/*
* Copyright (C) 2011 The Android Open Source Project
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#define LOG_TAG "nv_offload_visualizer"
//#define LOG_NDEBUG 1
#include <assert.h>
#include <math.h>
#include <stdlib.h>
#include <string.h>
#include <time.h>
#include <sys/prctl.h>
#include <cutils/list.h>
#include <cutils/log.h>
#include <system/thread_defs.h>
#include <tinyalsa/asoundlib.h>
#include <audio_effects/effect_visualizer.h>
enum {
EFFECT_STATE_UNINITIALIZED,
EFFECT_STATE_INITIALIZED,
EFFECT_STATE_ACTIVE,
};
typedef struct effect_context_s effect_context_t;
typedef struct output_context_s output_context_t;
/* effect specific operations. Only the init() and process() operations must be defined.
* Others are optional.
*/
typedef struct effect_ops_s {
int (*init)(effect_context_t *context);
int (*release)(effect_context_t *context);
int (*reset)(effect_context_t *context);
int (*enable)(effect_context_t *context);
int (*disable)(effect_context_t *context);
int (*start)(effect_context_t *context, output_context_t *output);
int (*stop)(effect_context_t *context, output_context_t *output);
int (*process)(effect_context_t *context, audio_buffer_t *in, audio_buffer_t *out);
int (*set_parameter)(effect_context_t *context, effect_param_t *param, uint32_t size);
int (*get_parameter)(effect_context_t *context, effect_param_t *param, uint32_t *size);
int (*command)(effect_context_t *context, uint32_t cmdCode, uint32_t cmdSize,
void *pCmdData, uint32_t *replySize, void *pReplyData);
} effect_ops_t;
struct effect_context_s {
const struct effect_interface_s *itfe;
struct listnode effects_list_node;
struct listnode output_node;
effect_config_t config;
const effect_descriptor_t *desc;
audio_io_handle_t out_handle; /* io handle of the output the effect is attached to */
uint32_t state;
bool offload_enabled; /* when offload is enabled we process VISUALIZER_CMD_CAPTURE command.
Otherwise non offloaded visualizer has already processed the command
and we must not overwrite the reply. */
effect_ops_t ops;
};
typedef struct output_context_s {
struct listnode outputs_list_node; /* node in active_outputs_list */
audio_io_handle_t handle; /* io handle */
struct listnode effects_list; /* list of effects attached to this output */
} output_context_t;
/* maximum time since last capture buffer update before resetting capture buffer. This means
that the framework has stopped playing audio and we must start returning silence */
#define MAX_STALL_TIME_MS 1000
#define CAPTURE_BUF_SIZE 65536 /* "64k should be enough for everyone" */
#define DISCARD_MEASUREMENTS_TIME_MS 2000 /* discard measurements older than this number of ms */
/* maximum number of buffers for which we keep track of the measurements */
#define MEASUREMENT_WINDOW_MAX_SIZE_IN_BUFFERS 25 /* note: buffer index is stored in uint8_t */
typedef struct buffer_stats_s {
bool is_valid;
uint16_t peak_u16; /* the positive peak of the absolute value of the samples in a buffer */
float rms_squared; /* the average square of the samples in a buffer */
} buffer_stats_t;
typedef struct visualizer_context_s {
effect_context_t common;
uint32_t capture_idx;
uint32_t capture_size;
uint32_t scaling_mode;
uint32_t last_capture_idx;
uint32_t latency;
struct timespec buffer_update_time;
uint8_t capture_buf[CAPTURE_BUF_SIZE];
/* for measurements */
uint8_t channel_count; /* to avoid recomputing it every time a buffer is processed */
uint32_t meas_mode;
uint8_t meas_wndw_size_in_buffers;
uint8_t meas_buffer_idx;
buffer_stats_t past_meas[MEASUREMENT_WINDOW_MAX_SIZE_IN_BUFFERS];
} visualizer_context_t;
extern const struct effect_interface_s effect_interface;
/* Visualizer UUID: 09f673c0-10bc-11e4-9589-0002a5d5c51b */
const effect_descriptor_t visualizer_descriptor = {
{0xe46b26a0, 0xdddd, 0x11db, 0x8afd, {0x00, 0x02, 0xa5, 0xd5, 0xc5, 0x1b}},
{0x09f673c0, 0x10bc, 0x11e4, 0x9589, {0x00, 0x02, 0xa5, 0xd5, 0xc5, 0x1b}},
EFFECT_CONTROL_API_VERSION,
(EFFECT_FLAG_TYPE_INSERT | EFFECT_FLAG_HW_ACC_TUNNEL ),
0,
1,
"Nvidia offload visualizer",
"The Android Open Source Project",
};
const effect_descriptor_t *descriptors[] = {
&visualizer_descriptor,
NULL,
};
pthread_once_t once = PTHREAD_ONCE_INIT;
int init_status;
/* list of created effects. Updated by visualizer_hal_start_output()
* and visualizer_hal_stop_output() */
struct listnode created_effects_list;
/* list of active output streams. Updated by visualizer_hal_start_output()
* and visualizer_hal_stop_output() */
struct listnode active_outputs_list;
/* thread capturing PCM from Proxy port and calling the process function on each enabled effect
* attached to an active output stream */
pthread_t capture_thread;
/* lock must be held when modifying or accessing created_effects_list or active_outputs_list */
pthread_mutex_t lock;
/* thread_lock must be held when starting or stopping the capture thread.
