You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
83 lines
3.4 KiB
83 lines
3.4 KiB
/////////////////////////////////////////////////////////////////////////////////// |
|
// Copyright (C) 2026 Jon Beniston, M7RCE <jon@beniston.com> // |
|
// // |
|
// This program is free software; you can redistribute it and/or modify // |
|
// it under the terms of the GNU General Public License as published by // |
|
// the Free Software Foundation as version 3 of the License, or // |
|
// (at your option) any later version. // |
|
// // |
|
// This program is distributed in the hope that it will be useful, // |
|
// but WITHOUT ANY WARRANTY; without even the implied warranty of // |
|
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the // |
|
// GNU General Public License V3 for more details. // |
|
// // |
|
// You should have received a copy of the GNU General Public License // |
|
// along with this program. If not, see <http://www.gnu.org/licenses/>. // |
|
/////////////////////////////////////////////////////////////////////////////////// |
|
|
|
// LDPC decoder for 16 codewords at a time, with ARM NEON (always present on AArch64; on 32-bit ARM, when the |
|
// compiler targets it) |
|
|
|
#include "dvbt2ldpc.h" |
|
|
|
#if defined(DVBT2_LDPC_NEON) |
|
|
|
#include <arm_neon.h> |
|
|
|
#include "dvbt2ldpckernel.h" |
|
|
|
namespace dvbt2 { |
|
namespace { |
|
|
|
struct VectorNEON |
|
{ |
|
typedef int8x16_t T; |
|
static const int W = 16; |
|
|
|
static T zero() { return vdupq_n_s8(0); } |
|
static T set1(char x) { return vdupq_n_s8((int8_t) x); } |
|
static T adds(T a, T b) { return vqaddq_s8(a, b); } |
|
static T subs(T a, T b) { return vqsubq_s8(a, b); } |
|
static T addsu(T a, T b) { return vreinterpretq_s8_u8(vqaddq_u8(vreinterpretq_u8_s8(a), vreinterpretq_u8_s8(b))); } |
|
static T abs(T a) { return vabsq_s8(a); } |
|
static T min(T a, T b) { return vminq_s8(a, b); } |
|
static T max(T a, T b) { return vmaxq_s8(a, b); } |
|
static T cmpgt(T a, T b) { return vreinterpretq_s8_u8(vcgtq_s8(a, b)); } |
|
|
|
// As SSSE3's _mm_sign_epi8: a, negated where b is negative, zero where b is zero |
|
static T sign(T a, T b) |
|
{ |
|
const uint8x16_t negative = vcltq_s8(b, vdupq_n_s8(0)); |
|
const uint8x16_t isZero = vceqq_s8(b, vdupq_n_s8(0)); |
|
const T r = vbslq_s8(negative, vnegq_s8(a), a); |
|
return vbicq_s8(r, vreinterpretq_s8_u8(isZero)); |
|
} |
|
|
|
static T xor_(T a, T b) { return veorq_s8(a, b); } |
|
static T and_(T a, T b) { return vandq_s8(a, b); } |
|
static T or_(T a, T b) { return vorrq_s8(a, b); } |
|
static void store(uint8_t *p, T a) { vst1q_u8(p, vreinterpretq_u8_s8(a)); } |
|
|
|
// The sign bit of each lane, as SSE2's _mm_movemask_epi8. NEON has no equivalent, but this is only used now and |
|
// then, to see if decoding has finished |
|
static uint64_t movemask(T a) |
|
{ |
|
uint8_t bits[16]; |
|
vst1q_u8(bits, vshrq_n_u8(vreinterpretq_u8_s8(a), 7)); |
|
uint64_t mask = 0; |
|
|
|
for (int i = 0; i < 16; i++) { |
|
mask |= (uint64_t) bits[i] << i; |
|
} |
|
|
|
return mask; |
|
} |
|
}; |
|
|
|
} // namespace |
|
|
|
const LDPCKernel ldpcKernelNEON = {VectorNEON::W, decodeLDPC<VectorNEON>}; |
|
|
|
} // namespace dvbt2 |
|
|
|
#endif // DVBT2_LDPC_NEON
|
|
|