diff options
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_fp16_mla_6x32/a55.cpp')
| -rw-r--r-- | src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_fp16_mla_6x32/a55.cpp | 3285 |
1 files changed, 1644 insertions, 1641 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_fp16_mla_6x32/a55.cpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_fp16_mla_6x32/a55.cpp index b049ed45f9..aae6322b59 100644 --- a/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_fp16_mla_6x32/a55.cpp +++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_fp16_mla_6x32/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 * @@ -47,18 +47,19 @@ void a64_hybrid_fp16_mla_6x32_a55 ( size_t output_offset = {}; size_t input_initial_col = {}; size_t input_offset = {}; + void *output_ptr = {}; + const __fp16 *bias = {}; } 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; } @@ -79,6 +80,7 @@ void a64_hybrid_fp16_mla_6x32_a55 ( ka.string_lengths = string_lengths; ka.N = N; ka.B_ptr = B_ptr; + ka.bias = bias; switch(act.type) { default: case Activation::Type::None: @@ -101,10 +103,10 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "cmp %x[M], #0x2\n" "bgt 99f\n" "beq 50f\n" - "mov x7, %x[bias]\n" + "ldr x7, [%x[args_ptr], %[offsetof_bias]]\n" "ldr x8, [%x[args_ptr], %[offsetof_N]]\n" "ldr x17, [%x[args_ptr], %[offsetof_B_ptr]]\n" - "mov x16, %x[output_ptr]\n" + "ldr x16, [%x[args_ptr], %[offsetof_output_ptr]]\n" "2:" // Height 1: Column loop "cbz x7, 3f\n" "ldr q8, [x7, #0x0]\n" @@ -243,8 +245,8 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "mov x15, #0x0\n" "23:" // 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, 24f\n" "ldr x20, [%x[input_ptr], x15, LSL #0x3]\n" "add x20, x20, x21, LSL #3\n" @@ -266,150 +268,153 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "26:" // Height 1: Multiply loop: Main loop head "fmla v8.8h, v6.8h, v0.h[0]\n" "ldr d17, [x17, #0x20]\n" - "ldr x20, [x17, #0x28]\n" + "ldr x21, [x17, #0x28]\n" "fmla v9.8h, v7.8h, v0.h[0]\n" "ldr d16, [x17, #0x30]\n" - "mov v17.d[1], x20\n" + "add x13, x13, #0x10\n" "ldr x20, [x17, #0x38]\n" + "sub x14, x14, #0x8\n" + "mov v17.d[1], x21\n" + "ldr x21, [x17, #0x48]\n" + "ldr x22, [x13, #0x8]\n" + "cmp x14, #0x10\n" "mov v16.d[1], x20\n" "fmla v10.8h, v17.8h, v0.h[0]\n" "ldr d17, [x17, #0x40]\n" - "ldr x20, [x17, #0x48]\n" "fmla v11.8h, v16.8h, v0.h[0]\n" "ldr d16, [x17, #0x50]\n" - "mov v17.d[1], x20\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" "fmla v8.8h, v17.8h, v0.h[1]\n" "ldr d17, [x17, #0x60]\n" - "ldr x20, [x17, #0x68]\n" "fmla v9.8h, v16.8h, v0.h[1]\n" "ldr d16, [x17, #0x70]\n" - "mov v17.d[1], x20\n" "ldr x20, [x17, #0x78]\n" + "mov v17.d[1], x21\n" + "ldr x21, [x17, #0x88]\n" "mov v16.d[1], x20\n" "fmla v10.8h, v17.8h, v0.h[1]\n" "ldr d17, [x17, #0x80]\n" - "ldr x20, [x17, #0x88]\n" "fmla v11.8h, v16.8h, v0.h[1]\n" "ldr d16, [x17, #0x90]\n" - "mov v17.d[1], x20\n" "ldr x20, [x17, #0x98]\n" + "mov v17.d[1], x21\n" + "ldr x21, [x17, #0xa8]\n" "mov v16.d[1], x20\n" "fmla v8.8h, v17.8h, v0.h[2]\n" "ldr d17, [x17, #0xa0]\n" - "ldr x20, [x17, #0xa8]\n" "fmla v9.8h, v16.8h, v0.h[2]\n" "ldr d16, [x17, #0xb0]\n" - "mov v17.d[1], x20\n" "ldr x20, [x17, #0xb8]\n" + "mov v17.d[1], x21\n" + "ldr x21, [x17, #0xc8]\n" "mov v16.d[1], x20\n" "fmla v10.8h, v17.8h, v0.h[2]\n" "ldr d17, [x17, #0xc0]\n" - "ldr x20, [x17, #0xc8]\n" "fmla v11.8h, v16.8h, v0.h[2]\n" "ldr d16, [x17, #0xd0]\n" - "mov v17.d[1], x20\n" "ldr x20, [x17, #0xd8]\n" + "mov v17.d[1], x21\n" + "ldr x21, [x17, #0xe8]\n" "mov v16.d[1], x20\n" "fmla v8.8h, v17.8h, v0.h[3]\n" "ldr d17, [x17, #0xe0]\n" - "ldr x20, [x17, #0xe8]\n" "fmla v9.8h, v16.8h, v0.h[3]\n" "ldr d16, [x17, #0xf0]\n" - "mov v17.d[1], x20\n" "ldr x20, [x17, #0xf8]\n" + "mov v17.d[1], x21\n" + "ldr x21, [x17, #0x108]\n" "mov v16.d[1], x20\n" "fmla v10.8h, v17.8h, v0.h[3]\n" "ldr d17, [x17, #0x100]\n" - "ldr x20, [x17, #0x108]\n" "fmla v11.8h, v16.8h, v0.h[3]\n" "ldr d16, [x17, #0x110]\n" - "mov v17.d[1], x20\n" "ldr x20, [x17, #0x118]\n" + "mov v17.d[1], x21\n" + "ldr x21, [x17, #0x128]\n" "mov v16.d[1], x20\n" "fmla v8.8h, v17.8h, v0.h[4]\n" "ldr d17, [x17, #0x120]\n" - "ldr x20, [x17, #0x128]\n" "fmla v9.8h, v16.8h, v0.h[4]\n" "ldr d16, [x17, #0x130]\n" - "mov v17.d[1], x20\n" "ldr x20, [x17, #0x138]\n" + "mov v17.d[1], x21\n" + "ldr x21, [x17, #0x148]\n" "mov v16.d[1], x20\n" "fmla v10.8h, v17.8h, v0.h[4]\n" "ldr d17, [x17, #0x140]\n" - "ldr x20, [x17, #0x148]\n" "fmla v11.8h, v16.8h, v0.h[4]\n" "ldr d16, [x17, #0x150]\n" - "mov v17.d[1], x20\n" "ldr x20, [x17, #0x158]\n" + "mov v17.d[1], x21\n" + "ldr x21, [x17, #0x168]\n" "mov v16.d[1], x20\n" "fmla v8.8h, v17.8h, v0.h[5]\n" "ldr d17, [x17, #0x160]\n" - "ldr x20, [x17, #0x168]\n" "fmla v9.8h, v16.8h, v0.h[5]\n" "ldr d16, [x17, #0x170]\n" - "mov v17.d[1], x20\n" "ldr x20, [x17, #0x178]\n" + "mov v17.d[1], x21\n" + "ldr x21, [x17, #0x188]\n" "mov v16.d[1], x20\n" "fmla v10.8h, v17.8h, v0.h[5]\n" "ldr d17, [x17, #0x180]\n" - "ldr x20, [x17, #0x188]\n" "fmla v11.8h, v16.8h, v0.h[5]\n" "ldr d16, [x17, #0x190]\n" - "mov v17.d[1], x20\n" "ldr x20, [x17, #0x198]\n" + "mov v17.d[1], x21\n" + "ldr x21, [x17, #0x1a8]\n" "mov v16.d[1], x20\n" "fmla v8.8h, v17.8h, v0.h[6]\n" "ldr d17, [x17, #0x1a0]\n" - "ldr x20, [x17, #0x1a8]\n" "fmla v9.8h, v16.8h, v0.h[6]\n" "ldr d16, [x17, #0x1b0]\n" - "mov v17.d[1], x20\n" "ldr x20, [x17, #0x1b8]\n" + "mov v17.d[1], x21\n" + "ldr x21, [x17, #0x1c8]\n" "mov v16.d[1], x20\n" "fmla v10.8h, v17.8h, v0.h[6]\n" "ldr d17, [x17, #0x1c0]\n" - "ldr x20, [x17, #0x1c8]\n" "fmla v11.8h, v16.8h, v0.h[6]\n" "ldr d16, [x17, #0x1d0]\n" - "mov v17.d[1], x20\n" "ldr x20, [x17, #0x1d8]\n" + "mov v17.d[1], x21\n" + "ldr x21, [x17, #0x1e8]\n" "mov v16.d[1], x20\n" "fmla v8.8h, v17.8h, v0.h[7]\n" "ldr d17, [x17, #0x1e0]\n" - "ldr x20, [x17, #0x1e8]\n" "fmla v9.8h, v16.8h, v0.h[7]\n" "ldr d16, [x17, #0x1f0]\n" - "mov v17.d[1], x20\n" "ldr x20, [x17, #0x1f8]\n" - "mov v16.d[1], x20\n" - "add x13, x13, #0x10\n" "add x17, x17, #0x200\n" + "mov v17.d[1], x21\n" + "ldr x21, [x17, #0x8]\n" + "mov v16.d[1], x20\n" "fmla v10.8h, v17.8h, v0.h[7]\n" "ldr d6, [x17, #0x0]\n" - "ldr x20, [x17, #0x8]\n" "fmla v11.8h, v16.8h, v0.h[7]\n" "ldr d0, [x13, #0x0]\n" - "sub x14, x14, #0x8\n" "ldr d7, [x17, #0x10]\n" - "cmp x14, #0x10\n" - "ldr x21, [x13, #0x8]\n" - "mov v6.d[1], x20\n" "ldr x20, [x17, #0x18]\n" - "mov v0.d[1], x21\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 26b\n" "27:" // Height 1: Multiply loop: Single iteration only "fmla v8.8h, v6.8h, v0.h[0]\n" "ldr q17, [x17, #0x20]\n" "fmla v9.8h, v7.8h, v0.h[0]\n" "ldr q16, [x17, #0x30]\n" + "add x13, x13, #0x10\n" + "sub x14, x14, #0x8\n" "fmla v10.8h, v17.8h, v0.h[0]\n" "ldr q17, [x17, #0x40]\n" "fmla v11.8h, v16.8h, v0.h[0]\n" "ldr q16, [x17, #0x50]\n" + "prfm pldl1keep, [x13, #0x80]\n" "fmla v8.8h, v17.8h, v0.h[1]\n" "ldr q17, [x17, #0x60]\n" "fmla v9.8h, v16.8h, v0.h[1]\n" @@ -462,26 +467,23 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "ldr q17, [x17, #0x1e0]\n" "fmla v9.8h, v16.8h, v0.h[7]\n" "ldr q16, [x17, #0x1f0]\n" - "add x13, x13, #0x10\n" - "sub x14, x14, #0x8\n" + "add x17, x17, #0x200\n" "fmla v10.8h, v17.8h, v0.h[7]\n" - "prfm pldl1keep, [x13, #0x80]\n" "fmla v11.8h, v16.8h, v0.h[7]\n" - "add x17, x17, #0x200\n" "28:" // Height 1: Multiply loop: Main loop skip "cbz x14, 30f\n" "29:" // Height 1: Multiply loop: Odd block loop "ldr h0, [x13], #0x2\n" "sub x14, x14, #0x1\n" - "ldr q16, [x17, #0x0]\n" - "fmla v8.8h, v16.8h, v0.h[0]\n" + "ldr q17, [x17, #0x0]\n" "ldr q16, [x17, #0x10]\n" + "fmla v8.8h, v17.8h, v0.h[0]\n" + "ldr q17, [x17, #0x20]\n" "fmla v9.8h, v16.8h, v0.h[0]\n" - "ldr q16, [x17, #0x20]\n" - "fmla v10.8h, v16.8h, v0.h[0]\n" "ldr q16, [x17, #0x30]\n" - "fmla v11.8h, v16.8h, v0.h[0]\n" "add x17, x17, #0x40\n" + "fmla v10.8h, v17.8h, v0.h[0]\n" + "fmla v11.8h, v16.8h, v0.h[0]\n" "cbnz x14, 29b\n" "30:" // Height 1: Multiply loop: No odd multiplies "ldr w20, [%x[args_ptr], %[offsetof_num_strings]]\n" @@ -490,14 +492,14 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "bne 23b\n" "prfm pstl1keep, [x16, #0x0]\n" "tbz %x[flags], #1, 31f\n" - "add x20, %x[args_ptr], %[offset_max]\n" - "ld1r { v16.8h }, [x20]\n" - "fmin v8.8h, v8.8h, v16.8h\n" - "fmin v9.8h, v9.8h, v16.8h\n" - "fmin v10.8h, v10.8h, v16.8h\n" - "fmin v11.8h, v11.8h, v16.8h\n" + "add x21, %x[args_ptr], %[offset_max]\n" "add x20, %x[args_ptr], %[offset_min]\n" + "ld1r { v17.8h }, [x21]\n" "ld1r { v16.8h }, [x20]\n" + "fmin v8.8h, v8.8h, v17.8h\n" + "fmin v9.8h, v9.8h, v17.8h\n" + "fmin v10.8h, v10.8h, v17.8h\n" + "fmin v11.8h, v11.8h, v17.8h\n" "fmax v8.8h, v8.8h, v16.8h\n" "fmax v9.8h, v9.8h, v16.8h\n" "fmax v10.8h, v10.8h, v16.8h\n" @@ -610,168 +612,168 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "bgt 2b\n" "b 296f\n" "50:" // Height 2 - "mov x7, %x[bias]\n" + "ldr x7, [%x[args_ptr], %[offsetof_bias]]\n" "ldr x8, [%x[args_ptr], %[offsetof_N]]\n" "ldr x17, [%x[args_ptr], %[offsetof_B_ptr]]\n" - "mov x16, %x[output_ptr]\n" + "ldr x16, [%x[args_ptr], %[offsetof_output_ptr]]\n" "51:" // Height 2: Column loop "cbz x7, 52f\n" "ldr q8, [x7, #0x0]\n" - "mov v12.16b, v8.16b\n" "ldr q9, [x7, #0x10]\n" - "mov v13.16b, v9.16b\n" "ldr q10, [x7, #0x20]\n" - "mov v14.16b, v10.16b\n" "ldr q11, [x7, #0x30]\n" - "mov v15.16b, v11.16b\n" + "mov v12.16b, v8.16b\n" + "mov v13.16b, v9.16b\n" "add x7, x7, #0x40\n" + "mov v14.16b, v10.16b\n" + "mov v15.16b, v11.16b\n" "b 71f\n" "52:" // Height 2: no bias "tbz %x[flags], #0, 70f\n" "ldr x20, [%x[args_ptr], %[offsetof_output_offset]]\n" "cmp x8, #0x20\n" - "add x25, x16, x20, LSL #1\n" + "add x26, x16, x20, LSL #1\n" "bge 69f\n" "tbz x8, #4, 60f\n" "ld1 { v8.8h }, [x16], #0x10\n" - "ld1 { v12.8h }, [x25], #0x10\n" + "ld1 { v12.8h }, [x26], #0x10\n" "ld1 { v9.8h }, [x16], #0x10\n" - "ld1 { v13.8h }, [x25], #0x10\n" + "ld1 { v13.8h }, [x26], #0x10\n" "tbz x8, #3, 56f\n" "ld1 { v10.8h }, [x16], #0x10\n" - "ld1 { v14.8h }, [x25], #0x10\n" + "ld1 { v14.8h }, [x26], #0x10\n" "tbz x8, #2, 54f\n" "ldr d11, [x16], #0x8\n" - "ldr d15, [x25], #0x8\n" + "ldr d15, [x26], #0x8\n" "tbz x8, #1, 53f\n" "ld1 { v11.s }[2], [x16], #0x4\n" "mov x20, #0x3c\n" - "ld1 { v15.s }[2], [x25], #0x4\n" + "ld1 { v15.s }[2], [x26], #0x4\n" "tbz x8, #0, 68f\n" "ld1 { v11.h }[6], [x16]\n" - "ld1 { v15.h }[6], [x25]\n" + "ld1 { v15.h }[6], [x26]\n" "b 68f\n" "53:" // Height 2: Partial accumulate: partial_1_28 "mov x20, #0x38\n" "tbz x8, #0, 68f\n" "ld1 { v11.h }[4], [x16]\n" - "ld1 { v15.h }[4], [x25]\n" + "ld1 { v15.h }[4], [x26]\n" "b 68f\n" "54:" // Height 2: Partial accumulate: partial_2_24 "tbz x8, #1, 55f\n" "ldr s11, [x16], #0x4\n" "mov x20, #0x34\n" - "ldr s15, [x25], #0x4\n" + "ldr s15, [x26], #0x4\n" "tbz x8, #0, 68f\n" "ld1 { v11.h }[2], [x16]\n" - "ld1 { v15.h }[2], [x25]\n" + "ld1 { v15.h }[2], [x26]\n" "b 68f\n" "55:" // Height 2: Partial accumulate: partial_1_24 "mov x20, #0x30\n" "tbz x8, #0, 68f\n" "ldr h11, [x16, #0x0]\n" - "ldr h15, [x25, #0x0]\n" + "ldr h15, [x26, #0x0]\n" "b 68f\n" "56:" // Height 2: Partial accumulate: partial_4_16 "tbz x8, #2, 58f\n" "ldr d10, [x16], #0x8\n" - "ldr d14, [x25], #0x8\n" + "ldr d14, [x26], #0x8\n" "tbz x8, #1, 57f\n" "ld1 { v10.s }[2], [x16], #0x4\n" "mov x20, #0x2c\n" - "ld1 { v14.s }[2], [x25], #0x4\n" + "ld1 { v14.s }[2], [x26], #0x4\n" "tbz x8, #0, 68f\n" "ld1 { v10.h }[6], [x16]\n" - "ld1 { v14.h }[6], [x25]\n" + "ld1 { v14.h }[6], [x26]\n" "b 68f\n" "57:" // Height 2: Partial accumulate: partial_1_20 "mov x20, #0x28\n" "tbz x8, #0, 68f\n" "ld1 { v10.h }[4], [x16]\n" - "ld1 { v14.h }[4], [x25]\n" + "ld1 { v14.h }[4], [x26]\n" "b 68f\n" "58:" // Height 2: Partial accumulate: partial_2_16 "tbz x8, #1, 59f\n" "ldr s10, [x16], #0x4\n" "mov x20, #0x24\n" - "ldr s14, [x25], #0x4\n" + "ldr s14, [x26], #0x4\n" "tbz x8, #0, 68f\n" "ld1 { v10.h }[2], [x16]\n" - "ld1 { v14.h }[2], [x25]\n" + "ld1 { v14.h }[2], [x26]\n" "b 68f\n" "59:" // Height 2: Partial accumulate: partial_1_16 "mov x20, #0x20\n" "tbz x8, #0, 68f\n" "ldr h10, [x16, #0x0]\n" - "ldr h14, [x25, #0x0]\n" + "ldr h14, [x26, #0x0]\n" "b 68f\n" "60:" // Height 2: Partial accumulate: partial_8_0 "tbz x8, #3, 64f\n" "ld1 { v8.8h }, [x16], #0x10\n" - "ld1 { v12.8h }, [x25], #0x10\n" + "ld1 { v12.8h }, [x26], #0x10\n" "tbz x8, #2, 62f\n" "ldr d9, [x16], #0x8\n" - "ldr d13, [x25], #0x8\n" + "ldr d13, [x26], #0x8\n" "tbz x8, #1, 61f\n" "ld1 { v9.s }[2], [x16], #0x4\n" "mov x20, #0x1c\n" - "ld1 { v13.s }[2], [x25], #0x4\n" + "ld1 { v13.s }[2], [x26], #0x4\n" "tbz x8, #0, 68f\n" "ld1 { v9.h }[6], [x16]\n" - "ld1 { v13.h }[6], [x25]\n" + "ld1 { v13.h }[6], [x26]\n" "b 68f\n" "61:" // Height 2: Partial accumulate: partial_1_12 "mov x20, #0x18\n" "tbz x8, #0, 68f\n" "ld1 { v9.h }[4], [x16]\n" - "ld1 { v13.h }[4], [x25]\n" + "ld1 { v13.h }[4], [x26]\n" "b 68f\n" "62:" // Height 2: Partial accumulate: partial_2_8 "tbz x8, #1, 63f\n" "ldr s9, [x16], #0x4\n" "mov x20, #0x14\n" - "ldr s13, [x25], #0x4\n" + "ldr s13, [x26], #0x4\n" "tbz x8, #0, 68f\n" "ld1 { v9.h }[2], [x16]\n" - "ld1 { v13.h }[2], [x25]\n" + "ld1 { v13.h }[2], [x26]\n" "b 68f\n" "63:" // Height 2: Partial accumulate: partial_1_8 "mov x20, #0x10\n" "tbz x8, #0, 68f\n" "ldr h9, [x16, #0x0]\n" - "ldr h13, [x25, #0x0]\n" + "ldr h13, [x26, #0x0]\n" "b 68f\n" "64:" // Height 2: Partial accumulate: partial_4_0 "tbz x8, #2, 66f\n" "ldr d8, [x16], #0x8\n" - "ldr d12, [x25], #0x8\n" + "ldr d12, [x26], #0x8\n" "tbz x8, #1, 65f\n" "ld1 { v8.s }[2], [x16], #0x4\n" "mov x20, #0xc\n" - "ld1 { v12.s }[2], [x25], #0x4\n" + "ld1 { v12.s }[2], [x26], #0x4\n" "tbz x8, #0, 68f\n" "ld1 { v8.h }[6], [x16]\n" - "ld1 { v12.h }[6], [x25]\n" + "ld1 { v12.h }[6], [x26]\n" "b 68f\n" "65:" // Height 2: Partial accumulate: partial_1_4 "mov x20, #0x8\n" "tbz x8, #0, 68f\n" "ld1 { v8.h }[4], [x16]\n" - "ld1 { v12.h }[4], [x25]\n" + "ld1 { v12.h }[4], [x26]\n" "b 68f\n" "66:" // Height 2: Partial accumulate: partial_2_0 "tbz x8, #1, 67f\n" "ldr s8, [x16], #0x4\n" "mov x20, #0x4\n" - "ldr s12, [x25], #0x4\n" + "ldr s12, [x26], #0x4\n" "tbz x8, #0, 68f\n" "ld1 { v8.h }[2], [x16]\n" - "ld1 { v12.h }[2], [x25]\n" + "ld1 { v12.h }[2], [x26]\n" "b 68f\n" "67:" // Height 2: Partial accumulate: partial_1_0 "ldr h8, [x16, #0x0]\n" "mov x20, #0x0\n" - "ldr h12, [x25, #0x0]\n" + "ldr h12, [x26, #0x0]\n" "68:" // Height 2: Partial accumulate: Done "sub x16, x16, x20\n" "b 71f\n" @@ -780,10 +782,10 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "ldr q9, [x16, #0x10]\n" "ldr q10, [x16, #0x20]\n" "ldr q11, [x16, #0x30]\n" - "ldr q12, [x25, #0x0]\n" - "ldr q13, [x25, #0x10]\n" - "ldr q14, [x25, #0x20]\n" - "ldr q15, [x25, #0x30]\n" + "ldr q12, [x26, #0x0]\n" + "ldr q13, [x26, #0x10]\n" + "ldr q14, [x26, #0x20]\n" + "ldr q15, [x26, #0x30]\n" "b 71f\n" "70:" // Height 2: no accumulate "movi v8.16b, #0x0\n" @@ -798,8 +800,8 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "mov x15, #0x0\n" "72:" // 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, 73f\n" "ldr x20, [%x[input_ptr], x15, LSL #0x3]\n" "add x20, x20, x21, LSL #3\n" @@ -824,178 +826,178 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "blt 76f\n" "75:" // Height 2: Multiply loop: Main loop head "fmla v8.8h, v6.8h, v0.h[0]\n" - "ldr x21, [x17, #0x28]\n" + "ldr x20, [x17, #0x28]\n" "fmla v12.8h, v6.8h, v1.h[0]\n" "ldr d17, [x17, #0x20]\n" "fmla v9.8h, v7.8h, v0.h[0]\n" - "ldr x20, [x17, #0x38]\n" + "ldr x21, [x17, #0x38]\n" "fmla v13.8h, v7.8h, v1.h[0]\n" "ldr d16, [x17, #0x30]\n" - "mov v17.d[1], x21\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" "fmla v10.8h, v17.8h, v0.h[0]\n" - "mov v16.d[1], x20\n" "fmla v14.8h, v17.8h, v1.h[0]\n" "ldr d17, [x17, #0x40]\n" "fmla v11.8h, v16.8h, v0.h[0]\n" - "ldr x20, [x17, #0x48]\n" + "ldr x21, [x17, #0x58]\n" "fmla v15.8h, v16.8h, v1.h[0]\n" "ldr d16, [x17, #0x50]\n" "mov v17.d[1], x20\n" - "ldr x20, [x17, #0x58]\n" - "mov v16.d[1], x20\n" + "ldr x20, [x17, #0x68]\n" + "ldr x23, [x13, #0x8]\n" + "sub x14, x14, #0x8\n" + "mov v16.d[1], x21\n" "fmla v8.8h, v17.8h, v0.h[1]\n" - "ldr x21, [x17, #0x68]\n" "fmla v12.8h, v17.8h, v1.h[1]\n" "ldr d17, [x17, #0x60]\n" "fmla v9.8h, v16.8h, v0.h[1]\n" - "ldr x20, [x17, #0x78]\n" + "ldr x21, [x17, #0x78]\n" "fmla v13.8h, v16.8h, v1.h[1]\n" "ldr d16, [x17, #0x70]\n" - "mov v17.d[1], x21\n" + "mov v17.d[1], x20\n" + "ldr x20, [x17, #0x88]\n" + "ldr x22, [x12, #0x8]\n" + "cmp x14, #0x10\n" + "mov v16.d[1], x21\n" "fmla v10.8h, v17.8h, v0.h[1]\n" - "mov v16.d[1], x20\n" "fmla v14.8h, v17.8h, v1.h[1]\n" "ldr d17, [x17, #0x80]\n" "fmla v11.8h, v16.8h, v0.h[1]\n" - "ldr x20, [x17, #0x88]\n" + "ldr x21, [x17, #0x98]\n" "fmla v15.8h, v16.8h, v1.h[1]\n" "ldr d16, [x17, #0x90]\n" "mov v17.d[1], x20\n" - "ldr x20, [x17, #0x98]\n" - "mov v16.d[1], x20\n" + "ldr x20, [x17, #0xa8]\n" + "prfm pldl1keep, [x13, #0x80]\n" + "mov v16.d[1], x21\n" "fmla v8.8h, v17.8h, v0.h[2]\n" - "ldr x21, [x17, #0xa8]\n" "fmla v12.8h, v17.8h, v1.h[2]\n" "ldr d17, [x17, #0xa0]\n" "fmla v9.8h, v16.8h, v0.h[2]\n" - "ldr x20, [x17, #0xb8]\n" + "ldr x21, [x17, #0xb8]\n" "fmla v13.8h, v16.8h, v1.h[2]\n" "ldr d16, [x17, #0xb0]\n" - "mov v17.d[1], x21\n" + "mov v17.d[1], x20\n" + "ldr x20, [x17, #0xc8]\n" + "prfm pldl1keep, [x12, #0x80]\n" + "mov v16.d[1], x21\n" "fmla v10.8h, v17.8h, v0.h[2]\n" - "mov v16.d[1], x20\n" "fmla v14.8h, v17.8h, v1.h[2]\n" "ldr d17, [x17, #0xc0]\n" "fmla v11.8h, v16.8h, v0.h[2]\n" - "ldr x20, [x17, #0xc8]\n" + "ldr x21, [x17, #0xd8]\n" "fmla v15.8h, v16.8h, v1.h[2]\n" "ldr d16, [x17, #0xd0]\n" "mov v17.d[1], x20\n" - "ldr x20, [x17, #0xd8]\n" - "mov v16.d[1], x20\n" + "ldr x20, [x17, #0xe8]\n" + "mov v16.d[1], x21\n" "fmla v8.8h, v17.8h, v0.h[3]\n" - "ldr x21, [x17, #0xe8]\n" "fmla v12.8h, v17.8h, v1.h[3]\n" "ldr d17, [x17, #0xe0]\n" "fmla v9.8h, v16.8h, v0.h[3]\n" - "ldr x20, [x17, #0xf8]\n" + "ldr x21, [x17, #0xf8]\n" "fmla v13.8h, v16.8h, v1.h[3]\n" "ldr d16, [x17, #0xf0]\n" - "mov v17.d[1], x21\n" + "mov v17.d[1], x20\n" + "ldr x20, [x17, #0x108]\n" + "mov v16.d[1], x21\n" "fmla v10.8h, v17.8h, v0.h[3]\n" - "mov v16.d[1], x20\n" "fmla v14.8h, v17.8h, v1.h[3]\n" "ldr d17, [x17, #0x100]\n" "fmla v11.8h, v16.8h, v0.h[3]\n" - "ldr x20, [x17, #0x108]\n" + "ldr x21, [x17, #0x118]\n" "fmla v15.8h, v16.8h, v1.h[3]\n" "ldr d16, [x17, #0x110]\n" "mov v17.d[1], x20\n" - "ldr x20, [x17, #0x118]\n" - "mov v16.d[1], x20\n" + "ldr x20, [x17, #0x128]\n" + "mov v16.d[1], x21\n" "fmla v8.8h, v17.8h, v0.h[4]\n" - "ldr x21, [x17, #0x128]\n" "fmla v12.8h, v17.8h, v1.h[4]\n" "ldr d17, [x17, #0x120]\n" "fmla v9.8h, v16.8h, v0.h[4]\n" - "ldr x20, [x17, #0x138]\n" + "ldr x21, [x17, #0x138]\n" "fmla v13.8h, v16.8h, v1.h[4]\n" "ldr d16, [x17, #0x130]\n" - "mov v17.d[1], x21\n" + "mov v17.d[1], x20\n" + "ldr x20, [x17, #0x148]\n" + "mov v16.d[1], x21\n" "fmla v10.8h, v17.8h, v0.h[4]\n" - "mov v16.d[1], x20\n" "fmla v14.8h, v17.8h, v1.h[4]\n" "ldr d17, [x17, #0x140]\n" "fmla v11.8h, v16.8h, v0.h[4]\n" - "ldr x20, [x17, #0x148]\n" + "ldr x21, [x17, #0x158]\n" "fmla v15.8h, v16.8h, v1.h[4]\n" "ldr d16, [x17, #0x150]\n" "mov v17.d[1], x20\n" - "ldr x20, [x17, #0x158]\n" - "mov v16.d[1], x20\n" + "ldr x20, [x17, #0x168]\n" + "mov v16.d[1], x21\n" "fmla v8.8h, v17.8h, v0.h[5]\n" - "ldr x21, [x17, #0x168]\n" "fmla v12.8h, v17.8h, v1.h[5]\n" "ldr d17, [x17, #0x160]\n" "fmla v9.8h, v16.8h, v0.h[5]\n" - "ldr x20, [x17, #0x178]\n" + "ldr x21, [x17, #0x178]\n" "fmla v13.8h, v16.8h, v1.h[5]\n" "ldr d16, [x17, #0x170]\n" - "mov v17.d[1], x21\n" + "mov v17.d[1], x20\n" + "ldr x20, [x17, #0x188]\n" + "mov v16.d[1], x21\n" "fmla v10.8h, v17.8h, v0.h[5]\n" - "mov v16.d[1], x20\n" "fmla v14.8h, v17.8h, v1.h[5]\n" "ldr d17, [x17, #0x180]\n" "fmla v11.8h, v16.8h, v0.h[5]\n" - "ldr x20, [x17, #0x188]\n" + "ldr x21, [x17, #0x198]\n" "fmla v15.8h, v16.8h, v1.h[5]\n" "ldr d16, [x17, #0x190]\n" "mov v17.d[1], x20\n" - "ldr x20, [x17, #0x198]\n" - "mov v16.d[1], x20\n" + "ldr x20, [x17, #0x1a8]\n" + "mov v16.d[1], x21\n" "fmla v8.8h, v17.8h, v0.h[6]\n" - "ldr x21, [x17, #0x1a8]\n" "fmla v12.8h, v17.8h, v1.h[6]\n" "ldr d17, [x17, #0x1a0]\n" "fmla v9.8h, v16.8h, v0.h[6]\n" - "ldr x20, [x17, #0x1b8]\n" + "ldr x21, [x17, #0x1b8]\n" "fmla v13.8h, v16.8h, v1.h[6]\n" "ldr d16, [x17, #0x1b0]\n" - "mov v17.d[1], x21\n" + "mov v17.d[1], x20\n" + "ldr x20, [x17, #0x1c8]\n" + "mov v16.d[1], x21\n" "fmla v10.8h, v17.8h, v0.h[6]\n" - "mov v16.d[1], x20\n" "fmla v14.8h, v17.8h, v1.h[6]\n" "ldr d17, [x17, #0x1c0]\n" "fmla v11.8h, v16.8h, v0.h[6]\n" - "ldr x20, [x17, #0x1c8]\n" + "ldr x21, [x17, #0x1d8]\n" "fmla v15.8h, v16.8h, v1.h[6]\n" "ldr d16, [x17, #0x1d0]\n" "mov v17.d[1], x20\n" - "ldr x20, [x17, #0x1d8]\n" - "mov v16.d[1], x20\n" + "ldr x20, [x17, #0x1e8]\n" + "mov v16.d[1], x21\n" "fmla v8.8h, v17.8h, v0.h[7]\n" - "ldr x21, [x17, #0x1e8]\n" "fmla v12.8h, v17.8h, v1.h[7]\n" "ldr d17, [x17, #0x1e0]\n" "fmla v9.8h, v16.8h, v0.h[7]\n" - "ldr