diff options
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/transforms/a64_transpose_interleave_12_2x2.hpp')
| -rw-r--r-- | src/core/NEON/kernels/arm_gemm/transforms/a64_transpose_interleave_12_2x2.hpp | 461 |
1 files changed, 241 insertions, 220 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/transforms/a64_transpose_interleave_12_2x2.hpp b/src/core/NEON/kernels/arm_gemm/transforms/a64_transpose_interleave_12_2x2.hpp index cef468e9cc..94726570d4 100644 --- a/src/core/NEON/kernels/arm_gemm/transforms/a64_transpose_interleave_12_2x2.hpp +++ b/src/core/NEON/kernels/arm_gemm/transforms/a64_transpose_interleave_12_2x2.hpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2021, 2023 Arm Limited. + * Copyright (c) 2021, 2023-2024 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -40,287 +40,308 @@ void a64_transpose_interleave_12_2x2(uint16_t *out, const uint16_t *in, size_t w __asm__ __volatile__( "cmp %x[height], #0x8\n" - "blt 10f\n" + "blt 11f\n" "1:" // Main row loop: Head "mov x9, %x[in]\n" - "add x28, x9, %x[in_stride]\n" - "add x27, x28, %x[in_stride]\n" - "add x26, x27, %x[in_stride]\n" + "mov x28, %x[width]\n" + "mov x27, %x[out]\n" + "sub %x[height], %x[height], #0x8\n" + "add x26, x9, %x[in_stride]\n" "add x25, x26, %x[in_stride]\n" - "mov x24, %x[width]\n" - "add x23, x25, %x[in_stride]\n" + "add x24, x25, %x[in_stride]\n" + "cmp x28, #0x18\n" + "add x23, x24, %x[in_stride]\n" "add x22, x23, %x[in_stride]\n" - "add x20, x22, %x[in_stride]\n" - "cmp x24, #0x18\n" + "add x21, x22, %x[in_stride]\n" + "add x20, x21, %x[in_stride]\n" "add %x[in], x20, %x[in_stride]\n" - "mov x21, %x[out]\n" - "sub %x[height], %x[height], #0x8\n" "blt 3f\n" "2:" // Main row loop: Unroll column loop - "ldr q19, [x9], #0x10\n" - "ldr q18, [x28], #0x10\n" - "sub x24, x24, #0x18\n" - "zip1 v10.8h, v19.8h, v18.8h\n" - "ldr q17, [x27], #0x10\n" - "ldr q16, [x26], #0x10\n" - "zip2 v9.8h, v19.8h, v18.8h\n" - "zip1 v8.8h, v17.8h, v16.8h\n" - "ldr q19, [x25], #0x10\n" + "ldr q22, [x9], #0x10\n" + "ldr q21, [x26], #0x10\n" + "sub x28, x28, #0x18\n" + "ldr q20, [x25], #0x10\n" + "ldr q19, [x24], #0x10\n" + "cmp x28, #0x18\n" "ldr q18, [x23], #0x10\n" - "zip2 v7.8h, v17.8h, v16.8h\n" - "zip1 v6.8h, v19.8h, v18.8h\n" "ldr q17, [x22], #0x10\n" + "ldr q23, [x21], #0x10\n" "ldr q16, [x20], #0x10\n" - "zip2 v5.8h, v19.8h, v18.8h\n" - "zip1 v4.8h, v17.8h, v16.8h\n" - "ldr q21, [x9], #0x10\n" - "ldr q18, [x28], #0x10\n" - "zip1 v3.8h, v21.8h, v18.8h\n" - "zip2 v2.8h, v17.8h, v16.8h\n" - "ldr q17, [x27], #0x10\n" - "ldr q16, [x26], #0x10\n" - "zip1 v1.8h, v17.8h, v16.8h\n" - "cmp x24, #0x18\n" + "zip1 v8.8h, v22.8h, v21.8h\n" + "zip2 v7.8h, v22.8h, v21.8h\n" + "ldr