Saturation when moving data from 32b to 16b

In the IVAS fix point code, there is this function

Copy_Scale_sig_32_16()

which moves data from 32 bits to 16 bits vector. By default, this function allows for saturation. I went into it and checked is saturation could happen there and to what level it could.

Here is a very small subset of what I found :

Value (x[i])			Left shift	Multiple of	MAX_16		File		Line	Frame
x[2] =	-1508	tmp =	31	1508.05	x MAX_16	OVF in	acelp_core_dec_ivas_fx	at Ln	2274	at frame	5746
x[0] =	-5692	tmp =	29	1423.04	x MAX_16	OVF in	acelp_core_dec_ivas_fx	at Ln	2274	at frame	6492
x[0] =	-5261	tmp =	29	1315.29	x MAX_16	OVF in	acelp_core_dec_ivas_fx	at Ln	2274	at frame	6486
x[10] =	1163	tmp =	31	1163.04	x MAX_16	OVF in	acelp_core_dec_ivas_fx	at Ln	2274	at frame	5736
x[4] =	4514	tmp =	29	1128.53	x MAX_16	OVF in	acelp_core_dec_ivas_fx	at Ln	2274	at frame	6485
x[4] =	4299	tmp =	29	1074.78	x MAX_16	OVF in	acelp_core_dec_ivas_fx	at Ln	2274	at frame	6553
x[10] =	-4291	tmp =	29	1072.78	x MAX_16	OVF in	acelp_core_dec_ivas_fx	at Ln	2274	at frame	6528
x[10] =	1070	tmp =	31	1070.03	x MAX_16	OVF in	acelp_core_dec_ivas_fx	at Ln	2274	at frame	6399
x[6] =	4052	tmp =	29	1013.03	x MAX_16	OVF in	acelp_core_dec_ivas_fx	at Ln	2274	at frame	6509
x[10] =	3968	tmp =	29	992.03	x MAX_16	OVF in	acelp_core_dec_ivas_fx	at Ln	2274	at frame	6527
x[0] =	72136448	tmp =	5	1.07	x MAX_16	OVF in	ivas_core_dec_fx	at Ln	933	at frame	5733
x[208] =	-1234	tmp =	21	1.21	x MAX_16	OVF in	ivas_core_dec_fx	at Ln	1215	at frame	2126
x[11] =	371169539	tmp =	3	1.38	x MAX_16	OVF in	ivas_hq_core_dec_fx	at Ln	927	at frame	1789

When the Multiple of is close to 1.0, it could be normal clipping, but when it is above 1.1 (meaning the value is 10% bigger than MAX_16) it already looks like the scaling is wrong and here we got some vectors that are way above normal clipping

these overflow were found with ltv_STEREO_32k_-stereo_-dtx_32000_32.zip