On Thu, Jan 10, 2013 at 4:30 PM, minux wrote:

On Thu, Jan 10, 2013 at 11:26 PM, Patrick Gundlach wrote:

...

Luigi,

this looks very promising!!! I've been able to run a simple document with my publisher with it. Great.

Just a question: there seems to be a new bit library included, right? Is this documented anywhere? Is the library itself documented?

It has these functions:

bit = { ["rol"] = "function: builtin#71" ["rshift"] = "function: builtin#69" ["ror"] = "function: builtin#72" ["bswap"] = "function: builtin#67" ["bxor"] = "function: builtin#75" ["bor"] = "function: builtin#74" ["arshift"] = "function: builtin#70" ["bnot"] = "function: builtin#66" ["tobit"] = "function: builtin#65" ["lshift"] = "function: builtin#68" ["tohex"] = "function: builtin#76" ["band"] = "function: builtin#73" }

http://luajit.org/extensions.html

This is the standard bit library of luajit but there is also bit32 Why two bit library ? bit32 is the bit lib of lua5.2, so it helps the programmer You can check bit32 with the following code (from H. Hagen). Please, note that I've not checked that the bit gives the same results -- but I would like to know, of course. print("testing bitwise operations") assert(bit32.band() == bit32.bnot(0)) assert(bit32.btest() == true) assert(bit32.bor() == 0) assert(bit32.bxor() == 0) assert(bit32.band() == bit32.band(0xffffffff)) assert(bit32.band(1,2) == 0) -- out-of-range numbers assert(bit32.band(-1) == 0xffffffff) assert(bit32.band(2^33 - 1) == 0xffffffff) assert(bit32.band(-2^33 - 1) == 0xffffffff) assert(bit32.band(2^33 + 1) == 1) assert(bit32.band(-2^33 + 1) == 1) assert(bit32.band(-2^40) == 0) assert(bit32.band(2^40) == 0) assert(bit32.band(-2^40 - 2) == 0xfffffffe) assert(bit32.band(2^40 - 4) == 0xfffffffc) assert(bit32.lrotate(0, -1) == 0) assert(bit32.lrotate(0, 7) == 0) assert(bit32.lrotate(0x12345678, 4) == 0x23456781) assert(bit32.rrotate(0x12345678, -4) == 0x23456781) assert(bit32.lrotate(0x12345678, -8) == 0x78123456) assert(bit32.rrotate(0x12345678, 8) == 0x78123456) assert(bit32.lrotate(0xaaaaaaaa, 2) == 0xaaaaaaaa) assert(bit32.lrotate(0xaaaaaaaa, -2) == 0xaaaaaaaa) for i = -50, 50 do assert(bit32.lrotate(0x89abcdef, i) == bit32.lrotate(0x89abcdef, i%32)) end assert(bit32.lshift(0x12345678, 4) == 0x23456780) assert(bit32.lshift(0x12345678, 8) == 0x34567800) assert(bit32.lshift(0x12345678, -4) == 0x01234567) assert(bit32.lshift(0x12345678, -8) == 0x00123456) assert(bit32.lshift(0x12345678, 32) == 0) assert(bit32.lshift(0x12345678, -32) == 0) assert(bit32.rshift(0x12345678, 4) == 0x01234567) assert(bit32.rshift(0x12345678, 8) == 0x00123456) assert(bit32.rshift(0x12345678, 32) == 0) assert(bit32.rshift(0x12345678, -32) == 0) assert(bit32.arshift(0x12345678, 0) == 0x12345678) assert(bit32.arshift(0x12345678, 1) == 0x12345678 / 2) assert(bit32.arshift(0x12345678, -1) == 0x12345678 * 2) assert(bit32.arshift(-1, 1) == 0xffffffff) assert(bit32.arshift(-1, 24) == 0xffffffff) assert(bit32.arshift(-1, 32) == 0xffffffff) assert(bit32.arshift(-1, -1) == (-1 * 2) % 2^32) print("+") -- some special cases local c = {0, 1, 2, 3, 10, 0x80000000, 0xaaaaaaaa, 0x55555555, 0xffffffff, 0x7fffffff} for _, b in pairs(c) do assert(bit32.band(b) == b) assert(bit32.band(b, b) == b) assert(bit32.btest(b, b) == (b ~= 0)) assert(bit32.band(b, b, b) == b) assert(bit32.btest(b, b, b) == (b ~= 0)) assert(bit32.band(b, bit32.bnot(b)) == 0) assert(bit32.bor(b, bit32.bnot(b)) == bit32.bnot(0)) assert(bit32.bor(b) == b) assert(bit32.bor(b, b) == b) assert(bit32.bor(b, b, b) == b) assert(bit32.bxor(b) == b) assert(bit32.bxor(b, b) == 0) assert(bit32.bxor(b, 0) == b) assert(bit32.bnot(b) ~= b) assert(bit32.bnot(bit32.bnot(b)) == b) assert(bit32.bnot(b) == 2^32 - 1 - b) assert(bit32.lrotate(b, 32) == b) assert(bit32.rrotate(b, 32) == b) assert(bit32.lshift(bit32.lshift(b, -4), 4) == bit32.band(b, bit32.bnot(0xf))) assert(bit32.rshift(bit32.rshift(b, 4), -4) == bit32.band(b, bit32.bnot(0xf))) for i = -40, 40 do assert(bit32.lshift(b, i) == math.floor((b * 2^i) % 2^32)) end end assert(not pcall(bit32.band, {})) assert(not pcall(bit32.bnot, "a")) assert(not pcall(bit32.lshift, 45)) assert(not pcall(bit32.lshift, 45, print)) assert(not pcall(bit32.rshift, 45, print)) print("+") -- testing extract/replace assert(bit32.extract(0x12345678, 0, 4) == 8) assert(bit32.extract(0x12345678, 4, 4) == 7) assert(bit32.extract(0xa0001111, 28, 4) == 0xa) assert(bit32.extract(0xa0001111, 31, 1) == 1) assert(bit32.extract(0x50000111, 31, 1) == 0) assert(bit32.extract(0xf2345679, 0, 32) == 0xf2345679) assert(not pcall(bit32.extract, 0, -1)) assert(not pcall(bit32.extract, 0, 32)) assert(not pcall(bit32.extract, 0, 0, 33)) assert(not pcall(bit32.extract, 0, 31, 2)) assert(bit32.replace(0x12345678, 5, 28, 4) == 0x52345678) assert(bit32.replace(0x12345678, 0x87654321, 0, 32) == 0x87654321) assert(bit32.replace(0, 1, 2) == 2^2) assert(bit32.replace(0, -1, 4) == 2^4) assert(bit32.replace(-1, 0, 31) == 2^31 - 1) assert(bit32.replace(-1, 0, 1, 2) == 2^32 - 7) print'OK' -- luigi