x20, [x17, #0x1f8]\n" + "ldr x21, [x17, #0x1f8]\n" "fmla v13.8h, v16.8h, v1.h[7]\n" "ldr d16, [x17, #0x1f0]\n" - "mov v17.d[1], x21\n" - "add x13, x13, #0x10\n" - "mov v16.d[1], x20\n" - "add x12, x12, #0x10\n" + "mov v17.d[1], x20\n" "add x17, x17, #0x200\n" + "ldr x20, [x17, #0x8]\n" + "mov v16.d[1], x21\n" "fmla v10.8h, v17.8h, v0.h[7]\n" "fmla v14.8h, v17.8h, v1.h[7]\n" "ldr d6, [x17, #0x0]\n" - "ldr x21, [x17, #0x8]\n" "fmla v11.8h, v16.8h, v0.h[7]\n" "ldr d0, [x13, #0x0]\n" "fmla v15.8h, v16.8h, v1.h[7]\n" "ldr d1, [x12, #0x0]\n" - "sub x14, x14, #0x8\n" "ldr d7, [x17, #0x10]\n" - "cmp x14, #0x10\n" - "ldr x20, [x13, #0x8]\n" - "mov v6.d[1], x21\n" - "ldr x21, [x12, #0x8]\n" - "mov v0.d[1], x20\n" + "mov v6.d[1], x20\n" "ldr x20, [x17, #0x18]\n" - "mov v1.d[1], x21\n" - "prfm pldl1keep, [x13, #0x80]\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 75b\n" "76:" // Height 2: Multiply loop: Single iteration only "fmla v8.8h, v6.8h, v0.h[0]\n" @@ -1105,8 +1107,8 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "sub x14, x14, #0x1\n" "ldr h0, [x12], #0x2\n" "ldr q17, [x17, #0x0]\n" - "fmla v8.8h, v17.8h, v1.h[0]\n" "ldr q16, [x17, #0x10]\n" + "fmla v8.8h, v17.8h, v1.h[0]\n" "fmla v12.8h, v17.8h, v0.h[0]\n" "ldr q17, [x17, #0x20]\n" "fmla v9.8h, v16.8h, v1.h[0]\n" @@ -1124,22 +1126,22 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "cmp x15, x20\n" "bne 72b\n" "ldr x20, [%x[args_ptr], %[offsetof_output_offset]]\n" - "add x25, x16, x20, LSL #1\n" "prfm pstl1keep, [x16, #0x0]\n" - "prfm pstl1keep, [x25, #0x0]\n" + "add x26, x16, x20, LSL #1\n" + "prfm pstl1keep, [x26, #0x0]\n" "tbz %x[flags], #1, 80f\n" - "add x20, %x[args_ptr], %[offset_max]\n" - "ld1r { v16.8h }, [x20]\n" - "fmin v8.8h, v8.8h, v16.8h\n" - "fmin v9.8h, v9.8h, v16.8h\n" - "fmin v10.8h, v10.8h, v16.8h\n" - "fmin v11.8h, v11.8h, v16.8h\n" - "fmin v12.8h, v12.8h, v16.8h\n" - "fmin v13.8h, v13.8h, v16.8h\n" - "fmin v14.8h, v14.8h, v16.8h\n" - "fmin v15.8h, v15.8h, v16.8h\n" + "add x21, %x[args_ptr], %[offset_max]\n" "add x20, %x[args_ptr], %[offset_min]\n" + "ld1r { v17.8h }, [x21]\n" "ld1r { v16.8h }, [x20]\n" + "fmin v8.8h, v8.8h, v17.8h\n" + "fmin v9.8h, v9.8h, v17.8h\n" + "fmin v10.8h, v10.8h, v17.8h\n" + "fmin v11.8h, v11.8h, v17.8h\n" + "fmin v12.8h, v12.8h, v17.8h\n" + "fmin v13.8h, v13.8h, v17.8h\n" + "fmin v14.8h, v14.8h, v17.8h\n" + "fmin v15.8h, v15.8h, v17.8h\n" "fmax v8.8h, v8.8h, v16.8h\n" "fmax v9.8h, v9.8h, v16.8h\n" "fmax v10.8h, v10.8h, v16.8h\n" @@ -1154,127 +1156,127 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "tbz x8, #4, 88f\n" "st1 { v8.8h }, [x16], #0x10\n" "st1 { v9.8h }, [x16], #0x10\n" - "st1 { v12.8h }, [x25], #0x10\n" - "st1 { v13.8h }, [x25], #0x10\n" + "st1 { v12.8h }, [x26], #0x10\n" + "st1 { v13.8h }, [x26], #0x10\n" "tbz x8, #3, 84f\n" "st1 { v10.8h }, [x16], #0x10\n" - "st1 { v14.8h }, [x25], #0x10\n" + "st1 { v14.8h }, [x26], #0x10\n" "tbz x8, #2, 82f\n" "str d11, [x16], #0x8\n" - "str d15, [x25], #0x8\n" + "str d15, [x26], #0x8\n" "tbz x8, #1, 81f\n" "st1 { v11.s }[2], [x16], #0x4\n" - "st1 { v15.s }[2], [x25], #0x4\n" + "st1 { v15.s }[2], [x26], #0x4\n" "tbz x8, #0, 96f\n" "st1 { v11.h }[6], [x16]\n" - "st1 { v15.h }[6], [x25]\n" + "st1 { v15.h }[6], [x26]\n" "b 96f\n" "81:" // Height 2: Partial direct writeback: partial_1_28 "tbz x8, #0, 96f\n" "st1 { v11.h }[4], [x16]\n" - "st1 { v15.h }[4], [x25]\n" + "st1 { v15.h }[4], [x26]\n" "b 96f\n" "82:" // Height 2: Partial direct writeback: partial_2_24 "tbz x8, #1, 83f\n" "str s11, [x16], #0x4\n" - "str s15, [x25], #0x4\n" + "str s15, [x26], #0x4\n" "tbz x8, #0, 96f\n" "st1 { v11.h }[2], [x16]\n" - "st1 { v15.h }[2], [x25]\n" + "st1 { v15.h }[2], [x26]\n" "b 96f\n" "83:" // Height 2: Partial direct writeback: partial_1_24 "tbz x8, #0, 96f\n" "str h11, [x16, #0x0]\n" - "str h15, [x25, #0x0]\n" + "str h15, [x26, #0x0]\n" "b 96f\n" "84:" // Height 2: Partial direct writeback: partial_4_16 "tbz x8, #2, 86f\n" "str d10, [x16], #0x8\n" - "str d14, [x25], #0x8\n" + "str d14, [x26], #0x8\n" "tbz x8, #1, 85f\n" "st1 { v10.s }[2], [x16], #0x4\n" - "st1 { v14.s }[2], [x25], #0x4\n" + "st1 { v14.s }[2], [x26], #0x4\n" "tbz x8, #0, 96f\n" "st1 { v10.h }[6], [x16]\n" - "st1 { v14.h }[6], [x25]\n" + "st1 { v14.h }[6], [x26]\n" "b 96f\n" "85:" // Height 2: Partial direct writeback: partial_1_20 "tbz x8, #0, 96f\n" "st1 { v10.h }[4], [x16]\n" - "st1 { v14.h }[4], [x25]\n" + "st1 { v14.h }[4], [x26]\n" "b 96f\n" "86:" // Height 2: Partial direct writeback: partial_2_16 "tbz x8, #1, 87f\n" "str s10, [x16], #0x4\n" - "str s14, [x25], #0x4\n" + "str s14, [x26], #0x4\n" "tbz x8, #0, 96f\n" "st1 { v10.h }[2], [x16]\n" - "st1 { v14.h }[2], [x25]\n" + "st1 { v14.h }[2], [x26]\n" "b 96f\n" "87:" // Height 2: Partial direct writeback: partial_1_16 "tbz x8, #0, 96f\n" "str h10, [x16, #0x0]\n" - "str h14, [x25, #0x0]\n" + "str h14, [x26, #0x0]\n" "b 96f\n" "88:" // Height 2: Partial direct writeback: partial_8_0 "tbz x8, #3, 92f\n" "st1 { v8.8h }, [x16], #0x10\n" - "st1 { v12.8h }, [x25], #0x10\n" + "st1 { v12.8h }, [x26], #0x10\n" "tbz x8, #2, 90f\n" "str d9, [x16], #0x8\n" - "str d13, [x25], #0x8\n" + "str d13, [x26], #0x8\n" "tbz x8, #1, 89f\n" "st1 { v9.s }[2], [x16], #0x4\n" - "st1 { v13.s }[2], [x25], #0x4\n" + "st1 { v13.s }[2], [x26], #0x4\n" "tbz x8, #0, 96f\n" "st1 { v9.h }[6], [x16]\n" - "st1 { v13.h }[6], [x25]\n" + "st1 { v13.h }[6], [x26]\n" "b 96f\n" "89:" // Height 2: Partial direct writeback: partial_1_12 "tbz x8, #0, 96f\n" "st1 { v9.h }[4], [x16]\n" - "st1 { v13.h }[4], [x25]\n" + "st1 { v13.h }[4], [x26]\n" "b 96f\n" "90:" // Height 2: Partial direct writeback: partial_2_8 "tbz x8, #1, 91f\n" "str s9, [x16], #0x4\n" - "str s13, [x25], #0x4\n" + "str s13, [x26], #0x4\n" "tbz x8, #0, 96f\n" "st1 { v9.h }[2], [x16]\n" - "st1 { v13.h }[2], [x25]\n" + "st1 { v13.h }[2], [x26]\n" "b 96f\n" "91:" // Height 2: Partial direct writeback: partial_1_8 "tbz x8, #0, 96f\n" "str h9, [x16, #0x0]\n" - "str h13, [x25, #0x0]\n" + "str h13, [x26, #0x0]\n" "b 96f\n" "92:" // Height 2: Partial direct writeback: partial_4_0 "tbz x8, #2, 94f\n" "str d8, [x16], #0x8\n" - "str d12, [x25], #0x8\n" + "str d12, [x26], #0x8\n" "tbz x8, #1, 93f\n" "st1 { v8.s }[2], [x16], #0x4\n" - "st1 { v12.s }[2], [x25], #0x4\n" + "st1 { v12.s }[2], [x26], #0x4\n" "tbz x8, #0, 96f\n" "st1 { v8.h }[6], [x16]\n" - "st1 { v12.h }[6], [x25]\n" + "st1 { v12.h }[6], [x26]\n" "b 96f\n" "93:" // Height 2: Partial direct writeback: partial_1_4 "tbz x8, #0, 96f\n" "st1 { v8.h }[4], [x16]\n" - "st1 { v12.h }[4], [x25]\n" + "st1 { v12.h }[4], [x26]\n" "b 96f\n" "94:" // Height 2: Partial direct writeback: partial_2_0 "tbz x8, #1, 95f\n" "str s8, [x16], #0x4\n" - "str s12, [x25], #0x4\n" + "str s12, [x26], #0x4\n" "tbz x8, #0, 96f\n" "st1 { v8.h }[2], [x16]\n" - "st1 { v12.h }[2], [x25]\n" + "st1 { v12.h }[2], [x26]\n" "b 96f\n" "95:" // Height 2: Partial direct writeback: partial_1_0 "str h8, [x16, #0x0]\n" - "str h12, [x25, #0x0]\n" + "str h12, [x26, #0x0]\n" "96:" // Height 2: Partial direct writeback: Done "b 98f\n" "97:" // Height 2: Full writeback @@ -1283,31 +1285,31 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "str q10, [x16, #0x20]\n" "str q11, [x16, #0x30]\n" "add x16, x16, #0x40\n" - "str q12, [x25, #0x0]\n" - "str q13, [x25, #0x10]\n" - "str q14, [x25, #0x20]\n" - "str q15, [x25, #0x30]\n" + "str q12, [x26, #0x0]\n" + "str q13, [x26, #0x10]\n" + "str q14, [x26, #0x20]\n" + "str q15, [x26, #0x30]\n" "98:" // Height 2: Writeback done "subs x8, x8, #0x20\n" "bgt 51b\n" "b 296f\n" "99:" // Height 3 - "mov x7, %x[bias]\n" + "ldr x7, [%x[args_ptr], %[offsetof_bias]]\n" "ldr x8, [%x[args_ptr], %[offsetof_N]]\n" "ldr x17, [%x[args_ptr], %[offsetof_B_ptr]]\n" - "mov x16, %x[output_ptr]\n" + "ldr x16, [%x[args_ptr], %[offsetof_output_ptr]]\n" "100:" // Height 3: Column loop "cbz x7, 101f\n" "ldr q8, [x7, #0x0]\n" - "mov v12.16b, v8.16b\n" "ldr q9, [x7, #0x10]\n" - "mov v13.16b, v9.16b\n" "ldr q10, [x7, #0x20]\n" - "mov v14.16b, v10.16b\n" "ldr q11, [x7, #0x30]\n" + "mov v12.16b, v8.16b\n" + "mov v13.16b, v9.16b\n" + "add x7, x7, #0x40\n" + "mov v14.16b, v10.16b\n" "mov v15.16b, v11.16b\n" "mov v16.16b, v8.16b\n" - "add x7, x7, #0x40\n" "mov v17.16b, v9.16b\n" "mov v18.16b, v10.16b\n" "mov v19.16b, v11.16b\n" @@ -1315,182 +1317,182 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "101:" // Height 3: no bias "tbz %x[flags], #0, 119f\n" "ldr x20, [%x[args_ptr], %[offsetof_output_offset]]\n" - "add x25, x16, x20, LSL #1\n" "cmp x8, #0x20\n" - "add x24, x25, x20, LSL #1\n" + "add x26, x16, x20, LSL #1\n" + "add x25, x26, x20, LSL #1\n" "bge 118f\n" "tbz x8, #4, 109f\n" "ld1 { v8.8h }, [x16], #0x10\n" - "ld1 { v12.8h }, [x25], #0x10\n" - "ld1 { v16.8h }, [x24], #0x10\n" + "ld1 { v12.8h }, [x26], #0x10\n" + "ld1 { v16.8h }, [x25], #0x10\n" "ld1 { v9.8h }, [x16], #0x10\n" - "ld1 { v13.8h }, [x25], #0x10\n" - "ld1 { v17.8h }, [x24], #0x10\n" + "ld1 { v13.8h }, [x26], #0x10\n" + "ld1 { v17.8h }, [x25], #0x10\n" "tbz x8, #3, 105f\n" "ld1 { v10.8h }, [x16], #0x10\n" - "ld1 { v14.8h }, [x25], #0x10\n" - "ld1 { v18.8h }, [x24], #0x10\n" + "ld1 { v14.8h }, [x26], #0x10\n" + "ld1 { v18.8h }, [x25], #0x10\n" "tbz x8, #2, 103f\n" "ldr d11, [x16], #0x8\n" - "ldr d15, [x25], #0x8\n" - "ldr d19, [x24], #0x8\n" + "ldr d15, [x26], #0x8\n" + "ldr d19, [x25], #0x8\n" "tbz x8, #1, 102f\n" "ld1 { v11.s }[2], [x16], #0x4\n" "mov x20, #0x3c\n" - "ld1 { v15.s }[2], [x25], #0x4\n" - "ld1 { v19.s }[2], [x24], #0x4\n" + "ld1 { v15.s }[2], [x26], #0x4\n" + "ld1 { v19.s }[2], [x25], #0x4\n" "tbz x8, #0, 117f\n" "ld1 { v11.h }[6], [x16]\n" - "ld1 { v15.h }[6], [x25]\n" - "ld1 { v19.h }[6], [x24]\n" + "ld1 { v15.h }[6], [x26]\n" + "ld1 { v19.h }[6], [x25]\n" "b 117f\n" "102:" // Height 3: Partial accumulate: partial_1_28 "mov x20, #0x38\n" "tbz x8, #0, 117f\n" "ld1 { v11.h }[4], [x16]\n" - "ld1 { v15.h }[4], [x25]\n" - "ld1 { v19.h }[4], [x24]\n" + "ld1 { v15.h }[4], [x26]\n" + "ld1 { v19.h }[4], [x25]\n" "b 117f\n" "103:" // Height 3: Partial accumulate: partial_2_24 "tbz x8, #1, 104f\n" "ldr s11, [x16], #0x4\n" "mov x20, #0x34\n" - "ldr s15, [x25], #0x4\n" - "ldr s19, [x24], #0x4\n" + "ldr s15, [x26], #0x4\n" + "ldr s19, [x25], #0x4\n" "tbz x8, #0, 117f\n" "ld1 { v11.h }[2], [x16]\n" - "ld1 { v15.h }[2], [x25]\n" - "ld1 { v19.h }[2], [x24]\n" + "ld1 { v15.h }[2], [x26]\n" + "ld1 { v19.h }[2], [x25]\n" "b 117f\n" "104:" // Height 3: Partial accumulate: partial_1_24 "mov x20, #0x30\n" "tbz x8, #0, 117f\n" "ldr h11, [x16, #0x0]\n" - "ldr h15, [x25, #0x0]\n" - "ldr h19, [x24, #0x0]\n" + "ldr h15, [x26, #0x0]\n" + "ldr h19, [x25, #0x0]\n" "b 117f\n" "105:" // Height 3: Partial accumulate: partial_4_16 "tbz x8, #2, 107f\n" "ldr d10, [x16], #0x8\n" - "ldr d14, [x25], #0x8\n" - "ldr d18, [x24], #0x8\n" + "ldr d14, [x26], #0x8\n" + "ldr d18, [x25], #0x8\n" "tbz x8, #1, 106f\n" "ld1 { v10.s }[2], [x16], #0x4\n" "mov x20, #0x2c\n" - "ld1 { v14.s }[2], [x25], #0x4\n" - "ld1 { v18.s }[2], [x24], #0x4\n" + "ld1 { v14.s }[2], [x26], #0x4\n" + "ld1 { v18.s }[2], [x25], #0x4\n" "tbz x8, #0, 117f\n" "ld1 { v10.h }[6], [x16]\n" - "ld1 { v14.h }[6], [x25]\n" - "ld1 { v18.h }[6], [x24]\n" + "ld1 { v14.h }[6], [x26]\n" + "ld1 { v18.h }[6], [x25]\n" "b 117f\n" "106:" // Height 3: Partial accumulate: partial_1_20 "mov x20, #0x28\n" "tbz x8, #0, 117f\n" "ld1 { v10.h }[4], [x16]\n" - "ld1 { v14.h }[4], [x25]\n" - "ld1 { v18.h }[4], [x24]\n" + "ld1 { v14.h }[4], [x26]\n" + "ld1 { v18.h }[4], [x25]\n" "b 117f\n" "107:" // Height 3: Partial accumulate: partial_2_16 "tbz x8, #1, 108f\n" "ldr s10, [x16], #0x4\n" "mov x20, #0x24\n" - "ldr s14, [x25], #0x4\n" - "ldr s18, [x24], #0x4\n" + "ldr s14, [x26], #0x4\n" + "ldr s18, [x25], #0x4\n" "tbz x8, #0, 117f\n" "ld1 { v10.h }[2], [x16]\n" - "ld1 { v14.h }[2], [x25]\n" - "ld1 { v18.h }[2], [x24]\n" + "ld1 { v14.h }[2], [x26]\n" + "ld1 { v18.h }[2], [x25]\n" "b 117f\n" "108:" // Height 3: Partial accumulate: partial_1_16 "mov x20, #0x20\n" "tbz x8, #0, 117f\n" "ldr h10, [x16, #0x0]\n" - "ldr h14, [x25, #0x0]\n" - "ldr h18, [x24, #0x0]\n" + "ldr h14, [x26, #0x0]\n" + "ldr h18, [x25, #0x0]\n" "b 117f\n" "109:" // Height 3: Partial accumulate: partial_8_0 "tbz x8, #3, 113f\n" "ld1 { v8.8h }, [x16], #0x10\n" - "ld1 { v12.8h }, [x25], #0x10\n" - "ld1 { v16.8h }, [x24], #0x10\n" + "ld1 { v12.8h }, [x26], #0x10\n" + "ld1 { v16.8h }, [x25], #0x10\n" "tbz x8, #2, 111f\n" "ldr d9, [x16], #0x8\n" - "ldr d13, [x25], #0x8\n" - "ldr d17, [x24], #0x8\n" + "ldr d13, [x26], #0x8\n" + "ldr d17, [x25], #0x8\n" "tbz x8, #1, 110f\n" "ld1 { v9.s }[2], [x16], #0x4\n" "mov x20, #0x1c\n" - "ld1 { v13.s }[2], [x25], #0x4\n" - "ld1 { v17.s }[2], [x24], #0x4\n" + "ld1 { v13.s }[2], [x26], #0x4\n" + "ld1 { v17.s }[2], [x25], #0x4\n" "tbz x8, #0, 117f\n" "ld1 { v9.h }[6], [x16]\n" - "ld1 { v13.h }[6], [x25]\n" - "ld1 { v17.h }[6], [x24]\n" + "ld1 { v13.h }[6], [x26]\n" + "ld1 { v17.h }[6], [x25]\n" "b 117f\n" "110:" // Height 3: Partial accumulate: partial_1_12 "mov x20, #0x18\n" "tbz x8, #0, 117f\n" "ld1 { v9.h }[4], [x16]\n" - "ld1 { v13.h }[4], [x25]\n" - "ld1 { v17.h }[4], [x24]\n" + "ld1 { v13.h }[4], [x26]\n" + "ld1 { v17.h }[4], [x25]\n" "b 117f\n" "111:" // Height 3: Partial accumulate: partial_2_8 "tbz x8, #1, 112f\n" "ldr s9, [x16], #0x4\n" "mov x20, #0x14\n" - "ldr s13, [x25], #0x4\n" - "ldr s17, [x24], #0x4\n" + "ldr s13, [x26], #0x4\n" + "ldr s17, [x25], #0x4\n" "tbz x8, #0, 117f\n" "ld1 { v9.h }[2], [x16]\n" - "ld1 { v13.h }[2], [x25]\n" - "ld1 { v17.h }[2], [x24]\n" + "ld1 { v13.h }[2], [x26]\n" + "ld1 { v17.h }[2], [x25]\n" "b 117f\n" "112:" // Height 3: Partial accumulate: partial_1_8 "mov x20, #0x10\n" "tbz x8, #0, 117f\n" "ldr h9, [x16, #0x0]\n" - "ldr h13, [x25, #0x0]\n" - "ldr h17, [x24, #0x0]\n" + "ldr h13, [x26, #0x0]\n" + "ldr h17, [x25, #0x0]\n" "b 117f\n" "113:" // Height 3: Partial accumulate: partial_4_0 "tbz x8, #2, 115f\n" "ldr d8, [x16], #0x8\n" - "ldr d12, [x25], #0x8\n" - "ldr d16, [x24], #0x8\n" + "ldr d12, [x26], #0x8\n" + "ldr d16, [x25], #0x8\n" "tbz x8, #1, 114f\n" "ld1 { v8.s }[2], [x16], #0x4\n" "mov x20, #0xc\n" - "ld1 { v12.s }[2], [x25], #0x4\n" - "ld1 { v16.s }[2], [x24], #0x4\n" + "ld1 { v12.s }[2], [x26], #0x4\n" + "ld1 { v16.s }[2], [x25], #0x4\n" "tbz x8, #0, 117f\n" "ld1 { v8.h }[6], [x16]\n" - "ld1 { v12.h }[6], [x25]\n" - "ld1 { v16.h }[6], [x24]\n" + "ld1 { v12.h }[6], [x26]\n" + "ld1 { v16.h }[6], [x25]\n" "b 117f\n" "114:" // Height 3: Partial accumulate: partial_1_4 "mov x20, #0x8\n" "tbz x8, #0, 117f\n" "ld1 { v8.h }[4], [x16]\n" - "ld1 { v12.h }[4], [x25]\n" - "ld1 { v16.h }[4], [x24]\n" + "ld1 { v12.h }[4], [x26]\n" + "ld1 { v16.h }[4], [x25]\n" "b 117f\n" "115:" // Height 3: Partial accumulate: partial_2_0 "tbz x8, #1, 116f\n" "ldr s8, [x16], #0x4\n" "mov x20, #0x4\n" - "ldr s12, [x25], #0x4\n" - "ldr s16, [x24], #0x4\n" + "ldr s12, [x26], #0x4\n" + "ldr s16, [x25], #0x4\n" "tbz x8, #0, 117f\n" "ld1 { v8.h }[2], [x16]\n" - "ld1 { v12.h }[2], [x25]\n" - "ld1 { v16.h }[2], [x24]\n" + "ld1 { v12.h }[2], [x26]\n" + "ld1 { v16.h }[2], [x25]\n" "b 117f\n" "116:" // Height 3: Partial accumulate: partial_1_0 "ldr h8, [x16, #0x0]\n" "mov x20, #0x0\n" - "ldr h12, [x25, #0x0]\n" - "ldr h16, [x24, #0x0]\n" + "ldr h12, [x26, #0x0]\n" + "ldr h16, [x25, #0x0]\n" "117:" // Height 3: Partial accumulate: Done "sub x16, x16, x20\n" "b 120f\n" @@ -1499,14 +1501,14 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "ldr q9, [x16, #0x10]\n" "ldr q10, [x16, #0x20]\n" "ldr q11, [x16, #0x30]\n" - "ldr q12, [x25, #0x0]\n" - "ldr q13, [x25, #0x10]\n" - "ldr q14, [x25, #0x20]\n" - "ldr q15, [x25, #0x30]\n" - "ldr q16, [x24, #0x0]\n" - "ldr q17, [x24, #0x10]\n" - "ldr q18, [x24, #0x20]\n" - "ldr q19, [x24, #0x30]\n" + "ldr q12, [x26, #0x0]\n" + "ldr q13, [x26, #0x10]\n" + "ldr q14, [x26, #0x20]\n" + "ldr q15, [x26, #0x30]\n" + "ldr q16, [x25, #0x0]\n" + "ldr q17, [x25, #0x10]\n" + "ldr q18, [x25, #0x20]\n" + "ldr q19, [x25, #0x30]\n" "b 120f\n" "119:" // Height 3: no accumulate "movi v8.16b, #0x0\n" @@ -1525,8 +1527,8 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "mov x15, #0x0\n" "121:" // 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, 122f\n" "ldr x20, [%x[input_ptr], x15, LSL #0x3]\n" "add x20, x20, x21, LSL #3\n" @@ -1561,11 +1563,15 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "fmla v16.8h, v6.8h, v2.h[0]\n" "ldr d21, [x17, #0x20]\n" "fmla v9.8h, v7.8h, v0.h[0]\n" - "mov v21.d[1], x21\n" + "add x13, x13, #0x10\n" "fmla v13.8h, v7.8h, v1.h[0]\n" - "ldr x21, [x17, #0x48]\n" + "add x12, x12, #0x10\n" "fmla v17.8h, v7.8h, v2.h[0]\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" "fmla v10.8h, v21.8h, v0.h[0]\n" "fmla v14.8h, v21.8h, v1.h[0]\n" @@ -1573,11 +1579,15 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "fmla v18.8h, v21.8h, v2.h[0]\n" "ldr d21, [x17, #0x40]\n" "fmla v11.8h, v20.8h, v0.h[0]\n" - "mov v21.d[1], x21\n" + "ldr x23, [x12, #0x8]\n" "fmla v15.8h, v20.8h, v1.h[0]\n" - "ldr x21, [x17, #0x68]\n" + "ldr x22, [x11, #0x8]\n" "fmla v19.8h, v20.8h, v2.h[0]\n" "ldr d20, [x17, #0x50]\n" + "mov v21.d[1], x21\n" + "ldr x21, [x17, #0x68]\n" + "sub x14, x14, #0x8\n" + "prfm pldl1keep, [x13, #0x80]\n" "mov v20.d[1], x20\n" "fmla v8.8h, v21.8h, v0.h[1]\n" "fmla v12.8h, v21.8h, v1.h[1]\n" @@ -1585,11 +1595,14 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "fmla v16.8h, v21.8h, v2.h[1]\n" "ldr d21, [x17, #0x60]\n" "fmla v9.8h, v20.8h, v0.h[1]\n" - "mov v21.d[1], x21\n" + "cmp x14, #0x10\n" "fmla v13.8h, v20.8h, v1.h[1]\n" - "ldr x21, [x17, #0x88]\n" + "prfm pldl1keep, [x12, #0x80]\n" "fmla v17.8h, v20.8h, v2.h[1]\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" "fmla v10.8h, v21.8h, v0.h[1]\n" "fmla v14.8h, v21.8h, v1.h[1]\n" @@ -1597,11 +1610,11 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "fmla v18.8h, v21.8h, v2.h[1]\n" "ldr d21, [x17, #0x80]\n" "fmla v11.8h, v20.8h, v0.h[1]\n" - "mov v21.d[1], x21\n" "fmla v15.8h, v20.8h, v1.h[1]\n" - "ldr x21, [x17, #0xa8]\n" "fmla v19.8h, v20.8h, v2.h[1]\n" "ldr d20, [x17, #0x90]\n" + "mov v21.d[1], x21\n" + "ldr x21, [x17, #0xa8]\n" "mov v20.d[1], x20\n" "fmla v8.8h, v21.8h, v0.h[2]\n" "fmla v12.8h, v21.8h, v1.h[2]\n" @@ -1609,11 +1622,11 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "fmla v16.8h, v21.8h, v2.h[2]\n" "ldr d21, [x17, #0xa0]\n" "fmla v9.8h, v20.8h, v0.h[2]\n" - "mov v21.d[1], x21\n" "fmla v13.8h, v20.8h, v1.h[2]\n" - "ldr x21, [x17, #0xc8]\n" "fmla v17.8h, v20.8h, v2.h[2]\n" "ldr d20, [x17, #0xb0]\n" + "mov v21.d[1], x21\n" + "ldr x21, [x17, #0xc8]\n" "mov v20.d[1], x20\n" "fmla v10.8h, v21.8h, v0.h[2]\n" "fmla v14.8h, v21.8h, v1.h[2]\n" @@ -1621,11 +1634,11 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "fmla v18.8h, v21.8h, v2.h[2]\n" "ldr d21, [x17, #0xc0]\n" "fmla v11.8h, v20.8h, v0.h[2]\n" - "mov v21.d[1], x21\n" "fmla v15.8h, v20.8h, v1.h[2]\n" - "ldr x21, [x17, #0xe8]\n" "fmla v19.8h, v20.8h, v2.h[2]\n" "ldr d20, [x17, #0xd0]\n" + "mov v21.d[1], x21\n" + "ldr x21, [x17, #0xe8]\n" "mov v20.d[1], x20\n" "fmla v8.8h, v21.8h, v0.h[3]\n" "fmla v12.8h, v21.8h, v1.h[3]\n" @@ -1633,11 +1646,11 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "fmla v16.8h, v21.8h, v2.h[3]\n" "ldr d21, [x17, #0xe0]\n" "fmla v9.8h, v20.8h, v0.h[3]\n" - "mov v21.d[1], x21\n" "fmla v13.8h, v20.8h, v1.h[3]\n" - "ldr x21, [x17, #0x108]\n" "fmla v17.8h, v20.8h, v2.h[3]\n" "ldr d20, [x17, #0xf0]\n" + "mov v21.d[1], x21\n" + "ldr x21, [x17, #0x108]\n" "mov v20.d[1], x20\n" "fmla v10.8h, v21.8h, v0.h[3]\n" "fmla v14.8h, v21.8h, v1.h[3]\n" @@ -1645,11 +1658,11 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "fmla v18.8h, v21.8h, v2.h[3]\n" "ldr d21, [x17, #0x100]\n" "fmla v11.8h, v20.8h, v0.h[3]\n" - "mov v21.d[1], x21\n" "fmla v15.8h, v20.8h, v1.h[3]\n" - "ldr x21, [x17, #0x128]\n" "fmla v19.8h, v20.8h, v2.h[3]\n" "ldr d20, [x17, #0x110]\n" + "mov v21.d[1], x21\n" + "ldr x21, [x17, #0x128]\n" "mov v20.d[1], x20\n" "fmla v8.8h, v21.8h, v0.h[4]\n" "fmla v12.8h, v21.8h, v1.h[4]\n" @@ -1657,11 +1670,11 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "fmla v16.8h, v21.8h, v2.h[4]\n" "ldr d21, [x17, #0x120]\n" "fmla v9.8h, v20.8h, v0.h[4]\n" - "mov v21.d[1], x21\n" "fmla v13.8h, v20.8h, v1.h[4]\n" - "ldr x21, [x17, #0x148]\n" "fmla v17.8h, v20.8h, v2.h[4]\n" "ldr d20, [x17, #0x130]\n" + "mov v21.d[1], x21\n" + "ldr x21, [x17, #0x148]\n" "mov v20.d[1], x20\n" "fmla v10.8h, v21.8h, v0.h[4]\n" "fmla v14.8h, v21.8h, v1.h[4]\n" @@ -1669,11 +1682,11 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "fmla v18.8h, v21.8h, v2.h[4]\n" "ldr d21, [x17, #0x140]\n" "fmla v11.8h, v20.8h, v0.h[4]\n" - "mov v21.d[1], x21\n" "fmla v15.8h, v20.8h, v1.h[4]\n" - "ldr x21, [x17, #0x168]\n" "fmla v19.8h, v20.8h, v2.h[4]\n" "ldr d20, [x17, #0x150]\n" + "mov v21.d[1], x21\n" + "ldr x21, [x17, #0x168]\n" "mov v20.d[1], x20\n" "fmla v8.8h, v21.8h, v0.h[5]\n" "fmla