diff options
Diffstat (limited to 'src/core/NEON/kernels/arm_conv/depthwise/kernels/a64_fp32_nhwc_3x3_s1_output3x3_mla_depthfirst/generic_direct.cpp')
| -rw-r--r-- | src/core/NEON/kernels/arm_conv/depthwise/kernels/a64_fp32_nhwc_3x3_s1_output3x3_mla_depthfirst/generic_direct.cpp | 1126 |
1 files changed, 563 insertions, 563 deletions
diff --git a/src/core/NEON/kernels/arm_conv/depthwise/kernels/a64_fp32_nhwc_3x3_s1_output3x3_mla_depthfirst/generic_direct.cpp b/src/core/NEON/kernels/arm_conv/depthwise/kernels/a64_fp32_nhwc_3x3_s1_output3x3_mla_depthfirst/generic_direct.cpp index 6d185e7274..620319bc7c 100644 --- a/src/core/NEON/kernels/arm_conv/depthwise/kernels/a64_fp32_nhwc_3x3_s1_output3x3_mla_depthfirst/generic_direct.cpp +++ b/src/core/NEON/kernels/arm_conv/depthwise/kernels/a64_fp32_nhwc_3x3_s1_output3x3_mla_depthfirst/generic_direct.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2021 Arm Limited. + * Copyright (c) 2021, 2023 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -87,739 +87,739 @@ void a64_fp32_nhwc_3x3_s1_output3x3_mla_depthfirst_direct_impl( ); __asm__ __volatile__( - "mov x7, #0x0\n" - "mov x8, #0x0\n" + "mov x24, #0x0\n" + "mov x23, #0x0\n" "1:" // Tile loop - "str x7, [%x[params_struct], %[offsetof_args_tile_i]]\n" + "str x24, [%x[params_struct], %[offsetof_args_tile_i]]\n" + "mov x27, #0x3\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" - "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 #2\n" // inptr[0] += offset * sizeof(float) - "ld1r { v18.4s }, [x24]\n" - "add x12, x15, x22, LSL #2\n" - "ld1r { v17.4s }, [x21]\n" - "add x11, x12, x22, LSL #2\n" - "lsl x16, x16, #0x2\n" - "add x10, x11, x22, LSL #2\n" - "add x9, x10, x22, LSL #2\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 #2\n" // outptrs[0] += offset * sizeof(float) - "add x25, x13, x20, LSL #2\n" - "add x24, x25, x20, LSL #2\n" - "lsl x14, x14, #0x2\n" - "add x22, x14, x14\n" - "mov x21, #0x10\n" // cntb _, ALL, #1 - "sub x20, XZR, x21\n" - "lsr x19, %x[n_channels], #0x2\n" - "cbz x19, 4f\n" - "ldr 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" - "ld1 { v10.4s }, [x15]\n" - "ldr q11, [x15, x26]\n" - "ld1 { v12.4s }, [x9]\n" - "ldr q13, [x12, x28]\n" + "str x23, [%x[params_struct], %[offsetof_args_tile_j]]\n" + "ldr x25, [%x[params_struct], %[offsetof_args_ld_input_row]]\n" + "ldr x22, [%x[params_struct], %[offsetof_args_ld_output_row]]\n" + "mul x21, x24, x25\n" // offset = tile_i * ld_input_row + "ldr x8, [%x[params_struct], %[offsetof_args_ld_input_col]]\n" + "ldr x17, [%x[params_struct], %[offsetof_args_ld_output_col]]\n" + "mul x20, x24, x22\n" // offset = tile_i * ld_output_row + "mov x24, #0x10\n" // cntb _, ALL, #1 + "madd x21, x23, x8, x21\n" // offset += tile_j * ld_input_col + "ldr x16, [%x[params_struct], %[offsetof_args_inptr]]\n" + "lsl x8, x8, #0x2\n" + "ldr x15, [%x[params_struct], %[offsetof_args_outptr]]\n" + "madd x20, x23, x17, x20\n" // offset += tile_j * ld_output_col + "lsl x17, x17, #0x2\n" + "lsr x23, %x[n_channels], #0x2\n" + "ldr x14, [%x[params_struct], %[offsetof_args_params]]\n" + "mul x21, x21, x27\n" // offset *= kernel_stride * output_size + "add x16, x16, x21, LSL #2\n" // inptr[0] += offset * sizeof(float) + "add x13, x16, x25, LSL #2\n" + "mul x20, x20, x26\n" // offset *= output_tile_size + "add x12, x13, x25, LSL #2\n" + "add x11, x8, x8\n" + "add x15, x15, x20, LSL #2\n" // outptrs[0] += offset * sizeof(float) + "add x10, x12, x25, LSL #2\n" + "add x9, x11, x8\n" + "add x28, x15, x22, LSL #2\n" + "add x20, %x[params_struct], %[offsetof_args_min]\n" + "ld1r { v18.4s }, [x20]\n" + "add x20, %x[params_struct], %[offsetof_args_max]\n" + "ld1r { v17.4s }, [x20]\n" + "add x27, x10, x25, LSL #2\n" + "add x26, x9, x8\n" + "add x25, x28, x22, LSL #2\n" + "add x22, x17, x17\n" + "mov x21, #0x0\n" + "sub x20, XZR, x24\n" + "cbz x23, 4f\n" + "ldr q16, [x14, #0x0]\n" + "ldr q0, [x14, #0x10]\n" + "cmp x24, x23, LSL #4\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" + "add x14, x14, #0xa0\n" + "ldr q9, [x12, x11]\n" + "ld1 { v10.4s }, [x16]\n" + "ldr q11, [x16, x26]\n" + "ld1 { v12.4s }, [x27]\n" + "ldr q13, [x13, x11]\n" "bge 3f\n" "2:" // Tile loop: Channel loop - "mov v31.16b, v16.16b\n fmla v31.4s, v8.4s, v9.4s\n" + "mov v24.16b, v16.16b\n fmla v24.4s, v7.4s, v9.4s\n" + "mov v23.16b, v16.16b\n fmla v23.4s, v8.4s, v9.4s\n" + "add x24, x24, #0x10\n" + "cmp x24, x23, LSL #4\n" + "mov v25.16b, v16.16b\n fmla v25.4s, v6.4s, v9.4s\n" + "fmla v24.4s, v4.4s, v13.4s\n" "add x20, x20, #0x10\n" - "mov v30.16b, v16.16b\n fmla v30.4s, v7.4s, v9.4s\n" - "add x23, x23, #0x10\n" - "mov v29.16b, v16.16b\n fmla v29.4s, v6.4s, v9.4s\n" "add x21, x21, #0x10\n" - "mov v28.16b, v16.16b\n fmla v28.4s, v5.4s, v9.4s\n" - "cmp x21, x19, LSL #4\n" + "mov v26.16b, v16.16b\n fmla v26.4s, v5.4s, v9.4s\n" "mov v27.16b, v16.16b\n fmla v27.4s, v4.4s, v9.4s\n" - "mov v26.16b, v16.16b\n fmla v26.4s, v3.4s, v9.4s\n" - "mov v25.16b, v16.16b\n fmla v25.4s, v2.4s, v9.4s\n" - "mov v24.16b, v16.16b\n fmla v24.4s, v1.4s, v9.4s\n" - "mov v23.16b, v16.16b\n fmla v23.4s, v0.4s, v9.4s\n" - "ldr q16, [x17, #0x0]\n" - "fmla v31.4s, v0.4s, v10.4s\n" - "ldr q10, [x11, x27]\n" - "fmla v29.4s, v2.4s, v11.4s\n" - "ldr q11, [x11, x16]\n" - "fmla v25.4s, v6.4s, v12.4s\n" - "ldr q12, [x9, x26]\n" - "fmla