/*
  This file is part of Ingen.
  Copyright 2007-2016 David Robillard <http://drobilla.net/>

  Ingen is free software: you can redistribute it and/or modify it under the
  terms of the GNU Affero General Public License as published by the Free
  Software Foundation, either version 3 of the License, or any later version.

  Ingen 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 Affero General Public License for details.

  You should have received a copy of the GNU Affero General Public License
  along with Ingen.  If not, see <http://www.gnu.org/licenses/>.
*/

#include "Buffer.hpp"

#include "BufferFactory.hpp"
#include "Engine.hpp"
#include "RunContext.hpp"

#include "ingen/Atom.hpp"
#include "ingen/Log.hpp"
#include "ingen/URIs.hpp"
#include "ingen_config.h"
#include "lv2/atom/atom.h"
#include "lv2/atom/util.h"

#include <algorithm>
#include <cstdint>
#include <cstdio>
#include <cstdlib>
#include <cstring>
#include <new>

#ifdef __SSE__
#    include <xmmintrin.h>
#endif

namespace ingen {
namespace server {

Buffer::Buffer(BufferFactory& bufs,
               LV2_URID       type,
               LV2_URID       value_type,
               uint32_t       capacity,
               bool           external,
               void*)
	: _factory(bufs)
	, _next(nullptr)
	, _buf(external ? nullptr : aligned_alloc(capacity))
	, _latest_event(0)
	, _type(type)
	, _value_type(value_type)
	, _capacity(capacity)
	, _refs(0)
	, _external(external)
{
	if (!external && !_buf) {
		bufs.engine().log().rt_error("Failed to allocate buffer\n");
		throw std::bad_alloc();
	}

	if (type != bufs.uris().atom_Sound) {
		/* Audio buffers are not atoms, the buffer is the start of a float
		   array which is already silent since the buffer is zeroed.  All other
		   buffers are atoms. */
		if (_buf) {
			auto* atom = get<LV2_Atom>();
			atom->size = capacity - sizeof(LV2_Atom);
			atom->type = type;

			clear();
		}

		if (value_type && value_type != type) {
			/* Buffer with a different value type.  These buffers (probably
			   sequences) have a "value" that persists independently of the buffer
			   contents.  This is used to represent things like a Sequence of
			   Float, which acts like an individual float (has a value), but the
			   buffer itself only transmits changes and does not necessarily
			   contain the current value. */
			_value_buffer = bufs.get_buffer(value_type, 0, 0);
		}
	}
}

Buffer::~Buffer()
{
	if (!_external) {
		free(_buf);
	}
}

void
Buffer::recycle()
{
	_factory.recycle(this);
}

void
Buffer::set_type(GetFn get, LV2_URID type, LV2_URID value_type)
{
	_type       = type;
	_value_type = value_type;
	if (type == _factory.uris().atom_Sequence && value_type) {
		_value_buffer = (_factory.*get)(value_type, 0, 0);
	}
}

void
Buffer::clear()
{
	if (is_audio() && _buf) {
		memset(_buf, 0, _capacity);
	} else if (is_control()) {
		get<LV2_Atom_Float>()->body = 0;
	} else if (is_sequence()) {
		auto* seq = get<LV2_Atom_Sequence>();
		seq->atom.type = _factory.uris().atom_Sequence;
		seq->atom.size = sizeof(LV2_Atom_Sequence_Body);
		seq->body.unit = 0;
		seq->body.pad  = 0;
		_latest_event  = 0;
	}
}

void
Buffer::render_sequence(const RunContext& ctx, const Buffer* src, bool add)
{
	const LV2_URID atom_Float = _factory.uris().atom_Float;
	const auto*    seq        = src->get<const LV2_Atom_Sequence>();
	const auto*    init       = reinterpret_cast<const LV2_Atom_Float*>(src->value());
	float          value      = init ? init->body : 0.0f;
	SampleCount    offset     = ctx.offset();

	LV2_ATOM_SEQUENCE_FOREACH(seq, ev) {
		if (ev->time.frames >= offset && ev->body.type == atom_Float) {
			write_block(value, offset, ev->time.frames, add);
			value  = reinterpret_cast<const LV2_Atom_Float*>(&ev->body)->body;
			offset = ev->time.frames;
		}
	}
	write_block(value, offset, ctx.offset() + ctx.nframes(), add);
}

void
Buffer::copy(const RunContext& ctx, const Buffer* src)
{
	if (!_buf) {
		return;
	} else if (_type == src->type()) {
		const uint32_t src_size = src->size();
		if (src_size <= _capacity) {
			memcpy(_buf, src->_buf, src_size);
		} else {
			clear();
		}
	} else if (src->is_audio() && is_control()) {
		samples()[0] = src->samples()[0];
	} else if (src->is_control() && is_audio()) {
		set_block(src->samples()[0], 0, ctx.nframes());
	} else if (src->is_sequence() && is_audio() &&
	           src->value_type() == _factory.uris().atom_Float) {
		render_sequence(ctx, src, false);
	} else {
		clear();
	}
}

