ispc 1.1.4 is released

[Matt Pharr]

As always, binaries are available from http://ispc.github.com/downloads.html, and source is available from that page and the main github page, https://github.com/ispc/ispc/.

There are two major bugfixes for Windows in this release.  First, a number of failures in AVX code generation on Windows have been fixed; AVX on Windows now has no known issues.  Second, a longstanding bug in parsing 64-bit integer constants on Windows has been fixed.

This release features a new experimental scalar target, contributed by Gabe Weisz <gwe...@cs.cmu.edu>.  This target ("--target=generic-1") compiles gangs of single program instances (i.e. programCount == 1); it can be useful for debugging ispc programs.

The compiler now supports dynamic memory allocation in ispc programs (with "new" and "delete" operators based on those in C++).  See  in the documentation for more information.

ispc now performs "short circuit" evaluation of the || and && logical operators and the ? : selection operator.  (This represents the correction of a major incompatibility with C.)  Code like "(index < arraySize && array[index] == 1)" thus now executes as in C, where "array[index]" won't be evaluated unless "index" is less than "arraySize".

The standard library now provides "local" atomic operations, which are atomic across the gang of program instances (but not across other gangs or other hardware threads.  See the updated documentation on atomics for more information.

The standard library now offers a clock() function, which returns a uniform int64 value that counts processor cycles; it can be used for fine-resolution timing measurements.

Finally (of limited interest now): ispc now supports the forthcoming AVX2 instruction set, due with Haswell-generation CPUs.  All tests and examples compile and execute correctly with AVX2.  (Thanks specifically to Craig Topper and Nadav Rotem for work on AVX2 support in LLVM, which made this possible.)