v30.4s, v4.4s, v13.4s\n" - "fmla v31.4s, v5.4s, v13.4s\n" - "fmla v29.4s, v3.4s, v13.4s\n" - "fmla v28.4s, v2.4s, v13.4s\n" + "mov v28.16b, v16.16b\n fmla v28.4s, v3.4s, v9.4s\n" + "fmla v23.4s, v0.4s, v10.4s\n" + "ldr q10, [x12, x9]\n" + "fmla v25.4s, v2.4s, v11.4s\n" + "ldr q11, [x12, x8]\n" + "mov v29.16b, v16.16b\n fmla v29.4s, v2.4s, v9.4s\n" + "fmla v24.4s, v6.4s, v11.4s\n" + "mov v31.16b, v16.16b\n fmla v31.4s, v0.4s, v9.4s\n" + "fmla v23.4s, v5.4s, v13.4s\n" + "fmla v25.4s, v3.4s, v13.4s\n" + "fmla v26.4s, v2.4s, v13.4s\n" "fmla v27.4s, v1.4s, v13.4s\n" - "fmla v26.4s, v0.4s, v13.4s\n" - "ldr q13, [x15, x16]\n" - "fmla v23.4s, v8.4s, v12.4s\n" - "ldr q12, [x15, x27]\n" - "fmla v31.4s, v7.4s, v11.4s\n" - "fmla v30.4s, v6.4s, v11.4s\n" - "fmla v28.4s, v4.4s, v11.4s\n" + "fmla v28.4s, v0.4s, v13.4s\n" + "ldr q13, [x16, x8]\n" + "fmla v29.4s, v6.4s, v12.4s\n" + "ldr q12, [x27, x26]\n" + "mov v30.16b, v16.16b\n fmla v30.4s, v1.4s, v9.4s\n" + "ldr q16, [x14, #0x0]\n" + "fmla v24.4s, v0.4s, v13.4s\n" + "fmla v31.4s, v8.4s, v12.4s\n" + "ldr q12, [x16, x9]\n" + "fmla v23.4s, v7.4s, v11.4s\n" + "fmla v30.4s, v0.4s, v11.4s\n" + "fmla v26.4s, v4.4s, v11.4s\n" "fmla v27.4s, v3.4s, v11.4s\n" - "fmla v25.4s, v1.4s, v11.4s\n" - "fmla v24.4s, v0.4s, v11.4s\n" - "ld1 { v11.4s }, [x12]\n" - "fmla v31.4s, v1.4s, v13.4s\n" - "fmla v30.4s, v0.4s, v13.4s\n" - "ldr q13, [x12, x26]\n" - "fmla v29.4s, v1.4s, v12.4s\n" - "fmla v27.4s, v5.4s, v10.4s\n" - "fmla v26.4s, v4.4s, v10.4s\n" - "fmla v30.4s, v2.4s, v12.4s\n" + "fmla v29.4s, v1.4s, v11.4s\n" + "ld1 { v11.4s }, [x13]\n" + "fmla v24.4s, v2.4s, v12.4s\n" + "fmla v25.4s, v1.4s, v12.4s\n" "ld1 { v12.4s }, [x10]\n" - "fmla v29.4s, v7.4s, v10.4s\n" - "fmla v24.4s, v2.4s, v10.4s\n" - "fmla v23.4s, v1.4s, v10.4s\n" - "fmla v30.4s, v8.4s, v10.4s\n" - "ldr q10, [x10, x28]\n" - "fmla v31.4s, v3.4s, v11.4s\n" - "fmla v28.4s, v0.4s, v11.4s\n" + "fmla v28.4s, v4.4s, v10.4s\n" + "fmla v23.4s, v1.4s, v13.4s\n" + "ldr q13, [x13, x26]\n" + "fmla v30.4s, v2.4s, v10.4s\n" + "fmla v31.4s, v1.4s, v10.4s\n" + "fmla v24.4s, v8.4s, v10.4s\n" + "fmla v25.4s, v7.4s, v10.4s\n" + "fmla v27.4s, v5.4s, v10.4s\n" + "ldr q10, [x10, x11]\n" + "fmla v26.4s, v0.4s, v11.4s\n" + "fmla v29.4s, v3.4s, v12.4s\n" + "fmla v28.4s, v2.4s, v13.4s\n" + "fmla v30.4s, v4.4s, v10.4s\n" + "fmla v31.4s, v3.4s, v10.4s\n" + "fmla v23.4s, v3.4s, v11.4s\n" "ldr q11, [x10, x26]\n" - "fmla v29.4s, v5.4s, v13.4s\n" - "fmla v26.4s, v2.4s, v13.4s\n" - "ldr q13, [x9, x16]\n" - "fmla v25.4s, v3.4s, v12.4s\n" - "fmla v28.4s, v6.4s, v12.4s\n" - "ldr q12, [x12, x16]\n" + "fmla v25.4s, v5.4s, v13.4s\n" + "ldr q13, [x27, x8]\n" + "fmla v26.4s, v6.4s, v12.4s\n" + "ldr q12, [x13, x8]\n" "fmla v27.4s, v7.4s, v10.4s\n" - "fmla v26.4s, v6.4s, v10.4s\n" - "fmla v25.4s, v5.4s, v10.4s\n" - "fmla v28.4s, v8.4s, v10.4s\n" - "fmla v24.4s, v4.4s, v10.4s\n" - "fmla v23.4s, v3.4s, v10.4s\n" - "fmla v26.4s, v8.4s, v11.4s\n" - "fmla v25.4s, v7.4s, v13.4s\n" - "fmla v24.4s, v6.4s, v13.4s\n" - "ldr q13, [x9, x27]\n" - "fmla v23.4s, v5.4s, v11.4s\n" - "ldr q11, [x12, x27]\n" - "add x12, x12, #0x10\n" - "fmla v31.4s, v4.4s, v12.4s\n" - "fmla v30.4s, v3.4s, v12.4s\n" - "fmla v28.4s, v1.4s, v12.4s\n" + "fmla v29.4s, v5.4s, v10.4s\n" + "fmla v28.4s, v6.4s, v10.4s\n" + "fmla v31.4s, v5.4s, v11.4s\n" + "fmla v30.4s, v6.4s, v13.4s\n" + "fmla v26.4s, v8.4s, v10.4s\n" + "fmla v29.4s, v7.4s, v13.4s\n" + "ldr q13, [x27, x9]\n" + "fmla v24.4s, v3.4s, v12.4s\n" "fmla v27.4s, v0.4s, v12.4s\n" - "ldr q12, [x10, x16]\n" - "fmla v29.4s, v4.4s, v11.4s\n" - "fmla v30.4s, v5.4s, v11.4s\n" - "fmla v26.4s, v1.4s, v11.4s\n" - "fmla v27.4s, v2.4s, v11.4s\n" - "ldr q11, [x15, x28]\n" - "add x15, x15, #0x10\n" - "fmla v24.4s, v8.4s, v13.4s\n" - "ld1 { v10.4s }, [x15]\n" - "fmla v23.4s, v7.4s, v13.4s\n" - "ldr q13, [x10, x27]\n" + "fmla v28.4s, v8.4s, v11.4s\n" + "ldr q11, [x13, x9]\n" + "fmla v30.4s, v8.4s, v13.4s\n" + "add x13, x13, #0x10\n" + "fmla v31.4s, v7.4s, v13.4s\n" + "ldr q13, [x10, x9]\n" + "fmla v23.4s, v4.4s, v12.4s\n" + "fmla v26.4s, v1.4s, v12.4s\n" + "ldr q12, [x10, x8]\n" + "fmla v24.4s, v5.4s, v11.4s\n" "add x10, x10, #0x10\n" - "fmla v28.4s, v7.4s, v12.4s\n" + "fmla v25.4s, v4.4s, v11.4s\n" + "fmla v27.4s, v2.4s, v11.4s\n" + "fmla v28.4s, v1.4s, v11.4s\n" + "ldr q11, [x16, x11]\n" + "fmla v29.4s, v4.4s, v12.4s\n" + "add x16, x16, #0x10\n" + "ld1 { v10.4s }, [x16]\n" + "fmla v30.4s, v3.4s, v12.4s\n" + "fmla v31.4s, v4.4s, v13.4s\n" + "ldr q4, [x14, #0x50]\n" + "fmla v26.4s, v7.4s, v12.4s\n" "fmla v27.4s, v6.4s, v12.4s\n" - "fmla v25.4s, v4.4s, v12.4s\n" - "fmla v24.4s, v3.4s, v12.4s\n" - "ld1 { v12.4s }, [x11]\n" + "ld1 { v12.4s }, [x12]\n" + "fmla v23.4s, v2.4s, v11.4s\n" + "fmla v24.4s, v1.4s, v11.4s\n" + "ldr q1, [x14, #0x20]\n" + "fmax v24.4s, v24.4s, v18.4s\n" + "fmla v25.4s, v0.4s, v11.4s\n" + "ldr q11, [x12, x26]\n" + "fmla v28.4s, v7.4s, v13.4s\n" + "add x12, x12, #0x10\n" + "ldr q9, [x12, x11]\n" + "fmla v30.4s, v5.4s, v13.4s\n" + "fmla v29.4s, v0.4s, v12.4s\n" + "ldr q0, [x14, #0x10]\n" "fmla v31.4s, v2.4s, v11.4s\n" - "fmla v30.4s, v1.4s, v11.4s\n" - "ldr q1, [x17, #0x20]\n" - "fmla v29.4s, v0.4s, v11.4s\n" - "ldr q11, [x11, x26]\n" - "add x11, x11, #0x10\n" + "ldr q2, [x14, #0x30]\n" "fmla v27.4s, v8.4s, v13.4s\n" - "ldr