q22, [x9], #0x10\n" + "ldr q21, [x26], #0x10\n" + "zip1 v6.8h, v20.8h, v19.8h\n" + "zip2 v5.8h, v20.8h, v19.8h\n" "ldr q20, [x25], #0x10\n" - "ldr q19, [x23], #0x10\n" - "zip1 v0.8h, v20.8h, v19.8h\n" - "zip2 v31.8h, v21.8h, v18.8h\n" - "ldr q30, [x22], #0x10\n" - "ldr q29, [x20], #0x10\n" - "zip1 v28.8h, v30.8h, v29.8h\n" - "zip2 v27.8h, v17.8h, v16.8h\n" - "ldr q17, [x9], #0x10\n" - "ldr q16, [x28], #0x10\n" - "zip1 v26.8h, v17.8h, v16.8h\n" - "zip2 v25.8h, v17.8h, v16.8h\n" - "ldr q17, [x27], #0x10\n" - "ldr q16, [x26], #0x10\n" - "zip1 v24.8h, v17.8h, v16.8h\n" - "zip2 v23.8h, v17.8h, v16.8h\n" - "ldr q18, [x25], #0x10\n" - "ldr q17, [x23], #0x10\n" - "zip2 v22.8h, v20.8h, v19.8h\n" - "zip1 v21.8h, v18.8h, v17.8h\n" - "ldr q20, [x22], #0x10\n" + "ldr q19, [x24], #0x10\n" + "zip1 v4.8h, v18.8h, v17.8h\n" + "zip2 v3.8h, v18.8h, v17.8h\n" + "ldr q18, [x23], #0x10\n" + "ldr q17, [x22], #0x10\n" + "zip1 v2.8h, v23.8h, v16.8h\n" + "zip2 v1.8h, v23.8h, v16.8h\n" + "ldr q23, [x21], #0x10\n" "ldr q16, [x20], #0x10\n" - "str q10, [x21, #0x0]\n" - "zip2 v19.8h, v18.8h, v17.8h\n" - "str q9, [x21, #0x10]\n" - "zip2 v18.8h, v30.8h, v29.8h\n" - "zip1 v17.8h, v20.8h, v16.8h\n" - "str q3, [x21, #0x20]\n" - "zip2 v16.8h, v20.8h, v16.8h\n" - "str q8, [x21, #0x30]\n" - "str q7, [x21, #0x40]\n" - "str q1, [x21, #0x50]\n" - "str q6, [x21, #0x60]\n" - "str q5, [x21, #0x70]\n" - "str q0, [x21, #0x80]\n" - "str q4, [x21, #0x90]\n" - "str q2, [x21, #0xa0]\n" - "str q28, [x21, #0xb0]\n" - "add x21, x21, %x[out_stride]\n" - "str q31, [x21, #0x0]\n" - "str q26, [x21, #0x10]\n" - "str q25, [x21, #0x20]\n" - "str q27, [x21, #0x30]\n" - "str q24, [x21, #0x40]\n" - "str q23, [x21, #0x50]\n" - "str q22, [x21, #0x60]\n" - "str q21, [x21, #0x70]\n" - "str q19, [x21, #0x80]\n" - "str q18, [x21, #0x90]\n" - "str q17, [x21, #0xa0]\n" - "str q16, [x21, #0xb0]\n" - "add x21, x21, %x[out_stride]\n" + "zip1 v0.8h, v22.8h, v21.8h\n" + "zip2 v31.8h, v22.8h, v21.8h\n" + "ldr q22, [x9], #0x10\n" + "ldr q21, [x26], #0x10\n" + "zip1 v30.8h, v20.8h, v19.8h\n" + "zip2 v29.8h, v20.8h, v19.8h\n" + "ldr q20, [x25], #0x10\n" + "ldr q19, [x24], #0x10\n" + "zip1 v28.8h, v18.8h, v17.8h\n" + "zip2 v27.8h, v18.8h, v17.8h\n" + "ldr q18, [x23], #0x10\n" + "ldr q17, [x22], #0x10\n" + "zip1 v26.8h, v23.8h, v16.8h\n" + "zip2 v25.8h, v23.8h, v16.8h\n" + "ldr q24, [x21], #0x10\n" + "ldr q16, [x20], #0x10\n" + "str q8, [x27, #0x0]\n" + "zip1 v23.8h, v22.8h, v21.8h\n" + "str q7, [x27, #0x10]\n" + "zip2 v22.8h, v22.8h, v21.8h\n" + "zip1 v21.8h, v20.8h, v19.8h\n" + "str q0, [x27, #0x20]\n" + "zip2 v20.8h, v20.8h, v19.8h\n" + "zip1 v19.8h, v18.8h, v17.8h\n" + "str q6, [x27, #0x30]\n" + "zip2 v18.8h, v18.8h, v17.8h\n" + "zip1 