From ba209750abc1ac7e42bab9fef5db284384d70fb3 Mon Sep 17 00:00:00 2001 From: Michael Tyler Date: Thu, 15 Dec 2022 12:39:29 +0000 Subject: Update CPU kernels to remove x19 Resolves: COMPMID-5805 Signed-off-by: Michael Tyler Change-Id: I250f64531e209625e4ff176dd5a552c1c34bc484 Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8909 Tested-by: Arm Jenkins Comments-Addressed: Arm Jenkins Reviewed-by: Gunes Bayir Reviewed-by: Viet-Hoa Do Benchmark: Arm Jenkins --- .../transforms/sme_transpose_interleave_16VL.hpp | 162 ++++++++++----------- 1 file changed, 81 insertions(+), 81 deletions(-) (limited to 'src/core/NEON/kernels/arm_gemm/transforms/sme_transpose_interleave_16VL.hpp') diff --git a/src/core/NEON/kernels/arm_gemm/transforms/sme_transpose_interleave_16VL.hpp b/src/core/NEON/kernels/arm_gemm/transforms/sme_transpose_interleave_16VL.hpp index 3475d30abc..01921c5ad9 100644 --- a/src/core/NEON/kernels/arm_gemm/transforms/sme_transpose_interleave_16VL.hpp +++ b/src/core/NEON/kernels/arm_gemm/transforms/sme_transpose_interleave_16VL.hpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2022 Arm Limited. + * Copyright (c) 2022-2023 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. */ #pragma once @@ -37,82 +37,82 @@ void sme_transpose_interleave_16VL(uint32_t *out, const uint32_t *in, size_t wid ".inst 0xd503477f // SMSTART ZA\n" "ptrue p7.b\n" "1:" // Main row loop: Head - "mov x22, %x[in]\n" - "add %x[in], x22, %x[in_stride]\n" - "mov x21, %x[out]\n" + "mov x23, %x[in]\n" + "add %x[in], x23, %x[in_stride]\n" + "mov x22, %x[out]\n" "sub %x[height], %x[height], #0x1\n" - "mov x20, %x[width]\n" + "mov x21, %x[width]\n" "2:" // Main row loop: Column loop - "mov x19, x20\n" - "whilelt p0.s, XZR, x19\n" - "ld1w { z31.s }, p0/Z, [x22]\n" - "decw x19\n" - "whilelt p0.s, XZR, x19\n" - "ld1w { z30.s }, p0/Z, [x22, #1, MUL VL]\n" - "decw x19\n" - "whilelt p0.s, XZR, x19\n" - "ld1w { z29.s }, p0/Z, [x22, #2, MUL VL]\n" - "decw x19\n" - "whilelt p0.s, XZR, x19\n" - "ld1w { z28.s }, p0/Z, [x22, #3, MUL VL]\n" - "decw x19\n" - "whilelt p0.s, XZR, x19\n" - "ld1w { z27.s }, p0/Z, [x22, #4, MUL VL]\n" - "decw x19\n" - "whilelt p0.s, XZR, x19\n" - "ld1w { z26.s }, p0/Z, [x22, #5, MUL VL]\n" - "decw x19\n" - "whilelt p0.s, XZR, x19\n" - "ld1w { z25.s }, p0/Z, [x22, #6, MUL VL]\n" - "decw x19\n" - "whilelt p0.s, XZR, x19\n" - "ld1w { z24.s }, p0/Z, [x22, #7, MUL VL]\n" - "decw x19\n" - "whilelt p0.s, XZR, x19\n" - "decw x19\n" - "whilelt p6.s, XZR, x19\n" - "decw x19\n" - "whilelt p5.s, XZR, x19\n" - "decw x19\n" - "whilelt p4.s, XZR, x19\n" - "decw x19\n" - "whilelt p3.s, XZR, x19\n" - "decw x19\n" - "whilelt p2.s, XZR, x19\n" - "decw x19\n" - "whilelt p1.s, XZR, x19\n" - "decw x19\n" - "addvl x22, x22, #16\n" - "ld1w { z23.s }, p0/Z, [x22, #-8, MUL VL]\n" - "whilelt p0.s, XZR, x19\n" - "mov x19, x21\n" - "ld1w { z22.s }, p6/Z, [x22, #-7, MUL VL]\n" - "decw x20, ALL, MUL #16\n" - "ld1w { z21.s }, p5/Z, [x22, #-6, MUL VL]\n" - "cmp x20, #0x0\n" - "ld1w { z20.s }, p4/Z, [x22, #-5, MUL VL]\n" - "add x21, x21, %x[out_stride]\n" - "ld1w { z19.s }, p3/Z, [x22, #-4, MUL VL]\n" - "ld1w { z18.s }, p2/Z, [x22, #-3, MUL VL]\n" - "ld1w { z17.s }, p1/Z, [x22, #-2, MUL VL]\n" - "ld1w { z16.s }, p0/Z, [x22, #-1, MUL VL]\n" - "st1w { z31.s }, p7, [x19]\n" - "st1w { z30.s }, p7, [x19, #1, MUL VL]\n" - "st1w { z29.s }, p7, [x19, #2, MUL VL]\n" - "st1w { z28.s }, p7, [x19, #3, MUL VL]\n" - "st1w { z27.s }, p7, [x19, #4, MUL