LLVM with the emscripten fastcomp javascript backend https://git.amat.us/emscripten-fastcomp/atom/test/NaCl/Localmods?h=master 2013-08-30T23:42:36Z 2013-08-30T23:42:36Z Jan Voung jvoung@chromium.org 2013-08-30T23:42:36Z urn:sha1:121830a16cdb2685b1ac49bb88407644c044ef30 Localmods came from: https://codereview.chromium.org/10825082/, and earlier. (1) The original change was so that byval parameters always go on the stack. That part was added because the original ARM code was buggy, and did not actually make a copy of the value, modifying the caller's struct (ouch!). (2) Then came a localmod to make all arguments following a byval go on the stack and to make the var-args code aware of that. This is so that arguments stay in the correct order for var-args to pick up. For (1) there has been some work upstream to make it work better. In any case, clang with --target=armv7a-...-gnueabi only used byval in some limited cases -- when the size of the struct is > 64 bytes where the backend will know that part of it could be in regs, and the rest can be memcpy'ed to the stack. For le32, clang will still generate byval without satisfying the same ARM condition (only for structs bigger than 64 bytes), so it could be *very bad* if we didn't have the ABI simpification passes rewrite the byval and try to let the ARM backend do things with byval... TEST=the GCC torture tests: va-arg-4.c, and 20030914-2.c and the example in issue 2746 still pass. BUG=none, cleanup R=dschuff@chromium.org Review URL: https://codereview.chromium.org/23691009 2013-07-18T14:57:42Z Eli Bendersky eliben@chromium.org 2013-07-18T14:57:42Z urn:sha1:0592f0ef8bbc825243cb1860b09468a7605f8716 2013-07-09T22:29:37Z Eli Bendersky eliben@chromium.org 2013-07-09T22:29:37Z urn:sha1:299c82a21a9fd010534ea59c46ae0b88c21f25b7 GlobalOpt currently assumes only an external "main" is the "real main". This is no longer the case for PNaCl, where we internalize "main". Make the test more strict and PNaCl specific by checking that "main" is just used once - in a call from "_start", but does not have to be external. Note that this also addresses a possible bug in the optimization for C code, since C does not guarantee that main is not recursive. This CL's purpose is to address a SPEC performance regression - 10% in 183.equake. The regression appeared after our ABI change that made 'main' internal, which disabled this particular optimization. The CL addresses this by re-enabling the optimization and also being more C-standard conforming. BUG=None Review URL: https://codereview.chromium.org/18615015