* Locking order: thread_lock -> lock */
pthread_mutex_t thread_lock;
/* cond is signaled when an output is started or stopped or an effect is enabled or disable: the
* capture thread will reevaluate the capture and effect rocess conditions. */
pthread_cond_t cond;
/* true when requesting the capture thread to exit */
bool exit_thread;
/* 0 if the capture thread was created successfully */
int thread_status;
#define SOUND_CARD 0
#define CAPTURE_DEVICE 8 /* Effects capture node */
#define AUDIO_CAPTURE_CHANNEL_COUNT 2
#define AUDIO_CAPTURE_SMP_RATE 48000
#define AUDIO_CAPTURE_PERIOD_SIZE 1024
struct pcm_config pcm_config_capture = {
.channels = AUDIO_CAPTURE_CHANNEL_COUNT,
.rate = AUDIO_CAPTURE_SMP_RATE,
.period_size = AUDIO_CAPTURE_PERIOD_SIZE,
.period_count = 4,
.format = PCM_FORMAT_S16_LE,
.start_threshold = 4095,
.stop_threshold = 4096,
.avail_min = 1,
};
/*
* Local functions
*/
static void init_once() {
list_init(&created_effects_list);
list_init(&active_outputs_list);
pthread_mutex_init(&lock, NULL);
pthread_mutex_init(&thread_lock, NULL);
pthread_cond_init(&cond, NULL);
exit_thread = false;
thread_status = -1;
init_status = 0;
}
int lib_init() {
pthread_once(&once, init_once);
return init_status;
}
bool effect_exists(effect_context_t *context) {
struct listnode *node;
list_for_each(node, &created_effects_list) {
effect_context_t *fx_ctxt = node_to_item(node,
effect_context_t,
effects_list_node);
if (fx_ctxt == context) {
return true;
}
}
return false;
}
output_context_t *get_output(audio_io_handle_t output) {
struct listnode *node;
list_for_each(node, &active_outputs_list) {
output_context_t *out_ctxt = node_to_item(node,
output_context_t,
outputs_list_node);
if (out_ctxt->handle == output) {
return out_ctxt;
}
}
return NULL;
}
void add_effect_to_output(output_context_t * output, effect_context_t *context) {
struct listnode *fx_node;
list_for_each(fx_node, &output->effects_list) {
effect_context_t *fx_ctxt = node_to_item(fx_node,
effect_context_t,
output_node);
if (fx_ctxt == context)
return;
}
list_add_tail(&output->effects_list, &context->output_node);
if (context->ops.start)
context->ops.start(context, output);
}
void remove_effect_from_output(output_context_t * output, effect_context_t *context) {
struct listnode *fx_node;
list_for_each(fx_node, &output->effects_list) {
effect_context_t *fx_ctxt = node_to_item(fx_node,
effect_context_t,
output_node);
if (fx_ctxt == context) {
if (context->ops.stop)
context->ops.stop(context, output);
list_remove(&context->output_node);
return;
}
}
}
bool effects_enabled() {
struct listnode *out_node;
list_for_each(out_node, &active_outputs_list) {
struct listnode *fx_node;
output_context_t *out_ctxt = node_to_item(out_node,
output_context_t,
outputs_list_node);
list_for_each(fx_node, &out_ctxt->effects_list) {
effect_context_t *fx_ctxt = node_to_item(fx_node,
effect_context_t,
output_node);
if (fx_ctxt->state == EFFECT_STATE_ACTIVE && fx_ctxt->ops.process != NULL)
return true;
}
}
return false;
}
void *effects_capture_thread_loop(void *arg __unused)
{
int16_t data[AUDIO_CAPTURE_PERIOD_SIZE * AUDIO_CAPTURE_CHANNEL_COUNT * sizeof(int16_t)];
audio_buffer_t buf;
buf.frameCount = AUDIO_CAPTURE_PERIOD_SIZE;
buf.s16 = data;
bool capture_enabled = false;
struct pcm *pcm = NULL;
int ret;
int retry_num = 0;
prctl(PR_SET_NAME, (unsigned long)"visualizer capture", 0, 0, 0);
pthread_mutex_lock(&lock);
for (;;) {
if (exit_thread) {
break;
}
if (effects_enabled()) {
if (!capture_enabled) {
pcm = pcm_open(SOUND_CARD, CAPTURE_DEVICE,
PCM_IN, &pcm_config_capture);
if (pcm && !pcm_is_ready(pcm)) {
ALOGW("%s: %s", __func__, pcm_get_error(pcm));
pcm_close(pcm);
pcm = NULL;
} else {
capture_enabled = true;
ALOGD("%s: capture ENABLED", __func__);
}
}
} else {
if (capture_enabled) {
if (pcm != NULL) {
ALOGD("%s:Closing pcm\n", __func__);
pcm_close(pcm);
}
ALOGD("%s: capture DISABLED", __func__);
capture_enabled = false;
}
pthread_cond_wait(&cond, &lock);
}
if (!capture_enabled)
