diff options
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block4_s8_s8_summing.hpp')
| -rw-r--r-- | src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block4_s8_s8_summing.hpp | 370 |
1 files changed, 370 insertions, 0 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block4_s8_s8_summing.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block4_s8_s8_summing.hpp new file mode 100644 index 0000000000..6c41b5fdfb --- /dev/null +++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave8_block4_s8_s8_summing.hpp @@ -0,0 +1,370 @@ +/* + * Copyright (c) 2019-2021, 2023 Arm Limited. + * + * SPDX-License-Identifier: MIT + * + * Permission is hereby granted, free of charge, to any person obtaining a copy + * of this software and associated documentation files (the "Software"), to + * deal in the Software without restriction, including without limitation the + * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or + * sell copies of the Software, and to permit persons to whom the Software is + * furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in all + * copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE + * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ + +#ifdef __aarch64__ + +template<> +void interleave_block<8, 4, VLType::None, true>( + int8_t * &out_ptr, const int8_t * const * in, size_t width, size_t height, + size_t row_offset, bool first +) +{ + __asm__ __volatile__( + "ldr x28, [%x[in], #0x0]\n" + "ldr x27, [%x[in], #0x8]\n" + "cmp %x[height], #0x8\n" + "mov x20, #0x0\n" + "ldr x26, [%x[in], #0x10]\n" + "ldr x25, [%x[in], #0x18]\n" + "movi v2.8h, #0x0\n" + "movi v1.8h, #0x0\n" + "ldr x24, [%x[in], #0x20]\n" + "ldr x23, [%x[in], #0x28]\n" + "movi v0.4s, #0x0\n" + "movi v31.4s, #0x0\n" + "ldr x22, [%x[in], #0x30]\n" + "ldr x21, [%x[in], #0x38]\n" + "add x28, x28, %x[row_offset]\n" + "add x27, x27, %x[row_offset]\n" + "add x26, x26, %x[row_offset]\n" + "add x25, x25, %x[row_offset]\n" + "add x24, x24, %x[row_offset]\n" + "add x23, x23, %x[row_offset]\n" + "add x22, x22, %x[row_offset]\n" + "add x21, x21, %x[row_offset]\n" + "beq 1f\n" + "cmp %x[height], #0x2\n" + "csel x27, x27, x28, GE\n" + "csel x26, x26, x28, GT\n" + "cmp %x[height], #0x4\n" + "csel x25, x25, x28, GE\n" + "csel x24, x24, x28, GT\n" + "cmp %x[height], #0x6\n" + "mov x21, x28\n" + "csel x23, x23, x28, GE\n" + "csel x22, x22, x28, GT\n" + "1:" // no_pointer_adj + "prfm pldl1keep, [x28, #0x0]\n" + "prfm pldl1keep, [x27, #0x0]\n" + "prfm pldl1keep, [x26, #0x0]\n" + "prfm pldl1keep, [x25, #0x0]\n" + "prfm pldl1keep, [x24, #0x0]\n" + "prfm pldl1keep, [x23, #0x0]\n" + "prfm pldl1keep, [x22, #0x0]\n" + "prfm pldl1keep, [x21, #0x0]\n" + "prfm pldl1keep, [x28, #0x40]\n" + "prfm pldl1keep, [x27, #0x40]\n" + "prfm pldl1keep, [x26, #0x40]\n" + "prfm pldl1keep, [x25, #0x40]\n" + "prfm pldl1keep, [x24, #0x40]\n" + "prfm pldl1keep, [x23, #0x40]\n" + "prfm