diff options
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_u8u32_dot_6x16/a55.cpp')
| -rw-r--r-- | src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_u8u32_dot_6x16/a55.cpp | 1629 |
1 files changed, 815 insertions, 814 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_u8u32_dot_6x16/a55.cpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_u8u32_dot_6x16/a55.cpp index 7f0fad7fa7..4ec75191b9 100644 --- a/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_u8u32_dot_6x16/a55.cpp +++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_u8u32_dot_6x16/a55.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2019-2021, 2023 Arm Limited. + * Copyright (c) 2019-2021, 2023-2024 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -44,18 +44,18 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( size_t output_offset = {}; size_t input_initial_col = {}; size_t input_offset = {}; + void *output_ptr = {}; } ka; unsigned long flags=0; - void *output_ptr; void *input_ptr; if (output_arg.is_indirect) { - output_ptr=(void *)(output_arg.indirect.ptr); + ka.output_ptr=(void *)(output_arg.indirect.ptr); ka.output_offset=output_arg.indirect.offset; flags |= 0x4; } else { - output_ptr=(void *)(output_arg.direct.base); + ka.output_ptr=(void *)(output_arg.direct.base); ka.output_offset=output_arg.direct.stride; } @@ -87,72 +87,72 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "bgt 69f\n" "beq 35f\n" "ldr x8, [%x[args_ptr], %[offsetof_N]]\n" - "mov x17, %x[output_ptr]\n" - "ldr x16, [%x[args_ptr], %[offsetof_B_ptr]]\n" + "ldr x17, [%x[args_ptr], %[offsetof_B_ptr]]\n" + "ldr x16, [%x[args_ptr], %[offsetof_output_ptr]]\n" "2:" // Height 1: Column loop "tbz %x[flags], #0, 12f\n" "cmp x8, #0x10\n" "bge 11f\n" "tbz x8, #3, 6f\n" - "ld1 { v8.4s }, [x17], #0x10\n" - "ld1 { v9.4s }, [x17], #0x10\n" + "ld1 { v8.4s }, [x16], #0x10\n" + "ld1 { v9.4s }, [x16], #0x10\n" "tbz x8, #2, 4f\n" - "ld1 { v10.4s }, [x17], #0x10\n" + "ld1 { v10.4s }, [x16], #0x10\n" "tbz x8, #1, 3f\n" - "ldr d11, [x17], #0x8\n" + "ldr d11, [x16], #0x8\n" "mov x25, #0x38\n" "tbz x8, #0, 10f\n" - "ld1 { v11.s }[2], [x17]\n" + "ld1 { v11.s }[2], [x16]\n" "b 10f\n" "3:" // Height 1: Partial accumulate: partial_1_12 "mov x25, #0x30\n" "tbz x8, #0, 10f\n" - "ldr s11, [x17, #0x0]\n" + "ldr s11, [x16, #0x0]\n" "b 10f\n" "4:" // Height 1: Partial accumulate: partial_2_8 "tbz x8, #1, 5f\n" - "ldr d10, [x17], #0x8\n" + "ldr d10, [x16], #0x8\n" "mov x25, #0x28\n" "tbz x8, #0, 10f\n" - "ld1 { v10.s }[2], [x17]\n" + "ld1 { v10.s }[2], [x16]\n" "b 10f\n" "5:" // Height 1: Partial accumulate: partial_1_8 "mov x25, #0x20\n" "tbz x8, #0, 10f\n" - "ldr s10, [x17, #0x0]\n" + "ldr s10, [x16, #0x0]\n" "b 10f\n" "6:" // Height 1: Partial accumulate: partial_4_0 "tbz x8, #2, 8f\n" - "ld1 { v8.4s }, [x17], #0x10\n" + "ld1 { v8.4s }, [x16], #0x10\n" "tbz x8, #1, 7f\n" - "ldr d9, [x17], #0x8\n" + "ldr d9, [x16], #0x8\n" "mov x25, #0x18\n" "tbz x8, #0, 10f\n" - "ld1 { v9.s }[2], [x17]\n" + "ld1 { v9.s }[2], [x16]\n" "b 10f\n" "7:" // Height 1: Partial accumulate: partial_1_4 "mov x25, #0x10\n" "tbz x8, #0, 10f\n" - "ldr s9, [x17, #0x0]\n" + "ldr s9, [x16, #0x0]\n" "b 10f\n" "8:" // Height 1: Partial accumulate: partial_2_0 "tbz x8, #1, 9f\n" - "ldr d8, [x17], #0x8\n" + "ldr d8, [x16], #0x8\n" "mov x25, #0x8\n" "tbz x8, #0, 10f\n" - "ld1 { v8.s }[2], [x17]\n" + "ld1 { v8.s }[2], [x16]\n" "b 10f\n" "9:" // Height 1: Partial accumulate: partial_1_0 - "ldr s8, [x17, #0x0]\n" + "ldr s8, [x16, #0x0]\n" "mov x25, #0x0\n" "10:" // Height 1: Partial accumulate: Done - "sub x17, x17, x25\n" + "sub x16, x16, x25\n" "b 13f\n" "11:" // Height 1: full accumulate - "ldr q8, [x17, #0x0]\n" - "ldr q9, [x17, #0x10]\n" - "ldr q10, [x17, #0x20]\n" - "ldr q11, [x17, #0x30]\n" + "ldr q8, [x16, #0x0]\n" + "ldr q9, [x16, #0x10]\n" + "ldr q10, [x16, #0x20]\n" + "ldr q11, [x16, #0x30]\n" "b 13f\n" "12:" // Height 1: no accumulate "movi v8.4s, #0x0\n" @@ -163,8 +163,8 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "mov x15, #0x0\n" "14:" // Height 1: String loop "ldr x20, [%x[args_ptr], %[offsetof_string_lengths]]\n" - "ldr w14, [x20, x15, LSL #0x2]\n" "ldr x21, [%x[args_ptr], %[offsetof_input_offset]]\n" + "ldr w14, [x20, x15, LSL #0x2]\n" "tbz %x[flags], #3, 15f\n" "ldr x20, [%x[input_ptr], x15, LSL #0x3]\n" "add x20, x20, x21, LSL #3\n" @@ -180,118 +180,118 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "blt 19f\n" "ldr q0, [x13, #0x0]\n" "cmp x14, #0x20\n" - "ldr q6, [x16, #0x0]\n" - "ldr q7, [x16, #0x10]\n" + "ldr q6, [x17, #0x0]\n" + "ldr q7, [x17, #0x10]\n" "blt 18f\n" "17:" // Height 1: Multiply loop: Main loop head ".inst 0x6f80e0c8 // udot v8.4s, v6.16b, v0.4b[0]\n" - "ldr d17, [x16, #0x20]\n" - "ldr x20, [x16, #0x28]\n" + "ldr d17, [x17, #0x20]\n" + "ldr x21, [x17, #0x28]\n" ".inst 0x6f80e0e9 // udot v9.4s, v7.16b, v0.4b[0]\n" - "ldr d16, [x16, #0x30]\n" - "mov v17.d[1], x20\n" - "ldr x20, [x16, #0x38]\n" + "ldr d16, [x17, #0x30]\n" + "add x13, x13, #0x10\n" + "ldr x20, [x17, #0x38]\n" + "sub x14, x14, #0x10\n" + "mov v17.d[1], x21\n" + "ldr x21, [x17, #0x48]\n" + "ldr x22, [x13, #0x8]\n" + "cmp x14, #0x20\n" "mov v16.d[1], x20\n" ".inst 0x6f80e22a // udot v10.4s, v17.16b, v0.4b[0]\n" - "ldr d17, [x16, #0x40]\n" - "ldr x20, [x16, #0x48]\n" + "ldr d17, [x17, #0x40]\n" ".inst 0x6f80e20b // udot v11.4s, v16.16b, v0.4b[0]\n" - "ldr d16, [x16, #0x50]\n" - "mov v17.d[1], x20\n" - "ldr x20, [x16, #0x58]\n" + "ldr d16, [x17, #0x50]\n" + "ldr x20, [x17, #0x58]\n" + "mov v17.d[1], x21\n" + "ldr x21, [x17, #0x68]\n" + "prfm pldl1keep, [x13, #0x80]\n" "mov v16.d[1], x20\n" ".inst 0x6fa0e228 // udot v8.4s, v17.16b, v0.4b[1]\n" - "ldr d17, [x16, #0x60]\n" - "ldr x20, [x16, #0x68]\n" + "ldr d17, [x17, #0x60]\n" ".inst 0x6fa0e209 // udot v9.4s, v16.16b, v0.4b[1]\n" - "ldr d16, [x16, #0x70]\n" - "mov v17.d[1], x20\n" - "ldr x20, [x16, #0x78]\n" + "ldr d16, [x17, #0x70]\n" + "ldr x20, [x17, #0x78]\n" + "mov v17.d[1], x21\n" + "ldr x21, [x17, #0x88]\n" "mov v16.d[1], x20\n" ".inst 0x6fa0e22a // udot v10.4s, v17.16b, v0.4b[1]\n" - "ldr d17, [x16, #0x80]\n" - "ldr x20, [x16, #0x88]\n" + "ldr d17, [x17, #0x80]\n" ".inst 0x6fa0e20b // udot v11.4s, v16.16b, v0.4b[1]\n" - "ldr d16, [x16, #0x90]\n" - "mov v17.d[1], x20\n" - "ldr x20, [x16, #0x98]\n" + "ldr d16, [x17, #0x90]\n" + "ldr x20, [x17, #0x98]\n" + "mov v17.d[1], x21\n" + "ldr x21, [x17, #0xa8]\n" "mov v16.d[1], x20\n" ".inst 0x6f80ea28 // udot v8.4s, v17.16b, v0.4b[2]\n" - "ldr d17, [x16, #0xa0]\n" - "ldr x20, [x16, #0xa8]\n" + "ldr d17, [x17, #0xa0]\n" ".inst 0x6f80ea09 // udot v9.4s, v16.16b, v0.4b[2]\n" - "ldr d16, [x16, #0xb0]\n" - "mov v17.d[1], x20\n" - "ldr x20, [x16, #0xb8]\n" + "ldr d16, [x17, #0xb0]\n" + "ldr x20, [x17, #0xb8]\n" + "mov v17.d[1], x21\n" + "ldr x21, [x17, #0xc8]\n" "mov v16.d[1], x20\n" ".inst 0x6f80ea2a // udot v10.4s, v17.16b, v0.4b[2]\n" - "ldr d17, [x16, #0xc0]\n" - "ldr x20, [x16, #0xc8]\n" + "ldr d17, [x17, #0xc0]\n" ".inst 0x6f80ea0b // udot v11.4s, v16.16b, v0.4b[2]\n" - "ldr d16, [x16, #0xd0]\n" - "mov v17.d[1], x20\n" - "ldr x20, [x16, #0xd8]\n" + "ldr d16, [x17, #0xd0]\n" + "ldr x20, [x17, #0xd8]\n" + "mov v17.d[1], x21\n" + "ldr x21, [x17, #0xe8]\n" "mov v16.d[1], x20\n" ".inst 0x6fa0ea28 // udot v8.4s, v17.16b, v0.4b[3]\n" - "ldr d17, [x16, #0xe0]\n" - "ldr x20, [x16, #0xe8]\n" + "ldr d17, [x17, #0xe0]\n" ".inst 0x6fa0ea09 // udot v9.4s, v16.16b, v0.4b[3]\n" - "ldr d16, [x16, #0xf0]\n" - "mov v17.d[1], x20\n" - "ldr x20, [x16, #0xf8]\n" + "ldr d16, [x17, #0xf0]\n" + "ldr x20, [x17, #0xf8]\n" + "add x17, x17, #0x100\n" + "mov v17.d[1], x21\n" + "ldr x21, [x17, #0x8]\n" "mov v16.d[1], x20\n" - "add x13, x13, #0x10\n" - "add x16, x16, #0x100\n" ".inst 0x6fa0ea2a // udot v10.4s, v17.16b, v0.4b[3]\n" - "ldr d6, [x16, #0x0]\n" - "ldr x20, [x16, #0x8]\n" + "ldr d6, [x17, #0x0]\n" ".inst 0x6fa0ea0b // udot v11.4s, v16.16b, v0.4b[3]\n" "ldr d0, [x13, #0x0]\n" - "sub x14, x14, #0x10\n" - "ldr d7, [x16, #0x10]\n" - "cmp x14, #0x20\n" - "ldr x21, [x13, #0x8]\n" - "mov v6.d[1], x20\n" - "ldr x20, [x16, #0x18]\n" - "mov v0.d[1], x21\n" + "ldr d7, [x17, #0x10]\n" + "ldr x20, [x17, #0x18]\n" + "mov v6.d[1], x21\n" + "mov v0.d[1], x22\n" "mov v7.d[1], x20\n" - "prfm pldl1keep, [x13, #0x80]\n" "bge 17b\n" "18:" // Height 1: Multiply loop: Single iteration only ".inst 0x6f80e0c8 // udot v8.4s, v6.16b, v0.4b[0]\n" - "ldr q17, [x16, #0x20]\n" + "ldr q17, [x17, #0x20]\n" ".inst 0x6f80e0e9 // udot v9.4s, v7.16b, v0.4b[0]\n" - "ldr q16, [x16, #0x30]\n" + "ldr q16, [x17, #0x30]\n" + "add x13, x13, #0x10\n" + "sub x14, x14, #0x10\n" ".inst 0x6f80e22a // udot v10.4s, v17.16b, v0.4b[0]\n" - "ldr q17, [x16, #0x40]\n" + "ldr q17, [x17, #0x40]\n" ".inst 0x6f80e20b // udot v11.4s, v16.16b, v0.4b[0]\n" - "ldr q16, [x16, #0x50]\n" + "ldr q16, [x17, #0x50]\n" + "prfm pldl1keep, [x13, #0x80]\n" ".inst 0x6fa0e228 // udot v8.4s, v17.16b, v0.4b[1]\n" - "ldr q17, [x16, #0x60]\n" + "ldr q17, [x17, #0x60]\n" ".inst 0x6fa0e209 // udot v9.4s, v16.16b, v0.4b[1]\n" - "ldr q16, [x16, #0x70]\n" + "ldr q16, [x17, #0x70]\n" ".inst 0x6fa0e22a // udot v10.4s, v17.16b, v0.4b[1]\n" - "ldr q17, [x16, #0x80]\n" + "ldr q17, [x17, #0x80]\n" ".inst 0x6fa0e20b // udot v11.4s, v16.16b, v0.4b[1]\n" - "ldr q16, [x16, #0x90]\n" + "ldr q16, [x17, #0x90]\n" ".inst 0x6f80ea28 // udot v8.4s, v17.16b, v0.4b[2]\n" - "ldr q17, [x16, #0xa0]\n" + "ldr q17, [x17, #0xa0]\n" ".inst 0x6f80ea09 // udot v9.4s, v16.16b, v0.4b[2]\n" - "ldr q16, [x16, #0xb0]\n" + "ldr q16, [x17, #0xb0]\n" ".inst 0x6f80ea2a // udot v10.4s, v17.16b, v0.4b[2]\n" - "ldr q17, [x16, #0xc0]\n" + "ldr q17, [x17, #0xc0]\n" ".inst 0x6f80ea0b // udot v11.4s, v16.16b, v0.4b[2]\n" - "ldr q16, [x16, #0xd0]\n" + "ldr q16, [x17, #0xd0]\n" ".inst 0x6fa0ea28 // udot v8.4s, v17.16b, v0.4b[3]\n" - "ldr q17, [x16, #0xe0]\n" + "ldr q17, [x17, #0xe0]\n" ".inst 0x6fa0ea09 // udot v9.4s, v16.16b, v0.4b[3]\n" - "ldr q16, [x16, #0xf0]\n" - "add x13, x13, #0x10\n" - "sub x14, x14, #0x10\n" + "ldr q16, [x17, #0xf0]\n" + "add x17, x17, #0x100\n" ".inst 0x6fa0ea2a // udot v10.4s, v17.16b, v0.4b[3]\n" - "prfm pldl1keep, [x13, #0x80]\n" ".inst 0x6fa0ea0b // udot v11.4s, v16.16b, v0.4b[3]\n" - "add x16, x16, #0x100\n" "19:" // Height 1: Multiply loop: Main loop skip "cbz x14, 24f\n" "cmp x14, #0x4\n" @@ -299,16 +299,16 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "20:" // Height 1: Multiply loop: Odd block loop "ldr s18, [x13], #0x4\n" "sub x14, x14, #0x4\n" - "ldr q16, [x16, #0x0]\n" - ".inst 0x6f92e208 // udot v8.4s, v16.16b, v18.4b[0]\n" - "ldr q16, [x16, #0x10]\n" - ".inst 0x6f92e209 // udot v9.4s, v16.16b, v18.4b[0]\n" - "ldr q17, [x16, #0x20]\n" + "ldr q17, [x17, #0x0]\n" "cmp x14, #0x4\n" - "ldr q16, [x16, #0x30]\n" + "ldr q16, [x17, #0x10]\n" + ".inst 0x6f92e228 // udot v8.4s, v17.16b, v18.4b[0]\n" + "ldr q17, [x17, #0x20]\n" + ".inst 0x6f92e209 // udot v9.4s, v16.16b, v18.4b[0]\n" + "ldr q16, [x17, #0x30]\n" + "add x17, x17, #0x40\n" ".inst 0x6f92e22a // udot v10.4s, v17.16b, v18.4b[0]\n" ".inst 0x6f92e20b // udot v11.4s, v16.16b, v18.4b[0]\n" - "add x16, x16, #0x40\n" "bge 20b\n" "21:" // Height 1: Multiply loop: Skip odd blocks "cbz x14, 24f\n" @@ -320,165 +320,165 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "22:" // Height 1: Multiply loop: Ragged operand read: partial_1_0 "ldr b0, [x13, #0x0]\n" "23:" // Height 1: Multiply loop: Ragged operand read: Done - "ldr q16, [x16, #0x0]\n" - ".inst 0x6f80e208 // udot v8.4s, v16.16b, v0.4b[0]\n" - "ldr q16, [x16, #0x10]\n" + "ldr q17, [x17, #0x0]\n" + "ldr q16, [x17, #0x10]\n" + ".inst 0x6f80e228 // udot v8.4s, v17.16b, v0.4b[0]\n" + "ldr q17, [x17, #0x20]\n" ".inst 0x6f80e209 // udot v9.4s, v16.16b, v0.4b[0]\n" - "ldr q16, [x16, #0x20]\n" - ".inst 0x6f80e20a // udot v10.4s, v16.16b, v0.4b[0]\n" - "ldr q16, [x16, #0x30]\n" + "ldr q16, [x17, #0x30]\n" + "add x17, x17, #0x40\n" + ".inst 0x6f80e22a // udot v10.4s, v17.16b, v0.4b[0]\n" ".inst 0x6f80e20b // udot v11.4s, v16.16b, v0.4b[0]\n" - "add x16, x16, #0x40\n" "24:" // Height 1: Multiply loop: No odd multiplies "ldr w20, [%x[args_ptr], %[offsetof_num_strings]]\n" "add x15, x15, #0x1\n" "cmp x15, x20\n" "bne 14b\n" "cmp x8, #0x10\n" - "prfm pstl1keep, [x17, #0x0]\n" + "prfm pstl1keep, [x16, #0x0]\n" "bge 33f\n" "tbz x8, #3, 28f\n" - "st1 { v8.4s }, [x17], #0x10\n" - "st1 { v9.4s }, [x17], #0x10\n" + "st1 { v8.4s }, [x16], #0x10\n" + "st1 { v9.4s }, [x16], #0x10\n" "tbz x8, #2, 26f\n" - "st1 { v10.4s }, [x17], #0x10\n" + "st1 { v10.4s }, [x16], #0x10\n" "tbz x8, #1, 25f\n" - "str d11, [x17], #0x8\n" + "str d11, [x16], #0x8\n" "tbz x8, #0, 32f\n" - "st1 { v11.s }[2], [x17]\n" + "st1 { v11.s }[2], [x16]\n" "b 32f\n" "25:" // Height 1: Partial direct writeback: partial_1_12 "tbz x8, #0, 32f\n" - "str s11, [x17, #0x0]\n" + "str s11, [x16, #0x0]\n" "b 32f\n" "26:" // Height 1: Partial direct writeback: partial_2_8 "tbz x8, #1, 27f\n" - "str d10, [x17], #0x8\n" + "str d10, [x16], #0x8\n" "tbz x8, #0, 32f\n" - "st1 { v10.s }[2], [x17]\n" + "st1 { v10.s }[2], [x16]\n" "b 32f\n" "27:" // Height 1: Partial direct writeback: partial_1_8 "tbz x8, #0, 32f\n" - "str s10, [x17, #0x0]\n" + "str s10, [x16, #0x0]\n" "b 32f\n" "28:" // Height 1: Partial direct writeback: partial_4_0 "tbz x8, #2, 30f\n" - "st1 { v8.4s }, [x17], #0x10\n" + "st1 { v8.4s }, [x16], #0x10\n" "tbz x8, #1, 29f\n" - "str d9, [x17], #0x8\n" + "str d9, [x16], #0x8\n" "tbz x8, #0, 32f\n" - "st1 { v9.s }[2], [x17]\n" + "st1 { v9.s }[2], [x16]\n" "b 32f\n" "29:" // Height 1: Partial direct writeback: partial_1_4 "tbz x8, #0, 32f\n" - "str s9, [x17, #0x0]\n" + "str s9, [x16, #0x0]\n" "b 32f\n" "30:" // Height 1: Partial direct writeback: partial_2_0 "tbz x8, #1, 31f\n" - "str d8, [x17], #0x8\n" + "str d8, [x16], #0x8\n" "tbz x8, #0, 32f\n" - "st1 { v8.s }[2], [x17]\n" + "st1 { v8.s }[2], [x16]\n" "b 32f\n" "31:" // Height 1: Partial direct writeback: partial_1_0 - "str s8, [x17, #0x0]\n" + "str s8, [x16, #0x0]\n" "32:" // Height 1: Partial direct writeback: Done "b 34f\n" "33:" // Height 1: Full writeback - "str q8, [x17, #0x0]\n" - "str q9, [x17, #0x10]\n" - "str q10, [x17, #0x20]\n" - "str q11, [x17, #0x30]\n" - "add x17, x17, #0x40\n" + "str q8, [x16, #0x0]\n" + "str q9, [x16, #0x10]\n" + "str q10, [x16, #0x20]\n" + "str q11, [x16, #0x30]\n" + "add x16, x16, #0x40\n" "34:" // Height 1: Writeback done "subs x8, x8, #0x10\n" "bgt 2b\n" "b 206f\n" "35:" // Height 2 "ldr x8, [%x[args_ptr], %[offsetof_N]]\n" - "mov x17, %x[output_ptr]\n" - "ldr x16, [%x[args_ptr], %[offsetof_B_ptr]]\n" + "ldr x17, [%x[args_ptr], %[offsetof_B_ptr]]\n" + "ldr x16, [%x[args_ptr], %[offsetof_output_ptr]]\n" "36:" // Height 2: Column loop "tbz %x[flags], #0, 46f\n" "ldr x20, [%x[args_ptr], %[offsetof_output_offset]]\n" "cmp x8, #0x10\n" - "add x24, x17, x20, LSL #2\n" + "add x24, x16, x20, LSL #2\n" "bge 45f\n" "tbz x8, #3, 40f\n" - "ld1 { v8.4s }, [x17], #0x10\n" + "ld1 { v8.4s }, [x16], #0x10\n" "ld1 { v12.4s }, [x24], #0x10\n" - "ld1 { v9.4s }, [x17], #0x10\n" + "ld1 { v9.4s }, [x16], #0x10\n" "ld1 { v13.4s }, [x24], #0x10\n" "tbz x8, #2, 38f\n" - "ld1 { v10.4s }, [x17], #0x10\n" + "ld1 { v10.4s }, [x16], #0x10\n" "ld1 { v14.4s }, [x24], #0x10\n" "tbz x8, #1, 37f\n" - "ldr d11, [x17], #0x8\n" + "ldr d11, [x16], #0x8\n" "mov x25, #0x38\n" "ldr d15, [x24], #0x8\n" "tbz x8, #0, 44f\n" - "ld1 { v11.s }[2], [x17]\n" + "ld1 { v11.s }[2], [x16]\n" "ld1 { v15.s }[2], [x24]\n" "b 44f\n" "37:" // Height 2: Partial accumulate: partial_1_12 "mov x25, #0x30\n" "tbz x8, #0, 44f\n" - "ldr s11, [x17, #0x0]\n" + "ldr s11, [x16, #0x0]\n" "ldr s15, [x24, #0x0]\n" "b 44f\n" "38:" // Height 2: Partial accumulate: partial_2_8 "tbz x8, #1, 39f\n" - "ldr d10, [x17], #0x8\n" + "ldr d10, [x16], #0x8\n" "mov x25, #0x28\n" "ldr d14, [x24], #0x8\n" "tbz x8, #0, 44f\n" - "ld1 { v10.s }[2], [x17]\n" + "ld1 { v10.s }[2], [x16]\n" "ld1 { v14.s }[2], [x24]\n" "b 44f\n" "39:" // Height 2: Partial accumulate: partial_1_8 "mov x25, #0x20\n" "tbz x8, #0, 44f\n" - "ldr s10, [x17, #0x0]\n" + "ldr s10, [x16, #0x0]\n" "ldr s14, [x24, #0x0]\n" "b 