diff options
author | Michael Tyler <michael.tyler@arm.com> | 2023-01-17 11:04:14 +0000 |
---|---|---|
committer | Gian Marco Iodice <gianmarco.iodice@arm.com> | 2023-01-18 09:43:38 +0000 |
commit | be13cead34e566bdd561ad3ffc3f645b460e482e (patch) | |
tree | cdc086de205d5a07fdd816afa6333d0b2f38d4e9 /src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_fp32_fp32.hpp | |
parent | 13bab71a76096985752a9e12711507021e25858d (diff) | |
download | ComputeLibrary-be13cead34e566bdd561ad3ffc3f645b460e482e.tar.gz |
Revert "Update CPU kernels to remove x19"
This reverts commit 3c59f01c209d2732a15d97d65565ead964787a8b.
Resolves: COMPMID-5817
Change-Id: Ie2443a21854a95db1e3d0cafa2121c0187a5e237
Signed-off-by: Michael Tyler <michael.tyler@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8974
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_fp32_fp32.hpp')
-rw-r--r-- | src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_fp32_fp32.hpp | 268 |
1 files changed, 134 insertions, 134 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_fp32_fp32.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_fp32_fp32.hpp index 7de88543d7..874fc797a4 100644 --- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_fp32_fp32.hpp +++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/sme_interleave1VL_fp32_fp32.hpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2022-2023 Arm Limited. + * Copyright (c) 2022 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. */ #if defined(__ARM_FEATURE_SVE) @@ -34,174 +34,174 @@ void interleave_block<1, 1, VLType::SME, false>( __asm__ __volatile__( ".inst 0xd503477f // SMSTART ZA\n" - "mov x22, %x[width]\n" - "incw x22\n" "cntw x10\n" - "sub x22, x22, #0x1\n" - "udiv x22, x22, x10\n" // n_passes = ceildiv(width, VL<T>) - "mov x21, %x[width]\n" - "sub x9, x10, #0x1\n" - "sub x20, x22, #0x1\n" - "ands x9, x21, x9\n" - "sub x28, x10, #0x2\n" - "mov x27, #0x0\n" - "mov x26, %x[in]\n" - "ldr x25, [x26, #0x0]\n" - "lsr x20, x20, #0x1\n" // n_loops = (n_passes - 1) / 2 - "and x24, x22, #0x1\n" // odd_tail = bool(n_passes & 0x1) - "ldr x23, [x26, #0x8]\n" - "csel x9, x9, x10, NE\n" + "mov x19, %x[width]\n" + "incw x19\n" + "sub x19, x19, #0x1\n" + "udiv x19, x19, x10\n" // n_passes = ceildiv(width, VL<T>) + "sub x9, x19, #0x1\n" + "lsr x9, x9, #0x1\n" // n_loops = (n_passes - 1) / 2 + "and x28, x19, #0x1\n" // odd_tail = bool(n_passes & 0x1) + "mov x19, %x[width]\n" + "sub x27, x10, #0x1\n" + "ands x27, x19, x27\n" + "csel x27, x27, x10, NE\n" + "sub x26, x10, #0x2\n" "ptrue p11.s\n" "whilelt p10.s, XZR, %x[height]\n" - "mov x22, %x[row_offset]\n" - "mov x21, %x[out]\n" - "whilelt p9.s, x27, %x[width]\n" - "whilelt p8.s, x27, %x[width]\n" - "add