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 | 294 |
1 files changed, 147 insertions, 147 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 index 2db54126c0..6c41b5fdfb 100644 --- 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 @@ -153,202 +153,202 @@ void interleave_block<8, 4, VLType::None, true>( "5:" // Main loop skip "cbz %x[width], 14f\n" "tbz %x[width], #3, 9f\n" - "ldr d30, [x28], #0x8\n" - "ldr d29, [x27], #0x8\n" - "ldr d28, [x26], #0x8\n" - "ldr d27, [x25], #0x8\n" - "ldr d20, [x24], #0x8\n" - "ldr d26, [x23], #0x8\n" - "ldr d19, [x22], #0x8\n" - "ldr d18, [x21], #0x8\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 { v30.s }[2], [x28], #0x4\n" - "ld1 { v29.s }[2], [x27], #0x4\n" - "ld1 { v28.s }[2], [x26], #0x4\n" - "ld1 { v27.s }[2], [x25], #0x4\n" - "ld1 { v20.s }[2], [x24], #0x4\n" - "ld1 { v26.s }[2], [x23], #0x4\n" - "ld1 { v19.s }[2], [x22], #0x4\n" - "ld1 { v18.s }[2], [x21], #0x4\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 { v30.h }[6], [x28], #0x2\n" - "ld1 { v29.h }[6], [x27], #0x2\n" + "ld1 { v29.h }[6], [x28], #0x2\n" + "ld1 { v28.h }[6], [x27], #0x2\n" "mov x20, #0x4\n" - "ld1 { v28.h }[6], [x26], #0x2\n" - "ld1 { v27.h }[6], [x25], #0x2\n" - "ld1 { v20.h }[6], [x24], #0x2\n" - "ld1 { v26.h }[6], [x23], #0x2\n" - "ld1 { v19.h }[6], [x22], #0x2\n" - "ld1 { v18.h }[6], [x21], #0x2\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 { v30.b }[14], [x28]\n" - "ld1 { v29.b }[14], [x27]\n" - "ld1 { v28.b }[14], [x26]\n" - "ld1 { v27.b }[14], [x25]\n" - "ld1 { v20.b }[14], [x24]\n" - "ld1 { v26.b }[14], [x23]\n" - "ld1 { v19.b }[14], [x22]\n" - "ld1 { v18.b }[14], [x21]\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 { v30.b }[12], [x28]\n" - "ld1 { v29.b }[12], [x27]\n" + "ld1 { v29.b }[12], [x28]\n" + "ld1 { v28.b }[12], [x27]\n" "mov x20, #0x4\n" - "ld1 { v28.b }[12], [x26]\n" - "ld1 { v27.b }[12], [x25]\n" - "ld1 { v20.b }[12], [x24]\n" - "ld1 { v26.b }[12], [x23]\n" - "ld1 { v19.b }[12], [x22]\n" - "ld1 { v18.b }[12], [x21]\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 { v30.h }[4], [x28], #0x2\n" - "ld1 { v29.h }[4], [x27], #0x2\n" + "ld1 { v29.h }[4], [x28], #0x2\n" + "ld1 { v28.h }[4], [x27], #0x2\n" "mov x20, #0x3\n" - "ld1 { v28.h }[4], [x26], #0x2\n" - "ld1 { v27.h }[4], [x25], #0x2\n" - "ld1 { v20.h }[4], [x24], #0x2\n" - "ld1 { v26.h }[4], [x23], #0x2\n" - "ld1 { v19.h }[4], [x22], #0x2\n" - "ld1 { v18.h }[4], [x21], #0x2\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 { v30.b }[10], [x28]\n" - "ld1 { v29.b }[10], [x27]\n" - "ld1 { v28.b }[10], [x26]\n" - "ld1 { v27.b }[10], [x25]\n" - "ld1 { v20.b }[10], [x24]\n" - "ld1 { v26.b }[10], [x23]\n" - "ld1 { v19.b }[10], [x22]\n" - "ld1 { v18.b }[10], [x21]\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 { v30.b }[8], [x28]\n" - "ld1 { v29.b }[8], [x27]\n" + "ld1 { v29.b }[8], [x28]\n" + "ld1 { v28.b }[8], [x27]\n" "mov x20, #0x3\n" - "ld1 { v28.b }[8], [x26]\n" - "ld1 { v27.b }[8], [x25]\n" - "ld1 { v20.b }[8], [x24]\n" - "ld1 { v26.b }[8], [x23]\n" - "ld1 { v19.b }[8], [x22]\n" - "ld1 { v18.b }[8], [x21]\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 s30, [x28], #0x4\n" - "ldr s29, [x27], #0x4\n" - "ldr s28, [x26], #0x4\n" - "ldr s27, [x25], #0x4\n" - "ldr s20, [x24], #0x4\n" - "ldr s26, [x23], #0x4\n" - "ldr s19, [x22], #0x4\n" - "ldr