v17.8h, v24.8h, v16.8h\n" + "str q5, [x27, #0x40]\n" + "zip2 v16.8h, v24.8h, v16.8h\n" + "str q30, [x27, #0x50]\n" + "str q4, [x27, #0x60]\n" + "str q3, [x27, #0x70]\n" + "str q28, [x27, #0x80]\n" + "str q2, [x27, #0x90]\n" + "str q1, [x27, #0xa0]\n" + "str q26, [x27, #0xb0]\n" + "add x27, x27, %x[out_stride]\n" + "str q31, [x27, #0x0]\n" + "str q23, [x27, #0x10]\n" + "str q22, [x27, #0x20]\n" + "str q29, [x27, #0x30]\n" + "str q21, [x27, #0x40]\n" + "str q20, [x27, #0x50]\n" + "str q27, [x27, #0x60]\n" + "str q19, [x27, #0x70]\n" + "str q18, [x27, #0x80]\n" + "str q25, [x27, #0x90]\n" + "str q17, [x27, #0xa0]\n" + "str q16, [x27, #0xb0]\n" + "add x27, x27, %x[out_stride]\n" "bge 2b\n" "3:" // Main row loop: Unroll column loop skip - "cmp x24, #0xc\n" + "cmp x28, #0xc\n" "blt 5f\n" "4:" // Main row loop: Column loop - "ldr q17, [x9], #0x10\n" - "ldr q16, [x28], #0x10\n" - "sub x24, x24, #0xc\n" - "cmp x24, #0xc\n" - "ldr q19, [x27], #0x10\n" - "ldr q18, [x26], #0x10\n" - "zip1 v28.8h, v17.8h, v16.8h\n" - "zip2 v27.8h, v17.8h, v16.8h\n" - "ldr q17, [x25], #0x10\n" - "ldr q16, [x23], #0x10\n" - "zip1 v26.8h, v19.8h, v18.8h\n" - "zip2 v25.8h, v19.8h, v18.8h\n" - "ldr q19, [x22], #0x10\n" - "ldr q18, [x20], #0x10\n" - "zip1 v24.8h, v17.8h, v16.8h\n" - "zip2 v23.8h, v17.8h, v16.8h\n" - "ldr d17, [x9], #0x8\n" - "ldr d16, [x28], #0x8\n" - "zip1 v22.8h, v17.8h, v16.8h\n" - "zip1 v21.8h, v19.8h, v18.8h\n" - "ldr d17, [x27], #0x8\n" - "ldr d16, [x26], #0x8\n" - "zip1 v20.8h, v17.8h, v16.8h\n" - "zip2 v19.8h, v19.8h, v18.8h\n" - "ldr d17, [x25], #0x8\n" - "ldr d16, [x23], #0x8\n" - "zip1 v18.8h, v17.8h, v16.8h\n" + "ldr q23, [x9], #0x10\n" + "ldr q19, [x26], #0x10\n" + "sub x28, x28, #0xc\n" + "ldr q22, [x25], #0x10\n" + "ldr q18, [x24], #0x10\n" + "cmp x28, #0xc\n" + "ldr q21, [x23], #0x10\n" + "ldr q17, [x22], #0x10\n" + "ldr q20, [x21], #0x10\n" + "ldr q16, [x20], #0x10\n" + "zip1 v31.8h, v23.8h, v19.8h\n" + "zip2 v30.8h, v23.8h, v19.8h\n" + "ldr d29, [x9], #0x8\n" + "ldr d19, [x26], #0x8\n" + "zip1 v28.8h, v22.8h, v18.8h\n" + "zip2 v27.8h, v22.8h, v18.8h\n" + "ldr d26, [x25], #0x8\n" + "ldr d18, [x24], #0x8\n" + "zip1 v25.8h, v21.8h, v17.8h\n" + "zip2 v24.8h, v21.8h, v17.8h\n" + "ldr d23, [x23], #0x8\n" "ldr d17, [x22], #0x8\n" + "zip1 v22.8h, v20.8h, v16.8h\n" + "zip2 v21.8h, v20.8h, v16.8h\n" + "ldr d20, [x21], #0x8\n" "ldr d16, [x20], #0x8\n" - "zip1 v16.8h, v17.8h, v16.8h\n" - "str q28, [x21, #0x0]\n" - "str q27, [x21, #0x10]\n" - "str q22, [x21, #0x20]\n" - "str q26, [x21, #0x30]\n" - "str q25, [x21, #0x40]\n" - "str q20, [x21, #0x50]\n" - "str q24, [x21, #0x60]\n" - "str q23, [x21, #0x70]\n" - "str q18, [x21, #0x80]\n" - "str q21, [x21, #0x90]\n" - "str