mirror of
https://github.com/ARMSX2/ARMSX2.git
synced 2026-08-24 16:50:16 -07:00
Brian Degenhardt
43e3430b61
spu2_sve2_fir.h offered SVE2 versions of the reverb FIR behind SPU2_HAS_SVE2_COMPILER, which is off on every target we build. It has never been compiled by anyone, and it would not compile if tried: the upsample coefficient table declares 32768 in an int16_t initializer, which is a narrowing error, not a warning. Clang rejects it outright. The arithmetic is wrong too. ReverbDownsample_reference accumulates the products and then does out >>= 15; the SSE, AVX and NEON paths get that scale implicitly from mulhrs / vqrdmulhq_s16. The SVE2 version accumulates with svmlalb/svmlalt and hands the raw sum to clamp_mix with no shift at all, so every sample would saturate. The same coefficient the initializer rejects is one the reference clamps to 32767 in make_up_coefs, so the table was wrong on its own terms. That leaves spu2_mt6899_tuning.h unreferenced: its only consumer anywhere was GetFeatures() from inside the SVE2 block. It held Cortex-X925 cache geometry, prefetch distances and thread-pinning helpers for a device that is not one of our targets. RegisterNEONBackend now installs the NEON FIR unconditionally, which is what it already did in every build that exists.
135 lines
4.5 KiB
C++
135 lines
4.5 KiB
C++
// SPDX-FileCopyrightText: 2002-2026 PCSX2 Dev Team
|
|
// SPDX-License-Identifier: GPL-3.0+
|
|
//
|
|
// SPU2/spu2_neon.cpp — ARM64 backend registration for SPU2.
|
|
//
|
|
// Registers NEON (and optionally SVE2) optimized implementations by
|
|
// overriding the global function pointers at runtime.
|
|
// Called once from SPU2::Open() after Multi-ISA default init.
|
|
//
|
|
// Only compiled on ARM64 targets. On x86, this file produces no object code.
|
|
|
|
#if defined(__aarch64__) || defined(_M_ARM64)
|
|
|
|
#include "SPU2/spu2_neon.h"
|
|
// defs.h provides V_Core / StereoOut32 / clamp_mix and the
|
|
// ReverbDownsample/ReverbUpsample function pointers.
|
|
#include "SPU2/defs.h"
|
|
|
|
#include <arm_neon.h>
|
|
#include <array>
|
|
#include <algorithm>
|
|
#include <cstdint>
|
|
#include <cstdio>
|
|
|
|
// ============================================================================
|
|
// Reverb FIR — NEON implementations (from original spu2_neon.cpp)
|
|
// ============================================================================
|
|
|
|
static constexpr int NEON_NUM_TAPS = 39;
|
|
|
|
static constexpr std::array<int16_t, 48> neon_down_coefs alignas(16) = {
|
|
-1, 0, 2, 0, -10, 0, 35, 0,
|
|
-103, 0, 266, 0, -616, 0, 1332, 0,
|
|
-2960, 0, 10246, 16384, 10246, 0, -2960, 0,
|
|
1332, 0, -616, 0, 266, 0, -103, 0,
|
|
35, 0, -10, 0, 2, 0, -1,
|
|
};
|
|
|
|
static constexpr std::array<int16_t, 48> make_neon_up_coefs()
|
|
{
|
|
std::array<int16_t, 48> ret = {};
|
|
for (int i = 0; i < NEON_NUM_TAPS; i++)
|
|
{
|
|
ret[i] = static_cast<int16_t>(
|
|
std::clamp<int32_t>(neon_down_coefs[i] * 2, INT16_MIN, INT16_MAX));
|
|
}
|
|
return ret;
|
|
}
|
|
|
|
static constexpr std::array<int16_t, 48> neon_up_coefs alignas(16) = make_neon_up_coefs();
|
|
|
|
// NEON ReverbDownsample — 39-tap FIR
|
|
