llvm-svn: 162900

The dependency printing was accidentally removed in during a previous
restructuring.

llvm-svn: 162662

llvm-svn: 162555

llvm-svn: 162554

llvm-svn: 162553

Before we defined GPGPU_CODEGEN to '0', which does not disable the relevant code
as we just check if that value is defined at all. We now follow the cmake
approach and only define GPGPU_CODEGEN, if the feature should be enabled.

Reported by: Sebastian Pop <spop@codeaurora.org>

llvm-svn: 162275

llvm-svn: 161930

Added a file that explains how to load Polly in dragonegg.
Also fixed a typo in the document for clang.

Committed with a typo fix and a change to make this website available from
the documentation section.

Contributed by: Sameer Sahasrabuddhe  <Sameer.Sahasrabuddhe@amd.com>

llvm-svn: 161928

Translate the selected parallel loop body into a ptx string and run it with the
cuda driver API. We limit this preliminary implementation to target the
following special test cases:

  - Support only 2-dimensional parallel loops with or without only one innermost
    non-parallel loop.
  - Support write memory access to only one array in a SCoP.

The patch was committed with smaller changes to the build system:

There is now a flag to enable gpu code generation explictly. This was required
as we need the llvm.codegen() patch applied on the llvm sources, to compile this
feature correctly. Also, enabling gpu code generation does not require cuda.
This requirement was removed to allow 'make polly-test' runs, even without an
installed cuda runtime.

Contributed by:  Yabin Hu  <yabin.hwu@gmail.com>

llvm-svn: 161239

This fixes a conflict between polly::createIndVarSimplifyPass() and
llvm::createIndVarSimplifyPass(), which causes problems on windows.

Reported by:  Michael Kruse  <MichaelKruse@meinersbur.de

llvm-svn: 161235

The Apple linker fails by default, if some function calls can not be resolved at
link time. However, all functions that are part of LLVM itself will not be
linked into Polly, but will be provided by the compiler that Polly is loaded
into.  Hence, during linking we need to ignore failures due to unresolved
function calls.

llvm-svn: 161234

Contributed by: Michael Kruse  <MichaelKruse@meinersbur.de>

llvm-svn: 161174

Contributed by:  Yabin Hu <yabin.hwu@gmail.com>

llvm-svn: 161160

llvm-svn: 161158

llvm-svn: 161157

  operand tree.

This patch fix Bug 13491, and the original "FIXME" in IndependentBlocks.cpp.

Patched by Kevin Fan<kevin.fan@gmail.com>.

llvm-svn: 161105

Otherwise the script spams the home directory and, in case there are folders
of previous attempts lying around, it may fail in some unexpected way.

llvm-svn: 160677

Cast instruction do not have side effects and can consequently be part of a
scop. We special cased them earlier, as they may be problematic within array
subscripts or loop bounds. However, the scalar evolution validator already
checks for them such that there is no need to also check the instructions within
the basic blocks.  Checking them is actually overly conservative as the precence
of casts may invalidate a scop, even though scalar evolution is not influenced
by it.

llvm-svn: 160261

I did not take into account, that this patch fails to compile without the
llvm.codegen patch applied. This breaks buildbots.

I revert this until we found a solution to commit this without buildbots
complaining.

This reverts commit cb43ab80e94434e780a66be3b9a6ad466822fe33.

llvm-svn: 160165

Translate the selected parallel loop body into a ptx string and run it
with cuda driver API. We limit this preliminary implementation to
target the following special test cases:
  - Support only 2-dimensional parallel loops with or without only one
    innermost non-parallel loop.
  - Support write memory access to only one array in a SCoP.

Contributed by:  Yabin Hu <yabin.hwu@gmail.com>

llvm-svn: 160164

Contributed by: Yabin Hu  <yabin.hwu@gmail.com>

llvm-svn: 159815

  Patched by TangKK <dengjunqi06323011@hotmail.com>.

llvm-svn: 159808

  Patched by JunQi<dengjunqi06323011@hotmail.com>.

llvm-svn: 159735

llvm-svn: 159734

Contributed by:  Yabin Hu  <yabin.hwu@gmail.com>

llvm-svn: 159725

llvm-svn: 159650

Contributed by: Yabin Hu <yabin.hwu@gmail.com>

llvm-svn: 159649

Contributed by: Yabin Hu <yabin.hwu@gmail.com>

llvm-svn: 159507

llvm-svn: 159423

llvm-svn: 158987

llvm-svn: 158306

Contributed by: Yabin Hu <yabin.hwu@gmail.com>

llvm-svn: 158304

llvm-svn: 158081

We will use the cuda library for the upcoming automatic GPGPU code generation.

Contributed by: Yabin Hu  <yabin.hwu@gmail.com>

llvm-svn: 158064

CLooG and the CLooG based code generation does not yet correctly derive the
types of the expressions, but just uses i64 for everything. This is incorrect,
but works normally pretty well. However, the recent change of adding parameter
bounds to the context made CLooG generate expressions that contain a lot of very
large integers that possibly don't fit into an i64. This broke the code
generation for several benchmarks.

To get the CLooG based code generation working again, we just don't take into
account any constraints in the context. This brings us back to the theoretical
incorrect, but in practice generally correct code.

The next step will be the isl based code generation. Here we will derive
automatically correct types.

llvm-svn: 158015

llvm-svn: 157690

The "llvm.codegen" intrinsic patch is a patch to LLVM, which is used
to generate code for embedded LLVM-IR strings. In Polly, we use it
to generate ptx assembly text for GPGPU code generation.

llvm-svn: 157689

Store a pointer to each ScopStmt in the isl_id associated with the space of its
domain. This will later allow us to recover the statement during code
generation with isl.

llvm-svn: 157607

In general, all code that we produce is NSW.

llvm-svn: 157606

We will reuse this function for the isl code generator.

llvm-svn: 157605