diff options
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_fp32_mla_8x4/a55.cpp')
| -rw-r--r-- | src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_fp32_mla_8x4/a55.cpp | 538 |
1 files changed, 269 insertions, 269 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_fp32_mla_8x4/a55.cpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_fp32_mla_8x4/a55.cpp index 99920002b2..e2dae4b414 100644 --- a/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_fp32_mla_8x4/a55.cpp +++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_fp32_mla_8x4/a55.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2019-2021 Arm Limited. + * Copyright (c) 2019-2021, 2023 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -10,16 +10,16 @@ * sell copies of the Software, and to permit persons to whom the Software is * furnished to do so, subject to the following conditions: * - * The above copyright notice and this permission notice shall be included in - * all copies or substantial portions of the Software. + * The above copyright notice and this permission notice shall be included in all + * copies or substantial portions of the Software. * * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER - * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING - * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS - * IN THE SOFTWARE. + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. */ #ifdef __aarch64__ @@ -120,15 +120,15 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "bge 6f\n" "tbz x17, #1, 4f\n" "ldr d24, [x14], #0x8\n" - "mov x19, #0x8\n" + "mov x8, #0x8\n" "tbz x17, #0, 5f\n" "ld1 { v24.s }[2], [x14]\n" "b 5f\n" "4:" // Height 1: Partial accumulate: partial_1_0 "ldr s24, [x14, #0x0]\n" - "mov x19, #0x0\n" + "mov x8, #0x0\n" "5:" // Height 1: Partial accumulate: Done - "sub x14, x14, x19\n" + "sub x14, x14, x8\n" "b 8f\n" "6:" // Height 1: full accumulate "ldr q24, [x14, #0x0]\n" @@ -139,15 +139,15 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "mov x13, #0x0\n" "9:" // Height 1: String loop "ldr x20, [%x[args_ptr], %[offsetof_string_lengths]]\n" - "ldr x19, [%x[args_ptr], %[offsetof_input_offset]]\n" + "ldr x8, [%x[args_ptr], %[offsetof_input_offset]]\n" "ldr w12, [x20, x13, LSL #0x2]\n" "tbz %x[flags], #3, 10f\n" "ldr x20, [%x[input_ptr], x13, LSL #0x3]\n" - "add x20, x20, x19, LSL #3\n" + "add x20, x20, x8, LSL #3\n" "ldr x11, [x20, #0x0]\n" "cbnz x13, 11f\n" - "ldr x19, [%x[args_ptr], %[offsetof_input_initial_col]]\n" - "add x11, x11, x19, LSL #2\n" + "ldr x8, [%x[args_ptr], %[offsetof_input_initial_col]]\n" + "add x11, x11, x8, LSL #2\n" "b 11f\n" "10:" // Height 1: setup direct input "mov x11, %x[input_ptr]\n" @@ -161,20 +161,20 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "12:" // Height 1: Multiply loop: Main loop head "fmla v24.4s, v8.4s, v0.s[0]\n" "ldr d9, [x16, #0x10]\n" - "ldr x19, [x16, #0x18]\n" + "ldr x8, [x16, #0x18]\n" "add x11, x11, #0x10\n" "ldr d10, [x16, #0x20]\n" "sub x12, x12, #0x4\n" "ldr x21, [x16, #0x28]\n" "cmp x12, #0x8\n" - "mov v9.d[1], x19\n" + "mov v9.d[1], x8\n" "ldr d11, [x16, #0x30]\n" - "ldr x19, [x16, #0x38]\n" + "ldr x8, [x16, #0x38]\n" "add x16, x16, #0x40\n" "fmla v24.4s, v9.4s, v0.s[1]\n" "mov v10.d[1], x21\n" "prfm pldl1keep, [x11, #0x80]\n" - "mov v11.d[1], x19\n" + "mov v11.d[1], x8\n" "ldr d8, [x16, #0x0]\n" "ldr x26, [x16, #0x8]\n" "fmla v24.4s, v10.4s, v0.s[2]\n" @@ -208,14 +208,14 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "16:" // Height 1: Multiply loop: No odd multiplies "ldr w19, [%x[args_ptr], %[offsetof_num_strings]]\n" "add x13, x13, #0x1\n" - "cmp x13, x19\n" + "cmp x13, x8\n" "bne 9b\n" "prfm pstl1keep, [x14, #0x0]\n" "tbz %x[flags], #1, 17f\n" - "add x19, %x[args_ptr], %[offset_min]\n" - "ld1r { v17.4s }, [x19]\n" - "add x19, %x[args_ptr], %[offset_max]\n" - "ld1r { v16.4s }, [x19]\n" + "add x8, %x[args_ptr], %[offset_min]\n" + "ld1r { v17.4s }, [x8]\n" + "add x8, %x[args_ptr], %[offset_max]\n" + "ld1r { v16.4s }, [x8]\n" "fmin v24.4s, v24.4s, v16.4s\n" "fmax v24.4s, v24.4s, v17.4s\n" "17:" // Height 1: No activation @@ -250,24 +250,24 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "b 29f\n" "24:" // Height 2: no bias "tbz %x[flags], #0, 28f\n" - "ldr x19, [%x[args_ptr], %[offsetof_output_offset]]\n" + "ldr x8, [%x[args_ptr], %[offsetof_output_offset]]\n" "cmp x17, #0x4\n" - "add x27, x14, x19, LSL #2\n" + "add x27, x14, x8, LSL #2\n" "bge 27f\n" "tbz x17, #1, 25f\n" "ldr d24, [x14], #0x8\n" "ldr d25, [x27], #0x8\n" - "mov x19, #0x8\n" + "mov x8, #0x8\n" "tbz x17, #0, 26f\n" "ld1 { v24.s }[2], [x14]\n" "ld1 { v25.s }[2], [x27]\n" "b 26f\n" "25:" // Height 2: Partial accumulate: partial_1_0 "ldr s24, [x14, #0x0]\n" - "mov x19, #0x0\n" + "mov x8, #0x0\n" "ldr s25, [x27, #0x0]\n" "26:" // Height 2: Partial accumulate: Done - "sub x14, x14, x19\n" + "sub x14, x14, x8\n" "b 29f\n" "27:" // Height 2: full accumulate "ldr q24, [x14, #0x0]\n" @@ -280,21 +280,21 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "mov x13, #0x0\n" "30:" // Height 2: String loop "ldr x20, [%x[args_ptr], %[offsetof_string_lengths]]\n" - "ldr