continue;
pthread_mutex_unlock(&lock);
ret = pcm_read(pcm, data, sizeof(data));
pthread_mutex_lock(&lock);
if (ret == 0) {
struct listnode *out_node;
list_for_each(out_node, &active_outputs_list) {
output_context_t *out_ctxt = node_to_item(out_node,
output_context_t,
outputs_list_node);
struct listnode *fx_node;
list_for_each(fx_node, &out_ctxt->effects_list) {
effect_context_t *fx_ctxt = node_to_item(fx_node,
effect_context_t,
output_node);
if (fx_ctxt->ops.process != NULL)
fx_ctxt->ops.process(fx_ctxt, &buf, &buf);
}
}
} else {
ALOGW("%s: read status %d %s", __func__, ret, pcm_get_error(pcm));
}
}
if (capture_enabled) {
if (pcm != NULL)
pcm_close(pcm);
}
pthread_mutex_unlock(&lock);
ALOGD("thread exit");
return NULL;
}
/*
* Interface from audio HAL
*/
__attribute__ ((visibility ("default")))
int visualizer_hal_start_output(audio_io_handle_t output, int pcm_id) {
int ret;
struct listnode *node;
ALOGV("%s output %d pcm_id %d", __func__, output, pcm_id);
if (lib_init() != 0)
return init_status;
pthread_mutex_lock(&thread_lock);
pthread_mutex_lock(&lock);
if (get_output(output) != NULL) {
ALOGE("%s output already started", __func__);
ret = -ENOSYS;
goto exit;
}
output_context_t *out_ctxt = (output_context_t *)malloc(sizeof(output_context_t));
out_ctxt->handle = output;
list_init(&out_ctxt->effects_list);
list_for_each(node, &created_effects_list) {
effect_context_t *fx_ctxt = node_to_item(node,
effect_context_t,
effects_list_node);
if (fx_ctxt->out_handle == output) {
if (fx_ctxt->ops.start)
fx_ctxt->ops.start(fx_ctxt, out_ctxt);
list_add_tail(&out_ctxt->effects_list, &fx_ctxt->output_node);
}
}
if (list_empty(&active_outputs_list)) {
exit_thread = false;
thread_status = pthread_create(&capture_thread, (const pthread_attr_t *) NULL,
effects_capture_thread_loop, NULL);
}
list_add_tail(&active_outputs_list, &out_ctxt->outputs_list_node);
pthread_cond_signal(&cond);
exit:
pthread_mutex_unlock(&lock);
pthread_mutex_unlock(&thread_lock);
return ret;
}
__attribute__ ((visibility ("default")))
int visualizer_hal_stop_output(audio_io_handle_t output, int pcm_id) {
int ret;
struct listnode *node;
struct listnode *fx_node;
output_context_t *out_ctxt;
ALOGV("%s output %d pcm_id %d", __func__, output, pcm_id);
if (lib_init() != 0)
return init_status;
pthread_mutex_lock(&thread_lock);
pthread_mutex_lock(&lock);
out_ctxt = get_output(output);
if (out_ctxt == NULL) {
ALOGW("%s output not started", __func__);
ret = -ENOSYS;
goto exit;
}
list_for_each(fx_node, &out_ctxt->effects_list) {
effect_context_t *fx_ctxt = node_to_item(fx_node,
effect_context_t,
output_node);
if (fx_ctxt->ops.stop)
fx_ctxt->ops.stop(fx_ctxt, out_ctxt);
}
list_remove(&out_ctxt->outputs_list_node);
pthread_cond_signal(&cond);
if (list_empty(&active_outputs_list)) {
if (thread_status == 0) {
exit_thread = true;
pthread_cond_signal(&cond);
pthread_mutex_unlock(&lock);
pthread_join(capture_thread, (void **) NULL);
pthread_mutex_lock(&lock);
thread_status = -1;
}
}
free(out_ctxt);
exit:
pthread_mutex_unlock(&lock);
pthread_mutex_unlock(&thread_lock);
return ret;
}
/*
* Effect operations
*/
int set_config(effect_context_t *context, effect_config_t *config)
{
if (config->inputCfg.samplingRate != config->outputCfg.samplingRate) return -EINVAL;
if (config->inputCfg.channels != config->outputCfg.channels) return -EINVAL;
if (config->inputCfg.format != config->outputCfg.format) return -EINVAL;
if (config->inputCfg.channels != AUDIO_CHANNEL_OUT_STEREO) return -EINVAL;
if (config->outputCfg.accessMode != EFFECT_BUFFER_ACCESS_WRITE &&
config->outputCfg.accessMode != EFFECT_BUFFER_ACCESS_ACCUMULATE) return -EINVAL;
if (config->inputCfg.format != AUDIO_FORMAT_PCM_16_BIT) return -EINVAL;
context->config = *config;
if (context->ops.reset)
context->ops.reset(context);
return 0;
}
void get_config(effect_context_t *context, effect_config_t *config)
{
*config = context->config;
}
/*
* Visualizer operations
*/
uint32_t visualizer_get_delta_time_ms_from_updated_time(visualizer_context_t* visu_ctxt) {