void
Buffer::resize(uint32_t capacity)
{
	if (!_external) {
		_buf      = realloc(_buf, capacity);
		_capacity = capacity;
		clear();
	} else {
		_factory.engine().log().error("Attempt to resize external buffer\n");
	}
}

void*
Buffer::port_data(PortType port_type, SampleCount offset)
{
	switch (port_type.id()) {
	case PortType::ID::CONTROL:
		return &_value_buffer->get<LV2_Atom_Float>()->body;
	case PortType::ID::CV:
	case PortType::ID::AUDIO:
		if (_type == _factory.uris().atom_Float) {
			return &get<LV2_Atom_Float>()->body;
		} else if (_type == _factory.uris().atom_Sound) {
			return static_cast<Sample*>(_buf) + offset;
		}
		break;
	case PortType::ID::ATOM:
		if (_type != _factory.uris().atom_Sound) {
			return _buf;
		}
	default: break;
	}
	return nullptr;
}

const void*
Buffer::port_data(PortType port_type, SampleCount offset) const
{
	return const_cast<void*>(
		const_cast<Buffer*>(this)->port_data(port_type, offset));
}

#ifdef __SSE__
/** Vector fabsf */
static inline __m128
mm_abs_ps(__m128 x)
{
	const __m128 sign_mask = _mm_set1_ps(-0.0f);  // -0.0f = 1 << 31
	return _mm_andnot_ps(sign_mask, x);
}
#endif

float
Buffer::peak(const RunContext& ctx) const
{
#ifdef __SSE__
	const auto* const vbuf    = reinterpret_cast<const __m128*>(samples());
	__m128            vpeak   = mm_abs_ps(vbuf[0]);
	const SampleCount nblocks = ctx.nframes() / 4;

	// First, find the vector absolute max of the buffer
	for (SampleCount i = 1; i < nblocks; ++i) {
		vpeak = _mm_max_ps(vpeak, mm_abs_ps(vbuf[i]));
	}

	// Now we need the single max of vpeak
	// vpeak = ABCD
	// tmp   = CDAB
	auto tmp = _mm_shuffle_ps(vpeak, vpeak, _MM_SHUFFLE(2, 3, 0, 1));

	// vpeak = MAX(A,C) MAX(B,D) MAX(C,A) MAX(D,B)
	vpeak = _mm_max_ps(vpeak, tmp);

	// tmp = BADC of the new vpeak
	// tmp = MAX(B,D) MAX(A,C) MAX(D,B) MAX(C,A)
	tmp = _mm_shuffle_ps(vpeak, vpeak, _MM_SHUFFLE(1, 0, 3, 2));

	// vpeak = MAX(MAX(A,C), MAX(B,D)), ...
	vpeak = _mm_max_ps(vpeak, tmp);

	// peak = vpeak[0]
	float peak = 0.0f;
	_mm_store_ss(&peak, vpeak);

	return peak;
#else
	const Sample* const buf = samples();
	float peak = 0.0f;
	for (SampleCount i = 0; i < context.nframes(); ++i) {
		peak = fmaxf(peak, fabsf(buf[i]));
	}
	return peak;
#endif
}

void
Buffer::prepare_write(RunContext&)
{
	if (_type == _factory.uris().atom_Sequence) {
		auto* atom = get<LV2_Atom>();

		atom->type    = static_cast<LV2_URID>(_factory.uris().atom_Sequence);
		atom->size    = sizeof(LV2_Atom_Sequence_Body);
		_latest_event = 0;
	}
}

void
Buffer::prepare_output_write(RunContext&)
{
	if (_type == _factory.uris().atom_Sequence) {
		auto* atom = get<LV2_Atom>();

		atom->type    = static_cast<LV2_URID>(_factory.uris().atom_Chunk);
		atom->size    = _capacity - sizeof(LV2_Atom);
		_latest_event = 0;
	}
}

bool
Buffer::append_event(int64_t        frames,
                     uint32_t       size,
                     uint32_t       type,
                     const uint8_t* data)
{
	assert(frames >= _latest_event);

	auto* atom = get<LV2_Atom>();
	if (atom->type == _factory.uris().atom_Chunk) {
		clear();  // Chunk initialized with prepare_output_write(), clear
	}

	if (sizeof(LV2_Atom) + atom->size + lv2_atom_pad_size(size) > _capacity) {
		return false;
	}

	auto* seq = reinterpret_cast<LV2_Atom_Sequence*>(atom);
	auto* ev  = reinterpret_cast<LV2_Atom_Event*>(
        reinterpret_cast<uint8_t*>(seq) + lv2_atom_total_size(&seq->atom));

	ev->time.frames = frames;
	ev->body.size   = size;
	ev->body.type   = type;
	memcpy(ev + 1, data, size);

	atom->size += sizeof(LV2_Atom_Event) + lv2_atom_pad_size(size);

