Intel SPMD Program Compiler Developers Google Group

Re: optimization

gabrieltan...@gmail.com (Gabriel Tanase) — Fri, 21 Dec 2012 16:21:00 UT

Hi Matt,
thanks for your answer. My first thought was also to implement a compiler
optimization pass but due to lack of llvm knowledge I actually implemented
FMA inside the cbackend.cpp. Here is my patch; Pls let me know if you see
any drawback to this approach:
diff --git a/cbackend.cpp b/cbackend.cpp

Re: optimization

matt.ph...@gmail.com (Matt Pharr) — Tue, 18 Dec 2012 00:04:51 UT

You'd probably want to add another optimization pass to the ispc C++
backend; note in WriteCXXFile() at the bottom of cbackend.cpp it creates a
pass manager and then adds a bunch of optimizations passes there before
generating C++ code.

You'd probably want to add a BasicBlockPass (
[link]) that in turn iterated over all

optimization

gabrieltan...@gmail.com (Gabriel Tanase) — Mon, 17 Dec 2012 19:19:30 UT

Hi Everybody,
I recently started using ISPC to generate SIMD code for the IBM Power
architecture :). I do this by generating code for generic-4 target and
replace generic-4.h with my own header file where I use power gcc
intrinsics.
However recently I started to think on how I can extend ISPC to add

Re: Interface to compiler backend.

matt.ph...@gmail.com (Matt Pharr) — Mon, 03 Dec 2012 04:35:35 UT

Unfortunately the system wasn't designed to have a clean way to interface with it in that way. (It certainly should have been, but there's always that tension between getting it out the door sooner to do one thing well, versus getting it out later but better tuned to do multiple things.)

As such, your choices would be either to generate ispc code (via print statements or what have you), or to create ispc ASTs directly. To see how to do that, the best way is probably just to look at the parser and see how it creates the AST from the source program; it should be pretty straightforward.

Interface to compiler backend.

bo.joel.svens...@gmail.com (Joel Svensson) — Thu, 29 Nov 2012 10:25:12 UT

Hello.
I previously used the Intel ArBB system as a code generating backend of a
language embedded in Haskell. I am getting a feeling that
ISPC could potentially be a suitable replacement but I have some questions.
Is there any accessible interface to the code generating backend of ISPC?
Generally when doing embedded languages the frontend of the

LLVM issue with "DiagnosticCommonKinds.inc"

wals...@gmail.com (Matt Walsh) — Fri, 20 Jul 2012 23:21:10 UT

Not an ISPC issue per se, but in building it I ran into :
Compiling module.cpp
In file included from /usr/include/clang/Basic/Diagn ostic.h:17,
from /usr/include/clang/Frontend/Di agnosticOptions.h:13,
from /usr/include/clang/Frontend/Co mpilerInvocation.h:20,
from /usr/include/clang/Frontend/Co mpilerInstance.h:13,

ispc v1.3.0 is released

matt.ph...@gmail.com (Matt Pharr) — Fri, 29 Jun 2012 16:14:00 UT

This is a major new release of ispc, with support for more compilation targets and a number of additions to the language. As usual, the quality of generated code has also been improved in a number of cases and a variety of small bugs have been fixed.

New targets:

* This release provides "beta" support for compiling to Intel Xeon Phi (the "Many Integrated Core" architecture).

ispc v1.2.2 is released

matt.ph...@gmail.com (Matt Pharr) — Fri, 20 Apr 2012 23:31:18 UT

Binaries are available from [link]. Note also that building from source with LLVM 2.9 is no longer supported; you must use either LLVM 3.0 or the soon-to-be released LLVM 3.1. (In any case, these generate much better vector code than 2.9 did.)
Some highlights from the release notes:

ispc v1.2.1 is released

matt.ph...@gmail.com (Matt Pharr) — Fri, 06 Apr 2012 23:18:18 UT

Binaries are up at [link] now.
This release contains only minor new functionality and is mostly for many small bugfixes and improvements to error handling and error reporting.
The new functionality that is present is:
* Significantly more efficient versions of the float / half conversion routines are now available in the standard library, thanks to Fabian Giesen.

Re: Hit by assertion "m->errorCount > 0" on trivial code

matt.ph...@gmail.com (Matt Pharr) — Wed, 21 Mar 2012 17:16:07 UT

Thanks for reporting this--the bug is now fixed in top-of-tree, and will be in the next binary release.
The problem is that in the function:
The a[] parameter is an array of varying ints, and varying parameter types aren't allowed in functions that are export'ed to be callable from C/C++. The check for varying parameters on exported functions was missing a case, such that it didn't issue an error when it should have, and then at a later point that assertion hit. Sorry for the trouble!

Hit by assertion "m->errorCount > 0" on trivial code

henkjan.lebb...@gmail.com (Henk-Jan Lebbink) — Wed, 21 Mar 2012 15:07:21 UT

Hi,

I have been playing with ispc 1.2.0, but I seem to get hit this assertion:
"m->errorCount > 0" error check quite often,

Performing Custom Build Tools

1> type.cpp:385: Assertion failed: "m->errorCount > 0"

1> ***

1> *** Please file a bug report at [link]

ispc v1.2.0 is released

matt.ph...@gmail.com (Matt Pharr) — Tue, 20 Mar 2012 21:01:08 UT

This is a major new release of ispc, with a number of significant improvements to functionality, performance, and compiler robustness. It does, however, include three small changes to language syntax and semantics that may require changes to existing programs:
* Syntax for the "launch" keyword has been cleaned up; it's now no longer necessary to bracket the launched function call with angle brackets. (In other words, now use "launch foo();", rather than "launch < foo() >;".)

Re: Compile fails on Ubuntu 12.04

ezsch...@googlemail.com (Enrico Zschemisch) — Sun, 04 Mar 2012 21:01:11 UT

Hi Matt,
oh, I see! My mistake. I have everything installed as e.g. llvm-config-3.0,
llvm-as-3.0. I corrected a thing of these things in the Makefile, but not
all of them.
Works now!
Thanks,
Enrico

Re: Compile fails on Ubuntu 12.04

matt.ph...@gmail.com (Matt Pharr) — Sun, 04 Mar 2012 14:41:48 UT

Hi, Enrico--
I don't think that's an issue related to AVX support on your system; it looks to me like the build script isn't finding the 'llvm-as' program, which should be in your PATH as part of the regular LLVM distribution. What happens if you type "which llvm-as"?
Thanks,
-matt

Re: Compile fails on Ubuntu 12.04

ezsch...@googlemail.com (Enrico Zschemisch) — Sun, 04 Mar 2012 13:26:40 UT

Hi Matt,

thanks for the quick fix!
Now I have the next problem when compiling:
Using LLVM 3.0 from /usr/lib/llvm-3.0/lib
Creating objs/ directory
Compiling objs/builtins-target-avx1.cpp
Creating C++ source from builtins definition file builtins/target-avx1-
x2.ll
Traceback (most recent call last):