uint32_t delta_ms = 0;
if (visu_ctxt->buffer_update_time.tv_sec != 0) {
struct timespec ts;
if (clock_gettime(CLOCK_MONOTONIC, &ts) == 0) {
time_t secs = ts.tv_sec - visu_ctxt->buffer_update_time.tv_sec;
long nsec = ts.tv_nsec - visu_ctxt->buffer_update_time.tv_nsec;
if (nsec < 0) {
--secs;
nsec += 1000000000;
}
delta_ms = secs * 1000 + nsec / 1000000;
}
}
return delta_ms;
}
int visualizer_reset(effect_context_t *context)
{
visualizer_context_t * visu_ctxt = (visualizer_context_t *)context;
visu_ctxt->capture_idx = 0;
visu_ctxt->last_capture_idx = 0;
visu_ctxt->buffer_update_time.tv_sec = 0;
visu_ctxt->latency = 0;
memset(visu_ctxt->capture_buf, 0x80, CAPTURE_BUF_SIZE);
return 0;
}
int visualizer_init(effect_context_t *context)
{
int32_t i;
visualizer_context_t * visu_ctxt = (visualizer_context_t *)context;
context->config.inputCfg.accessMode = EFFECT_BUFFER_ACCESS_READ;
context->config.inputCfg.channels = AUDIO_CHANNEL_OUT_STEREO;
context->config.inputCfg.format = AUDIO_FORMAT_PCM_16_BIT;
context->config.inputCfg.samplingRate = 44100;
context->config.inputCfg.bufferProvider.getBuffer = NULL;
context->config.inputCfg.bufferProvider.releaseBuffer = NULL;
context->config.inputCfg.bufferProvider.cookie = NULL;
context->config.inputCfg.mask = EFFECT_CONFIG_ALL;
context->config.outputCfg.accessMode = EFFECT_BUFFER_ACCESS_ACCUMULATE;
context->config.outputCfg.channels = AUDIO_CHANNEL_OUT_STEREO;
context->config.outputCfg.format = AUDIO_FORMAT_PCM_16_BIT;
context->config.outputCfg.samplingRate = 44100;
context->config.outputCfg.bufferProvider.getBuffer = NULL;
context->config.outputCfg.bufferProvider.releaseBuffer = NULL;
context->config.outputCfg.bufferProvider.cookie = NULL;
context->config.outputCfg.mask = EFFECT_CONFIG_ALL;
visu_ctxt->capture_size = VISUALIZER_CAPTURE_SIZE_MAX;
visu_ctxt->scaling_mode = VISUALIZER_SCALING_MODE_NORMALIZED;
// measurement initialization
visu_ctxt->channel_count = popcount(context->config.inputCfg.channels);
visu_ctxt->meas_mode = MEASUREMENT_MODE_NONE;
visu_ctxt->meas_wndw_size_in_buffers = MEASUREMENT_WINDOW_MAX_SIZE_IN_BUFFERS;
visu_ctxt->meas_buffer_idx = 0;
for (i=0 ; i<visu_ctxt->meas_wndw_size_in_buffers ; i++) {
visu_ctxt->past_meas[i].is_valid = false;
visu_ctxt->past_meas[i].peak_u16 = 0;
visu_ctxt->past_meas[i].rms_squared = 0;
}
set_config(context, &context->config);
return 0;
}
int visualizer_get_parameter(effect_context_t *context, effect_param_t *p, uint32_t *size)
{
visualizer_context_t *visu_ctxt = (visualizer_context_t *)context;
p->status = 0;
*size = sizeof(effect_param_t) + sizeof(uint32_t);
if (p->psize != sizeof(uint32_t)) {
p->status = -EINVAL;
return 0;
}
switch (*(uint32_t *)p->data) {
case VISUALIZER_PARAM_CAPTURE_SIZE:
ALOGV("%s get capture_size = %d", __func__, visu_ctxt->capture_size);
*((uint32_t *)p->data + 1) = visu_ctxt->capture_size;
p->vsize = sizeof(uint32_t);
*size += sizeof(uint32_t);
break;
case VISUALIZER_PARAM_SCALING_MODE:
ALOGV("%s get scaling_mode = %d", __func__, visu_ctxt->scaling_mode);
*((uint32_t *)p->data + 1) = visu_ctxt->scaling_mode;
p->vsize = sizeof(uint32_t);
*size += sizeof(uint32_t);
break;
case VISUALIZER_PARAM_MEASUREMENT_MODE:
ALOGV("%s get meas_mode = %d", __func__, visu_ctxt->meas_mode);
*((uint32_t *)p->data + 1) = visu_ctxt->meas_mode;
p->vsize = sizeof(uint32_t);
*size += sizeof(uint32_t);
break;
default:
p->status = -EINVAL;
}
return 0;
}
int visualizer_set_parameter(effect_context_t *context, effect_param_t *p, uint32_t size __unused)
{
visualizer_context_t *visu_ctxt = (visualizer_context_t *)context;
if (p->psize != sizeof(uint32_t) || p->vsize != sizeof(uint32_t))
return -EINVAL;
switch (*(uint32_t *)p->data) {
case VISUALIZER_PARAM_CAPTURE_SIZE:
visu_ctxt->capture_size = *((uint32_t *)p->data + 1);
ALOGV("%s set capture_size = %d", __func__, visu_ctxt->capture_size);
break;
case VISUALIZER_PARAM_SCALING_MODE:
visu_ctxt->scaling_mode = *((uint32_t *)p->data + 1);