x19, [%x[args_ptr], %[offsetof_input_offset]]\n" + "ldr x8, [%x[args_ptr], %[offsetof_input_offset]]\n" "ldr w12, [x20, x13, LSL #0x2]\n" "tbz %x[flags], #3, 31f\n" "ldr x20, [%x[input_ptr], x13, LSL #0x3]\n" - "add x20, x20, x19, LSL #3\n" + "add x20, x20, x8, LSL #3\n" "ldr x11, [x20, #0x0]\n" "ldr x9, [x20, #0x8]\n" "cbnz x13, 32f\n" - "ldr x19, [%x[args_ptr], %[offsetof_input_initial_col]]\n" - "add x11, x11, x19, LSL #2\n" - "add x9, x9, x19, LSL #2\n" + "ldr x8, [%x[args_ptr], %[offsetof_input_initial_col]]\n" + "add x11, x11, x8, LSL #2\n" + "add x9, x9, x8, LSL #2\n" "b 32f\n" "31:" // Height 2: setup direct input "mov x11, %x[input_ptr]\n" - "add x9, x11, x19, LSL #2\n" + "add x9, x11, x8, LSL #2\n" "32:" // Height 2: input setup done "cmp x12, #0x4\n" "blt 35f\n" @@ -307,20 +307,20 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "fmla v24.4s, v8.4s, v0.s[0]\n" "ldr d9, [x16, #0x10]\n" "fmla v25.4s, v8.4s, v1.s[0]\n" - "ldr x19, [x16, #0x18]\n" + "ldr x8, [x16, #0x18]\n" "ldr d10, [x16, #0x20]\n" "add x11, x11, #0x10\n" "ldr x21, [x16, #0x28]\n" "add x9, x9, #0x10\n" - "mov v9.d[1], x19\n" + "mov v9.d[1], x8\n" "ldr d11, [x16, #0x30]\n" - "ldr x19, [x16, #0x38]\n" + "ldr x8, [x16, #0x38]\n" "sub x12, x12, #0x4\n" "fmla v24.4s, v9.4s, v0.s[1]\n" "mov v10.d[1], x21\n" "fmla v25.4s, v9.4s, v1.s[1]\n" "prfm pldl1keep, [x11, #0x80]\n" - "mov v11.d[1], x19\n" + "mov v11.d[1], x8\n" "prfm pldl1keep, [x9, #0x80]\n" "ldr x10, [x11, #0x8]\n" "cmp x12, #0x8\n" @@ -370,17 +370,17 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "37:" // Height 2: Multiply loop: No odd multiplies "ldr w19, [%x[args_ptr], %[offsetof_num_strings]]\n" "add x13, x13, #0x1\n" - "cmp x13, x19\n" + "cmp x13, x8\n" "bne 30b\n" - "ldr x19, [%x[args_ptr], %[offsetof_output_offset]]\n" + "ldr x8, [%x[args_ptr], %[offsetof_output_offset]]\n" "prfm pstl1keep, [x14, #0x0]\n" - "add x27, x14, x19, LSL #2\n" + "add x27, x14, x8, LSL #2\n" "prfm pstl1keep, [x27, #0x0]\n" "tbz %x[flags], #1, 38f\n" "add x20, %x[args_ptr], %[offset_min]\n" - "add x19, %x[args_ptr], %[offset_max]\n" + "add x8, %x[args_ptr], %[offset_max]\n" "ld1r { v17.4s }, [x20]\n" - "ld1r { v16.4s }, [x19]\n" + "ld1r { v16.4s }, [x8]\n" "fmin v24.4s, v24.4s, v16.4s\n" "fmin v25.4s, v25.4s, v16.4s\n" "fmax v24.4s, v24.4s, v17.4s\n" @@ -422,15 +422,15 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "b 50f\n" "45:" // Height 3: no bias "tbz %x[flags], #0, 49f\n" - "ldr x19, [%x[args_ptr], %[offsetof_output_offset]]\n" + "ldr x8, [%x[args_ptr], %[offsetof_output_offset]]\n" "cmp x17, #0x4\n" - "add x27, x14, x19, LSL #2\n" - "add x26, x27, x19, LSL #2\n" + "add x27, x14, x8, LSL #2\n" + "add x26, x27, x8, LSL #2\n" "bge 48f\n" "tbz x17, #1, 46f\n" "ldr d24, [x14], #0x8\n" "ldr d25, [x27], #0x8\n" - "mov x19, #0x8\n" + "mov x8, #0x8\n" "ldr d26, [x26], #0x8\n" "tbz x17, #0, 47f\n" "ld1 { v24.s }[2], [x14]\n" @@ -439,11 +439,11 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "b 47f\n" "46:" // Height 3: Partial accumulate: partial_1_0 "ldr s24, [x14, #0x0]\n" - "mov x19, #0x0\n" + "mov x8, #0x0\n" "ldr s25, [x27, #0x0]\n" "ldr s26, [x26, #0x0]\n" "47:" // Height 3: Partial accumulate: Done - "sub x14, x14, x19\n" + "sub x14, x14, x8\n" "b 50f\n" "48:" // Height 3: full accumulate "ldr q24, [x14, #0x0]\n" @@ -458,24 +458,24 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "mov x13, #0x0\n" "51:" // Height 3: String loop "ldr x20, [%x[args_ptr], %[offsetof_string_lengths]]\n" - "ldr x19, [%x[args_ptr], %[offsetof_input_offset]]\n" + "ldr x8, [%x[args_ptr], %[offsetof_input_offset]]\n" "ldr w12, [x20, x13, LSL #0x2]\n" "tbz %x[flags], #3, 52f\n" "ldr x20, [%x[input_ptr], x13, LSL #0x3]\n" - "add x20, x20, x19, LSL #3\n" + "add x20, x20, x8, LSL #3\n" "ldr x11, [x20, #0x0]\n" "ldr x9, [x20, #0x8]\n" "ldr x27, [x20, #0x10]\n" "cbnz x13, 53f\n" - "ldr x19, [%x[args_ptr], %[offsetof_input_initial_col]]\n" - "add x11, x11, x19, LSL #2\n" - "add x9, x9, x19, LSL #2\n" - "add x27, x27, x19, LSL #2\n" + "ldr x8, [%x[args_ptr], %[offsetof_input_initial_col]]\n" + "add x11, x11, x8, LSL #2\n" + "add x9, x9, x8, LSL #2\n" + "add x27, x27, x8, LSL #2\n" "b 53f\n" "52:" // Height 3: setup direct input "mov x11, %x[input_ptr]\n" - "add x9, x11, x19, LSL #2\n" - "add x27, x9, x19, LSL #2\n" + "add x9, x11, x8, LSL #2\n" + "add x27, x9, x8, LSL #2\n" "53:" // Height 3: input setup done "cmp x12, #0x4\n" "blt 56f\n" @@ -489,21 +489,21 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "fmla v24.4s, v8.4s, v0.s[0]\n" "ldr d9, [x16, #0x10]\n" "fmla v25.4s, v8.4s, v1.s[0]\n" - "ldr x19, [x16, #0x18]\n" + "ldr x8, [x16, #0x18]\n" "fmla v26.4s, v8.4s, v2.s[0]\n" "ldr d10, [x16, #0x20]\n" "ldr x21, [x16, #0x28]\n" "add x11, x11, #0x10\n" - "mov v9.d[1], x19\n" + "mov v9.d[1], x8\n" "ldr d11, [x16, #0x30]\n" - "ldr x19, [x16, #0x38]\n" + "ldr x8, [x16, #0x38]\n" "add x9, x9, #0x10\n" "fmla v24.4s, v9.4s, v0.s[1]\n" "mov v10.d[1], x21\n" "fmla v25.4s, v9.4s, v1.s[1]\n" "prfm pldl1keep, [x11, #0x80]\n" "fmla v26.4s, v9.4s, v2.s[1]\n" - "mov v11.d[1], x19\n" + "mov