Commit Graph

8 Commits

Author SHA1 Message Date
River Riddle c33862b0ed Refactor FunctionAttr to hold the internal function reference by name instead of pointer. The one downside to this is that the function reference held by a FunctionAttr needs to be explicitly looked up from the parent module. This provides several benefits though:
* There is no longer a need to explicitly remap function attrs.
      - This removes a potentially expensive call from the destructor of Function.
      - This will enable some interprocedural transformations to now run intraprocedurally.
      - This wasn't scalable and forces dialect defined attributes to override
        a virtual function.
    * Replacing a function is now a trivial operation.
    * This is a necessary first step to representing functions as operations.

--

PiperOrigin-RevId: 249510802
2019-06-01 19:56:54 -07:00
Thomas Joerg 43547ccfac Verify that kernel functions referenced by gpu.launch_func have a gpu.kernel attribute.
Also extract gpu.launch_func's function attribute name into a constant.

--

PiperOrigin-RevId: 247595352
2019-05-10 19:28:04 -07:00
Thomas Joerg b28bafd74f Add operations to produce block/index ids and dimensions to the gpu dialect.
--

PiperOrigin-RevId: 246976227
2019-05-10 19:23:32 -07:00
MLIR Team 535a845011 Fix the indentation of CHECKs in test/GPU/ops.mlir.
--

PiperOrigin-RevId: 246809906
2019-05-06 08:30:17 -07:00
MLIR Team 465df0d916 Add the "gpu.launch_func" op to the GPU dialect.
The idea is to lower `gpu.launch` operations into `gpu.launch_func` operations by outlining the kernel body into a function, which is closer to the NVVM model.

--

PiperOrigin-RevId: 246806890
2019-05-06 08:30:07 -07:00
Alex Zinenko b7b75cf74c GPU dialect: introduce custom syntax for gpu.launch
This syntax removes boilerplate and verbose list of region arguments in the
    header of the entry block.  It groups operands into segments related to GPU
    blocks, GPU threads as well as the operands that are forwarded to the kernel.
    The two former segments are also used to give names to the region arguments
    that are used for GPU blocks and threads inside the kernel body region.

--

PiperOrigin-RevId: 246792329
2019-05-06 08:29:57 -07:00
Alex Zinenko d3380a504f Change syntax of regions in the generic form of operations
The generic form of operations currently supports optional regions to be
    located after the operation type.  As we are going to add a type to each
    region in a leading position in the region syntax, similarly to functions, it
    becomes ambiguous to have regions immediately after the operation type.  Put
    regions between operands the optional list of successors in the generic
    operation syntax and wrap them in parentheses.  The effect on the exisitng IR
    syntax is minimal since only three operations (`affine.for`, `affine.if` and
    `gpu.kernel`) currently use regions.

--

PiperOrigin-RevId: 246787087
2019-05-06 08:29:48 -07:00
Alex Zinenko aae8a7446e Start GPU Dialect
Define a new dialect related to GPU kernels.  Currently, it only contains a
    single operation for launching a kernel on a three-dimensional grid of thread
    blocks, following a model similar to that of CUDA.  In particular, the body of
    the kernel contains operations executed by each thread and uses region
    arguments to accept thread and block identifiers (similar to how the loop body
    region accepts the induction value).

--

PiperOrigin-RevId: 245713728
2019-05-06 08:20:08 -07:00