q19, [x21, #0xa0]\n" - "str q16, [x21, #0xb0]\n" - "add x21, x21, %x[out_stride]\n" + "zip1 v19.8h, v29.8h, v19.8h\n" + "str q31, [x27, #0x0]\n" + "zip1 v18.8h, v26.8h, v18.8h\n" + "str q30, [x27, #0x10]\n" + "zip1 v17.8h, v23.8h, v17.8h\n" + "zip1 v16.8h, v20.8h, v16.8h\n" + "str q19, [x27, #0x20]\n" + "str q28, [x27, #0x30]\n" + "str q27, [x27, #0x40]\n" + "str q18, [x27, #0x50]\n" + "str q25, [x27, #0x60]\n" + "str q24, [x27, #0x70]\n" + "str q17, [x27, #0x80]\n" + "str q22, [x27, #0x90]\n" + "str q21, [x27, #0xa0]\n" + "str q16, [x27, #0xb0]\n" + "add x27, x27, %x[out_stride]\n" "bge 4b\n" "5:" // Main row loop: Column loop skip - "cmp x24, #0x4\n" + "cbz x28, 10f\n" + "cmp x28, #0x4\n" + "movi v16.8h, #0x0\n" + "str q16, [x27, #0x0]\n" + "str q16, [x27, #0x10]\n" + "str q16, [x27, #0x20]\n" + "str q16, [x27, #0x30]\n" + "str q16, [x27, #0x40]\n" + "str q16, [x27, #0x50]\n" + "str q16, [x27, #0x60]\n" + "str q16, [x27, #0x70]\n" + "str q16, [x27, #0x80]\n" + "str q16, [x27, #0x90]\n" + "str q16, [x27, #0xa0]\n" + "str q16, [x27, #0xb0]\n" "blt 7f\n" "6:" // Main row loop: width 4 loop: loop - "ldr d19, [x9], #0x8\n" - "ldr d18, [x28], #0x8\n" - "sub x24, x24, #0x4\n" - "cmp x24, #0x4\n" - "ldr d17, [x27], #0x8\n" - "ldr d16, [x26], #0x8\n" - "zip1 v20.8h, v19.8h, v18.8h\n" - "zip1 v19.8h, v17.8h, v16.8h\n" - "ldr d17, [x25], #0x8\n" - "ldr d16, [x23], #0x8\n" - "zip1 v18.8h, v17.8h, v16.8h\n" + "ldr d23, [x9], #0x8\n" + "ldr d18, [x26], #0x8\n" + "sub x28, x28, #0x4\n" + "ldr d22, [x25], #0x8\n" + "ldr d16, [x24], #0x8\n" + "cmp x28, #0x4\n" + "ldr d21, [x23], #0x8\n" "ldr d17, [x22], #0x8\n" - "ldr d16, [x20], #0x8\n" - "str q20, [x21, #0x0]\n" - "zip1 v16.8h, v17.8h, v16.8h\n" - "str q19, [x21, #0x30]\n" - "str q18, [x21, #0x60]\n" - "str q16, [x21, #0x90]\n" - "add x21, x21, #0x10\n" + "ldr d20, [x21], #0x8\n" + "ldr d19, [x20], #0x8\n" + "zip1 v18.8h, v23.8h, v18.8h\n" + "zip1 v16.8h, v22.8h, v16.8h\n" + "zip1 v17.8h, v21.8h, v17.8h\n" + "str q18, [x27, #0x0]\n" + "str q16, [x27, #0x30]\n" + "zip1 v16.8h, v20.8h, v19.8h\n" + "str q17, [x27, #0x60]\n" + "str q16, [x27, #0x90]\n" + "add x27, x27, #0x10\n" "bge 6b\n" "7:" // Main row loop: width 4 loop: skip - "cmp x24, #0x1\n" + "cmp x28, #0x1\n" "blt 9f\n" "8:" // Main row loop: width 1 loop: loop - "ldr h19, [x9], #0x2\n" - "ldr h18, [x28], #0x2\n" - "sub x24, x24, #0x1\n" - "cmp x24, #0x1\n" - "ldr h17, [x27], #0x2\n" - "ldr h16, [x26], #0x2\n" - "zip1 v20.8h, v19.8h, v18.8h\n" - "zip1 v19.8h, v17.8h, v16.8h\n" - "ldr h17, [x25], #0x2\n" - "ldr h16, [x23], #0x2\n" - "zip1 v18.8h, v17.8h, v16.8h\n" + "ldr h23, [x9], #0x2\n" + "ldr h18, [x26], #0x2\n" + "sub x28, x28, #0x1\n" + "ldr h22, [x25], #0x2\n" + "ldr