experiments/STT-tensorflow
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Propagate noalias and alignment properties of TensorFlow ABI into kernels.

Browse Source 
PiperOrigin-RevId: 317646198
Change-Id: I20a649c3127586106c250e213e6b6700fb302495
This commit is contained in:
Stephan Herhut 2020-06-22 06:59:05 -07:00 committed by TensorFlower Gardener
parent 3f28ac1b63
commit 12f5cd7dde
1 changed files with 16 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

23
tensorflow/compiler/mlir/tools/kernel_gen/cubin_creator.cc
View File

@ -135,11 +135,11 @@ Status LowerTfOpToLhloWithDynamicShapes(mlir::ModuleOp module) {
return Status::OK(); return Status::OK();
} }
struct PropagateStaticKnowledge struct PropagateTensorFlowABIKnowledge
: public mlir::PassWrapper<PropagateStaticKnowledge, : public mlir::PassWrapper<PropagateTensorFlowABIKnowledge,
mlir::OperationPass<mlir::LLVM::LLVMFuncOp>> { mlir::OperationPass<mlir::LLVM::LLVMFuncOp>> {
explicit PropagateStaticKnowledge(mlir::FunctionType type, explicit PropagateTensorFlowABIKnowledge(mlir::FunctionType type,
llvm::ArrayRef<uint32_t> same_shape_) llvm::ArrayRef<uint32_t> same_shape_)
: func_type(type), same_shape(same_shape_) {} : func_type(type), same_shape(same_shape_) {}
void runOnOperation() override { void runOnOperation() override {
@ -148,6 +148,11 @@ struct PropagateStaticKnowledge
// we insert constants into the code and replace usages accordingly. // we insert constants into the code and replace usages accordingly.
// We do not change the signature so that we keep a somewhat stable ABI // We do not change the signature so that we keep a somewhat stable ABI
// that is easy to undertand by tools. // that is easy to undertand by tools.
// We also know that tensorflow aligns all allocated pointers by 16, so
// we pass this on. Furthermore, we know that arguments never alias. More
// precicely, they may only alias (due to reuse) if the kernel does not
// read from a position it previously has written to. We express this with
// the noalias attribute.
mlir::LLVM::LLVMFuncOp func = getOperation(); mlir::LLVM::LLVMFuncOp func = getOperation();
// This only works if the function is local and we can rewrite it. // This only works if the function is local and we can rewrite it.
@ -172,6 +177,9 @@ struct PropagateStaticKnowledge
return; return;
} }
positions.push_back(arg_pos); positions.push_back(arg_pos);
// Set alignment and aliasing on the pointers.
func.setArgAttr(arg_pos + 1, "llvm.noalias", b.getBoolAttr(true));
func.setArgAttr(arg_pos + 1, "llvm.align", b.getIndexAttr(16));
// Replace the offset with zero. Offset is argument number 3. // Replace the offset with zero. Offset is argument number 3.
func.getArgument(arg_pos + 2).replaceAllUsesWith(zero); func.getArgument(arg_pos + 2).replaceAllUsesWith(zero);
// Forward over base_ptr, aligned_ptr, offset, size and stride arguments. // Forward over base_ptr, aligned_ptr, offset, size and stride arguments.
@ -213,7 +221,7 @@ struct PropagateStaticKnowledge
llvm::ArrayRef<uint32_t> same_shape; llvm::ArrayRef<uint32_t> same_shape;
}; };
Status PropagateStaticShapeKnowledgeToKernel( Status PropagateTensorFlowABIKnowledgeToKernel(
mlir::ModuleOp module, llvm::ArrayRef<uint32_t> same_shape) { mlir::ModuleOp module, llvm::ArrayRef<uint32_t> same_shape) {
// Grab the original signature from the single function. // Grab the original signature from the single function.
auto func = *module.getBody()->op_begin<mlir::FuncOp>(); auto func = *module.getBody()->op_begin<mlir::FuncOp>();
@ -228,7 +236,8 @@ Status PropagateStaticShapeKnowledgeToKernel(
/*printAfterOnlyOnChange=*/false, llvm::dbgs()); /*printAfterOnlyOnChange=*/false, llvm::dbgs());
auto& kernel_pm = pm.nest<::mlir::gpu::GPUModuleOp>(); auto& kernel_pm = pm.nest<::mlir::gpu::GPUModuleOp>();
kernel_pm.addNestedPass<mlir::LLVM::LLVMFuncOp>( kernel_pm.addNestedPass<mlir::LLVM::LLVMFuncOp>(
absl::make_unique<PropagateStaticKnowledge>(func.getType(), same_shape)); absl::make_unique<PropagateTensorFlowABIKnowledge>(func.getType(),
same_shape));
if (failed(pm.run(module))) { if (failed(pm.run(module))) {
return InternalError("Static knowledge propagation failed."); return InternalError("Static knowledge propagation failed.");
@ -263,7 +272,7 @@ StatusOr<std::vector<uint8_t>> tensorflow::kernel_gen::GenerateCubinForTfCode(
} }
TF_RETURN_IF_ERROR(xla::mlir_gpu::LowerKernelBodiesToNVVM(module.get())); TF_RETURN_IF_ERROR(xla::mlir_gpu::LowerKernelBodiesToNVVM(module.get()));
TF_RETURN_IF_ERROR( TF_RETURN_IF_ERROR(
PropagateStaticShapeKnowledgeToKernel(module.get(), same_shape)); PropagateTensorFlowABIKnowledgeToKernel(module.get(), same_shape));
mlir::OwningModuleRef kernel_module = mlir::OwningModuleRef kernel_module =
xla::mlir_gpu::ExtractKernelModule(*module).ValueOrDie(); xla::mlir_gpu::ExtractKernelModule(*module).ValueOrDie();