q9, [x11, x28]\n" - "fmla v26.4s, v7.4s, v13.4s\n" - "fmla v24.4s, v5.4s, v13.4s\n" - "fmla v23.4s, v4.4s, v13.4s\n" - "ldr q13, [x9, x28]\n" - "add x9, x9, #0x10\n" - "fmla v31.4s, v6.4s, v12.4s\n" - "ldr q4, [x17, #0x50]\n" - "fmla v28.4s, v3.4s, v12.4s\n" - "ldr q3, [x17, #0x40]\n" - "fmla v25.4s, v0.4s, v12.4s\n" - "ld1 { v12.4s }, [x9]\n" - "fmla v29.4s, v8.4s, v11.4s\n" - "ldr q0, [x17, #0x10]\n" - "fmla v26.4s, v5.4s, v11.4s\n" - "ldr q5, [x17, #0x60]\n" - "fmla v23.4s, v2.4s, v11.4s\n" - "ldr q11, [x15, x26]\n" - "fmla v25.4s, v8.4s, v13.4s\n" - "ldr q2, [x17, #0x30]\n" - "fmla v24.4s, v7.4s, v13.4s\n" - "ldr q7, [x17, #0x80]\n" - "fmax v31.4s, v31.4s, v18.4s\n" - "ldr q8, [x17, #0x90]\n" - "fmla v23.4s, v6.4s, v13.4s\n" - "ldr q13, [x12, x28]\n" - "fmax v30.4s, v30.4s, v18.4s\n" - "ldr q6, [x17, #0x70]\n" - "add x17, x17, #0xa0\n" - "fmin v31.4s, v31.4s, v17.4s\n" - "st1 { v31.4s }, [x13]\n" - "fmin v30.4s, v30.4s, v17.4s\n" - "fmax v29.4s, v29.4s, v18.4s\n" - "str q30, [x13, x14]\n" - "fmin v29.4s, v29.4s, v17.4s\n" - "fmax v28.4s, v28.4s, v18.4s\n" - "str q29, [x13, x22]\n" - "fmin v28.4s, v28.4s, v17.4s\n" - "add x13, x13, #0x10\n" - "fmax v27.4s, v27.4s, v18.4s\n" - "st1 { v28.4s }, [x25]\n" - "fmax v26.4s, v26.4s, v18.4s\n" + "ldr q13, [x27, x11]\n" + "fmla v23.4s, v6.4s, v12.4s\n" + "fmla v26.4s, v3.4s, v12.4s\n" + "ldr q3, [x14, #0x40]\n" + "fmax v23.4s, v23.4s, v18.4s\n" + "fmla v25.4s, v8.4s, v11.4s\n" + "fmla v28.4s, v5.4s, v11.4s\n" + "ldr q11, [x16, x26]\n" + "ldr q5, [x14, #0x60]\n" + "fmla v29.4s, v8.4s, v13.4s\n" + "ldr q8, [x14, #0x90]\n" + "fmla v30.4s, v7.4s, v13.4s\n" + "ldr q7, [x14, #0x80]\n" + "fmla v31.4s, v6.4s, v13.4s\n" + "ldr q13, [x13, x11]\n" + "ldr q6, [x14, #0x70]\n" "fmax v25.4s, v25.4s, v18.4s\n" - "fmin v27.4s, v27.4s, v17.4s\n" - "str q27, [x25, x14]\n" - "fmin v26.4s, v26.4s, v17.4s\n" + "fmax v26.4s, v26.4s, v18.4s\n" + "fmax v27.4s, v27.4s, v18.4s\n" + "add x27, x27, #0x10\n" + "ld1 { v12.4s }, [x27]\n" + "fmax v28.4s, v28.4s, v18.4s\n" + "fmax v29.4s, v29.4s, v18.4s\n" + "add x14, x14, #0xa0\n" + "fmax v30.4s, v30.4s, v18.4s\n" + "fmax v31.4s, v31.4s, v18.4s\n" + "fmin v23.4s, v23.4s, v17.4s\n" + "fmin v24.4s, v24.4s, v17.4s\n" + "st1 { v23.4s }, [x15]\n" "fmin v25.4s, v25.4s, v17.4s\n" - "str q26, [x25, x22]\n" - "fmax v24.4s, v24.4s, v18.4s\n" + "fmin v26.4s, v26.4s, v17.4s\n" + "str q24, [x15, x17]\n" + "fmin v27.4s, v27.4s, v17.4s\n" + "fmin v28.4s, v28.4s, v17.4s\n" + "str q25, [x15, x22]\n" + "add x15, x15, #0x10\n" + "fmin v29.4s, v29.4s, v17.4s\n" + "fmin v30.4s, v30.4s, v17.4s\n" + "st1 { v26.4s }, [x28]\n" + "fmin v31.4s, v31.4s, v17.4s\n" + "str q27, [x28, x17]\n" + "str q28, [x28, x22]\n" + "add x28, x28, #0x10\n" + "st1 { v29.4s }, [x25]\n" + "str q30, [x25, x17]\n" + "str q31, [x25, x22]\n" "add x25, x25, #0x10\n" - "fmax v23.4s, v23.4s, v18.4s\n" - "st1 { v25.4s }, [x24]\n" - "fmin v24.4s, v24.4s, v17.4s\n" - "str q24, [x24, x14]\n" - "fmin v23.4s, v23.4s, v17.4s\n" - "str q23, [x24, x22]\n" - "add x24, x24, #0x10\n" "blt 2b\n" "3:" // Tile loop: Channel tail - "mov v31.16b, v16.16b\n fmla v31.4s, v8.4s, v9.4s\n" - "mov v30.16b, v16.16b\n fmla v30.4s, v7.4s, v9.4s\n" - "mov v29.16b, v16.16b\n fmla v29.4s, v6.4s, v9.4s\n" - "mov v28.16b, v16.16b\n fmla v28.4s, v5.4s, v9.4s\n" + "mov v24.16b, v16.16b\n fmla v24.4s, v7.4s, v9.4s\n" + "mov v23.16b, v16.16b\n fmla v23.4s, v8.4s, v9.4s\n" + "mov v25.16b, v16.16b\n fmla v25.4s, v6.4s, v9.4s\n" + "fmla v24.4s, v4.4s, v13.4s\n" + "mov v26.16b, v16.16b\n fmla v26.4s, v5.4s, v9.4s\n" "mov v27.16b, v16.16b\n fmla v27.4s, v4.4s, v9.4s\n" - "mov v26.16b, v16.16b\n fmla v26.4s, v3.4s, v9.4s\n" - "mov v25.16b, v16.16b\n fmla v25.4s, v2.4s, v9.4s\n" - "mov v24.16b, v16.16b\n fmla v24.4s, v1.4s, v9.4s\n" - "mov v23.16b, v16.16b\n fmla v23.4s, v0.4s, v9.4s\n" - "fmla v31.4s, v0.4s, v10.4s\n" - "ldr q10, [x11, x27]\n" - "fmla v29.4s, v2.4s, v11.4s\n" - "ldr q11, [x11, x16]\n" - "fmla v25.4s, v6.4s, v12.4s\n" - "ldr q12, [x9, x26]\n" - "fmla v30.4s, v4.4s, v13.4s\n" - "fmla v31.4s, v5.4s, v13.4s\n" - "fmla v29.4s, v3.4s, v13.4s\n" - "fmla v28.4s, v2.4s, v13.4s\n" + "mov v28.16b, v16.16b\n fmla v28.4s, v3.4s, v9.4s\n" + "fmla v23.4s, v0.4s, v10.4s\n" + "ldr q10, [x12, x9]\n" + "fmla v25.4s, v2.4s, v11.4s\n" + "ldr q11, [x12, x8]\n" + "mov v29.16b, v16.16b\n fmla v29.4s, v2.4s, v9.4s\n" + "fmla v24.4s, v6.4s, v11.4s\n" + "mov v31.16b, v16.16b\n fmla v31.4s, v0.4s, v9.4s\n" + "fmla v23.4s, v5.4s, v13.4s\n" + "fmla v25.4s, v3.4s, v13.4s\n" + "fmla v26.4s, v2.4s, v13.4s\n" "fmla v27.4s, v1.4s, v13.4s\n" - "fmla v26.4s, v0.4s, v13.4s\n" - "ldr q13, [x15, x16]\n" - "fmla v23.4s, v8.4s, v12.4s\n" - "ldr q12, [x15, x27]\n" - "fmla v31.4s, v7.4s, v11.4s\n" - "fmla v30.4s, v6.4s, v11.4s\n" - "fmla v28.4s, v4.4s, v11.4s\n" + "fmla v28.4s, v0.4s, v13.4s\n" + "ldr q13, [x16, x8]\n" + "fmla v29.4s, v6.4s, v12.4s\n" + "ldr q12, [x27, x26]\n" + "mov v30.16b, v16.16b\n fmla v30.4s, v1.4s, v9.4s\n" + "fmla v24.4s, v0.4s, v13.4s\n" + "fmla v31.4s, v8.4s, v12.4s\n" + "ldr q12, [x16, x9]\n" + "fmla v23.4s, v7.4s, v11.4s\n" + "fmla v30.4s, v0.4s, v11.4s\n" + "fmla v26.4s, v4.4s, v11.4s\n" "fmla v27.4s, v3.4s, v11.4s\n" - "fmla v25.4s, v1.4s, v11.4s\n" - "fmla v24.4s, v0.4s, v11.4s\n" - "ld1 { v11.4s }, [x12]\n" - "fmla v31.4s, v1.4s, v13.4s\n" - "fmla