ALOGV("%s set scaling_mode = %d", __func__, visu_ctxt->scaling_mode);
break;
case VISUALIZER_PARAM_LATENCY:
ALOGV("%s set latency = %d", __func__, visu_ctxt->latency);
break;
case VISUALIZER_PARAM_MEASUREMENT_MODE:
visu_ctxt->meas_mode = *((uint32_t *)p->data + 1);
ALOGV("%s set meas_mode = %d", __func__, visu_ctxt->meas_mode);
break;
default:
return -EINVAL;
}
return 0;
}
/* Real process function called from capture thread. Called with lock held */
int visualizer_process(effect_context_t *context,
audio_buffer_t *inBuffer,
audio_buffer_t *outBuffer)
{
visualizer_context_t *visu_ctxt = (visualizer_context_t *)context;
if (!effect_exists(context))
return -EINVAL;
if (inBuffer == NULL || inBuffer->raw == NULL ||
outBuffer == NULL || outBuffer->raw == NULL ||
inBuffer->frameCount != outBuffer->frameCount ||
inBuffer->frameCount == 0) {
return -EINVAL;
}
// perform measurements if needed
if (visu_ctxt->meas_mode & MEASUREMENT_MODE_PEAK_RMS) {
// find the peak and RMS squared for the new buffer
uint32_t inIdx;
int16_t max_sample = 0;
float rms_squared_acc = 0;
for (inIdx = 0 ; inIdx < inBuffer->frameCount * visu_ctxt->channel_count ; inIdx++) {
if (inBuffer->s16[inIdx] > max_sample) {
max_sample = inBuffer->s16[inIdx];
} else if (-inBuffer->s16[inIdx] > max_sample) {
max_sample = -inBuffer->s16[inIdx];
}
rms_squared_acc += (inBuffer->s16[inIdx] * inBuffer->s16[inIdx]);
}
// store the measurement
visu_ctxt->past_meas[visu_ctxt->meas_buffer_idx].peak_u16 = (uint16_t)max_sample;
visu_ctxt->past_meas[visu_ctxt->meas_buffer_idx].rms_squared =
rms_squared_acc / (inBuffer->frameCount * visu_ctxt->channel_count);
visu_ctxt->past_meas[visu_ctxt->meas_buffer_idx].is_valid = true;
if (++visu_ctxt->meas_buffer_idx >= visu_ctxt->meas_wndw_size_in_buffers) {
visu_ctxt->meas_buffer_idx = 0;
}
}
/* all code below assumes stereo 16 bit PCM output and input */
int32_t shift;
if (visu_ctxt->scaling_mode == VISUALIZER_SCALING_MODE_NORMALIZED) {
/* derive capture scaling factor from peak value in current buffer
* this gives more interesting captures for display. */
shift = 32;
int len = inBuffer->frameCount * 2;
int i;
for (i = 0; i < len; i++) {
int32_t smp = inBuffer->s16[i];
if (smp < 0) smp = -smp - 1; /* take care to keep the max negative in range */
int32_t clz = __builtin_clz(smp);
if (shift > clz) shift = clz;
}
/* A maximum amplitude signal will have 17 leading zeros, which we want to
* translate to a shift of 8 (for converting 16 bit to 8 bit) */
shift = 25 - shift;
/* Never scale by less than 8 to avoid returning unaltered PCM signal. */
if (shift < 3) {
shift = 3;
}
/* add one to combine the division by 2 needed after summing
* left and right channels below */
shift++;
} else {
assert(visu_ctxt->scaling_mode == VISUALIZER_SCALING_MODE_AS_PLAYED);
shift = 9;
}
uint32_t capt_idx;
uint32_t in_idx;
uint8_t *buf = visu_ctxt->capture_buf;
for (in_idx = 0, capt_idx = visu_ctxt->capture_idx;
in_idx < inBuffer->frameCount;
in_idx++, capt_idx++) {
if (capt_idx >= CAPTURE_BUF_SIZE) {
/* wrap around */
capt_idx = 0;
}
int32_t smp = inBuffer->s16[2 * in_idx] + inBuffer->s16[2 * in_idx + 1];
smp = smp >> shift;
buf[capt_idx] = ((uint8_t)smp)^0x80;
}
/* XXX the following two should really be atomic, though it probably doesn't
* matter much for visualization purposes */
visu_ctxt->capture_idx = capt_idx;
/* update last buffer update time stamp */
if (clock_gettime(CLOCK_MONOTONIC, &visu_ctxt->buffer_update_time) < 0) {
visu_ctxt->buffer_update_time.tv_sec = 0;
}
if (context->state != EFFECT_STATE_ACTIVE) {
ALOGV("%s DONE inactive", __func__);
return -ENODATA;
}
return 0;
}
int visualizer_command(effect_context_t * context, uint32_t cmdCode, uint32_t cmdSize __unused,
void *pCmdData __unused, uint32_t *replySize, void *pReplyData)
{
visualizer_context_t * visu_ctxt = (visualizer_context_t *)context;
switch (cmdCode) {
case VISUALIZER_CMD_CAPTURE:
if (pReplyData == NULL || *replySize != visu_ctxt->capture_size) {
