You need to run HaplotypeCaller individually for each of your bam files using the "-ERC GVCF" flag. This will produce once gvcf file for each of your bam files. Then you would combine each of the GVCF files produced by HaplotypeCaller (e.g., with gatk's CombineGVCFs tool) into a single GVCF file. Finally, you run GenotypeGVCFs on the combined GVCF file to get your SNPs.

So for example, run Haplotype caller like this for each of your bams:

 gatk --java-options "-Xmx4g" HaplotypeCaller  \
   -R $reference \
   -I input.bam \
   -O output1.g.vcf.gz \
   -ERC GVCF

Then, combine all your gvcf files:

gatk --java-options "-Xmx96g -Xms96g" CombineGVCFs \
-R $reference \
--variant output1.g.vcf.gz \
--variant output2.g.vcf.gz \
--variant output3.g.vcf.gz \
-O combined.g.vcf.gz

Finally, perform joint-genotyping on the combined gvcf:

gatk --java-options "-Xmx96g -Xms96g" GenotypeGVCFs \
 -R $reference \
-V combined.g.vcf.gz \
-O final.vcf.gz

See the Best Practices guide for more detailed information.

run a loop:

ls -1 all/*.sorted.bam | while read BAM
do
   (...)
    samtools index ${OUT_DIR}/${SAMPLE}.gatk.bam
   echo "${OUT_DIR}/${SAMPLE}.gatk.bam" >> all_bams.list
done

gatk --java-options "-Xmx4G" HaplotypeCaller -R ${REF_DIR}/ref.fa -I all_bams.list  -O ${OUT_DIR}/final.vcf

anyway, you'd better use a worklfow manager like snakemake or nextflow.