lib.ctable/lib.hash.siphash: optimize traces

Remove some redundant casts that were introducing unsunk cnewi.

Reformulate make_multi_hash optimal path to reduce side trace generation.

src/lib/ctable.lua

@@ -509,7 +509,6 @@ end
 function LookupStreamer:stream()
    local width = self.width
    local entries = self.entries
-   local keys = self.keys
    local pointers = self.pointers
    local stream_entries = self.stream_entries
    local entries_per_lookup = self.entries_per_lookup

src/lib/hash/siphash.dasl

@@ -576,8 +576,6 @@ local function make_hash2(opts)
    local hash = make_hash1(opts)
    local stride = opts.stride or opts.size
    return function(ptr, result)
-      ptr = ffi.cast('uint8_t*', ptr)
-      result = ffi.cast('uint32_t*', result)
       result[0] = hash(ptr)
       result[1] = hash(ptr + stride)
    end
@@ -587,8 +585,6 @@ local function make_hash4(opts)
    local hash = make_hash2(opts)
    local stride = opts.stride or opts.size
    return function(ptr, result)
-      ptr = ffi.cast('uint8_t*', ptr)
-      result = ffi.cast('uint32_t*', result)
       hash(ptr, result)
       hash(ptr + stride*2, result + 2)
    end
@@ -612,19 +608,13 @@ function make_multi_hash(opts)
 
    if width % 4 == 0 then
       return function(input, output)
-         input = ffi.cast('uint8_t*', input)
-         output = ffi.cast('uint32_t*', output)
-         for i=1,width,4 do
-            hash4(input, output)
-            input = input + stride*4
-            output = output + 4
+         for i=0,width-1,4 do
+            hash4(input + stride*i, output + i)
          end
       end
    end
 
    return function(input, output)
-      input = ffi.cast('uint8_t*', input)
-      output = ffi.cast('uint32_t*', output)
       for i=1,bit.rshift(width, 2) do
          hash4(input, output)
          input = input + stride*4