44f\n" "40:" // Height 2: Partial accumulate: partial_4_0 "tbz x8, #2, 42f\n" - "ld1 { v8.4s }, [x17], #0x10\n" + "ld1 { v8.4s }, [x16], #0x10\n" "ld1 { v12.4s }, [x24], #0x10\n" "tbz x8, #1, 41f\n" - "ldr d9, [x17], #0x8\n" + "ldr d9, [x16], #0x8\n" "mov x25, #0x18\n" "ldr d13, [x24], #0x8\n" "tbz x8, #0, 44f\n" - "ld1 { v9.s }[2], [x17]\n" + "ld1 { v9.s }[2], [x16]\n" "ld1 { v13.s }[2], [x24]\n" "b 44f\n" "41:" // Height 2: Partial accumulate: partial_1_4 "mov x25, #0x10\n" "tbz x8, #0, 44f\n" - "ldr s9, [x17, #0x0]\n" + "ldr s9, [x16, #0x0]\n" "ldr s13, [x24, #0x0]\n" "b 44f\n" "42:" // Height 2: Partial accumulate: partial_2_0 "tbz x8, #1, 43f\n" - "ldr d8, [x17], #0x8\n" + "ldr d8, [x16], #0x8\n" "mov x25, #0x8\n" "ldr d12, [x24], #0x8\n" "tbz x8, #0, 44f\n" - "ld1 { v8.s }[2], [x17]\n" + "ld1 { v8.s }[2], [x16]\n" "ld1 { v12.s }[2], [x24]\n" "b 44f\n" "43:" // Height 2: Partial accumulate: partial_1_0 - "ldr s8, [x17, #0x0]\n" + "ldr s8, [x16, #0x0]\n" "mov x25, #0x0\n" "ldr s12, [x24, #0x0]\n" "44:" // Height 2: Partial accumulate: Done - "sub x17, x17, x25\n" + "sub x16, x16, x25\n" "b 47f\n" "45:" // Height 2: full accumulate - "ldr q8, [x17, #0x0]\n" - "ldr q9, [x17, #0x10]\n" - "ldr q10, [x17, #0x20]\n" - "ldr q11, [x17, #0x30]\n" + "ldr q8, [x16, #0x0]\n" + "ldr q9, [x16, #0x10]\n" + "ldr q10, [x16, #0x20]\n" + "ldr q11, [x16, #0x30]\n" "ldr q12, [x24, #0x0]\n" "ldr q13, [x24, #0x10]\n" "ldr q14, [x24, #0x20]\n" @@ -497,8 +497,8 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "mov x15, #0x0\n" "48:" // Height 2: String loop "ldr x20, [%x[args_ptr], %[offsetof_string_lengths]]\n" - "ldr w14, [x20, x15, LSL #0x2]\n" "ldr x21, [%x[args_ptr], %[offsetof_input_offset]]\n" + "ldr w14, [x20, x15, LSL #0x2]\n" "tbz %x[flags], #3, 49f\n" "ldr x20, [%x[input_ptr], x15, LSL #0x3]\n" "add x20, x20, x21, LSL #3\n" @@ -518,154 +518,154 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "ldr q0, [x13, #0x0]\n" "cmp x14, #0x20\n" "ldr q1, [x12, #0x0]\n" - "ldr q6, [x16, #0x0]\n" - "ldr q7, [x16, #0x10]\n" + "ldr q6, [x17, #0x0]\n" + "ldr q7, [x17, #0x10]\n" "blt 52f\n" "51:" // Height 2: Multiply loop: Main loop head ".inst 0x6f80e0c8 // udot v8.4s, v6.16b, v0.4b[0]\n" - "ldr x21, [x16, #0x28]\n" + "ldr x20, [x17, #0x28]\n" ".inst 0x6f81e0cc // udot v12.4s, v6.16b, v1.4b[0]\n" - "ldr d17, [x16, #0x20]\n" + "ldr d17, [x17, #0x20]\n" ".inst 0x6f80e0e9 // udot v9.4s, v7.16b, v0.4b[0]\n" - "ldr x20, [x16, #0x38]\n" + "ldr x21, [x17, #0x38]\n" ".inst 0x6f81e0ed // udot v13.4s, v7.16b, v1.4b[0]\n" - "ldr d16, [x16, #0x30]\n" - "mov v17.d[1], x21\n" + "ldr d16, [x17, #0x30]\n" + "mov v17.d[1], x20\n" + "ldr x20, [x17, #0x48]\n" + "add x13, x13, #0x10\n" + "add x12, x12, #0x10\n" + "mov v16.d[1], x21\n" ".inst 0x6f80e22a // udot v10.4s, v17.16b, v0.4b[0]\n" - "mov v16.d[1], x20\n" ".inst 0x6f81e22e // udot v14.4s, v17.16b, v1.4b[0]\n" - "ldr d17, [x16, #0x40]\n" + "ldr d17, [x17, #0x40]\n" ".inst 0x6f80e20b // udot v11.4s, v16.16b, v0.4b[0]\n" - "ldr x20, [x16, #0x48]\n" + "ldr x21, [x17, #0x58]\n" ".inst 0x6f81e20f // udot v15.4s, v16.16b, v1.4b[0]\n" - "ldr d16, [x16, #0x50]\n" + "ldr d16, [x17, #0x50]\n" "mov v17.d[1], x20\n" - "ldr x20, [x16, #0x58]\n" - "mov v16.d[1], x20\n" + "ldr x20, [x17, #0x68]\n" + "ldr x23, [x13, #0x8]\n" + "sub x14, x14, #0x10\n" + "mov v16.d[1], x21\n" ".inst 0x6fa0e228 // udot v8.4s, v17.16b, v0.4b[1]\n" - "ldr x21, [x16, #0x68]\n" ".inst 0x6fa1e22c // udot v12.4s, v17.16b, v1.4b[1]\n" - "ldr d17, [x16, #0x60]\n" + "ldr d17, [x17, #0x60]\n" ".inst 0x6fa0e209 // udot v9.4s, v16.16b, v0.4b[1]\n" - "ldr x20, [x16, #0x78]\n" + "ldr x21, [x17, #0x78]\n" ".inst 0x6fa1e20d // udot v13.4s, v16.16b, v1.4b[1]\n" - "ldr d16, [x16, #0x70]\n" - "mov v17.d[1], x21\n" + "ldr d16, [x17, #0x70]\n" + "mov v17.d[1], x20\n" + "ldr x20, [x17, #0x88]\n" + "ldr x22, [x12, #0x8]\n" + "cmp x14, #0x20\n" + "mov v16.d[1], x21\n" ".inst 0x6fa0e22a // udot v10.4s, v17.16b, v0.4b[1]\n" - "mov v16.d[1], x20\n" ".inst 0x6fa1e22e // udot v14.4s, v17.16b, v1.4b[1]\n" - "ldr d17, [x16, #0x80]\n" + "ldr d17, [x17, #0x80]\n" ".inst 0x6fa0e20b // udot v11.4s, v16.16b, v0.4b[1]\n" - "ldr x20, [x16, #0x88]\n" + "ldr x21, [x17, #0x98]\n" ".inst 0x6fa1e20f // udot v15.4s, v16.16b, v1.4b[1]\n" - "ldr d16, [x16, #0x90]\n" + "ldr d16, [x17, #0x90]\n" "mov v17.d[1], x20\n" - "ldr x20, [x16, #0x98]\n" - "mov v16.d[1], x20\n" + "ldr x20, [x17, #0xa8]\n" + "prfm pldl1keep, [x13, #0x80]\n" + "mov v16.d[1], x21\n" ".inst 0x6f80ea28 // udot v8.4s, v17.16b, v0.4b[2]\n" - "ldr x21, [x16, #0xa8]\n" ".inst 0x6f81ea2c // udot v12.4s, v17.16b, v1.4b[2]\n" - "ldr d17, [x16, #0xa0]\n" + "ldr d17, [x17, #0xa0]\n" ".inst 0x6f80ea09 // udot v9.4s, v16.16b, v0.4b[2]\n" - "ldr x20, [x16, #0xb8]\n" + "ldr x21, [x17, #0xb8]\n" ".inst 0x6f81ea0d // udot v13.4s, v16.16b, v1.4b[2]\n" - "ldr d16, [x16, #0xb0]\n" - "mov v17.d[1], x21\n" + "ldr d16, [x17, #0xb0]\n" + "mov v17.d[1], x20\n" + "ldr x20, [x17, #0xc8]\n" + "prfm pldl1keep, [x12, #0x80]\n" + "mov v16.d[1], x21\n" ".inst 0x6f80ea2a // udot v10.4s, v17.16b, v0.4b[2]\n" - "mov v16.d[1], x20\n" ".inst 0x6f81ea2e // udot v14.4s, v17.16b, v1.4b[2]\n" - "ldr d17, [x16, #0xc0]\n" + "ldr d17, [x17, #0xc0]\n" ".inst 0x6f80ea0b // udot v11.4s, v16.16b, v0.4b[2]\n" - "ldr x20, [x16, #0xc8]\n" + "ldr x21, [x17, #0xd8]\n" ".inst 0x6f81ea0f // udot v15.4s, v16.16b, v1.4b[2]\n" - "ldr d16, [x16, #0xd0]\n" + "ldr d16, [x17, #0xd0]\n" "mov v17.d[1], x20\n" - "ldr x20, [x16, #0xd8]\n" - "mov v16.d[1], x20\n" + "ldr x20, [x17, #0xe8]\n" + "mov v16.d[1], x21\n" ".inst 0x6fa0ea28 // udot v8.4s, v17.16b, v0.4b[3]\n" - "ldr x21, [x16, #0xe8]\n" ".inst 0x6fa1ea2c // udot v12.4s, v17.16b, v1.4b[3]\n" - "ldr d17, [x16, #0xe0]\n" + "ldr d17, [x17, #0xe0]\n" ".inst 0x6fa0ea09 // udot v9.4s, v16.16b, v0.4b[3]\n" - "ldr x20, [x16, #0xf8]\n" + "ldr x21, [x17, #0xf8]\n" ".inst 0x6fa1ea0d // udot v13.4s, v16.16b, v1.4b[3]\n" - "ldr d16, [x16, #0xf0]\n" - "mov v17.d[1], x21\n" - "add x13, x13, #0x10\n" - "mov v16.d[1], x20\n" - "add x12, x12, #0x10\n" - "add x16, x16, #0x100\n" + "ldr d16, [x17, #0xf0]\n" + "mov v17.d[1], x20\n" + "add x17, x17, #0x100\n" + "ldr x20, [x17, #0x8]\n" + "mov v16.d[1], x21\n" ".inst 0x6fa0ea2a // udot v10.4s, v17.16b, v0.4b[3]\n" ".inst 0x6fa1ea2e // udot v14.4s, v17.16b, v1.4b[3]\n" - "ldr d6, [x16, #0x0]\n" - "ldr x21, [x16, #0x8]\n" + "ldr d6, [x17, #0x0]\n" ".inst 0x6fa0ea0b // udot v11.4s, v16.16b, v0.4b[3]\n" "ldr d0, [x13, #0x0]\n" ".inst 0x6fa1ea0f // udot v15.4s, v16.16b, v1.4b[3]\n" "ldr d1, [x12, #0x0]\n" - "sub x14, x14, #0x10\n" - "ldr d7, [x16, #0x10]\n" - "cmp x14, #0x20\n" - "ldr x20, [x13, #0x8]\n" - "mov v6.d[1], x21\n" - "ldr x21, [x12, #0x8]\n" - "mov v0.d[1], x20\n" - "ldr x20, [x16, #0x18]\n" - "mov v1.d[1], x21\n" - "prfm pldl1keep, [x13, #0x80]\n" + "ldr d7, [x17, #0x10]\n" + "mov v6.d[1], x20\n" + "ldr x20, [x17, #0x18]\n" + "mov v0.d[1], x23\n" + "mov v1.d[1], x22\n" "mov v7.d[1], x20\n" - "prfm pldl1keep, [x12, #0x80]\n" "bge 51b\n" "52:" // Height 2: Multiply loop: Single iteration only ".inst 0x6f80e0c8 // udot v8.4s, v6.16b, v0.4b[0]\n" "add x13, x13, #0x10\n" ".inst 0x6f81e0cc // udot v12.4s, v6.16b, v1.4b[0]\n" - "ldr q17, [x16, #0x20]\n" + "ldr q17, [x17, #0x20]\n" ".inst 0x6f80e0e9 // udot v9.4s, v7.16b, v0.4b[0]\n" "add x12, x12, #0x10\n" ".inst 0x6f81e0ed // udot v13.4s, v7.16b, v1.4b[0]\n" - "ldr q16, [x16, #0x30]\n" + "ldr q16, [x17, #0x30]\n" ".inst 0x6f80e22a // udot v10.4s, v17.16b, v0.4b[0]\n" "sub x14, x14, #0x10\n" ".inst 0x6f81e22e // udot v14.4s, v17.16b, v1.4b[0]\n" - "ldr q17, [x16, #0x40]\n" + "ldr q17, [x17, #0x40]\n" ".inst 0x6f80e20b // udot v11.4s, v16.16b, v0.4b[0]\n" "prfm pldl1keep, [x13, #0x80]\n" ".inst 0x6f81e20f // udot v15.4s, v16.16b, v1.4b[0]\n" - "ldr q16, [x16, #0x50]\n" + "ldr q16, [x17, #0x50]\n" ".inst 0x6fa0e228 // udot v8.4s, v17.16b, v0.4b[1]\n" "prfm pldl1keep, [x12, #0x80]\n" ".inst 0x6fa1e22c // udot v12.4s, v17.16b, v1.4b[1]\n" - "ldr q17, [x16, #0x60]\n" + "ldr q17, [x17, #0x60]\n" ".inst 0x6fa0e209 // udot v9.4s, v16.16b, v0.4b[1]\n" ".inst 0x6fa1e20d // udot v13.4s, v16.16b, v1.4b[1]\n" - "ldr q16, [x16, #0x70]\n" + "ldr q16, [x17, #0x70]\n" ".inst 0x6fa0e22a // udot v10.4s, v17.16b, v0.4b[1]\n" ".inst 0x6fa1e22e // udot v14.4s, v17.16b, v1.4b[1]\n" - "ldr q17, [x16, #0x80]\n" + "ldr q17, [x17, #0x80]\n" ".inst 0x6fa0e20b // udot v11.4s, v16.16b, v0.4b[1]\n" ".inst 0x6fa1e20f // udot v15.4s, v16.16b, v1.4b[1]\n" - "ldr q16, [x16, #0x90]\n" + "ldr q16, [x17, #0x90]\n" ".inst 0x6f80ea28 // udot v8.4s, v17.16b, v0.4b[2]\n" ".inst 0x6f81ea2c // udot v12.4s, v17.16b, v1.4b[2]\n" - "ldr q17, [x16, #0xa0]\n" + "ldr q17, [x17, #0xa0]\n" ".inst 0x6f80ea09 // udot v9.4s, v16.16b, v0.4b[2]\n" ".inst 0x6f81ea0d // udot v13.4s, v16.16b, v1.4b[2]\n" - "ldr q16, [x16, #0xb0]\n" + "ldr q16, [x17, #0xb0]\n" ".inst 0x6f80ea2a // udot v10.4s, v17.16b, v0.4b[2]\n" ".inst 0x6f81ea2e // udot v14.4s, v17.16b, v1.4b[2]\n" - "ldr q17, [x16, #0xc0]\n" + "ldr q17, [x17, #0xc0]\n" ".inst 0x6f80ea0b // udot v11.4s, v16.16b, v0.4b[2]\n" ".inst 0x6f81ea0f // udot v15.4s, v16.16b, v1.4b[2]\n" - "ldr q16, [x16, #0xd0]\n" + "ldr q16, [x17, #0xd0]\n" ".inst 0x6fa0ea28 // udot v8.4s, v17.16b, v0.4b[3]\n" ".inst 0x6fa1ea2c // udot v12.4s, v17.16b, v1.4b[3]\n" - "ldr q17, [x16, #0xe0]\n" + "ldr q17, [x17, #0xe0]\n" ".inst 0x6fa0ea09 // udot v9.4s, v16.16b, v0.4b[3]\n" ".inst 0x6fa1ea0d // udot v13.4s, v16.16b, v1.4b[3]\n" - "ldr q16, [x16, #0xf0]\n" + "ldr q16, [x17, #0xf0]\n" ".inst 0x6fa0ea2a // udot v10.4s, v17.16b, v0.4b[3]\n" - "add x16, x16, #0x100\n" + "add x17, x17, #0x100\n" ".inst 0x6fa1ea2e // udot v14.4s, v17.16b, v1.4b[3]\n" ".inst 0x6fa0ea0b // udot v11.4s, v16.16b, v0.4b[3]\n" ".inst 0x6fa1ea0f // udot v15.4s, v16.16b, v1.4b[3]\n" @@ -678,16 +678,16 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "sub x14, x14, #0x4\n" "ldr s18, [x12], #0x4\n" "cmp x14, #0x4\n" - "ldr q17, [x16, #0x0]\n" + "ldr q17, [x17, #0x0]\n" + "ldr q16, [x17, #0x10]\n" ".inst 0x6f93e228 // udot v8.4s, v17.16b, v19.4b[0]\n" - "ldr q16, [x16, #0x10]\n" ".inst 0x6f92e22c // udot v12.4s, v17.16b, v18.4b[0]\n" - "ldr q17, [x16, #0x20]\n" + "ldr q17, [x17, #0x20]\n" ".inst 0x6f93e209 // udot v9.4s, v16.16b, v19.4b[0]\n" ".inst 0x6f92e20d // udot v13.4s, v16.16b, v18.4b[0]\n" - "ldr q16, [x16, #0x30]\n" + "ldr q16, [x17, #0x30]\n" ".inst 0x6f93e22a // udot v10.4s, v17.16b, v19.4b[0]\n" - "add x16, x16, #0x40\n" + "add x17, x17, #0x40\n" ".inst 0x6f92e22e // udot v14.4s, v17.16b, v18.4b[0]\n" ".inst 0x6f93e20b // udot v11.4s, v16.16b, v19.4b[0]\n" ".inst 0x6f92e20f // udot v15.4s, v16.16b, v18.4b[0]\n" @@ -705,16 +705,16 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "ldr b0, [x13, #0x0]\n" "ldr b1, [x12, #0x0]\n" "57:" // Height 2: Multiply loop: Ragged operand read: Done - "ldr q17, [x16, #0x0]\n" + "ldr q17, [x17, #0x0]\n" + "ldr q16, [x17, #0x10]\n" ".inst 0x6f80e228 // udot v8.4s, v17.16b, v0.4b[0]\n" - "ldr q16, [x16, #0x10]\n" ".inst 0x6f81e22c // udot v12.4s, v17.16b, v1.4b[0]\n" - "ldr q17, [x16, #0x20]\n" + "ldr q17, [x17, #0x20]\n" ".inst 0x6f80e209 // udot v9.4s, v16.16b, v0.4b[0]\n" ".inst 0x6f81e20d // udot v13.4s, v16.16b, v1.4b[0]\n" - "ldr q16, [x16, #0x30]\n" + "ldr q16, [x17, #0x30]\n" ".inst 0x6f80e22a // udot v10.4s, v17.16b, v0.4b[0]\n" - "add x16, x16, #0x40\n" + "add x17, x17, #0x40\n" ".inst 0x6f81e22e // udot v14.4s, v17.16b, v1.4b[0]\n" ".inst 0x6f80e20b // udot v11.4s, v16.16b, v0.4b[0]\n" ".inst 0x6f81e20f // udot v15.4s, v16.16b, v1.4b[0]\n" @@ -724,79 +724,79 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "cmp x15, x20\n" "bne 48b\n" "ldr x20, [%x[args_ptr], %[offsetof_output_offset]]\n" - "add x24, x17, x20, LSL #2\n" "cmp x8, #0x10\n" - "prfm pstl1keep, [x17, #0x0]\n" + "prfm pstl1keep, [x16, #0x0]\n" + "add x24, x16, x20, LSL #2\n" "prfm pstl1keep, [x24, #0x0]\n" "bge 67f\n" "tbz x8, #3, 62f\n" - "st1 { v8.4s }, [x17], #0x10\n" - "st1 { v9.4s }, [x17], #0x10\n" + "st1 { v8.4s }, [x16], #0x10\n" + "st1 { v9.4s }, [x16], #0x10\n" "st1 { v12.4s }, [x24], #0x10\n" "st1 { v13.4s }, [x24], #0x10\n" "tbz x8, #2, 60f\n" - "st1 { v10.4s }, [x17], #0x10\n" + "st1 { v10.4s }, [x16], #0x10\n" "st1 { v14.4s }, [x24], #0x10\n" "tbz x8, #1, 59f\n" - "str d11, [x17], #0x8\n" + "str d11, [x16], #0x8\n" "str d15, [x24], #0x8\n" "tbz x8, #0, 66f\n" - "st1 { v11.s }[2], [x17]\n" + "st1 { v11.s }[2], [x16]\n" "st1 { v15.s }[2], [x24]\n" "b 66f\n" "59:" // Height 2: Partial direct writeback: partial_1_12 "tbz x8, #0, 66f\n" - "str s11, [x17, #0x0]\n" + "str s11, [x16, #0x0]\n" "str s15, [x24, #0x0]\n" "b 66f\n" "60:" // Height 2: Partial direct writeback: partial_2_8 "tbz x8, #1, 61f\n" - "str d10, [x17], #0x8\n" + "str d10, [x16], #0x8\n" "str d14, [x24], #0x8\n" "tbz x8, #0, 66f\n" - "st1 { v10.s }[2], [x17]\n" + "st1 { v10.s }[2], [x16]\n" "st1 { v14.s }[2], [x24]\n" "b 66f\n" "61:" // Height 2: Partial direct writeback: partial_1_8 "tbz x8, #0, 66f\n" - "str s10, [x17, #0x0]\n" + "str s10, [x16, #0x0]\n" "str s14, [x24, #0x0]\n" "b 66f\n" "62:" // Height 2: Partial direct writeback: partial_4_0 "tbz x8, #2, 64f\n" - "st1 { v8.4s }, [x17], #0x10\n" + "st1 { v8.4s }, [x16], #0x10\n" "st1 { v12.4s }, [x24], #0x10\n" "tbz x8, #1, 63f\n" - "str d9, [x17], #0x8\n" + "str d9, [x16], #0x8\n" "str d13, [x24], #0x8\n" "tbz x8, #0, 66f\n" - "st1 { v9.s }[2], [x17]\n" + "st1 { v9.s }[2], [x16]\n" "st1 { v13.s }[2], [x24]\n" "b 66f\n" "63:" // Height 2: Partial direct writeback: partial_1_4 "tbz x8, #0, 66f\n" - "str s9, [x17, #0x0]\n" + "str s9, [x16, #0x0]\n" "str s13, [x24, #0x0]\n" "b 66f\n" "64:" // Height 2: Partial direct writeback: partial_2_0 "tbz x8, #1, 65f\n" - "str d8, [x17], #0x8\n" + "str d8, [x16], #0x8\n" "str d12, [x24], #0x8\n" "tbz x8, #0, 66f\n" - "st1 { v8.s }[2], [x17]\n" + "st1 { v8.s }[2], [x16]\n" "st1 { v12.s }[2], [x24]\n" "b 66f\n" "65:" // Height 2: Partial direct writeback: partial_1_0 - "str s8, [x17, #0x0]\n" + "str s8, [x16, #0x0]\n" "str s12, [x24, #0x0]\n" "66:" // Height 2: Partial direct writeback: Done "b 68f\n" "67:" // Height 2: Full writeback - "str q8, [x17, #0x0]\n" - "str q9, [x17, #0x10]\n" - "str q10, [x17, #0x20]\n" - "str q11, [x17, #0x30]\n" - "add x17, x17, #0x40\n" + "str q8, [x16, #0x0]\n" + "str q9, [x16, #0x10]\n" + "str q10, [x16, #0x20]\n" + "str q11, [x16, #0x30]\n" + "add x16, x16, #0x40\n" "str q12, [x24, #0x0]\n" "str q13, [x24, #0x10]\n" "str q14, [x24, #0x20]\n" @@ -807,107 +807,107 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "b 206f\n" "69:" // Height 3 "ldr x8, [%x[args_ptr], %[offsetof_N]]\n" - "mov x17, %x[output_ptr]\n" - "ldr x16, [%x[args_ptr], %[offsetof_B_ptr]]\n" + "ldr x17, [%x[args_ptr], %[offsetof_B_ptr]]\n" + "ldr x16, [%x[args_ptr], %[offsetof_output_ptr]]\n" "70:" // Height 3: Column loop "tbz %x[flags], #0, 80f\n" "ldr x20, [%x[args_ptr], %[offsetof_output_offset]]\n" - "add x24, x17, x20, LSL #2\n" "cmp x8, #0x10\n" + "add x24, x16, x20, LSL #2\n" "add x23, x24, x20, LSL #2\n" "bge 79f\n" "tbz x8, #3, 74f\n" - "ld1 { v8.4s }, [x17], #0x10\n" + "ld1 { v8.4s }, [x16], #0x10\n" "ld1 { v12.4s }, [x24], #0x10\n" "ld1 { v16.4s }, [x23], #0x10\n" - "ld1 { v9.4s }, [x17], #0x10\n" + "ld1 { v9.4s }, [x16], #0x10\n" "ld1 { v13.4s }, [x24], #0x10\n" "ld1 { v17.4s }, [x23], #0x10\n" "tbz x8, #2, 72f\n" - "ld1 { v10.4s }, [x17], #0x10\n" + "ld1 { v10.4s }, [x16], #0x10\n" "ld1 { v14.4s }, [x24], #0x10\n" "ld1 { v18.4s }, [x23], #0x10\n" "tbz x8, #1, 71f\n" - "ldr d11, [x17], #0x8\n" + "ldr d11, [x16], #0x8\n" "mov x25, #0x38\n" "ldr d15, [x24], #0x8\n" "ldr d19, [x23], #0x8\n" "tbz x8, #0, 78f\n" - "ld1 { v11.s }[2], [x17]\n" + "ld1 { v11.s }[2], [x16]\n" "ld1 { v15.s }[2], [x24]\n" "ld1 { v19.s }[2], [x23]\n" "b 78f\n" "71:" // Height 3: Partial accumulate: partial_1_12 "mov x25, #0x30\n" "tbz x8, #0, 78f\n" - "ldr s11, [x17, #0x0]\n" + "ldr s11, [x16, #0x0]\n" "ldr s15, [x24, #0x0]\n" "ldr s19, [x23, #0x0]\n" "b 78f\n" "72:" // Height 3: Partial accumulate: partial_2_8 "tbz x8, #1, 73f\n" - "ldr