VL]\n" - "st1w { z26.s }, p7, [x19, #5, MUL VL]\n" - "st1w { z25.s }, p7, [x19, #6, MUL VL]\n" - "st1w { z24.s }, p7, [x19, #7, MUL VL]\n" - "addvl x19, x19, #16\n" - "st1w { z23.s }, p7, [x19, #-8, MUL VL]\n" - "st1w { z22.s }, p7, [x19, #-7, MUL VL]\n" - "st1w { z21.s }, p7, [x19, #-6, MUL VL]\n" - "st1w { z20.s }, p7, [x19, #-5, MUL VL]\n" - "st1w { z19.s }, p7, [x19, #-4, MUL VL]\n" - "st1w { z18.s }, p7, [x19, #-3, MUL VL]\n" - "st1w { z17.s }, p7, [x19, #-2, MUL VL]\n" - "st1w { z16.s }, p7, [x19, #-1, MUL VL]\n" + "mov x20, x21\n" + "whilelt p0.s, XZR, x20\n" + "ld1w { z31.s }, p0/Z, [x23]\n" + "decw x20\n" + "whilelt p0.s, XZR, x20\n" + "ld1w { z30.s }, p0/Z, [x23, #1, MUL VL]\n" + "decw x20\n" + "whilelt p0.s, XZR, x20\n" + "ld1w { z29.s }, p0/Z, [x23, #2, MUL VL]\n" + "decw x20\n" + "whilelt p0.s, XZR, x20\n" + "ld1w { z28.s }, p0/Z, [x23, #3, MUL VL]\n" + "decw x20\n" + "whilelt p0.s, XZR, x20\n" + "ld1w { z27.s }, p0/Z, [x23, #4, MUL VL]\n" + "decw x20\n" + "whilelt p0.s, XZR, x20\n" + "ld1w { z26.s }, p0/Z, [x23, #5, MUL VL]\n" + "decw x20\n" + "whilelt p0.s, XZR, x20\n" + "ld1w { z25.s }, p0/Z, [x23, #6, MUL VL]\n" + "decw x20\n" + "whilelt p0.s, XZR, x20\n" + "ld1w { z24.s }, p0/Z, [x23, #7, MUL VL]\n" + "decw x20\n" + "whilelt p0.s, XZR, x20\n" + "decw x20\n" + "whilelt p6.s, XZR, x20\n" + "decw x20\n" + "whilelt p5.s, XZR, x20\n" + "decw x20\n" + "whilelt p4.s, XZR, x20\n" + "decw x20\n" + "whilelt p3.s, XZR, x20\n" + "decw x20\n" + "whilelt p2.s, XZR, x20\n" + "decw x20\n" + "whilelt p1.s, XZR, x20\n" + "decw x20\n" + "addvl x23, x23, #16\n" + "ld1w { z23.s }, p0/Z, [x23, #-8, MUL VL]\n" + "whilelt p0.s, XZR, x20\n" + "mov x20, x22\n" + "ld1w { z22.s }, p6/Z, [x23, #-7, MUL VL]\n" + "decw x21, ALL, MUL #16\n" + "ld1w { z21.s }, p5/Z, [x23, #-6, MUL VL]\n" + "cmp x21, #0x0\n" + "ld1w { z20.s }, p4/Z, [x23, #-5, MUL VL]\n" + "add x22, x22, %x[out_stride]\n" + "ld1w { z19.s }, p3/Z, [x23, #-4, MUL VL]\n" + "ld1w { z18.s }, p2/Z, [x23, #-3, MUL VL]\n" + "ld1w { z17.s }, p1/Z, [x23, #-2, MUL VL]\n" + "ld1w { z16.s }, p0/Z, [x23, #-1, MUL VL]\n" + "st1w { z31.s }, p7, [x20]\n" + "st1w { z30.s }, p7, [x20, #1, MUL VL]\n" + "st1w { z29.s }, p7, [x20, #2, MUL VL]\n" + "st1w { z28.s }, p7, [x20, #3, MUL VL]\n" + "st1w { z27.s }, p7, [x20, #4, MUL VL]\n" + "st1w { z26.s }, p7, [x20, #5, MUL VL]\n" + "st1w { z25.s }, p7, [x20, #6, MUL VL]\n" + "st1w { z24.s }, p7, [x20, #7, MUL VL]\n" + "addvl x20, x20, #16\n" + "st1w { z23.s }, p7, [x20, #-8, MUL VL]\n" + "st1w { z22.s }, p7, [x20, #-7, MUL VL]\n" + "st1w { z21.s }, p7, [x20, #-6, MUL VL]\n" + "st1w { z20.s }, p7, [x20, #-5, MUL VL]\n" + "st1w { z19.s }, p7, [x20, #-4, MUL VL]\n" + "st1w { z18.s }, p7, [x20, #-3, MUL VL]\n" + "st1w { z17.s }, p7, [x20, #-2, MUL VL]\n" + "st1w { z16.s }, p7, [x20, #-1, MUL VL]\n" "bgt 2b\n" "3:" // Main row loop: Column loop skip "cmp %x[height], #0x1\n" @@ -121,7 +121,7 @@ void sme_transpose_interleave_16VL(uint32_t *out, const uint32_t *in, size_t wid ".inst 0xd503467f // SMSTOP\n" : [height] "+&r" (height), [in] "+&r" (in), [out] "+&r" (out) : [in_stride] "r" (in_stride), [out_stride] "r" (out_stride), [width] "r" (width) - : "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x19", "x20", "x21", "x22", "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", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x20", "x21", "x22", "x23", "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" ); } -- cgit v1.2.1