v12.8h, v21.8h, v1.h[5]\n" @@ -1681,11 +1694,11 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "fmla v16.8h, v21.8h, v2.h[5]\n" "ldr d21, [x17, #0x160]\n" "fmla v9.8h, v20.8h, v0.h[5]\n" - "mov v21.d[1], x21\n" "fmla v13.8h, v20.8h, v1.h[5]\n" - "ldr x21, [x17, #0x188]\n" "fmla v17.8h, v20.8h, v2.h[5]\n" "ldr d20, [x17, #0x170]\n" + "mov v21.d[1], x21\n" + "ldr x21, [x17, #0x188]\n" "mov v20.d[1], x20\n" "fmla v10.8h, v21.8h, v0.h[5]\n" "fmla v14.8h, v21.8h, v1.h[5]\n" @@ -1693,11 +1706,11 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "fmla v18.8h, v21.8h, v2.h[5]\n" "ldr d21, [x17, #0x180]\n" "fmla v11.8h, v20.8h, v0.h[5]\n" - "mov v21.d[1], x21\n" "fmla v15.8h, v20.8h, v1.h[5]\n" - "ldr x21, [x17, #0x1a8]\n" "fmla v19.8h, v20.8h, v2.h[5]\n" "ldr d20, [x17, #0x190]\n" + "mov v21.d[1], x21\n" + "ldr x21, [x17, #0x1a8]\n" "mov v20.d[1], x20\n" "fmla v8.8h, v21.8h, v0.h[6]\n" "fmla v12.8h, v21.8h, v1.h[6]\n" @@ -1705,11 +1718,11 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "fmla v16.8h, v21.8h, v2.h[6]\n" "ldr d21, [x17, #0x1a0]\n" "fmla v9.8h, v20.8h, v0.h[6]\n" - "mov v21.d[1], x21\n" "fmla v13.8h, v20.8h, v1.h[6]\n" - "ldr x21, [x17, #0x1c8]\n" "fmla v17.8h, v20.8h, v2.h[6]\n" "ldr d20, [x17, #0x1b0]\n" + "mov v21.d[1], x21\n" + "ldr x21, [x17, #0x1c8]\n" "mov v20.d[1], x20\n" "fmla v10.8h, v21.8h, v0.h[6]\n" "fmla v14.8h, v21.8h, v1.h[6]\n" @@ -1717,11 +1730,11 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "fmla v18.8h, v21.8h, v2.h[6]\n" "ldr d21, [x17, #0x1c0]\n" "fmla v11.8h, v20.8h, v0.h[6]\n" - "mov v21.d[1], x21\n" "fmla v15.8h, v20.8h, v1.h[6]\n" - "ldr x21, [x17, #0x1e8]\n" "fmla v19.8h, v20.8h, v2.h[6]\n" "ldr d20, [x17, #0x1d0]\n" + "mov v21.d[1], x21\n" + "ldr x21, [x17, #0x1e8]\n" "mov v20.d[1], x20\n" "fmla v8.8h, v21.8h, v0.h[7]\n" "fmla v12.8h, v21.8h, v1.h[7]\n" @@ -1729,40 +1742,29 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "fmla v16.8h, v21.8h, v2.h[7]\n" "ldr d21, [x17, #0x1e0]\n" "fmla v9.8h, v20.8h, v0.h[7]\n" - "mov v21.d[1], x21\n" "fmla v13.8h, v20.8h, v1.h[7]\n" - "add x13, x13, #0x10\n" "fmla v17.8h, v20.8h, v2.h[7]\n" "ldr d20, [x17, #0x1f0]\n" - "mov v20.d[1], x20\n" - "add x12, x12, #0x10\n" - "add x11, x11, #0x10\n" + "mov v21.d[1], x21\n" "add x17, x17, #0x200\n" + "ldr x21, [x17, #0x8]\n" + "mov v20.d[1], x20\n" "fmla v10.8h, v21.8h, v0.h[7]\n" - "ldr x20, [x17, #0x8]\n" "fmla v14.8h, v21.8h, v1.h[7]\n" - "ldr x23, [x13, #0x8]\n" + "ldr x20, [x17, #0x18]\n" "fmla v18.8h, v21.8h, v2.h[7]\n" "ldr d6, [x17, #0x0]\n" "fmla v11.8h, v20.8h, v0.h[7]\n" "ldr d0, [x13, #0x0]\n" "fmla v15.8h, v20.8h, v1.h[7]\n" "ldr d1, [x12, #0x0]\n" - "ldr x22, [x12, #0x8]\n" "fmla v19.8h, v20.8h, v2.h[7]\n" "ldr d2, [x11, #0x0]\n" - "sub x14, x14, #0x8\n" "ldr d7, [x17, #0x10]\n" - "cmp x14, #0x10\n" - "ldr x21, [x11, #0x8]\n" - "mov v6.d[1], x20\n" - "ldr x20, [x17, #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" + "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 124b\n" "125:" // Height 3: Multiply loop: Single iteration only @@ -1908,8 +1910,8 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "ldr h1, [x12], #0x2\n" "ldr h0, [x11], #0x2\n" "ldr q21, [x17, #0x0]\n" - "fmla v8.8h, v21.8h, v2.h[0]\n" "ldr q20, [x17, #0x10]\n" + "fmla v8.8h, v21.8h, v2.h[0]\n" "fmla v12.8h, v21.8h, v1.h[0]\n" "fmla v16.8h, v21.8h, v0.h[0]\n" "ldr q21, [x17, #0x20]\n" @@ -1931,28 +1933,28 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "cmp x15, x20\n" "bne 121b\n" "ldr x20, [%x[args_ptr], %[offsetof_output_offset]]\n" - "add x25, x16, x20, LSL #1\n" - "add x24, x25, x20, LSL #1\n" "prfm pstl1keep, [x16, #0x0]\n" + "add x26, x16, x20, LSL #1\n" + "add x25, x26, x20, LSL #1\n" + "prfm pstl1keep, [x26, #0x0]\n" "prfm pstl1keep, [x25, #0x0]\n" - "prfm pstl1keep, [x24, #0x0]\n" "tbz %x[flags], #1, 129f\n" - "add x20, %x[args_ptr], %[offset_max]\n" - "ld1r { v20.8h }, [x20]\n" - "fmin v8.8h, v8.8h, v20.8h\n" - "fmin v9.8h, v9.8h, v20.8h\n" - "fmin v10.8h, v10.8h, v20.8h\n" - "fmin v11.8h, v11.8h, v20.8h\n" - "fmin v12.8h, v12.8h, v20.8h\n" - "fmin v13.8h, v13.8h, v20.8h\n" - "fmin v14.8h, v14.8h, v20.8h\n" - "fmin v15.8h, v15.8h, v20.8h\n" - "fmin v16.8h, v16.8h, v20.8h\n" - "fmin v17.8h, v17.8h, v20.8h\n" - "fmin v18.8h, v18.8h, v20.8h\n" - "fmin v19.8h, v19.8h, v20.8h\n" + "add x21, %x[args_ptr], %[offset_max]\n" "add x20, %x[args_ptr], %[offset_min]\n" + "ld1r { v21.8h }, [x21]\n" "ld1r { v20.8h }, [x20]\n" + "fmin v8.8h, v8.8h, v21.8h\n" + "fmin v9.8h, v9.8h, v21.8h\n" + "fmin v10.8h, v10.8h, v21.8h\n" + "fmin v11.8h, v11.8h, v21.8h\n" + "fmin v12.8h, v12.8h, v21.8h\n" + "fmin v13.8h, v13.8h, v21.8h\n" + "fmin v14.8h, v14.8h, v21.8h\n" + "fmin v15.8h, v15.8h, v21.8h\n" + "fmin v16.8h, v16.8h, v21.8h\n" + "fmin v17.8h, v17.8h, v21.8h\n" + "fmin v18.8h, v18.8h, v21.8h\n" + "fmin v19.8h, v19.8h, v21.8h\n" "fmax v8.8h, v8.8h, v20.8h\n" "fmax v9.8h, v9.8h, v20.8h\n" "fmax v10.8h, v10.8h, v20.8h\n" @@ -1971,159 +1973,159 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "tbz x8, #4, 137f\n" "st1 { v8.8h }, [x16], #0x10\n" "st1 { v9.8h }, [x16], #0x10\n" - "st1 { v12.8h }, [x25], #0x10\n" - "st1 { v13.8h }, [x25], #0x10\n" - "st1 { v16.8h }, [x24], #0x10\n" - "st1 { v17.8h }, [x24], #0x10\n" + "st1 { v12.8h }, [x26], #0x10\n" + "st1 { v13.8h }, [x26], #0x10\n" + "st1 { v16.8h }, [x25], #0x10\n" + "st1 { v17.8h }, [x25], #0x10\n" "tbz x8, #3, 133f\n" "st1 { v10.8h }, [x16], #0x10\n" - "st1 { v14.8h }, [x25], #0x10\n" - "st1 { v18.8h }, [x24], #0x10\n" + "st1 { v14.8h }, [x26], #0x10\n" + "st1 { v18.8h }, [x25], #0x10\n" "tbz x8, #2, 131f\n" "str d11, [x16], #0x8\n" - "str d15, [x25], #0x8\n" - "str d19, [x24], #0x8\n" + "str d15, [x26], #0x8\n" + "str d19, [x25], #0x8\n" "tbz x8, #1, 130f\n" "st1 { v11.s }[2], [x16], #0x4\n" - "st1 { v15.s }[2], [x25], #0x4\n" - "st1 { v19.s }[2], [x24], #0x4\n" + "st1 { v15.s }[2], [x26], #0x4\n" + "st1 { v19.s }[2], [x25], #0x4\n" "tbz x8, #0, 145f\n" "st1 { v11.h }[6], [x16]\n" - "st1 { v15.h }[6], [x25]\n" - "st1 { v19.h }[6], [x24]\n" + "st1 { v15.h }[6], [x26]\n" + "st1 { v19.h }[6], [x25]\n" "b 145f\n" "130:" // Height 3: Partial direct writeback: partial_1_28 "tbz x8, #0, 145f\n" "st1 { v11.h }[4], [x16]\n" - "st1 { v15.h }[4], [x25]\n" - "st1 { v19.h }[4], [x24]\n" + "st1 { v15.h }[4], [x26]\n" + "st1 { v19.h }[4], [x25]\n" "b 145f\n" "131:" // Height 3: Partial direct writeback: partial_2_24 "tbz x8, #1, 132f\n" "str s11, [x16], #0x4\n" - "str s15, [x25], #0x4\n" - "str s19, [x24], #0x4\n" + "str s15, [x26], #0x4\n" + "str s19, [x25], #0x4\n" "tbz x8, #0, 145f\n" "st1 { v11.h }[2], [x16]\n" - "st1 { v15.h }[2], [x25]\n" - "st1 { v19.h }[2], [x24]\n" + "st1 { v15.h }[2], [x26]\n" + "st1 { v19.h }[2], [x25]\n" "b 145f\n" "132:" // Height 3: Partial direct writeback: partial_1_24 "tbz x8, #0, 145f\n" "str h11, [x16, #0x0]\n" - "str h15, [x25, #0x0]\n" - "str h19, [x24, #0x0]\n" + "str h15, [x26, #0x0]\n" + "str h19, [x25, #0x0]\n" "b 145f\n" "133:" // Height 3: Partial direct writeback: partial_4_16 "tbz x8, #2, 135f\n" "str d10, [x16], #0x8\n" - "str d14, [x25], #0x8\n" - "str d18, [x24], #0x8\n" + "str d14, [x26], #0x8\n" + "str d18, [x25], #0x8\n" "tbz x8, #1, 134f\n" "st1 { v10.s }[2], [x16], #0x4\n" - "st1 { v14.s }[2], [x25], #0x4\n" - "st1 { v18.s }[2], [x24], #0x4\n" + "st1 { v14.s }[2], [x26], #0x4\n" + "st1 { v18.s }[2], [x25], #0x4\n" "tbz x8, #0, 145f\n" "st1 { v10.h }[6], [x16]\n" - "st1 { v14.h }[6], [x25]\n" - "st1 { v18.h }[6], [x24]\n" + "st1 { v14.h }[6], [x26]\n" + "st1 { v18.h }[6], [x25]\n" "b 145f\n" "134:" // Height 3: Partial direct writeback: partial_1_20 "tbz x8, #0, 145f\n" "st1 { v10.h }[4], [x16]\n" - "st1 { v14.h }[4], [x25]\n" - "st1 { v18.h }[4], [x24]\n" + "st1 { v14.h }[4], [x26]\n" + "st1 { v18.h }[4], [x25]\n" "b 145f\n" "135:" // Height 3: Partial direct writeback: partial_2_16 "tbz x8, #1, 136f\n" "str s10, [x16], #0x4\n" - "str s14, [x25], #0x4\n" - "str s18, [x24], #0x4\n" + "str s14, [x26], #0x4\n" + "str s18, [x25], #0x4\n" "tbz x8, #0, 145f\n" "st1 { v10.h }[2], [x16]\n" - "st1 { v14.h }[2], [x25]\n" - "st1 { v18.h }[2], [x24]\n" + "st1 { v14.h }[2], [x26]\n" + "st1 { v18.h }[2], [x25]\n" "b 145f\n" "136:" // Height 3: Partial direct writeback: partial_1_16 "tbz x8, #0, 145f\n" "str h10, [x16, #0x0]\n" - "str h14, [x25, #0x0]\n" - "str h18, [x24, #0x0]\n" + "str h14, [x26, #0x0]\n" + "str h18, [x25, #0x0]\n" "b 145f\n" "137:" // Height 3: Partial direct writeback: partial_8_0 "tbz x8, #3, 141f\n" "st1 { v8.8h }, [x16], #0x10\n" - "st1 { v12.8h }, [x25], #0x10\n" - "st1 { v16.8h }, [x24], #0x10\n" + "st1 { v12.8h }, [x26], #0x10\n" + "st1 { v16.8h }, [x25], #0x10\n" "tbz x8, #2, 139f\n" "str d9, [x16], #0x8\n" - "str d13, [x25], #0x8\n" - "str d17, [x24], #0x8\n" + "str d13, [x26], #0x8\n" + "str d17, [x25], #0x8\n" "tbz x8, #1, 138f\n" "st1 { v9.s }[2], [x16], #0x4\n" - "st1 { v13.s }[2], [x25], #0x4\n" - "st1 { v17.s }[2], [x24], #0x4\n" + "st1 { v13.s }[2], [x26], #0x4\n" + "st1 { v17.s }[2], [x25], #0x4\n" "tbz x8, #0, 145f\n" "st1 { v9.h }[6], [x16]\n" - "st1 { v13.h }[6], [x25]\n" - "st1 { v17.h }[6], [x24]\n" + "st1 { v13.h }[6], [x26]\n" + "st1 { v17.h }[6], [x25]\n" "b 145f\n" "138:" // Height 3: Partial direct writeback: partial_1_12 "tbz x8, #0, 145f\n" "st1 { v9.h }[4], [x16]\n" - "st1 { v13.h }[4], [x25]\n" - "st1 { v17.h }[4], [x24]\n" + "st1 { v13.h }[4], [x26]\n" + "st1 { v17.h }[4], [x25]\n" "b 145f\n" "139:" // Height 3: Partial direct writeback: partial_2_8 "tbz x8, #1, 140f\n" "str s9, [x16], #0x4\n" - "str s13, [x25], #0x4\n" - "str s17, [x24], #0x4\n" + "str s13, [x26], #0x4\n" + "str s17, [x25], #0x4\n" "tbz x8, #0, 145f\n" "st1 { v9.h }[2], [x16]\n" - "st1 { v13.h }[2], [x25]\n" - "st1 { v17.h }[2], [x24]\n" + "st1 { v13.h }[2], [x26]\n" + "st1 { v17.h }[2], [x25]\n" "b 145f\n" "140:" // Height 3: Partial direct writeback: partial_1_8 "tbz x8, #0, 145f\n" "str h9, [x16, #0x0]\n" - "str h13, [x25, #0x0]\n" - "str h17, [x24, #0x0]\n" + "str h13, [x26, #0x0]\n" + "str h17, [x25, #0x0]\n" "b 145f\n" "141:" // Height 3: Partial direct writeback: partial_4_0 "tbz x8, #2, 143f\n" "str d8, [x16], #0x8\n" - "str d12, [x25], #0x8\n" - "str d16, [x24], #0x8\n" + "str d12, [x26], #0x8\n" + "str d16, [x25], #0x8\n" "tbz x8, #1, 142f\n" "st1 { v8.s }[2], [x16], #0x4\n" - "st1 { v12.s }[2], [x25], #0x4\n" - "st1 { v16.s }[2], [x24], #0x4\n" + "st1 { v12.s }[2], [x26], #0x4\n" + "st1 { v16.s }[2], [x25], #0x4\n" "tbz x8, #0, 145f\n" "st1 { v8.h }[6], [x16]\n" - "st1 { v12.h }[6], [x25]\n" - "st1 { v16.h }[6], [x24]\n" + "st1 { v12.h }[6], [x26]\n" + "st1 { v16.h }[6], [x25]\n" "b 145f\n" "142:" // Height 3: Partial direct writeback: partial_1_4 "tbz x8, #0, 145f\n" "st1 { v8.h }[4], [x16]\n" - "st1 { v12.h }[4], [x25]\n" - "st1 { v16.h }[4], [x24]\n" + "st1 { v12.h }[4], [x26]\n" + "st1 { v16.h }[4], [x25]\n" "b 145f\n" "143:" // Height 3: Partial direct writeback: partial_2_0 "tbz x8, #1, 144f\n" "str s8, [x16], #0x4\n" - "str s12, [x25], #0x4\n" - "str s16, [x24], #0x4\n" + "str s12, [x26], #0x4\n" + "str s16, [x25], #0x4\n" "tbz x8, #0, 145f\n" "st1 { v8.h }[2], [x16]\n" - "st1 { v12.h }[2], [x25]\n" - "st1 { v16.h }[2], [x24]\n" + "st1 { v12.h }[2], [x26]\n" + "st1 { v16.h }[2], [x25]\n" "b 145f\n" "144:" // Height 3: Partial direct writeback: partial_1_0 "str h8, [x16, #0x0]\n" - "str h12, [x25, #0x0]\n" - "str h16, [x24, #0x0]\n" + "str h12, [x26, #0x0]\n" + "str h16, [x25, #0x0]\n" "145:" // Height 3: Partial direct writeback: Done "b 147f\n" "146:" // Height 3: Full writeback @@ -2132,35 +2134,35 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "str q10, [x16, #0x20]\n" "str q11, [x16, #0x30]\n" "add x16, x16, #0x40\n" - "str q12, [x25, #0x0]\n" - "str q13, [x25, #0x10]\n" - "str q14, [x25, #0x20]\n" - "str q15, [x25, #0x30]\n" - "str q16, [x24, #0x0]\n" - "str q17, [x24, #0x10]\n" - "str q18, [x24, #0x20]\n" - "str q19, [x24, #0x30]\n" + "str q12, [x26, #0x0]\n" + "str q13, [x26, #0x10]\n" + "str q14, [x26, #0x20]\n" + "str q15, [x26, #0x30]\n" + "str q16, [x25, #0x0]\n" + "str q17, [x25, #0x10]\n" + "str q18, [x25, #0x20]\n" + "str q19, [x25, #0x30]\n" "147:" // Height 3: Writeback done "subs x8, x8, #0x20\n" "bgt 100b\n" "b 296f\n" "148:" // Height 4 - "mov x7, %x[bias]\n" + "ldr x7, [%x[args_ptr], %[offsetof_bias]]\n" "ldr x8, [%x[args_ptr], %[offsetof_N]]\n" "ldr x17, [%x[args_ptr], %[offsetof_B_ptr]]\n" - "mov x16, %x[output_ptr]\n" + "ldr x16, [%x[args_ptr], %[offsetof_output_ptr]]\n" "149:" // Height 4: Column loop "cbz x7, 150f\n" "ldr q8, [x7, #0x0]\n" - "mov v12.16b, v8.16b\n" "ldr q9, [x7, #0x10]\n" - "mov v13.16b, v9.16b\n" "ldr q10, [x7, #0x20]\n" - "mov v14.16b, v10.16b\n" "ldr q11, [x7, #0x30]\n" + "mov v12.16b, v8.16b\n" + "mov v13.16b, v9.16b\n" + "add x7, x7, #0x40\n" + "mov v14.16b, v10.16b\n" "mov v15.16b, v11.16b\n" "mov v16.16b, v8.16b\n" - "add x7, x7, #0x40\n" "mov v17.16b, v9.16b\n" "mov v18.16b, v10.16b\n" "mov v19.16b, v11.16b\n" @@ -2172,215 +2174,215 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "150:" // Height 4: no bias "tbz %x[flags], #0, 168f\n" "ldr x20, [%x[args_ptr], %[offsetof_output_offset]]\n" - "add x25, x16, x20, LSL #1\n" - "add x24, x25, x20, LSL #1\n" "cmp x8, #0x20\n" - "add x23, x24, x20, LSL #1\n" + "add x26, x16, x20, LSL #1\n" + "add x25, x26, x20, LSL #1\n" + "add x24, x25, x20, LSL #1\n" "bge 167f\n" "tbz x8, #4, 158f\n" "ld1 { v8.8h }, [x16], #0x10\n" - "ld1 { v12.8h }, [x25], #0x10\n" - "ld1 { v16.8h }, [x24], #0x10\n" - "ld1 { v20.8h }, [x23], #0x10\n" + "ld1 { v12.8h }, [x26], #0x10\n" + "ld1 { v16.8h }, [x25], #0x10\n" + "ld1 { v20.8h }, [x24], #0x10\n" "ld1 { v9.8h }, [x16], #0x10\n" - "ld1 { v13.8h }, [x25], #0x10\n" - "ld1 { v17.8h }, [x24], #0x10\n" - "ld1 { v21.8h }, [x23], #0x10\n" + "ld1 { v13.8h }, [x26], #0x10\n" + "ld1 { v17.8h }, [x25], #0x10\n" + "ld1 { v21.8h }, [x24], #0x10\n" "tbz x8, #3, 154f\n" "ld1 { v10.8h }, [x16], #0x10\n" - "ld1 { v14.8h }, [x25], #0x10\n" - "ld1 { v18.8h }, [x24], #0x10\n" - "ld1 { v22.8h }, [x23], #0x10\n" + "ld1 { v14.8h }, [x26], #0x10\n" + "ld1 { v18.8h }, [x25], #0x10\n" + "ld1 { v22.8h }, [x24], #0x10\n" "tbz x8, #2, 152f\n" "ldr d11, [x16], #0x8\n" - "ldr d15, [x25], #0x8\n" - "ldr d19, [x24], #0x8\n" - "ldr d23, [x23], #0x8\n" + "ldr d15, [x26], #0x8\n" + "ldr d19, [x25], #0x8\n" + "ldr d23, [x24], #0x8\n" "tbz x8, #1, 151f\n" "ld1 { v11.s }[2], [x16], #0x4\n" "mov x20, #0x3c\n" - "ld1 { v15.s }[2], [x25], #0x4\n" - "ld1 { v19.s }[2], [x24], #0x4\n" - "ld1 { v23.s }[2], [x23], #0x4\n" + "ld1 { v15.s }[2], [x26], #0x4\n" + "ld1 { v19.s }[2], [x25], #0x4\n" + "ld1 { v23.s }[2], [x24], #0x4\n" "tbz x8, #0, 166f\n" "ld1 { v11.h }[6], [x16]\n" - "ld1 { v15.h }[6], [x25]\n" - "ld1 { v19.h }[6], [x24]\n" - "ld1 { v23.h }[6], [x23]\n" + "ld1 { v15.h }[6], [x26]\n" + "ld1 { v19.h }[6], [x25]\n" + "ld1 { v23.h }[6], [x24]\n" "b 166f\n" "151:" // Height 4: Partial accumulate: partial_1_28 "mov x20, #0x38\n" "tbz x8, #0, 166f\n" "ld1 { v11.h }[4], [x16]\n" - "ld1 { v15.h }[4], [x25]\n" - "ld1 { v19.h }[4], [x24]\n" - "ld1 { v23.h }[4], [x23]\n" + "ld1 { v15.h }[4], [x26]\n" + "ld1 { v19.h }[4], [x25]\n" + "ld1 { v23.h }[4], [x24]\n" "b 166f\n" "152:" // Height 4: Partial accumulate: partial_2_24 "tbz x8, #1, 153f\n" "ldr s11, [x16], #0x4\n" "mov x20, #0x34\n" - "ldr s15, [x25], #0x4\n" - "ldr s19, [x24], #0x4\n" - "ldr s23, [x23], #0x4\n" + "ldr s15, [x26], #0x4\n" + "ldr s19, [x25], #0x4\n" + "ldr s23, [x24], #0x4\n" "tbz x8, #0, 166f\n" "ld1 { v11.h }[2], [x16]\n" - "ld1 { v15.h }[2], [x25]\n" - "ld1 { v19.h }[2], [x24]\n" - "ld1 { v23.h }[2], [x23]\n" + "ld1 { v15.h }[2], [x26]\n" + "ld1 { v19.h }[2], [x25]\n" + "ld1 { v23.h }[2], [x24]\n" "b 166f\n" "153:" // Height 4: Partial accumulate: partial_1_24 "mov x20, #0x30\n" "tbz x8, #0, 166f\n" "ldr h11, [x16, #0x0]\n" - "ldr h15, [x25, #0x0]\n" - "ldr h19, [x24, #0x0]\n" - "ldr h23, [x23, #0x0]\n" + "ldr h15, [x26, #0x0]\n" + "ldr h19, [x25, #0x0]\n" + "ldr h23, [x24, #0x0]\n" "b 166f\n" "154:" // Height 4: Partial accumulate: partial_4_16 "tbz x8, #2, 156f\n" "ldr d10, [x16], #0x8\n" - "ldr d14, [x25], #0x8\n" - "ldr d18, [x24], #0x8\n" - "ldr d22, [x23], #0x8\n" + "ldr d14, [x26], #0x8\n" + "ldr d18, [x25], #0x8\n" + "ldr d22, [x24], #0x8\n" "tbz x8, #1, 155f\n" "ld1 { v10.s }[2], [x16], #0x4\n" "mov x20, #0x2c\n" - "ld1 { v14.s }[2], [x25], #0x4\n" - "ld1 { v18.s }[2], [x24], #0x4\n" - "ld1 { v22.s }[2], [x23], #0x4\n" + "ld1 { v14.s }[2], [x26], #0x4\n" + "ld1 { v18.s }[2], [x25], #0x4\n" + "ld1 { v22.s }[2], [x24], #0x4\n" "tbz x8, #0, 166f\n" "ld1 { v10.h }[6], [x16]\n" - "ld1 { v14.h }[6], [x25]\n" - "ld1 { v18.h }[6], [x24]\n" - "ld1 { v22.h }[6], [x23]\n" + "ld1 { v14.h }[6], [x26]\n" + "ld1 { v18.h }[6], [x25]\n" + "ld1 { v22.h }[6], [x24]\n" "b 166f\n" "155:" // Height 4: Partial accumulate: partial_1_20 "mov x20, #0x28\n" "tbz x8, #0, 166f\n" "ld1 { v10.h }[4], [x16]\n" - "ld1 { v14.h }[4], [x25]\n" - "ld1 { v18.h }[4], [x24]\n" - "ld1 { v22.h }[4], [x23]\n" + "ld1 { v14.h }[4], [x26]\n" + "ld1 { v18.h }[4], [x25]\n" + "ld1 { v22.h }[4], [x24]\n" "b 166f\n" "156:" // Height 4: Partial accumulate: partial_2_16 "tbz x8, #1, 157f\n" "ldr s10, [x16], #0x4\n" "mov x20, #0x24\n" - "ldr s14, [x25], #0x4\n" - "ldr s18, [x24], #0x4\n" - "ldr s22, [x23], #0x4\n" + "ldr s14, [x26], #0x4\n" + "ldr s18, [x25], #0x4\n" + "ldr s22, [x24], #0x4\n" "tbz x8, #0, 166f\n" "ld1 { v10.h }[2], [x16]\n" - "ld1 { v14.h }[2], [x25]\n" - "ld1 { v18.h }[2], [x24]\n" - "ld1 { v22.h }[2], [x23]\n" + "ld1 { v14.h }[2], [x26]\n" + "ld1 { v18.h }[2], [x25]\n" + "ld1 { v22.h }[2], [x24]\n" "b 166f\n" "157:" // Height 4: Partial accumulate: partial_1_16 "mov x20, #0x20\n" "tbz x8, #0, 166f\n" "ldr h10, [x16, #0x0]\n" - "ldr h14, [x25, #0x0]\n" - "ldr h18, [x24, #0x0]\n" - "ldr h22, [x23, #0x0]\n" + "ldr h14, [x26, #0x0]\n" + "ldr h18, [x25, #0x0]\n" + "ldr h22, [x24, #0x0]\n" "b 166f\n" "158:" // Height 4: Partial accumulate: partial_8_0 "tbz x8, #3, 162f\n" "ld1 { v8.8h }, [x16], #0x10\n" - "ld1 { v12.8h }, [x25], #0x10\n" - "ld1 { v16.8h }, [x24], #0x10\n" - "ld1 { v20.8h }, [x23], #0x10\n" + "ld1 { v12.8h }, [x26], #0x10\n" + "ld1 { v16.8h }, [x25], #0x10\n" + "ld1 { v20.8h }, [x24], #0x10\n" "tbz x8, #2, 160f\n" "ldr d9, [x16], #0x8\n" - "ldr d13, [x25], #0x8\n" - "ldr d17, [x24], #0x8\n" - "ldr d21, [x23], #0x8\n" + "ldr d13, [x26], #0x8\n" + "ldr d17, [x25], #0x8\n" + "ldr d21, [x24], #0x8\n" "tbz x8, #1, 159f\n" "ld1 { v9.s }[2], [x16], #0x4\n" "mov x20, #0x1c\n" - "ld1 { v13.s }[2], [x25], #0x4\n" - "ld1 { v17.s }[2], [x24], #0x4\n" - "ld1 { v21.s }[2], [x23], #0x4\n" + "ld1 { v13.s }[2], [x26], #0x4\n" + "ld1 { v17.s }[2], [x25], #0x4\n" + "ld1 { v21.s }[2], [x24], #0x4\n" "tbz x8, #0, 166f\n" "ld1 { v9.h }[6], [x16]\n" - "ld1 { v13.h }[6], [x25]\n" - "ld1 { v17.h }[6], [x24]\n" - "ld1 { v21.h }[6], [x23]\n" + "ld1 { v13.h }[6], [x26]\n" + "ld1 { v17.h }[6], [x25]\n" + "ld1 { v21.h }[6], [x24]\n" "b 166f\n" "159:" // Height 4: Partial accumulate: partial_1_12 "mov x20, #0x18\n" "tbz x8, #0, 166f\n" "ld1 { v9.h }[4], [x16]\n" - "ld1 { v13.h }[4], [x25]\n" - "ld1 { v17.h }[4], [x24]\n" - "ld1 { v21.h }[4], [x23]\n" + "ld1 { v13.h }[4], [x26]\n" + "ld1 { v17.h }[4], [x25]\n" + "ld1 { v21.h }[4], [x24]\n" "b 166f\n" "160:" // Height 4: Partial accumulate: partial_2_8 "tbz x8, #1, 161f\n" "ldr s9, [x16], #0x4\n" "mov x20, #0x14\n" - "ldr s13, [x25], #0x4\n" - "ldr s17, [x24], #0x4\n" - "ldr s21, [x23], #0x4\n" + "ldr s13, [x26], #0x4\n" + "ldr s17, [x25], #0x4\n" + "ldr s21, [x24], #0x4\n" "tbz x8, #0, 166f\n" "ld1 { v9.h }[2], [x16]\n" - "ld1 { v13.h }[2], [x25]\n" - "ld1 { v17.h }[2], [x24]\n" - "ld1 { v21.h }[2], [x23]\n" + "ld1 { v13.h }[2], [x26]\n" + "ld1 { v17.h }[2], [x25]\n" + "ld1 { v21.h }[2], [x24]\n" "b 166f\n" "161:" // Height 4: Partial accumulate: partial_1_8 "mov x20, #0x10\n" "tbz x8, #0, 166f\n" "ldr h9, [x16, #0x0]\n" - "ldr h13, [x25, #0x0]\n" - "ldr h17, [x24, #0x0]\n" - "ldr h21, [x23, #0x0]\n" + "ldr h13, [x26, #0x0]\n" + "ldr h17, [x25, #0x0]\n" + "ldr h21, [x24, #0x0]\n" "b 166f\n" "162:" // Height 4: Partial accumulate: partial_4_0 "tbz x8, #2, 164f\n" "ldr d8, [x16], #0x8\n" - "ldr d12, [x25], #0x8\n" - "ldr d16, [x24], #0x8\n" - "ldr d20, [x23], #0x8\n" + "ldr d12, [x26], #0x8\n" + "ldr d16, [x25], #0x8\n" + "ldr d20, [x24], #0x8\n" "tbz x8, #1, 163f\n" "ld1 { v8.s }[2], [x16], #0x4\n" "mov x20, #0xc\n" - "ld1 { v12.s }[2], [x25], #0x4\n" - "ld1 { v16.s }[2], [x24], #0x4\n" - "ld1 { v20.s }[2], [x23], #0x4\n" + "ld1 { v12.s }[2], [x26], #0x4\n" + "ld1 { v16.s }[2], [x25], #0x4\n" + "ld1 { v20.s }[2], [x24], #0x4\n" "tbz x8, #0, 166f\n" "ld1 { v8.h }[6], [x16]\n" - "ld1 { v12.h }[6], [x25]\n" - "ld1 { v16.h }[6], [x24]\n" - "ld1 { v20.h }[6], [x23]\n" + "ld1 { v12.h }[6], [x26]\n" + "ld1 { v16.h }[6], [x25]\n" + "ld1 { v20.h }[6], [x24]\n" "b 