x26, x26, #0x10\n" + "mov x25, %x[row_offset]\n" + "mov x24, %x[out]\n" + "mov x23, #0x0\n" + "whilelt p9.s, x23, %x[width]\n" + "whilelt p8.s, x23, %x[width]\n" + "mov x22, %x[in]\n" + "ldr x21, [x22, #0x0]\n" + "ldr x20, [x22, #0x8]\n" + "add x22, x22, #0x10\n" "mov x12, #0x0\n" - "cbz x28, 2f\n" + "cbz x26, 2f\n" "1:" // K loop: Charge: Loop - ".inst 0x25306140 // psel p0.s, p8.s/Z, p10.s[w12]\n" - ".inst 0xe0960320 // ld1w { za0h.s[x12] }, p0/Z, [x25, x22, LSL #2]\n" - ".inst 0x25706140 // psel p0.s, p8.s/Z, p10.s[w12, #1]\n" - "ldr x25, [x26, #0x0]\n" - ".inst 0xe09602e1 // ld1w { za0h.s[x12, #1] }, p0/Z, [x23, x22, LSL #2]\n" + ".inst 0x25306140 // dup p0.s, p8.s/Z, p10.s[w12]\n" + ".inst 0xe09902a0 // ld1w { za0h.s[x12] }, p0/Z, [x21, x25, LSL #2]\n" + ".inst 0x25706140 // dup p0.s, p8.s/Z, p10.s[w12, #1]\n" + "ldr x21, [x22, #0x0]\n" + ".inst 0xe0990281 // ld1w { za0h.s[x12, #1] }, p0/Z, [x20, x25, LSL #2]\n" + "ldr x20, [x22, #0x8]\n" + "add x22, x22, #0x10\n" "add x12, x12, #0x2\n" - "cmp x12, x28\n" - "ldr x23, [x26, #0x8]\n" - "add x26, x26, #0x10\n" + "cmp x12, x26\n" "blt 1b\n" "2:" // K loop: Charge: End - ".inst 0x25306140 // psel p0.s, p8.s/Z, p10.s[w12]\n" - ".inst 0xe0960320 // ld1w { za0h.s[x12] }, p0/Z, [x25, x22, LSL #2]\n" - ".inst 0x25706140 // psel p0.s, p8.s/Z, p10.s[w12, #1]\n" - "mov x26, %x[in]\n" - ".inst 0xe09602e1 // ld1w { za0h.s[x12, #1] }, p0/Z, [x23, x22, LSL #2]\n" - "ldr x25, [x26, #0x0]\n" - "incw x22\n" - "ldr x23, [x26, #0x8]\n" - "add x26, x26, #0x10\n" - "incw x27\n" - "cbz x20, 8f\n" - "mov x20, x20\n" + ".inst 0x25306140 // dup p0.s, p8.s/Z, p10.s[w12]\n" + ".inst 0xe09902a0 // ld1w { za0h.s[x12] }, p0/Z, [x21, x25, LSL #2]\n" + ".inst 0x25706140 // dup p0.s, p8.s/Z, p10.s[w12, #1]\n" + "mov x22, %x[in]\n" + ".inst 0xe0990281 // ld1w { za0h.s[x12, #1] }, p0/Z, [x20, x25, LSL #2]\n" + "ldr x21, [x22, #0x0]\n" + "ldr x20, [x22, #0x8]\n" + "add x22, x22, #0x10\n" + "incw x25\n" + "incw x23\n" + "cbz x9, 8f\n" + "mov x19, x9\n" "3:" // K loop: Main loop - "whilelt p8.s, x27, %x[width]\n" + "whilelt p8.s, x23, %x[width]\n" "mov x12, #0x0\n" - "cbz x28, 5f\n" + "cbz x26, 5f\n" "4:" // K loop: Main loop: First: Loop - ".inst 0x25306140 // psel p0.s, p8.s/Z, p10.s[w12]\n" - ".inst 0xe0960328 // ld1w { za2h.s[x12] }, p0/Z, [x25, x22, LSL #2]\n" - ".inst 0x25706141 // psel p1.s, p8.s/Z, p10.s[w12, #1]\n" - "ldr x25, [x26, #0x0]\n" - ".inst 0x25306d20 // psel p0.s, p11.s/Z, p9.s[w12]\n" - ".inst 0xe09606e9 // ld1w { za2h.s[x12, #1] }, p1/Z, [x23, x22, LSL #2]\n" - "ldr x23, [x26, #0x8]\n" - ".inst 0xe0bf82a0 // st1w { za0v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n" - ".inst 0x25706d20 // psel