d10, [x17], #0x8\n" + "ldr d10, [x16], #0x8\n" "mov x25, #0x28\n" "ldr d14, [x24], #0x8\n" "ldr d18, [x23], #0x8\n" "tbz x8, #0, 78f\n" - "ld1 { v10.s }[2], [x17]\n" + "ld1 { v10.s }[2], [x16]\n" "ld1 { v14.s }[2], [x24]\n" "ld1 { v18.s }[2], [x23]\n" "b 78f\n" "73:" // Height 3: Partial accumulate: partial_1_8 "mov x25, #0x20\n" "tbz x8, #0, 78f\n" - "ldr s10, [x17, #0x0]\n" + "ldr s10, [x16, #0x0]\n" "ldr s14, [x24, #0x0]\n" "ldr s18, [x23, #0x0]\n" "b 78f\n" "74:" // Height 3: Partial accumulate: partial_4_0 "tbz x8, #2, 76f\n" - "ld1 { v8.4s }, [x17], #0x10\n" + "ld1 { v8.4s }, [x16], #0x10\n" "ld1 { v12.4s }, [x24], #0x10\n" "ld1 { v16.4s }, [x23], #0x10\n" "tbz x8, #1, 75f\n" - "ldr d9, [x17], #0x8\n" + "ldr d9, [x16], #0x8\n" "mov x25, #0x18\n" "ldr d13, [x24], #0x8\n" "ldr d17, [x23], #0x8\n" "tbz x8, #0, 78f\n" - "ld1 { v9.s }[2], [x17]\n" + "ld1 { v9.s }[2], [x16]\n" "ld1 { v13.s }[2], [x24]\n" "ld1 { v17.s }[2], [x23]\n" "b 78f\n" "75:" // Height 3: Partial accumulate: partial_1_4 "mov x25, #0x10\n" "tbz x8, #0, 78f\n" - "ldr s9, [x17, #0x0]\n" + "ldr s9, [x16, #0x0]\n" "ldr s13, [x24, #0x0]\n" "ldr s17, [x23, #0x0]\n" "b 78f\n" "76:" // Height 3: Partial accumulate: partial_2_0 "tbz x8, #1, 77f\n" - "ldr d8, [x17], #0x8\n" + "ldr d8, [x16], #0x8\n" "mov x25, #0x8\n" "ldr d12, [x24], #0x8\n" "ldr d16, [x23], #0x8\n" "tbz x8, #0, 78f\n" - "ld1 { v8.s }[2], [x17]\n" + "ld1 { v8.s }[2], [x16]\n" "ld1 { v12.s }[2], [x24]\n" "ld1 { v16.s }[2], [x23]\n" "b 78f\n" "77:" // Height 3: Partial accumulate: partial_1_0 - "ldr s8, [x17, #0x0]\n" + "ldr s8, [x16, #0x0]\n" "mov x25, #0x0\n" "ldr s12, [x24, #0x0]\n" "ldr s16, [x23, #0x0]\n" "78:" // Height 3: Partial accumulate: Done - "sub x17, x17, x25\n" + "sub x16, x16, x25\n" "b 81f\n" "79:" // Height 3: full accumulate - "ldr q8, [x17, #0x0]\n" - "ldr q9, [x17, #0x10]\n" - "ldr q10, [x17, #0x20]\n" - "ldr q11, [x17, #0x30]\n" + "ldr q8, [x16, #0x0]\n" + "ldr q9, [x16, #0x10]\n" + "ldr q10, [x16, #0x20]\n" + "ldr q11, [x16, #0x30]\n" "ldr q12, [x24, #0x0]\n" "ldr q13, [x24, #0x10]\n" "ldr q14, [x24, #0x20]\n" @@ -934,8 +934,8 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "mov x15, #0x0\n" "82:" // Height 3: String loop "ldr x20, [%x[args_ptr], %[offsetof_string_lengths]]\n" - "ldr w14, [x20, x15, LSL #0x2]\n" "ldr x21, [%x[args_ptr], %[offsetof_input_offset]]\n" + "ldr w14, [x20, x15, LSL #0x2]\n" "tbz %x[flags], #3, 83f\n" "ldr x20, [%x[input_ptr], x15, LSL #0x3]\n" "add x20, x20, x21, LSL #3\n" @@ -959,123 +959,123 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "cmp x14, #0x20\n" "ldr q1, [x12, #0x0]\n" "ldr q2, [x11, #0x0]\n" - "ldr q6, [x16, #0x0]\n" - "ldr q7, [x16, #0x10]\n" + "ldr q6, [x17, #0x0]\n" + "ldr q7, [x17, #0x10]\n" "blt 86f\n" "85:" // Height 3: Multiply loop: Main loop head ".inst 0x6f80e0c8 // udot v8.4s, v6.16b, v0.4b[0]\n" - "ldr x21, [x16, #0x28]\n" + "ldr x21, [x17, #0x28]\n" ".inst 0x6f81e0cc // udot v12.4s, v6.16b, v1.4b[0]\n" - "ldr x20, [x16, #0x38]\n" + "ldr x20, [x17, #0x38]\n" ".inst 0x6f82e0d0 // udot v16.4s, v6.16b, v2.4b[0]\n" - "ldr d21, [x16, #0x20]\n" + "ldr d21, [x17, #0x20]\n" ".inst 0x6f80e0e9 // udot v9.4s, v7.16b, v0.4b[0]\n" - "mov v21.d[1], x21\n" + "add x13, x13, #0x10\n" ".inst 0x6f81e0ed // udot v13.4s, v7.16b, v1.4b[0]\n" - "ldr x21, [x16, #0x48]\n" + "add x12, x12, #0x10\n" ".inst 0x6f82e0f1 // udot v17.4s, v7.16b, v2.4b[0]\n" - "ldr d20, [x16, #0x30]\n" + "ldr d20, [x17, #0x30]\n" + "mov v21.d[1], x21\n" + "ldr x21, [x17, #0x48]\n" + "add x11, x11, #0x10\n" + "ldr x24, [x13, #0x8]\n" "mov v20.d[1], x20\n" ".inst 0x6f80e2aa // udot v10.4s, v21.16b, v0.4b[0]\n" ".inst 0x6f81e2ae // udot v14.4s, v21.16b, v1.4b[0]\n" - "ldr x20, [x16, #0x58]\n" + "ldr x20, [x17, #0x58]\n" ".inst 0x6f82e2b2 // udot v18.4s, v21.16b, v2.4b[0]\n" - "ldr d21, [x16, #0x40]\n" + "ldr d21, [x17, #0x40]\n" ".inst 0x6f80e28b // udot v11.4s, v20.16b, v0.4b[0]\n" - "mov v21.d[1], x21\n" + "ldr x23, [x12, #0x8]\n" ".inst 0x6f81e28f // udot v15.4s, v20.16b, v1.4b[0]\n" - "ldr x21, [x16, #0x68]\n" + "ldr x22, [x11, #0x8]\n" ".inst 0x6f82e293 // udot v19.4s, v20.16b, v2.4b[0]\n" - "ldr d20, [x16, #0x50]\n" + "ldr d20, [x17, #0x50]\n" + "mov v21.d[1], x21\n" + "ldr x21, [x17, #0x68]\n" + "sub x14, x14, #0x10\n" + "prfm pldl1keep, [x13, #0x80]\n" "mov v20.d[1], x20\n" ".inst 0x6fa0e2a8 // udot v8.4s, v21.16b, v0.4b[1]\n" ".inst 0x6fa1e2ac // udot v12.4s, v21.16b, v1.4b[1]\n" - "ldr x20, [x16, #0x78]\n" + "ldr x20, [x17, #0x78]\n" ".inst 0x6fa2e2b0 // udot v16.4s, v21.16b, v2.4b[1]\n" - "ldr d21, [x16, #0x60]\n" + "ldr d21, [x17, #0x60]\n" ".inst 0x6fa0e289 // udot v9.4s, v20.16b, v0.4b[1]\n" - "mov v21.d[1], x21\n" + "cmp x14, #0x20\n" ".inst 0x6fa1e28d // udot v13.4s, v20.16b, v1.4b[1]\n" - "ldr x21, [x16, #0x88]\n" + "prfm pldl1keep, [x12, #0x80]\n" ".inst 0x6fa2e291 // udot v17.4s, v20.16b, v2.4b[1]\n" - "ldr d20, [x16, #0x70]\n" + "ldr d20, [x17, #0x70]\n" + "mov v21.d[1], x21\n" + "ldr x21, [x17, #0x88]\n" + "prfm pldl1keep, [x11, #0x80]\n" "mov v20.d[1], x20\n" ".inst 0x6fa0e2aa // udot v10.4s, v21.16b, v0.4b[1]\n" ".inst 0x6fa1e2ae // udot v14.4s, v21.16b, v1.4b[1]\n" - "ldr x20, [x16, #0x98]\n" + "ldr x20, [x17, #0x98]\n" ".inst 0x6fa2e2b2 // udot v18.4s, v21.16b, v2.4b[1]\n" - "ldr d21, [x16, #0x80]\n" + "ldr d21, [x17, #0x80]\n" ".inst 0x6fa0e28b // udot v11.4s, v20.16b, v0.4b[1]\n" - "mov v21.d[1], x21\n" ".inst 0x6fa1e28f // udot v15.4s, v20.16b, v1.4b[1]\n" - "ldr x21, [x16, #0xa8]\n" ".inst 0x6fa2e293 // udot v19.4s, v20.16b, v2.4b[1]\n" - "ldr d20, [x16, #0x90]\n" + "ldr d20, [x17, #0x90]\n" + "mov v21.d[1], x21\n" + "ldr x21, [x17, #0xa8]\n" "mov v20.d[1], x20\n" ".inst 0x6f80eaa8 // udot v8.4s, v21.16b, v0.4b[2]\n" ".inst 0x6f81eaac // udot v12.4s, v21.16b, v1.4b[2]\n" - "ldr x20, [x16, #0xb8]\n" + "ldr x20, [x17, #0xb8]\n" ".inst 0x6f82eab0 // udot v16.4s, v21.16b, v2.4b[2]\n" - "ldr d21, [x16, #0xa0]\n" + "ldr d21, [x17, #0xa0]\n" ".inst 0x6f80ea89 // udot v9.4s, v20.16b, v0.4b[2]\n" - "mov v21.d[1], x21\n" ".inst 0x6f81ea8d // udot v13.4s, v20.16b, v1.4b[2]\n" - "ldr x21, [x16, #0xc8]\n" ".inst 0x6f82ea91 // udot v17.4s, v20.16b, v2.4b[2]\n" - "ldr d20, [x16, #0xb0]\n" + "ldr d20, [x17, #0xb0]\n" + "mov v21.d[1], x21\n" + "ldr x21, [x17, #0xc8]\n" "mov v20.d[1], x20\n" ".inst 0x6f80eaaa // udot v10.4s, v21.16b, v0.4b[2]\n" ".inst 0x6f81eaae // udot v14.4s, v21.16b, v1.4b[2]\n" - "ldr x20, [x16, #0xd8]\n" + "ldr x20, [x17, #0xd8]\n" ".inst 0x6f82eab2 // udot v18.4s, v21.16b, v2.4b[2]\n" - "ldr d21, [x16, #0xc0]\n" + "ldr d21, [x17, #0xc0]\n" ".inst 0x6f80ea8b // udot v11.4s, v20.16b, v0.4b[2]\n" - "mov v21.d[1], x21\n" ".inst 0x6f81ea8f // udot v15.4s, v20.16b, v1.4b[2]\n" - "ldr x21, [x16, #0xe8]\n" ".inst 0x6f82ea93 // udot v19.4s, v20.16b, v2.4b[2]\n" - "ldr d20, [x16, #0xd0]\n" + "ldr d20, [x17, #0xd0]\n" + "mov v21.d[1], x21\n" + "ldr x21, [x17, #0xe8]\n" "mov v20.d[1], x20\n" ".inst 0x6fa0eaa8 // udot v8.4s, v21.16b, v0.4b[3]\n" ".inst 0x6fa1eaac // udot v12.4s, v21.16b, v1.4b[3]\n" - "ldr x20, [x16, #0xf8]\n" + "ldr x20, [x17, #0xf8]\n" ".inst 0x6fa2eab0 // udot v16.4s, v21.16b, v2.4b[3]\n" - "ldr d21, [x16, #0xe0]\n" + "ldr d21, [x17, #0xe0]\n" ".inst 0x6fa0ea89 // udot v9.4s, v20.16b, v0.4b[3]\n" - "mov v21.d[1], x21\n" ".inst 0x6fa1ea8d // udot v13.4s, v20.16b, v1.4b[3]\n" - "add x13, x13, #0x10\n" ".inst 0x6fa2ea91 // udot v17.4s, v20.16b, v2.4b[3]\n" - "ldr d20, [x16, #0xf0]\n" + "ldr d20, [x17, #0xf0]\n" + "mov v21.d[1], x21\n" + "add x17, x17, #0x100\n" + "ldr x21, [x17, #0x8]\n" "mov v20.d[1], x20\n" - "add x12, x12, #0x10\n" - "add x11, x11, #0x10\n" - "add x16, x16, #0x100\n" ".inst 0x6fa0eaaa // udot v10.4s, v21.16b, v0.4b[3]\n" - "ldr x20, [x16, #0x8]\n" ".inst 0x6fa1eaae // udot v14.4s, v21.16b, v1.4b[3]\n" - "ldr x23, [x13, #0x8]\n" + "ldr x20, [x17, #0x18]\n" ".inst 0x6fa2eab2 // udot v18.4s, v21.16b, v2.4b[3]\n" - "ldr d6, [x16, #0x0]\n" + "ldr d6, [x17, #0x0]\n" ".inst 0x6fa0ea8b // udot v11.4s, v20.16b, v0.4b[3]\n" "ldr d0, [x13, #0x0]\n" ".inst 0x6fa1ea8f // udot v15.4s, v20.16b, v1.4b[3]\n" "ldr d1, [x12, #0x0]\n" - "ldr x22, [x12, #0x8]\n" ".inst 0x6fa2ea93 // udot v19.4s, v20.16b, v2.4b[3]\n" "ldr d2, [x11, #0x0]\n" - "sub x14, x14, #0x10\n" - "ldr d7, [x16, #0x10]\n" - "cmp x14, #0x20\n" - "ldr x21, [x11, #0x8]\n" - "mov v6.d[1], x20\n" - "ldr x20, [x16, #0x18]\n" - "mov v0.d[1], x23\n" - "prfm pldl1keep, [x13, #0x80]\n" - "mov v1.d[1], x22\n" - "prfm pldl1keep, [x12, #0x80]\n" - "mov v2.d[1], x21\n" - "prfm pldl1keep, [x11, #0x80]\n" + "ldr d7, [x17, #0x10]\n" + "mov v6.d[1], x21\n" + "mov v0.d[1], x24\n" + "mov v1.d[1], x23\n" + "mov v2.d[1], x22\n" "mov v7.d[1], x20\n" "bge 85b\n" "86:" // Height 3: Multiply loop: Single iteration only @@ -1084,66 +1084,66 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( ".inst 0x6f81e0cc // udot v12.4s, v6.16b, v1.4b[0]\n" "add x12, x12, #0x10\n" ".inst 0x6f82e0d0 // udot v16.4s, v6.16b, v2.4b[0]\n" - "ldr q21, [x16, #0x20]\n" + "ldr q21, [x17, #0x20]\n" ".inst 0x6f80e0e9 // udot v9.4s, v7.16b, v0.4b[0]\n" "add x11, x11, #0x10\n" ".inst 0x6f81e0ed // udot v13.4s, v7.16b, v1.4b[0]\n" "sub x14, x14, #0x10\n" ".inst 0x6f82e0f1 // udot v17.4s, v7.16b, v2.4b[0]\n" - "ldr q20, [x16, #0x30]\n" + "ldr q20, [x17, #0x30]\n" ".inst 0x6f80e2aa // udot v10.4s, v21.16b, v0.4b[0]\n" "prfm pldl1keep, [x13, #0x80]\n" ".inst 0x6f81e2ae // udot v14.4s, v21.16b, v1.4b[0]\n" "prfm pldl1keep, [x12, #0x80]\n" ".inst 0x6f82e2b2 // udot v18.4s, v21.16b, v2.4b[0]\n" - "ldr q21, [x16, #0x40]\n" + "ldr q21, [x17, #0x40]\n" ".inst 0x6f80e28b // udot v11.4s, v20.16b, v0.4b[0]\n" "prfm pldl1keep, [x11, #0x80]\n" ".inst 0x6f81e28f // udot v15.4s, v20.16b, v1.4b[0]\n" ".inst 0x6f82e293 // udot v19.4s, v20.16b, v2.4b[0]\n" - "ldr q20, [x16, #0x50]\n" + "ldr q20, [x17, #0x50]\n" ".inst 0x6fa0e2a8 // udot v8.4s, v21.16b, v0.4b[1]\n" ".inst 0x6fa1e2ac // udot v12.4s, v21.16b, v1.4b[1]\n" ".inst 0x6fa2e2b0 // udot v16.4s, v21.16b, v2.4b[1]\n" - "ldr q21, [x16, #0x60]\n" + "ldr q21, [x17, #0x60]\n" ".inst 0x6fa0e289 // udot v9.4s, v20.16b, v0.4b[1]\n" ".inst 0x6fa1e28d // udot v13.4s, v20.16b, v1.4b[1]\n" ".inst 0x6fa2e291 // udot v17.4s, v20.16b, v2.4b[1]\n" - "ldr q20, [x16, #0x70]\n" + "ldr q20, [x17, #0x70]\n" ".inst 0x6fa0e2aa // udot v10.4s, v21.16b, v0.4b[1]\n" ".inst 0x6fa1e2ae // udot v14.4s, v21.16b, v1.4b[1]\n" ".inst 0x6fa2e2b2 // udot v18.4s, v21.16b, v2.4b[1]\n" - "ldr q21, [x16, #0x80]\n" + "ldr q21, [x17, #0x80]\n" ".inst 0x6fa0e28b // udot v11.4s, v20.16b, v0.4b[1]\n" ".inst 0x6fa1e28f // udot v15.4s, v20.16b, v1.4b[1]\n" ".inst 0x6fa2e293 // udot v19.4s, v20.16b, v2.4b[1]\n" - "ldr q20, [x16, #0x90]\n" + "ldr q20, [x17, #0x90]\n" ".inst 0x6f80eaa8 // udot v8.4s, v21.16b, v0.4b[2]\n" ".inst 0x6f81eaac // udot v12.4s, v21.16b, v1.4b[2]\n" ".inst 0x6f82eab0 // udot v16.4s, v21.16b, v2.4b[2]\n" - "ldr q21, [x16, #0xa0]\n" + "ldr q21, [x17, #0xa0]\n" ".inst 0x6f80ea89 // udot v9.4s, v20.16b, v0.4b[2]\n" ".inst 0x6f81ea8d // udot v13.4s, v20.16b, v1.4b[2]\n" ".inst 0x6f82ea91 // udot v17.4s, v20.16b, v2.4b[2]\n" - "ldr q20, [x16, #0xb0]\n" + "ldr q20, [x17, #0xb0]\n" ".inst 0x6f80eaaa // udot v10.4s, v21.16b, v0.4b[2]\n" ".inst 0x6f81eaae // udot v14.4s, v21.16b, v1.4b[2]\n" ".inst 0x6f82eab2 // udot v18.4s, v21.16b, v2.4b[2]\n" - "ldr q21, [x16, #0xc0]\n" + "ldr q21, [x17, #0xc0]\n" ".inst 0x6f80ea8b // udot v11.4s, v20.16b, v0.4b[2]\n" ".inst 0x6f81ea8f // udot v15.4s, v20.16b, v1.4b[2]\n" ".inst 0x6f82ea93 // udot v19.4s, v20.16b, v2.4b[2]\n" - "ldr q20, [x16, #0xd0]\n" + "ldr q20, [x17, #0xd0]\n" ".inst 0x6fa0eaa8 // udot v8.4s, v21.16b, v0.4b[3]\n" ".inst 0x6fa1eaac // udot v12.4s, v21.16b, v1.4b[3]\n" ".inst 0x6fa2eab0 // udot v16.4s, v21.16b, v2.4b[3]\n" - "ldr q21, [x16, #0xe0]\n" + "ldr q21, [x17, #0xe0]\n" ".inst 0x6fa0ea89 // udot v9.4s, v20.16b, v0.4b[3]\n" ".inst 0x6fa1ea8d // udot v13.4s, v20.16b, v1.4b[3]\n" ".inst 0x6fa2ea91 // udot v17.4s, v20.16b, v2.4b[3]\n" - "ldr q20, [x16, #0xf0]\n" + "ldr q20, [x17, #0xf0]\n" ".inst 0x6fa0eaaa // udot v10.4s, v21.16b, v0.4b[3]\n" - "add x16, x16, #0x100\n" + "add x17, x17, #0x100\n" ".inst 0x6fa1eaae // udot v14.4s, v21.16b, v1.4b[3]\n" ".inst 0x6fa2eab2 // udot v18.4s, v21.16b, v2.4b[3]\n" ".inst 0x6fa0ea8b // udot v11.4s, v20.16b, v0.4b[3]\n" @@ -1159,18 +1159,18 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "ldr s23, [x12], #0x4\n" "cmp x14, #0x4\n" "ldr s22, [x11], #0x4\n" - "ldr q21, [x16, #0x0]\n" + "ldr q21, [x17, #0x0]\n" + "ldr q20, [x17, #0x10]\n" ".inst 0x6f98e2a8 // udot v8.4s, v21.16b, v24.4b[0]\n" - "ldr q20, [x16, #0x10]\n" ".inst 0x6f97e2ac // udot v12.4s, v21.16b, v23.4b[0]\n" ".inst 0x6f96e2b0 // udot v16.4s, v21.16b, v22.4b[0]\n" - "ldr q21, [x16, #0x20]\n" + "ldr q21, [x17, #0x20]\n" ".inst 0x6f98e289 // udot v9.4s, v20.16b, v24.4b[0]\n" ".inst 0x6f97e28d // udot v13.4s, v20.16b, v23.4b[0]\n" ".inst 0x6f96e291 // udot v17.4s, v20.16b, v22.4b[0]\n" - "ldr q20, [x16, #0x30]\n" + "ldr q20, [x17, #0x30]\n" ".inst 0x6f98e2aa // udot v10.4s, v21.16b, v24.4b[0]\n" - "add x16, x16, #0x40\n" + "add x17, x17, #0x40\n" ".inst 0x6f97e2ae // udot v14.4s, v21.16b, v23.4b[0]\n" ".inst 0x6f96e2b2 // udot v18.4s, v21.16b, v22.4b[0]\n" ".inst 0x6f98e28b // udot v11.4s, v20.16b, v24.4b[0]\n" @@ -1193,18 +1193,18 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "ldr b1, [x12, #0x0]\n" "ldr b2, [x11, #0x0]\n" "91:" // Height 3: Multiply loop: Ragged operand read: Done - "ldr q21, [x16, #0x0]\n" + "ldr q21, [x17, #0x0]\n" + "ldr q20, [x17, #0x10]\n" ".inst 0x6f80e2a8 // udot v8.4s, v21.16b, v0.4b[0]\n" - "ldr q20, [x16, #0x10]\n" ".inst 0x6f81e2ac // udot v12.4s, v21.16b, v1.4b[0]\n" ".inst 0x6f82e2b0 // udot v16.4s, v21.16b, v2.4b[0]\n" - "ldr q21, [x16, #0x20]\n" + "ldr q21, [x17, #0x20]\n" ".inst 0x6f80e289 // udot v9.4s, v20.16b, v0.4b[0]\n" ".inst 0x6f81e28d // udot v13.4s, v20.16b, v1.4b[0]\n" ".inst 0x6f82e291 // udot v17.4s, v20.16b, v2.4b[0]\n" - "ldr q20, [x16, #0x30]\n" + "ldr q20, [x17, #0x30]\n" ".inst 0x6f80e2aa // udot v10.4s, v21.16b, v0.4b[0]\n" - "add x16, x16, #0x40\n" + "add x17, x17, #0x40\n" ".inst 0x6f81e2ae // udot v14.4s, v21.16b, v1.4b[0]\n" ".inst 0x6f82e2b2 // udot v18.4s, v21.16b, v2.4b[0]\n" ".inst 0x6f80e28b // udot v11.4s, v20.16b, v0.4b[0]\n" @@ -1216,97 +1216,97 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "cmp x15, x20\n" "bne 82b\n" "ldr x20, [%x[args_ptr], %[offsetof_output_offset]]\n" - "add x24, x17, x20, LSL #2\n" - "add x23, x24, x20, LSL #2\n" "cmp x8, #0x10\n" - "prfm pstl1keep, [x17, #0x0]\n" + "prfm pstl1keep, [x16, #0x0]\n" + "add x24, x16, x20, LSL #2\n" + "add x23, x24, x20, LSL #2\n" "prfm pstl1keep, [x24, #0x0]\n" "prfm pstl1keep, [x23, #0x0]\n" "bge 101f\n" "tbz x8, #3, 96f\n" - "st1 { v8.4s }, [x17], #0x10\n" - "st1 { v9.4s }, [x17], #0x10\n" + "st1 { v8.4s }, [x16], #0x10\n" + "st1 { v9.4s }, [x16], #0x10\n" "st1 { v12.4s }, [x24], #0x10\n" "st1 { v13.4s }, [x24], #0x10\n" "st1 { v16.4s }, [x23], #0x10\n" "st1 { v17.4s }, [x23], #0x10\n" "tbz x8, #2, 94f\n" - "st1 { v10.4s }, [x17], #0x10\n" + "st1 { v10.4s }, [x16], #0x10\n" "st1 { v14.4s }, [x24], #0x10\n" "st1 { v18.4s }, [x23], #0x10\n" "tbz x8, #1, 93f\n" - "str d11, [x17], #0x8\n" + "str d11, [x16], #0x8\n" "str d15, [x24], #0x8\n" "str d19, [x23], #0x8\n" "tbz x8, #0, 100f\n" - "st1 { v11.s }[2], [x17]\n" + "st1 { v11.s }[2], [x16]\n" "st1 { v15.s }[2], [x24]\n" "st1 { v19.s }[2], [x23]\n" "b 100f\n" "93:" // Height 3: Partial direct writeback: partial_1_12 "tbz x8, #0, 100f\n" - "str s11, [x17, #0x0]\n" + "str s11, [x16, #0x0]\n" "str s15, [x24, #0x0]\n" "str s19, [x23, #0x0]\n" "b 100f\n" "94:" // Height 3: Partial direct writeback: partial_2_8 "tbz x8, #1, 95f\n" - "str d10, [x17], #0x8\n" + "str d10, [x16], #0x8\n" "str d14, [x24], #0x8\n" "str d18, [x23], #0x8\n" "tbz x8, #0, 100f\n" - "st1 { v10.s }[2], [x17]\n" + "st1 { v10.s }[2], [x16]\n" "st1 { v14.s }[2], [x24]\n" "st1 { v18.s }[2], [x23]\n" "b 100f\n" "95:" // Height 3: Partial direct writeback: partial_1_8 "tbz x8, #0, 100f\n" - "str