v11.d[1], x8\n" "prfm pldl1keep, [x9, #0x80]\n" "add x27, x27, #0x10\n" "fmla v24.4s, v10.4s, v0.s[2]\n" @@ -569,19 +569,19 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "58:" // Height 3: Multiply loop: No odd multiplies "ldr w19, [%x[args_ptr], %[offsetof_num_strings]]\n" "add x13, x13, #0x1\n" - "cmp x13, x19\n" + "cmp x13, x8\n" "bne 51b\n" - "ldr x19, [%x[args_ptr], %[offsetof_output_offset]]\n" + "ldr x8, [%x[args_ptr], %[offsetof_output_offset]]\n" "prfm pstl1keep, [x14, #0x0]\n" - "add x27, x14, x19, LSL #2\n" + "add x27, x14, x8, LSL #2\n" "prfm pstl1keep, [x27, #0x0]\n" - "add x26, x27, x19, LSL #2\n" + "add x26, x27, x8, LSL #2\n" "prfm pstl1keep, [x26, #0x0]\n" "tbz %x[flags], #1, 59f\n" "add x20, %x[args_ptr], %[offset_min]\n" - "add x19, %x[args_ptr], %[offset_max]\n" + "add x8, %x[args_ptr], %[offset_max]\n" "ld1r { v17.4s }, [x20]\n" - "ld1r { v16.4s }, [x19]\n" + "ld1r { v16.4s }, [x8]\n" "fmin v24.4s, v24.4s, v16.4s\n" "fmin v25.4s, v25.4s, v16.4s\n" "fmin v26.4s, v26.4s, v16.4s\n" @@ -630,16 +630,16 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "b 71f\n" "66:" // Height 4: no bias "tbz %x[flags], #0, 70f\n" - "ldr x19, [%x[args_ptr], %[offsetof_output_offset]]\n" + "ldr x8, [%x[args_ptr], %[offsetof_output_offset]]\n" "cmp x17, #0x4\n" - "add x27, x14, x19, LSL #2\n" - "add x26, x27, x19, LSL #2\n" - "add x25, x26, x19, LSL #2\n" + "add x27, x14, x8, LSL #2\n" + "add x26, x27, x8, LSL #2\n" + "add x25, x26, x8, LSL #2\n" "bge 69f\n" "tbz x17, #1, 67f\n" "ldr d24, [x14], #0x8\n" "ldr d25, [x27], #0x8\n" - "mov x19, #0x8\n" + "mov x8, #0x8\n" "ldr d26, [x26], #0x8\n" "ldr d27, [x25], #0x8\n" "tbz x17, #0, 68f\n" @@ -650,12 +650,12 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "b 68f\n" "67:" // Height 4: Partial accumulate: partial_1_0 "ldr s24, [x14, #0x0]\n" - "mov x19, #0x0\n" + "mov x8, #0x0\n" "ldr s25, [x27, #0x0]\n" "ldr s26, [x26, #0x0]\n" "ldr s27, [x25, #0x0]\n" "68:" // Height 4: Partial accumulate: Done - "sub x14, x14, x19\n" + "sub x14, x14, x8\n" "b 71f\n" "69:" // Height 4: full accumulate "ldr q24, [x14, #0x0]\n" @@ -672,27 +672,27 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "mov x13, #0x0\n" "72:" // Height 4: String loop "ldr x20, [%x[args_ptr], %[offsetof_string_lengths]]\n" - "ldr x19, [%x[args_ptr], %[offsetof_input_offset]]\n" + "ldr x8, [%x[args_ptr], %[offsetof_input_offset]]\n" "ldr w12, [x20, x13, LSL #0x2]\n" "tbz %x[flags], #3, 73f\n" "ldr x20, [%x[input_ptr], x13, LSL #0x3]\n" - "add x20, x20, x19, LSL #3\n" + "add x20, x20, x8, LSL #3\n" "ldr x11, [x20, #0x0]\n" "ldr x9, [x20, #0x8]\n" "ldr x27, [x20, #0x10]\n" "ldr x25, [x20, #0x18]\n" "cbnz x13, 74f\n" - "ldr x19, [%x[args_ptr], %[offsetof_input_initial_col]]\n" - "add x11, x11, x19, LSL #2\n" - "add x9, x9, x19, LSL #2\n" - "add x27, x27, x19, LSL #2\n" - "add x25, x25, x19, LSL #2\n" + "ldr x8, [%x[args_ptr], %[offsetof_input_initial_col]]\n" + "add x11, x11, x8, LSL #2\n" + "add x9, x9, x8, LSL #2\n" + "add x27, x27, x8, LSL #2\n" + "add x25, x25, x8, LSL #2\n" "b 74f\n" "73:" // Height 4: setup direct input "mov x11, %x[input_ptr]\n" - "add x9, x11, x19, LSL #2\n" - "add x27, x9, x19, LSL #2\n" - "add x25, x27, x19, LSL #2\n" + "add x9, x11, x8, LSL #2\n" + "add x27, x9, x8, LSL #2\n" + "add x25, x27, x8, LSL #2\n" "74:" // Height 4: input setup done "cmp x12, #0x4\n" "blt 77f\n" @@ -707,21 +707,21 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "fmla v24.4s, v8.4s, v0.s[0]\n" "ldr d9, [x16, #0x10]\n" "fmla v25.4s, v8.4s, v1.s[0]\n" - "ldr x19, [x16, #0x18]\n" + "ldr x8, [x16, #0x18]\n" "fmla v26.4s, v8.4s, v2.s[0]\n" "ldr d10, [x16, #0x20]\n" "fmla v27.4s, v8.4s, v3.s[0]\n" "ldr x21, [x16, #0x28]\n" - "mov v9.d[1], x19\n" + "mov v9.d[1], x8\n" "ldr d11, [x16, #0x30]\n" - "ldr x19, [x16, #0x38]\n" + "ldr x8, [x16, #0x38]\n" "add x11, x11, #0x10\n" "fmla v24.4s, v9.4s, v0.s[1]\n" "mov v10.d[1], x21\n" "fmla v25.4s, v9.4s, v1.s[1]\n" "prfm pldl1keep, [x11, #0x80]\n" "fmla v26.4s, v9.4s, v2.s[1]\n" - "mov v11.d[1], x19\n" + "mov v11.d[1], x8\n" "fmla v27.4s, v9.4s, v3.s[1]\n" "ldr x10, [x11, #0x8]\n" "fmla v24.4s, v10.4s, v0.s[2]\n" @@ -748,11 +748,11 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "sub x12, x12, #0x4\n" "mov v2.d[1], x26\n" "ldr d3, [x25, #0x0]\n" - "ldr x19, [x25, #0x8]\n" + "ldr x8, [x25, #0x8]\n" "cmp x12, #0x8\n" "add x16, x16, #0x40\n" "ldr d8, [x16, #0x0]\n" - "mov v3.d[1], x19\n" + "mov v3.d[1], x8\n" "ldr x26, [x16, #0x8]\n" "mov v8.d[1], x26\n" "bge 75b\n" @@ -804,21 +804,21 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "79:" // Height 4: Multiply loop: No odd multiplies "ldr w19, [%x[args_ptr], %[offsetof_num_strings]]\n" "add x13, x13, #0x1\n" - "cmp x13, x19\n" + "cmp x13, x8\n" "bne 72b\n" - "ldr x19, [%x[args_ptr], %[offsetof_output_offset]]\n" + "ldr x8, [%x[args_ptr], %[offsetof_output_offset]]\n" "prfm pstl1keep, [x14, #0x0]\n" - "add x27, x14, x19, LSL #2\n" + "add x27, x14, x8, LSL #2\n" "prfm pstl1keep, [x27, #0x0]\n" - "add x26, x27, x19, LSL #2\n" + "add x26, x27, x8, LSL #2\n" "prfm pstl1keep, [x26, #0x0]\n" - "add x25, x26, x19, LSL #2\n" + "add x25, x26, x8, LSL #2\n" "prfm pstl1keep, [x25, #0x0]\n" "tbz %x[flags], #1, 80f\n" "add x20, %x[args_ptr], %[offset_min]\n" - "add x19, %x[args_ptr], %[offset_max]\n" + "add x8, %x[args_ptr], %[offset_max]\n" "ld1r { v17.4s }, [x20]\n" - "ld1r { v16.4s }, [x19]\n" + "ld1r { v16.4s }, [x8]\n" "fmin v24.4s, v24.4s, v16.4s\n" "fmin v25.4s, v25.4s, v16.4s\n" "fmin v26.4s, v26.4s, v16.4s\n" @@ -874,17 +874,17 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "b 92f\n" "87:" // Height 5: no bias "tbz %x[flags], #0, 91f\n" - "ldr x19, [%x[args_ptr], %[offsetof_output_offset]]\n" + "ldr x8, [%x[args_ptr], %[offsetof_output_offset]]\n" "cmp x17, #0x4\n" - "add x27, x14, x19, LSL #2\n" - "add x26, x27, x19, LSL #2\n" - "add x25, x26, x19, LSL #2\n" - "add x24, x25, x19, LSL #2\n" + "add x27, x14, x8, LSL #2\n" + "add x26, x27, x8, LSL #2\n" + "add x25, x26, x8, LSL #2\n" + "add x24, x25, x8, LSL #2\n" "bge 90f\n" "tbz x17, #1, 88f\n" "ldr d24, [x14], #0x8\n" "ldr d25, [x27], #0x8\n" - "mov x19, #0x8\n" + "mov x8, #0x8\n" "ldr d26, [x26], #0x8\n" "ldr d27, [x25], #0x8\n" "ldr d28, [x24], #0x8\n" @@ -897,13 +897,13 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "b 89f\n" "88:" // Height 5: Partial accumulate: partial_1_0 "ldr s24, [x14, #0x0]\n" - "mov x19, #0x0\n" + "mov x8, #0x0\n" "ldr s25, [x27, #0x0]\n" "ldr s26, [x26, #0x0]\n" "ldr s27, [x25, #0x0]\n" "ldr s28, [x24, #0x0]\n" "89:" // Height 5: Partial accumulate: Done - "sub x14, x14, x19\n" + "sub x14, x14, x8\n" "b 92f\n" "90:" // Height 5: full accumulate "ldr q24, [x14, #0x0]\n" @@ -922,30 +922,30 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "mov x13, #0x0\n" "93:" // Height 5: String loop "ldr x20, [%x[args_ptr], %[offsetof_string_lengths]]\n" - "ldr x19, [%x[args_ptr], %[offsetof_input_offset]]\n" + "ldr x8, [%x[args_ptr], %[offsetof_input_offset]]\n" "ldr w12, [x20, x13, LSL #0x2]\n" "tbz %x[flags], #3, 94f\n" "ldr x20, [%x[input_ptr], x13, LSL #0x3]\n" - "add x20, x20, x19, LSL #3\n" + "add x20, x20, x8, LSL #3\n" "ldr x11, [x20, #0x0]\n" "ldr x9, [x20, #0x8]\n" "ldr x27, [x20, #0x10]\n" "ldr x25, [x20, #0x18]\n" "ldr x24, [x20, #0x20]\n" "cbnz x13, 95f\n" - "ldr x19, [%x[args_ptr], %[offsetof_input_initial_col]]\n" - "add x11, x11, x19, LSL #2\n" - "add x9, x9, x19, LSL #2\n" - "add x27, x27, x19, LSL #2\n" - "add x25, x25, x19, LSL #2\n" - "add x24, x24, x19, LSL #2\n" + "ldr x8, [%x[args_ptr], %[offsetof_input_initial_col]]\n" + "add x11, x11, x8, LSL #2\n" + "add x9, x9, x8, LSL #2\n" + "add x27, x27, x8, LSL #2\n" + "add x25, x25, x8, LSL #2\n" + "add x24, x24, x8, LSL #2\n" "b 95f\n" "94:" // Height 5: setup direct input "mov x11, %x[input_ptr]\n" - "add x9, x11, x19, LSL #2\n" - "add x27, x9, x19, LSL #2\n" - "add x25, x27, x19, LSL #2\n" - "add x24, x25, x19, LSL #2\n" + "add x9, x11, x8, LSL #2\n" + "add x27, x9, x8, LSL #2\n" + "add x25, x27, x8, LSL #2\n" + "add x24, x25, x8, LSL #2\n" "95:" // Height 5: input setup done "cmp x12, #0x4\n" "blt 98f\n" @@ -961,25 +961,25 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "fmla v24.4s, v8.4s, v0.s[0]\n" "ldr d9, [x16, #0x10]\n" "fmla v25.4s, v8.4s, v1.s[0]\n" - "ldr x19, [x16, #0x18]\n" + "ldr x8, [x16, #0x18]\n" "fmla v26.4s, v8.4s, v2.s[0]\n" "ldr d10, [x16, #0x20]\n" "fmla v27.4s, v8.4s, v3.s[0]\n" "ldr x21, [x16, #0x28]\n" "fmla v28.4s, v8.4s, v4.s[0]\n" - "mov v9.d[1], x19\n" + "mov v9.d[1], x8\n" "ldr d11, [x16, #0x30]\n" "add x11, x11, #0x10\n" "fmla v24.4s, v9.4s, v0.s[1]\n" "mov v10.d[1], x21\n" "fmla v25.4s, v9.4s, v1.s[1]\n" - "ldr x19, [x16, #0x38]\n" + "ldr x8, [x16, #0x38]\n" "fmla v26.4s, v9.4s, v2.s[1]\n" "prfm pldl1keep, [x11, #0x80]\n" "fmla v27.4s, v9.4s, v3.s[1]\n" "ldr x10, [x11, #0x8]\n" "fmla v28.4s, v9.4s, v4.s[1]\n" - "mov v11.d[1], x19\n" + "mov v11.d[1], x8\n" "fmla v24.4s, v10.4s, v0.s[2]\n" "add x9, x9, #0x10\n" "fmla v25.4s, v10.4s, v1.s[2]\n" @@ -1008,12 +1008,12 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "sub x12, x12, #0x4\n" "mov v2.d[1], x26\n" "ldr d3, [x25, #0x0]\n" - "ldr x19, [x25, #0x8]\n" + "ldr x8, [x25, #0x8]\n" "cmp x12, #0x8\n" "ldr d4, [x24, #0x0]\n" "add x16, x16, #0x40\n" "ldr x21, [x24, #0x8]\n" - "mov v3.d[1], x19\n" + "mov v3.d[1], x8\n" "ldr d8, [x16, #0x0]\n" "ldr x26, [x16, #0x8]\n" "mov v4.d[1], x21\n" @@ -1075,23 +1075,23 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "100:" // Height 5: Multiply loop: No odd multiplies "ldr w19, [%x[args_ptr], %[offsetof_num_strings]]\n" "add x13, x13, #0x1\n" - "cmp x13, x19\n" + "cmp x13, x8\n" "bne 93b\n" - "ldr x19, [%x[args_ptr], %[offsetof_output_offset]]\n" + "ldr x8, [%x[args_ptr], %[offsetof_output_offset]]\n" "prfm pstl1keep, [x14, #0x0]\n" - "add x27, x14, x19, LSL #2\n" + "add x27, x14, x8, LSL #2\n" "prfm pstl1keep, [x27, #0x0]\n" - "add x26, x27, x19, LSL #2\n" + "add x26, x27, x8, LSL #2\n" "prfm pstl1keep, [x26, #0x0]\n" - "add x25, x26, x19, LSL #2\n" + "add x25, x26, x8, LSL #2\n" "prfm pstl1keep, [x25, #0x0]\n" - "add x24, x25, x19, LSL #2\n" + "add x24, x25, x8, LSL #2\n" "prfm pstl1keep, [x24, #0x0]\n" "tbz %x[flags], #1, 101f\n" "add