166f\n" "163:" // Height 4: Partial accumulate: partial_1_4 "mov x20, #0x8\n" "tbz x8, #0, 166f\n" "ld1 { v8.h }[4], [x16]\n" - "ld1 { v12.h }[4], [x25]\n" - "ld1 { v16.h }[4], [x24]\n" - "ld1 { v20.h }[4], [x23]\n" + "ld1 { v12.h }[4], [x26]\n" + "ld1 { v16.h }[4], [x25]\n" + "ld1 { v20.h }[4], [x24]\n" "b 166f\n" "164:" // Height 4: Partial accumulate: partial_2_0 "tbz x8, #1, 165f\n" "ldr s8, [x16], #0x4\n" "mov x20, #0x4\n" - "ldr s12, [x25], #0x4\n" - "ldr s16, [x24], #0x4\n" - "ldr s20, [x23], #0x4\n" + "ldr s12, [x26], #0x4\n" + "ldr s16, [x25], #0x4\n" + "ldr s20, [x24], #0x4\n" "tbz x8, #0, 166f\n" "ld1 { v8.h }[2], [x16]\n" - "ld1 { v12.h }[2], [x25]\n" - "ld1 { v16.h }[2], [x24]\n" - "ld1 { v20.h }[2], [x23]\n" + "ld1 { v12.h }[2], [x26]\n" + "ld1 { v16.h }[2], [x25]\n" + "ld1 { v20.h }[2], [x24]\n" "b 166f\n" "165:" // Height 4: Partial accumulate: partial_1_0 "ldr h8, [x16, #0x0]\n" "mov x20, #0x0\n" - "ldr h12, [x25, #0x0]\n" - "ldr h16, [x24, #0x0]\n" - "ldr h20, [x23, #0x0]\n" + "ldr h12, [x26, #0x0]\n" + "ldr h16, [x25, #0x0]\n" + "ldr h20, [x24, #0x0]\n" "166:" // Height 4: Partial accumulate: Done "sub x16, x16, x20\n" "b 169f\n" @@ -2389,18 +2391,18 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "ldr q9, [x16, #0x10]\n" "ldr q10, [x16, #0x20]\n" "ldr q11, [x16, #0x30]\n" - "ldr q12, [x25, #0x0]\n" - "ldr q13, [x25, #0x10]\n" - "ldr q14, [x25, #0x20]\n" - "ldr q15, [x25, #0x30]\n" - "ldr q16, [x24, #0x0]\n" - "ldr q17, [x24, #0x10]\n" - "ldr q18, [x24, #0x20]\n" - "ldr q19, [x24, #0x30]\n" - "ldr q20, [x23, #0x0]\n" - "ldr q21, [x23, #0x10]\n" - "ldr q22, [x23, #0x20]\n" - "ldr q23, [x23, #0x30]\n" + "ldr q12, [x26, #0x0]\n" + "ldr q13, [x26, #0x10]\n" + "ldr q14, [x26, #0x20]\n" + "ldr q15, [x26, #0x30]\n" + "ldr q16, [x25, #0x0]\n" + "ldr q17, [x25, #0x10]\n" + "ldr q18, [x25, #0x20]\n" + "ldr q19, [x25, #0x30]\n" + "ldr q20, [x24, #0x0]\n" + "ldr q21, [x24, #0x10]\n" + "ldr q22, [x24, #0x20]\n" + "ldr q23, [x24, #0x30]\n" "b 169f\n" "168:" // Height 4: no accumulate "movi v8.16b, #0x0\n" @@ -2423,8 +2425,8 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "mov x15, #0x0\n" "170:" // 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, 171f\n" "ldr x20, [%x[input_ptr], x15, LSL #0x3]\n" "add x20, x20, x21, LSL #3\n" @@ -2457,234 +2459,233 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "blt 174f\n" "173:" // Height 4: Multiply loop: Main loop head "fmla v8.8h, v6.8h, v0.h[0]\n" - "ldr x21, [x17, #0x28]\n" + "ldr x20, [x17, #0x28]\n" "fmla v12.8h, v6.8h, v1.h[0]\n" - "ldr x20, [x17, #0x38]\n" + "ldr x21, [x17, #0x38]\n" "fmla v16.8h, v6.8h, v2.h[0]\n" "add x13, x13, #0x10\n" "fmla v20.8h, v6.8h, v3.h[0]\n" "ldr d25, [x17, #0x20]\n" "fmla v9.8h, v7.8h, v0.h[0]\n" - "mov v25.d[1], x21\n" + "add x12, x12, #0x10\n" "fmla v13.8h, v7.8h, v1.h[0]\n" - "ldr x21, [x17, #0x48]\n" + "add x11, x11, #0x10\n" "fmla v17.8h, v7.8h, v2.h[0]\n" - "add x12, x12, #0x10\n" + "mov v25.d[1], x20\n" "fmla v21.8h, v7.8h, v3.h[0]\n" "ldr d24, [x17, #0x30]\n" - "mov v24.d[1], x20\n" "fmla v10.8h, v25.8h, v0.h[0]\n" + "ldr x20, [x17, #0x48]\n" "fmla v14.8h, v25.8h, v1.h[0]\n" - "ldr x20, [x17, #0x58]\n" + "add x10, x10, #0x10\n" + "mov v24.d[1], x21\n" "fmla v18.8h, v25.8h, v2.h[0]\n" - "add x11, x11, #0x10\n" "fmla v22.8h, v25.8h, v3.h[0]\n" "ldr d25, [x17, #0x40]\n" "fmla v11.8h, v24.8h, v0.h[0]\n" - "mov v25.d[1], x21\n" + "ldr x21, [x17, #0x58]\n" "fmla v15.8h, v24.8h, v1.h[0]\n" - "ldr x21, [x17, #0x68]\n" + "ldr x25, [x13, #0x8]\n" "fmla v19.8h, v24.8h, v2.h[0]\n" - "add x10, x10, #0x10\n" + "mov v25.d[1], x20\n" "fmla v23.8h, v24.8h, v3.h[0]\n" "ldr d24, [x17, #0x50]\n" - "mov v24.d[1], x20\n" "fmla v8.8h, v25.8h, v0.h[1]\n" + "ldr x20, [x17, #0x68]\n" "fmla v12.8h, v25.8h, v1.h[1]\n" - "ldr x20, [x17, #0x78]\n" + "ldr x24, [x12, #0x8]\n" + "mov v24.d[1], x21\n" "fmla v16.8h, v25.8h, v2.h[1]\n" - "ldr x25, [x13, #0x8]\n" "fmla v20.8h, v25.8h, v3.h[1]\n" "ldr d25, [x17, #0x60]\n" "fmla v9.8h, v24.8h, v0.h[1]\n" - "mov v25.d[1], x21\n" + "ldr x21, [x17, #0x78]\n" "fmla v13.8h, v24.8h, v1.h[1]\n" - "ldr x21, [x17, #0x88]\n" + "ldr x23, [x11, #0x8]\n" "fmla v17.8h, v24.8h, v2.h[1]\n" - "ldr x24, [x12, #0x8]\n" + "mov v25.d[1], x20\n" "fmla v21.8h, v24.8h, v3.h[1]\n" "ldr d24, [x17, #0x70]\n" - "mov v24.d[1], x20\n" "fmla v10.8h, v25.8h, v0.h[1]\n" + "ldr x20, [x17, #0x88]\n" "fmla v14.8h, v25.8h, v1.h[1]\n" - "ldr x20, [x17, #0x98]\n" + "ldr x22, [x10, #0x8]\n" + "mov v24.d[1], x21\n" "fmla v18.8h, v25.8h, v2.h[1]\n" - "ldr x23, [x11, #0x8]\n" "fmla v22.8h, v25.8h, v3.h[1]\n" "ldr d25, [x17, #0x80]\n" "fmla v11.8h, v24.8h, v0.h[1]\n" - "mov v25.d[1], x21\n" + "ldr x21, [x17, #0x98]\n" "fmla v15.8h, v24.8h, v1.h[1]\n" - "ldr x21, [x17, #0xa8]\n" + "sub x14, x14, #0x8\n" "fmla v19.8h, v24.8h, v2.h[1]\n" - "ldr x22, [x10, #0x8]\n" + "mov v25.d[1], x20\n" "fmla v23.8h, v24.8h, v3.h[1]\n" "ldr d24, [x17, #0x90]\n" - "mov v24.d[1], x20\n" "fmla v8.8h, v25.8h, v0.h[2]\n" + "ldr x20, [x17, #0xa8]\n" "fmla v12.8h, v25.8h, v1.h[2]\n" - "ldr x20, [x17, #0xb8]\n" + "cmp x14, #0x10\n" + "mov v24.d[1], x21\n" "fmla v16.8h, v25.8h, v2.h[2]\n" - "sub x14, x14, #0x8\n" "fmla v20.8h, v25.8h, v3.h[2]\n" "ldr d25, [x17, #0xa0]\n" "fmla v9.8h, v24.8h, v0.h[2]\n" - "mov v25.d[1], x21\n" + "ldr x21, [x17, #0xb8]\n" "fmla v13.8h, v24.8h, v1.h[2]\n" - "ldr x21, [x17, #0xc8]\n" + "prfm pldl1keep, [x13, #0x80]\n" "fmla v17.8h, v24.8h, v2.h[2]\n" - "cmp x14, #0x10\n" + "mov v25.d[1], x20\n" "fmla v21.8h, v24.8h, v3.h[2]\n" "ldr d24, [x17, #0xb0]\n" - "mov v24.d[1], x20\n" "fmla v10.8h, v25.8h, v0.h[2]\n" + "ldr x20, [x17, #0xc8]\n" "fmla v14.8h, v25.8h, v1.h[2]\n" - "ldr x20, [x17, #0xd8]\n" + "prfm pldl1keep, [x12, #0x80]\n" + "mov v24.d[1], x21\n" "fmla v18.8h, v25.8h, v2.h[2]\n" - "prfm pldl1keep, [x13, #0x80]\n" "fmla v22.8h, v25.8h, v3.h[2]\n" "ldr d25, [x17, #0xc0]\n" "fmla v11.8h, v24.8h, v0.h[2]\n" - "mov v25.d[1], x21\n" + "ldr x21, [x17, #0xd8]\n" "fmla v15.8h, v24.8h, v1.h[2]\n" - "ldr x21, [x17, #0xe8]\n" + "prfm pldl1keep, [x11, #0x80]\n" "fmla v19.8h, v24.8h, v2.h[2]\n" - "prfm pldl1keep, [x12, #0x80]\n" + "mov v25.d[1], x20\n" "fmla v23.8h, v24.8h, v3.h[2]\n" "ldr d24, [x17, #0xd0]\n" - "mov v24.d[1], x20\n" "fmla v8.8h, v25.8h, v0.h[3]\n" + "ldr x20, [x17, #0xe8]\n" "fmla v12.8h, v25.8h, v1.h[3]\n" - "ldr x20, [x17, #0xf8]\n" + "prfm pldl1keep, [x10, #0x80]\n" + "mov v24.d[1], x21\n" "fmla v16.8h, v25.8h, v2.h[3]\n" - "prfm pldl1keep, [x11, #0x80]\n" "fmla v20.8h, v25.8h, v3.h[3]\n" "ldr d25, [x17, #0xe0]\n" "fmla v9.8h, v24.8h, v0.h[3]\n" - "mov v25.d[1], x21\n" + "ldr x21, [x17, #0xf8]\n" "fmla v13.8h, v24.8h, v1.h[3]\n" - "ldr x21, [x17, #0x108]\n" "fmla v17.8h, v24.8h, v2.h[3]\n" - "prfm pldl1keep, [x10, #0x80]\n" + "mov v25.d[1], x20\n" "fmla v21.8h, v24.8h, v3.h[3]\n" "ldr d24, [x17, #0xf0]\n" - "mov v24.d[1], x20\n" "fmla v10.8h, v25.8h, v0.h[3]\n" + "ldr x20, [x17, #0x108]\n" "fmla v14.8h, v25.8h, v1.h[3]\n" - "ldr x20, [x17, #0x118]\n" + "mov v24.d[1], x21\n" "fmla v18.8h, v25.8h, v2.h[3]\n" "fmla v22.8h, v25.8h, v3.h[3]\n" "ldr d25, [x17, #0x100]\n" "fmla v11.8h, v24.8h, v0.h[3]\n" - "mov v25.d[1], x21\n" + "ldr x21, [x17, #0x118]\n" "fmla v15.8h, v24.8h, v1.h[3]\n" - "ldr x21, [x17, #0x128]\n" "fmla v19.8h, v24.8h, v2.h[3]\n" + "mov v25.d[1], x20\n" "fmla v23.8h, v24.8h, v3.h[3]\n" "ldr d24, [x17, #0x110]\n" - "mov v24.d[1], x20\n" "fmla v8.8h, v25.8h, v0.h[4]\n" + "ldr x20, [x17, #0x128]\n" "fmla v12.8h, v25.8h, v1.h[4]\n" - "ldr x20, [x17, #0x138]\n" + "mov v24.d[1], x21\n" "fmla v16.8h, v25.8h, v2.h[4]\n" "fmla v20.8h, v25.8h, v3.h[4]\n" "ldr d25, [x17, #0x120]\n" "fmla v9.8h, v24.8h, v0.h[4]\n" - "mov v25.d[1], x21\n" + "ldr x21, [x17, #0x138]\n" "fmla v13.8h, v24.8h, v1.h[4]\n" - "ldr x21, [x17, #0x148]\n" "fmla v17.8h, v24.8h, v2.h[4]\n" + "mov v25.d[1], x20\n" "fmla v21.8h, v24.8h, v3.h[4]\n" "ldr d24, [x17, #0x130]\n" - "mov v24.d[1], x20\n" "fmla v10.8h, v25.8h, v0.h[4]\n" + "ldr x20, [x17, #0x148]\n" "fmla v14.8h, v25.8h, v1.h[4]\n" - "ldr x20, [x17, #0x158]\n" + "mov v24.d[1], x21\n" "fmla v18.8h, v25.8h, v2.h[4]\n" "fmla v22.8h, v25.8h, v3.h[4]\n" "ldr d25, [x17, #0x140]\n" "fmla v11.8h, v24.8h, v0.h[4]\n" - "mov v25.d[1], x21\n" + "ldr x21, [x17, #0x158]\n" "fmla v15.8h, v24.8h, v1.h[4]\n" - "ldr x21, [x17, #0x168]\n" "fmla v19.8h, v24.8h, v2.h[4]\n" + "mov v25.d[1], x20\n" "fmla v23.8h, v24.8h, v3.h[4]\n" "ldr d24, [x17, #0x150]\n" - "mov v24.d[1], x20\n" "fmla v8.8h, v25.8h, v0.h[5]\n" + "ldr x20, [x17, #0x168]\n" "fmla v12.8h, v25.8h, v1.h[5]\n" - "ldr x20, [x17, #0x178]\n" + "mov v24.d[1], x21\n" "fmla v16.8h, v25.8h, v2.h[5]\n" "fmla v20.8h, v25.8h, v3.h[5]\n" "ldr d25, [x17, #0x160]\n" "fmla v9.8h, v24.8h, v0.h[5]\n" - "mov v25.d[1], x21\n" + "ldr x21, [x17, #0x178]\n" "fmla v13.8h, v24.8h, v1.h[5]\n" - "ldr x21, [x17, #0x188]\n" "fmla v17.8h, v24.8h, v2.h[5]\n" + "mov v25.d[1], x20\n" "fmla v21.8h, v24.8h, v3.h[5]\n" "ldr d24, [x17, #0x170]\n" - "mov v24.d[1], x20\n" "fmla v10.8h, v25.8h, v0.h[5]\n" + "ldr x20, [x17, #0x188]\n" "fmla v14.8h, v25.8h, v1.h[5]\n" - "ldr x20, [x17, #0x198]\n" + "mov v24.d[1], x21\n" "fmla v18.8h, v25.8h, v2.h[5]\n" "fmla v22.8h, v25.8h, v3.h[5]\n" "ldr d25, [x17, #0x180]\n" "fmla v11.8h, v24.8h, v0.h[5]\n" - "mov v25.d[1], x21\n" + "ldr x21, [x17, #0x198]\n" "fmla v15.8h, v24.8h, v1.h[5]\n" - "ldr x21, [x17, #0x1a8]\n" "fmla v19.8h, v24.8h, v2.h[5]\n" + "mov v25.d[1], x20\n" "fmla v23.8h, v24.8h, v3.h[5]\n" "ldr d24, [x17, #0x190]\n" - "mov v24.d[1], x20\n" "fmla v8.8h, v25.8h, v0.h[6]\n" + "ldr x20, [x17, #0x1a8]\n" "fmla v12.8h, v25.8h, v1.h[6]\n" - "ldr x20, [x17, #0x1b8]\n" + "mov v24.d[1], x21\n" "fmla v16.8h, v25.8h, v2.h[6]\n" "fmla v20.8h, v25.8h, v3.h[6]\n" "ldr d25, [x17, #0x1a0]\n" "fmla v9.8h, v24.8h, v0.h[6]\n" - "mov v25.d[1], x21\n" + "ldr x21, [x17, #0x1b8]\n" "fmla v13.8h, v24.8h, v1.h[6]\n" - "ldr x21, [x17, #0x1c8]\n" "fmla v17.8h, v24.8h, v2.h[6]\n" + "mov v25.d[1], x20\n" "fmla v21.8h, v24.8h, v3.h[6]\n" "ldr d24, [x17, #0x1b0]\n" - "mov v24.d[1], x20\n" "fmla v10.8h, v25.8h, v0.h[6]\n" + "ldr x20, [x17, #0x1c8]\n" "fmla v14.8h, v25.8h, v1.h[6]\n" - "ldr x20, [x17, #0x1d8]\n" + "mov v24.d[1], x21\n" "fmla v18.8h, v25.8h, v2.h[6]\n" "fmla v22.8h, v25.8h, v3.h[6]\n" "ldr d25, [x17, #0x1c0]\n" "fmla v11.8h, v24.8h, v0.h[6]\n" - "mov v25.d[1], x21\n" + "ldr x21, [x17, #0x1d8]\n" "fmla v15.8h, v24.8h, v1.h[6]\n" - "ldr x21, [x17, #0x1e8]\n" "fmla v19.8h, v24.8h, v2.h[6]\n" + "mov v25.d[1], x20\n" "fmla v23.8h, v24.8h, v3.h[6]\n" "ldr d24, [x17, #0x1d0]\n" - "mov v24.d[1], x20\n" "fmla v8.8h, v25.8h, v0.h[7]\n" + "ldr x20, [x17, #0x1e8]\n" "fmla v12.8h, v25.8h, v1.h[7]\n" - "ldr x20, [x17, #0x1f8]\n" + "mov v24.d[1], x21\n" "fmla v16.8h, v25.8h, v2.h[7]\n" "fmla v20.8h, v25.8h, v3.h[7]\n" "ldr d25, [x17, #0x1e0]\n" "fmla v9.8h, v24.8h, v0.h[7]\n" - "mov v25.d[1], x21\n" + "ldr x21, [x17, #0x1f8]\n" "fmla v13.8h, v24.8h, v1.h[7]\n" "fmla v17.8h, v24.8h, v2.h[7]\n" + "mov v25.d[1], x20\n" "fmla v21.8h, v24.8h, v3.h[7]\n" "ldr d24, [x17, #0x1f0]\n" - "mov v24.d[1], x20\n" "add x17, x17, #0x200\n" "fmla v10.8h, v25.8h, v0.h[7]\n" - "ldr x21, [x17, #0x8]\n" "fmla v14.8h, v25.8h, v1.h[7]\n" - "ldr x20, [x17, #0x18]\n" + "ldr x20, [x17, #0x8]\n" + "mov v24.d[1], x21\n" "fmla v18.8h, v25.8h, v2.h[7]\n" "fmla v22.8h, v25.8h, v3.h[7]\n" "ldr d6, [x17, #0x0]\n" @@ -2697,7 +2698,8 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "fmla v23.8h, v24.8h, v3.h[7]\n" "ldr d3, [x10, #0x0]\n" "ldr d7, [x17, #0x10]\n" - "mov v6.d[1], x21\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" @@ -2882,8 +2884,8 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "ldr h1, [x11], #0x2\n" "ldr h0, [x10], #0x2\n" "ldr q25, [x17, #0x0]\n" - "fmla v8.8h, v25.8h, v3.h[0]\n" "ldr q24, [x17, #0x10]\n" + "fmla v8.8h, v25.8h, v3.h[0]\n" "fmla v12.8h, v25.8h, v2.h[0]\n" "fmla v16.8h, v25.8h, v1.h[0]\n" "fmla v20.8h, v25.8h, v0.h[0]\n" @@ -2909,34 +2911,34 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "cmp x15, x20\n" "bne 170b\n" "ldr x20, [%x[args_ptr], %[offsetof_output_offset]]\n" - "add x25, x16, x20, LSL #1\n" - "add x24, x25, x20, LSL #1\n" - "add x23, x24, x20, LSL #1\n" "prfm pstl1keep, [x16, #0x0]\n" + "add x26, x16, x20, LSL #1\n" + "add x25, x26, x20, LSL #1\n" + "prfm pstl1keep, [x26, #0x0]\n" + "add x24, x25, x20, LSL #1\n" "prfm pstl1keep, [x25, #0x0]\n" "prfm pstl1keep, [x24, #0x0]\n" - "prfm pstl1keep, [x23, #0x0]\n" "tbz %x[flags], #1, 178f\n" - "add x20, %x[args_ptr], %[offset_max]\n" - "ld1r { v24.8h }, [x20]\n" - "fmin v8.8h, v8.8h, v24.8h\n" - "fmin v9.8h, v9.8h, v24.8h\n" - "fmin v10.8h, v10.8h, v24.8h\n" - "fmin v11.8h, v11.8h, v24.8h\n" - "fmin v12.8h, v12.8h, v24.8h\n" - "fmin v13.8h, v13.8h, v24.8h\n" - "fmin v14.8h, v14.8h, v24.8h\n" - "fmin v15.8h, v15.8h, v24.8h\n" - "fmin v16.8h, v16.8h, v24.8h\n" - "fmin v17.8h, v17.8h, v24.8h\n" - "fmin v18.8h, v18.8h, v24.8h\n" - "fmin v19.8h, v19.8h, v24.8h\n" - "fmin v20.8h, v20.8h, v24.8h\n" - "fmin v21.8h, v21.8h, v24.8h\n" - "fmin v22.8h, v22.8h, v24.8h\n" - "fmin v23.8h, v23.8h, v24.8h\n" + "add x21, %x[args_ptr], %[offset_max]\n" "add x20, %x[args_ptr], %[offset_min]\n" + "ld1r { v25.8h }, [x21]\n" "ld1r { v24.8h }, [x20]\n" + "fmin v8.8h, v8.8h, v25.8h\n" + "fmin v9.8h, v9.8h, v25.8h\n" + "fmin v10.8h, v10.8h, v25.8h\n" + "fmin v11.8h, v11.8h, v25.8h\n" + "fmin v12.8h, v12.8h, v25.8h\n" + "fmin v13.8h, v13.8h, v25.8h\n" + "fmin v14.8h, v14.8h, v25.8h\n" + "fmin v15.8h, v15.8h, v25.8h\n" + "fmin v16.8h, v16.8h, v25.8h\n" + "fmin v17.8h, v17.8h, v25.8h\n" + "fmin v18.8h, v18.8h, v25.8h\n" + "fmin v19.8h, v19.8h, v25.8h\n" + "fmin v20.8h, v20.8h, v25.8h\n" + "fmin v21.8h, v21.8h, v25.8h\n" + "fmin v22.8h, v22.8h, v25.8h\n" + "fmin v23.8h, v23.8h, v25.8h\n" "fmax v8.8h, v8.8h, v24.8h\n" "fmax v9.8h, v9.8h, v24.8h\n" "fmax v10.8h, v10.8h, v24.8h\n" @@ -2959,191 +2961,191 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "tbz x8, #4, 186f\n" "st1 { v8.8h }, [x16], #0x10\n" "st1 { v9.8h }, [x16], #0x10\n" - "st1 { v12.8h }, [x25], #0x10\n" - "st1 { v13.8h }, [x25], #0x10\n" - "st1 { v16.8h }, [x24], #0x10\n" - "st1 { v17.8h }, [x24], #0x10\n" - "st1 { v20.8h }, [x23], #0x10\n" - "st1 { v21.8h }, [x23], #0x10\n" + "st1 { v12.8h }, [x26], #0x10\n" + "st1 { v13.8h }, [x26], #0x10\n" + "st1 { v16.8h }, [x25], #0x10\n" + "st1 { v17.8h }, [x25], #0x10\n" + "st1 { v20.8h }, [x24], #0x10\n" + "st1 { v21.8h }, [x24], #0x10\n" "tbz x8, #3, 182f\n" "st1 { v10.8h }, [x16], #0x10\n" - "st1 { v14.8h }, [x25], #0x10\n" - "st1 { v18.8h }, [x24], #0x10\n" - "st1 { v22.8h }, [x23], #0x10\n" + "st1 { v14.8h }, [x26], #0x10\n" + "st1 { v18.8h }, [x25], #0x10\n" + "st1 { v22.8h }, [x24], #0x10\n" "tbz x8, #2, 180f\n" "str d11, [x16], #0x8\n" - "str d15, [x25], #0x8\n" - "str d19, [x24], #0x8\n" - "str d23, [x23], #0x8\n" + "str d15, [x26], #0x8\n" + "str d19, [x25], #0x8\n" + "str d23, [x24], #0x8\n" "tbz x8, #1, 179f\n" "st1 { v11.s }[2], [x16], #0x4\n" - "st1 { v15.s }[2], [x25], #0x4\n" - "st1 { v19.s }[2], [x24], #0x4\n" - "st1 { v23.s }[2], [x23], #0x4\n" + "st1 { v15.s }[2], [x26], #0x4\n" + "st1 { v19.s }[2], [x25], #0x4\n" + "st1 { v23.s }[2], [x24], #0x4\n" "tbz x8, #0, 194f\n" "st1 { v11.h }[6], [x16]\n" - "st1 { v15.h }[6], [x25]\n" - "st1 { v19.h }[6], [x24]\n" - "st1 { v23.h }[6], [x23]\n" + "st1 { v15.h }[6], [x26]\n" + "st1 { v19.h }[6], [x25]\n" + "st1 { v23.h }[6], [x24]\n" "b 194f\n" "179:" // Height 4: Partial direct writeback: partial_1_28 "tbz x8, #0, 194f\n" "st1 { v11.h }[4], [x16]\n" - "st1 { v15.h }[4], [x25]\n" - "st1 { v19.h }[4], [x24]\n" - "st1 { v23.h }[4], [x23]\n" + "st1 { v15.h }[4], [x26]\n" + "st1 { v19.h }[4], [x25]\n" + "st1 { v23.h }[4], [x24]\n" "b 194f\n" "180:" // Height 4: Partial direct writeback: partial_2_24 "tbz x8, #1, 181f\n" "str s11, [x16], #0x4\n" - "str s15, [x25], #0x4\n" - "str s19, [x24], #0x4\n" - "str s23, [x23], #0x4\n" + "str s15, [x26], #0x4\n" + "str s19, [x25], #0x4\n" + "str s23, [x24], #0x4\n" "tbz x8, #0, 194f\n" "st1 { v11.h }[2], [x16]\n" - "st1 { v15.h }[2], [x25]\n" - "st1 { v19.h }[2], [x24]\n" - "st1 { v23.h }[2], [x23]\n" + "st1 { v15.h }[2], [x26]\n" + "st1 { v19.h }[2], [x25]\n" + "st1 { v23.h }[2], [x24]\n" "b 194f\n" "181:" // Height 4: Partial direct writeback: partial_1_24 "tbz x8, #0, 194f\n" "str h11, [x16, #0x0]\n" - "str h15, [x25, #0x0]\n" - "str h19, [x24, #0x0]\n" - "str h23, [x23, #0x0]\n" + "str h15, [x26, #0x0]\n" + "str h19, [x25, #0x0]\n" + "str h23, [x24, #0x0]\n" "b 194f\n" "182:" // Height 4: Partial direct writeback: partial_4_16 "tbz x8, #2, 184f\n" "str d10, [x16], #0x8\n" - "str d14, [x25], #0x8\n" - "str d18, [x24], #0x8\n" - "str d22, [x23], #0x8\n" + "str d14, [x26], #0x8\n" + "str d18, [x25], #0x8\n" + "str d22, [x24], #0x8\n" "tbz x8, #1, 183f\n" "st1 { v10.s }[2], [x16], #0x4\n" - "st1 { v14.s }[2], [x25], #0x4\n" - "st1 { v18.s }[2], [x24], #0x4\n" - "st1 { v22.s }[2], [x23], #0x4\n" + "st1 { v14.s }[2], [x26], #0x4\n" + "st1 { v18.s }[2], [x25], #0x4\n" + "st1 { v22.s }[2], [x24], #0x4\n" "tbz x8, #0, 194f\n" "st1 { v10.h }[6], [x16]\n" - "st1 { v14.h }[6], [x25]\n" - "st1 { v18.h }[6], [x24]\n" - "st1 { v22.h }[6], [x23]\n" + "st1 { v14.h }[6], [x26]\n" + "st1 { v18.h }[6], [x25]\n" + "st1 { v22.h }[6], [x24]\n" "b 194f\n" "183:" // Height 4: Partial direct writeback: partial_1_20 "tbz x8, #0, 194f\n" "st1 { v10.h }[4], [x16]\n" - "st1 { v14.h }[4], [x25]\n" - "st1 { v18.h }[4], [x24]\n" - "st1 { v22.h }[4], [x23]\n" + "st1 { v14.h }[4], [x26]\n" + "st1 { v18.h }[4], [x25]\n" + "st1 { v22.h }[4], [x24]\n" "b 194f\n" "184:" // Height 4: Partial direct writeback: partial_2_16 "tbz x8, #1, 185f\n" "str s10, [x16], #0x4\n" - "str s14, [x25], #0x4\n" - "str s18, [x24], #0x4\n" - "str s22, [x23], #0x4\n" + "str s14, [x26], #0x4\n" + "str s18, [x25], #0x4\n" + "str s22, [x24], #0x4\n" "tbz x8, #0, 194f\n" "st1 { v10.h }[2], [x16]\n" - "st1 { v14.h }[2], [x25]\n" - "st1 { v18.h }[2], [x24]\n" - "st1 { v22.h }[2], [x23]\n" + "st1 { v14.h }[2], [x26]\n" + "st1 { v18.h }[2], [x25]\n" + "st1 { v22.h }[2], [x24]\n" "b 194f\n" "185:" // Height 4: Partial direct writeback: partial_1_16 "tbz x8, #0, 194f\n" "str h10, [x16, #0x0]\n" - "str h14, [x25, #0x0]\n" - "str h18, [x24, #0x0]\n" - "str h22, [x23, #0x0]\n" + "str h14, [x26, #0x0]\n" + "str h18, [x25, #0x0]\n" + "str h22, [x24, #0x0]\n" "b 194f\n" "186:" // Height 4: Partial direct writeback: partial_8_0 "tbz x8, #3, 190f\n" "st1 { v8.8h }, [x16], #0x10\n" - "st1 { v12.8h }, [x25], #0x10\n" - "st1 { v16.8h }, [x24], #0x10\n" - "st1 { v20.8h }, [x23], #0x10\n" + "st1 { v12.8h }, [x26], #0x10\n" + "st1 { v16.8h }, [x25], #0x10\n" + "st1 { v20.8h }, [x24], #0x10\n" "tbz x8, #2, 188f\n" "str d9, [x16], #0x8\n" - "str d13, [x25], #0x8\n" - "str d17, [x24], #0x8\n" - "str d21, [x23], #0x8\n" + "str d13, [x26], #0x8\n" + "str d17, [x25], #0x8\n" + "str d21, [x24], #0x8\n" "tbz x8, #1, 187f\n" "st1 { v9.s }[2], [x16], #0x4\n" - "st1 { v13.s }[2], [x25], #0x4\n" - "st1 { v17.s }[2], [x24], #0x4\n" - "st1 { v21.s }[2], [x23], #0x4\n" + "st1 { v13.s }[2], [x26], #0x4\n" + "st1 { v17.s }[2], [x25], #0x4\n" + "st1 { v21.s }[2], [x24], #0x4\n" "tbz x8, #0, 194f\n" "st1 { v9.h }[6], [x16]\n" - "st1 { v13.h }[6], [x25]\n" - "st1 { v17.h }[6], [x24]\n" - "st1 { v21.h }[6], [x23]\n" + "st1 { v13.h }[6], [x26]\n" + "st1 { v17.h }[6], [x25]\n" + "st1 { v21.h }[6], [x24]\n" "b 194f\n" "187:" // Height 4: Partial direct writeback: partial_1_12 "tbz x8, #0, 194f\n" "st1 { v9.h }[4], [x16]\n" - "st1 { v13.h }[4], [x25]\n" - "st1 { v17.h }[4], [x24]\n" - "st1 { v21.h }[4], [x23]\n" + "st1 { v13.h }[4], [x26]\n" + "st1 { v17.h }[4], [x25]\n" + "st1 { v21.h }[4], [x24]\n" "b 194f\n" "188:" // Height 4: Partial direct writeback: partial_2_8 "tbz x8, #1, 189f\n" "str s9, [x16], #0x4\n" - "str s13, [x25], #0x4\n" - "str s17, [x24], #0x4\n" - "str s21, [x23], #0x4\n" + "str s13, [x26], #0x4\n" + "str s17, [x25], #0x4\n" + "str s21, [x24], #0x4\n" "tbz x8, #0, 194f\n" "st1 { v9.h }[2], [x16]\n" - "st1 { v13.h }[2], [x25]\n" - "st1 { v17.h }[2], [x24]\n" - "st1 { v21.h }[2], [x23]\n" + "st1 { v13.h }[2], [x26]\n" + "st1 { v17.h }[2], [x25]\n" + "st1 { v21.h }[2], [x24]\n" "b 194f\n" "189:" // Height 4: Partial direct writeback: partial_1_8 "tbz x8, #0, 194f\n" "str h9, [x16, #0x0]\n" - "str h13, [x25, #0x0]\n" - "str h17, [x24, #0x0]\n" - "str h21, [x23, #0x0]\n" + "str h13, [x26, #0x0]\n" + "str h17, [x25, #0x0]\n" + "str h21, [x24, #0x0]\n" "b 194f\n" "190:" // Height 4: Partial direct writeback: partial_4_0 "tbz x8, #2, 192f\n" "str d8, [x16], #0x8\n" - "str d12, [x25], #0x8\n" - "str d16, [x24], #0x8\n" - "str d20, [x23], #0x8\n" + "str d12, [x26], #0x8\n" + "str d16, [x25], #0x8\n" + "str d20, [x24], #0x8\n" "tbz x8, #1, 191f\n" "st1 { v8.s }[2], [x16], #0x4\n" - "st1 { v12.s }[2], [x25], #0x4\n" - "st1 { v16.s }[2], [x24], #0x4\n" - "st1 { v20.s }[2], [x23], #0x4\n" + "st1 { v12.s }[2], [x26], #0x4\n" + "st1 { v16.s }[2], [x25], #0x4\n" + "st1 { v20.s }[2], [x24], #0x4\n" "tbz x8, #0, 194f\n" "st1 { v8.h }[6], [x16]\n" - "st1 { v12.h }[6], [x25]\n" - "st1 { v16.h }[6], [x24]\n" - "st1 { v20.h }[6], [x23]\n" + "st1 { v12.h }[6], [x26]\n" + "st1 { v16.h }[6], [x25]\n" + "st1 { v20.h }[6], [x24]\n" "b 194f\n" "191:" // Height 4: Partial direct writeback: partial_1_4 "tbz x8, #0, 194f\n" "st1 { v8.h }[4], [x16]\n" - "st1 { v12.h }[4], [x25]\n" - "st1 { v16.h }[4], [x24]\n" - "st1 { v20.h }[4], [x23]\n" + "st1 { v12.h }[4], [x26]\n" + "st1 { v16.h }[4], [x25]\n" + "st1 { v20.h }[4], [x24]\n" "b 194f\n" "192:" // Height 4: Partial direct writeback: partial_2_0 "tbz x8, #1, 193f\n" "str s8, [x16], #0x4\n" - "str s12, [x25], #0x4\n" - "str s16, [x24], #0x4\n" - "str s20, [x23], #0x4\n" + "str s12, [x26], #0x4\n" + "str s16, [x25], #0x4\n" + "str s20, [x24], #0x4\n" "tbz x8, #0, 194f\n" "st1 { v8.h }[2], [x16]\n" - "st1 { v12.h }[2], [x25]\n" - "st1 { v16.h }[2], [x24]\n" - "st1 { v20.h }[2], [x23]\n" + "st1 { v12.h }[2], [x26]\n" + "st1 { v16.h }[2], [x25]\n" + "st1 { v20.h }[2], [x24]\n" "b 194f\n" "193:" // Height 4: Partial direct writeback: partial_1_0 "str h8, [x16, #0x0]\n" - "str h12, [x25, #0x0]\n" - "str h16, [x24, #0x0]\n" - "str h20, [x23, #0x0]\n" + "str h12, [x26, #0x0]\n" + "str h16, [x25, #0x0]\n" + "str h20, [x24, #0x0]\n" "194:" // Height 4: Partial direct writeback: Done "b 196f\n" "195:" // Height 4: Full writeback @@ -3152,39 +3154,39 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "str q10, [x16, #0x20]\n" "str q11, [x16, #0x30]\n" "add x16, x16, #0x40\n" - "str q12, [x25, #0x0]\n" - "str q13, [x25, #0x10]\n" - "str q14, [x25, #0x20]\n" - "str q15, [x25, #0x30]\n" - "str q16, [x24, #0x0]\n" - "str q17, [x24, #0x10]\n" - "str q18, [x24, #0x20]\n" - "str q19, [x24, #0x30]\n" - "str q20, [x23, #0x0]\n" - "str q21, [x23, #0x10]\n" - "str q22, [x23, #0x20]\n" - "str q23, [x23, #0x30]\n" + "str q12, [x26, #0x0]\n" + "str q13, [x26, #0x10]\n" + "str q14, [x26, #0x20]\n" + "str q15, [x26, #0x30]\n" + "str q16, [x25, #0x0]\n" + "str q17, [x25, #0x10]\n" + "str q18, [x25, #0x20]\n" + "str q19, [x25, #0x30]\n" + "str q20, [x24, #0x0]\n" + "str q21, [x24, #0x10]\n" + "str q22, [x24, #0x20]\n" + "str q23, [x24, #0x30]\n" "196:" // Height 4: Writeback done "subs x8, x8, #0x20\n" "bgt 149b\n" "b 296f\n" "197:" // Height 5 - "mov x7, %x[bias]\n" + "ldr x7, [%x[args_ptr], %[offsetof_bias]]\n" "ldr x8, [%x[args_ptr], %[offsetof_N]]\n" "ldr x17, [%x[args_ptr], %[offsetof_B_ptr]]\n" - "mov x16, %x[output_ptr]\n" + "ldr x16, [%x[args_ptr], %[offsetof_output_ptr]]\n" "198:" // Height 5: Column loop "cbz x7, 199f\n" "ldr q8, [x7, #0x0]\n" - "mov v12.16b, v8.16b\n" "ldr q9, [x7, #0x10]\n" - "mov v13.16b, v9.16b\n" "ldr q10, [x7, #0x20]\n" - "mov v14.16b, v10.16b\n" "ldr q11, [x7, #0x30]\n" + "mov v12.16b, v8.16b\n" + "mov v13.16b, v9.16b\n" + "add x7, x7, #0x40\n" + "mov v14.16b, v10.16b\n" "mov v15.16b, v11.16b\n" "mov v16.16b, v8.16b\n" - "add x7, x7, #0x40\n" "mov v17.16b, v9.16b\n" "mov v18.16b, v10.16b\n" "mov v19.16b, v11.16b\n" @@ -3200,248 +3202,248 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "199:" // Height 5: no bias "tbz %x[flags], #0, 217f\n" "ldr x20, [%x[args_ptr], %[offsetof_output_offset]]\n" - "add x25, x16, x20, LSL #1\n" + "cmp x8, #0x20\n" + "add x26, x16, x20, LSL #1\n" + "add x25, x26, x20, LSL #1\n" "add x24, x25, x20, LSL #1\n" "add x23, x24, x20, LSL #1\n" - "cmp x8, #0x20\n" - "add x22, x23, x20, LSL #1\n" "bge 216f\n" "tbz x8, #4, 207f\n" "ld1 { v8.8h }, [x16], #0x10\n" - "ld1 { v12.8h }, [x25], #0x10\n" - "ld1 { v16.8h }, [x24], #0x10\n" - "ld1 { v20.8h }, [x23], #0x10\n" - "ld1 { v24.8h }, [x22], #0x10\n" + "ld1 { v12.8h }, [x26], #0x10\n" + "ld1 { v16.8h }, [x25], #0x10\n" + "ld1 { v20.8h }, [x24], #0x10\n" + "ld1 { v24.8h }, [x23], #0x10\n" "ld1 { v9.8h }, [x16], #0x10\n" - "ld1 { v13.8h }, [x25], #0x10\n" - "ld1 { v17.8h }, [x24], #0x10\n" - "ld1 { v21.8h }, [x23], #0x10\n" - "ld1 { v25.8h }, [x22], #0x10\n" + "ld1 { v13.8h }, [x26], #0x10\n" + "ld1 { v17.8h }, [x25], #0x10\n" + "ld1 { v21.8h }, [x24], #0x10\n" + "ld1 { v25.8h }, [x23], #0x10\n" "tbz x8, #3, 203f\n" "ld1 { v10.8h }, [x16], #0x10\n" - "ld1 { v14.8h }, [x25], #0x10\n" - "ld1 { v18.8h }, [x24], #0x10\n" - "ld1 { v22.8h }, [x23], #0x10\n" - "ld1 { v26.8h }, [x22], #0x10\n" + "ld1 { v14.8h }, [x26], #0x10\n" + "ld1 { v18.8h }, [x25], #0x10\n" + "ld1 { v22.8h }, [x24], #0x10\n" + "ld1 { v26.8h }, [x23], #0x10\n" "tbz x8, #2, 201f\n" "ldr d11, [x16], #0x8\n" - "ldr d15, [x25], #0x8\n" - "ldr d19, [x24], #0x8\n" - "ldr d23, [x23], #0x8\n" - "ldr d27, [x22], #0x8\n" + "ldr d15, [x26], #0x8\n" + "ldr d19, [x25], #0x8\n" + "ldr d23, [x24], #0x8\n" + "ldr d27, [x23], #0x8\n" "tbz x8, #1, 200f\n" "ld1 { v11.s }[2], [x16], #0x4\n" "mov x20, #0x3c\n" - "ld1 { v15.s }[2], [x25], #0x4\n" - "ld1 { v19.s }[2], [x24], #0x4\n" - "ld1 { v23.s }[2], [x23], #0x4\n" - "ld1 { v27.s }[2], [x22], #0x4\n" + "ld1 { v15.s }[2], [x26], #0x4\n" + "ld1 { v19.s }[2], [x25], #0x4\n" + "ld1 { v23.s }[2], [x24], #0x4\n" + "ld1 { v27.s }[2], [x23], #0x4\n" "tbz x8, #0, 215f\n" "ld1 { v11.h }[6], [x16]\n" - "ld1 { v15.h }[6], [x25]\n" - "ld1 { v19.h }[6], [x24]\n" - "ld1 { v23.h }[6], [x23]\n" - "ld1 { v27.h }[6], [x22]\n" + "ld1 { v15.h }[6], [x26]\n" + "ld1 { v19.h }[6], [x25]\n" + "ld1 { v23.h }[6], [x24]\n" + "ld1 { v27.h }[6], [x23]\n" "b 215f\n" "200:" // Height 5: Partial accumulate: partial_1_28 "mov x20, #0x38\n" "tbz x8, #0, 215f\n" "ld1 { v11.h }[4], [x16]\n" - "ld1 { v15.h }[4], [x25]\n" - "ld1 { v19.h }[4], [x24]\n" - "ld1 { v23.h }[4], [x23]\n" - "ld1 { v27.h }[4], [x22]\n" + "ld1 { v15.h }[4], [x26]\n" + "ld1 { v19.h }[4], [x25]\n" + "ld1 { v23.h }[4], [x24]\n" + "ld1 { v27.h }[4], [x23]\n" "b 215f\n" "201:" // Height 5: Partial accumulate: partial_2_24 "tbz x8, #1, 202f\n" "ldr s11, [x16], #0x4\n" "mov x20, #0x34\n" - "ldr s15, [x25], #0x4\n" - "ldr s19, [x24], #0x4\n" - "ldr s23, [x23], #0x4\n" - "ldr s27, [x22], #0x4\n" + "ldr s15, [x26], #0x4\n" + "ldr s19, [x25], #0x4\n" + "ldr s23, [x24], #0x4\n" + "ldr s27, [x23], #0x4\n" "tbz x8, #0, 215f\n" "ld1 { v11.h }[2], [x16]\n" - "ld1 { v15.h }[2], [x25]\n" - "ld1 { v19.h }[2], [x24]\n" - "ld1 { v23.h }[2], [x23]\n" - "ld1 { v27.h }[2], [x22]\n" + "ld1 { v15.h }[2], [x26]\n" + "ld1 { v19.h }[2], [x25]\n" + "ld1 { v23.h }[2], [x24]\n" + "ld1 { v27.h }[2], [x23]\n" "b 215f\n" "202:" // Height 5: Partial accumulate: partial_1_24 "mov x20, #0x30\n" "tbz x8, #0, 215f\n" "ldr h11, [x16, #0x0]\n" - "ldr h15, [x25, #0x0]\n" - "ldr h19, [x24, #0x0]\n" - "ldr h23, [x23, #0x0]\n" - "ldr h27, [x22, #0x0]\n" + "ldr h15, [x26, #0x0]\n" + "ldr h19, [x25, #0x0]\n" + "ldr h23, [x24, #0x0]\n" + "ldr h27, [x23, #0x0]\n" "b 215f\n" "203:" // Height 5: Partial accumulate: partial_4_16 "tbz x8, #2, 205f\n" "ldr d10, [x16], #0x8\n" - "ldr d14, [x25], #0x8\n" - "ldr d18, [x24], #0x8\n" - "ldr d22, [x23], #0x8\n" - "ldr d26, [x22], #0x8\n" + "ldr d14, [x26], #0x8\n" + "ldr d18, [x25], #0x8\n" + "ldr d22, [x24], #0x8\n" + "ldr d26, [x23], #0x8\n" "tbz x8, #1, 204f\n" "ld1 { v10.s }[2], [x16], #0x4\n" "mov x20, #0x2c\n" - "ld1 { v14.s }[2], [x25], #0x4\n" - "ld1 { v18.s }[2], [x24], #0x4\n" - "ld1 { v22.s }[2], [x23], #0x4\n" - "ld1 { v26.s }[2], [x22], #0x4\n" + "ld1 { v14.s }[2], [x26], #0x4\n" + "ld1 { v18.s }[2], [x25], #0x4\n" + "ld1 { v22.s }[2], [x24], #0x4\n" + "ld1 { v26.s }[2], [x23], #0x4\n" "tbz x8, #0, 215f\n" "ld1 { v10.h }[6], [x16]\n" - "ld1 { v14.h }[6], [x25]\n" - "ld1 { v18.h }[6], [x24]\n" - "ld1 { v22.h }[6], [x23]\n" - "ld1 { v26.h }[6], [x22]\n" + "ld1 { v14.h }[6], [x26]\n" + "ld1 { v18.h }[6], [x25]\n" + "ld1 { v22.h }[6], [x24]\n" + "ld1 { v26.h }[6], [x23]\n" "b 215f\n" "204:" // Height 5: Partial accumulate: partial_1_20 "mov x20, #0x28\n" "tbz x8, #0, 215f\n" "ld1 { v10.h }[4], [x16]\n" - "ld1 { v14.h }[4], [x25]\n" - "ld1 { v18.h }[4], [x24]\n" - "ld1 { v22.h }[4], [x23]\n" - "ld1 { v26.h }[4], [x22]\n" + "ld1 { v14.h }[4], [x26]\n" + "ld1 { v18.h }[4], [x25]\n" + "ld1 { v22.h }[4], [x24]\n" + "ld1 { v26.h }[4], [x23]\n" "b 215f\n" "205:" // Height 5: Partial accumulate: partial_2_16 "tbz x8, #1, 206f\n" "ldr s10, [x16], #0x4\n" "mov x20, #0x24\n" - "ldr s14, [x25], #0x4\n" - "ldr s18, [x24], #0x4\n" - "ldr s22, [x23], #0x4\n" - "ldr s26, [x22], #0x4\n" + "ldr s14, [x26], #0x4\n" + "ldr s18, [x25], #0x4\n" + "ldr s22, [x24], #0x4\n" + "ldr s26, [x23], #0x4\n" "tbz x8, #0, 215f\n" "ld1 { v10.h }[2], [x16]\n" - "ld1 { v14.h }[2], [x25]\n" - "ld1 { v18.h }[2], [x24]\n" - "ld1 { v22.h }[2], [x23]\n" - "ld1 { v26.h }[2], [x22]\n" + "ld1 { v14.h }[2], [x26]\n" + "ld1 { v18.h }[2], [x25]\n" + "ld1 { v22.h }[2], [x24]\n" + "ld1 { v26.h }[2], [x23]\n" "b 215f\n" "206:" // Height 5: Partial accumulate: partial_1_16 "mov x20, #0x20\n" "tbz x8, #0, 215f\n" "ldr h10, [x16, #0x0]\n" - "ldr h14, [x25, #0x0]\n" - "ldr h18, [x24, #0x0]\n" - "ldr h22, [x23, #0x0]\n" - "ldr h26, [x22, #0x0]\n" + "ldr h14, [x26, #0x0]\n" + "ldr h18, [x25, #0x0]\n" + "ldr h22, [x24, #0x0]\n" + "ldr h26, [x23, #0x0]\n" "b 215f\n" "207:" // Height 5: Partial accumulate: partial_8_0 "tbz x8, #3, 211f\n" "ld1 { v8.8h }, [x16], #0x10\n" - "ld1 { v12.8h }, [x25], #0x10\n" - "ld1 { v16.8h }, [x24], #0x10\n" - "ld1 { v20.8h }, [x23], #0x10\n" - "ld1 { v24.8h }, [x22], #0x10\n" + "ld1 { v12.8h }, [x26], #0x10\n" + "ld1 { v16.8h }, [x25], #0x10\n" + "ld1 { v20.8h }, [x24], #0x10\n" + "ld1 { v24.8h }, [x23], #0x10\n" "tbz x8, #2, 209f\n" "ldr d9, [x16], #0x8\n" - "ldr d13, [x25], #0x8\n" - "ldr d17, [x24], #0x8\n" - "ldr d21, [x23], #0x8\n" - "ldr d25, [x22], #0x8\n" + "ldr d13, [x26], #0x8\n" + "ldr d17, [x25], #0x8\n" + "ldr d21, [x24], #0x8\n" + "ldr d25, [x23], #0x8\n" "tbz x8, #1, 208f\n" "ld1 { v9.s }[2], [x16], #0x4\n" "mov x20, #0x1c\n" - "ld1 { v13.s }[2], [x25], #0x4\n" - "ld1 { v17.s }[2], [x24], #0x4\n" - "ld1 { v21.s }[2], [x23], #0x4\n" - "ld1 { v25.s }[2], [x22], #0x4\n" + "ld1 { v13.s }[2], [x26], #0x4\n" + "ld1 { v17.s }[2], [x25], #0x4\n" + "ld1 { v21.s }[2], [x24], #0x4\n" + "ld1 { v25.s }[2], [x23], #0x4\n" "tbz x8, #0, 215f\n" "ld1 { v9.h }[6], [x16]\n" - "ld1 { v13.h }[6], [x25]\n" - "ld1 { v17.h }[6], [x24]\n" - "ld1 { v21.h }[6], [x23]\n" - "ld1 { v25.h }[6], [x22]\n" + "ld1 { v13.h }[6], [x26]\n" + "ld1 { v17.h }[6], [x25]\n" + "ld1 { v21.h }[6], [x24]\n" + "ld1 { v25.h }[6], [x23]\n" "b 215f\n" "208:" // Height 5: Partial accumulate: partial_1_12 "mov x20, #0x18\n" "tbz x8, #0, 215f\n" "ld1 { v9.h }[4], [x16]\n" - "ld1 { v13.h }[4], [x25]\n" - "ld1 { v17.h }[4], [x24]\n" - "ld1 { v21.h }[4], [x23]\n" - "ld1 { v25.h }[4], [x22]\n" + "ld1 { v13.h }[4], [x26]\n" + "ld1 { v17.h }[4], [x25]\n" + "ld1 { v21.h }[4], [x24]\n" + "ld1 { v25.h }[4], [x23]\n" "b 215f\n" "209:" // Height 5: Partial accumulate: partial_2_8 "tbz x8, #1, 210f\n" "ldr s9, [x16], #0x4\n" "mov x20, #0x14\n" - "ldr s13, [x25], #0x4\n" - "ldr s17, [x24], #0x4\n" - "ldr s21, [x23], #0x4\n" - "ldr s25, [x22], #0x4\n" + "ldr s13, [x26], #0x4\n" + "ldr s17, [x25], #0x4\n" + "ldr s21, [x24], #0x4\n" + "ldr s25, [x23], #0x4\n" "tbz x8, #0, 215f\n" "ld1 { v9.h }[2], [x16]\n" - "ld1 { v13.h }[2], [x25]\n" - "ld1 { v17.h }[2], [x24]\n" - "ld1 { v21.h }[2], [x23]\n" - "ld1 { v25.h }[2], [x22]\n" + "ld1 { v13.h }[2], [x26]\n" + "ld1 { v17.h }[2], [x25]\n" + "ld1 { v21.h }[2], [x24]\n" + "ld1 { v25.h }[2], [x23]\n" "b 215f\n" "210:" // Height 5: Partial accumulate: partial_1_8 "mov x20, #0x10\n" "tbz x8, #0, 215f\n" "ldr h9, [x16, #0x0]\n" - "ldr h13, [x25, #0x0]\n" - "ldr h17, [x24, #0x0]\n" - "ldr h21, [x23, #0x0]\n" - "ldr h25, [x22, #0x0]\n" + "ldr h13, [x26, #0x0]\n" + "ldr h17, [x25, #0x0]\n" + "ldr h21, [x24, #0x0]\n" + "ldr h25, [x23, #0x0]\n" "b 215f\n" "211:" // Height 5: Partial accumulate: partial_4_0 "tbz x8, #2, 213f\n" "ldr d8, [x16], #0x8\n" - "ldr d12, [x25], #0x8\n" - "ldr d16, [x24], #0x8\n" - "ldr d20, [x23], #0x8\n" - "ldr d24, [x22], #0x8\n" + "ldr d12, [x26], #0x8\n" + "ldr d16, [x25], #0x8\n" + "ldr d20, [x24], #0x8\n" + "ldr d24, [x23], #0x8\n" "tbz x8, #1, 212f\n" "ld1 { v8.s }[2], [x16], #0x4\n" "mov x20, #0xc\n" - "ld1 { v12.s }[2], [x25], #0x4\n" - "ld1 { v16.s }[2], [x24], #0x4\n" - "ld1 { v20.s }[2], [x23], #0x4\n" - "ld1 { v24.s }[2], [x22], #0x4\n" + "ld1 { v12.s }[2], [x26], #0x4\n" + "ld1 { v16.s }[2], [x25], #0x4\n" + "ld1 { v20.s }[2], [x24], #0x4\n" + "ld1 { v24.s }[2], [x23], #0x4\n" "tbz x8, #0, 215f\n" "ld1 { v8.h }[6], [x16]\n" - "ld1 { v12.h }[6], [x25]\n" - "ld1 { v16.h }[6], [x24]\n" - "ld1 { v20.h }[6], [x23]\n" - "ld1 { v24.h }[6], [x22]\n" + "ld1 { v12.h }[6], [x26]\n" + "ld1 { v16.h }[6], [x25]\n" + "ld1 { v20.h }[6], [x24]\n" + "ld1 { v24.h }[6], [x23]\n" "b 215f\n" "212:" // Height 5: Partial accumulate: partial_1_4 "mov x20, #0x8\n" "tbz x8, #0, 215f\n" "ld1 { v8.h }[4], [x16]\n" - "ld1 { v12.h }[4], [x25]\n" - "ld1 { v16.h }[4], [x24]\n" - "ld1 { v20.h }[4], [x23]\n" - "ld1 { v24.h }[4], [x22]\n" + "ld1 { v12.h }[4], [x26]\n" + "ld1 { v16.h }[4], [x25]\n" + "ld1 { v20.h }[4], [x24]\n" + "ld1 { v24.h }[4], [x23]\n" "b 215f\n" "213:" // Height 5: Partial accumulate: partial_2_0 "tbz x8, #1, 214f\n" "ldr s8, [x16], #0x4\n" "mov x20, #0x4\n" - "ldr s12, [x25], #0x4\n" - "ldr s16, [x24], #0x4\n" - "ldr s20, [x23], #0x4\n" - "ldr s24, [x22], #0x4\n" + "ldr s12, [x26], #0x4\n" + "ldr s16, [x25], #0x4\n" + "ldr s20, [x24], #0x4\n" + "ldr s24, [x23], #0x4\n" "tbz x8, #0, 215f\n" "ld1 { v8.h }[2], [x16]\n" - "ld1 { v12.h }[2], [x25]\n" - "ld1 { v16.h }[2], [x24]\n" - "ld1 { v20.h }[2], [x23]\n" - "ld1 { v24.h }[2], [x22]\n" + "ld1 { v12.h }[2], [x26]\n" + "ld1 { v16.h }[2], [x25]\n" + "ld1 { v20.h }[2], [x24]\n" + "ld1 { v24.h }[2], [x23]\n" "b 215f\n" "214:" // Height 5: Partial accumulate: partial_1_0 "ldr h8, [x16, #0x0]\n" "mov x20, #0x0\n" - "ldr h12, [x25, #0x0]\n" - "ldr h16, [x24, #0x0]\n" - "ldr h20, [x23, #0x0]\n" - "ldr h24, [x22, #0x0]\n" + "ldr h12, [x26, #0x0]\n" + "ldr h16, [x25, #0x0]\n" + "ldr h20, [x24, #0x0]\n" + "ldr h24, [x23, #0x0]\n" "215:" // Height 5: Partial accumulate: Done "sub x16, x16, x20\n" "b 218f\n" @@ -3450,22 +3452,22 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "ldr q9, [x16, #0x10]\n" "ldr q10, [x16, #0x20]\n" "ldr q11, [x16, #0x30]\n" - "ldr q12, [x25, #0x0]\n" - "ldr q13, [x25, #0x10]\n" - "ldr q14, [x25, #0x20]\n" - "ldr q15, [x25, #0x30]\n" - "ldr q16, [x24, #0x0]\n" - "ldr q17, [x24, #0x10]\n" - "ldr q18, [x24, #0x20]\n" - "ldr q19, [x24, #0x30]\n" - "ldr q20, [x23, #0x0]\n" - "ldr q21, [x23, #0x10]\n" - "ldr q22, [x23, #0x20]\n" - "ldr q23, [x23, #0x30]\n" - "ldr q24, [x22, #0x0]\n" - "ldr q25, [x22, #0x10]\n" - "ldr q26, [x22, #0x20]\n" - "ldr q27, [x22, #0x30]\n" + "ldr q12, [x26, #0x0]\n" + "ldr q13, [x26, #0x10]\n" + "ldr q14, [x26, #0x20]\n" + "ldr q15, [x26, #0x30]\n" + "ldr q16, [x25, #0x0]\n" + "ldr q17, [x25, #0x10]\n" + "ldr q18, [x25, #0x20]\n" + "ldr q19, [x25, #0x30]\n" + "ldr q20, [x24, #0x0]\n" + "ldr q21, [x24, #0x10]\n" + "ldr q22, [x24, #0x20]\n" + "ldr q23, [x24, #0x30]\n" + "ldr q24, [x23, #0x0]\n" + "ldr q25, [x23, #0x10]\n" + "ldr q26, [x23, #0x20]\n" + "ldr q27, [x23, #0x30]\n" "b 218f\n" "217:" // Height 5: no accumulate "movi v8.16b, #0x0\n" @@ -3492,8 +3494,8 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "mov x15, #0x0\n" "219:" // 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, 220f\n" "ldr x20, [%x[input_ptr], x15, LSL #0x3]\n" "add x20, x20, x21, LSL #3\n" @@ -3540,259 +3542,259 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "fmla v24.8h, v6.8h, v4.h[0]\n" "ldr d29, [x17, #0x20]\n" "fmla v9.8h, v7.8h, v0.h[0]\n" - "mov v29.d[1], x21\n" + "add x11, x11, #0x10\n" "fmla v13.8h, v7.8h, v1.h[0]\n" - "ldr x21, [x17, #0x48]\n" + "add x10, x10, #0x10\n" "fmla v17.8h, v7.8h, v2.h[0]\n" - "add x11, x11, #0x10\n" + "mov v29.d[1], x21\n" "fmla v21.8h, v7.8h, v3.h[0]\n" - "add x10, x10, #0x10\n" + "ldr x21, [x17, #0x48]\n" "fmla v25.8h, v7.8h, v4.h[0]\n" "ldr d28, [x17, #0x30]\n" - "mov v28.d[1], x20\n" "fmla v10.8h, v29.8h, v0.h[0]\n" + "add x9, x9, #0x10\n" "fmla v14.8h, v29.8h, v1.h[0]\n" - "ldr x20, [x17, #0x58]\n" + "ldr x26, [x13, #0x8]\n" + "mov v28.d[1], x20\n" "fmla v18.8h, v29.8h, v2.h[0]\n" - "add x9, x9, #0x10\n" "fmla v22.8h, v29.8h, v3.h[0]\n" - "ldr x26, [x13, #0x8]\n" + "ldr x20, [x17, #0x58]\n" "fmla v26.8h, v29.8h, v4.h[0]\n" "ldr d29, [x17, #0x40]\n" "fmla v11.8h, v28.8h, v0.h[0]\n" - "mov v29.d[1], x21\n" + "ldr x25, [x12, #0x8]\n" "fmla v15.8h, v28.8h, v1.h[0]\n" - "ldr x21, [x17, #0x68]\n" + "ldr x24, [x11, #0x8]\n" "fmla v19.8h, v28.8h, v2.h[0]\n" - "ldr x25, [x12, #0x8]\n" + "mov v29.d[1], x21\n" "fmla v23.8h, v28.8h, v3.h[0]\n" - "ldr x24, [x11, #0x8]\n" + "ldr x21, [x17, #0x68]\n" "fmla v27.8h, v28.8h, v4.h[0]\n" "ldr d28, [x17, #0x50]\n" - "mov v28.d[1], x20\n" "fmla v8.8h, v29.8h, v0.h[1]\n" + "ldr x23, [x10, #0x8]\n" "fmla v12.8h, v29.8h, v1.h[1]\n" - "ldr x20, [x17, #0x78]\n" + "ldr x22, [x9, #0x8]\n" + "mov v28.d[1], x20\n" "fmla v16.8h, v29.8h, v2.h[1]\n" - "ldr x23, [x10, #0x8]\n" "fmla v20.8h, v29.8h, v3.h[1]\n" - "ldr x22, [x9, #0x8]\n" + "ldr x20, [x17, #0x78]\n" "fmla v24.8h, v29.8h, v4.h[1]\n" "ldr d29, [x17, #0x60]\n" "fmla v9.8h, v28.8h, v0.h[1]\n" - "mov v29.d[1], x21\n" + "sub x14, x14, #0x8\n" "fmla v13.8h, v28.8h, v1.h[1]\n" - "ldr x21, [x17, #0x88]\n" + "cmp x14, #0x10\n" "fmla v17.8h, v28.8h, v2.h[1]\n" - "sub x14, x14, #0x8\n" + "mov v29.d[1], x21\n" "fmla v21.8h, v28.8h, v3.h[1]\n" - "cmp x14, #0x10\n" + "ldr x21, [x17, #0x88]\n" "fmla v25.8h, v28.8h, v4.h[1]\n" "ldr d28, [x17, #0x70]\n" - "mov v28.d[1], x20\n" "fmla v10.8h, v29.8h, v0.h[1]\n" + "prfm pldl1keep, [x13, #0x80]\n" "fmla v14.8h, v29.8h, v1.h[1]\n" - "ldr x20, [x17, #0x98]\n" + "prfm pldl1keep, [x12, #0x80]\n" + "mov v28.d[1], x20\n" "fmla v18.8h, v29.8h, v2.h[1]\n" - "prfm pldl1keep, [x13, #0x80]\n" "fmla v22.8h, v29.8h, v3.h[1]\n" - "prfm pldl1keep, [x12, #0x80]\n" + "ldr x20, [x17, #0x98]\n" "fmla v26.8h, v29.8h, v4.h[1]\n" "ldr d29, [x17, #0x80]\n" "fmla v11.8h, v28.8h, v0.h[1]\n" - "mov v29.d[1], x21\n" + "prfm pldl1keep, [x11, #0x80]\n" "fmla v15.8h, v28.8h, v1.h[1]\n" - "ldr x21, [x17, #0xa8]\n" + "prfm pldl1keep, [x10, #0x80]\n" "fmla v19.8h, v28.8h, v2.h[1]\n" - "prfm pldl1keep, [x11, #0x80]\n" + "mov v29.d[1], x21\n" "fmla v23.8h, v28.8h, v3.h[1]\n" - "prfm pldl1keep, [x10, #0x80]\n" + "ldr x21, [x17, #0xa8]\n" "fmla v27.8h, v28.8h, v4.h[1]\n" "ldr d28, [x17, #0x90]\n" - "mov v28.d[1], x20\n" "fmla v8.8h, v29.8h, v0.h[2]\n" + "prfm pldl1keep, [x9, #0x80]\n" "fmla v12.8h, v29.8h, v1.h[2]\n" - "ldr x20, [x17, #0xb8]\n" + "mov v28.d[1], x20\n" "fmla v16.8h, v29.8h, v2.h[2]\n" - "prfm pldl1keep, [x9, #0x80]\n" "fmla v20.8h, v29.8h, v3.h[2]\n" + "ldr x20, [x17, #0xb8]\n" "fmla v24.8h, v29.8h, v4.h[2]\n" "ldr d29, [x17, #0xa0]\n" "fmla v9.8h, v28.8h, v0.h[2]\n" - "mov v29.d[1], x21\n" "fmla v13.8h, v28.8h, v1.h[2]\n" - "ldr x21, [x17, #0xc8]\n" "fmla v17.8h, v28.8h, v2.h[2]\n" + "mov v29.d[1], x21\n" "fmla v21.8h, v28.8h, v3.h[2]\n" + "ldr x21, [x17, #0xc8]\n" "fmla v25.8h, v28.8h, v4.h[2]\n" "ldr d28, [x17, #0xb0]\n" - "mov v28.d[1], x20\n" "fmla v10.8h, v29.8h, v0.h[2]\n" "fmla v14.8h, v29.8h, v1.h[2]\n" - "ldr x20, [x17, #0xd8]\n" + "mov v28.d[1], x20\n" "fmla v18.8h, v29.8h, v2.h[2]\n" "fmla v22.8h, v29.8h, v3.h[2]\n" + "ldr x20, [x17, #0xd8]\n" "fmla v26.8h, v29.8h, v4.h[2]\n" "ldr d29, [x17, #0xc0]\n" "fmla v11.8h, v28.8h, v0.h[2]\n" - "mov v29.d[1], x21\n" "fmla v15.8h, v28.8h, v1.h[2]\n" - "ldr x21, [x17, #0xe8]\n" "fmla v19.8h, v28.8h, v2.h[2]\n" + "mov v29.d[1], x21\n" "fmla v23.8h, v28.8h, v3.h[2]\n" + "ldr x21, [x17, #0xe8]\n" "fmla v27.8h, v28.8h, v4.h[2]\n" "ldr d28, [x17, #0xd0]\n" - "mov v28.d[1], x20\n" "fmla v8.8h, v29.8h, v0.h[3]\n" "fmla v12.8h, v29.8h, v1.h[3]\n" - "ldr x20, [x17, #0xf8]\n" + "mov v28.d[1], x20\n" "fmla v16.8h, v29.8h, v2.h[3]\n" "fmla v20.8h, v29.8h, v3.h[3]\n" + "ldr x20, [x17, #0xf8]\n" "fmla v24.8h, v29.8h, v4.h[3]\n" "ldr d29, [x17, #0xe0]\n" "fmla v9.8h, v28.8h, v0.h[3]\n" - "mov v29.d[1], x21\n" "fmla v13.8h, v28.8h, v1.h[3]\n" - "ldr x21, [x17, #0x108]\n" "fmla v17.8h, v28.8h, v2.h[3]\n" + "mov v29.d[1], x21\n" "fmla v21.8h, v28.8h, v3.h[3]\n" + "ldr x21, [x17, #0x108]\n" "fmla v25.8h, v28.8h, v4.h[3]\n" "ldr d28, [x17, #0xf0]\n" - "mov v28.d[1], x20\n" "fmla v10.8h, v29.8h, v0.h[3]\n" "fmla v14.8h, v29.8h, v1.h[3]\n" - "ldr x20, [x17, #0x118]\n" + "mov v28.d[1], x20\n" "fmla v18.8h, v29.8h, v2.h[3]\n" "fmla v22.8h, v29.8h, v3.h[3]\n" + "ldr x20, [x17, #0x118]\n" "fmla v26.8h, v29.8h, v4.h[3]\n" "ldr d29, [x17, #0x100]\n" "fmla v11.8h, v28.8h, v0.h[3]\n" - "mov v29.d[1], x21\n" "fmla v15.8h, v28.8h, v1.h[3]\n" - "ldr x21, [x17, #0x128]\n" "fmla v19.8h, v28.8h, v2.h[3]\n" + "mov v29.d[1], x21\n" "fmla v23.8h, v28.8h, v3.h[3]\n" + "ldr