Don't create ispc-callable symbols for other functions that we find in the LLVM
bitcode files that are loaded up and linked into the module so that they can be
called from ispc stdlib functions. This fixes an issue where we had a clash
between the declared versions of double sin(double) and the corresponding
ispc stdlib routines for uniform doubles, which in turn led to bogus code
being generated for calls to those ispc stdlib functions.
- In the ispc-generated header files, a #define now indicates which compilation target
was used.
- The examples use utility routines from the new file examples/cpuid.h to check the
system's CPU's capabilities to see if it supports the ISA that was used for
compiling the example code and print error messages if things aren't going to
work...
This way, we match C/C++ in that casting a bool to an int gives either the value
zero or the value one. There is a new stdlib function int sign_extend(bool)
that does sign extension for cases where that's desired.
MSVC 2010 issues an error if given a string larger than 64k characters
long. To work around this, the pre-processed stdlib.ispc code is now
stored as an array of characters terminated with a NUL (i.e. the same thing
in the end); MSVC is fine with arrays larger than 64k characters.
Fixed the implementations of these builtin functions for targets that don't have native masked load instructions so that they do no loads if the vector mask is all off, and only do an (unaligned) vector load if both the first and last element of the mask are on. Otherwise they serialize and do scalar loads for only the active lanes. This fixes a number of potential sources of crashes due to accessing invalid memory.
lex.ll(397): warning C4244: '=' : conversion from 'long' to 'char', possible loss of data
lex.ll(402): warning C4244: '=' : conversion from 'long' to 'char', possible loss of data
by explicit cast to 'char'.
See: http://msdn.microsoft.com/en-us/library/w4z2wdyc(v=vs.80).aspx
long strtol(
const char *nptr,
char **endptr,
int base
);
Add much more suppport for doubles and in64 types in the standard library, basically supporting everything for them that are supported for floats and int32s. (The notable exceptions being the approximate rcp() and rsqrt() functions, which don't really have sensible analogs for doubles (or at least not built-in instructions).)
Fixes bug #55. A number of tests were crashing on Windows due to the task
launch code using alloca to allocate space for the tasks' parameters. On
Windows, the stack isn't generally big enough for this to be a good idea.
Also added an alignment parmaeter to ISPCMalloc() to pass the alignment
requirement along.
A few more productions to recover from parse errors (in function parameter lists and in statement lists). These eliminate some of the massive cascading error messages from a single parse error that the previous error recovery strategy would sometimes cause. Fixes issue #44.
This checkin provides the standard set of atomic operations and a memory barrier in the ispc standard library. Both signed and unsigned 32- and 64-bit integer types are supported.
Link the appropriate clang libraries to make the preprocessor
stuff work on Windows builds. Also updated the solution files
for the examples to stop using cl.exe for preprocessing but to
just call ispc directly. Finishes fixes for issue #32.
When creating function Symbols for functions that were defined in LLVM bitcode for the standard library, if any of the function parameters are integer types, create two ispc-side Symbols: one where the integer types are all signed and the other where they are all unsigned. This allows us to provide, for example, both store_to_int16(reference int a[], uniform int offset, int val) as well as store_to_int16(reference unsigned int a[], uniform int offset, unsigned int val). functions.
Added some additional tests to exercise the new variants of these.
Also fixed some cases where the __{load,store}_int{8,16} builtins would read from/write to memory even if the mask was all off (which could cause crashes in some cases.)
- Call SSE versions for all the various scalar intrinsics
- Fix names of many (all?) AVX intrinsics; all were missing .256 suffix, others had additional issues.
scalar values (that ispc used to smear across the array/struct
elements). Now, initializers in variable declarations must be
{ }-delimited lists, with one element per struct member or array
element, respectively.
There were a few problems with the previous implementation of the
functionality to initialize from scalars. First, the expression
would be evaluated once per value initialized, so if it had side-effects,
the wrong thing would happen. Next, for large multidimensional arrays,
the generated code would be a long series of move instructions, rather
than loops (and this in turn made LLVM take a long time.)
While both of these problems are fixable, it's a non-trivial
amount of re-plumbing for a questionable feature anyway.
Fixes issue #50.
build to handle recent change to API. If building with LLVM tot, a
version starting with or after this change must be used:
commit 276365dd4bc0c2160f91fd8062ae1fc90c86c324
Author: Evan Cheng <evan.cheng@apple.com>
Date: Thu Jun 30 01:53:36 2011 +0000
Fix the ridiculous SubtargetFeatures API where it implicitly expects CPU name to
be the first encoded as the first feature. It then uses the CPU name to look up
features / scheduling itineray even though clients know full well the CPU name
being used to query these properties.
The fix is to just have the clients explictly pass the CPU name!
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@134127 91177308-0d34-0410-b5e6-96231b3b80d8