ALOGV("%s VISUALIZER_CMD_CAPTURE error *replySize %d context->capture_size %d",
__func__, *replySize, visu_ctxt->capture_size);
return -EINVAL;
}
if (!context->offload_enabled)
break;
if (context->state == EFFECT_STATE_ACTIVE) {
int32_t latency_ms = visu_ctxt->latency;
const uint32_t delta_ms = visualizer_get_delta_time_ms_from_updated_time(visu_ctxt);
latency_ms -= delta_ms;
if (latency_ms < 0) {
latency_ms = 0;
}
const uint32_t delta_smp = context->config.inputCfg.samplingRate * latency_ms / 1000;
int32_t capture_point = visu_ctxt->capture_idx - visu_ctxt->capture_size - delta_smp;
int32_t capture_size = visu_ctxt->capture_size;
if (capture_point < 0) {
int32_t size = -capture_point;
if (size > capture_size)
size = capture_size;
memcpy(pReplyData,
visu_ctxt->capture_buf + CAPTURE_BUF_SIZE + capture_point,
size);
pReplyData = (void *)((size_t)pReplyData + size);
capture_size -= size;
capture_point = 0;
}
memcpy(pReplyData,
visu_ctxt->capture_buf + capture_point,
capture_size);
/* if audio framework has stopped playing audio although the effect is still
* active we must clear the capture buffer to return silence */
if ((visu_ctxt->last_capture_idx == visu_ctxt->capture_idx) &&
(visu_ctxt->buffer_update_time.tv_sec != 0)) {
if (delta_ms > MAX_STALL_TIME_MS) {
ALOGV("%s capture going to idle", __func__);
visu_ctxt->buffer_update_time.tv_sec = 0;
memset(pReplyData, 0x80, visu_ctxt->capture_size);
}
}
visu_ctxt->last_capture_idx = visu_ctxt->capture_idx;
} else {
memset(pReplyData, 0x80, visu_ctxt->capture_size);
}
break;
case VISUALIZER_CMD_MEASURE: {
uint16_t peak_u16 = 0;
float sum_rms_squared = 0.0f;
uint8_t nb_valid_meas = 0;
/* reset measurements if last measurement was too long ago (which implies stored
* measurements aren't relevant anymore and shouldn't bias the new one) */
const int32_t delay_ms = visualizer_get_delta_time_ms_from_updated_time(visu_ctxt);
if (delay_ms > DISCARD_MEASUREMENTS_TIME_MS) {
uint32_t i;
ALOGV("Discarding measurements, last measurement is %dms old", delay_ms);
for (i=0 ; i<visu_ctxt->meas_wndw_size_in_buffers ; i++) {
visu_ctxt->past_meas[i].is_valid = false;
visu_ctxt->past_meas[i].peak_u16 = 0;
visu_ctxt->past_meas[i].rms_squared = 0;
}
visu_ctxt->meas_buffer_idx = 0;
} else {
/* only use actual measurements, otherwise the first RMS measure happening before
* MEASUREMENT_WINDOW_MAX_SIZE_IN_BUFFERS have been played will always be artificially
* low */
uint32_t i;
for (i=0 ; i < visu_ctxt->meas_wndw_size_in_buffers ; i++) {
if (visu_ctxt->past_meas[i].is_valid) {
if (visu_ctxt->past_meas[i].peak_u16 > peak_u16) {
peak_u16 = visu_ctxt->past_meas[i].peak_u16;
}
sum_rms_squared += visu_ctxt->past_meas[i].rms_squared;
nb_valid_meas++;
}
}
}
float rms = nb_valid_meas == 0 ? 0.0f : sqrtf(sum_rms_squared / nb_valid_meas);
int32_t* p_int_reply_data = (int32_t*)pReplyData;
/* convert from I16 sample values to mB and write results */
if (rms < 0.000016f) {
p_int_reply_data[MEASUREMENT_IDX_RMS] = -9600; //-96dB
} else {
p_int_reply_data[MEASUREMENT_IDX_RMS] = (int32_t) (2000 * log10(rms / 32767.0f));
}
if (peak_u16 == 0) {
p_int_reply_data[MEASUREMENT_IDX_PEAK] = -9600; //-96dB
} else {
p_int_reply_data[MEASUREMENT_IDX_PEAK] = (int32_t) (2000 * log10(peak_u16 / 32767.0f));
}
ALOGV("VISUALIZER_CMD_MEASURE peak=%d (%dmB), rms=%.1f (%dmB)",
peak_u16, p_int_reply_data[MEASUREMENT_IDX_PEAK],
rms, p_int_reply_data[MEASUREMENT_IDX_RMS]);
}
break;
default:
ALOGW("%s invalid command %d", __func__, cmdCode);
return -EINVAL;
}
return 0;
}
/*
* Effect Library Interface Implementation
*/
int effect_lib_create(const effect_uuid_t *uuid,
int32_t sessionId __unused,
int32_t ioId,
effect_handle_t *pHandle) {
int ret;
int i;
if (lib_init() != 0)
return init_status;
if (pHandle == NULL || uuid == NULL)
return -EINVAL;
for (i = 0; descriptors[i] != NULL; i++) {
if (memcmp(uuid, &descriptors[i]->uuid, sizeof(effect_uuid_t)) == 0)
break;
}
if (descriptors[i] == NULL)
return -EINVAL;
effect_context_t *context;