x21, [x17, #0x128]\n" "fmla v27.8h, v28.8h, v4.h[3]\n" "ldr d28, [x17, #0x110]\n" - "mov v28.d[1], x20\n" "fmla v8.8h, v29.8h, v0.h[4]\n" "fmla v12.8h, v29.8h, v1.h[4]\n" - "ldr x20, [x17, #0x138]\n" + "mov v28.d[1], x20\n" "fmla v16.8h, v29.8h, v2.h[4]\n" "fmla v20.8h, v29.8h, v3.h[4]\n" + "ldr x20, [x17, #0x138]\n" "fmla v24.8h, v29.8h, v4.h[4]\n" "ldr d29, [x17, #0x120]\n" "fmla v9.8h, v28.8h, v0.h[4]\n" - "mov v29.d[1], x21\n" "fmla v13.8h, v28.8h, v1.h[4]\n" - "ldr x21, [x17, #0x148]\n" "fmla v17.8h, v28.8h, v2.h[4]\n" + "mov v29.d[1], x21\n" "fmla v21.8h, v28.8h, v3.h[4]\n" + "ldr x21, [x17, #0x148]\n" "fmla v25.8h, v28.8h, v4.h[4]\n" "ldr d28, [x17, #0x130]\n" - "mov v28.d[1], x20\n" "fmla v10.8h, v29.8h, v0.h[4]\n" "fmla v14.8h, v29.8h, v1.h[4]\n" - "ldr x20, [x17, #0x158]\n" + "mov v28.d[1], x20\n" "fmla v18.8h, v29.8h, v2.h[4]\n" "fmla v22.8h, v29.8h, v3.h[4]\n" + "ldr x20, [x17, #0x158]\n" "fmla v26.8h, v29.8h, v4.h[4]\n" "ldr d29, [x17, #0x140]\n" "fmla v11.8h, v28.8h, v0.h[4]\n" - "mov v29.d[1], x21\n" "fmla v15.8h, v28.8h, v1.h[4]\n" - "ldr x21, [x17, #0x168]\n" "fmla v19.8h, v28.8h, v2.h[4]\n" + "mov v29.d[1], x21\n" "fmla v23.8h, v28.8h, v3.h[4]\n" + "ldr x21, [x17, #0x168]\n" "fmla v27.8h, v28.8h, v4.h[4]\n" "ldr d28, [x17, #0x150]\n" - "mov v28.d[1], x20\n" "fmla v8.8h, v29.8h, v0.h[5]\n" "fmla v12.8h, v29.8h, v1.h[5]\n" - "ldr x20, [x17, #0x178]\n" + "mov v28.d[1], x20\n" "fmla v16.8h, v29.8h, v2.h[5]\n" "fmla v20.8h, v29.8h, v3.h[5]\n" + "ldr x20, [x17, #0x178]\n" "fmla v24.8h, v29.8h, v4.h[5]\n" "ldr d29, [x17, #0x160]\n" "fmla v9.8h, v28.8h, v0.h[5]\n" - "mov v29.d[1], x21\n" "fmla v13.8h, v28.8h, v1.h[5]\n" - "ldr x21, [x17, #0x188]\n" "fmla v17.8h, v28.8h, v2.h[5]\n" + "mov v29.d[1], x21\n" "fmla v21.8h, v28.8h, v3.h[5]\n" + "ldr x21, [x17, #0x188]\n" "fmla v25.8h, v28.8h, v4.h[5]\n" "ldr d28, [x17, #0x170]\n" - "mov v28.d[1], x20\n" "fmla v10.8h, v29.8h, v0.h[5]\n" "fmla v14.8h, v29.8h, v1.h[5]\n" - "ldr x20, [x17, #0x198]\n" + "mov v28.d[1], x20\n" "fmla v18.8h, v29.8h, v2.h[5]\n" "fmla v22.8h, v29.8h, v3.h[5]\n" + "ldr x20, [x17, #0x198]\n" "fmla v26.8h, v29.8h, v4.h[5]\n" "ldr d29, [x17, #0x180]\n" "fmla v11.8h, v28.8h, v0.h[5]\n" - "mov v29.d[1], x21\n" "fmla v15.8h, v28.8h, v1.h[5]\n" - "ldr x21, [x17, #0x1a8]\n" "fmla v19.8h, v28.8h, v2.h[5]\n" + "mov v29.d[1], x21\n" "fmla v23.8h, v28.8h, v3.h[5]\n" + "ldr x21, [x17, #0x1a8]\n" "fmla v27.8h, v28.8h, v4.h[5]\n" "ldr d28, [x17, #0x190]\n" - "mov v28.d[1], x20\n" "fmla v8.8h, v29.8h, v0.h[6]\n" "fmla v12.8h, v29.8h, v1.h[6]\n" - "ldr x20, [x17, #0x1b8]\n" + "mov v28.d[1], x20\n" "fmla v16.8h, v29.8h, v2.h[6]\n" "fmla v20.8h, v29.8h, v3.h[6]\n" + "ldr x20, [x17, #0x1b8]\n" "fmla v24.8h, v29.8h, v4.h[6]\n" "ldr d29, [x17, #0x1a0]\n" "fmla v9.8h, v28.8h, v0.h[6]\n" - "mov v29.d[1], x21\n" "fmla v13.8h, v28.8h, v1.h[6]\n" - "ldr x21, [x17, #0x1c8]\n" "fmla v17.8h, v28.8h, v2.h[6]\n" + "mov v29.d[1], x21\n" "fmla v21.8h, v28.8h, v3.h[6]\n" + "ldr x21, [x17, #0x1c8]\n" "fmla v25.8h, v28.8h, v4.h[6]\n" "ldr d28, [x17, #0x1b0]\n" - "mov v28.d[1], x20\n" "fmla v10.8h, v29.8h, v0.h[6]\n" "fmla v14.8h, v29.8h, v1.h[6]\n" - "ldr x20, [x17, #0x1d8]\n" + "mov v28.d[1], x20\n" "fmla v18.8h, v29.8h, v2.h[6]\n" "fmla v22.8h, v29.8h, v3.h[6]\n" + "ldr x20, [x17, #0x1d8]\n" "fmla v26.8h, v29.8h, v4.h[6]\n" "ldr d29, [x17, #0x1c0]\n" "fmla v11.8h, v28.8h, v0.h[6]\n" - "mov v29.d[1], x21\n" "fmla v15.8h, v28.8h, v1.h[6]\n" - "ldr x21, [x17, #0x1e8]\n" "fmla v19.8h, v28.8h, v2.h[6]\n" + "mov v29.d[1], x21\n" "fmla v23.8h, v28.8h, v3.h[6]\n" + "ldr x21, [x17, #0x1e8]\n" "fmla v27.8h, v28.8h, v4.h[6]\n" "ldr d28, [x17, #0x1d0]\n" - "mov v28.d[1], x20\n" "fmla v8.8h, v29.8h, v0.h[7]\n" "fmla v12.8h, v29.8h, v1.h[7]\n" - "ldr x20, [x17, #0x1f8]\n" + "mov v28.d[1], x20\n" "fmla v16.8h, v29.8h, v2.h[7]\n" "fmla v20.8h, v29.8h, v3.h[7]\n" + "ldr x20, [x17, #0x1f8]\n" "fmla v24.8h, v29.8h, v4.h[7]\n" "ldr d29, [x17, #0x1e0]\n" "fmla v9.8h, v28.8h, v0.h[7]\n" - "mov v29.d[1], x21\n" "fmla v13.8h, v28.8h, v1.h[7]\n" "fmla v17.8h, v28.8h, v2.h[7]\n" + "mov v29.d[1], x21\n" "fmla v21.8h, v28.8h, v3.h[7]\n" "fmla v25.8h, v28.8h, v4.h[7]\n" "ldr d28, [x17, #0x1f0]\n" - "mov v28.d[1], x20\n" "add x17, x17, #0x200\n" "fmla v10.8h, v29.8h, v0.h[7]\n" - "ldr x21, [x17, #0x8]\n" "fmla v14.8h, v29.8h, v1.h[7]\n" - "ldr x20, [x17, #0x18]\n" + "ldr x21, [x17, #0x8]\n" + "mov v28.d[1], x20\n" "fmla v18.8h, v29.8h, v2.h[7]\n" "fmla v22.8h, v29.8h, v3.h[7]\n" + "ldr x20, [x17, #0x18]\n" "fmla v26.8h, v29.8h, v4.h[7]\n" "ldr d6, [x17, #0x0]\n" "fmla v11.8h, v28.8h, v0.h[7]\n" @@ -4027,8 +4029,8 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "ldr h1, [x10], #0x2\n" "ldr h0, [x9], #0x2\n" "ldr q29, [x17, #0x0]\n" - "fmla v8.8h, v29.8h, v4.h[0]\n" "ldr q28, [x17, #0x10]\n" + "fmla v8.8h, v29.8h, v4.h[0]\n" "fmla v12.8h, v29.8h, v3.h[0]\n" "fmla v16.8h, v29.8h, v2.h[0]\n" "fmla v20.8h, v29.8h, v1.h[0]\n" @@ -4058,40 +4060,40 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "cmp x15, x20\n" "bne 219b\n" "ldr x20, [%x[args_ptr], %[offsetof_output_offset]]\n" - "add x25, x16, x20, LSL #1\n" - "add x24, x25, x20, LSL #1\n" - "add x23, x24, x20, LSL #1\n" - "add x22, x23, x20, LSL #1\n" "prfm pstl1keep, [x16, #0x0]\n" + "add x26, x16, x20, LSL #1\n" + "add x25, x26, x20, LSL #1\n" + "prfm pstl1keep, [x26, #0x0]\n" + "add x24, x25, x20, LSL #1\n" "prfm pstl1keep, [x25, #0x0]\n" + "add x23, x24, x20, LSL #1\n" "prfm pstl1keep, [x24, #0x0]\n" "prfm pstl1keep, [x23, #0x0]\n" - "prfm pstl1keep, [x22, #0x0]\n" "tbz %x[flags], #1, 227f\n" - "add x20, %x[args_ptr], %[offset_max]\n" - "ld1r { v28.8h }, [x20]\n" - "fmin v8.8h, v8.8h, v28.8h\n" - "fmin v9.8h, v9.8h, v28.8h\n" - "fmin v10.8h, v10.8h, v28.8h\n" - "fmin v11.8h, v11.8h, v28.8h\n" - "fmin v12.8h, v12.8h, v28.8h\n" - "fmin v13.8h, v13.8h, v28.8h\n" - "fmin v14.8h, v14.8h, v28.8h\n" - "fmin v15.8h, v15.8h, v28.8h\n" - "fmin v16.8h, v16.8h, v28.8h\n" - "fmin v17.8h, v17.8h, v28.8h\n" - "fmin v18.8h, v18.8h, v28.8h\n" - "fmin v19.8h, v19.8h, v28.8h\n" - "fmin v20.8h, v20.8h, v28.8h\n" - "fmin v21.8h, v21.8h, v28.8h\n" - "fmin v22.8h, v22.8h, v28.8h\n" - "fmin v23.8h, v23.8h, v28.8h\n" - "fmin v24.8h, v24.8h, v28.8h\n" - "fmin v25.8h, v25.8h, v28.8h\n" - "fmin v26.8h, v26.8h, v28.8h\n" - "fmin v27.8h, v27.8h, v28.8h\n" + "add x21, %x[args_ptr], %[offset_max]\n" "add x20, %x[args_ptr], %[offset_min]\n" + "ld1r { v29.8h }, [x21]\n" "ld1r { v28.8h }, [x20]\n" + "fmin v8.8h, v8.8h, v29.8h\n" + "fmin v9.8h, v9.8h, v29.8h\n" + "fmin v10.8h, v10.8h, v29.8h\n" + "fmin v11.8h, v11.8h, v29.8h\n" + "fmin v12.8h, v12.8h, v29.8h\n" + "fmin v13.8h, v13.8h, v29.8h\n" + "fmin v14.8h, v14.8h, v29.8h\n" + "fmin v15.8h, v15.8h, v29.8h\n" + "fmin v16.8h, v16.8h, v29.8h\n" + "fmin v17.8h, v17.8h, v29.8h\n" + "fmin v18.8h, v18.8h, v29.8h\n" + "fmin v19.8h, v19.8h, v29.8h\n" + "fmin v20.8h, v20.8h, v29.8h\n" + "fmin v21.8h, v21.8h, v29.8h\n" + "fmin v22.8h, v22.8h, v29.8h\n" + "fmin v23.8h, v23.8h, v29.8h\n" + "fmin v24.8h, v24.8h, v29.8h\n" + "fmin v25.8h, v25.8h, v29.8h\n" + "fmin v26.8h, v26.8h, v29.8h\n" + "fmin v27.8h, v27.8h, v29.8h\n" "fmax v8.8h, v8.8h, v28.8h\n" "fmax v9.8h, v9.8h, v28.8h\n" "fmax v10.8h, v10.8h, v28.8h\n" @@ -4118,223 +4120,223 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "tbz x8, #4, 235f\n" "st1 { v8.8h }, [x16], #0x10\n" "st1 { v9.8h }, [x16], #0x10\n" - "st1 { v12.8h }, [x25], #0x10\n" - "st1 { v13.8h }, [x25], #0x10\n" - "st1 { v16.8h }, [x24], #0x10\n" - "st1 { v17.8h }, [x24], #0x10\n" - "st1 { v20.8h }, [x23], #0x10\n" - "st1 { v21.8h }, [x23], #0x10\n" - "st1 { v24.8h }, [x22], #0x10\n" - "st1 { v25.8h }, [x22], #0x10\n" + "st1 { v12.8h }, [x26], #0x10\n" + "st1 { v13.8h }, [x26], #0x10\n" + "st1 { v16.8h }, [x25], #0x10\n" + "st1 { v17.8h }, [x25], #0x10\n" + "st1 { v20.8h }, [x24], #0x10\n" + "st1 { v21.8h }, [x24], #0x10\n" + "st1 { v24.8h }, [x23], #0x10\n" + "st1 { v25.8h }, [x23], #0x10\n" "tbz x8, #3, 231f\n" "st1 { v10.8h }, [x16], #0x10\n" - "st1 { v14.8h }, [x25], #0x10\n" - "st1 { v18.8h }, [x24], #0x10\n" - "st1 { v22.8h }, [x23], #0x10\n" - "st1 { v26.8h }, [x22], #0x10\n" + "st1 { v14.8h }, [x26], #0x10\n" + "st1 { v18.8h }, [x25], #0x10\n" + "st1 { v22.8h }, [x24], #0x10\n" + "st1 { v26.8h }, [x23], #0x10\n" "tbz x8, #2, 229f\n" "str d11, [x16], #0x8\n" - "str d15, [x25], #0x8\n" - "str d19, [x24], #0x8\n" - "str d23, [x23], #0x8\n" - "str d27, [x22], #0x8\n" + "str d15, [x26], #0x8\n" + "str d19, [x25], #0x8\n" + "str d23, [x24], #0x8\n" + "str d27, [x23], #0x8\n" "tbz x8, #1, 228f\n" "st1 { v11.s }[2], [x16], #0x4\n" - "st1 { v15.s }[2], [x25], #0x4\n" - "st1 { v19.s }[2], [x24], #0x4\n" - "st1 { v23.s }[2], [x23], #0x4\n" - "st1 { v27.s }[2], [x22], #0x4\n" + "st1 { v15.s }[2], [x26], #0x4\n" + "st1 { v19.s }[2], [x25], #0x4\n" + "st1 { v23.s }[2], [x24], #0x4\n" + "st1 { v27.s }[2], [x23], #0x4\n" "tbz x8, #0, 243f\n" "st1 { v11.h }[6], [x16]\n" - "st1 { v15.h }[6], [x25]\n" - "st1 { v19.h }[6], [x24]\n" - "st1 { v23.h }[6], [x23]\n" - "st1 { v27.h }[6], [x22]\n" + "st1 { v15.h }[6], [x26]\n" + "st1 { v19.h }[6], [x25]\n" + "st1 { v23.h }[6], [x24]\n" + "st1 { v27.h }[6], [x23]\n" "b 243f\n" "228:" // Height 5: Partial direct writeback: partial_1_28 "tbz x8, #0, 243f\n" "st1 { v11.h }[4], [x16]\n" - "st1 { v15.h }[4], [x25]\n" - "st1 { v19.h }[4], [x24]\n" - "st1 { v23.h }[4], [x23]\n" - "st1 { v27.h }[4], [x22]\n" + "st1 { v15.h }[4], [x26]\n" + "st1 { v19.h }[4], [x25]\n" + "st1 { v23.h }[4], [x24]\n" + "st1 { v27.h }[4], [x23]\n" "b 243f\n" "229:" // Height 5: Partial direct writeback: partial_2_24 "tbz x8, #1, 230f\n" "str s11, [x16], #0x4\n" - "str s15, [x25], #0x4\n" - "str s19, [x24], #0x4\n" - "str s23, [x23], #0x4\n" - "str s27, [x22], #0x4\n" + "str s15, [x26], #0x4\n" + "str s19, [x25], #0x4\n" + "str s23, [x24], #0x4\n" + "str s27, [x23], #0x4\n" "tbz x8, #0, 243f\n" "st1 { v11.h }[2], [x16]\n" - "st1 { v15.h }[2], [x25]\n" - "st1 { v19.h }[2], [x24]\n" - "st1 { v23.h }[2], [x23]\n" - "st1 { v27.h }[2], [x22]\n" + "st1 { v15.h }[2], [x26]\n" + "st1 { v19.h }[2], [x25]\n" + "st1 { v23.h }[2], [x24]\n" + "st1 { v27.h }[2], [x23]\n" "b 243f\n" "230:" // Height 5: Partial direct writeback: partial_1_24 "tbz x8, #0, 243f\n" "str h11, [x16, #0x0]\n" - "str h15, [x25, #0x0]\n" - "str h19, [x24, #0x0]\n" - "str h23, [x23, #0x0]\n" - "str h27, [x22, #0x0]\n" + "str h15, [x26, #0x0]\n" + "str h19, [x25, #0x0]\n" + "str h23, [x24, #0x0]\n" + "str h27, [x23, #0x0]\n" "b 243f\n" "231:" // Height 5: Partial direct writeback: partial_4_16 "tbz x8, #2, 233f\n" "str d10, [x16], #0x8\n" - "str d14, [x25], #0x8\n" - "str d18, [x24], #0x8\n" - "str d22, [x23], #0x8\n" - "str d26, [x22], #0x8\n" + "str d14, [x26], #0x8\n" + "str d18, [x25], #0x8\n" + "str d22, [x24], #0x8\n" + "str d26, [x23], #0x8\n" "tbz x8, #1, 232f\n" "st1 { v10.s }[2], [x16], #0x4\n" - "st1 { v14.s }[2], [x25], #0x4\n" - "st1 { v18.s }[2], [x24], #0x4\n" - "st1 { v22.s }[2], [x23], #0x4\n" - "st1 { v26.s }[2], [x22], #0x4\n" + "st1 { v14.s }[2], [x26], #0x4\n" + "st1 { v18.s }[2], [x25], #0x4\n" + "st1 { v22.s }[2], [x24], #0x4\n" + "st1 { v26.s }[2], [x23], #0x4\n" "tbz x8, #0, 243f\n" "st1 { v10.h }[6], [x16]\n" - "st1 { v14.h }[6], [x25]\n" - "st1 { v18.h }[6], [x24]\n" - "st1 { v22.h }[6], [x23]\n" - "st1 { v26.h }[6], [x22]\n" + "st1 { v14.h }[6], [x26]\n" + "st1 { v18.h }[6], [x25]\n" + "st1 { v22.h }[6], [x24]\n" + "st1 { v26.h }[6], [x23]\n" "b 243f\n" "232:" // Height 5: Partial direct writeback: partial_1_20 "tbz x8, #0, 243f\n" "st1 { v10.h }[4], [x16]\n" - "st1 { v14.h }[4], [x25]\n" - "st1 { v18.h }[4], [x24]\n" - "st1 { v22.h }[4], [x23]\n" - "st1 { v26.h }[4], [x22]\n" + "st1 { v14.h }[4], [x26]\n" + "st1 { v18.h }[4], [x25]\n" + "st1 { v22.h }[4], [x24]\n" + "st1 { v26.h }[4], [x23]\n" "b 243f\n" "233:" // Height 5: Partial direct writeback: partial_2_16 "tbz x8, #1, 234f\n" "str s10, [x16], #0x4\n" - "str s14, [x25], #0x4\n" - "str s18, [x24], #0x4\n" - "str s22, [x23], #0x4\n" - "str s26, [x22], #0x4\n" + "str s14, [x26], #0x4\n" + "str s18, [x25], #0x4\n" + "str s22, [x24], #0x4\n" + "str s26, [x23], #0x4\n" "tbz x8, #0, 243f\n" "st1 { v10.h }[2], [x16]\n" - "st1 { v14.h }[2], [x25]\n" - "st1 { v18.h }[2], [x24]\n" - "st1 { v22.h }[2], [x23]\n" - "st1 { v26.h }[2], [x22]\n" + "st1 { v14.h }[2], [x26]\n" + "st1 { v18.h }[2], [x25]\n" + "st1 { v22.h }[2], [x24]\n" + "st1 { v26.h }[2], [x23]\n" "b 243f\n" "234:" // Height 5: Partial direct writeback: partial_1_16 "tbz x8, #0, 243f\n" "str h10, [x16, #0x0]\n" - "str h14, [x25, #0x0]\n" - "str h18, [x24, #0x0]\n" - "str h22, [x23, #0x0]\n" - "str h26, [x22, #0x0]\n" + "str h14, [x26, #0x0]\n" + "str h18, [x25, #0x0]\n" + "str h22, [x24, #0x0]\n" + "str h26, [x23, #0x0]\n" "b 243f\n" "235:" // Height 5: Partial direct writeback: partial_8_0 "tbz x8, #3, 239f\n" "st1 { v8.8h }, [x16], #0x10\n" - "st1 { v12.8h }, [x25], #0x10\n" - "st1 { v16.8h }, [x24], #0x10\n" - "st1 { v20.8h }, [x23], #0x10\n" - "st1 { v24.8h }, [x22], #0x10\n" + "st1 { v12.8h }, [x26], #0x10\n" + "st1 { v16.8h }, [x25], #0x10\n" + "st1 { v20.8h }, [x24], #0x10\n" + "st1 { v24.8h }, [x23], #0x10\n" "tbz x8, #2, 237f\n" "str d9, [x16], #0x8\n" - "str d13, [x25], #0x8\n" - "str d17, [x24], #0x8\n" - "str d21, [x23], #0x8\n" - "str d25, [x22], #0x8\n" + "str d13, [x26], #0x8\n" + "str d17, [x25], #0x8\n" + "str d21, [x24], #0x8\n" + "str d25, [x23], #0x8\n" "tbz x8, #1, 236f\n" "st1 { v9.s }[2], [x16], #0x4\n" - "st1 { v13.s }[2], [x25], #0x4\n" - "st1 { v17.s }[2], [x24], #0x4\n" - "st1 { v21.s }[2], [x23], #0x4\n" - "st1 { v25.s }[2], [x22], #0x4\n" + "st1 { v13.s }[2], [x26], #0x4\n" + "st1 { v17.s }[2], [x25], #0x4\n" + "st1 { v21.s }[2], [x24], #0x4\n" + "st1 { v25.s }[2], [x23], #0x4\n" "tbz x8, #0, 243f\n" "st1 { v9.h }[6], [x16]\n" - "st1 { v13.h }[6], [x25]\n" - "st1 { v17.h }[6], [x24]\n" - "st1 { v21.h }[6], [x23]\n" - "st1 { v25.h }[6], [x22]\n" + "st1 { v13.h }[6], [x26]\n" + "st1 { v17.h }[6], [x25]\n" + "st1 { v21.h }[6], [x24]\n" + "st1 { v25.h }[6], [x23]\n" "b 243f\n" "236:" // Height 5: Partial direct writeback: partial_1_12 "tbz x8, #0, 243f\n" "st1 { v9.h }[4], [x16]\n" - "st1 { v13.h }[4], [x25]\n" - "st1 { v17.h }[4], [x24]\n" - "st1 { v21.h }[4], [x23]\n" - "st1 { v25.h }[4], [x22]\n" + "st1 { v13.h }[4], [x26]\n" + "st1 { v17.h }[4], [x25]\n" + "st1 { v21.h }[4], [x24]\n" + "st1 { v25.h }[4], [x23]\n" "b 243f\n" "237:" // Height 5: Partial direct writeback: partial_2_8 "tbz x8, #1, 238f\n" "str s9, [x16], #0x4\n" - "str s13, [x25], #0x4\n" - "str s17, [x24], #0x4\n" - "str s21, [x23], #0x4\n" - "str s25, [x22], #0x4\n" + "str s13, [x26], #0x4\n" + "str s17, [x25], #0x4\n" + "str s21, [x24], #0x4\n" + "str s25, [x23], #0x4\n" "tbz x8, #0, 243f\n" "st1 { v9.h }[2], [x16]\n" - "st1 { v13.h }[2], [x25]\n" - "st1 { v17.h }[2], [x24]\n" - "st1 { v21.h }[2], [x23]\n" - "st1 { v25.h }[2], [x22]\n" + "st1 { v13.h }[2], [x26]\n" + "st1 { v17.h }[2], [x25]\n" + "st1 { v21.h }[2], [x24]\n" + "st1 { v25.h }[2], [x23]\n" "b 243f\n" "238:" // Height 5: Partial direct writeback: partial_1_8 "tbz x8, #0, 243f\n" "str h9, [x16, #0x0]\n" - "str h13, [x25, #0x0]\n" - "str h17, [x24, #0x0]\n" - "str h21, [x23, #0x0]\n" - "str h25, [x22, #0x0]\n" + "str h13, [x26, #0x0]\n" + "str h17, [x25, #0x0]\n" + "str h21, [x24, #0x0]\n" + "str h25, [x23, #0x0]\n" "b 243f\n" "239:" // Height 5: Partial direct writeback: partial_4_0 "tbz x8, #2, 241f\n" "str d8, [x16], #0x8\n" - "str d12, [x25], #0x8\n" - "str d16, [x24], #0x8\n" - "str d20, [x23], #0x8\n" - "str d24, [x22], #0x8\n" + "str d12, [x26], #0x8\n" + "str d16, [x25], #0x8\n" + "str d20, [x24], #0x8\n" + "str d24, [x23], #0x8\n" "tbz x8, #1, 240f\n" "st1 { v8.s }[2], [x16], #0x4\n" - "st1 { v12.s }[2], [x25], #0x4\n" - "st1 { v16.s }[2], [x24], #0x4\n" - "st1 { v20.s }[2], [x23], #0x4\n" - "st1 { v24.s }[2], [x22], #0x4\n" + "st1 { v12.s }[2], [x26], #0x4\n" + "st1 { v16.s }[2], [x25], #0x4\n" + "st1 { v20.s }[2], [x24], #0x4\n" + "st1 { v24.s }[2], [x23], #0x4\n" "tbz x8, #0, 243f\n" "st1 { v8.h }[6], [x16]\n" - "st1 { v12.h }[6], [x25]\n" - "st1 { v16.h }[6], [x24]\n" - "st1 { v20.h }[6], [x23]\n" - "st1 { v24.h }[6], [x22]\n" + "st1 { v12.h }[6], [x26]\n" + "st1 { v16.h }[6], [x25]\n" + "st1 { v20.h }[6], [x24]\n" + "st1 { v24.h }[6], [x23]\n" "b 243f\n" "240:" // Height 5: Partial direct writeback: partial_1_4 "tbz x8, #0, 243f\n" "st1 { v8.h }[4], [x16]\n" - "st1 { v12.h }[4], [x25]\n" - "st1 { v16.h }[4], [x24]\n" - "st1 { v20.h }[4], [x23]\n" - "st1 { v24.h }[4], [x22]\n" + "st1 { v12.h }[4], [x26]\n" + "st1 { v16.h }[4], [x25]\n" + "st1 { v20.h }[4], [x24]\n" + "st1 { v24.h }[4], [x23]\n" "b 243f\n" "241:" // Height 5: Partial direct writeback: partial_2_0 "tbz x8, #1, 242f\n" "str s8, [x16], #0x4\n" - "str s12, [x25], #0x4\n" - "str s16, [x24], #0x4\n" - "str s20, [x23], #0x4\n" - "str s24, [x22], #0x4\n" + "str s12, [x26], #0x4\n" + "str s16, [x25], #0x4\n" + "str s20, [x24], #0x4\n" + "str s24, [x23], #0x4\n" "tbz x8, #0, 243f\n" "st1 { v8.h }[2], [x16]\n" - "st1 { v12.h }[2], [x25]\n" - "st1 { v16.h }[2], [x24]\n" - "st1 { v20.h }[2], [x23]\n" - "st1 { v24.h }[2], [x22]\n" + "st1 { v12.h }[2], [x26]\n" + "st1 { v16.h }[2], [x25]\n" + "st1 { v20.h }[2], [x24]\n" + "st1 { v24.h }[2], [x23]\n" "b 243f\n" "242:" // Height 5: Partial direct writeback: partial_1_0 "str h8, [x16, #0x0]\n" - "str h12, [x25, #0x0]\n" - "str h16, [x24, #0x0]\n" - "str h20, [x23, #0x0]\n" - "str h24, [x22, #0x0]\n" + "str h12, [x26, #0x0]\n" + "str h16, [x25, #0x0]\n" + "str h20, [x24, #0x0]\n" + "str h24, [x23, #0x0]\n" "243:" // Height 5: Partial direct writeback: Done "b 245f\n" "244:" // Height 5: Full writeback @@ -4343,22 +4345,22 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "str q10, [x16, #0x20]\n" "str q11, [x16, #0x30]\n" "add x16, x16, #0x40\n" - "str q12, [x25, #0x0]\n" - "str q13, [x25, #0x10]\n" - "str q14, [x25, #0x20]\n" - "str q15, [x25, #0x30]\n" - "str q16, [x24, #0x0]\n" - "str q17, [x24, #0x10]\n" - "str q18, [x24, #0x20]\n" - "str q19, [x24, #0x30]\n" - "str q20, [x23, #0x0]\n" - "str q21, [x23, #0x10]\n" - "str q22, [x23, #0x20]\n" - "str q23, [x23, #0x30]\n" - "str q24, [x22, #0x0]\n" - "str q25, [x22, #0x10]\n" - "str q26, [x22, #0x20]\n" - "str q27, [x22, #0x30]\n" + "str q12, [x26, #0x0]\n" + "str q13, [x26, #0x10]\n" + "str q14, [x26, #0x20]\n" + "str q15, [x26, #0x30]\n" + "str q16, [x25, #0x0]\n" + "str q17, [x25, #0x10]\n" + "str q18, [x25, #0x20]\n" + "str q19, [x25, #0x30]\n" + "str q20, [x24, #0x0]\n" + "str q21, [x24, #0x10]\n" + "str q22, [x24, #0x20]\n" + "str q23, [x24, #0x30]\n" + "str q24, [x23, #0x0]\n" + "str q25, [x23, #0x10]\n" + "str q26, [x23, #0x20]\n" + "str q27, [x23, #0x30]\n" "245:" // Height 5: Writeback done "subs x8, x8, #0x20\n" "bgt 198b\n" @@ -4366,23 +4368,24 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "246:" // Height 6 "ldr x21, [%x[args_ptr], %[offsetof_output_offset]]\n" "mov x20, #0xc\n" - "mov x7, %x[bias]\n" + "ldr x16, [%x[args_ptr], %[offsetof_output_ptr]]\n" + "ldr x7, [%x[args_ptr], %[offsetof_bias]]\n" "ldr x8, [%x[args_ptr], %[offsetof_N]]\n" + "madd x20, x21, x20, x16\n" "ldr x17, [%x[args_ptr], %[offsetof_B_ptr]]\n" - "mov x16, %x[output_ptr]\n" - "madd %x[output_ptr], x21, x20, %x[output_ptr]\n" + "str x20, [%x[args_ptr], %[offsetof_output_ptr]]\n" "247:" // Height 6: Column loop "cbz x7, 248f\n" "ldr q8, [x7, #0x0]\n" - "mov v12.16b, v8.16b\n" "ldr q9, [x7, #0x10]\n" - "mov v13.16b, v9.16b\n" "ldr q10, [x7, #0x20]\n" - "mov v14.16b, v10.16b\n" "ldr q11, [x7, #0x30]\n" + "mov v12.16b, v8.16b\n" + "mov v13.16b, v9.16b\n" + "add x7, x7, #0x40\n" + "mov v14.16b, v10.16b\n" "mov v15.16b, v11.16b\n" "mov v16.16b, v8.16b\n" - "add x7, x7, #0x40\n" "mov v17.16b, v9.16b\n" "mov v18.16b, v10.16b\n" "mov v19.16b, v11.16b\n" @@ -4402,281 +4405,281 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "248:" // Height 6: no bias "tbz %x[flags], #0, 266f\n" "ldr x20, [%x[args_ptr], %[offsetof_output_offset]]\n" - "add x25, x16, x20, LSL #1\n" + "cmp x8, #0x20\n" + "add x26, x16, x20, LSL #1\n" + "add x25, x26, x20, LSL #1\n" "add x24, x25, x20, LSL #1\n" "add x23, x24, x20, LSL #1\n" "add x22, x23, x20, LSL #1\n" - "cmp x8, #0x20\n" - "add x21, x22, x20, LSL #1\n" "bge 265f\n" "tbz x8, #4, 256f\n" "ld1 { v8.8h }, [x16], #0x10\n" - "ld1 { v12.8h }, [x25], #0x10\n" - "ld1 { v16.8h }, [x24], #0x10\n" - "ld1 { v20.8h }, [x23], #0x10\n" - "ld1 { v24.8h }, [x22], #0x10\n" - "ld1 { v28.8h }, [x21], #0x10\n" + "ld1 { v12.8h }, [x26], #0x10\n" + "ld1 { v16.8h }, [x25], #0x10\n" + "ld1 { v20.8h }, [x24], #0x10\n" + "ld1 { v24.8h }, [x23], #0x10\n" + "ld1 { v28.8h }, [x22], #0x10\n" "ld1 { v9.8h }, [x16], #0x10\n" - "ld1 { v13.8h }, [x25], #0x10\n" - "ld1 { v17.8h }, [x24], #0x10\n" - "ld1 { v21.8h }, [x23], #0x10\n" - "ld1 { v25.8h }, [x22], #0x10\n" - "ld1 { v29.8h }, [x21], #0x10\n" + "ld1 { v13.8h }, [x26], #0x10\n" + "ld1 { v17.8h }, [x25], #0x10\n" + "ld1 { v21.8h }, [x24], #0x10\n" + "ld1 { v25.8h }, [x23], #0x10\n" + "ld1 { v29.8h }, [x22], #0x10\n" "tbz x8, #3, 252f\n" "ld1 { v10.8h }, [x16], #0x10\n" - "ld1 { v14.8h }, [x25], #0x10\n" - "ld1 { v18.8h }, [x24], #0x10\n" - "ld1 { v22.8h }, [x23], #0x10\n" - "ld1 { v26.8h }, [x22], #0x10\n" - "ld1 { v30.8h }, [x21], #0x10\n" + "ld1 { v14.8h }, [x26], #0x10\n" + "ld1 { v18.8h }, [x25], #0x10\n" + "ld1 { v22.8h }, [x24], #0x10\n" + "ld1 { v26.8h }, [x23], #0x10\n" + "ld1 { v30.8h }, [x22], #0x10\n" "tbz x8, #2, 250f\n" "ldr d11, [x16], #0x8\n" - "ldr d15, [x25], #0x8\n" - "ldr d19, [x24], #0x8\n" - "ldr d23, [x23], #0x8\n" - "ldr d27, [x22], #0x8\n" - "ldr d31, [x21], #0x8\n" + "ldr d15, [x26], #0x8\n" + "ldr d19, [x25], #0x8\n" + "ldr d23, [x24], #0x8\n" + "ldr d27, [x23], #0x8\n" + "ldr d31, [x22], #0x8\n" "tbz x8, #1, 249f\n" "ld1 { v11.s }[2], [x16], #0x4\n" "mov x20, #0x3c\n" - "ld1 { v15.s }[2], [x25], #0x4\n" - "ld1 { v19.s }[2], [x24], #0x4\n" - "ld1 { v23.s }[2], [x23], #0x4\n" - "ld1 { v27.s }[2], [x22], #0x4\n" - "ld1 { v31.s }[2], [x21], #0x4\n" + "ld1 { v15.s }[2], [x26], #0x4\n" + "ld1 { v19.s }[2], [x25], #0x4\n" + "ld1 { v23.s }[2], [x24], #0x4\n" + "ld1 { v27.s }[2], [x23], #0x4\n" + "ld1 { v31.s }[2], [x22], #0x4\n" "tbz x8, #0, 264f\n" "ld1 { v11.h }[6], [x16]\n" - "ld1 { v15.h }[6], [x25]\n" - "ld1 { v19.h }[6], [x24]\n" - "ld1 { v23.h }[6], [x23]\n" - "ld1 { v27.h }[6], [x22]\n" - "ld1 { v31.h }[6], [x21]\n" + "ld1 { v15.h }[6], [x26]\n" + "ld1 { v19.h }[6], [x25]\n" + "ld1 { v23.h }[6], [x24]\n" + "ld1 { v27.h }[6], [x23]\n" + "ld1 { v31.h }[6], [x22]\n" "b 264f\n" "249:" // Height 6: Partial accumulate: partial_1_28 "mov x20, #0x38\n" "tbz x8, #0, 264f\n" "ld1 { v11.h }[4], [x16]\n" - "ld1 { v15.h }[4], [x25]\n" - "ld1 { v19.h }[4], [x24]\n" - "ld1 { v23.h }[4], [x23]\n" - "ld1 { v27.h }[4], [x22]\n" - "ld1 { v31.h }[4], [x21]\n" + "ld1 { v15.h }[4], [x26]\n" + "ld1 { v19.h }[4], [x25]\n" + "ld1 { v23.h }[4], [x24]\n" + "ld1 { v27.h }[4], [x23]\n" + "ld1 { v31.h }[4], [x22]\n" "b 264f\n" "250:" // Height 6: Partial accumulate: partial_2_24 "tbz x8, #1, 251f\n" "ldr s11, [x16], #0x4\n" "mov x20, #0x34\n" - "ldr s15, [x25], #0x4\n" - "ldr s19, [x24], #0x4\n" - "ldr s23, [x23], #0x4\n" - "ldr s27, [x22], #0x4\n" - "ldr s31, [x21], #0x4\n" + "ldr s15, [x26], #0x4\n" + "ldr s19, [x25], #0x4\n" + "ldr s23, [x24], #0x4\n" + "ldr s27, [x23], #0x4\n" + "ldr s31, [x22], #0x4\n" "tbz x8, #0, 264f\n" "ld1 { v11.h }[2], [x16]\n" - "ld1 { v15.h }[2], [x25]\n" - "ld1 { v19.h }[2], [x24]\n" - "ld1 { v23.h }[2], [x23]\n" - "ld1 { v27.h }[2], [x22]\n" - "ld1 { v31.h }[2], [x21]\n" + "ld1 { v15.h }[2], [x26]\n" + "ld1 { v19.h }[2], [x25]\n" + "ld1 { v23.h }[2], [x24]\n" + "ld1 { v27.h }[2], [x23]\n" + "ld1 { v31.h }[2], [x22]\n" "b 264f\n" "251:" // Height 6: Partial accumulate: partial_1_24 "mov x20, #0x30\n" "tbz x8, #0, 264f\n" "ldr h11, [x16, #0x0]\n" - "ldr h15, [x25, #0x0]\n" - "ldr h19, [x24, #0x0]\n" - "ldr h23, [x23, #0x0]\n" - "ldr h27, [x22, #0x0]\n" - "ldr h31, [x21, #0x0]\n" + "ldr h15, [x26, #0x0]\n" + "ldr h19, [x25, #0x0]\n" + "ldr h23, [x24, #0x0]\n" + "ldr h27, [x23, #0x0]\n" + "ldr h31, [x22, #0x0]\n" "b 264f\n" "252:" // Height 6: Partial accumulate: partial_4_16 "tbz x8, #2, 254f\n" "ldr d10, [x16], #0x8\n" - "ldr d14, [x25], #0x8\n" - "ldr d18, [x24], #0x8\n" - "ldr d22, [x23], #0x8\n" - "ldr d26, [x22], #0x8\n" - "ldr d30, [x21], #0x8\n" + "ldr d14, [x26], #0x8\n" + "ldr d18, [x25], #0x8\n" + "ldr d22, [x24], #0x8\n" + "ldr d26, [x23], #0x8\n" + "ldr d30, [x22], #0x8\n" "tbz x8, #1, 253f\n" "ld1 { v10.s }[2], [x16], #0x4\n" "mov x20, #0x2c\n" - "ld1 { v14.s }[2], [x25], #0x4\n" - "ld1 { v18.s }[2], [x24], #0x4\n" - "ld1 { v22.s }[2], [x23], #0x4\n" - "ld1 { v26.s }[2], [x22], #0x4\n" - "ld1 { v30.s }[2], [x21], #0x4\n" + "ld1 { v14.s }[2], [x26], #0x4\n" + "ld1 { v18.s }[2], [x25], #0x4\n" + "ld1 { v22.s }[2], [x24], #0x4\n" + "ld1 { v26.s }[2], [x23], #0x4\n" + "ld1 { v30.s }[2], [x22], #0x4\n" "tbz x8, #0, 264f\n" "ld1 { v10.h }[6], [x16]\n" - "ld1 { v14.h }[6], [x25]\n" - "ld1 { v18.h }[6], [x24]\n" - "ld1 { v22.h }[6], [x23]\n" - "ld1 { v26.h }[6], [x22]\n" - "ld1 { v30.h }[6], [x21]\n" + "ld1 { v14.h }[6], [x26]\n" + "ld1 { v18.h }[6], [x25]\n" + "ld1 { v22.h }[6], [x24]\n" + "ld1 { v26.h }[6], [x23]\n" + "ld1 { v30.h }[6], [x22]\n" "b 264f\n" "253:" // Height 6: Partial accumulate: partial_1_20 "mov x20, #0x28\n" "tbz x8, #0, 264f\n" "ld1 { v10.h }[4], [x16]\n" - "ld1 { v14.h }[4], [x25]\n" - "ld1 { v18.h }[4], [x24]\n" - "ld1 { v22.h }[4], [x23]\n" - "ld1 { v26.h }[4], [x22]\n" - "ld1 { v30.h }[4], [x21]\n" + "ld1 { v14.h }[4], [x26]\n" + "ld1 { v18.h }[4], [x25]\n" + "ld1 { v22.h }[4], [x24]\n" + "ld1 { v26.h }[4], [x23]\n" + "ld1 { v30.h }[4], [x22]\n" "b 264f\n" "254:" // Height 6: Partial accumulate: partial_2_16 "tbz x8, #1, 255f\n" "ldr s10, [x16], #0x4\n" "mov x20, #0x24\n" - "ldr s14, [x25], #0x4\n" - "ldr s18, [x24], #0x4\n" - "ldr s22, [x23], #0x4\n" - "ldr s26, [x22], #0x4\n" - "ldr s30, [x21], #0x4\n" + "ldr s14, [x26], #0x4\n" + "ldr s18, [x25], #0x4\n" + "ldr s22, [x24], #0x4\n" + "ldr s26, [x23], #0x4\n" + "ldr s30, [x22], #0x4\n" "tbz x8, #0, 264f\n" "ld1 { v10.h }[2], [x16]\n" - "ld1 { v14.h }[2], [x25]\n" - "ld1 { v18.h }[2], [x24]\n" - "ld1 { v22.h }[2], [x23]\n" - "ld1 { v26.h }[2], [x22]\n" - "ld1 { v30.h }[2], [x21]\n" + "ld1 { v14.h }[2], [x26]\n" + "ld1 { v18.h }[2], [x25]\n" + "ld1 { v22.h }[2], [x24]\n" + "ld1 { v26.h }[2], [x23]\n" + "ld1 { v30.h }[2], [x22]\n" "b 264f\n" "255:" // Height 6: Partial accumulate: partial_1_16 "mov x20, #0x20\n" "tbz x8, #0, 264f\n" "ldr h10, [x16, #0x0]\n" - "ldr h14, [x25, #0x0]\n" - "ldr h18, [x24, #0x0]\n" - "ldr h22, [x23, #0x0]\n" - "ldr h26, [x22, #0x0]\n" - "ldr h30, [x21, #0x0]\n" + "ldr h14, [x26, #0x0]\n" + "ldr h18, [x25, #0x0]\n" + "ldr h22, [x24, #0x0]\n" + "ldr h26, [x23, #0x0]\n" + "ldr h30, [x22, #0x0]\n" "b 264f\n" "256:" // Height 6: Partial accumulate: partial_8_0 "tbz x8, #3, 260f\n" "ld1 { v8.8h }, [x16], #0x10\n" - "ld1 { v12.8h }, [x25], #0x10\n" - "ld1 { v16.8h }, [x24], #0x10\n" - "ld1 { v20.8h }, [x23], #0x10\n" - "ld1 { v24.8h }, [x22], #0x10\n" - "ld1 { v28.8h }, [x21], #0x10\n" + "ld1 { v12.8h }, [x26], #0x10\n" + "ld1 { v16.8h }, [x25], #0x10\n" + "ld1 { v20.8h }, [x24], #0x10\n" + "ld1 { v24.8h }, [x23], #0x10\n" + "ld1 { v28.8h }, [x22], #0x10\n" "tbz x8, #2, 258f\n" "ldr d9, [x16], #0x8\n" - "ldr d13, [x25], #0x8\n" - "ldr d17, [x24], #0x8\n" - "ldr d21, [x23], #0x8\n" - "ldr d25, [x22], #0x8\n" - "ldr d29, [x21], #0x8\n" + "ldr d13, [x26], #0x8\n" + "ldr d17, [x25], #0x8\n" + "ldr d21, [x24], #0x8\n" + "ldr d25, [x23], #0x8\n" + "ldr d29, [x22], #0x8\n" "tbz x8, #1, 257f\n" "ld1 { v9.s }[2], [x16], #0x4\n" "mov x20, #0x1c\n" - "ld1 { v13.s }[2], [x25], #0x4\n" - "ld1 { v17.s }[2], [x24], #0x4\n" - "ld1 { v21.s }[2], [x23], #0x4\n" - "ld1 { v25.s }[2], [x22], #0x4\n" - "ld1 { v29.s }[2], [x21], #0x4\n" + "ld1 { v13.s }[2], [x26], #0x4\n" + "ld1 { v17.s }[2], [x25], #0x4\n" + "ld1 { v21.s }[2], [x24], #0x4\n" + "ld1 { v25.s }[2], [x23], #0x4\n" + "ld1 { v29.s }[2], [x22], #0x4\n" "tbz x8, #0, 264f\n" "ld1 { v9.h }[6], [x16]\n" - "ld1 { v13.h }[6], [x25]\n" - "ld1 { v17.h }[6], [x24]\n" - "ld1 { v21.h }[6], [x23]\n" - "ld1 { v25.h }[6], [x22]\n" - "ld1 { v29.h }[6], [x21]\n" + "ld1 { v13.h }[6], [x26]\n" + "ld1 { v17.h }[6], [x25]\n" + "ld1 { v21.h }[6], [x24]\n" + "ld1 { v25.h }[6], [x23]\n" + "ld1 { v29.h }[6], [x22]\n" "b 264f\n" "257:" // Height 6: Partial accumulate: partial_1_12 "mov x20, #0x18\n" "tbz x8, #0, 264f\n" "ld1 { v9.h }[4], [x16]\n" - "ld1 { v13.h }[4], [x25]\n" - "ld1 { v17.h }[4], [x24]\n" - "ld1 { v21.h }[4], [x23]\n" - "ld1 { v25.h }[4], [x22]\n" - "ld1 { v29.h }[4], [x21]\n" + "ld1 { v13.h }[4], [x26]\n" + "ld1 { v17.h }[4], [x25]\n" + "ld1 { v21.h }[4], [x24]\n" + "ld1 { v25.h }[4], [x23]\n" + "ld1 { v29.h }[4], [x22]\n" "b 264f\n" "258:" // Height 6: Partial accumulate: partial_2_8 "tbz x8, #1, 259f\n" "ldr s9, [x16], #0x4\n" "mov x20, #0x14\n" - "ldr s13, [x25], #0x4\n" - "ldr s17, [x24], #0x4\n" - "ldr s21, [x23], #0x4\n" - "ldr s25, [x22], #0x4\n" - "ldr s29, [x21], #0x4\n" + "ldr s13, [x26], #0x4\n" + "ldr s17, [x25], #0x4\n" + "ldr s21, [x24], #0x4\n" + "ldr s25, [x23], #0x4\n" + "ldr s29, [x22], #0x4\n" "tbz x8, #0, 264f\n" "ld1 { v9.h }[2], [x16]\n" - "ld1 { v13.h }[2], [x25]\n" - "ld1 { v17.h }[2], [x24]\n" - "ld1 { v21.h }[2], [x23]\n" - "ld1 { v25.h }[2], [x22]\n" - "ld1 { v29.h }[2], [x21]\n" + "ld1 { v13.h }[2], [x26]\n" + "ld1 { v17.h }[2], [x25]\n" + "ld1 { v21.h }[2], [x24]\n" + "ld1 { v25.h }[2], [x23]\n" + "ld1 { v29.h }[2], [x22]\n" "b 264f\n" "259:" // Height 6: Partial accumulate: partial_1_8 "mov x20, #0x10\n" "tbz x8, #0, 264f\n" "ldr h9, [x16, #0x0]\n" - "ldr h13, [x25, #0x0]\n" - "ldr h17, [x24, #0x0]\n" - "ldr h21, [x23, #0x0]\n" - "ldr h25, [x22, #0x0]\n" - "ldr h29, [x21, #0x0]\n" + "ldr h13, [x26, #0x0]\n" + "ldr h17, [x25, #0x0]\n" + "ldr h21, [x24, #0x0]\n" + "ldr h25, [x23, #0x0]\n" + "ldr h29, [x22, #0x0]\n" "b 264f\n" "260:" // Height 6: Partial accumulate: partial_4_0 "tbz x8, #2, 262f\n" "ldr d8, [x16], #0x8\n" - "ldr d12, [x25], #0x8\n" - "ldr d16, [x24], #0x8\n" - "ldr d20, [x23], #0x8\n" - "ldr d24, [x22], #0x8\n" - "ldr d28, [x21], #0x8\n" + "ldr d12, [x26], #0x8\n" + "ldr d16, [x25], #0x8\n" + "ldr d20, [x24], #0x8\n" + "ldr d24, [x23], #0x8\n" + "ldr d28, [x22], #0x8\n" "tbz x8, #1, 261f\n" "ld1 { v8.s }[2], [x16], #0x4\n" "mov x20, #0xc\n" - "ld1 { v12.s }[2], [x25], #0x4\n" - "ld1 { v16.s }[2], [x24], #0x4\n" - "ld1 { v20.s }[2], [x23], #0x4\n" - "ld1 { v24.s }[2], [x22], #0x4\n" - "ld1 { v28.s }[2], [x21], #0x4\n" + "ld1 { v12.s }[2], [x26], #0x4\n" + "ld1 { v16.s }[2], [x25], #0x4\n" + "ld1 { v20.s }[2], [x24], #0x4\n" + "ld1 { v24.s }[2], [x23], #0x4\n" + "ld1 { v28.s }[2], [x22], #0x4\n" "tbz x8, #0, 264f\n" "ld1 { v8.h }[6], [x16]\n" - "ld1 { v12.h }[6], [x25]\n" - "ld1 { v16.h }[6], [x24]\n" - "ld1 { v20.h }[6], [x23]\n" - "ld1 { v24.h }[6], [x22]\n" - "ld1 { v28.h }[6], [x21]\n" + "ld1 { v12.h }[6], [x26]\n" + "ld1 { v16.h }[6], [x25]\n" + "ld1 { v20.h }[6], [x24]\n" + "ld1 { v24.h }[6], [x23]\n" + "ld1 { v28.h }[6], [x22]\n" "b 264f\n" "261:" // Height 6: Partial accumulate: partial_1_4 "mov x20, #0x8\n" "tbz x8, #0, 264f\n" "ld1 { v8.h }[4], [x16]\n" - "ld1 { v12.h }[4], [x25]\n" - "ld1 { v16.h }[4], [x24]\n" - "ld1 { v20.h }[4], [x23]\n" - "ld1 { v24.h }[4], [x22]\n" - "ld1 { v28.h }[4], [x21]\n" + "ld1 { v12.h }[4], [x26]\n" + "ld1 { v16.h }[4], [x25]\n" + "ld1 { v20.h }[4], [x24]\n" + "ld1 { v24.h }[4], [x23]\n" + "ld1 { v28.h }[4], [x22]\n" "b 264f\n" "262:" // Height 6: Partial accumulate: partial_2_0 "tbz x8, #1, 263f\n" "ldr s8, [x16], #0x4\n" "mov x20, #0x4\n" - "ldr s12, [x25], #0x4\n" - "ldr s16, [x24], #0x4\n" - "ldr s20, [x23], #0x4\n" - "ldr s24, [x22], #0x4\n" - "ldr s28, [x21], #0x4\n" + "ldr s12, [x26], #0x4\n" + "ldr s16, [x25], #0x4\n" + "ldr s20, [x24], #0x4\n" + "ldr s24, [x23], #0x4\n" + "ldr s28, [x22], #0x4\n" "tbz x8, #0, 264f\n" "ld1 { v8.h }[2], [x16]\n" - "ld1 { v12.h }[2], [x25]\n" - "ld1 { v16.h }[2], [x24]\n" - "ld1 { v20.h }[2], [x23]\n" - "ld1 { v24.h }[2], [x22]\n" - "ld1 { v28.h }[2], [x21]\n" + "ld1 { v12.h }[2], [x26]\n" + "ld1 { v16.h }[2], [x25]\n" + "ld1 { v20.h }[2], [x24]\n" + "ld1 { v24.h }[2], [x23]\n" + "ld1 { v28.h }[2], [x22]\n" "b 264f\n" "263:" // Height 6: Partial accumulate: partial_1_0 "ldr h8, [x16, #0x0]\n" "mov x20, #0x0\n" - "ldr h12, [x25, #0x0]\n" - "ldr h16, [x24, #0x0]\n" - "ldr h20, [x23, #0x0]\n" - "ldr h24, [x22, #0x0]\n" - "ldr h28, [x21, #0x0]\n" + "ldr h12, [x26, #0x0]\n" + "ldr h16, [x25, #0x0]\n" + "ldr h20, [x24, #0x0]\n" + "ldr h24, [x23, #0x0]\n" + "ldr h28, [x22, #0x0]\n" "264:" // Height 6: Partial accumulate: Done "sub x16, x16, x20\n" "b 267f\n" @@ -4685,26 +4688,26 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "ldr q9, [x16, #0x10]\n" "ldr q10, [x16, #0x20]\n" "ldr q11, [x16, #0x30]\n" - "ldr q12, [x25, #0x0]\n" - "ldr q13, [x25, #0x10]\n" - "ldr q14, [x25, #0x20]\n" - "ldr q15, [x25, #0x30]\n" - "ldr q16, [x24, #0x0]\n" - "ldr q17, [x24, #0x10]\n" - "ldr q18, [x24, #0x20]\n" - "ldr q19, [x24, #0x30]\n" - "ldr q20, [x23, #0x0]\n" - "ldr q21, [x23, #0x10]\n" - "ldr q22, [x23, #0x20]\n" - "ldr q23, [x23, #0x30]\n" - "ldr q24, [x22, #0x0]\n" - "ldr q25, [x22, #0x10]\n" - "ldr q26, [x22, #0x20]\n" - "ldr q27, [x22, #0x30]\n" - "ldr q28, [x21, #0x0]\n" - "ldr q29, [x21, #0x10]\n" - "ldr q30, [x21, #0x20]\n" - "ldr q31, [x21, #0x30]\n" + "ldr q12, [x26, #0x0]\n" + "ldr q13, [x26, #0x10]\n" + "ldr q14, [x26, #0x20]\n" + "ldr q15, [x26, #0x30]\n" + "ldr q16, [x25, #0x0]\n" + "ldr q17, [x25, #0x10]\n" + "ldr q18, [x25, #0x20]\n" + "ldr q19, [x25, #0x30]\n" + "ldr q20, [x24, #0x0]\n" + "ldr q21, [x24, #0x10]\n" + "ldr q22, [x24, #0x20]\n" + "ldr q23, [x24, #0x30]\n" + "ldr q24, [x23, #0x0]\n" + "ldr q25, [x23, #0x10]\n" + "ldr q26, [x23, #0x20]\n" + "ldr q27, [x23, #0x30]\n" + "ldr q28, [x22, #0x0]\n" + "ldr q29, [x22, #0x10]\n" + "ldr q30, [x22, #0x20]\n" + "ldr q31, [x22, #0x30]\n" "b 267f\n" "266:" // Height 6: no accumulate "movi v8.16b, #0x0\n" @@ -4735,8 +4738,8 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "mov x15, #0x0\n" "268:" // 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, 269f\n" "ldr x20, [%x[input_ptr], x15, LSL #0x3]\n" "add x20, x20, x21, LSL #3\n" @@ -4789,290 +4792,290 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "fmla v28.8h, v6.8h, v5.h[0]\n" "ldr d6, [x17, #0x20]\n" "fmla v9.8h, v7.8h, v0.h[0]\n" - "mov v6.d[1], x21\n" + "add x10, x10, #0x10\n" "fmla v13.8h, v7.8h, v1.h[0]\n" - "ldr x21, [x17, #0x48]\n" + "add x9, x9, #0x10\n" "fmla v17.8h, v7.8h, v2.h[0]\n" - "add x10, x10, #0x10\n" + "mov v6.d[1], x21\n" "fmla v21.8h, v7.8h, v3.h[0]\n" - "add x9, x9, #0x10\n" + "ldr x21, [x17, #0x48]\n" "fmla v25.8h, v7.8h, v4.h[0]\n" "add x28, x28, #0x10\n" "fmla v29.8h, v7.8h, v5.h[0]\n" "ldr d7, [x17, #0x30]\n" - "mov v7.d[1], x20\n" "fmla v10.8h, v6.8h, v0.h[0]\n" + "ldr x27, [x13, #0x8]\n" "fmla v14.8h, v6.8h, v1.h[0]\n" - "ldr x20, [x17, #0x58]\n" + "ldr x26, [x12, #0x8]\n" + "mov v7.d[1], x20\n" "fmla v18.8h, v6.8h, v2.h[0]\n" - "ldr x27, [x13, #0x8]\n" "fmla v22.8h, v6.8h, v3.h[0]\n" - "ldr x26, [x12, #0x8]\n" + "ldr x20, [x17, #0x58]\n" "fmla v26.8h, v6.8h, v4.h[0]\n" "ldr x25, [x11, #0x8]\n" "fmla v30.8h, v6.8h, v5.h[0]\n" "ldr d6, [x17, #0x40]\n" "fmla v11.8h, v7.8h, v0.h[0]\n" - "mov v6.d[1], x21\n" + "ldr x24, [x10, #0x8]\n" "fmla v15.8h, v7.8h, v1.h[0]\n" - "ldr x21, [x17, #0x68]\n" + "ldr x23, [x9, #0x8]\n" "fmla v19.8h, v7.8h, v2.h[0]\n" - "ldr x24, [x10, #0x8]\n" + "mov v6.d[1], x21\n" "fmla v23.8h, v7.8h, v3.h[0]\n" - "ldr x23, [x9, #0x8]\n" + "ldr x21, [x17, #0x68]\n" "fmla v27.8h, v7.8h, v4.h[0]\n" "ldr x22, [x28, #0x8]\n" "fmla v31.8h, v7.8h, v5.h[0]\n" "ldr d7, [x17, #0x50]\n" - "mov v7.d[1], x20\n" "fmla v8.8h, v6.8h, v0.h[1]\n" + "sub x14, x14, #0x8\n" "fmla v12.8h, v6.8h, v1.h[1]\n" - "ldr x20, [x17, #0x78]\n" + "cmp x14, #0x10\n" + "mov v7.d[1], x20\n" "fmla v16.8h, v6.8h, v2.h[1]\n" - "sub x14, x14, #0x8\n" "fmla v20.8h, v6.8h, v3.h[1]\n" - "cmp x14, #0x10\n" + "ldr x20, [x17, #0x78]\n" "fmla v24.8h, v6.8h, v4.h[1]\n" "prfm pldl1keep, [x13, #0x80]\n" "fmla v28.8h, v6.8h, v5.h[1]\n" "ldr d6, [x17, #0x60]\n" "fmla v9.8h, v7.8h, v0.h[1]\n" - "mov v6.d[1], x21\n" + "prfm pldl1keep, [x12, #0x80]\n" "fmla v13.8h, v7.8h, v1.h[1]\n" - "ldr x21, [x17, #0x88]\n" + "prfm pldl1keep, [x11, #0x80]\n" "fmla v17.8h, v7.8h, v2.h[1]\n" - "prfm pldl1keep, [x12, #0x80]\n" + "mov v6.d[1], x21\n" "fmla v21.8h, v7.8h, v3.h[1]\n" - "prfm pldl1keep, [x11, #0x80]\n" + "ldr x21, [x17, #0x88]\n" "fmla v25.8h, v7.8h, v4.h[1]\n" "prfm pldl1keep, [x10, #0x80]\n" "fmla v29.8h, v7.8h, v5.h[1]\n" "ldr d7, [x17, #0x70]\n" - "mov v7.d[1], x20\n" "fmla v10.8h, v6.8h, v0.h[1]\n" + "prfm pldl1keep, [x9, #0x80]\n" "fmla v14.8h, v6.8h, v1.h[1]\n" - "ldr x20, [x17, #0x98]\n" + "prfm pldl1keep, [x28, #0x80]\n" + "mov v7.d[1], x20\n" "fmla v18.8h, v6.8h, v2.h[1]\n" - "prfm pldl1keep, [x9, #0x80]\n" "fmla v22.8h, v6.8h, v3.h[1]\n" - "prfm pldl1keep, [x28, #0x80]\n" + "ldr x20, [x17, #0x98]\n" "fmla v26.8h, v6.8h, v4.h[1]\n" "fmla v30.8h, v6.8h, v5.h[1]\n" "ldr d6, [x17, #0x80]\n" "fmla v11.8h, v7.8h, v0.h[1]\n" - "mov v6.d[1], x21\n" "fmla v15.8h, v7.8h, v1.h[1]\n" - "ldr x21, [x17, #0xa8]\n" "fmla v19.8h, v7.8h, v2.h[1]\n" + "mov v6.d[1], x21\n" "fmla v23.8h, v7.8h, v3.h[1]\n" + "ldr x21, [x17, #0xa8]\n" "fmla v27.8h, v7.8h, v4.h[1]\n" "fmla v31.8h, v7.8h, v5.h[1]\n" "ldr d7, [x17, #0x90]\n" - "mov v7.d[1], x20\n" "fmla v8.8h, v6.8h, v0.h[2]\n" "fmla v12.8h, v6.8h, v1.h[2]\n" - "ldr x20, [x17, #0xb8]\n" + "mov v7.d[1], x20\n" "fmla v16.8h, v6.8h, v2.h[2]\n" "fmla v20.8h, v6.8h, v3.h[2]\n" + "ldr x20, [x17, #0xb8]\n" "fmla v24.8h, v6.8h, v4.h[2]\n" "fmla v28.8h, v6.8h, v5.h[2]\n" "ldr d6, [x17, #0xa0]\n" "fmla v9.8h, v7.8h, v0.h[2]\n" - "mov v6.d[1], x21\n" "fmla v13.8h, v7.8h, v1.h[2]\n" - "ldr x21, [x17, #0xc8]\n" "fmla v17.8h, v7.8h, v2.h[2]\n" + "mov v6.d[1], x21\n" "fmla v21.8h, v7.8h, v3.h[2]\n" + "ldr x21, [x17, #0xc8]\n" "fmla v25.8h, v7.8h, v4.h[2]\n" "fmla v29.8h, v7.8h, v5.h[2]\n" "ldr d7, [x17, #0xb0]\n" - "mov v7.d[1], x20\n" "fmla v10.8h, v6.8h, v0.h[2]\n" "fmla v14.8h, v6.8h, v1.h[2]\n" - "ldr x20, [x17, #0xd8]\n" + "mov v7.d[1], x20\n" "fmla v18.8h, v6.8h, v2.h[2]\n" "fmla v22.8h, v6.8h, v3.h[2]\n" + "ldr x20, [x17, #0xd8]\n" "fmla v26.8h, v6.8h, v4.h[2]\n" "fmla v30.8h, v6.8h, v5.h[2]\n" "ldr d6, [x17, #0xc0]\n" "fmla v11.8h, v7.8h, v0.h[2]\n" - "mov v6.d[1], x21\n" "fmla v15.8h, v7.8h, v1.h[2]\n" - "ldr x21, [x17, #0xe8]\n" "fmla v19.8h, v7.8h, v2.h[2]\n" + "mov v6.d[1], x21\n" "fmla v23.8h, v7.8h, v3.h[2]\n" + "ldr x21, [x17, #0xe8]\n" "fmla v27.8h, v7.8h, v4.h[2]\n" "fmla v31.8h, v7.8h, v5.h[2]\n" "ldr d7, [x17, #0xd0]\n" - "mov v7.d[1], x20\n" "fmla v8.8h, v6.8h, v0.h[3]\n" "fmla v12.8h, v6.8h, v1.h[3]\n" - "ldr x20, [x17, #0xf8]\n" + "mov v7.d[1], x20\n" "fmla v16.8h, v6.8h, v2.h[3]\n" "fmla v20.8h, v6.8h, v3.h[3]\n" + "ldr x20, [x17, #0xf8]\n" "fmla v24.8h, v6.8h, v4.h[3]\n" "fmla v28.8h, v6.8h, v5.h[3]\n" "ldr d6, [x17, #0xe0]\n" "fmla v9.8h, v7.8h, v0.h[3]\n" - "mov v6.d[1], x21\n" "fmla v13.8h, v7.8h, v1.h[3]\n" - "ldr x21, [x17, #0x108]\n" "fmla v17.8h, v7.8h, v2.h[3]\n" + "mov v6.d[1], x21\n" "fmla v21.8h, v7.8h, v3.h[3]\n" + "ldr x21, [x17, #0x108]\n" "fmla v25.8h, v7.8h, v4.h[3]\n" "fmla v29.8h, v7.8h, v5.h[3]\n" "ldr d7, [x17, #0xf0]\n" - "mov v7.d[1], x20\n" "fmla v10.8h, v6.8h, v0.h[3]\n" "fmla v14.8h, v6.8h, v1.h[3]\n" - "ldr x20, [x17, #0x118]\n" + "mov v7.d[1], x20\n" "fmla v18.8h, v6.8h, v2.h[3]\n" "fmla v22.8h, v6.8h, v3.h[3]\n" + "ldr x20, [x17, #0x118]\n" "fmla v26.8h, v6.8h, v4.h[3]\n" "fmla v30.8h, v6.8h, v5.h[3]\n" "ldr d6, [x17, #0x100]\n" "fmla v11.8h, v7.8h, v0.h[3]\n" - "mov v6.d[1], x21\n" "fmla v15.8h, v7.8h, v1.h[3]\n" - "ldr x21, [x17, #0x128]\n" "fmla v19.8h, v7.8h, v2.h[3]\n" + "mov v6.d[1], x21\n" "fmla v23.8h, v7.8h, v3.h[3]\n" + "ldr x21, [x17, #0x128]\n" "fmla v27.8h, v7.8h, v4.h[3]\n" "fmla v31.8h, v7.8h, v5.h[3]\n" "ldr d7, [x17, #0x110]\n" - "mov v7.d[1], x20\n" "fmla v8.8h, v6.8h, v0.h[4]\n" "fmla v12.8h, v6.8h, v1.h[4]\n" - "ldr x20, [x17, #0x138]\n" + "mov v7.d[1], x20\n" "fmla v16.8h, v6.8h, v2.h[4]\n" "fmla v20.8h, v6.8h, v3.h[4]\n" + "ldr x20, [x17, #0x138]\n" "fmla v24.8h, v6.8h, v4.h[4]\n" "fmla v28.8h, v6.8h, v5.h[4]\n" "ldr d6, [x17, #0x120]\n" "fmla v9.8h, v7.8h, v0.h[4]\n" - "mov v6.d[1], x21\n" "fmla v13.8h, v7.8h, v1.h[4]\n" - "ldr x21, [x17, #0x148]\n" "fmla v17.8h, v7.8h, v2.h[4]\n" + "mov v6.d[1], x21\n" "fmla v21.8h, v7.8h, v3.h[4]\n" + "ldr x21, [x17, #0x148]\n" "fmla v25.8h, v7.8h, v4.h[4]\n" "fmla v29.8h, v7.8h, v5.h[4]\n" "ldr d7, [x17, #0x130]\n" - "mov v7.d[1], x20\n" "fmla v10.8h, v6.8h, v0.h[4]\n" "fmla v14.8h, v6.8h, v1.h[4]\n" - "ldr x20, [x17, #0x158]\n" + "mov v7.d[1], x20\n" "fmla v18.8h, v6.8h, v2.h[4]\n" "fmla v22.8h, v6.8h, v3.h[4]\n" + "ldr x20, [x17, #0x158]\n" "fmla v26.8h, v6.8h, v4.h[4]\n" "fmla v30.8h, v6.8h, v5.h[4]\n" "ldr d6, [x17, #0x140]\n" "fmla v11.8h, v7.8h, v0.h[4]\n" - "mov v6.d[1], x21\n" "fmla v15.8h, v7.8h, v1.h[4]\n" - "ldr x21, [x17, #0x168]\n" "fmla v19.8h, v7.8h, v2.h[4]\n" + "mov v6.d[1], x21\n" "fmla v23.8h, v7.8h, v3.h[4]\n" + "ldr x21, [x17, #0x168]\n" "fmla v27.8h, v7.8h, v4.h[4]\n" "fmla v31.8h, v7.8h, v5.h[4]\n" "ldr d7, [x17, #0x150]\n" - "mov v7.d[1], x20\n" "fmla v8.8h, v6.8h, v0.h[5]\n" "fmla v12.8h, v6.8h, v1.h[5]\n" - "ldr x20, [x17, #0x178]\n" + "mov v7.d[1], x20\n" "fmla v16.8h, v6.8h, v2.h[5]\n" "fmla v20.8h, v6.8h, v3.h[5]\n" + "ldr x20, [x17, #0x178]\n" "fmla v24.8h, v6.8h, v4.h[5]\n" "fmla