[Clang][SME2] Enable multi-vector loads & stores for SME2 (#75821)
[llvm-project.git] / compiler-rt / lib / builtins / popcountti2.c
blob79cbb2fb34c003bda0678ed01a9ac390bd00625a
1 //===-- popcountti2.c - Implement __popcountti2
2 //----------------------------===//
3 //
4 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
5 // See https://llvm.org/LICENSE.txt for license information.
6 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
7 //
8 //===----------------------------------------------------------------------===//
9 //
10 // This file implements __popcountti2 for the compiler_rt library.
12 //===----------------------------------------------------------------------===//
14 #include "int_lib.h"
16 #ifdef CRT_HAS_128BIT
18 // Returns: count of 1 bits
20 COMPILER_RT_ABI int __popcountti2(ti_int a) {
21 tu_int x3 = (tu_int)a;
22 x3 = x3 - ((x3 >> 1) &
23 (((tu_int)0x5555555555555555uLL << 64) | 0x5555555555555555uLL));
24 // Every 2 bits holds the sum of every pair of bits (64)
25 x3 = ((x3 >> 2) &
26 (((tu_int)0x3333333333333333uLL << 64) | 0x3333333333333333uLL)) +
27 (x3 & (((tu_int)0x3333333333333333uLL << 64) | 0x3333333333333333uLL));
28 // Every 4 bits holds the sum of every 4-set of bits (3 significant bits) (32)
29 x3 = (x3 + (x3 >> 4)) &
30 (((tu_int)0x0F0F0F0F0F0F0F0FuLL << 64) | 0x0F0F0F0F0F0F0F0FuLL);
31 // Every 8 bits holds the sum of every 8-set of bits (4 significant bits) (16)
32 du_int x2 = (du_int)(x3 + (x3 >> 64));
33 // Every 8 bits holds the sum of every 8-set of bits (5 significant bits) (8)
34 su_int x = (su_int)(x2 + (x2 >> 32));
35 // Every 8 bits holds the sum of every 8-set of bits (6 significant bits) (4)
36 x = x + (x >> 16);
37 // Every 8 bits holds the sum of every 8-set of bits (7 significant bits) (2)
39 // Upper 16 bits are garbage
40 return (x + (x >> 8)) & 0xFF; // (8 significant bits)
43 #endif // CRT_HAS_128BIT