }[2], [x19]\n" + "b 92f\n" + "91:" // Tile loop: Oddments: Store: Bit 2: Unset: Bit 1: Unset + "mov x21, x14\n" + "mov x20, x27\n" + "st1 { v23.h }[0], [x21], x16\n" "mov x19, x24\n" - "st1 { v30.h }[0], [x21], x14\n" - "st1 { v28.h }[0], [x20], x14\n" - "st1 { v29.h }[0], [x21]\n" - "st1 { v27.h }[0], [x20], x14\n" - "st1 { v26.h }[0], [x20]\n" - "st1 { v25.h }[0], [x19], x14\n" - "st1 { v24.h }[0], [x19], x14\n" - "st1 { v23.h }[0], [x19]\n" - "48:" // Tile loop: Oddments: Store: Bit 1: End + "st1 { v26.h }[0], [x20], x16\n" + "st1 { v29.h }[0], [x19], x16\n" + "st1 { v24.h }[0], [x21], x16\n" + "st1 { v27.h }[0], [x20], x16\n" + "st1 { v30.h }[0], [x19], x16\n" + "st1 { v25.h }[0], [x21]\n" + "st1 { v28.h }[0], [x20]\n" + "st1 { v31.h }[0], [x19]\n" + "92:" // Tile loop: Oddments: Store: Bit 2: End - "49:" // Tile loop: End - "ldr x7, [%x[params_struct], %[offsetof_args_tile_i]]\n" - "add x21, x7, #0x1\n" - "ldr x8, [%x[params_struct], %[offsetof_args_tile_j]]\n" - "ldr x20, [%x[params_struct], %[offsetof_args_n_tile_rows]]\n" - "add x8, x8, #0x1\n" + "93:" // Tile loop: End + "ldr x22, [%x[params_struct], %[offsetof_args_tile_j]]\n" + "ldr x23, [%x[params_struct], %[offsetof_args_tile_i]]\n" + "add x22, x22, #0x1\n" + "add x20, x23, #0x1\n" "ldr x19, [%x[params_struct], %[offsetof_args_n_tile_cols]]\n" - "cmp x8, x19\n" - "csel x8, x8, XZR, LT\n" - "csel x7, x7, x21, LT\n" - "cmp x7, x20\n" + "cmp x22, x19\n" + "ldr x19, [%x[params_struct], %[offsetof_args_n_tile_rows]]\n" + "csel x23, x23, x20, LT\n" + "csel x22, x22, XZR, LT\n" + "cmp x23, x19\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", "v16", "v17", "v18", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31", "x7", "x8", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x16", "x17", "x19", "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", "v16", "v17", "v18", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x16", "x17", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28" ); } diff --git a/src/core/NEON/kernels/arm_conv/depthwise/kernels/a64_fp16_nhwc_3x3_s1_output3x3_mla_depthfirst/generic_indirect.cpp b/src/core/NEON/kernels/arm_conv/depthwise/kernels/a64_fp16_nhwc_3x3_s1_output3x3_mla_depthfirst/generic_indirect.cpp index ed47c308c4..fc988913be 100644 --- a/src/core/NEON/kernels/arm_conv/depthwise/kernels/a64_fp16_nhwc_3x3_s1_output3x3_mla_depthfirst/generic_indirect.cpp +++ b/src/core/NEON/kernels/arm_conv/depthwise/kernels/a64_fp16_nhwc_3x3_s1_output3x3_mla_depthfirst/generic_indirect.cpp @@ -87,818 +87,1203 @@ void a64_fp16_nhwc_3x3_s1_output3x3_mla_depthfirst_indirect_impl( activation_min, activation_max); __asm__ __volatile__( - "ldr x17, [%x[params_struct], %[offsetof_args_outptrs]]\n" - "add x16, %x[params_struct], %[offsetof_Args_inptrs]\n" - "ldr x15, [%x[params_struct], %[offsetof_args_params]]\n" + "mov x17, #0x10\n" // cntb _, ALL, #1 + "lsr x16, %x[n_channels], #0x3\n" + "ldr x15, [%x[params_struct], %[offsetof_args_outptrs]]\n" + "ldr x14, [%x[params_struct], %[offsetof_args_params]]\n" "add x20, %x[params_struct], %[offsetof_args_min]\n" "add x19, %x[params_struct], %[offsetof_args_max]\n" "ld1r { v18.8h }, [x20]\n" "ld1r { v17.8h }, [x19]\n" - "mov x14, #0x0\n" - "mov x13, #0x10\n" // cntb _, ALL, #1 - "sub x12, XZR, x13\n" - "lsr x11, %x[n_channels], #0x3\n" - "cbz x11, 3f\n" - "ldr q16, [x15, #0x0]\n" - "ldr q0, [x15, #0x10]\n" - "cmp x13, x11, LSL #4\n" - "ldr q1, [x15, #0x20]\n" - "ldr q2, [x15, #0x30]\n" - "ldr q3, [x15, #0x40]\n" - "ldr q4, [x15, #0x50]\n" - "ldr q5, [x15, #0x60]\n" - "ldr q6, [x15, #0x70]\n" - "ldr q7, [x15, #0x80]\n" - "ldr q8, [x15, #0x90]\n" - "add x15, x15, #0xa0\n" - "ldp x10, x9, [x16, #0x0]\n" - "ldp x28, x27, [x16, #0x10]\n" - "ldr x26, [x16, #0x20]\n" - "ldr q9, [x10, x14]\n" - "ldr q10, [x9, x14]\n" - "ldr q11, [x28, x14]\n" - "ldr q12, [x27, x14]\n" - "ldr q13, [x26, x14]\n" + "add x13, %x[params_struct], %[offsetof_Args_inptrs]\n" + "mov x12, #0x0\n" + "sub x11, XZR, x17\n" + "cbz x16, 3f\n" + "ldp x10, x9, [x13, #0x0]\n" + "ldp x28, x27, [x13, #0x10]\n" + "ldr x26, [x13, #0x20]\n" + "cmp x17, x16, LSL #4\n" + "ldr q16, [x14, #0x0]\n" + "ldr q0, [x14, #0x10]\n" + "ldr q1, [x14, #0x20]\n" + "ldr q2, [x14, #0x30]\n" + "ldr q3, [x14, #0x40]\n" + "ldr q4, [x14, #0x50]\n" + "ldr q5, [x14, #0x60]\n" + "ldr q6, [x14, #0x70]\n" + "ldr q7, [x14, #0x80]\n" + "ldr q8, [x14, #0x90]\n" + "ldr q9, [x10, x12]\n" + "add x14, x14, #0xa0\n" + "ldr q10, [x9, x12]\n" + "ldr q11, [x28, x12]\n" + "ldr q12, [x27, x12]\n" + "ldr q13, [x26, x12]\n" "bge 2f\n" "1:" // Channel loop - "mov v31.16b, v16.16b\n fmla v31.8h, v8.8h, v9.8h\n" - "ldr x25, [x16, #0x28]\n" - "add x12, x12, #0x10\n" - "mov v30.16b, v16.16b\n fmla v30.8h, v7.8h, v9.8h\n" - "ldr x24, [x16, #0x30]\n" - "mov v29.16b, v16.16b\n fmla v29.8h, v6.8h, v9.8h\n" - "ldr x23, [x16, #0x38]\n" - "mov v28.16b, v16.16b\n fmla v28.8h, v5.8h, v9.8h\n" - "ldr x10, [x16, #0x40]\n" + "mov v23.16b, v16.16b\n fmla v23.8h, v8.8h, v9.8h\n" + "mov v24.16b, v16.16b\n fmla v24.8h, v7.8h, v9.8h\n" + "ldr x25, [x13, #0x30]\n" + "ldr x24, [x13, #0x38]\n" + "mov v25.16b, v16.16b\n fmla v25.8h, v6.8h, v9.8h\n" + "fmla v23.8h, v0.8h, v10.8h\n" + "ldr x23, [x13, #0x28]\n" + "ldr x9, [x13, #0x48]\n" + "fmla v24.8h, v4.8h, v13.8h\n" + "mov v26.16b, v16.16b\n fmla v26.8h, v5.8h, v9.8h\n" + "ldr x10, [x13, #0x40]\n" + "ldr q10, [x9, x12]\n" "mov v27.16b, v16.16b\n fmla v27.8h, v4.8h, v9.8h\n" - "ldr x9, [x16, #0x48]\n" - "mov v26.16b, v16.16b\n fmla v26.8h, v3.8h, v9.8h\n" - "ldr x28, [x16, #0x50]\n" - "mov v25.16b, v16.16b\n fmla v25.8h, v2.8h, v9.8h\n" - "ldr x27, [x16, #0x58]\n" - "mov v24.16b, v16.16b\n fmla v24.8h, v1.8h, v9.8h\n" - "ldr x26, [x16, #0x60]\n" - "mov v23.16b, v16.16b\n fmla v23.8h, v0.8h, v9.8h\n" - "ldr x22, [x17, #0x0]\n" - "fmla v31.8h, v0.8h, v10.8h\n" - "ldr q10, [x9, x14]\n" - "fmla v29.8h, v2.8h, v11.8h\n" - "ldr q11, [x24, x14]\n" - "fmla v25.8h, v6.8h, v12.8h\n" - "ldr q12, [x25, x14]\n" - "fmla v30.8h, v4.8h, v13.8h\n" - "ldr x25, [x16, #0x68]\n" - "fmla v31.8h, v5.8h, v13.8h\n" - "ldr x24, [x16, #0x70]\n" - "fmla v29.8h, v3.8h, v13.8h\n" - "ldr x9, [x16, #0x88]\n" - "fmla v28.8h, v2.8h, v13.8h\n" - "ldr x21, [x17, #0x8]\n" + "mov v28.16b, v16.16b\n fmla v28.8h, v3.8h, v9.8h\n" + "ldr x28, [x13, #0x50]\n" + "ldr x27, [x13, #0x58]\n" + "fmla v25.8h, v2.8h, v11.8h\n" + "ldr q11, [x25, x12]\n" + "mov v29.16b, v16.16b\n fmla v29.8h, v2.8h, v9.8h\n" + "ldr x26, [x13, #0x60]\n" + "fmla v23.8h, v5.8h, v13.8h\n" + "fmla v24.8h, v6.8h, v11.8h\n" + "ldr x25, [x13, #0x70]\n" + "ldr x9, [x13, #0x88]\n" + "mov v31.16b, v16.16b\n fmla v31.8h, v0.8h, v9.8h\n" + "fmla v25.8h, v3.8h, v13.8h\n" + "ldr x22, [x15, #0x0]\n" + "add x11, x11, #0x10\n" + "fmla v26.8h, v2.8h, v13.8h\n" "fmla v27.8h, v1.8h, v13.8h\n" - "ldr x20, [x17, #0x10]\n" - "fmla v26.8h, v0.8h, v13.8h\n" - "ldr q13, [x23, x14]\n" - "fmla v23.8h, v8.8h, v12.8h\n" - "ldr q12, [x10, x14]\n" - "fmla v31.8h, v7.8h, v11.8h\n" - "ldr x23, [x16, #0x78]\n" - "fmla v30.8h, v6.8h, v11.8h\n" - "ldr x10, [x16, #0x80]\n" - "fmla v28.8h, v4.8h, v11.8h\n" - "ldr x19, [x17, #0x18]\n" + "ldr x21, [x15, #0x8]\n" + "ldr x20, [x15, #0x10]\n" + "fmla v28.8h, v0.8h, v13.8h\n" + "ldr q13, [x24, x12]\n" + "fmla v29.8h, v6.8h, v12.8h\n" + "ldr q12, [x23, x12]\n" + "mov v30.16b, v16.16b\n fmla v30.8h, v1.8h, v9.8h\n" + "fmla v23.8h, v7.8h, v11.8h\n" + "ldr x23, [x13, #0x68]\n" + "ldr x24, [x13, #0x78]\n" + "fmla v24.8h, v0.8h, v13.8h\n" + "fmla v31.8h, v8.8h, v12.8h\n" + "ldr q12, [x10, x12]\n" + "ldr x10, [x13, #0x80]\n" + "fmla v26.8h, v4.8h, v11.8h\n" "fmla v27.8h, v3.8h, v11.8h\n" - "ldr q16, [x15, #0x0]\n" - "fmla v25.8h, v1.8h, v11.8h\n" - "fmla v24.8h, v0.8h, v11.8h\n" - "ldr q11, [x28, x14]\n" - "fmla v31.8h, v1.8h, v13.8h\n" - "ldr x28, [x16, #0x90]\n" - "fmla v30.8h, v0.8h, v13.8h\n" - "ldr q13, [x27, x14]\n" - "fmla v29.8h, v1.8h, v12.8h\n" - "ldr x27, [x16, #0x98]\n" + "ldr x19, [x15, #0x18]\n" + "ldr q16, [x14, #0x0]\n" + "fmla v30.8h, v0.8h, v11.8h\n" + "fmla v28.8h, v4.8h, v10.8h\n" + "fmla v29.8h, v1.8h, v11.8h\n" + "fmla