s10, [x17, #0x0]\n" + "str s10, [x16, #0x0]\n" "str s14, [x24, #0x0]\n" "str s18, [x23, #0x0]\n" "b 100f\n" "96:" // Height 3: Partial direct writeback: partial_4_0 "tbz x8, #2, 98f\n" - "st1 { v8.4s }, [x17], #0x10\n" + "st1 { v8.4s }, [x16], #0x10\n" "st1 { v12.4s }, [x24], #0x10\n" "st1 { v16.4s }, [x23], #0x10\n" "tbz x8, #1, 97f\n" - "str d9, [x17], #0x8\n" + "str d9, [x16], #0x8\n" "str d13, [x24], #0x8\n" "str d17, [x23], #0x8\n" "tbz x8, #0, 100f\n" - "st1 { v9.s }[2], [x17]\n" + "st1 { v9.s }[2], [x16]\n" "st1 { v13.s }[2], [x24]\n" "st1 { v17.s }[2], [x23]\n" "b 100f\n" "97:" // Height 3: Partial direct writeback: partial_1_4 "tbz x8, #0, 100f\n" - "str s9, [x17, #0x0]\n" + "str s9, [x16, #0x0]\n" "str s13, [x24, #0x0]\n" "str s17, [x23, #0x0]\n" "b 100f\n" "98:" // Height 3: Partial direct writeback: partial_2_0 "tbz x8, #1, 99f\n" - "str d8, [x17], #0x8\n" + "str d8, [x16], #0x8\n" "str d12, [x24], #0x8\n" "str d16, [x23], #0x8\n" "tbz x8, #0, 100f\n" - "st1 { v8.s }[2], [x17]\n" + "st1 { v8.s }[2], [x16]\n" "st1 { v12.s }[2], [x24]\n" "st1 { v16.s }[2], [x23]\n" "b 100f\n" "99:" // Height 3: Partial direct writeback: partial_1_0 - "str s8, [x17, #0x0]\n" + "str s8, [x16, #0x0]\n" "str s12, [x24, #0x0]\n" "str s16, [x23, #0x0]\n" "100:" // Height 3: Partial direct writeback: Done "b 102f\n" "101:" // Height 3: Full writeback - "str q8, [x17, #0x0]\n" - "str q9, [x17, #0x10]\n" - "str q10, [x17, #0x20]\n" - "str q11, [x17, #0x30]\n" - "add x17, x17, #0x40\n" + "str q8, [x16, #0x0]\n" + "str q9, [x16, #0x10]\n" + "str q10, [x16, #0x20]\n" + "str q11, [x16, #0x30]\n" + "add x16, x16, #0x40\n" "str q12, [x24, #0x0]\n" "str q13, [x24, #0x10]\n" "str q14, [x24, #0x20]\n" @@ -1321,38 +1321,38 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "b 206f\n" "103:" // Height 4 "ldr x8, [%x[args_ptr], %[offsetof_N]]\n" - "mov x17, %x[output_ptr]\n" - "ldr x16, [%x[args_ptr], %[offsetof_B_ptr]]\n" + "ldr x17, [%x[args_ptr], %[offsetof_B_ptr]]\n" + "ldr x16, [%x[args_ptr], %[offsetof_output_ptr]]\n" "104:" // Height 4: Column loop "tbz %x[flags], #0, 114f\n" "ldr x20, [%x[args_ptr], %[offsetof_output_offset]]\n" - "add x24, x17, x20, LSL #2\n" - "add x23, x24, x20, LSL #2\n" "cmp x8, #0x10\n" + "add x24, x16, x20, LSL #2\n" + "add x23, x24, x20, LSL #2\n" "add x22, x23, x20, LSL #2\n" "bge 113f\n" "tbz x8, #3, 108f\n" - "ld1 { v8.4s }, [x17], #0x10\n" + "ld1 { v8.4s }, [x16], #0x10\n" "ld1 { v12.4s }, [x24], #0x10\n" "ld1 { v16.4s }, [x23], #0x10\n" "ld1 { v20.4s }, [x22], #0x10\n" - "ld1 { v9.4s }, [x17], #0x10\n" + "ld1 { v9.4s }, [x16], #0x10\n" "ld1 { v13.4s }, [x24], #0x10\n" "ld1 { v17.4s }, [x23], #0x10\n" "ld1 { v21.4s }, [x22], #0x10\n" "tbz x8, #2, 106f\n" - "ld1 { v10.4s }, [x17], #0x10\n" + "ld1 { v10.4s }, [x16], #0x10\n" "ld1 { v14.4s }, [x24], #0x10\n" "ld1 { v18.4s }, [x23], #0x10\n" "ld1 { v22.4s }, [x22], #0x10\n" "tbz x8, #1, 105f\n" - "ldr d11, [x17], #0x8\n" + "ldr d11, [x16], #0x8\n" "mov x25, #0x38\n" "ldr d15, [x24], #0x8\n" "ldr d19, [x23], #0x8\n" "ldr d23, [x22], #0x8\n" "tbz x8, #0, 112f\n" - "ld1 { v11.s }[2], [x17]\n" + "ld1 { v11.s }[2], [x16]\n" "ld1 { v15.s }[2], [x24]\n" "ld1 { v19.s }[2], [x23]\n" "ld1 { v23.s }[2], [x22]\n" @@ -1360,20 +1360,20 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "105:" // Height 4: Partial accumulate: partial_1_12 "mov x25, #0x30\n" "tbz x8, #0, 112f\n" - "ldr s11, [x17, #0x0]\n" + "ldr s11, [x16, #0x0]\n" "ldr s15, [x24, #0x0]\n" "ldr s19, [x23, #0x0]\n" "ldr s23, [x22, #0x0]\n" "b 112f\n" "106:" // Height 4: Partial accumulate: partial_2_8 "tbz x8, #1, 107f\n" - "ldr d10, [x17], #0x8\n" + "ldr d10, [x16], #0x8\n" "mov x25, #0x28\n" "ldr d14, [x24], #0x8\n" "ldr d18, [x23], #0x8\n" "ldr d22, [x22], #0x8\n" "tbz x8, #0, 112f\n" - "ld1 { v10.s }[2], [x17]\n" + "ld1 { v10.s }[2], [x16]\n" "ld1 { v14.s }[2], [x24]\n" "ld1 { v18.s }[2], [x23]\n" "ld1 { v22.s }[2], [x22]\n" @@ -1381,25 +1381,25 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "107:" // Height 4: Partial accumulate: partial_1_8 "mov x25, #0x20\n" "tbz x8, #0, 112f\n" - "ldr s10, [x17, #0x0]\n" + "ldr s10, [x16, #0x0]\n" "ldr s14, [x24, #0x0]\n" "ldr s18, [x23, #0x0]\n" "ldr s22, [x22, #0x0]\n" "b 112f\n" "108:" // Height 4: Partial accumulate: partial_4_0 "tbz x8, #2, 110f\n" - "ld1 { v8.4s }, [x17], #0x10\n" + "ld1 { v8.4s }, [x16], #0x10\n" "ld1 { v12.4s }, [x24], #0x10\n" "ld1 { v16.4s }, [x23], #0x10\n" "ld1 { v20.4s }, [x22], #0x10\n" "tbz x8, #1, 109f\n" - "ldr d9, [x17], #0x8\n" + "ldr d9, [x16], #0x8\n" "mov x25, #0x18\n" "ldr d13, [x24], #0x8\n" "ldr d17, [x23], #0x8\n" "ldr d21, [x22], #0x8\n" "tbz x8, #0, 112f\n" - "ld1 { v9.s }[2], [x17]\n" + "ld1 { v9.s }[2], [x16]\n" "ld1 { v13.s }[2], [x24]\n" "ld1 { v17.s }[2], [x23]\n" "ld1 { v21.s }[2], [x22]\n" @@ -1407,38 +1407,38 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "109:" // Height 4: Partial accumulate: partial_1_4 "mov x25, #0x10\n" "tbz x8, #0, 112f\n" - "ldr s9, [x17, #0x0]\n" + "ldr s9, [x16, #0x0]\n" "ldr s13, [x24, #0x0]\n" "ldr s17, [x23, #0x0]\n" "ldr s21, [x22, #0x0]\n" "b 112f\n" "110:" // Height 4: Partial accumulate: partial_2_0 "tbz x8, #1, 111f\n" - "ldr d8, [x17], #0x8\n" + "ldr d8, [x16], #0x8\n" "mov x25, #0x8\n" "ldr d12, [x24], #0x8\n" "ldr d16, [x23], #0x8\n" "ldr d20, [x22], #0x8\n" "tbz x8, #0, 112f\n" - "ld1 { v8.s }[2], [x17]\n" + "ld1 { v8.s }[2], [x16]\n" "ld1 { v12.s }[2], [x24]\n" "ld1 { v16.s }[2], [x23]\n" "ld1 { v20.s }[2], [x22]\n" "b 112f\n" "111:" // Height 4: Partial accumulate: partial_1_0 - "ldr s8, [x17, #0x0]\n" + "ldr s8, [x16, #0x0]\n" "mov x25, #0x0\n" "ldr s12, [x24, #0x0]\n" "ldr s16, [x23, #0x0]\n" "ldr s20, [x22, #0x0]\n" "112:" // Height 4: Partial accumulate: Done - "sub x17, x17, x25\n" + "sub x16, x16, x25\n" "b 115f\n" "113:" // Height 4: full accumulate - "ldr q8, [x17, #0x0]\n" - "ldr q9, [x17, #0x10]\n" - "ldr q10, [x17, #0x20]\n" - "ldr q11, [x17, #0x30]\n" + "ldr q8, [x16, #0x0]\n" + "ldr q9, [x16, #0x10]\n" + "ldr q10, [x16, #0x20]\n" + "ldr q11, [x16, #0x30]\n" "ldr q12, [x24, #0x0]\n" "ldr q13, [x24, #0x10]\n" "ldr q14, [x24, #0x20]\n" @@ -1473,8 +1473,8 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "mov x15, #0x0\n" "116:" // Height 4: String loop "ldr x20, [%x[args_ptr], %[offsetof_string_lengths]]\n" - "ldr w14, [x20, x15, LSL #0x2]\n" "ldr x21, [%x[args_ptr], %[offsetof_input_offset]]\n" + "ldr w14, [x20, x15, LSL #0x2]\n" "tbz %x[flags], #3, 117f\n" "ldr x20, [%x[input_ptr], x15, LSL #0x3]\n" "add x20, x20, x21, LSL #3\n" @@ -1502,130 +1502,129 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "ldr q1, [x12, #0x0]\n" "ldr q2, [x11, #0x0]\n" "ldr q3, [x10, #0x0]\n" - "ldr q6, [x16, #0x0]\n" - "ldr q7, [x16, #0x10]\n" + "ldr q6, [x17, #0x0]\n" + "ldr q7, [x17, #0x10]\n" "blt 120f\n" "119:" // Height 4: Multiply loop: Main loop head ".inst 0x6f80e0c8 // udot v8.4s, v6.16b, v0.4b[0]\n" - "ldr x21, [x16, #0x28]\n" + "ldr x20, [x17, #0x28]\n" ".inst 0x6f81e0cc // udot v12.4s, v6.16b, v1.4b[0]\n" - "ldr x20, [x16, #0x38]\n" + "ldr x21, [x17, #0x38]\n" ".inst 0x6f82e0d0 // udot v16.4s, v6.16b, v2.4b[0]\n" "add x13, x13, #0x10\n" ".inst 0x6f83e0d4 // udot v20.4s, v6.16b, v3.4b[0]\n" - "ldr d25, [x16, #0x20]\n" + "ldr d25, [x17, #0x20]\n" ".inst 0x6f80e0e9 // udot v9.4s, v7.16b, v0.4b[0]\n" - "mov v25.d[1], x21\n" + "add x12, x12, #0x10\n" ".inst 0x6f81e0ed // udot v13.4s, v7.16b, v1.4b[0]\n" - "ldr x21, [x16, #0x48]\n" + "add x11, x11, #0x10\n" ".inst 0x6f82e0f1 // udot v17.4s, v7.16b, v2.4b[0]\n" - "add x12, x12, #0x10\n" + "mov v25.d[1], x20\n" ".inst 0x6f83e0f5 // udot v21.4s, v7.16b, v3.4b[0]\n" - "ldr d24, [x16, #0x30]\n" - "mov v24.d[1], x20\n" + "ldr d24, [x17, #0x30]\n" ".inst 0x6f80e32a // udot v10.4s, v25.16b, v0.4b[0]\n" + "ldr x20, [x17, #0x48]\n" ".inst 0x6f81e32e // udot v14.4s, v25.16b, v1.4b[0]\n" - "ldr x20, [x16, #0x58]\n" + "add x10, x10, #0x10\n" + "mov v24.d[1], x21\n" ".inst 0x6f82e332 // udot v18.4s, v25.16b, v2.4b[0]\n" - "add x11, x11, #0x10\n" ".inst 0x6f83e336 // udot v22.4s, v25.16b, v3.4b[0]\n" - "ldr d25, [x16, #0x40]\n" + "ldr d25, [x17, #0x40]\n" ".inst 0x6f80e30b // udot v11.4s, v24.16b, v0.4b[0]\n" - "mov v25.d[1], x21\n" + "ldr x21, [x17, #0x58]\n" ".inst 0x6f81e30f // udot v15.4s, v24.16b, v1.4b[0]\n" - "ldr x21, [x16, #0x68]\n" + "ldr x25, [x13, #0x8]\n" ".inst 0x6f82e313 // udot v19.4s, v24.16b, v2.4b[0]\n" - "add x10, x10, #0x10\n" + "mov v25.d[1], x20\n" ".inst 0x6f83e317 // udot v23.4s, v24.16b, v3.4b[0]\n" - "ldr d24, [x16, #0x50]\n" - "mov v24.d[1], x20\n" + "ldr d24, [x17, #0x50]\n" ".inst 0x6fa0e328 // udot v8.4s, v25.16b, v0.4b[1]\n" + "ldr x20, [x17, #0x68]\n" ".inst 0x6fa1e32c // udot v12.4s, v25.16b, v1.4b[1]\n" - "ldr x20, [x16, #0x78]\n" + "ldr x24, [x12, #0x8]\n" + "mov v24.d[1], x21\n" ".inst 0x6fa2e330 // udot v16.4s, v25.16b, v2.4b[1]\n" - "ldr x25, [x13, #0x8]\n" ".inst 0x6fa3e334 // udot v20.4s, v25.16b, v3.4b[1]\n" - "ldr d25, [x16, #0x60]\n" + "ldr d25, [x17, #0x60]\n" ".inst 0x6fa0e309 // udot v9.4s, v24.16b, v0.4b[1]\n" - "mov v25.d[1], x21\n" + "ldr x21, [x17, #0x78]\n" ".inst 0x6fa1e30d // udot v13.4s, v24.16b, v1.4b[1]\n" - "ldr x21, [x16, #0x88]\n" + "ldr x23, [x11, #0x8]\n" ".inst 0x6fa2e311 // udot v17.4s, v24.16b, v2.4b[1]\n" - "ldr x24, [x12, #0x8]\n" + "mov v25.d[1], x20\n" ".inst 0x6fa3e315 // udot v21.4s, v24.16b, v3.4b[1]\n" - "ldr d24, [x16, #0x70]\n" - "mov v24.d[1], x20\n" + "ldr d24, [x17, #0x70]\n" ".inst 0x6fa0e32a // udot v10.4s, v25.16b, v0.4b[1]\n" + "ldr x20, [x17, #0x88]\n" ".inst 0x6fa1e32e // udot v14.4s, v25.16b, v1.4b[1]\n" - "ldr x20, [x16, #0x98]\n" + "ldr x22, [x10, #0x8]\n" + "mov v24.d[1], x21\n" ".inst 0x6fa2e332 // udot v18.4s, v25.16b, v2.4b[1]\n" - "ldr x23, [x11, #0x8]\n" ".inst 0x6fa3e336 // udot v22.4s, v25.16b, v3.4b[1]\n" - "ldr d25, [x16, #0x80]\n" + "ldr d25, [x17, #0x80]\n" ".inst 0x6fa0e30b // udot v11.4s, v24.16b, v0.4b[1]\n" - "mov v25.d[1], x21\n" + "ldr x21, [x17, #0x98]\n" ".inst 0x6fa1e30f // udot v15.4s, v24.16b, v1.4b[1]\n" - "ldr x21, [x16, #0xa8]\n" + "sub x14, x14, #0x10\n" ".inst 0x6fa2e313 // udot v19.4s, v24.16b, v2.4b[1]\n" - "ldr x22, [x10, #0x8]\n" + "mov v25.d[1], x20\n" ".inst 0x6fa3e317 // udot v23.4s, v24.16b, v3.4b[1]\n" - "ldr d24, [x16, #0x90]\n" - "mov v24.d[1], x20\n" + "ldr d24, [x17, #0x90]\n" ".inst 0x6f80eb28 // udot v8.4s, v25.16b, v0.4b[2]\n" + "ldr x20, [x17, #0xa8]\n" ".inst 0x6f81eb2c // udot v12.4s, v25.16b, v1.4b[2]\n" - "ldr x20, [x16, #0xb8]\n" + "cmp x14, #0x20\n" + "mov v24.d[1], x21\n" ".inst 0x6f82eb30 // udot v16.4s, v25.16b, v2.4b[2]\n" - "sub x14, x14, #0x10\n" ".inst 0x6f83eb34 // udot v20.4s, v25.16b, v3.4b[2]\n" - "ldr d25, [x16, #0xa0]\n" + "ldr d25, [x17, #0xa0]\n" ".inst 0x6f80eb09 // udot v9.4s, v24.16b, v0.4b[2]\n" - "mov v25.d[1], x21\n" + "ldr x21, [x17, #0xb8]\n" ".inst 0x6f81eb0d // udot v13.4s, v24.16b, v1.4b[2]\n" - "ldr x21, [x16, #0xc8]\n" + "prfm pldl1keep, [x13, #0x80]\n" ".inst 0x6f82eb11 // udot v17.4s, v24.16b, v2.4b[2]\n" - "cmp x14, #0x20\n" + "mov v25.d[1], x20\n" ".inst 0x6f83eb15 // udot v21.4s, v24.16b, v3.4b[2]\n" - "ldr d24, [x16, #0xb0]\n" - "mov v24.d[1], x20\n" + "ldr d24, [x17, #0xb0]\n" ".inst 0x6f80eb2a // udot v10.4s, v25.16b, v0.4b[2]\n" + "ldr x20, [x17, #0xc8]\n" ".inst 0x6f81eb2e // udot v14.4s, v25.16b, v1.4b[2]\n" - "ldr x20, [x16, #0xd8]\n" + "prfm pldl1keep, [x12, #0x80]\n" + "mov v24.d[1], x21\n" ".inst 0x6f82eb32 // udot v18.4s, v25.16b, v2.4b[2]\n" - "prfm pldl1keep, [x13, #0x80]\n" ".inst 0x6f83eb36 // udot v22.4s, v25.16b, v3.4b[2]\n" - "ldr d25, [x16, #0xc0]\n" + "ldr d25, [x17, #0xc0]\n" ".inst 0x6f80eb0b // udot v11.4s, v24.16b, v0.4b[2]\n" - "mov v25.d[1], x21\n" + "ldr x21, [x17, #0xd8]\n" ".inst 0x6f81eb0f // udot v15.4s, v24.16b, v1.4b[2]\n" - "ldr x21, [x16, #0xe8]\n" + "prfm pldl1keep, [x11, #0x80]\n" ".inst 0x6f82eb13 // udot v19.4s, v24.16b, v2.4b[2]\n" - "prfm pldl1keep, [x12, #0x80]\n" + "mov v25.d[1], x20\n" ".inst 0x6f83eb17 // udot v23.4s, v24.16b, v3.4b[2]\n" - "ldr d24, [x16, #0xd0]\n" - "mov v24.d[1], x20\n" + "ldr d24, [x17, #0xd0]\n" ".inst 0x6fa0eb28 // udot v8.4s, v25.16b, v0.4b[3]\n" + "ldr x20, [x17, #0xe8]\n" ".inst 0x6fa1eb2c // udot v12.4s, v25.16b, v1.4b[3]\n" - "ldr x20, [x16, #0xf8]\n" + "prfm pldl1keep, [x10, #0x80]\n" + "mov v24.d[1], x21\n" ".inst 0x6fa2eb30 // udot v16.4s, v25.16b, v2.4b[3]\n" - "prfm pldl1keep, [x11, #0x80]\n" ".inst 0x6fa3eb34 // udot v20.4s, v25.16b, v3.4b[3]\n" - "ldr d25, [x16, #0xe0]\n" + "ldr d25, [x17, #0xe0]\n" ".inst 0x6fa0eb09 // udot v9.4s, v24.16b, v0.4b[3]\n" - "mov v25.d[1], x21\n" + "ldr x21, [x17, #0xf8]\n" ".inst 0x6fa1eb0d // udot v13.4s, v24.16b, v1.4b[3]\n" - "prfm pldl1keep, [x10, #0x80]\n" ".inst 0x6fa2eb11 // udot v17.4s, v24.16b, v2.4b[3]\n" + "mov v25.d[1], x20\n" ".inst 0x6fa3eb15 // udot v21.4s, v24.16b, v3.4b[3]\n" - "ldr d24, [x16, #0xf0]\n" - "mov v24.d[1], x20\n" - "add x16, x16, #0x100\n" + "ldr d24, [x17, #0xf0]\n" + "add x17, x17, #0x100\n" ".inst 0x6fa0eb2a // udot v10.4s, v25.16b, v0.4b[3]\n" - "ldr x21, [x16, #0x8]\n" ".inst 0x6fa1eb2e // udot v14.4s, v25.16b, v1.4b[3]\n" - "ldr x20, [x16, #0x18]\n" + "ldr x20, [x17, #0x8]\n" + "mov v24.d[1], x21\n" ".inst 0x6fa2eb32 // udot v18.4s, v25.16b, v2.4b[3]\n" ".inst 0x6fa3eb36 // udot v22.4s, v25.16b, v3.4b[3]\n" - "ldr d6, [x16, #0x0]\n" + "ldr d6, [x17, #0x0]\n" ".inst 0x6fa0eb0b // udot v11.4s, v24.16b, v0.4b[3]\n" "ldr d0, [x13, #0x0]\n" ".inst 0x6fa1eb0f // udot v15.4s, v24.16b, v1.4b[3]\n" @@ -1634,8 +1633,9 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "ldr d2, [x11, #0x0]\n" ".inst 0x6fa3eb17 // udot v23.4s, v24.16b, v3.4b[3]\n" "ldr d3, [x10, #0x0]\n" - "ldr d7, [x16, #0x10]\n" - "mov v6.d[1], x21\n" + "ldr d7, [x17, #0x10]\n" + "mov v6.d[1], x20\n" + "ldr x20, [x17, #0x18]\n" "mov v0.d[1], x25\n" "mov v1.d[1], x24\n" "mov v2.d[1], x23\n" @@ -1650,7 +1650,7 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( ".inst 0x6f82e0d0 // udot v16.4s, v6.16b, v2.4b[0]\n" "add x11, x11, #0x10\n" ".inst 0x6f83e0d4 // udot v20.4s, v6.16b, v3.4b[0]\n" - "ldr q25, [x16, #0x20]\n" + "ldr q25, [x17, #0x20]\n" ".inst 0x6f80e0e9 // udot v9.4s, v7.16b, v0.4b[0]\n" "add x10, x10, #0x10\n" ".inst 0x6f81e0ed // udot v13.4s, v7.16b, v1.4b[0]\n" @@ -1658,7 +1658,7 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( ".inst 0x6f82e0f1 // udot v17.4s, v7.16b, v2.4b[0]\n" "prfm pldl1keep, [x13, #0x80]\n" ".inst 0x6f83e0f5 // udot v21.4s, v7.16b, v3.4b[0]\n" - "ldr q24, [x16, #0x30]\n" + "ldr q24, [x17, #0x30]\n" ".inst 0x6f80e32a // udot v10.4s, v25.16b, v0.4b[0]\n" "prfm pldl1keep, [x12, #0x80]\n" ".inst 0x6f81e32e // udot v14.4s, v25.16b, v1.4b[0]\n" @@ -1666,64 +1666,64 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( ".inst 0x6f82e332 // udot v18.4s, v25.16b, v2.4b[0]\n" "prfm pldl1keep, [x10, #0x80]\n" ".inst 0x6f83e336 // udot v22.4s, v25.16b, v3.4b[0]\n" - "ldr q25, [x16, #0x40]\n" + "ldr q25, [x17, #0x40]\n" ".inst 0x6f80e30b // udot v11.4s, v24.16b, v0.4b[0]\n" ".inst 0x6f81e30f // udot v15.4s, v24.16b, v1.4b[0]\n" ".inst 0x6f82e313 // udot v19.4s, v24.16b, v2.4b[0]\n" ".inst 0x6f83e317 // udot v23.4s, v24.16b, v3.4b[0]\n" - "ldr q24, [x16, #0x50]\n" + "ldr q24, [x17, #0x50]\n" ".inst 0x6fa0e328 // udot v8.4s, v25.16b, v0.4b[1]\n" ".inst 0x6fa1e32c // udot v12.4s, v25.16b, v1.4b[1]\n" ".inst 0x6fa2e330 // udot v16.4s, v25.16b, v2.4b[1]\n" ".inst 0x6fa3e334 // udot v20.4s, v25.16b, v3.4b[1]\n" - "ldr q25, [x16, #0x60]\n" + "ldr q25, [x17, #0x60]\n" ".inst 0x6fa0e309 // udot v9.4s, v24.16b, v0.4b[1]\n" ".inst 0x6fa1e30d // udot v13.4s, v24.16b, v1.4b[1]\n" ".inst 0x6fa2e311 // udot v17.4s, v24.16b, v2.4b[1]\n" ".inst 0x6fa3e315 // udot v21.4s, v24.16b, v3.4b[1]\n" - "ldr q24, [x16, #0x70]\n" + "ldr q24, [x17, #0x70]\n" ".inst 0x6fa0e32a // udot v10.4s, v25.16b, v0.4b[1]\n" ".inst 0x6fa1e32e // udot v14.4s, v25.16b, v1.4b[1]\n" ".inst 0x6fa2e332 // udot v18.4s, v25.16b, v2.4b[1]\n" ".inst 0x6fa3e336 // udot v22.4s, v25.16b, v3.4b[1]\n" - "ldr q25, [x16, #0x80]\n" + "ldr q25, [x17, #0x80]\n" ".inst 0x6fa0e30b // udot v11.4s, v24.16b, v0.4b[1]\n" ".inst 