x20, %x[args_ptr], %[offset_min]\n" - "add x19, %x[args_ptr], %[offset_max]\n" + "add x8, %x[args_ptr], %[offset_max]\n" "ld1r { v17.4s }, [x20]\n" - "ld1r { v16.4s }, [x19]\n" + "ld1r { v16.4s }, [x8]\n" "fmin v24.4s, v24.4s, v16.4s\n" "fmin v25.4s, v25.4s, v16.4s\n" "fmin v26.4s, v26.4s, v16.4s\n" @@ -1154,18 +1154,18 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "b 113f\n" "108:" // Height 6: no bias "tbz %x[flags], #0, 112f\n" - "ldr x19, [%x[args_ptr], %[offsetof_output_offset]]\n" + "ldr x8, [%x[args_ptr], %[offsetof_output_offset]]\n" "cmp x17, #0x4\n" - "add x27, x14, x19, LSL #2\n" - "add x26, x27, x19, LSL #2\n" - "add x25, x26, x19, LSL #2\n" - "add x24, x25, x19, LSL #2\n" - "add x23, x24, x19, LSL #2\n" + "add x27, x14, x8, LSL #2\n" + "add x26, x27, x8, LSL #2\n" + "add x25, x26, x8, LSL #2\n" + "add x24, x25, x8, LSL #2\n" + "add x23, x24, x8, LSL #2\n" "bge 111f\n" "tbz x17, #1, 109f\n" "ldr d24, [x14], #0x8\n" "ldr d25, [x27], #0x8\n" - "mov x19, #0x8\n" + "mov x8, #0x8\n" "ldr d26, [x26], #0x8\n" "ldr d27, [x25], #0x8\n" "ldr d28, [x24], #0x8\n" @@ -1180,14 +1180,14 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "b 110f\n" "109:" // Height 6: Partial accumulate: partial_1_0 "ldr s24, [x14, #0x0]\n" - "mov x19, #0x0\n" + "mov x8, #0x0\n" "ldr s25, [x27, #0x0]\n" "ldr s26, [x26, #0x0]\n" "ldr s27, [x25, #0x0]\n" "ldr s28, [x24, #0x0]\n" "ldr s29, [x23, #0x0]\n" "110:" // Height 6: Partial accumulate: Done - "sub x14, x14, x19\n" + "sub x14, x14, x8\n" "b 113f\n" "111:" // Height 6: full accumulate "ldr q24, [x14, #0x0]\n" @@ -1208,11 +1208,11 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "mov x13, #0x0\n" "114:" // Height 6: String loop "ldr x20, [%x[args_ptr], %[offsetof_string_lengths]]\n" - "ldr x19, [%x[args_ptr], %[offsetof_input_offset]]\n" + "ldr x8, [%x[args_ptr], %[offsetof_input_offset]]\n" "ldr w12, [x20, x13, LSL #0x2]\n" "tbz %x[flags], #3, 115f\n" "ldr x20, [%x[input_ptr], x13, LSL #0x3]\n" - "add x20, x20, x19, LSL #3\n" + "add x20, x20, x8, LSL #3\n" "ldr x11, [x20, #0x0]\n" "ldr x9, [x20, #0x8]\n" "ldr x27, [x20, #0x10]\n" @@ -1220,21 +1220,21 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "ldr x24, [x20, #0x20]\n" "ldr x23, [x20, #0x28]\n" "cbnz x13, 116f\n" - "ldr x19, [%x[args_ptr], %[offsetof_input_initial_col]]\n" - "add x11, x11, x19, LSL #2\n" - "add x9, x9, x19, LSL #2\n" - "add x27, x27, x19, LSL #2\n" - "add x25, x25, x19, LSL #2\n" - "add x24, x24, x19, LSL #2\n" - "add x23, x23, x19, LSL #2\n" + "ldr x8, [%x[args_ptr], %[offsetof_input_initial_col]]\n" + "add x11, x11, x8, LSL #2\n" + "add x9, x9, x8, LSL #2\n" + "add x27, x27, x8, LSL #2\n" + "add x25, x25, x8, LSL #2\n" + "add x24, x24, x8, LSL #2\n" + "add x23, x23, x8, LSL #2\n" "b 116f\n" "115:" // Height 6: setup direct input "mov x11, %x[input_ptr]\n" - "add x9, x11, x19, LSL #2\n" - "add x27, x9, x19, LSL #2\n" - "add x25, x27, x19, LSL #2\n" - "add x24, x25, x19, LSL #2\n" - "add x23, x24, x19, LSL #2\n" + "add x9, x11, x8, LSL #2\n" + "add x27, x9, x8, LSL #2\n" + "add x25, x27, x8, LSL #2\n" + "add x24, x25, x8, LSL #2\n" + "add x23, x24, x8, LSL #2\n" "116:" // Height 6: input setup done "cmp x12, #0x4\n" "blt 119f\n" @@ -1251,25 +1251,25 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "fmla v24.4s, v8.4s, v0.s[0]\n" "ldr d9, [x16, #0x10]\n" "fmla v25.4s, v8.4s, v1.s[0]\n" - "ldr x19, [x16, #0x18]\n" + "ldr x8, [x16, #0x18]\n" "fmla v26.4s, v8.4s, v2.s[0]\n" "ldr d10, [x16, #0x20]\n" "fmla v27.4s, v8.4s, v3.s[0]\n" "ldr x21, [x16, #0x28]\n" "fmla v28.4s, v8.4s, v4.s[0]\n" - "mov v9.d[1], x19\n" + "mov v9.d[1], x8\n" "fmla v29.4s, v8.4s, v5.s[0]\n" "ldr d11, [x16, #0x30]\n" "fmla v24.4s, v9.4s, v0.s[1]\n" "mov v10.d[1], x21\n" "fmla v25.4s, v9.4s, v1.s[1]\n" - "ldr x19, [x16, #0x38]\n" + "ldr x8, [x16, #0x38]\n" "fmla v26.4s, v9.4s, v2.s[1]\n" "add x11, x11, #0x10\n" "fmla v27.4s, v9.4s, v3.s[1]\n" "prfm pldl1keep, [x11, #0x80]\n" "fmla v28.4s, v9.4s, v4.s[1]\n" - "mov v11.d[1], x19\n" + "mov v11.d[1], x8\n" "fmla v29.4s, v9.4s, v5.s[1]\n" "ldr x10, [x11, #0x8]\n" "fmla v24.4s, v10.4s, v0.s[2]\n" @@ -1306,16 +1306,16 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "cmp x12, #0x8\n" "ldr d3, [x25, #0x0]\n" "add x16, x16, #0x40\n" - "ldr x19, [x25, #0x8]\n" + "ldr x8, [x25, #0x8]\n" "ldr d4, [x24, #0x0]\n" "ldr x21, [x24, #0x8]\n" - "mov v3.d[1], x19\n" + "mov v3.d[1], x8\n" "ldr d5, [x23, #0x0]\n" - "ldr x19, [x23, #0x8]\n" + "ldr x8, [x23, #0x8]\n" "mov v4.d[1], x21\n" "ldr d8, [x16, #0x0]\n" "ldr x26, [x16, #0x8]\n" - "mov v5.d[1], x19\n" + "mov v5.d[1], x8\n" "mov v8.d[1], x26\n" "bge 117b\n" "118:" // Height 6: Multiply loop: Single iteration only @@ -1382,25 +1382,25 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "121:" // Height 6: Multiply loop: No odd multiplies "ldr w19, [%x[args_ptr], %[offsetof_num_strings]]\n" "add x13, x13, #0x1\n" - "cmp x13, x19\n" + "cmp x13, x8\n" "bne 114b\n" - "ldr x19, [%x[args_ptr], %[offsetof_output_offset]]\n" + "ldr x8, [%x[args_ptr], %[offsetof_output_offset]]\n" "prfm pstl1keep, [x14, #0x0]\n" - "add x27, x14, x19, LSL #2\n" + "add x27, x14, x8, LSL #2\n" "prfm pstl1keep, [x27, #0x0]\n" - "add x26, x27, x19, LSL #2\n" + "add x26, x27, x8, LSL #2\n" "prfm pstl1keep, [x26, #0x0]\n" - "add x25, x26, x19, LSL #2\n" + "add x25, x26, x8, LSL #2\n" "prfm pstl1keep, [x25, #0x0]\n" - "add x24, x25, x19, LSL #2\n" + "add x24, x25, x8, LSL #2\n" "prfm pstl1keep, [x24, #0x0]\n" - "add x23, x24, x19, LSL #2\n" + "add x23, x24, x8, LSL #2\n" "prfm pstl1keep, [x23, #0x0]\n" "tbz %x[flags], #1, 122f\n" "add x20, %x[args_ptr], %[offset_min]\n" - "add x19, %x[args_ptr], %[offset_max]\n" + "add x8, %x[args_ptr], %[offset_max]\n" "ld1r { v17.4s }, [x20]\n" - "ld1r { v16.4s }, [x19]\n" + "ld1r { v16.4s }, [x8]\n" "fmin v24.4s, v24.4s, v16.4s\n" "fmin v25.4s, v25.4s, v16.4s\n" "fmin v26.4s, v26.4s, v16.4s\n" @@ -1470,19 +1470,19 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "b 134f\n" "129:" // Height 7: no bias "tbz %x[flags], #0, 133f\n" - "ldr x19, [%x[args_ptr], %[offsetof_output_offset]]\n" + "ldr x8, [%x[args_ptr], %[offsetof_output_offset]]\n" "cmp x17, #0x4\n" - "add x27, x14, x19, LSL #2\n" - "add x26, x27, x19, LSL #2\n" - "add x25, x26, x19, LSL #2\n" - "add x24, x25, x19, LSL #2\n" - "add x23, x24, x19, LSL #2\n" - "add x22, x23, x19, LSL #2\n" + "add x27, x14, x8, LSL #2\n" + "add x26, x27, x8, LSL #2\n" + "add x25, x26, x8, LSL #2\n" + "add x24, x25, x8, LSL #2\n" + "add x23, x24, x8, LSL #2\n" + "add x22, x23, x8, LSL #2\n" "bge 132f\n" "tbz x17, #1, 130f\n" "ldr d24, [x14], #0x8\n" "ldr d25, [x27], #0x8\n" - "mov x19, #0x8\n" + "mov x8, #0x8\n" "ldr d26, [x26], #0x8\n" "ldr d27, [x25], #0x8\n" "ldr d28, [x24], #0x8\n" @@ -1499,7 +1499,7 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "b 131f\n" "130:" // Height 7: Partial accumulate: partial_1_0 "ldr s24, [x14, #0x0]\n" - "mov x19, #0x0\n" + "mov x8, #0x0\n" "ldr s25, [x27, #0x0]\n" "ldr s26, [x26, #0x0]\n" "ldr s27, [x25, #0x0]\n" @@ -1507,7 +1507,7 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "ldr s29, [x23, #0x0]\n" "ldr s30, [x22, #0x0]\n" "131:" // Height 7: Partial accumulate: Done - "sub x14, x14, x19\n" + "sub x14, x14, x8\n" "b 134f\n" "132:" // Height 7: full accumulate "ldr q24, [x14, #0x0]\n" @@ -1530,11 +1530,11 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "mov x13, #0x0\n" "135:" // Height 7: String loop "ldr x20, [%x[args_ptr], %[offsetof_string_lengths]]\n" - "ldr x19, [%x[args_ptr], %[offsetof_input_offset]]\n" + "ldr x8, [%x[args_ptr], %[offsetof_input_offset]]\n" "ldr w12, [x20, x13, LSL #0x2]\n" "tbz %x[flags], #3, 136f\n" "ldr x20, [%x[input_ptr], x13, LSL #0x3]\n" - "add x20, x20, x19, LSL #3\n" + "add x20, x20, x8, LSL #3\n" "ldr x11, [x20, #0x0]\n" "ldr x9, [x20, #0x8]\n" "ldr x27, [x20, #0x10]\n" @@ -1543,23 +1543,23 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "ldr x23, [x20, #0x28]\n" "ldr x22, [x20, #0x30]\n" "cbnz x13, 137f\n" - "ldr x19, [%x[args_ptr], %[offsetof_input_initial_col]]\n" - "add x11, x11, x19, LSL #2\n" - "add x9, x9, x19, LSL #2\n" - "add x27, x27, x19, LSL #2\n" - "add x25, x25, x19, LSL #2\n" - "add x24, x24, x19, LSL #2\n" - "add x23, x23, x19, LSL #2\n" - "add x22, x22, x19, LSL #2\n" + "ldr x8, [%x[args_ptr], %[offsetof_input_initial_col]]\n" + "add x11, x11, x8, LSL #2\n" + "add x9, x9, x8, LSL #2\n" + "add x27, x27, x8, LSL #2\n" + "add x25, x25, x8, LSL #2\n" + "add x24, x24, x8, LSL #2\n" + "add x23, x23, x8, LSL #2\n" + "add x22, x22, x8, LSL #2\n" "b 137f\n" "136:" // Height 7: setup direct input "mov x11, %x[input_ptr]\n" - "add x9, x11, x19, LSL #2\n" - "add x27, x9, x19, LSL #2\n" - "add x25, x27, x19, LSL #2\n" - "add x24, x25, x19, LSL #2\n" - "add x23, x24, x19, LSL #2\n" - "add x22, x23, x19, LSL #2\n" + "add x9, x11, x8, LSL #2\n" + "add x27, x9, x8, LSL #2\n" + "add x25, x27, x8, LSL #2\n" + "add x24, x25, x8, LSL #2\n" + "add x23, x24, x8, LSL #2\n" + "add x22, x23, x8, LSL #2\n" "137:" // Height 7: input setup done "cmp x12, #0x4\n" "blt 140f\n" @@ -1577,25 +1577,25 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "fmla v24.4s, v8.4s, v0.s[0]\n" "ldr d9, [x16, #0x10]\n" "fmla v25.4s, v8.4s, v1.s[0]\n" - "ldr x19, [x16, #0x18]\n" + "ldr x8, [x16, #0x18]\n" "fmla v26.4s, v8.4s, v2.s[0]\n" "ldr d10, [x16, #0x20]\n" "fmla v27.4s, v8.4s, v3.s[0]\n" "ldr x21, [x16, #0x28]\n" "fmla v28.4s, v8.4s, v4.s[0]\n" - "mov v9.d[1], x19\n" + "mov v9.d[1], x8\n" "fmla v29.4s, v8.4s, v5.s[0]\n" "ldr d11, [x16, #0x30]\n" "fmla v30.4s, v8.4s, v6.s[0]\n" "mov v10.d[1], x21\n" "fmla v24.4s, v9.4s, v0.s[1]\n" - "ldr x19, [x16, #0x38]\n" + "ldr x8, [x16, #0x38]\n" "fmla v25.4s, v9.4s, v1.s[1]\n" "add x11, x11, #0x10\n" "fmla v26.4s, v9.4s, v2.s[1]\n" "prfm pldl1keep, [x11, #0x80]\n" "fmla v27.4s, v9.4s, v3.s[1]\n" - "mov v11.d[1], x19\n" + "mov v11.d[1], x8\n" "fmla v28.4s, v9.4s, v4.s[1]\n" "ldr x10, [x11, #0x8]\n" "fmla v29.4s, v9.4s, v5.s[1]\n" @@ -1615,7 +1615,7 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "fmla v29.4s, v10.4s, v5.s[2]\n" "prfm pldl1keep, [x25, #0x80]\n" "fmla v30.4s, v10.4s, v6.s[2]\n" - "ldr x19, [x25, #0x8]\n" + "ldr x8, [x25, #0x8]\n" "fmla v24.4s, v11.4s, v0.s[3]\n" "ldr