v23.8h, v1.8h, v13.8h\n" + "ldr q11, [x28, x12]\n" + "ldr q13, [x27, x12]\n" + "fmla v24.8h, v2.8h, v12.8h\n" + "fmla v25.8h, v1.8h, v12.8h\n" + "ldr q12, [x26, x12]\n" + "ldr x28, [x13, #0x90]\n" "fmla v27.8h, v5.8h, v10.8h\n" - "fmla v26.8h, v4.8h, v10.8h\n" - "fmla v30.8h, v2.8h, v12.8h\n" - "ldr q12, [x26, x14]\n" - "fmla v29.8h, v7.8h, v10.8h\n" - "ldr x26, [x16, #0xa0]\n" - "fmla v24.8h, v2.8h, v10.8h\n" - "fmla v23.8h, v1.8h, v10.8h\n" - "fmla v30.8h, v8.8h, v10.8h\n" - "ldr q10, [x25, x14]\n" - "fmla v31.8h, v3.8h, v11.8h\n" - "ldr x25, [x16, #0xa8]\n" - "fmla v28.8h, v0.8h, v11.8h\n" - "ldr q11, [x24, x14]\n" - "fmla v29.8h, v5.8h, v13.8h\n" - "ldr x24, [x16, #0xb0]\n" - "fmla v26.8h, v2.8h, v13.8h\n" - "ldr q13, [x23, x14]\n" - "fmla v25.8h, v3.8h, v12.8h\n" - "ldr x23, [x16, #0xb8]\n" - "fmla v28.8h, v6.8h, v12.8h\n" - "ldr q12, [x10, x14]\n" + "fmla v30.8h, v2.8h, v10.8h\n" + "ldr x26, [x13, #0xa0]\n" + "ldr x27, [x13, #0x98]\n" + "fmla v26.8h, v0.8h, v11.8h\n" + "fmla v28.8h, v2.8h, v13.8h\n" + "fmla v24.8h, v8.8h, v10.8h\n" + "fmla v25.8h, v7.8h, v10.8h\n" + "fmla v31.8h, v1.8h, v10.8h\n" + "fmla v29.8h, v3.8h, v12.8h\n" + "ldr q10, [x23, x12]\n" + "ldr x23, [x13, #0xa8]\n" + "fmla v26.8h, v6.8h, v12.8h\n" "fmla v27.8h, v7.8h, v10.8h\n" - "ldr x10, [x16, #0xc0]\n" - "fmla v26.8h, v6.8h, v10.8h\n" - "fmla v25.8h, v5.8h, v10.8h\n" - "fmla v28.8h, v8.8h, v10.8h\n" - "fmla v24.8h, v4.8h, v10.8h\n" - "fmla v23.8h, v3.8h, v10.8h\n" - "fmla v26.8h, v8.8h, v11.8h\n" - "fmla v25.8h, v7.8h, v13.8h\n" - "fmla v24.8h, v6.8h, v13.8h\n" - "ldr q13, [x28, x14]\n" - "fmla v23.8h, v5.8h, v11.8h\n" - "ldr q11, [x9, x14]\n" - "fmla v31.8h, v4.8h, v12.8h\n" - "fmla v30.8h, v3.8h, v12.8h\n" - "fmla v28.8h, v1.8h, v12.8h\n" + "ldr q12, [x10, x12]\n" + "ldr x10, [x13, #0xc0]\n" + "fmla v28.8h, v6.8h, v10.8h\n" + "fmla v30.8h, v4.8h, v10.8h\n" + "fmla v23.8h, v3.8h, v11.8h\n" + "fmla v25.8h, v5.8h, v13.8h\n" + "ldr q11, [x25, x12]\n" + "ldr q13, [x24, x12]\n" + "fmla v29.8h, v5.8h, v10.8h\n" + "fmla v31.8h, v3.8h, v10.8h\n" + "ldr x25, [x13, #0xb0]\n" + "ldr x24, [x13, #0xb8]\n" + "fmla v26.8h, v8.8h, v10.8h\n" + "fmla v28.8h, v8.8h, v11.8h\n" + "fmla v30.8h, v6.8h, v13.8h\n" + "fmla v24.8h, v3.8h, v12.8h\n" "fmla v27.8h, v0.8h, v12.8h\n" - "ldr q12, [x27, x14]\n" - "fmla v29.8h, v4.8h, v11.8h\n" - "fmla v30.8h, v5.8h, v11.8h\n" - "fmla v26.8h, v1.8h, v11.8h\n" + "fmla v31.8h, v5.8h, v11.8h\n" + "ldr q11, [x9, x12]\n" + "fmla v29.8h, v7.8h, v13.8h\n" + "ldr q13, [x28, x12]\n" + "fmla v23.8h, v4.8h, v12.8h\n" + "fmla v26.8h, v1.8h, v12.8h\n" + "fmla v24.8h, v5.8h, v11.8h\n" + "ldr q12, [x27, x12]\n" + "fmla v25.8h, v4.8h, v11.8h\n" "fmla v27.8h, v2.8h, v11.8h\n" - "ldr q11, [x26, x14]\n" - "fmla v24.8h, v8.8h, v13.8h\n" - "ldr x26, [x16, #0x20]\n" - "fmla v23.8h, v7.8h, v13.8h\n" - "ldr q13, [x25, x14]\n" - "fmla v28.8h, v7.8h, v12.8h\n" + "fmla v28.8h, v1.8h, v11.8h\n" + "fmla v30.8h, v8.8h, v13.8h\n" + "ldr q11, [x26, x12]\n" + "ldr x26, [x13, #0x20]\n" + "fmla v31.8h, v7.8h, v13.8h\n" + "ldr q13, [x23, x12]\n" + "fmla v23.8h, v2.8h, v11.8h\n" + "fmla v26.8h, v7.8h, v12.8h\n" "fmla v27.8h, v6.8h, v12.8h\n" - "fmla v25.8h, v4.8h, v12.8h\n" - "fmla v24.8h, v3.8h, v12.8h\n" - "ldr q12, [x24, x14]\n" + "fmla v29.8h, v4.8h, v12.8h\n" + "fmla v30.8h, v3.8h, v12.8h\n" + "ldr q12, [x25, x12]\n" + "fmla v31.8h, v4.8h, v13.8h\n" + "fmla v24.8h, v1.8h, v11.8h\n" + "fmax v24.8h, v24.8h, v18.8h\n" + "ldr q1, [x14, #0x20]\n" + "fmla v25.8h, v0.8h, v11.8h\n" + "ldr q11, [x24, x12]\n" + "fmla v23.8h, v6.8h, v12.8h\n" + "fmax v23.8h, v23.8h, v18.8h\n" + "fmla v28.8h, v7.8h, v13.8h\n" + "fmla v30.8h, v5.8h, v13.8h\n" + "fmin v23.8h, v23.8h, v17.8h\n" + "str q23, [x22, x11]\n" + "fmla v29.8h, v0.8h, v12.8h\n" "fmla v31.8h, v2.8h, v11.8h\n" - "fmla v30.8h, v1.8h, v11.8h\n" - "ldr q1, [x15, #0x20]\n" - "fmla v29.8h, v0.8h, v11.8h\n" - "ldr q11, [x23, x14]\n" + "ldr x22, [x15, #0x20]\n" + "fmin v24.8h, v24.8h, v17.8h\n" "fmla v27.8h, v8.8h, v13.8h\n" - "fmla v26.8h, v7.8h, v13.8h\n" - "fmla v24.8h, v5.8h, v13.8h\n" - "fmla v23.8h, v4.8h, v13.8h\n" - "ldr q13, [x10, x14]\n" - "add x14, x14, #0x10\n" - "fmla v31.8h, v6.8h, v12.8h\n" - "ldp x10, x9, [x16, #0x0]\n" - "fmla v28.8h, v3.8h, v12.8h\n" - "ldp x28, x27, [x16, #0x10]\n" - "fmla v25.8h, v0.8h, v12.8h\n" - "ldr q0, [x15, #0x10]\n" - "fmla v29.8h, v8.8h, v11.8h\n" - "ldr q9, [x10, x13]\n" - "fmla v26.8h, v5.8h, v11.8h\n" - "ldr q10, [x9, x13]\n" - "fmla v23.8h, v2.8h, v11.8h\n" - "ldr q11, [x28, x13]\n" - "fmla v25.8h, v8.8h, v13.8h\n" - "ldr q12, [x27, x13]\n" - "fmla v24.8h, v7.8h, v13.8h\n" - "ldr q2, [x15, #0x30]\n" - "fmax v31.8h, v31.8h, v18.8h\n" - "ldr q3, [x15, #0x40]\n" - "fmla v23.8h, v6.8h, v13.8h\n" - "ldr q13, [x26, x13]\n" - "add x13, x13, #0x10\n" - "fmin v31.8h, v31.8h, v17.8h\n" - "ldr q4, [x15, #0x50]\n" - "cmp x13, x11, LSL #4\n" - "fmax v30.8h, v30.8h, v18.8h\n" - "ldr q5, [x15, #0x60]\n" - "fmax v29.8h, v29.8h, v18.8h\n" - "ldr q6, [x15, #0x70]\n" - "fmax v28.8h, v28.8h, v18.8h\n" - "str q31, [x22, x12]\n" - "fmax v27.8h, v27.8h, v18.8h\n" - "ldr x22, [x17, #0x20]\n" - "fmin v30.8h, v30.8h, v17.8h\n" - "ldr q7, [x15, #0x80]\n" - "fmin v29.8h, v29.8h, v17.8h\n" - "ldr q8, [x15, #0x90]\n" - "add x15, x15, #0xa0\n" - "fmin v28.8h, v28.8h, v17.8h\n" - "str q30, [x21, x12]\n" - "fmin v27.8h, v27.8h, v17.8h\n" - "str q29, [x20, x12]\n" + "fmla v26.8h, v3.8h, v12.8h\n" + "ldr q13, [x10, x12]\n" "fmax v26.8h, v26.8h, v18.8h\n" - "ldr x21, [x17, #0x28]\n" + "fmla v25.8h, v8.8h, v11.8h\n" + "fmla v28.8h, v5.8h, v11.8h\n" "fmax v25.8h, v25.8h, v18.8h\n" - "str q28, [x19, x12]\n" - "fmax v24.8h, v24.8h, v18.8h\n" - "str q27, [x22, x12]\n" - "fmin v26.8h, v26.8h, v17.8h\n" - "ldr x20, [x17, #0x30]\n" + "ldp x10, x9, [x13, #0x0]\n" + "fmla v29.8h, v8.8h, v13.8h\n" + "fmla v30.8h, v7.8h, v13.8h\n" + "fmax v27.8h, v27.8h, v18.8h\n" + "ldp x28, x27, [x13, #0x10]\n" + "fmla v31.8h, v6.8h, v13.8h\n" "fmin v25.8h, v25.8h, v17.8h\n" - "ldr x19, [x17, #0x38]\n" - "fmin v24.8h, v24.8h, v17.8h\n" - "str q26, [x21, x12]\n" - "fmax v23.8h, v23.8h, v18.8h\n" - "str q25, [x20, x12]\n" - "ldr x22, [x17, #0x40]\n" - "fmin v23.8h, v23.8h, v17.8h\n" - "str q24, [x19, x12]\n" - "str q23, [x22, x12]\n" + "str q24, [x21, x11]\n" + "ldr x21, [x15, #0x28]\n" + "fmin v26.8h, v26.8h, v17.8h\n" + "fmin v27.8h, v27.8h, v17.8h\n" + "str q25, [x20, x11]\n" + "ldr x20, [x15, #0x30]\n" + "fmax v28.8h, v28.8h, v18.8h\n" + "fmax v29.8h, v29.8h, v18.8h\n" + "str q26, [x19, x11]\n" + "ldr x19, [x15, #0x38]\n" + "fmax v30.8h, v30.8h, v18.8h\n" + "fmax v31.8h, v31.8h, v18.8h\n" + "str q27, [x22, x11]\n" + "ldr x22, [x15, #0x40]\n" + "ldr q9, [x10, x17]\n" + "ldr q10, [x9, x17]\n" + "fmin v28.8h, v28.8h, v17.8h\n" + "fmin v29.8h, v29.8h, v17.8h\n" + "ldr q11, [x28, x17]\n" + "ldr q12, [x27, x17]\n" + "fmin v30.8h, v30.8h, v17.8h\n" + "fmin v31.8h, v31.8h, v17.8h\n" + "ldr q13, [x26, x17]\n" + "add x17, x17, #0x10\n" + "cmp x17, x16, LSL #4\n" + "str q28, [x21, x11]\n" + "add x12, x12, #0x10\n" + "str q29, [x20, x11]\n" + "ldr q0, [x14, #0x10]\n" + "ldr q2, [x14, #0x30]\n" + "str q30, [x19, x11]\n" + "ldr q3, [x14, #0x40]\n" + "ldr q4, [x14, #0x50]\n" + "str q31, [x22, x11]\n" + "ldr q5, [x14, #0x60]\n" + "ldr q6, [x14, #0x70]\n" + "ldr q7, [x14, #0x80]\n" + "ldr q8, [x14, #0x90]\n" + "add x14, x14, #0xa0\n" "blt 1b\n" "2:" // Channel tail - "mov v31.16b, v16.16b\n fmla v31.8h, v8.8h, v9.8h\n" - "ldr x25, [x16, #0x28]\n" - "add x12, x12, #0x10\n" - "mov v30.16b, v16.16b\n fmla v30.8h, v7.8h, v9.8h\n" - "ldr x24, [x16, #0x30]\n" - "mov v29.16b, v16.16b\n fmla v29.8h, v6.8h, v9.8h\n" - "ldr x23, [x16, #0x38]\n" - "mov v28.16b, v16.16b\n fmla v28.8h, v5.8h, v9.8h\n" - "ldr x10, [x16, #0x40]\n" + "mov v23.16b, v16.16b\n fmla v23.8h, v8.8h, v9.8h\n" + "mov v24.16b, v16.16b\n fmla v24.8h, v7.8h, v9.8h\n" + "ldr x25, [x13, #0x30]\n" + "ldr x24, [x13, #0x38]\n" + "mov v25.16b, v16.16b\n fmla v25.8h, v6.8h, v9.8h\n" + "fmla v23.8h, v0.8h, v10.8h\n" + "ldr x23, [x13, #0x28]\n" + "ldr x9, [x13, #0x48]\n" + "fmla v24.8h, v4.8h, v13.8h\n" + "mov v26.16b, v16.16b\n fmla