summaryrefslogtreecommitdiffstats
path: root/src
diff options
context:
space:
mode:
Diffstat (limited to 'src')
-rw-r--r--src/odr-audioenc.cpp45
1 files changed, 23 insertions, 22 deletions
diff --git a/src/odr-audioenc.cpp b/src/odr-audioenc.cpp
index 6b82c5f..bfc7738 100644
--- a/src/odr-audioenc.cpp
+++ b/src/odr-audioenc.cpp
@@ -754,22 +754,23 @@ int AudioEnc::run()
zmq_output->set_encoder_type(selected_encoder, bitrate);
}
- int outbuf_size;
- vec_u8 zmqframebuf;
+ int outbuf_size = 0;
vec_u8 outbuf;
- if (selected_encoder == encoder_selection_t::fdk_dabplus) {
- outbuf_size = bitrate/8*120;
- outbuf.resize(24*120);
+ switch (selected_encoder) {
+ case encoder_selection_t::fdk_dabplus:
+ outbuf_size = bitrate/8*120;
+ outbuf.resize(24*120);
- if(outbuf_size % 5 != 0) {
- fprintf(stderr, "Warning: (outbuf_size mod 5) = %d\n", outbuf_size % 5);
- }
- }
- else if (selected_encoder == encoder_selection_t::toolame_dab) {
- outbuf_size = 4092;
- outbuf.resize(outbuf_size);
- fprintf(stderr, "Setting outbuf size to %zu\n", outbuf.size());
+ if(outbuf_size % 5 != 0) {
+ fprintf(stderr, "Warning: (outbuf_size mod 5) = %d\n", outbuf_size % 5);
+ }
+ break;
+ case encoder_selection_t::toolame_dab:
+ outbuf_size = 4092;
+ outbuf.resize(outbuf_size);
+ fprintf(stderr, "Setting outbuf size to %zu\n", outbuf.size());
+ break;
}
unsigned char pad_buf[padlen + 1];
@@ -828,8 +829,8 @@ int AudioEnc::run()
// -------------- Read Data
- memset(&outbuf[0], 0x00, outbuf_size);
- memset(&input_buf[0], 0x00, input_buf.size());
+ memset(outbuf.data(), 0x00, outbuf_size);
+ memset(input_buf.data(), 0x00, input_buf.size());
/*! \section DataInput
* We read data input either in a blocking way (file input, VLC or ALSA
@@ -880,7 +881,7 @@ int AudioEnc::run()
if (drift_compensation) {
size_t overruns = 0;
- size_t bytes_from_queue = queue.pop(&input_buf[0], input_buf.size(), &overruns); // returns bytes
+ size_t bytes_from_queue = queue.pop(input_buf.data(), input_buf.size(), &overruns); // returns bytes
if (bytes_from_queue != input_buf.size()) {
expand_missing_samples(input_buf, channels, bytes_from_queue);
}
@@ -909,7 +910,7 @@ int AudioEnc::run()
/*! pop_wait() must return after a timeout, otherwise the silence detector cannot do
* its job. */
- ssize_t bytes_from_queue = queue.pop_wait(&input_buf[0], read_bytes, timeout_ms, &overruns); // returns bytes
+ ssize_t bytes_from_queue = queue.pop_wait(input_buf.data(), read_bytes, timeout_ms, &overruns); // returns bytes
if (overruns) {
throw logic_error("Queue overrun in non-drift compensation!");
@@ -1014,7 +1015,7 @@ int AudioEnc::run()
int in_size[2], in_elem_size[2];
int out_size, out_elem_size;
- in_ptr[0] = &input_buf[0];
+ in_ptr[0] = input_buf.data();
in_ptr[1] = pad_buf + (padlen - calculated_padlen); // offset due to unused PAD bytes
in_size[0] = read_bytes;
in_size[1] = calculated_padlen;
@@ -1029,7 +1030,7 @@ int AudioEnc::run()
in_buf.bufSizes = in_size;
in_buf.bufElSizes = in_elem_size;
- out_ptr = &outbuf[0];
+ out_ptr = outbuf.data();
out_size = outbuf.size();
out_elem_size = 1;
out_buf.numBufs = 1;
@@ -1060,7 +1061,7 @@ int AudioEnc::run()
short input_buffers[2][1152];
if (channels == 1) {
- memcpy(input_buffers[0], &input_buf[0], 1152 * BYTES_PER_SAMPLE);
+ memcpy(input_buffers[0], input_buf.data(), 1152 * BYTES_PER_SAMPLE);
}
else if (channels == 2) {
for (int i = 0; i < 1152; i++) {
@@ -1076,10 +1077,10 @@ int AudioEnc::run()
}
if (read_bytes) {
- numOutBytes = toolame_encode_frame(input_buffers, pad_buf, calculated_padlen, &outbuf[0], outbuf.size());
+ numOutBytes = toolame_encode_frame(input_buffers, pad_buf, calculated_padlen, outbuf.data(), outbuf.size());
}
else {
- numOutBytes = toolame_finish(&outbuf[0], outbuf.size());
+ numOutBytes = toolame_finish(outbuf.data(), outbuf.size());
}
}