v30.4s, v0.4s, v13.4s\n" - "ldr q13, [x12, x26]\n" - "fmla v29.4s, v1.4s, v12.4s\n" - "fmla v27.4s, v5.4s, v10.4s\n" - "fmla v26.4s, v4.4s, v10.4s\n" - "fmla v30.4s, v2.4s, v12.4s\n" + "fmla v29.4s, v1.4s, v11.4s\n" + "ld1 { v11.4s }, [x13]\n" + "fmla v24.4s, v2.4s, v12.4s\n" + "fmla v25.4s, v1.4s, v12.4s\n" "ld1 { v12.4s }, [x10]\n" - "fmla v29.4s, v7.4s, v10.4s\n" - "fmla v24.4s, v2.4s, v10.4s\n" - "fmla v23.4s, v1.4s, v10.4s\n" - "fmla v30.4s, v8.4s, v10.4s\n" - "ldr q10, [x10, x28]\n" - "fmla v31.4s, v3.4s, v11.4s\n" - "fmla v28.4s, v0.4s, v11.4s\n" + "fmla v28.4s, v4.4s, v10.4s\n" + "fmla v23.4s, v1.4s, v13.4s\n" + "ldr q13, [x13, x26]\n" + "fmla v30.4s, v2.4s, v10.4s\n" + "fmla v31.4s, v1.4s, v10.4s\n" + "fmla v24.4s, v8.4s, v10.4s\n" + "fmla v25.4s, v7.4s, v10.4s\n" + "fmla v27.4s, v5.4s, v10.4s\n" + "ldr q10, [x10, x11]\n" + "fmla v26.4s, v0.4s, v11.4s\n" + "fmla v29.4s, v3.4s, v12.4s\n" + "fmla v28.4s, v2.4s, v13.4s\n" + "fmla v30.4s, v4.4s, v10.4s\n" + "fmla v31.4s, v3.4s, v10.4s\n" + "fmla v23.4s, v3.4s, v11.4s\n" "ldr q11, [x10, x26]\n" - "fmla v29.4s, v5.4s, v13.4s\n" - "fmla v26.4s, v2.4s, v13.4s\n" - "ldr q13, [x9, x16]\n" - "fmla v25.4s, v3.4s, v12.4s\n" - "fmla v28.4s, v6.4s, v12.4s\n" - "ldr q12, [x12, x16]\n" + "fmla v25.4s, v5.4s, v13.4s\n" + "ldr q13, [x27, x8]\n" + "fmla v26.4s, v6.4s, v12.4s\n" + "ldr q12, [x13, x8]\n" "fmla v27.4s, v7.4s, v10.4s\n" - "fmla v26.4s, v6.4s, v10.4s\n" - "fmla v25.4s, v5.4s, v10.4s\n" - "fmla v28.4s, v8.4s, v10.4s\n" - "fmla v24.4s, v4.4s, v10.4s\n" - "fmla v23.4s, v3.4s, v10.4s\n" - "fmla v26.4s, v8.4s, v11.4s\n" - "fmla v25.4s, v7.4s, v13.4s\n" - "fmla v24.4s, v6.4s, v13.4s\n" - "ldr q13, [x9, x27]\n" - "fmla v23.4s, v5.4s, v11.4s\n" - "ldr q11, [x12, x27]\n" - "add x12, x12, #0x10\n" - "fmla v31.4s, v4.4s, v12.4s\n" - "fmla v30.4s, v3.4s, v12.4s\n" - "fmla v28.4s, v1.4s, v12.4s\n" + "fmla v29.4s, v5.4s, v10.4s\n" + "fmla v28.4s, v6.4s, v10.4s\n" + "fmla v31.4s, v5.4s, v11.4s\n" + "fmla v30.4s, v6.4s, v13.4s\n" + "fmla v26.4s, v8.4s, v10.4s\n" + "fmla v29.4s, v7.4s, v13.4s\n" + "ldr q13, [x27, x9]\n" + "fmla v24.4s, v3.4s, v12.4s\n" "fmla v27.4s, v0.4s, v12.4s\n" - "ldr q12, [x10, x16]\n" - "fmla v29.4s, v4.4s, v11.4s\n" - "fmla v30.4s, v5.4s, v11.4s\n" - "fmla v26.4s, v1.4s, v11.4s\n" - "fmla v27.4s, v2.4s, v11.4s\n" - "ldr q11, [x15, x28]\n" - "add x15, x15, #0x10\n" - "fmla v24.4s, v8.4s, v13.4s\n" - "fmla v23.4s, v7.4s, v13.4s\n" - "ldr q13, [x10, x27]\n" + "fmla v28.4s, v8.4s, v11.4s\n" + "ldr q11, [x13, x9]\n" + "fmla v30.4s, v8.4s, v13.4s\n" + "add x13, x13, #0x10\n" + "fmla v31.4s, v7.4s, v13.4s\n" + "ldr q13, [x10, x9]\n" + "fmla v23.4s, v4.4s, v12.4s\n" + "fmla v26.4s, v1.4s, v12.4s\n" + "ldr q12, [x10, x8]\n" + "fmla v24.4s, v5.4s, v11.4s\n" "add x10, x10, #0x10\n" - "fmla v28.4s, v7.4s, v12.4s\n" + "fmla v25.4s, v4.4s, v11.4s\n" + "fmla v27.4s, v2.4s, v11.4s\n" + "fmla v28.4s, v1.4s, v11.4s\n" + "ldr q11, [x16, x11]\n" + "fmla v29.4s, v4.4s, v12.4s\n" + "add x16, x16, #0x10\n" + "fmla v30.4s, v3.4s, v12.4s\n" + "fmla v31.4s, v4.4s, v13.4s\n" + "fmla v26.4s, v7.4s, v12.4s\n" "fmla v27.4s, v6.4s, v12.4s\n" - "fmla v25.4s, v4.4s, v12.4s\n" - "fmla v24.4s, v3.4s, v12.4s\n" - "ld1 { v12.4s }, [x11]\n" + "ld1 { v12.4s }, [x12]\n" + "fmla v23.4s, v2.4s, v11.4s\n" + "fmla v24.4s, v1.4s, v11.4s\n" + "fmax v24.4s, v24.4s, v18.4s\n" + "fmla v25.4s, v0.4s, v11.4s\n" + "ldr q11, [x12, x26]\n" + "fmla v28.4s, v7.4s, v13.4s\n" + "fmin v24.4s, v24.4s, v17.4s\n" + "fmla v30.4s, v5.4s, v13.4s\n" + "fmla v29.4s, v0.4s, v12.4s\n" + "add x12, x12, #0x10\n" "fmla v31.4s, v2.4s, v11.4s\n" - "fmla v30.4s, v1.4s, v11.4s\n" - "fmla v29.4s, v0.4s, v11.4s\n" - "ldr q11, [x11, x26]\n" - "add x11, x11, #0x10\n" "fmla v27.4s, v8.4s, v13.4s\n" - "fmla v26.4s, v7.4s, v13.4s\n" - "fmla v24.4s, v5.4s, v13.4s\n" - "fmla v23.4s, v4.4s, v13.4s\n" - "ldr q13, [x9, x28]\n" - "add x9, x9, #0x10\n" - "fmla v31.4s, v6.4s, v12.4s\n" - "fmla v28.4s, v3.4s, v12.4s\n" - "fmla v25.4s, v0.4s, v12.4s\n" - "fmla v29.4s, v8.4s, v11.4s\n" - "fmla v26.4s, v5.4s, v11.4s\n" - "fmla v23.4s, v2.4s, v11.4s\n" - "fmla v25.4s, v8.4s, v13.4s\n" - "fmla v24.4s, v7.4s, v13.4s\n" - "fmax v31.4s, v31.4s, v18.4s\n" - "fmla v23.4s, v6.4s, v13.4s\n" - "fmax v30.4s, v30.4s, v18.4s\n" - "fmin v31.4s, v31.4s, v17.4s\n" - "st1 { v31.4s }, [x13]\n" - "fmin v30.4s, v30.4s, v17.4s\n" - "fmax v29.4s, v29.4s, v18.4s\n" - "str q30, [x13, x14]\n" - "fmin v29.4s, v29.4s, v17.4s\n" - "fmax v28.4s, v28.4s, v18.4s\n" - "str q29, [x13, x22]\n" - "fmin v28.4s, v28.4s, v17.4s\n" - "add x13, x13, #0x10\n" + "ldr q13, [x27, x11]\n" "fmax v27.4s, v27.4s, v18.4s\n" - "st1 { v28.4s }, [x25]\n" - "fmax v26.4s, v26.4s, v18.4s\n" + "fmla v23.4s, v6.4s, v12.4s\n" + "fmla v26.4s, v3.4s, v12.4s\n" + "fmax v23.4s, v23.4s, v18.4s\n" + "add x27, x27, #0x10\n" + "fmla v25.4s, v8.4s, v11.4s\n" + "fmla v28.4s, v5.4s, v11.4s\n" "fmax v25.4s, v25.4s, v18.4s\n" - "fmin v27.4s, v27.4s, v17.4s\n" - "str q27, [x25, x14]\n" - "fmin v26.4s, v26.4s, v17.4s\n" + "fmla v29.4s, v8.4s, v13.4s\n" + "fmla v30.4s, v7.4s, v13.4s\n" + "fmax v26.4s, v26.4s, v18.4s\n" + "fmla v31.4s, v6.4s, v13.4s\n" + "fmax v28.4s, v28.4s, v18.4s\n" + "fmax v29.4s, v29.4s, v18.4s\n" + "fmax v30.4s, v30.4s, v18.4s\n" + "fmax v31.4s, v31.4s, v18.4s\n" + "fmin