0x6fa1e30f // udot v15.4s, v24.16b, v1.4b[1]\n" ".inst 0x6fa2e313 // udot v19.4s, v24.16b, v2.4b[1]\n" ".inst 0x6fa3e317 // udot v23.4s, v24.16b, v3.4b[1]\n" - "ldr q24, [x16, #0x90]\n" + "ldr q24, [x17, #0x90]\n" ".inst 0x6f80eb28 // udot v8.4s, v25.16b, v0.4b[2]\n" ".inst 0x6f81eb2c // udot v12.4s, v25.16b, v1.4b[2]\n" ".inst 0x6f82eb30 // udot v16.4s, v25.16b, v2.4b[2]\n" ".inst 0x6f83eb34 // udot v20.4s, v25.16b, v3.4b[2]\n" - "ldr q25, [x16, #0xa0]\n" + "ldr q25, [x17, #0xa0]\n" ".inst 0x6f80eb09 // udot v9.4s, v24.16b, v0.4b[2]\n" ".inst 0x6f81eb0d // udot v13.4s, v24.16b, v1.4b[2]\n" ".inst 0x6f82eb11 // udot v17.4s, v24.16b, v2.4b[2]\n" ".inst 0x6f83eb15 // udot v21.4s, v24.16b, v3.4b[2]\n" - "ldr q24, [x16, #0xb0]\n" + "ldr q24, [x17, #0xb0]\n" ".inst 0x6f80eb2a // udot v10.4s, v25.16b, v0.4b[2]\n" ".inst 0x6f81eb2e // udot v14.4s, v25.16b, v1.4b[2]\n" ".inst 0x6f82eb32 // udot v18.4s, v25.16b, v2.4b[2]\n" ".inst 0x6f83eb36 // udot v22.4s, v25.16b, v3.4b[2]\n" - "ldr q25, [x16, #0xc0]\n" + "ldr q25, [x17, #0xc0]\n" ".inst 0x6f80eb0b // udot v11.4s, v24.16b, v0.4b[2]\n" ".inst 0x6f81eb0f // udot v15.4s, v24.16b, v1.4b[2]\n" ".inst 0x6f82eb13 // udot v19.4s, v24.16b, v2.4b[2]\n" ".inst 0x6f83eb17 // udot v23.4s, v24.16b, v3.4b[2]\n" - "ldr q24, [x16, #0xd0]\n" + "ldr q24, [x17, #0xd0]\n" ".inst 0x6fa0eb28 // udot v8.4s, v25.16b, v0.4b[3]\n" ".inst 0x6fa1eb2c // udot v12.4s, v25.16b, v1.4b[3]\n" ".inst 0x6fa2eb30 // udot v16.4s, v25.16b, v2.4b[3]\n" ".inst 0x6fa3eb34 // udot v20.4s, v25.16b, v3.4b[3]\n" - "ldr q25, [x16, #0xe0]\n" + "ldr q25, [x17, #0xe0]\n" ".inst 0x6fa0eb09 // udot v9.4s, v24.16b, v0.4b[3]\n" ".inst 0x6fa1eb0d // udot v13.4s, v24.16b, v1.4b[3]\n" ".inst 0x6fa2eb11 // udot v17.4s, v24.16b, v2.4b[3]\n" ".inst 0x6fa3eb15 // udot v21.4s, v24.16b, v3.4b[3]\n" - "ldr q24, [x16, #0xf0]\n" + "ldr q24, [x17, #0xf0]\n" ".inst 0x6fa0eb2a // udot v10.4s, v25.16b, v0.4b[3]\n" - "add x16, x16, #0x100\n" + "add x17, x17, #0x100\n" ".inst 0x6fa1eb2e // udot v14.4s, v25.16b, v1.4b[3]\n" ".inst 0x6fa2eb32 // udot v18.4s, v25.16b, v2.4b[3]\n" ".inst 0x6fa3eb36 // udot v22.4s, v25.16b, v3.4b[3]\n" @@ -1742,20 +1742,20 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "cmp x14, #0x4\n" "ldr s27, [x11], #0x4\n" "ldr s26, [x10], #0x4\n" - "ldr q25, [x16, #0x0]\n" + "ldr q25, [x17, #0x0]\n" + "ldr q24, [x17, #0x10]\n" ".inst 0x6f9de328 // udot v8.4s, v25.16b, v29.4b[0]\n" - "ldr q24, [x16, #0x10]\n" ".inst 0x6f9ce32c // udot v12.4s, v25.16b, v28.4b[0]\n" ".inst 0x6f9be330 // udot v16.4s, v25.16b, v27.4b[0]\n" ".inst 0x6f9ae334 // udot v20.4s, v25.16b, v26.4b[0]\n" - "ldr q25, [x16, #0x20]\n" + "ldr q25, [x17, #0x20]\n" ".inst 0x6f9de309 // udot v9.4s, v24.16b, v29.4b[0]\n" ".inst 0x6f9ce30d // udot v13.4s, v24.16b, v28.4b[0]\n" ".inst 0x6f9be311 // udot v17.4s, v24.16b, v27.4b[0]\n" ".inst 0x6f9ae315 // udot v21.4s, v24.16b, v26.4b[0]\n" - "ldr q24, [x16, #0x30]\n" + "ldr q24, [x17, #0x30]\n" ".inst 0x6f9de32a // udot v10.4s, v25.16b, v29.4b[0]\n" - "add x16, x16, #0x40\n" + "add x17, x17, #0x40\n" ".inst 0x6f9ce32e // udot v14.4s, v25.16b, v28.4b[0]\n" ".inst 0x6f9be332 // udot v18.4s, v25.16b, v27.4b[0]\n" ".inst 0x6f9ae336 // udot v22.4s, v25.16b, v26.4b[0]\n" @@ -1783,20 +1783,20 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "ldr b2, [x11, #0x0]\n" "ldr b3, [x10, #0x0]\n" "125:" // Height 4: Multiply loop: Ragged operand read: Done - "ldr q25, [x16, #0x0]\n" + "ldr q25, [x17, #0x0]\n" + "ldr q24, [x17, #0x10]\n" ".inst 0x6f80e328 // udot v8.4s, v25.16b, v0.4b[0]\n" - "ldr q24, [x16, #0x10]\n" ".inst 0x6f81e32c // udot v12.4s, v25.16b, v1.4b[0]\n" ".inst 0x6f82e330 // udot v16.4s, v25.16b, v2.4b[0]\n" ".inst 0x6f83e334 // udot v20.4s, v25.16b, v3.4b[0]\n" - "ldr q25, [x16, #0x20]\n" + "ldr q25, [x17, #0x20]\n" ".inst 0x6f80e309 // udot v9.4s, v24.16b, v0.4b[0]\n" ".inst 0x6f81e30d // udot v13.4s, v24.16b, v1.4b[0]\n" ".inst 0x6f82e311 // udot v17.4s, v24.16b, v2.4b[0]\n" ".inst 0x6f83e315 // udot v21.4s, v24.16b, v3.4b[0]\n" - "ldr q24, [x16, #0x30]\n" + "ldr q24, [x17, #0x30]\n" ".inst 0x6f80e32a // udot v10.4s, v25.16b, v0.4b[0]\n" - "add x16, x16, #0x40\n" + "add x17, x17, #0x40\n" ".inst 0x6f81e32e // udot v14.4s, v25.16b, v1.4b[0]\n" ".inst 0x6f82e332 // udot v18.4s, v25.16b, v2.4b[0]\n" ".inst 0x6f83e336 // udot v22.4s, v25.16b, v3.4b[0]\n" @@ -1810,18 +1810,18 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "cmp x15, x20\n" "bne 116b\n" "ldr x20, [%x[args_ptr], %[offsetof_output_offset]]\n" - "add x24, x17, x20, LSL #2\n" - "add x23, x24, x20, LSL #2\n" - "add x22, x23, x20, LSL #2\n" "cmp x8, #0x10\n" - "prfm pstl1keep, [x17, #0x0]\n" + "prfm pstl1keep, [x16, #0x0]\n" + "add x24, x16, x20, LSL #2\n" + "add x23, x24, x20, LSL #2\n" "prfm pstl1keep, [x24, #0x0]\n" + "add x22, x23, x20, LSL #2\n" "prfm pstl1keep, [x23, #0x0]\n" "prfm pstl1keep, [x22, #0x0]\n" "bge 135f\n" "tbz x8, #3, 130f\n" - "st1 { v8.4s }, [x17], #0x10\n" - "st1 { v9.4s }, [x17], #0x10\n" + "st1 { v8.4s }, [x16], #0x10\n" + "st1 { v9.4s }, [x16], #0x10\n" "st1 { v12.4s }, [x24], #0x10\n" "st1 { v13.4s }, [x24], #0x10\n" "st1 { v16.4s }, [x23], #0x10\n" @@ -1829,96 +1829,96 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "st1 { v20.4s }, [x22], #0x10\n" "st1 { v21.4s }, [x22], #0x10\n" "tbz x8, #2, 128f\n" - "st1 { v10.4s }, [x17], #0x10\n" + "st1 { v10.4s }, [x16], #0x10\n" "st1 { v14.4s }, [x24], #0x10\n" "st1 { v18.4s }, [x23], #0x10\n" "st1 { v22.4s }, [x22], #0x10\n" "tbz x8, #1, 127f\n" - "str d11, [x17], #0x8\n" + "str d11, [x16], #0x8\n" "str d15, [x24], #0x8\n" "str d19, [x23], #0x8\n" "str d23, [x22], #0x8\n" "tbz x8, #0, 134f\n" - "st1 { v11.s }[2], [x17]\n" + "st1 { v11.s }[2], [x16]\n" "st1 { v15.s }[2], [x24]\n" "st1 { v19.s }[2], [x23]\n" "st1 { v23.s }[2], [x22]\n" "b 134f\n" "127:" // Height 4: Partial direct writeback: partial_1_12 "tbz x8, #0, 134f\n" - "str s11, [x17, #0x0]\n" + "str s11, [x16, #0x0]\n" "str s15, [x24, #0x0]\n" "str s19, [x23, #0x0]\n" "str s23, [x22, #0x0]\n" "b 134f\n" "128:" // Height 4: Partial direct writeback: partial_2_8 "tbz x8, #1, 129f\n" - "str d10, [x17], #0x8\n" + "str d10, [x16], #0x8\n" "str d14, [x24], #0x8\n" "str d18, [x23], #0x8\n" "str d22, [x22], #0x8\n" "tbz x8, #0, 134f\n" - "st1 { v10.s }[2], [x17]\n" + "st1 { v10.s }[2], [x16]\n" "st1 { v14.s }[2], [x24]\n" "st1 { v18.s }[2], [x23]\n" "st1 { v22.s }[2], [x22]\n" "b 134f\n" "129:" // Height 4: Partial direct writeback: partial_1_8 "tbz x8, #0, 134f\n" - "str s10, [x17, #0x0]\n" + "str s10, [x16, #0x0]\n" "str s14, [x24, #0x0]\n" "str s18, [x23, #0x0]\n" "str s22, [x22, #0x0]\n" "b 134f\n" "130:" // Height 4: Partial direct writeback: partial_4_0 "tbz x8, #2, 132f\n" - "st1 { v8.4s }, [x17], #0x10\n" + "st1 { v8.4s }, [x16], #0x10\n" "st1 { v12.4s }, [x24], #0x10\n" "st1 { v16.4s }, [x23], #0x10\n" "st1 { v20.4s }, [x22], #0x10\n" "tbz x8, #1, 131f\n" - "str d9, [x17], #0x8\n" + "str d9, [x16], #0x8\n" "str d13, [x24], #0x8\n" "str d17, [x23], #0x8\n" "str d21, [x22], #0x8\n" "tbz x8, #0, 134f\n" - "st1 { v9.s }[2], [x17]\n" + "st1 { v9.s }[2], [x16]\n" "st1 { v13.s }[2], [x24]\n" "st1 { v17.s }[2], [x23]\n" "st1 { v21.s }[2], [x22]\n" "b 134f\n" "131:" // Height 4: Partial direct writeback: partial_1_4 "tbz x8, #0, 134f\n" - "str s9, [x17, #0x0]\n" + "str s9, [x16, #0x0]\n" "str s13, [x24, #0x0]\n" "str s17, [x23, #0x0]\n" "str s21, [x22, #0x0]\n" "b 134f\n" "132:" // Height 4: Partial direct writeback: partial_2_0 "tbz x8, #1, 133f\n" - "str d8, [x17], #0x8\n" + "str d8, [x16], #0x8\n" "str d12, [x24], #0x8\n" "str d16, [x23], #0x8\n" "str d20, [x22], #0x8\n" "tbz x8, #0, 134f\n" - "st1 { v8.s }[2], [x17]\n" + "st1 { v8.s }[2], [x16]\n" "st1 { v12.s }[2], [x24]\n" "st1 { v16.s }[2], [x23]\n" "st1 { v20.s }[2], [x22]\n" "b 134f\n" "133:" // Height 4: Partial direct writeback: partial_1_0 - "str s8, [x17, #0x0]\n" + "str s8, [x16, #0x0]\n" "str s12, [x24, #0x0]\n" "str s16, [x23, #0x0]\n" "str s20, [x22, #0x0]\n" "134:" // Height 4: Partial direct writeback: Done "b 136f\n" "135:" // Height 4: Full writeback - "str q8, [x17, #0x0]\n" - "str q9, [x17, #0x10]\n" - "str q10, [x17, #0x20]\n" - "str q11, [x17, #0x30]\n" - "add x17, x17, #0x40\n" + "str q8, [x16, #0x0]\n" + "str q9, [x16, #0x10]\n" + "str q10, [x16, #0x20]\n" + "str q11, [x16, #0x30]\n" + "add x16, x16, #0x40\n" "str q12, [x24, #0x0]\n" "str q13, [x24, #0x10]\n" "str q14, [x24, #0x20]\n" @@ -1937,43 +1937,43 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "b 206f\n" "137:" // Height 5 "ldr x8, [%x[args_ptr], %[offsetof_N]]\n" - "mov x17, %x[output_ptr]\n" - "ldr x16, [%x[args_ptr], %[offsetof_B_ptr]]\n" + "ldr x17, [%x[args_ptr], %[offsetof_B_ptr]]\n" + "ldr x16, [%x[args_ptr], %[offsetof_output_ptr]]\n" "138:" // Height 5: Column loop "tbz %x[flags], #0, 148f\n" "ldr x20, [%x[args_ptr], %[offsetof_output_offset]]\n" - "add x24, x17, x20, LSL #2\n" + "cmp x8, #0x10\n" + "add x24, x16, x20, LSL #2\n" "add x23, x24, x20, LSL #2\n" "add x22, x23, x20, LSL #2\n" - "cmp x8, #0x10\n" "add x21, x22, x20, LSL #2\n" "bge 147f\n" "tbz x8, #3, 142f\n" - "ld1 { v8.4s }, [x17], #0x10\n" + "ld1 { v8.4s }, [x16], #0x10\n" "ld1 { v12.4s }, [x24], #0x10\n" "ld1 { v16.4s }, [x23], #0x10\n" "ld1 { v20.4s }, [x22], #0x10\n" "ld1 { v24.4s }, [x21], #0x10\n" - "ld1 { v9.4s }, [x17], #0x10\n" + "ld1 { v9.4s }, [x16], #0x10\n" "ld1 { v13.4s }, [x24], #0x10\n" "ld1 { v17.4s }, [x23], #0x10\n" "ld1 { v21.4s }, [x22], #0x10\n" "ld1 { v25.4s }, [x21], #0x10\n" "tbz x8, #2, 140f\n" - "ld1 { v10.4s }, [x17], #0x10\n" + "ld1 { v10.4s }, [x16], #0x10\n" "ld1 { v14.4s }, [x24], #0x10\n" "ld1 { v18.4s }, [x23], #0x10\n" "ld1 { v22.4s }, [x22], #0x10\n" "ld1 { v26.4s }, [x21], #0x10\n" "tbz x8, #1, 139f\n" - "ldr d11, [x17], #0x8\n" + "ldr d11, [x16], #0x8\n" "mov x25, #0x38\n" "ldr d15, [x24], #0x8\n" "ldr d19, [x23], #0x8\n" "ldr d23, [x22], #0x8\n" "ldr d27, [x21], #0x8\n" "tbz x8, #0, 146f\n" - "ld1 { v11.s }[2], [x17]\n" + "ld1 { v11.s }[2], [x16]\n" "ld1 { v15.s }[2], [x24]\n" "ld1 { v19.s }[2], [x23]\n" "ld1 { v23.s }[2], [x22]\n" @@ -1982,7 +1982,7 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "139:" // Height 5: Partial accumulate: partial_1_12 "mov x25, #0x30\n" "tbz x8, #0, 146f\n" - "ldr s11, [x17, #0x0]\n" + "ldr s11, [x16, #0x0]\n" "ldr s15, [x24, #0x0]\n" "ldr s19, [x23, #0x0]\n" "ldr s23, [x22, #0x0]\n" @@ -1990,14 +1990,14 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "b 146f\n" "140:" // Height 5: Partial accumulate: partial_2_8 "tbz x8, #1, 141f\n" - "ldr d10, [x17], #0x8\n" + "ldr d10, [x16], #0x8\n" "mov x25, #0x28\n" "ldr d14, [x24], #0x8\n" "ldr d18, [x23], #0x8\n" "ldr d22, [x22], #0x8\n" "ldr d26, [x21], #0x8\n" "tbz x8, #0, 146f\n" - "ld1 { v10.s }[2], [x17]\n" + "ld1 { v10.s }[2], [x16]\n" "ld1 { v14.s }[2], [x24]\n" "ld1 { v18.s }[2], [x23]\n" "ld1 { v22.s }[2], [x22]\n" @@ -2006,7 +2006,7 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "141:" // Height 5: Partial accumulate: partial_1_8 "mov x25, #0x20\n" "tbz x8, #0, 146f\n" - "ldr s10, [x17, #0x0]\n" + "ldr s10, [x16, #0x0]\n" "ldr s14, [x24, #0x0]\n" "ldr s18, [x23, #0x0]\n" "ldr s22, [x22, #0x0]\n" @@ -2014,20 +2014,20 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "b 146f\n" "142:" // Height 5: Partial accumulate: partial_4_0 "tbz x8, #2, 144f\n" - "ld1 { v8.4s }, [x17], #0x10\n" + "ld1 { v8.4s }, [x16], #0x10\n" "ld1 { v12.4s }, [x24], #0x10\n" "ld1 { v16.4s }, [x23], #0x10\n" "ld1 { v20.4s }, [x22], #0x10\n" "ld1 { v24.4s }, [x21], #0x10\n" "tbz x8, #1, 143f\n" - "ldr d9, [x17], #0x8\n" + "ldr d9, [x16], #0x8\n" "mov x25, #0x18\n" "ldr d13, [x24], #0x8\n" "ldr d17, [x23], #0x8\n" "ldr d21, [x22], #0x8\n" "ldr d25, [x21], #0x8\n" "tbz x8, #0, 146f\n" - "ld1 { v9.s }[2], [x17]\n" + "ld1 { v9.s }[2], [x16]\n" "ld1 { v13.s }[2], [x24]\n" "ld1 { v17.s }[2], [x23]\n" "ld1 { v21.s }[2], [x22]\n" @@ -2036,7 +2036,7 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "143:" // Height 5: Partial accumulate: partial_1_4 "mov x25, #0x10\n" "tbz x8, #0, 146f\n" - "ldr s9, [x17, #0x0]\n" + "ldr s9, [x16, #0x0]\n" "ldr s13, [x24, #0x0]\n" "ldr s17, [x23, #0x0]\n" "ldr s21, [x22, #0x0]\n" @@ -2044,34 +2044,34 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "b 146f\n" "144:" // Height 5: Partial accumulate: partial_2_0 "tbz x8, #1, 145f\n" - "ldr d8, [x17], #0x8\n" + "ldr d8, [x16], #0x8\n" "mov x25, #0x8\n" "ldr d12, [x24], #0x8\n" "ldr d16, [x23], #0x8\n" "ldr d20, [x22], #0x8\n" "ldr d24, [x21], #0x8\n" "tbz x8, #0, 146f\n" - "ld1 { v8.s }[2], [x17]\n" + "ld1 { v8.s }[2], [x16]\n" "ld1 { v12.s }[2], [x24]\n" "ld1 { v16.s }[2], [x23]\n" "ld1 { v20.s }[2], [x22]\n" "ld1 { v24.s }[2], [x21]\n" "b 146f\n" "145:" // Height 5: Partial accumulate: partial_1_0 - "ldr s8, [x17, #0x0]\n" + "ldr s8, [x16, #0x0]\n" "mov x25, #0x0\n" "ldr s12, [x24, #0x0]\n" "ldr s16, [x23, #0x0]\n" "ldr s20, [x22, #0x0]\n" "ldr s24, [x21, #0x0]\n" "146:" // Height 5: Partial accumulate: Done - "sub x17, x17, x25\n" + "sub x16, x16, x25\n" "b 149f\n" "147:" // Height 5: full accumulate - "ldr q8, [x17, #0x0]\n" - "ldr q9, [x17, #0x10]\n" - "ldr q10, [x17, #0x20]\n" - "ldr q11, [x17, #0x30]\n" + "ldr q8, [x16, #0x0]\n" + "ldr q9, [x16, #0x10]\n" + "ldr q10, [x16, #0x20]\n" + "ldr q11, [x16, #0x30]\n" "ldr q12, [x24, #0x0]\n" "ldr q13, [x24, #0x10]\n" "ldr q14, [x24, #0x20]\n" @@ -2114,8 +2114,8 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "mov x15, #0x0\n" "150:" // Height 5: String loop "ldr x20, [%x[args_ptr], %[offsetof_string_lengths]]\n" - "ldr w14, [x20, x15, LSL #0x2]\n" "ldr x21, [%x[args_ptr], %[offsetof_input_offset]]\n" + "ldr w14, [x20, x15, LSL #0x2]\n" "tbz %x[flags], #3, 151f\n" "ldr x20, [%x[input_ptr], x15, LSL #0x3]\n" "add x20, x20, x21, LSL #3\n" @@ -2147,148 +2147,148 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "ldr q2, [x11, #0x0]\n" "ldr q3, [x10, #0x0]\n" "ldr q4, [x9, #0x0]\n" - "ldr q6, [x16, #0x0]\n" - "ldr q7, [x16, #0x10]\n" + "ldr q6, [x17, #0x0]\n" + "ldr q7, [x17, #0x10]\n" "blt 154f\n" "153:" // Height 5: Multiply loop: Main loop head ".inst 0x6f80e0c8 // udot v8.4s, v6.16b, v0.4b[0]\n" - "ldr x21, [x16, #0x28]\n" + "ldr x21, [x17, #0x28]\n" ".inst 0x6f81e0cc // udot v12.4s, v6.16b, v1.4b[0]\n" - "ldr x20, [x16, #0x38]\n" + "ldr x20, [x17, #0x38]\n" ".inst 0x6f82e0d0 // udot v16.4s, v6.16b, v2.4b[0]\n" "add x13, x13, #0x10\n" ".inst 0x6f83e0d4 // udot v20.4s, v6.16b, v3.4b[0]\n" "add x12, x12, #0x10\n" ".inst 0x6f84e0d8 // udot v24.4s, v6.16b, v4.4b[0]\n" - "ldr d29, [x16, #0x20]\n" + "ldr d29, [x17, #0x20]\n" ".inst 0x6f80e0e9 // udot v9.4s, v7.16b, v0.4b[0]\n" - "mov v29.d[1], x21\n" + "add x11, x11, #0x10\n" ".inst 0x6f81e0ed // udot v13.4s, v7.16b, v1.4b[0]\n" - "ldr x21, [x16, #0x48]\n" + "add x10, x10, #0x10\n" ".inst 0x6f82e0f1 // udot v17.4s, v7.16b, v2.4b[0]\n" - "add x11, x11, #0x10\n" + "mov v29.d[1], x21\n" ".inst 0x6f83e0f5 // udot v21.4s, v7.16b, v3.4b[0]\n" - "add x10, x10, #0x10\n" + "ldr x21, [x17, #0x48]\n" ".inst 0x6f84e0f9 // udot v25.4s, v7.16b, v4.4b[0]\n" - "ldr d28, [x16, #0x30]\n" - "mov v28.d[1], x20\n" + "ldr d28, [x17, #0x30]\n" ".inst 0x6f80e3aa // udot v10.4s, v29.16b, v0.4b[0]\n" + "add x9, x9, #0x10\n" ".inst 0x6f81e3ae // udot v14.4s, v29.16b, v1.4b[0]\n" - "ldr x20, [x16, #0x58]\n" + "ldr x26, [x13, #0x8]\n" + "mov v28.d[1], x20\n" ".inst 0x6f82e3b2 // udot v18.4s, v29.16b, v2.4b[0]\n" - "add x9, x9, #0x10\n" ".inst 0x6f83e3b6 // udot v22.4s, v29.16b, v3.4b[0]\n" - "ldr x26, [x13, #0x8]\n" + "ldr x20, [x17, #0x58]\n" ".inst 0x6f84e3ba // udot v26.4s, v29.16b, v4.4b[0]\n" - "ldr d29, [x16, #0x40]\n" + "ldr d29, [x17, #0x40]\n" ".inst 0x6f80e38b // udot v11.4s, v28.16b, v0.4b[0]\n" - "mov v29.d[1], x21\n" + "ldr x25, [x12, #0x8]\n" ".inst 0x6f81e38f // udot v15.4s, v28.16b, v1.4b[0]\n" - "ldr x21, [x16, #0x68]\n" + "ldr x24, [x11, #0x8]\n" ".inst 0x6f82e393 // udot v19.4s, v28.16b, v2.4b[0]\n" - "ldr x25, [x12, #0x8]\n" + "mov v29.d[1], x21\n" ".inst 0x6f83e397 // udot v23.4s, v28.16b, v3.4b[0]\n" - "ldr x24, [x11, #0x8]\n" + "ldr