p0.s, p11.s/Z, p9.s[w12, #1]\n" - ".inst 0xe0aa82a1 // st1w { za0v.s[x12, #1] }, p0/Z, [x21, x10, LSL #2]\n" + ".inst 0x25306140 // dup p0.s, p8.s/Z, p10.s[w12]\n" + ".inst 0xe09902a8 // ld1w { za2h.s[x12] }, p0/Z, [x21, x25, LSL #2]\n" + ".inst 0x25706140 // dup p0.s, p8.s/Z, p10.s[w12, #1]\n" + "ldr x21, [x22, #0x0]\n" + ".inst 0xe0990289 // ld1w { za2h.s[x12, #1] }, p0/Z, [x20, x25, LSL #2]\n" + "ldr x20, [x22, #0x8]\n" + "add x22, x22, #0x10\n" + ".inst 0x25306d20 // dup p0.s, p11.s/Z, p9.s[w12]\n" + ".inst 0xe0bf8300 // st1w { za0v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n" + ".inst 0x25706d20 // dup p0.s, p11.s/Z, p9.s[w12, #1]\n" + ".inst 0xe0aa8301 // st1w { za0v.s[x12, #1] }, p0/Z, [x24, x10, LSL #2]\n" + "addvl x24, x24, #2\n" "add x12, x12, #0x2\n" - "cmp x12, x28\n" - "add x26, x26, #0x10\n" - "addvl x21, x21, #2\n" + "cmp x12, x26\n" "blt 4b\n" "5:" // K loop: Main loop: First: Tail - ".inst 0x25306140 // psel p0.s, p8.s/Z, p10.s[w12]\n" - ".inst 0xe0960328 // ld1w { za2h.s[x12] }, p0/Z, [x25, x22, LSL #2]\n" - "mov x26, %x[in]\n" - "ldr x25, [x26, #0x0]\n" - ".inst 0x25706141 // psel p1.s, p8.s/Z, p10.s[w12, #1]\n" - ".inst 0x25306d20 // psel p0.s, p11.s/Z, p9.s[w12]\n" - ".inst 0xe09606e9 // ld1w { za2h.s[x12, #1] }, p1/Z, [x23, x22, LSL #2]\n" - "ldr x23, [x26, #0x8]\n" - ".inst 0xe0bf82a0 // st1w { za0v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n" - ".inst 0x25706d20 // psel p0.s, p11.s/Z, p9.s[w12, #1]\n" - "whilelt p9.s, x27, %x[width]\n" - "incw x27\n" - "add x26, x26, #0x10\n" - ".inst 0xe0aa82a1 // st1w { za0v.s[x12, #1] }, p0/Z, [x21, x10, LSL #2]\n" - "addvl x21, x21, #2\n" - "incw x22\n" - "whilelt p8.s, x27, %x[width]\n" + "mov x22, %x[in]\n" + ".inst 0x25306140 // dup p0.s, p8.s/Z, p10.s[w12]\n" + ".inst 0xe09902a8 // ld1w { za2h.s[x12] }, p0/Z, [x21, x25, LSL #2]\n" + ".inst 0x25706140 // dup p0.s, p8.s/Z, p10.s[w12, #1]\n" + "ldr x21, [x22, #0x0]\n" + ".inst 0xe0990289 // ld1w { za2h.s[x12, #1] }, p0/Z, [x20, x25, LSL #2]\n" + "ldr x20, [x22, #0x8]\n" + "add x22, x22, #0x10\n" + ".inst 0x25306d20 // dup p0.s, p11.s/Z, p9.s[w12]\n" + ".inst 0xe0bf8300 // st1w { za0v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n" + ".inst 0x25706d20 // dup p0.s, p11.s/Z, p9.s[w12, #1]\n" + "whilelt p9.s, x23, %x[width]\n" + ".inst 0xe0aa8301 // st1w { za0v.s[x12, #1] }, p0/Z, [x24, x10, LSL #2]\n" + "addvl x24, x24, #2\n" + "incw x23\n" + "incw x25\n" + "whilelt p8.s, x23, %x[width]\n" "mov x12, #0x0\n" - "cbz x28, 7f\n" + "cbz x26, 7f\n" "6:" // K loop: Main loop: Second: Loop - ".inst 0x25306140 // psel p0.s, p8.s/Z, p10.s[w12]\n" - ".inst 0xe0960320 // ld1w { za0h.s[x12] }, p0/Z, [x25, x22, LSL #2]\n" - ".inst 0x25706141 // psel p1.s, p8.s/Z, p10.s[w12, #1]\n" - "ldr x25, [x26, #0x0]\n" - ".inst 0x25306d20 // psel p0.s, p11.s/Z, p9.s[w12]\n" - ".inst 0xe09606e1 // ld1w { za0h.s[x12, #1] }, p1/Z, [x23, x22, LSL #2]\n" - "ldr x23, [x26, #0x8]\n" - ".inst 0xe0bf82a8 // st1w { za2v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n" - ".inst 0x25706d20 // psel p0.s, p11.s/Z, p9.s[w12, #1]\n" - ".inst 0xe0aa82a9 // st1w { za2v.s[x12, #1] }, p0/Z, [x21, x10, LSL #2]\n" + ".inst 0x25306140 // dup p0.s, p8.s/Z, p10.s[w12]\n" + ".inst 0xe09902a0 // ld1w { za0h.s[x12] }, p0/Z, [x21, x25, LSL #2]\n" + ".inst 0x25706140 // dup p0.s, p8.s/Z, p10.s[w12, #1]\n" + "ldr x21, [x22, #0x0]\n" + ".inst 0xe0990281 // ld1w { za0h.s[x12, #1] }, p0/Z, [x20, x25, LSL #2]\n" + "ldr x20, [x22, #0x8]\n" + "add x22, x22, #0x10\n" + ".inst 0x25306d20 // dup p0.s, p11.s/Z, p9.s[w12]\n" + ".inst 0xe0bf8308 // st1w { za2v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n" + ".inst 0x25706d20 // dup p0.s, p11.s/Z, p9.s[w12, #1]\n" + ".inst 0xe0aa8309 // st1w { za2v.s[x12, #1] }, p0/Z, [x24, x10, LSL #2]\n" + "addvl x24, x24, #2\n" "add x12, x12, #0x2\n" - "cmp x12, x28\n" - "add x26, x26, #0x10\n" - "addvl x21, x21, #2\n" + "cmp x12, x26\n" "blt 6b\n" "7:" // K loop: Main loop: Second: Tail - ".inst 0x25306140 // psel p0.s, p8.s/Z, p10.s[w12]\n" - ".inst 0xe0960320 // ld1w { za0h.s[x12] }, p0/Z, [x25, x22, LSL #2]\n" - "mov x26, %x[in]\n" - "ldr x25, [x26, #0x0]\n" - ".inst 0x25706141 // psel p1.s, p8.s/Z, p10.s[w12, #1]\n" - ".inst 0x25306d20 // psel p0.s, p11.s/Z, p9.s[w12]\n" - ".inst 0xe09606e1 // ld1w { za0h.s[x12, #1] }, p1/Z, [x23, x22, LSL #2]\n" - "ldr x23, [x26, #0x8]\n" - ".inst 0xe0bf82a8 // st1w { za2v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n" - ".inst 0x25706d20 // psel p0.s, p11.s/Z, p9.s[w12, #1]\n" - "whilelt p9.s, x27, %x[width]\n" - "subs x20, x20, #0x1\n" - "add x26, x26, #0x10\n" - ".inst 0xe0aa82a9 // st1w { za2v.s[x12, #1] }, p0/Z, [x21, x10, LSL #2]\n" - "addvl x21, x21, #2\n" - "incw x27\n" - "incw x22\n" + "mov x22, %x[in]\n" + ".inst 0x25306140 // dup p0.s, p8.s/Z, p10.s[w12]\n" + ".inst 0xe09902a0 // ld1w { za0h.s[x12] }, p0/Z, [x21, x25, LSL #2]\n" + ".inst 0x25706140 // dup p0.s, p8.s/Z, p10.s[w12, #1]\n" + "ldr x21, [x22, #0x0]\n" + ".inst 0xe0990281 // ld1w { za0h.s[x12, #1] }, p0/Z, [x20, x25, LSL #2]\n" + "ldr x20, [x22, #0x8]\n" + "add x22, x22, #0x10\n" + ".inst 0x25306d20 // dup p0.s, p11.s/Z, p9.s[w12]\n" + ".inst 0xe0bf8308 // st1w { za2v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n" + ".inst 0x25706d20 // dup p0.s, p11.s/Z, p9.s[w12, #1]\n" + "whilelt p9.s, x23, %x[width]\n" + ".inst 0xe0aa8309 // st1w { za2v.s[x12, #1] }, p0/Z, [x24, x10, LSL #2]\n" + "addvl x24, x24, #2\n" + "incw x23\n" + "incw x25\n" + "subs x19, x19, #0x1\n" "bgt 3b\n" "8:" // K loop: Tails - "cbnz x24, 11f\n" - "mov x26, %x[in]\n" - "whilelt p8.s, x27, %x[width]\n" + "cbnz x28, 11f\n" + "mov x22, %x[in]\n" + "whilelt p8.s, x23, %x[width]\n" "mov x12, #0x0\n" "9:" // K loop: Tails: Even: First - ".inst 0x25306d20 // psel p0.s, p11.s/Z, p9.s[w12]\n" - ".inst 0xe0bf82a0 // st1w { za0v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n" - "ldr x25, [x26, #0x0]\n" - ".inst 0x25306140 // psel p0.s, p8.s/Z, p10.s[w12]\n" - ".inst 0xe0960328 // ld1w { za2h.s[x12] }, p0/Z, [x25, x22, LSL #2]\n" + ".inst 0x25306d20 // dup p0.s, p11.s/Z, p9.s[w12]\n" + ".inst 0xe0bf8300 // st1w { za0v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n" + ".inst 0x25306140 // dup p0.s, p8.s/Z, p10.s[w12]\n" + "addvl x24, x24, #1\n" + "ldr x21, [x22, #0x0]\n" + ".inst 0xe09902a8 // ld1w { za2h.s[x12] }, p0/Z, [x21, x25, LSL #2]\n" + "add x22, x22, #0x8\n" "add x12, x12, #0x1\n" "cmp x12, x10\n" - "add x26, x26, #0x8\n" - "addvl x21, x21, #1\n" "blt 9b\n" - "whilelt p9.s, x27, %x[width]\n" - "whilelt p8.s, x27, %x[width]\n" + "whilelt p9.s, x23, %x[width]\n" + "whilelt p8.s, x23, %x[width]\n" "mov x12, #0x0\n" "10:" // K loop: Tails: Even: Second - ".inst 0x25306d20 // psel p0.s, p11.s/Z, p9.s[w12]\n" - ".inst 0xe0bf82a8 // st1w { za2v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n" + ".inst 0x25306d20 // dup p0.s, p11.s/Z, p9.s[w12]\n" + ".inst 0xe0bf8308 // st1w { za2v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n" + "addvl x24, x24, #1\n" "add x12, x12, #0x1\n" - "cmp x12, x9\n" - "addvl x21, x21, #1\n" + "cmp x12, x27\n" "blt 10b\n" - "whilelt p9.s, x27, %x[width]\n" + "whilelt p9.s, x23, %x[width]\n" "b 13f\n" "11:" // K loop: Tails: Odd "mov x12, #0x0\n" "12:" // K loop: Tails: Odd: Loop - ".inst 0x25306d20 // psel p0.s, p11.s/Z, p9.s[w12]\n" - ".inst 0xe0bf82a0 // st1w { za0v.s[x12] }, p0/Z, [x21, XZR, LSL #2]\n" + ".inst 0x25306d20 // dup p0.s, p11.s/Z, p9.s[w12]\n" + ".inst 0xe0bf8300 // st1w { za0v.s[x12] }, p0/Z, [x24, XZR, LSL #2]\n" + "addvl x24, x24, #1\n" "add x12, x12, #0x1\n" - "cmp x12, x9\n" - "addvl x21, x21, #1\n" + "cmp x12, x27\n" "blt 12b\n" "13:" // K loop: End - "mov %x[out], x21\n" + "mov %x[out], x24\n" ".inst 0xd503467f // SMSTOP\n" : [out] "+&r" (out) : [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset), [width] "r" (width) - : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x9", "x10", "x12", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31" + : "cc", "memory", "p0", "p8", "p9", "p10", "p11", "x9", "x10", "x12", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31" ); } |