static int pa_stream_writeShort(lua_State *L)
{
pa_Stream *stream = NULL;
THShortTensor *data = NULL;
long nelem = 0;
PaError err = 0;
int narg = lua_gettop(L);
if(narg == 2 && luaT_isudata(L, 1, "pa.Stream") && luaT_isudata(L, 2, "torch.ShortTensor"))
{
stream = luaT_toudata(L, 1, "pa.Stream");
data = luaT_toudata(L, 2, "torch.ShortTensor");
}
else
luaL_error(L, "expected arguments: Stream ShortTensor");
if(!stream->id)
luaL_error(L, "attempt to operate on a closed stream");
nelem = THShortTensor_nElement(data);
luaL_argcheck(L, (nelem > 0) && (nelem % stream->noutchannel == 0), 2, "invalid data: number of elements must be > 0 and divisible by the number of channels");
luaL_argcheck(L, stream->outsampleformat & paInt16, 1, "stream does not support short data");
data = THShortTensor_newContiguous(data);
err = Pa_WriteStream(stream->id, THShortTensor_data(data), nelem/stream->noutchannel);
THShortTensor_free(data);
if(err == paOutputUnderflowed)
lua_pushboolean(L, 0);
else if(err == paNoError)
lua_pushboolean(L, 1);
else
pa_checkerror(L, err);
return 1;
}
static void load_array_to_lua(lua_State *L, cnpy::NpyArray& arr){
int ndims = arr.shape.size();
//based on code from mattorch with stride fix
int k;
THLongStorage *size = THLongStorage_newWithSize(ndims);
THLongStorage *stride = THLongStorage_newWithSize(ndims);
for (k=0; k<ndims; k++) {
THLongStorage_set(size, k, arr.shape[k]);
if (k > 0)
THLongStorage_set(stride, ndims-k-1, arr.shape[ndims-k]*THLongStorage_get(stride,ndims-k));
else
THLongStorage_set(stride, ndims-k-1, 1);
}
void * tensorDataPtr = NULL;
size_t numBytes = 0;
if ( arr.arrayType == 'f' ){ // float32/64
if ( arr.word_size == 4 ){ //float32
THFloatTensor *tensor = THFloatTensor_newWithSize(size, stride);
tensorDataPtr = (void *)(THFloatTensor_data(tensor));
numBytes = THFloatTensor_nElement(tensor) * arr.word_size;
luaT_pushudata(L, tensor, luaT_checktypename2id(L, "torch.FloatTensor"));
}else if ( arr.word_size == 8){ //float 64
THDoubleTensor *tensor = THDoubleTensor_newWithSize(size, stride);
tensorDataPtr = (void *)(THDoubleTensor_data(tensor));
numBytes = THDoubleTensor_nElement(tensor) * arr.word_size;
luaT_pushudata(L, tensor, luaT_checktypename2id(L, "torch.DoubleTensor"));
}
}else if ( arr.arrayType == 'i' || arr.arrayType == 'u' ){ // does torch have unsigned types .. need to look
if ( arr.word_size == 1 ){ //int8
THByteTensor *tensor = THByteTensor_newWithSize(size, stride);
tensorDataPtr = (void *)(THByteTensor_data(tensor));
numBytes = THByteTensor_nElement(tensor) * arr.word_size;
luaT_pushudata(L, tensor, luaT_checktypename2id(L, "torch.ByteTensor"));
}else if ( arr.word_size == 2 ){ //int16
THShortTensor *tensor = THShortTensor_newWithSize(size, stride);
tensorDataPtr = (void *)(THShortTensor_data(tensor));
numBytes = THShortTensor_nElement(tensor) * arr.word_size;
luaT_pushudata(L, tensor, luaT_checktypename2id(L, "torch.ShortTensor"));
}else if ( arr.word_size == 4 ){ //int32
THIntTensor *tensor = THIntTensor_newWithSize(size, stride);
tensorDataPtr = (void *)(THIntTensor_data(tensor));
numBytes = THIntTensor_nElement(tensor) * arr.word_size;
luaT_pushudata(L, tensor, luaT_checktypename2id(L, "torch.IntTensor"));
}else if ( arr.word_size == 8){ //long 64
THLongTensor *tensor = THLongTensor_newWithSize(size, stride);
tensorDataPtr = (void *)(THLongTensor_data(tensor));
numBytes = THLongTensor_nElement(tensor) * arr.word_size;
luaT_pushudata(L, tensor, luaT_checktypename2id(L, "torch.LongTensor"));
}
}else{
printf("array type unsupported");
throw std::runtime_error("unsupported data type");
}
// now copy the data
assert(tensorDataPtr);
memcpy(tensorDataPtr, (void *)(arr.data<void>()), numBytes);
}