v23.4s, v23.4s, v17.4s\n" + "st1 { v23.4s }, [x15]\n" "fmin v25.4s, v25.4s, v17.4s\n" - "str q26, [x25, x22]\n" - "fmax v24.4s, v24.4s, v18.4s\n" + "fmin v26.4s, v26.4s, v17.4s\n" + "str q24, [x15, x17]\n" + "fmin v27.4s, v27.4s, v17.4s\n" + "fmin v28.4s, v28.4s, v17.4s\n" + "str q25, [x15, x22]\n" + "add x15, x15, #0x10\n" + "fmin v29.4s, v29.4s, v17.4s\n" + "fmin v30.4s, v30.4s, v17.4s\n" + "st1 { v26.4s }, [x28]\n" + "fmin v31.4s, v31.4s, v17.4s\n" + "str q27, [x28, x17]\n" + "str q28, [x28, x22]\n" + "add x28, x28, #0x10\n" + "st1 { v29.4s }, [x25]\n" + "str q30, [x25, x17]\n" + "str q31, [x25, x22]\n" "add x25, x25, #0x10\n" - "fmax v23.4s, v23.4s, v18.4s\n" - "st1 { v25.4s }, [x24]\n" - "fmin v24.4s, v24.4s, v17.4s\n" - "str q24, [x24, x14]\n" - "fmin v23.4s, v23.4s, v17.4s\n" - "str q23, [x24, x22]\n" - "add x24, x24, #0x10\n" "4:" // Tile loop: Oddments "tst %x[n_channels], #0x3\n" "beq 49f\n" - "ldr q16, [x17, #0x0]\n" - "ldr q0, [x17, #0x10]\n" - "add x23, x11, x28\n" - "ldr q1, [x17, #0x20]\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 q16, [x14, #0x0]\n" + "ldr q0, [x14, #0x10]\n" + "add x24, x12, x11\n" + "add x23, x16, XZR\n" + "ldr q1, [x14, #0x20]\n" + "ldr q2, [x14, #0x30]\n" + "add x22, x16, x26\n" + "add x21, x27, XZR\n" + "ldr q3, [x14, #0x40]\n" + "ldr q4, [x14, #0x50]\n" + "add x20, x13, x11\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], #1, 5f\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" + "ldr d9, [x24], #0x8\n" + "ldr d10, [x23], #0x8\n" + "ldr d11, [x22], #0x8\n" + "ldr d12, [x21], #0x8\n" + "ldr d13, [x20], #0x8\n" "tbz %x[n_channels], #0, 6f\n" - "ld1 { v9.s }[2], [x23]\n" - "ld1 { v10.s }[2], [x22]\n" - "ld1 { v11.s }[2], [x21]\n" - "ld1 { v12.s }[2], [x20]\n" - "ld1 { v13.s }[2], [x19]\n" + "ld1 { v9.s }[2], [x24]\n" + "ld1 { v10.s }[2], [x23]\n" + "ld1 { v11.s }[2], [x22]\n" + "ld1 { v12.s }[2], [x21]\n" + "ld1 { v13.s }[2], [x20]\n" "b 6f\n" "5:" // Tile loop: Oddments: Load inputs: (2, 2), (0, 0), (0, 4), (4, 0), (1, 2): Bit 1: Unset - "ldr s9, [x23, #0x0]\n" - "ldr s10, [x22, #0x0]\n" - "ldr s11, [x21, #0x0]\n" - "ldr s12, [x20, #0x0]\n" - "ldr s13, [x19, #0x0]\n" + "ldr s9, [x24, #0x0]\n" + "ldr s10, [x23, #0x0]\n" + "ldr s11, [x22, #0x0]\n" + "ldr s12, [x21, #0x0]\n" + "ldr s13, [x20, #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 v31.4s, v8.4s, v9.4s\n" - "add x19, x9, x26\n" - "mov v30.16b, v16.16b\n fmla v30.4s, v7.4s, v9.4s\n" - "mov v29.16b, v16.16b\n fmla v29.4s, v6.4s, v9.4s\n" - "mov v28.16b, v16.16b\n fmla v28.4s, v5.4s, v9.4s\n" + "mov v23.16b, v16.16b\n fmla v23.4s, v8.4s, v9.4s\n" + "mov v25.16b, v16.16b\n fmla v25.4s, v6.4s, v9.4s\n" + "add x20, x27, x26\n" + "mov v24.16b, v16.16b\n fmla v24.4s, v7.4s, v9.4s\n" + "mov v26.16b, v16.16b\n fmla v26.4s, v5.4s, v9.4s\n" "mov v27.16b, v16.16b\n fmla v27.4s, v4.4s, v9.4s\n" - "mov v26.16b, v16.16b\n fmla v26.4s, v3.4s, v9.4s\n" - "mov v25.16b, v16.16b\n fmla v25.4s, v2.4s, v9.4s\n" - "mov v24.16b, v16.16b\n fmla v24.4s, v1.4s, v9.4s\n" - "mov v23.16b, v16.16b\n fmla v23.4s, v0.4s, v9.4s\n" - "fmla v31.4s, v0.4s, v10.4s\n" - "fmla v29.4s, v2.4s, v11.4s\n" - "fmla v25.4s, v6.4s, v12.4s\n" - "fmla v30.4s, v4.4s, v13.4s\n" - "fmla v31.4s, v5.4s, v13.4s\n" - "fmla v29.4s, v3.4s, v13.4s\n" - "fmla v28.4s, v2.4s, v13.4s\n" + "mov v28.16b, v16.16b\n fmla v28.4s, v3.4s, v9.4s\n" + "mov v29.16b, v16.16b\n fmla v29.4s, v2.4s, v9.4s\n" + "fmla v23.4s, v0.4s, v10.4s\n" + "fmla v25.4s, v2.4s, v11.4s\n" + "mov v30.16b, v16.16b\n fmla v30.4s, v1.4s, v9.4s\n" + "mov v31.16b, v16.16b\n fmla v31.4s, v0.4s, v9.4s\n" + "fmla v29.4s, v6.4s, v12.4s\n" + "fmla v23.4s, v5.4s, v13.4s\n" + "fmla v24.4s, v4.4s, v13.4s\n" + "fmla v25.4s, v3.4s, v13.4s\n" + "fmla v26.4s, v2.4s, v13.4s\n" "fmla v27.4s, v1.4s, v13.4s\n" - "fmla v26.4s, v0.4s, v13.4s\n" + "fmla v28.4s, v0.4s, v13.4s\n" "tbz %x[n_channels], #1, 7f\n" - "ldr d12, [x19], #0x8\n" + "ldr d12, [x20], #0x8\n" "tbz %x[n_channels], #0, 8f\n" - "ld1 { v12.s }[2], [x19]\n" + "ld1 { v12.s }[2], [x20]\n" "b 8f\n" "7:" // Tile loop: Oddments: Load inputs: (4, 4): Bit 1: Unset - "ldr s12, [x19, #0x0]\n" + "ldr s12, [x20, #0x0]\n" "8:" // Tile loop: Oddments: Load inputs: (4, 4): Bit 1: End - "fmla v23.4s, v8.4s, v12.4s\n" - "add x19, x11, x16\n" + "fmla v31.4s, v8.4s, v12.4s\n" + "add x20, x12, x8\n" "tbz %x[n_channels], #1, 9f\n" - "ldr d11, [x19], #0x8\n" + "ldr d11, [x20], #0x8\n" "tbz %x[n_channels], #0, 10f\n" - "ld1 { v11.s }[2], [x19]\n" + "ld1 { v11.s }[2], [x20]\n" "b 10f\n" "9:" // Tile loop: Oddments: Load inputs: (2, 1): Bit 1: Unset - "ldr s11, [x19, #0x0]\n" + "ldr s11, [x20, #0x0]\n" "10:" // Tile loop: Oddments: Load inputs: (2, 1): Bit 1: End - "fmla v31.4s, v7.4s, v11.4s\n" - "add x19, x15, x16\n" - "fmla v30.4s, v6.4s, v11.4s\n" - "fmla v28.4s, v4.4s, v11.4s\n" + "fmla v23.4s, v7.4s, v11.4s\n" + "fmla v24.4s, v6.4s, v11.4s\n" + "add x20, x16, x8\n" + "fmla v26.4s, v4.4s, v11.4s\n" "fmla v27.4s, v3.4s, v11.4s\n" - "fmla v25.4s, v1.4s, v11.4s\n" - "fmla v24.4s, v0.4s, v11.4s\n" + "fmla v29.4s, v1.4s, v11.4s\n" + "fmla v30.4s, v0.4s, v11.4s\n" "tbz %x[n_channels], #1, 11f\n" - "ldr d13, [x19], #0x8\n" + "ldr d13, [x20], #0x8\n" "tbz %x[n_channels], #0, 12f\n" - "ld1 { v13.s }[2], [x19]\n" + "ld1 { v13.s }[2], [x20]\n" "b 12f\n" "11:" // Tile