x21, [x17, #0x68]\n" ".inst 0x6f84e39b // udot v27.4s, v28.16b, v4.4b[0]\n" - "ldr d28, [x16, #0x50]\n" - "mov v28.d[1], x20\n" + "ldr d28, [x17, #0x50]\n" ".inst 0x6fa0e3a8 // udot v8.4s, v29.16b, v0.4b[1]\n" + "ldr x23, [x10, #0x8]\n" ".inst 0x6fa1e3ac // udot v12.4s, v29.16b, v1.4b[1]\n" - "ldr x20, [x16, #0x78]\n" + "ldr x22, [x9, #0x8]\n" + "mov v28.d[1], x20\n" ".inst 0x6fa2e3b0 // udot v16.4s, v29.16b, v2.4b[1]\n" - "ldr x23, [x10, #0x8]\n" ".inst 0x6fa3e3b4 // udot v20.4s, v29.16b, v3.4b[1]\n" - "ldr x22, [x9, #0x8]\n" + "ldr x20, [x17, #0x78]\n" ".inst 0x6fa4e3b8 // udot v24.4s, v29.16b, v4.4b[1]\n" - "ldr d29, [x16, #0x60]\n" + "ldr d29, [x17, #0x60]\n" ".inst 0x6fa0e389 // udot v9.4s, v28.16b, v0.4b[1]\n" - "mov v29.d[1], x21\n" + "sub x14, x14, #0x10\n" ".inst 0x6fa1e38d // udot v13.4s, v28.16b, v1.4b[1]\n" - "ldr x21, [x16, #0x88]\n" + "cmp x14, #0x20\n" ".inst 0x6fa2e391 // udot v17.4s, v28.16b, v2.4b[1]\n" - "sub x14, x14, #0x10\n" + "mov v29.d[1], x21\n" ".inst 0x6fa3e395 // udot v21.4s, v28.16b, v3.4b[1]\n" - "cmp x14, #0x20\n" + "ldr x21, [x17, #0x88]\n" ".inst 0x6fa4e399 // udot v25.4s, v28.16b, v4.4b[1]\n" - "ldr d28, [x16, #0x70]\n" - "mov v28.d[1], x20\n" + "ldr d28, [x17, #0x70]\n" ".inst 0x6fa0e3aa // udot v10.4s, v29.16b, v0.4b[1]\n" + "prfm pldl1keep, [x13, #0x80]\n" ".inst 0x6fa1e3ae // udot v14.4s, v29.16b, v1.4b[1]\n" - "ldr x20, [x16, #0x98]\n" + "prfm pldl1keep, [x12, #0x80]\n" + "mov v28.d[1], x20\n" ".inst 0x6fa2e3b2 // udot v18.4s, v29.16b, v2.4b[1]\n" - "prfm pldl1keep, [x13, #0x80]\n" ".inst 0x6fa3e3b6 // udot v22.4s, v29.16b, v3.4b[1]\n" - "prfm pldl1keep, [x12, #0x80]\n" + "ldr x20, [x17, #0x98]\n" ".inst 0x6fa4e3ba // udot v26.4s, v29.16b, v4.4b[1]\n" - "ldr d29, [x16, #0x80]\n" + "ldr d29, [x17, #0x80]\n" ".inst 0x6fa0e38b // udot v11.4s, v28.16b, v0.4b[1]\n" - "mov v29.d[1], x21\n" + "prfm pldl1keep, [x11, #0x80]\n" ".inst 0x6fa1e38f // udot v15.4s, v28.16b, v1.4b[1]\n" - "ldr x21, [x16, #0xa8]\n" + "prfm pldl1keep, [x10, #0x80]\n" ".inst 0x6fa2e393 // udot v19.4s, v28.16b, v2.4b[1]\n" - "prfm pldl1keep, [x11, #0x80]\n" + "mov v29.d[1], x21\n" ".inst 0x6fa3e397 // udot v23.4s, v28.16b, v3.4b[1]\n" - "prfm pldl1keep, [x10, #0x80]\n" + "ldr x21, [x17, #0xa8]\n" ".inst 0x6fa4e39b // udot v27.4s, v28.16b, v4.4b[1]\n" - "ldr d28, [x16, #0x90]\n" - "mov v28.d[1], x20\n" + "ldr d28, [x17, #0x90]\n" ".inst 0x6f80eba8 // udot v8.4s, v29.16b, v0.4b[2]\n" + "prfm pldl1keep, [x9, #0x80]\n" ".inst 0x6f81ebac // udot v12.4s, v29.16b, v1.4b[2]\n" - "ldr x20, [x16, #0xb8]\n" + "mov v28.d[1], x20\n" ".inst 0x6f82ebb0 // udot v16.4s, v29.16b, v2.4b[2]\n" - "prfm pldl1keep, [x9, #0x80]\n" ".inst 0x6f83ebb4 // udot v20.4s, v29.16b, v3.4b[2]\n" + "ldr x20, [x17, #0xb8]\n" ".inst 0x6f84ebb8 // udot v24.4s, v29.16b, v4.4b[2]\n" - "ldr d29, [x16, #0xa0]\n" + "ldr d29, [x17, #0xa0]\n" ".inst 0x6f80eb89 // udot v9.4s, v28.16b, v0.4b[2]\n" - "mov v29.d[1], x21\n" ".inst 0x6f81eb8d // udot v13.4s, v28.16b, v1.4b[2]\n" - "ldr x21, [x16, #0xc8]\n" ".inst 0x6f82eb91 // udot v17.4s, v28.16b, v2.4b[2]\n" + "mov v29.d[1], x21\n" ".inst 0x6f83eb95 // udot v21.4s, v28.16b, v3.4b[2]\n" + "ldr x21, [x17, #0xc8]\n" ".inst 0x6f84eb99 // udot v25.4s, v28.16b, v4.4b[2]\n" - "ldr d28, [x16, #0xb0]\n" - "mov v28.d[1], x20\n" + "ldr d28, [x17, #0xb0]\n" ".inst 0x6f80ebaa // udot v10.4s, v29.16b, v0.4b[2]\n" ".inst 0x6f81ebae // udot v14.4s, v29.16b, v1.4b[2]\n" - "ldr x20, [x16, #0xd8]\n" + "mov v28.d[1], x20\n" ".inst 0x6f82ebb2 // udot v18.4s, v29.16b, v2.4b[2]\n" ".inst 0x6f83ebb6 // udot v22.4s, v29.16b, v3.4b[2]\n" + "ldr x20, [x17, #0xd8]\n" ".inst 0x6f84ebba // udot v26.4s, v29.16b, v4.4b[2]\n" - "ldr d29, [x16, #0xc0]\n" + "ldr d29, [x17, #0xc0]\n" ".inst 0x6f80eb8b // udot v11.4s, v28.16b, v0.4b[2]\n" - "mov v29.d[1], x21\n" ".inst 0x6f81eb8f // udot v15.4s, v28.16b, v1.4b[2]\n" - "ldr x21, [x16, #0xe8]\n" ".inst 0x6f82eb93 // udot v19.4s, v28.16b, v2.4b[2]\n" + "mov v29.d[1], x21\n" ".inst 0x6f83eb97 // udot v23.4s, v28.16b, v3.4b[2]\n" + "ldr x21, [x17, #0xe8]\n" ".inst 0x6f84eb9b // udot v27.4s, v28.16b, v4.4b[2]\n" - "ldr d28, [x16, #0xd0]\n" - "mov v28.d[1], x20\n" + "ldr d28, [x17, #0xd0]\n" ".inst 0x6fa0eba8 // udot v8.4s, v29.16b, v0.4b[3]\n" ".inst 0x6fa1ebac // udot v12.4s, v29.16b, v1.4b[3]\n" - "ldr x20, [x16, #0xf8]\n" + "mov v28.d[1], x20\n" ".inst 0x6fa2ebb0 // udot v16.4s, v29.16b, v2.4b[3]\n" ".inst 0x6fa3ebb4 // udot v20.4s, v29.16b, v3.4b[3]\n" + "ldr x20, [x17, #0xf8]\n" ".inst 0x6fa4ebb8 // udot v24.4s, v29.16b, v4.4b[3]\n" - "ldr d29, [x16, #0xe0]\n" + "ldr d29, [x17, #0xe0]\n" ".inst 0x6fa0eb89 // udot v9.4s, v28.16b, v0.4b[3]\n" - "mov v29.d[1], x21\n" ".inst 0x6fa1eb8d // udot v13.4s, v28.16b, v1.4b[3]\n" ".inst 0x6fa2eb91 // udot v17.4s, v28.16b, v2.4b[3]\n" + "mov v29.d[1], x21\n" ".inst 0x6fa3eb95 // udot v21.4s, v28.16b, v3.4b[3]\n" ".inst 0x6fa4eb99 // udot v25.4s, v28.16b, v4.4b[3]\n" - "ldr d28, [x16, #0xf0]\n" - "mov v28.d[1], x20\n" - "add x16, x16, #0x100\n" + "ldr d28, [x17, #0xf0]\n" + "add x17, x17, #0x100\n" ".inst 0x6fa0ebaa // udot v10.4s, v29.16b, v0.4b[3]\n" - "ldr x21, [x16, #0x8]\n" ".inst 0x6fa1ebae // udot v14.4s, v29.16b, v1.4b[3]\n" - "ldr x20, [x16, #0x18]\n" + "ldr x21, [x17, #0x8]\n" + "mov v28.d[1], x20\n" ".inst 0x6fa2ebb2 // udot v18.4s, v29.16b, v2.4b[3]\n" ".inst 0x6fa3ebb6 // udot v22.4s, v29.16b, v3.4b[3]\n" + "ldr x20, [x17, #0x18]\n" ".inst 0x6fa4ebba // udot v26.4s, v29.16b, v4.4b[3]\n" - "ldr d6, [x16, #0x0]\n" + "ldr d6, [x17, #0x0]\n" ".inst 0x6fa0eb8b // udot v11.4s, v28.16b, v0.4b[3]\n" "ldr d0, [x13, #0x0]\n" ".inst 0x6fa1eb8f // udot v15.4s, v28.16b, v1.4b[3]\n" @@ -2299,7 +2299,7 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "ldr d3, [x10, #0x0]\n" ".inst 0x6fa4eb9b // udot v27.4s, v28.16b, v4.4b[3]\n" "ldr d4, [x9, #0x0]\n" - "ldr d7, [x16, #0x10]\n" + "ldr d7, [x17, #0x10]\n" "mov v6.d[1], x21\n" "mov v0.d[1], x26\n" "mov v1.d[1], x25\n" @@ -2318,7 +2318,7 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( ".inst 0x6f83e0d4 // udot v20.4s, v6.16b, v3.4b[0]\n" "add x10, x10, #0x10\n" ".inst 0x6f84e0d8 // udot v24.4s, v6.16b, v4.4b[0]\n" - "ldr q29, [x16, #0x20]\n" + "ldr q29, [x17, #0x20]\n" ".inst 0x6f80e0e9 // udot v9.4s, v7.16b, v0.4b[0]\n" "add x9, x9, #0x10\n" ".inst 0x6f81e0ed // udot v13.4s, v7.16b, v1.4b[0]\n" @@ -2328,7 +2328,7 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( ".inst 0x6f83e0f5 // udot v21.4s, v7.16b, v3.4b[0]\n" "prfm pldl1keep, [x12, #0x80]\n" ".inst 0x6f84e0f9 // udot v25.4s, v7.16b, v4.4b[0]\n" - "ldr q28, [x16, #0x30]\n" + "ldr q28, [x17, #0x30]\n" ".inst 0x6f80e3aa // udot v10.4s, v29.16b, v0.4b[0]\n" "prfm pldl1keep, [x11, #0x80]\n" ".inst 0x6f81e3ae // udot v14.4s, v29.16b, v1.4b[0]\n" @@ -2337,75 +2337,75 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "prfm pldl1keep, [x9, #0x80]\n" ".inst 0x6f83e3b6 // udot v22.4s, v29.16b, v3.4b[0]\n" ".inst 0x6f84e3ba // udot v26.4s, v29.16b, v4.4b[0]\n" - "ldr q29, [x16, #0x40]\n" + "ldr q29, [x17, #0x40]\n" ".inst 0x6f80e38b // udot v11.4s, v28.16b, v0.4b[0]\n" ".inst 0x6f81e38f // udot v15.4s, v28.16b, v1.4b[0]\n" ".inst 0x6f82e393 // udot v19.4s, v28.16b, v2.4b[0]\n" ".inst 0x6f83e397 // udot v23.4s, v28.16b, v3.4b[0]\n" ".inst 0x6f84e39b // udot v27.4s, v28.16b, v4.4b[0]\n" - "ldr q28, [x16, #0x50]\n" + "ldr q28, [x17, #0x50]\n" ".inst 0x6fa0e3a8 // udot v8.4s, v29.16b, v0.4b[1]\n" ".inst 0x6fa1e3ac // udot v12.4s, v29.16b, v1.4b[1]\n" ".inst 0x6fa2e3b0 // udot v16.4s, v29.16b, v2.4b[1]\n" ".inst 0x6fa3e3b4 // udot v20.4s, v29.16b, v3.4b[1]\n" ".inst 0x6fa4e3b8 // udot v24.4s, v29.16b, v4.4b[1]\n" - "ldr q29, [x16, #0x60]\n" + "ldr q29, [x17, #0x60]\n" ".inst 0x6fa0e389 // udot v9.4s, v28.16b, v0.4b[1]\n" ".inst 0x6fa1e38d // udot v13.4s, v28.16b, v1.4b[1]\n" ".inst 0x6fa2e391 // udot v17.4s, v28.16b, v2.4b[1]\n" ".inst 0x6fa3e395 // udot v21.4s, v28.16b, v3.4b[1]\n" ".inst 0x6fa4e399 // udot v25.4s, v28.16b, v4.4b[1]\n" - "ldr q28, [x16, #0x70]\n" + "ldr q28, [x17, #0x70]\n" ".inst 0x6fa0e3aa // udot v10.4s, v29.16b, v0.4b[1]\n" ".inst 0x6fa1e3ae // udot v14.4s, v29.16b, v1.4b[1]\n" ".inst 0x6fa2e3b2 // udot v18.4s, v29.16b, v2.4b[1]\n" ".inst 0x6fa3e3b6 // udot v22.4s, v29.16b, v3.4b[1]\n" ".inst 0x6fa4e3ba // udot v26.4s, v29.16b, v4.4b[1]\n" - "ldr q29, [x16, #0x80]\n" + "ldr q29, [x17, #0x80]\n" ".inst 0x6fa0e38b // udot v11.4s, v28.16b, v0.4b[1]\n" ".inst 0x6fa1e38f // udot v15.4s, v28.16b, v1.4b[1]\n" ".inst 0x6fa2e393 // udot v19.4s, v28.16b, v2.4b[1]\n" ".inst 0x6fa3e397 // udot v23.4s, v28.16b, v3.4b[1]\n" ".inst 0x6fa4e39b // udot v27.4s, v28.16b, v4.4b[1]\n" - "ldr q28, [x16, #0x90]\n" + "ldr q28, [x17, #0x90]\n" ".inst 0x6f80eba8 // udot v8.4s, v29.16b, v0.4b[2]\n" ".inst 0x6f81ebac // udot v12.4s, v29.16b, v1.4b[2]\n" ".inst 0x6f82ebb0 // udot v16.4s, v29.16b, v2.4b[2]\n" ".inst 0x6f83ebb4 // udot v20.4s, v29.16b, v3.4b[2]\n" ".inst 0x6f84ebb8 // udot v24.4s, v29.16b, v4.4b[2]\n" - "ldr q29, [x16, #0xa0]\n" + "ldr q29, [x17, #0xa0]\n" ".inst 0x6f80eb89 // udot v9.4s, v28.16b, v0.4b[2]\n" ".inst 0x6f81eb8d // udot v13.4s, v28.16b, v1.4b[2]\n" ".inst 0x6f82eb91 // udot v17.4s, v28.16b, v2.4b[2]\n" ".inst 0x6f83eb95 // udot v21.4s, v28.16b, v3.4b[2]\n" ".inst 0x6f84eb99 // udot v25.4s, v28.16b, v4.4b[2]\n" - "ldr q28, [x16, #0xb0]\n" + "ldr q28, [x17, #0xb0]\n" ".inst 0x6f80ebaa // udot v10.4s, v29.16b, v0.4b[2]\n" ".inst 0x6f81ebae // udot v14.4s, v29.16b, v1.4b[2]\n" ".inst 0x6f82ebb2 // udot v18.4s, v29.16b, v2.4b[2]\n" ".inst 0x6f83ebb6 // udot v22.4s, v29.16b, v3.4b[2]\n" ".inst 0x6f84ebba // udot v26.4s, v29.16b, v4.4b[2]\n" - "ldr q29, [x16, #0xc0]\n" + "ldr q29, [x17, #0xc0]\n" ".inst 0x6f80eb8b // udot v11.4s, v28.16b, v0.4b[2]\n" ".inst 0x6f81eb8f // udot v15.4s, v28.16b, v1.4b[2]\n" ".inst 0x6f82eb93 // udot v19.4s, v28.16b, v2.4b[2]\n" ".inst 0x6f83eb97 // udot v23.4s, v28.16b, v3.4b[2]\n" ".inst 0x6f84eb9b // udot v27.4s, v28.16b, v4.4b[2]\n" - "ldr q28, [x16, #0xd0]\n" + "ldr q28, [x17, #0xd0]\n" ".inst 0x6fa0eba8 // udot v8.4s, v29.16b, v0.4b[3]\n" ".inst 0x6fa1ebac // udot v12.4s, v29.16b, v1.4b[3]\n" ".inst 0x6fa2ebb0 // udot v16.4s, v29.16b, v2.4b[3]\n" ".inst 0x6fa3ebb4 // udot v20.4s, v29.16b, v3.4b[3]\n" ".inst 0x6fa4ebb8 // udot v24.4s, v29.16b, v4.4b[3]\n" - "ldr q29, [x16, #0xe0]\n" + "ldr q29, [x17, #0xe0]\n" ".inst 0x6fa0eb89 // udot v9.4s, v28.16b, v0.4b[3]\n" ".inst 0x6fa1eb8d // udot v13.4s, v28.16b, v1.4b[3]\n" ".inst 0x6fa2eb91 // udot v17.4s, v28.16b, v2.4b[3]\n" ".inst 0x6fa3eb95 // udot v21.4s, v28.16b, v3.4b[3]\n" ".inst 0x6fa4eb99 // udot v25.4s, v28.16b, v4.4b[3]\n" - "ldr q28, [x16, #0xf0]\n" + "ldr q28, [x17, #0xf0]\n" ".inst 0x6fa0ebaa // udot v10.4s, v29.16b, v0.4b[3]\n" - "add x16, x16, #0x100\n" + "add x17, x17, #0x100\n" ".inst 0x6fa1ebae // udot v14.4s, v29.16b, v1.4b[3]\n" ".inst 0x6fa2ebb2 // udot v18.4s, v29.16b, v2.4b[3]\n" ".inst 0x6fa3ebb6 // udot v22.4s, v29.16b, v3.4b[3]\n" @@ -2427,22 +2427,22 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "ldr s0, [x11], #0x4\n" "ldr s31, [x10], #0x4\n" "ldr s30, [x9], #0x4\n" - "ldr q29, [x16, #0x0]\n" + "ldr q29, [x17, #0x0]\n" + "ldr q28, [x17, #0x10]\n" ".inst 0x6f82e3a8 // udot v8.4s, v29.16b, v2.4b[0]\n" - "ldr q28, [x16, #0x10]\n" ".inst 0x6f81e3ac // udot v12.4s, v29.16b, v1.4b[0]\n" ".inst 0x6f80e3b0 // udot v16.4s, v29.16b, v0.4b[0]\n" ".inst 0x6f9fe3b4 // udot v20.4s, v29.16b, v31.4b[0]\n" ".inst 0x6f9ee3b8 // udot v24.4s, v29.16b, v30.4b[0]\n" - "ldr q29, [x16, #0x20]\n" + "ldr q29, [x17, #0x20]\n" ".inst 0x6f82e389 // udot v9.4s, v28.16b, v2.4b[0]\n" ".inst 0x6f81e38d // udot v13.4s, v28.16b, v1.4b[0]\n" ".inst 0x6f80e391 // udot v17.4s, v28.16b, v0.4b[0]\n" ".inst 0x6f9fe395 // udot v21.4s, v28.16b, v31.4b[0]\n" ".inst 0x6f9ee399 // udot v25.4s, v28.16b, v30.4b[0]\n" - "ldr q28, [x16, #0x30]\n" + "ldr q28, [x17, #0x30]\n" ".inst 0x6f82e3aa // udot v10.4s, v29.16b, v2.4b[0]\n" - "add x16, x16, #0x40\n" + "add x17, x17, #0x40\n" ".inst 0x6f81e3ae // udot v14.4s, v29.16b, v1.4b[0]\n" ".inst 0x6f80e3b2 // udot v18.4s, v29.16b, v0.4b[0]\n" ".inst 0x6f9fe3b6 // udot v22.4s, v29.16b, v31.4b[0]\n" @@ -2475,22 +2475,22 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "ldr b3, [x10, #0x0]\n" "ldr b4, [x9, #0x0]\n" "159:" // Height 5: Multiply loop: Ragged operand read: Done - "ldr q29, [x16, #0x0]\n" + "ldr q29, [x17, #0x0]\n" + "ldr q28, [x17, #0x10]\n" ".inst 0x6f80e3a8 // udot v8.4s, v29.16b, v0.4b[0]\n" - "ldr q28, [x16, #0x10]\n" ".inst 0x6f81e3ac // udot v12.4s, v29.16b, v1.4b[0]\n" ".inst 0x6f82e3b0 // udot v16.4s, v29.16b, v2.4b[0]\n" ".inst 0x6f83e3b4 // udot v20.4s, v29.16b, v3.4b[0]\n" ".inst 0x6f84e3b8 // udot v24.4s, v29.16b, v4.4b[0]\n" - "ldr q29, [x16, #0x20]\n" + "ldr q29, [x17, #0x20]\n" ".inst 0x6f80e389 // udot v9.4s, v28.16b, v0.4b[0]\n" ".inst 0x6f81e38d // udot v13.4s, v28.16b, v1.4b[0]\n" ".inst 0x6f82e391 // udot v17.4s, v28.16b, v2.4b[0]\n" ".inst 0x6f83e395 // udot v21.4s, v28.16b, v3.4b[0]\n" ".inst 0x6f84e399 // udot v25.4s, v28.16b, v4.4b[0]\n" - "ldr q28, [x16, #0x30]\n" + "ldr q28, [x17, #0x30]\n" ".inst 0x6f80e3aa // udot v10.4s, v29.16b, v0.4b[0]\n" - "add x16, x16, #0x40\n" + "add x17, x17, #0x40\n" ".inst 0x6f81e3ae // udot v14.4s, v29.16b, v1.4b[0]\n" ".inst 0x6f82e3b2 // udot v18.4s, v29.16b, v2.4b[0]\n" ".inst 0x6f83e3b6 // udot v22.4s, v29.16b, v3.4b[0]\n" @@ -2506,20 +2506,20 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "cmp x15, x20\n" "bne 150b\n" "ldr x20, [%x[args_ptr], %[offsetof_output_offset]]\n" - "add x24, x17, x20, LSL #2\n" - "add x23, x24, x20, LSL #2\n" - "add x22, x23, x20, LSL #2\n" - "add x21, x22, x20, LSL #2\n" "cmp x8, #0x10\n" - "prfm pstl1keep, [x17, #0x0]\n" + "prfm pstl1keep, [x16, #0x0]\n" + "add x24, x16, x20, LSL #2\n" + "add x23, x24, x20, LSL #2\n" "prfm pstl1keep, [x24, #0x0]\n" + "add x22, x23, x20, LSL #2\n" "prfm pstl1keep, [x23, #0x0]\n" + "add x21, x22, x20, LSL #2\n" "prfm pstl1keep, [x22, #0x0]\n" "prfm pstl1keep, [x21, #0x0]\n" "bge 169f\n" "tbz x8, #3, 164f\n" - "st1 { v8.4s }, [x17], #0x10\n" - "st1 { v9.4s }, [x17], #0x10\n" + "st1 { v8.4s }, [x16], #0x10\n" + "st1 { v9.4s }, [x16], #0x10\n" "st1 { v12.4s }, [x24], #0x10\n" "st1 { v13.4s }, [x24], #0x10\n" "st1 { v16.4s }, [x23], #0x10\n" @@ -2529,19 +2529,19 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "st1 { v24.4s }, [x21], #0x10\n" "st1 { v25.4s }, [x21], #0x10\n" "tbz x8, #2, 162f\n" - "st1 { v10.4s }, [x17], #0x10\n" + "st1 { v10.4s }, [x16], #0x10\n" "st1 { v14.4s }, [x24], #0x10\n" "st1 { v18.4s }, [x23], #0x10\n" "st1 { v22.4s }, [x22], #0x10\n" "st1 { v26.4s }, [x21], #0x10\n" "tbz x8, #1, 161f\n" - "str d11, [x17], #0x8\n" + "str d11, [x16], #0x8\n" "str d15, [x24], #0x8\n" "str d19, [x23], #0x8\n" "str d23, [x22], #0x8\n" "str d27, [x21], #0x8\n" "tbz x8, #0, 168f\n" - "st1 { v11.s }[2], [x17]\n" + "st1 { v11.s }[2], [x16]\n" "st1 { v15.s }[2], [x24]\n" "st1 { v19.s }[2], [x23]\n" "st1 { v23.s }[2], [x22]\n" @@ -2549,7 +2549,7 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "b 168f\n" "161:" // Height 5: Partial direct writeback: partial_1_12 "tbz x8, #0, 168f\n" - "str s11, [x17, #0x0]\n" + "str s11, [x16, #0x0]\n" "str s15, [x24, #0x0]\n" "str s19, [x23, #0x0]\n" "str s23, [x22, #0x0]\n" @@ -2557,13 +2557,13 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "b 168f\n" "162:" // Height 5: Partial direct writeback: partial_2_8 "tbz x8, #1, 163f\n" - "str d10, [x17], #0x8\n" + "str d10, [x16], #0x8\n" "str d14, [x24], #0x8\n" "str d18, [x23], #0x8\n" "str d22, [x22], #0x8\n" "str d26, [x21], #0x8\n" "tbz x8, #0, 168f\n" - "st1 { v10.s }[2], [x17]\n" + "st1 { v10.s }[2], [x16]\n" "st1 { v14.s }[2], [x24]\n" "st1 { v18.s }[2], [x23]\n" "st1 { v22.s }[2], [x22]\n" @@ -2571,7 +2571,7 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "b 168f\n" "163:" // Height 5: Partial direct writeback: partial_1_8 "tbz x8, #0, 168f\n" - "str s10, [x17, #0x0]\n" + "str s10, [x16, #0x0]\n" "str s14, [x24, #0x0]\n" "str s18, [x23, #0x0]\n" "str s22, [x22, #0x0]\n" @@ -2579,19 +2579,19 