static int32_t ReverbDownsample_neon(V_Core& core, bool right)
|
|
{
|
|
const int index = (core.RevbSampleBufPos - NEON_NUM_TAPS) & 63;
|
|
int16x8_t acc = vdupq_n_s16(0);
|
|
|
|
int16x8_t coef, samp;
|
|
|
|
coef = vld1q_s16(&neon_down_coefs[0]);
|
|
samp = vld1q_s16(&core.RevbDownBuf[right][index]);
|
|
acc = vqaddq_s16(acc, vqrdmulhq_s16(samp, coef));
|
|
|
|
coef = vld1q_s16(&neon_down_coefs[8]);
|
|
samp = vld1q_s16(&core.RevbDownBuf[right][index + 8]);
|
|
acc = vqaddq_s16(acc, vqrdmulhq_s16(samp, coef));
|
|
|
|
coef = vld1q_s16(&neon_down_coefs[16]);
|
|
samp = vld1q_s16(&core.RevbDownBuf[right][index + 16]);
|
|
acc = vqaddq_s16(acc, vqrdmulhq_s16(samp, coef));
|
|
|
|
coef = vld1q_s16(&neon_down_coefs[24]);
|
|
samp = vld1q_s16(&core.RevbDownBuf[right][index + 24]);
|
|
acc = vqaddq_s16(acc, vqrdmulhq_s16(samp, coef));
|
|
|
|
coef = vld1q_s16(&neon_down_coefs[32]);
|
|
samp = vld1q_s16(&core.RevbDownBuf[right][index + 32]);
|
|
acc = vqaddq_s16(acc, vqrdmulhq_s16(samp, coef));
|
|
|
|
int32x4_t sum32 = vpaddlq_s16(acc);
|
|
int32x2_t pair = vadd_s32(vget_low_s32(sum32), vget_high_s32(sum32));
|
|
int32_t sum = vget_lane_s32(pair, 0) + vget_lane_s32(pair, 1);
|
|
|
|
return clamp_mix(sum);
|
|
}
|
|
|
|
// NEON ReverbUpsample — 39-tap FIR, L/R channels
|
|
static StereoOut32 ReverbUpsample_neon(V_Core& core)
|
|
{
|
|
const int index = (core.RevbSampleBufPos - NEON_NUM_TAPS) & 63;
|
|
int16x8_t l_acc = vdupq_n_s16(0);
|
|
int16x8_t r_acc = vdupq_n_s16(0);
|
|
|
|
struct { int offset; } groups[] = {{0}, {8}, {16}, {24}, {32}};
|
|
|
|
for (auto& g : groups)
|
|
{
|
|
int16x8_t coef = vld1q_s16(&neon_up_coefs[g.offset]);
|
|
int16x8_t l_s = vld1q_s16(&core.RevbUpBuf[0][index + g.offset]);
|
|
int16x8_t r_s = vld1q_s16(&core.RevbUpBuf[1][index + g.offset]);
|
|
l_acc = vqaddq_s16(l_acc, vqrdmulhq_s16(l_s, coef));
|
|
r_acc = vqaddq_s16(r_acc, vqrdmulhq_s16(r_s, coef));
|
|
}
|
|
|
|
int32x4_t l_s32 = vpaddlq_s16(l_acc);
|
|
int32x2_t l_p = vadd_s32(vget_low_s32(l_s32), vget_high_s32(l_s32));
|
|
int32_t l = vget_lane_s32(l_p, 0) + vget_lane_s32(l_p, 1);
|
|
|
|
int32x4_t r_s32 = vpaddlq_s16(r_acc);
|
|
int32x2_t r_p = vadd_s32(vget_low_s32(r_s32), vget_high_s32(r_s32));
|
|
int32_t r = vget_lane_s32(r_p, 0) + vget_lane_s32(r_p, 1);
|
|
|
|
return {clamp_mix(l), clamp_mix(r)};
|
|
}
|
|
|
|
// ============================================================================
|
|
// Registration — override global function pointers
|
|
// ============================================================================
|
|
//
|
|
// NOTE (ARMSX2): the original contributor build also pinned the calling thread
|
|
// to a fixed "prime" core (sched_setaffinity to CPU 0-3) and forced SCHED_FIFO.
|
|
// That is dropped here: this runs on the VM/EE thread (not the Oboe audio
|
|
// thread), the CPU-topology assumption is device-specific, and ARMSX2 already
|
|
// manages audio-thread affinity. Only the FIR pointer override is kept.
|
|
|
|
namespace SPU2
|
|
{
|
|
void RegisterNEONBackend()
|
|
{
|
|
ReverbDownsample = ReverbDownsample_neon;
|
|
ReverbUpsample = ReverbUpsample_neon;
|
|
}
|
|
} // namespace SPU2
|
|
|
|
#endif // __aarch64__ || _M_ARM64
|