32 Commits

Author SHA1 Message Date
Stephan Herhut 69040d5b0b [MLIR] Allow for multiple gpu modules during translation.
This change makes the ModuleTranslation threadsafe by locking on the
LLVMContext. Furthermore, we now clone the llvm module into a new
context when compiling to PTX similar to what the OrcJit does.

Differential Revision: https://reviews.llvm.org/D78207
2020-04-16 14:18:31 +02:00
River Riddle 80aca1eaf7 [mlir][Pass] Remove the use of CRTP from the Pass classes
This revision removes all of the CRTP from the pass hierarchy in preparation for using the tablegen backend instead. This creates a much cleaner interface in the C++ code, and naturally fits with the rest of the infrastructure. A new utility class, PassWrapper, is added to replicate the existing behavior for passes not suitable for using the tablegen backend.

Differential Revision: https://reviews.llvm.org/D77350
2020-04-07 14:08:52 -07:00
River Riddle e8f5c072f6 [mlir] Move the testing pass for GpuKernelToCubin to the test/ directory
Summary:
This removes the static pass registration, and also cleans up some lingering technical debt.

Differential Revision: https://reviews.llvm.org/D76554
2020-03-22 03:38:09 -07:00
Mehdi Amini 850cb135a3 Do not build the CUBIN conversion pass when NVPTX Backend isn't configured
This pass would currently build, but fail to run when this backend isn't
linked in. On the other hand, we'd like it to initialize only the NVPTX
backend, which isn't possible if we continue to build it without the
backend available. Instead of building a broken configuration, let's
skip building the pass entirely.

Differential Revision: https://reviews.llvm.org/D74592
2020-02-14 09:33:12 +00:00
Mehdi Amini 7b635880ab Fix MLIR build when the NVPTX target isn't configured
Differential Revision: https://reviews.llvm.org/D74472
2020-02-12 12:38:45 +00:00
Mehdi Amini 308571074c Mass update the MLIR license header to mention "Part of the LLVM project"
This is an artifact from merging MLIR into LLVM, the file headers are
now aligned with the rest of the project.
2020-01-26 03:58:30 +00:00
Tres Popp 9a52ea5cf9 Create a gpu.module operation for the GPU Dialect.
Summary:
This is based on the use of code constantly checking for an attribute on
a model and instead represents the distinct operaion with a different
op. Instead, this op can be used to provide better filtering.

Reverts "Revert "[mlir] Create a gpu.module operation for the GPU Dialect.""

This reverts commit ac446302ca4145cdc89f377c0c364c29ee303be5 after
fixing internal Google issues.

This additionally updates ROCDL lowering to use the new gpu.module.

Reviewers: herhut, mravishankar, antiagainst, nicolasvasilache

Subscribers: jholewinski, mgorny, mehdi_amini, jpienaar, burmako, shauheen, csigg, arpith-jacob, mgester, lucyrfox, aartbik, liufengdb, llvm-commits, mravishankar, rriddle, antiagainst, bkramer

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D72921
2020-01-21 14:05:03 +01:00
Benjamin Kramer 0133cc60e4 Revert "[mlir] Create a gpu.module operation for the GPU Dialect."
This reverts commit 4624a1e8ac. Causing
problems downstream.
2020-01-15 17:52:17 +01:00
Tres Popp 4624a1e8ac [mlir] Create a gpu.module operation for the GPU Dialect.
Summary:
This is based on the use of code constantly checking for an attribute on
a model and instead represents the distinct operaion with a different
op. Instead, this op can be used to provide better filtering.

Reviewers: herhut, mravishankar, antiagainst, rriddle

Reviewed By: herhut, antiagainst, rriddle

Subscribers: liufengdb, aartbik, jholewinski, mgorny, mehdi_amini, rriddle, jpienaar, burmako, shauheen, antiagainst, nicolasvasilache, csigg, arpith-jacob, mgester, lucyrfox, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D72336
2020-01-14 12:05:47 +01:00
Mehdi Amini 56222a0694 Adjust License.txt file to use the LLVM license
PiperOrigin-RevId: 286906740
2019-12-23 15:33:37 -08:00
MLIR Team 62d5b1de45 Adapt code to LLVM API updates.
PiperOrigin-RevId: 280431812
2019-11-14 08:27:19 -08:00
Kazuaki Ishizaki 8bfedb3ca5 Fix minor spelling tweaks (NFC)
Closes tensorflow/mlir#177