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "b 168f\n" "164:" // Height 5: Partial direct writeback: partial_4_0 "tbz x8, #2, 166f\n" - "st1 { v8.4s }, [x17], #0x10\n" + "st1 { v8.4s }, [x16], #0x10\n" "st1 { v12.4s }, [x24], #0x10\n" "st1 { v16.4s }, [x23], #0x10\n" "st1 { v20.4s }, [x22], #0x10\n" "st1 { v24.4s }, [x21], #0x10\n" "tbz x8, #1, 165f\n" - "str d9, [x17], #0x8\n" + "str d9, [x16], #0x8\n" "str d13, [x24], #0x8\n" "str d17, [x23], #0x8\n" "str d21, [x22], #0x8\n" "str d25, [x21], #0x8\n" "tbz x8, #0, 168f\n" - "st1 { v9.s }[2], [x17]\n" + "st1 { v9.s }[2], [x16]\n" "st1 { v13.s }[2], [x24]\n" "st1 { v17.s }[2], [x23]\n" "st1 { v21.s }[2], [x22]\n" @@ -2599,7 +2599,7 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "b 168f\n" "165:" // Height 5: Partial direct writeback: partial_1_4 "tbz x8, #0, 168f\n" - "str s9, [x17, #0x0]\n" + "str s9, [x16, #0x0]\n" "str s13, [x24, #0x0]\n" "str s17, [x23, #0x0]\n" "str s21, [x22, #0x0]\n" @@ -2607,20 +2607,20 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "b 168f\n" "166:" // Height 5: Partial direct writeback: partial_2_0 "tbz x8, #1, 167f\n" - "str d8, [x17], #0x8\n" + "str d8, [x16], #0x8\n" "str d12, [x24], #0x8\n" "str d16, [x23], #0x8\n" "str d20, [x22], #0x8\n" "str d24, [x21], #0x8\n" "tbz x8, #0, 168f\n" - "st1 { v8.s }[2], [x17]\n" + "st1 { v8.s }[2], [x16]\n" "st1 { v12.s }[2], [x24]\n" "st1 { v16.s }[2], [x23]\n" "st1 { v20.s }[2], [x22]\n" "st1 { v24.s }[2], [x21]\n" "b 168f\n" "167:" // Height 5: Partial direct writeback: partial_1_0 - "str s8, [x17, #0x0]\n" + "str s8, [x16, #0x0]\n" "str s12, [x24, #0x0]\n" "str s16, [x23, #0x0]\n" "str s20, [x22, #0x0]\n" @@ -2628,11 +2628,11 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "168:" // Height 5: Partial direct writeback: Done "b 170f\n" "169:" // Height 5: Full writeback - "str q8, [x17, #0x0]\n" - "str q9, [x17, #0x10]\n" - "str q10, [x17, #0x20]\n" - "str q11, [x17, #0x30]\n" - "add x17, x17, #0x40\n" + "str q8, [x16, #0x0]\n" + "str q9, [x16, #0x10]\n" + "str q10, [x16, #0x20]\n" + "str q11, [x16, #0x30]\n" + "add x16, x16, #0x40\n" "str q12, [x24, #0x0]\n" "str q13, [x24, #0x10]\n" "str q14, [x24, #0x20]\n" @@ -2656,42 +2656,43 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "171:" // Height 6 "ldr x21, [%x[args_ptr], %[offsetof_output_offset]]\n" "mov x20, #0x18\n" + "ldr x16, [%x[args_ptr], %[offsetof_output_ptr]]\n" "ldr x8, [%x[args_ptr], %[offsetof_N]]\n" - "mov x17, %x[output_ptr]\n" - "ldr x16, [%x[args_ptr], %[offsetof_B_ptr]]\n" - "madd %x[output_ptr], x21, x20, %x[output_ptr]\n" + "ldr x17, [%x[args_ptr], %[offsetof_B_ptr]]\n" + "madd x20, x21, x20, x16\n" + "str x20, [%x[args_ptr], %[offsetof_output_ptr]]\n" "172:" // Height 6: Column loop "tbz %x[flags], #0, 182f\n" "ldr x20, [%x[args_ptr], %[offsetof_output_offset]]\n" - "add x24, x17, x20, LSL #2\n" + "cmp x8, #0x10\n" + "add x24, x16, x20, LSL #2\n" "add x23, x24, x20, LSL #2\n" "add x22, x23, x20, LSL #2\n" "add x21, x22, x20, LSL #2\n" - "cmp x8, #0x10\n" "add x20, x21, x20, LSL #2\n" "bge 181f\n" "tbz x8, #3, 176f\n" - "ld1 { v8.4s }, [x17], #0x10\n" + "ld1 { v8.4s }, [x16], #0x10\n" "ld1 { v12.4s }, [x24], #0x10\n" "ld1 { v16.4s }, [x23], #0x10\n" "ld1 { v20.4s }, [x22], #0x10\n" "ld1 { v24.4s }, [x21], #0x10\n" "ld1 { v28.4s }, [x20], #0x10\n" - "ld1 { v9.4s }, [x17], #0x10\n" + "ld1 { v9.4s }, [x16], #0x10\n" "ld1 { v13.4s }, [x24], #0x10\n" "ld1 { v17.4s }, [x23], #0x10\n" "ld1 { v21.4s }, [x22], #0x10\n" "ld1 { v25.4s }, [x21], #0x10\n" "ld1 { v29.4s }, [x20], #0x10\n" "tbz x8, #2, 174f\n" - "ld1 { v10.4s }, [x17], #0x10\n" + "ld1 { v10.4s }, [x16], #0x10\n" "ld1 { v14.4s }, [x24], #0x10\n" "ld1 { v18.4s }, [x23], #0x10\n" "ld1 { v22.4s }, [x22], #0x10\n" "ld1 { v26.4s }, [x21], #0x10\n" "ld1 { v30.4s }, [x20], #0x10\n" "tbz x8, #1, 173f\n" - "ldr d11, [x17], #0x8\n" + "ldr d11, [x16], #0x8\n" "mov x25, #0x38\n" "ldr d15, [x24], #0x8\n" "ldr d19, [x23], #0x8\n" @@ -2699,7 +2700,7 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "ldr d27, [x21], #0x8\n" "ldr d31, [x20], #0x8\n" "tbz x8, #0, 180f\n" - "ld1 { v11.s }[2], [x17]\n" + "ld1 { v11.s }[2], [x16]\n" "ld1 { v15.s }[2], [x24]\n" "ld1 { v19.s }[2], [x23]\n" "ld1 { v23.s }[2], [x22]\n" @@ -2709,7 +2710,7 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "173:" // Height 6: Partial accumulate: partial_1_12 "mov x25, #0x30\n" "tbz x8, #0, 180f\n" - "ldr s11, [x17, #0x0]\n" + "ldr s11, [x16, #0x0]\n" "ldr s15, [x24, #0x0]\n" "ldr s19, [x23, #0x0]\n" "ldr s23, [x22, #0x0]\n" @@ -2718,7 +2719,7 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "b 180f\n" "174:" // Height 6: Partial accumulate: partial_2_8 "tbz x8, #1, 175f\n" - "ldr d10, [x17], #0x8\n" + "ldr d10, [x16], #0x8\n" "mov x25, #0x28\n" "ldr d14, [x24], #0x8\n" "ldr d18, [x23], #0x8\n" @@ -2726,7 +2727,7 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "ldr d26, [x21], #0x8\n" "ldr d30, [x20], #0x8\n" "tbz x8, #0, 180f\n" - "ld1 { v10.s }[2], [x17]\n" + "ld1 { v10.s }[2], [x16]\n" "ld1 { v14.s }[2], [x24]\n" "ld1 { v18.s }[2], [x23]\n" "ld1 { v22.s }[2], [x22]\n" @@ -2736,7 +2737,7 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "175:" // Height 6: Partial accumulate: partial_1_8 "mov x25, #0x20\n" "tbz x8, #0, 180f\n" - "ldr s10, [x17, #0x0]\n" + "ldr s10, [x16, #0x0]\n" "ldr s14, [x24, #0x0]\n" "ldr s18, [x23, #0x0]\n" "ldr s22, [x22, #0x0]\n" @@ -2745,14 +2746,14 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "b 180f\n" "176:" // Height 6: Partial accumulate: partial_4_0 "tbz x8, #2, 178f\n" - "ld1 { v8.4s }, [x17], #0x10\n" + "ld1 { v8.4s }, [x16], #0x10\n" "ld1 { v12.4s }, [x24], #0x10\n" "ld1 { v16.4s }, [x23], #0x10\n" "ld1 { v20.4s }, [x22], #0x10\n" "ld1 { v24.4s }, [x21], #0x10\n" "ld1 { v28.4s }, [x20], #0x10\n" "tbz x8, #1, 177f\n" - "ldr d9, [x17], #0x8\n" + "ldr d9, [x16], #0x8\n" "mov x25, #0x18\n" "ldr d13, [x24], #0x8\n" "ldr d17, [x23], #0x8\n" @@ -2760,7 +2761,7 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "ldr d25, [x21], #0x8\n" "ldr d29, [x20], #0x8\n" "tbz x8, #0, 180f\n" - "ld1 { v9.s }[2], [x17]\n" + "ld1 { v9.s }[2], [x16]\n" "ld1 { v13.s }[2], [x24]\n" "ld1 { v17.s }[2], [x23]\n" "ld1 { v21.s }[2], [x22]\n" @@ -2770,7 +2771,7 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "177:" // Height 6: Partial accumulate: partial_1_4 "mov x25, #0x10\n" "tbz x8, #0, 180f\n" - "ldr s9, [x17, #0x0]\n" + "ldr s9, [x16, #0x0]\n" "ldr s13, [x24, #0x0]\n" "ldr s17, [x23, #0x0]\n" "ldr s21, [x22, #0x0]\n" @@ -2779,7 +2780,7 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "b 180f\n" "178:" // Height 6: Partial accumulate: partial_2_0 "tbz x8, #1, 179f\n" - "ldr d8, [x17], #0x8\n" + "ldr d8, [x16], #0x8\n" "mov x25, #0x8\n" "ldr d12, [x24], #0x8\n" "ldr d16, [x23], #0x8\n" @@ -2787,7 +2788,7 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "ldr d24, [x21], #0x8\n" "ldr d28, [x20], #0x8\n" "tbz x8, #0, 180f\n" - "ld1 { v8.s }[2], [x17]\n" + "ld1 { v8.s }[2], [x16]\n" "ld1 { v12.s }[2], [x24]\n" "ld1 { v16.s }[2], [x23]\n" "ld1 { v20.s }[2], [x22]\n" @@ -2795,7 +2796,7 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "ld1 { v28.s }[2], [x20]\n" "b 180f\n" "179:" // Height 6: Partial accumulate: partial_1_0 - "ldr s8, [x17, #0x0]\n" + "ldr s8, [x16, #0x0]\n" "mov x25, #0x0\n" "ldr s12, [x24, #0x0]\n" "ldr s16, [x23, #0x0]\n" @@ -2803,13 +2804,13 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "ldr s24, [x21, #0x0]\n" "ldr s28, [x20, #0x0]\n" "180:" // Height 6: Partial accumulate: Done - "sub x17, x17, x25\n" + "sub x16, x16, x25\n" "b 183f\n" "181:" // Height 6: full accumulate - "ldr q8, [x17, #0x0]\n" - "ldr q9, [x17, #0x10]\n" - "ldr q10, [x17, #0x20]\n" - "ldr q11, [x17, #0x30]\n" + "ldr q8, [x16, #0x0]\n" + "ldr q9, [x16, #0x10]\n" + "ldr q10, [x16, #0x20]\n" + "ldr q11, [x16, #0x30]\n" "ldr q12, [x24, #0x0]\n" "ldr q13, [x24, #0x10]\n" "ldr q14, [x24, #0x20]\n" @@ -2860,8 +2861,8 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "mov x15, #0x0\n" "184:" // Height 6: String loop "ldr x20, [%x[args_ptr], %[offsetof_string_lengths]]\n" - "ldr w14, [x20, x15, LSL #0x2]\n" "ldr x21, [%x[args_ptr], %[offsetof_input_offset]]\n" + "ldr w14, [x20, x15, LSL #0x2]\n" "tbz %x[flags], #3, 185f\n" "ldr x20, [%x[input_ptr], x15, LSL #0x3]\n" "add x20, x20, x21, LSL #3\n" @@ -2897,14 +2898,14 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "ldr q3, [x10, #0x0]\n" "ldr q4, [x9, #0x0]\n" "ldr q5, [x28, #0x0]\n" - "ldr q6, [x16, #0x0]\n" - "ldr q7, [x16, #0x10]\n" + "ldr q6, [x17, #0x0]\n" + "ldr q7, [x17, #0x10]\n" "blt 188f\n" "187:" // Height 6: Multiply loop: Main loop head ".inst 0x6f80e0c8 // udot v8.4s, v6.16b, v0.4b[0]\n" - "ldr x21, [x16, #0x28]\n" + "ldr x21, [x17, #0x28]\n" ".inst 0x6f81e0cc // udot v12.4s, v6.16b, v1.4b[0]\n" - "ldr x20, [x16, #0x38]\n" + "ldr x20, [x17, #0x38]\n" ".inst 0x6f82e0d0 // udot v16.4s, v6.16b, v2.4b[0]\n" "add x13, x13, #0x10\n" ".inst 0x6f83e0d4 // udot v20.4s, v6.16b, v3.4b[0]\n" @@ -2912,151 +2913,151 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( ".inst 0x6f84e0d8 // udot v24.4s, v6.16b, v4.4b[0]\n" "add x11, x11, #0x10\n" ".inst 0x6f85e0dc // udot v28.4s, v6.16b, v5.4b[0]\n" - "ldr d6, [x16, #0x20]\n" + "ldr d6, [x17, #0x20]\n" ".inst 0x6f80e0e9 // udot v9.4s, v7.16b, v0.4b[0]\n" - "mov v6.d[1], x21\n" + "add x10, x10, #0x10\n" ".inst 0x6f81e0ed // udot v13.4s, v7.16b, v1.4b[0]\n" - "ldr x21, [x16, #0x48]\n" + "add x9, x9, #0x10\n" ".inst 0x6f82e0f1 // udot v17.4s, v7.16b, v2.4b[0]\n" - "add x10, x10, #0x10\n" + "mov v6.d[1], x21\n" ".inst 0x6f83e0f5 // udot v21.4s, v7.16b, v3.4b[0]\n" - "add x9, x9, #0x10\n" + "ldr x21, [x17, #0x48]\n" ".inst 0x6f84e0f9 // udot v25.4s, v7.16b, v4.4b[0]\n" "add x28, x28, #0x10\n" ".inst 0x6f85e0fd // udot v29.4s, v7.16b, v5.4b[0]\n" - "ldr d7, [x16, #0x30]\n" - "mov v7.d[1], x20\n" + "ldr d7, [x17, #0x30]\n" ".inst 0x6f80e0ca // udot v10.4s, v6.16b, v0.4b[0]\n" + "ldr x27, [x13, #0x8]\n" ".inst 0x6f81e0ce // udot v14.4s, v6.16b, v1.4b[0]\n" - "ldr x20, [x16, #0x58]\n" + "ldr x26, [x12, #0x8]\n" + "mov v7.d[1], x20\n" ".inst 0x6f82e0d2 // udot v18.4s, v6.16b, v2.4b[0]\n" - "ldr x27, [x13, #0x8]\n" ".inst 0x6f83e0d6 // udot v22.4s, v6.16b, v3.4b[0]\n" - "ldr x26, [x12, #0x8]\n" + "ldr x20, [x17, #0x58]\n" ".inst 0x6f84e0da // udot v26.4s, v6.16b, v4.4b[0]\n" "ldr x25, [x11, #0x8]\n" ".inst 0x6f85e0de // udot v30.4s, v6.16b, v5.4b[0]\n" - "ldr d6, [x16, #0x40]\n" + "ldr d6, [x17, #0x40]\n" ".inst 0x6f80e0eb // udot v11.4s, v7.16b, v0.4b[0]\n" - "mov v6.d[1], x21\n" + "ldr x24, [x10, #0x8]\n" ".inst 0x6f81e0ef // udot v15.4s, v7.16b, v1.4b[0]\n" - "ldr x21, [x16, #0x68]\n" + "ldr x23, [x9, #0x8]\n" ".inst 0x6f82e0f3 // udot v19.4s, v7.16b, v2.4b[0]\n" - "ldr x24, [x10, #0x8]\n" + "mov v6.d[1], x21\n" ".inst 0x6f83e0f7 // udot v23.4s, v7.16b, v3.4b[0]\n" - "ldr x23, [x9, #0x8]\n" + "ldr x21, [x17, #0x68]\n" ".inst 0x6f84e0fb // udot v27.4s, v7.16b, v4.4b[0]\n" "ldr x22, [x28, #0x8]\n" ".inst 0x6f85e0ff // udot v31.4s, v7.16b, v5.4b[0]\n" - "ldr d7, [x16, #0x50]\n" - "mov v7.d[1], x20\n" + "ldr d7, [x17, #0x50]\n" ".inst 0x6fa0e0c8 // udot v8.4s, v6.16b, v0.4b[1]\n" + "sub x14, x14, #0x10\n" ".inst 0x6fa1e0cc // udot v12.4s, v6.16b, v1.4b[1]\n" - "ldr x20, [x16, #0x78]\n" + "cmp x14, #0x20\n" + "mov v7.d[1], x20\n" ".inst 0x6fa2e0d0 // udot v16.4s, v6.16b, v2.4b[1]\n" - "sub x14, x14, #0x10\n" ".inst 0x6fa3e0d4 // udot v20.4s, v6.16b, v3.4b[1]\n" - "cmp x14, #0x20\n" + "ldr x20, [x17, #0x78]\n" ".inst 0x6fa4e0d8 // udot v24.4s, v6.16b, v4.4b[1]\n" "prfm pldl1keep, [x13, #0x80]\n" ".inst 0x6fa5e0dc // udot v28.4s, v6.16b, v5.4b[1]\n" - "ldr d6, [x16, #0x60]\n" + "ldr d6, [x17, #0x60]\n" ".inst 0x6fa0e0e9 // udot v9.4s, v7.16b, v0.4b[1]\n" - "mov v6.d[1], x21\n" + "prfm pldl1keep, [x12, #0x80]\n" ".inst 0x6fa1e0ed // udot v13.4s, v7.16b, v1.4b[1]\n" - "ldr x21, [x16, #0x88]\n" + "prfm pldl1keep, [x11, #0x80]\n" ".inst 0x6fa2e0f1 // udot v17.4s, v7.16b, v2.4b[1]\n" - "prfm pldl1keep, [x12, #0x80]\n" + "mov v6.d[1], x21\n" ".inst 0x6fa3e0f5 // udot v21.4s, v7.16b, v3.4b[1]\n" - "prfm pldl1keep, [x11, #0x80]\n" + "ldr x21, [x17, #0x88]\n" ".inst 0x6fa4e0f9 // udot v25.4s, v7.16b, v4.4b[1]\n" "prfm pldl1keep, [x10, #0x80]\n" ".inst 0x6fa5e0fd // udot v29.4s, v7.16b, v5.4b[1]\n" - "ldr d7, [x16, #0x70]\n" - "mov v7.d[1], x20\n" + "ldr d7, [x17, #0x70]\n" ".inst 0x6fa0e0ca // udot v10.4s, v6.16b, v0.4b[1]\n" + "prfm pldl1keep, [x9, #0x80]\n" ".inst 0x6fa1e0ce // udot v14.4s, v6.16b, v1.4b[1]\n" - "ldr x20, [x16, #0x98]\n" + "prfm pldl1keep, [x28, #0x80]\n" + "mov v7.d[1], x20\n" ".inst 0x6fa2e0d2 // udot v18.4s, v6.16b, v2.4b[1]\n" - "prfm pldl1keep, [x9, #0x80]\n" ".inst 0x6fa3e0d6 // udot v22.4s, v6.16b, v3.4b[1]\n" - "prfm pldl1keep, [x28, #0x80]\n" + "ldr x20, [x17, #0x98]\n" ".inst 0x6fa4e0da // udot v26.4s, v6.16b, v4.4b[1]\n" ".inst 0x6fa5e0de // udot v30.4s, v6.16b, v5.4b[1]\n" - "ldr d6, [x16, #0x80]\n" + "ldr d6, [x17, #0x80]\n" ".inst 0x6fa0e0eb // udot v11.4s, v7.16b, v0.4b[1]\n" - "mov v6.d[1], x21\n" ".inst 0x6fa1e0ef // udot v15.4s, v7.16b, v1.4b[1]\n" - "ldr x21, [x16, #0xa8]\n" ".inst 0x6fa2e0f3 // udot v19.4s, v7.16b, v2.4b[1]\n" + "mov v6.d[1], x21\n" ".inst 0x6fa3e0f7 // udot v23.4s, v7.16b, v3.4b[1]\n" + "ldr x21, [x17, #0xa8]\n" ".inst 0x6fa4e0fb // udot v27.4s, v7.16b, v4.4b[1]\n" ".inst 0x6fa5e0ff // udot v31.4s, v7.16b, v5.4b[1]\n" - "ldr d7, [x16, #0x90]\n" - "mov v7.d[1], x20\n" + "ldr d7, [x17, #0x90]\n" ".inst 0x6f80e8c8 // udot v8.4s, v6.16b, v0.4b[2]\n" ".inst 0x6f81e8cc // udot v12.4s, v6.16b, v1.4b[2]\n" - "ldr x20, [x16, #0xb8]\n" + "mov v7.d[1], x20\n" ".inst 0x6f82e8d0 // udot v16.4s, v6.16b, v2.4b[2]\n" ".inst 0x6f83e8d4 // udot v20.4s, v6.16b, v3.4b[2]\n" + "ldr x20, [x17, #0xb8]\n" ".inst 0x6f84e8d8 // udot v24.4s, v6.16b, v4.4b[2]\n" ".inst 0x6f85e8dc // udot v28.4s, v6.16b, v5.4b[2]\n" - "ldr d6, [x16, #0xa0]\n" + "ldr d6, [x17, #0xa0]\n" ".inst 0x6f80e8e9 // udot v9.4s, v7.16b, v0.4b[2]\n" - "mov v6.d[1], x21\n" ".inst 0x6f81e8ed // udot v13.4s, v7.16b, v1.4b[2]\n" - "ldr x21, [x16, #0xc8]\n" ".inst 0x6f82e8f1 // udot v17.4s, v7.16b, v2.4b[2]\n" + "mov v6.d[1], x21\n" ".inst 0x6f83e8f5 // udot v21.4s, v7.16b, v3.4b[2]\n" + "ldr x21, [x17, #0xc8]\n" ".inst 0x6f84e8f9 // udot v25.4s, v7.16b, v4.4b[2]\n" ".inst 0x6f85e8fd // udot v29.4s, v7.16b, v5.4b[2]\n" - "ldr d7, [x16, #0xb0]\n" - "mov v7.d[1], x20\n" + "ldr d7, [x17, #0xb0]\n" ".inst 0x6f80e8ca // udot v10.4s, v6.16b, v0.4b[2]\n" ".inst 0x6f81e8ce // udot v14.4s, v6.16b, v1.4b[2]\n" - "ldr x20, [x16, #0xd8]\n" + "mov v7.d[1], x20\n" ".inst 0x6f82e8d2 // udot v18.4s, v6.16b, v2.4b[2]\n" ".inst 0x6f83e8d6 // udot v22.4s, v6.16b, v3.4b[2]\n" + "ldr x20, [x17, #0xd8]\n" ".inst 0x6f84e8da // udot v26.4s, v6.16b, v4.4b[2]\n" ".inst 0x6f85e8de // udot v30.4s, v6.16b, v5.4b[2]\n" - "ldr d6, [x16, #0xc0]\n" + "ldr d6, [x17, #0xc0]\n" ".inst 0x6f80e8eb // udot v11.4s, v7.16b, v0.4b[2]\n" - "mov v6.d[1], x21\n" ".inst 0x6f81e8ef // udot v15.4s, v7.16b, v1.4b[2]\n" - "ldr x21, [x16, #0xe8]\n" ".inst 0x6f82e8f3 // udot v19.4s, v7.16b, v2.4b[2]\n" + "mov v6.d[1], x21\n" ".inst 0x6f83e8f7 // udot v23.4s, v7.16b, v3.4b[2]\n" + "ldr x21, [x17, #0xe8]\n" ".inst 0x6f84e8fb // udot v27.4s, v7.16b, v4.4b[2]\n" ".inst 0x6f85e8ff // udot v31.4s, v7.16b, v5.4b[2]\n" - "ldr d7, [x16, #0xd0]\n" - "mov v7.d[1], x20\n" + "ldr d7, [x17, #0xd0]\n" ".inst 0x6fa0e8c8 // udot v8.4s, v6.16b, v0.4b[3]\n" ".inst 0x6fa1e8cc // udot v12.4s, v6.16b, v1.4b[3]\n" - "ldr x20, [x16, #0xf8]\n" + "mov v7.d[1], x20\n" ".inst 0x6fa2e8d0 // udot v16.4s, v6.16b, v2.4b[3]\n" ".inst 0x6fa3e8d4 // udot v20.4s, v6.16b, v3.4b[3]\n" + "ldr x20, [x17, #0xf8]\n" ".inst 0x6fa4e8d8 // udot v24.4s, v6.16b, v4.4b[3]\n" ".inst 0x6fa5e8dc // udot v28.4s, v6.16b, v5.4b[3]\n" - "ldr d6, [x16, #0xe0]\n" + "ldr d6, [x17, #0xe0]\n" ".inst 0x6fa0e8e9 // udot v9.4s, v7.16b, v0.4b[3]\n" - "mov v6.d[1], x21\n" ".inst 0x6fa1e8ed // udot v13.4s, v7.16b, v1.4b[3]\n" ".inst 0x6fa2e8f1 // udot v17.4s, v7.16b, v2.4b[3]\n" + "mov v6.d[1], x21\n" ".inst 