d0, [x11, #0x0]\n" "fmla v25.4s, v11.4s, v1.s[3]\n" @@ -1634,7 +1634,7 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "add x23, x23, #0x10\n" "prfm pldl1keep, [x24, #0x80]\n" "add x22, x22, #0x10\n" - "mov v3.d[1], x19\n" + "mov v3.d[1], x8\n" "prfm pldl1keep, [x23, #0x80]\n" "prfm pldl1keep, [x22, #0x80]\n" "sub x12, x12, #0x4\n" @@ -1646,11 +1646,11 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "ldr x26, [x16, #0x8]\n" "mov v4.d[1], x21\n" "ldr d5, [x23, #0x0]\n" - "ldr x19, [x23, #0x8]\n" + "ldr x8, [x23, #0x8]\n" "mov v8.d[1], x26\n" "ldr d6, [x22, #0x0]\n" "ldr x21, [x22, #0x8]\n" - "mov v5.d[1], x19\n" + "mov v5.d[1], x8\n" "mov v6.d[1], x21\n" "bge 138b\n" "139:" // Height 7: Multiply loop: Single iteration only @@ -1725,27 +1725,27 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "142:" // Height 7: Multiply loop: No odd multiplies "ldr w19, [%x[args_ptr], %[offsetof_num_strings]]\n" "add x13, x13, #0x1\n" - "cmp x13, x19\n" + "cmp x13, x8\n" "bne 135b\n" - "ldr x19, [%x[args_ptr], %[offsetof_output_offset]]\n" + "ldr x8, [%x[args_ptr], %[offsetof_output_offset]]\n" "prfm pstl1keep, [x14, #0x0]\n" - "add x27, x14, x19, LSL #2\n" + "add x27, x14, x8, LSL #2\n" "prfm pstl1keep, [x27, #0x0]\n" - "add x26, x27, x19, LSL #2\n" + "add x26, x27, x8, LSL #2\n" "prfm pstl1keep, [x26, #0x0]\n" - "add x25, x26, x19, LSL #2\n" + "add x25, x26, x8, LSL #2\n" "prfm pstl1keep, [x25, #0x0]\n" - "add x24, x25, x19, LSL #2\n" + "add x24, x25, x8, LSL #2\n" "prfm pstl1keep, [x24, #0x0]\n" - "add x23, x24, x19, LSL #2\n" + "add x23, x24, x8, LSL #2\n" "prfm pstl1keep, [x23, #0x0]\n" - "add x22, x23, x19, LSL #2\n" + "add x22, x23, x8, LSL #2\n" "prfm pstl1keep, [x22, #0x0]\n" "tbz %x[flags], #1, 143f\n" "add x20, %x[args_ptr], %[offset_min]\n" - "add x19, %x[args_ptr], %[offset_max]\n" + "add x8, %x[args_ptr], %[offset_max]\n" "ld1r { v17.4s }, [x20]\n" - "ld1r { v16.4s }, [x19]\n" + "ld1r { v16.4s }, [x8]\n" "fmin v24.4s, v24.4s, v16.4s\n" "fmin v25.4s, v25.4s, v16.4s\n" "fmin v26.4s, v26.4s, v16.4s\n" @@ -1809,8 +1809,8 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "ldr x16, [%x[args_ptr], %[offsetof_B_ptr]]\n" "mov x14, %x[output_ptr]\n" "ldr x20, [%x[args_ptr], %[offsetof_output_offset]]\n" - "mov x19, #0x20\n" - "madd %x[output_ptr], x20, x19, %x[output_ptr]\n" + "mov x8, #0x20\n" + "madd %x[output_ptr], x20, x8, %x[output_ptr]\n" "149:" // Height 8: Column loop "cbz x15, 150f\n" "ldr q24, [x15, #0x0]\n" @@ -1825,20 +1825,20 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "b 155f\n" "150:" // Height 8: no bias "tbz %x[flags], #0, 154f\n" - "ldr x19, [%x[args_ptr], %[offsetof_output_offset]]\n" + "ldr x8, [%x[args_ptr], %[offsetof_output_offset]]\n" "cmp x17, #0x4\n" - "add x27, x14, x19, LSL #2\n" - "add x26, x27, x19, LSL #2\n" - "add x25, x26, x19, LSL #2\n" - "add x24, x25, x19, LSL #2\n" - "add x23, x24, x19, LSL #2\n" - "add x22, x23, x19, LSL #2\n" - "add x21, x22, x19, LSL #2\n" + "add x27, x14, x8, LSL #2\n" + "add x26, x27, x8, LSL #2\n" + "add x25, x26, x8, LSL #2\n" + "add x24, x25, x8, LSL #2\n" + "add x23, x24, x8, LSL #2\n" + "add x22, x23, x8, LSL #2\n" + "add x21, x22, x8, LSL #2\n" "bge 153f\n" "tbz x17, #1, 151f\n" "ldr d24, [x14], #0x8\n" "ldr d25, [x27], #0x8\n" - "mov x19, #0x8\n" + "mov x8, #0x8\n" "ldr d26, [x26], #0x8\n" "ldr d27, [x25], #0x8\n" "ldr d28, [x24], #0x8\n" @@ -1857,7 +1857,7 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "b 152f\n" "151:" // Height 8: Partial accumulate: partial_1_0 "ldr s24, [x14, #0x0]\n" - "mov x19, #0x0\n" + "mov x8, #0x0\n" "ldr s25, [x27, #0x0]\n" "ldr s26, [x26, #0x0]\n" "ldr s27, [x25, #0x0]\n" @@ -1866,7 +1866,7 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "ldr s30, [x22, #0x0]\n" "ldr s31, [x21, #0x0]\n" "152:" // Height 8: Partial accumulate: Done - "sub x14, x14, x19\n" + "sub x14, x14, x8\n" "b 155f\n" "153:" // Height 8: full accumulate "ldr q24, [x14, #0x0]\n" @@ -1891,11 +1891,11 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "mov x13, #0x0\n" "156:" // Height 8: String loop "ldr x20, [%x[args_ptr], %[offsetof_string_lengths]]\n" - "ldr x19, [%x[args_ptr], %[offsetof_input_offset]]\n" + "ldr x8, [%x[args_ptr], %[offsetof_input_offset]]\n" "ldr w12, [x20, x13, LSL #0x2]\n" "tbz %x[flags], #3, 157f\n" "ldr x20, [%x[input_ptr], x13, LSL #0x3]\n" - "add x20, x20, x19, LSL #3\n" + "add x20, x20, x8, LSL #3\n" "ldr x11, [x20, #0x0]\n" "ldr x9, [x20, #0x8]\n" "ldr x27, [x20, #0x10]\n" @@ -1905,25 +1905,25 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "ldr x22, [x20, #0x30]\n" "ldr x20, [x20, #0x38]\n" "cbnz x13, 158f\n" - "ldr x19, [%x[args_ptr], %[offsetof_input_initial_col]]\n" - "add x11, x11, x19, LSL #2\n" - "add x9, x9, x19, LSL #2\n" - "add x27, x27, x19, LSL #2\n" - "add x25, x25, x19, LSL #2\n" - "add x24, x24, x19, LSL #2\n" - "add x23, x23, x19, LSL #2\n" - "add x22, x22, x19, LSL #2\n" - "add x20, x20, x19, LSL #2\n" + "ldr x8, [%x[args_ptr], %[offsetof_input_initial_col]]\n" + "add x11, x11, x8, LSL #2\n" + "add x9, x9, x8, LSL #2\n" + "add x27, x27, x8, LSL #2\n" + "add