h16, [x24], #0x2\n" + "cmp x28, #0x1\n" + "ldr h21, [x23], #0x2\n" "ldr h17, [x22], #0x2\n" - "ldr h16, [x20], #0x2\n" - "str s20, [x21, #0x0]\n" - "zip1 v16.8h, v17.8h, v16.8h\n" - "str s19, [x21, #0x30]\n" - "str s18, [x21, #0x60]\n" - "str s16, [x21, #0x90]\n" - "add x21, x21, #0x4\n" + "ldr h20, [x21], #0x2\n" + "ldr h19, [x20], #0x2\n" + "zip1 v18.8h, v23.8h, v18.8h\n" + "zip1 v16.8h, v22.8h, v16.8h\n" + "zip1 v17.8h, v21.8h, v17.8h\n" + "str s18, [x27, #0x0]\n" + "str s16, [x27, #0x30]\n" + "zip1 v16.8h, v20.8h, v19.8h\n" + "str s17, [x27, #0x60]\n" + "str s16, [x27, #0x90]\n" + "add x27, x27, #0x4\n" "bge 8b\n" "9:" // Main row loop: width 1 loop: skip + "10:" // Main row loop: odd col skip "cmp %x[height], #0x8\n" "add %x[out], %x[out], #0xc0\n" "bge 1b\n" - "cbz %x[height], 20f\n" - "10:" // Main loop skip - "11:" // Tail row loop: Head + "cbz %x[height], 22f\n" + "11:" // Main loop skip + "12:" // Tail row loop: Head "mov x9, %x[in]\n" "mov x20, %x[width]\n" - "add x28, x9, %x[in_stride]\n" "cmp %x[height], #0x1\n" - "add %x[in], x28, %x[in_stride]\n" - "csel x28, x28, %x[pad_row], GT\n" - "cmp x20, #0x18\n" - "mov x21, %x[out]\n" + "mov x27, %x[out]\n" "sub %x[height], %x[height], #0x2\n" - "blt 13f\n" - "12:" // Tail row loop: Unroll column loop - "ldr q17, [x9], #0x10\n" - "ldr q16, [x28], #0x10\n" - "sub x20, x20, #0x18\n" - "zip1 v22.8h, v17.8h, v16.8h\n" - "ldr q21, [x9], #0x10\n" - "ldr q18, [x28], #0x10\n" - "zip2 v17.8h, v17.8h, v16.8h\n" - "zip1 v20.8h, v21.8h, v18.8h\n" + "add x26, x9, %x[in_stride]\n" + "add %x[in], x26, %x[in_stride]\n" + "csel x26, x26, %x[pad_row], GT\n" + "cmp x20, #0x18\n" + "blt 14f\n" + "13:" // Tail row loop: Unroll column loop "ldr q19, [x9], #0x10\n" - "ldr q16, [x28], #0x10\n" - "str q22, [x21, #0x0]\n" + "ldr q16, [x26], #0x10\n" + "sub x20, x20, #0x18\n" + "ldr q22, [x9], #0x10\n" "cmp x20, #0x18\n" - "str q17, [x21, #0x10]\n" - "zip2 v18.8h, v21.8h, v18.8h\n" + "ldr q18, [x26], #0x10\n" + "ldr q21, [x9], #0x10\n" + "ldr q20, [x26], #0x10\n" "zip1 v17.8h, v19.8h, v16.8h\n" - "str q20, [x21, #0x20]\n" - "add x21, x21, %x[out_stride]\n" "zip2 v16.8h, v19.8h, v16.8h\n" - "str q18, [x21, #0x0]\n" - "str q17, [x21, #0x10]\n" - "str q16, [x21, #0x20]\n" - "add x21, x21, %x[out_stride]\n" - "bge 12b\n" - "13:" // Tail row loop: Unroll column loop skip + "zip1 v19.8h, v22.8h, v18.8h\n" + "zip2 v18.8h, v22.8h, v18.8h\n" + "str q17, [x27, #0x0]\n" + "str q16, [x27, #0x10]\n" + "zip1 v17.8h, v21.8h, v20.8h\n" + "zip2 v16.8h, v21.8h, v20.8h\n" + "str q19, [x27, #0x20]\n" + "add x27, x27, %x[out_stride]\n" + "str q18, [x27, #0x0]\n" + "str q17, [x27, #0x10]\n" + "str q16, [x27, #0x20]\n" + "add x27, x27, %x[out_stride]\n" + "bge 