0x6fa3e8f5 // udot v21.4s, v7.16b, v3.4b[3]\n" ".inst 0x6fa4e8f9 // udot v25.4s, v7.16b, v4.4b[3]\n" ".inst 0x6fa5e8fd // udot v29.4s, v7.16b, v5.4b[3]\n" - "ldr d7, [x16, #0xf0]\n" - "mov v7.d[1], x20\n" - "add x16, x16, #0x100\n" + "ldr d7, [x17, #0xf0]\n" + "add x17, x17, #0x100\n" ".inst 0x6fa0e8ca // udot v10.4s, v6.16b, v0.4b[3]\n" - "ldr x21, [x16, #0x8]\n" ".inst 0x6fa1e8ce // udot v14.4s, v6.16b, v1.4b[3]\n" - "ldr x20, [x16, #0x18]\n" + "ldr x21, [x17, #0x8]\n" + "mov v7.d[1], x20\n" ".inst 0x6fa2e8d2 // udot v18.4s, v6.16b, v2.4b[3]\n" ".inst 0x6fa3e8d6 // udot v22.4s, v6.16b, v3.4b[3]\n" + "ldr x20, [x17, #0x18]\n" ".inst 0x6fa4e8da // udot v26.4s, v6.16b, v4.4b[3]\n" ".inst 0x6fa5e8de // udot v30.4s, v6.16b, v5.4b[3]\n" - "ldr d6, [x16, #0x0]\n" + "ldr d6, [x17, #0x0]\n" ".inst 0x6fa0e8eb // udot v11.4s, v7.16b, v0.4b[3]\n" "ldr d0, [x13, #0x0]\n" ".inst 0x6fa1e8ef // udot v15.4s, v7.16b, v1.4b[3]\n" @@ -3069,7 +3070,7 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "ldr d4, [x9, #0x0]\n" ".inst 0x6fa5e8ff // udot v31.4s, v7.16b, v5.4b[3]\n" "ldr d5, [x28, #0x0]\n" - "ldr d7, [x16, #0x10]\n" + "ldr d7, [x17, #0x10]\n" "mov v6.d[1], x21\n" "mov v0.d[1], x27\n" "mov v1.d[1], x26\n" @@ -3091,7 +3092,7 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( ".inst 0x6f84e0d8 // udot v24.4s, v6.16b, v4.4b[0]\n" "add x9, x9, #0x10\n" ".inst 0x6f85e0dc // udot v28.4s, v6.16b, v5.4b[0]\n" - "ldr q6, [x16, #0x20]\n" + "ldr q6, [x17, #0x20]\n" ".inst 0x6f80e0e9 // udot v9.4s, v7.16b, v0.4b[0]\n" "add x28, x28, #0x10\n" ".inst 0x6f81e0ed // udot v13.4s, v7.16b, v1.4b[0]\n" @@ -3103,7 +3104,7 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( ".inst 0x6f84e0f9 // udot v25.4s, v7.16b, v4.4b[0]\n" "prfm pldl1keep, [x11, #0x80]\n" ".inst 0x6f85e0fd // udot v29.4s, v7.16b, v5.4b[0]\n" - "ldr q7, [x16, #0x30]\n" + "ldr q7, [x17, #0x30]\n" ".inst 0x6f80e0ca // udot v10.4s, v6.16b, v0.4b[0]\n" "prfm pldl1keep, [x10, #0x80]\n" ".inst 0x6f81e0ce // udot v14.4s, v6.16b, v1.4b[0]\n" @@ -3113,86 +3114,86 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( ".inst 0x6f83e0d6 // udot v22.4s, v6.16b, v3.4b[0]\n" ".inst 0x6f84e0da // udot v26.4s, v6.16b, v4.4b[0]\n" ".inst 0x6f85e0de // udot v30.4s, v6.16b, v5.4b[0]\n" - "ldr q6, [x16, #0x40]\n" + "ldr q6, [x17, #0x40]\n" ".inst 0x6f80e0eb // udot v11.4s, v7.16b, v0.4b[0]\n" ".inst 0x6f81e0ef // udot v15.4s, v7.16b, v1.4b[0]\n" ".inst 0x6f82e0f3 // udot v19.4s, v7.16b, v2.4b[0]\n" ".inst 0x6f83e0f7 // udot v23.4s, v7.16b, v3.4b[0]\n" ".inst 0x6f84e0fb // udot v27.4s, v7.16b, v4.4b[0]\n" ".inst 0x6f85e0ff // udot v31.4s, v7.16b, v5.4b[0]\n" - "ldr q7, [x16, #0x50]\n" + "ldr q7, [x17, #0x50]\n" ".inst 0x6fa0e0c8 // udot v8.4s, v6.16b, v0.4b[1]\n" ".inst 0x6fa1e0cc // udot v12.4s, v6.16b, v1.4b[1]\n" ".inst 0x6fa2e0d0 // udot v16.4s, v6.16b, v2.4b[1]\n" ".inst 0x6fa3e0d4 // udot v20.4s, v6.16b, v3.4b[1]\n" ".inst 0x6fa4e0d8 // udot v24.4s, v6.16b, v4.4b[1]\n" ".inst 0x6fa5e0dc // udot v28.4s, v6.16b, v5.4b[1]\n" - "ldr q6, [x16, #0x60]\n" + "ldr q6, [x17, #0x60]\n" ".inst 0x6fa0e0e9 // udot v9.4s, v7.16b, v0.4b[1]\n" ".inst 0x6fa1e0ed // udot v13.4s, v7.16b, v1.4b[1]\n" ".inst 0x6fa2e0f1 // udot v17.4s, v7.16b, v2.4b[1]\n" ".inst 0x6fa3e0f5 // udot v21.4s, v7.16b, v3.4b[1]\n" ".inst 0x6fa4e0f9 // udot v25.4s, v7.16b, v4.4b[1]\n" ".inst 0x6fa5e0fd // udot v29.4s, v7.16b, v5.4b[1]\n" - "ldr q7, [x16, #0x70]\n" + "ldr q7, [x17, #0x70]\n" ".inst 0x6fa0e0ca // udot v10.4s, v6.16b, v0.4b[1]\n" ".inst 0x6fa1e0ce // udot v14.4s, v6.16b, v1.4b[1]\n" ".inst 0x6fa2e0d2 // udot v18.4s, v6.16b, v2.4b[1]\n" ".inst 0x6fa3e0d6 // udot v22.4s, v6.16b, v3.4b[1]\n" ".inst 0x6fa4e0da // udot v26.4s, v6.16b, v4.4b[1]\n" ".inst 0x6fa5e0de // udot v30.4s, v6.16b, v5.4b[1]\n" - "ldr q6, [x16, #0x80]\n" + "ldr q6, [x17, #0x80]\n" ".inst 0x6fa0e0eb // udot v11.4s, v7.16b, v0.4b[1]\n" ".inst 0x6fa1e0ef // udot v15.4s, v7.16b, v1.4b[1]\n" ".inst 0x6fa2e0f3 // udot v19.4s, v7.16b, v2.4b[1]\n" ".inst 0x6fa3e0f7 // udot v23.4s, v7.16b, v3.4b[1]\n" ".inst 0x6fa4e0fb // udot v27.4s, v7.16b, v4.4b[1]\n" ".inst 0x6fa5e0ff // udot v31.4s, v7.16b, v5.4b[1]\n" - "ldr q7, [x16, #0x90]\n" + "ldr q7, [x17, #0x90]\n" ".inst 0x6f80e8c8 // udot v8.4s, v6.16b, v0.4b[2]\n" ".inst 0x6f81e8cc // udot v12.4s, v6.16b, v1.4b[2]\n" ".inst 0x6f82e8d0 // udot v16.4s, v6.16b, v2.4b[2]\n" ".inst 0x6f83e8d4 // udot v20.4s, v6.16b, v3.4b[2]\n" ".inst 0x6f84e8d8 // udot v24.4s, v6.16b, v4.4b[2]\n" ".inst 0x6f85e8dc // udot v28.4s, v6.16b, v5.4b[2]\n" - "ldr q6, [x16, #0xa0]\n" + "ldr q6, [x17, #0xa0]\n" ".inst 0x6f80e8e9 // udot v9.4s, v7.16b, v0.4b[2]\n" ".inst 0x6f81e8ed // udot v13.4s, v7.16b, v1.4b[2]\n" ".inst 0x6f82e8f1 // udot v17.4s, v7.16b, v2.4b[2]\n" ".inst 0x6f83e8f5 // udot v21.4s, v7.16b, v3.4b[2]\n" ".inst 0x6f84e8f9 // udot v25.4s, v7.16b, v4.4b[2]\n" ".inst 0x6f85e8fd // udot v29.4s, v7.16b, v5.4b[2]\n" - "ldr q7, [x16, #0xb0]\n" + "ldr q7, [x17, #0xb0]\n" ".inst 0x6f80e8ca // udot v10.4s, v6.16b, v0.4b[2]\n" ".inst 0x6f81e8ce // udot v14.4s, v6.16b, v1.4b[2]\n" ".inst 0x6f82e8d2 // udot v18.4s, v6.16b, v2.4b[2]\n" ".inst 0x6f83e8d6 // udot v22.4s, v6.16b, v3.4b[2]\n" ".inst 0x6f84e8da // udot v26.4s, v6.16b, v4.4b[2]\n" ".inst 0x6f85e8de // udot v30.4s, v6.16b, v5.4b[2]\n" - "ldr q6, [x16, #0xc0]\n" + "ldr q6, [x17, #0xc0]\n" ".inst 0x6f80e8eb // udot v11.4s, v7.16b, v0.4b[2]\n" ".inst 0x6f81e8ef // udot v15.4s, v7.16b, v1.4b[2]\n" ".inst 0x6f82e8f3 // udot v19.4s, v7.16b, v2.4b[2]\n" ".inst 0x6f83e8f7 // udot v23.4s, v7.16b, v3.4b[2]\n" ".inst 0x6f84e8fb // udot v27.4s, v7.16b, v4.4b[2]\n" ".inst 0x6f85e8ff // udot v31.4s, v7.16b, v5.4b[2]\n" - "ldr q7, [x16, #0xd0]\n" + "ldr q7, [x17, #0xd0]\n" ".inst 0x6fa0e8c8 // udot v8.4s, v6.16b, v0.4b[3]\n" ".inst 0x6fa1e8cc // udot v12.4s, v6.16b, v1.4b[3]\n" ".inst 0x6fa2e8d0 // udot v16.4s, v6.16b, v2.4b[3]\n" ".inst 0x6fa3e8d4 // udot v20.4s, v6.16b, v3.4b[3]\n" ".inst 0x6fa4e8d8 // udot v24.4s, v6.16b, v4.4b[3]\n" ".inst 0x6fa5e8dc // udot v28.4s, v6.16b, v5.4b[3]\n" - "ldr q6, [x16, #0xe0]\n" + "ldr q6, [x17, #0xe0]\n" ".inst 0x6fa0e8e9 // udot v9.4s, v7.16b, v0.4b[3]\n" ".inst 0x6fa1e8ed // udot v13.4s, v7.16b, v1.4b[3]\n" ".inst 0x6fa2e8f1 // udot v17.4s, v7.16b, v2.4b[3]\n" ".inst 0x6fa3e8f5 // udot v21.4s, v7.16b, v3.4b[3]\n" ".inst 0x6fa4e8f9 // udot v25.4s, v7.16b, v4.4b[3]\n" ".inst 0x6fa5e8fd // udot v29.4s, v7.16b, v5.4b[3]\n" - "ldr q7, [x16, #0xf0]\n" + "ldr q7, [x17, #0xf0]\n" ".inst 0x6fa0e8ca // udot v10.4s, v6.16b, v0.4b[3]\n" - "add x16, x16, #0x100\n" + "add x17, x17, #0x100\n" ".inst 0x6fa1e8ce // udot v14.4s, v6.16b, v1.4b[3]\n" ".inst 0x6fa2e8d2 // udot v18.4s, v6.16b, v2.4b[3]\n" ".inst 0x6fa3e8d6 // udot v22.4s, v6.16b, v3.4b[3]\n" @@ -3217,24 +3218,24 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "ldr s4, [x10], #0x4\n" "ldr s3, [x9], #0x4\n" "ldr s2, [x28], #0x4\n" - "ldr q1, [x16, #0x0]\n" + "ldr q1, [x17, #0x0]\n" + "ldr q0, [x17, #0x10]\n" ".inst 0x6f87e028 // udot v8.4s, v1.16b, v7.4b[0]\n" - "ldr q0, [x16, #0x10]\n" ".inst 0x6f86e02c // udot v12.4s, v1.16b, v6.4b[0]\n" ".inst 0x6f85e030 // udot v16.4s, v1.16b, v5.4b[0]\n" ".inst 0x6f84e034 // udot v20.4s, v1.16b, v4.4b[0]\n" ".inst 0x6f83e038 // udot v24.4s, v1.16b, v3.4b[0]\n" ".inst 0x6f82e03c // udot v28.4s, v1.16b, v2.4b[0]\n" - "ldr q1, [x16, #0x20]\n" + "ldr q1, [x17, #0x20]\n" ".inst 0x6f87e009 // udot v9.4s, v0.16b, v7.4b[0]\n" ".inst 0x6f86e00d // udot v13.4s, v0.16b, v6.4b[0]\n" ".inst 0x6f85e011 // udot v17.4s, v0.16b, v5.4b[0]\n" ".inst 0x6f84e015 // udot v21.4s, v0.16b, v4.4b[0]\n" ".inst 0x6f83e019 // udot v25.4s, v0.16b, v3.4b[0]\n" ".inst 0x6f82e01d // udot v29.4s, v0.16b, v2.4b[0]\n" - "ldr q0, [x16, #0x30]\n" + "ldr q0, [x17, #0x30]\n" ".inst 0x6f87e02a // udot v10.4s, v1.16b, v7.4b[0]\n" - "add x16, x16, #0x40\n" + "add x17, x17, #0x40\n" ".inst 0x6f86e02e // udot v14.4s, v1.16b, v6.4b[0]\n" ".inst 0x6f85e032 // udot v18.4s, v1.16b, v5.4b[0]\n" ".inst 0x6f84e036 // udot v22.4s, v1.16b, v4.4b[0]\n" @@ -3272,24 +3273,24 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "ldr b4, [x9, #0x0]\n" "ldr b5, [x28, #0x0]\n" "193:" // Height 6: Multiply loop: Ragged operand read: Done - "ldr q7, [x16, #0x0]\n" + "ldr q7, [x17, #0x0]\n" + "ldr q6, [x17, #0x10]\n" ".inst 0x6f80e0e8 // udot v8.4s, v7.16b, v0.4b[0]\n" - "ldr q6, [x16, #0x10]\n" ".inst 0x6f81e0ec // udot v12.4s, v7.16b, v1.4b[0]\n" ".inst 0x6f82e0f0 // udot v16.4s, v7.16b, v2.4b[0]\n" ".inst 0x6f83e0f4 // udot v20.4s, v7.16b, v3.4b[0]\n" ".inst 0x6f84e0f8 // udot v24.4s, v7.16b, v4.4b[0]\n" ".inst 0x6f85e0fc // udot v28.4s, v7.16b, v5.4b[0]\n" - "ldr q7, [x16, #0x20]\n" + "ldr q7, [x17, #0x20]\n" ".inst 0x6f80e0c9 // udot v9.4s, v6.16b, v0.4b[0]\n" ".inst 0x6f81e0cd // udot v13.4s, v6.16b, v1.4b[0]\n" ".inst 0x6f82e0d1 // udot v17.4s, v6.16b, v2.4b[0]\n" ".inst 0x6f83e0d5 // udot v21.4s, v6.16b, v3.4b[0]\n" ".inst 0x6f84e0d9 // udot v25.4s, v6.16b, v4.4b[0]\n" ".inst 0x6f85e0dd // udot v29.4s, v6.16b, v5.4b[0]\n" - "ldr q6, [x16, #0x30]\n" + "ldr q6, [x17, #0x30]\n" ".inst 0x6f80e0ea // udot v10.4s, v7.16b, v0.4b[0]\n" - "add x16, x16, #0x40\n" + "add x17, x17, #0x40\n" ".inst 0x6f81e0ee // udot v14.4s, v7.16b, v1.4b[0]\n" ".inst 0x6f82e0f2 // udot v18.4s, v7.16b, v2.4b[0]\n" ".inst 0x6f83e0f6 // udot v22.4s, v7.16b, v3.4b[0]\n" @@ -3307,22 +3308,22 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "cmp x15, x20\n" "bne 184b\n" "ldr x20, [%x[args_ptr], %[offsetof_output_offset]]\n" - "add x24, x17, x20, LSL #2\n" - "add x23, x24, x20, LSL #2\n" - "add x22, x23, x20, LSL #2\n" - "add x21, x22, x20, LSL #2\n" - "add x20, x21, x20, LSL #2\n" "cmp x8, #0x10\n" - "prfm pstl1keep, [x17, #0x0]\n" + "prfm pstl1keep, [x16, #0x0]\n" + "add x24, x16, x20, LSL #2\n" + "add x23, x24, x20, LSL #2\n" "prfm pstl1keep, [x24, #0x0]\n" + "add x22, x23, x20, LSL #2\n" "prfm pstl1keep, [x23, #0x0]\n" + "add x21, x22, x20, LSL #2\n" "prfm pstl1keep, [x22, #0x0]\n" + "add x20, x21, x20, LSL #2\n" "prfm pstl1keep, [x21, #0x0]\n" "prfm pstl1keep, [x20, #0x0]\n" "bge 203f\n" "tbz x8, #3, 198f\n" - "st1 { v8.4s }, [x17], #0x10\n" - "st1 { v9.4s }, [x17], #0x10\n" + "st1 { v8.4s }, [x16], #0x10\n" + "st1 { v9.4s }, [x16], #0x10\n" "st1 { v12.4s }, [x24], #0x10\n" "st1 { v13.4s }, [x24], #0x10\n" "st1 { v16.4s }, [x23], #0x10\n" @@ -3334,21 +3335,21 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "st1 { v28.4s }, [x20], #0x10\n" "st1 { v29.4s }, [x20], #0x10\n" "tbz x8, #2, 196f\n" - "st1 { v10.4s }, [x17], #0x10\n" + "st1 { v10.4s }, [x16], #0x10\n" "st1 { v14.4s }, [x24], #0x10\n" "st1 { v18.4s }, [x23], #0x10\n" "st1 { v22.4s }, [x22], #0x10\n" "st1 { v26.4s }, [x21], #0x10\n" "st1 { v30.4s }, [x20], #0x10\n" "tbz x8, #1, 195f\n" - "str d11, [x17], #0x8\n" + "str d11, [x16], #0x8\n" "str d15, [x24], #0x8\n" "str d19, [x23], #0x8\n" "str d23, [x22], #0x8\n" "str d27, [x21], #0x8\n" "str d31, [x20], #0x8\n" "tbz x8, #0, 202f\n" - "st1 { v11.s }[2], [x17]\n" + "st1 { v11.s }[2], [x16]\n" "st1 { v15.s }[2], [x24]\n" "st1 { v19.s }[2], [x23]\n" "st1 { v23.s }[2], [x22]\n" @@ -3357,7 +3358,7 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "b 202f\n" "195:" // Height 6: Partial direct writeback: partial_1_12 "tbz x8, #0, 202f\n" - "str s11, [x17, #0x0]\n" + "str s11, [x16, #0x0]\n" "str s15, [x24, #0x0]\n" "str s19, [x23, #0x0]\n" "str s23, [x22, #0x0]\n" @@ -3366,14 +3367,14 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "b 202f\n" "196:" // Height 6: Partial direct writeback: partial_2_8 "tbz x8, #1, 197f\n" - "str d10, [x17], #0x8\n" + "str d10, [x16], #0x8\n" "str d14, [x24], #0x8\n" "str d18, [x23], #0x8\n" "str d22, [x22], #0x8\n" "str d26, [x21], #0x8\n" "str d30, [x20], #0x8\n" "tbz x8, #0, 202f\n" - "st1 { v10.s }[2], [x17]\n" + "st1 { v10.s }[2], [x16]\n" "st1 { v14.s }[2], [x24]\n" "st1 { v18.s }[2], [x23]\n" "st1 { v22.s }[2], [x22]\n" @@ -3382,7 +3383,7 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "b 202f\n" "197:" // Height 6: Partial direct writeback: partial_1_8 "tbz x8, #0, 202f\n" - "str s10, [x17, #0x0]\n" + "str s10, [x16, #0x0]\n" "str s14, [x24, #0x0]\n" "str s18, [x23, #0x0]\n" "str s22, [x22, #0x0]\n" @@ -3391,21 +3392,21 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "b 202f\n" "198:" // Height 6: Partial direct writeback: partial_4_0 "tbz x8, #2, 200f\n" - "st1 { v8.4s }, [x17], #0x10\n" + "st1 { v8.4s }, [x16], #0x10\n" "st1 { v12.4s }, [x24], #0x10\n" "st1 { v16.4s }, [x23], #0x10\n" "st1 { v20.4s }, [x22], #0x10\n" "st1 { v24.4s }, [x21], #0x10\n" "st1 { v28.4s }, [x20], #0x10\n" "tbz x8, #1, 199f\n" - "str d9, [x17], #0x8\n" + "str d9, [x16], #0x8\n" "str d13, [x24], #0x8\n" "str d17, [x23], #0x8\n" "str d21, [x22], #0x8\n" "str d25, [x21], #0x8\n" "str d29, [x20], #0x8\n" "tbz x8, #0, 202f\n" - "st1 { v9.s }[2], [x17]\n" + "st1 { v9.s }[2], [x16]\n" "st1 { v13.s }[2], [x24]\n" "st1 { v17.s }[2], [x23]\n" "st1 { v21.s }[2], [x22]\n" @@ -3414,7 +3415,7 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "b 202f\n" "199:" // Height 6: Partial direct writeback: partial_1_4 "tbz x8, #0, 202f\n" - "str s9, [x17, #0x0]\n" + "str s9, [x16, #0x0]\n" "str s13, [x24, #0x0]\n" "str s17, [x23, #0x0]\n" "str s21, [x22, #0x0]\n" @@ -3423,14 +3424,14 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "b 202f\n" "200:" // Height 6: Partial direct writeback: partial_2_0 "tbz x8, #1, 201f\n" - "str d8, [x17], #0x8\n" + "str d8, [x16], #0x8\n" "str d12, [x24], #0x8\n" "str d16, [x23], #0x8\n" "str d20, [x22], #0x8\n" "str d24, [x21], #0x8\n" "str d28, [x20], #0x8\n" "tbz x8, #0, 202f\n" - "st1 { v8.s }[2], [x17]\n" + "st1 { v8.s }[2], [x16]\n" "st1 { v12.s }[2], [x24]\n" "st1 { v16.s }[2], [x23]\n" "st1 { v20.s }[2], [x22]\n" @@ -3438,7 +3439,7 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "st1 { v28.s }[2], [x20]\n" "b 202f\n" "201:" // Height 6: Partial direct writeback: partial_1_0 - "str s8, [x17, #0x0]\n" + "str s8, [x16, #0x0]\n" "str s12, [x24, #0x0]\n" "str s16, [x23, #0x0]\n" "str s20, [x22, #0x0]\n" @@ -3447,11 +3448,11 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "202:" // Height 6: Partial direct writeback: Done "b 204f\n" "203:" // Height 6: Full writeback - "str q8, [x17, #0x0]\n" - "str q9, [x17, #0x10]\n" - "str q10, [x17, #0x20]\n" - "str q11, [x17, #0x30]\n" - "add x17, x17, #0x40\n" + "str q8, [x16, #0x0]\n" + "str q9, [x16, #0x10]\n" + "str q10, [x16, #0x20]\n" + "str q11, [x16, #0x30]\n" + "add x16, x16, #0x40\n" "str q12, [x24, #0x0]\n" "str q13, [x24, #0x10]\n" "str q14, [x24, #0x20]\n" @@ -3487,8 +3488,8 @@ void a64_hybrid_u8u32_dot_6x16_a55 ( "madd %x[input_ptr], x20, x21, %x[input_ptr]\n" "b 1b\n" "206:" // Exit - : [M] "+&r" (M), [input_ptr] "+&r" (input_ptr), [output_ptr] "+&r" (output_ptr) - : [args_ptr] "r" (&ka), [flags] "r" (flags), [offsetof_B_ptr] "I" (offsetof(KernelArgs, B_ptr)), [offsetof_N] "I" (offsetof(KernelArgs, N)), [offsetof_input_initial_col] "I" (offsetof(KernelArgs, input_initial_col)), [offsetof_input_offset] "I" (offsetof(KernelArgs, input_offset)), [offsetof_num_strings] "I" (offsetof(KernelArgs, num_strings)), [offsetof_output_offset] "I" (offsetof(KernelArgs, output_offset)), [offsetof_string_lengths] "I" (offsetof(KernelArgs, string_lengths)) + : [M] "+&r" (M), [input_ptr] "+&r" (input_ptr) + : [args_ptr] "r" (&ka), [flags] "r" (flags), [offsetof_B_ptr] "I" (offsetof(KernelArgs, B_ptr)), [offsetof_N] "I" (offsetof(KernelArgs, N)), [offsetof_input_initial_col] "I" (offsetof(KernelArgs, input_initial_col)), [offsetof_input_offset] "I" (offsetof(KernelArgs, input_offset)), [offsetof_num_strings] "I" (offsetof(KernelArgs, num_strings)), [offsetof_output_offset] "I" (offsetof(KernelArgs, output_offset)), [offsetof_output_ptr] "I" (offsetof(KernelArgs, output_ptr)), [offsetof_string_lengths] "I" (offsetof(KernelArgs, string_lengths)) : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v8", "v9", "v10", "v11", "v12", "v13", "v14", "v15", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "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" ); } |