v26.8h, v5.8h, v9.8h\n" + "ldr x10, [x13, #0x40]\n" + "ldr q10, [x9, x12]\n" "mov v27.16b, v16.16b\n fmla v27.8h, v4.8h, v9.8h\n" - "ldr x9, [x16, #0x48]\n" - "mov v26.16b, v16.16b\n fmla v26.8h, v3.8h, v9.8h\n" - "ldr x28, [x16, #0x50]\n" - "mov v25.16b, v16.16b\n fmla v25.8h, v2.8h, v9.8h\n" - "ldr x27, [x16, #0x58]\n" - "mov v24.16b, v16.16b\n fmla v24.8h, v1.8h, v9.8h\n" - "ldr x26, [x16, #0x60]\n" - "mov v23.16b, v16.16b\n fmla v23.8h, v0.8h, v9.8h\n" - "ldr x22, [x17, #0x0]\n" - "fmla v31.8h, v0.8h, v10.8h\n" - "ldr q10, [x9, x14]\n" - "fmla v29.8h, v2.8h, v11.8h\n" - "ldr q11, [x24, x14]\n" - "fmla v25.8h, v6.8h, v12.8h\n" - "ldr q12, [x25, x14]\n" - "fmla v30.8h, v4.8h, v13.8h\n" - "ldr x25, [x16, #0x68]\n" - "fmla v31.8h, v5.8h, v13.8h\n" - "ldr x24, [x16, #0x70]\n" - "fmla v29.8h, v3.8h, v13.8h\n" - "ldr x9, [x16, #0x88]\n" - "fmla v28.8h, v2.8h, v13.8h\n" - "ldr x21, [x17, #0x8]\n" + "mov v28.16b, v16.16b\n fmla v28.8h, v3.8h, v9.8h\n" + "ldr x28, [x13, #0x50]\n" + "ldr x27, [x13, #0x58]\n" + "fmla v25.8h, v2.8h, v11.8h\n" + "ldr q11, [x25, x12]\n" + "mov v29.16b, v16.16b\n fmla v29.8h, v2.8h, v9.8h\n" + "ldr x26, [x13, #0x60]\n" + "fmla v23.8h, v5.8h, v13.8h\n" + "fmla v24.8h, v6.8h, v11.8h\n" + "ldr x25, [x13, #0x70]\n" + "ldr x9, [x13, #0x88]\n" + "mov v31.16b, v16.16b\n fmla v31.8h, v0.8h, v9.8h\n" + "fmla v25.8h, v3.8h, v13.8h\n" + "ldr x22, [x15, #0x0]\n" + "add x11, x11, #0x10\n" + "fmla v26.8h, v2.8h, v13.8h\n" "fmla v27.8h, v1.8h, v13.8h\n" - "ldr x20, [x17, #0x10]\n" - "fmla v26.8h, v0.8h, v13.8h\n" - "ldr q13, [x23, x14]\n" - "fmla v23.8h, v8.8h, v12.8h\n" - "ldr q12, [x10, x14]\n" - "fmla v31.8h, v7.8h, v11.8h\n" - "ldr x23, [x16, #0x78]\n" - "fmla v30.8h, v6.8h, v11.8h\n" - "ldr x10, [x16, #0x80]\n" - "fmla v28.8h, v4.8h, v11.8h\n" - "ldr x19, [x17, #0x18]\n" + "ldr x21, [x15, #0x8]\n" + "ldr x20, [x15, #0x10]\n" + "fmla v28.8h, v0.8h, v13.8h\n" + "ldr q13, [x24, x12]\n" + "fmla v29.8h, v6.8h, v12.8h\n" + "ldr q12, [x23, x12]\n" + "mov v30.16b, v16.16b\n fmla v30.8h, v1.8h, v9.8h\n" + "fmla v23.8h, v7.8h, v11.8h\n" + "ldr x23, [x13, #0x68]\n" + "ldr x24, [x13, #0x78]\n" + "fmla v24.8h, v0.8h, v13.8h\n" + "fmla v31.8h, v8.8h, v12.8h\n" + "ldr q12, [x10, x12]\n" + "ldr x10, [x13, #0x80]\n" + "fmla v26.8h, v4.8h, v11.8h\n" "fmla v27.8h, v3.8h, v11.8h\n" - "fmla v25.8h, v1.8h, v11.8h\n" - "fmla v24.8h, v0.8h, v11.8h\n" - "ldr q11, [x28, x14]\n" - "fmla v31.8h, v1.8h, v13.8h\n" - "ldr x28, [x16, #0x90]\n" - "fmla v30.8h, v0.8h, v13.8h\n" - "ldr q13, [x27, x14]\n" - "fmla v29.8h, v1.8h, v12.8h\n" - "ldr x27, [x16, #0x98]\n" + "ldr x19, [x15, #0x18]\n" + "fmla v30.8h, v0.8h, v11.8h\n" + "fmla v28.8h, v4.8h, v10.8h\n" + "fmla v29.8h, v1.8h, v11.8h\n" + "fmla v23.8h, v1.8h, v13.8h\n" + "ldr q11, [x28, x12]\n" + "ldr q13, [x27, x12]\n" + "fmla v24.8h, v2.8h, v12.8h\n" + "fmla v25.8h, v1.8h, v12.8h\n" + "ldr q12, [x26, x12]\n" + "ldr x28, [x13, #0x90]\n" "fmla v27.8h, v5.8h, v10.8h\n" - "fmla v26.8h, v4.8h, v10.8h\n" - "fmla v30.8h, v2.8h, v12.8h\n" - "ldr q12, [x26, x14]\n" - "fmla v29.8h, v7.8h, v10.8h\n" - "ldr x26, [x16, #0xa0]\n" - "fmla v24.8h, v2.8h, v10.8h\n" - "fmla v23.8h, v1.8h, v10.8h\n" - "fmla v30.8h, v8.8h, v10.8h\n" - "ldr q10, [x25, x14]\n" - "fmla v31.8h, v3.8h, v11.8h\n" - "ldr x25, [x16, #0xa8]\n" - "fmla v28.8h, v0.8h, v11.8h\n" - "ldr q11, [x24, x14]\n" - "fmla v29.8h, v5.8h, v13.8h\n" - "ldr x24, [x16, #0xb0]\n" - "fmla v26.8h, v2.8h, v13.8h\n" - "ldr q13, [x23, x14]\n" - "fmla v25.8h, v3.8h, v12.8h\n" - "ldr x23, [x16, #0xb8]\n" - "fmla v28.8h, v6.8h, v12.8h\n" - "ldr q12, [x10, x14]\n" + "fmla v30.8h, v2.8h, v10.8h\n" + "ldr x26, [x13, #0xa0]\n" + "ldr x27, [x13, #0x98]\n" + "fmla v26.8h, v0.8h, v11.8h\n" + "fmla v28.8h, v2.8h, v13.8h\n" + "fmla v24.8h, v8.8h, v10.8h\n" + "fmla v25.8h, v7.8h, v10.8h\n" + "fmla v31.8h, v1.8h, v10.8h\n" + "fmla v29.8h, v3.8h, v12.8h\n" + "ldr q10, [x23, x12]\n" + "ldr x23, [x13, #0xa8]\n" + "fmla v26.8h, v6.8h, v12.8h\n" "fmla v27.8h, v7.8h, v10.8h\n" - "ldr x10, [x16, #0xc0]\n" - "fmla v26.8h, v6.8h, v10.8h\n" - "fmla v25.8h, v5.8h, v10.8h\n" - "fmla v28.8h, v8.8h, v10.8h\n" - "fmla v24.8h, v4.8h, v10.8h\n" - "fmla v23.8h, v3.8h, v10.8h\n" - "fmla v26.8h, v8.8h, v11.8h\n" - "fmla v25.8h, v7.8h, v13.8h\n" - "fmla v24.8h, v6.8h, v13.8h\n" - "ldr q13, [x28, x14]\n" - "fmla v23.8h, v5.8h, v11.8h\n" - "ldr q11, [x9, x14]\n" - "fmla v31.8h, v4.8h, v12.8h\n" - "fmla v30.8h, v3.8h, v12.8h\n" - "fmla v28.8h, v1.8h, v12.8h\n" + "ldr q12, [x10, x12]\n" + "ldr x10, [x13, #0xc0]\n" + "fmla v28.8h, v6.8h, v10.8h\n" + "fmla v30.8h, v4.8h, v10.8h\n" + "fmla v23.8h, v3.8h, v11.8h\n" + "fmla v25.8h, v5.8h, v13.8h\n" + "ldr q11, [x25, x12]\n" + "ldr q13, [x24, x12]\n" + "fmla v29.8h, v5.8h, v10.8h\n" + "fmla v31.8h, v3.8h, v10.8h\n" + "ldr x25, [x13, #0xb0]\n" + "ldr x24, [x13, #0xb8]\n" + "fmla v26.8h, v8.8h, v10.8h\n" + "fmla v28.8h, v8.8h, v11.8h\n" + "fmla v30.8h, v6.8h, v13.8h\n" + "fmla v24.8h, v3.8h, v12.8h\n" "fmla v27.8h, v0.8h, v12.8h\n" - "ldr q12, [x27, x14]\n" - "fmla v29.8h, v4.8h, v11.8h\n" - "fmla v30.8h, v5.8h, v11.8h\n" - "fmla v26.8h, v1.8h, v11.8h\n" + "fmla v31.8h, v5.8h, v11.8h\n" + "ldr q11, [x9, x12]\n" + "fmla v29.8h, v7.8h, v13.8h\n" + "ldr q13, [x28, x12]\n" + "fmla v23.8h, v4.8h, v12.8h\n" + "fmla v26.8h, v1.8h, v12.8h\n" + "fmla v24.8h, v5.8h, v11.8h\n" + "ldr q12, [x27, x12]\n" + "fmla v25.8h, v4.8h, v11.8h\n" "fmla v27.8h, v2.8h, v11.8h\n" - "ldr q11, [x26, x14]\n" - "fmla v24.8h, v8.8h, v13.8h\n" - "fmla v23.8h, v7.8h, v13.8h\n" - "ldr q13, [x25, x14]\n" - "fmla v28.8h, v7.8h, v12.8h\n" + "fmla v28.8h, v1.8h, v11.8h\n" + "fmla v30.8h, v8.8h, v13.8h\n" + "ldr q11, [x26, x12]\n" + "fmla v31.8h, v7.8h, v13.8h\n" + "ldr q13, [x23, x12]\n" + "fmla v23.8h, v2.8h, v11.8h\n" + "fmla v26.8h, v7.8h, v12.8h\n" "fmla v27.8h, v6.8h, v12.8h\n" - "fmla v25.8h, v4.8h, v12.8h\n" - "fmla v24.8h, v3.8h, v12.8h\n" - "ldr q12, [x24, x14]\n" + "fmla v29.8h, v4.8h, v12.8h\n" + "fmla v30.8h, v3.8h, v12.8h\n" + "ldr q12, [x25, x12]\n" + "fmla v31.8h, v4.8h, v13.8h\n" + "fmla v24.8h, v1.8h, v11.8h\n" + "fmax v24.8h, v24.8h, v18.8h\n" + "fmla v25.8h, v0.8h, v11.8h\n" + "ldr q11, [x24, x12]\n" + "fmla v23.8h, v6.8h, v12.8h\n" + "fmax v23.8h, v23.8h, v18.8h\n" + "fmla v28.8h, v7.8h, v13.8h\n" + "fmla v30.8h, v5.8h, v13.8h\n" + "fmin v23.8h, v23.8h, v17.8h\n" + "str q23, [x22, x11]\n" + "fmla v29.8h, v0.8h, v12.8h\n" "fmla v31.8h, v2.8h, v11.8h\n" - "fmla v30.8h, v1.8h, v11.8h\n" - "fmla v29.8h, v0.8h, v11.8h\n" - "ldr q11, [x23, x14]\n" + "ldr x22, [x15, #0x20]\n" + "fmin v24.8h, v24.8h, v17.8h\n" "fmla v27.8h, v8.8h, v13.8h\n" - "fmla v26.8h, v7.8h, v13.8h\n" - "fmla v24.8h, v5.8h, v13.8h\n" - "fmla v23.8h, v4.8h, v13.8h\n" - "ldr q13, [x10, x14]\n" - "add x14, x14, #0x10\n" - "fmla v31.8h, v6.8h, v12.8h\n" - "fmla v28.8h, v3.8h, v12.8h\n" - "fmla v25.8h, v0.8h, v12.8h\n" - "fmla v29.8h, v8.8h, v11.8h\n" - "fmla v26.8h, v5.8h, v11.8h\n" - "fmla v23.8h, v2.8h, v11.8h\n" - "fmla v25.8h, v8.8h, v13.8h\n" - "fmla v24.8h, v7.8h, v13.8h\n" - "fmax v31.8h, v31.8h, v18.8h\n" - "fmla v23.8h, v6.8h, v13.8h\n" - "fmax v30.8h, v30.8h, v18.8h\n" - "fmin v31.8h, v31.8h, v17.8h\n" - "str q31, [x22, x12]\n" - "fmin v30.8h, v30.8h, v17.8h\n" - "fmax v29.8h, v29.8h, v18.8h\n" - "ldr x22, [x17, #0x20]\n" - "fmax v28.8h, v28.8h, v18.8h\n" - "str q30, [x21, x12]\n" - "fmin v29.8h, v29.8h, v17.8h\n" - "fmax v27.8h, v27.8h, v18.8h\n" - "ldr x21, [x17, #0x28]\n" - "fmin v28.8h, v28.8h, v17.8h\n" - "str q29, [x20, x12]\n" - "fmin v27.8h, v27.8h, v17.8h\n" + "fmla v26.8h, v3.8h, v12.8h\n" + "ldr q13, [x10, x12]\n" "fmax v26.8h, v26.8h, v18.8h\n" - "str q28, [x19, x12]\n" + "fmla v25.8h, v8.8h, v11.8h\n" + "fmla v28.8h, v5.8h, v11.8h\n" "fmax v25.8h, v25.8h, v18.8h\n" - "ldr x20, [x17, #0x30]\n" - "fmax v24.8h, v24.8h, v18.8h\n" - "str q27, [x22, x12]\n" - "fmin v26.8h, v26.8h, v17.8h\n" - "ldr x19, [x17, #0x38]\n" + "str q24, [x21, x11]\n" + "fmla v29.8h, v8.8h, v13.8h\n" + "fmla v30.8h, v7.8h, v13.8h\n" + "fmax v27.8h, v27.8h, v18.8h\n" + "ldr x21, [x15, #0x28]\n" + "fmla v31.8h, v6.8h, v13.8h\n" "fmin