diff --git a/src/lib/newchecksum.dasl b/src/lib/newchecksum.dasl
index 19ff4c78c7..6487e27c9f 100644
--- a/src/lib/newchecksum.dasl
+++ b/src/lib/newchecksum.dasl
@@ -146,12 +146,15 @@ function selftest ()
    end
    local function benchmark (fn, times)
       local now = os.clock()
-      local temp
+      local csum
       for i=1,times do
-         temp = fn()
+         csum, pkt = fn()
       end
-      local ret = {os.clock() - now, temp}
-      return ret[1]
+      local elapse = os.clock() - now
+      local ns_per_byte = elapse * 10e9 / pkt.length
+      local ns_per_csum = elapse * 10e9 / times
+      local ret = {elapse = elapse, ns_per_byte = ns_per_byte, ns_per_csum = ns_per_csum, csum = csum}
+      return ("elapse: %.6f; ns_per_byte: %.2f; ns_per_csum: %.2f"):format(ret.elapse, ret.ns_per_byte, ret.ns_per_csum)
    end
    local function hex (num)
       return ("0x%.2x"):format(num)
@@ -167,10 +170,10 @@ function selftest ()
    assert(hex(ntohs(newchecksum(pkt.data, pkt.length))) == hex(ntohs(checksum_lua(pkt.data, pkt.length))))
    assert(hex(ntohs(newchecksum(pkt.data, pkt.length))) == hex(C.cksum_generic(pkt.data, pkt.length, 0)))
    -- Benchmark for different architectures.
-   print("Gen: ", benchmark(function() return C.cksum_generic(pkt.data, pkt.length, 0) end, times))
-   print("SSE2: ", benchmark(function() return C.cksum_sse2(pkt.data, pkt.length, 0) end, times))
-   print("AVX2: ", benchmark(function() return C.cksum_avx2(pkt.data, pkt.length, 0) end, times))
-   print("New: ", benchmark(function() return newchecksum(pkt.data, pkt.length) end, times))
+   print("Gen: ", benchmark(function() return C.cksum_generic(pkt.data, pkt.length, 0), pkt end, times))
+   print("SSE2: ", benchmark(function() return C.cksum_sse2(pkt.data, pkt.length, 0), pkt end, times))
+   print("AVX2: ", benchmark(function() return C.cksum_avx2(pkt.data, pkt.length, 0), pkt end, times))
+   print("New: ", benchmark(function() return newchecksum(pkt.data, pkt.length), pkt end, times))
 
    size = 550
    print("2M; "..size.." bytes")
@@ -180,10 +183,10 @@ function selftest ()
    assert(hex(ntohs(newchecksum(pkt.data, pkt.length))) == hex(ntohs(checksum_lua(pkt.data, pkt.length))))
    assert(hex(ntohs(newchecksum(pkt.data, pkt.length))) == hex(C.cksum_generic(pkt.data, pkt.length, 0)))
    -- Benchmark for different architectures.
-   print("Gen: ", benchmark(function() return C.cksum_generic(pkt.data, pkt.length, 0) end, times))
-   print("SSE2: ", benchmark(function() return C.cksum_sse2(pkt.data, pkt.length, 0) end, times))
-   print("AVX2: ", benchmark(function() return C.cksum_avx2(pkt.data, pkt.length, 0) end, times))
-   print("New: ", benchmark(function() return newchecksum(pkt.data, pkt.length) end, times))
+   print("Gen: ", benchmark(function() return C.cksum_generic(pkt.data, pkt.length, 0), pkt end, times))
+   print("SSE2: ", benchmark(function() return C.cksum_sse2(pkt.data, pkt.length, 0), pkt end, times))
+   print("AVX2: ", benchmark(function() return C.cksum_avx2(pkt.data, pkt.length, 0), pkt end, times))
+   print("New: ", benchmark(function() return newchecksum(pkt.data, pkt.length), pkt end, times))
 
    size = 1500
    print("1M; "..size.." bytes")
@@ -193,8 +196,8 @@ function selftest ()
    assert(hex(ntohs(newchecksum(pkt.data, pkt.length))) == hex(ntohs(checksum_lua(pkt.data, pkt.length))))
    assert(hex(ntohs(newchecksum(pkt.data, pkt.length))) == hex(C.cksum_generic(pkt.data, pkt.length, 0)))
    -- Benchmark for different architectures.
-   print("Gen: ", benchmark(function() return C.cksum_generic(pkt.data, pkt.length, 0) end, times))
-   print("SSE2: ", benchmark(function() return C.cksum_sse2(pkt.data, pkt.length, 0) end, times))
-   print("AVX2: ", benchmark(function() return C.cksum_avx2(pkt.data, pkt.length, 0) end, times))
-   print("New: ", benchmark(function() return newchecksum(pkt.data, pkt.length) end, times))
+   print("Gen: ", benchmark(function() return C.cksum_generic(pkt.data, pkt.length, 0), pkt end, times))
+   print("SSE2: ", benchmark(function() return C.cksum_sse2(pkt.data, pkt.length, 0), pkt end, times))
+   print("AVX2: ", benchmark(function() return C.cksum_avx2(pkt.data, pkt.length, 0), pkt end, times))
+   print("New: ", benchmark(function() return newchecksum(pkt.data, pkt.length), pkt end, times))
 end