if (memcmp(uuid, &visualizer_descriptor.uuid, sizeof(effect_uuid_t)) == 0) {
visualizer_context_t *visu_ctxt = (visualizer_context_t *)calloc(1,
sizeof(visualizer_context_t));
context = (effect_context_t *)visu_ctxt;
context->ops.init = visualizer_init;
context->ops.reset = visualizer_reset;
context->ops.process = visualizer_process;
context->ops.set_parameter = visualizer_set_parameter;
context->ops.get_parameter = visualizer_get_parameter;
context->ops.command = visualizer_command;
context->desc = &visualizer_descriptor;
} else {
return -EINVAL;
}
context->itfe = &effect_interface;
context->state = EFFECT_STATE_UNINITIALIZED;
context->out_handle = (audio_io_handle_t)ioId;
ret = context->ops.init(context);
if (ret < 0) {
ALOGW("%s init failed", __func__);
free(context);
return ret;
}
context->state = EFFECT_STATE_INITIALIZED;
pthread_mutex_lock(&lock);
list_add_tail(&created_effects_list, &context->effects_list_node);
output_context_t *out_ctxt = get_output(ioId);
if (out_ctxt != NULL)
add_effect_to_output(out_ctxt, context);
pthread_mutex_unlock(&lock);
*pHandle = (effect_handle_t)context;
ALOGV("%s created context %p", __func__, context);
return 0;
}
int effect_lib_release(effect_handle_t handle) {
effect_context_t *context = (effect_context_t *)handle;
int status;
if (lib_init() != 0)
return init_status;
ALOGV("%s context %p", __func__, handle);
pthread_mutex_lock(&lock);
status = -EINVAL;
if (effect_exists(context)) {
output_context_t *out_ctxt = get_output(context->out_handle);
if (out_ctxt != NULL)
remove_effect_from_output(out_ctxt, context);
list_remove(&context->effects_list_node);
if (context->ops.release)
context->ops.release(context);
free(context);
status = 0;
}
pthread_mutex_unlock(&lock);
return status;
}
int effect_lib_get_descriptor(const effect_uuid_t *uuid,
effect_descriptor_t *descriptor) {
int i;
if (lib_init() != 0)
return init_status;
if (descriptor == NULL || uuid == NULL) {
ALOGV("%s called with NULL pointer", __func__);
return -EINVAL;
}
for (i = 0; descriptors[i] != NULL; i++) {
if (memcmp(uuid, &descriptors[i]->uuid, sizeof(effect_uuid_t)) == 0) {
*descriptor = *descriptors[i];
return 0;
}
}
return -EINVAL;
}
/*
* Effect Control Interface Implementation
*/
/* Stub function for effect interface: never called for offloaded effects */
int effect_process(effect_handle_t self,
audio_buffer_t *inBuffer __unused,
audio_buffer_t *outBuffer __unused)
{
effect_context_t * context = (effect_context_t *)self;
int status = 0;
//ALOGW("%s Called ?????", __func__);
pthread_mutex_lock(&lock);
if (!effect_exists(context)) {
status = -EINVAL;
goto exit;
}
if (context->state != EFFECT_STATE_ACTIVE) {
status = -EINVAL;
goto exit;
}
exit:
pthread_mutex_unlock(&lock);
return status;
}
int effect_command(effect_handle_t self, uint32_t cmdCode, uint32_t cmdSize,
void *pCmdData, uint32_t *replySize, void *pReplyData)
{
effect_context_t * context = (effect_context_t *)self;
int retsize;
int status = 0;
pthread_mutex_lock(&lock);
if (!effect_exists(context)) {
status = -EINVAL;
goto exit;
}
if (context == NULL || context->state == EFFECT_STATE_UNINITIALIZED) {
status = -EINVAL;
goto exit;
}
// ALOGV_IF(cmdCode != VISUALIZER_CMD_CAPTURE,
// "%s command %d cmdSize %d", __func__, cmdCode, cmdSize);
switch (cmdCode) {
case EFFECT_CMD_INIT:
if (pReplyData == NULL || *replySize != sizeof(int)) {
status = -EINVAL;
goto exit;
}
if (context->ops.init)
*(int *) pReplyData = context->ops.init(context);
else
*(int *) pReplyData = 0;
break;
case EFFECT_CMD_SET_CONFIG:
if (pCmdData == NULL || cmdSize != sizeof(effect_config_t)
|| pReplyData == NULL || *replySize != sizeof(int)) {
status = -EINVAL;
goto exit;
}
*(int *) pReplyData = set_config(context, (effect_config_t *) pCmdData);
break;
case EFFECT_CMD_GET_CONFIG:
if (pReplyData == NULL ||
*replySize != sizeof(effect_config_t)) {
status = -EINVAL;
goto exit;
}
if (!context->offload_enabled) {
status = -EINVAL;
goto exit;
}
get_config(context, (effect_config_t *)pReplyData);