v25.8h, v25.8h, v17.8h\n" - "ldr x22, [x17, #0x40]\n" - "fmin v24.8h, v24.8h, v17.8h\n" - "str q26, [x21, x12]\n" - "fmax v23.8h, v23.8h, v18.8h\n" - "str q25, [x20, x12]\n" - "str q24, [x19, x12]\n" - "fmin v23.8h, v23.8h, v17.8h\n" - "str q23, [x22, x12]\n" + "str q25, [x20, x11]\n" + "ldr x20, [x15, #0x30]\n" + "fmin v26.8h, v26.8h, v17.8h\n" + "fmin v27.8h, v27.8h, v17.8h\n" + "str q26, [x19, x11]\n" + "ldr x19, [x15, #0x38]\n" + "fmax v28.8h, v28.8h, v18.8h\n" + "fmax v29.8h, v29.8h, v18.8h\n" + "str q27, [x22, x11]\n" + "ldr x22, [x15, #0x40]\n" + "fmax v30.8h, v30.8h, v18.8h\n" + "fmax v31.8h, v31.8h, v18.8h\n" + "add x12, x12, #0x10\n" + "fmin v28.8h, v28.8h, v17.8h\n" + "fmin v29.8h, v29.8h, v17.8h\n" + "str q28, [x21, x11]\n" + "fmin v30.8h, v30.8h, v17.8h\n" + "fmin v31.8h, v31.8h, v17.8h\n" + "str q29, [x20, x11]\n" + "str q30, [x19, x11]\n" + "str q31, [x22, x11]\n" "3:" // Oddments - "tst %x[n_channels], #0x1\n" - "beq 48f\n" - "ldr q16, [x15, #0x0]\n" - "ldr q0, [x15, #0x10]\n" - "mov x12, x14\n" - "ldr q1, [x15, #0x20]\n" - "ldr q2, [x15, #0x30]\n" - "ldr q3, [x15, #0x40]\n" - "ldr q4, [x15, #0x50]\n" - "ldr q5, [x15, #0x60]\n" - "ldr q6, [x15, #0x70]\n" - "ldr q7, [x15, #0x80]\n" - "ldr q8, [x15, #0x90]\n" - "ldr x10, [x16, #0x0]\n" - "add x10, x10, x14\n" - "ldr x9, [x16, #0x8]\n" - "ldr x28, [x16, #0x10]\n" - "add x9, x9, x14\n" - "ldr x27, [x16, #0x18]\n" - "ldr x26, [x16, #0x20]\n" - "add x28, x28, x14\n" - "add x27, x27, x14\n" - "add x26, x26, x14\n" + "tst %x[n_channels], #0x7\n" + "beq 92f\n" + "ldr x10, [x13, #0x0]\n" + "ldr x9, [x13, #0x8]\n" + "ldr x28, [x13, #0x10]\n" + "ldr x27, [x13, #0x18]\n" + "mov x11, x12\n" + "add x10, x10, x12\n" + "ldr x26, [x13, #0x20]\n" + "ldr q16, [x14, #0x0]\n" + "add x9, x9, x12\n" + "add x28, x28, x12\n" + "ldr q0, [x14, #0x10]\n" + "ldr q1, [x14, #0x20]\n" + "add x27, x27, x12\n" + "add x26, x26, x12\n" + "ldr q2, [x14, #0x30]\n" + "ldr q3, [x14, #0x40]\n" + "ldr q4, [x14, #0x50]\n" + "ldr q5, [x14, #0x60]\n" + "ldr q6, [x14, #0x70]\n" + "ldr q7, [x14, #0x80]\n" + "ldr q8, [x14, #0x90]\n" + "tbz %x[n_channels], #2, 5f\n" + "ld1 { v9.d }[0], [x10], #0x8\n" + "ld1 { v10.d }[0], [x9], #0x8\n" + "ld1 { v11.d }[0], [x28], #0x8\n" + "ld1 { v12.d }[0], [x27], #0x8\n" + "ld1 { v13.d }[0], [x26], #0x8\n" "tbz %x[n_channels], #1, 4f\n" + "ld1 { v9.s }[2], [x10], #0x4\n" + "ld1 { v10.s }[2], [x9], #0x4\n" + "ld1 { v11.s }[2], [x28], #0x4\n" + "ld1 { v12.s }[2], [x27], #0x4\n" + "ld1 { v13.s }[2], [x26], #0x4\n" + "tbz %x[n_channels], #0, 7f\n" + "ld1 { v9.h }[6], [x10], #0x2\n" + "ld1 { v10.h }[6], [x9], #0x2\n" + "ld1 { v11.h }[6], [x28], #0x2\n" + "ld1 { v12.h }[6], [x27], #0x2\n" + "ld1 { v13.h }[6], [x26], #0x2\n" + "b 7f\n" + "4:" // Oddments: Load inputs (2, 2), (0, 0), (0, 4), (4, 0), (1, 2): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 7f\n" + "ld1 { v9.h }[4], [x10], #0x2\n" + "ld1 { v10.h }[4], [x9], #0x2\n" + "ld1 { v11.h }[4], [x28], #0x2\n" + "ld1 { v12.h }[4], [x27], #0x2\n" + "ld1 { v13.h }[4], [x26], #0x2\n" + "b 7f\n" + "5:" // Oddments: Load inputs (2, 2), (0, 0), (0, 4), (4, 0), (1, 2): Bit 2: Unset + "tbz %x[n_channels], #1, 6f\n" "ld1 { v9.s }[0], [x10], #0x4\n" "ld1 { v10.s }[0], [x9], #0x4\n" "ld1 { v11.s }[0], [x28], #0x4\n" "ld1 { v12.s }[0], [x27], #0x4\n" "ld1 { v13.s }[0], [x26], #0x4\n" - "tbz %x[n_channels], #0, 5f\n" + "tbz %x[n_channels], #0, 7f\n" "ld1 { v9.h }[2], [x10], #0x2\n" "ld1 { v10.h }[2], [x9], #0x2\n" "ld1 { v11.h }[2], [x28], #0x2\n" "ld1 { v12.h }[2], [x27], #0x2\n" "ld1 { v13.h }[2], [x26], #0x2\n" - "b 5f\n" - "4:" // Oddments: Load inputs (2, 2), (0, 0), (0, 4), (4, 0), (1, 2): Bit 1: Unset + "b 7f\n" + "6:" // Oddments: Load inputs (2, 2), (0, 0), (0, 4), (4, 0), (1, 2): Bit 2: Unset: Bit 1: Unset "ld1 { v9.h }[0], [x10], #0x2\n" "ld1 { v10.h }[0], [x9], #0x2\n" "ld1 { v11.h }[0], [x28], #0x2\n" "ld1 { v12.h }[0], [x27], #0x2\n" "ld1 { v13.h }[0], [x26], #0x2\n" - "5:" // Oddments: Load inputs (2, 2), (0, 0), (0, 4), (4, 0), (1, 2): Bit 1: End - "mov v31.16b, v16.16b\n fmla v31.8h, v8.8h, v9.8h\n" - "ldr x25, [x16, #0x28]\n" - "add x25, x25, x14\n" - "mov v30.16b, v16.16b\n fmla v30.8h, v7.8h, v9.8h\n" - "mov v29.16b, v16.16b\n fmla v29.8h, v6.8h, v9.8h\n" - "mov v28.16b, v16.16b\n fmla v28.8h, v5.8h, v9.8h\n" + "7:" // Oddments: Load inputs (2, 2), (0, 0), (0, 4), (4, 0), (1, 2): Bit 2: End + "mov v23.16b, v16.16b\n fmla v23.8h, v8.8h, v9.8h\n" + "mov v25.16b, v16.16b\n fmla v25.8h, v6.8h, v9.8h\n" + "ldr x23, [x13, #0x28]\n" + "add x23, x23, x12\n" + "mov v24.16b, v16.16b\n fmla v24.8h, v7.8h, v9.8h\n" + "mov v26.16b, v16.16b\n fmla v26.8h, v5.8h, v9.8h\n" "mov v27.16b, v16.16b\n fmla v27.8h, v4.8h, v9.8h\n" - "mov v26.16b, v16.16b\n fmla v26.8h, v3.8h, v9.8h\n" - "mov v25.16b, v16.16b\n fmla v25.8h, v2.8h, v9.8h\n" - "mov v24.16b, v16.16b\n fmla v24.8h, v1.8h, v9.8h\n" - "mov v23.16b, v16.16b\n fmla v23.8h, v0.8h, v9.8h\n" - "fmla v31.8h, v0.8h, v10.8h\n" - "fmla v29.8h, v2.8h, v11.8h\n" - "fmla v25.8h, v6.8h, v12.8h\n" - "fmla v30.8h, v4.8h, v13.8h\n" - "fmla v31.8h, v5.8h, v13.8h\n" - "fmla v29.8h, v3.8h, v13.8h\n" - "fmla v28.8h, v2.8h, v13.8h\n" + "mov v28.16b, v16.16b\n fmla v28.8h, v3.8h, v9.8h\n" + "mov v29.16b, v16.16b\n fmla v29.8h, v2.8h, v9.8h\n" + "fmla v23.8h, v0.8h, v10.8h\n" + "fmla v25.8h, v2.8h, v11.8h\n" + "mov v30.16b, v16.16b\n fmla v30.8h, v1.8h, v9.8h\n" + "mov v31.16b, v16.16b\n fmla v31.8h, v0.8h, v9.8h\n" + "fmla v29.8h, v6.8h, v12.8h\n" + "fmla v23.8h, v5.8h, v13.8h\n" + "fmla v24.8h, v4.8h, v13.8h\n" + "fmla v25.8h, v3.8h, v13.8h\n" + "fmla v26.8h, v2.8h, v13.8h\n" "fmla v27.8h, v1.8h, v13.8h\n" - "fmla v26.8h, v0.8h, v13.8h\n" - "tbz %x[n_channels], #1, 6f\n" - "ld1 { v12.s }[0], [x25], #0x4\n" - "tbz %x[n_channels], #0, 7f\n" - "ld1 { v12.h }[2], [x25], #0x2\n" - "b 7f\n" - "6:" // Oddments: Load input (4, 4): Bit 1: Unset - "ld1 { v12.h }[0], [x25], #0x2\n" - "7:" // Oddments: Load input (4, 4): Bit 1: End - "fmla v23.8h, v8.8h, v12.8h\n" - "ldr x24, [x16, #0x30]\n" - "add x24, x24, x14\n" + "fmla v28.8h, v0.8h, v13.8h\n" + "tbz %x[n_channels], #2, 9f\n" + "ld1 { v12.d }[0], [x23], #0x8\n" "tbz %x[n_channels], #1, 8f\n" - "ld1 { v11.s }[0], [x24], #0x4\n" - "tbz %x[n_channels], #0, 9f\n" - "ld1 { v11.h }[2], [x24], #0x2\n" - "b 9f\n" - "8:" // Oddments: Load input (2, 1): Bit 1: Unset - "ld1 { v11.h }[0], [x24], #0x2\n" - "9:" // Oddments: Load input (2, 1): Bit 1: End - "fmla v31.8h, v7.8h, v11.8h\n" - "ldr x23, [x16, #0x38]\n" - "fmla v30.8h, v6.8h, v11.8h\n" - "add x23, x23, x14\n" - "fmla v28.8h, v4.8h, v11.8h\n" - "fmla v27.8h, v3.8h, v11.8h\n" - "fmla v25.8h, v1.8h, v11.8h\n" - "fmla v24.8h, v0.8h, v11.8h\n" + "ld1 { v12.s }[2], [x23], #0x4\n" + "tbz %x[n_channels], #0, 11f\n" + "ld1 { v12.h }[6], [x23], #0x2\n" + "b 11f\n" + "8:" // Oddments: Load input (4, 4): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 11f\n" + "ld1 { v12.h }[4], [x23], #0x2\n" + "b 11f\n" + "9:" // Oddments: Load input (4, 4): Bit 2: Unset "tbz %x[n_channels], #1, 10f\n" - "ld1 { v13.s }[0], [x23], #0x4\n" + "ld1 { v12.s }[0], [x23], #0x4\n" "tbz %x[n_channels], #0, 11f\n" - "ld1 { v13.h }[2], [x23], #0x2\n" + "ld1 { v12.h }[2], [x23], #0x2\n" "b 11f\n" - "10:" // Oddments: Load input (0, 1): Bit 1: Unset - "ld1 { v13.h }[0], [x23], #0x2\n" - "11:" // Oddments: Load input (0, 1): Bit 1: End - "fmla v31.8h, v1.8h, v13.8h\n" - "ldr x10, [x16, #0x40]\n" - "fmla v30.8h, v0.8h, v13.8h\n" - "add x10, x10, x14\n" + "10:" // Oddments: Load input (4, 4): Bit 2: Unset: Bit 1: Unset + "ld1 { v12.h }[0], [x23], #0x2\n" + "11:" // Oddments: Load input (4, 4): Bit 2: End + "ldr x25, [x13, #0x30]\n" + "fmla v31.8h, v8.8h, v12.8h\n" + "add x25, x25, x12\n" + "tbz %x[n_channels], #2, 13f\n" + "ld1 { v11.d }[0], [x25], #0x8\n" "tbz %x[n_channels], #1, 12f\n" + "ld1 { v11.s }[2], [x25], #0x4\n" + "tbz %x[n_channels], #0, 15f\n" + "ld1 { v11.h }[6], [x25], #0x2\n" + "b 15f\n" + "12:" // Oddments: Load input (2, 1): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 15f\n" + "ld1 { v11.h }[4], [x25], #0x2\n" + "b 15f\n" + "13:" // Oddments: Load input (2, 1): Bit 2: Unset + "tbz %x[n_channels], #1, 14f\n" + "ld1 { v11.s }[0], [x25], #0x4\n" + "tbz %x[n_channels], #0, 15f\n" + "ld1 { v11.h }[2], [x25], #0x2\n" + "b 15f\n" + "14:" // Oddments: Load input (2, 1): Bit 2: Unset: Bit 1: Unset + "ld1 { v11.h }[0], [x25], #0x2\n" + "15:" // Oddments: Load input (2, 1): Bit 2: End + "ldr x24, [x13, #0x38]\n" + "fmla v23.8h, v7.8h, v11.8h\n" + "fmla v24.8h, v6.8h, v11.8h\n" + "add x24, x24, x12\n" + "fmla v26.8h, v4.8h, v11.8h\n" + "fmla v27.8h, v3.8h, v11.8h\n" + "fmla v29.8h, v1.8h, v11.8h\n" + "fmla v30.8h, v0.8h, v11.8h\n" + "tbz %x[n_channels], #2, 17f\n" + "ld1 { v13.d }[0], [x24], #0x8\n" + "tbz %x[n_channels], #1, 16f\n" + "ld1 { v13.s }[2], [x24], #0x4\n" + "tbz %x[n_channels], #0, 19f\n" + "ld1 { v13.h }[6], [x24], #0x2\n" + "b 19f\n" + "16:" // Oddments: Load input (0, 1): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 19f\n" + "ld1 { v13.h }[4], [x24], #0x2\n" + "b 19f\n" + "17:" // Oddments: Load input (0, 1): Bit 2: Unset + "tbz %x[n_channels], #1, 18f\n" + "ld1 { v13.s }[0], [x24], #0x4\n" + "tbz %x[n_channels], #0, 19f\n" + "ld1 { v13.h }[2], [x24], #0x2\n" + "b 19f\n" + "18:" // Oddments: Load input (0, 1): Bit 2: Unset: Bit 1: Unset + "ld1 { v13.h }[0], [x24], #0x2\n" + "19:" // Oddments: Load input (0, 1): Bit 2: End + "ldr x10, [x13, #0x40]\n" + "fmla v23.8h, v1.8h, v13.8h\n" + "fmla v24.8h, v0.8h, v13.8h\n" + "add x10, x10, x12\n" + "tbz %x[n_channels], #2, 21f\n" + "ld1 { v12.d }[0], [x10], #0x8\n" + "tbz %x[n_channels], #1, 20f\n" + "ld1 { v12.s }[2], [x10], #0x4\n" + "tbz %x[n_channels], #0, 23f\n" + "ld1 { v12.h }[6], [x10], #0x2\n" + "b 23f\n" + "20:" // Oddments: Load input (0, 3): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 23f\n" + "ld1 { v12.h }[4], [x10], #0x2\n" + "b 23f\n" + "21:" // Oddments: Load input (0, 3): Bit 2: Unset + "tbz %x[n_channels], #1, 22f\n" "ld1 { v12.s }[0], [x10], #0x4\n" - "tbz %x[n_channels], #0, 13f\n" + "tbz %x[n_channels], #0, 23f\n" "ld1 { v12.h }[2], [x10], #0x2\n" - "b 13f\n" - "12:" // Oddments: Load input (0, 3): Bit 1: Unset + "b 23f\n" + "22:" // Oddments: Load input (0, 3): Bit 2: Unset: Bit 1: Unset "ld1 { v12.h }[0], [x10], #0x2\n" - "13:" // Oddments: Load input (0, 3): Bit 1: End - "fmla v30.8h, v2.8h, v12.8h\n" - "ldr x9, [x16, #0x48]\n" - "fmla v29.8h, v1.8h, v12.8h\n" - "add x9, x9, x14\n" - "tbz %x[n_channels], #1, 14f\n" + "23:" // Oddments: Load input (0, 3): Bit 2: End + "ldr x9, [x13, #0x48]\n" + "fmla v24.8h, v2.8h, v12.8h\n" + "fmla v25.8h, v1.8h, v12.8h\n" + "add x9, x9, x12\n" + "tbz %x[n_channels], #2, 25f\n" + "ld1 { v10.d }[0], [x9], #0x8\n" + "tbz %x[n_channels], #1, 24f\n" + "ld1 { v10.s }[2], [x9], #0x4\n" + "tbz %x[n_channels], #0, 27f\n" + "ld1 { v10.h }[6], [x9], #0x2\n" + "b 27f\n" + "24:" // Oddments: Load input (2, 3): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 27f\n" + "ld1 { v10.h }[4], [x9], #0x2\n" + "b 27f\n" + "25:" // Oddments: Load input (2, 3): Bit 2: Unset + "tbz %x[n_channels], #1, 26f\n" "ld1 { v10.s }[0], [x9], #0x4\n" - "tbz %x[n_channels], #0, 15f\n" + "tbz %x[n_channels], #0, 27f\n" "ld1 { v10.h }[2], [x9], #0x2\n" - "b 15f\n" - "14:" // Oddments: Load input (2, 3): Bit 1: Unset + "b 27f\n" + "26:" // Oddments: Load input (2, 3): Bit 2: Unset: Bit 1: Unset "ld1 { v10.h }[0], [x9], #0x2\n" - "15:" // Oddments: Load input (2, 3): Bit 1: End - "fmla v30.8h, v8.8h, v10.8h\n" - "ldr x28, [x16, #0x50]\n" - "fmla v29.8h, v7.8h, v10.8h\n" - "add x28, x28, x14\n" + "27:" // Oddments: Load input (2, 3): Bit 2: End + "ldr x28, [x13, #0x50]\n" + "fmla v24.8h, v8.8h, v10.8h\n" + "fmla v25.8h, v7.8h, v10.8h\n" + "add x28, x28, x12\n" "fmla v27.8h, v5.8h, v10.8h\n" - "fmla v26.8h, v4.8h, v10.8h\n" - "fmla v24.8h, v2.8h, v10.8h\n" - "fmla v23.8h, v1.8h, v10.8h\n" - "tbz %x[n_channels], #1, 16f\n" + "fmla v28.8h, v4.8h, v10.8h\n" + "fmla v30.8h, v2.8h, v10.8h\n" + "fmla v31.8h, v1.8h, v10.8h\n" + "tbz %x[n_channels], #2, 29f\n" + "ld1 { v11.d }[0], [x28], #0x8\n" + "tbz %x[n_channels], #1, 28f\n" + "ld1 { v11.s }[2], [x28], #0x4\n" + "tbz %x[n_channels], #0, 31f\n" + "ld1 { v11.h }[6], [x28], #0x2\n" + "b 31f\n" + "28:" // Oddments: Load input (1, 0): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 31f\n" + "ld1 { v11.h }[4], [x28], #0x2\n" + "b 31f\n" + "29:" // Oddments: Load input (1, 0): Bit 2: Unset + "tbz %x[n_channels], #1, 30f\n" "ld1 { v11.s }[0], [x28], #0x4\n" - "tbz %x[n_channels], #0, 17f\n" + "tbz %x[n_channels], #0, 31f\n" "ld1 { v11.h }[2], [x28], #0x2\n" - "b 17f\n" - "16:" // Oddments: Load input (1, 0): Bit 1: Unset + "b 31f\n" + "30:" // Oddments: Load input (1, 0): Bit 2: Unset: Bit 1: Unset "ld1 { v11.h }[0], [x28], #0x2\n" - "17:" // Oddments: Load input (1, 0): Bit 1: End - "fmla v31.8h, v3.8h, v11.8h\n" - "ldr x27, [x16, #0x58]\n" - "fmla v28.8h, v0.8h, v11.8h\n" - "add x27, x27, x14\n" - "tbz %x[n_channels], #1, 18f\n" + "31:" // Oddments: Load input (1, 0): Bit 2: End + "ldr x27, [x13, #0x58]\n" + "fmla v23.8h, v3.8h, v11.8h\n" + "fmla v26.8h, v0.8h, v11.8h\n" + "add x27, x27, x12\n" + "tbz %x[n_channels], #2, 33f\n" + "ld1 { v13.d }[0], [x27], #0x8\n" + "tbz %x[n_channels], #1, 32f\n" + "ld1 { v13.s }[2], [x27], #0x4\n" + "tbz %x[n_channels], #0, 35f\n" + "ld1 { v13.h }[6], [x27], #0x2\n" + "b 35f\n" + "32:" // Oddments: Load input (1, 4): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 35f\n" + "ld1 { v13.h }[4], [x27], #0x2\n" + "b 35f\n" + "33:" // Oddments: Load input (1, 4): Bit 2: Unset + "tbz %x[n_channels], #1, 34f\n" "ld1 { v13.s }[0], [x27], #0x4\n" - "tbz %x[n_channels], #0, 19f\n" + "tbz %x[n_channels], #0, 35f\n" "ld1 { v13.h }[2], [x27], #0x2\n" - "b 19f\n" - "18:" // Oddments: Load input (1, 4): Bit 1: Unset + "b 35f\n" + "34:" // Oddments: Load input (1, 4): Bit 2: Unset: Bit 1: Unset "ld1 { v13.h }[0], [x27], #0x2\n" - "19:" // Oddments: Load input (1, 4): Bit 1: End - "fmla v29.8h, v5.8h, v13.8h\n" - "ldr x26, [x16, #0x60]\n" - "fmla v26.8h, v2.8h, v13.8h\n" - "add x26, x26, x14\n" - "tbz %x[n_channels], #1, 20f\n" + "35:" // Oddments: Load input (1, 4): Bit 2: End + "ldr x26, [x13, #0x60]\n" + "fmla v25.8h, v5.8h, v13.8h\n" + "fmla v28.8h, v2.8h, v13.8h\n" + "add x26, x26, x12\n" + "tbz %x[n_channels], #2, 37f\n" + "ld1 { v12.d }[0], [x26], #0x8\n" + "tbz %x[n_channels], #1, 36f\n" + "ld1 { v12.s }[2], [x26], #0x4\n" + "tbz %x[n_channels], #0, 39f\n" + "ld1 { v12.h }[6], [x26], #0x2\n" + "b 39f\n" + "36:" // Oddments: Load input (3, 0): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 39f\n" + "ld1 { v12.h }[4], [x26], #0x2\n" + "b 39f\n" + "37:" // Oddments: Load input (3, 0): Bit 2: Unset + "tbz %x[n_channels], #1, 38f\n" "ld1 { v12.s }[0], [x26], #0x4\n" - "tbz %x[n_channels], #0, 21f\n" + "tbz %x[n_channels], #0, 39f\n" "ld1 { v12.h }[2], [x26], #0x2\n" - "b 21f\n" - "20:" // Oddments: Load input (3, 0): Bit 1: Unset + "b 39f\n" + "38:" // Oddments: Load input (3, 0): Bit 2: Unset: Bit 1: Unset "ld1 { v12.h }[0], [x26], #0x2\n" - "21:" // Oddments: Load input (3, 0): Bit 1: End - "fmla v28.8h, v6.8h, v12.8h\n" - "ldr x25, [x16, #0x68]\n" - "fmla v25.8h, v3.8h, v12.8h\n" - "add x25, x25, x14\n" - "tbz %x[n_channels], #1, 22f\n" - "ld1 { v10.s }[0], [x25], #0x4\n" - "tbz %x[n_channels], #0, 23f\n" - "ld1 { v10.h }[2], [x25], #0x2\n" - "b 23f\n" - "22:" // Oddments: Load input (3, 2): Bit 1: Unset - "ld1 { v10.h }[0], [x25], #0x2\n" - "23:" // Oddments: Load input (3, 2): Bit 1: End - "fmla v28.8h, v8.8h, v10.8h\n" - "ldr x24, [x16, #0x70]\n" + "39:" // Oddments: Load input (3, 0): Bit 2: End + "ldr x23, [x13, #0x68]\n" + "fmla v26.8h, v6.8h, v12.8h\n" + "fmla v29.8h, v3.8h, v12.8h\n" + "add x23, x23, x12\n" + "tbz %x[n_channels], #2, 41f\n" + "ld1 { v10.d }[0], [x23], #0x8\n" + "tbz %x[n_channels], #1, 40f\n" + "ld1 { v10.s }[2], [x23], #0x4\n" + "tbz %x[n_channels], #0, 43f\n" + "ld1 { v10.h }[6], [x23], #0x2\n" + "b 43f\n" + "40:" // Oddments: Load input (3, 2): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 43f\n" + "ld1 { v10.h }[4], [x23], #0x2\n" + "b 43f\n" + "41:" // Oddments: Load input (3, 2): Bit 