13b\n" + "14:" // Tail row loop: Unroll column loop skip "cmp x20, #0xc\n" - "blt 15f\n" - "14:" // Tail row loop: Column loop + "blt 16f\n" + "15:" // Tail row loop: Column loop "ldr q20, [x9], #0x10\n" - "ldr q17, [x28], #0x10\n" + "ldr q17, [x26], #0x10\n" "sub x20, x20, #0xc\n" - "cmp x20, #0xc\n" "ldr d19, [x9], #0x8\n" - "ldr d16, [x28], #0x8\n" + "cmp x20, #0xc\n" + "ldr d16, [x26], #0x8\n" "zip1 v18.8h, v20.8h, v17.8h\n" "zip2 v17.8h, v20.8h, v17.8h\n" "zip1 v16.8h, v19.8h, v16.8h\n" - "str q18, [x21, #0x0]\n" - "str q17, [x21, #0x10]\n" - "str q16, [x21, #0x20]\n" - "add x21, x21, %x[out_stride]\n" - "bge 14b\n" - "15:" // Tail row loop: Column loop skip + "str q18, [x27, #0x0]\n" + "str q17, [x27, #0x10]\n" + "str q16, [x27, #0x20]\n" + "add x27, x27, %x[out_stride]\n" + "bge 15b\n" + "16:" // Tail row loop: Column loop skip + "cbz x20, 21f\n" "cmp x20, #0x4\n" - "blt 17f\n" - "16:" // Tail row loop: width 4 loop: loop + "movi v16.8h, #0x0\n" + "str q16, [x27, #0x0]\n" + "str q16, [x27, #0x10]\n" + "str q16, [x27, #0x20]\n" + "blt 18f\n" + "17:" // Tail row loop: width 4 loop: loop "ldr d17, [x9], #0x8\n" - "ldr d16, [x28], #0x8\n" + "ldr d16, [x26], #0x8\n" "sub x20, x20, #0x4\n" "cmp x20, #0x4\n" "zip1 v16.8h, v17.8h, v16.8h\n" - "str q16, [x21, #0x0]\n" - "add x21, x21, #0x10\n" - "bge 16b\n" - "17:" // Tail row loop: width 4 loop: skip + "str q16, [x27, #0x0]\n" + "add x27, x27, #0x10\n" + "bge 17b\n" + "18:" // Tail row loop: width 4 loop: skip "cmp x20, #0x1\n" - "blt 19f\n" - "18:" // Tail row loop: width 1 loop: loop + "blt 20f\n" + "19:" // Tail row loop: width 1 loop: loop "ldr h17, [x9], #0x2\n" - "ldr h16, [x28], #0x2\n" + "ldr h16, [x26], #0x2\n" "sub x20, x20, #0x1\n" "cmp x20, #0x1\n" "zip1 v16.8h, v17.8h, v16.8h\n" - "str s16, [x21, #0x0]\n" - "add x21, x21, #0x4\n" - "bge 18b\n" - "19:" // Tail row loop: width 1 loop: skip + "str s16, [x27, #0x0]\n" + "add x27, x27, #0x4\n" + "bge 19b\n" + "20:" // Tail row loop: width 1 loop: skip + "21:" // Tail row loop: odd col skip "cmp %x[height], #0x1\n" "add %x[out], %x[out], #0x30\n" - "bge 11b\n" - "20:" // Done + "bge 12b\n" + "22:" // Done : [height] "+&r" (height), [in] "+&r" (in), [out] "+&r" (out) : [in_stride] "r" (in_stride), [out_stride] "r" (out_stride), [pad_row] "r" (pad_row), [width] "r" (width) - : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v8", "v9", "v10", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31", "x9", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28" + : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v8", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31", "x9", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28" ); } |