Commit Graph

21 Commits

Author SHA1 Message Date
Matt Pharr
3b95452481 Add memcpy(), memmove() and memset() to the standard library.
Issue #183.
2012-03-05 16:09:00 -08:00
Matt Pharr
c152ae3c32 Add single-precision asin() and acos() to stdlib.
Issue #184.
2012-03-05 13:32:13 -08:00
Matt Pharr
5b4673e8eb Fix build with LLVM 2.9. 2012-02-07 08:37:13 -08:00
Matt Pharr
f2fbc168af Scalar target builtins bugfixes.
Typo in __max_varying_double.
Add declarations for half functions.
Use the gen_scatter macro to get the scatter functions.
2012-01-29 13:47:44 -08:00
Gabe Weisz
c67a286aa6 Add support for 1-wide scalar target.
Issue #40.
2012-01-29 06:36:07 -08:00
Matt Pharr
664dc3bdda Add support for "new" and "delete" to the language.
Issue #139.
2012-01-27 14:47:06 -08:00
Matt Pharr
d9c0f9315a Fix generic targets: half conversion functions weren't declared.
(Broken by 1867b5b31).
2012-01-27 14:44:43 -08:00
Matt Pharr
24f58fa16a Update per_lane macro to not use ID for lane number in macro expansion
This was leading to unintended consequences if WIDTH was used in macro code,
which was undesirable.
2012-01-27 09:12:13 -08:00
Matt Pharr
1867b5b317 Use native float/half conversion instructions with the AVX2 target. 2012-01-24 15:33:38 -08:00
Matt Pharr
a5b7fca7e0 Extract constant offsets from gather/scatter base+offsets offset vectors.
When we're able to turn a general gather/scatter into the "base + offsets"
form, we now try to extract out any constant components of the offsets and
then pass them as a separate parameter to the gather/scatter function
implementation.

We then in turn carefully emit code for the addressing calculation so that
these constant offsets match LLVM's patterns to detect this case, such that
we get the constant offsets directly encoded in the instruction's addressing
calculation in many cases, saving arithmetic instructions to do these
calculations.

Improves performance of stencil by ~15%.  Other workloads unchanged.
2012-01-24 14:41:15 -08:00
Matt Pharr
d805e8b183 Add clock() function to standard library.
Also corrected the declaration of num_cores() to return a
uniform value.
2012-01-22 13:05:27 -08:00
Matt Pharr
1bba9d4307 Improve atomic_swap_global() to take advantage of associativity.
We now do a single atomic hardware swap and then effectively do 
swaps between the running program instances such that the result
is the same as if they had happened to run a particular ordering
of hardware swaps themselves.

Also cleaned up __atomic_swap_uniform_* built-in implementations
to not take the mask, which they weren't using anyway.

Finishes Issue #56.
2012-01-20 10:37:33 -08:00
Matt Pharr
d14a2de168 Fix generic code emission when building with LLVM3.0/2.9.
Specifically, don't use vector select for masked store blend there,
but emit a call to a undefined __masked_store_blend_*() functions.

Added implementations of these functions to the sse4.h and generic-16.h
in examples/instrinsics.  (Calls to these will never be generated with
LLVM 3.1).
2012-01-17 23:42:22 -07:00
Matt Pharr
58a0b4a20d Add separate set of builtins for AVX2.
(i.e., stop just reusing the ones for AVX1).

For now the only difference is that the int/uint min/max
functions call the new intrinsic for that.  Once gather is
available from LLVM, that will go here as well.
2012-01-13 14:40:01 -08:00
Matt Pharr
652215861e Update dynamic target dispatch code to support AVX2. 2012-01-12 08:37:18 -08:00
Pierre-Antoine Lacaze
d84cf781da Mingw does not have sysconf, use the msc way of finding processors. 2012-01-09 09:45:40 +01:00
Matt Pharr
234e5cd3e1 Use vector select for masked store blend if building with LLVM3.1 2012-01-04 12:59:03 -08:00
Matt Pharr
f75c94a8f1 Have aos/soa and broadcast/shuffle/rotate functions provided by the target.
The SSE/AVX targets use the old versions from util.m4, but these functions are
now passed through to the generic targets.
2012-01-04 12:59:03 -08:00
Matt Pharr
848a432640 Fix various small things that were broken with single-bit-per-lane masks.
Also small cleanups to declarations, "no captures" added, etc.
2012-01-04 12:59:03 -08:00
Matt Pharr
562d61caff Added masked load optimization pass.
This pass handles the "all on" and "all off" mask cases appropriately.

Also renamed load_masked stuff in built-ins to masked_load for consistency with
masked_store.
2012-01-04 11:51:26 -08:00
Matt Pharr
1d9201fe3d Add "generic" 4, 8, and 16-wide targets.
When used, these targets end up with calls to undefined functions for all
of the various special vector stuff ispc needs to compile ispc programs
(masked store, gather, min/max, sqrt, etc.).

These targets are not yet useful for anything, but are a step toward
having an option to C++ code with calls out to intrinsics.

Reorganized the directory structure a bit and put the LLVM bitcode used
to define target-specific stuff (as well as some generic built-ins stuff)
into a builtins/ directory.

Note that for building on Windows, it's now necessary to set a LLVM_VERSION
environment variable (with values like LLVM_2_9, LLVM_3_0, LLVM_3_1svn, etc.)
2011-12-19 13:46:50 -08:00