2: Unset + "tbz %x[n_channels], #1, 42f\n" + "ld1 { v10.s }[0], [x23], #0x4\n" + "tbz %x[n_channels], #0, 43f\n" + "ld1 { v10.h }[2], [x23], #0x2\n" + "b 43f\n" + "42:" // Oddments: Load input (3, 2): Bit 2: Unset: Bit 1: Unset + "ld1 { v10.h }[0], [x23], #0x2\n" + "43:" // Oddments: Load input (3, 2): Bit 2: End + "ldr x25, [x13, #0x70]\n" + "fmla v26.8h, v8.8h, v10.8h\n" "fmla v27.8h, v7.8h, v10.8h\n" - "add x24, x24, x14\n" - "fmla v26.8h, v6.8h, v10.8h\n" - "fmla v25.8h, v5.8h, v10.8h\n" - "fmla v24.8h, v4.8h, v10.8h\n" - "fmla v23.8h, v3.8h, v10.8h\n" - "tbz %x[n_channels], #1, 24f\n" - "ld1 { v11.s }[0], [x24], #0x4\n" - "tbz %x[n_channels], #0, 25f\n" - "ld1 { v11.h }[2], [x24], #0x2\n" - "b 25f\n" - "24:" // Oddments: Load input (3, 4): Bit 1: Unset - "ld1 { v11.h }[0], [x24], #0x2\n" - "25:" // Oddments: Load input (3, 4): Bit 1: End - "fmla v26.8h, v8.8h, v11.8h\n" - "ldr x23, [x16, #0x78]\n" - "fmla v23.8h, v5.8h, v11.8h\n" - "add x23, x23, x14\n" - "tbz %x[n_channels], #1, 26f\n" - "ld1 { v13.s }[0], [x23], #0x4\n" - "tbz %x[n_channels], #0, 27f\n" - "ld1 { v13.h }[2], [x23], #0x2\n" - "b 27f\n" - "26:" // Oddments: Load input (4, 1): Bit 1: Unset - "ld1 { v13.h }[0], [x23], #0x2\n" - "27:" // Oddments: Load input (4, 1): Bit 1: End - "fmla v25.8h, v7.8h, v13.8h\n" - "ldr x10, [x16, #0x80]\n" - "fmla v24.8h, v6.8h, v13.8h\n" - "add x10, x10, x14\n" - "tbz %x[n_channels], #1, 28f\n" + "add x25, x25, x12\n" + "fmla v28.8h, v6.8h, v10.8h\n" + "fmla v29.8h, v5.8h, v10.8h\n" + "fmla v30.8h, v4.8h, v10.8h\n" + "fmla v31.8h, v3.8h, v10.8h\n" + "tbz %x[n_channels], #2, 45f\n" + "ld1 { v11.d }[0], [x25], #0x8\n" + "tbz %x[n_channels], #1, 44f\n" + "ld1 { v11.s }[2], [x25], #0x4\n" + "tbz %x[n_channels], #0, 47f\n" + "ld1 { v11.h }[6], [x25], #0x2\n" + "b 47f\n" + "44:" // Oddments: Load input (3, 4): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 47f\n" + "ld1 { v11.h }[4], [x25], #0x2\n" + "b 47f\n" + "45:" // Oddments: Load input (3, 4): Bit 2: Unset + "tbz %x[n_channels], #1, 46f\n" + "ld1 { v11.s }[0], [x25], #0x4\n" + "tbz %x[n_channels], #0, 47f\n" + "ld1 { v11.h }[2], [x25], #0x2\n" + "b 47f\n" + "46:" // Oddments: Load input (3, 4): Bit 2: Unset: Bit 1: Unset + "ld1 { v11.h }[0], [x25], #0x2\n" + "47:" // Oddments: Load input (3, 4): Bit 2: End + "ldr x24, [x13, #0x78]\n" + "fmla v28.8h, v8.8h, v11.8h\n" + "fmla v31.8h, v5.8h, v11.8h\n" + "add x24, x24, x12\n" + "tbz %x[n_channels], #2, 49f\n" + "ld1 { v13.d }[0], [x24], #0x8\n" + "tbz %x[n_channels], #1, 48f\n" + "ld1 { v13.s }[2], [x24], #0x4\n" + "tbz %x[n_channels], #0, 51f\n" + "ld1 { v13.h }[6], [x24], #0x2\n" + "b 51f\n" + "48:" // Oddments: Load input (4, 1): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 51f\n" + "ld1 { v13.h }[4], [x24], #0x2\n" + "b 51f\n" + "49:" // Oddments: Load input (4, 1): Bit 2: Unset + "tbz %x[n_channels], #1, 50f\n" + "ld1 { v13.s }[0], [x24], #0x4\n" + "tbz %x[n_channels], #0, 51f\n" + "ld1 { v13.h }[2], [x24], #0x2\n" + "b 51f\n" + "50:" // Oddments: Load input (4, 1): Bit 2: Unset: Bit 1: Unset + "ld1 { v13.h }[0], [x24], #0x2\n" + "51:" // Oddments: Load input (4, 1): Bit 2: End + "ldr x10, [x13, #0x80]\n" + "fmla v29.8h, v7.8h, v13.8h\n" + "fmla v30.8h, v6.8h, v13.8h\n" + "add x10, x10, x12\n" + "tbz %x[n_channels], #2, 53f\n" + "ld1 { v12.d }[0], [x10], #0x8\n" + "tbz %x[n_channels], #1, 52f\n" + "ld1 { v12.s }[2], [x10], #0x4\n" + "tbz %x[n_channels], #0, 55f\n" + "ld1 { v12.h }[6], [x10], #0x2\n" + "b 55f\n" + "52:" // Oddments: Load input (1, 1): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 55f\n" + "ld1 { v12.h }[4], [x10], #0x2\n" + "b 55f\n" + "53:" // Oddments: Load input (1, 1): Bit 2: Unset + "tbz %x[n_channels], #1, 54f\n" "ld1 { v12.s }[0], [x10], #0x4\n" - "tbz %x[n_channels], #0, 29f\n" + "tbz %x[n_channels], #0, 55f\n" "ld1 { v12.h }[2], [x10], #0x2\n" - "b 29f\n" - "28:" // Oddments: Load input (1, 1): Bit 1: Unset + "b 55f\n" + "54:" // Oddments: Load input (1, 1): Bit 2: Unset: Bit 1: Unset "ld1 { v12.h }[0], [x10], #0x2\n" - "29:" // Oddments: Load input (1, 1): Bit 1: End - "fmla v31.8h, v4.8h, v12.8h\n" - "ldr x9, [x16, #0x88]\n" - "fmla v30.8h, v3.8h, v12.8h\n" - "add x9, x9, x14\n" - "fmla v28.8h, v1.8h, v12.8h\n" + "55:" // Oddments: Load input (1, 1): Bit 2: End + "ldr x9, [x13, #0x88]\n" + "fmla v23.8h, v4.8h, v12.8h\n" + "fmla v24.8h, v3.8h, v12.8h\n" + "add x9, x9, x12\n" + "fmla v26.8h, v1.8h, v12.8h\n" "fmla v27.8h, v0.8h, v12.8h\n" - "tbz %x[n_channels], #1, 30f\n" + "tbz %x[n_channels], #2, 57f\n" + "ld1 { v11.d }[0], [x9], #0x8\n" + "tbz %x[n_channels], #1, 56f\n" + "ld1 { v11.s }[2], [x9], #0x4\n" + "tbz %x[n_channels], #0, 59f\n" + "ld1 { v11.h }[6], [x9], #0x2\n" + "b 59f\n" + "56:" // Oddments: Load input (1, 3): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 59f\n" + "ld1 { v11.h }[4], [x9], #0x2\n" + "b 59f\n" + "57:" // Oddments: Load input (1, 3): Bit 2: Unset + "tbz %x[n_channels], #1, 58f\n" "ld1 { v11.s }[0], [x9], #0x4\n" - "tbz %x[n_channels], #0, 31f\n" + "tbz %x[n_channels], #0, 59f\n" "ld1 { v11.h }[2], [x9], #0x2\n" - "b 31f\n" - "30:" // Oddments: Load input (1, 3): Bit 1: Unset + "b 59f\n" + "58:" // Oddments: Load input (1, 3): Bit 2: Unset: Bit 1: Unset "ld1 { v11.h }[0], [x9], #0x2\n" - "31:" // Oddments: Load input (1, 3): Bit 1: End - "fmla v30.8h, v5.8h, v11.8h\n" - "ldr x28, [x16, #0x90]\n" - "fmla v29.8h, v4.8h, v11.8h\n" - "add x28, x28, x14\n" + "59:" // Oddments: Load input (1, 3): Bit 2: End + "ldr x28, [x13, #0x90]\n" + "fmla v24.8h, v5.8h, v11.8h\n" + "fmla v25.8h, v4.8h, v11.8h\n" + "add x28, x28, x12\n" "fmla v27.8h, v2.8h, v11.8h\n" - "fmla v26.8h, v1.8h, v11.8h\n" - "tbz %x[n_channels], #1, 32f\n" + "fmla v28.8h, v1.8h, v11.8h\n" + "tbz %x[n_channels], #2, 61f\n" + "ld1 { v13.d }[0], [x28], #0x8\n" + "tbz %x[n_channels], #1, 60f\n" + "ld1 { v13.s }[2], [x28], #0x4\n" + "tbz %x[n_channels], #0, 63f\n" + "ld1 { v13.h }[6], [x28], #0x2\n" + "b 63f\n" + "60:" // Oddments: Load input (4, 3): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 63f\n" + "ld1 { v13.h }[4], [x28], #0x2\n" + "b 63f\n" + "61:" // Oddments: Load input (4, 3): Bit 2: Unset + "tbz %x[n_channels], #1, 62f\n" "ld1 { v13.s }[0], [x28], #0x4\n" - "tbz %x[n_channels], #0, 33f\n" + "tbz %x[n_channels], #0, 63f\n" "ld1 { v13.h }[2], [x28], #0x2\n" - "b 33f\n" - "32:" // Oddments: Load input (4, 3): Bit 1: Unset + "b 63f\n" + "62:" // Oddments: Load input (4, 3): Bit 2: Unset: Bit 1: Unset "ld1 { v13.h }[0], [x28], #0x2\n" - "33:" // Oddments: Load input (4, 3): Bit 1: End - "fmla v24.8h, v8.8h, v13.8h\n" - "ldr x27, [x16, #0x98]\n" - "fmla v23.8h, v7.8h, v13.8h\n" - "add x27, x27, x14\n" - "tbz %x[n_channels], #1, 34f\n" + "63:" // Oddments: Load input (4, 3): Bit 2: End + "ldr x27, [x13, #0x98]\n" + "fmla v30.8h, v8.8h, v13.8h\n" + "fmla v31.8h, v7.8h, v13.8h\n" + "add x27, x27, x12\n" + "tbz %x[n_channels], #2, 65f\n" + "ld1 { v12.d }[0], [x27], #0x8\n" + "tbz %x[n_channels], #1, 64f\n" + "ld1 { v12.s }[2], [x27], #0x4\n" + "tbz %x[n_channels], #0, 67f\n" + "ld1 { v12.h }[6], [x27], #0x2\n" + "b 67f\n" + "64:" // Oddments: Load input (3, 1): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 67f\n" + "ld1 { v12.h }[4], [x27], #0x2\n" + "b 67f\n" + "65:" // Oddments: Load input (3, 1): Bit 2: Unset + "tbz %x[n_channels], #1, 66f\n" "ld1 { v12.s }[0], [x27], #0x4\n" - "tbz %x[n_channels], #0, 35f\n" + "tbz %x[n_channels], #0, 67f\n" "ld1 { v12.h }[2], [x27], #0x2\n" - "b 35f\n" - "34:" // Oddments: Load input (3, 1): Bit 1: Unset + "b 67f\n" + "66:" // Oddments: Load input (3, 1): Bit 2: Unset: Bit 1: Unset "ld1 { v12.h }[0], [x27], #0x2\n" - "35:" // Oddments: Load input (3, 1): Bit 1: End - "fmla v28.8h, v7.8h, v12.8h\n" - "ldr x26, [x16, #0xa0]\n" + "67:" // Oddments: Load input (3, 1): Bit 2: End + "ldr x26, [x13, #0xa0]\n" + "fmla v26.8h, v7.8h, v12.8h\n" "fmla v27.8h, v6.8h, v12.8h\n" - "add x26, x26, x14\n" - "fmla v25.8h, v4.8h, v12.8h\n" - "fmla v24.8h, v3.8h, v12.8h\n" - "tbz %x[n_channels], #1, 36f\n" + "add x26, x26, x12\n" + "fmla v29.8h, v4.8h, v12.8h\n" + "fmla