loop: Oddments: Load inputs: (0, 1): Bit 1: Unset - "ldr s13, [x19, #0x0]\n" + "ldr s13, [x20, #0x0]\n" "12:" // Tile loop: Oddments: Load inputs: (0, 1): Bit 1: End - "fmla v31.4s, v1.4s, v13.4s\n" - "add x19, x15, x27\n" - "fmla v30.4s, v0.4s, v13.4s\n" + "fmla v23.4s, v1.4s, v13.4s\n" + "fmla v24.4s, v0.4s, v13.4s\n" + "add x20, x16, x9\n" "tbz %x[n_channels], #1, 13f\n" - "ldr d12, [x19], #0x8\n" + "ldr d12, [x20], #0x8\n" "tbz %x[n_channels], #0, 14f\n" - "ld1 { v12.s }[2], [x19]\n" + "ld1 { v12.s }[2], [x20]\n" "b 14f\n" "13:" // Tile loop: Oddments: Load inputs: (0, 3): Bit 1: Unset - "ldr s12, [x19, #0x0]\n" + "ldr s12, [x20, #0x0]\n" "14:" // Tile loop: Oddments: Load inputs: (0, 3): Bit 1: End - "fmla v30.4s, v2.4s, v12.4s\n" - "add x19, x11, x27\n" - "fmla v29.4s, v1.4s, v12.4s\n" + "fmla v24.4s, v2.4s, v12.4s\n" + "fmla v25.4s, v1.4s, v12.4s\n" + "add x20, x12, x9\n" "tbz %x[n_channels], #1, 15f\n" - "ldr d10, [x19], #0x8\n" + "ldr d10, [x20], #0x8\n" "tbz %x[n_channels], #0, 16f\n" - "ld1 { v10.s }[2], [x19]\n" + "ld1 { v10.s }[2], [x20]\n" "b 16f\n" "15:" // Tile loop: Oddments: Load inputs: (2, 3): Bit 1: Unset - "ldr s10, [x19, #0x0]\n" + "ldr s10, [x20, #0x0]\n" "16:" // Tile loop: Oddments: Load inputs: (2, 3): Bit 1: End - "fmla v30.4s, v8.4s, v10.4s\n" - "add x19, x12, XZR\n" - "fmla v29.4s, v7.4s, v10.4s\n" + "fmla v24.4s, v8.4s, v10.4s\n" + "fmla v25.4s, v7.4s, v10.4s\n" + "add x20, x13, XZR\n" "fmla v27.4s, v5.4s, v10.4s\n" - "fmla v26.4s, v4.4s, v10.4s\n" - "fmla v24.4s, v2.4s, v10.4s\n" - "fmla v23.4s, v1.4s, v10.4s\n" + "fmla v28.4s, v4.4s, v10.4s\n" + "fmla v30.4s, v2.4s, v10.4s\n" + "fmla v31.4s, v1.4s, v10.4s\n" "tbz %x[n_channels], #1, 17f\n" - "ldr d11, [x19], #0x8\n" + "ldr d11, [x20], #0x8\n" "tbz %x[n_channels], #0, 18f\n" - "ld1 { v11.s }[2], [x19]\n" + "ld1 { v11.s }[2], [x20]\n" "b 18f\n" "17:" // Tile loop: Oddments: Load inputs: (1, 0): Bit 1: Unset - "ldr s11, [x19, #0x0]\n" + "ldr s11, [x20, #0x0]\n" "18:" // Tile loop: Oddments: Load inputs: (1, 0): Bit 1: End - "fmla v31.4s, v3.4s, v11.4s\n" - "add x19, x12, x26\n" - "fmla v28.4s, v0.4s, v11.4s\n" + "fmla v23.4s, v3.4s, v11.4s\n" + "fmla v26.4s, v0.4s, v11.4s\n" + "add x20, x13, x26\n" "tbz %x[n_channels], #1, 19f\n" - "ldr d13, [x19], #0x8\n" + "ldr d13, [x20], #0x8\n" "tbz %x[n_channels], #0, 20f\n" - "ld1 { v13.s }[2], [x19]\n" + "ld1 { v13.s }[2], [x20]\n" "b 20f\n" "19:" // Tile loop: Oddments: Load inputs: (1, 4): Bit 1: Unset - "ldr s13, [x19, #0x0]\n" + "ldr s13, [x20, #0x0]\n" "20:" // Tile loop: Oddments: Load inputs: (1, 4): Bit 1: End - "fmla v29.4s, v5.4s, v13.4s\n" - "add x19, x10, XZR\n" - "fmla v26.4s, v2.4s, v13.4s\n" + "fmla v25.4s, v5.4s, v13.4s\n" + "fmla v28.4s, v2.4s, v13.4s\n" + "add x20, x10, XZR\n" "tbz %x[n_channels], #1, 21f\n" - "ldr d12, [x19], #0x8\n" + "ldr d12, [x20], #0x8\n" "tbz %x[n_channels], #0, 22f\n" - "ld1 { v12.s }[2], [x19]\n" + "ld1 { v12.s }[2], [x20]\n" "b 22f\n" "21:" // Tile loop: Oddments: Load inputs: (3, 0): Bit 1: Unset - "ldr s12, [x19, #0x0]\n" + "ldr s12, [x20, #0x0]\n" "22:" // Tile loop: Oddments: Load inputs: (3, 0): Bit 1: End - "fmla v28.4s, v6.4s, v12.4s\n" - "add x19, x10, x28\n" - "fmla v25.4s, v3.4s, v12.4s\n" + "fmla v26.4s, v6.4s, v12.4s\n" + "fmla v29.4s, v3.4s, v12.4s\n" + "add x20, x10, x11\n" "tbz %x[n_channels], #1, 23f\n" - "ldr d10, [x19], #0x8\n" + "ldr d10, [x20], #0x8\n" "tbz %x[n_channels], #0, 24f\n" - "ld1 { v10.s }[2], [x19]\n" + "ld1 { v10.s }[2], [x20]\n" "b 24f\n" "23:" // Tile loop: Oddments: Load inputs: (3, 2): Bit 1: Unset - "ldr s10, [x19, #0x0]\n" + "ldr s10, [x20, #0x0]\n" "24:" // Tile loop: Oddments: Load inputs: (3, 2): Bit 1: End - "fmla v28.4s, v8.4s, v10.4s\n" - "add x19, x10, x26\n" + "fmla v26.4s, v8.4s, v10.4s\n" "fmla v27.4s, v7.4s, v10.4s\n" - "fmla v26.4s, v6.4s, v10.4s\n" - "fmla v25.4s, v5.4s, v10.4s\n" - "fmla v24.4s, v4.4s, v10.4s\n" - "fmla v23.4s, v3.4s, v10.4s\n" + "add x20, x10, x26\n" + "fmla v28.4s, v6.4s, v10.4s\n" + "fmla v29.4s, v5.4s, v10.4s\n" + "fmla v30.4s, v4.4s, v10.4s\n" + "fmla v31.4s, v3.4s, v10.4s\n" "tbz %x[n_channels], #1, 25f\n" - "ldr d11, [x19], #0x8\n" + "ldr d11, [x20], #0x8\n" "tbz %x[n_channels], #0, 26f\n" - "ld1 { v11.s }[2], [x19]\n" + "ld1 { v11.s }[2], [x20]\n" "b 26f\n" "25:" // Tile loop: Oddments: Load inputs: (3, 4): Bit 1: Unset - "ldr s11, [x19, #0x0]\n" + "ldr s11, [x20, #0x0]\n" "26:" // Tile loop: Oddments: Load inputs: (3, 4): Bit 1: End - "fmla v26.4s, v8.4s, v11.4s\n" - "add x19, x9, x16\n" - "fmla v23.4s, v5.4s, v11.4s\n" + "fmla v28.4s, v8.4s, v11.4s\n" + "fmla v31.4s, v5.4s, v11.4s\n" + "add x20, x27, x8\n" "tbz %x[n_channels], #1, 27f\n" - "ldr d13, [x19], #0x8\n" + "ldr d13, [x20], #0x8\n" "tbz %x[n_channels], #0, 28f\n" - "ld1 { v13.s }[2], [x19]\n" + "ld1 { v13.s }[2], [x20]\n" "b 28f\n" "27:" // Tile loop: Oddments: Load inputs: (4, 1): Bit 1: Unset - "ldr s13, [x19, #0x0]\n" + "ldr s13, [x20, #0x0]\n" "28:" // Tile loop: Oddments: Load inputs: (4, 1): Bit 1: End - "fmla v25.4s, v7.4s, v13.4s\n" - "add x19, x12, x16\n" - "fmla v24.4s, v6.4s, v13.4s\n" + "fmla v29.4s, v7.4s, v13.4s\n" + "fmla v30.4s, v6.4s, v13.4s\n" + "add x20, x13, x8\n" "tbz %x[n_channels], #1, 29f\n" - "ldr d12, [x19], #0x8\n" + "ldr d12, [x20], #0x8\n" "tbz %x[n_channels], #0, 