PiperOrigin-RevId: 275692653
2019-10-20 00:11:34 -07:00
Alex Zinenko 11d12670da GPUToCUDA: attach CUBIN to the nested module rather than to the function
Originally, we were attaching attributes containing CUBIN blobs to the kernel
function called by `gpu.launch_func`. This kernel is now contained in a nested
module that is used as a compilation unit. Attach compiled CUBIN blobs to the
module rather than to the function since we were compiling the module. This
also avoids duplication of the attribute on multiple kernels within the same
module.

PiperOrigin-RevId: 273497303
2019-10-08 05:11:26 -07:00
Christian Sigg b8676da1fc Outline GPU kernel function into a nested module.
Roll forward of commit 5684a12.

When outlining GPU kernels, put the kernel function inside a nested module. Then use a nested pipeline to generate the cubins, independently per kernel. In a final pass, move the cubins back to the parent module.

PiperOrigin-RevId: 270639748
2019-09-23 03:17:01 -07:00
George Karpenkov 2df646bef6 Automated rollback of commit 5684a12434
PiperOrigin-RevId: 270126672
2019-09-19 14:34:30 -07:00
MLIR Team 5684a12434 Outline GPU kernel function into a nested module.
When outlining GPU kernels, put the kernel function inside a nested module. Then use a nested pipeline to generate the cubins, independently per kernel. In a final pass, move the cubins back to the parent module.

PiperOrigin-RevId: 269987720
2019-09-19 01:51:28 -07:00
MLIR Team 1c73be76d8 Unify error messages to start with lower-case.
PiperOrigin-RevId: 269803466
2019-09-18 07:45:17 -07:00
River Riddle f1b100c77b NFC: Finish replacing FunctionPassBase/ModulePassBase with OpPassBase.
These directives were temporary during the generalization of FunctionPass/ModulePass to OpPass.

PiperOrigin-RevId: 268970259
2019-09-13 13:34:27 -07:00
Jacques Pienaar 79f53b0cf1 Change from llvm::make_unique to std::make_unique
Switch to C++14 standard method as llvm::make_unique has been removed (
https://reviews.llvm.org/D66259). Also mark some targets as c++14 to ease next
integrates.

PiperOrigin-RevId: 263953918
2019-08-17 11:06:03 -07:00
Mehdi Amini 926fb685de Express ownership transfer in PassManager API through std::unique_ptr (NFC)
Since raw pointers are always passed around for IR construct without
implying any ownership transfer, it can be error prone to have implicit
ownership transferred the same way.
For example this code can seem harmless:

  Pass *pass = ....
  pm.addPass(pass);
  pm.addPass(pass);
  pm.run(module);

PiperOrigin-RevId: 263053082
2019-08-12 19:13:12 -07:00
Alex Zinenko 60965b4612 Move GPU dialect to {lib,include/mlir}/Dialect
Per tacit agreement, individual dialects should now live in lib/Dialect/Name
with headers in include/mlir/Dialect/Name and tests in test/Dialect/Name.

PiperOrigin-RevId: 259896851
2019-07-25 00:41:17 -07:00
River Riddle fec20e590f NFC: Rename Module to ModuleOp.
Module is a legacy name that only exists as a typedef of ModuleOp.

PiperOrigin-RevId: 257427248
2019-07-10 10:11:21 -07:00
River Riddle 6b6dc59f30 Update ModuleOp::create(...) to take a Location instead of a context.
This allows for giving a Module a more interesting location than 'Unknown'.

PiperOrigin-RevId: 257310117
2019-07-10 10:11:00 -07:00
River Riddle 8c44367891 NFC: Rename Function to FuncOp.
PiperOrigin-RevId: 257293379
2019-07-10 10:10:53 -07:00
River Riddle 626b8b6a5d NFC: Remove Module::getFunctions in favor of a general getOps<T>.
Modules can now contain more than just Functions, this just updates the iteration API to reflect that. The 'begin'/'end' methods have also been updated to iterate over opaque Operations.

PiperOrigin-RevId: 257099084
2019-07-08 18:28:17 -07:00