split cpu+gpu work queues

Split work queues for CPU and GPU, assigning using a --backend
whitelist.  This lets us up the CPU process limit to NumCPUs,
while keeping the GPU process limit at 8.

On my desktop this speeds up this mixed workload about 2x:

cat <<EOF | go run tools/fm/fm_bot.go -quiet -script - out/fm
    b=cpu gms ct=8888 w=$WHERE/8888 legacy=true  # old and busted
    b=cpu gms ct=8888 w=$WHERE/srgb              # new hotness
    b=gl  gms ct=8888 w=$WHERE/gl
    b=mtl gms ct=8888 w=$WHERE/mtl
    b=cpu gms ct=565  w=$WHERE/565
    b=cpu gms ct=f16  w=$WHERE/f16
    b=skp gms         w=$WHERE/skp
EOF

Before:        44.37 real       303.87 user        15.40 sys
After:         21.87 real       480.05 user        27.10 sys

Change-Id: Ibea4537c8cdf278f8e4621f26ebacd942952eec3
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/205833
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: Mike Klein <mtklein@google.com>
1 file changed