30f\n" - "ld1 { v12.s }[2], [x19]\n" + "ld1 { v12.s }[2], [x20]\n" "b 30f\n" "29:" // Tile loop: Oddments: Load inputs: (1, 1): Bit 1: Unset - "ldr s12, [x19, #0x0]\n" + "ldr s12, [x20, #0x0]\n" "30:" // Tile loop: Oddments: Load inputs: (1, 1): Bit 1: End - "fmla v31.4s, v4.4s, v12.4s\n" - "add x19, x12, x27\n" - "fmla v30.4s, v3.4s, v12.4s\n" - "fmla v28.4s, v1.4s, v12.4s\n" + "fmla v23.4s, v4.4s, v12.4s\n" + "fmla v24.4s, v3.4s, v12.4s\n" + "add x20, x13, x9\n" + "fmla v26.4s, v1.4s, v12.4s\n" "fmla v27.4s, v0.4s, v12.4s\n" "tbz %x[n_channels], #1, 31f\n" - "ldr d11, [x19], #0x8\n" + "ldr d11, [x20], #0x8\n" "tbz %x[n_channels], #0, 32f\n" - "ld1 { v11.s }[2], [x19]\n" + "ld1 { v11.s }[2], [x20]\n" "b 32f\n" "31:" // Tile loop: Oddments: Load inputs: (1, 3): Bit 1: Unset - "ldr s11, [x19, #0x0]\n" + "ldr s11, [x20, #0x0]\n" "32:" // Tile loop: Oddments: Load inputs: (1, 3): Bit 1: End - "fmla v30.4s, v5.4s, v11.4s\n" - "add x19, x9, x27\n" - "fmla v29.4s, v4.4s, v11.4s\n" + "fmla v24.4s, v5.4s, v11.4s\n" + "fmla v25.4s, v4.4s, v11.4s\n" + "add x20, x27, x9\n" "fmla v27.4s, v2.4s, v11.4s\n" - "fmla v26.4s, v1.4s, v11.4s\n" + "fmla v28.4s, v1.4s, v11.4s\n" "tbz %x[n_channels], #1, 33f\n" - "ldr d13, [x19], #0x8\n" + "ldr d13, [x20], #0x8\n" "tbz %x[n_channels], #0, 34f\n" - "ld1 { v13.s }[2], [x19]\n" + "ld1 { v13.s }[2], [x20]\n" "b 34f\n" "33:" // Tile loop: Oddments: Load inputs: (4, 3): Bit 1: Unset - "ldr s13, [x19, #0x0]\n" + "ldr s13, [x20, #0x0]\n" "34:" // Tile loop: Oddments: Load inputs: (4, 3): Bit 1: End - "fmla v24.4s, v8.4s, v13.4s\n" - "add x19, x10, x16\n" - "fmla v23.4s, v7.4s, v13.4s\n" + "fmla v30.4s, v8.4s, v13.4s\n" + "fmla v31.4s, v7.4s, v13.4s\n" + "add x20, x10, x8\n" "tbz %x[n_channels], #1, 35f\n" - "ldr d12, [x19], #0x8\n" + "ldr d12, [x20], #0x8\n" "tbz %x[n_channels], #0, 36f\n" - "ld1 { v12.s }[2], [x19]\n" + "ld1 { v12.s }[2], [x20]\n" "b 36f\n" "35:" // Tile loop: Oddments: Load inputs: (3, 1): Bit 1: Unset - "ldr s12, [x19, #0x0]\n" + "ldr s12, [x20, #0x0]\n" "36:" // Tile loop: Oddments: Load inputs: (3, 1): Bit 1: End - "fmla v28.4s, v7.4s, v12.4s\n" - "add x19, x15, x28\n" + "fmla v26.4s, v7.4s, v12.4s\n" "fmla v27.4s, v6.4s, v12.4s\n" - "fmla v25.4s, v4.4s, v12.4s\n" - "fmla v24.4s, v3.4s, v12.4s\n" + "add x20, x16, x11\n" + "fmla v29.4s, v4.4s, v12.4s\n" + "fmla v30.4s, v3.4s, v12.4s\n" "tbz %x[n_channels], #1, 37f\n" - "ldr d11, [x19], #0x8\n" + "ldr d11, [x20], #0x8\n" "tbz %x[n_channels], #0, 38f\n" - "ld1 { v11.s }[2], [x19]\n" + "ld1 { v11.s }[2], [x20]\n" "b 38f\n" "37:" // Tile loop: Oddments: Load inputs: (0, 2): Bit 1: Unset - "ldr s11, [x19, #0x0]\n" + "ldr s11, [x20, #0x0]\n" "38:" // Tile loop: Oddments: Load inputs: (0, 2): Bit 1: End - "fmla v31.4s, v2.4s, v11.4s\n" - "add x19, x10, x27\n" - "fmla v30.4s, v1.4s, v11.4s\n" - "fmla v29.4s, v0.4s, v11.4s\n" + "fmla v23.4s, v2.4s, v11.4s\n" + "fmla v24.4s, v1.4s, v11.4s\n" + "add x20, x10, x9\n" + "fmla v25.4s, v0.4s, v11.4s\n" "tbz %x[n_channels], #1, 39f\n" - "ldr d13, [x19], #0x8\n" + "ldr d13, [x20], #0x8\n" "tbz %x[n_channels], #0, 40f\n" - "ld1 { v13.s }[2], [x19]\n" + "ld1 { v13.s }[2], [x20]\n" "b 40f\n" "39:" // Tile loop: Oddments: Load inputs: (3, 3): Bit 1: Unset - "ldr s13, [x19, #0x0]\n" + "ldr s13, [x20, #0x0]\n" "40:" // Tile loop: Oddments: Load inputs: (3, 3): Bit 1: End "fmla v27.4s, v8.4s, v13.4s\n" - "add x19, x11, XZR\n" - "fmla v26.4s, v7.4s, v13.4s\n" - "fmla v24.4s, v5.4s, v13.4s\n" - "fmla v23.4s, v4.4s, v13.4s\n" + "fmla v28.4s, v7.4s, v13.4s\n" + "add x20, x12, XZR\n" + "fmla v30.4s, v5.4s, v13.4s\n" + "fmla v31.4s, v4.4s, v13.4s\n" "tbz %x[n_channels], #1, 41f\n" - "ldr d12, [x19], #0x8\n" + "ldr d12, [x20], #0x8\n" "tbz %x[n_channels], #0, 42f\n" - "ld1 { v12.s }[2], [x19]\n" + "ld1 { v12.s }[2], [x20]\n" "b 42f\n" "41:" // Tile loop: Oddments: Load inputs: (2, 0): Bit 1: Unset - "ldr s12, [x19, #0x0]\n" + "ldr s12, [x20, #0x0]\n" "42:" // Tile loop: Oddments: Load inputs: (2, 0): Bit 1: End - "fmla v31.4s, v6.4s, v12.4s\n" - "add x19, x11, x26\n" - "fmla v28.4s, v3.4s, v12.4s\n" - "fmla v25.4s, v0.4s, v12.4s\n" + "fmla v23.4s, v6.4s, v12.4s\n" + "fmla v26.4s, v3.4s, v12.4s\n" + "add x20, x12, x26\n" + "fmla v29.4s, v0.4s, v12.4s\n" "tbz %x[n_channels], #1, 43f\n" - "ldr d11, [x19], #0x8\n" + "ldr d11, [x20], #0x8\n" "tbz %x[n_channels], #0, 44f\n" - "ld1 { v11.s }[2], [x19]\n" + "ld1 { v11.s }[2], [x20]\n" "b 44f\n" "43:" // Tile loop: Oddments: Load inputs: (2, 4): Bit 1: Unset - "ldr s11, [x19, #0x0]\n" + "ldr s11, [x20, #0x0]\n" "44:" // Tile loop: Oddments: Load inputs: (2, 4): Bit 1: End - "fmla v29.4s, v8.4s, v11.4s\n" - "add x19, x9, x28\n" - "fmla v26.4s, v5.4s, v11.4s\n" - "fmla v23.4s, v2.4s, v11.4s\n" + "fmla v25.4s, v8.4s, v11.4s\n" + "fmla v28.4s, v5.4s, v11.4s\n" + "add x20, x27, x11\n" + "fmla v31.4s, v2.4s, v11.4s\n" "tbz %x[n_channels], #1, 45f\n" - "ldr d13, [x19], #0x8\n" + "ldr d13, [x20], #0x8\n" "tbz %x[n_channels], #0, 46f\n" - "ld1 { v13.s }[2], [x19]\n" + "ld1 { v13.s }[2], [x20]\n" "b 46f\n" "45:" // Tile loop: Oddments: Load inputs: (4, 2): Bit 1: Unset - "ldr s13, [x19, #0x0]\n" + "ldr s13, [x20, #0x0]\n" "46:" // Tile loop: Oddments: Load inputs: (4, 2): Bit 1: End - "fmla v25.4s, v8.4s, v13.4s\n" - "fmla v24.4s, v7.4s, v13.4s\n" - "fmla