break;
case EFFECT_CMD_RESET:
if (context->ops.reset)
context->ops.reset(context);
break;
case EFFECT_CMD_ENABLE:
if (pReplyData == NULL || *replySize != sizeof(int)) {
status = -EINVAL;
goto exit;
}
if (context->state != EFFECT_STATE_INITIALIZED) {
status = -ENOSYS;
goto exit;
}
context->state = EFFECT_STATE_ACTIVE;
if (context->ops.enable) {
context->ops.enable(context);
}
pthread_cond_signal(&cond);
*(int *)pReplyData = 0;
break;
case EFFECT_CMD_DISABLE:
if (pReplyData == NULL || *replySize != sizeof(int)) {
status = -EINVAL;
goto exit;
}
if (context->state != EFFECT_STATE_ACTIVE) {
status = -ENOSYS;
goto exit;
}
context->state = EFFECT_STATE_INITIALIZED;
if (context->ops.disable)
context->ops.disable(context);
pthread_cond_signal(&cond);
ALOGV("%s EFFECT_CMD_DISABLE", __func__);
*(int *)pReplyData = 0;
break;
case EFFECT_CMD_GET_PARAM: {
if (pCmdData == NULL ||
cmdSize != (int)(sizeof(effect_param_t) + sizeof(uint32_t)) ||
pReplyData == NULL ||
*replySize < (int)(sizeof(effect_param_t) + sizeof(uint32_t) + sizeof(uint32_t))) {
status = -EINVAL;
goto exit;
}
if (!context->offload_enabled) {
status = -EINVAL;
goto exit;
}
memcpy(pReplyData, pCmdData, sizeof(effect_param_t) + sizeof(uint32_t));
effect_param_t *p = (effect_param_t *)pReplyData;
if (context->ops.get_parameter)
context->ops.get_parameter(context, p, replySize);
} break;
case EFFECT_CMD_SET_PARAM: {
if (pCmdData == NULL ||
cmdSize != (int)(sizeof(effect_param_t) + sizeof(uint32_t) + sizeof(uint32_t)) ||
pReplyData == NULL || *replySize != sizeof(int32_t)) {
status = -EINVAL;
goto exit;
}
*(int32_t *)pReplyData = 0;
effect_param_t *p = (effect_param_t *)pCmdData;
if (context->ops.set_parameter)
*(int32_t *)pReplyData = context->ops.set_parameter(context, p, *replySize);
} break;
case EFFECT_CMD_SET_DEVICE:
case EFFECT_CMD_SET_VOLUME:
case EFFECT_CMD_SET_AUDIO_MODE:
break;
case EFFECT_CMD_OFFLOAD: {
output_context_t *out_ctxt;
if (cmdSize != sizeof(effect_offload_param_t) || pCmdData == NULL
|| pReplyData == NULL || *replySize != sizeof(int)) {
ALOGV("%s EFFECT_CMD_OFFLOAD bad format", __func__);
status = -EINVAL;
break;
}
effect_offload_param_t* offload_param = (effect_offload_param_t*)pCmdData;
ALOGV("%s EFFECT_CMD_OFFLOAD offload %d output %d",
__func__, offload_param->isOffload, offload_param->ioHandle);
*(int *)pReplyData = 0;
context->offload_enabled = offload_param->isOffload;
if (context->out_handle == offload_param->ioHandle)
break;
out_ctxt = get_output(context->out_handle);
if (out_ctxt != NULL)
remove_effect_from_output(out_ctxt, context);
context->out_handle = offload_param->ioHandle;
out_ctxt = get_output(offload_param->ioHandle);
if (out_ctxt != NULL)
add_effect_to_output(out_ctxt, context);
} break;
default:
if (cmdCode >= EFFECT_CMD_FIRST_PROPRIETARY && context->ops.command)
status = context->ops.command(context, cmdCode, cmdSize,
pCmdData, replySize, pReplyData);
else {
ALOGW("%s invalid command %d", __func__, cmdCode);
status = -EINVAL;
}
break;
}
exit:
pthread_mutex_unlock(&lock);
// ALOGV_IF(cmdCode != VISUALIZER_CMD_CAPTURE,"%s DONE", __func__);
return status;
}
/* Effect Control Interface Implementation: get_descriptor */
int effect_get_descriptor(effect_handle_t self,
effect_descriptor_t *descriptor)
{
effect_context_t *context = (effect_context_t *)self;
if (!effect_exists(context))
return -EINVAL;
if (descriptor == NULL)
return -EINVAL;
*descriptor = *context->desc;
return 0;
}
/* effect_handle_t interface implementation for visualizer effect */
const struct effect_interface_s effect_interface = {
effect_process,
effect_command,
effect_get_descriptor,
NULL,
};
__attribute__ ((visibility ("default")))
audio_effect_library_t AUDIO_EFFECT_LIBRARY_INFO_SYM = {
tag : AUDIO_EFFECT_LIBRARY_TAG,
version : EFFECT_LIBRARY_API_VERSION,
name : "Visualizer Library",
implementor : "Nvidia",
create_effect : effect_lib_create,
release_effect : effect_lib_release,
get_descriptor : effect_lib_get_descriptor,
};