	_latest_event = frames;

	return true;
}

bool
Buffer::append_event(int64_t frames, const LV2_Atom* body)
{
	return append_event(frames,
	                    body->size,
	                    body->type,
	                    reinterpret_cast<const uint8_t*>(body + 1));
}

bool
Buffer::append_event_buffer(const Buffer* buf)
{
	auto*       seq = reinterpret_cast<LV2_Atom_Sequence*>(get<LV2_Atom>());
	const auto* bseq =
	    reinterpret_cast<const LV2_Atom_Sequence*>(buf->get<LV2_Atom>());

	if (seq->atom.type == _factory.uris().atom_Chunk) {
		clear();  // Chunk initialized with prepare_output_write(), clear
	}

	const uint32_t total_size = lv2_atom_total_size(&seq->atom);
	uint8_t* const end        = reinterpret_cast<uint8_t*>(seq) + total_size;
	const uint32_t n_bytes    = bseq->atom.size - sizeof(bseq->body);
	if (sizeof(LV2_Atom) + total_size + n_bytes >= _capacity) {
		return false;  // Not enough space
	}

	memcpy(end, bseq + 1, n_bytes);
	seq->atom.size += n_bytes;

	_latest_event = std::max(_latest_event, buf->_latest_event);

	return true;
}

SampleCount
Buffer::next_value_offset(SampleCount offset, SampleCount end) const
{
	if (_type == _factory.uris().atom_Sequence && _value_type) {
		const auto* seq = get<const LV2_Atom_Sequence>();
		LV2_ATOM_SEQUENCE_FOREACH(seq, ev) {
			if (ev->time.frames >  offset   &&
			    ev->time.frames <  end &&
			    ev->body.type   == _value_type) {
				return ev->time.frames;
			}
		}
	}

	/* For CV buffers, it's possible to scan for a value change here, which for
	   stepped CV would do the right thing, but in the worst case (e.g. with
	   sine waves), when connected to a control port would split the cycle for
	   every frame which isn't feasible.  Instead, just return end, so the
	   cycle will not be split.

	   A plugin that takes CV and emits discrete change events, possibly with a
	   maximum rate or fuzz factor, would allow the user to choose which
	   behaviour, at the cost of some overhead.
	*/

	return end;
}

const LV2_Atom*
Buffer::value() const
{
	return _value_buffer ? _value_buffer->get<const LV2_Atom>() : nullptr;
}

void
Buffer::set_value(const Atom& value)
{
	if (!value.is_valid() || !_value_buffer) {
		return;
	}

	const uint32_t total_size = sizeof(LV2_Atom) + value.size();
	if (total_size > _value_buffer->capacity()) {
		_value_buffer = _factory.claim_buffer(value.type(), 0, total_size);
	}

	memcpy(_value_buffer->get<LV2_Atom*>(), value.atom(), total_size);
}

void
Buffer::update_value_buffer(SampleCount offset)
{
	if (!_value_buffer || !_value_type) {
		return;
	}

	auto*           seq    = get<LV2_Atom_Sequence>();
	LV2_Atom_Event* latest = nullptr;
	LV2_ATOM_SEQUENCE_FOREACH(seq, ev) {
		if (ev->time.frames > offset) {
			break;
		} else if (ev->body.type == _value_type) {
			latest = ev;
		}
	}

	if (latest) {
		memcpy(_value_buffer->get<LV2_Atom>(),
		       &latest->body,
		       lv2_atom_total_size(&latest->body));
	}
}

#ifndef NDEBUG
void
Buffer::dump_cv(const RunContext& ctx) const
{
	float value = samples()[0];
	fprintf(stderr, "{ 0000: %.02f\n", value);
	for (uint32_t i = 0; i < ctx.nframes(); ++i) {
		if (samples()[i] != value) {
			value = samples()[i];
			fprintf(stderr, "  %4u: %.02f\n", i, value);
		}
	}
	fprintf(stderr, "}\n");
}
#endif

void* Buffer::aligned_alloc(size_t size)
{
#ifdef HAVE_POSIX_MEMALIGN
	void* buf = nullptr;
	if (!posix_memalign(static_cast<void**>(&buf), 16, size)) {
		memset(buf, 0, size);
		return buf;
	}
#else
	return (LV2_buf*)calloc(1, size);
#endif
	return nullptr;
}

void
intrusive_ptr_add_ref(Buffer* b)
{
	b->ref();
}

void
intrusive_ptr_release(Buffer* b)
{
	b->deref();
}

} // namespace server
} // namespace ingen