x25, x25, x8, LSL #2\n" + "add x24, x24, x8, LSL #2\n" + "add x23, x23, x8, LSL #2\n" + "add x22, x22, x8, LSL #2\n" + "add x20, x20, x8, LSL #2\n" "b 158f\n" "157:" // Height 8: setup direct input "mov x11, %x[input_ptr]\n" - "add x9, x11, x19, LSL #2\n" - "add x27, x9, x19, LSL #2\n" - "add x25, x27, x19, LSL #2\n" - "add x24, x25, x19, LSL #2\n" - "add x23, x24, x19, LSL #2\n" - "add x22, x23, x19, LSL #2\n" - "add x20, x22, x19, LSL #2\n" + "add x9, x11, x8, LSL #2\n" + "add x27, x9, x8, LSL #2\n" + "add x25, x27, x8, LSL #2\n" + "add x24, x25, x8, LSL #2\n" + "add x23, x24, x8, LSL #2\n" + "add x22, x23, x8, LSL #2\n" + "add x20, x22, x8, LSL #2\n" "158:" // Height 8: input setup done "cmp x12, #0x4\n" "blt 161f\n" @@ -1942,25 +1942,25 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "fmla v24.4s, v8.4s, v0.s[0]\n" "ldr d9, [x16, #0x10]\n" "fmla v25.4s, v8.4s, v1.s[0]\n" - "ldr x19, [x16, #0x18]\n" + "ldr x8, [x16, #0x18]\n" "fmla v26.4s, v8.4s, v2.s[0]\n" "ldr d10, [x16, #0x20]\n" "fmla v27.4s, v8.4s, v3.s[0]\n" "ldr x21, [x16, #0x28]\n" "fmla v28.4s, v8.4s, v4.s[0]\n" - "mov v9.d[1], x19\n" + "mov v9.d[1], x8\n" "fmla v29.4s, v8.4s, v5.s[0]\n" "ldr d11, [x16, #0x30]\n" "fmla v30.4s, v8.4s, v6.s[0]\n" "mov v10.d[1], x21\n" "fmla v31.4s, v8.4s, v7.s[0]\n" - "ldr x19, [x16, #0x38]\n" + "ldr x8, [x16, #0x38]\n" "fmla v24.4s, v9.4s, v0.s[1]\n" "add x11, x11, #0x10\n" "fmla v25.4s, v9.4s, v1.s[1]\n" "prfm pldl1keep, [x11, #0x80]\n" "fmla v26.4s, v9.4s, v2.s[1]\n" - "mov v11.d[1], x19\n" + "mov v11.d[1], x8\n" "fmla v27.4s, v9.4s, v3.s[1]\n" "ldr x10, [x11, #0x8]\n" "fmla v28.4s, v9.4s, v4.s[1]\n" @@ -1980,7 +1980,7 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "fmla v27.4s, v10.4s, v3.s[2]\n" "prfm pldl1keep, [x25, #0x80]\n" "fmla v28.4s, v10.4s, v4.s[2]\n" - "ldr x19, [x25, #0x8]\n" + "ldr x8, [x25, #0x8]\n" "fmla v29.4s, v10.4s, v5.s[2]\n" "add x24, x24, #0x10\n" "fmla v30.4s, v10.4s, v6.s[2]\n" @@ -2005,7 +2005,7 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "ldr d4, [x24, #0x0]\n" "add x23, x23, #0x10\n" "add x22, x22, #0x10\n" - "mov v3.d[1], x19\n" + "mov v3.d[1], x8\n" "prfm pldl1keep, [x23, #0x80]\n" "mov v4.d[1], x21\n" "prfm pldl1keep, [x22, #0x80]\n" @@ -2013,19 +2013,19 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "add x20, x20, #0x10\n" "prfm pldl1keep, [x20, #0x80]\n" "sub x12, x12, #0x4\n" - "ldr x19, [x23, #0x8]\n" + "ldr x8, [x23, #0x8]\n" "cmp x12, #0x8\n" "ldr d6, [x22, #0x0]\n" "add x16, x16, #0x40\n" "ldr d8, [x16, #0x0]\n" - "mov v5.d[1], x19\n" + "mov v5.d[1], x8\n" "ldr x26, [x16, #0x8]\n" "ldr x21, [x22, #0x8]\n" "ldr d7, [x20, #0x0]\n" "mov v8.d[1], x26\n" - "ldr x19, [x20, #0x8]\n" + "ldr x8, [x20, #0x8]\n" "mov v6.d[1], x21\n" - "mov v7.d[1], x19\n" + "mov v7.d[1], x8\n" "bge 159b\n" "160:" // Height 8: Multiply loop: Single iteration only "fmla v24.4s, v8.4s, v0.s[0]\n" @@ -2107,29 +2107,29 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "163:" // Height 8: Multiply loop: No odd multiplies "ldr w19, [%x[args_ptr], %[offsetof_num_strings]]\n" "add x13, x13, #0x1\n" - "cmp x13, x19\n" + "cmp x13, x8\n" "bne 156b\n" - "ldr x19, [%x[args_ptr], %[offsetof_output_offset]]\n" + "ldr x8, [%x[args_ptr], %[offsetof_output_offset]]\n" "prfm pstl1keep, [x14, #0x0]\n" - "add x27, x14, x19, LSL #2\n" + "add x27, x14, x8, LSL #2\n" "prfm pstl1keep, [x27, #0x0]\n" - "add x26, x27, x19, LSL #2\n" + "add x26, x27, x8, LSL #2\n" "prfm pstl1keep, [x26, #0x0]\n" - "add x25, x26, x19, LSL #2\n" + "add x25, x26, x8, LSL #2\n" "prfm pstl1keep, [x25, #0x0]\n" - "add x24, x25, x19, LSL #2\n" + "add x24, x25, x8, LSL #2\n" "prfm pstl1keep, [x24, #0x0]\n" - "add x23, x24, x19, LSL #2\n" + "add x23, x24, x8, LSL #2\n" "prfm pstl1keep, [x23, #0x0]\n" - "add x22, x23, x19, LSL #2\n" + "add x22, x23, x8, LSL #2\n" "prfm pstl1keep, [x22, #0x0]\n" - "add x21, x22, x19, LSL #2\n" + "add x21, x22, x8, LSL #2\n" "prfm pstl1keep, [x21, #0x0]\n" "tbz %x[flags], #1, 164f\n" "add x20, %x[args_ptr], %[offset_min]\n" - "add x19, %x[args_ptr], %[offset_max]\n" + "add x8, %x[args_ptr], %[offset_max]\n" "ld1r { v17.4s }, [x20]\n" - "ld1r { v16.4s }, [x19]\n" + "ld1r { v16.4s }, [x8]\n" "fmin v24.4s, v24.4s, v16.4s\n" "fmin v25.4s, v25.4s, v16.4s\n" "fmin v26.4s, v26.4s, v16.4s\n" @@ -2200,14 +2200,14 @@ void a64_hybrid_fp32_mla_8x4_a55 ( "str x20, [%x[args_ptr], %[offsetof_input_offset]]\n" "b 1b\n" "169:" // Update direct input - "mov x19, #0x20\n" - "madd %x[input_ptr], x19, x20, %x[input_ptr]\n" + "mov x8, #0x20\n" + "madd %x[input_ptr], x8, x20, %x[input_ptr]\n" "b 1b\n" "170:" // 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)) - : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v8", "v9", "v10", "v11", "v12", "v16", "v17", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x16", "x17", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28" + : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v8", "v9", "v10", "v11", "v12", "v16", "v17", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x16", "x17", "x8", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28" ); } |