As far as I understood the FPKM values from different isoforms are summed together to represent a single FPKM value for a gene. However, cufflinks produces two tracks for one gene:

I run cufflinks twice: with --compatible-hits-norm and without. Reference annotation contained exons and transcripts.

isoforms.fpkm

ENSMUST00000021056.7    -    -    ENSMUSG00000001027.7    Scn4a    -    chr11:106179905-106210703    6598    0.0314068    0.00365992    0    0.0109798    OK
ENSMUST00000174877.1    -    -    ENSMUSG00000001027.7    Scn4a    -    chr11:106210768-106214602    443    0    0    0    0    OK

genes.fpkm

ENSMUSG00000001027.7    -    -    ENSMUSG00000001027.7    Scn4a    -    chr11:106179905-106210703    -    -    0.00365992    0    0.0109798    OK
ENSMUSG00000001027.7    -    -    ENSMUSG00000001027.7    Scn4a    -    chr11:106210768-106214602    -    -    0    0    0    OK

I also run cuffdiff with the same reference annotation and it produced only one track per gene:

ENSMUSG00000001027.7    -    -    ENSMUSG00000001027.7    Scn4a    -    chr11:106179905-106210703    -    -    0.00444207    0    0.0137161    OK    0.00839146    0    0.0227746    OK    0    0    0    OK    0.00973267    0    0.0263986    OK​

Why do cuffdiff and cufflinks produce different results for the same annotation?