v28.8h, v6.8h, v5.h[5]\n" "ldr d6, [x17, #0x160]\n" "fmla v9.8h, v7.8h, v0.h[5]\n" - "mov v6.d[1], x21\n" "fmla v13.8h, v7.8h, v1.h[5]\n" - "ldr x21, [x17, #0x188]\n" "fmla v17.8h, v7.8h, v2.h[5]\n" + "mov v6.d[1], x21\n" "fmla v21.8h, v7.8h, v3.h[5]\n" + "ldr x21, [x17, #0x188]\n" "fmla v25.8h, v7.8h, v4.h[5]\n" "fmla v29.8h, v7.8h, v5.h[5]\n" "ldr d7, [x17, #0x170]\n" - "mov v7.d[1], x20\n" "fmla v10.8h, v6.8h, v0.h[5]\n" "fmla v14.8h, v6.8h, v1.h[5]\n" - "ldr x20, [x17, #0x198]\n" + "mov v7.d[1], x20\n" "fmla v18.8h, v6.8h, v2.h[5]\n" "fmla v22.8h, v6.8h, v3.h[5]\n" + "ldr x20, [x17, #0x198]\n" "fmla v26.8h, v6.8h, v4.h[5]\n" "fmla v30.8h, v6.8h, v5.h[5]\n" "ldr d6, [x17, #0x180]\n" "fmla v11.8h, v7.8h, v0.h[5]\n" - "mov v6.d[1], x21\n" "fmla v15.8h, v7.8h, v1.h[5]\n" - "ldr x21, [x17, #0x1a8]\n" "fmla v19.8h, v7.8h, v2.h[5]\n" + "mov v6.d[1], x21\n" "fmla v23.8h, v7.8h, v3.h[5]\n" + "ldr x21, [x17, #0x1a8]\n" "fmla v27.8h, v7.8h, v4.h[5]\n" "fmla v31.8h, v7.8h, v5.h[5]\n" "ldr d7, [x17, #0x190]\n" - "mov v7.d[1], x20\n" "fmla v8.8h, v6.8h, v0.h[6]\n" "fmla v12.8h, v6.8h, v1.h[6]\n" - "ldr x20, [x17, #0x1b8]\n" + "mov v7.d[1], x20\n" "fmla v16.8h, v6.8h, v2.h[6]\n" "fmla v20.8h, v6.8h, v3.h[6]\n" + "ldr x20, [x17, #0x1b8]\n" "fmla v24.8h, v6.8h, v4.h[6]\n" "fmla v28.8h, v6.8h, v5.h[6]\n" "ldr d6, [x17, #0x1a0]\n" "fmla v9.8h, v7.8h, v0.h[6]\n" - "mov v6.d[1], x21\n" "fmla v13.8h, v7.8h, v1.h[6]\n" - "ldr x21, [x17, #0x1c8]\n" "fmla v17.8h, v7.8h, v2.h[6]\n" + "mov v6.d[1], x21\n" "fmla v21.8h, v7.8h, v3.h[6]\n" + "ldr x21, [x17, #0x1c8]\n" "fmla v25.8h, v7.8h, v4.h[6]\n" "fmla v29.8h, v7.8h, v5.h[6]\n" "ldr d7, [x17, #0x1b0]\n" - "mov v7.d[1], x20\n" "fmla v10.8h, v6.8h, v0.h[6]\n" "fmla v14.8h, v6.8h, v1.h[6]\n" - "ldr x20, [x17, #0x1d8]\n" + "mov v7.d[1], x20\n" "fmla v18.8h, v6.8h, v2.h[6]\n" "fmla v22.8h, v6.8h, v3.h[6]\n" + "ldr x20, [x17, #0x1d8]\n" "fmla v26.8h, v6.8h, v4.h[6]\n" "fmla v30.8h, v6.8h, v5.h[6]\n" "ldr d6, [x17, #0x1c0]\n" "fmla v11.8h, v7.8h, v0.h[6]\n" - "mov v6.d[1], x21\n" "fmla v15.8h, v7.8h, v1.h[6]\n" - "ldr x21, [x17, #0x1e8]\n" "fmla v19.8h, v7.8h, v2.h[6]\n" + "mov v6.d[1], x21\n" "fmla v23.8h, v7.8h, v3.h[6]\n" + "ldr x21, [x17, #0x1e8]\n" "fmla v27.8h, v7.8h, v4.h[6]\n" "fmla v31.8h, v7.8h, v5.h[6]\n" "ldr d7, [x17, #0x1d0]\n" - "mov v7.d[1], x20\n" "fmla v8.8h, v6.8h, v0.h[7]\n" "fmla v12.8h, v6.8h, v1.h[7]\n" - "ldr x20, [x17, #0x1f8]\n" + "mov v7.d[1], x20\n" "fmla v16.8h, v6.8h, v2.h[7]\n" "fmla v20.8h, v6.8h, v3.h[7]\n" + "ldr x20, [x17, #0x1f8]\n" "fmla v24.8h, v6.8h, v4.h[7]\n" "fmla v28.8h, v6.8h, v5.h[7]\n" "ldr d6, [x17, #0x1e0]\n" "fmla v9.8h, v7.8h, v0.h[7]\n" - "mov v6.d[1], x21\n" "fmla v13.8h, v7.8h, v1.h[7]\n" "fmla v17.8h, v7.8h, v2.h[7]\n" + "mov v6.d[1], x21\n" "fmla v21.8h, v7.8h, v3.h[7]\n" "fmla v25.8h, v7.8h, v4.h[7]\n" "fmla v29.8h, v7.8h, v5.h[7]\n" "ldr d7, [x17, #0x1f0]\n" - "mov v7.d[1], x20\n" "add x17, x17, #0x200\n" "fmla v10.8h, v6.8h, v0.h[7]\n" - "ldr x21, [x17, #0x8]\n" "fmla v14.8h, v6.8h, v1.h[7]\n" - "ldr x20, [x17, #0x18]\n" + "ldr x21, [x17, #0x8]\n" + "mov v7.d[1], x20\n" "fmla v18.8h, v6.8h, v2.h[7]\n" "fmla v22.8h, v6.8h, v3.h[7]\n" + "ldr x20, [x17, #0x18]\n" "fmla v26.8h, v6.8h, v4.h[7]\n" "fmla v30.8h, v6.8h, v5.h[7]\n" "ldr d6, [x17, #0x0]\n" @@ -5346,8 +5349,8 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "ldr h3, [x9], #0x2\n" "ldr h2, [x28], #0x2\n" "ldr q1, [x17, #0x0]\n" - "fmla v8.8h, v1.8h, v7.h[0]\n" "ldr q0, [x17, #0x10]\n" + "fmla v8.8h, v1.8h, v7.h[0]\n" "fmla v12.8h, v1.8h, v6.h[0]\n" "fmla v16.8h, v1.8h, v5.h[0]\n" "fmla v20.8h, v1.8h, v4.h[0]\n" @@ -5381,46 +5384,46 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "cmp x15, x20\n" "bne 268b\n" "ldr x20, [%x[args_ptr], %[offsetof_output_offset]]\n" - "add x25, x16, x20, LSL #1\n" - "add x24, x25, x20, LSL #1\n" - "add x23, x24, x20, LSL #1\n" - "add x22, x23, x20, LSL #1\n" - "add x21, x22, x20, LSL #1\n" "prfm pstl1keep, [x16, #0x0]\n" + "add x26, x16, x20, LSL #1\n" + "add x25, x26, x20, LSL #1\n" + "prfm pstl1keep, [x26, #0x0]\n" + "add x24, x25, x20, LSL #1\n" "prfm pstl1keep, [x25, #0x0]\n" + "add x23, x24, x20, LSL #1\n" "prfm pstl1keep, [x24, #0x0]\n" + "add x22, x23, x20, LSL #1\n" "prfm pstl1keep, [x23, #0x0]\n" "prfm pstl1keep, [x22, #0x0]\n" - "prfm pstl1keep, [x21, #0x0]\n" "tbz %x[flags], #1, 276f\n" - "add x20, %x[args_ptr], %[offset_max]\n" - "ld1r { v0.8h }, [x20]\n" - "fmin v8.8h, v8.8h, v0.8h\n" - "fmin v9.8h, v9.8h, v0.8h\n" - "fmin v10.8h, v10.8h, v0.8h\n" - "fmin v11.8h, v11.8h, v0.8h\n" - "fmin v12.8h, v12.8h, v0.8h\n" - "fmin v13.8h, v13.8h, v0.8h\n" - "fmin v14.8h, v14.8h, v0.8h\n" - "fmin v15.8h, v15.8h, v0.8h\n" - "fmin v16.8h, v16.8h, v0.8h\n" - "fmin v17.8h, v17.8h, v0.8h\n" - "fmin v18.8h, v18.8h, v0.8h\n" - "fmin v19.8h, v19.8h, v0.8h\n" - "fmin v20.8h, v20.8h, v0.8h\n" - "fmin v21.8h, v21.8h, v0.8h\n" - "fmin v22.8h, v22.8h, v0.8h\n" - "fmin v23.8h, v23.8h, v0.8h\n" - "fmin v24.8h, v24.8h, v0.8h\n" - "fmin v25.8h, v25.8h, v0.8h\n" - "fmin v26.8h, v26.8h, v0.8h\n" - "fmin v27.8h, v27.8h, v0.8h\n" - "fmin v28.8h, v28.8h, v0.8h\n" - "fmin v29.8h, v29.8h, v0.8h\n" - "fmin v30.8h, v30.8h, v0.8h\n" - "fmin v31.8h, v31.8h, v0.8h\n" + "add x21, %x[args_ptr], %[offset_max]\n" "add x20, %x[args_ptr], %[offset_min]\n" + "ld1r { v1.8h }, [x21]\n" "ld1r { v0.8h }, [x20]\n" + "fmin v8.8h, v8.8h, v1.8h\n" + "fmin v9.8h, v9.8h, v1.8h\n" + "fmin v10.8h, v10.8h, v1.8h\n" + "fmin v11.8h, v11.8h, v1.8h\n" + "fmin v12.8h, v12.8h, v1.8h\n" + "fmin v13.8h, v13.8h, v1.8h\n" + "fmin v14.8h, v14.8h, v1.8h\n" + "fmin v15.8h, v15.8h, v1.8h\n" + "fmin v16.8h, v16.8h, v1.8h\n" + "fmin v17.8h, v17.8h, v1.8h\n" + "fmin v18.8h, v18.8h, v1.8h\n" + "fmin v19.8h, v19.8h, v1.8h\n" + "fmin v20.8h, v20.8h, v1.8h\n" + "fmin v21.8h, v21.8h, v1.8h\n" + "fmin v22.8h, v22.8h, v1.8h\n" + "fmin v23.8h, v23.8h, v1.8h\n" + "fmin v24.8h, v24.8h, v1.8h\n" + "fmin v25.8h, v25.8h, v1.8h\n" + "fmin v26.8h, v26.8h, v1.8h\n" + "fmin v27.8h, v27.8h, v1.8h\n" + "fmin v28.8h, v28.8h, v1.8h\n" + "fmin v29.8h, v29.8h, v1.8h\n" + "fmin v30.8h, v30.8h, v1.8h\n" + "fmin v31.8h, v31.8h, v1.8h\n" "fmax v8.8h, v8.8h, v0.8h\n" "fmax v9.8h, v9.8h, v0.8h\n" "fmax v10.8h, v10.8h, v0.8h\n" @@ -5451,255 +5454,255 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "tbz x8, #4, 284f\n" "st1 { v8.8h }, [x16], #0x10\n" "st1 { v9.8h }, [x16], #0x10\n" - "st1 { v12.8h }, [x25], #0x10\n" - "st1 { v13.8h }, [x25], #0x10\n" - "st1 { v16.8h }, [x24], #0x10\n" - "st1 { v17.8h }, [x24], #0x10\n" - "st1 { v20.8h }, [x23], #0x10\n" - "st1 { v21.8h }, [x23], #0x10\n" - "st1 { v24.8h }, [x22], #0x10\n" - "st1 { v25.8h }, [x22], #0x10\n" - "st1 { v28.8h }, [x21], #0x10\n" - "st1 { v29.8h }, [x21], #0x10\n" + "st1 { v12.8h }, [x26], #0x10\n" + "st1 { v13.8h }, [x26], #0x10\n" + "st1 { v16.8h }, [x25], #0x10\n" + "st1 { v17.8h }, [x25], #0x10\n" + "st1 { v20.8h }, [x24], #0x10\n" + "st1 { v21.8h }, [x24], #0x10\n" + "st1 { v24.8h }, [x23], #0x10\n" + "st1 { v25.8h }, [x23], #0x10\n" + "st1 { v28.8h }, [x22], #0x10\n" + "st1 { v29.8h }, [x22], #0x10\n" "tbz x8, #3, 280f\n" "st1 { v10.8h }, [x16], #0x10\n" - "st1 { v14.8h }, [x25], #0x10\n" - "st1 { v18.8h }, [x24], #0x10\n" - "st1 { v22.8h }, [x23], #0x10\n" - "st1 { v26.8h }, [x22], #0x10\n" - "st1 { v30.8h }, [x21], #0x10\n" + "st1 { v14.8h }, [x26], #0x10\n" + "st1 { v18.8h }, [x25], #0x10\n" + "st1 { v22.8h }, [x24], #0x10\n" + "st1 { v26.8h }, [x23], #0x10\n" + "st1 { v30.8h }, [x22], #0x10\n" "tbz x8, #2, 278f\n" "str d11, [x16], #0x8\n" - "str d15, [x25], #0x8\n" - "str d19, [x24], #0x8\n" - "str d23, [x23], #0x8\n" - "str d27, [x22], #0x8\n" - "str d31, [x21], #0x8\n" + "str d15, [x26], #0x8\n" + "str d19, [x25], #0x8\n" + "str d23, [x24], #0x8\n" + "str d27, [x23], #0x8\n" + "str d31, [x22], #0x8\n" "tbz x8, #1, 277f\n" "st1 { v11.s }[2], [x16], #0x4\n" - "st1 { v15.s }[2], [x25], #0x4\n" - "st1 { v19.s }[2], [x24], #0x4\n" - "st1 { v23.s }[2], [x23], #0x4\n" - "st1 { v27.s }[2], [x22], #0x4\n" - "st1 { v31.s }[2], [x21], #0x4\n" + "st1 { v15.s }[2], [x26], #0x4\n" + "st1 { v19.s }[2], [x25], #0x4\n" + "st1 { v23.s }[2], [x24], #0x4\n" + "st1 { v27.s }[2], [x23], #0x4\n" + "st1 { v31.s }[2], [x22], #0x4\n" "tbz x8, #0, 292f\n" "st1 { v11.h }[6], [x16]\n" - "st1 { v15.h }[6], [x25]\n" - "st1 { v19.h }[6], [x24]\n" - "st1 { v23.h }[6], [x23]\n" - "st1 { v27.h }[6], [x22]\n" - "st1 { v31.h }[6], [x21]\n" + "st1 { v15.h }[6], [x26]\n" + "st1 { v19.h }[6], [x25]\n" + "st1 { v23.h }[6], [x24]\n" + "st1 { v27.h }[6], [x23]\n" + "st1 { v31.h }[6], [x22]\n" "b 292f\n" "277:" // Height 6: Partial direct writeback: partial_1_28 "tbz x8, #0, 292f\n" "st1 { v11.h }[4], [x16]\n" - "st1 { v15.h }[4], [x25]\n" - "st1 { v19.h }[4], [x24]\n" - "st1 { v23.h }[4], [x23]\n" - "st1 { v27.h }[4], [x22]\n" - "st1 { v31.h }[4], [x21]\n" + "st1 { v15.h }[4], [x26]\n" + "st1 { v19.h }[4], [x25]\n" + "st1 { v23.h }[4], [x24]\n" + "st1 { v27.h }[4], [x23]\n" + "st1 { v31.h }[4], [x22]\n" "b 292f\n" "278:" // Height 6: Partial direct writeback: partial_2_24 "tbz x8, #1, 279f\n" "str s11, [x16], #0x4\n" - "str s15, [x25], #0x4\n" - "str s19, [x24], #0x4\n" - "str s23, [x23], #0x4\n" - "str s27, [x22], #0x4\n" - "str s31, [x21], #0x4\n" + "str s15, [x26], #0x4\n" + "str s19, [x25], #0x4\n" + "str s23, [x24], #0x4\n" + "str s27, [x23], #0x4\n" + "str s31, [x22], #0x4\n" "tbz x8, #0, 292f\n" "st1 { v11.h }[2], [x16]\n" - "st1 { v15.h }[2], [x25]\n" - "st1 { v19.h }[2], [x24]\n" - "st1 { v23.h }[2], [x23]\n" - "st1 { v27.h }[2], [x22]\n" - "st1 { v31.h }[2], [x21]\n" + "st1 { v15.h }[2], [x26]\n" + "st1 { v19.h }[2], [x25]\n" + "st1 { v23.h }[2], [x24]\n" + "st1 { v27.h }[2], [x23]\n" + "st1 { v31.h }[2], [x22]\n" "b 292f\n" "279:" // Height 6: Partial direct writeback: partial_1_24 "tbz x8, #0, 292f\n" "str h11, [x16, #0x0]\n" - "str h15, [x25, #0x0]\n" - "str h19, [x24, #0x0]\n" - "str h23, [x23, #0x0]\n" - "str h27, [x22, #0x0]\n" - "str h31, [x21, #0x0]\n" + "str h15, [x26, #0x0]\n" + "str h19, [x25, #0x0]\n" + "str h23, [x24, #0x0]\n" + "str h27, [x23, #0x0]\n" + "str h31, [x22, #0x0]\n" "b 292f\n" "280:" // Height 6: Partial direct writeback: partial_4_16 "tbz x8, #2, 282f\n" "str d10, [x16], #0x8\n" - "str d14, [x25], #0x8\n" - "str d18, [x24], #0x8\n" - "str d22, [x23], #0x8\n" - "str d26, [x22], #0x8\n" - "str d30, [x21], #0x8\n" + "str d14, [x26], #0x8\n" + "str d18, [x25], #0x8\n" + "str d22, [x24], #0x8\n" + "str d26, [x23], #0x8\n" + "str d30, [x22], #0x8\n" "tbz x8, #1, 281f\n" "st1 { v10.s }[2], [x16], #0x4\n" - "st1 { v14.s }[2], [x25], #0x4\n" - "st1 { v18.s }[2], [x24], #0x4\n" - "st1 { v22.s }[2], [x23], #0x4\n" - "st1 { v26.s }[2], [x22], #0x4\n" - "st1 { v30.s }[2], [x21], #0x4\n" + "st1 { v14.s }[2], [x26], #0x4\n" + "st1 { v18.s }[2], [x25], #0x4\n" + "st1 { v22.s }[2], [x24], #0x4\n" + "st1 { v26.s }[2], [x23], #0x4\n" + "st1 { v30.s }[2], [x22], #0x4\n" "tbz x8, #0, 292f\n" "st1 { v10.h }[6], [x16]\n" - "st1 { v14.h }[6], [x25]\n" - "st1 { v18.h }[6], [x24]\n" - "st1 { v22.h }[6], [x23]\n" - "st1 { v26.h }[6], [x22]\n" - "st1 { v30.h }[6], [x21]\n" + "st1 { v14.h }[6], [x26]\n" + "st1 { v18.h }[6], [x25]\n" + "st1 { v22.h }[6], [x24]\n" + "st1 { v26.h }[6], [x23]\n" + "st1 { v30.h }[6], [x22]\n" "b 292f\n" "281:" // Height 6: Partial direct writeback: partial_1_20 "tbz x8, #0, 292f\n" "st1 { v10.h }[4], [x16]\n" - "st1 { v14.h }[4], [x25]\n" - "st1 { v18.h }[4], [x24]\n" - "st1 { v22.h }[4], [x23]\n" - "st1 { v26.h }[4], [x22]\n" - "st1 { v30.h }[4], [x21]\n" + "st1 { v14.h }[4], [x26]\n" + "st1 { v18.h }[4], [x25]\n" + "st1 { v22.h }[4], [x24]\n" + "st1 { v26.h }[4], [x23]\n" + "st1 { v30.h }[4], [x22]\n" "b 292f\n" "282:" // Height 6: Partial direct writeback: partial_2_16 "tbz x8, #1, 283f\n" "str s10, [x16], #0x4\n" - "str s14, [x25], #0x4\n" - "str s18, [x24], #0x4\n" - "str s22, [x23], #0x4\n" - "str s26, [x22], #0x4\n" - "str s30, [x21], #0x4\n" + "str s14, [x26], #0x4\n" + "str s18, [x25], #0x4\n" + "str s22, [x24], #0x4\n" + "str s26, [x23], #0x4\n" + "str s30, [x22], #0x4\n" "tbz x8, #0, 292f\n" "st1 { v10.h }[2], [x16]\n" - "st1 { v14.h }[2], [x25]\n" - "st1 { v18.h }[2], [x24]\n" - "st1 { v22.h }[2], [x23]\n" - "st1 { v26.h }[2], [x22]\n" - "st1 { v30.h }[2], [x21]\n" + "st1 { v14.h }[2], [x26]\n" + "st1 { v18.h }[2], [x25]\n" + "st1 { v22.h }[2], [x24]\n" + "st1 { v26.h }[2], [x23]\n" + "st1 { v30.h }[2], [x22]\n" "b 292f\n" "283:" // Height 6: Partial direct writeback: partial_1_16 "tbz x8, #0, 292f\n" "str h10, [x16, #0x0]\n" - "str h14, [x25, #0x0]\n" - "str h18, [x24, #0x0]\n" - "str h22, [x23, #0x0]\n" - "str h26, [x22, #0x0]\n" - "str h30, [x21, #0x0]\n" + "str h14, [x26, #0x0]\n" + "str h18, [x25, #0x0]\n" + "str h22, [x24, #0x0]\n" + "str h26, [x23, #0x0]\n" + "str h30, [x22, #0x0]\n" "b 292f\n" "284:" // Height 6: Partial direct writeback: partial_8_0 "tbz x8, #3, 288f\n" "st1 { v8.8h }, [x16], #0x10\n" - "st1 { v12.8h }, [x25], #0x10\n" - "st1 { v16.8h }, [x24], #0x10\n" - "st1 { v20.8h }, [x23], #0x10\n" - "st1 { v24.8h }, [x22], #0x10\n" - "st1 { v28.8h }, [x21], #0x10\n" + "st1 { v12.8h }, [x26], #0x10\n" + "st1 { v16.8h }, [x25], #0x10\n" + "st1 { v20.8h }, [x24], #0x10\n" + "st1 { v24.8h }, [x23], #0x10\n" + "st1 { v28.8h }, [x22], #0x10\n" "tbz x8, #2, 286f\n" "str d9, [x16], #0x8\n" - "str d13, [x25], #0x8\n" - "str d17, [x24], #0x8\n" - "str d21, [x23], #0x8\n" - "str d25, [x22], #0x8\n" - "str d29, [x21], #0x8\n" + "str d13, [x26], #0x8\n" + "str d17, [x25], #0x8\n" + "str d21, [x24], #0x8\n" + "str d25, [x23], #0x8\n" + "str d29, [x22], #0x8\n" "tbz x8, #1, 285f\n" "st1 { v9.s }[2], [x16], #0x4\n" - "st1 { v13.s }[2], [x25], #0x4\n" - "st1 { v17.s }[2], [x24], #0x4\n" - "st1 { v21.s }[2], [x23], #0x4\n" - "st1 { v25.s }[2], [x22], #0x4\n" - "st1 { v29.s }[2], [x21], #0x4\n" + "st1 { v13.s }[2], [x26], #0x4\n" + "st1 { v17.s }[2], [x25], #0x4\n" + "st1 { v21.s }[2], [x24], #0x4\n" + "st1 { v25.s }[2], [x23], #0x4\n" + "st1 { v29.s }[2], [x22], #0x4\n" "tbz x8, #0, 292f\n" "st1 { v9.h }[6], [x16]\n" - "st1 { v13.h }[6], [x25]\n" - "st1 { v17.h }[6], [x24]\n" - "st1 { v21.h }[6], [x23]\n" - "st1 { v25.h }[6], [x22]\n" - "st1 { v29.h }[6], [x21]\n" + "st1 { v13.h }[6], [x26]\n" + "st1 { v17.h }[6], [x25]\n" + "st1 { v21.h }[6], [x24]\n" + "st1 { v25.h }[6], [x23]\n" + "st1 { v29.h }[6], [x22]\n" "b 292f\n" "285:" // Height 6: Partial direct writeback: partial_1_12 "tbz x8, #0, 292f\n" "st1 { v9.h }[4], [x16]\n" - "st1 { v13.h }[4], [x25]\n" - "st1 { v17.h }[4], [x24]\n" - "st1 { v21.h }[4], [x23]\n" - "st1 { v25.h }[4], [x22]\n" - "st1 { v29.h }[4], [x21]\n" + "st1 { v13.h }[4], [x26]\n" + "st1 { v17.h }[4], [x25]\n" + "st1 { v21.h }[4], [x24]\n" + "st1 { v25.h }[4], [x23]\n" + "st1 { v29.h }[4], [x22]\n" "b 292f\n" "286:" // Height 6: Partial direct writeback: partial_2_8 "tbz x8, #1, 287f\n" "str s9, [x16], #0x4\n" - "str s13, [x25], #0x4\n" - "str s17, [x24], #0x4\n" - "str s21, [x23], #0x4\n" - "str s25, [x22], #0x4\n" - "str s29, [x21], #0x4\n" + "str s13, [x26], #0x4\n" + "str s17, [x25], #0x4\n" + "str s21, [x24], #0x4\n" + "str s25, [x23], #0x4\n" + "str s29, [x22], #0x4\n" "tbz x8, #0, 292f\n" "st1 { v9.h }[2], [x16]\n" - "st1 { v13.h }[2], [x25]\n" - "st1 { v17.h }[2], [x24]\n" - "st1 { v21.h }[2], [x23]\n" - "st1 { v25.h }[2], [x22]\n" - "st1 { v29.h }[2], [x21]\n" + "st1 { v13.h }[2], [x26]\n" + "st1 { v17.h }[2], [x25]\n" + "st1 { v21.h }[2], [x24]\n" + "st1 { v25.h }[2], [x23]\n" + "st1 { v29.h }[2], [x22]\n" "b 292f\n" "287:" // Height 6: Partial direct writeback: partial_1_8 "tbz x8, #0, 292f\n" "str h9, [x16, #0x0]\n" - "str h13, [x25, #0x0]\n" - "str h17, [x24, #0x0]\n" - "str h21, [x23, #0x0]\n" - "str h25, [x22, #0x0]\n" - "str h29, [x21, #0x0]\n" + "str h13, [x26, #0x0]\n" + "str h17, [x25, #0x0]\n" + "str h21, [x24, #0x0]\n" + "str h25, [x23, #0x0]\n" + "str h29, [x22, #0x0]\n" "b 292f\n" "288:" // Height 6: Partial direct writeback: partial_4_0 "tbz x8, #2, 290f\n" "str d8, [x16], #0x8\n" - "str d12, [x25], #0x8\n" - "str d16, [x24], #0x8\n" - "str d20, [x23], #0x8\n" - "str d24, [x22], #0x8\n" - "str d28, [x21], #0x8\n" + "str d12, [x26], #0x8\n" + "str d16, [x25], #0x8\n" + "str d20, [x24], #0x8\n" + "str d24, [x23], #0x8\n" + "str d28, [x22], #0x8\n" "tbz x8, #1, 289f\n" "st1 { v8.s }[2], [x16], #0x4\n" - "st1 { v12.s }[2], [x25], #0x4\n" - "st1 { v16.s }[2], [x24], #0x4\n" - "st1 { v20.s }[2], [x23], #0x4\n" - "st1 { v24.s }[2], [x22], #0x4\n" - "st1 { v28.s }[2], [x21], #0x4\n" + "st1 { v12.s }[2], [x26], #0x4\n" + "st1 { v16.s }[2], [x25], #0x4\n" + "st1 { v20.s }[2], [x24], #0x4\n" + "st1 { v24.s }[2], [x23], #0x4\n" + "st1 { v28.s }[2], [x22], #0x4\n" "tbz x8, #0, 292f\n" "st1 { v8.h }[6], [x16]\n" - "st1 { v12.h }[6], [x25]\n" - "st1 { v16.h }[6], [x24]\n" - "st1 { v20.h }[6], [x23]\n" - "st1 { v24.h }[6], [x22]\n" - "st1 { v28.h }[6], [x21]\n" + "st1 { v12.h }[6], [x26]\n" + "st1 { v16.h }[6], [x25]\n" + "st1 { v20.h }[6], [x24]\n" + "st1 { v24.h }[6], [x23]\n" + "st1 { v28.h }[6], [x22]\n" "b 292f\n" "289:" // Height 6: Partial direct writeback: partial_1_4 "tbz x8, #0, 292f\n" "st1 { v8.h }[4], [x16]\n" - "st1 { v12.h }[4], [x25]\n" - "st1 { v16.h }[4], [x24]\n" - "st1 { v20.h }[4], [x23]\n" - "st1 { v24.h }[4], [x22]\n" - "st1 { v28.h }[4], [x21]\n" + "st1 { v12.h }[4], [x26]\n" + "st1 { v16.h }[4], [x25]\n" + "st1 { v20.h }[4], [x24]\n" + "st1 { v24.h }[4], [x23]\n" + "st1 { v28.h }[4], [x22]\n" "b 292f\n" "290:" // Height 6: Partial direct writeback: partial_2_0 "tbz x8, #1, 291f\n" "str s8, [x16], #0x4\n" - "str s12, [x25], #0x4\n" - "str s16, [x24], #0x4\n" - "str s20, [x23], #0x4\n" - "str s24, [x22], #0x4\n" - "str s28, [x21], #0x4\n" + "str s12, [x26], #0x4\n" + "str s16, [x25], #0x4\n" + "str s20, [x24], #0x4\n" + "str s24, [x23], #0x4\n" + "str s28, [x22], #0x4\n" "tbz x8, #0, 292f\n" "st1 { v8.h }[2], [x16]\n" - "st1 { v12.h }[2], [x25]\n" - "st1 { v16.h }[2], [x24]\n" - "st1 { v20.h }[2], [x23]\n" - "st1 { v24.h }[2], [x22]\n" - "st1 { v28.h }[2], [x21]\n" + "st1 { v12.h }[2], [x26]\n" + "st1 { v16.h }[2], [x25]\n" + "st1 { v20.h }[2], [x24]\n" + "st1 { v24.h }[2], [x23]\n" + "st1 { v28.h }[2], [x22]\n" "b 292f\n" "291:" // Height 6: Partial direct writeback: partial_1_0 "str h8, [x16, #0x0]\n" - "str h12, [x25, #0x0]\n" - "str h16, [x24, #0x0]\n" - "str h20, [x23, #0x0]\n" - "str h24, [x22, #0x0]\n" - "str h28, [x21, #0x0]\n" + "str h12, [x26, #0x0]\n" + "str h16, [x25, #0x0]\n" + "str h20, [x24, #0x0]\n" + "str h24, [x23, #0x0]\n" + "str h28, [x22, #0x0]\n" "292:" // Height 6: Partial direct writeback: Done "b 294f\n" "293:" // Height 6: Full writeback @@ -5708,26 +5711,26 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "str q10, [x16, #0x20]\n" "str q11, [x16, #0x30]\n" "add x16, x16, #0x40\n" - "str q12, [x25, #0x0]\n" - "str q13, [x25, #0x10]\n" - "str q14, [x25, #0x20]\n" - "str q15, [x25, #0x30]\n" - "str q16, [x24, #0x0]\n" - "str q17, [x24, #0x10]\n" - "str q18, [x24, #0x20]\n" - "str q19, [x24, #0x30]\n" - "str q20, [x23, #0x0]\n" - "str q21, [x23, #0x10]\n" - "str q22, [x23, #0x20]\n" - "str q23, [x23, #0x30]\n" - "str q24, [x22, #0x0]\n" - "str q25, [x22, #0x10]\n" - "str q26, [x22, #0x20]\n" - "str q27, [x22, #0x30]\n" - "str q28, [x21, #0x0]\n" - "str q29, [x21, #0x10]\n" - "str q30, [x21, #0x20]\n" - "str q31, [x21, #0x30]\n" + "str q12, [x26, #0x0]\n" + "str q13, [x26, #0x10]\n" + "str q14, [x26, #0x20]\n" + "str q15, [x26, #0x30]\n" + "str q16, [x25, #0x0]\n" + "str q17, [x25, #0x10]\n" + "str q18, [x25, #0x20]\n" + "str q19, [x25, #0x30]\n" + "str q20, [x24, #0x0]\n" + "str q21, [x24, #0x10]\n" + "str q22, [x24, #0x20]\n" + "str q23, [x24, #0x30]\n" + "str q24, [x23, #0x0]\n" + "str q25, [x23, #0x10]\n" + "str q26, [x23, #0x20]\n" + "str q27, [x23, #0x30]\n" + "str q28, [x22, #0x0]\n" + "str q29, [x22, #0x10]\n" + "str q30, [x22, #0x20]\n" + "str q31, [x22, #0x30]\n" "294:" // Height 6: Writeback done "subs x8, x8, #0x20\n" "bgt 247b\n" @@ -5743,8 +5746,8 @@ void a64_hybrid_fp16_mla_6x32_a55 ( "madd %x[input_ptr], x20, x21, %x[input_ptr]\n" "b 1b\n" "296:" // Exit - : [M] "+&r" (M), [input_ptr] "+&r" (input_ptr), [output_ptr] "+&r" (output_ptr) - : [args_ptr] "r" (&ka), [bias] "r" (bias), [flags] "r" (flags), [offset_max] "I" (offsetof(KernelArgs, maxval)), [offset_min] "I" (offsetof(KernelArgs, minval)), [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), [offset_max] "I" (offsetof(KernelArgs, maxval)), [offset_min] "I" (offsetof(KernelArgs, minval)), [offsetof_B_ptr] "I" (offsetof(KernelArgs, B_ptr)), [offsetof_N] "I" (offsetof(KernelArgs, N)), [offsetof_bias] "I" (offsetof(KernelArgs, bias)), [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", "x7", "x8", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x16", "x17", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28" ); } |