s18, [x21], #0x4\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 { v30.h }[2], [x28], #0x2\n" - "ld1 { v29.h }[2], [x27], #0x2\n" + "ld1 { v29.h }[2], [x28], #0x2\n" + "ld1 { v28.h }[2], [x27], #0x2\n" "mov x20, #0x2\n" - "ld1 { v28.h }[2], [x26], #0x2\n" - "ld1 { v27.h }[2], [x25], #0x2\n" - "ld1 { v20.h }[2], [x24], #0x2\n" - "ld1 { v26.h }[2], [x23], #0x2\n" - "ld1 { v19.h }[2], [x22], #0x2\n" - "ld1 { v18.h }[2], [x21], #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 { v30.b }[6], [x28]\n" - "ld1 { v29.b }[6], [x27]\n" - "ld1 { v28.b }[6], [x26]\n" - "ld1 { v27.b }[6], [x25]\n" - "ld1 { v20.b }[6], [x24]\n" - "ld1 { v26.b }[6], [x23]\n" - "ld1 { v19.b }[6], [x22]\n" - "ld1 { v18.b }[6], [x21]\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 { v30.b }[4], [x28]\n" - "ld1 { v29.b }[4], [x27]\n" + "ld1 { v29.b }[4], [x28]\n" + "ld1 { v28.b }[4], [x27]\n" "mov x20, #0x2\n" - "ld1 { v28.b }[4], [x26]\n" - "ld1 { v27.b }[4], [x25]\n" - "ld1 { v20.b }[4], [x24]\n" - "ld1 { v26.b }[4], [x23]\n" - "ld1 { v19.b }[4], [x22]\n" - "ld1 { v18.b }[4], [x21]\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 h30, [x28], #0x2\n" - "ldr h29, [x27], #0x2\n" + "ldr h29, [x28], #0x2\n" + "ldr h28, [x27], #0x2\n" "mov x20, #0x1\n" - "ldr h28, [x26], #0x2\n" - "ldr h27, [x25], #0x2\n" - "ldr h20, [x24], #0x2\n" - "ldr h26, [x23], #0x2\n" - "ldr h19, [x22], #0x2\n" - "ldr h18, [x21], #0x2\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 { v30.b }[2], [x28]\n" - "ld1 { v29.b }[2], [x27]\n" - "ld1 { v28.b }[2], [x26]\n" - "ld1 { v27.b }[2], [x25]\n" - "ld1 { v20.b }[2], [x24]\n" - "ld1 { v26.b }[2], [x23]\n" - "ld1 { v19.b }[2], [x22]\n" - "ld1 { v18.b }[2], [x21]\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 b30, [x28, #0x0]\n" - "ldr b29, [x27, #0x0]\n" + "ldr b29, [x28, #0x0]\n" + "ldr b28, [x27, #0x0]\n" "mov x20, #0x1\n" - "ldr b28, [x26, #0x0]\n" - "ldr b27, [x25, #0x0]\n" - "ldr b20, [x24, #0x0]\n" - "ldr b26, [x23, #0x0]\n" - "ldr b19, [x22, #0x0]\n" - "ldr b18, [x21, #0x0]\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 v22.4s, v30.4s, v28.4s\n" "zip1 v21.4s, v29.4s, v27.4s\n" + "zip1 v20.4s, v28.4s, v26.4s\n" "subs x20, x20, #0x1\n" - "zip1 v17.4s, v20.4s, v19.4s\n" - "zip1 v16.4s, v26.4s, v18.4s\n" - "zip1 v25.4s, v22.4s, v21.4s\n" - "zip1 v24.4s, v17.4s, v16.4s\n" - "str q25, [%x[out_ptr], #0x0]\n" - "sadalp v2.8h, v25.16b\n" - "str q24, [%x[out_ptr], #0x10]\n" - "sadalp v1.8h, v24.16b\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 v23.4s, v22.4s, v21.4s\n" - "zip2 v22.4s, v17.4s, v16.4s\n" + "zip2 v17.4s, v21.4s, v20.4s\n" + "zip2 v16.4s, v19.4s, v18.4s\n" "subs x20, x20, #0x1\n" - "str q23, [%x[out_ptr], #0x0]\n" - "sadalp v2.8h, v23.16b\n" - "str q22, [%x[out_ptr], #0x10]\n" - "sadalp v1.8h, v22.16b\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, v30.4s, v28.4s\n" - "zip2 v17.4s, v29.4s, v27.4s\n" + "zip2 v21.4s, v29.4s, v27.4s\n" + "zip2 v20.4s, v28.4s, v26.4s\n" "subs x20, x20, #0x1\n" - "zip2 v20.4s, v20.4s, v19.4s\n" - "zip2 v16.4s, v26.4s, v18.4s\n" - "zip1 v19.4s, v21.4s, v17.4s\n" - "zip1 v18.4s, v20.4s, v16.4s\n" - "str q19, [%x[out_ptr], #0x0]\n" - "sadalp v2.8h, v19.16b\n" - "str q18, [%x[out_ptr], #0x10]\n" - "sadalp v1.8h, v18.16b\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, v17.4s\n" - "zip2 v16.4s, v20.4s, v16.4s\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" |