pldl1keep, [x22, #0x40]\n" + "prfm pldl1keep, [x21, #0x40]\n" + "cbnz %w[first], 2f\n" + "sub %x[out_ptr], %x[out_ptr], #0x20\n" + "ld1 { v0.4s }, [%x[out_ptr]]\n" + "ldr q31, [%x[out_ptr], #0x10]\n" + "2:" // first_pass + "cmp %x[width], #0x10\n" + "blt 5f\n" + "3:" // Main loop head + "cmp x20, #0x1e\n" + "ble 4f\n" + "sadalp v0.4s, v2.8h\n" + "movi v2.8h, #0x0\n" + "mov x20, #0x0\n" + "sadalp v31.4s, v1.8h\n" + "movi v1.8h, #0x0\n" + "4:" // no_accumulate_16 + "ldr q30, [x28], #0x10\n" + "ldr q29, [x27], #0x10\n" + "subs %x[width], %x[width], #0x10\n" + "cmp %x[width], #0x10\n" + "ldr q28, [x26], #0x10\n" + "ldr q27, [x25], #0x10\n" + "zip1 v22.4s, v30.4s, v28.4s\n" + "zip1 v21.4s, v29.4s, v27.4s\n" + "ldr q20, [x24], #0x10\n" + "ldr q26, [x23], #0x10\n" + "zip1 v25.4s, v22.4s, v21.4s\n" + "sadalp v2.8h, v25.16b\n" + "ldr q19, [x22], #0x10\n" + "ldr q18, [x21], #0x10\n" + "zip1 v17.4s, v20.4s, v19.4s\n" + "zip1 v16.4s, v26.4s, v18.4s\n" + "zip1 v24.4s, v17.4s, v16.4s\n" + "sadalp v1.8h, v24.16b\n" + "prfm pldl1keep, [x28, #0x70]\n" + "prfm pldl1keep, [x27, #0x70]\n" + "zip2 v23.4s, v22.4s, v21.4s\n" + "zip2 v22.4s, v17.4s, v16.4s\n" + "prfm pldl1keep, [x26, #0x70]\n" + "prfm pldl1keep, [x25, #0x70]\n" + "zip2 v21.4s, v30.4s, v28.4s\n" + "zip2 v17.4s, v29.4s, v27.4s\n" + "prfm pldl1keep, [x24, #0x70]\n" + "prfm pldl1keep, [x23, #0x70]\n" + "zip2 v20.4s, v20.4s, v19.4s\n" + "zip2 v16.4s, v26.4s, v18.4s\n" + "prfm pldl1keep, [x22, #0x70]\n" + "prfm pldl1keep, [x21, #0x70]\n" + "sadalp v2.8h, v23.16b\n" + "sadalp v1.8h, v22.16b\n" + "str q25, [%x[out_ptr], #0x0]\n" + "add x20, x20, #0x1\n" + "zip1 v19.4s, v21.4s, v17.4s\n" + "zip1 v18.4s, v20.4s, v16.4s\n" + "str q24, [%x[out_ptr], #0x10]\n" + "sadalp v2.8h, v19.16b\n" + "sadalp v1.8h, v18.16b\n" + "str q23, [%x[out_ptr], #0x20]\n" + "zip2 v17.4s, v21.4s, v17.4s\n" + "zip2 v16.4s, v20.4s, v16.4s\n" + "str q22, [%x[out_ptr], #0x30]\n" + "str q19, [%x[out_ptr], #0x40]\n" + "sadalp v2.8h, v17.16b\n" + "sadalp v1.8h, v16.16b\n" + "str q18, [%x[out_ptr], #0x50]\n" + "str q17, [%x[out_ptr], #0x60]\n" + "str q16, [%x[out_ptr], #0x70]\n" + "add %x[out_ptr], %x[out_ptr], #0x80\n" + "bge 3b\n" + "5:" // Main loop skip + "cbz %x[width], 14f\n" + "tbz %x[width], #3, 9f\n" + "ldr d29, [x28], #0x8\n" + "ldr d28, [x27], #0x8\n" + "ldr d27, [x26], #0x8\n" + "ldr d26, [x25], #0x8\n" + "ldr d25, [x24], #0x8\n" + "ldr d24, [x23], #0x8\n" + "ldr d23, [x22], #0x8\n" + "ldr d22, [x21], #0x8\n" + "tbz %x[width], #2, 7f\n" + "ld1 { v29.s }[2], [x28], #0x4\n" + "ld1 { v28.s }[2], [x27], #0x4\n" + "ld1 { v27.s }[2], [x26], #0x4\n" + "ld1 { v26.s }[2], [x25], #0x4\n" + "ld1 { v25.s }[2], [x24], #0x4\n" + "ld1 { v24.s }[2], [x23], #0x4\n" + "ld1 { v23.s }[2], [x22], #0x4\n" + "ld1 { v22.s }[2], [x21], #0x4\n" + "tbz %x[width], #1, 6f\n" + "ld1 { v29.h }[6], [x28], #0x2\n" + "ld1 { v28.h }[6], [x27], #0x2\n" + "mov x20, #0x4\n" + "ld1 { v27.h }[6], [x26], #0x2\n" + "ld1 { v26.h }[6], [x25], #0x2\n" + "ld1 { v25.h }[6], [x24], #0x2\n" + "ld1 { v24.h }[6], [x23], #0x2\n" + "ld1 { v23.h }[6], [x22], #0x2\n" + "ld1 { v22.h }[6], [x21], #0x2\n" + "tbz %x[width], #0, 13f\n" + "ld1 { v29.b }[14], [x28]\n" + "ld1 { v28.b }[14], [x27]\n" + "ld1 { v27.b }[14], [x26]\n" + "ld1 { v26.b }[14], [x25]\n" + "ld1 { v25.b }[14], [x24]\n" + "ld1 { v24.b }[14], [x23]\n" + "ld1 { v23.b }[14], [x22]\n" + "ld1 { v22.b }[14], [x21]\n" + "b 13f\n" + "6:" // odd_loads_1_12 + "mov x20, #0x3\n" + "tbz %x[width], #0, 13f\n" + "ld1 { v29.b }[12], [x28]\n" + "ld1 { v28.b }[12], [x27]\n" + "mov x20, #0x4\n" + "ld1 { v27.b }[12], [x26]\n" + "ld1 { v26.b }[12], [x25]\n" + "ld1 { v25.b }[12], [x24]\n" + "ld1 { v24.b }[12], [x23]\n" + "ld1 { v23.b }[12], [x22]\n" + "ld1 { v22.b }[12], [x21]\n" + "b 13f\n" + "7:" // odd_loads_2_8 + "tbz %x[width], #1, 8f\n" + "ld1 { v29.h }[4], [x28], #0x2\n" + "ld1 { v28.h }[4], [x27], #0x2\n" + "mov x20, #0x3\n" + "ld1 { v27.h }[4], [x26], #0x2\n" + "ld1 { v26.h }[4], [x25], #0x2\n" + "ld1 { v25.h }[4], [x24], #0x2\n" + "ld1 { v24.h }[4], [x23], #0x2\n" + "ld1 { v23.h }[4], [x22], #0x2\n" + "ld1 { v22.h }[4], [x21], #0x2\n" + "tbz %x[width], #0, 13f\n" + "ld1 { v29.b }[10], [x28]\n" + "ld1 { v28.b }[10], [x27]\n" + "ld1 { v27.b }[10], [x26]\n" + "ld1 { v26.b }[10], [x25]\n" + "ld1 { v25.b }[10], [x24]\n" + "ld1 { v24.b }[10], [x23]\n" + "ld1 { v23.b }[10], [x22]\n" + "ld1 { v22.b }[10], [x21]\n" + "b 13f\n" + "8:" // odd_loads_1_8 + "mov x20, #0x2\n" + "tbz %x[width], #0, 13f\n" + "ld1 { v29.b }[8], [x28]\n" + "ld1 { v28.b }[8], [x27]\n" + "mov x20, #0x3\n" + "ld1 { v27.b }[8], [x26]\n" + "ld1 { v26.b }[8], [x25]\n" + "ld1 { v25.b }[8], [x24]\n" + "ld1 { v24.b }[8], [x23]\n" + "ld1 { v23.b }[8], [x22]\n" + "ld1 { v22.b }[8], [x21]\n" + "b 13f\n" + "9:" // odd_loads_4_0 + "tbz %x[width], #2, 11f\n" + "ldr s29, [x28], #0x4\n" + "ldr s28, [x27], #0x4\n" + "ldr s27, [x26], #0x4\n" + "ldr s26, [x25], #0x4\n" + "ldr s25, [x24], #0x4\n" + "ldr s24, [x23], #0x4\n" + "ldr s23, [x22], #0x4\n" + "ldr s22, [x21], #0x4\n" + "tbz %x[width], #1, 10f\n" + "ld1 { v29.h }[2], [x28], #0x2\n" + "ld1 { v28.h }[2], [x27], #0x2\n" + "mov x20, #0x2\n" + "ld1 { v27.h }[2], [x26], #0x2\n" + "ld1 { v26.h }[2], [x25], #0x2\n" + "ld1 { v25.h }[2], [x24], #0x2\n" + "ld1 { v24.h }[2], [x23], #0x2\n" + "ld1 { v23.h }[2], [x22], #0x2\n" + "ld1 { v22.h }[2], [x21], #0x2\n" + "tbz %x[width], #0, 13f\n" + "ld1 { v29.b }[6], [x28]\n" + "ld1 { v28.b }[6], [x27]\n" + "ld1 { v27.b }[6], [x26]\n" + "ld1 { v26.b }[6], [x25]\n" + "ld1 { v25.b }[6], [x24]\n" + "ld1 { v24.b }[6], [x23]\n" + "ld1 { v23.b }[6], [x22]\n" + "ld1 { v22.b }[6], [x21]\n" + "b 13f\n" + "10:" // odd_loads_1_4 + "mov x20, #0x1\n" + "tbz %x[width], #0, 13f\n" + "ld1 { v29.b }[4], [x28]\n" + "ld1 { v28.b }[4], [x27]\n" + "mov x20, #0x2\n" + "ld1 { v27.b }[4], [x26]\n" + "ld1 { v26.b }[4], [x25]\n" + "ld1 { v25.b }[4], [x24]\n" + "ld1 { v24.b }[4], [x23]\n" + "ld1 { v23.b }[4], [x22]\n" + "ld1 { v22.b }[4], [x21]\n" + "b 13f\n" + "11:" // odd_loads_2_0 + "tbz %x[width], #1, 12f\n" + "ldr h29, [x28], #0x2\n" + "ldr h28, [x27], #0x2\n" + "mov x20, #0x1\n" + "ldr h27, [x26], #0x2\n" + "ldr h26, [x25], #0x2\n" + "ldr h25, [x24], #0x2\n" + "ldr h24, [x23], #0x2\n" + "ldr h23, [x22], #0x2\n" + "ldr h22, [x21], #0x2\n" + "tbz %x[width], #0, 13f\n" + "ld1 { v29.b }[2], [x28]\n" + "ld1 { v28.b }[2], [x27]\n" + "ld1 { v27.b }[2], [x26]\n" + "ld1 { v26.b }[2], [x25]\n" + "ld1 { v25.b }[2], [x24]\n" + "ld1 { v24.b }[2], [x23]\n" + "ld1 { v23.b }[2], [x22]\n" + "ld1 { v22.b }[2], [x21]\n" + "b 13f\n" + "12:" // odd_loads_1_0 + "ldr b29, [x28, #0x0]\n" + "ldr b28, [x27, #0x0]\n" + "mov x20, #0x1\n" + "ldr b27, [x26, #0x0]\n" + "ldr b26, [x25, #0x0]\n" + "ldr b25, [x24, #0x0]\n" + "ldr b24, [x23, #0x0]\n" + "ldr b23, [x22, #0x0]\n" + "ldr b22, [x21, #0x0]\n" + "13:" // Odd load end + "zip1 v21.4s, v29.4s, v27.4s\n" + "zip1 v20.4s, v28.4s, v26.4s\n" + "subs x20, x20, #0x1\n" + "zip1 v19.4s, v25.4s, v23.4s\n" + "zip1 v18.4s, v24.4s, v22.4s\n" + "zip1 v17.4s, v21.4s, v20.4s\n" + "zip1 v16.4s, v19.4s, v18.4s\n" + "str q17, [%x[out_ptr], #0x0]\n" + "sadalp v2.8h, v17.16b\n" + "str q16, [%x[out_ptr], #0x10]\n" + "sadalp v1.8h, v16.16b\n" + "add %x[out_ptr], %x[out_ptr], #0x20\n" + "beq 14f\n" + "zip2 v17.4s, v21.4s, v20.4s\n" + "zip2 v16.4s, v19.4s, v18.4s\n" + "subs x20, x20, #0x1\n" + "str q17, [%x[out_ptr], #0x0]\n" + "sadalp v2.8h, v17.16b\n" + "str q16, [%x[out_ptr], #0x10]\n" + "sadalp v1.8h, v16.16b\n" + "add %x[out_ptr], %x[out_ptr], #0x20\n" + "beq 14f\n" + "zip2 v21.4s, v29.4s, v27.4s\n" + "zip2 v20.4s, v28.4s, v26.4s\n" + "subs x20, x20, #0x1\n" + "zip2 v19.4s, v25.4s, v23.4s\n" + "zip2 v18.4s, v24.4s, v22.4s\n" + "zip1 v17.4s, v21.4s, v20.4s\n" + "zip1 v16.4s, v19.4s, v18.4s\n" + "str q17, [%x[out_ptr], #0x0]\n" + "sadalp v2.8h, v17.16b\n" + "str q16, [%x[out_ptr], #0x10]\n" + "sadalp v1.8h, v16.16b\n" + "add %x[out_ptr], %x[out_ptr], #0x20\n" + "beq 14f\n" + "zip2 v17.4s, v21.4s, v20.4s\n" + "zip2 v16.4s, v19.4s, v18.4s\n" + "str q17, [%x[out_ptr], #0x0]\n" + "sadalp v2.8h, v17.16b\n" + "str q16, [%x[out_ptr], #0x10]\n" + "sadalp v1.8h, v16.16b\n" + "add %x[out_ptr], %x[out_ptr], #0x20\n" + "14:" // Odds skip + "sadalp v0.4s, v2.8h\n" + "sadalp v31.4s, v1.8h\n" + "str q0, [%x[out_ptr], #0x0]\n" + "str q31, [%x[out_ptr], #0x10]\n" + "add %x[out_ptr], %x[out_ptr], #0x20\n" + : [out_ptr] "+&r" (out_ptr), [width] "+&r" (width) + : [first] "r" (first), [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset) + : "cc", "memory", "v0", "v1", "v2", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28" + ); +} + + +#endif // __aarch64__ |