v30.8h, v3.8h, v12.8h\n" + "tbz %x[n_channels], #2, 69f\n" + "ld1 { v11.d }[0], [x26], #0x8\n" + "tbz %x[n_channels], #1, 68f\n" + "ld1 { v11.s }[2], [x26], #0x4\n" + "tbz %x[n_channels], #0, 71f\n" + "ld1 { v11.h }[6], [x26], #0x2\n" + "b 71f\n" + "68:" // Oddments: Load input (0, 2): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 71f\n" + "ld1 { v11.h }[4], [x26], #0x2\n" + "b 71f\n" + "69:" // Oddments: Load input (0, 2): Bit 2: Unset + "tbz %x[n_channels], #1, 70f\n" "ld1 { v11.s }[0], [x26], #0x4\n" - "tbz %x[n_channels], #0, 37f\n" + "tbz %x[n_channels], #0, 71f\n" "ld1 { v11.h }[2], [x26], #0x2\n" - "b 37f\n" - "36:" // Oddments: Load input (0, 2): Bit 1: Unset + "b 71f\n" + "70:" // Oddments: Load input (0, 2): Bit 2: Unset: Bit 1: Unset "ld1 { v11.h }[0], [x26], #0x2\n" - "37:" // Oddments: Load input (0, 2): Bit 1: End - "fmla v31.8h, v2.8h, v11.8h\n" - "ldr x25, [x16, #0xa8]\n" - "fmla v30.8h, v1.8h, v11.8h\n" - "add x25, x25, x14\n" - "fmla v29.8h, v0.8h, v11.8h\n" - "tbz %x[n_channels], #1, 38f\n" - "ld1 { v13.s }[0], [x25], #0x4\n" - "tbz %x[n_channels], #0, 39f\n" - "ld1 { v13.h }[2], [x25], #0x2\n" - "b 39f\n" - "38:" // Oddments: Load input (3, 3): Bit 1: Unset - "ld1 { v13.h }[0], [x25], #0x2\n" - "39:" // Oddments: Load input (3, 3): Bit 1: End - "fmla v27.8h, v8.8h, v13.8h\n" - "ldr x24, [x16, #0xb0]\n" - "fmla v26.8h, v7.8h, v13.8h\n" - "add x24, x24, x14\n" - "fmla v24.8h, v5.8h, v13.8h\n" - "fmla v23.8h, v4.8h, v13.8h\n" - "tbz %x[n_channels], #1, 40f\n" - "ld1 { v12.s }[0], [x24], #0x4\n" - "tbz %x[n_channels], #0, 41f\n" - "ld1 { v12.h }[2], [x24], #0x2\n" - "b 41f\n" - "40:" // Oddments: Load input (2, 0): Bit 1: Unset - "ld1 { v12.h }[0], [x24], #0x2\n" - "41:" // Oddments: Load input (2, 0): Bit 1: End - "fmla v31.8h, v6.8h, v12.8h\n" - "ldr x23, [x16, #0xb8]\n" - "fmla v28.8h, v3.8h, v12.8h\n" - "add x23, x23, x14\n" - "fmla v25.8h, v0.8h, v12.8h\n" - "tbz %x[n_channels], #1, 42f\n" - "ld1 { v11.s }[0], [x23], #0x4\n" - "tbz %x[n_channels], #0, 43f\n" - "ld1 { v11.h }[2], [x23], #0x2\n" - "b 43f\n" - "42:" // Oddments: Load input (2, 4): Bit 1: Unset - "ld1 { v11.h }[0], [x23], #0x2\n" - "43:" // Oddments: Load input (2, 4): Bit 1: End - "fmla v29.8h, v8.8h, v11.8h\n" - "ldr x10, [x16, #0xc0]\n" - "fmla v26.8h, v5.8h, v11.8h\n" - "add x10, x10, x14\n" + "71:" // Oddments: Load input (0, 2): Bit 2: End + "ldr x23, [x13, #0xa8]\n" "fmla v23.8h, v2.8h, v11.8h\n" - "tbz %x[n_channels], #1, 44f\n" + "fmla v24.8h, v1.8h, v11.8h\n" + "add x23, x23, x12\n" + "fmla v25.8h, v0.8h, v11.8h\n" + "tbz %x[n_channels], #2, 73f\n" + "ld1 { v13.d }[0], [x23], #0x8\n" + "tbz %x[n_channels], #1, 72f\n" + "ld1 { v13.s }[2], [x23], #0x4\n" + "tbz %x[n_channels], #0, 75f\n" + "ld1 { v13.h }[6], [x23], #0x2\n" + "b 75f\n" + "72:" // Oddments: Load input (3, 3): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 75f\n" + "ld1 { v13.h }[4], [x23], #0x2\n" + "b 75f\n" + "73:" // Oddments: Load input (3, 3): Bit 2: Unset + "tbz %x[n_channels], #1, 74f\n" + "ld1 { v13.s }[0], [x23], #0x4\n" + "tbz %x[n_channels], #0, 75f\n" + "ld1 { v13.h }[2], [x23], #0x2\n" + "b 75f\n" + "74:" // Oddments: Load input (3, 3): Bit 2: Unset: Bit 1: Unset + "ld1 { v13.h }[0], [x23], #0x2\n" + "75:" // Oddments: Load input (3, 3): Bit 2: End + "ldr x25, [x13, #0xb0]\n" + "fmla v27.8h, v8.8h, v13.8h\n" + "fmla v28.8h, v7.8h, v13.8h\n" + "add x25, x25, x12\n" + "fmla v30.8h, v5.8h, v13.8h\n" + "fmla v31.8h, v4.8h, v13.8h\n" + "tbz %x[n_channels], #2, 77f\n" + "ld1 { v12.d }[0], [x25], #0x8\n" + "tbz %x[n_channels], #1, 76f\n" + "ld1 { v12.s }[2], [x25], #0x4\n" + "tbz %x[n_channels], #0, 79f\n" + "ld1 { v12.h }[6], [x25], #0x2\n" + "b 79f\n" + "76:" // Oddments: Load input (2, 0): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 79f\n" + "ld1 { v12.h }[4], [x25], #0x2\n" + "b 79f\n" + "77:" // Oddments: Load input (2, 0): Bit 2: Unset + "tbz %x[n_channels], #1, 78f\n" + "ld1 { v12.s }[0], [x25], #0x4\n" + "tbz %x[n_channels], #0, 79f\n" + "ld1 { v12.h }[2], [x25], #0x2\n" + "b 79f\n" + "78:" // Oddments: Load input (2, 0): Bit 2: Unset: Bit 1: Unset + "ld1 { v12.h }[0], [x25], #0x2\n" + "79:" // Oddments: Load input (2, 0): Bit 2: End + "ldr x24, [x13, #0xb8]\n" + "fmla v23.8h, v6.8h, v12.8h\n" + "fmla v26.8h, v3.8h, v12.8h\n" + "add x24, x24, x12\n" + "fmla v29.8h, v0.8h, v12.8h\n" + "tbz %x[n_channels], #2, 81f\n" + "ld1 { v11.d }[0], [x24], #0x8\n" + "tbz %x[n_channels], #1, 80f\n" + "ld1 { v11.s }[2], [x24], #0x4\n" + "tbz %x[n_channels], #0, 83f\n" + "ld1 { v11.h }[6], [x24], #0x2\n" + "b 83f\n" + "80:" // Oddments: Load input (2, 4): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 83f\n" + "ld1 { v11.h }[4], [x24], #0x2\n" + "b 83f\n" + "81:" // Oddments: Load input (2, 4): Bit 2: Unset + "tbz %x[n_channels], #1, 82f\n" + "ld1 { v11.s }[0], [x24], #0x4\n" + "tbz %x[n_channels], #0, 83f\n" + "ld1 { v11.h }[2], [x24], #0x2\n" + "b 83f\n" + "82:" // Oddments: Load input (2, 4): Bit 2: Unset: Bit 1: Unset + "ld1 { v11.h }[0], [x24], #0x2\n" + "83:" // Oddments: Load input (2, 4): Bit 2: End + "ldr x10, [x13, #0xc0]\n" + "fmla v25.8h, v8.8h, v11.8h\n" + "fmla v28.8h, v5.8h, v11.8h\n" + "add x10, x10, x12\n" + "fmla v31.8h, v2.8h, v11.8h\n" + "tbz %x[n_channels], #2, 85f\n" + "ld1 { v13.d }[0], [x10], #0x8\n" + "tbz %x[n_channels], #1, 84f\n" + "ld1 { v13.s }[2], [x10], #0x4\n" + "tbz %x[n_channels], #0, 87f\n" + "ld1 { v13.h }[6], [x10], #0x2\n" + "b 87f\n" + "84:" // Oddments: Load input (4, 2): Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 87f\n" + "ld1 { v13.h }[4], [x10], #0x2\n" + "b 87f\n" + "85:" // Oddments: Load input (4, 2): Bit 2: Unset + "tbz %x[n_channels], #1, 86f\n" "ld1 { v13.s }[0], [x10], #0x4\n" - "tbz %x[n_channels], #0, 45f\n" + "tbz %x[n_channels], #0, 87f\n" "ld1 { v13.h }[2], [x10], #0x2\n" - "b 45f\n" - "44:" // Oddments: Load input (4, 2): Bit 1: Unset + "b 87f\n" + "86:" // Oddments: Load input (4, 2): Bit 2: Unset: Bit 1: Unset "ld1 { v13.h }[0], [x10], #0x2\n" - "45:" // Oddments: Load input (4, 2): Bit 1: End - "fmla v25.8h, v8.8h, v13.8h\n" - "fmla v24.8h, v7.8h, v13.8h\n" - "fmla v23.8h, v6.8h, v13.8h\n" - "fmax v31.8h, v31.8h, v18.8h\n" - "fmax v30.8h, v30.8h, v18.8h\n" - "fmax v29.8h, v29.8h, v18.8h\n" - "fmin v31.8h, v31.8h, v17.8h\n" - "fmin v30.8h, v30.8h, v17.8h\n" - "fmin v29.8h, v29.8h, v17.8h\n" - "fmax v28.8h, v28.8h, v18.8h\n" - "fmax v27.8h, v27.8h, v18.8h\n" - "fmax v26.8h, v26.8h, v18.8h\n" - "fmin v28.8h, v28.8h, v17.8h\n" - "fmin v27.8h, v27.8h, v17.8h\n" - "fmin v26.8h, v26.8h, v17.8h\n" - "fmax v25.8h, v25.8h, v18.8h\n" - "fmax v24.8h, v24.8h, v18.8h\n" + "87:" // Oddments: Load input (4, 2): Bit 2: End + "fmla v29.8h, v8.8h, v13.8h\n" + "fmla v30.8h, v7.8h, v13.8h\n" "fmax v23.8h, v23.8h, v18.8h\n" - "fmin v25.8h, v25.8h, v17.8h\n" - "fmin v24.8h, v24.8h, v17.8h\n" + "fmla v31.8h, v6.8h, v13.8h\n" + "fmax v24.8h, v24.8h, v18.8h\n" + "fmax v25.8h, v25.8h, v18.8h\n" + "fmax v26.8h, v26.8h, v18.8h\n" + "fmax v27.8h, v27.8h, v18.8h\n" + "fmax v28.8h, v28.8h, v18.8h\n" + "fmax v29.8h, v29.8h, v18.8h\n" + "fmax v30.8h, v30.8h, v18.8h\n" + "fmax v31.8h, v31.8h, v18.8h\n" "fmin v23.8h, v23.8h, v17.8h\n" - "tbz %x[n_channels], #1, 46f\n" - "ldr x22, [x17, #0x0]\n" - "ldr x21, [x17, #0x8]\n" - "add x22, x22, x12\n" - "ldr x20, [x17, #0x10]\n" - "ldr x19, [x17, #0x18]\n" - "add x21, x21, x12\n" - "st1 { v31.s }[0], [x22]\n" - "add x20, x20, x12\n" - "st1 { v30.s }[0], [x21]\n" - "ldr x22, [x17, #0x20]\n" - "add x19, x19, x12\n" - "st1 { v29.s }[0], [x20]\n" - "add x22, x22, x12\n" - "st1 { v28.s }[0], [x19]\n" - "ldr x21, [x17, #0x28]\n" - "add x21, x21, x12\n" - "st1 { v27.s }[0], [x22]\n" - "ldr x20, [x17, #0x30]\n" - "add x20, x20, x12\n" - "st1 { v26.s }[0], [x21]\n" - "ldr x19, [x17, #0x38]\n" - "add x19, x19, x12\n" - "st1 { v25.s }[0], [x20]\n" - "ldr x22, [x17, #0x40]\n" - "add x22, x22, x12\n" - "st1 { v24.s }[0], [x19]\n" - "add x12, x12, #0x4\n" + "fmin v24.8h, v24.8h, v17.8h\n" + "fmin v25.8h, v25.8h, v17.8h\n" + "fmin v26.8h, v26.8h, v17.8h\n" + "fmin v27.8h, v27.8h, v17.8h\n" + "fmin