From e28cf395b57a091d0850cd28cecc81046153b843 Mon Sep 17 00:00:00 2001 From: Georgios Pinitas Date: Sun, 31 Jan 2021 05:18:43 +0000 Subject: Regenerate kernels and update A55 versions Signed-off-by: Georgios Pinitas Change-Id: I9eae76c77db03b8806af65729da34ab2d77f95f2 Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/4965 Tested-by: Arm Jenkins Reviewed-by: Michele Di Giorgio Comments-Addressed: Arm Jenkins --- .../a64_interleave4_block16_s8_s8_summing.hpp | 24 +++++++++++----------- 1 file changed, 12 insertions(+), 12 deletions(-) (limited to 'src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave4_block16_s8_s8_summing.hpp') diff --git a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave4_block16_s8_s8_summing.hpp b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave4_block16_s8_s8_summing.hpp index 1650916f9f..954a86656e 100644 --- a/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave4_block16_s8_s8_summing.hpp +++ b/src/core/NEON/kernels/arm_gemm/indirect-interleaves/a64_interleave4_block16_s8_s8_summing.hpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2019-2020 Arm Limited. + * Copyright (c) 2019-2021 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -84,24 +84,24 @@ void interleave_block<4, 16, VLType::None, true>( "mov x22, #0x0\n" "4:" // no_accumulate_16 "ldr q19, [x23], #0x10\n" - "prfm pldl1keep, [x23, #0x70]\n" + "add x22, x22, #0x1\n" "ldr q18, [x21], #0x10\n" + "subs %x[width], %x[width], #0x10\n" "ldr q17, [x20], #0x10\n" - "prfm pldl1keep, [x21, #0x70]\n" + "cmp %x[width], #0x10\n" "ldr q16, [x19], #0x10\n" - "prfm pldl1keep, [x20, #0x70]\n" - "str q19, [%x[out_ptr], #0x0]\n" "sadalp v28.8h, v19.16b\n" + "prfm pldl1keep, [x23, #0x70]\n" + "prfm pldl1keep, [x21, #0x70]\n" + "sadalp v27.8h, v18.16b\n" + "prfm pldl1keep, [x20, #0x70]\n" + "sadalp v26.8h, v17.16b\n" "prfm pldl1keep, [x19, #0x70]\n" + "sadalp v25.8h, v16.16b\n" + "str q19, [%x[out_ptr], #0x0]\n" "str q18, [%x[out_ptr], #0x10]\n" - "sadalp v27.8h, v18.16b\n" "str q17, [%x[out_ptr], #0x20]\n" - "sadalp v26.8h, v17.16b\n" "str q16, [%x[out_ptr], #0x30]\n" - "sadalp v25.8h, v16.16b\n" - "add x22, x22, #0x1\n" - "subs %x[width], %x[width], #0x10\n" - "cmp %x[width], #0x10\n" "add %x[out_ptr], %x[out_ptr], #0x40\n" "bge 3b\n" "5:" // Main loop skip @@ -215,7 +215,7 @@ void interleave_block<4, 16, VLType::None, true>( "add v24.4s, v24.4s, v20.4s\n" "str q24, [%x[out_ptr], #0x0]\n" "add %x[out_ptr], %x[out_ptr], #0x10\n" - : [out_ptr] "+r" (out_ptr), [width] "+r" (width) + : [out_ptr] "+&r" (out_ptr), [width] "+&r" (width) : [first] "r" (first), [height] "r" (height), [in] "r" (in), [row_offset] "r" (row_offset) : "cc", "memory", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "x19", "x20", "x21", "x22", "x23" ); -- cgit v1.2.1