v23.4s, v6.4s, v13.4s\n" - "fmax v31.4s, v31.4s, v18.4s\n" - "fmax v30.4s, v30.4s, v18.4s\n" - "fmax v29.4s, v29.4s, v18.4s\n" - "fmin v31.4s, v31.4s, v17.4s\n" - "fmin v30.4s, v30.4s, v17.4s\n" - "fmin v29.4s, v29.4s, v17.4s\n" - "fmax v28.4s, v28.4s, v18.4s\n" - "fmax v27.4s, v27.4s, v18.4s\n" - "fmax v26.4s, v26.4s, v18.4s\n" - "fmin v28.4s, v28.4s, v17.4s\n" - "fmin v27.4s, v27.4s, v17.4s\n" - "fmin v26.4s, v26.4s, v17.4s\n" - "fmax v25.4s, v25.4s, v18.4s\n" - "fmax v24.4s, v24.4s, v18.4s\n" + "fmla v29.4s, v8.4s, v13.4s\n" + "fmla v30.4s, v7.4s, v13.4s\n" "fmax v23.4s, v23.4s, v18.4s\n" - "fmin v25.4s, v25.4s, v17.4s\n" - "fmin v24.4s, v24.4s, v17.4s\n" + "fmla v31.4s, v6.4s, v13.4s\n" + "fmax v24.4s, v24.4s, v18.4s\n" + "fmax v25.4s, v25.4s, v18.4s\n" + "fmax v26.4s, v26.4s, v18.4s\n" + "fmax v27.4s, v27.4s, v18.4s\n" + "fmax v28.4s, v28.4s, v18.4s\n" + "fmax v29.4s, v29.4s, v18.4s\n" + "fmax v30.4s, v30.4s, v18.4s\n" + "fmax v31.4s, v31.4s, v18.4s\n" "fmin v23.4s, v23.4s, v17.4s\n" + "fmin v24.4s, v24.4s, v17.4s\n" + "fmin v25.4s, v25.4s, v17.4s\n" + "fmin v26.4s, v26.4s, v17.4s\n" + "fmin v27.4s, v27.4s, v17.4s\n" + "fmin v28.4s, v28.4s, v17.4s\n" + "fmin v29.4s, v29.4s, v17.4s\n" + "fmin v30.4s, v30.4s, v17.4s\n" + "fmin v31.4s, v31.4s, v17.4s\n" "tbz %x[n_channels], #1, 47f\n" - "mov x19, x13\n" - "st1 { v31.d }[0], [x19], x14\n" - "add x13, x13, #0x8\n" - "st1 { v30.d }[0], [x19], x14\n" + "mov x22, x15\n" + "mov x21, x28\n" + "st1 { v23.d }[0], [x22], x17\n" "mov x20, x25\n" - "st1 { v29.d }[0], [x19]\n" - "st1 { v28.d }[0], [x20], x14\n" + "st1 { v26.d }[0], [x21], x17\n" + "add x15, x15, #0x8\n" + "st1 { v29.d }[0], [x20], x17\n" + "add x28, x28, #0x8\n" "add x25, x25, #0x8\n" - "st1 { v27.d }[0], [x20], x14\n" - "mov x19, x24\n" - "st1 { v26.d }[0], [x20]\n" - "add x24, x24, #0x8\n" - "st1 { v25.d }[0], [x19], x14\n" - "st1 { v24.d }[0], [x19], x14\n" - "st1 { v23.d }[0], [x19]\n" + "st1 { v24.d }[0], [x22], x17\n" + "st1 { v27.d }[0], [x21], x17\n" + "st1 { v30.d }[0], [x20], x17\n" + "st1 { v25.d }[0], [x22]\n" + "st1 { v28.d }[0], [x21]\n" + "st1 { v31.d }[0], [x20]\n" "tbz %x[n_channels], #0, 48f\n" - "mov x21, x13\n" - "st1 { v31.s }[2], [x21], x14\n" + "mov x22, x15\n" + "mov x21, x28\n" + "st1 { v23.s }[2], [x22], x17\n" "mov x20, x25\n" - "st1 { v30.s }[2], [x21], x14\n" - "st1 { v28.s }[2], [x20], x14\n" - "mov x19, x24\n" - "st1 { v29.s }[2], [x21]\n" - "st1 { v27.s }[2], [x20], x14\n" - "st1 { v26.s }[2], [x20]\n" - "st1 { v25.s }[2], [x19], x14\n" - "st1 { v24.s }[2], [x19], x14\n" - "st1 { v23.s }[2], [x19]\n" + "st1 { v26.s }[2], [x21], x17\n" + "st1 { v29.s }[2], [x20], x17\n" + "st1 { v24.s }[2], [x22], x17\n" + "st1 { v27.s }[2], [x21], x17\n" + "st1 { v30.s }[2], [x20], x17\n" + "st1 { v25.s }[2], [x22]\n" + "st1 { v28.s }[2], [x21]\n" + "st1 { v31.s }[2], [x20]\n" "b 48f\n" "47:" // Tile loop: Oddments: Store: Bit 1: Unset - "mov x21, x13\n" - "st1 { v31.s }[0], [x21], x14\n" + "mov x22, x15\n" + "mov x21, x28\n" + "st1 { v23.s }[0], [x22], x17\n" "mov x20, x25\n" - "mov x19, x24\n" - "st1 { v30.s }[0], [x21], x14\n" - "st1 { v28.s }[0], [x20], x14\n" - "st1 { v29.s }[0], [x21]\n" - "st1 { v27.s }[0], [x20], x14\n" - "st1 { v26.s }[0], [x20]\n" - "st1 { v25.s }[0], [x19], x14\n" - "st1 { v24.s }[0], [x19], x14\n" - "st1 { v23.s }[0], [x19]\n" + "st1 { v26.s }[0], [x21], x17\n" + "st1 { v29.s }[0], [x20], x17\n" + "st1 { v24.s }[0], [x22], x17\n" + "st1 { v27.s }[0], [x21], x17\n" + "st1 { v30.s }[0], [x20], x17\n" + "st1 { v25.s }[0], [x22]\n" + "st1 { v28.s }[0], [x21]\n" + "st1 { v31.s }[0], [x20]\n" "48:" // Tile loop: Oddments: Store: Bit 1: 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 x23, [%x[params_struct], %[offsetof_args_tile_j]]\n" + "ldr x24, [%x[params_struct], %[offsetof_args_tile_i]]\n" + "add x23, x23, #0x1\n" + "add x21, x24, #0x1\n" + "ldr x20, [%x[params_struct], %[offsetof_args_n_tile_cols]]\n" + "cmp x23, x20\n" "ldr x20, [%x[params_struct], %[offsetof_args_n_tile_rows]]\n" - "add x8, x8, #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" + "csel x24, x24, x21, LT\n" + "csel x23, x23, XZR, LT\n" + "cmp x24, x20\n" "blt 1b\n" : : [n_channels] "r" ((unsigned long) n_channels), [offsetof_args_inptr] "I" (offsetof(Args, inptr)), [offsetof_args_ld_input_col] "I" (offsetof(Args, ld_input_col)), [offsetof_args_ld_input_row] "I" (offsetof(Args, ld_input_row)), [offsetof_args_ld_output_col] "I" (offsetof(Args, ld_output_col)), [offsetof_args_ld_output_row] "I" (offsetof(Args, ld_output_row)), [offsetof_args_max] "I" (offsetof(Args, max)), [offsetof_args_min] "I" (offsetof(Args, min)), [offsetof_args_n_tile_cols] "I" (offsetof(Args, n_tile_cols)), [offsetof_args_n_tile_rows] "I" (offsetof(Args, n_tile_rows)), [offsetof_args_outptr] "I" (offsetof(Args, outptr)), [offsetof_args_params] "I" (offsetof(Args, params)), [offsetof_args_tile_i] "I" (offsetof(Args, tile_i)), [offsetof_args_tile_j] "I" (offsetof(Args, tile_j)), [params_struct] "r" (¶ms_struct) - : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v8", "v9", "v10", "v11", "v12", "v13", "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", "x8", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x16", "x17", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28" ); } |