v28.8h, v28.8h, v17.8h\n" + "fmin v29.8h, v29.8h, v17.8h\n" + "fmin v30.8h, v30.8h, v17.8h\n" + "fmin v31.8h, v31.8h, v17.8h\n" + "tbz %x[n_channels], #2, 89f\n" + "ldr x22, [x15, #0x0]\n" + "add x22, x22, x11\n" + "st1 { v23.d }[0], [x22]\n" + "ldr x21, [x15, #0x8]\n" + "ldr x20, [x15, #0x10]\n" + "ldr x19, [x15, #0x18]\n" + "ldr x22, [x15, #0x20]\n" + "add x21, x21, x11\n" + "add x20, x20, x11\n" + "add x19, x19, x11\n" + "add x22, x22, x11\n" + "st1 { v24.d }[0], [x21]\n" + "ldr x21, [x15, #0x28]\n" + "st1 { v25.d }[0], [x20]\n" + "ldr x20, [x15, #0x30]\n" + "add x21, x21, x11\n" + "add x20, x20, x11\n" + "st1 { v26.d }[0], [x19]\n" + "ldr x19, [x15, #0x38]\n" + "add x19, x19, x11\n" + "st1 { v27.d }[0], [x22]\n" + "ldr x22, [x15, #0x40]\n" + "add x22, x22, x11\n" + "add x11, x11, #0x8\n" + "st1 { v28.d }[0], [x21]\n" + "st1 { v29.d }[0], [x20]\n" + "st1 { v30.d }[0], [x19]\n" + "st1 { v31.d }[0], [x22]\n" + "tbz %x[n_channels], #1, 88f\n" + "ldr x22, [x15, #0x0]\n" + "add x22, x22, x11\n" + "st1 { v23.s }[2], [x22]\n" + "ldr x21, [x15, #0x8]\n" + "ldr x20, [x15, #0x10]\n" + "add x21, x21, x11\n" + "ldr x19, [x15, #0x18]\n" + "ldr x22, [x15, #0x20]\n" + "add x20, x20, x11\n" + "add x19, x19, x11\n" + "add x22, x22, x11\n" + "st1 { v24.s }[2], [x21]\n" + "ldr x21, [x15, #0x28]\n" + "add x21, x21, x11\n" + "st1 { v25.s }[2], [x20]\n" + "ldr x20, [x15, #0x30]\n" + "add x20, x20, x11\n" + "st1 { v26.s }[2], [x19]\n" + "ldr x19, [x15, #0x38]\n" + "add x19, x19, x11\n" + "st1 { v27.s }[2], [x22]\n" + "ldr x22, [x15, #0x40]\n" + "add x22, x22, x11\n" + "add x11, x11, #0x4\n" + "st1 { v28.s }[2], [x21]\n" + "st1 { v29.s }[2], [x20]\n" + "st1 { v30.s }[2], [x19]\n" + "st1 { v31.s }[2], [x22]\n" + "tbz %x[n_channels], #0, 91f\n" + "ldr x22, [x15, #0x0]\n" + "add x22, x22, x11\n" + "st1 { v23.h }[6], [x22]\n" + "ldr x21, [x15, #0x8]\n" + "ldr x20, [x15, #0x10]\n" + "add x21, x21, x11\n" + "ldr x19, [x15, #0x18]\n" + "ldr x22, [x15, #0x20]\n" + "add x20, x20, x11\n" + "add x19, x19, x11\n" + "add x22, x22, x11\n" + "st1 { v24.h }[6], [x21]\n" + "ldr x21, [x15, #0x28]\n" + "add x21, x21, x11\n" + "st1 { v25.h }[6], [x20]\n" + "ldr x20, [x15, #0x30]\n" + "add x20, x20, x11\n" + "st1 { v26.h }[6], [x19]\n" + "ldr x19, [x15, #0x38]\n" + "add x19, x19, x11\n" + "st1 { v27.h }[6], [x22]\n" + "ldr x22, [x15, #0x40]\n" + "add x22, x22, x11\n" + "st1 { v28.h }[6], [x21]\n" + "st1 { v29.h }[6], [x20]\n" + "st1 { v30.h }[6], [x19]\n" + "st1 { v31.h }[6], [x22]\n" + "b 91f\n" + "88:" // Oddments: Store: Bit 2: Bit 1: Unset + "tbz %x[n_channels], #0, 91f\n" + "ldr x22, [x15, #0x0]\n" + "add x22, x22, x11\n" + "st1 { v23.h }[4], [x22]\n" + "ldr x21, [x15, #0x8]\n" + "ldr x20, [x15, #0x10]\n" + "add x21, x21, x11\n" + "add x20, x20, x11\n" + "ldr x19, [x15, #0x18]\n" + "ldr x22, [x15, #0x20]\n" + "add x19, x19, x11\n" + "add x22, x22, x11\n" + "st1 { v24.h }[4], [x21]\n" + "ldr x21, [x15, #0x28]\n" + "add x21, x21, x11\n" + "st1 { v25.h }[4], [x20]\n" + "ldr x20, [x15, #0x30]\n" + "add x20, x20, x11\n" + "st1 { v26.h }[4], [x19]\n" + "ldr x19, [x15, #0x38]\n" + "add x19, x19, x11\n" + "st1 { v27.h }[4], [x22]\n" + "ldr x22, [x15, #0x40]\n" + "add x22, x22, x11\n" + "st1 { v28.h }[4], [x21]\n" + "st1 { v29.h }[4], [x20]\n" + "st1 { v30.h }[4], [x19]\n" + "st1 { v31.h }[4], [x22]\n" + "b 91f\n" + "89:" // Oddments: Store: Bit 2: Unset + "tbz %x[n_channels], #1, 90f\n" + "ldr x22, [x15, #0x0]\n" + "add x22, x22, x11\n" "st1 { v23.s }[0], [x22]\n" - "tbz %x[n_channels], #0, 47f\n" - "ldr x22, [x17, #0x0]\n" - "ldr x21, [x17, #0x8]\n" - "add x22, x22, x12\n" - "ldr x20, [x17, #0x10]\n" - "ldr x19, [x17, #0x18]\n" - "add x21, x21, x12\n" - "st1 { v31.h }[2], [x22]\n" - "add x20, x20, x12\n" - "st1 { v30.h }[2], [x21]\n" - "ldr x22, [x17, #0x20]\n" - "add x19, x19, x12\n" - "st1 { v29.h }[2], [x20]\n" - "add x22, x22, x12\n" - "st1 { v28.h }[2], [x19]\n" - "ldr x21, [x17, #0x28]\n" - "add x21, x21, x12\n" - "st1 { v27.h }[2], [x22]\n" - "ldr x20, [x17, #0x30]\n" - "add x20, x20, x12\n" - "st1 { v26.h }[2], [x21]\n" - "ldr x19, [x17, #0x38]\n" - "add x19, x19, x12\n" - "st1 { v25.h }[2], [x20]\n" - "ldr x22, [x17, #0x40]\n" - "add x22, x22, x12\n" - "st1 { v24.h }[2], [x19]\n" + "ldr x21, [x15, #0x8]\n" + "ldr x20, [x15, #0x10]\n" + "add x21, x21, x11\n" + "add x20, x20, x11\n" + "ldr x19, [x15, #0x18]\n" + "ldr x22, [x15, #0x20]\n" + "add x19, x19, x11\n" + "add x22, x22, x11\n" + "st1 { v24.s }[0], [x21]\n" + "ldr x21, [x15, #0x28]\n" + "add x21, x21, x11\n" + "st1 { v25.s }[0], [x20]\n" + "ldr x20, [x15, #0x30]\n" + "add x20, x20, x11\n" + "st1 { v26.s }[0], [x19]\n" + "ldr x19, [x15, #0x38]\n" + "add x19, x19, x11\n" + "st1 { v27.s }[0], [x22]\n" + "ldr x22, [x15, #0x40]\n" + "add x22, x22, x11\n" + "add x11, x11, #0x4\n" + "st1 { v28.s }[0], [x21]\n" + "st1 { v29.s }[0], [x20]\n" + "st1 { v30.s }[0], [x19]\n" + "st1 { v31.s }[0], [x22]\n" + "tbz %x[n_channels], #0, 91f\n" + "ldr x22, [x15, #0x0]\n" + "add x22, x22, x11\n" "st1 { v23.h }[2], [x22]\n" - "b 47f\n" - "46:" // Oddments: Store: Bit 1: Unset - "ldr x22, [x17, #0x0]\n" - "add x22, x22, x12\n" - "ldr x21, [x17, #0x8]\n" - "ldr x20, [x17, #0x10]\n" - "add x21, x21, x12\n" - "st1 { v31.h }[0], [x22]\n" - "ldr x19, [x17, #0x18]\n" - "add x20, x20, x12\n" - "st1 { v30.h }[0], [x21]\n" - "add x19, x19, x12\n" - "st1 { v29.h }[0], [x20]\n" - "ldr x22, [x17, #0x20]\n" - "add x22, x22, x12\n" - "st1 { v28.h }[0], [x19]\n" - "ldr x21, [x17, #0x28]\n" - "add x21, x21, x12\n" - "st1 { v27.h }[0], [x22]\n" - "ldr x20, [x17, #0x30]\n" - "add x20, x20, x12\n" - "st1 { v26.h }[0], [x21]\n" - "ldr x19, [x17, #0x38]\n" - "add x19, x19, x12\n" - "st1 { v25.h }[0], [x20]\n" - "ldr x22, [x17, #0x40]\n" - "add x22, x22, x12\n" - "st1 { v24.h }[0], [x19]\n" + "ldr x21, [x15, #0x8]\n" + "ldr x20, [x15, #0x10]\n" + "add x21, x21, x11\n" + "ldr x19, [x15, #0x18]\n" + "ldr x22, [x15, #0x20]\n" + "add x20, x20, x11\n" + "add x19, x19, x11\n" + "add x22, x22, x11\n" + "st1 { v24.h }[2], [x21]\n" + "ldr x21, [x15, #0x28]\n" + "add x21, x21, x11\n" + "st1 { v25.h }[2], [x20]\n" + "ldr x20, [x15, #0x30]\n" + "add x20, x20, x11\n" + "st1 { v26.h }[2], [x19]\n" + "ldr x19, [x15, #0x38]\n" + "add x19, x19, x11\n" + "st1 { v27.h }[2], [x22]\n" + "ldr x22, [x15, #0x40]\n" + "add x22, x22, x11\n" + "st1 { v28.h }[2], [x21]\n" + "st1 { v29.h }[2], [x20]\n" + "st1 { v30.h }[2], [x19]\n" + "st1 { v31.h }[2], [x22]\n" + "b 91f\n" + "90:" // Oddments: Store: Bit 2: Unset: Bit 1: Unset + "ldr x22, [x15, #0x0]\n" + "add x22, x22, x11\n" "st1 { v23.h }[0], [x22]\n" - "47:" // Oddments: Store: Bit 1: End + "ldr x21, [x15, #0x8]\n" + "ldr x20, [x15, #0x10]\n" + "ldr x19, [x15, #0x18]\n" + "add x21, x21, x11\n" + "add x20, x20, x11\n" + "ldr x22, [x15, #0x20]\n" + "add x19, x19, x11\n" + "add x22, x22, x11\n" + "st1 { v24.h }[0], [x21]\n" + "st1 { v25.h }[0], [x20]\n" + "ldr x21, [x15, #0x28]\n" + "ldr x20, [x15, #0x30]\n" + "add x21, x21, x11\n" + "st1 { v26.h }[0], [x19]\n" + "ldr x19, [x15, #0x38]\n" + "add x20, x20, x11\n" + "add x19, x19, x11\n" + "st1 { v27.h }[0], [x22]\n" + "ldr x22, [x15, #0x40]\n" + "add x22, x22, x11\n" + "st1 { v28.h }[0], [x21]\n" + "st1 { v29.h }[0], [x20]\n" + "st1 { v30.h }[0], [x19]\n" + "st1 { v31.h }[0], [x22]\n" + "91:" // Oddments: Store: Bit 2: End - "48:" // End + "92:" // End : : [n_channels] "r" ((unsigned long) n_channels), [offsetof_Args_inptrs] "I" (offsetof(Args, inptrs)), [offsetof_args_max] "I" (offsetof(Args, max)), [offsetof_args_min] "I" (offsetof(Args, min)), [offsetof_args_outptrs] "I" (offsetof(Args, outptrs)), [offsetof_args_params] "I" (offsetof(Args, params)), [params_struct] "r" (¶ms_struct) - : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v8", "v9", "v10", "v11", "v12", "v13", "v16", "v17", "v18", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x16", "x17", "x19", "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", "v16", "v17", "v18", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x16", "x17", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28" ); } } // namespace depthwise |