On Thu, Jan 10, 2013 at 5:20 PM, Hans Hagen wrote:

On 1/10/2013 4:26 PM, Patrick Gundlach wrote:

...

Luigi,

this looks very promising!!! I've been able to run a simple document with my publisher with it. Great.

Are you using clang or gcc ? Did you checked some C module -- i.e. the binding for libxml ? Remember that it's lua5.1 !!!

PS the really fun for you is the ffi module... -- luigi

On Thu, Jan 10, 2013 at 8:08 PM, Patrick Gundlach wrote:

not yet. I will need to make my publisher run with luajittex smoothly first. But I am lying in bed and unable to do serious work. sorry to stress you.. I've added a check of Darwin X86_64, so that the export LDFLAGS=.. should be not needed anymore.. can you check ? (It assumes the gcc compiler)

-- luigi

On 1/11/2013 6:29 PM, luigi scarso wrote:

I've added the script buildjit-clang.sh

I've checked under linux 64bit, and it works even if there are some warnings on luajit. The performances are slightly different vs gcc (gcc is better) but nothing of important.

btw, we measured just a few % slower processing on our benchmarks so it's neglectable, but then: using -O3 on gcc is also faster so speed is hard to define Hans ----------------------------------------------------------------- Hans Hagen | PRAGMA ADE Ridderstraat 27 | 8061 GH Hasselt | The Netherlands tel: 038 477 53 69 | voip: 087 875 68 74 | www.pragma-ade.com | www.pragma-pod.nl -----------------------------------------------------------------